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 static void nfsd_fh(FILE *f)
 356 {
 357         /* request are:
 358          *  domain fsidtype fsid
 359          * interpret fsid, find export point and options, and write:
 360          *  domain fsidtype fsid expiry path
 361          */
 362         char *cp;
 363         char *dom;
 364         int fsidtype;
 365         int fsidlen;
 366         unsigned int dev, major=0, minor=0;
 367         unsigned int inode=0;
 368         unsigned long long inode64;
 369         unsigned int fsidnum=0;
 370         char fsid[32];
 371         struct exportent *found = NULL;
 372         struct addrinfo *ai = NULL;
 373         char *found_path = NULL;
 374         nfs_export *exp;
 375         int i;
 376         int dev_missing = 0;
 377         int uuidlen = 0;
 378         char *fhuuid = NULL;
 379
 380         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 381                 return;
 382
 383         xlog(D_CALL, "nfsd_fh: inbuf '%s'", lbuf);
 384
 385         cp = lbuf;
 386
 387         dom = malloc(strlen(cp));
 388         if (dom == NULL)
 389                 return;
 390         if (qword_get(&cp, dom, strlen(cp)) <= 0)
 391                 goto out;
 392         if (qword_get_int(&cp, &fsidtype) != 0)
 393                 goto out;
 394         if (fsidtype < 0 || fsidtype > 7)
 395                 goto out; /* unknown type */
 396         if ((fsidlen = qword_get(&cp, fsid, 32)) <= 0)
 397                 goto out;
 398         switch(fsidtype) {
 399         case FSID_DEV: /* 4 bytes: 2 major, 2 minor, 4 inode */
 400                 if (fsidlen != 8)
 401                         goto out;
 402                 memcpy(&dev, fsid, 4);
 403                 memcpy(&inode, fsid+4, 4);
 404                 major = ntohl(dev)>>16;
 405                 minor = ntohl(dev) & 0xFFFF;
 406                 break;
 407
 408         case FSID_NUM: /* 4 bytes - fsid */
 409                 if (fsidlen != 4)
 410                         goto out;
 411                 memcpy(&fsidnum, fsid, 4);
 412                 break;
 413
 414         case FSID_MAJOR_MINOR: /* 12 bytes: 4 major, 4 minor, 4 inode
 415                  * This format is never actually used but was
 416                  * an historical accident
 417                  */
 418                 if (fsidlen != 12)
 419                         goto out;
 420                 memcpy(&dev, fsid, 4); major = ntohl(dev);
 421                 memcpy(&dev, fsid+4, 4); minor = ntohl(dev);
 422                 memcpy(&inode, fsid+8, 4);
 423                 break;
 424
 425         case FSID_ENCODE_DEV: /* 8 bytes: 4 byte packed device number, 4 inode */
 426                 /* This is *host* endian, not net-byte-order, because
 427                  * no-one outside this host has any business interpreting it
 428                  */
 429                 if (fsidlen != 8)
 430                         goto out;
 431                 memcpy(&dev, fsid, 4);
 432                 memcpy(&inode, fsid+4, 4);
 433                 major = (dev & 0xfff00) >> 8;
 434                 minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
 435                 break;
 436
 437         case FSID_UUID4_INUM: /* 4 byte inode number and 4 byte uuid */
 438                 if (fsidlen != 8)
 439                         goto out;
 440                 memcpy(&inode, fsid, 4);
 441                 uuidlen = 4;
 442                 fhuuid = fsid+4;
 443                 break;
 444         case FSID_UUID8: /* 8 byte uuid */
 445                 if (fsidlen != 8)
 446                         goto out;
 447                 uuidlen = 8;
 448                 fhuuid = fsid;
 449                 break;
 450         case FSID_UUID16: /* 16 byte uuid */
 451                 if (fsidlen != 16)
 452                         goto out;
 453                 uuidlen = 16;
 454                 fhuuid = fsid;
 455                 break;
 456         case FSID_UUID16_INUM: /* 8 byte inode number and 16 byte uuid */
 457                 if (fsidlen != 24)
 458                         goto out;
 459                 memcpy(&inode64, fsid, 8);
 460                 inode = inode64;
 461                 uuidlen = 16;
 462                 fhuuid = fsid+8;
 463                 break;
 464         }
 465
 466         auth_reload();
 467
 468         /* Now determine export point for this fsid/domain */
 469         for (i=0 ; i < MCL_MAXTYPES; i++) {
 470                 nfs_export *next_exp;
 471                 for (exp = exportlist[i].p_head; exp; exp = next_exp) {
 472                         struct stat stb;
 473                         char u[16];
 474                         char *path;
 475                         int type;
 476
 477                         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
 478                                 static nfs_export *prev = NULL;
 479                                 static void *mnt = NULL;
 480
 481                                 if (prev == exp) {
 482                                         /* try a submount */
 483                                         path = next_mnt(&mnt, exp->m_export.e_path);
 484                                         if (!path) {
 485                                                 next_exp = exp->m_next;
 486                                                 prev = NULL;
 487                                                 continue;
 488                                         }
 489                                         next_exp = exp;
 490                                 } else {
 491                                         prev = exp;
 492                                         mnt = NULL;
 493                                         path = exp->m_export.e_path;
 494                                         next_exp = exp;
 495                                 }
 496                         } else {
 497                                 path = exp->m_export.e_path;
 498                                 next_exp = exp->m_next;
 499                         }
 500
 501                         if (!use_ipaddr && !client_member(dom, exp->m_client->m_hostname))
 502                                 continue;
 503                         if (exp->m_export.e_mountpoint &&
 504                             !is_mountpoint(exp->m_export.e_mountpoint[0]?
 505                                            exp->m_export.e_mountpoint:
 506                                            exp->m_export.e_path))
 507                                 dev_missing ++;
 508                         if (stat(path, &stb) != 0)
 509                                 continue;
 510                         if (!S_ISDIR(stb.st_mode) && !S_ISREG(stb.st_mode)) {
 511                                 continue;
 512                         }
 513                         switch(fsidtype){
 514                         case FSID_DEV:
 515                         case FSID_MAJOR_MINOR:
 516                         case FSID_ENCODE_DEV:
 517                                 if (stb.st_ino != inode)
 518                                         continue;
 519                                 if (major != major(stb.st_dev) ||
 520                                     minor != minor(stb.st_dev))
 521                                         continue;
 522                                 break;
 523                         case FSID_NUM:
 524                                 if (((exp->m_export.e_flags & NFSEXP_FSID) == 0 ||
 525                                      exp->m_export.e_fsid != fsidnum))
 526                                         continue;
 527                                 break;
 528                         case FSID_UUID4_INUM:
 529                         case FSID_UUID16_INUM:
 530                                 if (stb.st_ino != inode)
 531                                         continue;
 532                                 goto check_uuid;
 533                         case FSID_UUID8:
 534                         case FSID_UUID16:
 535                                 if (!is_mountpoint(path))
 536                                         continue;
 537                         check_uuid:
 538                                 if (exp->m_export.e_uuid)
 539                                         get_uuid(exp->m_export.e_uuid,
 540                                                  uuidlen, u);
 541                                 else
 542                                         for (type = 0;
 543                                              uuid_by_path(path, type, uuidlen, u);
 544                                              type++)
 545                                                 if (memcmp(u, fhuuid, uuidlen) == 0)
 546                                                         break;
 547
 548                                 if (memcmp(u, fhuuid, uuidlen) != 0)
 549                                         continue;
 550                                 break;
 551                         }
 552                         if (use_ipaddr) {
 553                                 if (ai == NULL) {
 554                                         struct addrinfo *tmp;
 555                                         tmp = host_pton(dom);
 556                                         if (tmp == NULL)
 557                                                 goto out;
 558                                         ai = client_resolve(tmp->ai_addr);
 559                                         freeaddrinfo(tmp);
 560                                 }
 561                                 if (!client_check(exp->m_client, ai))
 562                                         continue;
 563                         }
 564                         if (!found || subexport(&exp->m_export, found)) {
 565                                 found = &exp->m_export;
 566                                 free(found_path);
 567                                 found_path = strdup(path);
 568                                 if (found_path == NULL)
 569                                         goto out;
 570                         } else if (strcmp(found->e_path, exp->m_export.e_path)
 571                                    && !subexport(found, &exp->m_export))
 572                         {
 573                                 xlog(L_WARNING, "%s and %s have same filehandle for %s, using first",
 574                                      found_path, path, dom);
 575                         }
 576                 }
 577         }
 578         if (found &&
 579             found->e_mountpoint &&
 580             !is_mountpoint(found->e_mountpoint[0]?
 581                            found->e_mountpoint:
 582                            found->e_path)) {
 583                 /* Cannot export this yet
 584                  * should log a warning, but need to rate limit
 585                    xlog(L_WARNING, "%s not exported as %d not a mountpoint",
 586                    found->e_path, found->e_mountpoint);
 587                  */
 588                 /* FIXME we need to make sure we re-visit this later */
 589                 goto out;
 590         }
 591         if (!found && dev_missing) {
 592                 /* The missing dev could be what we want, so just be
 593                  * quite rather than returning stale yet
 594                  */
 595                 goto out;
 596         }
 597
 598         if (found)
 599                 if (cache_export_ent(dom, found, found_path) < 0)
 600                         found = 0;
 601
 602         qword_print(f, dom);
 603         qword_printint(f, fsidtype);
 604         qword_printhex(f, fsid, fsidlen);
 605         /* The fsid -> path lookup can be quite expensive as it
 606          * potentially stats and reads lots of devices, and some of those
 607          * might have spun-down.  The Answer is not likely to
 608          * change underneath us, and an 'exportfs -f' can always
 609          * remove this from the kernel, so use a really log
 610          * timeout.  Maybe this should be configurable on the command
 611          * line.
 612          */
 613         qword_printint(f, 0x7fffffff);
 614         if (found)
 615                 qword_print(f, found_path);
 616         qword_eol(f);
 617  out:
 618         if (found_path)
 619                 free(found_path);
 620         freeaddrinfo(ai);
 621         free(dom);
 622         xlog(D_CALL, "nfsd_fh: found %p path %s", found, found ? found->e_path : NULL);
 623         return;
 624 }
 625
 626 static void write_fsloc(FILE *f, struct exportent *ep)
 627 {
 628         struct servers *servers;
 629
 630         if (ep->e_fslocmethod == FSLOC_NONE)
 631                 return;
 632
 633         servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
 634         if (!servers)
 635                 return;
 636         qword_print(f, "fsloc");
 637         qword_printint(f, servers->h_num);
 638         if (servers->h_num >= 0) {
 639                 int i;
 640                 for (i=0; i<servers->h_num; i++) {
 641                         qword_print(f, servers->h_mp[i]->h_host);
 642                         qword_print(f, servers->h_mp[i]->h_path);
 643                 }
 644         }
 645         qword_printint(f, servers->h_referral);
 646         release_replicas(servers);
 647 }
 648
 649 static void write_secinfo(FILE *f, struct exportent *ep, int flag_mask)
 650 {
 651         struct sec_entry *p;
 652
 653         for (p = ep->e_secinfo; p->flav; p++)
 654                 ; /* Do nothing */
 655         if (p == ep->e_secinfo) {
 656                 /* There was no sec= option */
 657                 return;
 658         }
 659         qword_print(f, "secinfo");
 660         qword_printint(f, p - ep->e_secinfo);
 661         for (p = ep->e_secinfo; p->flav; p++) {
 662                 qword_printint(f, p->flav->fnum);
 663                 qword_printint(f, p->flags & flag_mask);
 664         }
 665
 666 }
 667
 668 static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exp)
 669 {
 670         qword_print(f, domain);
 671         qword_print(f, path);
 672         if (exp) {
 673                 int different_fs = strcmp(path, exp->e_path) != 0;
 674                 int flag_mask = different_fs ? ~NFSEXP_FSID : ~0;
 675
 676                 qword_printuint(f, time(0) + exp->e_ttl);
 677                 qword_printint(f, exp->e_flags & flag_mask);
 678                 qword_printint(f, exp->e_anonuid);
 679                 qword_printint(f, exp->e_anongid);
 680                 qword_printint(f, exp->e_fsid);
 681                 write_fsloc(f, exp);
 682                 write_secinfo(f, exp, flag_mask);
 683                 if (exp->e_uuid == NULL || different_fs) {
 684                         char u[16];
 685                         if (uuid_by_path(path, 0, 16, u)) {
 686                                 qword_print(f, "uuid");
 687                                 qword_printhex(f, u, 16);
 688                         }
 689                 } else {
 690                         char u[16];
 691                         get_uuid(exp->e_uuid, 16, u);
 692                         qword_print(f, "uuid");
 693                         qword_printhex(f, u, 16);
 694                 }
 695         } else
 696                 qword_printuint(f, time(0) + DEFAULT_TTL);
 697         return qword_eol(f);
 698 }
 699
 700 static int is_subdirectory(char *child, char *parent)
 701 {
 702         int l = strlen(parent);
 703
 704         return strcmp(child, parent) == 0
 705                 || (strncmp(child, parent, l) == 0 && child[l] == '/');
 706 }
 707
 708 static int path_matches(nfs_export *exp, char *path)
 709 {
 710         if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)
 711                 return is_subdirectory(path, exp->m_export.e_path);
 712         return strcmp(path, exp->m_export.e_path) == 0;
 713 }
 714
 715 static int
 716 client_matches(nfs_export *exp, char *dom, struct addrinfo *ai)
 717 {
 718         if (use_ipaddr)
 719                 return client_check(exp->m_client, ai);
 720         return client_member(dom, exp->m_client->m_hostname);
 721 }
 722
 723 static int
 724 export_matches(nfs_export *exp, char *dom, char *path, struct addrinfo *ai)
 725 {
 726         return path_matches(exp, path) && client_matches(exp, dom, ai);
 727 }
 728
 729 static nfs_export *
 730 lookup_export(char *dom, char *path, struct addrinfo *ai)
 731 {
 732         nfs_export *exp;
 733         nfs_export *found = NULL;
 734         int found_type = 0;
 735         int i;
 736
 737         for (i=0 ; i < MCL_MAXTYPES; i++) {
 738                 for (exp = exportlist[i].p_head; exp; exp = exp->m_next) {
 739                         if (!export_matches(exp, dom, path, ai))
 740                                 continue;
 741                         if (!found) {
 742                                 found = exp;
 743                                 found_type = i;
 744                                 continue;
 745                         }
 746
 747                         /* Always prefer non-V4ROOT mounts */
 748                         if (found->m_export.e_flags & NFSEXP_V4ROOT)
 749                                 continue;
 750
 751                         /* If one is a CROSSMOUNT, then prefer the longest path */
 752                         if (((found->m_export.e_flags & NFSEXP_CROSSMOUNT) ||
 753                              (exp->m_export.e_flags & NFSEXP_CROSSMOUNT)) &&
 754                             strlen(found->m_export.e_path) !=
 755                             strlen(exp->m_export.e_path)) {
 756
 757                                 if (strlen(exp->m_export.e_path) >
 758                                     strlen(found->m_export.e_path)) {
 759                                         found = exp;
 760                                         found_type = i;
 761                                 }
 762                                 continue;
 763
 764                         } else if (found_type == i && found->m_warned == 0) {
 765                                 xlog(L_WARNING, "%s exported to both %s and %s, "
 766                                      "arbitrarily choosing options from first",
 767                                      path, found->m_client->m_hostname, exp->m_client->m_hostname,
 768                                      dom);
 769                                 found->m_warned = 1;
 770                         }
 771                 }
 772         }
 773         return found;
 774 }
 775
 776 static void nfsd_export(FILE *f)
 777 {
 778         /* requests are:
 779          *  domain path
 780          * determine export options and return:
 781          *  domain path expiry flags anonuid anongid fsid
 782          */
 783
 784         char *cp;
 785         char *dom, *path;
 786         nfs_export *found = NULL;
 787         struct addrinfo *ai = NULL;
 788
 789         if (readline(fileno(f), &lbuf, &lbuflen) != 1)
 790                 return;
 791
 792         xlog(D_CALL, "nfsd_export: inbuf '%s'", lbuf);
 793
 794         cp = lbuf;
 795         dom = malloc(strlen(cp));
 796         path = malloc(strlen(cp));
 797
 798         if (!dom || !path)
 799                 goto out;
 800
 801         if (qword_get(&cp, dom, strlen(lbuf)) <= 0)
 802                 goto out;
 803         if (qword_get(&cp, path, strlen(lbuf)) <= 0)
 804                 goto out;
 805
 806         auth_reload();
 807
 808         if (use_ipaddr) {
 809                 struct addrinfo *tmp;
 810                 tmp = host_pton(dom);
 811                 if (tmp == NULL)
 812                         goto out;
 813                 ai = client_resolve(tmp->ai_addr);
 814                 freeaddrinfo(tmp);
 815                         goto out;
 816         }
 817
 818         found = lookup_export(dom, path, ai);
 819
 820         if (found) {
 821                 if (dump_to_cache(f, dom, path, &found->m_export) < 0) {
 822                         xlog(L_WARNING,
 823                              "Cannot export %s, possibly unsupported filesystem"
 824                              " or fsid= required", path);
 825                         dump_to_cache(f, dom, path, NULL);
 826                 }
 827         } else {
 828                 dump_to_cache(f, dom, path, NULL);
 829         }
 830  out:
 831         xlog(D_CALL, "nfsd_export: found %p path %s", found, path ? path : NULL);
 832         if (dom) free(dom);
 833         if (path) free(path);
 834         freeaddrinfo(ai);
 835 }
 836
 837
 838 struct {
 839         char *cache_name;
 840         void (*cache_handle)(FILE *f);
 841         FILE *f;
 842         char vbuf[RPC_CHAN_BUF_SIZE];
 843 } cachelist[] = {
 844         { "auth.unix.ip", auth_unix_ip, NULL, ""},
 845         { "auth.unix.gid", auth_unix_gid, NULL, ""},
 846         { "nfsd.export", nfsd_export, NULL, ""},
 847         { "nfsd.fh", nfsd_fh, NULL, ""},
 848         { NULL, NULL, NULL, ""}
 849 };
 850
 851 extern int manage_gids;
 852
 853 /**
 854  * cache_open - prepare communications channels with kernel RPC caches
 855  *
 856  */
 857 void cache_open(void)
 858 {
 859         int i;
 860         for (i=0; cachelist[i].cache_name; i++ ) {
 861                 char path[100];
 862                 if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
 863                         continue;
 864                 sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
 865                 cachelist[i].f = fopen(path, "r+");
 866                 if (cachelist[i].f != NULL) {
 867                         setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF,
 868                                 RPC_CHAN_BUF_SIZE);
 869                 }
 870         }
 871 }
 872
 873 /**
 874  * cache_set_fds - prepare cache file descriptors for one iteration of the service loop
 875  * @fdset: pointer to fd_set to prepare
 876  */
 877 void cache_set_fds(fd_set *fdset)
 878 {
 879         int i;
 880         for (i=0; cachelist[i].cache_name; i++) {
 881                 if (cachelist[i].f)
 882                         FD_SET(fileno(cachelist[i].f), fdset);
 883         }
 884 }
 885
 886 /**
 887  * cache_process_req - process any active cache file descriptors during service loop iteration
 888  * @fdset: pointer to fd_set to examine for activity
 889  */
 890 int cache_process_req(fd_set *readfds)
 891 {
 892         int i;
 893         int cnt = 0;
 894         for (i=0; cachelist[i].cache_name; i++) {
 895                 if (cachelist[i].f != NULL &&
 896                     FD_ISSET(fileno(cachelist[i].f), readfds)) {
 897                         cnt++;
 898                         cachelist[i].cache_handle(cachelist[i].f);
 899                         FD_CLR(fileno(cachelist[i].f), readfds);
 900                 }
 901         }
 902         return cnt;
 903 }
 904
 905
 906 /*
 907  * Give IP->domain and domain+path->options to kernel
 908  * % echo nfsd $IP  $[now+DEFAULT_TTL] $domain > /proc/net/rpc/auth.unix.ip/channel
 909  * % echo $domain $path $[now+DEFAULT_TTL] $options $anonuid $anongid $fsid > /proc/net/rpc/nfsd.export/channel
 910  */
 911
 912 static int cache_export_ent(char *domain, struct exportent *exp, char *path)
 913 {
 914         int err;
 915         FILE *f = fopen("/proc/net/rpc/nfsd.export/channel", "w");
 916         if (!f)
 917                 return -1;
 918
 919         err = dump_to_cache(f, domain, exp->e_path, exp);
 920         if (err) {
 921                 xlog(L_WARNING,
 922                      "Cannot export %s, possibly unsupported filesystem or"
 923                      " fsid= required", exp->e_path);
 924         }
 925
 926         while (err == 0 && (exp->e_flags & NFSEXP_CROSSMOUNT) && path) {
 927                 /* really an 'if', but we can break out of
 928                  * a 'while' more easily */
 929                 /* Look along 'path' for other filesystems
 930                  * and export them with the same options
 931                  */
 932                 struct stat stb;
 933                 size_t l = strlen(exp->e_path);
 934                 __dev_t dev;
 935
 936                 if (strlen(path) <= l || path[l] != '/' ||
 937                     strncmp(exp->e_path, path, l) != 0)
 938                         break;
 939                 if (stat(exp->e_path, &stb) != 0)
 940                         break;
 941                 dev = stb.st_dev;
 942                 while(path[l] == '/') {
 943                         char c;
 944                         /* errors for submount should fail whole filesystem */
 945                         int err2;
 946
 947                         l++;
 948                         while (path[l] != '/' && path[l])
 949                                 l++;
 950                         c = path[l];
 951                         path[l] = 0;
 952                         err2 = lstat(path, &stb);
 953                         path[l] = c;
 954                         if (err2 < 0)
 955                                 break;
 956                         if (stb.st_dev == dev)
 957                                 continue;
 958                         dev = stb.st_dev;
 959                         path[l] = 0;
 960                         dump_to_cache(f, domain, path, exp);
 961                         path[l] = c;
 962                 }
 963                 break;
 964         }
 965
 966         fclose(f);
 967         return err;
 968 }
 969
 970 /**
 971  * cache_export - Inform kernel of a new nfs_export
 972  * @exp: target nfs_export
 973  * @path: NUL-terminated C string containing export path
 974  */
 975 int cache_export(nfs_export *exp, char *path)
 976 {
 977         char buf[INET6_ADDRSTRLEN];
 978         int err;
 979         FILE *f;
 980
 981         f = fopen("/proc/net/rpc/auth.unix.ip/channel", "w");
 982         if (!f)
 983                 return -1;
 984
 985
 986         qword_print(f, "nfsd");
 987         qword_print(f,
 988                 host_ntop(get_addrlist(exp->m_client, 0), buf, sizeof(buf)));
 989         qword_printuint(f, time(0) + exp->m_export.e_ttl);
 990         qword_print(f, exp->m_client->m_hostname);
 991         err = qword_eol(f);
 992
 993         fclose(f);
 994
 995         err = cache_export_ent(exp->m_client->m_hostname, &exp->m_export, path)
 996                 || err;
 997         return err;
 998 }
 999
1000 /**
1001  * cache_get_filehandle - given an nfs_export, get its root filehandle
1002  * @exp: target nfs_export
1003  * @len: length of requested file handle
1004  * @p: NUL-terminated C string containing export path
1005  *
1006  * Returns pointer to NFS file handle of root directory of export
1007  *
1008  * {
1009  *   echo $domain $path $length
1010  *   read filehandle <&0
1011  * } <> /proc/fs/nfsd/filehandle
1012  */
1013 struct nfs_fh_len *
1014 cache_get_filehandle(nfs_export *exp, int len, char *p)
1015 {
1016         FILE *f = fopen("/proc/fs/nfsd/filehandle", "r+");
1017         char buf[200];
1018         char *bp = buf;
1019         int failed;
1020         static struct nfs_fh_len fh;
1021
1022         if (!f)
1023                 f = fopen("/proc/fs/nfs/filehandle", "r+");
1024         if (!f)
1025                 return NULL;
1026
1027         qword_print(f, exp->m_client->m_hostname);
1028         qword_print(f, p);
1029         qword_printint(f, len);
1030         failed = qword_eol(f);
1031
1032         if (!failed)
1033                 failed = (fgets(buf, sizeof(buf), f) == NULL);
1034         fclose(f);
1035         if (failed)
1036                 return NULL;
1037         memset(fh.fh_handle, 0, sizeof(fh.fh_handle));
1038         fh.fh_size = qword_get(&bp, (char *)fh.fh_handle, NFS3_FHSIZE);
1039         return &fh;
1040 }