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 */
59 #include <sys/types.h>
60 #include <sys/socket.h>
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 /** how much space for child PIDs we have by default. Dynamically
93 allocated, and will be realloc()ed if out of space, so this should
94 probably be fair for most situations. */
95 #define DEFAULT_CHILD_ARRAY 256
97 /** Logging macros, now nothing goes to syslog unless you say ISSERVER */
99 #define msg2(a,b) syslog(a,b)
100 #define msg3(a,b,c) syslog(a,b,c)
101 #define msg4(a,b,c,d) syslog(a,b,c,d)
103 #define msg2(a,b) g_message(b)
104 #define msg3(a,b,c) g_message(b,c)
105 #define msg4(a,b,c,d) g_message(b,c,d)
108 /* Debugging macros */
111 #define DEBUG( a ) printf( a )
112 #define DEBUG2( a,b ) printf( a,b )
113 #define DEBUG3( a,b,c ) printf( a,b,c )
116 #define DEBUG2( a,b )
117 #define DEBUG3( a,b,c )
119 #ifndef PACKAGE_VERSION
120 #define PACKAGE_VERSION ""
123 * The highest value a variable of type off_t can reach.
125 /* This is starting to get ugly. If someone knows a better way to find
126 * the maximum value of a signed type *without* relying on overflow
127 * (doing so breaks on 64bit architectures), that would be nice.
129 * Actually, do we need this at all? Can't we just say '0 is autodetect', and
130 * live with it? Or better yet, use an extra flag, or so?
132 #define OFFT_MAX (((((off_t)1)<<((sizeof(off_t)-1)*8))-1)<<7)+127
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 GIGA (1*1024*1024*1024) /**< 1 Gigabyte. Used as hunksize when doing
137 the multiple file thingy */
138 #define DIFFPAGESIZE 4096 /**< diff file uses those chunks */
139 #define F_READONLY 1 /**< flag to tell us a file is readonly */
140 #define F_MULTIFILE 2 /**< flag to tell us a file is exported using -m */
141 #define F_COPYONWRITE 4 /**< flag to tell us a file is exported using copyonwrite */
142 #define F_AUTOREADONLY 8 /**< flag to tell us a file is set to autoreadonly */
143 //char difffilename[1024]; /**< filename of the copy-on-write file. Doesn't belong here! */
144 //unsigned int timeout = 0; /**< disconnect timeout */
145 //int autoreadonly = 0; /**< 1 = switch to readonly if opening readwrite isn't
147 //char *auth_file="nbd_server.allow"; /**< authorization file */
148 //char exportname2[1024]; /**< File I'm exporting, with virtualhost resolved */
149 //off_t lastpoint = (off_t)-1; /**< keep track of where we are in the file, to
150 // avoid an lseek if possible */
151 //char pagebuf[DIFFPAGESIZE]; /**< when doing copyonwrite, this is
152 // used as a temporary buffer to store
153 // the exported block in. @todo this is
154 // a great example of namespace
155 // pollution. Throw it out. */
156 //unsigned int port; /**< Port I'm listening at */
157 //char *exportname; /**< File I'm exporting */
158 //off_t exportsize = OFFT_MAX; /**< length of file I'm exporting */
159 //off_t hunksize = OFFT_MAX; /**< size of each exported file in case of -m */
160 //int flags = 0; /**< flags associated with this exported file */
161 //int export[1024];/**< array of filedescriptors of exported files; only first is
162 // used unless -m option is activated */
163 //int difffile=-1; /**< filedescriptor for copyonwrite file */
164 //u32 difffilelen=0 ; /**< number of pages in difffile */
165 //u32 *difmap=NULL ; /**< Determine whether a block is in the original file
166 // (difmap[block]==-1) or in the copyonwrite file (in which
167 // case it contains the offset where it is to be found in the
168 // copyonwrite file). @todo the kernel knows about sparse
169 // files, we should use those instead. Should also be off_t
170 // instead of u32; copyonwrite is probably broken wrt LFS */
171 char clientname[256] ;
172 int child_arraysize=DEFAULT_CHILD_ARRAY; /**< number of available slots for
174 pid_t *children; /**< child array */
175 char pidfname[256]; /**< name of our PID file */
178 * Variables associated with a server.
181 char* exportname; /**< (unprocessed) filename of the file we're exporting */
182 off_t hunksize; /**< size of a hunk of an exported file */
183 off_t expected_size; /**< size of the exported file as it was told to
184 us through configuration */
185 unsigned int port; /**< port we're exporting this file at */
186 char* authname; /**< filename of the authorization file */
187 int flags; /**< flags associated with this exported file */
188 unsigned int timeout;/**< how long a connection may be idle
193 * Variables associated with a client socket.
196 off_t exportsize; /**< size of the file we're exporting */
197 char *clientname; /**< peer */
198 char *exportname; /**< (processed) filename of the file we're exporting */
199 int export[1024]; /**< array of filedescriptors of exported files;
200 only the first is actually used unless we're
201 doing the multiple file option */
202 int lastpoint; /**< For keeping track of where we are in a file.
203 This code is BUGGY currently, at least in
204 combination with the multiple file option. */
205 int net; /**< The actual client socket */
206 SERVER *server; /**< The server this client is getting data from */
207 char* difffilename; /**< filename of the copy-on-write file, if any */
208 int difffile; /**< filedescriptor of copyonwrite file. @todo
209 shouldn't this be an array too? (cfr
210 nbd_server_opts::export) Or make -m and -c
211 mutually exclusive */
212 u32 difffilelen; /**< number of pages in difffile */
213 u32 *difmap; /**< see comment on the global difmap for this one */
217 * Check whether a client is allowed to connect. Works with an authorization
218 * file which contains one line per machine, no wildcards.
220 * @param name IP address of client trying to connect (in human-readable form)
221 * @return 0 - authorization refused, 1 - OK
223 int authorized_client(CLIENT *opts) {
228 if ((f=fopen(opts->server->authname,"r"))==NULL) {
229 msg4(LOG_INFO,"Can't open authorization file %s (%s).",
230 opts->server->authname,strerror(errno)) ;
234 while (fgets(line,LINELEN,f)!=NULL) {
235 if (strncmp(line,opts->clientname,strlen(opts->clientname))==0) {
245 * Read data from a file descriptor into a buffer
247 * @param f a file descriptor
248 * @param buf a buffer
249 * @param len the number of bytes to be read
251 inline void readit(int f, void *buf, size_t len)
256 if ((res = read(f, buf, len)) <= 0)
257 err("Read failed: %m");
264 * Write data from a buffer into a filedescriptor
266 * @param f a file descriptor
267 * @param buf a buffer containing data
268 * @param len the number of bytes to be written
270 inline void writeit(int f, void *buf, size_t len)
275 if ((res = write(f, buf, len)) <= 0)
276 err("Send failed: %m");
283 * Print out a message about how to use nbd-server. Split out to a separate
284 * function so that we can call it from multiple places
287 printf("This is nbd-server version " VERSION "\n");
288 printf("Usage: port file_to_export [size][kKmM] [-r] [-m] [-c] [-a timeout_sec]\n"
290 " -m multiple file\n"
291 " -c copy on write\n"
292 " -l file with list of hosts that are allowed to connect.\n"
293 " -a maximum idle seconds, terminates when idle time exceeded\n"
294 " if port is set to 0, stdin is used (for running from inetd)\n"
295 " if file_to_export contains '%%s', it is substituted with IP\n"
296 " address of machine trying to connect\n" );
300 * Parse the command line.
302 * @todo getopt() is a great thing, and easy to use. Also, we want to
303 * create a configuration file which nbd-server will read. Maybe do (as in,
306 * @param argc the argc argument to main()
307 * @param argv the argv argument to main()
309 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'},
322 serve=g_malloc(sizeof(SERVER));
323 serve->hunksize=OFFT_MAX;
324 while((c=getopt_long(argc, argv, "a:cl:mr", long_options, &i))>=0) {
327 serve->flags |= F_READONLY;
330 serve->flags |= F_MULTIFILE;
331 serve->hunksize = 1*GIGA;
334 serve->flags |=F_COPYONWRITE;
337 serve->authname=optarg;
340 serve->timeout=strtol(optarg, NULL, 0);
348 /* What's left: the port to export, the name of the to be exported
349 * file, and, optionally, the size of the file, in that order. */
354 serve->port=strtol(argv[i], NULL, 0);
359 serve->exportname = argv[i];
362 size_t last = strlen(argv[i])-1;
363 char suffix = argv[i][last];
364 if (suffix == 'k' || suffix == 'K' ||
365 suffix == 'm' || suffix == 'M')
366 argv[i][last] = '\0';
367 es = (off_t)atol(argv[i]);
375 serve->expected_size = es;
381 * Signal handler for SIGCHLD
382 * @param s the signal we're handling (must be SIGCHLD, or something
385 void sigchld_handler(int s)
392 while((pid=wait(status)) > 0) {
393 if(WIFEXITED(status)) {
394 msg3(LOG_INFO, "Child exited with %d", WEXITSTATUS(status));
396 for(i=0;children[i]!=pid&&i<child_arraysize;i++);
397 if(i>=child_arraysize) {
398 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID %ld", (long)pid);
400 children[i]=(pid_t)0;
401 DEBUG2("Removing %d from the list of children", pid);
407 * Handle SIGTERM and dispatch it to our children
408 * @param s the signal we're handling (must be SIGTERM, or something
409 * is severely wrong).
411 void sigterm_handler(int s) {
415 for(i=0;i<child_arraysize;i++) {
417 kill(children[i], s);
430 * Detect the size of a file.
432 * @param export An open filedescriptor
433 * @return the size of the file, or OFFT_MAX if detection was
436 off_t size_autodetect(int export)
440 struct stat stat_buf;
443 #ifdef HAVE_SYS_MOUNT_H
444 #ifdef HAVE_SYS_IOCTL_H
446 DEBUG("looking for export size with ioctl BLKGETSIZE\n");
447 if (!ioctl(export, BLKGETSIZE, &es32) && es32) {
448 es = (off_t)es32 * (off_t)512;
451 #endif /* BLKGETSIZE */
452 #endif /* HAVE_SYS_IOCTL_H */
453 #endif /* HAVE_SYS_MOUNT_H */
455 DEBUG("looking for export size with fstat\n");
456 stat_buf.st_size = 0;
457 error = fstat(export, &stat_buf);
459 if(stat_buf.st_size > 0)
460 return (off_t)stat_buf.st_size;
462 err("fstat failed: %m");
465 DEBUG("looking for export size with lseek SEEK_END\n");
466 es = lseek(export, (off_t)0, SEEK_END);
467 if (es > ((off_t)0)) {
470 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
473 err("Could not find size of exported block device: %m");
478 * Seek to a position in a file, unless we're already there.
479 * @param handle a filedescriptor
480 * @param a position to seek to
481 * @param client the client we're working for
483 void maybeseek(int handle, off_t a, CLIENT* client) {
484 if (a < 0 || a > client->exportsize) {
485 err("Can not happen\n");
487 if (client->lastpoint != a) {
488 if (lseek(handle, a, SEEK_SET) < 0) {
489 err("Can not seek locally!\n");
491 client->lastpoint = a;
498 * Write an amount of bytes at a given offset to the right file. This
499 * abstracts the write-side of the multiple file option.
501 * @param a The offset where the write should start
502 * @param buf The buffer to write from
503 * @param len The length of buf
504 * @return The number of bytes actually written, or -1 in case of an error
506 int rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client)
510 maybeseek(client->export[a/client->server->hunksize],
511 a%client->server->hunksize, client);
512 res = write(client->export[a/client->server->hunksize], buf, len);
513 return (res < 0 || (size_t)res != len);
517 * seek to a position in a file, no matter what. Used when using maybeseek is a
518 * bad idea (for instance, because we're reading the copyonwrite file instead
519 * of the exported file).
520 * @param handle a filedescriptor
521 * @param a position to seek to
522 * @todo get rid of this; lastpoint is a global variable right now, but it
523 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
526 void myseek(int handle,off_t a) {
527 if (lseek(handle, a, SEEK_SET) < 0) {
528 err("Can not seek locally!\n");
533 * Read an amount of bytes at a given offset from the right file. This
534 * abstracts the read-side of the multiple files option.
536 * @param a The offset where the read should start
537 * @param buf A buffer to read into
538 * @param len The size of buf
539 * @return The number of bytes actually read, or -1 in case of an
542 int rawexpread(off_t a, char *buf, size_t len, CLIENT *client)
546 maybeseek(client->export[a/client->server->hunksize],
547 a%client->server->hunksize, client);
548 res = read(client->export[a/client->server->hunksize], buf, len);
549 return (res < 0 || (size_t)res != len);
553 * Read an amount of bytes at a given offset from the right file. This
554 * abstracts the read-side of the copyonwrite stuff, and calls
555 * rawexpread() with the right parameters to do the actual work.
556 * @param a The offset where the read should start
557 * @param buf A buffer to read into
558 * @param len The size of buf
559 * @return The number of bytes actually read, or -1 in case of an error
561 int expread(off_t a, char *buf, size_t len, CLIENT *client)
564 off_t mapcnt, mapl, maph, pagestart;
566 if (!(client->server->flags & F_COPYONWRITE))
567 return rawexpread(a, buf, len, client);
568 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
570 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
572 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
573 pagestart=mapcnt*DIFFPAGESIZE;
575 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
576 len : (size_t)DIFFPAGESIZE-offset;
577 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
578 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
579 (unsigned long)(client->difmap[mapcnt]));
580 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
581 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
582 } else { /* the block is not there */
583 DEBUG2("Page %Lu is not here, we read the original one\n",
584 (unsigned long long)mapcnt);
585 return rawexpread(a, buf, rdlen, client);
587 len-=rdlen; a+=rdlen; buf+=rdlen;
593 * Write an amount of bytes at a given offset to the right file. This
594 * abstracts the write-side of the copyonwrite option, and calls
595 * rawexpwrite() with the right parameters to do the actual work.
597 * @param a The offset where the write should start
598 * @param buf The buffer to write from
599 * @param len The length of buf
600 * @return The number of bytes actually written, or -1 in case of an error
602 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
603 char pagebuf[DIFFPAGESIZE];
604 off_t mapcnt,mapl,maph;
609 if (!(client->server->flags & F_COPYONWRITE))
610 return(rawexpwrite(a,buf,len, client));
611 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
613 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
615 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
616 pagestart=mapcnt*DIFFPAGESIZE ;
618 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
619 len : (size_t)DIFFPAGESIZE-offset;
621 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
622 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
623 (unsigned long)(client->difmap[mapcnt])) ;
624 myseek(client->difffile,
625 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
626 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
627 } else { /* the block is not there */
628 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
629 client->difmap[mapcnt]=client->difffilelen++ ;
630 DEBUG3("Page %Lu is not here, we put it at %lu\n",
631 (unsigned long long)mapcnt,
632 (unsigned long)(client->difmap[mapcnt]));
634 if (rdlen+pagestart%(client->server->hunksize) >
635 (client->server->hunksize))
636 rdlen=client->server->hunksize -
637 (pagestart%client->server->hunksize);
638 if (rawexpread(pagestart, pagebuf, rdlen, client))
640 memcpy(pagebuf+offset,buf,wrlen) ;
641 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
645 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
651 * Do the initial negotiation.
653 * @param net A socket to do the negotiation over
655 void negotiate(CLIENT *client) {
659 memset(zeros, 0, 290);
660 if (write(client->net, INIT_PASSWD, 8) < 0)
661 err("Negotiation failed: %m");
662 cliserv_magic = htonll(cliserv_magic);
663 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
664 err("Negotiation failed: %m");
665 size_host = htonll((u64)(client->exportsize));
666 if (write(client->net, &size_host, 8) < 0)
667 err("Negotiation failed: %m");
668 if (write(client->net, zeros, 128) < 0)
669 err("Negotiation failed: %m");
672 /** sending macro. */
673 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
675 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; client->lastpoint = -1; }
677 * Serve a file to a single client.
679 * @todo This beast needs to be split up in many tiny little manageable
680 * pieces. Preferably with a chainsaw.
682 * @param net A network socket, connected to an nbd client
685 int mainloop(CLIENT *client)
687 struct nbd_request request;
688 struct nbd_reply reply;
693 DEBUG("Entering request loop!\n");
694 reply.magic = htonl(NBD_REPLY_MAGIC);
703 if (client->server->timeout)
704 alarm(client->server->timeout);
705 readit(client->net, &request, sizeof(request));
706 request.from = ntohll(request.from);
707 request.type = ntohl(request.type);
709 if (request.type==NBD_CMD_DISC) { /* Disconnect request */
710 if (client->difmap) free(client->difmap) ;
711 if (client->difffile>=0) {
712 close(client->difffile) ; unlink(client->difffilename) ; }
713 err("Disconnect request received.") ;
716 len = ntohl(request.len);
718 if (request.magic != htonl(NBD_REQUEST_MAGIC))
719 err("Not enough magic.");
721 err("Request too big!");
723 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
724 "READ", (unsigned long long)request.from,
725 (unsigned long long)request.from / 512, len);
727 memcpy(reply.handle, request.handle, sizeof(reply.handle));
728 if ((request.from + len) > (OFFT_MAX)) {
729 DEBUG("[Number too large!]");
730 ERROR(client, reply);
734 if (((ssize_t)((off_t)request.from + len) > client->exportsize) ||
735 ((client->server->flags & F_READONLY) && request.type)) {
737 ERROR(client, reply);
741 if (request.type==1) { /* WRITE */
742 DEBUG("wr: net->buf, ");
743 readit(client->net, buf, len);
745 if ((client->server->flags & F_AUTOREADONLY) ||
746 expwrite(request.from, buf, len,
748 DEBUG("Write failed: %m" );
749 ERROR(client, reply);
752 client->lastpoint += len;
753 SEND(client->net, reply);
760 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
761 client->lastpoint = -1;
762 DEBUG("Read failed: %m");
763 ERROR(client, reply);
766 client->lastpoint += len;
769 memcpy(buf, &reply, sizeof(struct nbd_reply));
770 writeit(client->net, buf, len + sizeof(struct nbd_reply));
776 * Split a single exportfile into multiple ones, if that was asked.
777 * @return 0 on success, -1 on failure
778 * @param client information on the client which we want to split
780 int splitexport(CLIENT* client) {
783 for (i=0; i<client->exportsize; i+=client->server->hunksize) {
786 if(client->server->flags & F_MULTIFILE) {
787 snprintf(tmpname, 1024, "%s.%d", client->exportname,
788 (int)(i/client->server->hunksize));
790 strncpy(tmpname, client->exportname, 1024);
793 DEBUG2( "Opening %s\n", tmpname );
794 if ((client->export[ i/ client->server->hunksize ] =
795 open(tmpname, (client->server->flags &
796 F_READONLY) ? O_RDONLY
798 /* Read WRITE ACCESS was requested by media is only read only */
799 client->server->flags |= F_AUTOREADONLY;
800 client->server->flags |= F_READONLY;
801 if ((client->export[i/client->server->hunksize] =
802 open(tmpname, O_RDONLY)) == -1)
803 err("Could not open exported file: %m");
807 if (client->server->flags & F_COPYONWRITE) {
808 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
810 client->difffilename[1023]='\0';
811 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
812 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
813 if (client->difffile<0) err("Could not create diff file (%m)") ;
814 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
815 err("Could not allocate memory") ;
816 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
823 * Serve a connection.
825 * @todo allow for multithreading, perhaps use libevent.
827 * @param net A network socket connected to an nbd client
829 void serveconnection(CLIENT *client) {
832 if (!client->server->expected_size) {
833 client->exportsize = size_autodetect(client->export[0]);
835 /* Perhaps we should check first. Not now. */
836 client->exportsize = client->server->expected_size;
838 if (client->exportsize > OFFT_MAX) {
839 /* uhm, well... In a parallel universe, this *might* be
841 err("Size of exported file is too big\n");
844 msg3(LOG_INFO, "size of exported file/device is %Lu", (unsigned long long)client->exportsize);
847 setmysockopt(client->net);
853 * Find the name of the file we have to serve. This will use snprintf()
854 * to put the IP address of the client inside a filename containing
855 * "%s". That name is then written to exportname2
857 * @param net A socket connected to an nbd client
858 * @param client information about the client. The IP address in human-readable
859 * format will be written to a new char* buffer, the address of which will be
860 * stored in client->clientname.
862 void set_peername(int net, CLIENT *client) {
863 struct sockaddr_in addrin;
864 int addrinlen = sizeof( addrin );
867 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
868 err("getsockname failed: %m");
869 peername = inet_ntoa(addrin.sin_addr);
870 client->exportname=g_strdup_printf(client->server->exportname, peername);
872 msg4(LOG_INFO, "connect from %s, assigned file is %s",
873 peername, client->exportname);
874 strncpy(clientname,peername,255) ;
878 * Connect the socket, and start to serve. This function will fork()
879 * if a connection from an authorized client is received, and will
882 * @todo modularize this giant beast. Preferably with a chainsaw. Also,
883 * it has no business starting mainloop(); it should connect, and be
886 * @param port the port where we will listen
888 void connectme(SERVER* serve) {
889 struct sockaddr_in addrin;
891 int addrinlen = sizeof(addrin);
892 int net, sock, newpid, i;
906 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
907 pidf=fopen(pidfname, "w");
909 fprintf(pidf,"%d", (int)getpid());
913 fprintf(stderr, "Not fatal; continuing");
917 #endif /* NODAEMON */
919 if ((sock = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
922 /* lose the pesky "Address already in use" error message */
923 if (setsockopt(sock,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
924 err("setsockopt SO_REUSEADDR");
926 if (setsockopt(sock,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
927 err("setsockopt SO_KEEPALIVE");
930 DEBUG("Waiting for connections... bind, ");
931 addrin.sin_family = AF_INET;
932 addrin.sin_port = htons(serve->port);
933 addrin.sin_addr.s_addr = 0;
934 if (bind(sock, (struct sockaddr *) &addrin, addrinlen) < 0)
937 if (listen(sock, 1) < 0)
940 sa.sa_handler = sigchld_handler;
941 sigemptyset(&sa.sa_mask);
942 sa.sa_flags = SA_RESTART;
943 if(sigaction(SIGCHLD, &sa, NULL) == -1)
944 err("sigaction: %m");
945 sa.sa_handler = sigterm_handler;
946 sigemptyset(&sa.sa_mask);
947 sa.sa_flags = SA_RESTART;
948 if(sigaction(SIGTERM, &sa, NULL) == -1)
949 err("sigaction: %m");
950 children=g_malloc(sizeof(pid_t)*child_arraysize);
951 memset(children, 0, sizeof(pid_t)*DEFAULT_CHILD_ARRAY);
952 for(;;) { /* infinite loop */
954 if ((net = accept(sock, (struct sockaddr *) &addrin, &addrinlen)) < 0)
957 client = g_malloc(sizeof(CLIENT));
958 client->server=serve;
959 client->exportsize=OFFT_MAX;
961 set_peername(net, client);
962 if (!authorized_client(client)) {
963 msg2(LOG_INFO,"Unauthorized client") ;
967 msg2(LOG_INFO,"Authorized client") ;
968 for(i=0;children[i]&&i<child_arraysize;i++);
969 if(i>=child_arraysize) {
972 ptr=realloc(children, sizeof(pid_t)*child_arraysize);
975 memset(children+child_arraysize, 0, sizeof(pid_t)*DEFAULT_CHILD_ARRAY);
977 child_arraysize+=DEFAULT_CHILD_ARRAY;
979 msg2(LOG_INFO,"Not enough memory to store child PID");
985 if ((children[i]=fork())<0) {
986 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
990 if (children[i]>0) { /* parent */
991 close(net) ; continue ; }
997 msg2(LOG_INFO,"Starting to serve") ;
998 serveconnection(client);
1003 * Main entry point...
1005 int main(int argc, char *argv[]) {
1007 if (sizeof( struct nbd_request )!=28) {
1008 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1012 serve=cmdline(argc, argv);
1014 if (!(serve->port)) {
1017 /* You really should define ISSERVER if you're going to use inetd
1018 * mode, but if you don't, closing stdout and stderr (which inetd
1019 * had connected to the client socket) will let it work. */
1022 open("/dev/null", O_WRONLY);
1023 open("/dev/null", O_WRONLY);
1025 client=g_malloc(sizeof(CLIENT));
1026 client->server=serve;
1028 client->exportsize=OFFT_MAX;
1029 set_peername(0,client);
1033 connectme(serve); /* serve infinitely */