* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Return how many pages have been written.
*/
int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
- enum cl_fsync_mode mode)
+ enum cl_fsync_mode mode, int ignore_layout)
{
struct cl_env_nest nest;
struct lu_env *env;
io = ccc_env_thread_io(env);
io->ci_obj = cl_i2info(inode)->lli_clob;
- io->ci_ignore_layout = 1;
+ io->ci_ignore_layout = ignore_layout;
/* initialize parameters for sync */
fio = &io->u.ci_fsync;
struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
- CL_FSYNC_ALL);
+ CL_FSYNC_ALL, 0);
if (rc == 0 && err < 0)
rc = err;
if (rc < 0)
RETURN(result);
}
+/* Fetch layout from MDT with getxattr request, if it's not ready yet */
+static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
+
+{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ struct obd_capa *oc;
+ struct ptlrpc_request *req;
+ struct mdt_body *body;
+ void *lvbdata;
+ void *lmm;
+ int lmmsize;
+ int rc;
+ ENTRY;
+
+ if (lock->l_lvb_data != NULL)
+ RETURN(0);
+
+ /* if layout lock was granted right away, the layout is returned
+ * within DLM_LVB of dlm reply; otherwise if the lock was ever
+ * blocked and then granted via completion ast, we have to fetch
+ * layout here. Please note that we can't use the LVB buffer in
+ * completion AST because it doesn't have a large enough buffer */
+ oc = ll_mdscapa_get(inode);
+ rc = ll_get_max_mdsize(sbi, &lmmsize);
+ if (rc == 0)
+ rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
+ OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
+ lmmsize, 0, &req);
+ capa_put(oc);
+ if (rc < 0)
+ RETURN(rc);
+
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+ if (body == NULL || body->eadatasize > lmmsize)
+ GOTO(out, rc = -EPROTO);
+
+ lmmsize = body->eadatasize;
+ if (lmmsize == 0) /* empty layout */
+ GOTO(out, rc = 0);
+
+ lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
+ if (lmm == NULL)
+ GOTO(out, rc = -EFAULT);
+
+ OBD_ALLOC_LARGE(lvbdata, lmmsize);
+ if (lvbdata == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ memcpy(lvbdata, lmm, lmmsize);
+ lock_res_and_lock(lock);
+ if (lock->l_lvb_data == NULL) {
+ lock->l_lvb_data = lvbdata;
+ lock->l_lvb_len = lmmsize;
+ lvbdata = NULL;
+ }
+ unlock_res_and_lock(lock);
+
+ if (lvbdata != NULL)
+ OBD_FREE_LARGE(lvbdata, lmmsize);
+ EXIT;
+
+out:
+ ptlrpc_req_finished(req);
+ return rc;
+}
+
/**
* Apply the layout to the inode. Layout lock is held and will be released
* in this function.
struct cl_object_conf conf;
int rc = 0;
bool lvb_ready;
+ bool wait_layout = false;
ENTRY;
LASSERT(lustre_handle_is_used(lockh));
LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n",
inode, PFID(&lli->lli_fid), reconf);
+ /* in case this is a caching lock and reinstate with new inode */
+ md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
+
lock_res_and_lock(lock);
lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY);
unlock_res_and_lock(lock);
/* checking lvb_ready is racy but this is okay. The worst case is
* that multi processes may configure the file on the same time. */
if (lvb_ready || !reconf) {
- LDLM_LOCK_PUT(lock);
-
rc = -ENODATA;
if (lvb_ready) {
/* layout_gen must be valid if layout lock is not
*gen = lli->lli_layout_gen;
rc = 0;
}
- ldlm_lock_decref(lockh, mode);
- RETURN(rc);
+ GOTO(out, rc);
}
+ rc = ll_layout_fetch(inode, lock);
+ if (rc < 0)
+ GOTO(out, rc);
+
/* for layout lock, lmm is returned in lock's lvb.
* lvb_data is immutable if the lock is held so it's safe to access it
* without res lock. See the description in ldlm_lock_decref_internal()
PFID(&lli->lli_fid), rc);
}
}
- if (rc < 0) {
- LDLM_LOCK_PUT(lock);
- ldlm_lock_decref(lockh, mode);
- RETURN(rc);
- }
+ if (rc < 0)
+ GOTO(out, rc);
/* set layout to file. Unlikely this will fail as old layout was
* surely eliminated */
conf.coc_lock = lock;
conf.u.coc_md = &md;
rc = ll_layout_conf(inode, &conf);
- LDLM_LOCK_PUT(lock);
-
- ldlm_lock_decref(lockh, mode);
if (md.lsm != NULL)
obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
+ /* refresh layout failed, need to wait */
+ wait_layout = rc == -EBUSY;
+ EXIT;
+
+out:
+ LDLM_LOCK_PUT(lock);
+ ldlm_lock_decref(lockh, mode);
+
/* wait for IO to complete if it's still being used. */
- if (rc == -EBUSY) {
+ if (wait_layout) {
CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n",
ll_get_fsname(inode->i_sb, NULL, 0),
inode, PFID(&lli->lli_fid));
CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n",
PFID(&lli->lli_fid), rc);
}
-
RETURN(rc);
}
int rc;
ENTRY;
- *gen = LL_LAYOUT_GEN_NONE;
+ *gen = lli->lli_layout_gen;
if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
RETURN(0);
ll_finish_md_op_data(op_data);
- md_set_lock_data(sbi->ll_md_exp, &it.d.lustre.it_lock_handle, inode, NULL);
-
mode = it.d.lustre.it_lock_mode;
it.d.lustre.it_lock_mode = 0;
ll_intent_drop_lock(&it);