* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
plus the procfs overhead :( */
static int filter_export_stats_init(struct obd_device *obd,
struct obd_export *exp,
+ int reconnect,
void *client_nid)
{
int rc, newnid = 0;
/* Self-export gets no proc entry */
RETURN(0);
- rc = lprocfs_exp_setup(exp, client_nid, &newnid);
+ rc = lprocfs_exp_setup(exp, client_nid, reconnect, &newnid);
if (rc) {
/* Mask error for already created
* /proc entries */
RETURN(0);
clean:
- lprocfs_exp_cleanup(exp);
return rc;
}
fed = &exp->exp_filter_data;
*fed->fed_ted.ted_lcd = *lcd;
fed->fed_group = 0; /* will be assigned at connect */
- filter_export_stats_init(obd, exp, NULL);
+ filter_export_stats_init(obd, exp, 0, NULL);
rc = filter_client_add(obd, exp, cl_idx);
/* can't fail for existing client */
LASSERTF(rc == 0, "rc = %d\n", rc);
exp->exp_connecting = 0;
exp->exp_in_recovery = 0;
cfs_spin_unlock(&exp->exp_lock);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
obd->obd_max_recoverable_clients++;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
class_export_put(exp);
if (last_rcvd > le64_to_cpu(lsd->lsd_last_transno))
obd->obd_last_committed = le64_to_cpu(lsd->lsd_last_transno);
out:
- lut->lut_mount_count = mount_count + 1;
- lsd->lsd_mount_count = cpu_to_le64(lut->lut_mount_count);
+ obd->u.obt.obt_mount_count = mount_count + 1;
+ lsd->lsd_mount_count = cpu_to_le64(obd->u.obt.obt_mount_count);
/* save it, so mount count and last_transno is current */
rc = filter_update_server_data(obd);
loff_t off = 0;
ENTRY;
- O_dentry = simple_mkdir(current->fs->pwd, obd->u.obt.obt_vfsmnt,
+ O_dentry = simple_mkdir(cfs_fs_pwd(current->fs), obd->u.obt.obt_vfsmnt,
"O", 0700, 1);
CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
if (IS_ERR(O_dentry)) {
if (dir_dentry == NULL)
filter_parent_unlock(dparent);
if (IS_ERR(dchild)) {
- CERROR("%s: child lookup error %ld\n", obd->obd_name,
- PTR_ERR(dchild));
+ CERROR("%s: object "LPU64":"LPU64" lookup error: rc %ld\n",
+ obd->obd_name, id, group, PTR_ERR(dchild));
RETURN(dchild);
}
GOTO(out, rc = -EPERM);
/* check_sticky() */
- if ((dentry->d_inode->i_uid != current->fsuid &&
+ if ((dentry->d_inode->i_uid != cfs_curproc_fsuid() &&
!cfs_capable(CFS_CAP_FOWNER)) || IS_APPEND(dentry->d_inode) ||
IS_IMMUTABLE(dentry->d_inode))
GOTO(out, rc = -EPERM);
/* NOTE: This might need to go outside i_mutex, though it isn't clear if
* that was done because of journal_start (which is already done
* here) or some other ordering issue. */
- DQUOT_INIT(dir);
+ ll_vfs_dq_init(dir);
rc = ll_security_inode_unlink(dir, dentry, mnt);
if (rc)
struct inode *inode = dchild->d_inode;
int rc;
- if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
+ /* There should be 2 references to the inode:
+ * 1) taken by filter_prepare_destroy
+ * 2) taken by filter_destroy */
+ if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 2) {
CERROR("destroying objid %.*s ino %lu nlink %lu count %d\n",
dchild->d_name.len, dchild->d_name.name, inode->i_ino,
(unsigned long)inode->i_nlink,
* lock, and should not be granted if the lock will be blocked.
*/
- LASSERT(ns == res->lr_namespace);
+ LASSERT(ns == ldlm_res_to_ns(res));
lock_res(res);
rc = policy(lock, &tmpflags, 0, &err, &rpc_list);
check_res_locked(res);
if (rc == LDLM_ITER_CONTINUE) {
/* do not grant locks to the liblustre clients: they cannot
* handle ASTs robustly. We need to do this while still
- * holding ns_lock to avoid the lock remaining on the res_link
+ * holding lr_lock to avoid the lock remaining on the res_link
* list (and potentially being added to l_pending_list by an
* AST) when we are going to drop this lock ASAP. */
if (lock->l_export->exp_libclient ||
*reply_lvb = *res_lvb;
/*
- * ->ns_lock guarantees that no new locks are granted, and,
+ * lr_lock guarantees that no new locks are granted, and,
* therefore, that res->lr_lvb_data cannot increase beyond the
* end of already granted lock. As a result, it is safe to
* check against "stale" reply_lvb->lvb_size value without
LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
- /* Update the LVB from disk if the AST failed (this is a legal race) */
- /*
- * XXX nikita: situation when ldlm_server_glimpse_ast() failed before
- * sending ast is not handled. This can result in lost client writes.
- */
- if (rc != 0)
- ldlm_res_lvbo_update(res, NULL, 1);
lock_res(res);
*reply_lvb = *res_lvb;
__u8 *uuid_ptr;
char *str, *label;
char ns_name[48];
- request_queue_t *q;
+ struct request_queue *q;
int rc, i;
ENTRY;
obd->u.obt.obt_vfsmnt = mnt;
obd->u.obt.obt_sb = mnt->mnt_sb;
+ obd->u.obt.obt_magic = OBT_MAGIC;
filter->fo_fstype = mnt->mnt_sb->s_type->name;
CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
CFS_INIT_LIST_HEAD(&filter->fo_export_list);
cfs_sema_init(&filter->fo_alloc_lock, 1);
init_brw_stats(&filter->fo_filter_stats);
+ cfs_spin_lock_init(&filter->fo_flags_lock);
filter->fo_read_cache = 1; /* enable read-only cache by default */
filter->fo_writethrough_cache = 1; /* enable writethrough cache */
filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
filter->fo_fmd_max_num = FILTER_FMD_MAX_NUM_DEFAULT;
filter->fo_fmd_max_age = FILTER_FMD_MAX_AGE_DEFAULT;
+ filter->fo_syncjournal = 0; /* Don't sync journals on i/o by default */
+ filter_slc_set(filter); /* initialize sync on lock cancel */
rc = filter_prep(obd);
if (rc)
GOTO(err_post, rc = -ENOMEM);
sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
- obd->obd_namespace = ldlm_namespace_new(obd, ns_name, LDLM_NAMESPACE_SERVER,
- LDLM_NAMESPACE_GREEDY);
+ obd->obd_namespace = ldlm_namespace_new(obd, ns_name,
+ LDLM_NAMESPACE_SERVER,
+ LDLM_NAMESPACE_GREEDY,
+ LDLM_NS_TYPE_OST);
if (obd->obd_namespace == NULL)
GOTO(err_post, rc = -ENOMEM);
obd->obd_namespace->ns_lvbp = obd;
GOTO(err_post, rc);
q = bdev_get_queue(mnt->mnt_sb->s_bdev);
- if (q->max_sectors < q->max_hw_sectors &&
- q->max_sectors < PTLRPC_MAX_BRW_SIZE >> 9)
+ if (queue_max_sectors(q) < queue_max_hw_sectors(q) &&
+ queue_max_sectors(q) < PTLRPC_MAX_BRW_SIZE >> 9)
LCONSOLE_INFO("%s: underlying device %s should be tuned "
"for larger I/O requests: max_sectors = %u "
"could be up to max_hw_sectors=%u\n",
obd->obd_name, mnt->mnt_sb->s_id,
- q->max_sectors, q->max_hw_sectors);
+ queue_max_sectors(q), queue_max_hw_sectors(q));
uuid_ptr = fsfilt_uuid(obd, obd->u.obt.obt_sb);
if (uuid_ptr != NULL) {
obd->obd_name, body->lgdc_logid.lgl_oid,
body->lgdc_logid.lgl_oseq, body->lgdc_logid.lgl_ogen);
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->u.filter.fo_flags_lock);
obd->u.filter.fo_mds_ost_sync = 1;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->u.filter.fo_flags_lock);
rc = llog_connect(ctxt, &body->lgdc_logid,
&body->lgdc_gen, NULL);
llog_ctxt_put(ctxt);
filter_post(obd);
- LL_DQUOT_OFF(obd->u.obt.obt_sb);
+ ll_vfs_dq_off(obd->u.obt.obt_sb, 0);
shrink_dcache_sb(obd->u.obt.obt_sb);
server_put_mount(obd->obd_name, obd->u.obt.obt_vfsmnt);
rc = filter_connect_internal(exp, data, 1);
if (rc == 0)
- filter_export_stats_init(obd, exp, localdata);
+ filter_export_stats_init(obd, exp, 1, localdata);
RETURN(rc);
}
if (rc)
GOTO(cleanup, rc);
- filter_export_stats_init(obd, lexp, localdata);
+ filter_export_stats_init(obd, lexp, 0, localdata);
if (obd->obd_replayable) {
struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
LASSERT(lcd);
cleanup:
if (rc) {
class_disconnect(lexp);
- lprocfs_exp_cleanup(lexp);
*exp = NULL;
} else {
*exp = lexp;
*fcc = oa->o_lcookie;
}
if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) {
- DQUOT_INIT(inode);
+ unsigned long now = jiffies;
+ ll_vfs_dq_init(inode);
/* Filter truncates and writes are serialized by
* i_alloc_sem, see the comment in
* filter_preprw_write.*/
if (ia_valid & ATTR_SIZE)
down_write(&inode->i_alloc_sem);
LOCK_INODE_MUTEX(inode);
+ fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex");
old_size = i_size_read(inode);
}
if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_CREDITS))
fsfilt_extend(exp->exp_obd, inode, 0, handle);
- /* The truncate might have used up our transaction credits. Make
- * sure we have one left for the last_rcvd update. */
- err = fsfilt_extend(exp->exp_obd, inode, 1, handle);
+ /* The truncate might have used up our transaction credits. Make sure
+ * we have two left for the last_rcvd and VBR inode version updates. */
+ err = fsfilt_extend(exp->exp_obd, inode, 2, handle);
+
+ /* Update inode version only if data has changed => size has changed */
+ rc = filter_finish_transno(exp, ia_valid & ATTR_SIZE ? inode : NULL,
+ oti, rc, sync);
- rc = filter_finish_transno(exp, inode, oti, rc, sync);
if (sync) {
filter_cancel_cookies_cb(exp->exp_obd, 0, fcc, rc);
fcc = NULL;
*/
if (oa->o_valid &
(OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME)) {
+ unsigned long now = jiffies;
down_write(&dentry->d_inode->i_alloc_sem);
+ fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem");
fmd = filter_fmd_get(exp, oa->o_id, oa->o_seq);
if (fmd && fmd->fmd_mactime_xid < oti->oti_xid)
fmd->fmd_mactime_xid = oti->oti_xid;
filter_set_last_id(filter, id, doa.o_seq);
rc = filter_update_last_objid(exp->exp_obd, doa.o_seq, 1);
} else {
- /* don't reuse orphan object, return last used objid */
+ /*
+ * We have destroyed orphan objects, but don't want to reuse
+ * them. Therefore we don't reset last_id to the last created
+ * objects. Instead, we report back to the MDS the object id
+ * of the last orphan, so that the MDS can restart allocating
+ * objects from this id + 1 and thus skip the whole orphan
+ * object id range
+ */
oa->o_id = last;
rc = 0;
}
OBD_ALLOC(osfs, sizeof(*osfs));
if (osfs == NULL)
RETURN(-ENOMEM);
- rc = filter_statfs(obd, osfs, cfs_time_current_64() - CFS_HZ,
+ rc = filter_statfs(obd, osfs,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
0);
if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) {
CDEBUG(D_RPCTRACE,"%s: not enough space for create "
} else
next_id = filter_last_id(filter, group) + 1;
- /* Temporary solution for oid in CMD before fid-on-OST */
- if ((fid_seq_is_mdt0(oa->o_seq) && next_id >= IDIF_MAX_OID) &&
- (fid_seq_is_cmd(oa->o_seq) && next_id >= OBIF_MAX_OID)) {
- CERROR("%s:"POSTID" hit the max IDIF_MAX_OID(1<<48)!\n",
+ /* Don't create objects beyond the valid range for this SEQ */
+ if (unlikely(fid_seq_is_mdt0(group) &&
+ next_id >= IDIF_MAX_OID)) {
+ CERROR("%s:"POSTID" hit the IDIF_MAX_OID (1<<48)!\n",
+ obd->obd_name, next_id, group);
+ GOTO(cleanup, rc = -ENOSPC);
+ } else if (unlikely(!fid_seq_is_mdt0(group) &&
+ next_id >= OBIF_MAX_OID)) {
+ CERROR("%s:"POSTID" hit the OBIF_MAX_OID (1<<32)!\n",
obd->obd_name, next_id, group);
GOTO(cleanup, rc = -ENOSPC);
}
if (rc)
break;
if (cfs_time_after(jiffies, enough_time)) {
+ i++;
CDEBUG(D_RPCTRACE,
"%s: precreate slow - want %d got %d \n",
obd->obd_name, *num, i);
RETURN(rc);
}
-static int filter_create(struct obd_export *exp, struct obdo *oa,
- struct lov_stripe_md **ea, struct obd_trans_info *oti)
+int filter_create(struct obd_export *exp, struct obdo *oa,
+ struct lov_stripe_md **ea, struct obd_trans_info *oti)
{
struct obd_device *obd = exp->exp_obd;
struct filter_export_data *fed;
if ((oa->o_valid & OBD_MD_FLFLAGS) &&
(oa->o_flags & OBD_FL_RECREATE_OBJS)) {
- if (oa->o_id > filter_last_id(filter, oa->o_seq)) {
+ if (!obd->obd_recovering ||
+ oa->o_id > filter_last_id(filter, oa->o_seq)) {
CERROR("recreate objid "LPU64" > last id "LPU64"\n",
oa->o_id, filter_last_id(filter, oa->o_seq));
rc = -EINVAL;
struct llog_cookie *fcc = NULL;
int rc, rc2, cleanup_phase = 0, sync = 0;
struct iattr iattr;
+ unsigned long now;
ENTRY;
rc = filter_auth_capa(exp, NULL, oa->o_seq,
if (fcc != NULL)
*fcc = oa->o_lcookie;
}
- DQUOT_INIT(dchild->d_inode);
+ ll_vfs_dq_init(dchild->d_inode);
/* we're gonna truncate it first in order to avoid possible deadlock:
* P1 P2
* between page lock, i_mutex & starting new journal handle.
* (see bug 20321) -johann
*/
+ now = jiffies;
down_write(&dchild->d_inode->i_alloc_sem);
LOCK_INODE_MUTEX(dchild->d_inode);
+ fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex");
/* VBR: version recovery check */
rc = filter_version_get_check(exp, oti, dchild->d_inode);
RETURN(rc);
}
+ if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
+ *((__u32 *) val) = obd->u.filter.fo_sync_lock_cancel;
+ *vallen = sizeof(__u32);
+ RETURN(0);
+ }
+
CDEBUG(D_IOCTL, "invalid key\n");
RETURN(-EINVAL);
}