X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=inline;f=lustre%2Fmdd%2Fmdd_object.c;h=7de10177b9c90fb97fda43cf6f4a75f327660d60;hb=40daa59ac41f450b60b42eb2bb0ff42ebd3c998b;hp=1025a0d5b20ed44eec2016fc0fb024de2b94ca3b;hpb=27c51b5623123ba3208569ecdef343f61646c7e4;p=fs%2Flustre-release.git diff --git a/lustre/mdd/mdd_object.c b/lustre/mdd/mdd_object.c index 1025a0d..7de1017 100644 --- a/lustre/mdd/mdd_object.c +++ b/lustre/mdd/mdd_object.c @@ -27,7 +27,6 @@ */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/mdd/mdd_object.c * @@ -55,7 +54,7 @@ static const struct lu_object_operations mdd_lu_obj_ops; struct mdd_object_user { struct list_head mou_list; /**< linked off mod_users */ - int mou_flags; /**< open mode by client */ + u64 mou_open_flags; /**< open mode by client */ __u64 mou_uidgid; /**< uid_gid on client */ int mou_opencount; /**< # opened */ ktime_t mou_deniednext; /**< time of next access denied @@ -70,28 +69,30 @@ static int mdd_xattr_get(const struct lu_env *env, static int mdd_changelog_data_store_by_fid(const struct lu_env *env, struct mdd_device *mdd, enum changelog_rec_type type, - int flags, const struct lu_fid *fid, + enum changelog_rec_flags clf_flags, + const struct lu_fid *fid, + const struct lu_fid *pfid, const char *xattr_name, struct thandle *handle); static inline bool has_prefix(const char *str, const char *prefix); -static unsigned int flags_helper(int flags) +static u32 flags_helper(u64 open_flags) { - unsigned int mflags = 0; + u32 open_mode = 0; - if (flags & MDS_FMODE_EXEC) { - mflags = MDS_FMODE_EXEC; + if (open_flags & MDS_FMODE_EXEC) { + open_mode = MDS_FMODE_EXEC; } else { - if (flags & MDS_FMODE_READ) - mflags = MDS_FMODE_READ; - if (flags & + if (open_flags & MDS_FMODE_READ) + open_mode = MDS_FMODE_READ; + if (open_flags & (MDS_FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND)) - mflags |= MDS_FMODE_WRITE; + open_mode |= MDS_FMODE_WRITE; } - return mflags; + return open_mode; } /** Allocate/init a user and its sub-structures. @@ -102,7 +103,7 @@ static unsigned int flags_helper(int flags) * \retval mou [OUT] success valid structure * \retval mou [OUT] */ -static struct mdd_object_user *mdd_obj_user_alloc(int flags, +static struct mdd_object_user *mdd_obj_user_alloc(u64 open_flags, uid_t uid, gid_t gid) { struct mdd_object_user *mou; @@ -113,7 +114,7 @@ static struct mdd_object_user *mdd_obj_user_alloc(int flags, if (mou == NULL) RETURN(ERR_PTR(-ENOMEM)); - mou->mou_flags = flags; + mou->mou_open_flags = open_flags; mou->mou_uidgid = ((__u64)uid << 32) | gid; mou->mou_opencount = 0; mou->mou_deniednext = ktime_set(0, 0); @@ -142,7 +143,8 @@ static void mdd_obj_user_free(struct mdd_object_user *mou) */ static struct mdd_object_user *mdd_obj_user_find(struct mdd_object *mdd_obj, - uid_t uid, gid_t gid, int flags) + uid_t uid, gid_t gid, + u64 open_flags) { struct mdd_object_user *mou; __u64 uidgid; @@ -152,7 +154,8 @@ struct mdd_object_user *mdd_obj_user_find(struct mdd_object *mdd_obj, uidgid = ((__u64)uid << 32) | gid; list_for_each_entry(mou, &mdd_obj->mod_users, mou_list) { if (mou->mou_uidgid == uidgid && - flags_helper(mou->mou_flags) == flags_helper(flags)) + flags_helper(mou->mou_open_flags) == + flags_helper(open_flags)) RETURN(mou); } RETURN(NULL); @@ -177,9 +180,7 @@ static int mdd_obj_user_add(struct mdd_object *mdd_obj, __u32 gid = mou->mou_uidgid & ((1UL << 32) - 1); ENTRY; - - tmp = mdd_obj_user_find(mdd_obj, uid, - gid, mou->mou_flags); + tmp = mdd_obj_user_find(mdd_obj, uid, gid, mou->mou_open_flags); if (tmp != NULL) RETURN(-EEXIST); @@ -234,8 +235,7 @@ int mdd_la_get(const struct lu_env *env, struct mdd_object *obj, return rc; } - if (la->la_valid & LA_FLAGS && - la->la_flags & LUSTRE_ORPHAN_FL) + if (la->la_valid & LA_FLAGS && la->la_flags & LUSTRE_ORPHAN_FL) obj->mod_flags |= ORPHAN_OBJ | DEAD_OBJ; return 0; @@ -243,19 +243,14 @@ int mdd_la_get(const struct lu_env *env, struct mdd_object *obj, struct mdd_thread_info *mdd_env_info(const struct lu_env *env) { - struct mdd_thread_info *info; - - lu_env_refill((struct lu_env *)env); - info = lu_context_key_get(&env->le_ctx, &mdd_thread_key); - LASSERT(info != NULL); - return info; + return lu_env_info(env, &mdd_thread_key); } 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; @@ -266,7 +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; @@ -340,6 +335,7 @@ static void mdd_object_free(const struct lu_env *env, struct lu_object *o) } lu_object_fini(o); + /* mdd doesn't contain an lu_object_header, so don't need call_rcu */ OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem); } @@ -421,14 +417,14 @@ static int mdd_xattr_get(const struct lu_env *env, strcmp(name, XATTR_NAME_LINK) == 0)) RETURN(-ENOENT); - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_read_lock(env, mdd_obj, DT_TGT_CHILD); rc = mdo_xattr_get(env, mdd_obj, buf, name); mdd_read_unlock(env, mdd_obj); mdd = mdo2mdd(obj); /* record only getting user xattrs and acls */ - if (rc >= 0 && + if (rc >= 0 && buf->lb_buf && mdd_changelog_enabled(env, mdd, CL_GETXATTR) && (has_prefix(name, XATTR_USER_PREFIX) || has_prefix(name, XATTR_NAME_POSIX_ACL_ACCESS) || @@ -436,7 +432,7 @@ static int mdd_xattr_get(const struct lu_env *env, struct thandle *handle; int rc2; - LASSERT(mdo2fid(mdd_obj) != NULL); + LASSERT(mdd_object_fid(mdd_obj) != NULL); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) @@ -452,8 +448,8 @@ static int mdd_xattr_get(const struct lu_env *env, GOTO(stop, rc2); rc2 = mdd_changelog_data_store_by_fid(env, mdd, CL_GETXATTR, 0, - mdo2fid(mdd_obj), name, - handle); + mdd_object_fid(mdd_obj), + NULL, name, handle); stop: rc2 = mdd_trans_stop(env, mdd, rc2, handle); @@ -487,7 +483,7 @@ int mdd_readlink(const struct lu_env *env, struct md_object *obj, LASSERT(next != NULL); LASSERT(next->do_body_ops != NULL); LASSERT(next->do_body_ops->dbo_read != NULL); - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_read_lock(env, mdd_obj, DT_TGT_CHILD); rc = dt_read(env, next, buf, &pos); mdd_read_unlock(env, mdd_obj); RETURN(rc); @@ -504,7 +500,7 @@ static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj, ENTRY; - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_read_lock(env, mdd_obj, DT_TGT_CHILD); rc = mdo_xattr_list(env, mdd_obj, buf); mdd_read_unlock(env, mdd_obj); @@ -554,7 +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; @@ -586,7 +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; @@ -605,7 +601,7 @@ int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj, ENTRY; rc = mdo_attr_set(env, obj, attr, handle); -#ifdef CONFIG_FS_POSIX_ACL +#ifdef CONFIG_LUSTRE_FS_POSIX_ACL if (!rc && (attr->la_valid & LA_MODE) && needacl) rc = mdd_acl_chmod(env, obj, attr->la_mode, handle); #endif @@ -637,6 +633,22 @@ int mdd_update_time(const struct lu_env *env, struct mdd_object *obj, RETURN(rc); } + +static bool is_project_state_change(const struct lu_attr *oattr, + struct lu_attr *la) +{ + if (la->la_valid & LA_PROJID && + oattr->la_projid != la->la_projid) + return true; + + if ((la->la_valid & LA_FLAGS) && + (la->la_flags & LUSTRE_PROJINHERIT_FL) != + (oattr->la_flags & LUSTRE_PROJINHERIT_FL)) + return true; + + return false; +} + /* * This gives the same functionality as the code between * sys_chmod and inode_setattr @@ -646,10 +658,12 @@ int mdd_update_time(const struct lu_env *env, struct mdd_object *obj, */ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, const struct lu_attr *oattr, struct lu_attr *la, - const unsigned long flags) + const struct md_attr *ma) { struct lu_ucred *uc; int rc = 0; + const unsigned long flags = ma->ma_attr_flags; + ENTRY; if (!la->la_valid) @@ -669,6 +683,14 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, if (uc == NULL) RETURN(0); + if (is_project_state_change(oattr, la)) { + 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))) + RETURN(-EPERM); + } + if (la->la_valid == LA_CTIME) { if (!(flags & MDS_PERM_BYPASS)) /* This is only for set ctime when rename's source is @@ -680,11 +702,19 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, RETURN(rc); } - if (la->la_valid == LA_ATIME) { - /* This is an atime-only attribute update for close RPCs. */ - if (la->la_atime < (oattr->la_atime + + if (flags & MDS_CLOSE_UPDATE_TIMES && + la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) { + /* This is an atime/mtime/ctime attribute update for + * close RPCs. + */ + if (la->la_valid & LA_ATIME && + la->la_atime <= (oattr->la_atime + mdd_obj2mdd_dev(obj)->mdd_atime_diff)) la->la_valid &= ~LA_ATIME; + if (la->la_valid & LA_CTIME && la->la_ctime <= oattr->la_ctime) + la->la_valid &= ~LA_CTIME; + if (la->la_valid & LA_MTIME && la->la_mtime <= oattr->la_mtime) + la->la_valid &= ~LA_MTIME; RETURN(0); } @@ -696,17 +726,26 @@ 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, CFS_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, CFS_CAP_LINUX_IMMUTABLE)) + !cap_raised(uc->uc_cap, CAP_LINUX_IMMUTABLE)) RETURN(-EPERM); - if (!S_ISDIR(oattr->la_mode)) + if (!S_ISDIR(oattr->la_mode)) { la->la_flags &= ~(LUSTRE_DIRSYNC_FL | LUSTRE_TOPDIR_FL); + } else if (la->la_flags & LUSTRE_ENCRYPT_FL) { + /* when trying to add encryption flag on dir, + * make sure it is empty + */ + rc = mdd_dir_is_empty(env, obj); + if (rc) + RETURN(rc); + rc = 0; + } } if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) && @@ -718,7 +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, CFS_CAP_FOWNER)) { + !cap_raised(uc->uc_cap, CAP_FOWNER)) { rc = mdd_permission_internal(env, obj, oattr, MAY_WRITE); if (rc) @@ -751,7 +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, CFS_CAP_FOWNER)) + !cap_raised(uc->uc_cap, CAP_FOWNER)) RETURN(-EPERM); if (la->la_mode == (umode_t) -1) @@ -763,7 +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, CFS_CAP_FSETID)) + !cap_raised(uc->uc_cap, CAP_FSETID)) la->la_mode &= ~S_ISGID; } else { la->la_mode = oattr->la_mode; @@ -775,7 +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, CFS_CAP_CHOWN)) + !cap_raised(uc->uc_cap, CAP_CHOWN)) RETURN(-EPERM); /* If the user or group of a non-directory has been @@ -801,7 +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, CFS_CAP_CHOWN)) + !cap_raised(uc->uc_cap, CAP_CHOWN)) RETURN(-EPERM); /* Likewise, if the user or group of a non-directory @@ -831,8 +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); } @@ -843,22 +884,24 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, static int mdd_changelog_data_store_by_fid(const struct lu_env *env, struct mdd_device *mdd, enum changelog_rec_type type, - int flags, const struct lu_fid *fid, + enum changelog_rec_flags clf_flags, + const struct lu_fid *fid, + const struct lu_fid *pfid, const char *xattr_name, struct thandle *handle) { const struct lu_ucred *uc = lu_ucred(env); + enum changelog_rec_extra_flags xflags = CLFE_INVALID; struct llog_changelog_rec *rec; struct lu_buf *buf; int reclen; - int xflags = CLFE_INVALID; int rc; - flags = (flags & CLF_FLAGMASK) | CLF_VERSION | CLF_EXTRA_FLAGS; + clf_flags = (clf_flags & CLF_FLAGMASK) | CLF_VERSION | CLF_EXTRA_FLAGS; if (uc) { if (uc->uc_jobid[0] != '\0') - flags |= CLF_JOBID; + clf_flags |= CLF_JOBID; xflags |= CLFE_UIDGID; xflags |= CLFE_NID; } @@ -868,23 +911,25 @@ static int mdd_changelog_data_store_by_fid(const struct lu_env *env, xflags |= CLFE_XATTR; reclen = llog_data_len(LLOG_CHANGELOG_HDR_SZ + - changelog_rec_offset(flags & CLF_SUPPORTED, + 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; rec->cr_hdr.lrh_len = reclen; - rec->cr.cr_flags = flags; + rec->cr.cr_flags = clf_flags; rec->cr.cr_type = (__u32)type; rec->cr.cr_tfid = *fid; + if (pfid) + rec->cr.cr_pfid = *pfid; rec->cr.cr_namelen = 0; - if (flags & CLF_JOBID) + if (clf_flags & CLF_JOBID) mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid); - if (flags & CLF_EXTRA_FLAGS) { + if (clf_flags & CLF_EXTRA_FLAGS) { mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags); if (xflags & CLFE_UIDGID) mdd_changelog_rec_extra_uidgid(&rec->cr, @@ -892,7 +937,7 @@ static int mdd_changelog_data_store_by_fid(const struct lu_env *env, if (xflags & CLFE_NID) mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid); if (xflags & CLFE_OPEN) - mdd_changelog_rec_extra_omode(&rec->cr, flags); + mdd_changelog_rec_extra_omode(&rec->cr, clf_flags); if (xflags & CLFE_XATTR) { if (xattr_name == NULL) RETURN(-EINVAL); @@ -910,10 +955,13 @@ static int mdd_changelog_data_store_by_fid(const struct lu_env *env, * want the change to commit without the log entry. * \param mdd_obj - mdd_object of change * \param handle - transaction handle + * \param pfid - parent FID for CL_MTIME changelogs */ int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd, - enum changelog_rec_type type, int flags, - struct mdd_object *mdd_obj, struct thandle *handle) + enum changelog_rec_type type, + enum changelog_rec_flags clf_flags, + struct mdd_object *mdd_obj, struct thandle *handle, + const struct lu_fid *pfid) { int rc; @@ -934,8 +982,9 @@ int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd, RETURN(0); } - rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags, - mdo2fid(mdd_obj), NULL, handle); + rc = mdd_changelog_data_store_by_fid(env, mdd, type, clf_flags, + mdd_object_fid(mdd_obj), pfid, + NULL, handle); if (rc == 0) mdd_obj->mod_cltime = ktime_get(); @@ -945,11 +994,12 @@ int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd, int mdd_changelog_data_store_xattr(const struct lu_env *env, struct mdd_device *mdd, enum changelog_rec_type type, - int flags, struct mdd_object *mdd_obj, + enum changelog_rec_flags clf_flags, + struct mdd_object *mdd_obj, const char *xattr_name, struct thandle *handle) { - int rc; + int rc; LASSERT(mdd_obj != NULL); LASSERT(handle != NULL); @@ -969,17 +1019,20 @@ int mdd_changelog_data_store_xattr(const struct lu_env *env, RETURN(0); } - rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags, - mdo2fid(mdd_obj), xattr_name, - handle); + rc = mdd_changelog_data_store_by_fid(env, mdd, type, clf_flags, + mdd_object_fid(mdd_obj), NULL, + xattr_name, handle); if (rc == 0) mdd_obj->mod_cltime = ktime_get(); RETURN(rc); } +/* only the bottom CLF_FLAGSHIFT bits of @flags are stored in the record, + * except for open flags have a dedicated record to store 32 bits of flags */ static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type, - int flags, struct md_device *m, const struct lu_fid *fid) + enum changelog_rec_flags clf_flags, + struct md_device *m, const struct lu_fid *fid) { struct thandle *handle; struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); @@ -1005,8 +1058,8 @@ static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type, if (rc) GOTO(stop, rc); - rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags, - fid, NULL, handle); + rc = mdd_changelog_data_store_by_fid(env, mdd, type, clf_flags, + fid, NULL, NULL, handle); stop: rc = mdd_trans_stop(env, mdd, rc, handle); @@ -1027,28 +1080,28 @@ stop: * atime and ctime are independent.) */ static int mdd_attr_set_changelog(const struct lu_env *env, struct md_object *obj, struct thandle *handle, - __u64 valid) + const struct lu_fid *pfid, __u64 valid) { struct mdd_device *mdd = mdo2mdd(obj); int bits, type = 0; - bits = (valid & LA_SIZE) ? 1 << CL_TRUNC : 0; - bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0; - bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0; - bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0; - bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0; - bits = bits & mdd->mdd_cl.mc_mask; + bits = (valid & LA_SIZE) ? BIT(CL_TRUNC) : 0; + bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? BIT(CL_SETATTR) : 0; + 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_current_mask; /* This is an implementation limit rather than a protocol limit */ - CLASSERT(CL_LAST <= sizeof(int) * 8); + BUILD_BUG_ON(CL_LAST > sizeof(int) * 8); if (bits == 0) return 0; /* The record type is the lowest non-masked set bit */ type = __ffs(bits); - /* FYI we only store the first CLF_FLAGMASK bits of la_valid */ - return mdd_changelog_data_store(env, mdd, type, (int)valid, - md2mdd_obj(obj), handle); + /* XXX: we only store the low CLF_FLAGMASK bits of la_valid */ + return mdd_changelog_data_store(env, mdd, type, valid, md2mdd_obj(obj), + handle, pfid); } static int mdd_declare_attr_set(const struct lu_env *env, @@ -1063,9 +1116,9 @@ static int mdd_declare_attr_set(const struct lu_env *env, if (rc) return rc; -#ifdef CONFIG_FS_POSIX_ACL +#ifdef CONFIG_LUSTRE_FS_POSIX_ACL if (attr->la_valid & LA_MODE) { - mdd_read_lock(env, obj, MOR_TGT_CHILD); + mdd_read_lock(env, obj, DT_TGT_CHILD); rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_ACL_ACCESS); mdd_read_unlock(env, obj); @@ -1092,6 +1145,7 @@ static int mdd_declare_attr_set(const struct lu_env *env, /* * LU-3671 + * LU-7239 * * permission changes may require sync operation, to mitigate performance * impact, only do this for dir and when permission is reduced. @@ -1106,17 +1160,24 @@ static inline bool permission_needs_sync(const struct lu_attr *old, if (!S_ISDIR(old->la_mode)) return false; - if (new->la_valid & (LA_UID | LA_GID)) + if (new->la_valid & LA_UID && old->la_uid != new->la_uid) return true; - if (new->la_valid & LA_MODE && - new->la_mode & (S_ISUID | S_ISGID | S_ISVTX)) + if (new->la_valid & LA_GID && old->la_gid != new->la_gid) return true; - if ((new->la_valid & LA_MODE) && - ((new->la_mode & old->la_mode) & S_IRWXUGO) != - (old->la_mode & S_IRWXUGO)) - return true; + if (new->la_valid & LA_MODE) { + /* turned on sticky bit */ + if (!(old->la_mode & S_ISVTX) && (new->la_mode & S_ISVTX)) + return true; + + /* set-GID has no impact on what is allowed, not checked */ + + /* turned off setuid bit, or one of rwx for someone */ + if (((new->la_mode & old->la_mode) & (0777 | S_ISUID)) != + (old->la_mode & (0777 | S_ISUID))) + return true; + } return false; } @@ -1152,10 +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; + struct lquota_id_info qi; + bool quota_reserved = false; + bool chrgrp_by_unprivileged_user = false; int rc; ENTRY; @@ -1168,15 +1232,16 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, RETURN(rc); *la_copy = ma->ma_attr; - rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags); + rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma); if (rc) RETURN(rc); /* no need to setattr anymore */ if (la_copy->la_valid == 0) { - CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous" - "valid is %#llx\n", mdd2obd_dev(mdd)->obd_name, - PFID(mdo2fid(mdd_obj)), la->la_valid); + CDEBUG(D_INODE, + "%s: no valid attribute on "DFID", previous was %#llx\n", + mdd_obj_dev_name(mdd_obj), + PFID(mdd_object_fid(mdd_obj)), la->la_valid); RETURN(0); } @@ -1185,17 +1250,54 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, * the setattr operation will be processed synchronously to * honor the quota limit of the corresponding group. see LU-5152 */ uc = lu_ucred_check(env); + memset(&qi, 0, sizeof(qi)); if (S_ISREG(attr->la_mode) && la->la_valid & LA_GID && la->la_gid != attr->la_gid && uc != NULL && uc->uc_fsuid != 0) { - 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) + qi.lqi_space = la->la_blocks << 9; + else if (la->la_valid & LA_SIZE) + qi.lqi_space = la->la_size; + /* use local attr gotten above */ + else if (attr->la_valid & LA_BLOCKS) + qi.lqi_space = attr->la_blocks << 9; + else if (attr->la_valid & LA_SIZE) + qi.lqi_space = attr->la_size; + + if (qi.lqi_space > 0) { + qi.lqi_id.qid_gid = la->la_gid; + qi.lqi_type = GRPQUOTA; + qi.lqi_space = toqb(qi.lqi_space); + qi.lqi_is_blk = true; + rc = dt_reserve_or_free_quota(env, mdd->mdd_bottom, + &qi); + + if (rc) { + CDEBUG(D_QUOTA, "%s: failed to reserve quota for gid %d size %llu\n", + mdd2obd_dev(mdd)->obd_name, + la->la_gid, qi.lqi_space); + + GOTO(out, rc); + } + + quota_reserved = true; + la_copy->la_valid |= LA_FLAGS; + } - /* Flush the possible existing sync requests to OSTs to - * keep the order of sync for the current setattr operation + chrgrp_by_unprivileged_user = true; + + /* Flush the possible existing client setattr requests to OSTs + * to keep the order with the current setattr operation that * will be sent directly to OSTs. see LU-5152 */ + /* LU-11303 disable sync as this is too heavyweight. + * This should be replaced with a sync only for the object + * being modified here, not the whole filesystem. rc = dt_sync(env, mdd->mdd_child); if (rc) GOTO(out, rc); + */ } handle = mdd_trans_create(env, mdd); @@ -1214,14 +1316,15 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, if (rc) GOTO(out, rc); - if (mdd->mdd_sync_permission && permission_needs_sync(attr, la)) + if (!chrgrp_by_unprivileged_user && mdd->mdd_sync_permission && + permission_needs_sync(attr, la)) handle->th_sync = 1; if (la->la_valid & (LA_MTIME | LA_CTIME)) CDEBUG(D_INODE, "setting mtime %llu, ctime %llu\n", la->la_mtime, la->la_ctime); - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_write_lock(env, mdd_obj, DT_TGT_CHILD); /* LU-10509: setattr of LA_SIZE should be skipped case of DOM, * otherwise following truncate will do nothing and truncated @@ -1255,9 +1358,24 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, out: if (rc == 0) - rc = mdd_attr_set_changelog(env, obj, handle, + 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))) { + qi.lqi_space *= -1; + dt_reserve_or_free_quota(env, mdd->mdd_bottom, &qi); + } else { + sub_th->th_reserved_quota = qi; + } + } else if (quota_reserved) { + qi.lqi_space *= -1; + dt_reserve_or_free_quota(env, mdd->mdd_bottom, &qi); + } + if (handle != NULL) rc = mdd_trans_stop(env, mdd, rc, handle); @@ -1281,11 +1399,11 @@ 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, CFS_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, CFS_CAP_FOWNER)) { + !cap_raised(uc->uc_cap, CAP_FOWNER)) { RETURN(-EPERM); } @@ -1309,7 +1427,7 @@ static inline bool has_prefix(const char *str, const char *prefix) * * \param[in] xattr_name Full extended attribute name. * - * \return The type of changelog to use, or -1 if no changelog is to be emitted. + * \return type of changelog to use, or CL_NONE if no changelog is to be emitted */ static enum changelog_rec_type mdd_xattr_changelog_type(const struct lu_env *env, struct mdd_device *mdd, @@ -1317,8 +1435,8 @@ mdd_xattr_changelog_type(const struct lu_env *env, struct mdd_device *mdd, { /* Layout changes systematically recorded */ if (strcmp(XATTR_NAME_LOV, xattr_name) == 0 || - strncmp(XATTR_LUSTRE_LOV, xattr_name, - strlen(XATTR_LUSTRE_LOV)) == 0) + strcmp(XATTR_LUSTRE_LOV, xattr_name) == 0 || + allowed_lustre_lov(xattr_name)) return CL_LAYOUT; /* HSM information changes systematically recorded */ @@ -1372,23 +1490,26 @@ static int mdd_declare_xattr_set(const struct lu_env *env, static int mdd_hsm_update_locked(const struct lu_env *env, struct md_object *obj, const struct lu_buf *buf, - struct thandle *handle, int *cl_flags) + struct thandle *handle, + enum changelog_rec_flags *clf_flags) { struct mdd_thread_info *info = mdd_env_info(env); - struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct lu_buf *current_buf; - struct md_hsm *current_mh; - struct md_hsm *new_mh; - int rc; - ENTRY; + struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct lu_buf *current_buf; + struct md_hsm *current_mh; + struct md_hsm *new_mh; + int rc; + ENTRY; OBD_ALLOC_PTR(current_mh); if (current_mh == NULL) RETURN(-ENOMEM); /* Read HSM attrs from disk */ - current_buf = lu_buf_check_and_alloc(&info->mti_xattr_buf, - mdo2mdd(obj)->mdd_dt_conf.ddp_max_ea_size); + 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)); rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM); rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh); if (rc < 0 && rc != -ENODATA) @@ -1406,9 +1527,9 @@ static int mdd_hsm_update_locked(const struct lu_env *env, /* Flags differ, set flags for the changelog that will be added */ if (current_mh->mh_flags != new_mh->mh_flags) { - hsm_set_cl_event(cl_flags, HE_STATE); + hsm_set_cl_event(clf_flags, HE_STATE); if (new_mh->mh_flags & HS_DIRTY) - hsm_set_cl_flags(cl_flags, CLF_HSM_DIRTY); + hsm_set_cl_flags(clf_flags, CLF_HSM_DIRTY); } OBD_FREE_PTR(new_mh); @@ -1467,64 +1588,64 @@ 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(mdo2fid(obj), mdo2fid(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, MOR_TGT_CHILD); - mdd_write_lock(env, vic, MOR_TGT_CHILD); - } else { - mdd_write_lock(env, vic, MOR_TGT_CHILD); - mdd_write_lock(env, obj, MOR_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); + + 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_LUSTRE_LOV, LU_XATTR_MERGE, + rc = mdo_xattr_set(env, obj, buf_vic, XATTR_NAME_LOV, LU_XATTR_MERGE, handle); if (rc) GOTO(out, rc); @@ -1533,8 +1654,10 @@ static int mdd_xattr_merge(const struct lu_env *env, struct md_object *md_obj, if (rc) /* wtf? */ GOTO(out_restore, rc); - (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle); - (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle); + (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle, + NULL); + (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle, + NULL); EXIT; out_restore: @@ -1542,15 +1665,16 @@ out_restore: int rc2 = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV, LU_XATTR_REPLACE, handle); if (rc2) - CERROR("%s: failed to rollback of layout of: "DFID - ": %d, file state unknown\n", - mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2); + CERROR("%s: failed rollback of "DFID" layout: file state unknown: rc = %d\n", + mdd_obj_dev_name(obj), + PFID(mdd_object_fid(obj)), rc2); } 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); @@ -1686,102 +1810,175 @@ static int mdd_split_ea(struct lov_comp_md_v1 *comp_v1, __u16 mirror_id, return 0; } +static int mdd_dom_data_truncate(const struct lu_env *env, + struct mdd_device *mdd, struct mdd_object *mo); + static int mdd_xattr_split(const struct lu_env *env, struct md_object *md_obj, struct md_rejig_data *mrd) { struct mdd_device *mdd = mdo2mdd(md_obj); struct mdd_object *obj = md2mdd_obj(md_obj); - struct mdd_object *vic = md2mdd_obj(mrd->mrd_obj); - 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 mdd_object *vic = NULL; + 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; + bool dom_stripe = false; + ENTRY; - rc = lu_fid_cmp(mdo2fid(obj), mdo2fid(vic)); - if (rc == 0) /* same fid */ - RETURN(-EPERM); + /** + * NULL @mrd_obj means mirror deleting, and use NULL vic to indicate + * mirror deleting + */ + if (mrd->mrd_obj) + vic = md2mdd_obj(mrd->mrd_obj); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - if (rc > 0) { - mdd_write_lock(env, obj, MOR_TGT_CHILD); - mdd_write_lock(env, vic, MOR_TGT_CHILD); - } else { - mdd_write_lock(env, vic, MOR_TGT_CHILD); - mdd_write_lock(env, obj, MOR_TGT_CHILD); - } - /* 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 - * mirror layout to the victim file. + * mirror layout to the victim buffer. */ memset(buf, 0, sizeof(*buf)); 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, rc); - rc = mdd_declare_xattr_set(env, mdd, vic, buf_vic, XATTR_NAME_LOV, - LU_XATTR_SPLIT, handle); - if (rc) - GOTO(out, rc); + dom_stripe = mdd_lmm_dom_size(buf_vic->lb_buf) > 0; + + if (vic) { + /** + * non delete mirror split + * + * declare obj set remaining layout in @buf, will set obj's + * in-memory layout + */ + rc = mdd_declare_xattr_set(env, mdd, obj, buf, XATTR_NAME_LOV, + LU_XATTR_SPLIT, handle); + if (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(stop, rc); + } else { + /** + * declare delete mirror objects in @buf_vic, will change obj's + * in-memory layout + */ + rc = mdd_declare_xattr_set(env, mdd, obj, buf_vic, + XATTR_NAME_LOV, LU_XATTR_PURGE, + handle); + if (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(stop, rc); + } rc = mdd_trans_start(env, mdd, handle); if (rc) - GOTO(out, rc); + GOTO(stop, rc); - rc = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV, LU_XATTR_REPLACE, - handle); - if (rc) - GOTO(out, 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); - rc = mdo_xattr_set(env, vic, buf_vic, XATTR_NAME_LOV, LU_XATTR_CREATE, + 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_SPLIT, handle); if (rc) - GOTO(out_restore, rc); + GOTO(unlock, rc); - rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle); - if (rc) - GOTO(out, rc); + if (vic) { + /* set vic's layout in @buf_vic */ + rc = mdo_xattr_set(env, vic, buf_vic, XATTR_NAME_LOV, + LU_XATTR_CREATE, handle); + if (rc) + GOTO(out_restore, rc); + } else { + /* delete mirror objects */ + rc = mdo_xattr_set(env, obj, buf_vic, XATTR_NAME_LOV, + LU_XATTR_PURGE, handle); + if (rc) + GOTO(out_restore, rc); + } - rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle); + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle, + NULL); if (rc) - GOTO(out, rc); - EXIT; + GOTO(out_restore, rc); + + if (vic) { + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, + handle, NULL); + if (rc) + GOTO(out_restore, rc); + } out_restore: if (rc) { - /* restore obj's layout */ + /* restore obj's in-memory and on-disk layout */ int rc2 = mdo_xattr_set(env, obj, buf_save, XATTR_NAME_LOV, LU_XATTR_REPLACE, handle); if (rc2) - CERROR("%s: failed to rollback of layout of: "DFID - ": %d, file state unkonwn.\n", - mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2); + CERROR("%s: failed rollback "DFID + " layout: file state unknown: rc = %d\n", + mdd_obj_dev_name(obj), + PFID(mdd_object_fid(obj)), rc); } -out: - mdd_trans_stop(env, mdd, rc, handle); + +unlock: mdd_write_unlock(env, obj); - mdd_write_unlock(env, vic); + 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); + lu_buf_free(buf_save); lu_buf_free(buf); lu_buf_free(buf_vic); @@ -1816,13 +2013,13 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, const struct lu_buf *buf, const char *name, int fl) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct lu_attr *attr = MDD_ENV_VAR(env, cattr); - struct mdd_device *mdd = mdo2mdd(obj); - struct thandle *handle; + struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); + struct mdd_device *mdd = mdo2mdd(obj); + struct thandle *handle; enum changelog_rec_type cl_type; - int cl_flags = 0; - int rc; + enum changelog_rec_flags clf_flags = 0; + int rc; ENTRY; rc = mdd_la_get(env, mdd_obj, attr); @@ -1841,20 +2038,16 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, if (buf->lb_len != sizeof(*mrd)) RETURN(-EINVAL); - rc = mdd_layout_merge_allowed(env, obj, victim); - if (rc) - RETURN(rc); - if (fl == LU_XATTR_MERGE) + if (fl == LU_XATTR_MERGE) { + rc = mdd_layout_merge_allowed(env, obj, victim); + if (rc) + RETURN(rc); /* merge layout of victim as a mirror of obj's. */ rc = mdd_xattr_merge(env, obj, victim); - else + } else { rc = mdd_xattr_split(env, obj, mrd); - RETURN(rc); - } - - if (strcmp(name, XATTR_NAME_LMV) == 0) { - rc = mdd_dir_layout_shrink(env, obj, buf); + } RETURN(rc); } @@ -1892,10 +2085,10 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, if (rc) GOTO(stop, rc); - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_write_lock(env, mdd_obj, DT_TGT_CHILD); if (strcmp(XATTR_NAME_HSM, name) == 0) { - rc = mdd_hsm_update_locked(env, obj, buf, handle, &cl_flags); + rc = mdd_hsm_update_locked(env, obj, buf, handle, &clf_flags); if (rc) { mdd_write_unlock(env, mdd_obj); GOTO(stop, rc); @@ -1911,7 +2104,7 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, if (cl_type < 0) GOTO(stop, rc = 0); - rc = mdd_changelog_data_store_xattr(env, mdd, cl_type, cl_flags, + rc = mdd_changelog_data_store_xattr(env, mdd, cl_type, clf_flags, mdd_obj, name, handle); EXIT; @@ -1973,7 +2166,7 @@ static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, if (rc) GOTO(stop, rc); - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_write_lock(env, mdd_obj, DT_TGT_CHILD); rc = mdo_xattr_del(env, mdd_obj, name, handle); mdd_write_unlock(env, mdd_obj); if (rc) @@ -1997,7 +2190,7 @@ 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; @@ -2011,9 +2204,9 @@ 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); @@ -2030,7 +2223,7 @@ 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); @@ -2042,8 +2235,8 @@ static int mdd_xattr_hsm_replace(const struct lu_env *env, struct thandle *handle) { struct hsm_attrs *attrs; - __u32 hsm_flags; - int flags = 0; + enum hsm_states hsm_flags; + enum changelog_rec_flags clf_flags = 0; int rc; ENTRY; @@ -2058,9 +2251,9 @@ static int mdd_xattr_hsm_replace(const struct lu_env *env, RETURN(0); /* Add a changelog record for release. */ - hsm_set_cl_event(&flags, HE_RELEASE); + hsm_set_cl_event(&clf_flags, HE_RELEASE); rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM, - flags, o, handle); + clf_flags, o, handle, NULL); RETURN(rc); } @@ -2079,11 +2272,11 @@ static int mdd_layout_swap_allowed(const struct lu_env *env, const struct lu_attr *attr2, __u64 flags) { - const struct lu_fid *fid1, *fid2; + const struct lu_fid *fid1, *fid2; ENTRY; - fid1 = mdo2fid(o1); - fid2 = mdo2fid(o2); + fid1 = mdd_object_fid(o1); + fid2 = mdd_object_fid(o2); if (!fid_is_norm(fid1) && (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1)))) @@ -2224,8 +2417,8 @@ stop: out: /* Ignore failure but report the error */ if (rc) - CERROR("%s: "DFID" can't truncate DOM inode data, rc = %d\n", - mdd_obj_dev_name(mo), PFID(mdo2fid(mo)), rc); + CERROR("%s: can't truncate DOM inode "DFID" data: rc = %d\n", + mdd_obj_dev_name(mo), PFID(mdd_object_fid(mo)), rc); return rc; } @@ -2242,10 +2435,10 @@ 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; @@ -2256,12 +2449,12 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, ENTRY; - CLASSERT(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 */ - rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o)); + rc = lu_fid_cmp(mdd_object_fid(fst_o), mdd_object_fid(snd_o)); if (rc == 0) /* same fid ? */ RETURN(-EPERM); @@ -2286,8 +2479,8 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, RETURN(PTR_ERR(handle)); /* objects are already sorted */ - mdd_write_lock(env, fst_o, MOR_TGT_CHILD); - mdd_write_lock(env, snd_o, MOR_TGT_CHILD); + mdd_write_lock(env, fst_o, DT_TGT_CHILD); + mdd_write_lock(env, snd_o, DT_TGT_CHILD); rc = mdd_stripe_get(env, fst_o, fst_buf, XATTR_NAME_LOV); if (rc < 0 && rc != -ENODATA) @@ -2326,15 +2519,17 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, if (domsize_vlt > 0 && domsize_dom == 0) { rc = -EOPNOTSUPP; - CDEBUG(D_LAYOUT, "cannot swap layout for "DFID": OST to DOM " - "migration is not supported: rc = %d\n", - PFID(mdo2fid(snd_o)), rc); + CDEBUG(D_LAYOUT, + "%s: cannot swap "DFID" layout: OST to DOM migration not supported: rc = %d\n", + mdd_obj_dev_name(snd_o), + PFID(mdd_object_fid(snd_o)), rc); GOTO(stop, rc); } else if (domsize_vlt > 0 && domsize_dom != domsize_vlt) { rc = -EOPNOTSUPP; - CDEBUG(D_LAYOUT, "cannot swap layout for "DFID": new layout " - "must have the same DoM component size: rc = %d\n", - PFID(mdo2fid(fst_o)), rc); + CDEBUG(D_LAYOUT, + "%s: cannot swap "DFID" layout: new layout must have same DoM component size: rc = %d\n", + mdd_obj_dev_name(fst_o), + PFID(mdd_object_fid(fst_o)), rc); GOTO(stop, rc); } else if (domsize_vlt > 0) { /* Migration with the same DOM component size, no need to @@ -2392,10 +2587,13 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, /* set the file specific informations in lmm */ if (fst_lmm != NULL) { - saved_oi = &info->mti_oa.o_oi; + struct ost_id temp_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); - mdd_set_lmm_oi(fst_lmm, &snd_lmm->lmm_oi); + mdd_set_lmm_oi(fst_lmm, &temp_oi); mdd_set_lmm_oi(snd_lmm, saved_oi); } else { if ((snd_lmm->lmm_magic & cpu_to_le32(LOV_MAGIC_MASK)) == @@ -2467,9 +2665,9 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle); if (rc2 < 0) - CERROR("%s: restore "DFID" HSM error: %d/%d\n", + CERROR("%s: HSM error restoring "DFID": rc = %d/%d\n", mdd_obj_dev_name(fst_o), - PFID(mdo2fid(fst_o)), rc, rc2); + PFID(mdd_object_fid(fst_o)), rc, rc2); GOTO(stop, rc); } } @@ -2491,11 +2689,13 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, GOTO(out_restore, rc); /* Issue one changelog record per file */ - rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle); + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle, + NULL); if (rc) GOTO(stop, rc); - rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle); + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle, + NULL); if (rc) GOTO(stop, rc); EXIT; @@ -2517,21 +2717,25 @@ 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) { /* very bad day */ - CERROR("%s: unable to roll back layout swap. FIDs: " - DFID" and "DFID "error: %d/%d, steps: %d\n", + CERROR("%s: unable to roll back layout swap of "DFID" and "DFID", steps: %d: rc = %d/%d\n", mdd_obj_dev_name(fst_o), - PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)), + PFID(mdd_object_fid(snd_o)), + PFID(mdd_object_fid(fst_o)), rc, rc2, steps); /* a solution to avoid journal commit is to panic, * but it has strong consequences so we use LBUG to @@ -2605,13 +2809,14 @@ mdd_layout_instantiate_component(const struct lu_env *env, if (rc) RETURN(rc); - mdd_write_lock(env, obj, MOR_TGT_CHILD); + mdd_write_lock(env, obj, DT_TGT_CHILD); rc = mdo_layout_change(env, obj, mlc, handle); mdd_write_unlock(env, obj); if (rc) RETURN(rc); - rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle); + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle, + NULL); RETURN(rc); } @@ -2629,7 +2834,7 @@ 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; @@ -2657,6 +2862,12 @@ mdd_layout_update_rdonly(const struct lu_env *env, struct mdd_object *obj, 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); @@ -2682,7 +2893,7 @@ mdd_layout_update_rdonly(const struct lu_env *env, struct mdd_object *obj, /* it needs a sync tx to make FLR to work properly */ handle->th_sync = 1; - mdd_write_lock(env, obj, MOR_TGT_CHILD); + mdd_write_lock(env, obj, DT_TGT_CHILD); rc = mdo_layout_change(env, obj, mlc, handle); if (!rc && fl) { /* SOM state transition from STRICT to STALE */ @@ -2695,7 +2906,7 @@ mdd_layout_update_rdonly(const struct lu_env *env, struct mdd_object *obj, if (rc) GOTO(out, rc); - rc = mdd_changelog_data_store(env, mdd, CL_FLRW, 0, obj, handle); + rc = mdd_changelog_data_store(env, mdd, CL_FLRW, 0, obj, handle, NULL); if (rc) GOTO(out, rc); @@ -2719,6 +2930,9 @@ 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)->mdi_buf[1]; + struct lustre_som_attrs *som = &mlc->mlc_som; + int fl = 0; int rc; ENTRY; @@ -2731,8 +2945,25 @@ mdd_layout_update_write_pending(const struct lu_env *env, * 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); @@ -2742,6 +2973,13 @@ mdd_layout_update_write_pending(const struct lu_env *env, 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); @@ -2749,8 +2987,14 @@ mdd_layout_update_write_pending(const struct lu_env *env, /* it needs a sync tx to make FLR to work properly */ handle->th_sync = 1; - mdd_write_lock(env, obj, MOR_TGT_CHILD); + 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); @@ -2778,7 +3022,7 @@ 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; @@ -2845,7 +3089,8 @@ mdd_object_update_sync_pending(const struct lu_env *env, struct mdd_object *obj, GOTO(out, rc); } - rc = mdd_changelog_data_store(env, mdd, CL_RESYNC, 0, obj, handle); + rc = mdd_changelog_data_store(env, mdd, CL_RESYNC, 0, obj, handle, + NULL); if (rc) GOTO(out, rc); EXIT; @@ -2870,8 +3115,24 @@ mdd_layout_change(const struct lu_env *env, struct md_object *o, struct thandle *handle; int flr_state; int rc; + ENTRY; + if (S_ISDIR(mdd_object_type(obj))) { + switch (mlc->mlc_opc) { + case MD_LAYOUT_SHRINK: + rc = mdd_dir_layout_shrink(env, o, mlc); + break; + case MD_LAYOUT_SPLIT: + rc = mdd_dir_layout_split(env, o, mlc); + break; + default: + LBUG(); + } + + RETURN(rc); + } + /* Verify acceptable operations */ switch (mlc->mlc_opc) { case MD_LAYOUT_WRITE: @@ -2932,6 +3193,7 @@ void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent, struct dt_allocation_hint *hint) { struct dt_object *np = parent ? mdd_object_child(parent) : NULL; + struct mdd_device *mdd = mdd_obj2mdd_dev(child); struct dt_object *nc = mdd_object_child(child); memset(hint, 0, sizeof(*hint)); @@ -2942,9 +3204,10 @@ 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; + } 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)), @@ -2953,71 +3216,64 @@ void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent, nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT); } -/* - * do NOT or the MAY_*'s, you'll get the weakest - */ -int accmode(const struct lu_env *env, const struct lu_attr *la, int flags) +static int mdd_accmode(const struct lu_env *env, const struct lu_attr *la, + u64 open_flags) { - int res = 0; - /* Sadly, NFSD reopens a file repeatedly during operation, so the * "acc_mode = 0" allowance for newly-created files isn't honoured. * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file * owner can write to a file even if it is marked readonly to hide * its brokenness. (bug 5781) */ - if (flags & MDS_OPEN_OWNEROVERRIDE) { + if (open_flags & MDS_OPEN_OWNEROVERRIDE) { struct lu_ucred *uc = lu_ucred_check(env); if ((uc == NULL) || (la->la_uid == uc->uc_fsuid)) return 0; } - if (flags & MDS_FMODE_READ) - res |= MAY_READ; - if (flags & (MDS_FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND)) - res |= MAY_WRITE; - if (flags & MDS_FMODE_EXEC) - res = MAY_EXEC; - return res; + return mds_accmode(open_flags); } static int mdd_open_sanity_check(const struct lu_env *env, - struct mdd_object *obj, - const struct lu_attr *attr, int flag) + struct mdd_object *obj, + const struct lu_attr *attr, u64 open_flags, + int is_replay) { - int mode, rc; + unsigned int may_mask; + int rc; ENTRY; /* EEXIST check, also opening of *open* orphans is allowed so we can * open-by-handle unlinked files */ - if (mdd_is_dead_obj(obj) && + if (mdd_is_dead_obj(obj) && !is_replay && likely(!(mdd_is_orphan_obj(obj) && obj->mod_count > 0))) RETURN(-ENOENT); if (S_ISLNK(attr->la_mode)) RETURN(-ELOOP); - mode = accmode(env, attr, flag); + may_mask = mdd_accmode(env, attr, open_flags); - if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE)) + if (S_ISDIR(attr->la_mode) && (may_mask & MAY_WRITE)) RETURN(-EISDIR); - if (!(flag & MDS_OPEN_CREATED)) { - rc = mdd_permission_internal(env, obj, attr, mode); + if (!(open_flags & MDS_OPEN_CREATED)) { + rc = mdd_permission_internal(env, obj, attr, may_mask); if (rc) RETURN(rc); } if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) || S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode)) - flag &= ~MDS_OPEN_TRUNC; + open_flags &= ~MDS_OPEN_TRUNC; /* For writing append-only file must open it with append mode. */ if (attr->la_flags & LUSTRE_APPEND_FL) { - if ((flag & MDS_FMODE_WRITE) && !(flag & MDS_OPEN_APPEND)) + if ((open_flags & MDS_FMODE_WRITE) && + !(open_flags & MDS_OPEN_APPEND)) RETURN(-EPERM); - if (flag & MDS_OPEN_TRUNC) + if (open_flags & MDS_OPEN_TRUNC) RETURN(-EPERM); } @@ -3025,7 +3281,7 @@ static int mdd_open_sanity_check(const struct lu_env *env, } static int mdd_open(const struct lu_env *env, struct md_object *obj, - int flags) + u64 open_flags, struct md_op_spec *spec) { struct mdd_object *mdd_obj = md2mdd_obj(obj); struct md_device *md_dev = lu2md_dev(mdd2lu_dev(mdo2mdd(obj))); @@ -3035,15 +3291,17 @@ static int mdd_open(const struct lu_env *env, struct md_object *obj, struct mdd_device *mdd = mdo2mdd(obj); enum changelog_rec_type type = CL_OPEN; int rc = 0; + ENTRY; - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_write_lock(env, mdd_obj, DT_TGT_CHILD); rc = mdd_la_get(env, mdd_obj, attr); if (rc != 0) GOTO(out, rc); - rc = mdd_open_sanity_check(env, mdd_obj, attr, flags); - if ((rc == -EACCES) && (mdd->mdd_cl.mc_mask & (1 << CL_DN_OPEN))) + rc = mdd_open_sanity_check(env, mdd_obj, attr, open_flags, + spec->no_create); + if ((rc == -EACCES) && (mdd->mdd_cl.mc_current_mask & BIT(CL_DN_OPEN))) type = CL_DN_OPEN; else if (rc != 0) GOTO(out, rc); @@ -3055,13 +3313,13 @@ static int mdd_open(const struct lu_env *env, struct md_object *obj, find: /* look for existing opener in list under mdd_write_lock */ - mou = mdd_obj_user_find(mdd_obj, uc->uc_uid, uc->uc_gid, flags); + mou = mdd_obj_user_find(mdd_obj, uc->uc_uid, uc->uc_gid, open_flags); if (!mou) { int rc2; /* add user to list */ - mou = mdd_obj_user_alloc(flags, uc->uc_uid, uc->uc_gid); + mou = mdd_obj_user_alloc(open_flags, uc->uc_uid, uc->uc_gid); if (IS_ERR(mou)) { if (rc == 0) rc = PTR_ERR(mou); @@ -3097,7 +3355,8 @@ find: } } - mdd_changelog(env, type, flags, md_dev, mdo2fid(mdd_obj)); + /* FYI, only the bottom 32 bits of open_flags are recorded */ + mdd_changelog(env, type, open_flags, md_dev, mdd_object_fid(mdd_obj)); EXIT; out: @@ -3121,7 +3380,7 @@ static int mdd_declare_close(const struct lu_env *env, struct mdd_object *obj, * No permission check is needed. */ static int mdd_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma, int mode) + struct md_attr *ma, u64 open_flags) { struct mdd_object *mdd_obj = md2mdd_obj(obj); struct mdd_device *mdd = mdo2mdd(obj); @@ -3129,12 +3388,12 @@ static int mdd_close(const struct lu_env *env, struct md_object *obj, int is_orphan = 0; int rc; bool blocked = false; - int last_close_by_uid = 0; + bool last_close_by_uid = false; const struct lu_ucred *uc = lu_ucred(env); ENTRY; if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) { - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_write_lock(env, mdd_obj, DT_TGT_CHILD); mdd_obj->mod_count--; mdd_write_unlock(env, mdd_obj); @@ -3189,7 +3448,7 @@ again: } cont: - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_write_lock(env, mdd_obj, DT_TGT_CHILD); rc = mdd_la_get(env, mdd_obj, &ma->ma_attr); if (rc != 0) { CERROR("%s: failed to get lu_attr of "DFID": rc = %d\n", @@ -3211,8 +3470,9 @@ cont: mdd_obj->mod_count--; /*release open count */ /* under mdd write lock */ - /* If recording, see if we need to remove UID from list */ - if (mdd_changelog_enabled(env, mdd, CL_OPEN)) { + /* If recording, see if we need to remove UID from list. uc is not + * initialized if the client has been evicted. */ + if (mdd_changelog_enabled(env, mdd, CL_OPEN) && uc) { struct mdd_object_user *mou; /* look for UID in list */ @@ -3220,12 +3480,13 @@ cont: * the user had the file open. So the corresponding close * will not be logged. */ - mou = mdd_obj_user_find(mdd_obj, uc->uc_uid, uc->uc_gid, mode); + mou = mdd_obj_user_find(mdd_obj, uc->uc_uid, uc->uc_gid, + open_flags); if (mou) { mou->mou_opencount--; if (mou->mou_opencount == 0) { mdd_obj_user_remove(mdd_obj, mou); - last_close_by_uid = 1; + last_close_by_uid = true; } } } @@ -3278,9 +3539,11 @@ 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 & (1 << CL_OPEN)) && - (mode & (MDS_FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC))) || - ((mdd->mdd_cl.mc_mask & (1 << CL_OPEN)) && last_close_by_uid)) && + 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_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)); @@ -3292,14 +3555,15 @@ out: if (rc) GOTO(stop, rc); - rc = mdd_trans_start(env, mdo2mdd(obj), handle); - if (rc) - GOTO(stop, rc); - } + rc = mdd_trans_start(env, mdo2mdd(obj), handle); + if (rc) + GOTO(stop, rc); + } - mdd_changelog_data_store(env, mdd, CL_CLOSE, mode, - mdd_obj, handle); - } + /* FYI, only the bottom 32 bits of open_flags are recorded */ + mdd_changelog_data_store(env, mdd, CL_CLOSE, open_flags, + mdd_obj, handle, NULL); + } stop: if (handle != NULL && !IS_ERR(handle)) @@ -3327,89 +3591,96 @@ 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, @@ -3425,7 +3696,7 @@ int mdd_readpage(const struct lu_env *env, struct md_object *obj, return -ENOENT; } - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); + mdd_read_lock(env, mdd_obj, DT_TGT_CHILD); rc = mdd_readpage_sanity_check(env, mdd_obj); if (rc) GOTO(out_unlock, rc);