Whamcloud - gitweb
LU-11085 nodemap: switch interval tree to in-kernel impl.
[fs/lustre-release.git] / lustre / ptlrpc / nodemap_handler.c
index 0cd0366..5fc80bf 100644 (file)
 /*
  * Copyright (C) 2013, Trustees of Indiana University
  *
- * Copyright (c) 2014, Intel Corporation.
+ * Copyright (c) 2017, Intel Corporation.
  *
  * Author: Joshua Walgenbach <jjw@iu.edu>
  */
 #include <linux/module.h>
 #include <linux/sort.h>
-#include <lnet/nidstr.h>
+#include <uapi/linux/lnet/nidstr.h>
 #include <lustre_net.h>
 #include <lustre_acl.h>
 #include <lustre_eacl.h>
@@ -71,9 +71,9 @@ static void nodemap_destroy(struct lu_nodemap *nodemap)
        nm_member_reclassify_nodemap(nodemap);
        up_read(&active_config->nmc_range_tree_lock);
 
-       write_lock(&nodemap->nm_idmap_lock);
+       down_write(&nodemap->nm_idmap_lock);
        idmap_delete_tree(nodemap);
-       write_unlock(&nodemap->nm_idmap_lock);
+       up_write(&nodemap->nm_idmap_lock);
 
        mutex_unlock(&active_config_lock);
 
@@ -257,14 +257,14 @@ struct lu_nodemap *nodemap_lookup(const char *name)
  */
 struct lu_nodemap *nodemap_classify_nid(lnet_nid_t nid)
 {
-       struct lu_nid_range     *range;
-       struct lu_nodemap       *nodemap;
+       struct lu_nid_range *range;
+       struct lu_nodemap *nodemap;
        int rc;
 
        ENTRY;
 
        /* don't use 0@lo, use the first non-lo local NID instead */
-       if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND) {
+       if (nid == LNET_NID_LO_0) {
                struct lnet_process_id id;
                int i = 0;
 
@@ -272,7 +272,7 @@ struct lu_nodemap *nodemap_classify_nid(lnet_nid_t nid)
                        rc = LNetGetId(i++, &id);
                        if (rc < 0)
                                RETURN(ERR_PTR(-EINVAL));
-               } while (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND);
+               } while (id.nid == LNET_NID_LO_0);
 
                nid = id.nid;
                CDEBUG(D_INFO, "found nid %s\n", libcfs_nid2str(nid));
@@ -448,28 +448,56 @@ EXPORT_SYMBOL(nodemap_del_member);
 /**
  * add an idmap to the proper nodemap trees
  *
- * \param      name            name of nodemap
+ * \param      nodemap         nodemap to add idmap to
  * \param      id_type         NODEMAP_UID or NODEMAP_GID
  * \param      map             array[2] __u32 containing the map values
  *                             map[0] is client id
  *                             map[1] is the filesystem id
  *
- * \retval     0 on success
+ * \retval     0       on success
+ * \retval     < 0     if error occurs
  */
 int nodemap_add_idmap_helper(struct lu_nodemap *nodemap,
                             enum nodemap_id_type id_type,
                             const __u32 map[2])
 {
        struct lu_idmap         *idmap;
+       struct lu_idmap         *temp;
        int                     rc = 0;
 
        idmap = idmap_create(map[0], map[1]);
        if (idmap == NULL)
                GOTO(out, rc = -ENOMEM);
 
-       write_lock(&nodemap->nm_idmap_lock);
-       idmap_insert(id_type, idmap, nodemap);
-       write_unlock(&nodemap->nm_idmap_lock);
+       down_write(&nodemap->nm_idmap_lock);
+       temp = idmap_insert(id_type, idmap, nodemap);
+       /* If the new id_client or id_fs is matched, the old idmap and its
+        * index should be deleted according to its id_client before the new
+        * idmap is added again.
+        */
+       if (IS_ERR(temp))
+               GOTO(out_insert, rc = PTR_ERR(temp));
+       if (temp) {
+               __u32 del_map[2];
+
+               del_map[0] = temp->id_client;
+               idmap_delete(id_type, temp, nodemap);
+               rc = nodemap_idx_idmap_del(nodemap, id_type, del_map);
+               /* In case there is any corrupted idmap */
+               if (!rc || unlikely(rc == -ENOENT)) {
+                       temp = idmap_insert(id_type, idmap, nodemap);
+                       if (IS_ERR(temp))
+                               rc = PTR_ERR(temp);
+                       else if (!temp)
+                               rc = 0;
+                       else
+                               rc = -EPERM;
+               }
+       }
+out_insert:
+       if (rc)
+               OBD_FREE_PTR(idmap);
+       up_write(&nodemap->nm_idmap_lock);
        nm_member_revoke_locks(nodemap);
 
 out:
@@ -482,6 +510,8 @@ int nodemap_add_idmap(const char *name, enum nodemap_id_type id_type,
        struct lu_nodemap       *nodemap = NULL;
        int                      rc;
 
+       ENTRY;
+
        mutex_lock(&active_config_lock);
        nodemap = nodemap_lookup(name);
        if (IS_ERR(nodemap)) {
@@ -500,7 +530,7 @@ int nodemap_add_idmap(const char *name, enum nodemap_id_type id_type,
        nodemap_putref(nodemap);
 
 out:
-       return rc;
+       RETURN(rc);
 }
 EXPORT_SYMBOL(nodemap_add_idmap);
 
@@ -522,6 +552,8 @@ int nodemap_del_idmap(const char *name, enum nodemap_id_type id_type,
        struct lu_idmap         *idmap = NULL;
        int                     rc = 0;
 
+       ENTRY;
+
        mutex_lock(&active_config_lock);
        nodemap = nodemap_lookup(name);
        if (IS_ERR(nodemap)) {
@@ -532,7 +564,7 @@ int nodemap_del_idmap(const char *name, enum nodemap_id_type id_type,
        if (is_default_nodemap(nodemap))
                GOTO(out_putref, rc = -EINVAL);
 
-       write_lock(&nodemap->nm_idmap_lock);
+       down_write(&nodemap->nm_idmap_lock);
        idmap = idmap_search(nodemap, NODEMAP_CLIENT_TO_FS, id_type,
                             map[0]);
        if (idmap == NULL) {
@@ -541,7 +573,7 @@ int nodemap_del_idmap(const char *name, enum nodemap_id_type id_type,
                idmap_delete(id_type, idmap, nodemap);
                rc = nodemap_idx_idmap_del(nodemap, id_type, map);
        }
-       write_unlock(&nodemap->nm_idmap_lock);
+       up_write(&nodemap->nm_idmap_lock);
 
 out_putref:
        mutex_unlock(&active_config_lock);
@@ -550,7 +582,7 @@ out_putref:
        nodemap_putref(nodemap);
 
 out:
-       return rc;
+       RETURN(rc);
 }
 EXPORT_SYMBOL(nodemap_del_idmap);
 
@@ -615,7 +647,7 @@ EXPORT_SYMBOL(nodemap_get_from_exp);
  * is, return 0. Otherwise, return the squash uid or gid.
  *
  * if the nodemap is configured to trusted the ids from the client system, just
- * return the passwd id without mapping.
+ * return the passed id without mapping.
  *
  * if by this point, we haven't returned and the nodemap in question is the
  * default nodemap, return the squash uid or gid.
@@ -638,6 +670,12 @@ __u32 nodemap_map_id(struct lu_nodemap *nodemap,
        if (unlikely(nodemap == NULL))
                goto out;
 
+       if (nodemap->nmf_map_uid_only && id_type == NODEMAP_GID)
+               goto out;
+
+       if (nodemap->nmf_map_gid_only && id_type == NODEMAP_UID)
+               goto out;
+
        if (id == 0) {
                if (nodemap->nmf_allow_root_access)
                        goto out;
@@ -651,10 +689,10 @@ __u32 nodemap_map_id(struct lu_nodemap *nodemap,
        if (is_default_nodemap(nodemap))
                goto squash;
 
-       read_lock(&nodemap->nm_idmap_lock);
+       down_read(&nodemap->nm_idmap_lock);
        idmap = idmap_search(nodemap, tree_type, id_type, id);
        if (idmap == NULL) {
-               read_unlock(&nodemap->nm_idmap_lock);
+               up_read(&nodemap->nm_idmap_lock);
                goto squash;
        }
 
@@ -662,7 +700,7 @@ __u32 nodemap_map_id(struct lu_nodemap *nodemap,
                found_id = idmap->id_client;
        else
                found_id = idmap->id_fs;
-       read_unlock(&nodemap->nm_idmap_lock);
+       up_read(&nodemap->nm_idmap_lock);
        RETURN(found_id);
 
 squash:
@@ -690,22 +728,24 @@ ssize_t nodemap_map_acl(struct lu_nodemap *nodemap, void *buf, size_t size,
                        enum nodemap_tree_type tree_type)
 {
        posix_acl_xattr_header  *header = buf;
-       posix_acl_xattr_entry   *entry = &header->a_entries[0];
+       posix_acl_xattr_entry   *entry = GET_POSIX_ACL_XATTR_ENTRY(header);
        posix_acl_xattr_entry   *new_entry = entry;
        posix_acl_xattr_entry   *end;
        int                      count;
 
+       ENTRY;
+
        if (!nodemap_active)
-               return size;
+               RETURN(size);
 
        if (unlikely(nodemap == NULL))
-               return size;
+               RETURN(size);
 
        count = posix_acl_xattr_count(size);
        if (count < 0)
-               return -EINVAL;
+               RETURN(-EINVAL);
        if (count == 0)
-               return 0;
+               RETURN(0);
 
        for (end = entry + count; entry != end; entry++) {
                __u16 tag = le16_to_cpu(entry->e_tag);
@@ -735,7 +775,7 @@ ssize_t nodemap_map_acl(struct lu_nodemap *nodemap, void *buf, size_t size,
                new_entry++;
        }
 
-       return (void *)new_entry - (void *)header;
+       RETURN((void *)new_entry - (void *)header);
 }
 EXPORT_SYMBOL(nodemap_map_acl);
 
@@ -822,7 +862,7 @@ EXPORT_SYMBOL(nodemap_add_range);
 /**
  * delete a range
  * \param      name            nodemap name
- * \param      range_str       string containing range
+ * \param      nid             nid range
  * \retval     0 on success
  *
  * Delete range from global range tree, and remove it
@@ -850,6 +890,10 @@ int nodemap_del_range(const char *name, const lnet_nid_t nid[2])
                up_write(&active_config->nmc_range_tree_lock);
                GOTO(out_putref, rc = -EINVAL);
        }
+       if (range->rn_nodemap != nodemap) {
+               up_write(&active_config->nmc_range_tree_lock);
+               GOTO(out_putref, rc = -EINVAL);
+       }
        rc = nodemap_idx_range_del(range);
        range_delete(&active_config->nmc_range_tree, range);
        nm_member_reclassify_nodemap(nodemap);
@@ -880,8 +924,20 @@ static int nodemap_set_fileset_helper(struct nodemap_config *config,
 {
        int rc = 0;
 
-       /* we allow fileset = "" which means clear fileset info */
-       if (fileset == NULL || (fileset[0] != 0 && fileset[0] != '/'))
+       /* Allow 'fileset=clear' in addition to 'fileset=""' to clear fileset
+        * because either command 'lctl set_param -P *.*.fileset=""' or
+        * 'lctl nodemap_set_fileset --fileset ""' can only work correctly
+        * on MGS, while on other servers, both commands will invoke upcall
+        * "/usr/sbin/lctl set_param nodemap.default.fileset=" by function
+        * process_param2_config(), which will cause "no value" error and
+        * won't clear fileset.
+        * 'fileset=""' is still kept for compatibility reason.
+        */
+       if (fileset == NULL)
+               rc = -EINVAL;
+       else if (fileset[0] == '\0' || strcmp(fileset, "clear") == 0)
+               nodemap->nm_fileset[0] = '\0';
+       else if (fileset[0] != '/')
                rc = -EINVAL;
        else if (strlcpy(nodemap->nm_fileset, fileset,
                         sizeof(nodemap->nm_fileset)) >=
@@ -903,11 +959,7 @@ int nodemap_set_fileset(const char *name, const char *fileset)
                GOTO(out, rc = PTR_ERR(nodemap));
        }
 
-       if (is_default_nodemap(nodemap))
-               rc = -EINVAL;
-       else
-               rc = nodemap_set_fileset_helper(active_config, nodemap,
-                                               fileset);
+       rc = nodemap_set_fileset_helper(active_config, nodemap, fileset);
        mutex_unlock(&active_config_lock);
 
        nodemap_putref(nodemap);
@@ -925,19 +977,124 @@ EXPORT_SYMBOL(nodemap_set_fileset);
  */
 char *nodemap_get_fileset(const struct lu_nodemap *nodemap)
 {
-       if (!nodemap_active || is_default_nodemap(nodemap))
+       if (!nodemap_active)
                return NULL;
-       else
-               return (char *)nodemap->nm_fileset;
+
+       return (char *)nodemap->nm_fileset;
 }
 EXPORT_SYMBOL(nodemap_get_fileset);
 
+static int nodemap_validate_sepol(const char *sepol)
+{
+       char buf[LUSTRE_NODEMAP_SEPOL_LENGTH + 1];
+       char *p = (char *)sepol;
+       char *q = buf;
+       char polname[NAME_MAX + 1] = "";
+       char hash[SELINUX_POLICY_HASH_LEN + 1] = "";
+       unsigned char mode;
+       unsigned short ver;
+
+       BUILD_BUG_ON(sizeof(buf) != sizeof(((struct lu_nodemap *)0)->nm_sepol));
+
+       if (sepol == NULL)
+               return -EINVAL;
+
+       /* we allow sepol = "" which means clear SELinux policy info */
+       if (sepol[0] == '\0')
+               return 0;
+
+       /* make a copy of sepol, by replacing ':' with space
+        * so that we can use sscanf over the string
+        */
+       while (p-sepol < sizeof(buf)) {
+               if (*p == ':')
+                       *q = ' ';
+               else
+                       *q = *p;
+               if (*p == '\0')
+                       break;
+               p++;
+               q++;
+       }
+       if (p-sepol == sizeof(buf))
+               return -ENAMETOOLONG;
+
+       if (sscanf(buf, "%1hhu %s %hu %s", &mode, polname, &ver, hash) != 4)
+               return -EINVAL;
+
+       if (mode != 0 && mode != 1)
+               return -EINVAL;
+
+       return 0;
+}
+
+/**
+ * set SELinux policy on nodemap
+ * \param      name            nodemap to set SELinux policy info on
+ * \param      sepol           string containing SELinux policy info
+ * \retval     0 on success
+ *
+ * set SELinux policy info on the named nodemap
+ */
+int nodemap_set_sepol(const char *name, const char *sepol)
+{
+       struct lu_nodemap       *nodemap = NULL;
+       int                      rc;
+
+       rc = nodemap_validate_sepol(sepol);
+       if (rc < 0)
+               GOTO(out, rc);
+
+       mutex_lock(&active_config_lock);
+       nodemap = nodemap_lookup(name);
+       if (IS_ERR(nodemap)) {
+               mutex_unlock(&active_config_lock);
+               GOTO(out, rc = PTR_ERR(nodemap));
+       }
+
+       if (is_default_nodemap(nodemap)) {
+               /* We do not want nodes in the default nodemap to have
+                * SELinux restrictions. Sec admin should create dedicated
+                * nodemap entries for this.
+                */
+               GOTO(out_putref, rc = -EINVAL);
+       }
+
+       /* truncation cannot happen, as string length was checked in
+        * nodemap_validate_sepol()
+        */
+       strlcpy(nodemap->nm_sepol, sepol, sizeof(nodemap->nm_sepol));
+
+out_putref:
+       mutex_unlock(&active_config_lock);
+       nodemap_putref(nodemap);
+out:
+       return rc;
+}
+EXPORT_SYMBOL(nodemap_set_sepol);
+
+/**
+ * get SELinux policy info defined on nodemap
+ * \param      nodemap         nodemap to get SELinux policy info from
+ * \retval     SELinux policy info, or NULL if not defined or not activated
+ *
+ * get the SELinux policy info defined on the nodemap
+ */
+const char *nodemap_get_sepol(const struct lu_nodemap *nodemap)
+{
+       if (is_default_nodemap(nodemap))
+               return NULL;
+       else
+               return (char *)nodemap->nm_sepol;
+}
+EXPORT_SYMBOL(nodemap_get_sepol);
+
 /**
  * Nodemap constructor
  *
  * Creates an lu_nodemap structure and assigns sane default
  * member values. If this is the default nodemap, the defaults
- * are the most restictive in xterms of mapping behavior. Otherwise
+ * are the most restrictive in terms of mapping behavior. Otherwise
  * the default flags should be inherited from the default nodemap.
  * The adds nodemap to nodemap_hash.
  *
@@ -958,6 +1115,7 @@ struct lu_nodemap *nodemap_create(const char *name,
        struct lu_nodemap       *default_nodemap;
        struct cfs_hash         *hash = config->nmc_nodemap_hash;
        int                      rc = 0;
+       ENTRY;
 
        default_nodemap = config->nmc_default_nodemap;
 
@@ -971,9 +1129,8 @@ struct lu_nodemap *nodemap_create(const char *name,
 
        OBD_ALLOC_PTR(nodemap);
        if (nodemap == NULL) {
-               CERROR("cannot allocate memory (%zu bytes)"
-                      "for nodemap '%s'\n", sizeof(*nodemap),
-                      name);
+               CERROR("cannot allocate memory (%zu bytes) for nodemap '%s'\n",
+                      sizeof(*nodemap), name);
                GOTO(out, rc = -ENOMEM);
        }
 
@@ -994,7 +1151,7 @@ struct lu_nodemap *nodemap_create(const char *name,
        INIT_LIST_HEAD(&nodemap->nm_member_list);
 
        mutex_init(&nodemap->nm_member_list_lock);
-       rwlock_init(&nodemap->nm_idmap_lock);
+       init_rwsem(&nodemap->nm_idmap_lock);
        nodemap->nm_fs_to_client_uidmap = RB_ROOT;
        nodemap->nm_client_to_fs_uidmap = RB_ROOT;
        nodemap->nm_fs_to_client_gidmap = RB_ROOT;
@@ -1012,9 +1169,15 @@ struct lu_nodemap *nodemap_create(const char *name,
                nodemap->nmf_trust_client_ids = 0;
                nodemap->nmf_allow_root_access = 0;
                nodemap->nmf_deny_unknown = 0;
+               nodemap->nmf_map_uid_only = 0;
+               nodemap->nmf_map_gid_only = 0;
+               nodemap->nmf_enable_audit = 1;
+               nodemap->nmf_forbid_encryption = 0;
 
                nodemap->nm_squash_uid = NODEMAP_NOBODY_UID;
                nodemap->nm_squash_gid = NODEMAP_NOBODY_GID;
+               nodemap->nm_fileset[0] = '\0';
+               nodemap->nm_sepol[0] = '\0';
                if (!is_default)
                        CWARN("adding nodemap '%s' to config without"
                              " default nodemap\n", nodemap->nm_name);
@@ -1025,17 +1188,26 @@ struct lu_nodemap *nodemap_create(const char *name,
                                default_nodemap->nmf_allow_root_access;
                nodemap->nmf_deny_unknown =
                                default_nodemap->nmf_deny_unknown;
+               nodemap->nmf_map_uid_only =
+                               default_nodemap->nmf_map_uid_only;
+               nodemap->nmf_map_gid_only =
+                               default_nodemap->nmf_map_gid_only;
+               nodemap->nmf_enable_audit =
+                       default_nodemap->nmf_enable_audit;
+               nodemap->nmf_forbid_encryption =
+                       default_nodemap->nmf_forbid_encryption;
 
                nodemap->nm_squash_uid = default_nodemap->nm_squash_uid;
                nodemap->nm_squash_gid = default_nodemap->nm_squash_gid;
-               nodemap->nm_fileset[0] = 0;
+               nodemap->nm_fileset[0] = '\0';
+               nodemap->nm_sepol[0] = '\0';
        }
 
-       return nodemap;
+       RETURN(nodemap);
 
 out:
        CERROR("cannot add nodemap: '%s': rc = %d\n", name, rc);
-       return ERR_PTR(rc);
+       RETURN(ERR_PTR(rc));
 }
 
 /**
@@ -1123,6 +1295,42 @@ out:
 }
 EXPORT_SYMBOL(nodemap_set_trust_client_ids);
 
+int nodemap_set_mapping_mode(const char *name, enum nodemap_mapping_modes mode)
+{
+       struct lu_nodemap       *nodemap = NULL;
+       int                     rc = 0;
+
+       mutex_lock(&active_config_lock);
+       nodemap = nodemap_lookup(name);
+       mutex_unlock(&active_config_lock);
+       if (IS_ERR(nodemap))
+               GOTO(out, rc = PTR_ERR(nodemap));
+
+       switch (mode) {
+       case NODEMAP_MAP_BOTH:
+               nodemap->nmf_map_uid_only = 0;
+               nodemap->nmf_map_gid_only = 0;
+               break;
+       case NODEMAP_MAP_UID_ONLY:
+               nodemap->nmf_map_uid_only = 1;
+               nodemap->nmf_map_gid_only = 0;
+               break;
+       case NODEMAP_MAP_GID_ONLY:
+               nodemap->nmf_map_uid_only = 0;
+               nodemap->nmf_map_gid_only = 1;
+               break;
+       default:
+               CWARN("cannot set unknown mapping mode, mode = %d\n", mode);
+       }
+       rc = nodemap_idx_nodemap_update(nodemap);
+
+       nm_member_revoke_locks(nodemap);
+       nodemap_putref(nodemap);
+out:
+       return rc;
+}
+EXPORT_SYMBOL(nodemap_set_mapping_mode);
+
 /**
  * Update the squash_uid for a nodemap.
  *
@@ -1202,6 +1410,63 @@ bool nodemap_can_setquota(const struct lu_nodemap *nodemap)
 EXPORT_SYMBOL(nodemap_can_setquota);
 
 /**
+ * Set the nmf_enable_audit flag to true or false.
+ * \param      name            nodemap name
+ * \param      audit_mode      if true, allow audit
+ * \retval     0 on success
+ *
+ */
+int nodemap_set_audit_mode(const char *name, bool enable_audit)
+{
+       struct lu_nodemap       *nodemap = NULL;
+       int                     rc = 0;
+
+       mutex_lock(&active_config_lock);
+       nodemap = nodemap_lookup(name);
+       mutex_unlock(&active_config_lock);
+       if (IS_ERR(nodemap))
+               GOTO(out, rc = PTR_ERR(nodemap));
+
+       nodemap->nmf_enable_audit = enable_audit;
+       rc = nodemap_idx_nodemap_update(nodemap);
+
+       nm_member_revoke_locks(nodemap);
+       nodemap_putref(nodemap);
+out:
+       return rc;
+}
+EXPORT_SYMBOL(nodemap_set_audit_mode);
+
+/**
+ * Set the nmf_forbid_encryption flag to true or false.
+ * \param      name                    nodemap name
+ * \param      forbid_encryption       if true, forbid encryption
+ * \retval     0 on success
+ *
+ */
+int nodemap_set_forbid_encryption(const char *name, bool forbid_encryption)
+{
+       struct lu_nodemap       *nodemap = NULL;
+       int                     rc = 0;
+
+       mutex_lock(&active_config_lock);
+       nodemap = nodemap_lookup(name);
+       mutex_unlock(&active_config_lock);
+       if (IS_ERR(nodemap))
+               GOTO(out, rc = PTR_ERR(nodemap));
+
+       nodemap->nmf_forbid_encryption = forbid_encryption;
+       rc = nodemap_idx_nodemap_update(nodemap);
+
+       nm_member_revoke_locks(nodemap);
+       nodemap_putref(nodemap);
+out:
+       return rc;
+}
+EXPORT_SYMBOL(nodemap_set_forbid_encryption);
+
+
+/**
  * Add a nodemap
  *
  * \param      name            name of nodemap
@@ -1357,6 +1622,8 @@ struct nodemap_config *nodemap_config_alloc(void)
 
        init_rwsem(&config->nmc_range_tree_lock);
 
+       config->nmc_range_tree.nmrt_range_interval_root = INTERVAL_TREE_ROOT;
+
        return config;
 }
 EXPORT_SYMBOL(nodemap_config_alloc);