RETURN(0);
if (is_project_state_change(oattr, la)) {
- if (!md_capable(uc, CAP_SYS_RESOURCE) &&
+ if (!cap_raised(uc->uc_cap, CAP_SYS_RESOURCE) &&
!lustre_in_group_p(uc, ma->ma_enable_chprojid_gid) &&
!(ma->ma_enable_chprojid_gid == -1 &&
mdd_permission_internal(env, obj, oattr, MAY_WRITE)))
(LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
if ((uc->uc_fsuid != oattr->la_uid) &&
- !md_capable(uc, CAP_FOWNER))
+ !cap_raised(uc->uc_cap, CAP_FOWNER))
RETURN(-EPERM);
/* The IMMUTABLE and APPEND_ONLY flags can
* only be changed by the relevant capability. */
if ((oldflags ^ newflags) &&
- !md_capable(uc, CAP_LINUX_IMMUTABLE))
+ !cap_raised(uc->uc_cap, CAP_LINUX_IMMUTABLE))
RETURN(-EPERM);
if (!S_ISDIR(oattr->la_mode)) {
if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
!(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
if ((uc->uc_fsuid != oattr->la_uid) &&
- !md_capable(uc, CAP_FOWNER)) {
+ !cap_raised(uc->uc_cap, CAP_FOWNER)) {
rc = mdd_permission_internal(env, obj, oattr,
MAY_WRITE);
if (rc)
if (la->la_valid & LA_MODE) {
if (!(flags & MDS_PERM_BYPASS) &&
(uc->uc_fsuid != oattr->la_uid) &&
- !md_capable(uc, CAP_FOWNER))
+ !cap_raised(uc->uc_cap, CAP_FOWNER))
RETURN(-EPERM);
if (la->la_mode == (umode_t) -1)
/* Also check the setgid bit! */
if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
la->la_gid : oattr->la_gid) &&
- !md_capable(uc, CAP_FSETID))
+ !cap_raised(uc->uc_cap, CAP_FSETID))
la->la_mode &= ~S_ISGID;
} else {
la->la_mode = oattr->la_mode;
la->la_uid = oattr->la_uid;
if (((uc->uc_fsuid != oattr->la_uid) ||
(la->la_uid != oattr->la_uid)) &&
- !md_capable(uc, CAP_CHOWN))
+ !cap_raised(uc->uc_cap, CAP_CHOWN))
RETURN(-EPERM);
/* If the user or group of a non-directory has been
if (((uc->uc_fsuid != oattr->la_uid) ||
((la->la_gid != oattr->la_gid) &&
!lustre_in_group_p(uc, la->la_gid))) &&
- !md_capable(uc, CAP_CHOWN))
+ !cap_raised(uc->uc_cap, CAP_CHOWN))
RETURN(-EPERM);
/* Likewise, if the user or group of a non-directory
}
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);
}
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);
* can write attributes. */
if (S_ISDIR(attr->la_mode) && (attr->la_mode & S_ISVTX) &&
(uc->uc_fsuid != attr->la_uid) &&
- !md_capable(uc, CAP_FOWNER))
+ !cap_raised(uc->uc_cap, CAP_FOWNER))
RETURN(-EPERM);
} else if (strcmp(name, XATTR_NAME_SOM) != 0 &&
(uc->uc_fsuid != attr->la_uid) &&
- !md_capable(uc, CAP_FOWNER)) {
+ !cap_raised(uc->uc_cap, CAP_FOWNER)) {
RETURN(-EPERM);
}
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_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);
+
+ 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);
}
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);
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) {
unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
hint->dah_eadata = spec->u.sp_ea.eadata;
hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
- } else {
- hint->dah_eadata = NULL;
- hint->dah_eadata_len = 0;
- if (spec->sp_cr_flags & MDS_OPEN_APPEND) {
- if (mdd->mdd_append_stripe_count != 0 ||
- mdd->mdd_append_pool[0])
- CDEBUG(D_INFO,
- "using O_APPEND file striping\n");
- if (mdd->mdd_append_stripe_count)
- hint->dah_append_stripes =
- mdd->mdd_append_stripe_count;
- if (mdd->mdd_append_pool[0])
- hint->dah_append_pool = mdd->mdd_append_pool;
- } else {
- hint->dah_append_stripes = 0;
- }
+ } else if (S_ISREG(attr->la_mode) &&
+ spec->sp_cr_flags & MDS_OPEN_APPEND) {
+ hint->dah_append_stripe_count = mdd->mdd_append_stripe_count;
+ hint->dah_append_pool = mdd->mdd_append_pool;
}
CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
RETURN(rc);
}
-static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
- size_t nob, const struct dt_it_ops *iops,
+static int mdd_dir_page_build(const struct lu_env *env, struct dt_object *obj,
+ union lu_page *lp, size_t bytes,
+ const struct dt_it_ops *iops,
struct dt_it *it, __u32 attr, void *arg)
{
- struct lu_dirpage *dp = &lp->lp_dir;
- void *area = dp;
- int result;
- __u64 hash = 0;
- struct lu_dirent *ent;
- struct lu_dirent *last = NULL;
- struct lu_fid fid;
- int first = 1;
-
- if (nob < sizeof(*dp))
- return -EINVAL;
-
- memset(area, 0, sizeof (*dp));
- area += sizeof (*dp);
- nob -= sizeof (*dp);
-
- ent = area;
- do {
- int len;
+ struct lu_dirpage *dp = &lp->lp_dir;
+ void *area = dp;
+ __u64 hash = 0;
+ struct lu_dirent *ent;
+ struct lu_dirent *last = NULL;
+ struct lu_fid fid;
+ int first = 1;
+ int result;
+
+ if (bytes < sizeof(*dp))
+ GOTO(out_err, result = -EOVERFLOW);
+
+ memset(area, 0, sizeof(*dp));
+ area += sizeof(*dp);
+ bytes -= sizeof(*dp);
+
+ ent = area;
+ do {
+ int len;
size_t recsize;
- len = iops->key_size(env, it);
+ len = iops->key_size(env, it);
- /* IAM iterator can return record with zero len. */
- if (len == 0)
- goto next;
+ /* IAM iterator can return record with zero len. */
+ if (len == 0)
+ GOTO(next, 0);
- hash = iops->store(env, it);
- if (unlikely(first)) {
- first = 0;
- dp->ldp_hash_start = cpu_to_le64(hash);
- }
+ hash = iops->store(env, it);
+ if (unlikely(first)) {
+ first = 0;
+ dp->ldp_hash_start = cpu_to_le64(hash);
+ }
- /* calculate max space required for lu_dirent */
- recsize = lu_dirent_calc_size(len, attr);
+ /* calculate max space required for lu_dirent */
+ recsize = lu_dirent_calc_size(len, attr);
- if (nob >= recsize) {
- result = iops->rec(env, it, (struct dt_rec *)ent, attr);
- if (result == -ESTALE)
- goto next;
- if (result != 0)
- goto out;
+ if (bytes >= recsize &&
+ !OBD_FAIL_CHECK(OBD_FAIL_MDS_DIR_PAGE_WALK)) {
+ result = iops->rec(env, it, (struct dt_rec *)ent, attr);
+ if (result == -ESTALE)
+ GOTO(next, result);
+ if (result != 0)
+ GOTO(out, result);
- /* osd might not able to pack all attributes,
- * so recheck rec length */
- recsize = le16_to_cpu(ent->lde_reclen);
+ /* OSD might not able to pack all attributes, so
+ * recheck record length had room to store FID
+ */
+ recsize = le16_to_cpu(ent->lde_reclen);
if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
fid_le_to_cpu(&fid, &ent->lde_fid);
if (fid_is_dot_lustre(&fid))
- goto next;
+ GOTO(next, recsize);
}
- } else {
- result = (last != NULL) ? 0 :-EINVAL;
- goto out;
- }
- last = ent;
- ent = (void *)ent + recsize;
- nob -= recsize;
+ } else {
+ result = (last != NULL) ? 0 : -EBADSLT;
+ GOTO(out, result);
+ }
+ last = ent;
+ ent = (void *)ent + recsize;
+ bytes -= recsize;
next:
- result = iops->next(env, it);
- if (result == -ESTALE)
- goto next;
- } while (result == 0);
+ result = iops->next(env, it);
+ if (result == -ESTALE)
+ GOTO(next, result);
+ } while (result == 0);
out:
- dp->ldp_hash_end = cpu_to_le64(hash);
- if (last != NULL) {
- if (last->lde_hash == dp->ldp_hash_end)
- dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
- last->lde_reclen = 0; /* end mark */
- }
+ dp->ldp_hash_end = cpu_to_le64(hash);
+ if (last != NULL) {
+ if (last->lde_hash == dp->ldp_hash_end)
+ dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
+ last->lde_reclen = 0; /* end mark */
+ }
+out_err:
if (result > 0)
/* end of directory */
dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
else if (result < 0)
- CWARN("build page failed: %d!\n", result);
- return result;
+ CWARN("%s: build page failed for "DFID": rc = %d\n",
+ lu_dev_name(obj->do_lu.lo_dev),
+ PFID(lu_object_fid(&obj->do_lu)), result);
+
+ return result;
}
int mdd_readpage(const struct lu_env *env, struct md_object *obj,