X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_lib.c;h=eb08efc53fa02e23f4386f9a83b8bb5f9503ff69;hp=051421bbd0303bbd4134808e95b1e28e9a7e3d58;hb=abedce670d4f396443164bb582c7b58ec78ae7b7;hpb=f4547f0569774eb794fb143362e201f658415f4c diff --git a/lustre/mdt/mdt_lib.c b/lustre/mdt/mdt_lib.c index 051421bb..eb08efc 100644 --- a/lustre/mdt/mdt_lib.c +++ b/lustre/mdt/mdt_lib.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -48,9 +48,13 @@ #define DEBUG_SUBSYSTEM S_MDS +#include +#ifdef HAVE_UIDGID_HEADER +# include +#endif #include "mdt_internal.h" -#include - +#include +#include typedef enum ucred_init_type { NONE_INIT = 0, @@ -67,7 +71,7 @@ void mdt_exit_ucred(struct mdt_thread_info *info) if (uc->uc_valid != UCRED_INIT) { uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1; if (uc->uc_ginfo) { - cfs_put_group_info(uc->uc_ginfo); + put_group_info(uc->uc_ginfo); uc->uc_ginfo = NULL; } if (uc->uc_identity) { @@ -80,7 +84,7 @@ void mdt_exit_ucred(struct mdt_thread_info *info) } static int match_nosquash_list(struct rw_semaphore *sem, - cfs_list_t *nidlist, + struct list_head *nidlist, lnet_nid_t peernid) { int rc; @@ -95,28 +99,28 @@ static int match_nosquash_list(struct rw_semaphore *sem, static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid) { struct lu_ucred *ucred = mdt_ucred(info); + struct root_squash_info *squash = &info->mti_mdt->mdt_squash; ENTRY; LASSERT(ucred != NULL); - if (!info->mti_mdt->mdt_squash_uid || ucred->uc_fsuid) + if (!squash->rsi_uid || ucred->uc_fsuid) RETURN(0); - if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem, - &info->mti_mdt->mdt_nosquash_nids, - peernid)) { - CDEBUG(D_OTHER, "%s is in nosquash_nids list\n", - libcfs_nid2str(peernid)); - RETURN(0); - } + if (match_nosquash_list(&squash->rsi_sem, + &squash->rsi_nosquash_nids, + peernid)) { + CDEBUG(D_OTHER, "%s is in nosquash_nids list\n", + libcfs_nid2str(peernid)); + RETURN(0); + } CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n", libcfs_nid2str(peernid), ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap, - info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid, - 0); + squash->rsi_uid, squash->rsi_gid, 0); - ucred->uc_fsuid = info->mti_mdt->mdt_squash_uid; - ucred->uc_fsgid = info->mti_mdt->mdt_squash_gid; + ucred->uc_fsuid = squash->rsi_uid; + ucred->uc_fsgid = squash->rsi_gid; ucred->uc_cap = 0; ucred->uc_suppgids[0] = -1; ucred->uc_suppgids[1] = -1; @@ -124,8 +128,20 @@ static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid) RETURN(0); } +static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc) +{ + struct ptlrpc_request *req = mdt_info_req(info); + const char *jobid = mdt_req_get_jobid(req); + + /* set jobid if specified. */ + if (jobid) + strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid)); + else + uc->uc_jobid[0] = '\0'; +} + static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type, - void *buf) + void *buf, bool drop_fs_cap) { struct ptlrpc_request *req = mdt_info_req(info); struct mdt_device *mdt = info->mti_mdt; @@ -133,7 +149,6 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type, struct lu_ucred *ucred = mdt_ucred(info); lnet_nid_t peernid = req->rq_peer.nid; __u32 perm = 0; - __u32 remote = exp_connect_rmtclient(info->mti_exp); int setuid; int setgid; int rc = 0; @@ -155,75 +170,48 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type, if (type == BODY_INIT) { struct mdt_body *body = (struct mdt_body *)buf; - ucred->uc_suppgids[0] = body->suppgid; + ucred->uc_suppgids[0] = body->mbo_suppgid; ucred->uc_suppgids[1] = -1; } - /* sanity check: we expect the uid which client claimed is true */ - if (remote) { - if (req->rq_auth_mapped_uid == INVALID_UID) { - CDEBUG(D_SEC, "remote user not mapped, deny access!\n"); - RETURN(-EACCES); - } - - if (ptlrpc_user_desc_do_idmap(req, pud)) - RETURN(-EACCES); - - if (req->rq_auth_mapped_uid != pud->pud_uid) { - CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u " - "while client claims %u:%u/%u:%u\n", - libcfs_nid2str(peernid), req->rq_auth_uid, - req->rq_auth_mapped_uid, - pud->pud_uid, pud->pud_gid, - pud->pud_fsuid, pud->pud_fsgid); - RETURN(-EACCES); - } - } else { - if (req->rq_auth_uid != pud->pud_uid) { - CDEBUG(D_SEC, "local client %s: auth uid %u " - "while client claims %u:%u/%u:%u\n", - libcfs_nid2str(peernid), req->rq_auth_uid, - pud->pud_uid, pud->pud_gid, - pud->pud_fsuid, pud->pud_fsgid); - RETURN(-EACCES); - } - } + if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) && + req->rq_auth_uid != pud->pud_uid) { + CDEBUG(D_SEC, "local client %s: auth uid %u " + "while client claims %u:%u/%u:%u\n", + libcfs_nid2str(peernid), req->rq_auth_uid, + pud->pud_uid, pud->pud_gid, + pud->pud_fsuid, pud->pud_fsgid); + RETURN(-EACCES); + } - if (is_identity_get_disabled(mdt->mdt_identity_cache)) { - if (remote) { - CDEBUG(D_SEC, "remote client must run with identity_get " - "enabled!\n"); - RETURN(-EACCES); - } else { - ucred->uc_identity = NULL; - perm = CFS_SETUID_PERM | CFS_SETGID_PERM | - CFS_SETGRP_PERM; - } - } else { - struct md_identity *identity; + if (is_identity_get_disabled(mdt->mdt_identity_cache)) { + ucred->uc_identity = NULL; + perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM; + } else { + struct md_identity *identity; - identity = mdt_identity_get(mdt->mdt_identity_cache, - pud->pud_uid); - if (IS_ERR(identity)) { - if (unlikely(PTR_ERR(identity) == -EREMCHG && - !remote)) { + identity = mdt_identity_get(mdt->mdt_identity_cache, + pud->pud_uid); + if (IS_ERR(identity)) { + if (unlikely(PTR_ERR(identity) == -EREMCHG)) { ucred->uc_identity = NULL; - perm = CFS_SETUID_PERM | CFS_SETGID_PERM | - CFS_SETGRP_PERM; - } else { - CDEBUG(D_SEC, "Deny access without identity: uid %u\n", - pud->pud_uid); - RETURN(-EACCES); - } - } else { + perm = CFS_SETUID_PERM | CFS_SETGID_PERM | + CFS_SETGRP_PERM; + } else { + CDEBUG(D_SEC, + "Deny access without identity: uid %u\n", + pud->pud_uid); + RETURN(-EACCES); + } + } else { ucred->uc_identity = identity; perm = mdt_identity_get_perm(ucred->uc_identity, - remote, peernid); - } - } + peernid); + } + } - /* find out the setuid/setgid attempt */ - setuid = (pud->pud_uid != pud->pud_fsuid); + /* find out the setuid/setgid attempt */ + setuid = (pud->pud_uid != pud->pud_fsuid); setgid = ((pud->pud_gid != pud->pud_fsgid) || (ucred->uc_identity && (pud->pud_gid != ucred->uc_identity->mi_gid))); @@ -244,13 +232,10 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type, GOTO(out, rc = -EACCES); } - /* - * NB: remote client not allowed to setgroups anyway. - */ - if (!remote && perm & CFS_SETGRP_PERM) { + if (perm & CFS_SETGRP_PERM) { if (pud->pud_ngroups) { /* setgroups for local client */ - ucred->uc_ginfo = cfs_groups_alloc(pud->pud_ngroups); + ucred->uc_ginfo = groups_alloc(pud->pud_ngroups); if (!ucred->uc_ginfo) { CERROR("failed to alloc %d groups\n", pud->pud_ngroups); @@ -278,21 +263,19 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type, mdt_root_squash(info, peernid); /* remove fs privilege for non-root user. */ - if (ucred->uc_fsuid) + if (ucred->uc_fsuid && drop_fs_cap) ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK; else ucred->uc_cap = pud->pud_cap; - if (remote && !(perm & CFS_RMTOWN_PERM)) - ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK | - CFS_CAP_CHOWN_MASK); ucred->uc_valid = UCRED_NEW; + ucred_set_jobid(info, ucred); EXIT; out: if (rc) { if (ucred->uc_ginfo) { - cfs_put_group_info(ucred->uc_ginfo); + put_group_info(ucred->uc_ginfo); ucred->uc_ginfo = NULL; } if (ucred->uc_identity) { @@ -305,6 +288,38 @@ out: return rc; } +/** + * Check whether allow the client to set supplementary group IDs or not. + * + * \param[in] info pointer to the thread context + * \param[in] uc pointer to the RPC user descriptor + * + * \retval true if allow to set supplementary group IDs + * \retval false for other cases + */ +bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc) +{ + __u32 perm; + + /* 1. If identity_upcall is disabled, + * permit local client to do anything. */ + if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache)) + return true; + + /* 2. If fail to get related identities, then forbid any client to + * set supplementary group IDs. */ + if (uc->uc_identity == NULL) + return false; + + /* 3. Check the permission in the identities. */ + perm = mdt_identity_get_perm(uc->uc_identity, + mdt_info_req(info)->rq_peer.nid); + if (perm & CFS_SETGRP_PERM) + return true; + + return false; +} + int mdt_check_ucred(struct mdt_thread_info *info) { struct ptlrpc_request *req = mdt_info_req(info); @@ -314,7 +329,6 @@ int mdt_check_ucred(struct mdt_thread_info *info) struct md_identity *identity = NULL; lnet_nid_t peernid = req->rq_peer.nid; __u32 perm = 0; - __u32 remote = exp_connect_rmtclient(info->mti_exp); int setuid; int setgid; int rc = 0; @@ -325,62 +339,36 @@ int mdt_check_ucred(struct mdt_thread_info *info) if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW)) RETURN(0); - if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc) - RETURN(0); - - /* sanity check: if we use strong authentication, we expect the - * uid which client claimed is true */ - if (remote) { - if (req->rq_auth_mapped_uid == INVALID_UID) { - CDEBUG(D_SEC, "remote user not mapped, deny access!\n"); - RETURN(-EACCES); - } + if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc) + RETURN(0); - if (ptlrpc_user_desc_do_idmap(req, pud)) - RETURN(-EACCES); - - if (req->rq_auth_mapped_uid != pud->pud_uid) { - CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u " - "while client claims %u:%u/%u:%u\n", - libcfs_nid2str(peernid), req->rq_auth_uid, - req->rq_auth_mapped_uid, - pud->pud_uid, pud->pud_gid, - pud->pud_fsuid, pud->pud_fsgid); - RETURN(-EACCES); - } - } else { - if (req->rq_auth_uid != pud->pud_uid) { - CDEBUG(D_SEC, "local client %s: auth uid %u " - "while client claims %u:%u/%u:%u\n", - libcfs_nid2str(peernid), req->rq_auth_uid, - pud->pud_uid, pud->pud_gid, - pud->pud_fsuid, pud->pud_fsgid); - RETURN(-EACCES); - } - } + /* sanity check: if we use strong authentication, we expect the + * uid which client claimed is true */ + if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) && + req->rq_auth_uid != pud->pud_uid) { + CDEBUG(D_SEC, "local client %s: auth uid %u " + "while client claims %u:%u/%u:%u\n", + libcfs_nid2str(peernid), req->rq_auth_uid, + pud->pud_uid, pud->pud_gid, + pud->pud_fsuid, pud->pud_fsgid); + RETURN(-EACCES); + } - if (is_identity_get_disabled(mdt->mdt_identity_cache)) { - if (remote) { - CDEBUG(D_SEC, "remote client must run with identity_get " - "enabled!\n"); - RETURN(-EACCES); - } - RETURN(0); - } + if (is_identity_get_disabled(mdt->mdt_identity_cache)) + RETURN(0); - identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid); - if (IS_ERR(identity)) { - if (unlikely(PTR_ERR(identity) == -EREMCHG && - !remote)) { - RETURN(0); - } else { - CDEBUG(D_SEC, "Deny access without identity: uid %u\n", - pud->pud_uid); - RETURN(-EACCES); - } - } + identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid); + if (IS_ERR(identity)) { + if (unlikely(PTR_ERR(identity) == -EREMCHG)) { + RETURN(0); + } else { + CDEBUG(D_SEC, "Deny access without identity: uid %u\n", + pud->pud_uid); + RETURN(-EACCES); + } + } - perm = mdt_identity_get_perm(identity, remote, peernid); + perm = mdt_identity_get_perm(identity, peernid); /* find out the setuid/setgid attempt */ setuid = (pud->pud_uid != pud->pud_fsuid); setgid = (pud->pud_gid != pud->pud_fsgid || @@ -409,29 +397,20 @@ out: return rc; } -static int old_init_ucred(struct mdt_thread_info *info, - struct mdt_body *body) +static int old_init_ucred_common(struct mdt_thread_info *info, + struct lu_nodemap *nodemap, + bool drop_fs_cap) { - struct lu_ucred *uc = mdt_ucred(info); - struct mdt_device *mdt = info->mti_mdt; - struct md_identity *identity = NULL; + struct lu_ucred *uc = mdt_ucred(info); + struct mdt_device *mdt = info->mti_mdt; + struct md_identity *identity = NULL; - ENTRY; - - LASSERT(uc != NULL); - uc->uc_valid = UCRED_INVALID; - uc->uc_o_uid = uc->uc_uid = body->uid; - uc->uc_o_gid = uc->uc_gid = body->gid; - uc->uc_o_fsuid = uc->uc_fsuid = body->fsuid; - uc->uc_o_fsgid = uc->uc_fsgid = body->fsgid; - uc->uc_suppgids[0] = body->suppgid; - uc->uc_suppgids[1] = -1; - uc->uc_ginfo = NULL; if (!is_identity_get_disabled(mdt->mdt_identity_cache)) { identity = mdt_identity_get(mdt->mdt_identity_cache, uc->uc_fsuid); if (IS_ERR(identity)) { - if (unlikely(PTR_ERR(identity) == -EREMCHG)) { + if (unlikely(PTR_ERR(identity) == -EREMCHG || + uc->uc_cap & CFS_CAP_FS_MASK)) { identity = NULL; } else { CDEBUG(D_SEC, "Deny access without identity: " @@ -442,114 +421,208 @@ static int old_init_ucred(struct mdt_thread_info *info, } uc->uc_identity = identity; + if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) { + uc->uc_fsuid = nodemap->nm_squash_uid; + uc->uc_fsgid = nodemap->nm_squash_gid; + uc->uc_cap = 0; + uc->uc_suppgids[0] = -1; + uc->uc_suppgids[1] = -1; + } + /* process root_squash here. */ mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid); /* remove fs privilege for non-root user. */ - if (uc->uc_fsuid) - uc->uc_cap = body->capability & ~CFS_CAP_FS_MASK; - else - uc->uc_cap = body->capability; + if (uc->uc_fsuid && drop_fs_cap) + uc->uc_cap &= ~CFS_CAP_FS_MASK; uc->uc_valid = UCRED_OLD; + ucred_set_jobid(info, uc); - RETURN(0); + return 0; } -static int old_init_ucred_reint(struct mdt_thread_info *info) +static int old_init_ucred(struct mdt_thread_info *info, + struct mdt_body *body, bool drop_fs_cap) { struct lu_ucred *uc = mdt_ucred(info); - struct mdt_device *mdt = info->mti_mdt; - struct md_identity *identity = NULL; + struct lu_nodemap *nodemap; + int rc; + ENTRY; + + nodemap = nodemap_get_from_exp(info->mti_exp); + if (IS_ERR(nodemap)) + RETURN(PTR_ERR(nodemap)); + body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_CLIENT_TO_FS, body->mbo_uid); + body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_CLIENT_TO_FS, body->mbo_gid); + body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_CLIENT_TO_FS, body->mbo_fsuid); + body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_CLIENT_TO_FS, body->mbo_fsgid); + + LASSERT(uc != NULL); + uc->uc_valid = UCRED_INVALID; + uc->uc_o_uid = uc->uc_uid = body->mbo_uid; + uc->uc_o_gid = uc->uc_gid = body->mbo_gid; + uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid; + uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid; + uc->uc_suppgids[0] = body->mbo_suppgid; + uc->uc_suppgids[1] = -1; + uc->uc_ginfo = NULL; + uc->uc_cap = body->mbo_capability; + + rc = old_init_ucred_common(info, nodemap, drop_fs_cap); + nodemap_putref(nodemap); + + RETURN(rc); +} + +static int old_init_ucred_reint(struct mdt_thread_info *info) +{ + struct lu_ucred *uc = mdt_ucred(info); + struct lu_nodemap *nodemap; + int rc; ENTRY; + nodemap = nodemap_get_from_exp(info->mti_exp); + if (IS_ERR(nodemap)) + RETURN(PTR_ERR(nodemap)); + LASSERT(uc != NULL); + + uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_CLIENT_TO_FS, uc->uc_fsuid); + uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_CLIENT_TO_FS, uc->uc_fsgid); + uc->uc_valid = UCRED_INVALID; uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid; uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid; uc->uc_ginfo = NULL; - if (!is_identity_get_disabled(mdt->mdt_identity_cache)) { - identity = mdt_identity_get(mdt->mdt_identity_cache, - uc->uc_fsuid); - if (IS_ERR(identity)) { - if (unlikely(PTR_ERR(identity) == -EREMCHG)) { - identity = NULL; - } else { - CDEBUG(D_SEC, "Deny access without identity: " - "uid %u\n", uc->uc_fsuid); - RETURN(-EACCES); - } - } - } - uc->uc_identity = identity; - /* process root_squash here. */ - mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid); + rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */ + nodemap_putref(nodemap); - /* remove fs privilege for non-root user. */ - if (uc->uc_fsuid) - uc->uc_cap &= ~CFS_CAP_FS_MASK; - uc->uc_valid = UCRED_OLD; - - RETURN(0); + RETURN(rc); } -int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body) +static inline int __mdt_init_ucred(struct mdt_thread_info *info, + struct mdt_body *body, + bool drop_fs_cap) { - struct ptlrpc_request *req = mdt_info_req(info); - struct lu_ucred *uc = mdt_ucred(info); + struct ptlrpc_request *req = mdt_info_req(info); + struct lu_ucred *uc = mdt_ucred(info); LASSERT(uc != NULL); if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW)) return 0; - mdt_exit_ucred(info); + mdt_exit_ucred(info); - if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc) - return old_init_ucred(info, body); - else - return new_init_ucred(info, BODY_INIT, body); + if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc) + return old_init_ucred(info, body, drop_fs_cap); + else + return new_init_ucred(info, BODY_INIT, body, drop_fs_cap); +} + +int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body) +{ + return __mdt_init_ucred(info, body, true); +} + +/* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry() + * doesn't set correct fsuid explicitely, but raise capability to allow + * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for + * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it + * will fail permission check. */ +int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info, + struct mdt_body *body) +{ + return __mdt_init_ucred(info, body, false); } int mdt_init_ucred_reint(struct mdt_thread_info *info) { - struct ptlrpc_request *req = mdt_info_req(info); + struct ptlrpc_request *req = mdt_info_req(info); struct lu_ucred *uc = mdt_ucred(info); + struct md_attr *ma = &info->mti_attr; LASSERT(uc != NULL); if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW)) return 0; - mdt_exit_ucred(info); + /* LU-5564: for normal close request, skip permission check */ + if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE && + !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP))) + uc->uc_cap |= CFS_CAP_FS_MASK; - if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc) - return old_init_ucred_reint(info); - else - return new_init_ucred(info, REC_INIT, NULL); + mdt_exit_ucred(info); + + if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc) + return old_init_ucred_reint(info); + else + return new_init_ucred(info, REC_INIT, NULL, true); } /* copied from lov/lov_ea.c, just for debugging, will be removed later */ -void mdt_dump_lmm(int level, const struct lov_mds_md *lmm) +void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid) { - const struct lov_ost_data_v1 *lod; - int i; - __u16 count; - - count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count); - - CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n", - le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic), - le32_to_cpu(lmm->lmm_pattern)); - CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n", - le32_to_cpu(lmm->lmm_stripe_size), count); - if (count == LOV_ALL_STRIPES) - return; - LASSERT(count <= LOV_MAX_STRIPE_COUNT); - for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) - CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n", - i, le32_to_cpu(lod->l_ost_idx), - le64_to_cpu(lod->l_object_seq), - le64_to_cpu(lod->l_object_id)); + const struct lov_ost_data_v1 *lod; + int i; + __u16 count; + + if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM))) + return; + + count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count); + + CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n", + POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic), + le32_to_cpu(lmm->lmm_pattern)); + CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n", + le32_to_cpu(lmm->lmm_stripe_size), count); + + /* If it's a directory or a released file, then there are + * no actual objects to print, so bail out. */ + if (valid & OBD_MD_FLDIREA || + le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED) + return; + + LASSERT(count <= LOV_MAX_STRIPE_COUNT); + for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) { + struct ost_id oi; + + ostid_le_to_cpu(&lod->l_ost_oi, &oi); + CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", + i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi)); + } +} + +void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv) +{ + const struct lmv_mds_md_v1 *lmm1; + int i; + + if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM))) + return; + + lmm1 = &lmv->lmv_md_v1; + CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n", + le32_to_cpu(lmm1->lmv_magic), + le32_to_cpu(lmm1->lmv_master_mdt_index), + le32_to_cpu(lmm1->lmv_stripe_count)); + + if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE) + return; + + for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) { + struct lu_fid fid; + + fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]); + CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid)); + } } /* Shrink and/or grow reply buffers */ @@ -565,12 +638,13 @@ int mdt_fix_reply(struct mdt_thread_info *info) body = req_capsule_server_get(pill, &RMF_MDT_BODY); LASSERT(body != NULL); - if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME)) - md_size = body->eadatasize; - else - md_size = 0; + if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | + OBD_MD_LINKNAME)) + md_size = body->mbo_eadatasize; + else + md_size = 0; - acl_size = body->aclsize; + acl_size = body->mbo_aclsize; /* this replay - not send info to client */ if (info->mti_spec.no_create) { @@ -578,11 +652,8 @@ int mdt_fix_reply(struct mdt_thread_info *info) acl_size = 0; } - CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d" - " MDSCAPA = %llx, OSSCAPA = %llx\n", - md_size, acl_size, - (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA), - (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA)); + CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n", + md_size, acl_size); /* &RMF_MDT_BODY, &RMF_MDT_MD, @@ -595,17 +666,21 @@ int mdt_fix_reply(struct mdt_thread_info *info) /* MDT_MD buffer may be bigger than packed value, let's shrink all * buffers before growing it */ if (info->mti_big_lmm_used) { - LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)); - md_packed = req_capsule_get_size(pill, &RMF_MDT_MD, - RCL_SERVER); - LASSERT(md_packed > 0); - /* buffer must be allocated separately */ - LASSERT(info->mti_attr.ma_lmm != - req_capsule_server_get(pill, &RMF_MDT_MD)); - req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER); - /* free big lmm if md_size is not needed */ - if (md_size == 0) + /* big_lmm buffer may be used even without packing the result + * into reply, just for internal server needs */ + if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) + md_packed = req_capsule_get_size(pill, &RMF_MDT_MD, + RCL_SERVER); + + /* free big lmm if md_size is not needed */ + if (md_size == 0 || md_packed == 0) { info->mti_big_lmm_used = 0; + } else { + /* buffer must be allocated separately */ + LASSERT(info->mti_attr.ma_lmm != + req_capsule_server_get(pill, &RMF_MDT_MD)); + req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER); + } } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) { req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER); } @@ -617,12 +692,12 @@ int mdt_fix_reply(struct mdt_thread_info *info) acl_size, RCL_SERVER); if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) && - !(body->valid & OBD_MD_FLMDSCAPA)) - req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER); + !(body->mbo_valid & OBD_MD_FLMDSCAPA)) + req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER); - if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) && - !(body->valid & OBD_MD_FLOSSCAPA)) - req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER); + if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) && + !(body->mbo_valid & OBD_MD_FLOSSCAPA)) + req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER); /* * Some more field should be shrinked if needed. @@ -641,17 +716,25 @@ int mdt_fix_reply(struct mdt_thread_info *info) /* we can't answer with proper LOV EA, drop flags, * the rc is also returned so this request is * considered as failed */ - body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE); + body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE); /* don't return transno along with error */ lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0); } else { - /* now we need to pack right LOV EA */ - lmm = req_capsule_server_get(pill, &RMF_MDT_MD); - LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD, - RCL_SERVER) == - info->mti_attr.ma_lmm_size); - memcpy(lmm, info->mti_attr.ma_lmm, - info->mti_attr.ma_lmm_size); + /* now we need to pack right LOV/LMV EA */ + lmm = req_capsule_server_get(pill, &RMF_MDT_MD); + if (info->mti_attr.ma_valid & MA_LOV) { + LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD, + RCL_SERVER) == + info->mti_attr.ma_lmm_size); + memcpy(lmm, info->mti_attr.ma_lmm, + info->mti_attr.ma_lmm_size); + } else if (info->mti_attr.ma_valid & MA_LMV) { + LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD, + RCL_SERVER) == + info->mti_attr.ma_lmv_size); + memcpy(lmm, info->mti_attr.ma_lmv, + info->mti_attr.ma_lmv_size); + } } /* update mdt_max_mdsize so clients will be aware about that */ if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size) @@ -670,7 +753,6 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo, { struct mdt_body *repbody; const struct lu_attr *la = &ma->ma_attr; - int rc; ENTRY; repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); @@ -683,135 +765,96 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo, CERROR("No need in LOV EA upon unlink\n"); dump_stack(); } - repbody->eadatasize = 0; - - if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) { - repbody->aclsize = ma->ma_cookie_size; - repbody->valid |= OBD_MD_FLCOOKIE; - } - - if (info->mti_mdt->mdt_opts.mo_oss_capa && - info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA && - repbody->valid & OBD_MD_FLEASIZE) { - struct lustre_capa *capa; - - capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2); - LASSERT(capa); - capa->lc_opc = CAPA_OPC_OSS_DESTROY; - rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0); - if (rc) - RETURN(rc); - - repbody->valid |= OBD_MD_FLOSSCAPA; - } + repbody->mbo_eadatasize = 0; RETURN(0); } -static inline unsigned int attr_unpack(__u64 sa_valid) { - unsigned int ia_valid = 0; - - if (sa_valid & MDS_ATTR_MODE) - ia_valid |= ATTR_MODE; - if (sa_valid & MDS_ATTR_UID) - ia_valid |= ATTR_UID; - if (sa_valid & MDS_ATTR_GID) - ia_valid |= ATTR_GID; - if (sa_valid & MDS_ATTR_SIZE) - ia_valid |= ATTR_SIZE; - if (sa_valid & MDS_ATTR_ATIME) - ia_valid |= ATTR_ATIME; - if (sa_valid & MDS_ATTR_MTIME) - ia_valid |= ATTR_MTIME; - if (sa_valid & MDS_ATTR_CTIME) - ia_valid |= ATTR_CTIME; - if (sa_valid & MDS_ATTR_ATIME_SET) - ia_valid |= ATTR_ATIME_SET; - if (sa_valid & MDS_ATTR_MTIME_SET) - ia_valid |= ATTR_MTIME_SET; - if (sa_valid & MDS_ATTR_FORCE) - ia_valid |= ATTR_FORCE; - if (sa_valid & MDS_ATTR_ATTR_FLAG) - ia_valid |= ATTR_ATTR_FLAG; - if (sa_valid & MDS_ATTR_KILL_SUID) - ia_valid |= ATTR_KILL_SUID; - if (sa_valid & MDS_ATTR_KILL_SGID) - ia_valid |= ATTR_KILL_SGID; - if (sa_valid & MDS_ATTR_CTIME_SET) - ia_valid |= ATTR_CTIME_SET; - if (sa_valid & MDS_ATTR_FROM_OPEN) - ia_valid |= ATTR_FROM_OPEN; - if (sa_valid & MDS_ATTR_BLOCKS) - ia_valid |= ATTR_BLOCKS; - if (sa_valid & MDS_OPEN_OWNEROVERRIDE) - ia_valid |= MDS_OPEN_OWNEROVERRIDE; - return ia_valid; -} - static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr, struct md_attr *ma) { - __u64 out; - - out = 0; - if (in & ATTR_MODE) - out |= LA_MODE; - if (in & ATTR_UID) - out |= LA_UID; - if (in & ATTR_GID) - out |= LA_GID; - if (in & ATTR_SIZE) - out |= LA_SIZE; - if (in & ATTR_BLOCKS) - out |= LA_BLOCKS; - - if (in & ATTR_FROM_OPEN) - rr->rr_flags |= MRF_OPEN_TRUNC; - - if (in & ATTR_ATIME_SET) - out |= LA_ATIME; - - if (in & ATTR_CTIME_SET) - out |= LA_CTIME; + __u64 out; + + out = 0; + if (in & MDS_ATTR_MODE) + out |= LA_MODE; + if (in & MDS_ATTR_UID) + out |= LA_UID; + if (in & MDS_ATTR_GID) + out |= LA_GID; + if (in & MDS_ATTR_SIZE) + out |= LA_SIZE; + if (in & MDS_ATTR_BLOCKS) + out |= LA_BLOCKS; + if (in & MDS_ATTR_ATIME_SET) + out |= LA_ATIME; + if (in & MDS_ATTR_CTIME_SET) + out |= LA_CTIME; + if (in & MDS_ATTR_MTIME_SET) + out |= LA_MTIME; + if (in & MDS_ATTR_ATTR_FLAG) + out |= LA_FLAGS; + if (in & MDS_ATTR_KILL_SUID) + out |= LA_KILL_SUID; + if (in & MDS_ATTR_KILL_SGID) + out |= LA_KILL_SGID; + + if (in & MDS_ATTR_FROM_OPEN) + rr->rr_flags |= MRF_OPEN_TRUNC; + if (in & MDS_OPEN_OWNEROVERRIDE) + ma->ma_attr_flags |= MDS_OWNEROVERRIDE; + if (in & MDS_ATTR_FORCE) + ma->ma_attr_flags |= MDS_PERM_BYPASS; + + in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | + MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME | + MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET | + MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG | + MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID | + MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE); + if (in != 0) + CERROR("Unknown attr bits: "LPX64"\n", in); + return out; +} - if (in & ATTR_MTIME_SET) - out |= LA_MTIME; +/* unpacking */ - if (in & ATTR_ATTR_FLAG) - out |= LA_FLAGS; +int mdt_name_unpack(struct req_capsule *pill, + const struct req_msg_field *field, + struct lu_name *ln, + enum mdt_name_flags flags) +{ + ln->ln_name = req_capsule_client_get(pill, field); + ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1; - if (in & ATTR_KILL_SUID) - out |= LA_KILL_SUID; + if (!lu_name_is_valid(ln)) { + ln->ln_name = NULL; + ln->ln_namelen = 0; - if (in & ATTR_KILL_SGID) - out |= LA_KILL_SGID; + return -EPROTO; + } - if (in & MDS_OPEN_OWNEROVERRIDE) - ma->ma_attr_flags |= MDS_OWNEROVERRIDE; + if ((flags & MNF_FIX_ANON) && + ln->ln_namelen == 1 && ln->ln_name[0] == '/') { + /* Newer (3.x) kernels use a name of "/" for the + * "anonymous" disconnected dentries from NFS + * filehandle conversion. See d_obtain_alias(). */ + ln->ln_name = NULL; + ln->ln_namelen = 0; + } - if (in & ATTR_FORCE) - ma->ma_attr_flags |= MDS_PERM_BYPASS; - - /*XXX need ATTR_RAW?*/ - in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS| - ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN| - ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET| - ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE| - ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID); - if (in != 0) - CERROR("Unknown attr bits: "LPX64"\n", in); - return out; + return 0; } -/* unpacking */ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info) { - struct lu_ucred *uc = mdt_ucred(info); - struct md_attr *ma = &info->mti_attr; - struct lu_attr *la = &ma->ma_attr; - struct req_capsule *pill = info->mti_pill; - struct mdt_reint_record *rr = &info->mti_rr; - struct mdt_rec_setattr *rec; + struct lu_ucred *uc = mdt_ucred(info); + struct md_attr *ma = &info->mti_attr; + struct lu_attr *la = &ma->ma_attr; + struct req_capsule *pill = info->mti_pill; + struct mdt_reint_record *rr = &info->mti_rr; + struct mdt_rec_setattr *rec; + struct lu_nodemap *nodemap; ENTRY; CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint)); @@ -827,41 +870,62 @@ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info) uc->uc_suppgids[1] = -1; rr->rr_fid1 = &rec->sa_fid; - la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma); - la->la_mode = rec->sa_mode; - la->la_flags = rec->sa_attr_flags; - la->la_uid = rec->sa_uid; - la->la_gid = rec->sa_gid; - la->la_size = rec->sa_size; - la->la_blocks = rec->sa_blocks; - la->la_ctime = rec->sa_ctime; - la->la_atime = rec->sa_atime; - la->la_mtime = rec->sa_mtime; - ma->ma_valid = MA_INODE; + la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma); + la->la_mode = rec->sa_mode; + la->la_flags = rec->sa_attr_flags; + + nodemap = nodemap_get_from_exp(info->mti_exp); + if (IS_ERR(nodemap)) + RETURN(PTR_ERR(nodemap)); + + la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_CLIENT_TO_FS, rec->sa_uid); + la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_CLIENT_TO_FS, rec->sa_gid); + nodemap_putref(nodemap); + + la->la_size = rec->sa_size; + la->la_blocks = rec->sa_blocks; + la->la_ctime = rec->sa_ctime; + la->la_atime = rec->sa_atime; + la->la_mtime = rec->sa_mtime; + ma->ma_valid = MA_INODE; if (rec->sa_bias & MDS_DATA_MODIFIED) ma->ma_attr_flags |= MDS_DATA_MODIFIED; else ma->ma_attr_flags &= ~MDS_DATA_MODIFIED; - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); + if (rec->sa_bias & MDS_HSM_RELEASE) + ma->ma_attr_flags |= MDS_HSM_RELEASE; + else + ma->ma_attr_flags &= ~MDS_HSM_RELEASE; + + if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP) + ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP; + else + ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP; - RETURN(0); + RETURN(0); } -static int mdt_ioepoch_unpack(struct mdt_thread_info *info) +static int mdt_close_handle_unpack(struct mdt_thread_info *info) { - struct req_capsule *pill = info->mti_pill; - ENTRY; + struct req_capsule *pill = info->mti_pill; + struct mdt_ioepoch *ioepoch; + ENTRY; - if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT)) - info->mti_ioepoch = - req_capsule_client_get(pill, &RMF_MDT_EPOCH); - else - info->mti_ioepoch = NULL; - RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0); + if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT)) + ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH); + else + ioepoch = NULL; + + if (ioepoch == NULL) + RETURN(-EPROTO); + + info->mti_close_handle = ioepoch->mio_handle; + + RETURN(0); } static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) { @@ -888,36 +952,67 @@ static int mdt_setattr_unpack(struct mdt_thread_info *info) if (rc) RETURN(rc); - /* Epoch may be absent */ - mdt_ioepoch_unpack(info); - if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) { rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA); rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT); - ma->ma_lmm_size = rr->rr_eadatalen; - if (ma->ma_lmm_size > 0) { - ma->ma_lmm = (void *)rr->rr_eadata; - ma->ma_valid |= MA_LOV; - } - } + if (rr->rr_eadatalen > 0) { + const struct lmv_user_md *lum; + + lum = rr->rr_eadata; + /* Sigh ma_valid(from req) does not indicate whether + * it will set LOV/LMV EA, so we have to check magic */ + if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) { + ma->ma_valid |= MA_LMV; + ma->ma_lmv = (void *)rr->rr_eadata; + ma->ma_lmv_size = rr->rr_eadatalen; + } else { + ma->ma_valid |= MA_LOV; + ma->ma_lmm = (void *)rr->rr_eadata; + ma->ma_lmm_size = rr->rr_eadatalen; + } + } + } - rc = mdt_dlmreq_unpack(info); - RETURN(rc); + rc = mdt_dlmreq_unpack(info); + RETURN(rc); +} + +static int mdt_intent_close_unpack(struct mdt_thread_info *info) +{ + struct md_attr *ma = &info->mti_attr; + struct req_capsule *pill = info->mti_pill; + ENTRY; + + if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP))) + RETURN(0); + + req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE); + + if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) && + req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT))) + RETURN(-EFAULT); + + RETURN(0); } int mdt_close_unpack(struct mdt_thread_info *info) { - int rc; - ENTRY; + int rc; + ENTRY; - rc = mdt_ioepoch_unpack(info); + rc = mdt_close_handle_unpack(info); if (rc) RETURN(rc); rc = mdt_setattr_unpack_rec(info); if (rc) RETURN(rc); + + rc = mdt_intent_close_unpack(info); + if (rc) + RETURN(rc); + RETURN(mdt_init_ucred_reint(info)); } @@ -959,15 +1054,9 @@ static int mdt_create_unpack(struct mdt_thread_info *info) memset(&sp->u, 0, sizeof(sp->u)); sp->sp_cr_flags = get_mrc_cr_flags(rec); - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA); - - rr->rr_name = req_capsule_client_get(pill, &RMF_NAME); - rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, - RCL_CLIENT) - 1; - LASSERT(rr->rr_name && rr->rr_namelen > 0); + rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0); + if (rc < 0) + RETURN(rc); if (S_ISLNK(attr->la_mode)) { const char *tgt = NULL; @@ -980,11 +1069,20 @@ static int mdt_create_unpack(struct mdt_thread_info *info) if (tgt == NULL) RETURN(-EFAULT); } else { - req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL); - } + req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL); + if (S_ISDIR(attr->la_mode) && + req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) { + sp->u.sp_ea.eadata = + req_capsule_client_get(pill, &RMF_EADATA); + sp->u.sp_ea.eadatalen = + req_capsule_get_size(pill, &RMF_EADATA, + RCL_CLIENT); + sp->sp_cr_flags |= MDS_OPEN_HAS_EA; + } + } - rc = mdt_dlmreq_unpack(info); - RETURN(rc); + rc = mdt_dlmreq_unpack(info); + RETURN(rc); } static int mdt_link_unpack(struct mdt_thread_info *info) @@ -1017,22 +1115,13 @@ static int mdt_link_unpack(struct mdt_thread_info *info) attr->la_mtime = rec->lk_time; attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME; - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) - mdt_set_capainfo(info, 1, rr->rr_fid2, - req_capsule_client_get(pill, &RMF_CAPA2)); - - rr->rr_name = req_capsule_client_get(pill, &RMF_NAME); - if (rr->rr_name == NULL) - RETURN(-EFAULT); - rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1; + rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0); + if (rc < 0) + RETURN(rc); - LASSERT(rr->rr_namelen > 0); + rc = mdt_dlmreq_unpack(info); - rc = mdt_dlmreq_unpack(info); - RETURN(rc); + RETURN(rc); } static int mdt_unlink_unpack(struct mdt_thread_info *info) @@ -1067,26 +1156,27 @@ static int mdt_unlink_unpack(struct mdt_thread_info *info) attr->la_mode = rec->ul_mode; attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE; - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - - rr->rr_name = req_capsule_client_get(pill, &RMF_NAME); - rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1; - if (rr->rr_name == NULL || rr->rr_namelen == 0) - RETURN(-EFAULT); + rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0); + if (rc < 0) + RETURN(rc); if (rec->ul_bias & MDS_VTX_BYPASS) ma->ma_attr_flags |= MDS_VTX_BYPASS; else ma->ma_attr_flags &= ~MDS_VTX_BYPASS; - info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info)); + info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info)); rc = mdt_dlmreq_unpack(info); RETURN(rc); } +static int mdt_rmentry_unpack(struct mdt_thread_info *info) +{ + info->mti_spec.sp_rm_entry = 1; + return mdt_unlink_unpack(info); +} + static int mdt_rename_unpack(struct mdt_thread_info *info) { struct lu_ucred *uc = mdt_ucred(info); @@ -1120,30 +1210,33 @@ static int mdt_rename_unpack(struct mdt_thread_info *info) attr->la_mode = rec->rn_mode; attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE; - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) - mdt_set_capainfo(info, 1, rr->rr_fid2, - req_capsule_client_get(pill, &RMF_CAPA2)); + rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0); + if (rc < 0) + RETURN(rc); - rr->rr_name = req_capsule_client_get(pill, &RMF_NAME); - rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT); - if (rr->rr_name == NULL || rr->rr_tgt == NULL) - RETURN(-EFAULT); - rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1; - rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1; - LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0); + rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0); + if (rc < 0) + RETURN(rc); if (rec->rn_bias & MDS_VTX_BYPASS) ma->ma_attr_flags |= MDS_VTX_BYPASS; else ma->ma_attr_flags &= ~MDS_VTX_BYPASS; + if (rec->rn_bias & MDS_RENAME_MIGRATE) { + req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE); + rc = mdt_close_handle_unpack(info); + if (rc < 0) + RETURN(rc); + info->mti_spec.sp_migrate_close = 1; + } + info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info)); - rc = mdt_dlmreq_unpack(info); - RETURN(rc); + + rc = mdt_dlmreq_unpack(info); + + RETURN(rc); } /* @@ -1211,32 +1304,10 @@ static int mdt_open_unpack(struct mdt_thread_info *info) /* Do not trigger ASSERTION if client miss to set such flags. */ if (unlikely(info->mti_spec.sp_cr_flags == 0)) RETURN(-EPROTO); - info->mti_replayepoch = rec->cr_ioepoch; info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF); - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - if (req_is_replay(req) && - req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) { -#if 0 - mdt_set_capainfo(info, 1, rr->rr_fid2, - req_capsule_client_get(pill, &RMF_CAPA2)); -#else - /* - * FIXME: capa in replay open request might have expired, - * bypass capa check. Security hole? - */ - mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA); - mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA); -#endif - } - - rr->rr_name = req_capsule_client_get(pill, &RMF_NAME); - if (rr->rr_name == NULL) - RETURN(-EFAULT); - rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1; + mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON); if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) { rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA, @@ -1265,12 +1336,13 @@ static int mdt_open_unpack(struct mdt_thread_info *info) static int mdt_setxattr_unpack(struct mdt_thread_info *info) { - struct mdt_reint_record *rr = &info->mti_rr; - struct lu_ucred *uc = mdt_ucred(info); - struct lu_attr *attr = &info->mti_attr.ma_attr; - struct req_capsule *pill = info->mti_pill; - struct mdt_rec_setxattr *rec; - ENTRY; + struct mdt_reint_record *rr = &info->mti_rr; + struct lu_ucred *uc = mdt_ucred(info); + struct lu_attr *attr = &info->mti_attr.ma_attr; + struct req_capsule *pill = info->mti_pill; + struct mdt_rec_setxattr *rec; + int rc; + ENTRY; CLASSERT(sizeof(struct mdt_rec_setxattr) == @@ -1294,17 +1366,9 @@ static int mdt_setxattr_unpack(struct mdt_thread_info *info) attr->la_size = rec->sx_size; attr->la_flags = rec->sx_flags; - if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, rr->rr_fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - else - mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA); - - rr->rr_name = req_capsule_client_get(pill, &RMF_NAME); - if (rr->rr_name == NULL) - RETURN(-EFAULT); - rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1; - LASSERT(rr->rr_namelen > 0); + rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0); + if (rc < 0) + RETURN(rc); if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) { rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA, @@ -1322,6 +1386,9 @@ static int mdt_setxattr_unpack(struct mdt_thread_info *info) RETURN(-EFAULT); } + if (mdt_dlmreq_unpack(info) < 0) + RETURN(-EPROTO); + RETURN(0); } @@ -1329,13 +1396,15 @@ static int mdt_setxattr_unpack(struct mdt_thread_info *info) typedef int (*reint_unpacker)(struct mdt_thread_info *info); static reint_unpacker mdt_reint_unpackers[REINT_MAX] = { - [REINT_SETATTR] = mdt_setattr_unpack, - [REINT_CREATE] = mdt_create_unpack, - [REINT_LINK] = mdt_link_unpack, - [REINT_UNLINK] = mdt_unlink_unpack, - [REINT_RENAME] = mdt_rename_unpack, - [REINT_OPEN] = mdt_open_unpack, - [REINT_SETXATTR] = mdt_setxattr_unpack + [REINT_SETATTR] = mdt_setattr_unpack, + [REINT_CREATE] = mdt_create_unpack, + [REINT_LINK] = mdt_link_unpack, + [REINT_UNLINK] = mdt_unlink_unpack, + [REINT_RENAME] = mdt_rename_unpack, + [REINT_OPEN] = mdt_open_unpack, + [REINT_SETXATTR] = mdt_setxattr_unpack, + [REINT_RMENTRY] = mdt_rmentry_unpack, + [REINT_MIGRATE] = mdt_rename_unpack, }; int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)