#include "mdt_internal.h"
-
static unsigned int max_mod_rpcs_per_client = 8;
module_param(max_mod_rpcs_per_client, uint, 0644);
MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
{
struct mdt_device *mdt = info->mti_mdt;
struct lu_name *lname = &info->mti_name;
- char *name = NULL;
+ char *filename = info->mti_filename;
struct mdt_object *parent;
u32 mode;
int rc = 0;
LASSERT(!info->mti_cross_ref);
- OBD_ALLOC(name, NAME_MAX + 1);
- if (name == NULL)
- return -ENOMEM;
- lname->ln_name = name;
-
/*
* We may want to allow this to mount a completely separate
* fileset from the MDT in the future, but keeping it to
break;
}
- strncpy(name, s1, lname->ln_namelen);
- name[lname->ln_namelen] = '\0';
+ strncpy(filename, s1, lname->ln_namelen);
+ filename[lname->ln_namelen] = '\0';
+ lname->ln_name = filename;
parent = mdt_object_find(info->mti_env, mdt, fid);
if (IS_ERR(parent)) {
}
}
- OBD_FREE(name, NAME_MAX + 1);
-
return rc;
}
static int mdt_statfs(struct tgt_session_info *tsi)
{
- struct ptlrpc_request *req = tgt_ses_req(tsi);
- struct mdt_thread_info *info = tsi2mdt_info(tsi);
- struct mdt_device *mdt = info->mti_mdt;
- struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
- struct ptlrpc_service_part *svcpt;
- struct obd_statfs *osfs;
- int rc;
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct mdt_thread_info *info = tsi2mdt_info(tsi);
+ struct mdt_device *mdt = info->mti_mdt;
+ struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
+ struct md_device *next = mdt->mdt_child;
+ struct ptlrpc_service_part *svcpt;
+ struct obd_statfs *osfs;
+ struct mdt_body *reqbody = NULL;
+ struct mdt_statfs_cache *msf;
+ int rc;
ENTRY;
if (!osfs)
GOTO(out, rc = -EPROTO);
- rc = tgt_statfs_internal(tsi->tsi_env, &mdt->mdt_lut, osfs,
- ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
- NULL);
- if (unlikely(rc))
- GOTO(out, rc);
+ if (mdt_is_sum_statfs_client(req->rq_export) &&
+ lustre_packed_msg_size(req->rq_reqmsg) ==
+ req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
+ &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
+ req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
+ reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
+ }
+
+ if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
+ msf = &mdt->mdt_sum_osfs;
+ else
+ msf = &mdt->mdt_osfs;
+
+ if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
+ /** statfs data is too old, get up-to-date one */
+ if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
+ rc = next->md_ops->mdo_statfs(info->mti_env,
+ next, osfs);
+ else
+ rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
+ osfs);
+ if (rc)
+ GOTO(out, rc);
+ spin_lock(&mdt->mdt_lock);
+ msf->msf_osfs = *osfs;
+ msf->msf_age = ktime_get_seconds();
+ spin_unlock(&mdt->mdt_lock);
+ } else {
+ /** use cached statfs data */
+ spin_lock(&mdt->mdt_lock);
+ *osfs = msf->msf_osfs;
+ spin_unlock(&mdt->mdt_lock);
+ }
/* at least try to account for cached pages. its still racy and
* might be under-reporting if clients haven't announced their
exp_connect_large_acl(info->mti_exp) &&
buf->lb_buf != info->mti_big_acl) {
if (info->mti_big_acl == NULL) {
+ info->mti_big_aclsize =
+ MIN(mdt->mdt_max_ea_size,
+ XATTR_SIZE_MAX);
OBD_ALLOC_LARGE(info->mti_big_acl,
- mdt->mdt_max_ea_size);
+ info->mti_big_aclsize);
if (info->mti_big_acl == NULL) {
+ info->mti_big_aclsize = 0;
CERROR("%s: unable to grow "
DFID" ACL buffer\n",
mdt_obd_name(mdt),
PFID(mdt_object_fid(o)));
RETURN(-ENOMEM);
}
-
- info->mti_big_aclsize =
- mdt->mdt_max_ea_size;
}
CDEBUG(D_INODE, "%s: grow the "DFID
" ACL buffer to size %d\n",
mdt_obd_name(mdt),
PFID(mdt_object_fid(o)),
- mdt->mdt_max_ea_size);
+ info->mti_big_aclsize);
buf->lb_buf = info->mti_big_acl;
buf->lb_len = info->mti_big_aclsize;
return -EINVAL;
}
+ LASSERT(buf->lb_buf);
+
rc = mo_xattr_get(info->mti_env, next, buf, name);
if (rc > 0) {
/* permission check. Make sure the calling process having permission
* to write both files. */
rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
- MAY_WRITE);
+ MAY_WRITE);
if (rc < 0)
GOTO(put, rc);
rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
- MAY_WRITE);
+ MAY_WRITE);
if (rc < 0)
GOTO(put, rc);
struct lu_name *lname = NULL;
struct mdt_lock_handle *lhp = NULL;
struct ldlm_lock *lock;
+ struct req_capsule *pill = info->mti_pill;
__u64 try_bits = 0;
bool is_resent;
int ma_need = 0;
+ bool deal_with_dom = false;
int rc;
ENTRY;
mdt_lock_reg_init(lhc, LCK_PR);
/*
- * Object's name is on another MDS, no lookup or layout
- * lock is needed here but update lock is.
+ * Object's name entry is on another MDS, it will
+ * request PERM lock only because LOOKUP lock is owned
+ * by the MDS where name entry resides.
+ *
+ * TODO: it should try layout lock too. - Jinshan
*/
child_bits &= ~(MDS_INODELOCK_LOOKUP |
MDS_INODELOCK_LAYOUT);
- child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
+ child_bits |= MDS_INODELOCK_PERM;
rc = mdt_object_lock(info, child, lhc, child_bits);
if (rc < 0)
if (unlikely(rc != 0))
mdt_object_unlock(info, child, lhc, 1);
- RETURN(rc);
- }
+ mdt_pack_secctx_in_reply(info, child);
+
+ RETURN(rc);
+ }
lname = &info->mti_name;
- mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
+ mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
if (lu_name_is_valid(lname)) {
CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
"ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
PNAME(lname), ldlm_rep);
} else {
- reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
+ reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
if (unlikely(reqbody == NULL))
RETURN(err_serious(-EPROTO));
child_bits &= ~MDS_INODELOCK_UPDATE;
rc = mdt_object_lock(info, child, lhc, child_bits);
}
- if (unlikely(rc != 0))
- GOTO(out_child, rc);
- }
+ if (unlikely(rc != 0))
+ GOTO(out_child, rc);
+ }
- lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
+ lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
- /* finally, we can get attr for child. */
- rc = mdt_getattr_internal(info, child, ma_need);
- if (unlikely(rc != 0)) {
- mdt_object_unlock(info, child, lhc, 1);
+ /* finally, we can get attr for child. */
+ rc = mdt_getattr_internal(info, child, ma_need);
+ if (unlikely(rc != 0)) {
+ mdt_object_unlock(info, child, lhc, 1);
+ GOTO(out_lock, rc);
} else if (lock) {
/* Debugging code. */
LDLM_DEBUG(lock, "Returning lock to client");
if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
mdt_object_exists(child) && !mdt_object_remote(child) &&
- child != parent) {
- LDLM_LOCK_PUT(lock);
- mdt_object_put(info->mti_env, child);
- /* NB: call the mdt_pack_size2body always after
- * mdt_object_put(), that is why this speacial
- * exit path is used. */
- rc = mdt_pack_size2body(info, child_fid,
- &lhc->mlh_reg_lh);
- if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
- /* DOM lock was taken in advance but this is
- * not DoM file. Drop the lock. */
- lock_res_and_lock(lock);
- ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
- unlock_res_and_lock(lock);
- }
+ child != parent)
+ deal_with_dom = true;
+ }
- GOTO(out_parent, rc = 0);
- }
- }
- if (lock)
- LDLM_LOCK_PUT(lock);
+ mdt_pack_secctx_in_reply(info, child);
- EXIT;
+out_lock:
+ if (lock)
+ LDLM_LOCK_PUT(lock);
+
+ EXIT;
out_child:
- mdt_object_put(info->mti_env, child);
+ mdt_object_put(info->mti_env, child);
+ if (deal_with_dom) {
+ rc = mdt_pack_size2body(info, child_fid,
+ &lhc->mlh_reg_lh);
+ if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
+ /* DOM lock was taken in advance but this is
+ * not DoM file. Drop the lock.
+ */
+ lock_res_and_lock(lock);
+ ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
+ unlock_res_and_lock(lock);
+ }
+ rc = 0;
+ }
out_parent:
- if (lhp)
- mdt_object_unlock(info, parent, lhp, 1);
- return rc;
+ if (lhp)
+ mdt_object_unlock(info, parent, lhp, 1);
+ return rc;
}
/* normal handler: should release the child lock */
return 0;
}
+static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
+{
+ return op == REINT_OPEN &&
+ !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
+}
+
+static void mdt_preset_secctx_size(struct mdt_thread_info *info)
+{
+ struct req_capsule *pill = info->mti_pill;
+
+ if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
+ RCL_SERVER) &&
+ req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
+ RCL_CLIENT)) {
+ if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
+ RCL_CLIENT) != 0) {
+ /* pre-set size in server part with max size */
+ req_capsule_set_size(pill, &RMF_FILE_SECCTX,
+ RCL_SERVER,
+ info->mti_mdt->mdt_max_ea_size);
+ } else {
+ req_capsule_set_size(pill, &RMF_FILE_SECCTX,
+ RCL_SERVER, 0);
+ }
+ }
+
+}
+
static int mdt_reint_internal(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc,
__u32 op)
ENTRY;
- rc = mdt_reint_unpack(info, op);
- if (rc != 0) {
- CERROR("Can't unpack reint, rc %d\n", rc);
- RETURN(err_serious(rc));
- }
+ rc = mdt_reint_unpack(info, op);
+ if (rc != 0) {
+ CERROR("Can't unpack reint, rc %d\n", rc);
+ RETURN(err_serious(rc));
+ }
+
- /* for replay (no_create) lmm is not needed, client has it already */
- if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
- req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
+ /* check if the file system is set to readonly. O_RDONLY open
+ * is still allowed even the file system is set to readonly mode */
+ if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
+ RETURN(err_serious(-EROFS));
+
+ /* for replay (no_create) lmm is not needed, client has it already */
+ if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
+ req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
DEF_REP_MD_SIZE);
/* llog cookies are always 0, the field is kept for compatibility */
- if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
+ if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
/* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
- rc = req_capsule_server_pack(pill);
- if (rc != 0) {
- CERROR("Can't pack response, rc %d\n", rc);
- RETURN(err_serious(rc));
- }
+ mdt_preset_secctx_size(info);
+
+ rc = req_capsule_server_pack(pill);
+ if (rc != 0) {
+ CERROR("Can't pack response, rc %d\n", rc);
+ RETURN(err_serious(rc));
+ }
- if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
- repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
- LASSERT(repbody);
+ if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
+ repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
+ LASSERT(repbody);
repbody->mbo_eadatasize = 0;
repbody->mbo_aclsize = 0;
- }
+ }
- OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
- /* for replay no cookkie / lmm need, because client have this already */
- if (info->mti_spec.no_create)
- if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
- req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
+ /* for replay no cookkie / lmm need, because client have this already */
+ if (info->mti_spec.no_create)
+ if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
+ req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
- rc = mdt_init_ucred_reint(info);
- if (rc)
- GOTO(out_shrink, rc);
+ rc = mdt_init_ucred_reint(info);
+ if (rc)
+ GOTO(out_shrink, rc);
- rc = mdt_fix_attr_ucred(info, op);
- if (rc != 0)
- GOTO(out_ucred, rc = err_serious(rc));
+ rc = mdt_fix_attr_ucred(info, op);
+ if (rc != 0)
+ GOTO(out_ucred, rc = err_serious(rc));
rc = mdt_check_resent(info, mdt_reconstruct, lhc);
if (rc < 0) {
} else if (rc == 1) {
DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
- GOTO(out_ucred, rc);
- }
- rc = mdt_reint_rec(info, lhc);
- EXIT;
+ GOTO(out_ucred, rc);
+ }
+ rc = mdt_reint_rec(info, lhc);
+ EXIT;
out_ucred:
- mdt_exit_ucred(info);
+ mdt_exit_ucred(info);
out_shrink:
mdt_client_compatibility(info);
* Data-on-MDT optimization - read data along with OPEN and return it
* in reply. Do that only if we have both DOM and LAYOUT locks.
*/
- if (rc == 0 && op == REINT_OPEN &&
+ if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
info->mti_attr.ma_lmm != NULL &&
mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
rc = mdt_dom_read_on_open(info, info->mti_mdt,
[REINT_OPEN] = &RQF_MDS_REINT_OPEN,
[REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
[REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
- [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME,
+ [REINT_MIGRATE] = &RQF_MDS_REINT_MIGRATE,
[REINT_RESYNC] = &RQF_MDS_REINT_RESYNC,
};
}
/* this should sync the whole device */
-static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
+int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
{
struct dt_device *dt = mdt->mdt_bottom;
int rc;
{
struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+ struct ldlm_cb_set_arg *arg = data;
bool commit_async = false;
int rc;
ENTRY;
unlock_res_and_lock(lock);
RETURN(0);
}
- /* There is no lock conflict if l_blocking_lock == NULL,
- * it indicates a blocking ast sent from ldlm_lock_decref_internal
- * when the last reference to a local lock was released */
- if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
- lock->l_blocking_lock != NULL) {
+
+ /* A blocking ast may be sent from ldlm_lock_decref_internal
+ * when the last reference to a local lock was released and
+ * during blocking event from ldlm_work_bl_ast_lock().
+ * The 'data' parameter is l_ast_data in the first case and
+ * callback arguments in the second one. Distinguish them by that.
+ */
+ if (!data || data == lock->l_ast_data || !arg->bl_desc)
+ goto skip_cos_checks;
+
+ if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
if (mdt_cos_is_enabled(mdt)) {
- if (lock->l_client_cookie !=
- lock->l_blocking_lock->l_client_cookie)
+ if (!arg->bl_desc->bl_same_client)
mdt_set_lock_sync(lock);
} else if (mdt_slc_is_enabled(mdt) &&
- ldlm_is_cos_incompat(lock->l_blocking_lock)) {
+ arg->bl_desc->bl_cos_incompat) {
mdt_set_lock_sync(lock);
/*
* we may do extra commit here, but there is a small
*/
commit_async = true;
}
- } else if (lock->l_req_mode == LCK_COS &&
- lock->l_blocking_lock != NULL) {
+ } else if (lock->l_req_mode == LCK_COS) {
commit_async = true;
}
+skip_cos_checks:
rc = ldlm_blocking_ast_nocheck(lock);
if (commit_async) {
return 1;
}
+static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
+{
+ mdt_object_get(NULL, lock->l_ast_data);
+}
+
int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
struct mdt_object *o, const struct lu_fid *fid,
struct lustre_handle *lh, enum ldlm_mode mode,
* if we cache lock, couple lock with mdt_object, so that object
* can be easily found in lock ASTs.
*/
- mdt_object_get(mti->mti_env, o);
einfo->ei_cbdata = o;
+ einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
}
memset(policy, 0, sizeof(*policy));
rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
policy);
- if (rc < 0 && cache)
- mdt_object_put(mti->mti_env, o);
/* Return successfully acquired bits to a caller */
if (rc == 0) {
struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
union ldlm_policy_data *policy = &info->mti_policy;
struct ldlm_res_id *res_id = &info->mti_res_id;
- __u64 dlmflags = 0;
+ __u64 dlmflags = 0, *cookie = NULL;
int rc;
ENTRY;
/* Only enqueue LOOKUP lock for remote object */
LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
+ /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
+ if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
+ *ibits == MDS_INODELOCK_OPEN)
+ dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
+
if (lh->mlh_type == MDT_PDO_LOCK) {
/* check for exists after object is locked */
if (mdt_object_exists(o) == 0) {
}
}
-
fid_build_reg_res_name(mdt_object_fid(o), res_id);
dlmflags |= LDLM_FL_ATOMIC_CB;
+ if (info->mti_exp)
+ cookie = &info->mti_exp->exp_handle.h_cookie;
+
/*
* Take PDO lock on whole directory and build correct @res_id for lock
* on part of directory.
/* at least one of them should be set */
LASSERT(policy->l_inodebits.bits |
policy->l_inodebits.try_bits);
- rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
- policy, res_id, dlmflags,
- info->mti_exp == NULL ? NULL :
- &info->mti_exp->exp_handle.h_cookie);
+ rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
+ lh->mlh_pdo_mode, policy, res_id,
+ dlmflags, cookie);
if (unlikely(rc != 0))
GOTO(out_unlock, rc);
}
* going to be sent to client. If it is - mdt_intent_policy() path will
* fix it up and turn FL_LOCAL flag off.
*/
- rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
- res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
- info->mti_exp == NULL ? NULL :
- &info->mti_exp->exp_handle.h_cookie);
+ rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
+ policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
+ cookie);
out_unlock:
if (rc != 0)
mdt_object_unlock(info, o, lh, 1);
}
}
+ /* other components like LFSCK can use lockless access
+ * and populate cache, so we better invalidate it */
+ mo_invalidate(info->mti_env, mdt_object_child(o));
+
RETURN(0);
}
* - create lu_object, corresponding to the fid in mdt_body, and save it in
* @info;
*
- * - if HABEO_CORPUS flag is set for this request type check whether object
+ * - if HAS_BODY flag is set for this request type check whether object
* actually exists on storage (lu_object_exists()).
*
*/
obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
if (!IS_ERR(obj)) {
- if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
+ if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
mdt_object_put(env, obj);
rc = -ENOENT;
} else {
static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
enum tgt_handler_flags flags)
{
- struct req_capsule *pill = info->mti_pill;
- int rc;
- ENTRY;
+ struct req_capsule *pill = info->mti_pill;
+ int rc;
- if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
- rc = mdt_body_unpack(info, flags);
- else
- rc = 0;
+ ENTRY;
- if (rc == 0 && (flags & HABEO_REFERO)) {
- /* Pack reply. */
- if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
- req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
+ if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
+ rc = mdt_body_unpack(info, flags);
+ else
+ rc = 0;
+
+ if (rc == 0 && (flags & HAS_REPLY)) {
+ /* Pack reply. */
+ if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
+ req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
DEF_REP_MD_SIZE);
- if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
+ if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
req_capsule_set_size(pill, &RMF_LOGCOOKIES,
RCL_SERVER, 0);
req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
- rc = req_capsule_server_pack(pill);
- }
- RETURN(rc);
+ mdt_preset_secctx_size(info);
+
+ rc = req_capsule_server_pack(pill);
+ if (rc)
+ CWARN("%s: cannot pack response: rc = %d\n",
+ mdt_obd_name(info->mti_mdt), rc);
+ }
+ RETURN(rc);
}
void mdt_lock_handle_init(struct mdt_lock_handle *lh)
if (ldlm_rep == NULL ||
OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
mdt_object_unlock(info, info->mti_object, lhc, 1);
- RETURN(err_serious(-EFAULT));
+ if (is_serious(rc))
+ RETURN(rc);
+ else
+ RETURN(err_serious(-EFAULT));
}
ldlm_rep->lock_policy_res2 = clear_serious(rc);
if (layout_size > info->mti_mdt->mdt_max_mdsize)
info->mti_mdt->mdt_max_mdsize = layout_size;
}
+ CDEBUG(D_INFO, "%s: layout_size %d\n",
+ mdt_obd_name(info->mti_mdt), layout_size);
}
/*
out:
lhc->mlh_reg_lh.cookie = 0;
- return rc;
+ RETURN(rc);
}
static int mdt_intent_open(enum ldlm_intent_flags it_opc,
rc = mdt_reint_internal(info, lhc, opc);
- /* Check whether the reply has been packed successfully. */
- if (mdt_info_req(info)->rq_repmsg != NULL)
- rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
- if (rep == NULL)
- RETURN(err_serious(-EFAULT));
+ /* Check whether the reply has been packed successfully. */
+ if (mdt_info_req(info)->rq_repmsg != NULL)
+ rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
+ if (rep == NULL) {
+ if (is_serious(rc))
+ RETURN(rc);
+ else
+ RETURN(err_serious(-EFAULT));
+ }
/* MDC expects this in any case */
if (rc != 0)
case IT_OPEN:
case IT_OPEN|IT_CREAT:
/*
- * OCREAT is not a MUTABOR request since the file may
+ * OCREAT is not a IS_MUTABLE request since the file may
* already exist. We do the extra check of
* OBD_CONNECT_RDONLY in mdt_reint_open() when we
* really need to create the object.
case IT_LOOKUP:
it_format = &RQF_LDLM_INTENT_GETATTR;
it_handler = &mdt_intent_getattr;
- it_handler_flags = HABEO_REFERO;
+ it_handler_flags = HAS_REPLY;
break;
case IT_GETXATTR:
it_format = &RQF_LDLM_INTENT_GETXATTR;
it_handler = &mdt_intent_getxattr;
- it_handler_flags = HABEO_CORPUS;
+ it_handler_flags = HAS_BODY;
break;
case IT_LAYOUT:
it_format = &RQF_LDLM_INTENT_LAYOUT;
if (rc < 0)
RETURN(rc);
- if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
+ if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
RETURN(-EROFS);
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
}
-static int mdt_intent_policy(struct ldlm_namespace *ns,
- struct ldlm_lock **lockp, void *req_cookie,
- enum ldlm_mode mode, __u64 flags, void *data)
+static int mdt_intent_policy(const struct lu_env *env,
+ struct ldlm_namespace *ns,
+ struct ldlm_lock **lockp,
+ void *req_cookie,
+ enum ldlm_mode mode,
+ __u64 flags, void *data)
{
struct tgt_session_info *tsi;
struct mdt_thread_info *info;
LASSERT(req != NULL);
- tsi = tgt_ses_info(req->rq_svc_thread->t_env);
+ tsi = tgt_ses_info(env);
info = tsi2mdt_info(tsi);
LASSERT(info != NULL);
0, MDS_DISCONNECT, tgt_disconnect,
&RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
TGT_RPC_HANDLER(MDS_FIRST_OPC,
- HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
+ HAS_REPLY, MDS_SET_INFO, mdt_set_info,
&RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
-TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GET_ROOT, mdt_get_root),
-TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
+TGT_MDT_HDL(HAS_REPLY, MDS_GET_ROOT, mdt_get_root),
+TGT_MDT_HDL(HAS_BODY, MDS_GETATTR, mdt_getattr),
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_GETATTR_NAME,
mdt_getattr_name),
-TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
-TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
-TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
-TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
+TGT_MDT_HDL(HAS_BODY, MDS_GETXATTR, mdt_tgt_getxattr),
+TGT_MDT_HDL(HAS_REPLY, MDS_STATFS, mdt_statfs),
+TGT_MDT_HDL(IS_MUTABLE, MDS_REINT, mdt_reint),
+TGT_MDT_HDL(HAS_BODY, MDS_CLOSE, mdt_close),
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_READPAGE, mdt_readpage),
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_SYNC, mdt_sync),
TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
mdt_hsm_progress),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
mdt_hsm_ct_register),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
mdt_hsm_ct_unregister),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
mdt_hsm_state_get),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
mdt_hsm_state_set),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
-TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION, mdt_hsm_action),
+TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
mdt_hsm_request),
-TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
+TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
MDS_SWAP_LAYOUTS,
mdt_swap_layouts),
};
static struct tgt_handler mdt_io_ops[] = {
-TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
+TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
mdt_hp_brw),
-TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR, OST_BRW_WRITE, tgt_brw_write,
+TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write,
mdt_hp_brw),
-TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
+TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
OST_PUNCH, mdt_punch_hdl,
mdt_hp_punch),
-TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC, mdt_data_sync),
+TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, mdt_data_sync),
};
static struct tgt_handler mdt_sec_ctx_ops[] = {
};
static struct tgt_handler mdt_quota_ops[] = {
-TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
+TGT_QUOTA_HDL(HAS_REPLY, QUOTA_DQACQ, mdt_quota_dqacq),
};
static struct tgt_opc_slice mdt_common_slice[] = {
obd = class_name2obd(dev);
LASSERT(obd != NULL);
- m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
+ m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
m->mdt_opts.mo_evict_tgt_nids = 1;
m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
init_rwsem(&m->mdt_squash.rsi_sem);
spin_lock_init(&m->mdt_lock);
- m->mdt_enable_remote_dir = 0;
+ m->mdt_enable_remote_dir = 1;
+ m->mdt_enable_striped_dir = 1;
+ m->mdt_enable_dir_migration = 1;
m->mdt_enable_remote_dir_gid = 0;
+ m->mdt_enable_remote_rename = 1;
atomic_set(&m->mdt_mds_mds_conns, 0);
atomic_set(&m->mdt_async_commit_count, 0);
/* Remove mfd handle so it can't be found again.
* We are consuming the mfd_list reference here. */
- class_handle_unhash(&mfd->mfd_handle);
+ class_handle_unhash(&mfd->mfd_open_handle);
list_move_tail(&mfd->mfd_list, &closing_list);
}
spin_unlock(&med->med_open_lock);
* archive request into a noop if it's not actually
* dirty.
*/
- if (mfd->mfd_mode & MDS_FMODE_WRITE)
+ if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
/* Don't unlink orphan on failover umount, LU-184 */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff