*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
#define DEBUG_SUBSYSTEM S_MDS
#include <linux/module.h>
+#include <linux/pagemap.h>
#include <dt_object.h>
#include <lustre_acl.h>
#include <obd.h>
#include <obd_support.h>
+#include <llog_swab.h>
+
#include "mdt_internal.h"
static unsigned int max_mod_rpcs_per_client = 8;
-CFS_MODULE_PARM(max_mod_rpcs_per_client, "i", uint, 0644,
- "maximum number of modify RPCs in flight allowed per client");
+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");
mdl_mode_t mdt_mdl_lock_modes[] = {
[LCK_MINMODE] = MDL_MINMODE,
EXIT;
}
-static int mdt_getstatus(struct tgt_session_info *tsi)
+static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
+ struct lu_fid *fid)
+{
+ struct mdt_device *mdt = info->mti_mdt;
+ struct lu_name *lname = &info->mti_name;
+ char *name = NULL;
+ 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
+ * ROOT/ only for now avoid potential security issues.
+ */
+ *fid = mdt->mdt_md_root_fid;
+
+ while (rc == 0 && fileset != NULL && *fileset != '\0') {
+ const char *s1 = fileset;
+ const char *s2;
+
+ while (*++s1 == '/')
+ ;
+ s2 = s1;
+ while (*s2 != '/' && *s2 != '\0')
+ s2++;
+
+ if (s2 == s1)
+ break;
+
+ fileset = s2;
+
+ lname->ln_namelen = s2 - s1;
+ if (lname->ln_namelen > NAME_MAX) {
+ rc = -EINVAL;
+ break;
+ }
+
+ /* reject .. as a path component */
+ if (lname->ln_namelen == 2 &&
+ strncmp(s1, "..", 2) == 0) {
+ rc = -EINVAL;
+ break;
+ }
+
+ strncpy(name, s1, lname->ln_namelen);
+ name[lname->ln_namelen] = '\0';
+
+ parent = mdt_object_find(info->mti_env, mdt, fid);
+ if (IS_ERR(parent)) {
+ rc = PTR_ERR(parent);
+ break;
+ }
+ /* Only got the fid of this obj by name */
+ fid_zero(fid);
+ rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
+ fid, &info->mti_spec);
+ mdt_object_put(info->mti_env, parent);
+ }
+ if (!rc) {
+ parent = mdt_object_find(info->mti_env, mdt, fid);
+ if (IS_ERR(parent))
+ rc = PTR_ERR(parent);
+ else {
+ mode = lu_object_attr(&parent->mot_obj);
+ mdt_object_put(info->mti_env, parent);
+ if (!S_ISDIR(mode))
+ rc = -ENOTDIR;
+ }
+ }
+
+ OBD_FREE(name, NAME_MAX + 1);
+
+ return rc;
+}
+
+static int mdt_get_root(struct tgt_session_info *tsi)
{
struct mdt_thread_info *info = tsi2mdt_info(tsi);
struct mdt_device *mdt = info->mti_mdt;
struct mdt_body *repbody;
+ char *fileset = NULL, *buffer = NULL;
int rc;
+ struct obd_export *exp = info->mti_exp;
+ char *nodemap_fileset;
ENTRY;
if (rc)
GOTO(out, rc = err_serious(rc));
- if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
+ if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
GOTO(out, rc = err_serious(-ENOMEM));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
- repbody->mbo_fid1 = mdt->mdt_md_root_fid;
+ if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
+ fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
+ if (fileset == NULL)
+ GOTO(out, rc = err_serious(-EFAULT));
+ }
+
+ nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
+ if (nodemap_fileset && nodemap_fileset[0]) {
+ CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
+ if (fileset) {
+ /* consider fileset from client as a sub-fileset
+ * of the nodemap one */
+ OBD_ALLOC(buffer, PATH_MAX + 1);
+ if (buffer == NULL)
+ GOTO(out, rc = err_serious(-ENOMEM));
+ if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
+ nodemap_fileset, fileset) >= PATH_MAX + 1)
+ GOTO(out, rc = err_serious(-EINVAL));
+ fileset = buffer;
+ } else {
+ /* enforce fileset as specified in the nodemap */
+ fileset = nodemap_fileset;
+ }
+ }
+
+ if (fileset) {
+ CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
+ rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
+ if (rc < 0)
+ GOTO(out, rc = err_serious(rc));
+ } else {
+ repbody->mbo_fid1 = mdt->mdt_md_root_fid;
+ }
repbody->mbo_valid |= OBD_MD_FLID;
EXIT;
out:
mdt_thread_info_fini(info);
+ if (buffer)
+ OBD_FREE(buffer, PATH_MAX+1);
return rc;
}
rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
if (rc)
GOTO(out, rc);
- spin_lock(&info->mti_mdt->mdt_osfs_lock);
+ spin_lock(&info->mti_mdt->mdt_lock);
info->mti_mdt->mdt_osfs = *osfs;
info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
- spin_unlock(&info->mti_mdt->mdt_osfs_lock);
+ spin_unlock(&info->mti_mdt->mdt_lock);
} else {
/** use cached statfs data */
- spin_lock(&info->mti_mdt->mdt_osfs_lock);
+ spin_lock(&info->mti_mdt->mdt_lock);
*osfs = info->mti_mdt->mdt_osfs;
- spin_unlock(&info->mti_mdt->mdt_osfs_lock);
+ spin_unlock(&info->mti_mdt->mdt_lock);
}
if (rc == 0)
void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
const struct lu_attr *attr, const struct lu_fid *fid)
{
- struct md_attr *ma = &info->mti_attr;
- struct obd_export *exp = info->mti_exp;
- struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
+ struct md_attr *ma = &info->mti_attr;
+ struct obd_export *exp = info->mti_exp;
+ struct lu_nodemap *nodemap = NULL;
LASSERT(ma->ma_valid & MA_INODE);
b->mbo_nlink = attr->la_nlink;
b->mbo_valid |= OBD_MD_FLNLINK;
}
+ if (attr->la_valid & (LA_UID|LA_GID)) {
+ nodemap = nodemap_get_from_exp(exp);
+ if (IS_ERR(nodemap))
+ goto out;
+ }
if (attr->la_valid & LA_UID) {
b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
NODEMAP_FS_TO_CLIENT,
attr->la_gid);
b->mbo_valid |= OBD_MD_FLGID;
}
+
b->mbo_mode = attr->la_mode;
if (attr->la_valid & LA_MODE)
b->mbo_valid |= OBD_MD_FLMODE;
if (fid != NULL) {
b->mbo_fid1 = *fid;
b->mbo_valid |= OBD_MD_FLID;
- CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
+ CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
}
- if (info != NULL)
- mdt_body_reverse_idmap(info, b);
-
if (!(attr->la_valid & LA_TYPE))
return;
if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
PFID(fid), (unsigned long long)b->mbo_size);
+
+out:
+ if (!IS_ERR_OR_NULL(nodemap))
+ nodemap_putref(nodemap);
}
static inline int mdt_body_has_lov(const struct lu_attr *la,
#endif
out:
ma->ma_need = need;
- CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
+ CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
rc, ma->ma_valid, ma->ma_lmm);
RETURN(rc);
}
struct mdt_body *repbody;
struct lu_buf *buffer = &info->mti_buf;
struct obd_export *exp = info->mti_exp;
- struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
int rc;
int is_root;
ENTRY;
rc = mdt_attr_get_complex(info, o, ma);
if (unlikely(rc)) {
- CERROR("%s: getattr error for "DFID": rc = %d\n",
+ CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
+ "%s: getattr error for "DFID": rc = %d\n",
mdt_obd_name(info->mti_mdt),
PFID(mdt_object_fid(o)), rc);
RETURN(rc);
PFID(mdt_object_fid(o)), rc);
rc = -EFAULT;
} else {
- int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
+ int print_limit = min_t(int, PAGE_SIZE - 128, rc);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
rc -= 2;
repbody->mbo_max_mdsize);
}
- if (exp_connect_rmtclient(info->mti_exp) &&
- reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
- void *buf = req_capsule_server_get(pill, &RMF_ACL);
-
- /* mdt_getattr_lock only */
- rc = mdt_pack_remote_perm(info, o, buf);
- if (rc) {
- repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
- repbody->mbo_aclsize = 0;
- RETURN(rc);
- } else {
- repbody->mbo_valid |= OBD_MD_FLRMTPERM;
- repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
- }
- }
#ifdef CONFIG_FS_POSIX_ACL
- else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
- (reqbody->mbo_valid & OBD_MD_FLACL))
+ if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
+ (reqbody->mbo_valid & OBD_MD_FLACL)) {
+ struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
+ if (IS_ERR(nodemap))
+ RETURN(PTR_ERR(nodemap));
+
rc = mdt_pack_acl2body(info, repbody, o, nodemap);
+ nodemap_putref(nodemap);
+ }
#endif
out:
repbody->mbo_eadatasize = 0;
repbody->mbo_aclsize = 0;
- if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
- rc = mdt_init_ucred(info, reqbody);
- else
- rc = mdt_check_ucred(info);
+ rc = mdt_check_ucred(info);
if (unlikely(rc))
GOTO(out_shrink, rc);
info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
rc = mdt_getattr_internal(info, obj, 0);
- if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
- mdt_exit_ucred(info);
EXIT;
out_shrink:
mdt_client_compatibility(info);
*/
static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
{
- __u64 o1_flags;
+ unsigned int o1_lov_created = o1->mot_lov_created;
mutex_lock(&o1->mot_lov_mutex);
mutex_lock(&o2->mot_lov_mutex);
- o1_flags = o1->mot_flags;
- o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
- (o2->mot_flags & MOF_LOV_CREATED);
-
- o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
- (o1_flags & MOF_LOV_CREATED);
+ o1->mot_lov_created = o2->mot_lov_created;
+ o2->mot_lov_created = o1_lov_created;
mutex_unlock(&o2->mot_lov_mutex);
mutex_unlock(&o1->mot_lov_mutex);
if (!mdt_object_exists(child)) {
LU_OBJECT_DEBUG(D_INFO, info->mti_env,
&child->mot_obj,
- "remote object doesn't exist.\n");
+ "remote object doesn't exist.");
mdt_object_unlock(info, child, lhc, 1);
RETURN(-ENOENT);
}
if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
LU_OBJECT_DEBUG(D_INODE, info->mti_env,
&parent->mot_obj,
- "Parent doesn't exist!\n");
+ "Parent doesn't exist!");
RETURN(-ESTALE);
}
if (!mdt_object_exists(child)) {
LU_OBJECT_DEBUG(D_INODE, info->mti_env,
&child->mot_obj,
- "Object doesn't exist!\n");
+ "Object doesn't exist!");
GOTO(out_child, rc = -ENOENT);
}
*/
rdpg->rp_hash = reqbody->mbo_size;
if (rdpg->rp_hash != reqbody->mbo_size) {
- CERROR("Invalid hash: "LPX64" != "LPX64"\n",
+ CERROR("Invalid hash: %#llx != %#llx\n",
rdpg->rp_hash, reqbody->mbo_size);
RETURN(-EFAULT);
}
rdpg->rp_attrs |= LUDA_64BITHASH;
rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
exp_max_brw_size(tsi->tsi_exp));
- rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
- PAGE_CACHE_SHIFT;
+ rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
+ PAGE_SHIFT;
OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
if (rdpg->rp_pages == NULL)
RETURN(-ENOMEM);
for (i = 0; i < rdpg->rp_npages; ++i) {
- rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
+ rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
if (rdpg->rp_pages[i] == NULL)
GOTO(free_rdpg, rc = -ENOMEM);
}
return rc;
}
+static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
+{
+ struct lu_ucred *uc = mdt_ucred_check(info);
+ struct lu_attr *attr = &info->mti_attr.ma_attr;
+
+ if (uc == NULL)
+ return -EINVAL;
+
+ if (op != REINT_SETATTR) {
+ if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
+ attr->la_uid = uc->uc_fsuid;
+ /* for S_ISGID, inherit gid from his parent, such work will be
+ * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
+ if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
+ attr->la_gid = uc->uc_fsgid;
+ }
+
+ return 0;
+}
+
static int mdt_reint_internal(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc,
__u32 op)
int id, rc;
struct mdt_device *mdt = mdt_exp2dev(exp);
struct lu_device *qmt = mdt->mdt_qmt_dev;
- struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
+ struct lu_nodemap *nodemap;
ENTRY;
oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
if (rc)
RETURN(err_serious(rc));
+ nodemap = nodemap_get_from_exp(exp);
+ if (IS_ERR(nodemap))
+ RETURN(PTR_ERR(nodemap));
+
switch (oqctl->qc_cmd) {
/* master quotactl */
case Q_SETINFO:
case Q_SETQUOTA:
if (!nodemap_can_setquota(nodemap))
- RETURN(-EPERM);
+ GOTO(out_nodemap, rc = -EPERM);
case Q_GETINFO:
case Q_GETQUOTA:
if (qmt == NULL)
- RETURN(-EOPNOTSUPP);
+ GOTO(out_nodemap, rc = -EOPNOTSUPP);
/* slave quotactl */
case Q_GETOINFO:
case Q_GETOQUOTA:
break;
default:
CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
- RETURN(-EFAULT);
+ GOTO(out_nodemap, rc = -EFAULT);
}
- /* map uid/gid for remote client */
id = oqctl->qc_id;
- if (exp_connect_rmtclient(exp)) {
- struct lustre_idmap_table *idmap;
-
- idmap = exp->exp_mdt_data.med_idmap;
-
- if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
- oqctl->qc_cmd != Q_GETINFO))
- RETURN(-EPERM);
-
- if (oqctl->qc_type == USRQUOTA)
- id = lustre_idmap_lookup_uid(NULL, idmap, 0,
- oqctl->qc_id);
- else if (oqctl->qc_type == GRPQUOTA)
- id = lustre_idmap_lookup_gid(NULL, idmap, 0,
- oqctl->qc_id);
- else
- RETURN(-EINVAL);
-
- if (id == CFS_IDMAP_NOTFOUND) {
- CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
- RETURN(-EACCES);
- }
- }
-
if (oqctl->qc_type == USRQUOTA)
id = nodemap_map_id(nodemap, NODEMAP_UID,
NODEMAP_CLIENT_TO_FS, id);
repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
if (repoqc == NULL)
- RETURN(err_serious(-EFAULT));
+ GOTO(out_nodemap, rc = err_serious(-EFAULT));
if (oqctl->qc_id != id)
swap(oqctl->qc_id, id);
default:
CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
- RETURN(-EFAULT);
+ GOTO(out_nodemap, rc = -EFAULT);
}
if (oqctl->qc_id != id)
swap(oqctl->qc_id, id);
*repoqc = *oqctl;
- RETURN(rc);
+
+ EXIT;
+
+out_nodemap:
+ nodemap_putref(nodemap);
+
+ return rc;
}
/** clone llog ctxt from child (mdd)
*/
static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
{
- int rc;
-
- rc = mdt_handle_idmap(tsi);
- if (unlikely(rc)) {
- struct ptlrpc_request *req = tgt_ses_req(tsi);
- __u32 opc;
-
- opc = lustre_msg_get_opc(req->rq_reqmsg);
- if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
- sptlrpc_svc_ctx_invalidate(req);
- }
-
CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
- return rc;
+ return 0;
}
/*
int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
void *data, int flag)
{
- struct lustre_handle lockh;
- int rc;
+ int rc = 0;
ENTRY;
switch (flag) {
- case LDLM_CB_BLOCKING:
+ case LDLM_CB_BLOCKING: {
+ struct lustre_handle lockh;
+
ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh,
ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
RETURN(rc);
}
break;
- case LDLM_CB_CANCELING:
+ }
+ case LDLM_CB_CANCELING: {
+ struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
+ struct mdt_device *mdt =
+ mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
+
LDLM_DEBUG(lock, "Revoke remote lock\n");
+
/* discard slc lock here so that it can be cleaned anytime,
* especially for cleanup_resource() */
- tgt_discard_slc_lock(lock);
+ tgt_discard_slc_lock(&mdt->mdt_lut, lock);
+
+ /* once we cache lock, l_ast_data is set to mdt_object */
+ if (lock->l_ast_data != NULL) {
+ struct mdt_object *mo = lock->l_ast_data;
+ struct lu_env env;
+
+ rc = lu_env_init(&env, LCT_MD_THREAD);
+ if (unlikely(rc != 0)) {
+ CWARN("%s: lu_env initialization failed, object"
+ "%p "DFID" is leaked!\n",
+ obd->obd_name, mo,
+ PFID(mdt_object_fid(mo)));
+ RETURN(rc);
+ }
+
+ if (lock->l_policy_data.l_inodebits.bits &
+ (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
+ rc = mo_invalidate(&env, mdt_object_child(mo));
+ mo->mot_cache_attr = 0;
+ }
+ mdt_object_put(&env, mo);
+ lu_env_fini(&env);
+ }
break;
+ }
default:
LBUG();
}
- RETURN(0);
+ RETURN(rc);
}
int mdt_check_resent_lock(struct mdt_thread_info *info,
/* Lock is pinned by ldlm_handle_enqueue0() as it is
* a resend case, however, it could be already destroyed
* due to client eviction or a raced cancel RPC. */
- LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
+ LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
lhc->mlh_reg_lh.cookie);
RETURN(-ESTALE);
}
int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
const struct lu_fid *fid, struct lustre_handle *lh,
- enum ldlm_mode mode, __u64 ibits, bool nonblock)
+ enum ldlm_mode mode, __u64 ibits, bool nonblock,
+ bool cache)
{
struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
union ldlm_policy_data *policy = &mti->mti_policy;
einfo->ei_res_id = res_id;
if (nonblock)
einfo->ei_nonblock = 1;
+ if (cache) {
+ /*
+ * 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;
+ }
memset(policy, 0, sizeof(*policy));
policy->l_inodebits.bits = ibits;
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);
+ einfo->ei_cbdata = NULL;
+ }
RETURN(rc);
}
rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
&lh->mlh_rreg_lh,
lh->mlh_rreg_mode,
- MDS_INODELOCK_UPDATE, nonblock);
+ MDS_INODELOCK_UPDATE, nonblock,
+ false);
if (rc != ELDLM_OK) {
if (local_lh != NULL)
mdt_object_unlock(info, o, local_lh, rc);
cos = (mdt_cos_is_enabled(mdt) ||
mdt_slc_is_enabled(mdt));
- LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
+ LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
h->cookie);
/* there is no request if mdt_object_unlock() is called
* from mdt_export_cleanup()->mdt_add_dirty_flag() */
if (likely(req != NULL)) {
CDEBUG(D_HA, "request = %p reply state = %p"
- " transno = "LPD64"\n", req,
+ " transno = %lld\n", req,
req->rq_reply_state, req->rq_transno);
if (cos) {
ldlm_lock_downgrade(lock, LCK_COS);
* \param decref force immediate lock releasing
*/
static void mdt_save_remote_lock(struct mdt_thread_info *info,
- struct lustre_handle *h, enum ldlm_mode mode,
- int decref)
+ struct mdt_object *o, struct lustre_handle *h,
+ enum ldlm_mode mode, int decref)
{
ENTRY;
if (lustre_handle_is_used(h)) {
+ struct ldlm_lock *lock = ldlm_handle2lock(h);
+
+ if (o != NULL &&
+ (lock->l_policy_data.l_inodebits.bits &
+ (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
+ mo_invalidate(info->mti_env, mdt_object_child(o));
+
if (decref || !info->mti_has_trans ||
!(mode & (LCK_PW | LCK_EX))) {
ldlm_lock_decref_and_cancel(h, mode);
+ LDLM_LOCK_PUT(lock);
} else {
- struct ldlm_lock *lock = ldlm_handle2lock(h);
struct ptlrpc_request *req = mdt_info_req(info);
LASSERT(req != NULL);
- tgt_save_slc_lock(lock, req->rq_transno);
+ tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
+ req->rq_transno);
ldlm_lock_decref(h, mode);
}
h->cookie = 0ull;
mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
- mdt_save_remote_lock(info, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, decref);
+ mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
+ decref);
EXIT;
}
static int mdt_tgt_connect(struct tgt_session_info *tsi)
{
- struct ptlrpc_request *req = tgt_ses_req(tsi);
- int rc;
-
- ENTRY;
-
if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
cfs_fail_val ==
tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
}
- rc = tgt_connect(tsi);
- if (rc != 0)
- RETURN(rc);
-
- rc = mdt_init_idmap(tsi);
- if (rc != 0)
- GOTO(err, rc);
- RETURN(0);
-err:
- obd_disconnect(class_export_get(req->rq_export));
- return rc;
+ return tgt_connect(tsi);
}
enum mdt_it_code {
mdt_intent_lock_replace(struct mdt_thread_info *info,
struct ldlm_lock **lockp,
struct mdt_lock_handle *lh,
- __u64 flags)
+ __u64 flags, int result)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct ldlm_lock *lock = *lockp;
RETURN(0);
}
- LASSERTF(new_lock != NULL,
- "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
+ if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
+ /* Lock is pinned by ldlm_handle_enqueue0() as it is
+ * a resend case, however, it could be already destroyed
+ * due to client eviction or a raced cancel RPC. */
+ LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
+ lh->mlh_reg_lh.cookie);
+ lh->mlh_reg_lh.cookie = 0;
+ RETURN(-ESTALE);
+ }
+
+ LASSERTF(new_lock != NULL,
+ "lockh %#llx flags %#llx : rc = %d\n",
+ lh->mlh_reg_lh.cookie, flags, result);
/*
* If we've already given this lock to a client once, then we should
lh->mlh_reg_mode = new_lock->l_granted_mode;
LDLM_DEBUG(new_lock, "Restoring lock cookie");
- DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
+ DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
lh->mlh_reg_lh.cookie);
return;
}
*/
lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
- DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
+ DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
dlmreq->lock_handle[0].cookie);
}
{
struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
struct ldlm_reply *ldlm_rep = NULL;
- int rc, grc;
+ int rc;
+ ENTRY;
/*
* Initialize lhc->mlh_reg_lh either from a previously granted lock
return rc;
}
- grc = mdt_getxattr(info);
-
- rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
+ rc = mdt_getxattr(info);
if (mdt_info_req(info)->rq_repmsg != NULL)
ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
- if (ldlm_rep == NULL)
+
+ 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));
+ }
+
+ ldlm_rep->lock_policy_res2 = clear_serious(rc);
- ldlm_rep->lock_policy_res2 = grc;
+ /* This is left for interop instead of adding a new interop flag.
+ * LU-7433 */
+#if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
+ if (ldlm_rep->lock_policy_res2) {
+ mdt_object_unlock(info, info->mti_object, lhc, 1);
+ RETURN(ELDLM_LOCK_ABORTED);
+ }
+#endif
- return rc;
+ rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
+ RETURN(rc);
}
static int mdt_intent_getattr(enum mdt_it_code opcode,
GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
}
- rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
+ rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
EXIT;
out_ucred:
mdt_exit_ucred(info);
mdt_object_put(info->mti_env, obj);
if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
- rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
+ rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
out:
lhc->mlh_reg_lh.cookie = 0;
if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
(rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
rep->lock_policy_res2 = 0;
- rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
+ rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
RETURN(rc);
}
HABEO_REFERO, 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_GETSTATUS, mdt_getstatus),
+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,
mdt_getattr_name),
next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
- target_recovery_fini(obd);
ping_evictor_stop();
if (m->mdt_opts.mo_coordinator)
d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
}
+ if (m->mdt_md_root != NULL) {
+ mdt_object_put(env, m->mdt_md_root);
+ m->mdt_md_root = NULL;
+ }
+
mdt_quota_fini(env, m);
cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
m->mdt_squash.rsi_gid = 0;
INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
init_rwsem(&m->mdt_squash.rsi_sem);
- spin_lock_init(&m->mdt_osfs_lock);
+ spin_lock_init(&m->mdt_lock);
m->mdt_osfs_age = cfs_time_shift_64(-1000);
m->mdt_enable_remote_dir = 0;
m->mdt_enable_remote_dir_gid = 0;
{ "mdt.group_upcall", NULL },
{ "mdt.quota_type", NULL },
{ "mdd.quota_type", NULL },
+ { "mdt.som", NULL },
{ "mdt.rootsquash", "mdt.root_squash" },
{ "mdt.nosquash_nid", "mdt.nosquash_nids" },
{ NULL }
struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
return (*p)(env, cookie,
- LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
- mdto, mdto->mot_flags, mdto->mot_write_count);
+ LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
+ mdto, mdto->mot_lov_created ? "lov_created" : "",
+ mdto->mot_cache_attr ? "cache_attr" : "",
+ mdto->mot_write_count);
}
static int mdt_prepare(const struct lu_env *env,
* connect flags from the obd_connect_data::ocd_connect_flags field of the
* reply. \see mdt_connect().
*
+ * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
+ * full struct obd_connect_data. So care must be taken when accessing fields
+ * that are not present in struct obd_connect_data_v1. See LU-16.
+ *
* \param exp the obd_export associated with this client/target pair
* \param mdt the target device for the connection
* \param data stores data for this connect request
LASSERT(data != NULL);
data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
+
+ if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
+ data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
+
data->ocd_ibits_known &= MDS_INODELOCK_FULL;
if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
!(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
CWARN("%s: client %s does not support ibits lock, either "
- "very old or an invalid client: flags "LPX64"\n",
+ "very old or an invalid client: flags %#llx\n",
mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
data->ocd_connect_flags);
return -EBADE;
data->ocd_brw_size = min(data->ocd_brw_size,
(__u32)MD_MAX_BRW_SIZE);
if (data->ocd_brw_size == 0) {
- CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
+ CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
" ocd_version: %x ocd_grant: %d "
"ocd_index: %u ocd_brw_size is "
"unexpectedly zero, network data "
ENTRY;
LASSERT(env != NULL);
+ LASSERT(data != NULL);
+
if (!exp || !obd || !cluuid)
RETURN(-EINVAL);
* XXX: probably not very appropriate method is used now
* at some point we should find a better one
*/
- if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
+ if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
!(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
rc = obd_get_info(env, mdt->mdt_child_exp,
INIT_LIST_HEAD(&med->med_open_head);
spin_lock_init(&med->med_open_lock);
- mutex_init(&med->med_idmap_mutex);
- med->med_idmap = NULL;
spin_lock(&exp->exp_lock);
exp->exp_connecting = 1;
spin_unlock(&exp->exp_lock);
{
ENTRY;
- if (exp_connect_rmtclient(exp))
- mdt_cleanup_idmap(&exp->exp_mdt_data);
-
target_destroy_export(exp);
/* destroy can be called from failed obd_setup, so
* checking uuid is safer than obd_self_export */
* \param[in] info Per-thread common data shared by MDT level handlers.
* \param[in] obj Object to do path lookup of
* \param[in,out] fp User-provided struct to store path information
+ * \param[in] root_fid Root FID of current path should reach
*
* \retval 0 Lookup successful, path information stored in fp
* \retval -EAGAIN Lookup failed, usually because object is being moved
*/
static int mdt_path_current(struct mdt_thread_info *info,
struct mdt_object *obj,
- struct getinfo_fid2path *fp)
+ struct getinfo_fid2path *fp,
+ struct lu_fid *root_fid)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *mdt_obj;
RETURN(-ENOMEM);
ldata.ld_buf = buf;
- ptr = fp->gf_path + fp->gf_pathlen - 1;
+ ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
*ptr = 0;
--ptr;
*tmpfid = fp->gf_fid = *mdt_object_fid(obj);
- /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
- * so checking root_fid can only happen on MDT0. */
- while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
+ while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
struct lu_buf lmv_buf;
+ if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
+ lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
+ GOTO(out, rc = -ENOENT);
+
mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
if (IS_ERR(mdt_obj))
GOTO(out, rc = PTR_ERR(mdt_obj));
/* Pack the name in the end of the buffer */
ptr -= tmpname->ln_namelen;
- if (ptr - 1 <= fp->gf_path)
+ if (ptr - 1 <= fp->gf_u.gf_path)
GOTO(out, rc = -EOVERFLOW);
strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
*(--ptr) = '/';
remote_out:
ptr++; /* skip leading / */
- memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
+ memmove(fp->gf_u.gf_path, ptr,
+ fp->gf_u.gf_path + fp->gf_pathlen - ptr);
out:
RETURN(rc);
* \retval negative errno if there was a problem
*/
static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
- struct getinfo_fid2path *fp)
+ struct getinfo_fid2path *fp, struct lu_fid *root_fid)
{
struct mdt_device *mdt = info->mti_mdt;
int tries = 3;
if (fp->gf_pathlen < 3)
RETURN(-EOVERFLOW);
- if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
- fp->gf_path[0] = '\0';
+ if (root_fid == NULL)
+ root_fid = &mdt->mdt_md_root_fid;
+
+ if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
+ fp->gf_u.gf_path[0] = '\0';
RETURN(0);
}
/* Retry multiple times in case file is being moved */
while (tries-- && rc == -EAGAIN)
- rc = mdt_path_current(info, obj, fp);
+ rc = mdt_path_current(info, obj, fp, root_fid);
RETURN(rc);
}
* \retval negative errno if there was a problem
*/
static int mdt_fid2path(struct mdt_thread_info *info,
+ struct lu_fid *root_fid,
struct getinfo_fid2path *fp)
{
struct mdt_device *mdt = info->mti_mdt;
int rc;
ENTRY;
- CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
+ CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
if (!fid_is_sane(&fp->gf_fid))
RETURN(-EINVAL);
if (!fid_is_namespace_visible(&fp->gf_fid)) {
- CWARN("%s: "DFID" is invalid, sequence should be "
- ">= "LPX64"\n", mdt_obd_name(mdt),
- PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
+ CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
+ ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
+ PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
RETURN(-EINVAL);
}
RETURN(rc);
}
- rc = mdt_path(info, obj, fp);
+ rc = mdt_path(info, obj, fp, root_fid);
- CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
- PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
+ CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
+ PFID(&fp->gf_fid), fp->gf_u.gf_path,
+ fp->gf_recno, fp->gf_linkno);
mdt_object_put(info->mti_env, obj);
RETURN(rc);
}
-static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
+static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
void *val, int vallen)
{
struct getinfo_fid2path *fpout, *fpin;
+ struct lu_fid *root_fid = NULL;
int rc = 0;
fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
if (fpout->gf_pathlen != vallen - sizeof(*fpin))
RETURN(-EINVAL);
- rc = mdt_fid2path(info, fpout);
+ if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
+ sizeof(struct lu_fid)) {
+ /* client sent its root FID, which is normally fileset FID */
+ root_fid = fpin->gf_u.gf_root_fid;
+ if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
+ lustre_swab_lu_fid(root_fid);
+
+ if (root_fid != NULL && !fid_is_sane(root_fid))
+ RETURN(-EINVAL);
+ }
+
+ rc = mdt_fid2path(info, root_fid, fpout);
RETURN(rc);
}
if (KEY_IS(KEY_FID2PATH)) {
struct mdt_thread_info *info = tsi2mdt_info(tsi);
- rc = mdt_rpc_fid2path(info, key, valout, *vallen);
+ rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
mdt_thread_info_fini(info);
} else {
rc = -EINVAL;
if (rc)
RETURN(rc);
- switch (cmd) {
- case OBD_IOC_SYNC:
- rc = mdt_device_sync(&env, mdt);
- break;
- case OBD_IOC_SET_READONLY:
- rc = dt->dd_ops->dt_ro(&env, dt);
- break;
+ switch (cmd) {
+ case OBD_IOC_SYNC:
+ rc = mdt_device_sync(&env, mdt);
+ break;
+ case OBD_IOC_SET_READONLY:
+ rc = dt_sync(&env, dt);
+ if (rc == 0)
+ rc = dt_ro(&env, dt);
+ break;
case OBD_IOC_ABORT_RECOVERY:
CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
obd->obd_abort_recovery = 1;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff