/**
*
* cache_inode_lookup_sw: looks up for a name in a directory indicated by a
* cached entry.
*
* Looks up for a name in a directory indicated by a cached entry. The directory
* should have been cached before.
*
* @param pentry_parent [IN] entry for the parent directory to be managed.
* @param name [IN] name of the entry that we are looking for in the
* cache.
* @param pattr [OUT] attributes for the entry that we have found.
* @param ht [IN] hash table used for the cache, unused in this
* call.
* @param pclient [INOUT] ressource allocated by the client for the nfs
* management.
* @param pcontext [IN] FSAL credentials
* @param pstatus [OUT] returned status.
* @param use_mutex [IN] if TRUE, mutex management is done, not if equal
* to FALSE.
*
* @return CACHE_INODE_SUCCESS if operation is a success \n
* @return CACHE_INODE_LRU_ERROR if allocation error occured when validating the
* entry
*
*/
cache_entry_t *cache_inode_lookup_sw(cache_entry_t * pentry_parent,
fsal_name_t * pname,
cache_inode_policy_t policy,
fsal_attrib_list_t * pattr,
hash_table_t * ht,
cache_inode_client_t * pclient,
fsal_op_context_t * pcontext,
cache_inode_status_t * pstatus,
int use_mutex)
{
cache_inode_dir_entry_t dirent_key[1], *dirent;
struct avltree_node *dirent_node;
cache_inode_dir_entry_t *new_dir_entry;
cache_entry_t *pentry = NULL;
fsal_status_t fsal_status;
#ifdef _USE_MFSL
mfsl_object_t object_handle;
#else
fsal_handle_t object_handle;
#endif
fsal_handle_t dir_handle;
fsal_attrib_list_t object_attributes;
cache_inode_create_arg_t create_arg;
cache_inode_file_type_t type;
cache_inode_status_t cache_status;
cache_inode_fsal_data_t new_entry_fsdata;
fsal_accessflags_t access_mask = 0;
memset(&create_arg, 0, sizeof(create_arg));
memset( (char *)&new_entry_fsdata, 0, sizeof( new_entry_fsdata ) ) ;
/* Set the return default to CACHE_INODE_SUCCESS */
*pstatus = CACHE_INODE_SUCCESS;
/* stats */
(pclient->stat.nb_call_total)++;
(pclient->stat.func_stats.nb_call[CACHE_INODE_LOOKUP])++;
/* We should not renew entries when !use_mutex (because unless we
* make the flag explicit (shared vs. exclusive), we don't know
* whether a mutating operation is safe--and, the caller should have
* already renewed the entry */
if(use_mutex == TRUE) {
P_w(&pentry_parent->lock);
cache_status = cache_inode_renew_entry(pentry_parent, pattr, ht,
pclient, pcontext, pstatus);
if(cache_status != CACHE_INODE_SUCCESS)
{
V_w(&pentry_parent->lock);
inc_func_err_retryable(pclient, CACHE_INODE_GETATTR);
LogDebug(COMPONENT_CACHE_INODE,
"cache_inode_lookup: returning %d(%s) from cache_inode_renew_entry",
*pstatus, cache_inode_err_str(*pstatus));
return NULL;
}
/* RW Lock goes for writer to reader */
rw_lock_downgrade(&pentry_parent->lock);
}
if(pentry_parent->internal_md.type != DIRECTORY)
{
/* Parent is no directory base, return NULL */
*pstatus = CACHE_INODE_NOT_A_DIRECTORY;
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_LOOKUP])++;
if(use_mutex == TRUE)
V_r(&pentry_parent->lock);
return NULL;
}
/* if name is ".", use the input value */
if(!FSAL_namecmp(pname, (fsal_name_t *) & FSAL_DOT))
{
pentry = pentry_parent;
}
else if(!FSAL_namecmp(pname, (fsal_name_t *) & FSAL_DOT_DOT))
{
/* Directory do only have exactly one parent. This a limitation in all FS,
* which implies that hard link are forbidden on directories (so that
* they exists only in one dir). Because of this, the parent list is
* always limited to one element for a dir. Clients SHOULD never
* 'lookup( .. )' in something that is no dir. */
pentry =
cache_inode_lookupp_no_mutex(pentry_parent, ht, pclient, pcontext,
pstatus);
}
else
{
/* This is a "regular lookup" (not on "." or "..") */
/* Check is user (as specified by the credentials) is authorized to
* lookup the directory or not */
access_mask = FSAL_MODE_MASK_SET(FSAL_X_OK) |
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_LIST_DIR);
if(cache_inode_access_no_mutex(pentry_parent,
access_mask,
ht,
pclient,
pcontext,
pstatus) != CACHE_INODE_SUCCESS)
{
if(use_mutex == TRUE)
V_r(&pentry_parent->lock);
(pclient->stat.func_stats.nb_err_retryable[CACHE_INODE_GETATTR])++;
return NULL;
}
/* We first try avltree_lookup by name. If that fails, we dispatch to
* the fsal. */
FSAL_namecpy(&dirent_key->name, pname);
dirent_node = avltree_lookup(&dirent_key->node_n,
&pentry_parent->object.dir.dentries);
if (dirent_node) {
dirent = avltree_container_of(dirent_node, cache_inode_dir_entry_t,
node_n);
pentry = dirent->pentry;
}
if(pentry == NULL)
{
LogDebug(COMPONENT_CACHE_INODE, "Cache Miss detected");
dir_handle = pentry_parent->handle;
object_attributes.asked_attributes = pclient->attrmask;
#ifdef _USE_MFSL
#ifdef _USE_MFSL_ASYNC
if(!mfsl_async_is_object_asynchronous(&pentry_parent->mobject))
{
/* If the parent is asynchronous, rely on the content of the cache
* inode parent entry.
*
* /!\ If the fs behind the FSAL is touched in a non-nfs way,
* there will be huge incoherencies.
*/
#endif /* _USE_MFSL_ASYNC */
fsal_status = MFSL_lookup(&pentry_parent->mobject,
pname,
pcontext,
&pclient->mfsl_context,
&object_handle, &object_attributes, NULL);
#ifdef _USE_MFSL_ASYNC
}
else
{
LogMidDebug(COMPONENT_CACHE_INODE,
"cache_inode_lookup chose to bypass FSAL and trusted his cache for name=%s",
pname->name);
fsal_status.major = ERR_FSAL_NOENT;
fsal_status.minor = ENOENT;
}
#endif /* _USE_MFSL_ASYNC */
#else
fsal_status =
FSAL_lookup(&dir_handle, pname, pcontext, &object_handle,
&object_attributes);
#endif /* _USE_MFSL */
if(FSAL_IS_ERROR(fsal_status))
{
*pstatus = cache_inode_error_convert(fsal_status);
if(use_mutex == TRUE)
V_r(&pentry_parent->lock);
/* Stale File Handle to be detected and managed */
if(fsal_status.major == ERR_FSAL_STALE)
{
cache_inode_status_t kill_status;
LogEvent(COMPONENT_CACHE_INODE,
"cache_inode_lookup: Stale FSAL File Handle detected for pentry = %p, fsal_status=(%u,%u)",
pentry_parent, fsal_status.major, fsal_status.minor);
if(cache_inode_kill_entry(pentry_parent, NO_LOCK, ht, pclient, &kill_status) !=
CACHE_INODE_SUCCESS)
LogCrit(COMPONENT_CACHE_INODE,
"cache_inode_pentry_parent: Could not kill entry %p, status = %u",
pentry_parent, kill_status);
*pstatus = CACHE_INODE_FSAL_ESTALE;
}
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_LOOKUP])++;
return NULL;
}
type = cache_inode_fsal_type_convert(object_attributes.type);
/* If entry is a symlink, this value for be cached */
if(type == SYMBOLIC_LINK)
{
if( CACHE_INODE_KEEP_CONTENT( policy ) )
#ifdef _USE_MFSL
{
fsal_status =
MFSL_readlink(&object_handle, pcontext, &pclient->mfsl_context,
&create_arg.link_content, &object_attributes, NULL);
}
#else
{
fsal_status =
FSAL_readlink(&object_handle, pcontext, &create_arg.link_content,
&object_attributes);
}
else
{
fsal_status.major = ERR_FSAL_NO_ERROR ;
fsal_status.minor = 0 ;
}
#endif
if(FSAL_IS_ERROR(fsal_status))
{
*pstatus = cache_inode_error_convert(fsal_status);
if(use_mutex == TRUE)
V_r(&pentry_parent->lock);
/* Stale File Handle to be detected and managed */
if(fsal_status.major == ERR_FSAL_STALE)
{
cache_inode_status_t kill_status;
LogEvent(COMPONENT_CACHE_INODE,
"cache_inode_lookup: Stale FSAL File Handle detected for pentry = %p, fsal_status=(%u,%u)",
pentry_parent, fsal_status.major, fsal_status.minor);
if(cache_inode_kill_entry(pentry_parent, NO_LOCK, ht, pclient, &kill_status)
!= CACHE_INODE_SUCCESS)
LogCrit(COMPONENT_CACHE_INODE,
"cache_inode_pentry_parent: Could not kill entry %p, status = %u",
pentry_parent, kill_status);
*pstatus = CACHE_INODE_FSAL_ESTALE;
}
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_LOOKUP])++;
return NULL;
}
}
/* Allocation of a new entry in the cache */
#ifdef _USE_MFSL
new_entry_fsdata.handle = object_handle.handle;
#else
new_entry_fsdata.handle = object_handle;
#endif
new_entry_fsdata.cookie = 0;
if((pentry = cache_inode_new_entry( &new_entry_fsdata,
&object_attributes,
type,
policy,
&create_arg,
NULL,
ht,
pclient,
pcontext,
FALSE, /* This is a population and not a creation */
pstatus ) ) == NULL )
{
if(use_mutex == TRUE)
V_r(&pentry_parent->lock);
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_LOOKUP])++;
return NULL;
}
/* Entry was found in the FSAL, add this entry to the parent
* directory */
cache_status = cache_inode_add_cached_dirent(pentry_parent,
pname,
pentry,
ht,
&new_dir_entry,
pclient,
pcontext,
pstatus);
if(cache_status != CACHE_INODE_SUCCESS
&& cache_status != CACHE_INODE_ENTRY_EXISTS)
{
if(use_mutex == TRUE)
V_r(&pentry_parent->lock);
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_LOOKUP])++;
return NULL;
}
} /* cached lookup fail (try fsal) */
/**
*
* @brief Checks permissions on an entry for setattrs
*
* This function acquires the attribute lock on the supplied cache
* entry then checks if the supplied credentials are sufficient to
* perform the required setattrs.
*
* @param[in] entry The object to be checked
* @param[in] attr Attributes to set/result of set
*
* @return CACHE_INODE_SUCCESS if operation is a success
*/
cache_inode_status_t
cache_inode_check_setattr_perms(cache_entry_t *entry,
struct attrlist *attr,
bool is_open_write)
{
cache_inode_status_t status = CACHE_INODE_SUCCESS;
fsal_accessflags_t access_check = 0;
bool not_owner;
char *note = "";
const struct user_cred *creds = op_ctx->creds;
if (isDebug(COMPONENT_CACHE_INODE) || isDebug(COMPONENT_NFS_V4_ACL)) {
char *setattr_size = "";
char *setattr_owner = "";
char *setattr_owner_group = "";
char *setattr_mode = "";
char *setattr_acl = "";
char *setattr_mtime = "";
char *setattr_atime = "";
if (FSAL_TEST_MASK(attr->mask, ATTR_SIZE))
setattr_size = " SIZE";
if (FSAL_TEST_MASK(attr->mask, ATTR_OWNER))
setattr_owner = " OWNER";
if (FSAL_TEST_MASK(attr->mask, ATTR_GROUP))
setattr_owner_group = " GROUP";
if (FSAL_TEST_MASK(attr->mask, ATTR_MODE))
setattr_mode = " MODE";
if (FSAL_TEST_MASK(attr->mask, ATTR_ACL))
setattr_acl = " ACL";
if (FSAL_TEST_MASK(attr->mask, ATTR_ATIME))
setattr_atime = " ATIME";
else if (FSAL_TEST_MASK(attr->mask, ATTR_ATIME_SERVER))
setattr_atime = " ATIME_SERVER";
if (FSAL_TEST_MASK(attr->mask, ATTR_MTIME))
setattr_mtime = " MTIME";
else if (FSAL_TEST_MASK(attr->mask, ATTR_MTIME_SERVER))
setattr_mtime = " MTIME_SERVER";
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"SETATTR %s%s%s%s%s%s%s", setattr_size,
setattr_owner, setattr_owner_group, setattr_mode,
setattr_acl, setattr_mtime, setattr_atime);
}
/* Shortcut, if current user is root, then we can just bail out with
* success. */
if (creds->caller_uid == 0) {
note = " (Ok for root user)";
goto out;
}
not_owner = (creds->caller_uid != entry->obj_handle->attributes.owner);
/* Only ownership change need to be checked for owner */
if (FSAL_TEST_MASK(attr->mask, ATTR_OWNER)) {
/* non-root is only allowed to "take ownership of file" */
if (attr->owner != creds->caller_uid) {
status = CACHE_INODE_FSAL_EPERM;
note = " (new OWNER was not user)";
goto out;
}
/* Owner of file will always be able to "change" the owner to
* himself. */
if (not_owner) {
access_check |=
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_OWNER);
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Change OWNER requires FSAL_ACE_PERM_WRITE_OWNER");
}
}
if (FSAL_TEST_MASK(attr->mask, ATTR_GROUP)) {
/* non-root is only allowed to change group_owner to a group
* user is a member of. */
int not_in_group = not_in_group_list(attr->group);
if (not_in_group) {
status = CACHE_INODE_FSAL_EPERM;
note = " (user is not member of new GROUP)";
goto out;
}
/* Owner is always allowed to change the group_owner of a file
* to a group they are a member of.
*/
if (not_owner) {
access_check |=
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_OWNER);
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Change GROUP requires FSAL_ACE_PERM_WRITE_OWNER");
}
}
/* Any attribute after this is always changeable by the owner.
* And the above attributes have already been validated as a valid
* change for the file owner to make. Note that the owner may be
* setting ATTR_OWNER but at this point it MUST be to himself, and
* thus is no-op and does not need FSAL_ACE_PERM_WRITE_OWNER.
*/
if (!not_owner) {
note = " (Ok for owner)";
goto out;
}
if (FSAL_TEST_MASK(attr->mask, ATTR_MODE)
|| FSAL_TEST_MASK(attr->mask, ATTR_ACL)) {
/* Changing mode or ACL requires ACE4_WRITE_ACL */
access_check |= FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_ACL);
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Change MODE or ACL requires FSAL_ACE_PERM_WRITE_ACL");
}
if (FSAL_TEST_MASK(attr->mask, ATTR_SIZE) && !is_open_write) {
/* Changing size requires owner or write permission */
/** @todo: does FSAL_ACE_PERM_APPEND_DATA allow enlarging the file? */
access_check |= FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_DATA);
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Change SIZE requires FSAL_ACE_PERM_WRITE_DATA");
}
/* Check if just setting atime and mtime to "now" */
if ((FSAL_TEST_MASK(attr->mask, ATTR_MTIME_SERVER)
|| FSAL_TEST_MASK(attr->mask, ATTR_ATIME_SERVER))
&& !FSAL_TEST_MASK(attr->mask, ATTR_MTIME)
&& !FSAL_TEST_MASK(attr->mask, ATTR_ATIME)) {
/* If either atime and/or mtime are set to "now" then need only
* have write permission.
*
* Technically, client should not send atime updates, but if
* they really do, we'll let them to make the perm check a bit
* simpler. */
access_check |= FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_DATA);
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Change ATIME and MTIME to NOW requires FSAL_ACE_PERM_WRITE_DATA");
} else if (FSAL_TEST_MASK(attr->mask, ATTR_MTIME_SERVER)
|| FSAL_TEST_MASK(attr->mask, ATTR_ATIME_SERVER)
|| FSAL_TEST_MASK(attr->mask, ATTR_MTIME)
|| FSAL_TEST_MASK(attr->mask, ATTR_ATIME)) {
/* Any other changes to atime or mtime require owner, root, or
* ACES4_WRITE_ATTRIBUTES.
*
* NOTE: we explicity do NOT check for update of atime only to
* "now". Section 10.6 of both RFC 3530 and RFC 5661 document
* the reasons clients should not do atime updates.
*/
access_check |= FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_ATTR);
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Change ATIME and/or MTIME requires FSAL_ACE_PERM_WRITE_ATTR");
}
if (isDebug(COMPONENT_CACHE_INODE) || isDebug(COMPONENT_NFS_V4_ACL)) {
char *need_write_owner = "";
char *need_write_acl = "";
char *need_write_data = "";
char *need_write_attr = "";
if (access_check & FSAL_ACE_PERM_WRITE_OWNER)
need_write_owner = " WRITE_OWNER";
if (access_check & FSAL_ACE_PERM_WRITE_ACL)
need_write_acl = " WRITE_ACL";
if (access_check & FSAL_ACE_PERM_WRITE_DATA)
need_write_data = " WRITE_DATA";
if (access_check & FSAL_ACE_PERM_WRITE_ATTR)
need_write_attr = " WRITE_ATTR";
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Requires %s%s%s%s", need_write_owner,
need_write_acl, need_write_data, need_write_attr);
}
if (entry->obj_handle->attributes.acl) {
status =
cache_inode_access_no_mutex(entry, access_check);
note = " (checked ACL)";
goto out;
}
if (access_check != FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_WRITE_DATA)) {
/* Without an ACL, this user is not allowed some operation */
status = CACHE_INODE_FSAL_EPERM;
note = " (no ACL to check)";
goto out;
}
status = cache_inode_access_no_mutex(entry, FSAL_W_OK);
note = " (checked mode)";
out:
LogDebugCIA(COMPONENT_CACHE_INODE, COMPONENT_NFS_V4_ACL,
"Access check returned %s%s", cache_inode_err_str(status),
note);
return status;
}
/**
*
* cache_inode_readdir: Reads a directory.
*
* Looks up for a name in a directory indicated by a cached entry. The
* directory should have been cached before.
*
* NEW: pending new (C-language) callback based dirent unpacking into caller
* structures, we eliminate copies by returning dir entries by pointer. To
* permit this, we introduce lock donation. If new int pointer argument
* unlock is 1 on return, the calling thread holds pentry read-locked and
* must release this lock after dirent processing.
*
* This is the only function in the cache_inode_readdir.c file that manages MT
* safety on a directory cache entry.
*
* @param pentry [IN] entry for the parent directory to be read.
* @param cookie [IN] cookie for the readdir operation (basically the offset).
* @param nbwanted [IN] Maximum number of directory entries wanted.
* @param peod_met [OUT] A flag to know if end of directory was met during this call.
* @param dirent_array [OUT] the resulting array of found directory entries.
* @param ht [IN] hash table used for the cache, unused in this call.
* @param unlock [OUT] the caller shall release read-lock on pentry when done
* @param pclient [INOUT] ressource allocated by the client for the nfs management.
* @param pcontext [IN] FSAL credentials
* @param pstatus [OUT] returned status.
*
* @return CACHE_INODE_SUCCESS if operation is a success \n
* @return CACHE_INODE_BAD_TYPE if entry is not related to a directory\n
* @return CACHE_INODE_LRU_ERROR if allocation error occured when validating the entry
*
*/
cache_inode_status_t cache_inode_readdir(cache_entry_t * dir_pentry,
cache_inode_policy_t policy,
uint64_t cookie,
unsigned int nbwanted,
unsigned int *pnbfound,
uint64_t *pend_cookie,
cache_inode_endofdir_t *peod_met,
cache_inode_dir_entry_t **dirent_array,
hash_table_t *ht,
int *unlock,
cache_inode_client_t *pclient,
fsal_op_context_t *pcontext,
cache_inode_status_t *pstatus)
{
cache_inode_dir_entry_t dirent_key[1], *dirent;
struct avltree_node *dirent_node;
fsal_accessflags_t access_mask = 0;
uint64_t inoff = cookie;
int i = 0;
/* Guide to parameters:
* the first cookie is parameter 'cookie'
* number of entries queried is set by parameter 'nbwanted'
* number of found entries before eod is return is '*pnbfound'
* '*peod_met' is set if end of directory is encountered */
/* Set the return default to CACHE_INODE_SUCCESS */
*pstatus = CACHE_INODE_SUCCESS;
dirent = NULL;
/* Set initial value of unlock */
*unlock = FALSE;
/* end cookie initial value is the begin cookie */
LogFullDebug(COMPONENT_NFS_READDIR,
"--> Cache_inode_readdir: setting pend_cookie to cookie=%"
PRIu64,
cookie);
*pend_cookie = cookie;
/* stats */
pclient->stat.nb_call_total++;
(pclient->stat.func_stats.nb_call[CACHE_INODE_READDIR])++;
LogFullDebug(COMPONENT_NFS_READDIR,
"--> Cache_inode_readdir: parameters are cookie=%"PRIu64
"nbwanted=%u",
cookie, nbwanted);
/* Sanity check */
if(nbwanted == 0)
{
/* Asking for nothing is not a crime !!!!!
* build a 'dummy' return in this case */
*pstatus = CACHE_INODE_SUCCESS;
*pnbfound = 0;
*peod_met = TO_BE_CONTINUED;
/* stats */
(pclient->stat.func_stats.nb_success[CACHE_INODE_READDIR])++;
return *pstatus;
}
/* Force dir content invalidation if policy enforced no name cache */
if( !CACHE_INODE_KEEP_CONTENT( dir_pentry->policy ) )
return cache_inode_readdir_nonamecache( dir_pentry,
policy,
cookie,
nbwanted,
pnbfound,
pend_cookie,
peod_met,
dirent_array,
ht,
unlock,
pclient,
pcontext,
pstatus ) ;
P_w(&dir_pentry->lock);
/* Renew the entry (to avoid having it being garbagged */
if(cache_inode_renew_entry(dir_pentry, NULL, ht, pclient, pcontext,
pstatus) != CACHE_INODE_SUCCESS)
{
(pclient->stat.func_stats.nb_err_retryable[CACHE_INODE_GETATTR])++;
V_w(&dir_pentry->lock);
return *pstatus;
}
/* readdir can be done only with a directory */
if(dir_pentry->internal_md.type != DIRECTORY)
{
V_w(&dir_pentry->lock);
*pstatus = CACHE_INODE_BAD_TYPE;
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_READDIR])++;
return *pstatus;
}
/* Check is user (as specified by the credentials) is authorized to read
* the directory or not */
access_mask = FSAL_MODE_MASK_SET(FSAL_R_OK) |
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_LIST_DIR);
if(cache_inode_access_no_mutex(dir_pentry,
access_mask,
ht, pclient,
pcontext,
pstatus) != CACHE_INODE_SUCCESS)
{
V_w(&dir_pentry->lock);
(pclient->stat.func_stats.nb_err_retryable[CACHE_INODE_READDIR])++;
return *pstatus;
}
/* Is the directory fully cached (this is done if a readdir call is done on the directory) */
if(dir_pentry->object.dir.has_been_readdir != CACHE_INODE_YES)
{
/* populate the cache */
if(cache_inode_readdir_populate(dir_pentry,
policy,
ht,
pclient,
pcontext, pstatus) != CACHE_INODE_SUCCESS)
{
/* stats */
(pclient->stat.func_stats.nb_err_unrecover[CACHE_INODE_READDIR])++;
V_w(&dir_pentry->lock);
return *pstatus;
}
}
/* deal with dentry cache invalidates */
revalidate_cookie_cache(dir_pentry, pclient);
/* Downgrade Writer lock to a reader one. */
rw_lock_downgrade(&dir_pentry->lock);
/* deal with initial cookie value:
* 1. cookie is invalid (-should- be checked by caller)
* 2. cookie is 0 (first cookie) -- ok
* 3. cookie is > than highest dirent position (error)
* 4. cookie <= highest dirent position but > highest cached cookie
* (currently equivalent to #2, because we pre-populate the cookie avl)
* 5. cookie is in cached range -- ok */
if (cookie > 0) {
if (cookie < 3) {
*pstatus = CACHE_INODE_BAD_COOKIE;
V_r(&dir_pentry->lock);
return *pstatus;
}
if ((inoff-3) > avltree_size(&dir_pentry->object.dir.dentries)) {
LogCrit(COMPONENT_NFS_V4, "Bad initial cookie %"PRIu64,
inoff);
*pstatus = CACHE_INODE_BAD_COOKIE;
V_r(&dir_pentry->lock);
return *pstatus;
}
/* we assert this can now succeed */
dirent_key->cookie = inoff;
dirent_node = avltree_lookup(&dirent_key->node_c,
&dir_pentry->object.dir.cookies);
if (! dirent_node) {
LogCrit(COMPONENT_NFS_READDIR,
"%s: seek to cookie=%"PRIu64" fail",
__func__,
inoff);
*pstatus = CACHE_INODE_NOT_FOUND;
V_r(&dir_pentry->lock);
return *pstatus;
}
/* switch avls */
dirent = avltree_container_of(dirent_node,
cache_inode_dir_entry_t,
node_c);
dirent_node = &dirent->node_n;
/* client wants the cookie -after- the last we sent, and
* the Linux 3.0 and 3.1.0-rc7 clients misbehave if we
* resend the last one */
dirent_node = avltree_next(dirent_node);
} else {
/* initial readdir */
dirent_node = avltree_first(&dir_pentry->object.dir.dentries);
}
LogFullDebug(COMPONENT_NFS_READDIR,
"About to readdir in cache_inode_readdir: pentry=%p "
"cookie=%"PRIu64,
dir_pentry,
cookie);
/* Now satisfy the request from the cached readdir--stop when either
* the requested sequence or dirent sequence is exhausted */
*pnbfound = 0;
*peod_met = TO_BE_CONTINUED;
for(i = 0; i < nbwanted; ++i)
{
if (!dirent_node)
break;
dirent = avltree_container_of(dirent_node,
cache_inode_dir_entry_t,
node_n);
dirent_array[i] = dirent;
(*pnbfound)++;
dirent_node = avltree_next(dirent_node);
}
if (*pnbfound > 0)
{
if (!dirent)
{
LogCrit(COMPONENT_CACHE_INODE, "cache_inode_readdir: "
"UNEXPECTED CASE: dirent is NULL whereas nbfound>0");
*pstatus = CACHE_INODE_INCONSISTENT_ENTRY;
return CACHE_INODE_INCONSISTENT_ENTRY;
}
*pend_cookie = dirent->cookie;
}
if (! dirent_node)
*peod_met = END_OF_DIR;
*pstatus = cache_inode_valid(dir_pentry, CACHE_INODE_OP_GET, pclient);
/* stats */
if(*pstatus != CACHE_INODE_SUCCESS) {
(pclient->stat.func_stats.nb_err_retryable[CACHE_INODE_READDIR])++;
V_r(&dir_pentry->lock);
}
else {
(pclient->stat.func_stats.nb_success[CACHE_INODE_READDIR])++;
*unlock = TRUE;
}
return *pstatus;
} /* cache_inode_readdir */
cache_inode_status_t
cache_inode_readdir(cache_entry_t *directory,
uint64_t cookie, unsigned int *nbfound,
bool *eod_met,
attrmask_t attrmask,
cache_inode_getattr_cb_t cb,
void *opaque)
{
/* The entry being examined */
cache_inode_dir_entry_t *dirent = NULL;
/* The node in the tree being traversed */
struct avltree_node *dirent_node;
/* The access mask corresponding to permission to list directory
entries */
fsal_accessflags_t access_mask =
(FSAL_MODE_MASK_SET(FSAL_R_OK) |
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_LIST_DIR));
fsal_accessflags_t access_mask_attr =
(FSAL_MODE_MASK_SET(FSAL_R_OK) | FSAL_MODE_MASK_SET(FSAL_X_OK) |
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_LIST_DIR) |
FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_EXECUTE));
cache_inode_status_t status = CACHE_INODE_SUCCESS;
cache_inode_status_t attr_status;
struct cache_inode_readdir_cb_parms cb_parms = { opaque, NULL,
true, 0, true };
bool retry_stale = true;
LogFullDebug(COMPONENT_NFS_READDIR,
"Enter....");
/* readdir can be done only with a directory */
if (directory->type != DIRECTORY) {
status = CACHE_INODE_NOT_A_DIRECTORY;
/* no lock acquired so far, just return status */
LogFullDebug(COMPONENT_NFS_READDIR,
"Not a directory");
return status;
}
/* cache_inode_lock_trust_attrs can return an error, and no lock will
* be acquired */
status = cache_inode_lock_trust_attrs(directory, false);
if (status != CACHE_INODE_SUCCESS) {
LogDebug(COMPONENT_NFS_READDIR,
"cache_inode_lock_trust_attrs status=%s",
cache_inode_err_str(status));
return status;
}
/* Adjust access mask if ACL is asked for.
* NOTE: We intentionally do NOT check ACE4_READ_ATTR.
*/
if ((attrmask & ATTR_ACL) != 0) {
access_mask |= FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_READ_ACL);
access_mask_attr |= FSAL_ACE4_MASK_SET(FSAL_ACE_PERM_READ_ACL);
}
/* Check if user (as specified by the credentials) is authorized to read
* the directory or not */
status = cache_inode_access_no_mutex(directory, access_mask);
if (status != CACHE_INODE_SUCCESS) {
LogFullDebug(COMPONENT_NFS_READDIR,
"permission check for directory status=%s",
cache_inode_err_str(status));
PTHREAD_RWLOCK_unlock(&directory->attr_lock);
return status;
}
if (attrmask != 0) {
/* Check for access permission to get attributes */
attr_status = cache_inode_access_no_mutex(directory,
access_mask_attr);
if (attr_status != CACHE_INODE_SUCCESS) {
LogFullDebug(COMPONENT_NFS_READDIR,
"permission check for attributes "
"status=%s",
cache_inode_err_str(attr_status));
}
} else
/* No attributes requested, we don't need permission */
attr_status = CACHE_INODE_SUCCESS;
PTHREAD_RWLOCK_rdlock(&directory->content_lock);
PTHREAD_RWLOCK_unlock(&directory->attr_lock);
if (!
((directory->flags & CACHE_INODE_TRUST_CONTENT)
&& (directory->flags & CACHE_INODE_DIR_POPULATED))) {
PTHREAD_RWLOCK_unlock(&directory->content_lock);
PTHREAD_RWLOCK_wrlock(&directory->content_lock);
status = cache_inode_readdir_populate(directory);
if (status != CACHE_INODE_SUCCESS) {
LogFullDebug(COMPONENT_NFS_READDIR,
"cache_inode_readdir_populate status=%s",
cache_inode_err_str(status));
goto unlock_dir;
}
}
/* deal with initial cookie value:
* 1. cookie is invalid (-should- be checked by caller)
* 2. cookie is 0 (first cookie) -- ok
* 3. cookie is > than highest dirent position (error)
* 4. cookie <= highest dirent position but > highest cached cookie
* (currently equivalent to #2, because we pre-populate the cookie
* avl)
* 5. cookie is in cached range -- ok */
if (cookie > 0) {
/* N.B., cache_inode_avl_qp_insert_s ensures k > 2 */
if (cookie < 3) {
status = CACHE_INODE_BAD_COOKIE;
LogFullDebug(COMPONENT_NFS_READDIR,
"Bad cookie");
goto unlock_dir;
}
/* we assert this can now succeed */
dirent =
cache_inode_avl_lookup_k(directory, cookie,
CACHE_INODE_FLAG_NEXT_ACTIVE);
if (!dirent) {
/* Linux (3.4, etc) has been observed to send readdir
* at the offset of the last entry's cookie, and
* returns no dirents to userland if that readdir
* notfound or badcookie. */
if (cache_inode_avl_lookup_k
(directory, cookie, CACHE_INODE_FLAG_NONE)) {
/* yup, it was the last entry */
LogFullDebug(COMPONENT_NFS_READDIR,
"EOD because empty result");
*eod_met = true;
goto unlock_dir;
}
LogFullDebug(COMPONENT_NFS_READDIR,
"seek to cookie=%" PRIu64 " fail",
cookie);
status = CACHE_INODE_BAD_COOKIE;
goto unlock_dir;
}
/* dirent is the NEXT entry to return, since we sent
* CACHE_INODE_FLAG_NEXT_ACTIVE */
dirent_node = &dirent->node_hk;
} else {
/* initial readdir */
dirent_node = avltree_first(&directory->object.dir.avl.t);
}
LogFullDebug(COMPONENT_NFS_READDIR,
"About to readdir in cache_inode_readdir: directory=%p "
"cookie=%" PRIu64 " collisions %d", directory, cookie,
directory->object.dir.avl.collisions);
/* Now satisfy the request from the cached readdir--stop when either
* the requested sequence or dirent sequence is exhausted */
*nbfound = 0;
*eod_met = false;
for (; cb_parms.in_result && dirent_node;
dirent_node = avltree_next(dirent_node)) {
cache_entry_t *entry = NULL;
cache_inode_status_t tmp_status = 0;
dirent =
avltree_container_of(dirent_node, cache_inode_dir_entry_t,
node_hk);
estale_retry:
LogFullDebug(COMPONENT_NFS_READDIR,
"Lookup direct %s",
dirent->name);
entry =
cache_inode_get_keyed(&dirent->ckey,
CIG_KEYED_FLAG_NONE, &tmp_status);
if (!entry) {
LogFullDebug(COMPONENT_NFS_READDIR,
"Lookup returned %s",
cache_inode_err_str(tmp_status));
if (retry_stale
&& tmp_status == CACHE_INODE_ESTALE) {
LogDebug(COMPONENT_NFS_READDIR,
"cache_inode_get_keyed returned %s "
"for %s - retrying entry",
cache_inode_err_str(tmp_status),
dirent->name);
retry_stale = false; /* only one retry per
* dirent */
goto estale_retry;
}
if (tmp_status == CACHE_INODE_NOT_FOUND
|| tmp_status == CACHE_INODE_ESTALE) {
/* Directory changed out from under us.
Invalidate it, skip the name, and keep
going. */
atomic_clear_uint32_t_bits(
&directory->flags,
CACHE_INODE_TRUST_CONTENT);
LogDebug(COMPONENT_NFS_READDIR,
"cache_inode_get_keyed returned %s "
"for %s - skipping entry",
cache_inode_err_str(tmp_status),
dirent->name);
continue;
} else {
/* Something is more seriously wrong,
probably an inconsistency. */
status = tmp_status;
LogCrit(COMPONENT_NFS_READDIR,
"cache_inode_get_keyed returned %s "
"for %s - bailing out",
cache_inode_err_str(status),
dirent->name);
goto unlock_dir;
}
}
LogFullDebug(COMPONENT_NFS_READDIR,
"cache_inode_readdir: dirent=%p name=%s "
"cookie=%" PRIu64 " (probes %d)", dirent,
dirent->name, dirent->hk.k, dirent->hk.p);
cb_parms.name = dirent->name;
cb_parms.attr_allowed = attr_status == CACHE_INODE_SUCCESS;
cb_parms.cookie = dirent->hk.k;
tmp_status =
cache_inode_getattr(entry, &cb_parms, cb, CB_ORIGINAL);
if (tmp_status != CACHE_INODE_SUCCESS) {
cache_inode_lru_unref(entry, LRU_FLAG_NONE);
if (tmp_status == CACHE_INODE_ESTALE) {
if (retry_stale) {
LogDebug(COMPONENT_NFS_READDIR,
"cache_inode_getattr returned "
"%s for %s - retrying entry",
cache_inode_err_str
(tmp_status), dirent->name);
retry_stale = false; /* only one retry
* per dirent */
goto estale_retry;
}
/* Directory changed out from under us.
Invalidate it, skip the name, and keep
going. */
atomic_clear_uint32_t_bits(
&directory->flags,
CACHE_INODE_TRUST_CONTENT);
LogDebug(COMPONENT_NFS_READDIR,
"cache_inode_lock_trust_attrs "
"returned %s for %s - skipping entry",
cache_inode_err_str(tmp_status),
dirent->name);
continue;
}
status = tmp_status;
LogCrit(COMPONENT_NFS_READDIR,
"cache_inode_lock_trust_attrs returned %s for "
"%s - bailing out",
cache_inode_err_str(status), dirent->name);
goto unlock_dir;
}
(*nbfound)++;
cache_inode_lru_unref(entry, LRU_FLAG_NONE);
if (!cb_parms.in_result) {
LogDebug(COMPONENT_NFS_READDIR,
"bailing out due to entry not in result");
break;
}
}
/* We have reached the last node and every node traversed was
added to the result */
LogDebug(COMPONENT_NFS_READDIR,
"dirent_node = %p, nbfound = %u, in_result = %s", dirent_node,
*nbfound, cb_parms.in_result ? "TRUE" : "FALSE");
if (!dirent_node && cb_parms.in_result)
*eod_met = true;
else
*eod_met = false;
unlock_dir:
PTHREAD_RWLOCK_unlock(&directory->content_lock);
return status;
} /* cache_inode_readdir */
