/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* * There are well defined policies for mapping uid and gid values to and * from utf8 strings, as specified in RFC 3530. The protocol ops that are * most significantly affected by any changes in policy are GETATTR and * SETATTR, as these have different behavior depending on whether the id * mapping code is executing on the client or server. Thus, the following * rules represents the latest incantation of the id mapping policies. * * 1) For the case in which the nfsmapid(1m) daemon has _never_ been * started, the policy is to _always_ work with stringified uid's * and gid's * * 2) For the case in which the nfsmapid(1m) daemon _was_ started but * has either died or become unresponsive, the mapping policies are * as follows: * * Server Client * .-------------------------------.---------------------------------. * | | | * | . Respond to req by replying | . If attr string does not have | * | success and map the [u/g]id | '@' sign, attempt to decode | * | into its literal id string | a stringified id; map to | * | | *ID_NOBODY if not an encoded | * | | id. | * | | | * GETATTR | | . If attr string _does_ have | * | | '@' sign | * | | Map to *ID_NOBODY on failure. | * | | | * | nfs_idmap_*id_str | nfs_idmap_str_*id | * +-------------------------------+---------------------------------+ * | | | * | . Respond to req by returning | . _Must_ map the user's passed | * | ECOMM, which will be mapped | in [u/g]id into it's network | * | to NFS4ERR_DELAY to clnt | attr string, so contact the | * | | daemon, retrying forever if | * | Server must not allow the | necessary, unless interrupted | * SETATTR | mapping to *ID_NOBODY upon | | * | lack of communication with | Client _should_ specify the | * | the daemon, which could | correct attr string for a | * | result in the file being | SETATTR operation, otherwise | * | inadvertently given away ! | it can also result in the | * | | file being inadvertently | * | | given away ! | * | | | * | nfs_idmap_str_*id | nfs_idmap_*id_str | * `-------------------------------'---------------------------------' * * 3) Lastly, in order to leverage better cache utilization whenever * communication with nfsmapid(1m) is currently hindered, cache * entry eviction is throttled whenever nfsidmap_daemon_dh == NULL. * * * Server-side behavior for upcall communication errors * ==================================================== * * GETATTR - Server-side GETATTR *id to attr string conversion policies * for unresponsive/dead nfsmapid(1m) daemon * * a) If the *id is *ID_NOBODY, the string "nobody" is returned * * b) If the *id is not *ID_NOBODY _and_ the nfsmapid(1m) daemon * _is_ operational, the daemon is contacted to convert the * [u/g]id into a string of type "[user/group]@domain" * * c) If the nfsmapid(1m) daemon has died or has become unresponsive, * the server returns status == NFS4_OK for the GETATTR operation, * and returns a strigified [u/g]id to let the client map it into * the appropriate value. * * SETATTR - Server-side SETATTR attr string to *id conversion policies * for unresponsive/dead nfsmapid(1m) daemon * * a) If the otw string is a stringified uid (ie. does _not_ contain * an '@' sign and is of the form "12345") then the literal uid is * decoded and it is used to perform the mapping. * * b) If, on the other hand, the otw string _is_ of the form * "[user/group]@domain" and problems arise contacting nfsmapid(1m), * the SETATTR operation _must_ fail w/NFS4ERR_DELAY, as the server * cannot default to *ID_NOBODY, which would allow a file to be * given away by setting it's owner or owner_group to "nobody". */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Truly global modular globals */ zone_key_t nfsidmap_zone_key; static list_t nfsidmap_globals_list; static kmutex_t nfsidmap_globals_lock; static kmem_cache_t *nfsidmap_cache; static uint_t pkp_tab[NFSID_CACHE_ANCHORS]; static int nfs4_idcache_tout; /* * Some useful macros */ #define MOD2(a, pow_of_2) ((a) & ((pow_of_2) - 1)) #define _CACHE_TOUT (60*60) /* secs in 1 hour */ #define TIMEOUT(x) (gethrestime_sec() > \ ((x) + nfs4_idcache_tout)) /* * Max length of valid id string including the trailing null */ #define _MAXIDSTRLEN 11 /* * Pearson's string hash * * See: Communications of the ACM, June 1990 Vol 33 pp 677-680 * http://www.acm.org/pubs/citations/journals/cacm/1990-33-6/p677-pearson */ #define PS_HASH(msg, hash, len) \ { \ uint_t key = 0x12345678; /* arbitrary value */ \ int i; \ \ (hash) = MOD2((key + (len)), NFSID_CACHE_ANCHORS); \ \ for (i = 0; i < (len); i++) { \ (hash) = MOD2(((hash) + (msg)[i]), NFSID_CACHE_ANCHORS); \ (hash) = pkp_tab[(hash)]; \ } \ } #define ID_HASH(id, hash) \ { \ (hash) = MOD2(((id) ^ NFSID_CACHE_ANCHORS), NFSID_CACHE_ANCHORS); \ } /* * Prototypes */ static void *nfs_idmap_init_zone(zoneid_t); static void nfs_idmap_fini_zone(zoneid_t, void *); static int is_stringified_id(utf8string *); static void init_pkp_tab(void); static void nfs_idmap_i2s_literal(uid_t, utf8string *); static int nfs_idmap_s2i_literal(utf8string *, uid_t *, int); static void nfs_idmap_reclaim(void *); static void nfs_idmap_cache_reclaim(idmap_cache_info_t *); static void nfs_idmap_cache_create(idmap_cache_info_t *, const char *); static void nfs_idmap_cache_destroy(idmap_cache_info_t *); static void nfs_idmap_cache_flush(idmap_cache_info_t *); static uint_t nfs_idmap_cache_s2i_lkup(idmap_cache_info_t *, utf8string *, uint_t *, uid_t *); static uint_t nfs_idmap_cache_i2s_lkup(idmap_cache_info_t *, uid_t, uint_t *, utf8string *); static void nfs_idmap_cache_s2i_insert(idmap_cache_info_t *, uid_t, utf8string *, hash_stat, uint_t); static void nfs_idmap_cache_i2s_insert(idmap_cache_info_t *, uid_t, utf8string *, hash_stat, uint_t); static void nfs_idmap_cache_rment(nfsidmap_t *); /* * Initialization routine for NFSv4 id mapping */ void nfs_idmap_init(void) { /* * Initialize Pearson's Table */ init_pkp_tab(); /* * Initialize the kmem cache */ nfsidmap_cache = kmem_cache_create("NFS_idmap_cache", sizeof (nfsidmap_t), 0, NULL, NULL, nfs_idmap_reclaim, NULL, NULL, 0); /* * If not set in "/etc/system", set to default value */ if (!nfs4_idcache_tout) nfs4_idcache_tout = _CACHE_TOUT; /* * Initialize the list of nfsidmap_globals */ mutex_init(&nfsidmap_globals_lock, NULL, MUTEX_DEFAULT, NULL); list_create(&nfsidmap_globals_list, sizeof (struct nfsidmap_globals), offsetof(struct nfsidmap_globals, nig_link)); /* * Initialize the zone_key_t for per-zone idmaps */ zone_key_create(&nfsidmap_zone_key, nfs_idmap_init_zone, NULL, nfs_idmap_fini_zone); } /* * Called only when module was not loaded properly */ void nfs_idmap_fini(void) { (void) zone_key_delete(nfsidmap_zone_key); list_destroy(&nfsidmap_globals_list); mutex_destroy(&nfsidmap_globals_lock); kmem_cache_destroy(nfsidmap_cache); } /*ARGSUSED*/ static void * nfs_idmap_init_zone(zoneid_t zoneid) { struct nfsidmap_globals *nig; nig = kmem_alloc(sizeof (*nig), KM_SLEEP); nig->nig_msg_done = 0; mutex_init(&nig->nfsidmap_daemon_lock, NULL, MUTEX_DEFAULT, NULL); /* * nfsidmap certainly isn't running. */ nig->nfsidmap_pid = NOPID; nig->nfsidmap_daemon_dh = NULL; /* * Create the idmap caches */ nfs_idmap_cache_create(&nig->u2s_ci, "u2s_cache"); nig->u2s_ci.nfsidmap_daemon_dh = &nig->nfsidmap_daemon_dh; nfs_idmap_cache_create(&nig->s2u_ci, "s2u_cache"); nig->s2u_ci.nfsidmap_daemon_dh = &nig->nfsidmap_daemon_dh; nfs_idmap_cache_create(&nig->g2s_ci, "g2s_cache"); nig->g2s_ci.nfsidmap_daemon_dh = &nig->nfsidmap_daemon_dh; nfs_idmap_cache_create(&nig->s2g_ci, "s2g_cache"); nig->s2g_ci.nfsidmap_daemon_dh = &nig->nfsidmap_daemon_dh; /* * Add to global list. */ mutex_enter(&nfsidmap_globals_lock); list_insert_head(&nfsidmap_globals_list, nig); mutex_exit(&nfsidmap_globals_lock); return (nig); } /*ARGSUSED*/ static void nfs_idmap_fini_zone(zoneid_t zoneid, void *arg) { struct nfsidmap_globals *nig = arg; /* * Remove from list. */ mutex_enter(&nfsidmap_globals_lock); list_remove(&nfsidmap_globals_list, nig); /* * Destroy the idmap caches */ nfs_idmap_cache_destroy(&nig->u2s_ci); nfs_idmap_cache_destroy(&nig->s2u_ci); nfs_idmap_cache_destroy(&nig->g2s_ci); nfs_idmap_cache_destroy(&nig->s2g_ci); mutex_exit(&nfsidmap_globals_lock); /* * Cleanup */ if (nig->nfsidmap_daemon_dh) door_ki_rele(nig->nfsidmap_daemon_dh); mutex_destroy(&nig->nfsidmap_daemon_lock); kmem_free(nig, sizeof (*nig)); } /* * Convert a user utf-8 string identifier into its local uid. */ int nfs_idmap_str_uid(utf8string *u8s, uid_t *uid, bool_t isserver) { int error; uint_t hashno = 0; const char *whoami = "nfs_idmap_str_uid"; struct nfsidmap_globals *nig; struct mapid_arg *mapargp; struct mapid_res mapres; struct mapid_res *mapresp = &mapres; struct mapid_res *resp = mapresp; door_arg_t door_args; door_handle_t dh; nig = zone_getspecific(nfsidmap_zone_key, nfs_zone()); ASSERT(nig != NULL); if (!u8s || !u8s->utf8string_val || u8s->utf8string_len == 0 || (u8s->utf8string_val[0] == '\0')) { *uid = UID_NOBODY; return (isserver ? EINVAL : 0); } /* * If "nobody", just short circuit and bail */ if (bcmp(u8s->utf8string_val, "nobody", 6) == 0) { *uid = UID_NOBODY; return (0); } /* * Start-off with upcalls disabled, and once nfsmapid(1m) is up and * running, we'll leverage it's first flush to let the kernel know * when it's up and available to perform mappings. Also, on client * only, be smarter about when to issue upcalls by checking the * string for existence of an '@' sign. If no '@' sign, then we just * make our best effort to decode the string ourselves. */ retry: mutex_enter(&nig->nfsidmap_daemon_lock); dh = nig->nfsidmap_daemon_dh; if (dh) door_ki_hold(dh); mutex_exit(&nig->nfsidmap_daemon_lock); if (dh == NULL || nig->nfsidmap_pid == curproc->p_pid || (!utf8_strchr(u8s, '@') && !isserver)) { if (dh) door_ki_rele(dh); error = nfs_idmap_s2i_literal(u8s, uid, isserver); /* * If we get a numeric value, but we only do so because * we are nfsmapid, return ENOTSUP to indicate a valid * response, but not to cache it. */ if (!error && nig->nfsidmap_pid == curproc->p_pid) return (ENOTSUP); return (error); } /* cache hit */ if (nfs_idmap_cache_s2i_lkup(&nig->s2u_ci, u8s, &hashno, uid)) { door_ki_rele(dh); return (0); } /* cache miss */ mapargp = kmem_alloc(MAPID_ARG_LEN(u8s->utf8string_len), KM_SLEEP); mapargp->cmd = NFSMAPID_STR_UID; mapargp->u_arg.len = u8s->utf8string_len; (void) bcopy(u8s->utf8string_val, mapargp->str, mapargp->u_arg.len); mapargp->str[mapargp->u_arg.len] = '\0'; door_args.data_ptr = (char *)mapargp; door_args.data_size = MAPID_ARG_LEN(mapargp->u_arg.len); door_args.desc_ptr = NULL; door_args.desc_num = 0; door_args.rbuf = (char *)mapresp; door_args.rsize = sizeof (struct mapid_res); error = door_ki_upcall_limited(dh, &door_args, NULL, SIZE_MAX, 0); if (!error) { resp = (struct mapid_res *)door_args.rbuf; /* Should never provide daemon with bad args */ ASSERT(resp->status != NFSMAPID_INVALID); switch (resp->status) { case NFSMAPID_OK: /* * Valid mapping. Cache it. */ *uid = resp->u_res.uid; nfs_idmap_cache_s2i_insert(&nig->s2u_ci, *uid, u8s, HQ_HASH_HINT, hashno); break; case NFSMAPID_NUMSTR: /* * string came in as stringified id. Don't cache ! * * nfsmapid(1m) semantics have changed in order to * support diskless clients. Thus, for stringified * id's that have passwd/group entries, we'll go * ahead and map them, returning no error. */ *uid = resp->u_res.uid; break; case NFSMAPID_BADDOMAIN: /* * Make the offending "user@domain" string readily * available to D scripts that enable the probe. */ DTRACE_PROBE1(nfs4__str__uid, char *, mapargp->str); /* FALLTHROUGH */ case NFSMAPID_INVALID: case NFSMAPID_UNMAPPABLE: case NFSMAPID_INTERNAL: case NFSMAPID_BADID: case NFSMAPID_NOTFOUND: default: /* * For now, treat all of these errors as equal. * * Return error on the server side, then the * server returns NFS4_BADOWNER to the client. * On client side, just map to UID_NOBODY. */ if (isserver) error = EPERM; else *uid = UID_NOBODY; break; } kmem_free(mapargp, MAPID_ARG_LEN(u8s->utf8string_len)); if (resp != mapresp) kmem_free(door_args.rbuf, door_args.rsize); door_ki_rele(dh); return (error); } kmem_free(mapargp, MAPID_ARG_LEN(u8s->utf8string_len)); /* * We got some door error */ switch (error) { case EINTR: /* * If we took an interrupt we have to bail out. */ if (ttolwp(curthread) && ISSIG(curthread, JUSTLOOKING)) { door_ki_rele(dh); return (EINTR); } /* * We may have gotten EINTR for other reasons like the * door being revoked on us, instead of trying to * extract this out of the door handle, sleep * and try again, if still revoked we will get EBADF * next time through. */ /* FALLTHROUGH */ case EAGAIN: /* process may be forking */ door_ki_rele(dh); /* * Back off for a bit */ delay(hz); goto retry; default: /* Unknown must be fatal */ case EBADF: /* Invalid door */ case EINVAL: /* Not a door, wrong target */ /* * A fatal door error, if our failing door handle is the * current door handle, clean up our state and * mark the server dead. */ mutex_enter(&nig->nfsidmap_daemon_lock); if (dh == nig->nfsidmap_daemon_dh) { door_ki_rele(nig->nfsidmap_daemon_dh); nig->nfsidmap_daemon_dh = NULL; } mutex_exit(&nig->nfsidmap_daemon_lock); door_ki_rele(dh); if (isserver) return (ECOMM); /* * Note: We've already done optimizations above to check * for '@' sign, so if we can't comm w/nfsmapid, we * _know_ this _can't_ be a stringified uid. */ if (!nig->nig_msg_done) { zcmn_err(getzoneid(), CE_WARN, "!%s: Can't communicate with mapping daemon " "nfsmapid", whoami); nig->nig_msg_done = 1; } *uid = UID_NOBODY; return (0); } /* NOTREACHED */ } /* * Convert a uid into its utf-8 string representation. */ int nfs_idmap_uid_str(uid_t uid, utf8string *u8s, bool_t isserver) { int error; uint_t hashno = 0; const char *whoami = "nfs_idmap_uid_str"; struct nfsidmap_globals *nig; struct mapid_arg maparg; struct mapid_res mapres; struct mapid_res *mapresp = &mapres; struct mapid_res *resp = mapresp; door_arg_t door_args; door_handle_t dh; nig = zone_getspecific(nfsidmap_zone_key, nfs_zone()); ASSERT(nig != NULL); /* * If the supplied uid is "nobody", then we don't look at the * cache, since we DON'T cache it in the u2s_cache. We cannot * tell two strings apart from caching the same uid. */ if (uid == UID_NOBODY) { (void) str_to_utf8("nobody", u8s); return (0); } /* * Start-off with upcalls disabled, and once nfsmapid(1m) is * up and running, we'll leverage it's first flush to let the * kernel know when it's up and available to perform mappings. * We fall back to answering with stringified uid's. */ retry: mutex_enter(&nig->nfsidmap_daemon_lock); dh = nig->nfsidmap_daemon_dh; if (dh) door_ki_hold(dh); mutex_exit(&nig->nfsidmap_daemon_lock); if (dh == NULL || nig->nfsidmap_pid == curproc->p_pid) { if (dh) door_ki_rele(dh); nfs_idmap_i2s_literal(uid, u8s); return (0); } /* cache hit */ if (nfs_idmap_cache_i2s_lkup(&nig->u2s_ci, uid, &hashno, u8s)) { door_ki_rele(dh); return (0); } /* cache miss */ maparg.cmd = NFSMAPID_UID_STR; maparg.u_arg.uid = uid; door_args.data_ptr = (char *)&maparg; door_args.data_size = sizeof (struct mapid_arg); door_args.desc_ptr = NULL; door_args.desc_num = 0; door_args.rbuf = (char *)mapresp; door_args.rsize = sizeof (struct mapid_res); error = door_ki_upcall_limited(dh, &door_args, NULL, SIZE_MAX, 0); if (!error) { resp = (struct mapid_res *)door_args.rbuf; /* Should never provide daemon with bad args */ ASSERT(resp->status != NFSMAPID_INVALID); switch (resp->status) { case NFSMAPID_OK: /* * We now have a valid result from the * user-land daemon, so cache the result (if need be). * Load return value first then do the caches. */ (void) str_to_utf8(resp->str, u8s); nfs_idmap_cache_i2s_insert(&nig->u2s_ci, uid, u8s, HQ_HASH_HINT, hashno); break; case NFSMAPID_INVALID: case NFSMAPID_UNMAPPABLE: case NFSMAPID_INTERNAL: case NFSMAPID_BADDOMAIN: case NFSMAPID_BADID: case NFSMAPID_NOTFOUND: default: /* * For now, treat all of these errors as equal. */ error = EPERM; break; } if (resp != mapresp) kmem_free(door_args.rbuf, door_args.rsize); door_ki_rele(dh); return (error); } /* * We got some door error */ switch (error) { case EINTR: /* * If we took an interrupt we have to bail out. */ if (ttolwp(curthread) && ISSIG(curthread, JUSTLOOKING)) { door_ki_rele(dh); return (EINTR); } /* * We may have gotten EINTR for other reasons like the * door being revoked on us, instead of trying to * extract this out of the door handle, sleep * and try again, if still revoked we will get EBADF * next time through. */ /* FALLTHROUGH */ case EAGAIN: /* process may be forking */ door_ki_rele(dh); /* * Back off for a bit */ delay(hz); goto retry; default: /* Unknown must be fatal */ case EBADF: /* Invalid door */ case EINVAL: /* Not a door, wrong target */ /* * A fatal door error, if our failing door handle is the * current door handle, clean up our state and * mark the server dead. */ mutex_enter(&nig->nfsidmap_daemon_lock); if (dh == nig->nfsidmap_daemon_dh) { door_ki_rele(nig->nfsidmap_daemon_dh); nig->nfsidmap_daemon_dh = NULL; } mutex_exit(&nig->nfsidmap_daemon_lock); door_ki_rele(dh); /* * Log error on client-side only */ if (!nig->nig_msg_done && !isserver) { zcmn_err(getzoneid(), CE_WARN, "!%s: Can't communicate with mapping daemon " "nfsmapid", whoami); nig->nig_msg_done = 1; } nfs_idmap_i2s_literal(uid, u8s); return (0); } /* NOTREACHED */ } /* * Convert a group utf-8 string identifier into its local gid. */ int nfs_idmap_str_gid(utf8string *u8s, gid_t *gid, bool_t isserver) { int error; uint_t hashno = 0; const char *whoami = "nfs_idmap_str_gid"; struct nfsidmap_globals *nig; struct mapid_arg *mapargp; struct mapid_res mapres; struct mapid_res *mapresp = &mapres; struct mapid_res *resp = mapresp; door_arg_t door_args; door_handle_t dh; nig = zone_getspecific(nfsidmap_zone_key, nfs_zone()); ASSERT(nig != NULL); if (!u8s || !u8s->utf8string_val || u8s->utf8string_len == 0 || (u8s->utf8string_val[0] == '\0')) { *gid = GID_NOBODY; return (isserver ? EINVAL : 0); } /* * If "nobody", just short circuit and bail */ if (bcmp(u8s->utf8string_val, "nobody", 6) == 0) { *gid = GID_NOBODY; return (0); } /* * Start-off with upcalls disabled, and once nfsmapid(1m) is up and * running, we'll leverage it's first flush to let the kernel know * when it's up and available to perform mappings. Also, on client * only, be smarter about when to issue upcalls by checking the * string for existence of an '@' sign. If no '@' sign, then we just * make our best effort to decode the string ourselves. */ retry: mutex_enter(&nig->nfsidmap_daemon_lock); dh = nig->nfsidmap_daemon_dh; if (dh) door_ki_hold(dh); mutex_exit(&nig->nfsidmap_daemon_lock); if (dh == NULL || nig->nfsidmap_pid == curproc->p_pid || (!utf8_strchr(u8s, '@') && !isserver)) { if (dh) door_ki_rele(dh); error = nfs_idmap_s2i_literal(u8s, gid, isserver); /* * If we get a numeric value, but we only do so because * we are nfsmapid, return ENOTSUP to indicate a valid * response, but not to cache it. */ if (!error && nig->nfsidmap_pid == curproc->p_pid) return (ENOTSUP); return (error); } /* cache hit */ if (nfs_idmap_cache_s2i_lkup(&nig->s2g_ci, u8s, &hashno, gid)) { door_ki_rele(dh); return (0); } /* cache miss */ mapargp = kmem_alloc(MAPID_ARG_LEN(u8s->utf8string_len), KM_SLEEP); mapargp->cmd = NFSMAPID_STR_GID; mapargp->u_arg.len = u8s->utf8string_len; (void) bcopy(u8s->utf8string_val, mapargp->str, mapargp->u_arg.len); mapargp->str[mapargp->u_arg.len] = '\0'; door_args.data_ptr = (char *)mapargp; door_args.data_size = MAPID_ARG_LEN(mapargp->u_arg.len); door_args.desc_ptr = NULL; door_args.desc_num = 0; door_args.rbuf = (char *)mapresp; door_args.rsize = sizeof (struct mapid_res); error = door_ki_upcall_limited(dh, &door_args, NULL, SIZE_MAX, 0); if (!error) { resp = (struct mapid_res *)door_args.rbuf; /* Should never provide daemon with bad args */ ASSERT(resp->status != NFSMAPID_INVALID); switch (resp->status) { case NFSMAPID_OK: /* * Valid mapping. Cache it. */ *gid = resp->u_res.gid; error = 0; nfs_idmap_cache_s2i_insert(&nig->s2g_ci, *gid, u8s, HQ_HASH_HINT, hashno); break; case NFSMAPID_NUMSTR: /* * string came in as stringified id. Don't cache ! * * nfsmapid(1m) semantics have changed in order to * support diskless clients. Thus, for stringified * id's that have passwd/group entries, we'll go * ahead and map them, returning no error. */ *gid = resp->u_res.gid; break; case NFSMAPID_BADDOMAIN: /* * Make the offending "group@domain" string readily * available to D scripts that enable the probe. */ DTRACE_PROBE1(nfs4__str__gid, char *, mapargp->str); /* FALLTHROUGH */ case NFSMAPID_INVALID: case NFSMAPID_UNMAPPABLE: case NFSMAPID_INTERNAL: case NFSMAPID_BADID: case NFSMAPID_NOTFOUND: default: /* * For now, treat all of these errors as equal. * * Return error on the server side, then the * server returns NFS4_BADOWNER to the client. * On client side, just map to GID_NOBODY. */ if (isserver) error = EPERM; else *gid = GID_NOBODY; break; } kmem_free(mapargp, MAPID_ARG_LEN(u8s->utf8string_len)); if (resp != mapresp) kmem_free(door_args.rbuf, door_args.rsize); door_ki_rele(dh); return (error); } kmem_free(mapargp, MAPID_ARG_LEN(u8s->utf8string_len)); /* * We got some door error */ switch (error) { case EINTR: /* * If we took an interrupt we have to bail out. */ if (ttolwp(curthread) && ISSIG(curthread, JUSTLOOKING)) { door_ki_rele(dh); return (EINTR); } /* * We may have gotten EINTR for other reasons like the * door being revoked on us, instead of trying to * extract this out of the door handle, sleep * and try again, if still revoked we will get EBADF * next time through. */ /* FALLTHROUGH */ case EAGAIN: /* process may be forking */ door_ki_rele(dh); /* * Back off for a bit */ delay(hz); goto retry; default: /* Unknown must be fatal */ case EBADF: /* Invalid door */ case EINVAL: /* Not a door, wrong target */ /* * A fatal door error, clean up our state and * mark the server dead. */ mutex_enter(&nig->nfsidmap_daemon_lock); if (dh == nig->nfsidmap_daemon_dh) { door_ki_rele(nig->nfsidmap_daemon_dh); nig->nfsidmap_daemon_dh = NULL; } mutex_exit(&nig->nfsidmap_daemon_lock); door_ki_rele(dh); if (isserver) return (ECOMM); /* * Note: We've already done optimizations above to check * for '@' sign, so if we can't comm w/nfsmapid, we * _know_ this _can't_ be a stringified gid. */ if (!nig->nig_msg_done) { zcmn_err(getzoneid(), CE_WARN, "!%s: Can't communicate with mapping daemon " "nfsmapid", whoami); nig->nig_msg_done = 1; } *gid = GID_NOBODY; return (0); } /* NOTREACHED */ } /* * Convert a gid into its utf-8 string representation. */ int nfs_idmap_gid_str(gid_t gid, utf8string *u8s, bool_t isserver) { int error; uint_t hashno = 0; const char *whoami = "nfs_idmap_gid_str"; struct nfsidmap_globals *nig; struct mapid_arg maparg; struct mapid_res mapres; struct mapid_res *mapresp = &mapres; struct mapid_res *resp = mapresp; door_arg_t door_args; door_handle_t dh; nig = zone_getspecific(nfsidmap_zone_key, nfs_zone()); ASSERT(nig != NULL); /* * If the supplied gid is "nobody", then we don't look at the * cache, since we DON'T cache it in the u2s_cache. We cannot * tell two strings apart from caching the same gid. */ if (gid == GID_NOBODY) { (void) str_to_utf8("nobody", u8s); return (0); } /* * Start-off with upcalls disabled, and once nfsmapid(1m) is * up and running, we'll leverage it's first flush to let the * kernel know when it's up and available to perform mappings. * We fall back to answering with stringified gid's. */ retry: mutex_enter(&nig->nfsidmap_daemon_lock); dh = nig->nfsidmap_daemon_dh; if (dh) door_ki_hold(dh); mutex_exit(&nig->nfsidmap_daemon_lock); if (dh == NULL || nig->nfsidmap_pid == curproc->p_pid) { if (dh) door_ki_rele(dh); nfs_idmap_i2s_literal(gid, u8s); return (0); } /* cache hit */ if (nfs_idmap_cache_i2s_lkup(&nig->g2s_ci, gid, &hashno, u8s)) { door_ki_rele(dh); return (0); } /* cache miss */ maparg.cmd = NFSMAPID_GID_STR; maparg.u_arg.gid = gid; door_args.data_ptr = (char *)&maparg; door_args.data_size = sizeof (struct mapid_arg); door_args.desc_ptr = NULL; door_args.desc_num = 0; door_args.rbuf = (char *)mapresp; door_args.rsize = sizeof (struct mapid_res); error = door_ki_upcall_limited(dh, &door_args, NULL, SIZE_MAX, 0); if (!error) { resp = (struct mapid_res *)door_args.rbuf; /* Should never provide daemon with bad args */ ASSERT(resp->status != NFSMAPID_INVALID); switch (resp->status) { case NFSMAPID_OK: /* * We now have a valid result from the * user-land daemon, so cache the result (if need be). * Load return value first then do the caches. */ (void) str_to_utf8(resp->str, u8s); nfs_idmap_cache_i2s_insert(&nig->g2s_ci, gid, u8s, HQ_HASH_HINT, hashno); break; case NFSMAPID_INVALID: case NFSMAPID_UNMAPPABLE: case NFSMAPID_INTERNAL: case NFSMAPID_BADDOMAIN: case NFSMAPID_BADID: case NFSMAPID_NOTFOUND: default: /* * For now, treat all of these errors as equal. */ error = EPERM; break; } if (resp != mapresp) kmem_free(door_args.rbuf, door_args.rsize); door_ki_rele(dh); return (error); } /* * We got some door error */ switch (error) { case EINTR: /* * If we took an interrupt we have to bail out. */ if (ttolwp(curthread) && ISSIG(curthread, JUSTLOOKING)) { door_ki_rele(dh); return (EINTR); } /* * We may have gotten EINTR for other reasons like the * door being revoked on us, instead of trying to * extract this out of the door handle, sleep * and try again, if still revoked we will get EBADF * next time through. */ /* FALLTHROUGH */ case EAGAIN: /* process may be forking */ door_ki_rele(dh); /* * Back off for a bit */ delay(hz); goto retry; default: /* Unknown must be fatal */ case EBADF: /* Invalid door */ case EINVAL: /* Not a door, wrong target */ /* * A fatal door error, if our failing door handle is the * current door handle, clean up our state and * mark the server dead. */ mutex_enter(&nig->nfsidmap_daemon_lock); if (dh == nig->nfsidmap_daemon_dh) { door_ki_rele(nig->nfsidmap_daemon_dh); nig->nfsidmap_daemon_dh = NULL; } door_ki_rele(dh); mutex_exit(&nig->nfsidmap_daemon_lock); /* * Log error on client-side only */ if (!nig->nig_msg_done && !isserver) { zcmn_err(getzoneid(), CE_WARN, "!%s: Can't communicate with mapping daemon " "nfsmapid", whoami); nig->nig_msg_done = 1; } nfs_idmap_i2s_literal(gid, u8s); return (0); } /* NOTREACHED */ } /* -- idmap cache management -- */ /* * Cache creation and initialization routine */ static void nfs_idmap_cache_create(idmap_cache_info_t *cip, const char *name) { int i; nfsidhq_t *hq = NULL; cip->table = kmem_alloc((NFSID_CACHE_ANCHORS * sizeof (nfsidhq_t)), KM_SLEEP); for (i = 0, hq = cip->table; i < NFSID_CACHE_ANCHORS; i++, hq++) { hq->hq_que_forw = hq; hq->hq_que_back = hq; mutex_init(&(hq->hq_lock), NULL, MUTEX_DEFAULT, NULL); } cip->name = name; } /* * Cache destruction routine * * Ops per hash queue * * - dequeue cache entries * - release string storage per entry * - release cache entry storage * - destroy HQ lock when HQ is empty * - once all HQ's empty, release HQ storage */ static void nfs_idmap_cache_destroy(idmap_cache_info_t *cip) { int i; nfsidhq_t *hq; ASSERT(MUTEX_HELD(&nfsidmap_globals_lock)); nfs_idmap_cache_flush(cip); /* * We can safely destroy per-queue locks since the only * other entity that could be mucking with this table is the * kmem reaper thread which does everything under * nfsidmap_globals_lock (which we're holding). */ for (i = 0, hq = cip->table; i < NFSID_CACHE_ANCHORS; i++, hq++) mutex_destroy(&(hq->hq_lock)); kmem_free(cip->table, NFSID_CACHE_ANCHORS * sizeof (nfsidhq_t)); } void nfs_idmap_args(struct nfsidmap_args *idmp) { struct nfsidmap_globals *nig; nig = zone_getspecific(nfsidmap_zone_key, nfs_zone()); ASSERT(nig != NULL); nfs_idmap_cache_flush(&nig->u2s_ci); nfs_idmap_cache_flush(&nig->s2u_ci); nfs_idmap_cache_flush(&nig->g2s_ci); nfs_idmap_cache_flush(&nig->s2g_ci); /* * nfsmapid(1m) up and running; enable upcalls * State: * 0 Just flush caches * 1 Re-establish door knob */ if (idmp->state) { /* * When reestablishing the nfsmapid we need to * not only purge the idmap cache but also * the dnlc as it will have cached uid/gid's. * While heavyweight, this should almost never happen */ dnlc_purge(); /* * Invalidate the attrs of all rnodes to force new uid and gids */ nfs4_rnode_invalidate(NULL); mutex_enter(&nig->nfsidmap_daemon_lock); if (nig->nfsidmap_daemon_dh) door_ki_rele(nig->nfsidmap_daemon_dh); nig->nfsidmap_daemon_dh = door_ki_lookup(idmp->did); nig->nfsidmap_pid = curproc->p_pid; nig->nig_msg_done = 0; mutex_exit(&nig->nfsidmap_daemon_lock); } } /* * Cache flush routine * * The only serialization required it to hold the hash chain lock * when destroying cache entries. There is no need to prevent access * to all hash chains while flushing. It is possible that (valid) * entries could be cached in later hash chains after we start flushing. * It is unfortunate that the entry will be instantly destroyed, but * it isn't a major concern. This is only a cache. It'll be repopulated. * * Ops per hash queue * * - dequeue cache entries * - release string storage per entry * - release cache entry storage */ static void nfs_idmap_cache_flush(idmap_cache_info_t *cip) { int i; nfsidmap_t *p, *next; nfsidhq_t *hq; for (i = 0, hq = cip->table; i < NFSID_CACHE_ANCHORS; i++, hq++) { mutex_enter(&(hq->hq_lock)); /* * remove list from hash header so we can release * the lock early. */ p = hq->hq_lru_forw; hq->hq_que_forw = hq; hq->hq_que_back = hq; mutex_exit(&(hq->hq_lock)); /* * Iterate over the orphan'd list and free all elements. * There's no need to bother with remque since we're * freeing the entire list. */ while (p != (nfsidmap_t *)hq) { next = p->id_forw; if (p->id_val != 0) kmem_free(p->id_val, p->id_len); kmem_cache_free(nfsidmap_cache, p); p = next; } } } static void nfs_idmap_cache_reclaim(idmap_cache_info_t *cip) { nfsidhq_t *hq; nfsidmap_t *pprev = NULL; int i; nfsidmap_t *p; ASSERT(cip != NULL && cip->table != NULL); /* * If the daemon is down, do not flush anything */ if ((*cip->nfsidmap_daemon_dh) == NULL) return; for (i = 0, hq = cip->table; i < NFSID_CACHE_ANCHORS; i++, hq++) { if (!mutex_tryenter(&(hq->hq_lock))) continue; /* * Start at end of list and work backwards since LRU */ for (p = hq->hq_lru_back; p != (nfsidmap_t *)hq; p = pprev) { pprev = p->id_back; /* * List is LRU. If trailing end does not * contain stale entries, then no need to * continue. */ if (!TIMEOUT(p->id_time)) break; nfs_idmap_cache_rment(p); } mutex_exit(&(hq->hq_lock)); } } /* * Callback reclaim function for VM. We reap timed-out entries from all hash * tables in all zones. */ /* ARGSUSED */ void nfs_idmap_reclaim(void *arg) { struct nfsidmap_globals *nig; mutex_enter(&nfsidmap_globals_lock); for (nig = list_head(&nfsidmap_globals_list); nig != NULL; nig = list_next(&nfsidmap_globals_list, nig)) { nfs_idmap_cache_reclaim(&nig->u2s_ci); nfs_idmap_cache_reclaim(&nig->s2u_ci); nfs_idmap_cache_reclaim(&nig->g2s_ci); nfs_idmap_cache_reclaim(&nig->s2g_ci); } mutex_exit(&nfsidmap_globals_lock); } /* * Search the specified cache for the existence of the specified utf-8 * string. If found, the corresponding mapping is returned in id_buf and * the cache entry is updated to the head of the LRU list. The computed * hash queue number, is returned in hashno. */ static uint_t nfs_idmap_cache_s2i_lkup(idmap_cache_info_t *cip, /* cache info ptr */ utf8string *u8s, /* utf8 string to resolve */ uint_t *hashno, /* hash number, retval */ uid_t *id_buf) /* if found, id for u8s */ { nfsidmap_t *p; nfsidmap_t *pnext; nfsidhq_t *hq; uint_t hash; char *rqst_c_str; uint_t rqst_len; uint_t found_stat = 0; if ((rqst_c_str = utf8_to_str(u8s, &rqst_len, NULL)) == NULL) { /* * Illegal string, return not found. */ return (0); } /* * Compute hash queue */ PS_HASH(rqst_c_str, hash, rqst_len - 1); *hashno = hash; hq = &cip->table[hash]; /* * Look for the entry in the HQ */ mutex_enter(&(hq->hq_lock)); for (p = hq->hq_lru_forw; p != (nfsidmap_t *)hq; p = pnext) { pnext = p->id_forw; /* * Check entry for staleness first, as user's id * may have changed and may need to be remapped. * Note that we don't evict entries from the cache * if we're having trouble contacting nfsmapid(1m) */ if (TIMEOUT(p->id_time) && (*cip->nfsidmap_daemon_dh) != NULL) { nfs_idmap_cache_rment(p); continue; } /* * Compare equal length strings */ if (p->id_len == (rqst_len - 1)) { if (bcmp(p->id_val, rqst_c_str, (rqst_len - 1)) == 0) { /* * Found it. Update it and load return value. */ *id_buf = p->id_no; remque(p); insque(p, hq); p->id_time = gethrestime_sec(); found_stat = 1; break; } } } mutex_exit(&(hq->hq_lock)); if (rqst_c_str != NULL) kmem_free(rqst_c_str, rqst_len); return (found_stat); } /* * Search the specified cache for the existence of the specified utf8 * string, as it may have been inserted before this instance got a chance * to do it. If NOT found, then a new entry is allocated for the specified * cache, and inserted. The hash queue number is obtained from hash_number * if the behavior is HQ_HASH_HINT, or computed otherwise. */ static void nfs_idmap_cache_s2i_insert(idmap_cache_info_t *cip, /* cache info ptr */ uid_t id, /* id result from upcall */ utf8string *u8s, /* utf8 string to resolve */ hash_stat behavior, /* hash algorithm behavior */ uint_t hash_number) /* hash number iff hint */ { uint_t hashno; char *c_str; nfsidhq_t *hq; nfsidmap_t *newp; nfsidmap_t *p; nfsidmap_t *pnext; uint_t c_len; /* * This shouldn't fail, since already successful at lkup. * So, if it does happen, just drop the request-to-insert * on the floor. */ if ((c_str = utf8_to_str(u8s, &c_len, NULL)) == NULL) return; /* * Obtain correct hash queue to insert new entry in */ switch (behavior) { case HQ_HASH_HINT: hashno = hash_number; break; case HQ_HASH_FIND: default: PS_HASH(c_str, hashno, c_len - 1); break; } hq = &cip->table[hashno]; /* * Look for an existing entry in the cache. If one exists * update it, and return. Otherwise, allocate a new cache * entry, initialize it and insert it. */ mutex_enter(&(hq->hq_lock)); for (p = hq->hq_lru_forw; p != (nfsidmap_t *)hq; p = pnext) { pnext = p->id_forw; /* * Check entry for staleness first, as user's id * may have changed and may need to be remapped. * Note that we don't evict entries from the cache * if we're having trouble contacting nfsmapid(1m) */ if (TIMEOUT(p->id_time) && (*cip->nfsidmap_daemon_dh) != NULL) { nfs_idmap_cache_rment(p); continue; } /* * Compare equal length strings */ if (p->id_len == (c_len - 1)) { if (bcmp(p->id_val, c_str, (c_len - 1)) == 0) { /* * Move to front, and update time. */ remque(p); insque(p, hq); p->id_time = gethrestime_sec(); mutex_exit(&(hq->hq_lock)); kmem_free(c_str, c_len); return; } } } /* * Not found ! Alloc, init and insert new entry */ newp = kmem_cache_alloc(nfsidmap_cache, KM_SLEEP); newp->id_len = u8s->utf8string_len; newp->id_val = kmem_alloc(u8s->utf8string_len, KM_SLEEP); bcopy(u8s->utf8string_val, newp->id_val, u8s->utf8string_len); newp->id_no = id; newp->id_time = gethrestime_sec(); insque(newp, hq); mutex_exit(&(hq->hq_lock)); kmem_free(c_str, c_len); } /* * Search the specified cache for the existence of the specified id. * If found, the corresponding mapping is returned in u8s and the * cache entry is updated to the head of the LRU list. The computed * hash queue number, is returned in hashno. */ static uint_t nfs_idmap_cache_i2s_lkup(idmap_cache_info_t *cip, /* cache info ptr */ uid_t id, /* id to resolve */ uint_t *hashno, /* hash number, retval */ utf8string *u8s) /* if found, utf8 str for id */ { uint_t found_stat = 0; nfsidmap_t *p; nfsidmap_t *pnext; nfsidhq_t *hq; uint_t hash; /* * Compute hash queue */ ID_HASH(id, hash); *hashno = hash; hq = &cip->table[hash]; /* * Look for the entry in the HQ */ mutex_enter(&(hq->hq_lock)); for (p = hq->hq_lru_forw; p != (nfsidmap_t *)hq; p = pnext) { pnext = p->id_forw; /* * Check entry for staleness first, as user's id * may have changed and may need to be remapped. * Note that we don't evict entries from the cache * if we're having trouble contacting nfsmapid(1m) */ if (TIMEOUT(p->id_time) && (*cip->nfsidmap_daemon_dh) != NULL) { nfs_idmap_cache_rment(p); continue; } if (p->id_no == id) { /* * Found it. Load return value and move to head */ ASSERT(u8s->utf8string_val == NULL); u8s->utf8string_len = p->id_len; u8s->utf8string_val = kmem_alloc(p->id_len, KM_SLEEP); bcopy(p->id_val, u8s->utf8string_val, p->id_len); remque(p); insque(p, hq); p->id_time = gethrestime_sec(); found_stat = 1; break; } } mutex_exit(&(hq->hq_lock)); return (found_stat); } /* * Search the specified cache for the existence of the specified id, * as it may have been inserted before this instance got a chance to * do it. If NOT found, then a new entry is allocated for the specified * cache, and inserted. The hash queue number is obtained from hash_number * if the behavior is HQ_HASH_HINT, or computed otherwise. */ static void nfs_idmap_cache_i2s_insert(idmap_cache_info_t *cip, /* cache info ptr */ uid_t id, /* id to resolve */ utf8string *u8s, /* utf8 result from upcall */ hash_stat behavior, /* has algorithm behavior */ uint_t hash_number) /* hash number iff hint */ { uint_t hashno; nfsidhq_t *hq; nfsidmap_t *newp; nfsidmap_t *pnext; nfsidmap_t *p; /* * Obtain correct hash queue to insert new entry in */ switch (behavior) { case HQ_HASH_HINT: hashno = hash_number; break; case HQ_HASH_FIND: default: ID_HASH(id, hashno); break; } hq = &cip->table[hashno]; /* * Look for an existing entry in the cache. If one exists * update it, and return. Otherwise, allocate a new cache * entry, initialize and insert it. */ mutex_enter(&(hq->hq_lock)); for (p = hq->hq_lru_forw; p != (nfsidmap_t *)hq; p = pnext) { pnext = p->id_forw; /* * Check entry for staleness first, as user's id * may have changed and may need to be remapped. * Note that we don't evict entries from the cache * if we're having trouble contacting nfsmapid(1m) */ if (TIMEOUT(p->id_time) && (*cip->nfsidmap_daemon_dh) != NULL) { nfs_idmap_cache_rment(p); continue; } if ((p->id_no == id) && (p->id_len == u8s->utf8string_len)) { /* * Found It ! Move to front, and update time. */ remque(p); insque(p, hq); p->id_time = gethrestime_sec(); mutex_exit(&(hq->hq_lock)); return; } } /* * Not found ! Alloc, init and insert new entry */ newp = kmem_cache_alloc(nfsidmap_cache, KM_SLEEP); newp->id_len = u8s->utf8string_len; newp->id_val = kmem_alloc(u8s->utf8string_len, KM_SLEEP); bcopy(u8s->utf8string_val, newp->id_val, u8s->utf8string_len); newp->id_no = id; newp->id_time = gethrestime_sec(); insque(newp, hq); mutex_exit(&(hq->hq_lock)); } /* * Remove and free one cache entry */ static void nfs_idmap_cache_rment(nfsidmap_t *p) { remque(p); if (p->id_val != 0) kmem_free(p->id_val, p->id_len); kmem_cache_free(nfsidmap_cache, p); } #ifndef UID_MAX #define UID_MAX 2147483647 /* see limits.h */ #endif #ifndef isdigit #define isdigit(c) ((c) >= '0' && (c) <= '9') #endif static int is_stringified_id(utf8string *u8s) { int i; for (i = 0; i < u8s->utf8string_len; i++) if (!isdigit(u8s->utf8string_val[i])) return (0); return (1); } int nfs_idmap_s2i_literal(utf8string *u8s, uid_t *id, int isserver) { long tmp; int convd; char ids[_MAXIDSTRLEN]; /* * "nobody" unless we can actually decode it. */ *id = UID_NOBODY; /* * We're here because it has already been determined that the * string contains no '@' _or_ the nfsmapid daemon has yet to * be started. */ if (!is_stringified_id(u8s)) return (0); /* * If utf8string_len is greater than _MAXIDSTRLEN-1, then the id * is going to be greater than UID_MAX. Return id of "nobody" * right away. */ if (u8s->utf8string_len >= _MAXIDSTRLEN) return (isserver ? EPERM : 0); /* * Make sure we pass a NULL terminated 'C' string to ddi_strtol */ bcopy(u8s->utf8string_val, ids, u8s->utf8string_len); ids[u8s->utf8string_len] = '\0'; convd = ddi_strtol(ids, NULL, 10, &tmp); if (convd == 0 && tmp >= 0 && tmp <= UID_MAX) { *id = tmp; return (0); } return (isserver ? EPERM : 0); } static void nfs_idmap_i2s_literal(uid_t id, utf8string *u8s) { char ids[_MAXIDSTRLEN]; (void) snprintf(ids, _MAXIDSTRLEN, "%d", id); (void) str_to_utf8(ids, u8s); } /* -- Utility functions -- */ /* * Initialize table in pseudo-random fashion * for use in Pearson's string hash algorithm. * * See: Communications of the ACM, June 1990 Vol 33 pp 677-680 * http://www.acm.org/pubs/citations/journals/cacm/1990-33-6/p677-pearson */ static void init_pkp_tab(void) { int i; int j; int k = 7; uint_t s; for (i = 0; i < NFSID_CACHE_ANCHORS; i++) pkp_tab[i] = i; for (j = 0; j < 4; j++) for (i = 0; i < NFSID_CACHE_ANCHORS; i++) { s = pkp_tab[i]; k = MOD2((k + s), NFSID_CACHE_ANCHORS); pkp_tab[i] = pkp_tab[k]; pkp_tab[k] = s; } } char * utf8_strchr(utf8string *u8s, const char c) { int i; char *u8p = u8s->utf8string_val; int len = u8s->utf8string_len; for (i = 0; i < len; i++) if (u8p[i] == c) return (&u8p[i]); return (NULL); }