*
* 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
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2010, 2014, Intel Corporation.
+ * Copyright (c) 2010, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_MDS
#include <linux/module.h>
-/*
- * struct OBD_{ALLOC,FREE}*()
- */
-#include <obd_support.h>
-#include <lustre_ioctl.h>
-/* struct ptlrpc_request */
-#include <lustre_net.h>
-/* struct obd_export */
-#include <lustre_export.h>
-/* struct obd_device */
-#include <obd.h>
-/* lu2dt_dev() */
+#include <linux/pagemap.h>
+
#include <dt_object.h>
-#include <lustre_mds.h>
-#include <lustre_log.h>
-#include "mdt_internal.h"
#include <lustre_acl.h>
-#include <lustre_param.h>
-#include <lustre_quota.h>
+#include <lustre_export.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
#include <lustre_lfsck.h>
+#include <lustre_log.h>
#include <lustre_nodemap.h>
+#include <lustre_mds.h>
+#include <uapi/linux/lustre/lustre_param.h>
+#include <lustre_quota.h>
+#include <lustre_swab.h>
+#include <obd.h>
+#include <obd_support.h>
+#include <lustre_barrier.h>
+#include <obd_cksum.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,
[LCK_GROUP] = MDL_GROUP
};
-ldlm_mode_t mdt_dlm_lock_modes[] = {
- [MDL_MINMODE] = LCK_MINMODE,
- [MDL_EX] = LCK_EX,
- [MDL_PW] = LCK_PW,
- [MDL_PR] = LCK_PR,
- [MDL_CW] = LCK_CW,
- [MDL_CR] = LCK_CR,
- [MDL_NL] = LCK_NL,
- [MDL_GROUP] = LCK_GROUP
+enum ldlm_mode mdt_dlm_lock_modes[] = {
+ [MDL_MINMODE] = LCK_MINMODE,
+ [MDL_EX] = LCK_EX,
+ [MDL_PW] = LCK_PW,
+ [MDL_PR] = LCK_PR,
+ [MDL_CW] = LCK_CW,
+ [MDL_CR] = LCK_CR,
+ [MDL_NL] = LCK_NL,
+ [MDL_GROUP] = LCK_GROUP
};
static struct mdt_device *mdt_dev(struct lu_device *d);
rep->lock_policy_res1 |= op_flag;
}
-void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
+void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
{
- lh->mlh_pdo_hash = 0;
- lh->mlh_reg_mode = lm;
+ lh->mlh_pdo_hash = 0;
+ lh->mlh_reg_mode = lm;
lh->mlh_rreg_mode = lm;
- lh->mlh_type = MDT_REG_LOCK;
+ lh->mlh_type = MDT_REG_LOCK;
}
-void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
+void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
const struct lu_name *lname)
{
lh->mlh_reg_mode = lock_mode;
+ lh->mlh_pdo_mode = LCK_MINMODE;
lh->mlh_rreg_mode = lock_mode;
lh->mlh_type = MDT_PDO_LOCK;
if (lu_name_is_valid(lname)) {
- lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
- lname->ln_namelen);
+ lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
+ lname->ln_namelen);
/* XXX Workaround for LU-2856
*
* Zero is a valid return value of full_name_hash, but
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;
}
{
struct ptlrpc_request *req = tgt_ses_req(tsi);
struct mdt_thread_info *info = tsi2mdt_info(tsi);
- struct md_device *next = info->mti_mdt->mdt_child;
+ 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;
if (!osfs)
GOTO(out, rc = -EPROTO);
- /** statfs information are cached in the mdt_device */
- if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
- cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
- /** statfs data is too old, get up-to-date one */
- rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
- if (rc)
- GOTO(out, rc);
- spin_lock(&info->mti_mdt->mdt_osfs_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);
- } else {
- /** use cached statfs data */
- spin_lock(&info->mti_mdt->mdt_osfs_lock);
- *osfs = info->mti_mdt->mdt_osfs;
- spin_unlock(&info->mti_mdt->mdt_osfs_lock);
- }
+ rc = tgt_statfs_internal(tsi->tsi_env, &mdt->mdt_lut, osfs,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+ NULL);
+ if (unlikely(rc))
+ GOTO(out, rc);
+ /* at least try to account for cached pages. its still racy and
+ * might be under-reporting if clients haven't announced their
+ * caches with brw recently */
+ CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
+ " pending %llu free %llu avail %llu\n",
+ tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
+ tgd->tgd_tot_pending,
+ osfs->os_bfree << tgd->tgd_blockbits,
+ osfs->os_bavail << tgd->tgd_blockbits);
+
+ osfs->os_bavail -= min_t(u64, osfs->os_bavail,
+ ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
+ osfs->os_bsize - 1) >> tgd->tgd_blockbits));
+
+ tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
+ CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
+ "%llu objects: %llu free; state %x\n",
+ osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
+ osfs->os_files, osfs->os_ffree, osfs->os_state);
+
+ if (!exp_grant_param_supp(tsi->tsi_exp) &&
+ tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
+ /* clients which don't support OBD_CONNECT_GRANT_PARAM
+ * should not see a block size > page size, otherwise
+ * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
+ * block size which is the biggest block size known to work
+ * with all client's page size. */
+ osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
+ osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
+ osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
+ osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
+ }
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_STATFS);
out:
RETURN(rc);
}
+/**
+ * Pack size attributes into the reply.
+ */
+int mdt_pack_size2body(struct mdt_thread_info *info,
+ const struct lu_fid *fid, bool dom_lock)
+{
+ struct mdt_body *b;
+ struct md_attr *ma = &info->mti_attr;
+ int dom_stripe;
+
+ ENTRY;
+
+ LASSERT(ma->ma_attr.la_valid & LA_MODE);
+
+ if (!S_ISREG(ma->ma_attr.la_mode) ||
+ !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
+ RETURN(-ENODATA);
+
+ dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
+ /* no DoM stripe, no size in reply */
+ if (dom_stripe == LMM_NO_DOM)
+ RETURN(-ENOENT);
+
+ /* no DoM lock, no size in reply */
+ if (!dom_lock)
+ RETURN(0);
+
+ /* Either DoM lock exists or LMM has only DoM stripe then
+ * return size on body. */
+ b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
+
+ mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
+ RETURN(0);
+}
+
#ifdef CONFIG_FS_POSIX_ACL
/*
* Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
const struct lu_env *env = info->mti_env;
struct md_object *next = mdt_object_child(o);
struct lu_buf *buf = &info->mti_buf;
+ struct mdt_device *mdt = info->mti_mdt;
int rc;
+ ENTRY;
+
buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
RCL_SERVER);
if (buf->lb_len == 0)
- return 0;
+ RETURN(0);
+again:
rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
if (rc < 0) {
if (rc == -ENODATA) {
} else if (rc == -EOPNOTSUPP) {
rc = 0;
} else {
+ if (rc == -ERANGE &&
+ exp_connect_large_acl(info->mti_exp) &&
+ buf->lb_buf != info->mti_big_acl) {
+ if (info->mti_big_acl == NULL) {
+ OBD_ALLOC_LARGE(info->mti_big_acl,
+ mdt->mdt_max_ea_size);
+ if (info->mti_big_acl == NULL) {
+ 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);
+
+ buf->lb_buf = info->mti_big_acl;
+ buf->lb_len = info->mti_big_aclsize;
+
+ goto again;
+ }
+
CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
- mdt_obd_name(info->mti_mdt),
- PFID(mdt_object_fid(o)), rc);
+ mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
}
} else {
+ if (buf->lb_buf == info->mti_big_acl)
+ info->mti_big_acl_used = 1;
+
rc = nodemap_map_acl(nodemap, buf->lb_buf,
rc, NODEMAP_FS_TO_CLIENT);
/* if all ACLs mapped out, rc is still >= 0 */
if (rc < 0) {
CERROR("%s: nodemap_map_acl unable to parse "DFID
- " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
+ " ACL: rc = %d\n", mdt_obd_name(mdt),
PFID(mdt_object_fid(o)), rc);
} else {
repbody->mbo_aclsize = rc;
rc = 0;
}
}
- return rc;
+
+ RETURN(rc);
}
#endif
+/* XXX Look into layout in MDT layer. */
+static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
+{
+ struct lov_comp_md_v1 *comp_v1;
+ struct lov_mds_md *v1;
+ int i;
+
+ if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
+ comp_v1 = (struct lov_comp_md_v1 *)lmm;
+
+ for (i = 0; i < comp_v1->lcm_entry_count; i++) {
+ v1 = (struct lov_mds_md *)((char *)comp_v1 +
+ comp_v1->lcm_entries[i].lcme_offset);
+ /* We don't support partial release for now */
+ if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
+ return false;
+ }
+ return true;
+ } else {
+ return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
+ true : false;
+ }
+}
+
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;
}
+
+ if (attr->la_valid & LA_PROJID) {
+ /* TODO, nodemap for project id */
+ b->mbo_projid = attr->la_projid;
+ b->mbo_valid |= OBD_MD_FLPROJID;
+ }
+
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 no object is allocated on osts, the size on mds is valid.
* b=22272 */
b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
- } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
- ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
- /* A released file stores its size on MDS. */
- /* But return 1 block for released file, unless tools like tar
- * will consider it fully sparse. (LU-3864)
- */
- if (unlikely(b->mbo_size == 0))
- b->mbo_blocks = 0;
- else
- b->mbo_blocks = 1;
- b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+ } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
+ if (mdt_hsm_is_released(ma->ma_lmm)) {
+ /* A released file stores its size on MDS. */
+ /* But return 1 block for released file, unless tools
+ * like tar will consider it fully sparse. (LU-3864)
+ */
+ if (unlikely(b->mbo_size == 0))
+ b->mbo_blocks = 0;
+ else
+ b->mbo_blocks = 1;
+ b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+ } else if (info->mti_som_valid) { /* som is valid */
+ b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+ }
}
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,
rc = mo_attr_get(env, next, ma);
if (rc)
GOTO(out, rc);
+
+ if (S_ISREG(mode))
+ (void) mdt_get_som(info, o, &ma->ma_attr);
ma->ma_valid |= MA_INODE;
}
#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;
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
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);
repbody->mbo_t_state = MS_RESTORE;
}
- is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
-
- /* the Lustre protocol supposes to return default striping
- * on the user-visible root if explicitly requested */
- if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
- (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
- struct lu_fid rootfid;
- struct mdt_object *root;
- struct mdt_device *mdt = info->mti_mdt;
-
- rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
- if (rc)
- RETURN(rc);
- root = mdt_object_find(env, mdt, &rootfid);
- if (IS_ERR(root))
- RETURN(PTR_ERR(root));
- rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
- mdt_object_put(info->mti_env, root);
- if (unlikely(rc)) {
- CERROR("%s: getattr error for "DFID": rc = %d\n",
- mdt_obd_name(info->mti_mdt),
- PFID(mdt_object_fid(o)), rc);
- RETURN(rc);
- }
- }
-
if (likely(ma->ma_valid & MA_INODE))
mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
else
if (!mdt_is_striped_client(req->rq_export))
RETURN(-ENOTSUPP);
LASSERT(S_ISDIR(la->la_mode));
+ mdt_dump_lmv(D_INFO, ma->ma_lmv);
repbody->mbo_eadatasize = ma->ma_lmv_size;
repbody->mbo_valid |= (OBD_MD_FLDIREA |
OBD_MD_DEFAULT_MEA);
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:
req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
+ /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
+ * by default. If the target object has more ACL entries, then
+ * enlarge the buffer when necessary. */
+ req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
+ LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
+
rc = req_capsule_server_pack(pill);
if (unlikely(rc != 0))
GOTO(out, rc = err_serious(rc));
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);
}
/**
+ * Handler of layout intent RPC requiring the layout modification
+ *
+ * \param[in] info thread environment
+ * \param[in] obj object
+ * \param[in] layout layout change descriptor
+ *
+ * \retval 0 on success
+ * \retval < 0 error code
+ */
+static int mdt_layout_change(struct mdt_thread_info *info,
+ struct mdt_object *obj,
+ struct md_layout_change *layout)
+{
+ struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
+ struct layout_intent *intent = layout->mlc_intent;
+ int rc;
+ ENTRY;
+
+ CDEBUG(D_INFO, "got layout change request from client: "
+ "opc:%u flags:%#x extent "DEXT"\n",
+ intent->li_opc, intent->li_flags, PEXT(&intent->li_extent));
+
+ if (intent->li_extent.e_start >= intent->li_extent.e_end) {
+ CERROR(DFID ":invalid range of layout change "DEXT"\n",
+ PFID(mdt_object_fid(obj)), PEXT(&intent->li_extent));
+ RETURN(-EINVAL);
+ }
+
+ if (!mdt_object_exists(obj))
+ GOTO(out, rc = -ENOENT);
+
+ if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
+ GOTO(out, rc = -EINVAL);
+
+ rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
+ MAY_WRITE);
+ if (rc)
+ GOTO(out, rc);
+
+ /* take layout lock to prepare layout change */
+ mdt_lock_reg_init(lh, LCK_EX);
+ rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
+
+ mdt_object_unlock(info, obj, lh, 1);
+out:
+ RETURN(rc);
+}
+
+/**
* Exchange MOF_LOV_CREATED flags between two objects after a
* layout swap. No assumption is made on whether o1 or o2 have
* created objects or not.
*/
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);
__u64 child_bits,
struct ldlm_reply *ldlm_rep)
{
- struct ptlrpc_request *req = mdt_info_req(info);
- struct mdt_body *reqbody = NULL;
- struct mdt_object *parent = info->mti_object;
- struct mdt_object *child;
- struct lu_fid *child_fid = &info->mti_tmp_fid1;
- struct lu_name *lname = NULL;
- struct mdt_lock_handle *lhp = NULL;
- struct ldlm_lock *lock;
- bool is_resent;
- bool try_layout;
- int ma_need = 0;
- int rc;
+ struct ptlrpc_request *req = mdt_info_req(info);
+ struct mdt_body *reqbody = NULL;
+ struct mdt_object *parent = info->mti_object;
+ struct mdt_object *child;
+ struct lu_fid *child_fid = &info->mti_tmp_fid1;
+ struct lu_name *lname = NULL;
+ struct mdt_lock_handle *lhp = NULL;
+ struct ldlm_lock *lock;
+ __u64 try_bits = 0;
+ bool is_resent;
+ int ma_need = 0;
+ int rc;
+
ENTRY;
is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
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);
}
GOTO(out_parent, rc = PTR_ERR(child));
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
+ if (!mdt_object_exists(child)) {
+ LU_OBJECT_DEBUG(D_INODE, info->mti_env,
+ &child->mot_obj,
+ "Object doesn't exist!");
+ GOTO(out_child, rc = -ENOENT);
+ }
+
rc = mdt_check_resent_lock(info, child, lhc);
if (rc < 0) {
GOTO(out_child, rc);
} else if (rc > 0) {
- mdt_lock_handle_init(lhc);
+ mdt_lock_handle_init(lhc);
mdt_lock_reg_init(lhc, LCK_PR);
- try_layout = false;
-
- if (!mdt_object_exists(child)) {
- LU_OBJECT_DEBUG(D_INODE, info->mti_env,
- &child->mot_obj,
- "Object doesn't exist!\n");
- GOTO(out_child, rc = -ENOENT);
- }
if (!(child_bits & MDS_INODELOCK_UPDATE) &&
mdt_object_exists(child) && !mdt_object_remote(child)) {
- struct md_attr *ma = &info->mti_attr;
+ struct md_attr *ma = &info->mti_attr;
- ma->ma_valid = 0;
- ma->ma_need = MA_INODE;
+ ma->ma_valid = 0;
+ ma->ma_need = MA_INODE;
rc = mdt_attr_get_complex(info, child, ma);
- if (unlikely(rc != 0))
- GOTO(out_child, rc);
+ if (unlikely(rc != 0))
+ GOTO(out_child, rc);
/* If the file has not been changed for some time, we
* return not only a LOOKUP lock, but also an UPDATE
/* layout lock must be granted in a best-effort way
* for IT operations */
LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
- if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
- exp_connect_layout(info->mti_exp) &&
- S_ISREG(lu_object_attr(&child->mot_obj)) &&
+ if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
!mdt_object_remote(child) && ldlm_rep != NULL) {
- /* try to grant layout lock for regular file. */
- try_layout = true;
+ if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
+ exp_connect_layout(info->mti_exp)) {
+ /* try to grant layout lock for regular file. */
+ try_bits = MDS_INODELOCK_LAYOUT;
+ }
+ /* Acquire DOM lock in advance for data-on-mdt file */
+ if (child != parent)
+ try_bits |= MDS_INODELOCK_DOM;
}
- rc = 0;
- if (try_layout) {
- child_bits |= MDS_INODELOCK_LAYOUT;
+ if (try_bits != 0) {
/* try layout lock, it may fail to be granted due to
* contention at LOOKUP or UPDATE */
- if (!mdt_object_lock_try(info, child, lhc,
- child_bits)) {
- child_bits &= ~MDS_INODELOCK_LAYOUT;
- LASSERT(child_bits != 0);
- rc = mdt_object_lock(info, child, lhc,
- child_bits);
- } else {
+ rc = mdt_object_lock_try(info, child, lhc, &child_bits,
+ try_bits, false);
+ if (child_bits & MDS_INODELOCK_LAYOUT)
ma_need |= MA_LOV;
- }
} else {
/* Do not enqueue the UPDATE lock from MDT(cross-MDT),
* client will enqueue the lock to the remote MDT */
"Lock res_id: "DLDLMRES", fid: "DFID"\n",
PLDLMRES(lock->l_resource),
PFID(mdt_object_fid(child)));
+
+ 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,
+ child_bits & MDS_INODELOCK_DOM);
+ 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);
+ }
+
+ GOTO(out_parent, rc = 0);
+ }
}
if (lock)
LDLM_LOCK_PUT(lock);
repbody->mbo_eadatasize = 0;
repbody->mbo_aclsize = 0;
- rc = mdt_init_ucred(info, reqbody);
+ rc = mdt_init_ucred_intent_getattr(info, reqbody);
if (unlikely(rc))
GOTO(out_shrink, rc);
}
static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
- void *karg, void *uarg);
+ void *karg, void __user *uarg);
static int mdt_set_info(struct tgt_session_info *tsi)
{
rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
vallen, val, NULL);
+ } else if (KEY_IS(KEY_EVICT_BY_NID)) {
+ if (vallen > 0)
+ obd_export_evict_by_nid(req->rq_export->exp_obd, val);
} else {
RETURN(-EINVAL);
}
*/
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)
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
+ * by default. If the target object has more ACL entries, then
+ * enlarge the buffer when necessary. */
+ if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
+ 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);
}
/* this should sync this object */
-static int mdt_object_sync(struct mdt_thread_info *info)
+static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
+ struct mdt_object *mo)
{
- struct md_object *next;
int rc;
+
ENTRY;
- if (!mdt_object_exists(info->mti_object)) {
+ if (!mdt_object_exists(mo)) {
CWARN("%s: non existing object "DFID": rc = %d\n",
- mdt_obd_name(info->mti_mdt),
- PFID(mdt_object_fid(info->mti_object)), -ESTALE);
+ exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
+ -ESTALE);
RETURN(-ESTALE);
}
- next = mdt_object_child(info->mti_object);
- rc = mo_object_sync(info->mti_env, next);
+
+ rc = mo_object_sync(env, mdt_object_child(mo));
RETURN(rc);
}
struct mdt_thread_info *info = tsi2mdt_info(tsi);
/* sync an object */
- rc = mdt_object_sync(info);
+ rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
+ info->mti_object);
if (rc == 0) {
const struct lu_fid *fid;
struct lu_attr *la = &info->mti_attr.ma_attr;
RETURN(rc);
}
+static int mdt_data_sync(struct tgt_session_info *tsi)
+{
+ struct mdt_thread_info *info;
+ struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
+ struct ost_body *body = tsi->tsi_ost_body;
+ struct ost_body *repbody;
+ struct mdt_object *mo = NULL;
+ struct md_attr *ma;
+ int rc = 0;
+
+ ENTRY;
+
+ repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
+
+ /* if no fid is specified then do nothing,
+ * device sync is done via MDS_SYNC */
+ if (fid_is_zero(&tsi->tsi_fid))
+ RETURN(0);
+
+ mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
+ if (IS_ERR(mo))
+ RETURN(PTR_ERR(mo));
+
+ rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
+ if (rc)
+ GOTO(put, rc);
+
+ repbody->oa.o_oi = body->oa.o_oi;
+ repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
+
+ info = tsi2mdt_info(tsi);
+ ma = &info->mti_attr;
+ ma->ma_need = MA_INODE;
+ ma->ma_valid = 0;
+ rc = mdt_attr_get_complex(info, mo, ma);
+ if (rc == 0)
+ obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
+ else
+ rc = 0;
+ mdt_thread_info_fini(info);
+
+ EXIT;
+put:
+ if (mo != NULL)
+ mdt_object_put(tsi->tsi_env, mo);
+ return rc;
+}
+
/*
* Handle quota control requests to consult current usage/limit, but also
* to configure quota enforcement
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)
+ switch (oqctl->qc_type) {
+ case USRQUOTA:
id = nodemap_map_id(nodemap, NODEMAP_UID,
NODEMAP_CLIENT_TO_FS, id);
- else if (oqctl->qc_type == GRPQUOTA)
- id = nodemap_map_id(nodemap, NODEMAP_UID,
+ break;
+ case GRPQUOTA:
+ id = nodemap_map_id(nodemap, NODEMAP_GID,
NODEMAP_CLIENT_TO_FS, id);
-
+ break;
+ case PRJQUOTA:
+ /* todo: check/map project id */
+ id = oqctl->qc_id;
+ break;
+ default:
+ GOTO(out_nodemap, rc = -EOPNOTSUPP);
+ }
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_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
+ barrier_exit(tsi->tsi_tgt->lut_bottom);
if (oqctl->qc_id != id)
swap(oqctl->qc_id, id);
+ if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
+ if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
+ RETURN(-EINPROGRESS);
+ }
+
switch (oqctl->qc_cmd) {
case Q_GETINFO:
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;
}
/*
{
struct dt_device *dt = mdt->mdt_bottom;
int rc;
+ ENTRY;
rc = dt->dd_ops->dt_commit_async(env, dt);
if (unlikely(rc != 0))
CWARN("%s: async commit start failed: rc = %d\n",
mdt_obd_name(mdt), rc);
+ atomic_inc(&mdt->mdt_async_commit_count);
+ EXIT;
}
/**
* \see ldlm_blocking_ast_nocheck
*/
int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
- void *data, int flag)
+ void *data, int flag)
{
- struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
- struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
- int rc;
- ENTRY;
+ struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
+ struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+ bool commit_async = false;
+ int rc;
+ ENTRY;
- if (flag == LDLM_CB_CANCELING)
- RETURN(0);
- lock_res_and_lock(lock);
- if (lock->l_blocking_ast != mdt_blocking_ast) {
- unlock_res_and_lock(lock);
- RETURN(0);
- }
- if (mdt_cos_is_enabled(mdt) &&
- lock->l_req_mode & (LCK_PW | LCK_EX) &&
- lock->l_blocking_lock != NULL &&
- lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
- mdt_set_lock_sync(lock);
- }
- rc = ldlm_blocking_ast_nocheck(lock);
+ if (flag == LDLM_CB_CANCELING)
+ RETURN(0);
+
+ lock_res_and_lock(lock);
+ if (lock->l_blocking_ast != mdt_blocking_ast) {
+ 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) {
+ if (mdt_cos_is_enabled(mdt)) {
+ if (lock->l_client_cookie !=
+ lock->l_blocking_lock->l_client_cookie)
+ mdt_set_lock_sync(lock);
+ } else if (mdt_slc_is_enabled(mdt) &&
+ ldlm_is_cos_incompat(lock->l_blocking_lock)) {
+ mdt_set_lock_sync(lock);
+ /*
+ * we may do extra commit here, but there is a small
+ * window to miss a commit: lock was unlocked (saved),
+ * then a conflict lock queued and we come here, but
+ * REP-ACK not received, so lock was not converted to
+ * COS mode yet.
+ * Fortunately this window is quite small, so the
+ * extra commit should be rare (not to say distributed
+ * operation is rare too).
+ */
+ commit_async = true;
+ }
+ } else if (lock->l_req_mode == LCK_COS &&
+ lock->l_blocking_lock != NULL) {
+ commit_async = true;
+ }
- /* 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_COS && lock->l_blocking_lock != NULL) {
- struct lu_env env;
+ rc = ldlm_blocking_ast_nocheck(lock);
+
+ if (commit_async) {
+ struct lu_env env;
rc = lu_env_init(&env, LCT_LOCAL);
if (unlikely(rc != 0))
CWARN("%s: lu_env initialization failed, cannot "
"start asynchronous commit: rc = %d\n",
obd->obd_name, rc);
- else
- mdt_device_commit_async(&env, mdt);
- lu_env_fini(&env);
- }
- RETURN(rc);
+ else
+ mdt_device_commit_async(&env, mdt);
+ lu_env_fini(&env);
+ }
+ RETURN(rc);
}
-/* Used for cross-MDT lock */
+/*
+ * Blocking AST for cross-MDT lock
+ *
+ * Discard lock from uncommitted_slc_locks and cancel it.
+ *
+ * \param lock the lock which blocks a request or cancelling lock
+ * \param desc unused
+ * \param data unused
+ * \param flag indicates whether this cancelling or blocking callback
+ * \retval 0 on success
+ * \retval negative number on error
+ */
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, LCF_ASYNC);
+ rc = ldlm_cli_cancel(&lockh,
+ ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
if (rc < 0) {
CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
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(&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);
}
return 1;
}
-int mdt_remote_object_lock(struct mdt_thread_info *mti,
- struct mdt_object *o, const struct lu_fid *fid,
- struct lustre_handle *lh, ldlm_mode_t mode,
- __u64 ibits, bool nonblock)
+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,
+ __u64 *ibits, __u64 trybits, bool cache)
{
struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
- ldlm_policy_data_t *policy = &mti->mti_policy;
+ union ldlm_policy_data *policy = &mti->mti_policy;
struct ldlm_res_id *res_id = &mti->mti_res_id;
int rc = 0;
ENTRY;
einfo->ei_cb_cp = ldlm_completion_ast;
einfo->ei_enq_slave = 0;
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;
+ policy->l_inodebits.bits = *ibits;
+ policy->l_inodebits.try_bits = trybits;
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 successfully acquired bits to a caller */
+ if (rc == 0) {
+ struct ldlm_lock *lock = ldlm_handle2lock(lh);
+
+ LASSERT(lock);
+ *ibits = lock->l_policy_data.l_inodebits.bits;
+ LDLM_LOCK_PUT(lock);
+ }
RETURN(rc);
}
+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 cache)
+{
+ return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
+ cache);
+}
+
static int mdt_object_local_lock(struct mdt_thread_info *info,
struct mdt_object *o,
- struct mdt_lock_handle *lh, __u64 ibits,
- bool nonblock)
+ struct mdt_lock_handle *lh, __u64 *ibits,
+ __u64 trybits, bool cos_incompat)
{
- struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
- ldlm_policy_data_t *policy = &info->mti_policy;
- struct ldlm_res_id *res_id = &info->mti_res_id;
- __u64 dlmflags;
- int rc;
- ENTRY;
+ 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;
+ int rc;
+ ENTRY;
LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
LASSERT(lh->mlh_type != MDT_NUL_LOCK);
+ if (cos_incompat) {
+ LASSERT(lh->mlh_reg_mode == LCK_PW ||
+ lh->mlh_reg_mode == LCK_EX);
+ dlmflags |= LDLM_FL_COS_INCOMPAT;
+ } else if (mdt_cos_is_enabled(info->mti_mdt)) {
+ dlmflags |= LDLM_FL_COS_ENABLED;
+ }
+
/* Only enqueue LOOKUP lock for remote object */
- if (mdt_object_remote(o))
- LASSERT(ibits == MDS_INODELOCK_LOOKUP);
+ if (mdt_object_remote(o)) {
+ LASSERT(*ibits == MDS_INODELOCK_LOOKUP);
+ }
if (lh->mlh_type == MDT_PDO_LOCK) {
/* check for exists after object is locked */
/* Non-dir object shouldn't have PDO lock */
if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
RETURN(-ENOTDIR);
- }
- }
+ }
+ }
- memset(policy, 0, sizeof(*policy));
- fid_build_reg_res_name(mdt_object_fid(o), res_id);
- dlmflags = LDLM_FL_ATOMIC_CB;
- if (nonblock)
- dlmflags |= LDLM_FL_BLOCK_NOWAIT;
+ fid_build_reg_res_name(mdt_object_fid(o), res_id);
+ dlmflags |= LDLM_FL_ATOMIC_CB;
- /*
- * Take PDO lock on whole directory and build correct @res_id for lock
- * on part of directory.
- */
- if (lh->mlh_pdo_hash != 0) {
- LASSERT(lh->mlh_type == MDT_PDO_LOCK);
- mdt_lock_pdo_mode(info, o, lh);
- if (lh->mlh_pdo_mode != LCK_NL) {
- /*
- * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
- * is never going to be sent to client and we do not
- * want it slowed down due to possible cancels.
- */
- policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ /*
+ * Take PDO lock on whole directory and build correct @res_id for lock
+ * on part of directory.
+ */
+ if (lh->mlh_pdo_hash != 0) {
+ LASSERT(lh->mlh_type == MDT_PDO_LOCK);
+ mdt_lock_pdo_mode(info, o, lh);
+ if (lh->mlh_pdo_mode != LCK_NL) {
+ /*
+ * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
+ * is never going to be sent to client and we do not
+ * want it slowed down due to possible cancels.
+ */
+ policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ policy->l_inodebits.try_bits = 0;
rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
policy, res_id, dlmflags,
info->mti_exp == NULL ? NULL :
res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
}
- policy->l_inodebits.bits = ibits;
+ policy->l_inodebits.bits = *ibits;
+ policy->l_inodebits.try_bits = trybits;
/*
* Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
+ /* Return successfully acquired bits to a caller */
+ if (rc == 0) {
+ struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
+
+ LASSERT(lock);
+ *ibits = lock->l_policy_data.l_inodebits.bits;
+ LDLM_LOCK_PUT(lock);
+ }
RETURN(rc);
}
static int
mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
- struct mdt_lock_handle *lh, __u64 ibits,
- bool nonblock)
+ struct mdt_lock_handle *lh, __u64 *ibits,
+ __u64 trybits, bool cos_incompat)
{
struct mdt_lock_handle *local_lh = NULL;
int rc;
ENTRY;
- if (!mdt_object_remote(o))
- return mdt_object_local_lock(info, o, lh, ibits, nonblock);
+ if (!mdt_object_remote(o)) {
+ rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
+ cos_incompat);
+ RETURN(rc);
+ }
/* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
- ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
- MDS_INODELOCK_XATTR);
+ *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
+ MDS_INODELOCK_XATTR);
/* Only enqueue LOOKUP lock for remote object */
- if (ibits & MDS_INODELOCK_LOOKUP) {
- rc = mdt_object_local_lock(info, o, lh,
- MDS_INODELOCK_LOOKUP,
- nonblock);
+ if (*ibits & MDS_INODELOCK_LOOKUP) {
+ __u64 local = MDS_INODELOCK_LOOKUP;
+
+ rc = mdt_object_local_lock(info, o, lh, &local, 0,
+ cos_incompat);
if (rc != ELDLM_OK)
RETURN(rc);
local_lh = lh;
}
- if (ibits & MDS_INODELOCK_UPDATE) {
+ if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
/* Sigh, PDO needs to enqueue 2 locks right now, but
* enqueue RPC can only request 1 lock, to avoid extra
* RPC, so it will instead enqueue EX lock for remote
lh->mlh_rreg_mode = LCK_EX;
lh->mlh_type = MDT_REG_LOCK;
}
- rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
- &lh->mlh_rreg_lh,
- lh->mlh_rreg_mode,
- MDS_INODELOCK_UPDATE, nonblock);
+
+ rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
+ &lh->mlh_rreg_lh,
+ lh->mlh_rreg_mode,
+ ibits, trybits, false);
if (rc != ELDLM_OK) {
if (local_lh != NULL)
mdt_object_unlock(info, o, local_lh, rc);
int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
struct mdt_lock_handle *lh, __u64 ibits)
{
- return mdt_object_lock_internal(info, o, lh, ibits, false);
+ return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
}
-int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
- struct mdt_lock_handle *lh, __u64 ibits)
+int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
+ struct mdt_lock_handle *lh, __u64 ibits,
+ bool cos_incompat)
{
- struct mdt_lock_handle tmp = *lh;
- int rc;
-
- rc = mdt_object_lock_internal(info, o, &tmp, ibits, true);
- if (rc == 0)
- *lh = tmp;
+ LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
+ return mdt_object_lock_internal(info, o, lh, &ibits, 0,
+ cos_incompat);
+}
- return rc == 0;
+int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
+ struct mdt_lock_handle *lh, __u64 *ibits,
+ __u64 trybits, bool cos_incompat)
+{
+ return mdt_object_lock_internal(info, o, lh, ibits, trybits,
+ cos_incompat);
}
/**
* \param mode lock mode
* \param decref force immediate lock releasing
*/
-static
void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
- ldlm_mode_t mode, int decref)
+ enum ldlm_mode mode, int decref)
{
- ENTRY;
+ ENTRY;
- if (lustre_handle_is_used(h)) {
- if (decref || !info->mti_has_trans ||
- !(mode & (LCK_PW | LCK_EX))){
- mdt_fid_unlock(h, mode);
- } else {
- struct mdt_device *mdt = info->mti_mdt;
- struct ldlm_lock *lock = ldlm_handle2lock(h);
- struct ptlrpc_request *req = mdt_info_req(info);
- int no_ack = 0;
+ if (lustre_handle_is_used(h)) {
+ if (decref || !info->mti_has_trans ||
+ !(mode & (LCK_PW | LCK_EX))) {
+ mdt_fid_unlock(h, mode);
+ } else {
+ struct mdt_device *mdt = info->mti_mdt;
+ struct ldlm_lock *lock = ldlm_handle2lock(h);
+ struct ptlrpc_request *req = mdt_info_req(info);
+ bool cos = mdt_cos_is_enabled(mdt);
+ bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
+
+ LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
+ h->cookie);
- LASSERTF(lock != NULL, "no lock for cookie "LPX64"\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,
- req->rq_reply_state, req->rq_transno);
- if (mdt_cos_is_enabled(mdt)) {
- no_ack = 1;
+ LDLM_DEBUG(lock, "save lock request %p reply "
+ "state %p transno %lld\n", req,
+ req->rq_reply_state, req->rq_transno);
+ if (cos) {
ldlm_lock_downgrade(lock, LCK_COS);
mode = LCK_COS;
}
- ptlrpc_save_lock(req, h, mode, no_ack);
+ if (req->rq_export->exp_disconnected)
+ mdt_fid_unlock(h, mode);
+ else
+ ptlrpc_save_lock(req, h, mode, cos,
+ convert_lock);
} else {
- ldlm_lock_decref(h, mode);
+ mdt_fid_unlock(h, mode);
}
if (mdt_is_lock_sync(lock)) {
CDEBUG(D_HA, "found sync-lock,"
}
/**
+ * Save cross-MDT lock in uncommitted_slc_locks
+ *
+ * Keep the lock referenced until transaction commit happens or release the lock
+ * immediately depending on input parameters.
+ *
+ * \param info thead info object
+ * \param h lock handle
+ * \param mode lock mode
+ * \param decref force immediate lock releasing
+ */
+static void mdt_save_remote_lock(struct mdt_thread_info *info,
+ 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 ptlrpc_request *req = mdt_info_req(info);
+
+ LASSERT(req != NULL);
+ tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
+ req->rq_transno);
+ ldlm_lock_decref(h, mode);
+ }
+ h->cookie = 0ull;
+ }
+
+ EXIT;
+}
+
+/**
* Unlock mdt object.
*
* Immeditely release the regular lock and the PDO lock or save the
* \param o mdt object
* \param lh mdt lock handle referencing regular and PDO locks
* \param decref force immediate lock releasing
+ *
+ * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
*/
void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
- struct mdt_lock_handle *lh, int decref)
+ struct mdt_lock_handle *lh, int decref)
{
- ENTRY;
-
- 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);
+ ENTRY;
- if (lustre_handle_is_used(&lh->mlh_rreg_lh))
- ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
+ 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, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
+ decref);
- EXIT;
+ EXIT;
}
struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
req_capsule_set_size(pill, &RMF_LOGCOOKIES,
RCL_SERVER, 0);
+ /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
+ * by default. If the target object has more ACL entries, then
+ * enlarge the buffer when necessary. */
+ if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
+ req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
+ LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
+
rc = req_capsule_server_pack(pill);
}
RETURN(rc);
info->mti_cross_ref = 0;
info->mti_opdata = 0;
info->mti_big_lmm_used = 0;
+ info->mti_big_acl_used = 0;
+ info->mti_som_valid = 0;
info->mti_spec.no_create = 0;
info->mti_spec.sp_rm_entry = 0;
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_IT_GETXATTR,
MDT_IT_LAYOUT,
MDT_IT_QUOTA,
- MDT_IT_NR
+ MDT_IT_GLIMPSE,
+ MDT_IT_BRW,
+ MDT_IT_NR
};
static int mdt_intent_getattr(enum mdt_it_code opcode,
- struct mdt_thread_info *info,
- struct ldlm_lock **,
- __u64);
+ struct mdt_thread_info *info,
+ struct ldlm_lock **, __u64);
static int mdt_intent_getxattr(enum mdt_it_code opcode,
struct mdt_thread_info *info,
struct mdt_thread_info *info,
struct ldlm_lock **,
__u64);
+static int mdt_intent_glimpse(enum mdt_it_code opcode,
+ struct mdt_thread_info *info,
+ struct ldlm_lock **lockp, __u64 flags)
+{
+ return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
+ lockp, flags);
+}
+static int mdt_intent_brw(enum mdt_it_code opcode,
+ struct mdt_thread_info *info,
+ struct ldlm_lock **lockp, __u64 flags)
+{
+ return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
+ lockp, flags);
+}
static struct mdt_it_flavor {
const struct req_format *it_fmt;
.it_fmt = &RQF_LDLM_INTENT_LAYOUT,
.it_flags = 0,
.it_act = mdt_intent_layout
- }
+ },
+ [MDT_IT_GLIMPSE] = {
+ .it_fmt = &RQF_LDLM_INTENT,
+ .it_flags = 0,
+ .it_act = mdt_intent_glimpse,
+ },
+ [MDT_IT_BRW] = {
+ .it_fmt = &RQF_LDLM_INTENT,
+ .it_flags = 0,
+ .it_act = mdt_intent_brw,
+ },
+
};
-static int
-mdt_intent_lock_replace(struct mdt_thread_info *info,
- struct ldlm_lock **lockp,
- struct mdt_lock_handle *lh,
- __u64 flags)
+int mdt_intent_lock_replace(struct mdt_thread_info *info,
+ struct ldlm_lock **lockp,
+ struct mdt_lock_handle *lh,
+ __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
new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
new_lock->l_blocking_ast = lock->l_blocking_ast;
new_lock->l_completion_ast = lock->l_completion_ast;
+ if (ldlm_has_dom(new_lock))
+ new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
new_lock->l_remote_handle = lock->l_remote_handle;
new_lock->l_flags &= ~LDLM_FL_LOCAL;
RETURN(ELDLM_LOCK_REPLACED);
}
-static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
- struct ldlm_lock *new_lock,
- struct mdt_lock_handle *lh,
- __u64 flags)
+void mdt_intent_fixup_resent(struct mdt_thread_info *info,
+ struct ldlm_lock *new_lock,
+ struct mdt_lock_handle *lh, __u64 flags)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct ldlm_request *dlmreq;
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 = grc;
+ ldlm_rep->lock_policy_res2 = clear_serious(rc);
- return rc;
+ /* 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
+
+ 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);
__u64 flags)
{
struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
- struct layout_intent *layout;
+ struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
+ struct layout_intent *intent;
struct lu_fid *fid;
struct mdt_object *obj = NULL;
+ int layout_size = 0;
int rc = 0;
ENTRY;
RETURN(-EINVAL);
}
+ intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
+ if (intent == NULL)
+ RETURN(-EPROTO);
+
+ switch (intent->li_opc) {
+ case LAYOUT_INTENT_TRUNC:
+ case LAYOUT_INTENT_WRITE:
+ layout.mlc_opc = MD_LAYOUT_WRITE;
+ layout.mlc_intent = intent;
+ break;
+ case LAYOUT_INTENT_ACCESS:
+ break;
+ case LAYOUT_INTENT_READ:
+ case LAYOUT_INTENT_GLIMPSE:
+ case LAYOUT_INTENT_RELEASE:
+ case LAYOUT_INTENT_RESTORE:
+ CERROR("%s: Unsupported layout intent opc %d\n",
+ mdt_obd_name(info->mti_mdt), intent->li_opc);
+ rc = -ENOTSUPP;
+ break;
+ default:
+ CERROR("%s: Unknown layout intent opc %d\n",
+ mdt_obd_name(info->mti_mdt), intent->li_opc);
+ rc = -EINVAL;
+ break;
+ }
+ if (rc < 0)
+ RETURN(rc);
+
fid = &info->mti_tmp_fid2;
fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
if (IS_ERR(obj))
- RETURN(PTR_ERR(obj));
+ GOTO(out, rc = PTR_ERR(obj));
if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
- /* get the length of lsm */
- rc = mdt_attr_get_eabuf_size(info, obj);
- if (rc < 0) {
- mdt_object_put(info->mti_env, obj);
- RETURN(rc);
- }
+ layout_size = mdt_attr_get_eabuf_size(info, obj);
+ if (layout_size < 0)
+ GOTO(out_obj, rc = layout_size);
- if (rc > info->mti_mdt->mdt_max_mdsize)
- info->mti_mdt->mdt_max_mdsize = rc;
+ if (layout_size > info->mti_mdt->mdt_max_mdsize)
+ info->mti_mdt->mdt_max_mdsize = layout_size;
}
- mdt_object_put(info->mti_env, obj);
-
+ /*
+ * set reply buffer size, so that ldlm_handle_enqueue0()->
+ * ldlm_lvbo_fill() will fill the reply buffer with lovea.
+ */
(*lockp)->l_lvb_type = LVB_T_LAYOUT;
- req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
+ req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
+ layout_size);
rc = req_capsule_server_pack(info->mti_pill);
- if (rc != 0)
- RETURN(-EINVAL);
+ if (rc)
+ GOTO(out_obj, rc);
- if (lustre_handle_is_used(&lhc->mlh_reg_lh))
- rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
- layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
- LASSERT(layout != NULL);
- if (layout->li_opc == LAYOUT_INTENT_ACCESS)
- /* return to normal/resent ldlm handling */
- RETURN(rc);
+ if (layout.mlc_opc != MD_LAYOUT_NOP) {
+ struct lu_buf *buf = &layout.mlc_buf;
+
+ /**
+ * mdt_layout_change is a reint operation, when the request
+ * is resent, layout write shouldn't reprocess it again.
+ */
+ rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
+ if (rc)
+ GOTO(out_obj, rc = rc < 0 ? rc : 0);
+
+ /**
+ * There is another resent case: the client's job has been
+ * done by another client, referring lod_declare_layout_change
+ * -EALREADY case, and it became a operation w/o transaction,
+ * so we should not do the layout change, otherwise
+ * mdt_layout_change() will try to cancel the granted server
+ * CR lock whose remote counterpart is still in hold on the
+ * client, and a deadlock ensues.
+ */
+ rc = mdt_check_resent_lock(info, obj, lhc);
+ if (rc <= 0)
+ GOTO(out_obj, rc);
+
+ buf->lb_buf = NULL;
+ buf->lb_len = 0;
+ if (unlikely(req_is_replay(mdt_info_req(info)))) {
+ buf->lb_buf = req_capsule_client_get(info->mti_pill,
+ &RMF_EADATA);
+ buf->lb_len = req_capsule_get_size(info->mti_pill,
+ &RMF_EADATA, RCL_CLIENT);
+ /*
+ * If it's a replay of layout write intent RPC, the
+ * client has saved the extended lovea when
+ * it get reply then.
+ */
+ if (buf->lb_len > 0)
+ mdt_fix_lov_magic(info, buf->lb_buf);
+ }
+
+ /*
+ * Instantiate some layout components, if @buf contains
+ * lovea, then it's a replay of the layout intent write
+ * RPC.
+ */
+ rc = mdt_layout_change(info, obj, &layout);
+ if (rc)
+ GOTO(out_obj, rc);
+ }
+out_obj:
+ 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);
+
+out:
+ lhc->mlh_reg_lh.cookie = 0;
- CERROR("%s: Unsupported layout intent (%d)\n",
- mdt_obd_name(info->mti_mdt), layout->li_opc);
- RETURN(-EINVAL);
+ return rc;
}
static int mdt_intent_reint(enum mdt_it_code opcode,
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);
}
rep->lock_policy_res2 = clear_serious(rc);
if (rep->lock_policy_res2 == -ENOENT &&
- mdt_get_disposition(rep, DISP_LOOKUP_NEG))
- rep->lock_policy_res2 = 0;
+ mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
+ !mdt_get_disposition(rep, DISP_OPEN_CREATE))
+ rep->lock_policy_res2 = 0;
lhc->mlh_reg_lh.cookie = 0ull;
if (rc == -ENOTCONN || rc == -ENODEV ||
RETURN(ELDLM_LOCK_ABORTED);
}
-static int mdt_intent_code(long itcode)
+static int mdt_intent_code(enum ldlm_intent_flags itcode)
{
- int rc;
+ int rc;
- switch(itcode) {
- case IT_OPEN:
- rc = MDT_IT_OPEN;
- break;
- case IT_OPEN|IT_CREAT:
- rc = MDT_IT_OCREAT;
- break;
- case IT_CREAT:
- rc = MDT_IT_CREATE;
- break;
- case IT_READDIR:
- rc = MDT_IT_READDIR;
- break;
- case IT_GETATTR:
- rc = MDT_IT_GETATTR;
- break;
- case IT_LOOKUP:
- rc = MDT_IT_LOOKUP;
- break;
- case IT_UNLINK:
- rc = MDT_IT_UNLINK;
- break;
- case IT_TRUNC:
- rc = MDT_IT_TRUNC;
- break;
- case IT_GETXATTR:
- rc = MDT_IT_GETXATTR;
- break;
- case IT_LAYOUT:
- rc = MDT_IT_LAYOUT;
- break;
+ switch (itcode) {
+ case IT_OPEN:
+ rc = MDT_IT_OPEN;
+ break;
+ case IT_OPEN|IT_CREAT:
+ rc = MDT_IT_OCREAT;
+ break;
+ case IT_CREAT:
+ rc = MDT_IT_CREATE;
+ break;
+ case IT_READDIR:
+ rc = MDT_IT_READDIR;
+ break;
+ case IT_GETATTR:
+ rc = MDT_IT_GETATTR;
+ break;
+ case IT_LOOKUP:
+ rc = MDT_IT_LOOKUP;
+ break;
+ case IT_UNLINK:
+ rc = MDT_IT_UNLINK;
+ break;
+ case IT_TRUNC:
+ rc = MDT_IT_TRUNC;
+ break;
+ case IT_GETXATTR:
+ rc = MDT_IT_GETXATTR;
+ break;
+ case IT_LAYOUT:
+ rc = MDT_IT_LAYOUT;
+ break;
case IT_QUOTA_DQACQ:
case IT_QUOTA_CONN:
rc = MDT_IT_QUOTA;
break;
- default:
- CERROR("Unknown intent opcode: %ld\n", itcode);
- rc = -EINVAL;
- break;
- }
- return rc;
+ case IT_GLIMPSE:
+ rc = MDT_IT_GLIMPSE;
+ break;
+ case IT_BRW:
+ rc = MDT_IT_BRW;
+ break;
+ default:
+ CERROR("Unknown intent opcode: 0x%08x\n", itcode);
+ rc = -EINVAL;
+ break;
+ }
+ return rc;
}
-static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
+static int mdt_intent_opc(enum ldlm_intent_flags itopc,
+ struct mdt_thread_info *info,
struct ldlm_lock **lockp, __u64 flags)
{
struct req_capsule *pill = info->mti_pill;
if (qmt == NULL)
RETURN(-EOPNOTSUPP);
+ if (mdt_rdonly(req->rq_export))
+ RETURN(-EROFS);
+
(*lockp)->l_lvb_type = LVB_T_LQUOTA;
/* pass the request to quota master */
rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
if (rc < 0)
RETURN(rc);
- if (flv->it_flags & MUTABOR &&
- exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
+ if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
RETURN(-EROFS);
if (flv->it_act != NULL) {
struct ldlm_reply *rep;
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
+
/* execute policy */
rc = flv->it_act(opc, info, lockp, flags);
RETURN(rc);
}
+static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
+ enum ldlm_intent_flags it_opc)
+{
+ struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
+
+ /* update stats when IT code is known */
+ if (srv_stats != NULL)
+ lprocfs_counter_incr(srv_stats,
+ PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
+ LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
+}
+
static int mdt_intent_policy(struct ldlm_namespace *ns,
- struct ldlm_lock **lockp, void *req_cookie,
- ldlm_mode_t mode, __u64 flags, void *data)
+ 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;
struct ptlrpc_request *req = req_cookie;
struct ldlm_intent *it;
struct req_capsule *pill;
+ const struct ldlm_lock_desc *ldesc;
int rc;
ENTRY;
tsi = tgt_ses_info(req->rq_svc_thread->t_env);
info = tsi2mdt_info(tsi);
- LASSERT(info != NULL);
- pill = info->mti_pill;
- LASSERT(pill->rc_req == req);
+ LASSERT(info != NULL);
+ pill = info->mti_pill;
+ LASSERT(pill->rc_req == req);
+ ldesc = &info->mti_dlm_req->lock_desc;
- if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
+ if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
- it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
- if (it != NULL) {
- rc = mdt_intent_opc(it->opc, info, lockp, flags);
- if (rc == 0)
- rc = ELDLM_OK;
-
- /* Lock without inodebits makes no sense and will oops
- * later in ldlm. Let's check it now to see if we have
- * ibits corrupted somewhere in mdt_intent_opc().
- * The case for client miss to set ibits has been
- * processed by others. */
- LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
- lr_type == LDLM_IBITS,
- info->mti_dlm_req->lock_desc.\
- l_policy_data.l_inodebits.bits != 0));
- } else
- rc = err_serious(-EFAULT);
- } else {
- /* No intent was provided */
- LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
+ it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
+ if (it != NULL) {
+ mdt_ptlrpc_stats_update(req, it->opc);
+ rc = mdt_intent_opc(it->opc, info, lockp, flags);
+ if (rc == 0)
+ rc = ELDLM_OK;
+
+ /* Lock without inodebits makes no sense and will oops
+ * later in ldlm. Let's check it now to see if we have
+ * ibits corrupted somewhere in mdt_intent_opc().
+ * The case for client miss to set ibits has been
+ * processed by others. */
+ LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
+ ldesc->l_policy_data.l_inodebits.bits != 0));
+ } else {
+ rc = err_serious(-EFAULT);
+ }
+ } else {
+ /* No intent was provided */
req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
- rc = req_capsule_server_pack(pill);
- if (rc)
- rc = err_serious(rc);
- }
+ rc = req_capsule_server_pack(pill);
+ if (rc)
+ rc = err_serious(rc);
+ }
mdt_thread_info_fini(info);
RETURN(rc);
}
* by osd only doesn't have mdt/mdd slices -bzzz */
lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
lustre_cfg_bufs_set_string(bufs, 1, NULL);
- lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
- if (lcfg == NULL)
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+ if (!lcfg)
RETURN_EXIT;
+ lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
top->ld_ops->ldo_process_config(env, top, lcfg);
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
EXIT;
}
if (obd->obd_fail)
strcat(flags, "A");
lustre_cfg_bufs_set_string(bufs, 1, flags);
- lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
- if (lcfg == NULL)
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+ if (!lcfg)
RETURN_EXIT;
+ lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
LASSERT(top);
top->ld_ops->ldo_process_config(env, top, lcfg);
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
lu_site_purge(env, top->ld_site, -1);
lustre_cfg_bufs_set_string(bufs, 2, uuid);
lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
- lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
- if (lcfg == NULL)
- GOTO(free_bufs, rc = -ENOMEM);
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+ if (!lcfg)
+ GOTO(put_profile, rc = -ENOMEM);
+ lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
rc = class_attach(lcfg);
if (rc)
GOTO(lcfg_cleanup, rc = -EINVAL);
}
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
lustre_cfg_bufs_reset(bufs, name);
lustre_cfg_bufs_set_string(bufs, 1, uuid);
lustre_cfg_bufs_set_string(bufs, 2, dev);
lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
- lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
- if (lcfg == NULL)
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+ if (!lcfg)
GOTO(class_detach, rc = -ENOMEM);
+ lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
rc = class_setup(obd, lcfg);
if (rc)
if (rc)
class_detach(obd, lcfg);
lcfg_cleanup:
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
+put_profile:
+ class_put_profile(lprof);
free_bufs:
OBD_FREE_PTR(bufs);
cleanup_mem:
lustre_cfg_bufs_set_string(bufs, 2, uuid);
lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
- lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
- if (lcfg == NULL)
- GOTO(cleanup_mem, rc = -ENOMEM);
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+ if (!lcfg)
+ GOTO(put_profile, rc = -ENOMEM);
+ lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
rc = class_attach(lcfg);
if (rc)
GOTO(lcfg_cleanup, rc = -EINVAL);
}
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
lustre_cfg_bufs_reset(bufs, qmtname);
lustre_cfg_bufs_set_string(bufs, 1, uuid);
lustre_cfg_bufs_set_string(bufs, 3,
mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
- lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
- if (lcfg == NULL)
+ OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+ if (!lcfg)
GOTO(class_detach, rc = -ENOMEM);
+ lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
rc = class_setup(obd, lcfg);
if (rc)
if (rc)
class_detach(obd, lcfg);
lcfg_cleanup:
- lustre_cfg_free(lcfg);
+ OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
+put_profile:
+ class_put_profile(lprof);
cleanup_mem:
if (bufs)
OBD_FREE_PTR(bufs);
return rc;
}
+#define OBD_FAIL_OST_READ_NET OBD_FAIL_OST_BRW_NET
+#define OBD_FAIL_OST_WRITE_NET OBD_FAIL_OST_BRW_NET
+#define OST_BRW_READ OST_READ
+#define OST_BRW_WRITE OST_WRITE
+
static struct tgt_handler mdt_tgt_handlers[] = {
TGT_RPC_HANDLER(MDS_FIRST_OPC,
0, MDS_CONNECT, mdt_tgt_connect,
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),
mdt_swap_layouts),
};
+static struct tgt_handler mdt_io_ops[] = {
+TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read),
+TGT_OST_HDL(HABEO_CORPUS | MUTABOR, OST_BRW_WRITE, tgt_brw_write),
+TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
+ OST_PUNCH, mdt_punch_hdl),
+TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC, mdt_data_sync),
+};
+
static struct tgt_handler mdt_sec_ctx_ops[] = {
TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
.tos_opc_end = LFSCK_LAST_OPC,
.tos_hs = tgt_lfsck_handlers
},
-
+ {
+ .tos_opc_start = OST_FIRST_OPC,
+ .tos_opc_end = OST_LAST_OPC,
+ .tos_hs = mdt_io_ops
+ },
{
.tos_hs = NULL
}
static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
{
- struct md_device *next = m->mdt_child;
- struct lu_device *d = &m->mdt_lu_dev;
- struct obd_device *obd = mdt2obd_dev(m);
- struct lfsck_stop stop;
- ENTRY;
+ struct md_device *next = m->mdt_child;
+ struct lu_device *d = &m->mdt_lu_dev;
+ struct obd_device *obd = mdt2obd_dev(m);
+ struct lfsck_stop stop;
+ ENTRY;
stop.ls_status = LS_PAUSED;
stop.ls_flags = 0;
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)
- mdt_hsm_cdt_stop(m);
-
- mdt_hsm_cdt_fini(m);
+ /* Remove the HSM /proc entry so the coordinator cannot be
+ * restarted by a user while it's shutting down. */
+ hsm_cdt_procfs_fini(m);
+ mdt_hsm_cdt_stop(m);
mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
- mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
+ mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
if (m->mdt_namespace != NULL)
ldlm_namespace_free_prior(m->mdt_namespace, NULL,
d->ld_obd->obd_force);
- obd_exports_barrier(obd);
- obd_zombie_barrier();
+ obd_exports_barrier(obd);
+ obd_zombie_barrier();
+
+ mdt_quota_fini(env, m);
+
+ cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
+
+ /* Calling the cleanup functions in the same order as in the mdt_init0
+ * error path
+ */
+ mdt_procfs_fini(m);
+
+ target_recovery_fini(obd);
+ upcall_cache_cleanup(m->mdt_identity_cache);
+ m->mdt_identity_cache = NULL;
+
+ mdt_fs_cleanup(env, m);
- mdt_procfs_fini(m);
+ tgt_fini(env, &m->mdt_lut);
- tgt_fini(env, &m->mdt_lut);
- mdt_fs_cleanup(env, m);
- upcall_cache_cleanup(m->mdt_identity_cache);
- m->mdt_identity_cache = NULL;
+ mdt_hsm_cdt_fini(m);
if (m->mdt_namespace != NULL) {
ldlm_namespace_free_post(m->mdt_namespace);
d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
}
- mdt_quota_fini(env, m);
+ if (m->mdt_md_root != NULL) {
+ mdt_object_put(env, m->mdt_md_root);
+ m->mdt_md_root = NULL;
+ }
- cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
+ mdt_seq_fini(env, m);
- mdt_seq_fini(env, m);
- mdt_fld_fini(env, m);
+ mdt_fld_fini(env, m);
/*
* Finish the stack
static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
- struct lu_device_type *ldt, struct lustre_cfg *cfg)
-{
- struct mdt_thread_info *info;
- struct obd_device *obd;
- const char *dev = lustre_cfg_string(cfg, 0);
- const char *num = lustre_cfg_string(cfg, 2);
- struct lustre_mount_info *lmi = NULL;
- struct lustre_sb_info *lsi;
- struct lu_site *s;
- struct seq_server_site *ss_site;
- const char *identity_upcall = "NONE";
- struct md_device *next;
- int rc;
- long node_id;
- mntopt_t mntopts;
- ENTRY;
+ struct lu_device_type *ldt, struct lustre_cfg *cfg)
+{
+ const struct dt_device_param *dt_conf;
+ struct mdt_thread_info *info;
+ struct obd_device *obd;
+ const char *dev = lustre_cfg_string(cfg, 0);
+ const char *num = lustre_cfg_string(cfg, 2);
+ struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
+ struct lustre_mount_info *lmi = NULL;
+ struct lustre_sb_info *lsi;
+ struct lu_site *s;
+ struct seq_server_site *ss_site;
+ const char *identity_upcall = "NONE";
+ struct md_device *next;
+ int rc;
+ long node_id;
+ mntopt_t mntopts;
+ ENTRY;
lu_device_init(&m->mdt_lu_dev, ldt);
- /*
- * Environment (env) might be missing mdt_thread_key values at that
- * point, if device is allocated when mdt_thread_key is in QUIESCENT
- * mode.
- *
- * Usually device allocation path doesn't use module key values, but
- * mdt has to do a lot of work here, so allocate key value.
- */
- rc = lu_env_refill((struct lu_env *)env);
- if (rc != 0)
- RETURN(rc);
-
- info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
- LASSERT(info != NULL);
-
- obd = class_name2obd(dev);
- LASSERT(obd != NULL);
+ /*
+ * Environment (env) might be missing mdt_thread_key values at that
+ * point, if device is allocated when mdt_thread_key is in QUIESCENT
+ * mode.
+ *
+ * Usually device allocation path doesn't use module key values, but
+ * mdt has to do a lot of work here, so allocate key value.
+ */
+ rc = lu_env_refill((struct lu_env *)env);
+ if (rc != 0)
+ RETURN(rc);
- m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
+ info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
+ LASSERT(info != NULL);
- m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
+ obd = class_name2obd(dev);
+ LASSERT(obd != NULL);
- /* default is coordinator off, it is started through conf_param
- * or /proc */
- m->mdt_opts.mo_coordinator = 0;
+ m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
+ m->mdt_opts.mo_evict_tgt_nids = 1;
+ m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
lmi = server_get_mount(dev);
- if (lmi == NULL) {
- CERROR("Cannot get mount info for %s!\n", dev);
- RETURN(-EFAULT);
- } else {
- lsi = s2lsi(lmi->lmi_sb);
- /* CMD is supported only in IAM mode */
- LASSERT(num);
- node_id = simple_strtol(num, NULL, 10);
+ if (lmi == NULL) {
+ CERROR("Cannot get mount info for %s!\n", dev);
+ RETURN(-EFAULT);
+ } else {
+ lsi = s2lsi(lmi->lmi_sb);
+ /* CMD is supported only in IAM mode */
+ LASSERT(num);
+ node_id = simple_strtol(num, NULL, 10);
obd->u.obt.obt_magic = OBT_MAGIC;
+ if (lsi->lsi_lmd != NULL &&
+ lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
+ m->mdt_skip_lfsck = 1;
}
+ /* DoM files get IO lock at open by default */
+ m->mdt_opts.mo_dom_lock = 1;
+
m->mdt_squash.rsi_uid = 0;
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);
- m->mdt_osfs_age = cfs_time_shift_64(-1000);
+ spin_lock_init(&m->mdt_lock);
m->mdt_enable_remote_dir = 0;
m->mdt_enable_remote_dir_gid = 0;
+ atomic_set(&m->mdt_mds_mds_conns, 0);
+ atomic_set(&m->mdt_async_commit_count, 0);
+
m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
m->mdt_lu_dev.ld_obd = obd;
/* Set this lu_device to obd for error handling purposes. */
s->ld_seq_site = ss_site;
ss_site->ss_lu = s;
- /* set server index */
+ /* set server index */
ss_site->ss_node_id = node_id;
/* failover is the default
* FIXME: we do not failout mds0/mgs, which may cause some problems.
* assumed whose ss_node_id == 0 XXX
* */
- obd->obd_replayable = 1;
- /* No connection accepted until configurations will finish */
- obd->obd_no_conn = 1;
+ obd->obd_replayable = 1;
+ /* No connection accepted until configurations will finish */
+ obd->obd_no_conn = 1;
if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
char *str = lustre_cfg_string(cfg, 4);
snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
- m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
- LDLM_NAMESPACE_SERVER,
- LDLM_NAMESPACE_GREEDY,
- LDLM_NS_TYPE_MDT);
- if (m->mdt_namespace == NULL)
- GOTO(err_fini_seq, rc = -ENOMEM);
+ m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
+ LDLM_NAMESPACE_SERVER,
+ LDLM_NAMESPACE_GREEDY,
+ LDLM_NS_TYPE_MDT);
+ if (m->mdt_namespace == NULL)
+ GOTO(err_fini_seq, rc = -ENOMEM);
m->mdt_namespace->ns_lvbp = m;
m->mdt_namespace->ns_lvbo = &mdt_lvbo;
- ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
- /* set obd_namespace for compatibility with old code */
- obd->obd_namespace = m->mdt_namespace;
+ ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
+ /* set obd_namespace for compatibility with old code */
+ obd->obd_namespace = m->mdt_namespace;
rc = mdt_hsm_cdt_init(m);
if (rc != 0) {
CERROR("%s: error initializing coordinator, rc %d\n",
mdt_obd_name(m), rc);
- GOTO(err_free_ns, rc);
+ GOTO(err_free_ns, rc);
}
rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
if (rc)
GOTO(err_free_hsm, rc);
+ /* Amount of available space excluded from granting and reserved
+ * for metadata. It is in percentage and 50% is default value. */
+ tgd->tgd_reserved_pcnt = 50;
+
+ if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
+ m->mdt_brw_size = 1U << tgd->tgd_blockbits;
+ else
+ m->mdt_brw_size = ONE_MB_BRW_SIZE;
+
rc = mdt_fs_setup(env, m, obd, lsi);
if (rc)
GOTO(err_tgt, rc);
- tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
+ tgt_adapt_sptlrpc_conf(&m->mdt_lut);
- next = m->mdt_child;
- rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
- &mntopts);
- if (rc)
- GOTO(err_fs_cleanup, rc);
+ next = m->mdt_child;
+ dt_conf = next->md_ops->mdo_dtconf_get(env, next);
- if (mntopts & MNTOPT_USERXATTR)
- m->mdt_opts.mo_user_xattr = 1;
- else
- m->mdt_opts.mo_user_xattr = 0;
+ mntopts = dt_conf->ddp_mntopts;
- rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
- if (rc)
- GOTO(err_fs_cleanup, rc);
+ if (mntopts & MNTOPT_USERXATTR)
+ m->mdt_opts.mo_user_xattr = 1;
+ else
+ m->mdt_opts.mo_user_xattr = 0;
- if (mntopts & MNTOPT_ACL)
- m->mdt_opts.mo_acl = 1;
- else
- m->mdt_opts.mo_acl = 0;
+ m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
+
+ if (mntopts & MNTOPT_ACL)
+ m->mdt_opts.mo_acl = 1;
+ else
+ m->mdt_opts.mo_acl = 0;
/* XXX: to support suppgid for ACL, we enable identity_upcall
* by default, otherwise, maybe got unexpected -EACCESS. */
GOTO(err_fs_cleanup, rc);
}
- rc = mdt_procfs_init(m, dev);
- if (rc) {
- CERROR("Can't init MDT lprocfs, rc %d\n", rc);
- GOTO(err_recovery, rc);
- }
+ rc = mdt_procfs_init(m, dev);
+ if (rc) {
+ CERROR("Can't init MDT lprocfs, rc %d\n", rc);
+ GOTO(err_recovery, rc);
+ }
rc = mdt_quota_init(env, m, cfg);
if (rc)
* when the whole stack is complete and ready
* to serve the requests */
- /* Reduce the initial timeout on an MDS because it doesn't need such
- * a long timeout as an OST does. Adaptive timeouts will adjust this
- * value appropriately. */
- if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
- ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
+ /* Reduce the initial timeout on an MDS because it doesn't need such
+ * a long timeout as an OST does. Adaptive timeouts will adjust this
+ * value appropriately. */
+ if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
+ ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
- RETURN(0);
+ RETURN(0);
err_procfs:
mdt_procfs_fini(m);
err_recovery:
{ "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 }
cfg);
}
- if (old_cfg != NULL)
- lustre_cfg_free(cfg);
-
+ if (old_cfg)
+ OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
+ cfg->lcfg_buflens));
break;
}
default:
o->lo_ops = &mdt_obj_ops;
spin_lock_init(&mo->mot_write_lock);
mutex_init(&mo->mot_lov_mutex);
+ init_rwsem(&mo->mot_dom_sem);
init_rwsem(&mo->mot_open_sem);
+ atomic_set(&mo->mot_open_count, 0);
RETURN(o);
}
RETURN(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,
ENTRY;
if (KEY_IS(KEY_SPTLRPC_CONF)) {
- rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
+ rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
RETURN(rc);
}
* 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
* \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
* \retval -EBADE client and server feature requirements are incompatible
*/
-static int mdt_connect_internal(struct obd_export *exp,
+static int mdt_connect_internal(const struct lu_env *env,
+ struct obd_export *exp,
struct mdt_device *mdt,
- struct obd_connect_data *data)
+ struct obd_connect_data *data, bool reconnect)
{
LASSERT(data != NULL);
data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
+
+ if (mdt->mdt_bottom->dd_rdonly &&
+ !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
+ !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
+ RETURN(-EACCES);
+
+ 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;
if (!mdt->mdt_opts.mo_user_xattr)
data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
- if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
+ if (OCD_HAS_FLAG(data, BRW_SIZE)) {
data->ocd_brw_size = min(data->ocd_brw_size,
- (__u32)MD_MAX_BRW_SIZE);
+ mdt->mdt_brw_size);
if (data->ocd_brw_size == 0) {
- CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
- " ocd_version: %x ocd_grant: %d "
- "ocd_index: %u ocd_brw_size is "
- "unexpectedly zero, network data "
- "corruption? Refusing connection of this"
- " client\n",
+ CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
+ "ocd_version: %x ocd_grant: %d ocd_index: %u "
+ "ocd_brw_size unexpectedly zero, network data "
+ "corruption? Refusing to connect this client\n",
mdt_obd_name(mdt),
exp->exp_client_uuid.uuid,
exp, data->ocd_connect_flags, data->ocd_version,
}
}
+ if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
+ struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
+
+ /* client is reporting its page size, for future use */
+ exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
+ data->ocd_grant_blkbits = mdt->mdt_lut.lut_tgd.tgd_blockbits;
+ /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
+ * it's LDISKFS_DIR_REC_LEN(20) = 28. */
+ data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
+ /* ocd_grant_tax_kb is in 1K byte blocks */
+ data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
+ data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
+ }
+
+ if (OCD_HAS_FLAG(data, GRANT)) {
+ /* Save connect_data we have so far because tgt_grant_connect()
+ * uses it to calculate grant. */
+ exp->exp_connect_data = *data;
+ tgt_grant_connect(env, exp, data, !reconnect);
+ }
+
+ if (OCD_HAS_FLAG(data, MAXBYTES))
+ data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
+
/* NB: Disregard the rule against updating
* exp_connect_data.ocd_connect_flags in this case, since
* tgt_client_new() needs to know if this is a lightweight
spin_unlock(&exp->exp_lock);
}
+ if (OCD_HAS_FLAG(data, CKSUM)) {
+ __u32 cksum_types = data->ocd_cksum_types;
+
+ /* The client set in ocd_cksum_types the checksum types it
+ * supports. We have to mask off the algorithms that we don't
+ * support */
+ data->ocd_cksum_types &= cksum_types_supported_server();
+
+ if (unlikely(data->ocd_cksum_types == 0)) {
+ CERROR("%s: Connect with checksum support but no "
+ "ocd_cksum_types is set\n",
+ exp->exp_obd->obd_name);
+ RETURN(-EPROTO);
+ }
+
+ CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
+ "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
+ cksum_types, data->ocd_cksum_types);
+ } else {
+ /* This client does not support OBD_CONNECT_CKSUM
+ * fall back to CRC32 */
+ CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
+ "OBD_CONNECT_CKSUM, CRC32 will be used\n",
+ exp->exp_obd->obd_name, obd_export_nid2str(exp));
+ }
+
return 0;
}
RETURN(rc);
}
+static inline void mdt_enable_slc(struct mdt_device *mdt)
+{
+ if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
+ mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
+}
+
+static inline void mdt_disable_slc(struct mdt_device *mdt)
+{
+ if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
+ mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
+}
+
static int mdt_obd_disconnect(struct obd_export *exp)
{
- int rc;
- ENTRY;
+ int rc;
+
+ ENTRY;
- LASSERT(exp);
- class_export_get(exp);
+ LASSERT(exp);
+ class_export_get(exp);
+
+ if (!(exp->exp_flags & OBD_OPT_FORCE))
+ tgt_grant_sanity_check(exp->exp_obd, __func__);
+
+ if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
+ !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
+ struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
+
+ if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
+ mdt_disable_slc(mdt);
+ }
- nodemap_del_member(exp);
rc = server_disconnect_export(exp);
if (rc != 0)
CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
+ tgt_grant_discard(exp);
+
rc = mdt_export_cleanup(exp);
+ nodemap_del_member(exp);
class_export_put(exp);
RETURN(rc);
}
ENTRY;
LASSERT(env != NULL);
+ LASSERT(data != NULL);
+
if (!exp || !obd || !cluuid)
RETURN(-EINVAL);
mdt = mdt_dev(obd->obd_lu_dev);
+ if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
+ !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
+ atomic_inc(&mdt->mdt_mds_mds_conns);
+ mdt_enable_slc(mdt);
+ }
+
/*
* first, check whether the stack is ready to handle requests
* 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,
lexp = class_conn2export(&conn);
LASSERT(lexp != NULL);
- rc = mdt_connect_internal(lexp, mdt, data);
+ rc = nodemap_add_member(*client_nid, lexp);
+ if (rc != 0 && rc != -EEXIST)
+ GOTO(out, rc);
+
+ rc = mdt_connect_internal(env, lexp, mdt, data, false);
if (rc == 0) {
struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
LASSERT(lcd);
memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
rc = tgt_client_new(env, lexp);
- if (rc == 0) {
- rc = nodemap_add_member(*client_nid, lexp);
- if (rc != 0 && rc != -EEXIST)
- goto out;
-
+ if (rc == 0)
mdt_export_stats_init(obd, lexp, localdata);
- }
}
out:
if (rc != 0) {
class_disconnect(lexp);
+ nodemap_del_member(lexp);
*exp = NULL;
} else {
*exp = lexp;
+ /* Because we do not want this export to be evicted by pinger,
+ * let's not add this export to the timed chain list. */
+ if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
+ spin_lock(&lexp->exp_obd->obd_dev_lock);
+ list_del_init(&lexp->exp_obd_chain_timed);
+ spin_unlock(&lexp->exp_obd->obd_dev_lock);
+ }
}
RETURN(rc);
if (exp == NULL || obd == NULL || cluuid == NULL)
RETURN(-EINVAL);
- rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
- if (rc == 0) {
- rc = nodemap_add_member(*client_nid, exp);
- if (rc == 0 || rc == -EEXIST)
- mdt_export_stats_init(obd, exp, localdata);
- }
+ rc = nodemap_add_member(*client_nid, exp);
+ if (rc != 0 && rc != -EEXIST)
+ RETURN(rc);
+
+ rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
+ true);
+ if (rc == 0)
+ mdt_export_stats_init(obd, exp, localdata);
+ else
+ nodemap_del_member(exp);
RETURN(rc);
}
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 */
LASSERT(list_empty(&exp->exp_outstanding_replies));
LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
+ /*
+ * discard grants once we're sure no more
+ * interaction with the client is possible
+ */
+ tgt_grant_discard(exp);
+ if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
+ exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
+
+ if (!(exp->exp_flags & OBD_OPT_FORCE))
+ tgt_grant_sanity_check(exp->exp_obd, __func__);
+
RETURN(0);
}
if (rc < 0)
return rc;
- return linkea_init(ldata);
+ return linkea_init_with_rec(ldata);
}
/**
* \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;
- mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
- if (IS_ERR(mdt_obj))
- GOTO(out, rc = PTR_ERR(mdt_obj));
+ 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);
+
+ if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
+ mdt_obj = obj;
+ mdt_object_get(info->mti_env, mdt_obj);
+ } else {
+ mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
+ if (IS_ERR(mdt_obj))
+ GOTO(out, rc = PTR_ERR(mdt_obj));
+ }
if (!mdt_object_exists(mdt_obj)) {
mdt_object_put(info->mti_env, 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;
RETURN(rc);
}
-/* Pass the ioc down */
-static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
- unsigned int cmd, int len, void *data)
-{
- struct lu_context ioctl_session;
- struct md_device *next = mdt->mdt_child;
- int rc;
- ENTRY;
-
- rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
- if (rc)
- RETURN(rc);
- ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
- lu_context_enter(&ioctl_session);
- env->le_ses = &ioctl_session;
-
- LASSERT(next->md_ops->mdo_iocontrol);
- rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
-
- lu_context_exit(&ioctl_session);
- lu_context_fini(&ioctl_session);
- RETURN(rc);
-}
-
static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
{
struct obd_ioctl_data *data = karg;
/* ioctls on obd dev */
static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
- void *karg, void *uarg)
+ void *karg, void __user *uarg)
{
struct lu_env env;
struct obd_device *obd = exp->exp_obd;
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_force_abort_recovery = 1;
+ obd->obd_abort_recovery = 1;
target_stop_recovery_thread(obd);
rc = 0;
break;
case OBD_IOC_CHANGELOG_REG:
case OBD_IOC_CHANGELOG_DEREG:
case OBD_IOC_CHANGELOG_CLEAR:
- rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
+ rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
+ mdt->mdt_child,
+ cmd, len, karg);
break;
case OBD_IOC_START_LFSCK: {
struct md_device *next = mdt->mdt_child;
rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
break;
}
+ case OBD_IOC_QUERY_LFSCK: {
+ struct md_device *next = mdt->mdt_child;
+ struct obd_ioctl_data *data = karg;
+
+ rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
+ data->ioc_inlbuf1);
+ break;
+ }
case OBD_IOC_GET_OBJ_VERSION: {
struct mdt_thread_info *mti;
mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
{
struct lu_device *ld = md2lu_dev(mdt->mdt_child);
- struct lfsck_start_param lsp;
int rc;
ENTRY;
- lsp.lsp_start = NULL;
- lsp.lsp_index_valid = 0;
- rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
- OBD_IOC_START_LFSCK,
- 0, &lsp);
- if (rc != 0 && rc != -EALREADY)
- CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
- mdt_obd_name(mdt), rc);
+ if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
+ struct lfsck_start_param lsp;
+
+ lsp.lsp_start = NULL;
+ lsp.lsp_index_valid = 0;
+ rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
+ OBD_IOC_START_LFSCK,
+ 0, &lsp);
+ if (rc != 0 && rc != -EALREADY)
+ CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
+ mdt_obd_name(mdt), rc);
+ }
rc = ld->ld_ops->ldo_recovery_complete(env, ld);
RETURN(rc);
.o_destroy_export = mdt_destroy_export,
.o_iocontrol = mdt_iocontrol,
.o_postrecov = mdt_obd_postrecov,
+ /* Data-on-MDT IO methods */
+ .o_preprw = mdt_obd_preprw,
+ .o_commitrw = mdt_obd_commitrw,
};
static struct lu_device* mdt_device_fini(const struct lu_env *env,
info->mti_big_lmm = NULL;
info->mti_big_lmmsize = 0;
}
+
+ if (info->mti_big_acl) {
+ OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
+ info->mti_big_acl = NULL;
+ info->mti_big_aclsize = 0;
+ }
+
OBD_FREE_PTR(info);
}
.ldt_ctx_tags = LCT_MD_THREAD
};
-static int __init mdt_mod_init(void)
+static int __init mdt_init(void)
{
int rc;
return rc;
}
-static void __exit mdt_mod_exit(void)
+static void __exit mdt_exit(void)
{
class_unregister_type(LUSTRE_MDT_NAME);
mds_mod_exit();
lu_kmem_fini(mdt_caches);
}
-MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
+MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");
-module_init(mdt_mod_init);
-module_exit(mdt_mod_exit);
+module_init(mdt_init);
+module_exit(mdt_exit);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff