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,"%s", b)
100 #define msg3(a,b,c) syslog(a,"%s %s", b,c)
101 #define msg4(a,b,c,d) syslog(a,"%s %s %s", b,c,d)
103 #define msg2(a,b) g_message(a,b)
104 #define msg3(a,b,c) g_message(a,b,c)
105 #define msg4(a,b,c,d) g_message(a,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 while((c=getopt_long(argc, argv, "a:cl:mr", long_options, &i))>=0) {
326 serve->flags |= F_READONLY;
329 serve->flags |= F_MULTIFILE;
330 serve->hunksize = 1*GIGA;
333 serve->flags |=F_COPYONWRITE;
336 serve->authname=optarg;
339 serve->timeout=strtol(optarg, NULL, 0);
347 /* What's left: the port to export, the name of the to be exported
348 * file, and, optionally, the size of the file, in that order. */
353 serve->port=strtol(argv[i], NULL, 0);
358 serve->exportname = argv[i];
361 size_t last = strlen(argv[i])-1;
362 char suffix = argv[i][last];
363 if (suffix == 'k' || suffix == 'K' ||
364 suffix == 'm' || suffix == 'M')
365 argv[i][last] = '\0';
366 es = (off_t)atol(argv[i]);
374 serve->expected_size = es;
380 * Signal handler for SIGCHLD
381 * @param s the signal we're handling (must be SIGCHLD, or something
384 void sigchld_handler(int s)
391 while((pid=wait(status)) > 0) {
392 if(WIFEXITED(status)) {
393 memset(buf,'\0', 80);
394 snprintf(buf, 79, "%d", WEXITSTATUS(status));
395 msg3(LOG_INFO, "Child exited with ", buf);
397 for(i=0;children[i]!=pid&&i<child_arraysize;i++);
398 if(i>=child_arraysize) {
399 memset(buf, '\0', 80);
400 snprintf(buf, 79, "%ld", (long)pid);
401 msg3(LOG_INFO, "SIGCHLD received for an unknown child with PID ", buf);
403 children[i]=(pid_t)0;
404 DEBUG2("Removing %d from the list of children", pid);
410 * Handle SIGTERM and dispatch it to our children
411 * @param s the signal we're handling (must be SIGTERM, or something
412 * is severely wrong).
414 void sigterm_handler(int s) {
418 for(i=0;i<child_arraysize;i++) {
420 kill(children[i], s);
433 * Detect the size of a file.
435 * @param export An open filedescriptor
436 * @return the size of the file, or OFFT_MAX if detection was
439 off_t size_autodetect(int export)
443 struct stat stat_buf;
446 #ifdef HAVE_SYS_MOUNT_H
447 #ifdef HAVE_SYS_IOCTL_H
449 DEBUG("looking for export size with ioctl BLKGETSIZE\n");
450 if (!ioctl(export, BLKGETSIZE, &es32) && es32) {
451 es = (off_t)es32 * (off_t)512;
454 #endif /* BLKGETSIZE */
455 #endif /* HAVE_SYS_IOCTL_H */
456 #endif /* HAVE_SYS_MOUNT_H */
458 DEBUG("looking for export size with fstat\n");
459 stat_buf.st_size = 0;
460 error = fstat(export, &stat_buf);
461 if (!error && stat_buf.st_size > 0) {
462 return (off_t)stat_buf.st_size;
464 err("fstat failed: %m");
467 DEBUG("looking for export size with lseek SEEK_END\n");
468 es = lseek(export, (off_t)0, SEEK_END);
469 if (es > ((off_t)0)) {
472 DEBUG2("lseek failed: %d", errno==EBADF?1:(errno==ESPIPE?2:(errno==EINVAL?3:4)));
475 err("Could not find size of exported block device: %m");
480 * Seek to a position in a file, unless we're already there.
481 * @param handle a filedescriptor
482 * @param a position to seek to
483 * @param client the client we're working for
485 void maybeseek(int handle, off_t a, CLIENT* client) {
486 if (a < 0 || a > client->exportsize) {
487 err("Can not happen\n");
489 if (client->lastpoint != a) {
490 if (lseek(handle, a, SEEK_SET) < 0) {
491 err("Can not seek locally!\n");
493 client->lastpoint = a;
500 * Write an amount of bytes at a given offset to the right file. This
501 * abstracts the write-side of the multiple file option.
503 * @param a The offset where the write should start
504 * @param buf The buffer to write from
505 * @param len The length of buf
506 * @return The number of bytes actually written, or -1 in case of an error
508 int rawexpwrite(off_t a, char *buf, size_t len, CLIENT *client)
512 maybeseek(client->export[a/client->server->hunksize],
513 a%client->server->hunksize, client);
514 res = write(client->export[a/client->server->hunksize], buf, len);
515 return (res < 0 || (size_t)res != len);
519 * seek to a position in a file, no matter what. Used when using maybeseek is a
520 * bad idea (for instance, because we're reading the copyonwrite file instead
521 * of the exported file).
522 * @param handle a filedescriptor
523 * @param a position to seek to
524 * @todo get rid of this; lastpoint is a global variable right now, but it
525 * shouldn't be. If we pass it on as a parameter, that makes things a *lot*
528 void myseek(int handle,off_t a) {
529 if (lseek(handle, a, SEEK_SET) < 0) {
530 err("Can not seek locally!\n");
535 * Read an amount of bytes at a given offset from the right file. This
536 * abstracts the read-side of the multiple files option.
538 * @param a The offset where the read should start
539 * @param buf A buffer to read into
540 * @param len The size of buf
541 * @return The number of bytes actually read, or -1 in case of an
544 int rawexpread(off_t a, char *buf, size_t len, CLIENT *client)
548 maybeseek(client->export[a/client->server->hunksize],
549 a%client->server->hunksize, client);
550 res = read(client->export[a/client->server->hunksize], buf, len);
551 return (res < 0 || (size_t)res != len);
555 * Read an amount of bytes at a given offset from the right file. This
556 * abstracts the read-side of the copyonwrite stuff, and calls
557 * rawexpread() with the right parameters to do the actual work.
558 * @param a The offset where the read should start
559 * @param buf A buffer to read into
560 * @param len The size of buf
561 * @return The number of bytes actually read, or -1 in case of an error
563 int expread(off_t a, char *buf, size_t len, CLIENT *client)
566 off_t mapcnt, mapl, maph, pagestart;
568 if (!(client->server->flags & F_COPYONWRITE))
569 return rawexpread(a, buf, len, client);
570 DEBUG3("Asked to read %d bytes at %Lu.\n", len, (unsigned long long)a);
572 mapl=a/DIFFPAGESIZE; maph=(a+len-1)/DIFFPAGESIZE;
574 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
575 pagestart=mapcnt*DIFFPAGESIZE;
577 rdlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
578 len : (size_t)DIFFPAGESIZE-offset;
579 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
580 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
581 (unsigned long)difmap[mapcnt]);
582 myseek(client->difffile, client->difmap[mapcnt]*DIFFPAGESIZE+offset);
583 if (read(client->difffile, buf, rdlen) != rdlen) return -1;
584 } else { /* the block is not there */
585 DEBUG2("Page %Lu is not here, we read the original one\n",
586 (unsigned long long)mapcnt);
587 return rawexpread(a, buf, rdlen, client);
589 len-=rdlen; a+=rdlen; buf+=rdlen;
595 * Write an amount of bytes at a given offset to the right file. This
596 * abstracts the write-side of the copyonwrite option, and calls
597 * rawexpwrite() with the right parameters to do the actual work.
599 * @param a The offset where the write should start
600 * @param buf The buffer to write from
601 * @param len The length of buf
602 * @return The number of bytes actually written, or -1 in case of an error
604 int expwrite(off_t a, char *buf, size_t len, CLIENT *client) {
605 char pagebuf[DIFFPAGESIZE];
606 off_t mapcnt,mapl,maph;
611 if (!(client->server->flags & F_COPYONWRITE))
612 return(rawexpwrite(a,buf,len, client));
613 DEBUG3("Asked to write %d bytes at %Lu.\n", len, (unsigned long long)a);
615 mapl=a/DIFFPAGESIZE ; maph=(a+len-1)/DIFFPAGESIZE ;
617 for (mapcnt=mapl;mapcnt<=maph;mapcnt++) {
618 pagestart=mapcnt*DIFFPAGESIZE ;
620 wrlen=(0<DIFFPAGESIZE-offset && len<(size_t)(DIFFPAGESIZE-offset)) ?
621 len : (size_t)DIFFPAGESIZE-offset;
623 if (client->difmap[mapcnt]!=(u32)(-1)) { /* the block is already there */
624 DEBUG3("Page %Lu is at %lu\n", (unsigned long long)mapcnt,
625 (unsigned long)difmap[mapcnt]) ;
626 myseek(client->difffile,
627 client->difmap[mapcnt]*DIFFPAGESIZE+offset);
628 if (write(client->difffile, buf, wrlen) != wrlen) return -1 ;
629 } else { /* the block is not there */
630 myseek(client->difffile,client->difffilelen*DIFFPAGESIZE) ;
631 client->difmap[mapcnt]=client->difffilelen++ ;
632 DEBUG3("Page %Lu is not here, we put it at %lu\n",
633 (unsigned long long)mapcnt,
634 (unsigned long)difmap[mapcnt]);
636 if (rdlen+pagestart%(client->server->hunksize) >
637 (client->server->hunksize))
638 rdlen=client->server->hunksize -
639 (pagestart%client->server->hunksize);
640 if (rawexpread(pagestart, pagebuf, rdlen, client))
642 memcpy(pagebuf+offset,buf,wrlen) ;
643 if (write(client->difffile, pagebuf, DIFFPAGESIZE) !=
647 len-=wrlen ; a+=wrlen ; buf+=wrlen ;
653 * Do the initial negotiation.
655 * @param net A socket to do the negotiation over
657 void negotiate(CLIENT *client) {
661 memset(zeros, 0, 290);
662 if (write(client->net, INIT_PASSWD, 8) < 0)
663 err("Negotiation failed: %m");
664 cliserv_magic = htonll(cliserv_magic);
665 if (write(client->net, &cliserv_magic, sizeof(cliserv_magic)) < 0)
666 err("Negotiation failed: %m");
667 size_host = htonll((u64)(client->exportsize));
668 if (write(client->net, &size_host, 8) < 0)
669 err("Negotiation failed: %m");
670 if (write(client->net, zeros, 128) < 0)
671 err("Negotiation failed: %m");
674 /** sending macro. */
675 #define SEND(net,reply) writeit( net, &reply, sizeof( reply ));
677 #define ERROR(client,reply) { reply.error = htonl(-1); SEND(client->net,reply); reply.error = 0; client->lastpoint = -1; }
679 * Serve a file to a single client.
681 * @todo This beast needs to be split up in many tiny little manageable
682 * pieces. Preferably with a chainsaw.
684 * @param net A network socket, connected to an nbd client
687 int mainloop(CLIENT *client)
689 struct nbd_request request;
690 struct nbd_reply reply;
695 DEBUG("Entering request loop!\n");
696 reply.magic = htonl(NBD_REPLY_MAGIC);
705 if (client->server->timeout)
706 alarm(client->server->timeout);
707 readit(client->net, &request, sizeof(request));
708 request.from = ntohll(request.from);
709 request.type = ntohl(request.type);
711 if (request.type==NBD_CMD_DISC) { /* Disconnect request */
712 if (client->difmap) free(client->difmap) ;
713 if (client->difffile>=0) {
714 close(client->difffile) ; unlink(client->difffilename) ; }
715 err("Disconnect request received.") ;
718 len = ntohl(request.len);
720 if (request.magic != htonl(NBD_REQUEST_MAGIC))
721 err("Not enough magic.");
723 err("Request too big!");
725 printf("%s from %Lu (%Lu) len %d, ", request.type ? "WRITE" :
726 "READ", (unsigned long long)request.from,
727 (unsigned long long)request.from / 512, len);
729 memcpy(reply.handle, request.handle, sizeof(reply.handle));
730 if ((request.from + len) > (OFFT_MAX)) {
731 DEBUG("[Number too large!]");
732 ERROR(client, reply);
736 if (((ssize_t)((off_t)request.from + len) > client->exportsize) ||
737 ((client->server->flags & F_READONLY) && request.type)) {
739 ERROR(client, reply);
743 if (request.type==1) { /* WRITE */
744 DEBUG("wr: net->buf, ");
745 readit(client->net, buf, len);
747 if ((client->server->flags & F_AUTOREADONLY) ||
748 expwrite(request.from, buf, len,
750 DEBUG("Write failed: %m" );
751 ERROR(client, reply);
754 client->lastpoint += len;
755 SEND(client->net, reply);
762 if (expread(request.from, buf + sizeof(struct nbd_reply), len, client)) {
763 client->lastpoint = -1;
764 DEBUG("Read failed: %m");
765 ERROR(client, reply);
768 client->lastpoint += len;
771 memcpy(buf, &reply, sizeof(struct nbd_reply));
772 writeit(client->net, buf, len + sizeof(struct nbd_reply));
778 * Split a single exportfile into multiple ones, if that was asked.
779 * @return 0 on success, -1 on failure
780 * @param client information on the client which we want to split
782 int splitexport(CLIENT* client) {
785 for (i=0; i<client->exportsize; i+=client->server->hunksize) {
788 if(client->server->flags & F_MULTIFILE) {
789 snprintf(tmpname, 1024, "%s.%d", client->exportname,
790 (int)(i/client->server->hunksize));
792 strncpy(client->exportname, client->server->exportname, 1024);
795 DEBUG2( "Opening %s\n", tmpname );
796 if ((client->export[ i/ client->server->hunksize ] =
797 open(tmpname, (client->server->flags &
798 F_READONLY) ? O_RDONLY
800 /* Read WRITE ACCESS was requested by media is only read only */
801 client->server->flags |= F_AUTOREADONLY;
802 client->server->flags |= F_READONLY;
803 if ((client->export[i/client->server->hunksize] =
804 open(tmpname, O_RDONLY)) == -1)
805 err("Could not open exported file: %m");
809 if (client->server->flags & F_COPYONWRITE) {
810 snprintf(client->difffilename, 1024, "%s-%s-%d.diff",client->exportname,client->clientname,
812 client->difffilename[1023]='\0';
813 msg3(LOG_INFO,"About to create map and diff file %s",client->difffilename) ;
814 client->difffile=open(client->difffilename,O_RDWR | O_CREAT | O_TRUNC,0600) ;
815 if (client->difffile<0) err("Could not create diff file (%m)") ;
816 if ((client->difmap=calloc(client->exportsize/DIFFPAGESIZE,sizeof(u32)))==NULL)
817 err("Could not allocate memory") ;
818 for (i=0;i<client->exportsize/DIFFPAGESIZE;i++) client->difmap[i]=(u32)-1 ;
825 * Serve a connection.
827 * @todo allow for multithreading, perhaps use libevent.
829 * @param net A network socket connected to an nbd client
831 void serveconnection(CLIENT *client) {
834 if (!client->server->expected_size) {
835 client->exportsize = size_autodetect(client->export[0]);
837 /* Perhaps we should check first. Not now. */
838 client->exportsize = client->server->expected_size;
840 if (client->exportsize > OFFT_MAX) {
841 /* uhm, well... In a parallel universe, this *might* be
843 err("Size of exported file is too big\n");
846 memset(buf, '\0', 80);
847 snprintf(buf, 79, "%Lu", (unsigned long long)client->exportsize);
848 msg3(LOG_INFO, "size of exported file/device is ", buf);
851 setmysockopt(client->net);
857 * Find the name of the file we have to serve. This will use snprintf()
858 * to put the IP address of the client inside a filename containing
859 * "%s". That name is then written to exportname2
861 * @param net A socket connected to an nbd client
862 * @param client information about the client. The IP address in human-readable
863 * format will be written to a new char* buffer, the address of which will be
864 * stored in client->clientname.
866 void set_peername(int net, CLIENT *client) {
867 struct sockaddr_in addrin;
868 int addrinlen = sizeof( addrin );
871 client->clientname=g_malloc(256);
872 if (getpeername(net, (struct sockaddr *) &addrin, (socklen_t *)&addrinlen) < 0)
873 err("getsockname failed: %m");
874 peername = inet_ntoa(addrin.sin_addr);
875 snprintf(client->exportname, 1024, client->server->exportname, peername);
876 client->exportname[1023]='\0';
878 msg4(LOG_INFO, "connect from %s, assigned file is %s",
879 peername, client->exportname);
880 strncpy(clientname,peername,255) ;
884 * Connect the socket, and start to serve. This function will fork()
885 * if a connection from an authorized client is received, and will
888 * @todo modularize this giant beast. Preferably with a chainsaw. Also,
889 * it has no business starting mainloop(); it should connect, and be
892 * @param port the port where we will listen
894 void connectme(SERVER* serve) {
895 struct sockaddr_in addrin;
897 int addrinlen = sizeof(addrin);
898 int net, sock, newpid, i;
912 snprintf(pidfname, sizeof(char)*255, "/var/run/nbd-server.%d.pid", serve->port);
913 pidf=fopen(pidfname, "w");
915 fprintf(pidf,"%d", (int)getpid());
919 fprintf(stderr, "Not fatal; continuing");
923 #endif /* NODAEMON */
925 if ((sock = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
928 /* lose the pesky "Address already in use" error message */
929 if (setsockopt(sock,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
930 err("setsockopt SO_REUSEADDR");
932 if (setsockopt(sock,SOL_SOCKET,SO_KEEPALIVE,&yes,sizeof(int)) == -1) {
933 err("setsockopt SO_KEEPALIVE");
936 DEBUG("Waiting for connections... bind, ");
937 addrin.sin_family = AF_INET;
938 addrin.sin_port = htons(serve->port);
939 addrin.sin_addr.s_addr = 0;
940 if (bind(sock, (struct sockaddr *) &addrin, addrinlen) < 0)
943 if (listen(sock, 1) < 0)
946 sa.sa_handler = sigchld_handler;
947 sigemptyset(&sa.sa_mask);
948 sa.sa_flags = SA_RESTART;
949 if(sigaction(SIGCHLD, &sa, NULL) == -1)
950 err("sigaction: %m");
951 sa.sa_handler = sigterm_handler;
952 sigemptyset(&sa.sa_mask);
953 sa.sa_flags = SA_RESTART;
954 if(sigaction(SIGTERM, &sa, NULL) == -1)
955 err("sigaction: %m");
956 children=g_malloc(sizeof(pid_t)*child_arraysize);
957 memset(children, 0, sizeof(pid_t)*DEFAULT_CHILD_ARRAY);
958 for(;;) { /* infinite loop */
960 if ((net = accept(sock, (struct sockaddr *) &addrin, &addrinlen)) < 0)
963 client = g_malloc(sizeof(CLIENT));
964 client->server=serve;
966 set_peername(net, client);
967 if (!authorized_client(client)) {
968 msg2(LOG_INFO,"Unauthorized client") ;
972 msg2(LOG_INFO,"Authorized client") ;
973 for(i=0;children[i]&&i<child_arraysize;i++);
974 if(i>=child_arraysize) {
977 ptr=realloc(children, sizeof(pid_t)*child_arraysize);
980 memset(children+child_arraysize, 0, sizeof(pid_t)*DEFAULT_CHILD_ARRAY);
982 child_arraysize+=DEFAULT_CHILD_ARRAY;
984 msg2(LOG_INFO,"Not enough memory to store child PID");
990 if ((children[i]=fork())<0) {
991 msg3(LOG_INFO,"Could not fork (%s)",strerror(errno)) ;
995 if (children[i]>0) { /* parent */
996 close(net) ; continue ; }
1002 msg2(LOG_INFO,"Starting to serve") ;
1003 serveconnection(client);
1008 * Main entry point...
1010 int main(int argc, char *argv[]) {
1012 if (sizeof( struct nbd_request )!=28) {
1013 fprintf(stderr,"Bad size of structure. Alignment problems?\n");
1017 serve=cmdline(argc, argv);
1019 if (!(serve->port)) {
1022 /* You really should define ISSERVER if you're going to use inetd
1023 * mode, but if you don't, closing stdout and stderr (which inetd
1024 * had connected to the client socket) will let it work. */
1027 open("/dev/null", O_WRONLY);
1028 open("/dev/null", O_WRONLY);
1030 client=g_malloc(sizeof(CLIENT));
1031 client->server=serve;
1033 set_peername(0,client);
1037 connectme(serve); /* serve infinitely */