2 * Copyright 2009 Oracle. All rights reserved.
4 * This file is part of nfs-utils.
6 * nfs-utils is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * nfs-utils is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with nfs-utils. If not, see <http://www.gnu.org/licenses/>.
23 * Callback information and NSM state is stored in files, usually
24 * under /var/lib/nfs. A database of information contained in local
25 * files stores NLM callback data and what remote peers to notify of
28 * For each monitored remote peer, a text file is created under the
29 * directory specified by NSM_MONITOR_DIR. The name of the file
30 * is a valid DNS hostname. The hostname string must be a valid
31 * ASCII DNS name, and must not contain slash characters, white space,
32 * or '\0' (ie. anything that might have some special meaning in a
35 * The contents of each file include seven blank-separated fields of
36 * text, finished with '\n'. The first field contains the network
37 * address of the NLM service to call back. The current implementation
38 * supports using only IPv4 addresses, so the only contents of this
39 * field are a network order IPv4 address expressed in 8 hexadecimal
42 * The next four fields are text strings of hexadecimal characters,
45 * 2. A 4 byte RPC program number of the NLM service to call back
46 * 3. A 4 byte RPC version number of the NLM service to call back
47 * 4. A 4 byte RPC procedure number of the NLM service to call back
48 * 5. A 16 byte opaque cookie that the NLM service uses to identify
51 * The sixth field is the monitored host's mon_name, passed to statd
52 * via an SM_MON request.
54 * The seventh field is the my_name for this peer, which is the
55 * hostname of the local NLM (currently on Linux, the result of
56 * `uname -n`). This can be used as the source address/hostname
57 * when sending SM_NOTIFY requests.
59 * The NSM protocol does not limit the contents of these strings
60 * in any way except that they must fit into 1024 bytes. Our
61 * implementation requires that these strings not contain
62 * white space or '\0'.
69 #include <sys/types.h>
70 #ifdef HAVE_SYS_CAPABILITY_H
71 #include <sys/capability.h>
73 #include <sys/prctl.h>
92 #define RPCARGSLEN (4 * (8 + 1))
93 #define LINELEN (RPCARGSLEN + SM_PRIV_SIZE * 2 + 1)
95 #define NSM_KERNEL_STATE_FILE "/proc/sys/fs/nfs/nsm_local_state"
97 static char nsm_base_dirname[PATH_MAX] = NSM_DEFAULT_STATEDIR;
99 #define NSM_MONITOR_DIR "sm"
100 #define NSM_NOTIFY_DIR "sm.bak"
101 #define NSM_STATE_FILE "state"
105 error_check(const int len, const size_t buflen)
107 return (len < 0) || ((size_t)len >= buflen);
111 exact_error_check(const ssize_t len, const size_t buflen)
113 return (len < 0) || ((size_t)len != buflen);
117 * Returns a dynamically allocated, '\0'-terminated buffer
118 * containing an appropriate pathname, or NULL if an error
119 * occurs. Caller must free the returned result with free(3).
121 __attribute__((__malloc__))
123 nsm_make_record_pathname(const char *directory, const char *hostname)
131 * Block hostnames that contain characters that have
132 * meaning to the file system (like '/'), or that can
133 * be confusing on visual inspection (like ' ').
135 for (c = hostname; *c != '\0'; c++)
136 if (*c == '/' || isspace((int)*c) != 0) {
137 xlog(D_GENERAL, "Hostname contains invalid characters");
141 size = strlen(nsm_base_dirname) + strlen(directory) + strlen(hostname) + 3;
142 if (size > PATH_MAX) {
143 xlog(D_GENERAL, "Hostname results in pathname that is too long");
149 xlog(D_GENERAL, "Failed to allocate memory for pathname");
153 len = snprintf(path, size, "%s/%s/%s",
154 nsm_base_dirname, directory, hostname);
155 if (error_check(len, size)) {
156 xlog(D_GENERAL, "Pathname did not fit in specified buffer");
165 * Returns a dynamically allocated, '\0'-terminated buffer
166 * containing an appropriate pathname, or NULL if an error
167 * occurs. Caller must free the returned result with free(3).
169 __attribute__((__malloc__))
171 nsm_make_pathname(const char *directory)
177 size = strlen(nsm_base_dirname) + strlen(directory) + 2;
185 len = snprintf(path, size, "%s/%s", nsm_base_dirname, directory);
186 if (error_check(len, size)) {
195 * Returns a dynamically allocated, '\0'-terminated buffer
196 * containing an appropriate pathname, or NULL if an error
197 * occurs. Caller must free the returned result with free(3).
199 __attribute__((__malloc__))
201 nsm_make_temp_pathname(const char *pathname)
207 size = strlen(pathname) + sizeof(".new") + 2;
215 len = snprintf(path, size, "%s.new", pathname);
216 if (error_check(len, size)) {
225 * Use "mktemp, write, rename" to update the contents of a file atomically.
227 * Returns true if completely successful, or false if some error occurred.
230 nsm_atomic_write(const char *path, const void *buf, const size_t buflen)
232 _Bool result = false;
237 temp = nsm_make_temp_pathname(path);
239 xlog(L_ERROR, "Failed to create new path for %s", path);
243 fd = open(temp, O_CREAT | O_TRUNC | O_SYNC | O_WRONLY, 0644);
245 xlog(L_ERROR, "Failed to create %s: %m", temp);
249 len = write(fd, buf, buflen);
250 if (exact_error_check(len, buflen)) {
251 xlog(L_ERROR, "Failed to write %s: %m", temp);
257 if (close(fd) == -1) {
258 xlog(L_ERROR, "Failed to close %s: %m", temp);
263 if (rename(temp, path) == -1) {
264 xlog(L_ERROR, "Failed to rename %s -> %s: %m",
270 /* Ostensibly, a sync(2) is not needed here because
271 * open(O_CREAT), write(O_SYNC), and rename(2) are
272 * already synchronous with persistent storage, for
273 * any file system we care about. */
283 * nsm_setup_pathnames - set up pathname
284 * @progname: C string containing name of program, for error messages
285 * @parentdir: C string containing pathname to on-disk state, or NULL
287 * This runs before logging is set up, so error messages are directed
290 * Returns true and sets up our pathnames, if @parentdir was valid
291 * and usable; otherwise false is returned.
294 nsm_setup_pathnames(const char *progname, const char *parentdir)
296 static char buf[PATH_MAX];
300 /* First: test length of name and whether it exists */
301 if (lstat(parentdir, &st) == -1) {
302 (void)fprintf(stderr, "%s: Failed to stat %s: %s",
303 progname, parentdir, strerror(errno));
307 /* Ensure we have a clean directory pathname */
308 strncpy(buf, parentdir, sizeof(buf));
311 (void)fprintf(stderr, "%s: Unusable directory %s",
312 progname, parentdir);
316 xlog(D_CALL, "Using %s as the state directory", parentdir);
317 strncpy(nsm_base_dirname, parentdir, sizeof(nsm_base_dirname));
322 * nsm_is_default_parentdir - check if parent directory is default
324 * Returns true if the active statd parent directory, set by
325 * nsm_change_pathname(), is the same as the built-in default
326 * parent directory; otherwise false is returned.
329 nsm_is_default_parentdir(void)
331 return strcmp(nsm_base_dirname, NSM_DEFAULT_STATEDIR) == 0;
335 * Clear all capabilities but CAP_NET_BIND_SERVICE. This permits
336 * callers to acquire privileged source ports, but all other root
337 * capabilities are disallowed.
339 * Returns true if successful, or false if some error occurred.
342 nsm_clear_capabilities(void)
344 #ifdef HAVE_SYS_CAPABILITY_H
347 caps = cap_from_text("cap_net_bind_service=ep");
349 xlog(L_ERROR, "Failed to allocate capability: %m");
353 if (cap_set_proc(caps) == -1) {
354 xlog(L_ERROR, "Failed to set capability flags: %m");
355 (void)cap_free(caps);
359 (void)cap_free(caps);
365 * nsm_drop_privileges - drop root privileges
366 * @pidfd: file descriptor of a pid file
368 * Returns true if successful, or false if some error occurred.
370 * Set our effective UID and GID to that of our on-disk database.
373 nsm_drop_privileges(const int pidfd)
377 (void)umask(S_IRWXO);
380 * XXX: If we can't stat dirname, or if dirname is owned by
381 * root, we should use "statduser" instead, which is set up
382 * by configure.ac. Nothing in nfs-utils seems to use
383 * "statduser," though.
385 if (lstat(nsm_base_dirname, &st) == -1) {
386 xlog(L_ERROR, "Failed to stat %s: %m", nsm_base_dirname);
390 if (chdir(nsm_base_dirname) == -1) {
391 xlog(L_ERROR, "Failed to change working directory to %s: %m",
396 if (st.st_uid == 0) {
397 xlog_warn("Running as root. "
398 "chown %s to choose different user", nsm_base_dirname);
403 * If the pidfile happens to reside on NFS, dropping privileges
404 * will probably cause us to lose access, even though we are
405 * holding it open. Chown it to prevent this.
408 if (fchown(pidfd, st.st_uid, st.st_gid) == -1)
409 xlog_warn("Failed to change owner of pidfile: %m");
412 * Don't clear capabilities when dropping root.
414 if (prctl(PR_SET_KEEPCAPS, 1, 0, 0, 0) == -1) {
415 xlog(L_ERROR, "prctl(PR_SET_KEEPCAPS) failed: %m");
419 if (setgroups(0, NULL) == -1) {
420 xlog(L_ERROR, "Failed to drop supplementary groups: %m");
427 * setgid(2) first, as setuid(2) may remove privileges needed
428 * to set the group id.
430 if (setgid(st.st_gid) == -1 || setuid(st.st_uid) == -1) {
431 xlog(L_ERROR, "Failed to drop privileges: %m");
435 xlog(D_CALL, "Effective UID, GID: %u, %u", st.st_uid, st.st_gid);
437 return nsm_clear_capabilities();
441 * nsm_get_state - retrieve on-disk NSM state number
443 * Returns an odd NSM state number read from disk, or an initial
444 * state number. Zero is returned if some error occurs.
447 nsm_get_state(_Bool update)
453 path = nsm_make_pathname(NSM_STATE_FILE);
455 xlog(L_ERROR, "Failed to allocate path for " NSM_STATE_FILE);
459 fd = open(path, O_RDONLY);
461 if (errno != ENOENT) {
462 xlog(L_ERROR, "Failed to open %s: %m", path);
466 xlog(L_NOTICE, "Initializing NSM state");
472 result = read(fd, &state, sizeof(state));
473 if (exact_error_check(result, sizeof(state))) {
474 xlog_warn("Failed to read %s: %m", path);
476 xlog(L_NOTICE, "Initializing NSM state");
482 if ((state & 1) == 0)
490 if (!nsm_atomic_write(path, &state, sizeof(state)))
500 * nsm_update_kernel_state - attempt to post new NSM state to kernel
501 * @state: NSM state number
505 nsm_update_kernel_state(const int state)
511 fd = open(NSM_KERNEL_STATE_FILE, O_WRONLY);
513 xlog(D_GENERAL, "Failed to open " NSM_KERNEL_STATE_FILE ": %m");
517 len = snprintf(buf, sizeof(buf), "%d", state);
518 if (error_check(len, sizeof(buf))) {
519 xlog_warn("Failed to form NSM state number string");
523 result = write(fd, buf, strlen(buf));
524 if (exact_error_check(result, strlen(buf)))
525 xlog_warn("Failed to write NSM state number: %m");
528 xlog(L_ERROR, "Failed to close NSM state file "
529 NSM_KERNEL_STATE_FILE ": %m");
533 * nsm_retire_monitored_hosts - back up all hosts from "sm/" to "sm.bak/"
535 * Returns the count of host records that were moved.
537 * Note that if any error occurs during this process, some monitor
538 * records may be left in the "sm" directory.
541 nsm_retire_monitored_hosts(void)
543 unsigned int count = 0;
548 path = nsm_make_pathname(NSM_MONITOR_DIR);
550 xlog(L_ERROR, "Failed to allocate path for " NSM_MONITOR_DIR);
557 xlog_warn("Failed to open " NSM_MONITOR_DIR ": %m");
561 while ((de = readdir(dir)) != NULL) {
565 if (de->d_name[0] == '.')
568 src = nsm_make_record_pathname(NSM_MONITOR_DIR, de->d_name);
570 xlog_warn("Bad monitor file name, skipping");
574 /* NB: not all file systems fill in d_type correctly */
575 if (lstat(src, &stb) == -1) {
576 xlog_warn("Bad monitor file %s, skipping: %m",
581 if (!S_ISREG(stb.st_mode)) {
582 xlog(D_GENERAL, "Skipping non-regular file %s",
588 dst = nsm_make_record_pathname(NSM_NOTIFY_DIR, de->d_name);
591 xlog_warn("Bad notify file name, skipping");
595 if (rename(src, dst) == -1)
596 xlog_warn("Failed to rename %s -> %s: %m",
599 xlog(D_GENERAL, "Retired record for mon_name %s",
613 * nsm_priv_to_hex - convert a NSM private cookie to a hex string.
615 * @priv: buffer holding the binary NSM private cookie
616 * @buf: output buffer for NULL terminated hex string
617 * @buflen: size of output buffer
619 * Returns the length of the resulting string or 0 on error
622 nsm_priv_to_hex(const char *priv, char *buf, const size_t buflen)
625 size_t remaining = buflen;
627 for (i = 0; i < SM_PRIV_SIZE; i++) {
628 len = snprintf(buf, remaining, "%02x",
629 (unsigned int)(0xff & priv[i]));
630 if (error_check(len, remaining))
633 remaining -= (size_t)len;
636 return buflen - remaining;
640 * Returns the length in bytes of the created record.
642 __attribute__((__noinline__))
644 nsm_create_monitor_record(char *buf, const size_t buflen,
645 const struct sockaddr *sap, const struct mon *m)
647 const struct sockaddr_in *sin = (const struct sockaddr_in *)sap;
648 size_t hexlen, remaining = buflen;
651 len = snprintf(buf, remaining, "%08x %08x %08x %08x ",
652 (unsigned int)sin->sin_addr.s_addr,
653 (unsigned int)m->mon_id.my_id.my_prog,
654 (unsigned int)m->mon_id.my_id.my_vers,
655 (unsigned int)m->mon_id.my_id.my_proc);
656 if (error_check(len, remaining))
659 remaining -= (size_t)len;
661 hexlen = nsm_priv_to_hex(m->priv, buf, remaining);
667 len = snprintf(buf, remaining, " %s %s\n",
668 m->mon_id.mon_name, m->mon_id.my_id.my_name);
669 if (error_check(len, remaining))
671 remaining -= (size_t)len;
673 return buflen - remaining;
677 nsm_append_monitored_host(const char *path, const char *line)
679 _Bool result = false;
686 if (stat(path, &stb) == -1) {
687 xlog(L_ERROR, "Failed to insert: "
688 "could not stat original file %s: %m", path);
691 buflen = (size_t)stb.st_size + strlen(line);
693 buf = malloc(buflen + 1);
695 xlog(L_ERROR, "Failed to insert: no memory");
698 memset(buf, 0, buflen + 1);
700 fd = open(path, O_RDONLY);
702 xlog(L_ERROR, "Failed to insert: "
703 "could not open original file %s: %m", path);
707 len = read(fd, buf, (size_t)stb.st_size);
708 if (exact_error_check(len, (size_t)stb.st_size)) {
709 xlog(L_ERROR, "Failed to insert: "
710 "could not read original file %s: %m", path);
718 if (nsm_atomic_write(path, buf, buflen))
727 * nsm_insert_monitored_host - write callback data for one host to disk
728 * @hostname: C string containing a hostname
729 * @sap: sockaddr containing NLM callback address
730 * @mon: SM_MON arguments to save
732 * Returns true if successful, otherwise false if some error occurs.
735 nsm_insert_monitored_host(const char *hostname, const struct sockaddr *sap,
738 static char buf[LINELEN + 1 + SM_MAXSTRLEN + 2];
740 _Bool result = false;
745 path = nsm_make_record_pathname(NSM_MONITOR_DIR, hostname);
747 xlog(L_ERROR, "Failed to insert: bad monitor hostname '%s'",
752 size = nsm_create_monitor_record(buf, sizeof(buf), sap, m);
754 xlog(L_ERROR, "Failed to insert: record too long");
759 * If exclusive create fails, we're adding a new line to an
762 fd = open(path, O_WRONLY | O_CREAT | O_EXCL | O_SYNC, S_IRUSR | S_IWUSR);
764 if (errno != EEXIST) {
765 xlog(L_ERROR, "Failed to insert: creating %s: %m", path);
769 result = nsm_append_monitored_host(path, buf);
774 len = write(fd, buf, size);
775 if (exact_error_check(len, size)) {
776 xlog_warn("Failed to insert: writing %s: %m", path);
781 if (close(fd) == -1) {
782 xlog(L_ERROR, "Failed to insert: closing %s: %m", path);
792 __attribute__((__noinline__))
794 nsm_parse_line(char *line, struct sockaddr_in *sin, struct mon *m)
800 c = strchr(line, '\n');
804 count = sscanf(line, "%8x %8x %8x %8x ",
805 (unsigned int *)&sin->sin_addr.s_addr,
806 (unsigned int *)&m->mon_id.my_id.my_prog,
807 (unsigned int *)&m->mon_id.my_id.my_vers,
808 (unsigned int *)&m->mon_id.my_id.my_proc);
812 c = line + RPCARGSLEN;
813 for (i = 0; i < SM_PRIV_SIZE; i++) {
814 if (sscanf(c, "%2x", &tmp) != 1)
816 m->priv[i] = (char)tmp;
821 m->mon_id.mon_name = c;
822 while (*c != '\0' && *c != ' ')
828 m->mon_id.my_id.my_name = c;
834 * Stuff a 'struct mon' with callback data, and call @func.
836 * Returns the count of in-core records created.
839 nsm_read_line(const char *hostname, const time_t timestamp, char *line,
842 struct sockaddr_in sin = {
843 .sin_family = AF_INET,
847 if (!nsm_parse_line(line, &sin, &m))
850 return func(hostname, (struct sockaddr *)(char *)&sin, &m, timestamp);
854 * Given a filename, reads data from a file under "directory"
855 * and invokes @func so caller can populate their in-core
856 * database with this data.
859 nsm_load_host(const char *directory, const char *filename, nsm_populate_t func)
861 char buf[LINELEN + 1 + SM_MAXSTRLEN + 2];
862 unsigned int result = 0;
867 path = nsm_make_record_pathname(directory, filename);
871 if (lstat(path, &stb) == -1) {
872 xlog(L_ERROR, "Failed to stat %s: %m", path);
875 if (!S_ISREG(stb.st_mode)) {
876 xlog(D_GENERAL, "Skipping non-regular file %s",
881 f = fopen(path, "r");
883 xlog(L_ERROR, "Failed to open %s: %m", path);
887 while (fgets(buf, (int)sizeof(buf), f) != NULL) {
888 buf[sizeof(buf) - 1] = '\0';
889 result += nsm_read_line(filename, stb.st_mtime, buf, func);
892 xlog(L_ERROR, "Failed to read monitor data from %s", path);
903 nsm_load_dir(const char *directory, nsm_populate_t func)
905 unsigned int count = 0;
910 path = nsm_make_pathname(directory);
912 xlog(L_ERROR, "Failed to allocate path for directory %s",
920 xlog(L_ERROR, "Failed to open directory %s: %m",
925 while ((de = readdir(dir)) != NULL) {
926 if (de->d_name[0] == '.')
929 count += nsm_load_host(directory, de->d_name, func);
937 * nsm_load_monitor_list - load list of hosts to monitor
938 * @func: callback function to create entry for one host
940 * Returns the count of hosts that were found in the directory.
943 nsm_load_monitor_list(nsm_populate_t func)
945 return nsm_load_dir(NSM_MONITOR_DIR, func);
949 * nsm_load_notify_list - load list of hosts to notify
950 * @func: callback function to create entry for one host
952 * Returns the count of hosts that were found in the directory.
955 nsm_load_notify_list(nsm_populate_t func)
957 return nsm_load_dir(NSM_NOTIFY_DIR, func);
961 nsm_delete_host(const char *directory, const char *hostname,
962 const char *mon_name, const char *my_name)
964 char line[LINELEN + 1 + SM_MAXSTRLEN + 2];
971 path = nsm_make_record_pathname(directory, hostname);
973 xlog(L_ERROR, "Bad filename, not deleting");
977 if (stat(path, &stb) == -1) {
978 xlog(L_ERROR, "Failed to delete: "
979 "could not stat original file %s: %m", path);
982 remaining = (size_t)stb.st_size + 1;
984 outbuf = malloc(remaining);
985 if (outbuf == NULL) {
986 xlog(L_ERROR, "Failed to delete: no memory");
990 f = fopen(path, "r");
992 xlog(L_ERROR, "Failed to delete: "
993 "could not open original file %s: %m", path);
998 * Walk the records in the file, and copy the non-matching
999 * ones to our output buffer.
1002 while (fgets(line, (int)sizeof(line), f) != NULL) {
1003 struct sockaddr_in sin;
1007 if (!nsm_parse_line(line, &sin, &m)) {
1008 xlog(L_ERROR, "Failed to delete: "
1009 "could not parse original file %s", path);
1014 if (strcmp(mon_name, m.mon_id.mon_name) == 0 &&
1015 strcmp(my_name, m.mon_id.my_id.my_name) == 0)
1018 /* nsm_parse_line destroys the contents of line[], so
1019 * reconstruct the copy in our output buffer. */
1020 len = nsm_create_monitor_record(next, remaining,
1021 (struct sockaddr *)(char *)&sin, &m);
1023 xlog(L_ERROR, "Failed to delete: "
1024 "could not construct output record");
1035 * If nothing was copied when we're done, then unlink the file.
1036 * Otherwise, atomically update the contents of the file.
1038 if (next != outbuf) {
1039 if (!nsm_atomic_write(path, outbuf, strlen(outbuf)))
1040 xlog(L_ERROR, "Failed to delete: "
1041 "could not write new file %s: %m", path);
1043 if (unlink(path) == -1)
1044 xlog(L_ERROR, "Failed to delete: "
1045 "could not unlink file %s: %m", path);
1054 * nsm_delete_monitored_host - delete on-disk record for monitored host
1055 * @hostname: '\0'-terminated C string containing hostname of record to delete
1056 * @mon_name: '\0'-terminated C string containing monname of record to delete
1057 * @my_name: '\0'-terminated C string containing myname of record to delete
1061 nsm_delete_monitored_host(const char *hostname, const char *mon_name,
1062 const char *my_name)
1064 nsm_delete_host(NSM_MONITOR_DIR, hostname, mon_name, my_name);
1068 * nsm_delete_notified_host - delete on-disk host record after notification
1069 * @hostname: '\0'-terminated C string containing hostname of record to delete
1070 * @mon_name: '\0'-terminated C string containing monname of record to delete
1071 * @my_name: '\0'-terminated C string containing myname of record to delete
1075 nsm_delete_notified_host(const char *hostname, const char *mon_name,
1076 const char *my_name)
1078 nsm_delete_host(NSM_NOTIFY_DIR, hostname, mon_name, my_name);