*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/mdt/mdt_reint.c
*
struct mdt_device *mdt = info->mti_mdt;
struct lu_fid *fid = &info->mti_tmp_fid2;
struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
+ struct lmv_user_md *lum = spec->u.sp_ea.eadata;
struct lmv_mds_md_v1 *lmv;
struct mdt_object *child;
struct mdt_lock_handle *lhp;
if (!mdt->mdt_enable_dir_restripe)
RETURN(-EPERM);
+ LASSERT(lum);
+ lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_FIXED);
+
rc = mdt_version_get_check_save(info, parent, 0);
if (rc)
RETURN(rc);
if (ma->ma_valid & MA_LMV) {
/* don't allow restripe if parent dir layout is changing */
lmv = &ma->ma_lmv->lmv_md_v1;
- if (!lmv_is_sane(lmv))
+ if (!lmv_is_sane2(lmv))
GOTO(unlock_parent, rc = -EBADF);
if (lmv_is_layout_changing(lmv))
GOTO(out_child, rc = -EREMOTE);
}
+ if (!S_ISDIR(lu_object_attr(&child->mot_obj)))
+ GOTO(out_child, rc = -ENOTDIR);
+
+ rc = mdt_stripe_get(info, child, ma, XATTR_NAME_LMV);
+ if (rc)
+ GOTO(out_child, rc);
+
+ /* race with migrate? */
+ if ((ma->ma_valid & MA_LMV) &&
+ lmv_is_migrating(&ma->ma_lmv->lmv_md_v1))
+ GOTO(out_child, rc = -EBUSY);
+
/* lock object */
lhc = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lhc, LCK_EX);
LMV_HASH_TYPE_CRUSH)
RETURN(-EPROTO);
- if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
+ if (!md_capable(uc, CAP_SYS_ADMIN) &&
uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
mdt->mdt_enable_remote_dir_gid != -1)
RETURN(-EPERM);
- /* restripe if later found dir exists */
- if (le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT)
+ /* restripe if later found dir exists, MDS_OPEN_CREAT means
+ * this is create only, don't try restripe.
+ */
+ if (mdt->mdt_enable_dir_restripe &&
+ le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT &&
+ !(spec->sp_cr_flags & MDS_OPEN_CREAT))
restripe = true;
}
if (rc)
GOTO(put_child, rc);
- /* Let lower layer know current lock mode. */
- info->mti_spec.sp_cr_mode = mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
-
/*
* Do not perform lookup sanity check. We know that name does
* not exist.
* set the HSM state to dirty.
*/
cap_saved = uc->uc_cap;
- uc->uc_cap |= MD_CAP_TO_MASK(CFS_CAP_FOWNER);
+ uc->uc_cap |= MD_CAP_TO_MASK(CAP_FOWNER);
rc = mdt_hsm_attr_set(info, mo, &ma->ma_hsm);
uc->uc_cap = cap_saved;
if (rc)
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_object *mo;
struct mdt_body *repbody;
+ ktime_t kstart = ktime_get();
int rc, rc2;
ENTRY;
/* MDT supports FMD for regular files due to Data-on-MDT */
if (S_ISREG(lu_object_attr(&mo->mot_obj)) &&
- ma->ma_attr.la_valid & (LA_ATIME | LA_MTIME | LA_CTIME))
+ ma->ma_attr.la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
tgt_fmd_update(info->mti_exp, mdt_object_fid(mo),
req->rq_xid);
+ if (ma->ma_attr.la_valid & LA_MTIME) {
+ rc = mdt_attr_get_pfid(info, mo, &ma->ma_pfid);
+ if (!rc)
+ ma->ma_valid |= MA_PFID;
+ }
+ }
+
rc = mdt_attr_set(info, mo, ma);
if (rc)
GOTO(out_put, rc);
!mdt->mdt_enable_striped_dir)
GOTO(out_put, rc = -EPERM);
- if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
+ if (!md_capable(uc, CAP_SYS_ADMIN) &&
uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
mdt->mdt_enable_remote_dir_gid != -1)
GOTO(out_put, rc = -EPERM);
mdt_object_put(info->mti_env, mo);
out:
if (rc == 0)
- mdt_counter_incr(req, LPROC_MDT_SETATTR);
+ mdt_counter_incr(req, LPROC_MDT_SETATTR,
+ ktime_us_delta(ktime_get(), kstart));
mdt_client_compatibility(info);
rc2 = mdt_fix_reply(info);
struct mdt_lock_handle *lhc)
{
struct ptlrpc_request *req = mdt_info_req(info);
+ ktime_t kstart = ktime_get();
int rc;
ENTRY;
switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
- mdt_counter_incr(req, LPROC_MDT_MKDIR);
- break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
- /* Special file should stay on the same node as parent. */
- mdt_counter_incr(req, LPROC_MDT_MKNOD);
break;
default:
CERROR("%s: Unsupported mode %o\n",
}
rc = mdt_create(info);
+ if (rc == 0) {
+ if ((info->mti_attr.ma_attr.la_mode & S_IFMT) == S_IFDIR)
+ mdt_counter_incr(req, LPROC_MDT_MKDIR,
+ ktime_us_delta(ktime_get(), kstart));
+ else
+ /* Special file should stay on the same node as parent*/
+ mdt_counter_incr(req, LPROC_MDT_MKNOD,
+ ktime_us_delta(ktime_get(), kstart));
+ }
+
RETURN(rc);
}
__u64 lock_ibits;
bool cos_incompat = false;
int no_name = 0;
+ ktime_t kstart = ktime_get();
int rc;
ENTRY;
GOTO(put_parent, rc);
}
+ OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
+ OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
relock:
parent_lh = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(parent_lh, LCK_PW, &rr->rr_name);
/* Return -ENOTSUPP for old client */
GOTO(put_child, rc = -ENOTSUPP);
- if (!md_capable(uc, CFS_CAP_SYS_ADMIN))
+ if (!md_capable(uc, CAP_SYS_ADMIN))
GOTO(put_child, rc = -EPERM);
ma->ma_need = MA_INODE;
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
- mdt_counter_incr(req, LPROC_MDT_RMDIR);
+ mdt_counter_incr(req, LPROC_MDT_RMDIR,
+ ktime_us_delta(ktime_get(), kstart));
break;
case S_IFREG:
case S_IFLNK:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
- mdt_counter_incr(req, LPROC_MDT_UNLINK);
+ mdt_counter_incr(req, LPROC_MDT_UNLINK,
+ ktime_us_delta(ktime_get(), kstart));
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
struct mdt_object *mp;
struct mdt_lock_handle *lhs;
struct mdt_lock_handle *lhp;
+ ktime_t kstart = ktime_get();
bool cos_incompat;
int rc;
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
RETURN(err_serious(-ENOENT));
- if (OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_RESEND_RACE)) {
+ if (OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_RESEND_RACE) ||
+ OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND)) {
req->rq_no_reply = 1;
RETURN(err_serious(-ENOENT));
}
mdt_object_child(ms), &rr->rr_name, ma);
if (rc == 0)
- mdt_counter_incr(req, LPROC_MDT_LINK);
+ mdt_counter_incr(req, LPROC_MDT_LINK,
+ ktime_us_delta(ktime_get(), kstart));
EXIT;
unlock_source:
EXIT;
out:
- if (rc)
+ if (rc) {
mdt_unlock_list(info, link_locks, rc);
- else if (local_lnkp_cnt > RS_MAX_LOCKS - 6)
+ } else if (local_lnkp_cnt > RS_MAX_LOCKS - 5) {
+ CDEBUG(D_INFO, "Too many links (%d), sync operations\n",
+ local_lnkp_cnt);
/*
* parent may have 3 local objects: master object and 2 stripes
- * (if it's being migrated too); source may have 2 local
- * objects: master and 1 stripe; target has 1 local object.
+ * (if it's being migrated too); source may have 1 local objects
+ * as regular file; target has 1 local object.
+ * Note, source may have 2 local locks if it is directory but it
+ * can't have hardlinks, so it is not considered here.
*/
rc = 1;
+ }
return rc;
}
if (!mdt->mdt_enable_remote_dir || !mdt->mdt_enable_dir_migration)
RETURN(-EPERM);
- if (uc && !md_capable(uc, CFS_CAP_SYS_ADMIN) &&
+ if (uc && !md_capable(uc, CAP_SYS_ADMIN) &&
uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
mdt->mdt_enable_remote_dir_gid != -1)
RETURN(-EPERM);
if (ma->ma_valid & MA_LOV && mdt_lmm_dom_stripesize(ma->ma_lmm))
info->mti_spec.sp_migrate_nsonly = 1;
+ } else if (S_ISDIR(lu_object_attr(&sobj->mot_obj))) {
+ rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LMV);
+ if (rc)
+ GOTO(unlock_links, rc);
+
+ /* race with restripe/auto-split? */
+ if ((ma->ma_valid & MA_LMV) &&
+ lmv_is_restriping(&ma->ma_lmv->lmv_md_v1))
+ GOTO(unlock_links, rc = -EBUSY);
}
/* if migration HSM is allowed */
}
/*
+ * lock rename source object.
+ *
+ * Both source and source parent may be remote, and source may be a remote
+ * object on source parent, to avoid overriding lock handle, store remote
+ * LOOKUP lock separately in @lhr.
+ *
+ * \retval 0 on success
+ * \retval -ev negative errno upon error
+ */
+static int mdt_rename_source_lock(struct mdt_thread_info *info,
+ struct mdt_object *parent,
+ struct mdt_object *child,
+ struct mdt_lock_handle *lhc,
+ struct mdt_lock_handle *lhr,
+ __u64 ibits,
+ bool cos_incompat)
+{
+ int rc;
+
+ rc = mdt_is_remote_object(info, parent, child);
+ if (rc < 0)
+ return rc;
+
+ if (rc) {
+ /* enqueue remote LOOKUP lock from the parent MDT */
+ __u64 rmt_ibits = MDS_INODELOCK_LOOKUP;
+
+ if (mdt_object_remote(parent)) {
+ rc = mdt_remote_object_lock(info, parent,
+ mdt_object_fid(child),
+ &lhr->mlh_rreg_lh,
+ lhr->mlh_rreg_mode,
+ rmt_ibits, false);
+ if (rc != ELDLM_OK)
+ return rc;
+ } else {
+ LASSERT(mdt_object_remote(child));
+ rc = mdt_object_local_lock(info, child, lhr,
+ &rmt_ibits, 0, true);
+ if (rc < 0)
+ return rc;
+ }
+
+ ibits &= ~MDS_INODELOCK_LOOKUP;
+ }
+
+ if (mdt_object_remote(child)) {
+ rc = mdt_remote_object_lock(info, child, mdt_object_fid(child),
+ &lhc->mlh_rreg_lh,
+ lhc->mlh_rreg_mode,
+ ibits, false);
+ if (rc == ELDLM_OK)
+ rc = 0;
+ } else {
+ rc = mdt_reint_object_lock(info, child, lhc, ibits,
+ cos_incompat);
+ }
+
+ if (!rc)
+ mdt_object_unlock(info, child, lhr, rc);
+
+ return rc;
+}
+
+/*
* VBR: rename versions in reply: 0 - srcdir parent; 1 - tgtdir parent;
* 2 - srcdir child; 3 - tgtdir child.
* Update on disk version of srcdir child.
struct mdt_lock_handle *lh_srcdirp;
struct mdt_lock_handle *lh_tgtdirp;
struct mdt_lock_handle *lh_oldp = NULL;
+ struct mdt_lock_handle *lh_rmt = NULL;
struct mdt_lock_handle *lh_newp = NULL;
struct lu_fid *old_fid = &info->mti_tmp_fid1;
struct lu_fid *new_fid = &info->mti_tmp_fid2;
__u64 lock_ibits;
bool reverse = false, discard = false;
bool cos_incompat;
+ ktime_t kstart = ktime_get();
int rc;
ENTRY;
mdt_object_remote(msrcdir))
GOTO(out_put_tgtdir, rc = -EXDEV);
- rc = mdt_rename_lock(info, &rename_lh);
- if (rc != 0) {
- CERROR("%s: can't lock FS for rename: rc = %d\n",
- mdt_obd_name(mdt), rc);
- GOTO(out_put_tgtdir, rc);
+ /* This might be further relaxed in the future for regular file
+ * renames in different source and target parents. Start with
+ * only same-directory renames for simplicity and because this
+ * is by far the most the common use case.
+ */
+ if (msrcdir != mtgtdir) {
+ rc = mdt_rename_lock(info, &rename_lh);
+ if (rc != 0) {
+ CERROR("%s: cannot lock for rename: rc = %d\n",
+ mdt_obd_name(mdt), rc);
+ GOTO(out_put_tgtdir, rc);
+ }
+ } else {
+ CDEBUG(D_INFO, "%s: samedir rename "DFID"/"DNAME"\n",
+ mdt_obd_name(mdt), PFID(rr->rr_fid1),
+ PNAME(&rr->rr_name));
}
}
lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
+ OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
+ OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
relock:
mdt_lock_pdo_init(lh_srcdirp, LCK_PW, &rr->rr_name);
mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, &rr->rr_tgt_name);
GOTO(out_put_new, rc = -EISDIR);
lh_oldp = &info->mti_lh[MDT_LH_OLD];
+ lh_rmt = &info->mti_lh[MDT_LH_RMT];
mdt_lock_reg_init(lh_oldp, LCK_EX);
+ mdt_lock_reg_init(lh_rmt, LCK_EX);
lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_XATTR;
- if (mdt_object_remote(msrcdir)) {
- /* Enqueue lookup lock from the parent MDT */
- rc = mdt_remote_object_lock(info, msrcdir,
- mdt_object_fid(mold),
- &lh_oldp->mlh_rreg_lh,
- lh_oldp->mlh_rreg_mode,
- MDS_INODELOCK_LOOKUP,
- false);
- if (rc != ELDLM_OK)
- GOTO(out_put_new, rc);
-
- lock_ibits &= ~MDS_INODELOCK_LOOKUP;
- }
-
- rc = mdt_reint_object_lock(info, mold, lh_oldp, lock_ibits,
- cos_incompat);
- if (rc != 0)
- GOTO(out_unlock_old, rc);
+ rc = mdt_rename_source_lock(info, msrcdir, mold, lh_oldp,
+ lh_rmt, lock_ibits, cos_incompat);
+ if (rc < 0)
+ GOTO(out_put_new, rc);
/* Check if @msrcdir is subdir of @mnew, before locking child
* to avoid reverse locking.
lh_newp = &info->mti_lh[MDT_LH_NEW];
mdt_lock_reg_init(lh_newp, LCK_EX);
- rc = mdt_reint_object_lock(info, mnew, lh_newp,
- MDS_INODELOCK_LOOKUP |
- MDS_INODELOCK_UPDATE,
+ lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
+ if (mdt_object_remote(mtgtdir)) {
+ rc = mdt_remote_object_lock(info, mtgtdir,
+ mdt_object_fid(mnew),
+ &lh_newp->mlh_rreg_lh,
+ lh_newp->mlh_rreg_mode,
+ MDS_INODELOCK_LOOKUP,
+ false);
+ if (rc != ELDLM_OK)
+ GOTO(out_unlock_old, rc);
+
+ lock_ibits &= ~MDS_INODELOCK_LOOKUP;
+ }
+ rc = mdt_reint_object_lock(info, mnew, lh_newp, lock_ibits,
cos_incompat);
if (rc != 0)
- GOTO(out_unlock_old, rc);
+ GOTO(out_unlock_new, rc);
/* get and save version after locking */
mdt_version_get_save(info, mnew, 3);
- } else if (rc != -EREMOTE && rc != -ENOENT) {
+ } else if (rc != -ENOENT) {
GOTO(out_put_old, rc);
} else {
lh_oldp = &info->mti_lh[MDT_LH_OLD];
+ lh_rmt = &info->mti_lh[MDT_LH_RMT];
mdt_lock_reg_init(lh_oldp, LCK_EX);
+ mdt_lock_reg_init(lh_rmt, LCK_EX);
lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_XATTR;
- if (mdt_object_remote(msrcdir)) {
- /* Enqueue lookup lock from the parent MDT */
- rc = mdt_remote_object_lock(info, msrcdir,
- mdt_object_fid(mold),
- &lh_oldp->mlh_rreg_lh,
- lh_oldp->mlh_rreg_mode,
- MDS_INODELOCK_LOOKUP,
- false);
- if (rc != ELDLM_OK)
- GOTO(out_put_old, rc);
-
- lock_ibits &= ~MDS_INODELOCK_LOOKUP;
- }
-
- rc = mdt_reint_object_lock(info, mold, lh_oldp, lock_ibits,
- cos_incompat);
+ rc = mdt_rename_source_lock(info, msrcdir, mold, lh_oldp,
+ lh_rmt, lock_ibits, cos_incompat);
if (rc != 0)
- GOTO(out_unlock_old, rc);
+ GOTO(out_put_old, rc);
mdt_enoent_version_save(info, 3);
}
/* handle last link of tgt object */
if (rc == 0) {
- mdt_counter_incr(req, LPROC_MDT_RENAME);
+ mdt_counter_incr(req, LPROC_MDT_RENAME,
+ ktime_us_delta(ktime_get(), kstart));
if (mnew) {
mdt_handle_last_unlink(info, mnew, ma);
discard = mdt_dom_check_for_discard(info, mnew);
}
mdt_rename_counter_tally(info, info->mti_mdt, req,
- msrcdir, mtgtdir);
+ msrcdir, mtgtdir,
+ ktime_us_delta(ktime_get(), kstart));
}
EXIT;
+out_unlock_new:
if (mnew != NULL)
mdt_object_unlock(info, mnew, lh_newp, rc);
out_unlock_old:
+ mdt_object_unlock(info, NULL, lh_rmt, rc);
mdt_object_unlock(info, mold, lh_oldp, rc);
out_put_new:
if (mnew && !discard)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff