c264e142803dedbbcfa67aaf42a05e72b60acdc3
[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 "fsloc.h"
33 #include "pseudoflavors.h"
34
35 #ifdef USE_BLKID
36 #include "blkid/blkid.h"
37 #endif
38
39 /*
40  * Invoked by RPC service loop
41  */
42 void    cache_set_fds(fd_set *fdset);
43 int     cache_process_req(fd_set *readfds);
44
45 enum nfsd_fsid {
46         FSID_DEV = 0,
47         FSID_NUM,
48         FSID_MAJOR_MINOR,
49         FSID_ENCODE_DEV,
50         FSID_UUID4_INUM,
51         FSID_UUID8,
52         FSID_UUID16,
53         FSID_UUID16_INUM,
54 };
55
56 /*
57  * Support routines for text-based upcalls.
58  * Fields are separated by spaces.
59  * Fields are either mangled to quote space tab newline slosh with slosh
60  * or a hexified with a leading \x
61  * Record is terminated with newline.
62  *
63  */
64 static int cache_export_ent(char *domain, struct exportent *exp, char *p);
65
66 #define INITIAL_MANAGED_GROUPS 100
67
68 char *lbuf  = NULL;
69 int lbuflen = 0;
70 extern int use_ipaddr;
71
72 static void auth_unix_ip(FILE *f)
73 {
74         /* requests are
75          *  class IP-ADDR
76          * Ignore if class != "nfsd"
77          * Otherwise find domainname and write back:
78          *
79          *  "nfsd" IP-ADDR expiry domainname
80          */
81         char *cp;
82         char class[20];
83         char ipaddr[INET6_ADDRSTRLEN];
84         char *client = NULL;
85         struct addrinfo *tmp = NULL;
86         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
87                 return;
88
89         xlog(D_CALL, "auth_unix_ip: inbuf '%s'", lbuf);
90
91         cp = lbuf;
92
93         if (qword_get(&cp, class, 20) <= 0 ||
94             strcmp(class, "nfsd") != 0)
95                 return;
96
97         if (qword_get(&cp, ipaddr, sizeof(ipaddr)) <= 0)
98                 return;
99
100         tmp = host_pton(ipaddr);
101         if (tmp == NULL)
102                 return;
103
104         auth_reload();
105
106         /* addr is a valid, interesting address, find the domain name... */
107         if (!use_ipaddr) {
108                 struct addrinfo *ai = NULL;
109
110                 ai = client_resolve(tmp->ai_addr);
111                 if (ai) {
112                         client = client_compose(ai);
113                         freeaddrinfo(ai);
114                 }
115         }
116         qword_print(f, "nfsd");
117         qword_print(f, ipaddr);
118         qword_printtimefrom(f, 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         freeaddrinfo(tmp);
128
129 }
130
131 static void auth_unix_gid(FILE *f)
132 {
133         /* Request are
134          *  uid
135          * reply is
136          *  uid expiry count list of group ids
137          */
138         uid_t uid;
139         struct passwd *pw;
140         static gid_t *groups = NULL;
141         static int groups_len = 0;
142         gid_t *more_groups;
143         int ngroups;
144         int rv, i;
145         char *cp;
146
147         if (groups_len == 0) {
148                 groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
149                 if (!groups)
150                         return;
151
152                 groups_len = INITIAL_MANAGED_GROUPS;
153         }
154
155         ngroups = groups_len;
156
157         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
158                 return;
159
160         cp = lbuf;
161         if (qword_get_uint(&cp, &uid) != 0)
162                 return;
163
164         pw = getpwuid(uid);
165         if (!pw)
166                 rv = -1;
167         else {
168                 rv = getgrouplist(pw->pw_name, pw->pw_gid, groups, &ngroups);
169                 if (rv == -1 && ngroups >= groups_len) {
170                         more_groups = realloc(groups, sizeof(gid_t)*ngroups);
171                         if (!more_groups)
172                                 rv = -1;
173                         else {
174                                 groups = more_groups;
175                                 groups_len = ngroups;
176                                 rv = getgrouplist(pw->pw_name, pw->pw_gid,
177                                                   groups, &ngroups);
178                         }
179                 }
180         }
181         qword_printuint(f, uid);
182         qword_printtimefrom(f, DEFAULT_TTL);
183         if (rv >= 0) {
184                 qword_printuint(f, ngroups);
185                 for (i=0; i<ngroups; i++)
186                         qword_printuint(f, groups[i]);
187         } else
188                 qword_printuint(f, 0);
189         qword_eol(f);
190 }
191
192 #if USE_BLKID
193 static const char *get_uuid_blkdev(char *path)
194 {
195         /* We set *safe if we know that we need the
196          * fsid from statfs too.
197          */
198         static blkid_cache cache = NULL;
199         struct stat stb;
200         char *devname;
201         blkid_tag_iterate iter;
202         blkid_dev dev;
203         const char *type;
204         const char *val, *uuid = NULL;
205
206         if (cache == NULL)
207                 blkid_get_cache(&cache, NULL);
208
209         if (stat(path, &stb) != 0)
210                 return NULL;
211         devname = blkid_devno_to_devname(stb.st_dev);
212         if (!devname)
213                 return NULL;
214         dev = blkid_get_dev(cache, devname, BLKID_DEV_NORMAL);
215         free(devname);
216         if (!dev)
217                 return NULL;
218         iter = blkid_tag_iterate_begin(dev);
219         if (!iter)
220                 return NULL;
221         while (blkid_tag_next(iter, &type, &val) == 0) {
222                 if (strcmp(type, "UUID") == 0)
223                         uuid = val;
224                 if (strcmp(type, "TYPE") == 0 &&
225                     strcmp(val, "btrfs") == 0) {
226                         uuid = NULL;
227                         break;
228                 }
229         }
230         blkid_tag_iterate_end(iter);
231         return uuid;
232 }
233 #else
234 #define get_uuid_blkdev(path) (NULL)
235 #endif
236
237 static int get_uuid(const char *val, size_t uuidlen, char *u)
238 {
239         /* extract hex digits from uuidstr and compose a uuid
240          * of the given length (max 16), xoring bytes to make
241          * a smaller uuid.
242          */
243         size_t i = 0;
244         
245         memset(u, 0, uuidlen);
246         for ( ; *val ; val++) {
247                 int c = *val;
248                 if (!isxdigit(c))
249                         continue;
250                 if (isalpha(c)) {
251                         if (isupper(c))
252                                 c = c - 'A' + 10;
253                         else
254                                 c = c - 'a' + 10;
255                 } else
256                         c = c - '0' + 0;
257                 if ((i&1) == 0)
258                         c <<= 4;
259                 u[i/2] ^= (char)c;
260                 i++;
261                 if (i == uuidlen*2)
262                         i = 0;
263         }
264         return 1;
265 }
266
267 static int uuid_by_path(char *path, int type, size_t uuidlen, char *uuid)
268 {
269         /* get a uuid for the filesystem found at 'path'.
270          * There are several possible ways of generating the
271          * uuids (types).
272          * Type 0 is used for new filehandles, while other types
273          * may be used to interpret old filehandle - to ensure smooth
274          * forward migration.
275          * We return 1 if a uuid was found (and it might be worth 
276          * trying the next type) or 0 if no more uuid types can be
277          * extracted.
278          */
279
280         /* Possible sources of uuid are
281          * - blkid uuid
282          * - statfs64 uuid
283          *
284          * On some filesystems (e.g. vfat) the statfs64 uuid is simply an
285          * encoding of the device that the filesystem is mounted from, so
286          * it we be very bad to use that (as device numbers change).  blkid
287          * must be preferred.
288          * On other filesystems (e.g. btrfs) the statfs64 uuid contains
289          * important info that the blkid uuid cannot contain:  This happens
290          * when multiple subvolumes are exported (they have the same
291          * blkid uuid but different statfs64 uuids).
292          * We rely on get_uuid_blkdev *knowing* which is which and not returning
293          * a uuid for filesystems where the statfs64 uuid is better.
294          *
295          */
296         struct statfs64 st;
297         char fsid_val[17];
298         const char *blkid_val;
299         const char *val;
300
301         blkid_val = get_uuid_blkdev(path);
302
303         if (statfs64(path, &st) == 0 &&
304             (st.f_fsid.__val[0] || st.f_fsid.__val[1]))
305                 snprintf(fsid_val, 17, "%08x%08x",
306                          st.f_fsid.__val[0], st.f_fsid.__val[1]);
307         else
308                 fsid_val[0] = 0;
309
310         if (blkid_val && (type--) == 0)
311                 val = blkid_val;
312         else if (fsid_val[0] && (type--) == 0)
313                 val = fsid_val;
314         else
315                 return 0;
316
317         get_uuid(val, uuidlen, uuid);
318         return 1;
319 }
320
321 /* Iterate through /etc/mtab, finding mountpoints
322  * at or below a given path
323  */
324 static char *next_mnt(void **v, char *p)
325 {
326         FILE *f;
327         struct mntent *me;
328         size_t l = strlen(p);
329         if (*v == NULL) {
330                 f = setmntent("/etc/mtab", "r");
331                 *v = f;
332         } else
333                 f = *v;
334         while ((me = getmntent(f)) != NULL &&
335                (strncmp(me->mnt_dir, p, l) != 0 ||
336                 me->mnt_dir[l] != '/'))
337                 ;
338         if (me == NULL) {
339                 endmntent(f);
340                 *v = NULL;
341                 return NULL;
342         }
343         return me->mnt_dir;
344 }
345
346 /* True iff e1 is a child of e2 and e2 has crossmnt set: */
347 static bool subexport(struct exportent *e1, struct exportent *e2)
348 {
349         char *p1 = e1->e_path, *p2 = e2->e_path;
350         size_t l2 = strlen(p2);
351
352         return e2->e_flags & NFSEXP_CROSSMOUNT
353                && strncmp(p1, p2, l2) == 0
354                && p1[l2] == '/';
355 }
356
357 struct parsed_fsid {
358         int fsidtype;
359         /* We could use a union for this, but it would be more
360          * complicated; why bother? */
361         unsigned int inode;
362         unsigned int minor;
363         unsigned int major;
364         unsigned int fsidnum;
365         size_t uuidlen;
366         char *fhuuid;
367 };
368
369 static int parse_fsid(int fsidtype, int fsidlen, char *fsid,
370                 struct parsed_fsid *parsed)
371 {
372         unsigned int dev;
373         unsigned long long inode64;
374
375         memset(parsed, 0, sizeof(*parsed));
376         parsed->fsidtype = fsidtype;
377         switch(fsidtype) {
378         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
379                 if (fsidlen != 8)
380                         return -1;
381                 memcpy(&dev, fsid, 4);
382                 memcpy(&parsed->inode, fsid+4, 4);
383                 parsed->major = ntohl(dev)>>16;
384                 parsed->minor = ntohl(dev) & 0xFFFF;
385                 break;
386
387         case FSID_NUM: /* 4 bytes - fsid */
388                 if (fsidlen != 4)
389                         return -1;
390                 memcpy(&parsed->fsidnum, fsid, 4);
391                 break;
392
393         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode 
394                  * This format is never actually used but was
395                  * an historical accident
396                  */
397                 if (fsidlen != 12)
398                         return -1;
399                 memcpy(&dev, fsid, 4);
400                 parsed->major = ntohl(dev);
401                 memcpy(&dev, fsid+4, 4);
402                 parsed->minor = ntohl(dev);
403                 memcpy(&parsed->inode, fsid+8, 4);
404                 break;
405
406         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
407                 /* This is *host* endian, not net-byte-order, because
408                  * no-one outside this host has any business interpreting it
409                  */
410                 if (fsidlen != 8)
411                         return -1;
412                 memcpy(&dev, fsid, 4);
413                 memcpy(&parsed->inode, fsid+4, 4);
414                 parsed->major = (dev & 0xfff00) >> 8;
415                 parsed->minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
416                 break;
417
418         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
419                 if (fsidlen != 8)
420                         return -1;
421                 memcpy(&parsed->inode, fsid, 4);
422                 parsed->uuidlen = 4;
423                 parsed->fhuuid = fsid+4;
424                 break;
425         case FSID_UUID8: /* 8 byte uuid */
426                 if (fsidlen != 8)
427                         return -1;
428                 parsed->uuidlen = 8;
429                 parsed->fhuuid = fsid;
430                 break;
431         case FSID_UUID16: /* 16 byte uuid */
432                 if (fsidlen != 16)
433                         return -1;
434                 parsed->uuidlen = 16;
435                 parsed->fhuuid = fsid;
436                 break;
437         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
438                 if (fsidlen != 24)
439                         return -1;
440                 memcpy(&inode64, fsid, 8);
441                 parsed->inode = inode64;
442                 parsed->uuidlen = 16;
443                 parsed->fhuuid = fsid+8;
444                 break;
445         }
446         return 0;
447 }
448
449 static bool match_fsid(struct parsed_fsid *parsed, nfs_export *exp, char *path)
450 {
451         struct stat stb;
452         int type;
453         char u[16];
454
455         if (stat(path, &stb) != 0)
456                 return false;
457         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode))
458                 return false;
459
460         switch (parsed->fsidtype) {
461         case FSID_DEV:
462         case FSID_MAJOR_MINOR:
463         case FSID_ENCODE_DEV:
464                 if (stb.st_ino != parsed->inode)
465                         return false;
466                 if (parsed->major != major(stb.st_dev) ||
467                     parsed->minor != minor(stb.st_dev))
468                         return false;
469                 return true;
470         case FSID_NUM:
471                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
472                      exp->m_export.e_fsid != parsed->fsidnum))
473                         return false;
474                 return true;
475         case FSID_UUID4_INUM:
476         case FSID_UUID16_INUM:
477                 if (stb.st_ino != parsed->inode)
478                         return false;
479                 goto check_uuid;
480         case FSID_UUID8:
481         case FSID_UUID16:
482                 if (!is_mountpoint(path))
483                         return false;
484         check_uuid:
485                 if (exp->m_export.e_uuid)
486                         get_uuid(exp->m_export.e_uuid, parsed->uuidlen, u);
487                 else
488                         for (type = 0;
489                              uuid_by_path(path, type, parsed->uuidlen, u);
490                              type++)
491                                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) == 0)
492                                         return true;
493
494                 if (memcmp(u, parsed->fhuuid, parsed->uuidlen) != 0)
495                         return false;
496                 return true;
497         }
498         /* Well, unreachable, actually: */
499         return false;
500 }
501
502 static struct addrinfo *lookup_client_addr(char *dom)
503 {
504         struct addrinfo *ret;
505         struct addrinfo *tmp;
506
507         dom++; /* skip initial "$" */
508
509         tmp = host_pton(dom);
510         if (tmp == NULL)
511                 return NULL;
512         ret = client_resolve(tmp->ai_addr);
513         freeaddrinfo(tmp);
514         return ret;
515 }
516
517 static void nfsd_fh(FILE *f)
518 {
519         /* request are:
520          *  domain fsidtype fsid
521          * interpret fsid, find export point and options, and write:
522          *  domain fsidtype fsid expiry path
523          */
524         char *cp;
525         char *dom;
526         int fsidtype;
527         int fsidlen;
528         char fsid[32];
529         struct parsed_fsid parsed;
530         struct exportent *found = NULL;
531         struct addrinfo *ai = NULL;
532         char *found_path = NULL;
533         nfs_export *exp;
534         int i;
535         int dev_missing = 0;
536
537         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
538                 return;
539
540         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
541
542         cp = lbuf;
543
544         dom = malloc(strlen(cp));
545         if (dom == NULL)
546                 return;
547         if (qword_get(&cp, dom, strlen(cp)) <= 0)
548                 goto out;
549         if (qword_get_int(&cp, &fsidtype) != 0)
550                 goto out;
551         if (fsidtype < 0 || fsidtype > 7)
552                 goto out; /* unknown type */
553         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
554                 goto out;
555         if (parse_fsid(fsidtype, fsidlen, fsid, &parsed))
556                 goto out;
557
558         auth_reload();
559
560         if (is_ipaddr_client(dom)) {
561                 ai = lookup_client_addr(dom);
562                 if (!ai)
563                         goto out;
564         }
565
566         /* Now determine export point for this fsid/domain */
567         for (i=0 ; i < MCL_MAXTYPES; i++) {
568                 nfs_export *next_exp;
569                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
570                         char *path;
571
572                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
573                                 static nfs_export *prev = NULL;
574                                 static void *mnt = NULL;
575                                 
576                                 if (prev == exp) {
577                                         /* try a submount */
578                                         path = next_mnt(&mnt, exp->m_export.e_path);
579                                         if (!path) {
580                                                 next_exp = exp->m_next;
581                                                 prev = NULL;
582                                                 continue;
583                                         }
584                                         next_exp = exp;
585                                 } else {
586                                         prev = exp;
587                                         mnt = NULL;
588                                         path = exp->m_export.e_path;
589                                         next_exp = exp;
590                                 }
591                         } else {
592                                 path = exp->m_export.e_path;
593                                 next_exp = exp->m_next;
594                         }
595
596                         if (!is_ipaddr_client(dom)
597                                         && !namelist_client_matches(exp, dom))
598                                 continue;
599                         if (exp->m_export.e_mountpoint &&
600                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
601                                            exp->m_export.e_mountpoint:
602                                            exp->m_export.e_path))
603                                 dev_missing ++;
604
605                         if (!match_fsid(&parsed, exp, path))
606                                 continue;
607                         if (is_ipaddr_client(dom)
608                                         && !ipaddr_client_matches(exp, ai))
609                                 continue;
610                         if (!found || subexport(&exp->m_export, found)) {
611                                 found = &exp->m_export;
612                                 free(found_path);
613                                 found_path = strdup(path);
614                                 if (found_path == NULL)
615                                         goto out;
616                         } else if (strcmp(found->e_path, exp->m_export.e_path) != 0
617                                    && !subexport(found, &exp->m_export))
618                         {
619                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
620                                      found_path, path, dom);
621                         } else {
622                                 /* same path, if one is V4ROOT, choose the other */
623                                 if (found->e_flags & NFSEXP_V4ROOT) {
624                                         found = &exp->m_export;
625                                         free(found_path);
626                                         found_path = strdup(path);
627                                         if (found_path == NULL)
628                                                 goto out;
629                                 }
630                         }
631                 }
632         }
633         if (found && 
634             found->e_mountpoint &&
635             !is_mountpoint(found->e_mountpoint[0]?
636                            found->e_mountpoint:
637                            found->e_path)) {
638                 /* Cannot export this yet 
639                  * should log a warning, but need to rate limit
640                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
641                    found->e_path, found->e_mountpoint);
642                  */
643                 /* FIXME we need to make sure we re-visit this later */
644                 goto out;
645         }
646         if (!found && dev_missing) {
647                 /* The missing dev could be what we want, so just be
648                  * quite rather than returning stale yet
649                  */
650                 goto out;
651         }
652
653         if (found)
654                 if (cache_export_ent(dom, found, found_path) < 0)
655                         found = 0;
656
657         qword_print(f, dom);
658         qword_printint(f, fsidtype);
659         qword_printhex(f, fsid, fsidlen);
660         /* The fsid -> path lookup can be quite expensive as it
661          * potentially stats and reads lots of devices, and some of those
662          * might have spun-down.  The Answer is not likely to
663          * change underneath us, and an 'exportfs -f' can always
664          * remove this from the kernel, so use a really log
665          * timeout.  Maybe this should be configurable on the command
666          * line.
667          */
668         qword_printint(f, 0x7fffffff);
669         if (found)
670                 qword_print(f, found_path);
671         qword_eol(f);
672  out:
673         if (found_path)
674                 free(found_path);
675         freeaddrinfo(ai);
676         free(dom);
677         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
678         return;         
679 }
680
681 static void write_fsloc(FILE *f, struct exportent *ep)
682 {
683         struct servers *servers;
684
685         if (ep->e_fslocmethod == FSLOC_NONE)
686                 return;
687
688         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
689         if (!servers)
690                 return;
691         qword_print(f, "fsloc");
692         qword_printint(f, servers->h_num);
693         if (servers->h_num >= 0) {
694                 int i;
695                 for (i=0; i<servers->h_num; i++) {
696                         qword_print(f, servers->h_mp[i]->h_host);
697                         qword_print(f, servers->h_mp[i]->h_path);
698                 }
699         }
700         qword_printint(f, servers->h_referral);
701         release_replicas(servers);
702 }
703
704 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
705 {
706         struct sec_entry *p;
707
708         for (p = ep->e_secinfo; p->flav; p++)
709                 ; /* Do nothing */
710         if (p == ep->e_secinfo) {
711                 /* There was no sec= option */
712                 return;
713         }
714         qword_print(f, "secinfo");
715         qword_printint(f, p - ep->e_secinfo);
716         for (p = ep->e_secinfo; p->flav; p++) {
717                 qword_printint(f, p->flav->fnum);
718                 qword_printint(f, p->flags & flag_mask);
719         }
720
721 }
722
723 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
724 {
725         qword_print(f, domain);
726         qword_print(f, path);
727         if (exp) {
728                 int different_fs = strcmp(path, exp->e_path) != 0;
729                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
730
731                 qword_printtimefrom(f, exp->e_ttl);
732                 qword_printint(f, exp->e_flags & flag_mask);
733                 qword_printint(f, exp->e_anonuid);
734                 qword_printint(f, exp->e_anongid);
735                 qword_printint(f, exp->e_fsid);
736                 write_fsloc(f, exp);
737                 write_secinfo(f, exp, flag_mask);
738                 if (exp->e_uuid == NULL || different_fs) {
739                         char u[16];
740                         if (uuid_by_path(path, 0, 16, u)) {
741                                 qword_print(f, "uuid");
742                                 qword_printhex(f, u, 16);
743                         }
744                 } else {
745                         char u[16];
746                         get_uuid(exp->e_uuid, 16, u);
747                         qword_print(f, "uuid");
748                         qword_printhex(f, u, 16);
749                 }
750         } else
751                 qword_printtimefrom(f, DEFAULT_TTL);
752         return qword_eol(f);
753 }
754
755 static int is_subdirectory(char *child, char *parent)
756 {
757         size_t l = strlen(parent);
758
759         return strcmp(child, parent) == 0
760                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
761 }
762
763 static int path_matches(nfs_export *exp, char *path)
764 {
765         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
766                 return is_subdirectory(path, exp->m_export.e_path);
767         return strcmp(path, exp->m_export.e_path) == 0;
768 }
769
770 static int
771 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
772 {
773         return path_matches(exp, path) && client_matches(exp, dom, ai);
774 }
775
776 static nfs_export *
777 lookup_export(char *dom, char *path, struct addrinfo *ai)
778 {
779         nfs_export *exp;
780         nfs_export *found = NULL;
781         int found_type = 0;
782         int i;
783
784         for (i=0 ; i < MCL_MAXTYPES; i++) {
785                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
786                         if (!export_matches(exp, dom, path, ai))
787                                 continue;
788                         if (!found) {
789                                 found = exp;
790                                 found_type = i;
791                                 continue;
792                         }
793                         /* Always prefer non-V4ROOT exports */
794                         if (exp->m_export.e_flags & NFSEXP_V4ROOT)
795                                 continue;
796                         if (found->m_export.e_flags & NFSEXP_V4ROOT) {
797                                 found = exp;
798                                 found_type = i;
799                                 continue;
800                         }
801
802                         /* If one is a CROSSMOUNT, then prefer the longest path */
803                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
804                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
805                             strlen(found->m_export.e_path) !=
806                             strlen(exp->m_export.e_path)) {
807
808                                 if (strlen(exp->m_export.e_path) >
809                                     strlen(found->m_export.e_path)) {
810                                         found = exp;
811                                         found_type = i;
812                                 }
813                                 continue;
814
815                         } else if (found_type == i && found->m_warned == 0) {
816                                 xlog(L_WARNING, "%s exported to both %s and %s, "
817                                      "arbitrarily choosing options from first",
818                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
819                                      dom);
820                                 found->m_warned = 1;
821                         }
822                 }
823         }
824         return found;
825 }
826
827 #ifdef HAVE_NFS_PLUGIN_H
828 #include <dlfcn.h>
829 #include <nfs-plugin.h>
830
831 /*
832  * Find the export entry for the parent of "pathname".
833  * Caller must not free returned exportent.
834  */
835 static struct exportent *lookup_parent_export(char *dom,
836                 const char *pathname, struct addrinfo *ai)
837 {
838         char *parent, *slash;
839         nfs_export *result;
840
841         parent = strdup(pathname);
842         if (parent == NULL) {
843                 xlog(D_GENERAL, "%s: failed to allocate parent path buffer",
844                         __func__);
845                 goto out_default;
846         }
847         xlog(D_CALL, "%s: pathname = '%s'", __func__, pathname);
848
849 again:
850         /* shorten pathname by one component */
851         slash = strrchr(parent, '/');
852         if (slash == NULL) {
853                 xlog(D_GENERAL, "%s: no slash found in pathname",
854                         __func__);
855                 goto out_default;
856         }
857         *slash = '\0';
858
859         if (strlen(parent) == 0) {
860                 result = lookup_export(dom, "/", ai);
861                 if (result == NULL) {
862                         xlog(L_ERROR, "%s: no root export found.", __func__);
863                         goto out_default;
864                 }
865                 goto out;
866         }
867
868         result = lookup_export(dom, parent, ai);
869         if (result == NULL) {
870                 xlog(D_GENERAL, "%s: lookup_export(%s) found nothing",
871                         __func__, parent);
872                 goto again;
873         }
874
875 out:
876         xlog(D_CALL, "%s: found export for %s", __func__, parent);
877         free(parent);
878         return &result->m_export;
879
880 out_default:
881         free(parent);
882         return mkexportent("*", "/", "insecure");
883 }
884
885 /*
886  * Walk through a set of FS locations and build an e_fslocdata string.
887  * Returns true if all went to plan; otherwise, false.
888  */
889 static bool locations_to_fslocdata(struct jp_ops *ops,
890                 nfs_fsloc_set_t locations, char *fslocdata,
891                 size_t remaining, int *ttl)
892 {
893         char *server, *last_path, *rootpath, *ptr;
894         _Bool seen = false;
895
896         last_path = NULL;
897         rootpath = NULL;
898         server = NULL;
899         ptr = fslocdata;
900         *ttl = 0;
901
902         for (;;) {
903                 enum jp_status status;
904                 int len;
905
906                 status = ops->jp_get_next_location(locations, &server,
907                                                         &rootpath, ttl);
908                 if (status == JP_EMPTY)
909                         break;
910                 if (status != JP_OK) {
911                         xlog(D_GENERAL, "%s: failed to parse location: %s",
912                                 __func__, ops->jp_error(status));
913                         goto out_false;
914                 }
915                 xlog(D_GENERAL, "%s: Location: %s:%s",
916                         __func__, server, rootpath);
917
918                 if (last_path && strcmp(rootpath, last_path) == 0) {
919                         len = snprintf(ptr, remaining, "+%s", server);
920                         if (len < 0) {
921                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
922                                 goto out_false;
923                         }
924                         if ((size_t)len >= remaining) {
925                                 xlog(D_GENERAL, "%s: fslocdata buffer overflow", __func__);
926                                 goto out_false;
927                         }
928                         remaining -= (size_t)len;
929                         ptr += len;
930                 } else {
931                         if (last_path == NULL)
932                                 len = snprintf(ptr, remaining, "%s@%s",
933                                                         rootpath, server);
934                         else
935                                 len = snprintf(ptr, remaining, ":%s@%s",
936                                                         rootpath, server);
937                         if (len < 0) {
938                                 xlog(D_GENERAL, "%s: snprintf: %m", __func__);
939                                 goto out_false;
940                         }
941                         if ((size_t)len >= remaining) {
942                                 xlog(D_GENERAL, "%s: fslocdata buffer overflow",
943                                         __func__);
944                                 goto out_false;
945                         }
946                         remaining -= (size_t)len;
947                         ptr += len;
948                         last_path = rootpath;
949                 }
950
951                 seen = true;
952                 free(rootpath);
953                 free(server);
954         }
955
956         xlog(D_CALL, "%s: fslocdata='%s', ttl=%d",
957                 __func__, fslocdata, *ttl);
958         return seen;
959
960 out_false:
961         free(rootpath);
962         free(server);
963         return false;
964 }
965
966 /*
967  * Duplicate the junction's parent's export options and graft in
968  * the fslocdata we constructed from the locations list.
969  */
970 static struct exportent *create_junction_exportent(struct exportent *parent,
971                 const char *junction, const char *fslocdata, int ttl)
972 {
973         static struct exportent *eep;
974
975         eep = (struct exportent *)malloc(sizeof(*eep));
976         if (eep == NULL)
977                 goto out_nomem;
978
979         dupexportent(eep, parent);
980         strcpy(eep->e_path, junction);
981         eep->e_hostname = strdup(parent->e_hostname);
982         if (eep->e_hostname == NULL) {
983                 free(eep);
984                 goto out_nomem;
985         }
986         free(eep->e_uuid);
987         eep->e_uuid = NULL;
988         eep->e_ttl = (unsigned int)ttl;
989
990         free(eep->e_fslocdata);
991         eep->e_fslocdata = strdup(fslocdata);
992         if (eep->e_fslocdata == NULL) {
993                 free(eep->e_hostname);
994                 free(eep);
995                 goto out_nomem;
996         }
997         eep->e_fslocmethod = FSLOC_REFER;
998         return eep;
999
1000 out_nomem:
1001         xlog(L_ERROR, "%s: No memory", __func__);
1002         return NULL;
1003 }
1004
1005 /*
1006  * Walk through the set of FS locations and build an exportent.
1007  * Returns pointer to an exportent if "junction" refers to a junction.
1008  */
1009 static struct exportent *locations_to_export(struct jp_ops *ops,
1010                 nfs_fsloc_set_t locations, const char *junction,
1011                 struct exportent *parent)
1012 {
1013         static char fslocdata[BUFSIZ];
1014         int ttl;
1015
1016         fslocdata[0] = '\0';
1017         if (!locations_to_fslocdata(ops, locations,
1018                                         fslocdata, sizeof(fslocdata), &ttl))
1019                 return NULL;
1020         return create_junction_exportent(parent, junction, fslocdata, ttl);
1021 }
1022
1023 /*
1024  * Retrieve locations information in "junction" and dump it to the
1025  * kernel.  Returns pointer to an exportent if "junction" refers
1026  * to a junction.
1027  */
1028 static struct exportent *invoke_junction_ops(void *handle, char *dom,
1029                 const char *junction, struct addrinfo *ai)
1030 {
1031         struct exportent *parent, *exp = NULL;
1032         nfs_fsloc_set_t locations;
1033         enum jp_status status;
1034         struct jp_ops *ops;
1035         char *error;
1036
1037         ops = (struct jp_ops *)dlsym(handle, "nfs_junction_ops");
1038         error = dlerror();
1039         if (error != NULL) {
1040                 xlog(D_GENERAL, "%s: dlsym(jp_junction_ops): %s",
1041                         __func__, error);
1042                 return NULL;
1043         }
1044         if (ops->jp_api_version != JP_API_VERSION) {
1045                 xlog(D_GENERAL, "%s: unrecognized junction API version: %u",
1046                         __func__, ops->jp_api_version);
1047                 return NULL;
1048         }
1049
1050         status = ops->jp_init(false);
1051         if (status != JP_OK) {
1052                 xlog(D_GENERAL, "%s: failed to resolve %s: %s",
1053                         __func__, junction, ops->jp_error(status));
1054                 return NULL;
1055         }
1056
1057         status = ops->jp_get_locations(junction, &locations);
1058         switch (status) {
1059         case JP_OK:
1060                 break;
1061         case JP_NOTJUNCTION:
1062                 xlog(D_GENERAL, "%s: %s is not a junction",
1063                         __func__, junction);
1064                 goto out;
1065         default:
1066                 xlog(L_WARNING, "Dangling junction %s: %s",
1067                         junction, ops->jp_error(status));
1068                 goto out;
1069         }
1070
1071         parent = lookup_parent_export(dom, junction, ai);
1072         if (parent == NULL)
1073                 goto out;
1074
1075         exp = locations_to_export(ops, locations, junction, parent);
1076
1077         ops->jp_put_locations(locations);
1078
1079 out:
1080         ops->jp_done();
1081         return exp;
1082 }
1083
1084 /*
1085  * Load the junction plug-in, then try to resolve "pathname".
1086  * Returns pointer to an initialized exportent if "junction"
1087  * refers to a junction, or NULL if not.
1088  */
1089 static struct exportent *lookup_junction(char *dom, const char *pathname,
1090                 struct addrinfo *ai)
1091 {
1092         struct exportent *exp;
1093         void *handle;
1094
1095         handle = dlopen("libnfsjunct.so", RTLD_NOW);
1096         if (handle == NULL) {
1097                 xlog(D_GENERAL, "%s: dlopen: %s", __func__, dlerror());
1098                 return NULL;
1099         }
1100         (void)dlerror();        /* Clear any error */
1101
1102         exp = invoke_junction_ops(handle, dom, pathname, ai);
1103
1104         /* We could leave it loaded to make junction resolution
1105          * faster next time.  However, if we want to replace the
1106          * library, that would require restarting mountd. */
1107         (void)dlclose(handle);
1108         return exp;
1109 }
1110
1111 static void lookup_nonexport(FILE *f, char *dom, char *path,
1112                 struct addrinfo *ai)
1113 {
1114         struct exportent *eep;
1115
1116         eep = lookup_junction(dom, path, ai);
1117         dump_to_cache(f, dom, path, eep);
1118         if (eep == NULL)
1119                 return;
1120         exportent_release(eep);
1121         free(eep);
1122 }
1123 #else   /* !HAVE_NFS_PLUGIN_H */
1124 static void lookup_nonexport(FILE *f, char *dom, char *path,
1125                 struct addrinfo *UNUSED(ai))
1126 {
1127         dump_to_cache(f, dom, path, NULL);
1128 }
1129 #endif  /* !HAVE_NFS_PLUGIN_H */
1130
1131 static void nfsd_export(FILE *f)
1132 {
1133         /* requests are:
1134          *  domain path
1135          * determine export options and return:
1136          *  domain path expiry flags anonuid anongid fsid
1137          */
1138
1139         char *cp;
1140         char *dom, *path;
1141         nfs_export *found = NULL;
1142         struct addrinfo *ai = NULL;
1143
1144         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
1145                 return;
1146
1147         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
1148
1149         cp = lbuf;
1150         dom = malloc(strlen(cp));
1151         path = malloc(strlen(cp));
1152
1153         if (!dom || !path)
1154                 goto out;
1155
1156         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
1157                 goto out;
1158         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
1159                 goto out;
1160
1161         auth_reload();
1162
1163         if (is_ipaddr_client(dom)) {
1164                 ai = lookup_client_addr(dom);
1165                 if (!ai)
1166                         goto out;
1167         }
1168
1169         found = lookup_export(dom, path, ai);
1170
1171         if (found) {
1172                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
1173                         xlog(L_WARNING,
1174                              "Cannot export %s, possibly unsupported filesystem"
1175                              " or fsid= required", path);
1176                         dump_to_cache(f, dom, path, NULL);
1177                 }
1178         } else
1179                 lookup_nonexport(f, dom, path, ai);
1180
1181  out:
1182         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
1183         if (dom) free(dom);
1184         if (path) free(path);
1185         freeaddrinfo(ai);
1186 }
1187
1188
1189 struct {
1190         char *cache_name;
1191         void (*cache_handle)(FILE *f);
1192         FILE *f;
1193         char vbuf[RPC_CHAN_BUF_SIZE];
1194 } cachelist[] = {
1195         { "auth.unix.ip", auth_unix_ip, NULL, ""},
1196         { "auth.unix.gid", auth_unix_gid, NULL, ""},
1197         { "nfsd.export", nfsd_export, NULL, ""},
1198         { "nfsd.fh", nfsd_fh, NULL, ""},
1199         { NULL, NULL, NULL, ""}
1200 };
1201
1202 extern int manage_gids;
1203
1204 /**
1205  * cache_open - prepare communications channels with kernel RPC caches
1206  *
1207  */
1208 void cache_open(void) 
1209 {
1210         int i;
1211         for (i=0; cachelist[i].cache_name; i++ ) {
1212                 char path[100];
1213                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
1214                         continue;
1215                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
1216                 cachelist[i].f = fopen(path, "r+");
1217                 if (cachelist[i].f != NULL) {
1218                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF, 
1219                                 RPC_CHAN_BUF_SIZE);
1220                 }
1221         }
1222 }
1223
1224 /**
1225  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
1226  * @fdset: pointer to fd_set to prepare
1227  */
1228 void cache_set_fds(fd_set *fdset)
1229 {
1230         int i;
1231         for (i=0; cachelist[i].cache_name; i++) {
1232                 if (cachelist[i].f)
1233                         FD_SET(fileno(cachelist[i].f), fdset);
1234         }
1235 }
1236
1237 /**
1238  * cache_process_req - process any active cache file descriptors during service loop iteration
1239  * @fdset: pointer to fd_set to examine for activity
1240  */
1241 int cache_process_req(fd_set *readfds) 
1242 {
1243         int i;
1244         int cnt = 0;
1245         for (i=0; cachelist[i].cache_name; i++) {
1246                 if (cachelist[i].f != NULL &&
1247                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
1248                         cnt++;
1249                         cachelist[i].cache_handle(cachelist[i].f);
1250                         FD_CLR(fileno(cachelist[i].f), readfds);
1251                 }
1252         }
1253         return cnt;
1254 }
1255
1256
1257 /*
1258  * Give IP->domain and domain+path->options to kernel
1259  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
1260  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
1261  */
1262
1263 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
1264 {
1265         int err;
1266         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
1267         if (!f)
1268                 return -1;
1269
1270         err = dump_to_cache(f, domain, exp->e_path, exp);
1271         if (err) {
1272                 xlog(L_WARNING,
1273                      "Cannot export %s, possibly unsupported filesystem or"
1274                      " fsid= required", exp->e_path);
1275         }
1276
1277         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
1278                 /* really an 'if', but we can break out of
1279                  * a 'while' more easily */
1280                 /* Look along 'path' for other filesystems
1281                  * and export them with the same options
1282                  */
1283                 struct stat stb;
1284                 size_t l = strlen(exp->e_path);
1285                 __dev_t dev;
1286
1287                 if (strlen(path) <= l || path[l] != '/' ||
1288                     strncmp(exp->e_path, path, l) != 0)
1289                         break;
1290                 if (stat(exp->e_path, &stb) != 0)
1291                         break;
1292                 dev = stb.st_dev;
1293                 while(path[l] == '/') {
1294                         char c;
1295                         /* errors for submount should fail whole filesystem */
1296                         int err2;
1297
1298                         l++;
1299                         while (path[l] != '/' && path[l])
1300                                 l++;
1301                         c = path[l];
1302                         path[l] = 0;
1303                         err2 = lstat(path, &stb);
1304                         path[l] = c;
1305                         if (err2 < 0)
1306                                 break;
1307                         if (stb.st_dev == dev)
1308                                 continue;
1309                         dev = stb.st_dev;
1310                         path[l] = 0;
1311                         dump_to_cache(f, domain, path, exp);
1312                         path[l] = c;
1313                 }
1314                 break;
1315         }
1316
1317         fclose(f);
1318         return err;
1319 }
1320
1321 /**
1322  * cache_export - Inform kernel of a new nfs_export
1323  * @exp: target nfs_export
1324  * @path: NUL-terminated C string containing export path
1325  */
1326 int cache_export(nfs_export *exp, char *path)
1327 {
1328         char buf[INET6_ADDRSTRLEN];
1329         int err;
1330         FILE *f;
1331
1332         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
1333         if (!f)
1334                 return -1;
1335
1336
1337         qword_print(f, "nfsd");
1338         qword_print(f,
1339                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
1340         qword_printtimefrom(f, exp->m_export.e_ttl);
1341         qword_print(f, exp->m_client->m_hostname);
1342         err = qword_eol(f);
1343         
1344         fclose(f);
1345
1346         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
1347                 || err;
1348         return err;
1349 }
1350
1351 /**
1352  * cache_get_filehandle - given an nfs_export, get its root filehandle
1353  * @exp: target nfs_export
1354  * @len: length of requested file handle
1355  * @p: NUL-terminated C string containing export path
1356  *
1357  * Returns pointer to NFS file handle of root directory of export
1358  *
1359  * { 
1360  *   echo $domain $path $length 
1361  *   read filehandle <&0
1362  * } <> /proc/fs/nfsd/filehandle
1363  */
1364 struct nfs_fh_len *
1365 cache_get_filehandle(nfs_export *exp, int len, char *p)
1366 {
1367         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1368         char buf[200];
1369         char *bp = buf;
1370         int failed;
1371         static struct nfs_fh_len fh;
1372
1373         if (!f)
1374                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1375         if (!f)
1376                 return NULL;
1377
1378         qword_print(f, exp->m_client->m_hostname);
1379         qword_print(f, p);
1380         qword_printint(f, len); 
1381         failed = qword_eol(f);
1382         
1383         if (!failed)
1384                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1385         fclose(f);
1386         if (failed)
1387                 return NULL;
1388         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1389         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1390         return &fh;
1391 }