/*
* Copyright 2012 Xyratex Technology Limited
*
+ * Copyright (c) 2013, 2017, Intel Corporation.
+ *
* Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
*
*/
#include <linux/sched.h>
#include <linux/mm.h>
#include <obd_support.h>
-#include <lustre_lite.h>
#include <lustre_dlm.h>
-#include <lustre_ver.h>
#include "llite_internal.h"
/* If we ever have hundreds of extended attributes, we might want to consider
LASSERT(lli != NULL);
- CFS_INIT_LIST_HEAD(&lli->lli_xattrs);
- lli->lli_flags |= LLIF_XATTR_CACHE;
+ INIT_LIST_HEAD(&lli->lli_xattrs);
+ set_bit(LLIF_XATTR_CACHE, &lli->lli_flags);
}
/**
ENTRY;
if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
+ if (!strcmp(xattr_name, LL_XATTR_NAME_ENCRYPTION_CONTEXT))
+ /* it means enc ctx was already in cache,
+ * ignore error as it cannot be modified
+ */
+ RETURN(0);
+
CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
RETURN(-EPROTO);
}
}
/**
- * Check if the xattr cache is initialized (filled).
+ * Check if the xattr cache is initialized.
*
* \retval 0 @cache is not initialized
* \retval 1 @cache is initialized
*/
-int ll_xattr_cache_valid(struct ll_inode_info *lli)
+static int ll_xattr_cache_valid(struct ll_inode_info *lli)
{
- return !!(lli->lli_flags & LLIF_XATTR_CACHE);
+ return test_bit(LLIF_XATTR_CACHE, &lli->lli_flags);
+}
+
+/**
+ * Check if the xattr cache is filled.
+ *
+ * \retval 0 @cache is not filled
+ * \retval 1 @cache is filled
+ */
+static int ll_xattr_cache_filled(struct ll_inode_info *lli)
+{
+ return test_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
}
/**
while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
/* empty loop */ ;
- lli->lli_flags &= ~LLIF_XATTR_CACHE;
+
+ clear_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
+ clear_bit(LLIF_XATTR_CACHE, &lli->lli_flags);
RETURN(0);
}
}
/**
+ * ll_xattr_cache_empty - empty xattr cache for @ino
+ *
+ * Similar to ll_xattr_cache_destroy(), but preserves encryption context.
+ * So only LLIF_XATTR_CACHE_FILLED flag is cleared, but not LLIF_XATTR_CACHE.
+ */
+int ll_xattr_cache_empty(struct inode *inode)
+{
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct ll_xattr_entry *entry, *n;
+
+ ENTRY;
+
+ down_write(&lli->lli_xattrs_list_rwsem);
+ if (!ll_xattr_cache_valid(lli) ||
+ !ll_xattr_cache_filled(lli))
+ GOTO(out_empty, 0);
+
+ list_for_each_entry_safe(entry, n, &lli->lli_xattrs, xe_list) {
+ if (strcmp(entry->xe_name,
+ LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0)
+ continue;
+
+ CDEBUG(D_CACHE, "delete: %s\n", entry->xe_name);
+ list_del(&entry->xe_list);
+ OBD_FREE(entry->xe_name, entry->xe_namelen);
+ OBD_FREE(entry->xe_value, entry->xe_vallen);
+ OBD_SLAB_FREE_PTR(entry, xattr_kmem);
+ }
+ clear_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
+
+out_empty:
+ up_write(&lli->lli_xattrs_list_rwsem);
+ RETURN(0);
+}
+
+/**
* Match or enqueue a PR lock.
*
* Find or request an LDLM lock with xattr data.
* Since LDLM does not provide API for atomic match_or_enqueue,
* the function handles it with a separate enq lock.
- * If successful, the function exits with the list lock held.
+ * If successful, the function exits with a write lock held
+ * on lli_xattrs_list_rwsem.
*
* \retval 0 no error occured
* \retval -ENOMEM not enough memory
struct lookup_intent *oit,
struct ptlrpc_request **req)
{
- ldlm_mode_t mode;
+ enum ldlm_mode mode;
struct lustre_handle lockh = { 0 };
struct md_op_data *op_data;
struct ll_inode_info *lli = ll_i2info(inode);
- struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
- .ei_mode = it_to_lock_mode(oit),
- .ei_cb_bl = ll_md_blocking_ast,
- .ei_cb_cp = ldlm_completion_ast };
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct obd_export *exp = sbi->ll_md_exp;
int rc;
ENTRY;
mutex_lock(&lli->lli_xattrs_enq_lock);
- /* Try matching first. */
- mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0, LCK_PR);
- if (mode != 0) {
- /* fake oit in mdc_revalidate_lock() manner */
- oit->d.lustre.it_lock_handle = lockh.cookie;
- oit->d.lustre.it_lock_mode = mode;
- goto out;
+ /* inode may have been shrunk and recreated, so data is gone, match lock
+ * only when data exists. */
+ if (ll_xattr_cache_filled(lli)) {
+ /* Try matching first. */
+ mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0,
+ LCK_PR);
+ if (mode != 0) {
+ /* fake oit in mdc_revalidate_lock() manner */
+ oit->it_lock_handle = lockh.cookie;
+ oit->it_lock_mode = mode;
+ goto out;
+ }
}
/* Enqueue if the lock isn't cached locally. */
op_data->op_valid = OBD_MD_FLXATTR | OBD_MD_FLXATTRLS;
- rc = md_enqueue(exp, &einfo, oit, op_data, &lockh, NULL, 0, NULL, 0);
+ rc = md_intent_lock(exp, op_data, oit, req, &ll_md_blocking_ast, 0);
ll_finish_md_op_data(op_data);
+ *req = oit->it_request;
if (rc < 0) {
CDEBUG(D_CACHE, "md_intent_lock failed with %d for fid "DFID"\n",
RETURN(rc);
}
- *req = (struct ptlrpc_request *)oit->d.lustre.it_data;
out:
down_write(&lli->lli_xattrs_list_rwsem);
mutex_unlock(&lli->lli_xattrs_enq_lock);
/**
* Refill the xattr cache.
*
- * Fetch and cache the whole of xattrs for @inode, acquiring
- * a read or a write xattr lock depending on operation in @oit.
- * Intent is dropped on exit unless the operation is setxattr.
+ * Fetch and cache the whole of xattrs for @inode, thanks to the write lock
+ * on lli_xattrs_list_rwsem obtained from ll_xattr_find_get_lock().
+ * If successful, this write lock is kept.
*
* \retval 0 no error occured
* \retval -EPROTO network protocol error
* \retval -ENOMEM not enough memory for the cache
*/
-static int ll_xattr_cache_refill(struct inode *inode, struct lookup_intent *oit)
+static int ll_xattr_cache_refill(struct inode *inode)
{
+ struct lookup_intent oit = { .it_op = IT_GETXATTR };
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ptlrpc_request *req = NULL;
const char *xdata, *xval, *xtail, *xvtail;
ENTRY;
- rc = ll_xattr_find_get_lock(inode, oit, &req);
+ rc = ll_xattr_find_get_lock(inode, &oit, &req);
if (rc)
- GOTO(out_no_unlock, rc);
+ GOTO(err_req, rc);
/* Do we have the data at this point? */
- if (ll_xattr_cache_valid(lli)) {
+ if (ll_xattr_cache_filled(lli)) {
ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
- GOTO(out_maybe_drop, rc = 0);
+ ll_intent_drop_lock(&oit);
+ GOTO(err_req, rc = 0);
}
/* Matched but no cache? Cancelled on error by a parallel refill. */
if (unlikely(req == NULL)) {
CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
- GOTO(out_maybe_drop, rc = -EIO);
- }
-
- if (oit->d.lustre.it_status < 0) {
- CDEBUG(D_CACHE, "getxattr intent returned %d for fid "DFID"\n",
- oit->d.lustre.it_status, PFID(ll_inode2fid(inode)));
- rc = oit->d.lustre.it_status;
- /* xattr data is so large that we don't want to cache it */
- if (rc == -ERANGE)
- rc = -EAGAIN;
- GOTO(out_destroy, rc);
+ ll_intent_drop_lock(&oit);
+ GOTO(err_unlock, rc = -EAGAIN);
}
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
if (body == NULL) {
CERROR("no MDT BODY in the refill xattr reply\n");
- GOTO(out_destroy, rc = -EPROTO);
+ GOTO(err_cancel, rc = -EPROTO);
}
/* do not need swab xattr data */
xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
- body->eadatasize);
+ body->mbo_eadatasize);
xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
- body->aclsize);
+ body->mbo_aclsize);
xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
- body->max_mdsize * sizeof(__u32));
+ body->mbo_max_mdsize *
+ sizeof(__u32));
if (xdata == NULL || xval == NULL || xsizes == NULL) {
CERROR("wrong setxattr reply\n");
- GOTO(out_destroy, rc = -EPROTO);
+ GOTO(err_cancel, rc = -EPROTO);
}
- xtail = xdata + body->eadatasize;
- xvtail = xval + body->aclsize;
+ xtail = xdata + body->mbo_eadatasize;
+ xvtail = xval + body->mbo_aclsize;
CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);
- ll_xattr_cache_init(lli);
+ if (!ll_xattr_cache_valid(lli))
+ ll_xattr_cache_init(lli);
- for (i = 0; i < body->max_mdsize; i++) {
+ for (i = 0; i < body->mbo_max_mdsize; i++) {
CDEBUG(D_CACHE, "caching [%s]=%.*s\n", xdata, *xsizes, xval);
/* Perform consistency checks: attr names and vals in pill */
if (memchr(xdata, 0, xtail - xdata) == NULL) {
CDEBUG(D_CACHE, "not caching %s\n",
XATTR_NAME_ACL_ACCESS);
rc = 0;
+ } else if (!strcmp(xdata, "security.selinux")) {
+ /* Filter out security.selinux, it is cached in slab */
+ CDEBUG(D_CACHE, "not caching security.selinux\n");
+ rc = 0;
} else {
rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
*xsizes);
}
if (rc < 0) {
ll_xattr_cache_destroy_locked(lli);
- GOTO(out_destroy, rc);
+ GOTO(err_cancel, rc);
}
xdata += strlen(xdata) + 1;
xval += *xsizes;
if (xdata != xtail || xval != xvtail)
CERROR("a hole in xattr data\n");
+ else
+ set_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
- ll_set_lock_data(sbi->ll_md_exp, inode, oit, NULL);
-
- GOTO(out_maybe_drop, rc);
-out_maybe_drop:
+ ll_set_lock_data(sbi->ll_md_exp, inode, &oit, NULL);
+ ll_intent_drop_lock(&oit);
- ll_intent_drop_lock(oit);
-
- if (rc != 0)
- up_write(&lli->lli_xattrs_list_rwsem);
-out_no_unlock:
ptlrpc_req_finished(req);
+ RETURN(0);
- return rc;
-
-out_destroy:
- up_write(&lli->lli_xattrs_list_rwsem);
-
+err_cancel:
ldlm_lock_decref_and_cancel((struct lustre_handle *)
- &oit->d.lustre.it_lock_handle,
- oit->d.lustre.it_lock_mode);
+ &oit.it_lock_handle,
+ oit.it_lock_mode);
+err_unlock:
+ up_write(&lli->lli_xattrs_list_rwsem);
+err_req:
+ if (rc == -ERANGE)
+ rc = -EAGAIN;
- goto out_no_unlock;
+ ptlrpc_req_finished(req);
+ RETURN(rc);
}
/**
size_t size,
__u64 valid)
{
- struct lookup_intent oit = { .it_op = IT_GETXATTR };
struct ll_inode_info *lli = ll_i2info(inode);
int rc = 0;
LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));
down_read(&lli->lli_xattrs_list_rwsem);
- if (!ll_xattr_cache_valid(lli)) {
+ /* For performance reasons, we do not want to refill complete xattr
+ * cache if we are just interested in encryption context.
+ */
+ if ((valid & OBD_MD_FLXATTRLS ||
+ strcmp(name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) != 0) &&
+ !ll_xattr_cache_filled(lli)) {
up_read(&lli->lli_xattrs_list_rwsem);
- rc = ll_xattr_cache_refill(inode, &oit);
+ rc = ll_xattr_cache_refill(inode);
if (rc)
RETURN(rc);
+ /* Turn the write lock obtained in ll_xattr_cache_refill()
+ * into a read lock.
+ */
downgrade_write(&lli->lli_xattrs_list_rwsem);
} else {
ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
}
+ if (!ll_xattr_cache_valid(lli))
+ GOTO(out, rc = -ENODATA);
+
if (valid & OBD_MD_FLXATTR) {
struct ll_xattr_entry *xattr;
out:
up_read(&lli->lli_xattrs_list_rwsem);
- return rc;
+ RETURN(rc);
}
+/**
+ * Insert an xattr value into the cache.
+ *
+ * Add @name xattr with @buffer value and @size length for @inode.
+ * Init cache for @inode if necessary.
+ *
+ * \retval 0 success
+ * \retval < 0 from ll_xattr_cache_add(), except -EPROTO is ignored for
+ * LL_XATTR_NAME_ENCRYPTION_CONTEXT xattr
+ */
+int ll_xattr_cache_insert(struct inode *inode,
+ const char *name,
+ char *buffer,
+ size_t size)
+{
+ struct ll_inode_info *lli = ll_i2info(inode);
+ int rc;
+
+ ENTRY;
+
+ down_write(&lli->lli_xattrs_list_rwsem);
+ if (!ll_xattr_cache_valid(lli))
+ ll_xattr_cache_init(lli);
+ rc = ll_xattr_cache_add(&lli->lli_xattrs, name, buffer, size);
+ up_write(&lli->lli_xattrs_list_rwsem);
+ RETURN(rc);
+}