X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fptlrpc%2Fsec.c;h=7966fe3a08690af8c1b1b69db264ecbe70036d4b;hp=1d64af4da0ed8036594558791c93430717b30614;hb=f2a404d1fec2287ef9ffda105727e8cd3f8e0b7b;hpb=30dda5e616718499f54d06a7e3723dea4c4a7e79 diff --git a/lustre/ptlrpc/sec.c b/lustre/ptlrpc/sec.c index 1d64af4..7966fe3 100644 --- a/lustre/ptlrpc/sec.c +++ b/lustre/ptlrpc/sec.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -38,9 +38,6 @@ * Author: Eric Mei */ -#ifndef EXPORT_SYMTAB -#define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_SEC #include @@ -66,7 +63,7 @@ * policy registers * ***********************************************/ -static cfs_rwlock_t policy_lock; +static rwlock_t policy_lock; static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = { NULL, }; @@ -82,13 +79,13 @@ int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy) if (number >= SPTLRPC_POLICY_MAX) return -EINVAL; - cfs_write_lock(&policy_lock); + write_lock(&policy_lock); if (unlikely(policies[number])) { - cfs_write_unlock(&policy_lock); + write_unlock(&policy_lock); return -EALREADY; } policies[number] = policy; - cfs_write_unlock(&policy_lock); + write_unlock(&policy_lock); CDEBUG(D_SEC, "%s: registered\n", policy->sp_name); return 0; @@ -101,16 +98,16 @@ int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy) LASSERT(number < SPTLRPC_POLICY_MAX); - cfs_write_lock(&policy_lock); + write_lock(&policy_lock); if (unlikely(policies[number] == NULL)) { - cfs_write_unlock(&policy_lock); + write_unlock(&policy_lock); CERROR("%s: already unregistered\n", policy->sp_name); return -EINVAL; } LASSERT(policies[number] == policy); policies[number] = NULL; - cfs_write_unlock(&policy_lock); + write_unlock(&policy_lock); CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name); return 0; @@ -120,7 +117,7 @@ EXPORT_SYMBOL(sptlrpc_unregister_policy); static struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor) { - static CFS_DECLARE_MUTEX(load_mutex); + static DEFINE_MUTEX(load_mutex); static cfs_atomic_t loaded = CFS_ATOMIC_INIT(0); struct ptlrpc_sec_policy *policy; __u16 number = SPTLRPC_FLVR_POLICY(flavor); @@ -130,29 +127,30 @@ struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor) return NULL; while (1) { - cfs_read_lock(&policy_lock); - policy = policies[number]; - if (policy && !cfs_try_module_get(policy->sp_owner)) - policy = NULL; - if (policy == NULL) - flag = cfs_atomic_read(&loaded); - cfs_read_unlock(&policy_lock); + read_lock(&policy_lock); + policy = policies[number]; + if (policy && !try_module_get(policy->sp_owner)) + policy = NULL; + if (policy == NULL) + flag = cfs_atomic_read(&loaded); + read_unlock(&policy_lock); if (policy != NULL || flag != 0 || number != SPTLRPC_POLICY_GSS) break; /* try to load gss module, once */ - cfs_mutex_down(&load_mutex); - if (cfs_atomic_read(&loaded) == 0) { - if (cfs_request_module("ptlrpc_gss") == 0) - CWARN("module ptlrpc_gss loaded on demand\n"); + mutex_lock(&load_mutex); + if (cfs_atomic_read(&loaded) == 0) { + if (request_module("ptlrpc_gss") == 0) + CDEBUG(D_SEC, + "module ptlrpc_gss loaded on demand\n"); else CERROR("Unable to load module ptlrpc_gss\n"); cfs_atomic_set(&loaded, 1); } - cfs_mutex_up(&load_mutex); + mutex_unlock(&load_mutex); } return policy; @@ -160,10 +158,12 @@ struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor) __u32 sptlrpc_name2flavor_base(const char *name) { - if (!strcmp(name, "null")) - return SPTLRPC_FLVR_NULL; - if (!strcmp(name, "plain")) - return SPTLRPC_FLVR_PLAIN; + if (!strcmp(name, "null")) + return SPTLRPC_FLVR_NULL; + if (!strcmp(name, "plain")) + return SPTLRPC_FLVR_PLAIN; + if (!strcmp(name, "gssnull")) + return SPTLRPC_FLVR_GSSNULL; if (!strcmp(name, "krb5n")) return SPTLRPC_FLVR_KRB5N; if (!strcmp(name, "krb5a")) @@ -172,8 +172,12 @@ __u32 sptlrpc_name2flavor_base(const char *name) return SPTLRPC_FLVR_KRB5I; if (!strcmp(name, "krb5p")) return SPTLRPC_FLVR_KRB5P; + if (!strcmp(name, "ski")) + return SPTLRPC_FLVR_SKI; + if (!strcmp(name, "skpi")) + return SPTLRPC_FLVR_SKPI; - return SPTLRPC_FLVR_INVALID; + return SPTLRPC_FLVR_INVALID; } EXPORT_SYMBOL(sptlrpc_name2flavor_base); @@ -185,6 +189,8 @@ const char *sptlrpc_flavor2name_base(__u32 flvr) return "null"; else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN)) return "plain"; + else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_GSSNULL)) + return "gssnull"; else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N)) return "krb5n"; else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A)) @@ -193,9 +199,13 @@ const char *sptlrpc_flavor2name_base(__u32 flvr) return "krb5i"; else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P)) return "krb5p"; + else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKI)) + return "ski"; + else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_SKPI)) + return "skpi"; - CERROR("invalid wire flavor 0x%x\n", flvr); - return "invalid"; + CERROR("invalid wire flavor 0x%x\n", flvr); + return "invalid"; } EXPORT_SYMBOL(sptlrpc_flavor2name_base); @@ -237,21 +247,20 @@ EXPORT_SYMBOL(sptlrpc_flavor2name); char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize) { - buf[0] = '\0'; + buf[0] = '\0'; - if (flags & PTLRPC_SEC_FL_REVERSE) - strncat(buf, "reverse,", bufsize); - if (flags & PTLRPC_SEC_FL_ROOTONLY) - strncat(buf, "rootonly,", bufsize); - if (flags & PTLRPC_SEC_FL_UDESC) - strncat(buf, "udesc,", bufsize); - if (flags & PTLRPC_SEC_FL_BULK) - strncat(buf, "bulk,", bufsize); - if (buf[0] == '\0') - strncat(buf, "-,", bufsize); + if (flags & PTLRPC_SEC_FL_REVERSE) + strlcat(buf, "reverse,", bufsize); + if (flags & PTLRPC_SEC_FL_ROOTONLY) + strlcat(buf, "rootonly,", bufsize); + if (flags & PTLRPC_SEC_FL_UDESC) + strlcat(buf, "udesc,", bufsize); + if (flags & PTLRPC_SEC_FL_BULK) + strlcat(buf, "bulk,", bufsize); + if (buf[0] == '\0') + strlcat(buf, "-,", bufsize); - buf[bufsize - 1] = '\0'; - return buf; + return buf; } EXPORT_SYMBOL(sptlrpc_secflags2str); @@ -276,18 +285,17 @@ struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec) create = 0; remove_dead = 0; } - } else { - vcred.vc_uid = cfs_curproc_uid(); - vcred.vc_gid = cfs_curproc_gid(); - } + } else { + vcred.vc_uid = current_uid(); + vcred.vc_gid = current_gid(); + } - return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, - create, remove_dead); + return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred, create, + remove_dead); } struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx) { - LASSERT(cfs_atomic_read(&ctx->cc_refcount) > 0); cfs_atomic_inc(&ctx->cc_refcount); return ctx; } @@ -298,7 +306,7 @@ void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync) struct ptlrpc_sec *sec = ctx->cc_sec; LASSERT(sec); - LASSERT(cfs_atomic_read(&ctx->cc_refcount)); + LASSERT_ATOMIC_POS(&ctx->cc_refcount); if (!cfs_atomic_dec_and_test(&ctx->cc_refcount)) return; @@ -307,9 +315,10 @@ void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync) } EXPORT_SYMBOL(sptlrpc_cli_ctx_put); -/* - * expire the context immediately. - * the caller must hold at least 1 ref on the ctx. +/** + * Expire the client context immediately. + * + * \pre Caller must hold at least 1 reference on the \a ctx. */ void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx) { @@ -318,17 +327,21 @@ void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx) } EXPORT_SYMBOL(sptlrpc_cli_ctx_expire); +/** + * To wake up the threads who are waiting for this client context. Called + * after some status change happened on \a ctx. + */ void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx) { - struct ptlrpc_request *req, *next; + struct ptlrpc_request *req, *next; - cfs_spin_lock(&ctx->cc_lock); - cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list, - rq_ctx_chain) { - cfs_list_del_init(&req->rq_ctx_chain); - ptlrpc_client_wake_req(req); - } - cfs_spin_unlock(&ctx->cc_lock); + spin_lock(&ctx->cc_lock); + cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list, + rq_ctx_chain) { + cfs_list_del_init(&req->rq_ctx_chain); + ptlrpc_client_wake_req(req); + } + spin_unlock(&ctx->cc_lock); } EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup); @@ -344,15 +357,15 @@ int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize) static int import_sec_check_expire(struct obd_import *imp) { - int adapt = 0; + int adapt = 0; - cfs_spin_lock(&imp->imp_lock); - if (imp->imp_sec_expire && - imp->imp_sec_expire < cfs_time_current_sec()) { - adapt = 1; - imp->imp_sec_expire = 0; - } - cfs_spin_unlock(&imp->imp_lock); + spin_lock(&imp->imp_lock); + if (imp->imp_sec_expire && + imp->imp_sec_expire < cfs_time_current_sec()) { + adapt = 1; + imp->imp_sec_expire = 0; + } + spin_unlock(&imp->imp_lock); if (!adapt) return 0; @@ -388,6 +401,13 @@ static int import_sec_validate_get(struct obd_import *imp, return 0; } +/** + * Given a \a req, find or allocate a appropriate context for it. + * \pre req->rq_cli_ctx == NULL. + * + * \retval 0 succeed, and req->rq_cli_ctx is set. + * \retval -ev error number, and req->rq_cli_ctx == NULL. + */ int sptlrpc_req_get_ctx(struct ptlrpc_request *req) { struct obd_import *imp = req->rq_import; @@ -414,9 +434,14 @@ int sptlrpc_req_get_ctx(struct ptlrpc_request *req) RETURN(0); } -/* - * if @sync == 0, this function should return quickly without sleep; - * otherwise might trigger ctx destroying rpc to server. +/** + * Drop the context for \a req. + * \pre req->rq_cli_ctx != NULL. + * \post req->rq_cli_ctx == NULL. + * + * If \a sync == 0, this function should return quickly without sleep; + * otherwise it might trigger and wait for the whole process of sending + * an context-destroying rpc to server. */ void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync) { @@ -429,9 +454,9 @@ void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync) * in the context waiting list. */ if (!cfs_list_empty(&req->rq_ctx_chain)) { - cfs_spin_lock(&req->rq_cli_ctx->cc_lock); - cfs_list_del_init(&req->rq_ctx_chain); - cfs_spin_unlock(&req->rq_cli_ctx->cc_lock); + spin_lock(&req->rq_cli_ctx->cc_lock); + cfs_list_del_init(&req->rq_ctx_chain); + spin_unlock(&req->rq_cli_ctx->cc_lock); } sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync); @@ -445,30 +470,32 @@ int sptlrpc_req_ctx_switch(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *newctx) { struct sptlrpc_flavor old_flvr; - char *reqmsg; + char *reqmsg = NULL; /* to workaround old gcc */ int reqmsg_size; - int rc; + int rc = 0; LASSERT(req->rq_reqmsg); LASSERT(req->rq_reqlen); LASSERT(req->rq_replen); - CWARN("req %p: switch ctx %p(%u->%s) -> %p(%u->%s), " - "switch sec %p(%s) -> %p(%s)\n", req, - oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec), - newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec), - oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name, - newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name); + CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), " + "switch sec %p(%s) -> %p(%s)\n", req, + oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec), + newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec), + oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name, + newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name); /* save flavor */ old_flvr = req->rq_flvr; /* save request message */ reqmsg_size = req->rq_reqlen; - OBD_ALLOC(reqmsg, reqmsg_size); - if (reqmsg == NULL) - return -ENOMEM; - memcpy(reqmsg, req->rq_reqmsg, reqmsg_size); + if (reqmsg_size != 0) { + OBD_ALLOC_LARGE(reqmsg, reqmsg_size); + if (reqmsg == NULL) + return -ENOMEM; + memcpy(reqmsg, req->rq_reqmsg, reqmsg_size); + } /* release old req/rep buf */ req->rq_cli_ctx = oldctx; @@ -481,28 +508,29 @@ int sptlrpc_req_ctx_switch(struct ptlrpc_request *req, /* alloc new request buffer * we don't need to alloc reply buffer here, leave it to the - * rest procedure of ptlrpc - */ - rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size); - if (!rc) { - LASSERT(req->rq_reqmsg); - memcpy(req->rq_reqmsg, reqmsg, reqmsg_size); - } else { - CWARN("failed to alloc reqbuf: %d\n", rc); - req->rq_flvr = old_flvr; - } + * rest procedure of ptlrpc */ + if (reqmsg_size != 0) { + rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size); + if (!rc) { + LASSERT(req->rq_reqmsg); + memcpy(req->rq_reqmsg, reqmsg, reqmsg_size); + } else { + CWARN("failed to alloc reqbuf: %d\n", rc); + req->rq_flvr = old_flvr; + } - OBD_FREE(reqmsg, reqmsg_size); + OBD_FREE_LARGE(reqmsg, reqmsg_size); + } return rc; } /** - * if current context has died, or if we resend after flavor switched, - * call this func to switch context. if no switch is needed, request - * will end up with the same context. + * If current context of \a req is dead somehow, e.g. we just switched flavor + * thus marked original contexts dead, we'll find a new context for it. if + * no switch is needed, \a req will end up with the same context. * - * request must have a context. in any case of failure, restore the - * restore the old one - a request must have a context. + * \note a request must have a context, to keep other parts of code happy. + * In any case of failure during the switching, we must restore the old one. */ int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req) { @@ -529,16 +557,17 @@ int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req) LASSERT(newctx); if (unlikely(newctx == oldctx && - cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) { + test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) { /* * still get the old dead ctx, usually means system too busy */ - CWARN("ctx (%p, fl %lx) doesn't switch, relax a little bit\n", - newctx, newctx->cc_flags); + CDEBUG(D_SEC, + "ctx (%p, fl %lx) doesn't switch, relax a little bit\n", + newctx, newctx->cc_flags); - cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE, - CFS_HZ); - } else { + schedule_timeout_and_set_state(TASK_INTERRUPTIBLE, + HZ); + } else { /* * it's possible newctx == oldctx if we're switching * subflavor with the same sec. @@ -591,33 +620,35 @@ int ctx_refresh_timeout(void *data) static void ctx_refresh_interrupt(void *data) { - struct ptlrpc_request *req = data; + struct ptlrpc_request *req = data; - cfs_spin_lock(&req->rq_lock); - req->rq_intr = 1; - cfs_spin_unlock(&req->rq_lock); + spin_lock(&req->rq_lock); + req->rq_intr = 1; + spin_unlock(&req->rq_lock); } static void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx) { - cfs_spin_lock(&ctx->cc_lock); - if (!cfs_list_empty(&req->rq_ctx_chain)) - cfs_list_del_init(&req->rq_ctx_chain); - cfs_spin_unlock(&ctx->cc_lock); + spin_lock(&ctx->cc_lock); + if (!cfs_list_empty(&req->rq_ctx_chain)) + cfs_list_del_init(&req->rq_ctx_chain); + spin_unlock(&ctx->cc_lock); } -/* - * the status of context could be subject to be changed by other threads at any - * time. we allow this race. but once we return with 0, the caller will - * suppose it's uptodated and keep using it until the owning rpc is done. +/** + * To refresh the context of \req, if it's not up-to-date. + * \param timeout + * - < 0: don't wait + * - = 0: wait until success or fatal error occur + * - > 0: timeout value (in seconds) * - * @timeout: - * < 0 - don't wait - * = 0 - wait until success or fatal error occur - * > 0 - timeout value + * The status of the context could be subject to be changed by other threads + * at any time. We allow this race, but once we return with 0, the caller will + * suppose it's uptodated and keep using it until the owning rpc is done. * - * return 0 only if the context is uptodated. + * \retval 0 only if the context is uptodated. + * \retval -ev error number. */ int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout) { @@ -634,7 +665,7 @@ int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout) /* * during the process a request's context might change type even - * (e.g. from gss ctx to plain ctx), so each loop we need to re-check + * (e.g. from gss ctx to null ctx), so each loop we need to re-check * everything */ again: @@ -654,11 +685,11 @@ again: if (cli_ctx_is_eternal(ctx)) RETURN(0); - if (unlikely(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) { + if (unlikely(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) { LASSERT(ctx->cc_ops->refresh); ctx->cc_ops->refresh(ctx); } - LASSERT(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0); + LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0); LASSERT(ctx->cc_ops->validate); if (ctx->cc_ops->validate(ctx) == 0) { @@ -666,65 +697,68 @@ again: RETURN(0); } - if (unlikely(cfs_test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) { - cfs_spin_lock(&req->rq_lock); - req->rq_err = 1; - cfs_spin_unlock(&req->rq_lock); + if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) { + spin_lock(&req->rq_lock); + req->rq_err = 1; + spin_unlock(&req->rq_lock); req_off_ctx_list(req, ctx); RETURN(-EPERM); } - /* This is subtle. For resent message we have to keep original - * context to survive following situation: - * 1. the request sent to server - * 2. recovery was kick start - * 3. recovery finished, the request marked as resent - * 4. resend the request - * 5. old reply from server received (because xid is the same) - * 6. verify reply (has to be success) - * 7. new reply from server received, lnet drop it + /* + * There's a subtle issue for resending RPCs, suppose following + * situation: + * 1. the request was sent to server. + * 2. recovery was kicked start, after finished the request was + * marked as resent. + * 3. resend the request. + * 4. old reply from server received, we accept and verify the reply. + * this has to be success, otherwise the error will be aware + * by application. + * 5. new reply from server received, dropped by LNet. * - * Note we can't simply change xid for resent request because - * server reply on it for reply reconstruction. + * Note the xid of old & new request is the same. We can't simply + * change xid for the resent request because the server replies on + * it for reply reconstruction. * * Commonly the original context should be uptodate because we - * have a expiry nice time; And server will keep their half part - * context because we at least hold a ref of old context which - * prevent the context destroy RPC be sent. So server still can - * accept the request and finish RPC. Two cases: + * have a expiry nice time; server will keep its context because + * we at least hold a ref of old context which prevent context + * destroying RPC being sent. So server still can accept the request + * and finish the RPC. But if that's not the case: * 1. If server side context has been trimmed, a NO_CONTEXT will * be returned, gss_cli_ctx_verify/unseal will switch to new * context by force. * 2. Current context never be refreshed, then we are fine: we * never really send request with old context before. */ - if (cfs_test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) && - unlikely(req->rq_reqmsg) && - lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) { - req_off_ctx_list(req, ctx); - RETURN(0); - } - - if (unlikely(cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) { - req_off_ctx_list(req, ctx); - /* - * don't switch ctx if import was deactivated - */ - if (req->rq_import->imp_deactive) { - cfs_spin_lock(&req->rq_lock); - req->rq_err = 1; - cfs_spin_unlock(&req->rq_lock); - RETURN(-EINTR); - } - - rc = sptlrpc_req_replace_dead_ctx(req); - if (rc) { - LASSERT(ctx == req->rq_cli_ctx); - CERROR("req %p: failed to replace dead ctx %p: %d\n", - req, ctx, rc); - cfs_spin_lock(&req->rq_lock); - req->rq_err = 1; - cfs_spin_unlock(&req->rq_lock); + if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) && + unlikely(req->rq_reqmsg) && + lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) { + req_off_ctx_list(req, ctx); + RETURN(0); + } + + if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) { + req_off_ctx_list(req, ctx); + /* + * don't switch ctx if import was deactivated + */ + if (req->rq_import->imp_deactive) { + spin_lock(&req->rq_lock); + req->rq_err = 1; + spin_unlock(&req->rq_lock); + RETURN(-EINTR); + } + + rc = sptlrpc_req_replace_dead_ctx(req); + if (rc) { + LASSERT(ctx == req->rq_cli_ctx); + CERROR("req %p: failed to replace dead ctx %p: %d\n", + req, ctx, rc); + spin_lock(&req->rq_lock); + req->rq_err = 1; + spin_unlock(&req->rq_lock); RETURN(rc); } @@ -732,33 +766,35 @@ again: goto again; } - /* Now we're sure this context is during upcall, add myself into + /* + * Now we're sure this context is during upcall, add myself into * waiting list */ - cfs_spin_lock(&ctx->cc_lock); - if (cfs_list_empty(&req->rq_ctx_chain)) - cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list); - cfs_spin_unlock(&ctx->cc_lock); - - if (timeout < 0) - RETURN(-EWOULDBLOCK); - - /* Clear any flags that may be present from previous sends */ - LASSERT(req->rq_receiving_reply == 0); - cfs_spin_lock(&req->rq_lock); - req->rq_err = 0; - req->rq_timedout = 0; - req->rq_resend = 0; - req->rq_restart = 0; - cfs_spin_unlock(&req->rq_lock); - - lwi = LWI_TIMEOUT_INTR(timeout * CFS_HZ, ctx_refresh_timeout, - ctx_refresh_interrupt, req); - rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi); - - /* following cases we could be here: + spin_lock(&ctx->cc_lock); + if (cfs_list_empty(&req->rq_ctx_chain)) + cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list); + spin_unlock(&ctx->cc_lock); + + if (timeout < 0) + RETURN(-EWOULDBLOCK); + + /* Clear any flags that may be present from previous sends */ + LASSERT(req->rq_receiving_reply == 0); + spin_lock(&req->rq_lock); + req->rq_err = 0; + req->rq_timedout = 0; + req->rq_resend = 0; + req->rq_restart = 0; + spin_unlock(&req->rq_lock); + + lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout, + ctx_refresh_interrupt, req); + rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi); + + /* + * following cases could lead us here: * - successfully refreshed; - * - interruptted; + * - interrupted; * - timedout, and we don't want recover from the failure; * - timedout, and waked up upon recovery finished; * - someone else mark this ctx dead by force; @@ -776,8 +812,10 @@ again: goto again; } -/* - * Note this could be called in two situations: +/** + * Initialize flavor settings for \a req, according to \a opcode. + * + * \note this could be called in two situations: * - new request from ptlrpc_pre_req(), with proper @opcode * - old request which changed ctx in the middle, with @opcode == 0 */ @@ -794,6 +832,8 @@ void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode) switch (opcode) { case OST_READ: case MDS_READPAGE: + case MGS_CONFIG_READ: + case OBD_IDX_READ: req->rq_bulk_read = 1; break; case OST_WRITE: @@ -819,9 +859,9 @@ void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode) sec = req->rq_cli_ctx->cc_sec; - cfs_spin_lock(&sec->ps_lock); - req->rq_flvr = sec->ps_flvr; - cfs_spin_unlock(&sec->ps_lock); + spin_lock(&sec->ps_lock); + req->rq_flvr = sec->ps_flvr; + spin_unlock(&sec->ps_lock); /* force SVC_NULL for context initiation rpc, SVC_INTG for context * destruction rpc */ @@ -855,24 +895,24 @@ void sptlrpc_request_out_callback(struct ptlrpc_request *req) req->rq_reqbuf_len = 0; } -/* - * check whether current user have valid context for an import or not. - * might repeatedly try in case of non-fatal errors. - * return 0 on success, < 0 on failure +/** + * Given an import \a imp, check whether current user has a valid context + * or not. We may create a new context and try to refresh it, and try + * repeatedly try in case of non-fatal errors. Return 0 means success. */ int sptlrpc_import_check_ctx(struct obd_import *imp) { - struct ptlrpc_sec *sec; - struct ptlrpc_cli_ctx *ctx; - struct ptlrpc_request *req = NULL; - int rc; - ENTRY; + struct ptlrpc_sec *sec; + struct ptlrpc_cli_ctx *ctx; + struct ptlrpc_request *req = NULL; + int rc; + ENTRY; - cfs_might_sleep(); + might_sleep(); - sec = sptlrpc_import_sec_ref(imp); - ctx = get_my_ctx(sec); - sptlrpc_sec_put(sec); + sec = sptlrpc_import_sec_ref(imp); + ctx = get_my_ctx(sec); + sptlrpc_sec_put(sec); if (!ctx) RETURN(-ENOMEM); @@ -888,27 +928,32 @@ int sptlrpc_import_check_ctx(struct obd_import *imp) RETURN(-EACCES); } - OBD_ALLOC_PTR(req); - if (!req) - RETURN(-ENOMEM); + req = ptlrpc_request_cache_alloc(GFP_NOFS); + if (!req) + RETURN(-ENOMEM); - cfs_spin_lock_init(&req->rq_lock); - cfs_atomic_set(&req->rq_refcount, 10000); - CFS_INIT_LIST_HEAD(&req->rq_ctx_chain); - cfs_waitq_init(&req->rq_reply_waitq); - cfs_waitq_init(&req->rq_set_waitq); - req->rq_import = imp; - req->rq_flvr = sec->ps_flvr; - req->rq_cli_ctx = ctx; + spin_lock_init(&req->rq_lock); + cfs_atomic_set(&req->rq_refcount, 10000); + CFS_INIT_LIST_HEAD(&req->rq_ctx_chain); + init_waitqueue_head(&req->rq_reply_waitq); + init_waitqueue_head(&req->rq_set_waitq); + req->rq_import = imp; + req->rq_flvr = sec->ps_flvr; + req->rq_cli_ctx = ctx; rc = sptlrpc_req_refresh_ctx(req, 0); LASSERT(cfs_list_empty(&req->rq_ctx_chain)); sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1); - OBD_FREE_PTR(req); + ptlrpc_request_cache_free(req); - RETURN(rc); + RETURN(rc); } +/** + * Used by ptlrpc client, to perform the pre-defined security transformation + * upon the request message of \a req. After this function called, + * req->rq_reqmsg is still accessible as clear text. + */ int sptlrpc_cli_wrap_request(struct ptlrpc_request *req) { struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; @@ -1013,9 +1058,13 @@ static int do_cli_unwrap_reply(struct ptlrpc_request *req) RETURN(rc); } -/* - * upon this be called, the reply buffer should have been un-posted, - * so nothing is going to change. +/** + * Used by ptlrpc client, to perform security transformation upon the reply + * message of \a req. After return successfully, req->rq_repmsg points to + * the reply message in clear text. + * + * \pre the reply buffer should have been un-posted from LNet, so nothing is + * going to change. */ int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req) { @@ -1043,71 +1092,76 @@ int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req) } /** - * Upon called, the receive buffer might be still posted, so the reply data - * might be changed at any time, no matter we're holding rq_lock or not. we - * expect the rq_reply_off be 0, rq_nob_received is the early reply size. + * Used by ptlrpc client, to perform security transformation upon the early + * reply message of \a req. We expect the rq_reply_off is 0, and + * rq_nob_received is the early reply size. + * + * Because the receive buffer might be still posted, the reply data might be + * changed at any time, no matter we're holding rq_lock or not. For this reason + * we allocate a separate ptlrpc_request and reply buffer for early reply + * processing. * - * we allocate separate ptlrpc_request and reply buffer for early reply - * processing, return 0 and \a req_ret is a duplicated ptlrpc_request. caller - * must call sptlrpc_cli_finish_early_reply() on the returned request to - * release it. if anything goes wrong \a req_ret will not be set. + * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request. + * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned + * \a *req_ret to release it. + * \retval -ev error number, and \a req_ret will not be set. */ int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req, - struct ptlrpc_request **req_ret) -{ - struct ptlrpc_request *early_req; - char *early_buf; - int early_bufsz, early_size; - int rc; - ENTRY; - - OBD_ALLOC_PTR(early_req); - if (early_req == NULL) - RETURN(-ENOMEM); - - early_size = req->rq_nob_received; - early_bufsz = size_roundup_power2(early_size); - OBD_ALLOC(early_buf, early_bufsz); - if (early_buf == NULL) - GOTO(err_req, rc = -ENOMEM); - - /* sanity checkings and copy data out, do it inside spinlock */ - cfs_spin_lock(&req->rq_lock); - - if (req->rq_replied) { - cfs_spin_unlock(&req->rq_lock); - GOTO(err_buf, rc = -EALREADY); - } - - LASSERT(req->rq_repbuf); - LASSERT(req->rq_repdata == NULL); - LASSERT(req->rq_repmsg == NULL); - - if (req->rq_reply_off != 0) { - CERROR("early reply with offset %u\n", req->rq_reply_off); - cfs_spin_unlock(&req->rq_lock); - GOTO(err_buf, rc = -EPROTO); - } - - if (req->rq_nob_received != early_size) { - /* even another early arrived the size should be the same */ - CERROR("data size has changed from %u to %u\n", - early_size, req->rq_nob_received); - cfs_spin_unlock(&req->rq_lock); - GOTO(err_buf, rc = -EINVAL); - } - - if (req->rq_nob_received < sizeof(struct lustre_msg)) { - CERROR("early reply length %d too small\n", - req->rq_nob_received); - cfs_spin_unlock(&req->rq_lock); - GOTO(err_buf, rc = -EALREADY); - } - - memcpy(early_buf, req->rq_repbuf, early_size); - cfs_spin_unlock(&req->rq_lock); - - cfs_spin_lock_init(&early_req->rq_lock); + struct ptlrpc_request **req_ret) +{ + struct ptlrpc_request *early_req; + char *early_buf; + int early_bufsz, early_size; + int rc; + ENTRY; + + early_req = ptlrpc_request_cache_alloc(GFP_NOFS); + if (early_req == NULL) + RETURN(-ENOMEM); + + early_size = req->rq_nob_received; + early_bufsz = size_roundup_power2(early_size); + OBD_ALLOC_LARGE(early_buf, early_bufsz); + if (early_buf == NULL) + GOTO(err_req, rc = -ENOMEM); + + /* sanity checkings and copy data out, do it inside spinlock */ + spin_lock(&req->rq_lock); + + if (req->rq_replied) { + spin_unlock(&req->rq_lock); + GOTO(err_buf, rc = -EALREADY); + } + + LASSERT(req->rq_repbuf); + LASSERT(req->rq_repdata == NULL); + LASSERT(req->rq_repmsg == NULL); + + if (req->rq_reply_off != 0) { + CERROR("early reply with offset %u\n", req->rq_reply_off); + spin_unlock(&req->rq_lock); + GOTO(err_buf, rc = -EPROTO); + } + + if (req->rq_nob_received != early_size) { + /* even another early arrived the size should be the same */ + CERROR("data size has changed from %u to %u\n", + early_size, req->rq_nob_received); + spin_unlock(&req->rq_lock); + GOTO(err_buf, rc = -EINVAL); + } + + if (req->rq_nob_received < sizeof(struct lustre_msg)) { + CERROR("early reply length %d too small\n", + req->rq_nob_received); + spin_unlock(&req->rq_lock); + GOTO(err_buf, rc = -EALREADY); + } + + memcpy(early_buf, req->rq_repbuf, early_size); + spin_unlock(&req->rq_lock); + + spin_lock_init(&early_req->rq_lock); early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx); early_req->rq_flvr = req->rq_flvr; early_req->rq_repbuf = early_buf; @@ -1131,21 +1185,26 @@ int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req, err_ctx: sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1); err_buf: - OBD_FREE(early_buf, early_bufsz); + OBD_FREE_LARGE(early_buf, early_bufsz); err_req: - OBD_FREE_PTR(early_req); - RETURN(rc); + ptlrpc_request_cache_free(early_req); + RETURN(rc); } +/** + * Used by ptlrpc client, to release a processed early reply \a early_req. + * + * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply(). + */ void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req) { - LASSERT(early_req->rq_repbuf); - LASSERT(early_req->rq_repdata); - LASSERT(early_req->rq_repmsg); + LASSERT(early_req->rq_repbuf); + LASSERT(early_req->rq_repdata); + LASSERT(early_req->rq_repmsg); - sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1); - OBD_FREE(early_req->rq_repbuf, early_req->rq_repbuf_len); - OBD_FREE_PTR(early_req); + sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1); + OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len); + ptlrpc_request_cache_free(early_req); } /************************************************** @@ -1182,8 +1241,8 @@ static void sec_cop_destroy_sec(struct ptlrpc_sec *sec) { struct ptlrpc_sec_policy *policy = sec->ps_policy; - LASSERT(cfs_atomic_read(&sec->ps_refcount) == 0); - LASSERT(cfs_atomic_read(&sec->ps_nctx) == 0); + LASSERT_ATOMIC_ZERO(&sec->ps_refcount); + LASSERT_ATOMIC_ZERO(&sec->ps_nctx); LASSERT(policy->sp_cops->destroy_sec); CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec); @@ -1200,7 +1259,7 @@ EXPORT_SYMBOL(sptlrpc_sec_destroy); static void sptlrpc_sec_kill(struct ptlrpc_sec *sec) { - LASSERT(cfs_atomic_read(&sec->ps_refcount) > 0); + LASSERT_ATOMIC_POS(&sec->ps_refcount); if (sec->ps_policy->sp_cops->kill_sec) { sec->ps_policy->sp_cops->kill_sec(sec); @@ -1211,10 +1270,8 @@ static void sptlrpc_sec_kill(struct ptlrpc_sec *sec) struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec) { - if (sec) { - LASSERT(cfs_atomic_read(&sec->ps_refcount) > 0); + if (sec) cfs_atomic_inc(&sec->ps_refcount); - } return sec; } @@ -1223,11 +1280,9 @@ EXPORT_SYMBOL(sptlrpc_sec_get); void sptlrpc_sec_put(struct ptlrpc_sec *sec) { if (sec) { - LASSERT(cfs_atomic_read(&sec->ps_refcount) > 0); + LASSERT_ATOMIC_POS(&sec->ps_refcount); if (cfs_atomic_dec_and_test(&sec->ps_refcount)) { - LASSERT(cfs_atomic_read(&sec->ps_nctx) == 0); - sptlrpc_gc_del_sec(sec); sec_cop_destroy_sec(sec); } @@ -1291,27 +1346,27 @@ struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp, struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp) { - struct ptlrpc_sec *sec; + struct ptlrpc_sec *sec; - cfs_spin_lock(&imp->imp_lock); - sec = sptlrpc_sec_get(imp->imp_sec); - cfs_spin_unlock(&imp->imp_lock); + spin_lock(&imp->imp_lock); + sec = sptlrpc_sec_get(imp->imp_sec); + spin_unlock(&imp->imp_lock); - return sec; + return sec; } EXPORT_SYMBOL(sptlrpc_import_sec_ref); static void sptlrpc_import_sec_install(struct obd_import *imp, struct ptlrpc_sec *sec) { - struct ptlrpc_sec *old_sec; + struct ptlrpc_sec *old_sec; - LASSERT(cfs_atomic_read(&sec->ps_refcount) > 0); + LASSERT_ATOMIC_POS(&sec->ps_refcount); - cfs_spin_lock(&imp->imp_lock); - old_sec = imp->imp_sec; - imp->imp_sec = sec; - cfs_spin_unlock(&imp->imp_lock); + spin_lock(&imp->imp_lock); + old_sec = imp->imp_sec; + imp->imp_sec = sec; + spin_unlock(&imp->imp_lock); if (old_sec) { sptlrpc_sec_kill(old_sec); @@ -1340,39 +1395,42 @@ static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp, char str1[32], str2[32]; if (sec->ps_flvr.sf_flags != sf->sf_flags) - CWARN("changing sec flags: %s -> %s\n", - sptlrpc_secflags2str(sec->ps_flvr.sf_flags, - str1, sizeof(str1)), - sptlrpc_secflags2str(sf->sf_flags, - str2, sizeof(str2))); + CDEBUG(D_SEC, "changing sec flags: %s -> %s\n", + sptlrpc_secflags2str(sec->ps_flvr.sf_flags, + str1, sizeof(str1)), + sptlrpc_secflags2str(sf->sf_flags, + str2, sizeof(str2))); - cfs_spin_lock(&sec->ps_lock); - flavor_copy(&sec->ps_flvr, sf); - cfs_spin_unlock(&sec->ps_lock); + spin_lock(&sec->ps_lock); + flavor_copy(&sec->ps_flvr, sf); + spin_unlock(&sec->ps_lock); } -/* - * for normal import, @svc_ctx should be NULL and @flvr is ignored; - * for reverse import, @svc_ctx and @flvr is from incoming request. +/** + * To get an appropriate ptlrpc_sec for the \a imp, according to the current + * configuration. Upon called, imp->imp_sec may or may not be NULL. + * + * - regular import: \a svc_ctx should be NULL and \a flvr is ignored; + * - reverse import: \a svc_ctx and \a flvr are obtained from incoming request. */ int sptlrpc_import_sec_adapt(struct obd_import *imp, struct ptlrpc_svc_ctx *svc_ctx, struct sptlrpc_flavor *flvr) { - struct ptlrpc_connection *conn; - struct sptlrpc_flavor sf; - struct ptlrpc_sec *sec, *newsec; - enum lustre_sec_part sp; - char str[24]; - int rc = 0; - ENTRY; + struct ptlrpc_connection *conn; + struct sptlrpc_flavor sf; + struct ptlrpc_sec *sec, *newsec; + enum lustre_sec_part sp; + char str[24]; + int rc = 0; + ENTRY; - cfs_might_sleep(); + might_sleep(); - if (imp == NULL) - RETURN(0); + if (imp == NULL) + RETURN(0); - conn = imp->imp_connection; + conn = imp->imp_connection; if (svc_ctx == NULL) { struct client_obd *cliobd = &imp->imp_obd->u.cli; @@ -1407,11 +1465,11 @@ int sptlrpc_import_sec_adapt(struct obd_import *imp, if (flavor_equal(&sf, &sec->ps_flvr)) GOTO(out, rc); - CWARN("import %s->%s: changing flavor %s -> %s\n", - imp->imp_obd->obd_name, - obd_uuid2str(&conn->c_remote_uuid), - sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)), - sptlrpc_flavor2name(&sf, str2, sizeof(str2))); + CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n", + imp->imp_obd->obd_name, + obd_uuid2str(&conn->c_remote_uuid), + sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)), + sptlrpc_flavor2name(&sf, str2, sizeof(str2))); if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) == SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) && @@ -1420,15 +1478,16 @@ int sptlrpc_import_sec_adapt(struct obd_import *imp, sptlrpc_import_sec_adapt_inplace(imp, sec, &sf); GOTO(out, rc); } - } else { - CWARN("import %s->%s netid %x: select flavor %s\n", - imp->imp_obd->obd_name, - obd_uuid2str(&conn->c_remote_uuid), - LNET_NIDNET(conn->c_self), - sptlrpc_flavor2name(&sf, str, sizeof(str))); + } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) != + SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) { + CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n", + imp->imp_obd->obd_name, + obd_uuid2str(&conn->c_remote_uuid), + LNET_NIDNET(conn->c_self), + sptlrpc_flavor2name(&sf, str, sizeof(str))); } - cfs_mutex_down(&imp->imp_sec_mutex); + mutex_lock(&imp->imp_sec_mutex); newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp); if (newsec) { @@ -1440,7 +1499,7 @@ int sptlrpc_import_sec_adapt(struct obd_import *imp, rc = -EPERM; } - cfs_mutex_up(&imp->imp_sec_mutex); + mutex_unlock(&imp->imp_sec_mutex); out: sptlrpc_sec_put(sec); RETURN(rc); @@ -1481,7 +1540,7 @@ void sptlrpc_import_flush_root_ctx(struct obd_import *imp) void sptlrpc_import_flush_my_ctx(struct obd_import *imp) { - import_flush_ctx_common(imp, cfs_curproc_uid(), 1, 1); + import_flush_ctx_common(imp, current_uid(), 1, 1); } EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx); @@ -1491,9 +1550,9 @@ void sptlrpc_import_flush_all_ctx(struct obd_import *imp) } EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx); -/* - * when complete successfully, req->rq_reqmsg should point to the - * right place. +/** + * Used by ptlrpc client to allocate request buffer of \a req. Upon return + * successfully, req->rq_reqmsg points to a buffer with size \a msgsize. */ int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize) { @@ -1502,10 +1561,10 @@ int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize) int rc; LASSERT(ctx); - LASSERT(cfs_atomic_read(&ctx->cc_refcount)); LASSERT(ctx->cc_sec); LASSERT(ctx->cc_sec->ps_policy); LASSERT(req->rq_reqmsg == NULL); + LASSERT_ATOMIC_POS(&ctx->cc_refcount); policy = ctx->cc_sec->ps_policy; rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize); @@ -1521,21 +1580,26 @@ int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize) return rc; } +/** + * Used by ptlrpc client to free request buffer of \a req. After this + * req->rq_reqmsg is set to NULL and should not be accessed anymore. + */ void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req) { struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; struct ptlrpc_sec_policy *policy; LASSERT(ctx); - LASSERT(cfs_atomic_read(&ctx->cc_refcount)); LASSERT(ctx->cc_sec); LASSERT(ctx->cc_sec->ps_policy); + LASSERT_ATOMIC_POS(&ctx->cc_refcount); if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL) return; policy = ctx->cc_sec->ps_policy; policy->sp_cops->free_reqbuf(ctx->cc_sec, req); + req->rq_reqmsg = NULL; } /* @@ -1582,13 +1646,14 @@ void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg, } EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace); -/* - * enlarge @segment of upper message req->rq_reqmsg to @newsize, all data - * will be preserved after enlargement. this must be called after rq_reqmsg has - * been intialized at least. +/** + * Used by ptlrpc client to enlarge the \a segment of request message pointed + * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be + * preserved after the enlargement. this must be called after original request + * buffer being allocated. * - * caller's attention: upon return, rq_reqmsg and rq_reqlen might have - * been changed. + * \note after this be called, rq_reqmsg and rq_reqlen might have been changed, + * so caller should refresh its local pointers if needed. */ int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req, int segment, int newsize) @@ -1611,6 +1676,11 @@ int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf); +/** + * Used by ptlrpc client to allocate reply buffer of \a req. + * + * \note After this, req->rq_repmsg is still not accessible. + */ int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize) { struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; @@ -1618,7 +1688,6 @@ int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize) ENTRY; LASSERT(ctx); - LASSERT(cfs_atomic_read(&ctx->cc_refcount)); LASSERT(ctx->cc_sec); LASSERT(ctx->cc_sec->ps_policy); @@ -1629,6 +1698,10 @@ int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize) RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize)); } +/** + * Used by ptlrpc client to free reply buffer of \a req. After this + * req->rq_repmsg is set to NULL and should not be accessed anymore. + */ void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req) { struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx; @@ -1636,9 +1709,9 @@ void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req) ENTRY; LASSERT(ctx); - LASSERT(cfs_atomic_read(&ctx->cc_refcount)); LASSERT(ctx->cc_sec); LASSERT(ctx->cc_sec->ps_policy); + LASSERT_ATOMIC_POS(&ctx->cc_refcount); if (req->rq_repbuf == NULL) return; @@ -1646,6 +1719,7 @@ void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req) policy = ctx->cc_sec->ps_policy; policy->sp_cops->free_repbuf(ctx->cc_sec, req); + req->rq_repmsg = NULL; EXIT; } @@ -1692,6 +1766,11 @@ static int flavor_allowed(struct sptlrpc_flavor *exp, #define EXP_FLVR_UPDATE_EXPIRE (OBD_TIMEOUT_DEFAULT + 10) +/** + * Given an export \a exp, check whether the flavor of incoming \a req + * is allowed by the export \a exp. Main logic is about taking care of + * changing configurations. Return 0 means success. + */ int sptlrpc_target_export_check(struct obd_export *exp, struct ptlrpc_request *req) { @@ -1709,7 +1788,7 @@ int sptlrpc_target_export_check(struct obd_export *exp, if (req->rq_ctx_fini) return 0; - cfs_spin_lock(&exp->exp_lock); + spin_lock(&exp->exp_lock); /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for * the first req with the new flavor, then treat it as current flavor, @@ -1736,17 +1815,19 @@ int sptlrpc_target_export_check(struct obd_export *exp, /* if it's gss, we only interested in root ctx init */ if (req->rq_auth_gss && - !(req->rq_ctx_init && (req->rq_auth_usr_root || - req->rq_auth_usr_mdt))) { - cfs_spin_unlock(&exp->exp_lock); - CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d)\n", - req->rq_auth_gss, req->rq_ctx_init, - req->rq_auth_usr_root, req->rq_auth_usr_mdt); - return 0; - } - - exp->exp_flvr_adapt = 0; - cfs_spin_unlock(&exp->exp_lock); + !(req->rq_ctx_init && + (req->rq_auth_usr_root || req->rq_auth_usr_mdt || + req->rq_auth_usr_ost))) { + spin_unlock(&exp->exp_lock); + CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n", + req->rq_auth_gss, req->rq_ctx_init, + req->rq_auth_usr_root, req->rq_auth_usr_mdt, + req->rq_auth_usr_ost); + return 0; + } + + exp->exp_flvr_adapt = 0; + spin_unlock(&exp->exp_lock); return sptlrpc_import_sec_adapt(exp->exp_imp_reverse, req->rq_svc_ctx, &flavor); @@ -1758,38 +1839,39 @@ int sptlrpc_target_export_check(struct obd_export *exp, /* most cases should return here, we only interested in * gss root ctx init */ if (!req->rq_auth_gss || !req->rq_ctx_init || - (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt)) { - cfs_spin_unlock(&exp->exp_lock); - return 0; - } - - /* if flavor just changed, we should not proceed, just leave - * it and current flavor will be discovered and replaced - * shortly, and let _this_ rpc pass through */ - if (exp->exp_flvr_changed) { - LASSERT(exp->exp_flvr_adapt); - cfs_spin_unlock(&exp->exp_lock); - return 0; - } - - if (exp->exp_flvr_adapt) { - exp->exp_flvr_adapt = 0; - CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n", - exp, exp->exp_flvr.sf_rpc, - exp->exp_flvr_old[0].sf_rpc, - exp->exp_flvr_old[1].sf_rpc); - flavor = exp->exp_flvr; - cfs_spin_unlock(&exp->exp_lock); - - return sptlrpc_import_sec_adapt(exp->exp_imp_reverse, - req->rq_svc_ctx, - &flavor); - } else { - CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, " - "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc, - exp->exp_flvr_old[0].sf_rpc, - exp->exp_flvr_old[1].sf_rpc); - cfs_spin_unlock(&exp->exp_lock); + (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt && + !req->rq_auth_usr_ost)) { + spin_unlock(&exp->exp_lock); + return 0; + } + + /* if flavor just changed, we should not proceed, just leave + * it and current flavor will be discovered and replaced + * shortly, and let _this_ rpc pass through */ + if (exp->exp_flvr_changed) { + LASSERT(exp->exp_flvr_adapt); + spin_unlock(&exp->exp_lock); + return 0; + } + + if (exp->exp_flvr_adapt) { + exp->exp_flvr_adapt = 0; + CDEBUG(D_SEC, "exp %p (%x|%x|%x): do delayed adapt\n", + exp, exp->exp_flvr.sf_rpc, + exp->exp_flvr_old[0].sf_rpc, + exp->exp_flvr_old[1].sf_rpc); + flavor = exp->exp_flvr; + spin_unlock(&exp->exp_lock); + + return sptlrpc_import_sec_adapt(exp->exp_imp_reverse, + req->rq_svc_ctx, + &flavor); + } else { + CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, " + "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc, + exp->exp_flvr_old[0].sf_rpc, + exp->exp_flvr_old[1].sf_rpc); + spin_unlock(&exp->exp_lock); return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse, req->rq_svc_ctx); @@ -1806,7 +1888,7 @@ int sptlrpc_target_export_check(struct obd_export *exp, exp->exp_flvr_old[1].sf_rpc, exp->exp_flvr_expire[0] - cfs_time_current_sec()); - cfs_spin_unlock(&exp->exp_lock); + spin_unlock(&exp->exp_lock); return 0; } } else { @@ -1831,7 +1913,7 @@ int sptlrpc_target_export_check(struct obd_export *exp, exp->exp_flvr_old[1].sf_rpc, exp->exp_flvr_expire[1] - cfs_time_current_sec()); - cfs_spin_unlock(&exp->exp_lock); + spin_unlock(&exp->exp_lock); return 0; } } else { @@ -1848,13 +1930,14 @@ int sptlrpc_target_export_check(struct obd_export *exp, exp->exp_flvr_old[1].sf_rpc); } - cfs_spin_unlock(&exp->exp_lock); + spin_unlock(&exp->exp_lock); - CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u) with " + CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with " "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n", exp, exp->exp_obd->obd_name, req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini, - req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_flvr.sf_rpc, + req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost, + req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[0].sf_rpc, exp->exp_flvr_expire[0] ? @@ -1876,16 +1959,16 @@ void sptlrpc_target_update_exp_flavor(struct obd_device *obd, LASSERT(obd); - cfs_spin_lock(&obd->obd_dev_lock); + spin_lock(&obd->obd_dev_lock); - cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) { - if (exp->exp_connection == NULL) - continue; + cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) { + if (exp->exp_connection == NULL) + continue; - /* note if this export had just been updated flavor - * (exp_flvr_changed == 1), this will override the - * previous one. */ - cfs_spin_lock(&exp->exp_lock); + /* note if this export had just been updated flavor + * (exp_flvr_changed == 1), this will override the + * previous one. */ + spin_lock(&exp->exp_lock); sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer, exp->exp_connection->c_peer.nid, &new_flvr); @@ -1901,57 +1984,66 @@ void sptlrpc_target_update_exp_flavor(struct obd_device *obd, exp->exp_flvr.sf_rpc, exp->exp_flvr_old[1].sf_rpc); } - cfs_spin_unlock(&exp->exp_lock); - } + spin_unlock(&exp->exp_lock); + } - cfs_spin_unlock(&obd->obd_dev_lock); + spin_unlock(&obd->obd_dev_lock); } EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor); static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc) { - if (svc_rc == SECSVC_DROP) - return SECSVC_DROP; + /* peer's claim is unreliable unless gss is being used */ + if (!req->rq_auth_gss || svc_rc == SECSVC_DROP) + return svc_rc; switch (req->rq_sp_from) { case LUSTRE_SP_CLI: + if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) { + DEBUG_REQ(D_ERROR, req, "faked source CLI"); + svc_rc = SECSVC_DROP; + } + break; case LUSTRE_SP_MDT: + if (!req->rq_auth_usr_mdt) { + DEBUG_REQ(D_ERROR, req, "faked source MDT"); + svc_rc = SECSVC_DROP; + } + break; case LUSTRE_SP_OST: - case LUSTRE_SP_MGC: + if (!req->rq_auth_usr_ost) { + DEBUG_REQ(D_ERROR, req, "faked source OST"); + svc_rc = SECSVC_DROP; + } + break; case LUSTRE_SP_MGS: - case LUSTRE_SP_ANY: + case LUSTRE_SP_MGC: + if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt && + !req->rq_auth_usr_ost) { + DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS"); + svc_rc = SECSVC_DROP; + } break; + case LUSTRE_SP_ANY: default: DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from); - return SECSVC_DROP; - } - - if (!req->rq_auth_gss) - return svc_rc; - - if (unlikely(req->rq_sp_from == LUSTRE_SP_ANY)) { - CERROR("not specific part\n"); - return SECSVC_DROP; - } - - /* from MDT, must be authenticated as MDT */ - if (unlikely(req->rq_sp_from == LUSTRE_SP_MDT && - !req->rq_auth_usr_mdt)) { - DEBUG_REQ(D_ERROR, req, "fake source MDT"); - return SECSVC_DROP; - } - - /* from OST, must be callback to MDT and CLI, the reverse sec - * was from mdt/root keytab, so it should be MDT or root FIXME */ - if (unlikely(req->rq_sp_from == LUSTRE_SP_OST && - !req->rq_auth_usr_mdt && !req->rq_auth_usr_root)) { - DEBUG_REQ(D_ERROR, req, "fake source OST"); - return SECSVC_DROP; + svc_rc = SECSVC_DROP; } return svc_rc; } +/** + * Used by ptlrpc server, to perform transformation upon request message of + * incoming \a req. This must be the first thing to do with a incoming + * request in ptlrpc layer. + * + * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in + * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set. + * \retval SECSVC_COMPLETE success, the request has been fully processed, and + * reply message has been prepared. + * \retval SECSVC_DROP failed, this request should be dropped. + */ int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req) { struct ptlrpc_sec_policy *policy; @@ -2007,8 +2099,12 @@ int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req) RETURN(rc); } -int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, - int msglen) +/** + * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed, + * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to + * a buffer of \a msglen size. + */ +int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen) { struct ptlrpc_sec_policy *policy; struct ptlrpc_reply_state *rs; @@ -2023,8 +2119,18 @@ int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, rc = policy->sp_sops->alloc_rs(req, msglen); if (unlikely(rc == -ENOMEM)) { + struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt; + if (svcpt->scp_service->srv_max_reply_size < + msglen + sizeof(struct ptlrpc_reply_state)) { + /* Just return failure if the size is too big */ + CERROR("size of message is too big (%zd), %d allowed", + msglen + sizeof(struct ptlrpc_reply_state), + svcpt->scp_service->srv_max_reply_size); + RETURN(-ENOMEM); + } + /* failed alloc, try emergency pool */ - rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_service); + rs = lustre_get_emerg_rs(svcpt); if (rs == NULL) RETURN(-ENOMEM); @@ -2042,6 +2148,12 @@ int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, RETURN(rc); } +/** + * Used by ptlrpc server, to perform transformation upon reply message. + * + * \post req->rq_reply_off is set to approriate server-controlled reply offset. + * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible. + */ int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req) { struct ptlrpc_sec_policy *policy; @@ -2060,6 +2172,9 @@ int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req) RETURN(rc); } +/** + * Used by ptlrpc server, to free reply_state. + */ void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs) { struct ptlrpc_sec_policy *policy; @@ -2084,11 +2199,8 @@ void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req) { struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx; - if (ctx == NULL) - return; - - LASSERT(cfs_atomic_read(&ctx->sc_refcount) > 0); - cfs_atomic_inc(&ctx->sc_refcount); + if (ctx != NULL) + cfs_atomic_inc(&ctx->sc_refcount); } void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req) @@ -2098,7 +2210,7 @@ void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req) if (ctx == NULL) return; - LASSERT(cfs_atomic_read(&ctx->sc_refcount) > 0); + LASSERT_ATOMIC_POS(&ctx->sc_refcount); if (cfs_atomic_dec_and_test(&ctx->sc_refcount)) { if (ctx->sc_policy->sp_sops->free_ctx) ctx->sc_policy->sp_sops->free_ctx(ctx); @@ -2113,7 +2225,7 @@ void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req) if (ctx == NULL) return; - LASSERT(cfs_atomic_read(&ctx->sc_refcount) > 0); + LASSERT_ATOMIC_POS(&ctx->sc_refcount); if (ctx->sc_policy->sp_sops->invalidate_ctx) ctx->sc_policy->sp_sops->invalidate_ctx(ctx); } @@ -2123,6 +2235,10 @@ EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate); * bulk security * ****************************************/ +/** + * Perform transformation upon bulk data pointed by \a desc. This is called + * before transforming the request message. + */ int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req, struct ptlrpc_bulk_desc *desc) { @@ -2140,7 +2256,8 @@ int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk); -/* +/** + * This is called after unwrap the reply message. * return nob of actual plain text size received, or error code. */ int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req, @@ -2165,7 +2282,8 @@ int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read); -/* +/** + * This is called after unwrap the reply message. * return 0 for success or error code. */ int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req, @@ -2200,6 +2318,10 @@ int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write); +#ifdef HAVE_SERVER_SUPPORT +/** + * Performe transformation upon outgoing bulk read. + */ int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req, struct ptlrpc_bulk_desc *desc) { @@ -2218,6 +2340,9 @@ int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk); +/** + * Performe transformation upon incoming bulk write. + */ int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req, struct ptlrpc_bulk_desc *desc) { @@ -2254,6 +2379,9 @@ int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk); +/** + * Prepare buffers for incoming bulk write. + */ int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req, struct ptlrpc_bulk_desc *desc) { @@ -2272,6 +2400,8 @@ int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req, } EXPORT_SYMBOL(sptlrpc_svc_prep_bulk); +#endif /* HAVE_SERVER_SUPPORT */ + /**************************************** * user descriptor helpers * ****************************************/ @@ -2294,27 +2424,27 @@ EXPORT_SYMBOL(sptlrpc_current_user_desc_size); int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset) { - struct ptlrpc_user_desc *pud; + struct ptlrpc_user_desc *pud; - pud = lustre_msg_buf(msg, offset, 0); + pud = lustre_msg_buf(msg, offset, 0); - pud->pud_uid = cfs_curproc_uid(); - pud->pud_gid = cfs_curproc_gid(); - pud->pud_fsuid = cfs_curproc_fsuid(); - pud->pud_fsgid = cfs_curproc_fsgid(); - pud->pud_cap = cfs_curproc_cap_pack(); - pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4; + pud->pud_uid = current_uid(); + pud->pud_gid = current_gid(); + pud->pud_fsuid = current_fsuid(); + pud->pud_fsgid = current_fsgid(); + pud->pud_cap = cfs_curproc_cap_pack(); + pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4; #ifdef __KERNEL__ - task_lock(current); - if (pud->pud_ngroups > current_ngroups) - pud->pud_ngroups = current_ngroups; - memcpy(pud->pud_groups, current_cred()->group_info->blocks[0], - pud->pud_ngroups * sizeof(__u32)); - task_unlock(current); + task_lock(current); + if (pud->pud_ngroups > current_ngroups) + pud->pud_ngroups = current_ngroups; + memcpy(pud->pud_groups, current_cred()->group_info->blocks[0], + pud->pud_ngroups * sizeof(__u32)); + task_unlock(current); #endif - return 0; + return 0; } EXPORT_SYMBOL(sptlrpc_pack_user_desc); @@ -2394,11 +2524,11 @@ EXPORT_SYMBOL(sptlrpc_flavor_has_bulk); * initialize/finalize * ****************************************/ -int __init sptlrpc_init(void) +int sptlrpc_init(void) { int rc; - cfs_rwlock_init(&policy_lock); + rwlock_init(&policy_lock); rc = sptlrpc_gc_init(); if (rc) @@ -2440,7 +2570,7 @@ out: return rc; } -void __exit sptlrpc_fini(void) +void sptlrpc_fini(void) { sptlrpc_lproc_fini(); sptlrpc_plain_fini();