static inline bool mdt_hsm_is_admin(struct mdt_thread_info *info)
{
bool is_admin;
int rc;
if (info->mti_body == NULL)
return false;
rc = mdt_init_ucred(info, (struct mdt_body *)info->mti_body);
if (rc < 0)
return false;
is_admin = md_capable(mdt_ucred(info), CFS_CAP_SYS_ADMIN);
mdt_exit_ucred(info);
return is_admin;
}
int mdt_reint_setxattr(struct mdt_thread_info *info,
struct mdt_lock_handle *unused)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct lu_ucred *uc = lu_ucred(info->mti_env);
struct mdt_lock_handle *lh;
const struct lu_env *env = info->mti_env;
struct lu_buf *buf = &info->mti_buf;
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct mdt_object *obj;
struct md_object *child;
__u64 valid = attr->la_valid;
const char *xattr_name = rr->rr_name.ln_name;
int xattr_len = rr->rr_eadatalen;
__u64 lockpart;
int rc;
posix_acl_xattr_header *new_xattr = NULL;
__u32 remote = exp_connect_rmtclient(info->mti_exp);
__u32 perm;
ENTRY;
CDEBUG(D_INODE, "setxattr for "DFID"\n", PFID(rr->rr_fid1));
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SETXATTR))
RETURN(err_serious(-ENOMEM));
CDEBUG(D_INODE, "%s xattr %s\n",
valid & OBD_MD_FLXATTR ? "set" : "remove", xattr_name);
rc = mdt_init_ucred_reint(info);
if (rc != 0)
RETURN(rc);
if (valid & OBD_MD_FLRMTRSETFACL) {
if (unlikely(!remote))
GOTO(out, rc = err_serious(-EINVAL));
perm = mdt_identity_get_perm(uc->uc_identity, remote,
req->rq_peer.nid);
if (!(perm & CFS_RMTACL_PERM))
GOTO(out, rc = err_serious(-EPERM));
}
if (strncmp(xattr_name, XATTR_USER_PREFIX,
sizeof(XATTR_USER_PREFIX) - 1) == 0) {
if (!(exp_connect_flags(req->rq_export) & OBD_CONNECT_XATTR))
GOTO(out, rc = -EOPNOTSUPP);
} else if (strncmp(xattr_name, XATTR_TRUSTED_PREFIX,
sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0) {
if (!md_capable(mdt_ucred(info), CFS_CAP_SYS_ADMIN))
GOTO(out, rc = -EPERM);
if (strcmp(xattr_name, XATTR_NAME_LOV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMA) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LINK) == 0 ||
strcmp(xattr_name, XATTR_NAME_FID) == 0 ||
strcmp(xattr_name, XATTR_NAME_VERSION) == 0 ||
strcmp(xattr_name, XATTR_NAME_SOM) == 0 ||
strcmp(xattr_name, XATTR_NAME_HSM) == 0 ||
strcmp(xattr_name, XATTR_NAME_LFSCK_NAMESPACE) == 0)
GOTO(out, rc = 0);
} else if ((valid & OBD_MD_FLXATTR) &&
(strcmp(xattr_name, XATTR_NAME_ACL_ACCESS) == 0 ||
strcmp(xattr_name, XATTR_NAME_ACL_DEFAULT) == 0)) {
/* currently lustre limit acl access size */
if (xattr_len > LUSTRE_POSIX_ACL_MAX_SIZE)
GOTO(out, rc = -ERANGE);
}
lockpart = MDS_INODELOCK_UPDATE;
/* Revoke all clients' lookup lock, since the access
* permissions for this inode is changed when ACL_ACCESS is
* set. This isn't needed for ACL_DEFAULT, since that does
* not change the access permissions of this inode, nor any
* other existing inodes. It is setting the ACLs inherited
* by new directories/files at create time. */
/* We need revoke both LOOKUP|PERM lock here, see mdt_attr_set. */
if (!strcmp(xattr_name, XATTR_NAME_ACL_ACCESS))
lockpart |= MDS_INODELOCK_PERM | MDS_INODELOCK_LOOKUP;
/* We need to take the lock on behalf of old clients so that newer
* clients flush their xattr caches */
else
lockpart |= MDS_INODELOCK_XATTR;
lh = &info->mti_lh[MDT_LH_PARENT];
/* ACLs were sent to clients under LCK_CR locks, so taking LCK_EX
* to cancel them. */
mdt_lock_reg_init(lh, LCK_EX);
obj = mdt_object_find_lock(info, rr->rr_fid1, lh, lockpart);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
tgt_vbr_obj_set(env, mdt_obj2dt(obj));
rc = mdt_version_get_check_save(info, obj, 0);
if (rc)
GOTO(out_unlock, rc);
if (unlikely(!(valid & OBD_MD_FLCTIME))) {
/* This isn't strictly an error, but all current clients
* should set OBD_MD_FLCTIME when setting attributes. */
CWARN("%s: client miss to set OBD_MD_FLCTIME when "
"setxattr %s: [object "DFID"] [valid "LPU64"]\n",
mdt_obd_name(info->mti_mdt), xattr_name,
PFID(rr->rr_fid1), valid);
attr->la_ctime = cfs_time_current_sec();
}
attr->la_valid = LA_CTIME;
child = mdt_object_child(obj);
if (valid & OBD_MD_FLXATTR) {
char *xattr = (void *)rr->rr_eadata;
if (xattr_len > 0) {
int flags = 0;
if (valid & OBD_MD_FLRMTLSETFACL) {
if (unlikely(!remote))
GOTO(out_unlock, rc = -EINVAL);
xattr_len = mdt_rmtlsetfacl(info, child,
xattr_name,
(ext_acl_xattr_header *)xattr,
&new_xattr);
if (xattr_len < 0)
GOTO(out_unlock, rc = xattr_len);
xattr = (char *)new_xattr;
}
if (attr->la_flags & XATTR_REPLACE)
flags |= LU_XATTR_REPLACE;
if (attr->la_flags & XATTR_CREATE)
flags |= LU_XATTR_CREATE;
mdt_fail_write(env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_SETXATTR_WRITE);
buf->lb_buf = xattr;
buf->lb_len = xattr_len;
rc = mo_xattr_set(env, child, buf, xattr_name, flags);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
}
} else if (valid & OBD_MD_FLXATTRRM) {
rc = mo_xattr_del(env, child, xattr_name);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
} else {
CDEBUG(D_INFO, "valid bits: "LPX64"\n", valid);
rc = -EINVAL;
}
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_SETXATTR);
EXIT;
out_unlock:
mdt_object_unlock_put(info, obj, lh, rc);
if (unlikely(new_xattr != NULL))
lustre_posix_acl_xattr_free(new_xattr, xattr_len);
out:
mdt_exit_ucred(info);
return rc;
}
/**
* Change HSM state and archive number of a file.
*
* Archive number is changed iif the value is not 0.
* The new flagset that will be computed should result in a coherent state.
* This function checks that flags are compatible.
*
* This is MDS_HSM_STATE_SET RPC handler.
*/
int mdt_hsm_state_set(struct tgt_session_info *tsi)
{
struct mdt_thread_info *info = tsi2mdt_info(tsi);
struct mdt_object *obj = info->mti_object;
struct md_attr *ma = &info->mti_attr;
struct hsm_state_set *hss;
struct mdt_lock_handle *lh;
int rc;
__u64 flags;
ENTRY;
hss = req_capsule_client_get(info->mti_pill, &RMF_HSM_STATE_SET);
if (info->mti_body == NULL || obj == NULL || hss == NULL)
GOTO(out, rc = -EPROTO);
/* Only valid if client is remote */
rc = mdt_init_ucred(info, (struct mdt_body *)info->mti_body);
if (rc < 0)
GOTO(out, rc = err_serious(rc));
lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lh, LCK_PW);
rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LOOKUP |
MDS_INODELOCK_XATTR);
if (rc < 0)
GOTO(out_ucred, rc);
/* Detect out-of range masks */
if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK) {
CDEBUG(D_HSM, "Incompatible masks provided (set "LPX64
", clear "LPX64") vs supported set (%#x).\n",
hss->hss_setmask, hss->hss_clearmask, HSM_FLAGS_MASK);
GOTO(out_unlock, rc = -EINVAL);
}
/* Non-root users are forbidden to set or clear flags which are
* NOT defined in HSM_USER_MASK. */
if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
!md_capable(mdt_ucred(info), CFS_CAP_SYS_ADMIN)) {
CDEBUG(D_HSM, "Incompatible masks provided (set "LPX64
", clear "LPX64") vs unprivileged set (%#x).\n",
hss->hss_setmask, hss->hss_clearmask, HSM_USER_MASK);
GOTO(out_unlock, rc = -EPERM);
}
/* Read current HSM info */
ma->ma_valid = 0;
ma->ma_need = MA_HSM;
rc = mdt_attr_get_complex(info, obj, ma);
if (rc)
GOTO(out_unlock, rc);
/* Change HSM flags depending on provided masks */
if (hss->hss_valid & HSS_SETMASK)
ma->ma_hsm.mh_flags |= hss->hss_setmask;
if (hss->hss_valid & HSS_CLEARMASK)
ma->ma_hsm.mh_flags &= ~hss->hss_clearmask;
/* Change archive_id if provided. */
if (hss->hss_valid & HSS_ARCHIVE_ID) {
if (!(ma->ma_hsm.mh_flags & HS_EXISTS)) {
CDEBUG(D_HSM, "Could not set an archive number for "
DFID "if HSM EXISTS flag is not set.\n",
PFID(&info->mti_body->mbo_fid1));
GOTO(out_unlock, rc);
}
/* Detect out-of range archive id */
if (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE) {
CDEBUG(D_HSM, "archive id %u exceeds maximum %zu.\n",
hss->hss_archive_id, LL_HSM_MAX_ARCHIVE);
GOTO(out_unlock, rc = -EINVAL);
}
ma->ma_hsm.mh_arch_id = hss->hss_archive_id;
}
/* Check for inconsistant HSM flagset.
* DIRTY without EXISTS: no dirty if no archive was created.
* DIRTY and RELEASED: a dirty file could not be released.
* RELEASED without ARCHIVED: do not release a non-archived file.
* LOST without ARCHIVED: cannot lost a non-archived file.
*/
flags = ma->ma_hsm.mh_flags;
if ((flags & HS_DIRTY && !(flags & HS_EXISTS)) ||
(flags & HS_RELEASED && flags & HS_DIRTY) ||
(flags & HS_RELEASED && !(flags & HS_ARCHIVED)) ||
(flags & HS_LOST && !(flags & HS_ARCHIVED))) {
CDEBUG(D_HSM, "Incompatible flag change on "DFID
"flags="LPX64"\n",
PFID(&info->mti_body->mbo_fid1), flags);
GOTO(out_unlock, rc = -EINVAL);
}
/* Save the modified flags */
rc = mdt_hsm_attr_set(info, obj, &ma->ma_hsm);
if (rc)
GOTO(out_unlock, rc);
EXIT;
out_unlock:
mdt_object_unlock(info, obj, lh, 1);
out_ucred:
mdt_exit_ucred(info);
out:
mdt_thread_info_fini(info);
return rc;
}
int mdt_reint_setxattr(struct mdt_thread_info *info,
struct mdt_lock_handle *unused)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_lock_handle *lh;
const struct lu_env *env = info->mti_env;
struct lu_buf *buf = &info->mti_buf;
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct mdt_object *obj;
struct md_object *child;
__u64 valid = attr->la_valid;
const char *xattr_name = rr->rr_name.ln_name;
int xattr_len = rr->rr_eadatalen;
__u64 lockpart = MDS_INODELOCK_UPDATE;
int rc;
ENTRY;
CDEBUG(D_INODE, "setxattr for "DFID": %s %s\n", PFID(rr->rr_fid1),
valid & OBD_MD_FLXATTR ? "set" : "remove", xattr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SETXATTR))
RETURN(err_serious(-ENOMEM));
rc = mdt_init_ucred_reint(info);
if (rc != 0)
RETURN(rc);
if (strncmp(xattr_name, XATTR_USER_PREFIX,
sizeof(XATTR_USER_PREFIX) - 1) == 0) {
if (!(exp_connect_flags(req->rq_export) & OBD_CONNECT_XATTR))
GOTO(out, rc = -EOPNOTSUPP);
} else if (strncmp(xattr_name, XATTR_TRUSTED_PREFIX,
sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0) {
/* setxattr(LMV) with lum is used to shrink dir layout */
if (strcmp(xattr_name, XATTR_NAME_LMV) == 0) {
__u32 *magic = rr->rr_eadata;
/* we don't let to remove LMV? */
if (!rr->rr_eadata)
GOTO(out, rc = 0);
if (le32_to_cpu(*magic) == LMV_USER_MAGIC ||
le32_to_cpu(*magic) == LMV_USER_MAGIC_SPECIFIC) {
rc = mdt_dir_layout_shrink(info);
GOTO(out, rc);
}
}
if (!md_capable(mdt_ucred(info), CFS_CAP_SYS_ADMIN))
GOTO(out, rc = -EPERM);
if (strcmp(xattr_name, XATTR_NAME_LOV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMA) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LINK) == 0 ||
strcmp(xattr_name, XATTR_NAME_FID) == 0 ||
strcmp(xattr_name, XATTR_NAME_VERSION) == 0 ||
strcmp(xattr_name, XATTR_NAME_SOM) == 0 ||
strcmp(xattr_name, XATTR_NAME_HSM) == 0 ||
strcmp(xattr_name, XATTR_NAME_LFSCK_NAMESPACE) == 0)
GOTO(out, rc = 0);
} else if ((valid & OBD_MD_FLXATTR) &&
(strcmp(xattr_name, XATTR_NAME_ACL_ACCESS) == 0 ||
strcmp(xattr_name, XATTR_NAME_ACL_DEFAULT) == 0)) {
rc = mdt_nodemap_map_acl(info, rr->rr_eadata, xattr_len,
xattr_name, NODEMAP_CLIENT_TO_FS);
if (rc < 0)
GOTO(out, rc);
/* ACLs were mapped out, return an error so the user knows */
if (rc != xattr_len)
GOTO(out, rc = -EPERM);
} else if ((strlen(xattr_name) > strlen(XATTR_LUSTRE_LOV) + 1) &&
strncmp(xattr_name, XATTR_LUSTRE_LOV,
strlen(XATTR_LUSTRE_LOV)) == 0) {
if (strncmp(xattr_name, XATTR_LUSTRE_LOV".add",
strlen(XATTR_LUSTRE_LOV".add")) &&
strncmp(xattr_name, XATTR_LUSTRE_LOV".set",
strlen(XATTR_LUSTRE_LOV".set")) &&
strncmp(xattr_name, XATTR_LUSTRE_LOV".del",
strlen(XATTR_LUSTRE_LOV".del"))) {
CERROR("%s: invalid xattr name: %s\n",
mdt_obd_name(info->mti_mdt), xattr_name);
GOTO(out, rc = -EINVAL);
}
lockpart |= MDS_INODELOCK_LAYOUT;
}
/* Revoke all clients' lookup lock, since the access
* permissions for this inode is changed when ACL_ACCESS is
* set. This isn't needed for ACL_DEFAULT, since that does
* not change the access permissions of this inode, nor any
* other existing inodes. It is setting the ACLs inherited
* by new directories/files at create time. */
/* We need revoke both LOOKUP|PERM lock here, see mdt_attr_set. */
if (!strcmp(xattr_name, XATTR_NAME_ACL_ACCESS))
lockpart |= MDS_INODELOCK_PERM | MDS_INODELOCK_LOOKUP;
/* We need to take the lock on behalf of old clients so that newer
* clients flush their xattr caches */
else
lockpart |= MDS_INODELOCK_XATTR;
lh = &info->mti_lh[MDT_LH_PARENT];
/* ACLs were sent to clients under LCK_CR locks, so taking LCK_EX
* to cancel them. */
mdt_lock_reg_init(lh, LCK_EX);
obj = mdt_object_find_lock(info, rr->rr_fid1, lh, lockpart);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
tgt_vbr_obj_set(env, mdt_obj2dt(obj));
rc = mdt_version_get_check_save(info, obj, 0);
if (rc)
GOTO(out_unlock, rc);
if (unlikely(!(valid & OBD_MD_FLCTIME))) {
/* This isn't strictly an error, but all current clients
* should set OBD_MD_FLCTIME when setting attributes. */
CWARN("%s: client miss to set OBD_MD_FLCTIME when "
"setxattr %s: [object "DFID"] [valid %llu]\n",
mdt_obd_name(info->mti_mdt), xattr_name,
PFID(rr->rr_fid1), valid);
attr->la_ctime = ktime_get_real_seconds();
}
attr->la_valid = LA_CTIME;
child = mdt_object_child(obj);
if (valid & OBD_MD_FLXATTR) {
int flags = 0;
if (attr->la_flags & XATTR_REPLACE)
flags |= LU_XATTR_REPLACE;
if (attr->la_flags & XATTR_CREATE)
flags |= LU_XATTR_CREATE;
mdt_fail_write(env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_SETXATTR_WRITE);
buf->lb_buf = rr->rr_eadata;
buf->lb_len = xattr_len;
rc = mo_xattr_set(env, child, buf, xattr_name, flags);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
} else if (valid & OBD_MD_FLXATTRRM) {
rc = mo_xattr_del(env, child, xattr_name);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
} else {
CDEBUG(D_INFO, "valid bits: %#llx\n", valid);
rc = -EINVAL;
}
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_SETXATTR);
EXIT;
out_unlock:
mdt_object_unlock_put(info, obj, lh, rc);
out:
mdt_exit_ucred(info);
return rc;
}
/* shrink dir layout after migration */
static int mdt_dir_layout_shrink(struct mdt_thread_info *info)
{
const struct lu_env *env = info->mti_env;
struct mdt_device *mdt = info->mti_mdt;
struct lu_ucred *uc = mdt_ucred(info);
struct mdt_reint_record *rr = &info->mti_rr;
struct lmv_user_md *lmu = rr->rr_eadata;
__u32 lum_stripe_count = lmu->lum_stripe_count;
struct lu_buf *buf = &info->mti_buf;
struct lmv_mds_md_v1 *lmv;
struct md_attr *ma = &info->mti_attr;
struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
struct mdt_object *pobj = NULL;
struct mdt_object *obj;
struct mdt_lock_handle *lhp = NULL;
struct mdt_lock_handle *lhc;
int rc;
ENTRY;
if (!mdt->mdt_enable_dir_migration)
RETURN(-EPERM);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
mdt->mdt_enable_remote_dir_gid != -1)
RETURN(-EPERM);
/* mti_big_lmm is used to save LMV, but it may be uninitialized. */
if (unlikely(!info->mti_big_lmm)) {
info->mti_big_lmmsize = lmv_mds_md_size(64, LMV_MAGIC);
OBD_ALLOC(info->mti_big_lmm, info->mti_big_lmmsize);
if (!info->mti_big_lmm)
RETURN(-ENOMEM);
}
obj = mdt_object_find(env, mdt, rr->rr_fid1);
if (IS_ERR(obj))
RETURN(PTR_ERR(obj));
/* get parent from PFID */
rc = mdt_attr_get_pfid(info, obj, &ma->ma_pfid);
if (rc)
GOTO(put_obj, rc);
pobj = mdt_object_find(env, mdt, &ma->ma_pfid);
if (IS_ERR(pobj))
GOTO(put_obj, rc = PTR_ERR(pobj));
/* revoke object remote LOOKUP lock */
if (mdt_object_remote(pobj)) {
rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
if (rc)
GOTO(put_pobj, rc);
}
/*
* lock parent if dir will be shrunk to 1 stripe, because dir will be
* converted to normal directory, as will change dir fid and update
* namespace of parent.
*/
lhp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_reg_init(lhp, LCK_PW);
if (le32_to_cpu(lmu->lum_stripe_count) < 2) {
rc = mdt_reint_object_lock(info, pobj, lhp,
MDS_INODELOCK_UPDATE, true);
if (rc)
GOTO(put_pobj, rc);
}
/* lock object */
lhc = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lhc, LCK_EX);
rc = mdt_reint_striped_lock(info, obj, lhc, MDS_INODELOCK_FULL, einfo,
true);
if (rc)
GOTO(unlock_pobj, rc);
ma->ma_lmv = info->mti_big_lmm;
ma->ma_lmv_size = info->mti_big_lmmsize;
ma->ma_valid = 0;
rc = mdt_stripe_get(info, obj, ma, XATTR_NAME_LMV);
if (rc)
GOTO(unlock_obj, rc);
/* user may run 'lfs migrate' multiple times, so it's shrunk already */
if (!(ma->ma_valid & MA_LMV))
GOTO(unlock_obj, rc = -EALREADY);
lmv = &ma->ma_lmv->lmv_md_v1;
/* ditto */
if (!(le32_to_cpu(lmv->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION))
GOTO(unlock_obj, rc = -EALREADY);
lum_stripe_count = lmu->lum_stripe_count;
if (!lum_stripe_count)
lum_stripe_count = cpu_to_le32(1);
if (lmv->lmv_migrate_offset != lum_stripe_count) {
CERROR("%s: "DFID" migrate mdt count mismatch %u != %u\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
lmv->lmv_migrate_offset, lmu->lum_stripe_count);
GOTO(unlock_obj, rc = -EINVAL);
}
if (lmv->lmv_master_mdt_index != lmu->lum_stripe_offset) {
CERROR("%s: "DFID" migrate mdt index mismatch %u != %u\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
lmv->lmv_master_mdt_index, lmu->lum_stripe_offset);
GOTO(unlock_obj, rc = -EINVAL);
}
if (lum_stripe_count > 1 &&
(lmv->lmv_hash_type & cpu_to_le32(LMV_HASH_TYPE_MASK)) !=
lmu->lum_hash_type) {
CERROR("%s: "DFID" migrate mdt hash mismatch %u != %u\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
lmv->lmv_hash_type, lmu->lum_hash_type);
GOTO(unlock_obj, rc = -EINVAL);
}
buf->lb_buf = rr->rr_eadata;
buf->lb_len = rr->rr_eadatalen;
rc = mo_xattr_set(env, mdt_object_child(obj), buf, XATTR_NAME_LMV, 0);
GOTO(unlock_obj, rc);
unlock_obj:
mdt_reint_striped_unlock(info, obj, lhc, einfo, rc);
unlock_pobj:
mdt_object_unlock(info, pobj, lhp, rc);
put_pobj:
mdt_object_put(env, pobj);
put_obj:
mdt_object_put(env, obj);
return rc;
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
if (mdt_object_obf(mp))
GOTO(put_parent, rc = -EPERM);
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0)
GOTO(unlock_parent, rc);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(unlock_parent, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
-EPERM);
GOTO(unlock_parent, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma);
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
mdt_object_unlock_put(info, mc, child_lh, rc);
GOTO(unlock_parent, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
CERROR("%s: lfs rmdir should not be used on local dir %s\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name);
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc = -EPERM);
}
rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc);
}
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
mdt_object_unlock_put(info, mc, child_lh, rc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}
/*
* VBR: we save three versions in reply:
* 0 - parent. Check that parent version is the same during replay.
* 1 - name. Version of 'name' if file exists with the same name or
* ENOENT_VERSION, it is needed because file may appear due to missed replays.
* 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
* check.
*/
static int mdt_md_create(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *parent;
struct mdt_object *child;
struct mdt_lock_handle *lh;
struct mdt_body *repbody;
struct md_attr *ma = &info->mti_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
lh = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
MDS_INODELOCK_UPDATE);
if (IS_ERR(parent))
RETURN(PTR_ERR(parent));
if (mdt_object_obf(parent))
GOTO(out_put_parent, rc = -EPERM);
rc = mdt_version_get_check_save(info, parent, 0);
if (rc)
GOTO(out_put_parent, rc);
/*
* Check child name version during replay.
* During create replay a file may exist with same name.
*/
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
rc = mdt_lookup_version_check(info, parent, lname,
&info->mti_tmp_fid1, 1);
if (rc == 0)
GOTO(out_put_parent, rc = -EEXIST);
/* -ENOENT is expected here */
if (rc != -ENOENT)
GOTO(out_put_parent, rc);
/* save version of file name for replay, it must be ENOENT here */
mdt_enoent_version_save(info, 1);
child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
if (likely(!IS_ERR(child))) {
struct md_object *next = mdt_object_child(parent);
if (mdt_object_remote(child)) {
struct seq_server_site *ss;
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: Creating remote dir is only "
"permitted for administrator: rc = %d\n",
mdt2obd_dev(mdt)->obd_name, -EPERM);
GOTO(out_put_child, rc = -EPERM);
}
ss = mdt_seq_site(mdt);
if (ss->ss_node_id != 0 &&
mdt->mdt_enable_remote_dir == 0) {
CERROR("%s: remote dir is only permitted on"
" MDT0 or set_param"
" mdt.*.enable_remote_dir=1\n",
mdt2obd_dev(mdt)->obd_name);
GOTO(out_put_child, rc = -EPERM);
}
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
/* capa for cross-ref will be stored here */
ma->ma_capa = req_capsule_server_get(info->mti_pill,
&RMF_CAPA1);
LASSERT(ma->ma_capa);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_CREATE_WRITE);
/* Version of child will be updated on disk. */
info->mti_mos = child;
rc = mdt_version_get_check_save(info, child, 2);
if (rc)
GOTO(out_put_child, rc);
/* Let lower layer know current lock mode. */
info->mti_spec.sp_cr_mode =
mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
/*
* Do not perform lookup sanity check. We know that name does
* not exist.
*/
info->mti_spec.sp_cr_lookup = 0;
info->mti_spec.sp_feat = &dt_directory_features;
rc = mdo_create(info->mti_env, next, lname,
mdt_object_child(child),
&info->mti_spec, ma);
if (rc == 0)
rc = mdt_attr_get_complex(info, child, ma);
if (rc == 0) {
/* Return fid & attr to client. */
if (ma->ma_valid & MA_INODE)
mdt_pack_attr2body(info, repbody, &ma->ma_attr,
mdt_object_fid(child));
}
out_put_child:
mdt_object_put(info->mti_env, child);
} else {
rc = PTR_ERR(child);
}
mdt_create_pack_capa(info, rc, child, repbody);
out_put_parent:
mdt_object_unlock_put(info, parent, lh, rc);
RETURN(rc);
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
int no_name = 0;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
RETURN(-EPERM);
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0) {
/* Name might not be able to find during resend of
* remote unlink, considering following case.
* dir_A is a remote directory, the name entry of
* dir_A is on MDT0, the directory is on MDT1,
*
* 1. client sends unlink req to MDT1.
* 2. MDT1 sends name delete update to MDT0.
* 3. name entry is being deleted in MDT0 synchronously.
* 4. MDT1 is restarted.
* 5. client resends unlink req to MDT1. So it can not
* find the name entry on MDT0 anymore.
* In this case, MDT1 only needs to destory the local
* directory.
* */
if (mdt_object_remote(mp) && rc == -ENOENT &&
!fid_is_zero(rr->rr_fid2) &&
lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
no_name = 1;
*child_fid = *rr->rr_fid2;
} else {
GOTO(unlock_parent, rc);
}
}
if (fid_is_obf(child_fid) || fid_is_dot_lustre(child_fid))
GOTO(unlock_parent, rc = -EPERM);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(put_child, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (!mdt_is_dne_client(req->rq_export))
/* Return -EIO for old client */
GOTO(put_child, rc = -EIO);
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt_obd_name(info->mti_mdt),
-EPERM);
GOTO(put_child, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma, no_name);
GOTO(put_child, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
GOTO(unlock_child, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
rc = -EPERM;
CDEBUG(D_INFO, "%s: no rm_entry on local dir '%s': rc = %d\n",
mdt_obd_name(info->mti_mdt), (char *)rr->rr_name, rc);
GOTO(put_child, rc);
}
/* We used to acquire MDS_INODELOCK_FULL here but we can't do
* this now because a running HSM restore on the child (unlink
* victim) will hold the layout lock. See LU-4002. */
rc = mdt_object_lock(info, mc, child_lh,
MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
MDT_CROSS_LOCK);
if (rc != 0)
GOTO(put_child, rc);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
mutex_lock(&mc->mot_lov_mutex);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma, no_name);
mutex_unlock(&mc->mot_lov_mutex);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
unlock_child:
mdt_object_unlock(info, mc, child_lh, rc);
put_child:
mdt_object_put(info->mti_env, mc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}