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/wait.h> /* wait */
62 #ifdef HAVE_SYS_IOCTL_H
63 #include <sys/ioctl.h>
65 #include <sys/param.h>
66 #ifdef HAVE_SYS_MOUNT_H
67 #include <sys/mount.h> /* For BLKGETSIZE */
69 #include <signal.h> /* sigaction */
70 #include <netinet/tcp.h>
71 #include <netinet/in.h> /* sockaddr_in, htons, in_addr */
72 #include <netdb.h> /* hostent, gethostby*, getservby* */
79 #include <arpa/inet.h>
87 /* used in cliserv.h, so must come first */
88 #define MY_NAME "nbd_server"
91 /** Default position of the config file */
93 #define SYSCONFDIR "/etc"
95 #define CFILE SYSCONFDIR "/nbd-server/config"
97 /** Where our config file actually is */
98 gchar* config_file_pos;
100 /** how much space for child PIDs we have by default. Dynamically
101 allocated, and will be realloc()ed if out of space, so this should
102 probably be fair for most situations. */
103 #define DEFAULT_CHILD_ARRAY 256
105 /** Logging macros, now nothing goes to syslog unless you say ISSERVER */
107 #define msg2(a,b) syslog(a,b)
108 #define msg3(a,b,c) syslog(a,b,c)
109 #define msg4(a,b,c,d) syslog(a,b,c,d)
111 #define msg2(a,b) g_message(b)
112 #define msg3(a,b,c) g_message(b,c)
113 #define msg4(a,b,c,d) g_message(b,c,d)
116 /* Debugging macros */
119 #define DEBUG( a ) printf( a )
120 #define DEBUG2( a,b ) printf( a,b )
121 #define DEBUG3( a,b,c ) printf( a,b,c )
124 #define DEBUG2( a,b )
125 #define DEBUG3( a,b,c )
127 #ifndef PACKAGE_VERSION
128 #define PACKAGE_VERSION ""
131 * The highest value a variable of type off_t can reach.
133 /* This is starting to get ugly. If someone knows a better way to find
134 * the maximum value of a signed type *without* relying on overflow
135 * (doing so breaks on 64bit architectures), that would be nice.
137 * Actually, do we need this at all? Can't we just say '0 is autodetect', and
138 * live with it? Or better yet, use an extra flag, or so?
139 * Answer: yes, we need it, as the hunksize is defined to this when the
140 * multiple file thingy isn't used.
142 #define OFFT_MAX (((((off_t)1)<<((sizeof(off_t)-1)*8))-1)<<7)+127
143 #define LINELEN 256 /**< Size of static buffer used to read the
144 authorization file (yuck) */
145 #define BUFSIZE (1024*1024) /**< Size of buffer that can hold requests */
146 #define GIGA (1*1024*1024*1024) /**< 1 Gigabyte. Used as hunksize when doing
147 the multiple file thingy. @todo: make this a
148 configuration option. */
149 #define DIFFPAGESIZE 4096 /**< diff file uses those chunks */
150 #define F_READONLY 1 /**< flag to tell us a file is readonly */
151 #define F_MULTIFILE 2 /**< flag to tell us a file is exported using -m */
152 #define F_COPYONWRITE 4 /**< flag to tell us a file is exported using
154 #define F_AUTOREADONLY 8 /**< flag to tell us a file is set to autoreadonly */
155 GHashTable *children;
156 char pidfname[256]; /**< name of our PID file */
157 char default_authname[] = "/etc/nbd_server.allow"; /**< default name of allow file */
160 * Variables associated with a server.
163 gchar* exportname; /**< (unprocessed) filename of the file we're exporting */
164 off_t hunksize; /**< size of a hunk of an exported file */
165 off_t expected_size; /**< size of the exported file as it was told to
166 us through configuration */
167 unsigned int port; /**< port we're exporting this file at */
168 char* authname; /**< filename of the authorization file */
169 int flags; /**< flags associated with this exported file */
170 unsigned int timeout;/**< how long a connection may be idle
172 int socket; /**< The socket of this server. */
176 * Variables associated with a client socket.
179 off_t exportsize; /**< size of the file we're exporting */
180 char *clientname; /**< peer */
181 char *exportname; /**< (processed) filename of the file we're exporting */
182 GArray *export; /**< array of filedescriptors of exported files;
183 only the first is actually used unless we're
184 doing the multiple file option */
185 int net; /**< The actual client socket */
186 SERVER *server; /**< The server this client is getting data from */
187 char* difffilename; /**< filename of the copy-on-write file, if any */
188 int difffile; /**< filedescriptor of copyonwrite file. @todo
189 shouldn't this be an array too? (cfr export) Or
190 make -m and -c mutually exclusive */
191 u32 difffilelen; /**< number of pages in difffile */
192 u32 *difmap; /**< see comment on the global difmap for this one */
196 * Type of configuration file values
199 PARAM_INT, /**< This parameter is an integer */
200 PARAM_STRING, /**< This parameter is a string */
201 PARAM_BOOL, /**< This parameter is a boolean */
204 * Configuration file values
207 gchar *paramname; /**< Name of the parameter, as it appears in
209 gboolean required; /**< Whether this is a required (as opposed to
210 optional) parameter */
211 PARAM_TYPE ptype; /**< Type of the parameter. */
212 gpointer target; /**< Pointer to where the data of this
213 parameter should be written. If ptype is
214 PARAM_BOOL, the data is or'ed rather than
216 gint flagval; /**< Flag mask for this parameter in case ptype
221 * Check whether a client is allowed to connect. Works with an authorization
222 * file which contains one line per machine, no wildcards.
224 * @param opts The client who's trying to connect.
225 * @return 0 - authorization refused, 1 - OK
227 int authorized_client(CLIENT *opts) {
232 if ((f=fopen(opts->server->authname,"r"))==NULL) {
233 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
234 opts->server->authname,strerror(errno)) ;
238 while (fgets(line,LINELEN,f)!=NULL) {
239 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
249 * Read data from a file descriptor into a buffer
251 * @param f a file descriptor
252 * @param buf a buffer
253 * @param len the number of bytes to be read
255 inline void readit(int f, void *buf, size_t len) {
259 if ((res = read(f, buf, len)) <= 0)
260 err("Read failed: %m");
267 * Write data from a buffer into a filedescriptor
269 * @param f a file descriptor
270 * @param buf a buffer containing data
271 * @param len the number of bytes to be written
273 inline void writeit(int f, void *buf, size_t len) {
277 if ((res = write(f, buf, len)) <= 0)
278 err("Send failed: %m");
285 * Print out a message about how to use nbd-server. Split out to a separate
286 * function so that we can call it from multiple places
289 printf("This is nbd-server version " VERSION "\n");
290 printf("Usage: port file_to_export [size][kKmM] [-l authorize_file] [-r] [-m] [-c] [-a timeout_sec]\n"
291 "\t-r|--read-only\t\tread only\n"
292 "\t-m|--multi-file\t\tmultiple file\n"
293 "\t-c|--copy-on-write\tcopy on write\n"
294 "\t-l|--authorize-file\tfile with list of hosts that are allowed to\n\t\t\t\tconnect.\n"
295 "\t-a|--idle-time\t\tmaximum idle seconds; server terminates when\n\t\t\t\tidle time exceeded\n\n"
296 "\tif port is set to 0, stdin is used (for running from inetd)\n"
297 "\tif file_to_export contains '%%s', it is substituted with the IP\n"
298 "\t\taddress of the machine trying to connect\n" );
299 printf("Using configuration file %s\n", CFILE);
303 * Parse the command line.
305 * @param argc the argc argument to main()
306 * @param argv the argv argument to main()
308 SERVER* cmdline(int argc, char *argv[]) {
312 struct option long_options[] = {
313 {"read-only", no_argument, NULL, 'r'},
314 {"multi-file", no_argument, NULL, 'm'},
315 {"copy-on-write", no_argument, NULL, 'c'},
316 {"authorize-file", required_argument, NULL, 'l'},
317 {"idle-time", required_argument, NULL, 'a'},
318 {"config-file", required_argument, NULL, 'C'},
329 serve=g_new0(SERVER, 1);
330 serve->hunksize=OFFT_MAX;
331 serve->authname = g_strdup(default_authname);
332 while((c=getopt_long(argc, argv, "-a:C:cl:mr", long_options, &i))>=0) {
335 /* non-option argument */
336 switch(nonspecial++) {
338 serve->port=strtol(optarg, NULL, 0);
341 serve->exportname = g_strdup(optarg);
342 if(serve->exportname[0] != '/') {
343 fprintf(stderr, "E: The to be exported file needs to be an absolute filename!\n");
348 last=strlen(optarg)-1;
350 if (suffix == 'k' || suffix == 'K' ||
351 suffix == 'm' || suffix == 'M')
353 es = (off_t)atol(optarg);
361 serve->expected_size = es;
366 serve->flags |= F_READONLY;
369 serve->flags |= F_MULTIFILE;
370 serve->hunksize = 1*GIGA;
373 serve->flags |=F_COPYONWRITE;
376 g_free(config_file_pos);
377 config_file_pos=g_strdup(optarg);
380 g_free(serve->authname);
381 serve->authname=g_strdup(optarg);
384 serve->timeout=strtol(optarg, NULL, 0);
392 /* What's left: the port to export, the name of the to be exported
393 * file, and, optionally, the size of the file, in that order. */
402 * Error codes for config file parsing
405 CFILE_NOTFOUND, /**< The configuration file is not found */
406 CFILE_MISSING_GENERIC, /**< The (required) group "generic" is missing */
407 CFILE_KEY_MISSING, /**< A (required) key is missing */
408 CFILE_VALUE_INVALID, /**< A value is syntactically invalid */
409 CFILE_PROGERR /**< Programmer error */
413 * Remove a SERVER from memory. Used from the hash table
415 void remove_server(gpointer s) {
419 g_free(server->exportname);
421 g_free(server->authname);
426 * Parse the config file.
428 * @param f the name of the config file
429 * @param e a GError. @see CFILE_ERRORS for what error values this function can
431 * @return a GHashTable of SERVER* pointers, with the port number as the hash
432 * key. If the config file is empty or does not exist, returns an empty
433 * GHashTable; if the config file contains an error, returns NULL, and
434 * e is set appropriately
436 GHashTable* parse_cfile(gchar* f, GError** e) {
439 { "exportname", TRUE, PARAM_STRING, NULL, 0 },
440 { "port", TRUE, PARAM_INT, NULL, 0 },
441 { "authfile", FALSE, PARAM_STRING, NULL, 0 },
442 { "timeout", FALSE, PARAM_INT, NULL, 0 },
443 { "filesize", FALSE, PARAM_INT, NULL, 0 },
444 { "readonly", FALSE, PARAM_BOOL, NULL, F_READONLY },
445 { "multifile", FALSE, PARAM_BOOL, NULL, F_MULTIFILE },
446 { "copyonwrite", FALSE, PARAM_BOOL, NULL, F_COPYONWRITE },
456 errdomain = g_quark_from_string("parse_cfile");
457 cfile = g_key_file_new();
458 retval = g_hash_table_new_full(g_int_hash, g_int_equal, NULL, remove_server);
459 if(!g_key_file_load_from_file(cfile, f, G_KEY_FILE_KEEP_COMMENTS |
460 G_KEY_FILE_KEEP_TRANSLATIONS, &err)) {
461 g_set_error(e, errdomain, CFILE_NOTFOUND, "Could not open config file.");
462 g_key_file_free(cfile);
465 if(strcmp(g_key_file_get_start_group(cfile), "generic")) {
466 g_set_error(e, errdomain, CFILE_MISSING_GENERIC, "Config file does not contain the [generic] group!");
467 g_key_file_free(cfile);
470 groups = g_key_file_get_groups(cfile, NULL);
471 for(i=0;groups[i];i++) {
473 p[0].target=&(s->exportname);
474 p[1].target=&(s->port);
475 p[2].target=&(s->authname);
476 p[3].target=&(s->timeout);
477 p[4].target=&(s->expected_size);
478 p[5].target=p[6].target=p[7].target=p[8].target=&(s->flags);
480 g_assert(p[j].target != NULL);
481 g_assert(p[j].ptype==PARAM_INT||p[j].ptype==PARAM_STRING||p[j].ptype==PARAM_BOOL);
484 *((gint*)p[j].target) = g_key_file_get_integer(cfile, groups[i], p[j].paramname, &err);
487 *((gchar**)p[j].target) = g_key_file_get_string(cfile, groups[i], p[j].paramname, &err);
490 value = g_key_file_get_boolean(cfile, groups[i], p[j].paramname, &err);
492 *((gint*)p[j].target) |= value;
497 if(err->code == G_KEY_FILE_ERROR_KEY_NOT_FOUND) {
499 g_set_error(e, errdomain, CFILE_KEY_MISSING, "Could not find required value %s in group %s: %s", p[j].paramname, groups[i], err->message);
500 g_hash_table_destroy(retval);
502 g_key_file_free(cfile);
509 g_set_error(e, errdomain, CFILE_VALUE_INVALID, "Could not parse %s in group %s: %s", p[j].paramname, groups[i], err->message);
510 g_hash_table_destroy(retval);
512 g_key_file_free(cfile);
518 g_hash_table_insert(retval, &(s->port), s);
524 * Signal handler for SIGCHLD
525 * @param s the signal we're handling (must be SIGCHLD, or something
528 void sigchld_handler(int s) {
533 while((pid=wait(status)) > 0) {
534 if(WIFEXITED(status)) {
535 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
537 i=g_hash_table_lookup(children, &pid);
539 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
541 DEBUG2("Removing %d from the list of children", pid);
542 g_hash_table_remove(children, &pid);
548 * Kill a child. Called from sigterm_handler::g_hash_table_foreach.
551 * @param value the value corresponding to the above key
552 * @param user_data a pointer which we always set to 1, so that we know what
555 void killchild(gpointer key, gpointer value, gpointer user_data) {
557 int *parent=user_data;
564 * Handle SIGTERM and dispatch it to our children
565 * @param s the signal we're handling (must be SIGTERM, or something
566 * is severely wrong).
568 void sigterm_handler(int s) {
571 g_hash_table_foreach(children, killchild, &parent);
581 * Detect the size of a file.
583 * @param export An open filedescriptor
584 * @return the size of the file, or OFFT_MAX if detection was
587 off_t size_autodetect(int export) {
590 struct stat stat_buf;
593 #ifdef HAVE_SYS_MOUNT_H
594 #ifdef HAVE_SYS_IOCTL_H
596 DEBUG("looking for export size with ioctl BLKGETSIZE\n");
597 if (!ioctl(export, BLKGETSIZE, &es32) && es32) {
598 es = (off_t)es32 * (off_t)512;
601 #endif /* BLKGETSIZE */
602 #endif /* HAVE_SYS_IOCTL_H */
603 #endif /* HAVE_SYS_MOUNT_H */
605 DEBUG("looking for export size with fstat\n");
606 stat_buf.st_size = 0;
607 error = fstat(export, &stat_buf);
609 if(stat_buf.st_size > 0)
610 return (off_t)stat_buf.st_size;
612 err("fstat failed: %m");
615 DEBUG("looking for export size with lseek SEEK_END\n");
616 es = lseek(export, (off_t)0, SEEK_END);
617 if (es > ((off_t)0)) {
620 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
623 err("Could not find size of exported block device: %m");
628 * seek to a position in a file, with error handling.
629 * @param handle a filedescriptor
630 * @param a position to seek to
631 * @todo get rid of this; lastpoint is a global variable right now, but it
632 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
635 void myseek(int handle,off_t a) {
636 if (lseek(handle, a, SEEK_SET) < 0) {
637 err("Can not seek locally!\n");
642 * Write an amount of bytes at a given offset to the right file. This
643 * abstracts the write-side of the multiple file option.
645 * @param a The offset where the write should start
646 * @param buf The buffer to write from
647 * @param len The length of buf
648 * @param client The client we're serving for
649 * @return The number of bytes actually written, or -1 in case of an error
651 int rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client) {
654 myseek(g_array_index(client->export, int, (int)(a/client->server->hunksize)), a%client->server->hunksize);
656 res = write(g_array_index(client->export, int, (int)((off_t)a/(off_t)(client->server->hunksize))), buf, len);
657 return (res < 0 || (size_t)res != len);
661 * Read an amount of bytes at a given offset from the right file. This
662 * abstracts the read-side of the multiple files option.
664 * @param a The offset where the read should start
665 * @param buf A buffer to read into
666 * @param len The size of buf
667 * @param client The client we're serving for
668 * @return The number of bytes actually read, or -1 in case of an
671 int rawexpread(off_t a, char *buf, size_t len, CLIENT *client) {
674 myseek(g_array_index(client->export,int,(int)a/client->server->hunksize),
675 a%client->server->hunksize);
676 res = read(g_array_index(client->export,int,(int)a/client->server->hunksize), buf, len);
677 return (res < 0 || (size_t)res != len);
681 * Read an amount of bytes at a given offset from the right file. This
682 * abstracts the read-side of the copyonwrite stuff, and calls
683 * rawexpread() with the right parameters to do the actual work.
684 * @param a The offset where the read should start
685 * @param buf A buffer to read into
686 * @param len The size of buf
687 * @param client The client we're going to read for
688 * @return The number of bytes actually read, or -1 in case of an error
690 int expread(off_t a, char *buf, size_t len, CLIENT *client) {
692 off_t mapcnt, mapl, maph, pagestart;
694 if (!(client->server->flags & F_COPYONWRITE))
695 return rawexpread(a, buf, len, client);
696 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
698 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
700 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
701 pagestart=mapcnt*DIFFPAGESIZE;
703 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
704 len : (size_t)DIFFPAGESIZE-offset;
705 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
706 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
707 (unsigned long)(client->difmap[mapcnt]));
708 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
709 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
710 } else { /* the block is not there */
711 DEBUG2("Page %Lu is not here, we read the original one\n",
712 (unsigned long long)mapcnt);
713 if(rawexpread(a, buf, rdlen, client)) return -1;
715 len-=rdlen; a+=rdlen; buf+=rdlen;
721 * Write an amount of bytes at a given offset to the right file. This
722 * abstracts the write-side of the copyonwrite option, and calls
723 * rawexpwrite() with the right parameters to do the actual work.
725 * @param a The offset where the write should start
726 * @param buf The buffer to write from
727 * @param len The length of buf
728 * @param client The client we're going to write for.
729 * @return The number of bytes actually written, or -1 in case of an error
731 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
732 char pagebuf[DIFFPAGESIZE];
733 off_t mapcnt,mapl,maph;
738 if (!(client->server->flags & F_COPYONWRITE))
739 return(rawexpwrite(a,buf,len, client));
740 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
742 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
744 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
745 pagestart=mapcnt*DIFFPAGESIZE ;
747 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
748 len : (size_t)DIFFPAGESIZE-offset;
750 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
751 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
752 (unsigned long)(client->difmap[mapcnt])) ;
753 myseek(client->difffile,
754 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
755 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
756 } else { /* the block is not there */
757 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
758 client->difmap[mapcnt]=client->difffilelen++ ;
759 DEBUG3("Page %Lu is not here, we put it at %lu\n",
760 (unsigned long long)mapcnt,
761 (unsigned long)(client->difmap[mapcnt]));
763 if (rdlen+pagestart%(client->server->hunksize) >
764 (client->server->hunksize))
765 rdlen=client->server->hunksize -
766 (pagestart%client->server->hunksize);
767 if (rawexpread(pagestart, pagebuf, rdlen, client))
769 memcpy(pagebuf+offset,buf,wrlen) ;
770 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
774 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
780 * Do the initial negotiation.
782 * @param client The client we're negotiating with.
784 void negotiate(CLIENT *client) {
788 memset(zeros, 0, 290);
789 if (write(client->net, INIT_PASSWD, 8) < 0)
790 err("Negotiation failed: %m");
791 cliserv_magic = htonll(cliserv_magic);
792 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
793 err("Negotiation failed: %m");
794 size_host = htonll((u64)(client->exportsize));
795 if (write(client->net, &size_host, 8) < 0)
796 err("Negotiation failed: %m");
797 if (write(client->net, zeros, 128) < 0)
798 err("Negotiation failed: %m");
801 /** sending macro. */
802 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
804 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; }
806 * Serve a file to a single client.
808 * @todo This beast needs to be split up in many tiny little manageable
809 * pieces. Preferably with a chainsaw.
811 * @param client The client we're going to serve to.
814 int mainloop(CLIENT *client) {
815 struct nbd_request request;
816 struct nbd_reply reply;
822 DEBUG("Entering request loop!\n");
823 reply.magic = htonl(NBD_REPLY_MAGIC);
832 if (client->server->timeout)
833 alarm(client->server->timeout);
834 readit(client->net, &request, sizeof(request));
835 request.from = ntohll(request.from);
836 request.type = ntohl(request.type);
838 if (request.type==NBD_CMD_DISC) {
839 msg2(LOG_INFO, "Disconnect request received.");
840 if (client->difmap) g_free(client->difmap) ;
841 if (client->difffile>=0) {
842 close(client->difffile);
843 unlink(client->difffilename);
844 free(client->difffilename);
850 len = ntohl(request.len);
852 if (request.magic != htonl(NBD_REQUEST_MAGIC))
853 err("Not enough magic.");
854 if (len > BUFSIZE + sizeof(struct nbd_reply))
855 err("Request too big!");
857 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
858 "READ", (unsigned long long)request.from,
859 (unsigned long long)request.from / 512, len);
861 memcpy(reply.handle, request.handle, sizeof(reply.handle));
862 if ((request.from + len) > (OFFT_MAX)) {
863 DEBUG("[Number too large!]");
864 ERROR(client, reply);
868 if (((ssize_t)((off_t)request.from + len) > client->exportsize) ||
869 ((client->server->flags & F_READONLY) && request.type)) {
871 ERROR(client, reply);
875 if (request.type==NBD_CMD_WRITE) {
876 DEBUG("wr: net->buf, ");
877 readit(client->net, buf, len);
879 if ((client->server->flags & F_AUTOREADONLY) ||
880 expwrite(request.from, buf, len,
882 DEBUG("Write failed: %m" );
883 ERROR(client, reply);
886 SEND(client->net, reply);
893 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
894 DEBUG("Read failed: %m");
895 ERROR(client, reply);
900 memcpy(buf, &reply, sizeof(struct nbd_reply));
901 writeit(client->net, buf, len + sizeof(struct nbd_reply));
908 * Split a single exportfile into multiple ones, if that was asked.
909 * @return 0 on success, -1 on failure
910 * @param client information on the client which we want to split
912 int splitexport(CLIENT* client) {
916 client->export = g_array_new(TRUE, TRUE, sizeof(int));
917 for (i=0; i<client->exportsize; i+=client->server->hunksize) {
920 if(client->server->flags & F_MULTIFILE) {
921 tmpname=g_strdup_printf("%s.%d", client->exportname,
922 (int)(i/client->server->hunksize));
924 tmpname=g_strdup(client->exportname);
926 DEBUG2( "Opening %s\n", tmpname );
927 if((fhandle = open(tmpname, (client->server->flags & F_READONLY) ? O_RDONLY : O_RDWR)) == -1) {
928 /* Read WRITE ACCESS was requested by media is only read only */
929 client->server->flags |= F_AUTOREADONLY;
930 client->server->flags |= F_READONLY;
931 if((fhandle = open(tmpname, O_RDONLY)) == -1)
932 err("Could not open exported file: %m");
934 g_array_insert_val(client->export,i/client->server->hunksize,fhandle);
939 int copyonwrite_prepare(CLIENT* client)
942 if ((client->difffilename = malloc(1024))==NULL)
943 err("Failed to allocate string for diff file name");
944 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
946 client->difffilename[1023]='\0';
947 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
948 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
949 if (client->difffile<0) err("Could not create diff file (%m)") ;
950 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
951 err("Could not allocate memory") ;
952 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
958 * Serve a connection.
960 * @todo allow for multithreading, perhaps use libevent. Not just yet, though;
961 * follow the road map.
963 * @param client a connected client
965 void serveconnection(CLIENT *client) {
968 if (!client->server->expected_size) {
969 client->exportsize = size_autodetect(g_array_index(client->export,int,0));
971 /* Perhaps we should check first. Not now. */
972 client->exportsize = client->server->expected_size;
974 if (client->exportsize > OFFT_MAX) {
975 /* uhm, well... In a parallel universe, this *might* be
977 err("Size of exported file is too big\n");
980 msg3(LOG_INFO, "size of exported file/device is %Lu", (unsigned long long)client->exportsize);
983 if (client->server->flags & F_COPYONWRITE) {
984 copyonwrite_prepare(client);
987 setmysockopt(client->net);
993 * Find the name of the file we have to serve. This will use g_strdup_printf
994 * to put the IP address of the client inside a filename containing
995 * "%s". That name is then written to client->exportname.
997 * @param net A socket connected to an nbd client
998 * @param client information about the client. The IP address in human-readable
999 * format will be written to a new char* buffer, the address of which will be
1000 * stored in client->clientname.
1002 void set_peername(int net, CLIENT *client) {
1003 struct sockaddr_in addrin;
1004 int addrinlen = sizeof( addrin );
1007 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
1008 err("getsockname failed: %m");
1009 peername = inet_ntoa(addrin.sin_addr);
1010 client->exportname=g_strdup_printf(client->server->exportname, peername);
1012 msg4(LOG_INFO, "connect from %s, assigned file is %s",
1013 peername, client->exportname);
1014 client->clientname=g_strdup(peername);
1019 * @param data a pointer to pid_t which should be freed
1021 void destroy_pid_t(gpointer data) {
1026 * Go daemon (unless we specified at compile time that we didn't want this)
1027 * @param serve the first server of our configuration. If its port is zero,
1028 * then do not daemonize, because we're doing inetd then. This parameter
1029 * is only used to create a PID file of the form
1030 * /var/run/nbd-server.<port>.pid; it's not modified in any way.
1032 #if !defined(NODAEMON) && !defined(NOFORK)
1033 void daemonize(SERVER* serve) {
1040 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
1042 strncpy(pidfname, "/var/run/nbd-server.pid", sizeof(char)*255);
1044 pidf=fopen(pidfname, "w");
1046 fprintf(pidf,"%d\n", (int)getpid());
1050 fprintf(stderr, "Not fatal; continuing");
1054 #define daemonize(serve)
1055 #endif /* !defined(NODAEMON) && !defined(NOFORK) */
1058 * Connect a server's socket.
1060 * @todo modularize this giant beast.
1062 * @param port the port we're connecting to (we don't need it, but
1063 * g_hash_table_foreach requires it)
1064 * @param serve the server we want to connect.
1065 * @param data (unused) user data.
1067 void setup_serve(gpointer port, gpointer s, gpointer data) {
1068 struct sockaddr_in addrin;
1069 struct sigaction sa;
1070 int addrinlen = sizeof(addrin);
1076 SERVER *serve=(SERVER*)s;
1078 g_assert(data==NULL);
1079 if ((serve->socket = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
1082 /* lose the pesky "Address already in use" error message */
1083 if (setsockopt(serve->socket,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
1084 err("setsockopt SO_REUSEADDR");
1086 if (setsockopt(serve->socket,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
1087 err("setsockopt SO_KEEPALIVE");
1090 DEBUG("Waiting for connections... bind, ");
1091 addrin.sin_family = AF_INET;
1092 addrin.sin_port = htons(serve->port);
1093 addrin.sin_addr.s_addr = 0;
1094 if (bind(serve->socket, (struct sockaddr *) &addrin, addrinlen) < 0)
1097 if (listen(serve->socket, 1) < 0)
1099 sa.sa_handler = sigchld_handler;
1100 sigemptyset(&sa.sa_mask);
1101 sa.sa_flags = SA_RESTART;
1102 if(sigaction(SIGCHLD, &sa, NULL) == -1)
1103 err("sigaction: %m");
1104 sa.sa_handler = sigterm_handler;
1105 sigemptyset(&sa.sa_mask);
1106 sa.sa_flags = SA_RESTART;
1107 if(sigaction(SIGTERM, &sa, NULL) == -1)
1108 err("sigaction: %m");
1109 children=g_hash_table_new_full(g_int_hash, g_int_equal, NULL, destroy_pid_t);
1113 * Connect our servers.
1115 void setup_servers(GHashTable* servers) {
1116 g_hash_table_foreach(servers, setup_serve, NULL);
1120 * Loop through the available servers, and serve them.
1122 * Actually, right now we only handle one server. Will change that for
1125 int serveloop(SERVER* serve) {
1126 struct sockaddr_in addrin;
1127 socklen_t addrinlen=sizeof(addrin);
1134 if ((net = accept(serve->socket, (struct sockaddr *) &addrin, &addrinlen)) < 0)
1137 client = g_malloc(sizeof(CLIENT));
1138 client->server=serve;
1139 client->exportsize=OFFT_MAX;
1141 set_peername(net, client);
1142 if (!authorized_client(client)) {
1143 msg2(LOG_INFO,"Unauthorized client") ;
1147 msg2(LOG_INFO,"Authorized client") ;
1148 pid=g_malloc(sizeof(pid_t));
1150 if ((*pid=fork())<0) {
1151 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
1155 if (*pid>0) { /* parent */
1157 g_hash_table_insert(children, pid, pid);
1161 g_hash_table_destroy(children);
1162 close(serve->socket) ;
1164 msg2(LOG_INFO,"Starting to serve") ;
1165 serveconnection(client);
1170 * Main entry point...
1172 int main(int argc, char *argv[]) {
1174 GHashTable *servers;
1177 if (sizeof( struct nbd_request )!=28) {
1178 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1183 config_file_pos = g_strdup(CFILE);
1184 serve=cmdline(argc, argv);
1185 servers = parse_cfile(config_file_pos, &err);
1187 g_critical("Could not parse command file: %s", err->message);
1190 g_hash_table_insert(servers, &serve->port, serve);
1193 /* We don't support this at this time */
1195 if (!(serve->port)) {
1198 /* You really should define ISSERVER if you're going to use
1199 * inetd mode, but if you don't, closing stdout and stderr
1200 * (which inetd had connected to the client socket) will let it
1204 open("/dev/null", O_WRONLY);
1205 open("/dev/null", O_WRONLY);
1207 client=g_malloc(sizeof(CLIENT));
1208 client->server=serve;
1210 client->exportsize=OFFT_MAX;
1211 set_peername(0,client);
1212 serveconnection(client);
1216 if((!serve) && (!servers)) {
1217 g_message("Nothing to do! Bye!");
1221 setup_servers(servers);