X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fptlrpc%2Fgss%2Fgss_svc_upcall.c;h=8096eaf71b910d518ec579d114d848b0948efd5c;hb=7a1cb859d83891c2d7d1a50891cede6bcb1a1106;hp=72438aeff3901f308c1c938b79894894b7125974;hpb=3192e52a89946f12fd36d28a686c169d01d36e64;p=fs%2Flustre-release.git diff --git a/lustre/ptlrpc/gss/gss_svc_upcall.c b/lustre/ptlrpc/gss/gss_svc_upcall.c index 72438ae..8096eaf 100644 --- a/lustre/ptlrpc/gss/gss_svc_upcall.c +++ b/lustre/ptlrpc/gss/gss_svc_upcall.c @@ -1,9 +1,10 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * Modifications for Lustre - * Copyright 2004 - 2006, Cluster File Systems, Inc. - * All rights reserved + * + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. + * + * Copyright (c) 2012, 2014, Intel Corporation. + * * Author: Eric Mei */ @@ -47,7 +48,6 @@ */ #define DEBUG_SUBSYSTEM S_SEC -#ifdef __KERNEL__ #include #include #include @@ -55,16 +55,15 @@ #include #include #include -#else -#include -#endif +#include #include #include #include #include -#include #include +#include +#include #include #include "gss_err.h" @@ -73,46 +72,68 @@ #define GSS_SVC_UPCALL_TIMEOUT (20) -static spinlock_t __ctx_index_lock = SPIN_LOCK_UNLOCKED; +static spinlock_t __ctx_index_lock; static __u64 __ctx_index; __u64 gss_get_next_ctx_index(void) { - __u64 idx; + __u64 idx; - spin_lock(&__ctx_index_lock); - idx = __ctx_index++; - spin_unlock(&__ctx_index_lock); + spin_lock(&__ctx_index_lock); + idx = __ctx_index++; + spin_unlock(&__ctx_index_lock); - return idx; + return idx; } static inline unsigned long hash_mem(char *buf, int length, int bits) { - unsigned long hash = 0; - unsigned long l = 0; - int len = 0; - unsigned char c; - - do { - if (len == length) { - c = (char) len; - len = -1; - } else - c = *buf++; - - l = (l << 8) | c; - len++; - - if ((len & (BITS_PER_LONG/8-1)) == 0) - hash = hash_long(hash^l, BITS_PER_LONG); - } while (len); - - return hash >> (BITS_PER_LONG - bits); + unsigned long hash = 0; + unsigned long l = 0; + int len = 0; + unsigned char c; + + do { + if (len == length) { + c = (char) len; + len = -1; + } else + c = *buf++; + + l = (l << 8) | c; + len++; + + if ((len & (BITS_PER_LONG/8-1)) == 0) + hash = hash_long(hash^l, BITS_PER_LONG); + } while (len); + + return hash >> (BITS_PER_LONG - bits); } +/* This compatibility can be removed once kernel 3.3 is used, + * since cache_register_net/cache_unregister_net are exported. + * Note that since kernel 3.4 cache_register and cache_unregister + * are removed. +*/ +static inline int _cache_register_net(struct cache_detail *cd, struct net *net) +{ +#ifdef HAVE_CACHE_REGISTER + return cache_register(cd); +#else + return cache_register_net(cd, net); +#endif +} +static inline void _cache_unregister_net(struct cache_detail *cd, + struct net *net) +{ +#ifdef HAVE_CACHE_REGISTER + cache_unregister(cd); +#else + cache_unregister_net(cd, net); +#endif +} /**************************************** - * rsi cache * + * rpc sec init (rsi) cache * ****************************************/ #define RSI_HASHBITS (6) @@ -120,39 +141,24 @@ static inline unsigned long hash_mem(char *buf, int length, int bits) #define RSI_HASHMASK (RSI_HASHMAX - 1) struct rsi { - struct cache_head h; - __u32 lustre_svc; - __u64 nid; - wait_queue_head_t waitq; - rawobj_t in_handle, in_token; - rawobj_t out_handle, out_token; - int major_status, minor_status; + struct cache_head h; + __u32 lustre_svc; + __u64 nid; + char nm_name[LUSTRE_NODEMAP_NAME_LENGTH + 1]; + wait_queue_head_t waitq; + rawobj_t in_handle, in_token; + rawobj_t out_handle, out_token; + int major_status, minor_status; }; +#ifdef HAVE_CACHE_HEAD_HLIST +static struct hlist_head rsi_table[RSI_HASHMAX]; +#else static struct cache_head *rsi_table[RSI_HASHMAX]; +#endif static struct cache_detail rsi_cache; -static struct rsi *rsi_lookup(struct rsi *item, int set); - -static void rsi_free(struct rsi *rsi) -{ - rawobj_free(&rsi->in_handle); - rawobj_free(&rsi->in_token); - rawobj_free(&rsi->out_handle); - rawobj_free(&rsi->out_token); -} - -static void rsi_put(struct cache_head *item, struct cache_detail *cd) -{ - struct rsi *rsi = container_of(item, struct rsi, h); - - LASSERT(atomic_read(&item->refcnt) > 0); - - if (cache_put(item, cd)) { - LASSERT(item->next == NULL); - rsi_free(rsi); - kfree(rsi); /* created by cache mgmt using kmalloc */ - } -} +static struct rsi *rsi_update(struct rsi *new, struct rsi *old); +static struct rsi *rsi_lookup(struct rsi *item); static inline int rsi_hash(struct rsi *item) { @@ -162,48 +168,73 @@ static inline int rsi_hash(struct rsi *item) RSI_HASHBITS); } -static inline int rsi_match(struct rsi *item, struct rsi *tmp) +static inline int __rsi_match(struct rsi *item, struct rsi *tmp) { return (rawobj_equal(&item->in_handle, &tmp->in_handle) && rawobj_equal(&item->in_token, &tmp->in_token)); } -static void rsi_request(struct cache_detail *cd, - struct cache_head *h, - char **bpp, int *blen) +static void rsi_free(struct rsi *rsi) { - struct rsi *rsi = container_of(h, struct rsi, h); - __u64 index = 0; - - /* if in_handle is null, provide kernel suggestion */ - if (rsi->in_handle.len == 0) - index = gss_get_next_ctx_index(); + rawobj_free(&rsi->in_handle); + rawobj_free(&rsi->in_token); + rawobj_free(&rsi->out_handle); + rawobj_free(&rsi->out_token); +} - qword_addhex(bpp, blen, (char *) &rsi->lustre_svc, - sizeof(rsi->lustre_svc)); - qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid)); - qword_addhex(bpp, blen, (char *) &index, sizeof(index)); - qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len); - qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len); - (*bpp)[-1] = '\n'; +/* See handle_channel_req() userspace for where the upcall data is read */ +static void rsi_request(struct cache_detail *cd, + struct cache_head *h, + char **bpp, int *blen) +{ + struct rsi *rsi = container_of(h, struct rsi, h); + __u64 index = 0; + + /* if in_handle is null, provide kernel suggestion */ + if (rsi->in_handle.len == 0) + index = gss_get_next_ctx_index(); + + qword_addhex(bpp, blen, (char *) &rsi->lustre_svc, + sizeof(rsi->lustre_svc)); + qword_addhex(bpp, blen, (char *) &rsi->nid, sizeof(rsi->nid)); + qword_addhex(bpp, blen, (char *) &index, sizeof(index)); + qword_addhex(bpp, blen, (char *) rsi->nm_name, + strlen(rsi->nm_name) + 1); + qword_addhex(bpp, blen, rsi->in_handle.data, rsi->in_handle.len); + qword_addhex(bpp, blen, rsi->in_token.data, rsi->in_token.len); + (*bpp)[-1] = '\n'; } -static inline void rsi_init(struct rsi *new, struct rsi *item) +#ifdef HAVE_SUNRPC_UPCALL_HAS_3ARGS +static int rsi_upcall(struct cache_detail *cd, struct cache_head *h) { - new->out_handle = RAWOBJ_EMPTY; - new->out_token = RAWOBJ_EMPTY; + return sunrpc_cache_pipe_upcall(cd, h, rsi_request); +} +#else - new->in_handle = item->in_handle; - item->in_handle = RAWOBJ_EMPTY; - new->in_token = item->in_token; - item->in_token = RAWOBJ_EMPTY; +static int rsi_upcall(struct cache_detail *cd, struct cache_head *h) +{ + return sunrpc_cache_pipe_upcall(cd, h); +} +#endif - new->lustre_svc = item->lustre_svc; - new->nid = item->nid; - init_waitqueue_head(&new->waitq); +static inline void __rsi_init(struct rsi *new, struct rsi *item) +{ + new->out_handle = RAWOBJ_EMPTY; + new->out_token = RAWOBJ_EMPTY; + + new->in_handle = item->in_handle; + item->in_handle = RAWOBJ_EMPTY; + new->in_token = item->in_token; + item->in_token = RAWOBJ_EMPTY; + + new->lustre_svc = item->lustre_svc; + new->nid = item->nid; + memcpy(new->nm_name, item->nm_name, sizeof(item->nm_name)); + init_waitqueue_head(&new->waitq); } -static inline void rsi_update(struct rsi *new, struct rsi *item) +static inline void __rsi_update(struct rsi *new, struct rsi *item) { LASSERT(new->out_handle.len == 0); LASSERT(new->out_token.len == 0); @@ -217,6 +248,54 @@ static inline void rsi_update(struct rsi *new, struct rsi *item) new->minor_status = item->minor_status; } +static void rsi_put(struct kref *ref) +{ + struct rsi *rsi = container_of(ref, struct rsi, h.ref); + +#ifdef HAVE_CACHE_HEAD_HLIST + LASSERT(rsi->h.cache_list.next == NULL); +#else + LASSERT(rsi->h.next == NULL); +#endif + rsi_free(rsi); + OBD_FREE_PTR(rsi); +} + +static int rsi_match(struct cache_head *a, struct cache_head *b) +{ + struct rsi *item = container_of(a, struct rsi, h); + struct rsi *tmp = container_of(b, struct rsi, h); + + return __rsi_match(item, tmp); +} + +static void rsi_init(struct cache_head *cnew, struct cache_head *citem) +{ + struct rsi *new = container_of(cnew, struct rsi, h); + struct rsi *item = container_of(citem, struct rsi, h); + + __rsi_init(new, item); +} + +static void update_rsi(struct cache_head *cnew, struct cache_head *citem) +{ + struct rsi *new = container_of(cnew, struct rsi, h); + struct rsi *item = container_of(citem, struct rsi, h); + + __rsi_update(new, item); +} + +static struct cache_head *rsi_alloc(void) +{ + struct rsi *rsi; + + OBD_ALLOC_PTR(rsi); + if (rsi) + return &rsi->h; + else + return NULL; +} + static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen) { char *buf = mesg; @@ -248,6 +327,11 @@ static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen) goto out; } + rsip = rsi_lookup(&rsii); + if (!rsip) + goto out; + + rsii.h.flags = 0; /* expiry */ expiry = get_expiry(&mesg); if (expiry == 0) @@ -289,33 +373,64 @@ static int rsi_parse(struct cache_detail *cd, char *mesg, int mlen) } rsii.h.expiry_time = expiry; - rsip = rsi_lookup(&rsii, 1); + rsip = rsi_update(&rsii, rsip); status = 0; out: - rsi_free(&rsii); - if (rsip) { - wake_up_all(&rsip->waitq); - rsi_put(&rsip->h, &rsi_cache); - } - - if (status) - CERROR("rsi parse error %d\n", status); - RETURN(status); + rsi_free(&rsii); + if (rsip) { + wake_up_all(&rsip->waitq); + cache_put(&rsip->h, &rsi_cache); + } else { + status = -ENOMEM; + } + + if (status) + CERROR("rsi parse error %d\n", status); + RETURN(status); } static struct cache_detail rsi_cache = { - .hash_size = RSI_HASHMAX, - .hash_table = rsi_table, - .name = "auth.sptlrpc.init", - .cache_put = rsi_put, - .cache_request = rsi_request, - .cache_parse = rsi_parse, + .hash_size = RSI_HASHMAX, + .hash_table = rsi_table, + .name = "auth.sptlrpc.init", + .cache_put = rsi_put, +#ifndef HAVE_SUNRPC_UPCALL_HAS_3ARGS + .cache_request = rsi_request, +#endif + .cache_upcall = rsi_upcall, + .cache_parse = rsi_parse, + .match = rsi_match, + .init = rsi_init, + .update = update_rsi, + .alloc = rsi_alloc, }; -static DefineSimpleCacheLookup(rsi, 0) +static struct rsi *rsi_lookup(struct rsi *item) +{ + struct cache_head *ch; + int hash = rsi_hash(item); + + ch = sunrpc_cache_lookup(&rsi_cache, &item->h, hash); + if (ch) + return container_of(ch, struct rsi, h); + else + return NULL; +} + +static struct rsi *rsi_update(struct rsi *new, struct rsi *old) +{ + struct cache_head *ch; + int hash = rsi_hash(new); + + ch = sunrpc_cache_update(&rsi_cache, &new->h, &old->h, hash); + if (ch) + return container_of(ch, struct rsi, h); + else + return NULL; +} /**************************************** - * rsc cache * + * rpc sec context (rsc) cache * ****************************************/ #define RSC_HASHBITS (10) @@ -329,9 +444,14 @@ struct rsc { struct gss_svc_ctx ctx; }; +#ifdef HAVE_CACHE_HEAD_HLIST +static struct hlist_head rsc_table[RSC_HASHMAX]; +#else static struct cache_head *rsc_table[RSC_HASHMAX]; +#endif static struct cache_detail rsc_cache; -static struct rsc *rsc_lookup(struct rsc *item, int set); +static struct rsc *rsc_update(struct rsc *new, struct rsc *old); +static struct rsc *rsc_lookup(struct rsc *item); static void rsc_free(struct rsc *rsci) { @@ -340,31 +460,18 @@ static void rsc_free(struct rsc *rsci) lgss_delete_sec_context(&rsci->ctx.gsc_mechctx); } -static void rsc_put(struct cache_head *item, struct cache_detail *cd) -{ - struct rsc *rsci = container_of(item, struct rsc, h); - - LASSERT(atomic_read(&item->refcnt) > 0); - - if (cache_put(item, cd)) { - LASSERT(item->next == NULL); - rsc_free(rsci); - kfree(rsci); /* created by cache mgmt using kmalloc */ - } -} - static inline int rsc_hash(struct rsc *rsci) { return hash_mem((char *)rsci->handle.data, rsci->handle.len, RSC_HASHBITS); } -static inline int rsc_match(struct rsc *new, struct rsc *tmp) +static inline int __rsc_match(struct rsc *new, struct rsc *tmp) { return rawobj_equal(&new->handle, &tmp->handle); } -static inline void rsc_init(struct rsc *new, struct rsc *tmp) +static inline void __rsc_init(struct rsc *new, struct rsc *tmp) { new->handle = tmp->handle; tmp->handle = RAWOBJ_EMPTY; @@ -374,23 +481,72 @@ static inline void rsc_init(struct rsc *new, struct rsc *tmp) new->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY; } -static inline void rsc_update(struct rsc *new, struct rsc *tmp) +static inline void __rsc_update(struct rsc *new, struct rsc *tmp) { new->ctx = tmp->ctx; tmp->ctx.gsc_rvs_hdl = RAWOBJ_EMPTY; tmp->ctx.gsc_mechctx = NULL; memset(&new->ctx.gsc_seqdata, 0, sizeof(new->ctx.gsc_seqdata)); - spin_lock_init(&new->ctx.gsc_seqdata.ssd_lock); + spin_lock_init(&new->ctx.gsc_seqdata.ssd_lock); +} + +static void rsc_put(struct kref *ref) +{ + struct rsc *rsci = container_of(ref, struct rsc, h.ref); + +#ifdef HAVE_CACHE_HEAD_HLIST + LASSERT(rsci->h.cache_list.next == NULL); +#else + LASSERT(rsci->h.next == NULL); +#endif + rsc_free(rsci); + OBD_FREE_PTR(rsci); +} + +static int rsc_match(struct cache_head *a, struct cache_head *b) +{ + struct rsc *new = container_of(a, struct rsc, h); + struct rsc *tmp = container_of(b, struct rsc, h); + + return __rsc_match(new, tmp); +} + +static void rsc_init(struct cache_head *cnew, struct cache_head *ctmp) +{ + struct rsc *new = container_of(cnew, struct rsc, h); + struct rsc *tmp = container_of(ctmp, struct rsc, h); + + __rsc_init(new, tmp); +} + +static void update_rsc(struct cache_head *cnew, struct cache_head *ctmp) +{ + struct rsc *new = container_of(cnew, struct rsc, h); + struct rsc *tmp = container_of(ctmp, struct rsc, h); + + __rsc_update(new, tmp); +} + +static struct cache_head * rsc_alloc(void) +{ + struct rsc *rsc; + + OBD_ALLOC_PTR(rsc); + if (rsc) + return &rsc->h; + else + return NULL; } static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen) { - char *buf = mesg; - int len, rv, tmp_int; - struct rsc rsci, *rscp = NULL; - time_t expiry; - int status = -EINVAL; + char *buf = mesg; + int len, rv, tmp_int; + struct rsc rsci, *rscp = NULL; + time_t expiry; + int status = -EINVAL; + struct gss_api_mech *gm = NULL; memset(&rsci, 0, sizeof(rsci)); @@ -416,21 +572,29 @@ static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen) } rsci.ctx.gsc_remote = (tmp_int != 0); - /* root user flag */ + /* root user flag */ + rv = get_int(&mesg, &tmp_int); + if (rv) { + CERROR("fail to get root user flag\n"); + goto out; + } + rsci.ctx.gsc_usr_root = (tmp_int != 0); + + /* mds user flag */ rv = get_int(&mesg, &tmp_int); if (rv) { - CERROR("fail to get oss user flag\n"); + CERROR("fail to get mds user flag\n"); goto out; } - rsci.ctx.gsc_usr_root = (tmp_int != 0); + rsci.ctx.gsc_usr_mds = (tmp_int != 0); - /* mds user flag */ + /* oss user flag */ rv = get_int(&mesg, &tmp_int); if (rv) { - CERROR("fail to get mds user flag\n"); + CERROR("fail to get oss user flag\n"); goto out; } - rsci.ctx.gsc_usr_mds = (tmp_int != 0); + rsci.ctx.gsc_usr_oss = (tmp_int != 0); /* mapped uid */ rv = get_int(&mesg, (int *) &rsci.ctx.gsc_mapped_uid); @@ -439,71 +603,119 @@ static int rsc_parse(struct cache_detail *cd, char *mesg, int mlen) goto out; } + rscp = rsc_lookup(&rsci); + if (!rscp) + goto out; + /* uid, or NEGATIVE */ rv = get_int(&mesg, (int *) &rsci.ctx.gsc_uid); if (rv == -EINVAL) goto out; if (rv == -ENOENT) { CERROR("NOENT? set rsc entry negative\n"); - set_bit(CACHE_NEGATIVE, &rsci.h.flags); + set_bit(CACHE_NEGATIVE, &rsci.h.flags); } else { - struct gss_api_mech *gm; - rawobj_t tmp_buf; - unsigned long ctx_expiry; - - /* gid */ - if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid)) - goto out; - - /* mech name */ - len = qword_get(&mesg, buf, mlen); - if (len < 0) - goto out; - gm = lgss_name_to_mech(buf); - status = -EOPNOTSUPP; - if (!gm) - goto out; - - status = -EINVAL; - /* mech-specific data: */ - len = qword_get(&mesg, buf, mlen); - if (len < 0) { - lgss_mech_put(gm); - goto out; - } - tmp_buf.len = len; - tmp_buf.data = (unsigned char *)buf; - if (lgss_import_sec_context(&tmp_buf, gm, - &rsci.ctx.gsc_mechctx)) { - lgss_mech_put(gm); - goto out; - } - - /* currently the expiry time passed down from user-space - * is invalid, here we retrive it from mech. */ - if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) { - CERROR("unable to get expire time, drop it\n"); - lgss_mech_put(gm); - goto out; - } - expiry = (time_t) ctx_expiry; - - lgss_mech_put(gm); + rawobj_t tmp_buf; + time64_t ctx_expiry; + + /* gid */ + if (get_int(&mesg, (int *) &rsci.ctx.gsc_gid)) + goto out; + + /* mech name */ + len = qword_get(&mesg, buf, mlen); + if (len < 0) + goto out; + gm = lgss_name_to_mech(buf); + status = -EOPNOTSUPP; + if (!gm) + goto out; + + status = -EINVAL; + /* mech-specific data: */ + len = qword_get(&mesg, buf, mlen); + if (len < 0) + goto out; + + tmp_buf.len = len; + tmp_buf.data = (unsigned char *)buf; + if (lgss_import_sec_context(&tmp_buf, gm, + &rsci.ctx.gsc_mechctx)) + goto out; + + /* set to seconds since machine booted */ + expiry = ktime_get_seconds(); + + /* currently the expiry time passed down from user-space + * is invalid, here we retrive it from mech. + */ + if (lgss_inquire_context(rsci.ctx.gsc_mechctx, + (unsigned long *)&ctx_expiry)) { + CERROR("unable to get expire time, drop it\n"); + goto out; + } + + /* ctx_expiry is the number of seconds since Jan 1 1970. + * We want just the number of seconds into the future. + */ + expiry += ctx_expiry - ktime_get_real_seconds(); } rsci.h.expiry_time = expiry; - rscp = rsc_lookup(&rsci, 1); + rscp = rsc_update(&rsci, rscp); status = 0; out: + if (gm) + lgss_mech_put(gm); rsc_free(&rsci); if (rscp) - rsc_put(&rscp->h, &rsc_cache); + cache_put(&rscp->h, &rsc_cache); + else + status = -ENOMEM; if (status) CERROR("parse rsc error %d\n", status); return status; } +static struct cache_detail rsc_cache = { + .hash_size = RSC_HASHMAX, + .hash_table = rsc_table, + .name = "auth.sptlrpc.context", + .cache_put = rsc_put, + .cache_parse = rsc_parse, + .match = rsc_match, + .init = rsc_init, + .update = update_rsc, + .alloc = rsc_alloc, +}; + +static struct rsc *rsc_lookup(struct rsc *item) +{ + struct cache_head *ch; + int hash = rsc_hash(item); + + ch = sunrpc_cache_lookup(&rsc_cache, &item->h, hash); + if (ch) + return container_of(ch, struct rsc, h); + else + return NULL; +} + +static struct rsc *rsc_update(struct rsc *new, struct rsc *old) +{ + struct cache_head *ch; + int hash = rsc_hash(new); + + ch = sunrpc_cache_update(&rsc_cache, &new->h, &old->h, hash); + if (ch) + return container_of(ch, struct rsc, h); + else + return NULL; +} + +#define COMPAT_RSC_PUT(item, cd) cache_put((item), (cd)) + /**************************************** * rsc cache flush * ****************************************/ @@ -512,31 +724,48 @@ typedef int rsc_entry_match(struct rsc *rscp, long data); static void rsc_flush(rsc_entry_match *match, long data) { - struct cache_head **ch; +#ifdef HAVE_CACHE_HEAD_HLIST + struct cache_head *ch = NULL; + struct hlist_head *head; +#else + struct cache_head **ch; +#endif struct rsc *rscp; int n; ENTRY; - write_lock(&rsc_cache.hash_lock); + write_lock(&rsc_cache.hash_lock); for (n = 0; n < RSC_HASHMAX; n++) { - for (ch = &rsc_cache.hash_table[n]; *ch;) { - rscp = container_of(*ch, struct rsc, h); +#ifdef HAVE_CACHE_HEAD_HLIST + head = &rsc_cache.hash_table[n]; + hlist_for_each_entry(ch, head, cache_list) { + rscp = container_of(ch, struct rsc, h); +#else + for (ch = &rsc_cache.hash_table[n]; *ch;) { + rscp = container_of(*ch, struct rsc, h); +#endif if (!match(rscp, data)) { - ch = &((*ch)->next); +#ifndef HAVE_CACHE_HEAD_HLIST + ch = &((*ch)->next); +#endif continue; } /* it seems simply set NEGATIVE doesn't work */ - *ch = (*ch)->next; - rscp->h.next = NULL; +#ifdef HAVE_CACHE_HEAD_HLIST + hlist_del_init(&ch->cache_list); +#else + *ch = (*ch)->next; + rscp->h.next = NULL; +#endif cache_get(&rscp->h); - set_bit(CACHE_NEGATIVE, &rscp->h.flags); - rsc_put(&rscp->h, &rsc_cache); + set_bit(CACHE_NEGATIVE, &rscp->h.flags); + COMPAT_RSC_PUT(&rscp->h, &rsc_cache); rsc_cache.entries--; } } - write_unlock(&rsc_cache.hash_lock); + write_unlock(&rsc_cache.hash_lock); EXIT; } @@ -569,17 +798,6 @@ void gss_secsvc_flush(struct obd_device *target) { rsc_flush_target(target); } -EXPORT_SYMBOL(gss_secsvc_flush); - -static struct cache_detail rsc_cache = { - .hash_size = RSC_HASHMAX, - .hash_table = rsc_table, - .name = "auth.sptlrpc.context", - .cache_put = rsc_put, - .cache_parse = rsc_parse, -}; - -static DefineSimpleCacheLookup(rsc, 0); static struct rsc *gss_svc_searchbyctx(rawobj_t *handle) { @@ -590,7 +808,7 @@ static struct rsc *gss_svc_searchbyctx(rawobj_t *handle) if (rawobj_dup(&rsci.handle, handle)) return NULL; - found = rsc_lookup(&rsci, 0); + found = rsc_lookup(&rsci); rsc_free(&rsci); if (!found) return NULL; @@ -603,54 +821,65 @@ int gss_svc_upcall_install_rvs_ctx(struct obd_import *imp, struct gss_sec *gsec, struct gss_cli_ctx *gctx) { - struct rsc rsci, *rscp; + struct rsc rsci, *rscp = NULL; unsigned long ctx_expiry; __u32 major; + int rc; ENTRY; memset(&rsci, 0, sizeof(rsci)); if (rawobj_alloc(&rsci.handle, (char *) &gsec->gs_rvs_hdl, - sizeof(gsec->gs_rvs_hdl))) { - CERROR("unable alloc handle\n"); - RETURN(-ENOMEM); - } + sizeof(gsec->gs_rvs_hdl))) + GOTO(out, rc = -ENOMEM); + + rscp = rsc_lookup(&rsci); + if (rscp == NULL) + GOTO(out, rc = -ENOMEM); major = lgss_copy_reverse_context(gctx->gc_mechctx, &rsci.ctx.gsc_mechctx); - if (major != GSS_S_COMPLETE) { - CERROR("unable to copy reverse context\n"); - rsc_free(&rsci); - RETURN(-ENOMEM); - } + if (major != GSS_S_COMPLETE) + GOTO(out, rc = -ENOMEM); if (lgss_inquire_context(rsci.ctx.gsc_mechctx, &ctx_expiry)) { CERROR("unable to get expire time, drop it\n"); - rsc_free(&rsci); - RETURN(-EINVAL); + GOTO(out, rc = -EINVAL); } - rsci.h.expiry_time = (time_t) ctx_expiry; - rsci.target = imp->imp_obd; - rscp = rsc_lookup(&rsci, 1); + switch (imp->imp_obd->u.cli.cl_sp_to) { + case LUSTRE_SP_MDT: + rsci.ctx.gsc_usr_mds = 1; + break; + case LUSTRE_SP_OST: + rsci.ctx.gsc_usr_oss = 1; + break; + case LUSTRE_SP_CLI: + rsci.ctx.gsc_usr_root = 1; + default: + break; + } + + rscp = rsc_update(&rsci, rscp); + if (rscp == NULL) + GOTO(out, rc = -ENOMEM); + + rscp->target = imp->imp_obd; + rawobj_dup(&gctx->gc_svc_handle, &rscp->handle); + + CWARN("create reverse svc ctx %p to %s: idx %#llx\n", + &rscp->ctx, obd2cli_tgt(imp->imp_obd), gsec->gs_rvs_hdl); + rc = 0; +out: + if (rscp) + cache_put(&rscp->h, &rsc_cache); rsc_free(&rsci); - if (rscp) { - /* FIXME */ - rscp->ctx.gsc_usr_root = 1; - rscp->ctx.gsc_usr_mds= 1; - rscp->ctx.gsc_reverse = 1; - rawobj_dup(&gctx->gc_svc_handle, &rscp->handle); - - CWARN("create reverse svc ctx %p to %s: idx "LPX64"\n", - &rscp->ctx, obd2cli_tgt(imp->imp_obd), - gsec->gs_rvs_hdl); - - rsc_put(&rscp->h, &rsc_cache); - } - - RETURN(0); + if (rc) + CERROR("create reverse svc ctx: idx %#llx, rc %d\n", + gsec->gs_rvs_hdl, rc); + RETURN(rc); } int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle) @@ -664,7 +893,7 @@ int gss_svc_upcall_expire_rvs_ctx(rawobj_t *handle) &rscp->ctx, rscp); rscp->h.expiry_time = cfs_time_current_sec() + expire; - rsc_put(&rscp->h, &rsc_cache); + COMPAT_RSC_PUT(&rscp->h, &rsc_cache); } return 0; } @@ -686,7 +915,7 @@ int gss_svc_upcall_update_sequence(rawobj_t *handle, __u32 seq) &rscp->ctx, rscp, seq + 1); rscp->ctx.gsc_rvs_seq = seq + 1; - rsc_put(&rscp->h, &rsc_cache); + COMPAT_RSC_PUT(&rscp->h, &rsc_cache); } return 0; } @@ -698,26 +927,28 @@ static struct cache_deferred_req* cache_upcall_defer(struct cache_req *req) static struct cache_req cache_upcall_chandle = { cache_upcall_defer }; int gss_svc_upcall_handle_init(struct ptlrpc_request *req, - struct gss_svc_reqctx *grctx, - struct gss_wire_ctx *gw, - struct obd_device *target, - __u32 lustre_svc, - rawobj_t *rvs_hdl, - rawobj_t *in_token) -{ - struct ptlrpc_reply_state *rs; - struct rsc *rsci = NULL; - struct rsi *rsip = NULL, rsikey; - wait_queue_t wait; - int replen = sizeof(struct ptlrpc_body); - struct gss_rep_header *rephdr; - int first_check = 1; - int rc = SECSVC_DROP; - ENTRY; - - memset(&rsikey, 0, sizeof(rsikey)); - rsikey.lustre_svc = lustre_svc; - rsikey.nid = (__u64) req->rq_peer.nid; + struct gss_svc_reqctx *grctx, + struct gss_wire_ctx *gw, + struct obd_device *target, + __u32 lustre_svc, + rawobj_t *rvs_hdl, + rawobj_t *in_token) +{ + struct ptlrpc_reply_state *rs; + struct rsc *rsci = NULL; + struct rsi *rsip = NULL, rsikey; + wait_queue_t wait; + int replen = sizeof(struct ptlrpc_body); + struct gss_rep_header *rephdr; + int first_check = 1; + int rc = SECSVC_DROP; + ENTRY; + + memset(&rsikey, 0, sizeof(rsikey)); + rsikey.lustre_svc = lustre_svc; + rsikey.nid = (__u64) req->rq_peer.nid; + nodemap_test_nid(req->rq_peer.nid, rsikey.nm_name, + sizeof(rsikey.nm_name)); /* duplicate context handle. for INIT it always 0 */ if (rawobj_dup(&rsikey.in_handle, &gw->gw_handle)) { @@ -731,7 +962,7 @@ int gss_svc_upcall_handle_init(struct ptlrpc_request *req, GOTO(out, rc); } - rsip = rsi_lookup(&rsikey, 0); + rsip = rsi_lookup(&rsikey); rsi_free(&rsikey); if (!rsip) { CERROR("error in rsi_lookup.\n"); @@ -742,63 +973,70 @@ int gss_svc_upcall_handle_init(struct ptlrpc_request *req, GOTO(out, rc); } - cache_get(&rsip->h); /* take an extra ref */ - init_waitqueue_head(&rsip->waitq); - init_waitqueue_entry(&wait, current); - add_wait_queue(&rsip->waitq, &wait); + cache_get(&rsip->h); /* take an extra ref */ + init_waitqueue_head(&rsip->waitq); + init_waitqueue_entry(&wait, current); + add_wait_queue(&rsip->waitq, &wait); cache_check: - /* Note each time cache_check() will drop a reference if return - * non-zero. We hold an extra reference on initial rsip, but must - * take care of following calls. */ - rc = cache_check(&rsi_cache, &rsip->h, &cache_upcall_chandle); - switch (rc) { - case -EAGAIN: { - int valid; - - if (first_check) { - first_check = 0; - - read_lock(&rsi_cache.hash_lock); - valid = test_bit(CACHE_VALID, &rsip->h.flags); - if (valid == 0) - set_current_state(TASK_INTERRUPTIBLE); - read_unlock(&rsi_cache.hash_lock); - - if (valid == 0) - schedule_timeout(GSS_SVC_UPCALL_TIMEOUT * HZ); - - cache_get(&rsip->h); - goto cache_check; - } - CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT); - break; - } - case -ENOENT: - CWARN("cache_check return ENOENT, drop\n"); - break; - case 0: - /* if not the first check, we have to release the extra - * reference we just added on it. */ - if (!first_check) - cache_put(&rsip->h, &rsi_cache); - CDEBUG(D_SEC, "cache_check is good\n"); - break; - } - - remove_wait_queue(&rsip->waitq, &wait); - cache_put(&rsip->h, &rsi_cache); - - if (rc) - GOTO(out, rc = SECSVC_DROP); + /* Note each time cache_check() will drop a reference if return + * non-zero. We hold an extra reference on initial rsip, but must + * take care of following calls. */ + rc = cache_check(&rsi_cache, &rsip->h, &cache_upcall_chandle); + switch (rc) { + case -ETIMEDOUT: + case -EAGAIN: { + int valid; + + if (first_check) { + first_check = 0; + + read_lock(&rsi_cache.hash_lock); + valid = test_bit(CACHE_VALID, &rsip->h.flags); + if (valid == 0) + set_current_state(TASK_INTERRUPTIBLE); + read_unlock(&rsi_cache.hash_lock); + + if (valid == 0) { + unsigned long jiffies; + jiffies = msecs_to_jiffies(MSEC_PER_SEC * + GSS_SVC_UPCALL_TIMEOUT); + schedule_timeout(jiffies); + } + cache_get(&rsip->h); + goto cache_check; + } + CWARN("waited %ds timeout, drop\n", GSS_SVC_UPCALL_TIMEOUT); + break; + } + case -ENOENT: + CDEBUG(D_SEC, "cache_check return ENOENT, drop\n"); + break; + case 0: + /* if not the first check, we have to release the extra + * reference we just added on it. */ + if (!first_check) + cache_put(&rsip->h, &rsi_cache); + CDEBUG(D_SEC, "cache_check is good\n"); + break; + } + + remove_wait_queue(&rsip->waitq, &wait); + cache_put(&rsip->h, &rsi_cache); + + if (rc) + GOTO(out, rc = SECSVC_DROP); rc = SECSVC_DROP; rsci = gss_svc_searchbyctx(&rsip->out_handle); if (!rsci) { CERROR("authentication failed\n"); - if (!gss_pack_err_notify(req, GSS_S_FAILURE, 0)) - rc = SECSVC_COMPLETE; + /* gss mechanism returned major and minor code so we return + * those in error message */ + if (!gss_pack_err_notify(req, rsip->major_status, + rsip->minor_status)) + rc = SECSVC_COMPLETE; GOTO(out, rc); } else { @@ -822,9 +1060,9 @@ cache_check: } grctx->src_init = 1; - grctx->src_reserve_len = size_round4(rsip->out_token.len); + grctx->src_reserve_len = cfs_size_round4(rsip->out_token.len); - rc = lustre_pack_reply_v2(req, 1, &replen, NULL); + rc = lustre_pack_reply_v2(req, 1, &replen, NULL, 0); if (rc) { CERROR("failed to pack reply: %d\n", rc); GOTO(out, rc = SECSVC_DROP); @@ -856,24 +1094,23 @@ cache_check: rc = SECSVC_OK; out: - /* it looks like here we should put rsip also, but this mess up - * with NFS cache mgmt code... FIXME */ -#if 0 - if (rsip) - rsi_put(&rsip->h, &rsi_cache); -#endif - - if (rsci) { - /* if anything went wrong, we don't keep the context too */ - if (rc != SECSVC_OK) - set_bit(CACHE_NEGATIVE, &rsci->h.flags); - else - CDEBUG(D_SEC, "create rsc with idx "LPX64"\n", - gss_handle_to_u64(&rsci->handle)); - - rsc_put(&rsci->h, &rsc_cache); - } - RETURN(rc); + /* it looks like here we should put rsip also, but this mess up + * with NFS cache mgmt code... FIXME + * something like: + * if (rsip) + * rsi_put(&rsip->h, &rsi_cache); */ + + if (rsci) { + /* if anything went wrong, we don't keep the context too */ + if (rc != SECSVC_OK) + set_bit(CACHE_NEGATIVE, &rsci->h.flags); + else + CDEBUG(D_SEC, "create rsc with idx %#llx\n", + gss_handle_to_u64(&rsci->handle)); + + COMPAT_RSC_PUT(&rsci->h, &rsc_cache); + } + RETURN(rc); } struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req, @@ -883,7 +1120,7 @@ struct gss_svc_ctx *gss_svc_upcall_get_ctx(struct ptlrpc_request *req, rsc = gss_svc_searchbyctx(&gw->gw_handle); if (!rsc) { - CWARN("Invalid gss ctx idx "LPX64" from %s\n", + CWARN("Invalid gss ctx idx %#llx from %s\n", gss_handle_to_u64(&gw->gw_handle), libcfs_nid2str(req->rq_peer.nid)); return NULL; @@ -896,7 +1133,7 @@ void gss_svc_upcall_put_ctx(struct gss_svc_ctx *ctx) { struct rsc *rsc = container_of(ctx, struct rsc, ctx); - rsc_put(&rsc->h, &rsc_cache); + COMPAT_RSC_PUT(&rsc->h, &rsc_cache); } void gss_svc_upcall_destroy_ctx(struct gss_svc_ctx *ctx) @@ -904,53 +1141,59 @@ void gss_svc_upcall_destroy_ctx(struct gss_svc_ctx *ctx) struct rsc *rsc = container_of(ctx, struct rsc, ctx); /* can't be found */ - set_bit(CACHE_NEGATIVE, &rsc->h.flags); + set_bit(CACHE_NEGATIVE, &rsc->h.flags); /* to be removed at next scan */ rsc->h.expiry_time = 1; } int __init gss_init_svc_upcall(void) { - int i; - - cache_register(&rsi_cache); - cache_register(&rsc_cache); - - /* FIXME this looks stupid. we intend to give lsvcgssd a chance to open - * the init upcall channel, otherwise there's big chance that the first - * upcall issued before the channel be opened thus nfsv4 cache code will - * drop the request direclty, thus lead to unnecessary recovery time. - * here we wait at miximum 1.5 seconds. */ - for (i = 0; i < 6; i++) { - if (atomic_read(&rsi_cache.readers) > 0) - break; - set_current_state(TASK_UNINTERRUPTIBLE); - LASSERT(HZ >= 4); - schedule_timeout(HZ / 4); - } - - if (atomic_read(&rsi_cache.readers) == 0) - CWARN("Init channel is not opened by lsvcgssd, following " - "request might be dropped until lsvcgssd is active\n"); - - /* this helps reducing context index confliction. after server reboot, - * conflicting request from clients might be filtered out by initial - * sequence number checking, thus no chance to sent error notification - * back to clients. */ - get_random_bytes(&__ctx_index, sizeof(__ctx_index)); - - return 0; + int i, rc; + + spin_lock_init(&__ctx_index_lock); + /* + * this helps reducing context index confliction. after server reboot, + * conflicting request from clients might be filtered out by initial + * sequence number checking, thus no chance to sent error notification + * back to clients. + */ + cfs_get_random_bytes(&__ctx_index, sizeof(__ctx_index)); + + rc = _cache_register_net(&rsi_cache, &init_net); + if (rc != 0) + return rc; + + rc = _cache_register_net(&rsc_cache, &init_net); + if (rc != 0) { + _cache_unregister_net(&rsi_cache, &init_net); + return rc; + } + + /* FIXME this looks stupid. we intend to give lsvcgssd a chance to open + * the init upcall channel, otherwise there's big chance that the first + * upcall issued before the channel be opened thus nfsv4 cache code will + * drop the request direclty, thus lead to unnecessary recovery time. + * here we wait at miximum 1.5 seconds. */ + for (i = 0; i < 6; i++) { + if (atomic_read(&rsi_cache.readers) > 0) + break; + set_current_state(TASK_UNINTERRUPTIBLE); + LASSERT(msecs_to_jiffies(MSEC_PER_SEC) >= 4); + schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC / 4)); + } + + if (atomic_read(&rsi_cache.readers) == 0) + CWARN("Init channel is not opened by lsvcgssd, following " + "request might be dropped until lsvcgssd is active\n"); + + return 0; } -void __exit gss_exit_svc_upcall(void) +void gss_exit_svc_upcall(void) { - int rc; - - cache_purge(&rsi_cache); - if ((rc = cache_unregister(&rsi_cache))) - CERROR("unregister rsi cache: %d\n", rc); + cache_purge(&rsi_cache); + _cache_unregister_net(&rsi_cache, &init_net); - cache_purge(&rsc_cache); - if ((rc = cache_unregister(&rsc_cache))) - CERROR("unregister rsc cache: %d\n", rc); + cache_purge(&rsc_cache); + _cache_unregister_net(&rsc_cache, &init_net); }