* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_MDS
+#include <linux/user_namespace.h>
+#ifdef HAVE_UIDGID_HEADER
+# include <linux/uidgid.h>
+#endif
#include "mdt_internal.h"
#include <lnet/nidstr.h>
#include <lustre_nodemap.h>
}
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;
RETURN(-EACCES);
}
} else {
- if (req->rq_auth_uid != pud->pud_uid) {
+ 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,
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;
RETURN(-EACCES);
}
} else {
- if (req->rq_auth_uid != pud->pud_uid) {
+ 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,
return rc;
}
-static void mdt_squash_nodemap_id(struct lu_ucred *ucred,
- struct lu_nodemap *nodemap)
+static int old_init_ucred_common(struct mdt_thread_info *info,
+ struct lu_nodemap *nodemap,
+ bool drop_fs_cap)
{
- if (ucred->uc_o_uid == nodemap->nm_squash_uid) {
- ucred->uc_fsuid = nodemap->nm_squash_uid;
- ucred->uc_fsgid = nodemap->nm_squash_gid;
- ucred->uc_cap = 0;
- ucred->uc_suppgids[0] = -1;
- ucred->uc_suppgids[1] = -1;
+ struct lu_ucred *uc = mdt_ucred(info);
+ struct mdt_device *mdt = info->mti_mdt;
+ struct md_identity *identity = 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 ||
+ uc->uc_cap & CFS_CAP_FS_MASK)) {
+ identity = NULL;
+ } else {
+ CDEBUG(D_SEC, "Deny access without identity: "
+ "uid %u\n", uc->uc_fsuid);
+ RETURN(-EACCES);
+ }
+ }
}
-}
+ 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 && drop_fs_cap)
+ uc->uc_cap &= ~CFS_CAP_FS_MASK;
+ uc->uc_valid = UCRED_OLD;
+ ucred_set_jobid(info, uc);
+ return 0;
+}
static int old_init_ucred(struct mdt_thread_info *info,
- struct mdt_body *body)
+ 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 =
- info->mti_exp->exp_target_data.ted_nodemap;
+ 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));
+
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,
uc->uc_suppgids[0] = body->mbo_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)) {
- identity = NULL;
- } else {
- CDEBUG(D_SEC, "Deny access without identity: "
- "uid %u\n", uc->uc_fsuid);
- RETURN(-EACCES);
- }
- }
- }
- uc->uc_identity = identity;
-
- mdt_squash_nodemap_id(uc, nodemap);
-
- /* process root_squash here. */
- mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
+ uc->uc_cap = body->mbo_capability;
- /* remove fs privilege for non-root user. */
- if (uc->uc_fsuid)
- uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
- else
- uc->uc_cap = body->mbo_capability;
- uc->uc_valid = UCRED_OLD;
- ucred_set_jobid(info, uc);
+ rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
+ nodemap_putref(nodemap);
- RETURN(0);
+ RETURN(rc);
}
static int old_init_ucred_reint(struct mdt_thread_info *info)
{
- struct lu_ucred *uc = mdt_ucred(info);
- struct mdt_device *mdt = info->mti_mdt;
- struct md_identity *identity = NULL;
- struct lu_nodemap *nodemap =
- info->mti_exp->exp_target_data.ted_nodemap;
+ 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,
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;
- ucred_set_jobid(info, uc);
-
- 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);
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_reint(info);
- else
- return new_init_ucred(info, REC_INIT, NULL);
+ 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 */
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->mbo_valid & OBD_MD_FLMDSCAPA),
- (unsigned long long)(body->mbo_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,
{
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);
}
repbody->mbo_eadatasize = 0;
- if (info->mti_mdt->mdt_lut.lut_oss_capa &&
- exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
- repbody->mbo_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->mbo_valid |= OBD_MD_FLOSSCAPA;
- }
-
RETURN(0);
}
return out;
}
-void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
- const struct lu_fid *fid, struct lustre_capa *capa)
-{
- struct lu_capainfo *lci;
-
- LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
- if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
- !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
- return;
-
- lci = lu_capainfo_get(info->mti_env);
- LASSERT(lci);
- lci->lci_fid[offset] = *fid;
- lci->lci_capa[offset] = capa;
-}
-
-#ifdef DEBUG_CAPA
-void mdt_dump_capainfo(struct mdt_thread_info *info)
-{
- struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
- int i;
-
- if (lci == NULL)
- return;
-
- for (i = 0; i < LU_CAPAINFO_MAX; i++) {
- if (lci->lci_capa[i] == NULL) {
- CERROR("no capa for index %d "DFID"\n",
- i, PFID(&lci->lci_fid[i]));
- continue;
- }
- if (lci->lci_capa[i] == BYPASS_CAPA) {
- CERROR("bypass for index %d "DFID"\n",
- i, PFID(&lci->lci_fid[i]));
- continue;
- }
- DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
- }
-}
-#endif /* DEBUG_CAPA */
-
/* unpacking */
int mdt_name_unpack(struct req_capsule *pill,
struct req_capsule *pill = info->mti_pill;
struct mdt_reint_record *rr = &info->mti_rr;
struct mdt_rec_setattr *rec;
- struct lu_nodemap *nodemap =
- info->mti_exp->exp_target_data.ted_nodemap;
+ struct lu_nodemap *nodemap;
ENTRY;
CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
rr->rr_fid1 = &rec->sa_fid;
la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
- /* If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
- * the client does not have OBD_CONNECT_FULL20, convert it
- * to LA_xTIME. LU-3036 */
- if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
- if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
- (rec->sa_valid & MDS_ATTR_ATIME))
- la->la_valid |= LA_ATIME;
- if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
- (rec->sa_valid & MDS_ATTR_MTIME))
- la->la_valid |= LA_MTIME;
- if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
- (rec->sa_valid & MDS_ATTR_CTIME))
- la->la_valid |= LA_CTIME;
- }
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;
else
ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
- 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_CLOSE_LAYOUT_SWAP)
+ ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
+ else
+ ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
RETURN(0);
}
if (ioepoch == NULL)
RETURN(-EPROTO);
- info->mti_close_handle = ioepoch->handle;
+ info->mti_close_handle = ioepoch->mio_handle;
RETURN(0);
}
RETURN(rc);
}
-static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
+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;
+ struct req_capsule *pill = info->mti_pill;
ENTRY;
- if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
+ if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
RETURN(0);
- req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
+ 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)))
if (rc)
RETURN(rc);
- rc = mdt_hsm_release_unpack(info);
+ rc = mdt_intent_close_unpack(info);
if (rc)
RETURN(rc);
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);
-
rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
if (rc < 0)
RETURN(rc);
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));
-
rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
if (rc < 0)
RETURN(rc);
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));
-
rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
if (rc < 0)
RETURN(rc);
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);
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));
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
- }
-
mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
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);
-
rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
if (rc < 0)
RETURN(rc);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff