*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/mdd/mdd_object.c
*
{
struct lu_buf *buf;
- buf = &mdd_env_info(env)->mti_buf[0];
+ buf = &mdd_env_info(env)->mdi_buf[0];
buf->lb_buf = area;
buf->lb_len = len;
return buf;
{
struct lu_buf *buf;
- buf = &mdd_env_info(env)->mti_buf[0];
+ buf = &mdd_env_info(env)->mdi_buf[0];
buf->lb_buf = (void *)area;
buf->lb_len = len;
return buf;
const struct md_op_spec *spec,
struct dt_allocation_hint *hint)
{
- struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
+ struct dt_object_format *dof = &mdd_env_info(env)->mdi_dof;
const struct dt_index_features *feat = spec->sp_feat;
int rc;
ENTRY;
const struct md_op_spec *spec,
struct dt_allocation_hint *hint)
{
- struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
+ struct dt_object_format *dof = &mdd_env_info(env)->mdi_dof;
int rc;
ENTRY;
}
if (la->la_valid & LA_CTIME) {
- /* The pure setattr, it has the priority over what is
- * already set, do not drop it if ctime is equal. */
+ /**
+ * The pure setattr, it has the priority over what is
+ * already set, do not drop it if ctime is equal.
+ */
if (la->la_ctime < oattr->la_ctime)
la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
}
reclen = llog_data_len(LLOG_CHANGELOG_HDR_SZ +
changelog_rec_offset(clf_flags & CLF_SUPPORTED,
xflags & CLFE_SUPPORTED));
- buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
if (buf->lb_buf == NULL)
RETURN(-ENOMEM);
rec = buf->lb_buf;
bits |= (valid & LA_MTIME) ? BIT(CL_MTIME) : 0;
bits |= (valid & LA_CTIME) ? BIT(CL_CTIME) : 0;
bits |= (valid & LA_ATIME) ? BIT(CL_ATIME) : 0;
- bits = bits & mdd->mdd_cl.mc_mask;
+ bits = bits & mdd->mdd_cl.mc_current_mask;
/* This is an implementation limit rather than a protocol limit */
BUILD_BUG_ON(CL_LAST > sizeof(int) * 8);
if (bits == 0)
struct mdd_object *mdd_obj = md2mdd_obj(obj);
struct mdd_device *mdd = mdo2mdd(obj);
struct thandle *handle = NULL;
- struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
+ struct lu_attr *la_copy = &mdd_env_info(env)->mdi_la_for_fix;
struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
const struct lu_attr *la = &ma->ma_attr;
struct lu_ucred *uc;
+ bool quota_reserved = false;
bool chrgrp_by_unprivileged_user = false;
+ __s64 quota_size = 0;
int rc;
ENTRY;
uc = lu_ucred_check(env);
if (S_ISREG(attr->la_mode) && la->la_valid & LA_GID &&
la->la_gid != attr->la_gid && uc != NULL && uc->uc_fsuid != 0) {
- /* LU-10048: disable synchronous chgrp operation for it will
- * cause deadlock between MDT and OST.
- la_copy->la_valid |= LA_FLAGS;
- la_copy->la_flags |= LUSTRE_SET_SYNC_FL;
- */
+ CDEBUG(D_QUOTA, "%s: reserve quota for changing group: gid=%u size=%llu\n",
+ mdd2obd_dev(mdd)->obd_name, la->la_gid, la->la_size);
+
+ if (la->la_valid & LA_BLOCKS)
+ quota_size = la->la_blocks << 9;
+ else if (la->la_valid & LA_SIZE)
+ quota_size = la->la_size;
+ /* use local attr gotten above */
+ else if (attr->la_valid & LA_BLOCKS)
+ quota_size = attr->la_blocks << 9;
+ else if (attr->la_valid & LA_SIZE)
+ quota_size = attr->la_size;
+
+ if (quota_size > 0) {
+ rc = dt_reserve_or_free_quota(env, mdd->mdd_bottom,
+ GRPQUOTA, attr->la_uid,
+ la->la_gid, quota_size,
+ false);
+
+ if (rc) {
+ CDEBUG(D_QUOTA, "%s: failed to reserve quota for gid %d size %llu\n",
+ mdd2obd_dev(mdd)->obd_name,
+ la->la_gid, quota_size);
+
+ GOTO(out, rc);
+ }
+
+ quota_reserved = true;
+ la_copy->la_valid |= LA_FLAGS;
+ }
+
chrgrp_by_unprivileged_user = true;
/* Flush the possible existing client setattr requests to OSTs
rc = mdd_attr_set_changelog(env, obj, handle, &ma->ma_pfid,
la_copy->la_valid);
+ if (rc == 0 && quota_reserved) {
+ struct thandle *sub_th;
+
+ sub_th = thandle_get_sub_by_dt(env, handle, mdd->mdd_bottom);
+ if (unlikely(IS_ERR(sub_th))) {
+ dt_reserve_or_free_quota(env, mdd->mdd_bottom, GRPQUOTA,
+ attr->la_uid, la->la_gid,
+ -quota_size, false);
+ } else {
+ sub_th->th_reserved_quota.qrr_type = GRPQUOTA;
+ sub_th->th_reserved_quota.qrr_id.qid_gid = la->la_gid;
+ sub_th->th_reserved_quota.qrr_count = quota_size;
+ }
+ }
+
if (handle != NULL)
rc = mdd_trans_stop(env, mdd, rc, handle);
RETURN(-ENOMEM);
/* Read HSM attrs from disk */
- current_buf = lu_buf_check_and_alloc(&info->mti_xattr_buf,
+ current_buf = lu_buf_check_and_alloc(&info->mdi_xattr_buf,
min_t(unsigned int,
mdd_obj2mdd_dev(mdd_obj)->mdd_dt_conf.ddp_max_ea_size,
XATTR_SIZE_MAX));
struct mdd_device *mdd = mdo2mdd(md_obj);
struct mdd_object *obj = md2mdd_obj(md_obj);
struct mdd_object *vic = md2mdd_obj(md_vic);
- struct lu_buf *buf = &mdd_env_info(env)->mti_buf[0];
- struct lu_buf *buf_vic = &mdd_env_info(env)->mti_buf[1];
+ struct lu_buf *buf = &mdd_env_info(env)->mdi_buf[0];
+ struct lu_buf *buf_vic = &mdd_env_info(env)->mdi_buf[1];
struct lov_mds_md *lmm;
struct thandle *handle;
- int rc;
+ int rc, lock_order;
ENTRY;
- rc = lu_fid_cmp(mdd_object_fid(obj), mdd_object_fid(vic));
- if (rc == 0) /* same fid */
+ lock_order = lu_fid_cmp(mdd_object_fid(obj), mdd_object_fid(vic));
+ if (lock_order == 0) /* same fid */
RETURN(-EPERM);
handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
RETURN(PTR_ERR(handle));
- if (rc > 0) {
- mdd_write_lock(env, obj, DT_TGT_CHILD);
- mdd_write_lock(env, vic, DT_TGT_CHILD);
- } else {
- mdd_write_lock(env, vic, DT_TGT_CHILD);
- mdd_write_lock(env, obj, DT_TGT_CHILD);
- }
-
/* get EA of victim file */
memset(buf_vic, 0, sizeof(*buf_vic));
rc = mdd_stripe_get(env, vic, buf_vic, XATTR_NAME_LOV);
if (rc < 0) {
if (rc == -ENODATA)
rc = 0;
- GOTO(out, rc);
+ GOTO(stop, rc);
}
/* parse the layout of victim file */
lmm = buf_vic->lb_buf;
if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
- GOTO(out, rc = -EINVAL);
+ GOTO(stop, rc = -EINVAL);
/* save EA of target file for restore */
memset(buf, 0, sizeof(*buf));
rc = mdd_stripe_get(env, obj, buf, XATTR_NAME_LOV);
if (rc < 0)
- GOTO(out, rc);
+ GOTO(stop, rc);
/* Get rid of the layout from victim object */
rc = mdd_declare_xattr_del(env, mdd, vic, XATTR_NAME_LOV, handle);
if (rc)
- GOTO(out, rc);
+ GOTO(stop, rc);
- rc = mdd_declare_xattr_set(env, mdd, obj, buf_vic, XATTR_LUSTRE_LOV,
+ rc = mdd_declare_xattr_set(env, mdd, obj, buf_vic, XATTR_NAME_LOV,
LU_XATTR_MERGE, handle);
if (rc)
- GOTO(out, rc);
+ GOTO(stop, rc);
rc = mdd_trans_start(env, mdd, handle);
if (rc != 0)
- GOTO(out, rc);
+ GOTO(stop, rc);
- rc = mdo_xattr_set(env, obj, buf_vic, XATTR_LUSTRE_LOV, LU_XATTR_MERGE,
+ if (lock_order > 0) {
+ mdd_write_lock(env, obj, DT_TGT_CHILD);
+ mdd_write_lock(env, vic, DT_TGT_CHILD);
+ } else {
+ mdd_write_lock(env, vic, DT_TGT_CHILD);
+ mdd_write_lock(env, obj, DT_TGT_CHILD);
+ }
+
+ rc = mdo_xattr_set(env, obj, buf_vic, XATTR_NAME_LOV, LU_XATTR_MERGE,
handle);
if (rc)
GOTO(out, rc);
}
out:
- mdd_trans_stop(env, mdd, rc, handle);
mdd_write_unlock(env, obj);
mdd_write_unlock(env, vic);
+stop:
+ mdd_trans_stop(env, mdd, rc, handle);
lu_buf_free(buf);
lu_buf_free(buf_vic);
struct mdd_device *mdd = mdo2mdd(md_obj);
struct mdd_object *obj = md2mdd_obj(md_obj);
struct mdd_object *vic = NULL;
- struct lu_buf *buf = &mdd_env_info(env)->mti_buf[0];
- struct lu_buf *buf_save = &mdd_env_info(env)->mti_buf[1];
- struct lu_buf *buf_vic = &mdd_env_info(env)->mti_buf[2];
+ struct lu_buf *buf = &mdd_env_info(env)->mdi_buf[0];
+ struct lu_buf *buf_save = &mdd_env_info(env)->mdi_buf[1];
+ struct lu_buf *buf_vic = &mdd_env_info(env)->mdi_buf[2];
struct lov_comp_md_v1 *lcm;
struct thandle *handle;
int rc;
if (mrd->mrd_obj)
vic = md2mdd_obj(mrd->mrd_obj);
- if (vic) {
- /* don't use the same file to save the splitted mirror */
- rc = lu_fid_cmp(mdd_object_fid(obj), mdd_object_fid(vic));
- if (rc == 0)
- RETURN(-EPERM);
-
- if (rc > 0) {
- mdd_write_lock(env, obj, DT_TGT_CHILD);
- mdd_write_lock(env, vic, DT_TGT_CHILD);
- } else {
- mdd_write_lock(env, vic, DT_TGT_CHILD);
- mdd_write_lock(env, obj, DT_TGT_CHILD);
- }
- } else {
- mdd_write_lock(env, obj, DT_TGT_CHILD);
- }
-
handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
- GOTO(unlock, rc = PTR_ERR(handle));
+ RETURN(PTR_ERR(handle));
/* get EA of mirrored file */
memset(buf_save, 0, sizeof(*buf));
rc = mdd_stripe_get(env, obj, buf_save, XATTR_NAME_LOV);
if (rc < 0)
- GOTO(out, rc);
+ GOTO(stop, rc);
lcm = buf_save->lb_buf;
if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
- GOTO(out, rc = -EINVAL);
+ GOTO(stop, rc = -EINVAL);
/**
* Extract the mirror with specified mirror id, and store the splitted
memset(buf_vic, 0, sizeof(*buf_vic));
rc = mdd_split_ea(lcm, mrd->mrd_mirror_id, buf, buf_vic);
if (rc < 0)
- GOTO(out, rc);
+ GOTO(stop, rc);
/**
* @buf stores layout w/o the specified mirror, @buf_vic stores the
* splitted mirror
rc = mdd_declare_xattr_set(env, mdd, obj, buf, XATTR_NAME_LOV,
LU_XATTR_SPLIT, handle);
if (rc)
- GOTO(out_restore, rc);
+ GOTO(stop, rc);
/* declare vic set splitted layout in @buf_vic */
rc = mdd_declare_xattr_set(env, mdd, vic, buf_vic,
XATTR_NAME_LOV, LU_XATTR_SPLIT,
handle);
if (rc)
- GOTO(out_restore, rc);
+ GOTO(stop, rc);
} else {
/**
* declare delete mirror objects in @buf_vic, will change obj's
XATTR_NAME_LOV, LU_XATTR_PURGE,
handle);
if (rc)
- GOTO(out_restore, rc);
+ GOTO(stop, rc);
/* declare obj set remaining layout in @buf */
rc = mdd_declare_xattr_set(env, mdd, obj, buf,
XATTR_NAME_LOV, LU_XATTR_SPLIT,
handle);
if (rc)
- GOTO(out_restore, rc);
+ GOTO(stop, rc);
}
rc = mdd_trans_start(env, mdd, handle);
if (rc)
- GOTO(out_restore, rc);
+ GOTO(stop, rc);
+
+ if (vic) {
+ /* don't use the same file to save the splitted mirror */
+ rc = lu_fid_cmp(mdd_object_fid(obj), mdd_object_fid(vic));
+ if (rc == 0)
+ GOTO(stop, rc = -EPERM);
+
+ if (rc > 0) {
+ mdd_write_lock(env, obj, DT_TGT_CHILD);
+ mdd_write_lock(env, vic, DT_TGT_CHILD);
+ } else {
+ mdd_write_lock(env, vic, DT_TGT_CHILD);
+ mdd_write_lock(env, obj, DT_TGT_CHILD);
+ }
+ } else {
+ mdd_write_lock(env, obj, DT_TGT_CHILD);
+ }
/* set obj's layout in @buf */
- rc = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV, LU_XATTR_REPLACE,
+ rc = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV, LU_XATTR_SPLIT,
handle);
if (rc)
- GOTO(out_restore, rc);
+ GOTO(unlock, rc);
if (vic) {
/* set vic's layout in @buf_vic */
rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle,
NULL);
if (rc)
- GOTO(out, rc);
+ GOTO(out_restore, rc);
if (vic) {
rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic,
handle, NULL);
if (rc)
- GOTO(out, rc);
+ GOTO(out_restore, rc);
}
out_restore:
PFID(mdd_object_fid(obj)), rc);
}
-out:
+unlock:
+ mdd_write_unlock(env, obj);
+ if (vic)
+ mdd_write_unlock(env, vic);
+stop:
rc = mdd_trans_stop(env, mdd, rc, handle);
/* Truncate local DOM data if all went well */
if (!rc && dom_stripe)
mdd_dom_data_truncate(env, mdd, obj);
-unlock:
- mdd_write_unlock(env, obj);
- if (vic)
- mdd_write_unlock(env, vic);
lu_buf_free(buf_save);
lu_buf_free(buf);
lu_buf_free(buf_vic);
int mdd_stripe_get(const struct lu_env *env, struct mdd_object *obj,
struct lu_buf *lmm_buf, const char *name)
{
- struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf;
+ struct lu_buf *buf = &mdd_env_info(env)->mdi_big_buf;
int rc;
ENTRY;
repeat:
rc = mdo_xattr_get(env, obj, buf, name);
if (rc == -ERANGE) {
- /* mti_big_buf is allocated but is too small
+ /* mdi_big_buf is allocated but is too small
* we need to increase it */
- buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
+ buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_big_buf,
buf->lb_len * 2);
if (buf->lb_buf == NULL)
RETURN(-ENOMEM);
/*
* we don't use lmm_buf directly, because we don't know xattr size, so
- * by using mti_big_buf we can avoid calling mdo_xattr_get() twice.
+ * by using mdi_big_buf we can avoid calling mdo_xattr_get() twice.
*/
memcpy(lmm_buf->lb_buf, buf->lb_buf, rc);
struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr);
struct mdd_device *mdd = mdo2mdd(obj1);
struct lov_mds_md *fst_lmm, *snd_lmm;
- struct lu_buf *fst_buf = &info->mti_buf[0];
- struct lu_buf *snd_buf = &info->mti_buf[1];
- struct lu_buf *fst_hsm_buf = &info->mti_buf[2];
- struct lu_buf *snd_hsm_buf = &info->mti_buf[3];
+ struct lu_buf *fst_buf = &info->mdi_buf[0];
+ struct lu_buf *snd_buf = &info->mdi_buf[1];
+ struct lu_buf *fst_hsm_buf = &info->mdi_buf[2];
+ struct lu_buf *snd_hsm_buf = &info->mdi_buf[3];
struct ost_id *saved_oi = NULL;
struct thandle *handle;
struct mdd_object *dom_o = NULL;
ENTRY;
- BUILD_BUG_ON(ARRAY_SIZE(info->mti_buf) < 4);
- memset(info->mti_buf, 0, sizeof(info->mti_buf));
+ BUILD_BUG_ON(ARRAY_SIZE(info->mdi_buf) < 4);
+ memset(info->mdi_buf, 0, sizeof(info->mdi_buf));
/* we have to sort the 2 obj, so locking will always
* be in the same order, even in case of 2 concurrent swaps */
if (fst_lmm != NULL) {
struct ost_id temp_oi;
- saved_oi = &info->mti_oa.o_oi;
+ saved_oi = &info->mdi_oa.o_oi;
mdd_get_lmm_oi(fst_lmm, saved_oi);
mdd_get_lmm_oi(snd_lmm, &temp_oi);
mdd_set_lmm_gen(fst_lmm, &snd_gen);
if (rc2 < 0)
goto do_lbug;
- ++steps;
- rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
- if (rc2 < 0)
- goto do_lbug;
+ if (flags & SWAP_LAYOUTS_MDS_HSM) {
+ ++steps;
+ rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
+ handle);
+ if (rc2 < 0)
+ goto do_lbug;
- ++steps;
- rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);
+ ++steps;
+ rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf,
+ handle);
+ }
do_lbug:
if (rc2 < 0) {
struct md_layout_change *mlc, struct thandle *handle)
{
struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
- struct lu_buf *som_buf = &mdd_env_info(env)->mti_buf[1];
+ struct lu_buf *som_buf = &mdd_env_info(env)->mdi_buf[1];
struct lustre_som_attrs *som = &mlc->mlc_som;
int fl = 0;
int rc;
lustre_som_swab(som);
if (som->lsa_valid & SOM_FL_STRICT)
fl = LU_XATTR_REPLACE;
+
+ if (mlc->mlc_opc == MD_LAYOUT_WRITE &&
+ mlc->mlc_intent->li_extent.e_end > som->lsa_size) {
+ som->lsa_size = mlc->mlc_intent->li_extent.e_end + 1;
+ fl = LU_XATTR_REPLACE;
+ }
}
rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
struct thandle *handle)
{
struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
+ struct lu_buf *som_buf = &mdd_env_info(env)->mdi_buf[1];
+ struct lustre_som_attrs *som = &mlc->mlc_som;
+ int fl = 0;
int rc;
ENTRY;
* resync state. */
break;
case MD_LAYOUT_WRITE:
- /* legal race for concurrent write, the file state has been
- * changed by another client. */
+ /**
+ * legal race for concurrent write, the file state has been
+ * changed by another client. Or a jump over file size and
+ * write.
+ */
+ som_buf->lb_buf = som;
+ som_buf->lb_len = sizeof(*som);
+ rc = mdo_xattr_get(env, obj, som_buf, XATTR_NAME_SOM);
+ if (rc < 0 && rc != -ENODATA)
+ RETURN(rc);
+
+ if (rc > 0) {
+ lustre_som_swab(som);
+ if (mlc->mlc_intent->li_extent.e_end > som->lsa_size) {
+ som->lsa_size =
+ mlc->mlc_intent->li_extent.e_end + 1;
+ fl = LU_XATTR_REPLACE;
+ }
+ }
break;
default:
RETURN(-EBUSY);
if (rc)
GOTO(out, rc);
+ if (fl) {
+ rc = mdd_declare_xattr_set(env, mdd, obj, som_buf,
+ XATTR_NAME_SOM, fl, handle);
+ if (rc)
+ GOTO(out, rc);
+ }
+
rc = mdd_trans_start(env, mdd, handle);
if (rc)
GOTO(out, rc);
mdd_write_lock(env, obj, DT_TGT_CHILD);
rc = mdo_layout_change(env, obj, mlc, handle);
+ if (!rc && fl) {
+ som->lsa_valid = SOM_FL_STALE;
+ lustre_som_swab(som);
+ rc = mdo_xattr_set(env, obj, som_buf, XATTR_NAME_SOM,
+ fl, handle);
+ }
mdd_write_unlock(env, obj);
if (rc)
GOTO(out, rc);
struct md_layout_change *mlc, struct thandle *handle)
{
struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
- struct lu_buf *som_buf = &mdd_env_info(env)->mti_buf[1];
+ struct lu_buf *som_buf = &mdd_env_info(env)->mdi_buf[1];
int fl = 0;
int rc;
ENTRY;
rc = mdd_open_sanity_check(env, mdd_obj, attr, open_flags,
spec->no_create);
- if ((rc == -EACCES) && (mdd->mdd_cl.mc_mask & BIT(CL_DN_OPEN)))
+ if ((rc == -EACCES) && (mdd->mdd_cl.mc_current_mask & BIT(CL_DN_OPEN)))
type = CL_DN_OPEN;
else if (rc != 0)
GOTO(out, rc);
* this is not a big deal if we have a CL_CLOSE entry with no matching
* CL_OPEN. Plus Changelogs mask may not change often.
*/
- if (((!(mdd->mdd_cl.mc_mask & BIT(CL_OPEN)) &&
+ if (((!(mdd->mdd_cl.mc_current_mask & BIT(CL_OPEN)) &&
(open_flags & (MDS_FMODE_WRITE | MDS_OPEN_APPEND |
MDS_OPEN_TRUNC))) ||
- ((mdd->mdd_cl.mc_mask & BIT(CL_OPEN)) && last_close_by_uid)) &&
+ ((mdd->mdd_cl.mc_current_mask & BIT(CL_OPEN)) &&
+ last_close_by_uid)) &&
!(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
if (handle == NULL) {
handle = mdd_trans_create(env, mdo2mdd(obj));