#include <lustre_mds.h>
#include <lustre_fid.h>
#include <lustre_lmv.h>
+#include <lustre_idmap.h>
#include "mdd_internal.h"
}
rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
- MOR_TGT_PARENT);
+ DT_TGT_PARENT);
if (rc)
RETURN(rc);
if (rc != 0)
GOTO(lookup, rc);
+ /* the obj is not locked, don't cache attributes */
+ mdd_invalidate(env, &obj->mod_obj);
+
LASSERT(ldata.ld_leh != NULL);
/* Directory should only have 1 parent */
if (ldata.ld_leh->leh_reccount > 1)
iops->put(env, it);
iops->fini(env, it);
- } else
+ } else {
result = PTR_ERR(it);
+ /* -ENODEV means no valid stripe */
+ if (result == -ENODEV)
+ RETURN(0);
+ }
RETURN(result);
}
if (check_perm)
rc = mdd_permission_internal_locked(env, pobj, pattr,
MAY_WRITE | MAY_EXEC,
- MOR_TGT_PARENT);
+ DT_TGT_PARENT);
RETURN(rc);
}
rc = mdd_permission_internal_locked(env, pobj, pattr,
MAY_WRITE | MAY_EXEC,
- MOR_TGT_PARENT);
+ DT_TGT_PARENT);
if (rc != 0)
RETURN(rc);
int rc;
rc = mdd_permission_internal_locked(env, pobj, pattr,
MAY_WRITE | MAY_EXEC,
- MOR_TGT_PARENT);
+ DT_TGT_PARENT);
if (rc)
RETURN(rc);
}
rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
if (rc == 0 && S_ISDIR(type)) {
- mdd_write_lock(env, pobj, MOR_TGT_PARENT);
+ mdd_write_lock(env, pobj, DT_TGT_PARENT);
mdo_ref_add(env, pobj, handle);
mdd_write_unlock(env, pobj);
}
const char *name, int is_dir,
struct thandle *handle)
{
- int rc;
- ENTRY;
+ int rc;
+ ENTRY;
rc = __mdd_index_delete_only(env, pobj, name, handle);
- if (rc == 0 && is_dir) {
- mdd_write_lock(env, pobj, MOR_TGT_PARENT);
- mdo_ref_del(env, pobj, handle);
- mdd_write_unlock(env, pobj);
- }
+ if (rc == 0 && is_dir) {
+ mdd_write_lock(env, pobj, DT_TGT_PARENT);
+ mdo_ref_del(env, pobj, handle);
+ mdd_write_unlock(env, pobj);
+ }
- RETURN(rc);
+ RETURN(rc);
}
static int mdd_llog_record_calc_size(const struct lu_env *env,
return rc;
}
+int mdd_changelog_write_rec(const struct lu_env *env,
+ struct llog_handle *loghandle,
+ struct llog_rec_hdr *r,
+ struct llog_cookie *cookie,
+ int idx, struct thandle *th)
+{
+ int rc;
+
+ if (r->lrh_type == CHANGELOG_REC) {
+ struct mdd_device *mdd;
+ struct llog_changelog_rec *rec;
+
+ mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
+ rec = container_of0(r, struct llog_changelog_rec, cr_hdr);
+
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+
+ rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
+ cookie, idx, th);
+
+ /*
+ * if current llog is full, we will generate a new
+ * llog, and since it's actually not an error, let's
+ * avoid increasing index so that userspace apps
+ * should not see a gap in the changelog sequence
+ */
+ if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ ++mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ }
+ } else {
+ rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
+ cookie, idx, th);
+ }
+
+ return rc;
+}
+
/** Add a changelog entry \a rec to the changelog llog
* \param mdd
* \param rec
rec->cr_hdr.lrh_type = CHANGELOG_REC;
rec->cr.cr_time = cl_time();
- spin_lock(&mdd->mdd_cl.mc_lock);
- /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
- * but as long as the MDD transactions are ordered correctly for e.g.
- * rename conflicts, I don't think this should matter. */
- rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
- spin_unlock(&mdd->mdd_cl.mc_lock);
-
ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
if (ctxt == NULL)
return -ENXIO;
if (IS_ERR(llog_th))
GOTO(out_put, rc = PTR_ERR(llog_th));
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
/* nested journal transaction */
rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
if (rc)
GOTO(stop, rc);
- mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
+ mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
cattr);
if (rc)
struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
int rc;
- if (!S_ISDIR(mdd_object_type(obj)))
- return 0;
-
attr->la_valid = LA_FLAGS;
attr->la_flags = LUSTRE_ORPHAN_FL;
GOTO(stop, rc);
if (likely(mdd_cobj != NULL))
- mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
+ mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
- if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
- le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
+ if (!lmv_magic_supported(le32_to_cpu(lum->lum_magic)) &&
le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
rc = -EINVAL;
CERROR("%s: invalid lmv_user_md: magic = %x, "
if (rc != 0)
RETURN(rc);
- /* sgid check */
+ /* sgid check */
if (pattr->la_mode & S_ISGID) {
+ struct lu_ucred *uc = lu_ucred(env);
+
cattr->la_gid = pattr->la_gid;
+
+ /* Directories are special, and always inherit S_ISGID */
if (S_ISDIR(cattr->la_mode)) {
cattr->la_mode |= S_ISGID;
cattr->la_valid |= LA_MODE;
+ } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
+ == (S_ISGID | S_IXGRP) &&
+ !lustre_in_group_p(uc,
+ (cattr->la_valid & LA_GID) ?
+ cattr->la_gid : pattr->la_gid) &&
+ !md_capable(uc, CFS_CAP_FSETID)) {
+ cattr->la_mode &= ~S_ISGID;
+ cattr->la_valid |= LA_MODE;
}
}
const struct md_op_spec *spec,
struct lu_buf *def_acl_buf,
struct lu_buf *acl_buf,
+ struct lu_buf *hsm_buf,
struct dt_allocation_hint *hint)
{
const struct lu_buf *buf;
if (rc)
GOTO(out, rc);
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
/* if dir, then can inherit default ACl */
rc = mdo_declare_xattr_set(env, c, def_acl_buf,
0, handle);
if (rc)
GOTO(out, rc);
+
+ if (spec->sp_cr_flags & MDS_OPEN_PCC) {
+ rc = mdo_declare_xattr_set(env, c, hsm_buf,
+ XATTR_NAME_HSM,
+ 0, handle);
+ if (rc)
+ GOTO(out, rc);
+ }
}
if (S_ISLNK(attr->la_mode)) {
struct linkea_data *ldata,
struct lu_buf *def_acl_buf,
struct lu_buf *acl_buf,
+ struct lu_buf *hsm_buf,
struct dt_allocation_hint *hint)
{
int rc;
rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
- def_acl_buf, acl_buf, hint);
+ def_acl_buf, acl_buf, hsm_buf, hint);
if (rc)
GOTO(out, rc);
RETURN(0);
}
- mdd_read_lock(env, pobj, MOR_TGT_PARENT);
+ mdd_read_lock(env, pobj, DT_TGT_PARENT);
rc = mdo_xattr_get(env, pobj, def_acl_buf,
XATTR_NAME_ACL_DEFAULT);
mdd_read_unlock(env, pobj);
struct mdd_object *son, struct lu_attr *attr,
struct md_op_spec *spec, struct lu_buf *acl_buf,
struct lu_buf *def_acl_buf,
+ struct lu_buf *hsm_buf,
struct dt_allocation_hint *hint,
- struct thandle *handle)
+ struct thandle *handle, bool initsecctx)
{
const struct lu_buf *buf;
int rc;
- mdd_write_lock(env, son, MOR_TGT_CHILD);
+ mdd_write_lock(env, son, DT_TGT_CHILD);
rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
hint);
if (rc)
GOTO(err_destroy, rc);
}
-#ifdef CONFIG_FS_POSIX_ACL
+ if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
+ struct md_hsm mh;
+
+ memset(&mh, 0, sizeof(mh));
+ mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
+ mh.mh_arch_id = spec->sp_archive_id;
+ lustre_hsm2buf(hsm_buf->lb_buf, &mh);
+ rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
+ 0, handle);
+ if (rc != 0)
+ GOTO(err_destroy, rc);
+ }
+
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
S_ISDIR(attr->la_mode)) {
/* set default acl */
GOTO(err_initlized, rc = -EFAULT);
}
- if (spec->sp_cr_file_secctx_name != NULL) {
+ if (initsecctx && spec->sp_cr_file_secctx_name != NULL) {
buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
spec->sp_cr_file_secctx_size);
rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
* \retval 0 on success
* \retval negative errno on failure
*/
-static int mdd_create(const struct lu_env *env, struct md_object *pobj,
+int mdd_create(const struct lu_env *env, struct md_object *pobj,
const struct lu_name *lname, struct md_object *child,
struct md_op_spec *spec, struct md_attr *ma)
{
struct lu_attr *pattr = &info->mti_pattr;
struct lu_buf acl_buf;
struct lu_buf def_acl_buf;
+ struct lu_buf hsm_buf;
struct linkea_data *ldata = &info->mti_link_data;
const char *name = lname->ln_name;
struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
GOTO(out_free, rc = PTR_ERR(handle));
lu_buf_check_and_alloc(&info->mti_xattr_buf,
- MIN(mdd->mdd_dt_conf.ddp_max_ea_size, XATTR_SIZE_MAX));
+ min_t(unsigned int, mdd->mdd_dt_conf.ddp_max_ea_size,
+ XATTR_SIZE_MAX));
acl_buf = info->mti_xattr_buf;
def_acl_buf.lb_buf = info->mti_key;
def_acl_buf.lb_len = sizeof(info->mti_key);
lname, 1, 0, ldata);
}
+ if (spec->sp_cr_flags & MDS_OPEN_PCC) {
+ LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
+
+ memset(&hsm_buf, 0, sizeof(hsm_buf));
+ lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
+ if (hsm_buf.lb_buf == NULL)
+ GOTO(out_stop, rc = -ENOMEM);
+ }
+
rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
handle, spec, ldata, &def_acl_buf, &acl_buf,
- hint);
+ &hsm_buf, hint);
if (rc)
GOTO(out_stop, rc);
GOTO(out_stop, rc);
rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
- &def_acl_buf, hint, handle);
+ &def_acl_buf, &hsm_buf, hint, handle, true);
if (rc != 0)
GOTO(out_stop, rc);
if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
- mdd_write_lock(env, son, MOR_TGT_CHILD);
+ mdd_write_lock(env, son, DT_TGT_CHILD);
son->mod_flags |= VOLATILE_OBJ;
rc = mdd_orphan_insert(env, son, handle);
GOTO(out_volatile, rc);
goto out_stop;
err_created:
- mdd_write_lock(env, son, MOR_TGT_CHILD);
+ mdd_write_lock(env, son, DT_TGT_CHILD);
if (S_ISDIR(attr->la_mode)) {
/* Drop the reference, no need to delete "."/"..",
* because the object is to be destroyed directly. */
/* if we vmalloced a large buffer drop it */
lu_buf_free(ldata->ld_buf);
+ if (spec->sp_cr_flags & MDS_OPEN_PCC)
+ lu_buf_free(&hsm_buf);
+
/* The child object shouldn't be cached anymore */
if (rc)
set_bit(LU_OBJECT_HEARD_BANSHEE,
GOTO(fixup_tpobj, rc);
/* Update the linkEA for the source object */
- mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
+ mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
mdo2fid(mdd_tpobj), ltname, handle, ldata,
0, 0);
* it must be local one.
*/
if (tobj && mdd_object_exists(mdd_tobj)) {
- mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
+ mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
tobj_locked = 1;
if (mdd_is_dead_obj(mdd_tobj)) {
/* shld not be dead, something is wrong */
ENTRY;
if (!mdd_object_remote(sobj)) {
- mdd_read_lock(env, sobj, MOR_SRC_CHILD);
+ mdd_read_lock(env, sobj, DT_SRC_CHILD);
if (sobj->mod_count > 0) {
CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(sobj)),
if (index < del_offset)
RETURN(0);
- mdd_write_lock(env, stripe, MOR_SRC_CHILD);
+ mdd_write_lock(env, stripe, DT_SRC_CHILD);
rc = __mdd_index_delete_only(env, stripe, dotdot, handle);
if (rc)
GOTO(out, rc);
RETURN(rc);
}
- mdd_write_lock(env, stripe, MOR_SRC_CHILD);
+ mdd_write_lock(env, stripe, DT_SRC_CHILD);
rc = mdo_ref_del(env, stripe, handle);
if (!rc)
rc = mdo_destroy(env, stripe, handle);
for (i = 0; i < le32_to_cpu(lmv->lmv_stripe_count); i++) {
fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[i]);
+ if (!fid_is_sane(fid))
+ continue;
+
stripe = mdd_object_find(env, mdd, fid);
if (IS_ERR(stripe))
RETURN(PTR_ERR(stripe));
struct mdd_thread_info *info = mdd_env_info(env);
char *xname;
struct lu_buf list_xbuf;
+ struct lu_buf cbxbuf;
struct lu_buf xbuf = { NULL };
int list_xsize;
int xlen;
if (rc < 0)
GOTO(out, rc);
+ cbxbuf = xbuf;
+ cbxbuf.lb_len = xsize;
repeat:
- rc = cb(env, tobj, &xbuf, xname, 0, handle);
+ rc = cb(env, tobj, &cbxbuf, xname, 0, handle);
if (unlikely(rc == -ENOSPC &&
strcmp(xname, XATTR_NAME_LINK) == 0)) {
rc = linkea_overflow_shrink(
- (struct linkea_data *)(xbuf.lb_buf));
+ (struct linkea_data *)(cbxbuf.lb_buf));
if (likely(rc > 0)) {
- xbuf.lb_len = rc;
+ cbxbuf.lb_len = rc;
goto repeat;
}
}
RETURN(-ENOENT);
}
- mdd_write_lock(env, pobj, MOR_TGT_PARENT);
+ mdd_write_lock(env, pobj, DT_TGT_PARENT);
rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
if (!rc)
rc = __mdd_index_insert_only(env, pobj, mdo2fid(tobj),
if (rc)
RETURN(rc);
- mdd_write_lock(env, tobj, MOR_TGT_CHILD);
+ mdd_write_lock(env, tobj, DT_TGT_CHILD);
rc = mdo_ref_add(env, tobj, handle);
mdd_write_unlock(env, tobj);
if (rc)
RETURN(rc);
- mdd_write_lock(env, sobj, MOR_SRC_CHILD);
+ mdd_write_lock(env, sobj, DT_SRC_CHILD);
rc = mdo_ref_del(env, sobj, handle);
mdd_write_unlock(env, sobj);
ENTRY;
- mdd_write_lock(env, obj, MOR_SRC_PARENT);
+ mdd_write_lock(env, obj, DT_SRC_PARENT);
if (!lmv_buf->lb_buf)
/* normal dir */
rc = __mdd_index_delete_only(env, obj, dotdot, handle);
rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
lname, attr, handle, spec, ldata, NULL, NULL,
- hint);
+ NULL, hint);
if (rc)
return rc;
* stripes again.
*/
if (sbuf->lb_buf) {
- mdd_write_lock(env, sobj, MOR_SRC_CHILD);
+ mdd_write_lock(env, sobj, DT_SRC_CHILD);
rc = mdo_xattr_del(env, sobj, XATTR_NAME_LMV, handle);
mdd_write_unlock(env, sobj);
if (rc)
/* don't set nlink from sobj */
attr->la_valid &= ~LA_NLINK;
- rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, hint,
- handle);
+ rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
+ hint, handle, false);
if (rc)
RETURN(rc);
- mdd_write_lock(env, tobj, MOR_TGT_CHILD);
+ mdd_write_lock(env, tobj, DT_TGT_CHILD);
rc = mdd_iterate_xattrs(env, sobj, tobj, true, handle, mdo_xattr_set);
mdd_write_unlock(env, tobj);
if (rc)
RETURN(rc);
/* delete LOV to avoid deleting OST objs when destroying sobj */
- mdd_write_lock(env, sobj, MOR_SRC_CHILD);
+ mdd_write_lock(env, sobj, DT_SRC_CHILD);
rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
mdd_write_unlock(env, sobj);
if (rc)
la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
la->la_valid = LA_CTIME | LA_MTIME;
- mdd_write_lock(env, spobj, MOR_SRC_PARENT);
+ mdd_write_lock(env, spobj, DT_SRC_PARENT);
rc = mdd_update_time(env, spobj, spattr, la, handle);
mdd_write_unlock(env, spobj);
if (rc)
if (tpobj != spobj) {
la->la_valid = LA_CTIME | LA_MTIME;
- mdd_write_lock(env, tpobj, MOR_TGT_PARENT);
+ mdd_write_lock(env, tpobj, DT_TGT_PARENT);
rc = mdd_update_time(env, tpobj, tpattr, la, handle);
mdd_write_unlock(env, tpobj);
if (rc)
* message, this won't cause any inconsistency or trouble.
*/
if (do_create && do_destroy) {
- mdd_write_lock(env, sobj, MOR_SRC_CHILD);
+ mdd_write_lock(env, sobj, DT_SRC_CHILD);
mdo_ref_del(env, sobj, handle);
rc = mdo_destroy(env, sobj, handle);
mdd_write_unlock(env, sobj);
if (rc)
GOTO(out, rc);
- mdd_object_make_hint(env, NULL, tobj, attr, spec, hint);
+ mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
lmu->lum_stripe_count = cpu_to_le32(1);
/* delete LMV to avoid deleting stripes again upon destroy */
- mdd_write_lock(env, obj, MOR_SRC_CHILD);
+ mdd_write_lock(env, obj, DT_SRC_CHILD);
rc = mdo_xattr_del(env, obj, XATTR_NAME_LMV, handle);
mdd_write_unlock(env, obj);
if (rc)
}
/* destroy stripes after lmu_stripe_count */
- mdd_write_lock(env, obj, MOR_SRC_PARENT);
+ mdd_write_lock(env, obj, DT_SRC_PARENT);
rc = mdd_dir_iterate_stripes(env, obj, lmv_buf, &shrink_buf, handle,
mdd_dir_destroy_stripe);
mdd_write_unlock(env, obj);
LASSERT(pobj);
LASSERT(stripe);
- mdd_write_lock(env, pobj, MOR_SRC_PARENT);
- mdd_write_lock(env, obj, MOR_SRC_CHILD);
+ mdd_write_lock(env, pobj, DT_SRC_PARENT);
+ mdd_write_lock(env, obj, DT_SRC_CHILD);
/* insert dotdot to stripe which points to parent */
rc = __mdd_index_insert_only(env, stripe, mdo2fid(pobj), S_IFDIR,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff