X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_object.c;h=3e4d7ed01ee90e2c6ff6c4d84951f5552175a8a2;hp=0ed28209dca4cfcbed8697ee5634e5fdbba3948c;hb=03963106926883cf322e085feb8caa3ea64db1d1;hpb=f437134e80a1b320e575d774061e693042f3eb4c diff --git a/lustre/mdd/mdd_object.c b/lustre/mdd/mdd_object.c index 0ed2820..3e4d7ed 100644 --- a/lustre/mdd/mdd_object.c +++ b/lustre/mdd/mdd_object.c @@ -250,7 +250,7 @@ struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len) { 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; @@ -261,7 +261,7 @@ const struct lu_buf *mdd_buf_get_const(const struct lu_env *env, { 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; @@ -550,7 +550,7 @@ int mdd_declare_create_object_internal(const struct lu_env *env, 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; @@ -582,7 +582,7 @@ int mdd_create_object_internal(const struct lu_env *env, struct mdd_object *p, 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; @@ -684,7 +684,7 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, 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))) @@ -726,13 +726,13 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, (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)) { @@ -757,7 +757,7 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, 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) @@ -790,7 +790,7 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, 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) @@ -802,7 +802,7 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, /* 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; @@ -814,7 +814,7 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, 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 @@ -840,7 +840,7 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, 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 @@ -870,8 +870,10 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, } 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); } @@ -911,7 +913,7 @@ static int mdd_changelog_data_store_by_fid(const struct lu_env *env, 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_chlg_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; @@ -1088,7 +1090,7 @@ static int mdd_attr_set_changelog(const struct lu_env *env, 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) @@ -1211,11 +1213,13 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, 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; @@ -1248,11 +1252,37 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, 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 @@ -1328,6 +1358,21 @@ out: 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); @@ -1351,11 +1396,11 @@ static int mdd_xattr_sanity_check(const struct lu_env *env, * 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); } @@ -1458,7 +1503,7 @@ static int mdd_hsm_update_locked(const struct lu_env *env, 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)); @@ -1540,64 +1585,64 @@ static int mdd_xattr_merge(const struct lu_env *env, struct md_object *md_obj, 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); @@ -1623,9 +1668,10 @@ out_restore: } 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); @@ -1770,9 +1816,9 @@ static int mdd_xattr_split(const struct lu_env *env, struct md_object *md_obj, 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; @@ -1876,7 +1922,7 @@ static int mdd_xattr_split(const struct lu_env *env, struct md_object *md_obj, } /* 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(unlock, rc); @@ -2141,7 +2187,7 @@ stop: 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; @@ -2155,9 +2201,9 @@ int mdd_stripe_get(const struct lu_env *env, struct mdd_object *obj, 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); @@ -2174,7 +2220,7 @@ repeat: /* * 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); @@ -2386,10 +2432,10 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, 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; @@ -2400,8 +2446,8 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, 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 */ @@ -2540,7 +2586,7 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, 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); @@ -2668,13 +2714,17 @@ out_restore: 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) { @@ -2781,7 +2831,7 @@ mdd_layout_update_rdonly(const struct lu_env *env, struct mdd_object *obj, 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; @@ -2877,7 +2927,7 @@ mdd_layout_update_write_pending(const struct lu_env *env, 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; @@ -2969,7 +3019,7 @@ mdd_object_update_sync_pending(const struct lu_env *env, struct mdd_object *obj, 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; @@ -3151,22 +3201,10 @@ void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent, 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)), @@ -3260,7 +3298,7 @@ static int mdd_open(const struct lu_env *env, struct md_object *obj, 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); @@ -3498,10 +3536,11 @@ out: * 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)); @@ -3549,89 +3588,96 @@ static int mdd_readpage_sanity_check(const struct lu_env *env, 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,