nfsd_fh: if two exports are possible, choose the one without V4ROOT
[nfs-utils.git] / utils / mountd / cache.c
1
2 /*
3  * Handle communication with knfsd internal cache
4  *
5  * We open /proc/net/rpc/{auth.unix.ip,nfsd.export,nfsd.fh}/channel
6  * and listen for requests (using my_svc_run)
7  * 
8  */
9
10 #ifdef HAVE_CONFIG_H
11 #include <config.h>
12 #endif
13
14 #include <sys/types.h>
15 #include <sys/select.h>
16 #include <sys/stat.h>
17 #include <sys/vfs.h>
18 #include <time.h>
19 #include <netinet/in.h>
20 #include <arpa/inet.h>
21 #include <unistd.h>
22 #include <fcntl.h>
23 #include <errno.h>
24 #include <ctype.h>
25 #include <pwd.h>
26 #include <grp.h>
27 #include <mntent.h>
28 #include "misc.h"
29 #include "nfslib.h"
30 #include "exportfs.h"
31 #include "mountd.h"
32 #include "xmalloc.h"
33 #include "fsloc.h"
34 #include "pseudoflavors.h"
35
36 #ifdef USE_BLKID
37 #include "blkid/blkid.h"
38 #endif
39
40 /*
41  * Invoked by RPC service loop
42  */
43 void    cache_set_fds(fd_set *fdset);
44 int     cache_process_req(fd_set *readfds);
45
46 enum nfsd_fsid {
47         FSID_DEV = 0,
48         FSID_NUM,
49         FSID_MAJOR_MINOR,
50         FSID_ENCODE_DEV,
51         FSID_UUID4_INUM,
52         FSID_UUID8,
53         FSID_UUID16,
54         FSID_UUID16_INUM,
55 };
56
57 /*
58  * Support routines for text-based upcalls.
59  * Fields are separated by spaces.
60  * Fields are either mangled to quote space tab newline slosh with slosh
61  * or a hexified with a leading \x
62  * Record is terminated with newline.
63  *
64  */
65 static int cache_export_ent(char *domain, struct exportent *exp, char *p);
66
67 #define INITIAL_MANAGED_GROUPS 100
68
69 char *lbuf  = NULL;
70 int lbuflen = 0;
71 extern int use_ipaddr;
72
73 static void auth_unix_ip(FILE *f)
74 {
75         /* requests are
76          *  class IP-ADDR
77          * Ignore if class != "nfsd"
78          * Otherwise find domainname and write back:
79          *
80          *  "nfsd" IP-ADDR expiry domainname
81          */
82         char *cp;
83         char class[20];
84         char ipaddr[INET6_ADDRSTRLEN];
85         char *client = NULL;
86         struct addrinfo *tmp = NULL;
87         struct addrinfo *ai = NULL;
88         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
89                 return;
90
91         xlog(D_CALL, "auth_unix_ip: inbuf '%s'", lbuf);
92
93         cp = lbuf;
94
95         if (qword_get(&cp, class, 20) <= 0 ||
96             strcmp(class, "nfsd") != 0)
97                 return;
98
99         if (qword_get(&cp, ipaddr, sizeof(ipaddr)) <= 0)
100                 return;
101
102         tmp = host_pton(ipaddr);
103         if (tmp == NULL)
104                 return;
105
106         auth_reload();
107
108         /* addr is a valid, interesting address, find the domain name... */
109         if (!use_ipaddr) {
110                 ai = client_resolve(tmp->ai_addr);
111                 client = client_compose(ai);
112                 freeaddrinfo(ai);
113         }
114         freeaddrinfo(tmp);
115
116         qword_print(f, "nfsd");
117         qword_print(f, ipaddr);
118         qword_printuint(f, time(0) + DEFAULT_TTL);
119         if (use_ipaddr)
120                 qword_print(f, ipaddr);
121         else if (client)
122                 qword_print(f, *client?client:"DEFAULT");
123         qword_eol(f);
124         xlog(D_CALL, "auth_unix_ip: client %p '%s'", client, client?client: "DEFAULT");
125
126         free(client);
127 }
128
129 static void auth_unix_gid(FILE *f)
130 {
131         /* Request are
132          *  uid
133          * reply is
134          *  uid expiry count list of group ids
135          */
136         uid_t uid;
137         struct passwd *pw;
138         static gid_t *groups = NULL;
139         static int groups_len = 0;
140         gid_t *more_groups;
141         int ngroups;
142         int rv, i;
143         char *cp;
144
145         if (groups_len == 0) {
146                 groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
147                 if (!groups)
148                         return;
149
150                 groups_len = INITIAL_MANAGED_GROUPS;
151         }
152
153         ngroups = groups_len;
154
155         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
156                 return;
157
158         cp = lbuf;
159         if (qword_get_uint(&cp, &uid) != 0)
160                 return;
161
162         pw = getpwuid(uid);
163         if (!pw)
164                 rv = -1;
165         else {
166                 rv = getgrouplist(pw->pw_name, pw->pw_gid, groups, &ngroups);
167                 if (rv == -1 && ngroups >= groups_len) {
168                         more_groups = realloc(groups, sizeof(gid_t)*ngroups);
169                         if (!more_groups)
170                                 rv = -1;
171                         else {
172                                 groups = more_groups;
173                                 groups_len = ngroups;
174                                 rv = getgrouplist(pw->pw_name, pw->pw_gid,
175                                                   groups, &ngroups);
176                         }
177                 }
178         }
179         qword_printuint(f, uid);
180         qword_printuint(f, time(0) + DEFAULT_TTL);
181         if (rv >= 0) {
182                 qword_printuint(f, ngroups);
183                 for (i=0; i<ngroups; i++)
184                         qword_printuint(f, groups[i]);
185         } else
186                 qword_printuint(f, 0);
187         qword_eol(f);
188 }
189
190 #if USE_BLKID
191 static const char *get_uuid_blkdev(char *path)
192 {
193         /* We set *safe if we know that we need the
194          * fsid from statfs too.
195          */
196         static blkid_cache cache = NULL;
197         struct stat stb;
198         char *devname;
199         blkid_tag_iterate iter;
200         blkid_dev dev;
201         const char *type;
202         const char *val, *uuid = NULL;
203
204         if (cache == NULL)
205                 blkid_get_cache(&cache, NULL);
206
207         if (stat(path, &stb) != 0)
208                 return NULL;
209         devname = blkid_devno_to_devname(stb.st_dev);
210         if (!devname)
211                 return NULL;
212         dev = blkid_get_dev(cache, devname, BLKID_DEV_NORMAL);
213         free(devname);
214         if (!dev)
215                 return NULL;
216         iter = blkid_tag_iterate_begin(dev);
217         if (!iter)
218                 return NULL;
219         while (blkid_tag_next(iter, &type, &val) == 0) {
220                 if (strcmp(type, "UUID") == 0)
221                         uuid = val;
222                 if (strcmp(type, "TYPE") == 0 &&
223                     strcmp(val, "btrfs") == 0) {
224                         uuid = NULL;
225                         break;
226                 }
227         }
228         blkid_tag_iterate_end(iter);
229         return uuid;
230 }
231 #else
232 #define get_uuid_blkdev(path) (NULL)
233 #endif
234
235 static int get_uuid(const char *val, int uuidlen, char *u)
236 {
237         /* extract hex digits from uuidstr and compose a uuid
238          * of the given length (max 16), xoring bytes to make
239          * a smaller uuid.
240          */
241         int i = 0;
242         
243         memset(u, 0, uuidlen);
244         for ( ; *val ; val++) {
245                 char c = *val;
246                 if (!isxdigit(c))
247                         continue;
248                 if (isalpha(c)) {
249                         if (isupper(c))
250                                 c = c - 'A' + 10;
251                         else
252                                 c = c - 'a' + 10;
253                 } else
254                         c = c - '0' + 0;
255                 if ((i&1) == 0)
256                         c <<= 4;
257                 u[i/2] ^= c;
258                 i++;
259                 if (i == uuidlen*2)
260                         i = 0;
261         }
262         return 1;
263 }
264
265 static int uuid_by_path(char *path, int type, int uuidlen, char *uuid)
266 {
267         /* get a uuid for the filesystem found at 'path'.
268          * There are several possible ways of generating the
269          * uuids (types).
270          * Type 0 is used for new filehandles, while other types
271          * may be used to interpret old filehandle - to ensure smooth
272          * forward migration.
273          * We return 1 if a uuid was found (and it might be worth 
274          * trying the next type) or 0 if no more uuid types can be
275          * extracted.
276          */
277
278         /* Possible sources of uuid are
279          * - blkid uuid
280          * - statfs64 uuid
281          *
282          * On some filesystems (e.g. vfat) the statfs64 uuid is simply an
283          * encoding of the device that the filesystem is mounted from, so
284          * it we be very bad to use that (as device numbers change).  blkid
285          * must be preferred.
286          * On other filesystems (e.g. btrfs) the statfs64 uuid contains
287          * important info that the blkid uuid cannot contain:  This happens
288          * when multiple subvolumes are exported (they have the same
289          * blkid uuid but different statfs64 uuids).
290          * We rely on get_uuid_blkdev *knowing* which is which and not returning
291          * a uuid for filesystems where the statfs64 uuid is better.
292          *
293          */
294         struct statfs64 st;
295         char fsid_val[17];
296         const char *blkid_val;
297         const char *val;
298
299         blkid_val = get_uuid_blkdev(path);
300
301         if (statfs64(path, &st) == 0 &&
302             (st.f_fsid.__val[0] || st.f_fsid.__val[1]))
303                 snprintf(fsid_val, 17, "%08x%08x",
304                          st.f_fsid.__val[0], st.f_fsid.__val[1]);
305         else
306                 fsid_val[0] = 0;
307
308         if (blkid_val && (type--) == 0)
309                 val = blkid_val;
310         else if (fsid_val[0] && (type--) == 0)
311                 val = fsid_val;
312         else
313                 return 0;
314
315         get_uuid(val, uuidlen, uuid);
316         return 1;
317 }
318
319 /* Iterate through /etc/mtab, finding mountpoints
320  * at or below a given path
321  */
322 static char *next_mnt(void **v, char *p)
323 {
324         FILE *f;
325         struct mntent *me;
326         int l = strlen(p);
327         if (*v == NULL) {
328                 f = setmntent("/etc/mtab", "r");
329                 *v = f;
330         } else
331                 f = *v;
332         while ((me = getmntent(f)) != NULL &&
333                (strncmp(me->mnt_dir, p, l) != 0 ||
334                 me->mnt_dir[l] != '/'))
335                 ;
336         if (me == NULL) {
337                 endmntent(f);
338                 *v = NULL;
339                 return NULL;
340         }
341         return me->mnt_dir;
342 }
343
344 /* True iff e1 is a child of e2 and e2 has crossmnt set: */
345 static bool subexport(struct exportent *e1, struct exportent *e2)
346 {
347         char *p1 = e1->e_path, *p2 = e2->e_path;
348         int l2 = strlen(p2);
349
350         return e2->e_flags & NFSEXP_CROSSMOUNT
351                && strncmp(p1, p2, l2) == 0
352                && p1[l2] == '/';
353 }
354
355 struct parsed_fsid {
356         int fsidtype;
357         /* We could use a union for this, but it would be more
358          * complicated; why bother? */
359         unsigned int inode;
360         unsigned int minor;
361         unsigned int major;
362         unsigned int fsidnum;
363         int uuidlen;
364         char *fhuuid;
365 };
366
367 int parse_fsid(int fsidtype, int fsidlen, char *fsid, struct parsed_fsid *parsed)
368 {
369         unsigned int dev;
370         unsigned long long inode64;
371
372         parsed->fsidtype = fsidtype;
373         switch(fsidtype) {
374         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
375                 if (fsidlen != 8)
376                         return -1;
377                 memcpy(&dev, fsid, 4);
378                 memcpy(&parsed->inode, fsid+4, 4);
379                 parsed->major = ntohl(dev)>>16;
380                 parsed->minor = ntohl(dev) & 0xFFFF;
381                 break;
382
383         case FSID_NUM: /* 4 bytes - fsid */
384                 if (fsidlen != 4)
385                         return -1;
386                 memcpy(&parsed->fsidnum, fsid, 4);
387                 break;
388
389         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode 
390                  * This format is never actually used but was
391                  * an historical accident
392                  */
393                 if (fsidlen != 12)
394                         return -1;
395                 memcpy(&dev, fsid, 4);
396                 parsed->major = ntohl(dev);
397                 memcpy(&dev, fsid+4, 4);
398                 parsed->minor = ntohl(dev);
399                 memcpy(&parsed->inode, fsid+8, 4);
400                 break;
401
402         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
403                 /* This is *host* endian, not net-byte-order, because
404                  * no-one outside this host has any business interpreting it
405                  */
406                 if (fsidlen != 8)
407                         return -1;
408                 memcpy(&dev, fsid, 4);
409                 memcpy(&parsed->inode, fsid+4, 4);
410                 parsed->major = (dev & 0xfff00) >> 8;
411                 parsed->minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
412                 break;
413
414         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
415                 if (fsidlen != 8)
416                         return -1;
417                 memcpy(&parsed->inode, fsid, 4);
418                 parsed->uuidlen = 4;
419                 parsed->fhuuid = fsid+4;
420                 break;
421         case FSID_UUID8: /* 8 byte uuid */
422                 if (fsidlen != 8)
423                         return -1;
424                 parsed->uuidlen = 8;
425                 parsed->fhuuid = fsid;
426                 break;
427         case FSID_UUID16: /* 16 byte uuid */
428                 if (fsidlen != 16)
429                         return -1;
430                 parsed->uuidlen = 16;
431                 parsed->fhuuid = fsid;
432                 break;
433         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
434                 if (fsidlen != 24)
435                         return -1;
436                 memcpy(&inode64, fsid, 8);
437                 parsed->inode = inode64;
438                 parsed->uuidlen = 16;
439                 parsed->fhuuid = fsid+8;
440                 break;
441         }
442         return 0;
443 }
444
445 static bool match_fsid(struct parsed_fsid *parsed, nfs_export *exp, char *path)
446 {
447         struct stat stb;
448         int type;
449         char u[16];
450
451         if (stat(path, &stb) != 0)
452                 return false;
453         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode))
454                 return false;
455
456         switch (parsed->fsidtype) {
457         case FSID_DEV:
458         case FSID_MAJOR_MINOR:
459         case FSID_ENCODE_DEV:
460                 if (stb.st_ino != parsed->inode)
461                         return false;
462                 if (parsed->major != major(stb.st_dev) ||
463                     parsed->minor != minor(stb.st_dev))
464                         return false;
465                 return true;
466         case FSID_NUM:
467                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
468                      exp->m_export.e_fsid != parsed->fsidnum))
469                         return false;
470                 return true;
471         case FSID_UUID4_INUM:
472         case FSID_UUID16_INUM:
473                 if (stb.st_ino != parsed->inode)
474                         return false;
475                 goto check_uuid;
476         case FSID_UUID8:
477         case FSID_UUID16:
478                 if (!is_mountpoint(path))
479                         return false;
480         check_uuid:
481                 if (exp->m_export.e_uuid)
482                         get_uuid(exp->m_export.e_uuid, parsed->uuidlen, u);
483                 else
484                         for (type = 0;
485                              uuid_by_path(path, type, parsed->uuidlen, u);
486                              type++)
487                                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) == 0)
488                                         return true;
489
490                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) != 0)
491                         return false;
492                 return true;
493         }
494         /* Well, unreachable, actually: */
495         return false;
496 }
497
498 static void nfsd_fh(FILE *f)
499 {
500         /* request are:
501          *  domain fsidtype fsid
502          * interpret fsid, find export point and options, and write:
503          *  domain fsidtype fsid expiry path
504          */
505         char *cp;
506         char *dom;
507         int fsidtype;
508         int fsidlen;
509         char fsid[32];
510         struct parsed_fsid parsed;
511         struct exportent *found = NULL;
512         struct addrinfo *ai = NULL;
513         char *found_path = NULL;
514         nfs_export *exp;
515         int i;
516         int dev_missing = 0;
517
518         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
519                 return;
520
521         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
522
523         cp = lbuf;
524
525         dom = malloc(strlen(cp));
526         if (dom == NULL)
527                 return;
528         if (qword_get(&cp, dom, strlen(cp)) <= 0)
529                 goto out;
530         if (qword_get_int(&cp, &fsidtype) != 0)
531                 goto out;
532         if (fsidtype < 0 || fsidtype > 7)
533                 goto out; /* unknown type */
534         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
535                 goto out;
536         if (parse_fsid(fsidtype, fsidlen, fsid, &parsed))
537                 goto out;
538
539         auth_reload();
540
541         /* Now determine export point for this fsid/domain */
542         for (i=0 ; i < MCL_MAXTYPES; i++) {
543                 nfs_export *next_exp;
544                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
545                         char *path;
546
547                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
548                                 static nfs_export *prev = NULL;
549                                 static void *mnt = NULL;
550                                 
551                                 if (prev == exp) {
552                                         /* try a submount */
553                                         path = next_mnt(&mnt, exp->m_export.e_path);
554                                         if (!path) {
555                                                 next_exp = exp->m_next;
556                                                 prev = NULL;
557                                                 continue;
558                                         }
559                                         next_exp = exp;
560                                 } else {
561                                         prev = exp;
562                                         mnt = NULL;
563                                         path = exp->m_export.e_path;
564                                         next_exp = exp;
565                                 }
566                         } else {
567                                 path = exp->m_export.e_path;
568                                 next_exp = exp->m_next;
569                         }
570
571                         if (!use_ipaddr && !client_member(dom, exp->m_client->m_hostname))
572                                 continue;
573                         if (exp->m_export.e_mountpoint &&
574                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
575                                            exp->m_export.e_mountpoint:
576                                            exp->m_export.e_path))
577                                 dev_missing ++;
578
579                         if (!match_fsid(&parsed, exp, path))
580                                 continue;
581                         if (use_ipaddr) {
582                                 if (ai == NULL) {
583                                         struct addrinfo *tmp;
584                                         tmp = host_pton(dom);
585                                         if (tmp == NULL)
586                                                 goto out;
587                                         ai = client_resolve(tmp->ai_addr);
588                                         freeaddrinfo(tmp);
589                                 }
590                                 if (!client_check(exp->m_client, ai))
591                                         continue;
592                         }
593                         if (!found || subexport(&exp->m_export, found)) {
594                                 found = &exp->m_export;
595                                 free(found_path);
596                                 found_path = strdup(path);
597                                 if (found_path == NULL)
598                                         goto out;
599                         } else if (strcmp(found->e_path, exp->m_export.e_path) != 0
600                                    && !subexport(found, &exp->m_export))
601                         {
602                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
603                                      found_path, path, dom);
604                         } else {
605                                 /* same path, if one is V4ROOT, choose the other */
606                                 if (found->e_flags & NFSEXP_V4ROOT) {
607                                         found = &exp->m_export;
608                                         free(found_path);
609                                         found_path = strdup(path);
610                                         if (found_path == NULL)
611                                                 goto out;
612                                 }
613                         }
614                 }
615         }
616         if (found && 
617             found->e_mountpoint &&
618             !is_mountpoint(found->e_mountpoint[0]?
619                            found->e_mountpoint:
620                            found->e_path)) {
621                 /* Cannot export this yet 
622                  * should log a warning, but need to rate limit
623                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
624                    found->e_path, found->e_mountpoint);
625                  */
626                 /* FIXME we need to make sure we re-visit this later */
627                 goto out;
628         }
629         if (!found && dev_missing) {
630                 /* The missing dev could be what we want, so just be
631                  * quite rather than returning stale yet
632                  */
633                 goto out;
634         }
635
636         if (found)
637                 if (cache_export_ent(dom, found, found_path) < 0)
638                         found = 0;
639
640         qword_print(f, dom);
641         qword_printint(f, fsidtype);
642         qword_printhex(f, fsid, fsidlen);
643         /* The fsid -> path lookup can be quite expensive as it
644          * potentially stats and reads lots of devices, and some of those
645          * might have spun-down.  The Answer is not likely to
646          * change underneath us, and an 'exportfs -f' can always
647          * remove this from the kernel, so use a really log
648          * timeout.  Maybe this should be configurable on the command
649          * line.
650          */
651         qword_printint(f, 0x7fffffff);
652         if (found)
653                 qword_print(f, found_path);
654         qword_eol(f);
655  out:
656         if (found_path)
657                 free(found_path);
658         freeaddrinfo(ai);
659         free(dom);
660         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
661         return;         
662 }
663
664 static void write_fsloc(FILE *f, struct exportent *ep)
665 {
666         struct servers *servers;
667
668         if (ep->e_fslocmethod == FSLOC_NONE)
669                 return;
670
671         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
672         if (!servers)
673                 return;
674         qword_print(f, "fsloc");
675         qword_printint(f, servers->h_num);
676         if (servers->h_num >= 0) {
677                 int i;
678                 for (i=0; i<servers->h_num; i++) {
679                         qword_print(f, servers->h_mp[i]->h_host);
680                         qword_print(f, servers->h_mp[i]->h_path);
681                 }
682         }
683         qword_printint(f, servers->h_referral);
684         release_replicas(servers);
685 }
686
687 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
688 {
689         struct sec_entry *p;
690
691         for (p = ep->e_secinfo; p->flav; p++)
692                 ; /* Do nothing */
693         if (p == ep->e_secinfo) {
694                 /* There was no sec= option */
695                 return;
696         }
697         qword_print(f, "secinfo");
698         qword_printint(f, p - ep->e_secinfo);
699         for (p = ep->e_secinfo; p->flav; p++) {
700                 qword_printint(f, p->flav->fnum);
701                 qword_printint(f, p->flags & flag_mask);
702         }
703
704 }
705
706 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
707 {
708         qword_print(f, domain);
709         qword_print(f, path);
710         if (exp) {
711                 int different_fs = strcmp(path, exp->e_path) != 0;
712                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
713
714                 qword_printuint(f, time(0) + exp->e_ttl);
715                 qword_printint(f, exp->e_flags & flag_mask);
716                 qword_printint(f, exp->e_anonuid);
717                 qword_printint(f, exp->e_anongid);
718                 qword_printint(f, exp->e_fsid);
719                 write_fsloc(f, exp);
720                 write_secinfo(f, exp, flag_mask);
721                 if (exp->e_uuid == NULL || different_fs) {
722                         char u[16];
723                         if (uuid_by_path(path, 0, 16, u)) {
724                                 qword_print(f, "uuid");
725                                 qword_printhex(f, u, 16);
726                         }
727                 } else {
728                         char u[16];
729                         get_uuid(exp->e_uuid, 16, u);
730                         qword_print(f, "uuid");
731                         qword_printhex(f, u, 16);
732                 }
733         } else
734                 qword_printuint(f, time(0) + DEFAULT_TTL);
735         return qword_eol(f);
736 }
737
738 static int is_subdirectory(char *child, char *parent)
739 {
740         int l = strlen(parent);
741
742         return strcmp(child, parent) == 0
743                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
744 }
745
746 static int path_matches(nfs_export *exp, char *path)
747 {
748         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
749                 return is_subdirectory(path, exp->m_export.e_path);
750         return strcmp(path, exp->m_export.e_path) == 0;
751 }
752
753 static int
754 client_matches(nfs_export *exp, char *dom, struct addrinfo *ai)
755 {
756         if (use_ipaddr)
757                 return client_check(exp->m_client, ai);
758         return client_member(dom, exp->m_client->m_hostname);
759 }
760
761 static int
762 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
763 {
764         return path_matches(exp, path) && client_matches(exp, dom, ai);
765 }
766
767 static nfs_export *
768 lookup_export(char *dom, char *path, struct addrinfo *ai)
769 {
770         nfs_export *exp;
771         nfs_export *found = NULL;
772         int found_type = 0;
773         int i;
774
775         for (i=0 ; i < MCL_MAXTYPES; i++) {
776                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
777                         if (!export_matches(exp, dom, path, ai))
778                                 continue;
779                         if (!found) {
780                                 found = exp;
781                                 found_type = i;
782                                 continue;
783                         }
784                         /* Always prefer non-V4ROOT exports */
785                         if (exp->m_export.e_flags & NFSEXP_V4ROOT)
786                                 continue;
787                         if (found->m_export.e_flags & NFSEXP_V4ROOT) {
788                                 found = exp;
789                                 found_type = i;
790                                 continue;
791                         }
792
793                         /* If one is a CROSSMOUNT, then prefer the longest path */
794                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
795                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
796                             strlen(found->m_export.e_path) !=
797                             strlen(exp->m_export.e_path)) {
798
799                                 if (strlen(exp->m_export.e_path) >
800                                     strlen(found->m_export.e_path)) {
801                                         found = exp;
802                                         found_type = i;
803                                 }
804                                 continue;
805
806                         } else if (found_type == i && found->m_warned == 0) {
807                                 xlog(L_WARNING, "%s exported to both %s and %s, "
808                                      "arbitrarily choosing options from first",
809                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
810                                      dom);
811                                 found->m_warned = 1;
812                         }
813                 }
814         }
815         return found;
816 }
817
818 #ifdef HAVE_NFS_PLUGIN_H
819 #include <dlfcn.h>
820 #include <nfs-plugin.h>
821
822 /*
823  * Walk through a set of FS locations and build a set of export options.
824  * Returns true if all went to plan; otherwise, false.
825  */
826 static _Bool
827 locations_to_options(struct jp_ops *ops, nfs_fsloc_set_t locations,
828                 char *options, size_t remaining, int *ttl)
829 {
830         char *server, *last_path, *rootpath, *ptr;
831         _Bool seen = false;
832
833         last_path = NULL;
834         rootpath = NULL;
835         server = NULL;
836         ptr = options;
837         *ttl = 0;
838
839         for (;;) {
840                 enum jp_status status;
841                 int len;
842
843                 status = ops->jp_get_next_location(locations, &server,
844                                                         &rootpath, ttl);
845                 if (status == JP_EMPTY)
846                         break;
847                 if (status != JP_OK) {
848                         xlog(D_GENERAL, "%s: failed to parse location: %s",
849                                 __func__, ops->jp_error(status));
850                         goto out_false;
851                 }
852                 xlog(D_GENERAL, "%s: Location: %s:%s",
853                         __func__, server, rootpath);
854
855                 if (last_path && strcmp(rootpath, last_path) == 0) {
856                         len = snprintf(ptr, remaining, "+%s", server);
857                         if (len < 0) {
858                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
859                                 goto out_false;
860                         }
861                         if ((size_t)len >= remaining) {
862                                 xlog(D_GENERAL, "%s: options buffer overflow", __func__);
863                                 goto out_false;
864                         }
865                         remaining -= (size_t)len;
866                         ptr += len;
867                 } else {
868                         if (last_path == NULL)
869                                 len = snprintf(ptr, remaining, "refer=%s@%s",
870                                                         rootpath, server);
871                         else
872                                 len = snprintf(ptr, remaining, ":%s@%s",
873                                                         rootpath, server);
874                         if (len < 0) {
875                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
876                                 goto out_false;
877                         }
878                         if ((size_t)len >= remaining) {
879                                 xlog(D_GENERAL, "%s: options buffer overflow",
880                                         __func__);
881                                 goto out_false;
882                         }
883                         remaining -= (size_t)len;
884                         ptr += len;
885                         last_path = rootpath;
886                 }
887
888                 seen = true;
889                 free(rootpath);
890                 free(server);
891         }
892
893         xlog(D_CALL, "%s: options='%s', ttl=%d",
894                 __func__, options, *ttl);
895         return seen;
896
897 out_false:
898         free(rootpath);
899         free(server);
900         return false;
901 }
902
903 /*
904  * Walk through the set of FS locations and build an exportent.
905  * Returns pointer to an exportent if "junction" refers to a junction.
906  *
907  * Returned exportent points to static memory.
908  */
909 static struct exportent *do_locations_to_export(struct jp_ops *ops,
910                 nfs_fsloc_set_t locations, const char *junction,
911                 char *options, size_t options_len)
912 {
913         struct exportent *exp;
914         int ttl;
915
916         if (!locations_to_options(ops, locations, options, options_len, &ttl))
917                 return NULL;
918
919         exp = mkexportent("*", (char *)junction, options);
920         if (exp == NULL) {
921                 xlog(L_ERROR, "%s: Failed to construct exportent", __func__);
922                 return NULL;
923         }
924
925         exp->e_uuid = NULL;
926         exp->e_ttl = ttl;
927         return exp;
928 }
929
930 /*
931  * Convert set of FS locations to an exportent.  Returns pointer to
932  * an exportent if "junction" refers to a junction.
933  *
934  * Returned exportent points to static memory.
935  */
936 static struct exportent *locations_to_export(struct jp_ops *ops,
937                 nfs_fsloc_set_t locations, const char *junction)
938 {
939         struct exportent *exp;
940         char *options;
941
942         options = malloc(BUFSIZ);
943         if (options == NULL) {
944                 xlog(D_GENERAL, "%s: failed to allocate options buffer",
945                         __func__);
946                 return NULL;
947         }
948         options[0] = '\0';
949
950         exp = do_locations_to_export(ops, locations, junction,
951                                                 options, BUFSIZ);
952
953         free(options);
954         return exp;
955 }
956
957 /*
958  * Retrieve locations information in "junction" and dump it to the
959  * kernel.  Returns pointer to an exportent if "junction" refers
960  * to a junction.
961  *
962  * Returned exportent points to static memory.
963  */
964 static struct exportent *invoke_junction_ops(void *handle,
965                 const char *junction)
966 {
967         nfs_fsloc_set_t locations;
968         struct exportent *exp;
969         enum jp_status status;
970         struct jp_ops *ops;
971         char *error;
972
973         ops = (struct jp_ops *)dlsym(handle, "nfs_junction_ops");
974         error = dlerror();
975         if (error != NULL) {
976                 xlog(D_GENERAL, "%s: dlsym(jp_junction_ops): %s",
977                         __func__, error);
978                 return NULL;
979         }
980         if (ops->jp_api_version != JP_API_VERSION) {
981                 xlog(D_GENERAL, "%s: unrecognized junction API version: %u",
982                         __func__, ops->jp_api_version);
983                 return NULL;
984         }
985
986         status = ops->jp_init(false);
987         if (status != JP_OK) {
988                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
989                         __func__, junction, ops->jp_error(status));
990                 return NULL;
991         }
992
993         status = ops->jp_get_locations(junction, &locations);
994         if (status != JP_OK) {
995                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
996                         __func__, junction, ops->jp_error(status));
997                 return NULL;
998         }
999
1000         exp = locations_to_export(ops, locations, junction);
1001
1002         ops->jp_put_locations(locations);
1003         ops->jp_done();
1004         return exp;
1005 }
1006
1007 /*
1008  * Load the junction plug-in, then try to resolve "pathname".
1009  * Returns pointer to an initialized exportent if "junction"
1010  * refers to a junction, or NULL if not.
1011  *
1012  * Returned exportent points to static memory.
1013  */
1014 static struct exportent *lookup_junction(const char *pathname)
1015 {
1016         struct exportent *exp;
1017         void *handle;
1018
1019         handle = dlopen("libnfsjunct.so", RTLD_NOW);
1020         if (handle == NULL) {
1021                 xlog(D_GENERAL, "%s: dlopen: %s", __func__, dlerror());
1022                 return NULL;
1023         }
1024         (void)dlerror();        /* Clear any error */
1025
1026         exp = invoke_junction_ops(handle, pathname);
1027
1028         /* We could leave it loaded to make junction resolution
1029          * faster next time.  However, if we want to replace the
1030          * library, that would require restarting mountd. */
1031         (void)dlclose(handle);
1032         return exp;
1033 }
1034 #else   /* !HAVE_NFS_PLUGIN_H */
1035 static inline struct exportent *lookup_junction(const char *UNUSED(pathname))
1036 {
1037         return NULL;
1038 }
1039 #endif  /* !HAVE_NFS_PLUGIN_H */
1040
1041 static void nfsd_export(FILE *f)
1042 {
1043         /* requests are:
1044          *  domain path
1045          * determine export options and return:
1046          *  domain path expiry flags anonuid anongid fsid
1047          */
1048
1049         char *cp;
1050         char *dom, *path;
1051         nfs_export *found = NULL;
1052         struct addrinfo *ai = NULL;
1053
1054         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
1055                 return;
1056
1057         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
1058
1059         cp = lbuf;
1060         dom = malloc(strlen(cp));
1061         path = malloc(strlen(cp));
1062
1063         if (!dom || !path)
1064                 goto out;
1065
1066         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
1067                 goto out;
1068         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
1069                 goto out;
1070
1071         auth_reload();
1072
1073         if (use_ipaddr) {
1074                 struct addrinfo *tmp;
1075                 tmp = host_pton(dom);
1076                 if (tmp == NULL)
1077                         goto out;
1078                 ai = client_resolve(tmp->ai_addr);
1079                 freeaddrinfo(tmp);
1080                         goto out;
1081         }
1082
1083         found = lookup_export(dom, path, ai);
1084
1085         if (found) {
1086                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
1087                         xlog(L_WARNING,
1088                              "Cannot export %s, possibly unsupported filesystem"
1089                              " or fsid= required", path);
1090                         dump_to_cache(f, dom, path, NULL);
1091                 }
1092         } else {
1093                 dump_to_cache(f, dom, path, lookup_junction(path));
1094         }
1095  out:
1096         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
1097         if (dom) free(dom);
1098         if (path) free(path);
1099         freeaddrinfo(ai);
1100 }
1101
1102
1103 struct {
1104         char *cache_name;
1105         void (*cache_handle)(FILE *f);
1106         FILE *f;
1107         char vbuf[RPC_CHAN_BUF_SIZE];
1108 } cachelist[] = {
1109         { "auth.unix.ip", auth_unix_ip, NULL, ""},
1110         { "auth.unix.gid", auth_unix_gid, NULL, ""},
1111         { "nfsd.export", nfsd_export, NULL, ""},
1112         { "nfsd.fh", nfsd_fh, NULL, ""},
1113         { NULL, NULL, NULL, ""}
1114 };
1115
1116 extern int manage_gids;
1117
1118 /**
1119  * cache_open - prepare communications channels with kernel RPC caches
1120  *
1121  */
1122 void cache_open(void) 
1123 {
1124         int i;
1125         for (i=0; cachelist[i].cache_name; i++ ) {
1126                 char path[100];
1127                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
1128                         continue;
1129                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
1130                 cachelist[i].f = fopen(path, "r+");
1131                 if (cachelist[i].f != NULL) {
1132                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF, 
1133                                 RPC_CHAN_BUF_SIZE);
1134                 }
1135         }
1136 }
1137
1138 /**
1139  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
1140  * @fdset: pointer to fd_set to prepare
1141  */
1142 void cache_set_fds(fd_set *fdset)
1143 {
1144         int i;
1145         for (i=0; cachelist[i].cache_name; i++) {
1146                 if (cachelist[i].f)
1147                         FD_SET(fileno(cachelist[i].f), fdset);
1148         }
1149 }
1150
1151 /**
1152  * cache_process_req - process any active cache file descriptors during service loop iteration
1153  * @fdset: pointer to fd_set to examine for activity
1154  */
1155 int cache_process_req(fd_set *readfds) 
1156 {
1157         int i;
1158         int cnt = 0;
1159         for (i=0; cachelist[i].cache_name; i++) {
1160                 if (cachelist[i].f != NULL &&
1161                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
1162                         cnt++;
1163                         cachelist[i].cache_handle(cachelist[i].f);
1164                         FD_CLR(fileno(cachelist[i].f), readfds);
1165                 }
1166         }
1167         return cnt;
1168 }
1169
1170
1171 /*
1172  * Give IP->domain and domain+path->options to kernel
1173  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
1174  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
1175  */
1176
1177 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
1178 {
1179         int err;
1180         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
1181         if (!f)
1182                 return -1;
1183
1184         err = dump_to_cache(f, domain, exp->e_path, exp);
1185         if (err) {
1186                 xlog(L_WARNING,
1187                      "Cannot export %s, possibly unsupported filesystem or"
1188                      " fsid= required", exp->e_path);
1189         }
1190
1191         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
1192                 /* really an 'if', but we can break out of
1193                  * a 'while' more easily */
1194                 /* Look along 'path' for other filesystems
1195                  * and export them with the same options
1196                  */
1197                 struct stat stb;
1198                 size_t l = strlen(exp->e_path);
1199                 __dev_t dev;
1200
1201                 if (strlen(path) <= l || path[l] != '/' ||
1202                     strncmp(exp->e_path, path, l) != 0)
1203                         break;
1204                 if (stat(exp->e_path, &stb) != 0)
1205                         break;
1206                 dev = stb.st_dev;
1207                 while(path[l] == '/') {
1208                         char c;
1209                         /* errors for submount should fail whole filesystem */
1210                         int err2;
1211
1212                         l++;
1213                         while (path[l] != '/' && path[l])
1214                                 l++;
1215                         c = path[l];
1216                         path[l] = 0;
1217                         err2 = lstat(path, &stb);
1218                         path[l] = c;
1219                         if (err2 < 0)
1220                                 break;
1221                         if (stb.st_dev == dev)
1222                                 continue;
1223                         dev = stb.st_dev;
1224                         path[l] = 0;
1225                         dump_to_cache(f, domain, path, exp);
1226                         path[l] = c;
1227                 }
1228                 break;
1229         }
1230
1231         fclose(f);
1232         return err;
1233 }
1234
1235 /**
1236  * cache_export - Inform kernel of a new nfs_export
1237  * @exp: target nfs_export
1238  * @path: NUL-terminated C string containing export path
1239  */
1240 int cache_export(nfs_export *exp, char *path)
1241 {
1242         char buf[INET6_ADDRSTRLEN];
1243         int err;
1244         FILE *f;
1245
1246         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
1247         if (!f)
1248                 return -1;
1249
1250
1251         qword_print(f, "nfsd");
1252         qword_print(f,
1253                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
1254         qword_printuint(f, time(0) + exp->m_export.e_ttl);
1255         qword_print(f, exp->m_client->m_hostname);
1256         err = qword_eol(f);
1257         
1258         fclose(f);
1259
1260         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1261                 || err;
1262         return err;
1263 }
1264
1265 /**
1266  * cache_get_filehandle - given an nfs_export, get its root filehandle
1267  * @exp: target nfs_export
1268  * @len: length of requested file handle
1269  * @p: NUL-terminated C string containing export path
1270  *
1271  * Returns pointer to NFS file handle of root directory of export
1272  *
1273  * { 
1274  *   echo $domain $path $length 
1275  *   read filehandle <&0
1276  * } <> /proc/fs/nfsd/filehandle
1277  */
1278 struct nfs_fh_len *
1279 cache_get_filehandle(nfs_export *exp, int len, char *p)
1280 {
1281         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1282         char buf[200];
1283         char *bp = buf;
1284         int failed;
1285         static struct nfs_fh_len fh;
1286
1287         if (!f)
1288                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1289         if (!f)
1290                 return NULL;
1291
1292         qword_print(f, exp->m_client->m_hostname);
1293         qword_print(f, p);
1294         qword_printint(f, len); 
1295         failed = qword_eol(f);
1296         
1297         if (!failed)
1298                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1299         fclose(f);
1300         if (failed)
1301                 return NULL;
1302         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1303         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1304         return &fh;
1305 }