X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_xattr.c;h=aaa94abb6f926d0eb27c0be7c8412fafcff1eaf4;hp=f66f9868d8c7351989cd995d5fc24fd0ff0b8fbf;hb=a046e879fcadd601c9a19fd906f82ecbd2d4efd5;hpb=c7e86172425f203ce122ff18bc6434f843374b4b diff --git a/lustre/osd-zfs/osd_xattr.c b/lustre/osd-zfs/osd_xattr.c index f66f986..aaa94ab 100644 --- a/lustre/osd-zfs/osd_xattr.c +++ b/lustre/osd-zfs/osd_xattr.c @@ -15,21 +15,15 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ /* * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. - */ -/* - * Copyright (c) 2012, 2013, Intel Corporation. - * Use is subject to license terms. + * + * Copyright (c) 2012, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -44,7 +38,6 @@ #define DEBUG_SUBSYSTEM S_OSD -#include #include #include #include @@ -70,79 +63,56 @@ #include -/* - * Copy an extended attribute into the buffer provided, or compute the - * required buffer size. - * - * If buf is NULL, it computes the required buffer size. - * - * Returns 0 on success or a negative error number on failure. - * On success, the number of bytes used / required is stored in 'size'. - * - * No locking is done here. - */ -int __osd_xattr_load(udmu_objset_t *uos, uint64_t dnode, nvlist_t **sa_xattr) + +int __osd_xattr_load(struct osd_device *osd, sa_handle_t *hdl, nvlist_t **sa) { - sa_handle_t *sa_hdl; char *buf; int rc, size; - if (unlikely(dnode == ZFS_NO_OBJECT)) - return -ENOENT; - - rc = -sa_handle_get(uos->os, dnode, NULL, SA_HDL_PRIVATE, &sa_hdl); - if (rc) - return rc; - - rc = -sa_size(sa_hdl, SA_ZPL_DXATTR(uos), &size); + rc = -sa_size(hdl, SA_ZPL_DXATTR(osd), &size); if (rc) { if (rc == -ENOENT) - rc = -nvlist_alloc(sa_xattr, NV_UNIQUE_NAME, KM_SLEEP); + rc = -nvlist_alloc(sa, NV_UNIQUE_NAME, KM_SLEEP); goto out_sa; } - buf = sa_spill_alloc(KM_SLEEP); + buf = osd_zio_buf_alloc(size); if (buf == NULL) { rc = -ENOMEM; goto out_sa; } - rc = -sa_lookup(sa_hdl, SA_ZPL_DXATTR(uos), buf, size); + rc = -sa_lookup(hdl, SA_ZPL_DXATTR(osd), buf, size); if (rc == 0) - rc = -nvlist_unpack(buf, size, sa_xattr, KM_SLEEP); - sa_spill_free(buf); + rc = -nvlist_unpack(buf, size, sa, KM_SLEEP); + osd_zio_buf_free(buf, size); out_sa: - sa_handle_destroy(sa_hdl); return rc; } -static inline int __osd_xattr_cache(const struct lu_env *env, - struct osd_object *obj) +static inline int __osd_xattr_cache(struct osd_object *obj) { - LASSERT(obj->oo_sa_xattr == NULL); - LASSERT(obj->oo_db != NULL); - - return __osd_xattr_load(&osd_obj2dev(obj)->od_objset, - obj->oo_db->db_object, &obj->oo_sa_xattr); + LASSERT(obj->oo_sa_hdl); + if (obj->oo_sa_xattr != NULL) + return 0; + return __osd_xattr_load(osd_obj2dev(obj), + obj->oo_sa_hdl, &obj->oo_sa_xattr); } -int __osd_sa_xattr_get(const struct lu_env *env, struct osd_object *obj, - const struct lu_buf *buf, const char *name, int *sizep) +static int +__osd_sa_xattr_get(const struct lu_env *env, struct osd_object *obj, + const struct lu_buf *buf, const char *name, int *sizep) { uchar_t *nv_value; - int rc; + int rc = 0; - LASSERT(obj->oo_sa_hdl); - - if (obj->oo_sa_xattr == NULL) { - rc = __osd_xattr_cache(env, obj); - if (rc) - return rc; - } + rc = __osd_xattr_cache(obj); + if (rc) + return rc; LASSERT(obj->oo_sa_xattr); - rc = -nvlist_lookup_byte_array(obj->oo_sa_xattr, name, &nv_value, - sizep); + rc = -nvlist_lookup_byte_array(obj->oo_sa_xattr, name, + &nv_value, sizep); if (rc) return rc; @@ -158,12 +128,12 @@ int __osd_sa_xattr_get(const struct lu_env *env, struct osd_object *obj, return 0; } -int __osd_xattr_get_large(const struct lu_env *env, udmu_objset_t *uos, +int __osd_xattr_get_large(const struct lu_env *env, struct osd_device *osd, uint64_t xattr, struct lu_buf *buf, const char *name, int *sizep) { - dmu_buf_t *xa_data_db; - sa_handle_t *sa_hdl = NULL; + dnode_t *xa_data_dn; + sa_handle_t *sa_hdl = NULL; uint64_t xa_data_obj, size; int rc; @@ -172,22 +142,22 @@ int __osd_xattr_get_large(const struct lu_env *env, udmu_objset_t *uos, return -ENOENT; /* Lookup the object number containing the xattr data */ - rc = -zap_lookup(uos->os, xattr, name, sizeof(uint64_t), 1, + rc = -zap_lookup(osd->od_os, xattr, name, sizeof(uint64_t), 1, &xa_data_obj); if (rc) return rc; - rc = __osd_obj2dbuf(env, uos->os, xa_data_obj, &xa_data_db, FTAG); + rc = __osd_obj2dnode(osd->od_os, xa_data_obj, &xa_data_dn); if (rc) return rc; - rc = -sa_handle_get(uos->os, xa_data_obj, NULL, SA_HDL_PRIVATE, + rc = -sa_handle_get(osd->od_os, xa_data_obj, NULL, SA_HDL_PRIVATE, &sa_hdl); if (rc) goto out_rele; /* Get the xattr value length / object size */ - rc = -sa_lookup(sa_hdl, SA_ZPL_SIZE(uos), &size, 8); + rc = -sa_lookup(sa_hdl, SA_ZPL_SIZE(osd), &size, 8); if (rc) goto out; @@ -207,53 +177,69 @@ int __osd_xattr_get_large(const struct lu_env *env, udmu_objset_t *uos, goto out; } - rc = -dmu_read(uos->os, xa_data_db->db_object, 0, + rc = -dmu_read(osd->od_os, xa_data_dn->dn_object, 0, size, buf->lb_buf, DMU_READ_PREFETCH); out: sa_handle_destroy(sa_hdl); out_rele: - dmu_buf_rele(xa_data_db, FTAG); + osd_dnode_rele(xa_data_dn); return rc; } +/** + * Copy an extended attribute into the buffer provided, or compute + * the required buffer size if \a buf is NULL. + * + * On success, the number of bytes used or required is stored in \a sizep. + * + * Note that no locking is done here. + * + * \param[in] env execution environment + * \param[in] obj object for which to retrieve xattr + * \param[out] buf buffer to store xattr value in + * \param[in] name name of xattr to copy + * \param[out] sizep bytes used or required to store xattr + * + * \retval 0 on success + * \retval negative negated errno on failure + */ int __osd_xattr_get(const struct lu_env *env, struct osd_object *obj, - struct lu_buf *buf, const char *name, int *sizep) + struct lu_buf *buf, const char *name, int *sizep) { int rc; + if (unlikely(!dt_object_exists(&obj->oo_dt) || obj->oo_destroyed)) + return -ENOENT; + /* check SA_ZPL_DXATTR first then fallback to directory xattr */ rc = __osd_sa_xattr_get(env, obj, buf, name, sizep); if (rc != -ENOENT) return rc; - rc = __osd_xattr_get_large(env, &osd_obj2dev(obj)->od_objset, - obj->oo_xattr, buf, name, sizep); - - return rc; + return __osd_xattr_get_large(env, osd_obj2dev(obj), obj->oo_xattr, + buf, name, sizep); } int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, const char *name, - struct lustre_capa *capa) + struct lu_buf *buf, const char *name) { struct osd_object *obj = osd_dt_obj(dt); int rc, size = 0; ENTRY; - LASSERT(obj->oo_db != NULL); + LASSERT(obj->oo_dn != NULL); LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); if (!osd_obj2dev(obj)->od_posix_acl && - (strcmp(name, POSIX_ACL_XATTR_ACCESS) == 0 || - strcmp(name, POSIX_ACL_XATTR_DEFAULT) == 0)) + (strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS) == 0 || + strcmp(name, XATTR_NAME_POSIX_ACL_DEFAULT) == 0)) RETURN(-EOPNOTSUPP); - down(&obj->oo_guard); + down_read(&obj->oo_guard); rc = __osd_xattr_get(env, obj, buf, name, &size); - up(&obj->oo_guard); + up_read(&obj->oo_guard); if (rc == -ENOENT) rc = -ENODATA; @@ -262,36 +248,18 @@ int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } -void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj, - int vallen, const char *name, struct osd_thandle *oh) +/* the function is used to declare EAs when SA is not supported */ +void __osd_xattr_declare_legacy(const struct lu_env *env, + struct osd_object *obj, + int vallen, const char *name, + struct osd_thandle *oh) { struct osd_device *osd = osd_obj2dev(obj); - udmu_objset_t *uos = &osd->od_objset; - dmu_buf_t *db = obj->oo_db; - dmu_tx_t *tx = oh->ot_tx; - uint64_t xa_data_obj; - int rc = 0; - int here; - - here = dt_object_exists(&obj->oo_dt); - - /* object may be not yet created */ - if (here) { - LASSERT(db); - LASSERT(obj->oo_sa_hdl); - /* we might just update SA_ZPL_DXATTR */ - dmu_tx_hold_sa(tx, obj->oo_sa_hdl, 1); - - if (obj->oo_xattr == ZFS_NO_OBJECT) - rc = -ENOENT; - } + dmu_tx_t *tx = oh->ot_tx; + uint64_t xa_data_obj; + int rc; - if (!here || rc == -ENOENT) { - /* we'll be updating SA_ZPL_XATTR */ - if (here) { - LASSERT(obj->oo_sa_hdl); - dmu_tx_hold_sa(tx, obj->oo_sa_hdl, 1); - } + if (obj->oo_xattr == ZFS_NO_OBJECT) { /* xattr zap + entry */ dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, (char *) name); /* xattr value obj */ @@ -300,7 +268,7 @@ void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj, return; } - rc = -zap_lookup(uos->os, obj->oo_xattr, name, sizeof(uint64_t), 1, + rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1, &xa_data_obj); if (rc == 0) { /* @@ -310,7 +278,6 @@ void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj, dmu_tx_hold_bonus(tx, xa_data_obj); dmu_tx_hold_free(tx, xa_data_obj, vallen, DMU_OBJECT_END); dmu_tx_hold_write(tx, xa_data_obj, 0, vallen); - return; } else if (rc == -ENOENT) { /* * Entry doesn't exist, we need to create a new one and a new @@ -320,16 +287,50 @@ void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj, dmu_tx_hold_zap(tx, obj->oo_xattr, TRUE, (char *) name); dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE); dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, vallen); + } +} + +void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj, + int vallen, const char *name, + struct osd_thandle *oh) +{ + dmu_tx_t *tx = oh->ot_tx; + int bonuslen; + + if (unlikely(obj->oo_destroyed)) + return; + + if (unlikely(!osd_obj2dev(obj)->od_xattr_in_sa)) { + __osd_xattr_declare_legacy(env, obj, vallen, name, oh); return; } - /* An error happened */ - tx->tx_err = -rc; + /* declare EA in SA */ + if (dt_object_exists(&obj->oo_dt)) { + LASSERT(obj->oo_sa_hdl); + /* XXX: it should be possible to skip spill + * declaration if specific EA is part of + * bonus and doesn't grow */ + dmu_tx_hold_spill(tx, obj->oo_dn->dn_object); + return; + } + + bonuslen = osd_obj_bonuslen(obj); + + /* the object doesn't exist, but we've declared bonus + * in osd_declare_object_create() yet */ + if (obj->oo_ea_in_bonus > bonuslen) { + /* spill has been declared already */ + } else if (obj->oo_ea_in_bonus + vallen > bonuslen) { + /* we're about to exceed bonus, let's declare spill */ + dmu_tx_hold_spill(tx, DMU_NEW_OBJECT); + } + obj->oo_ea_in_bonus += vallen; } int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_buf *buf, const char *name, - int fl, struct thandle *handle) + const struct lu_buf *buf, const char *name, + int fl, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thandle *oh; @@ -338,13 +339,134 @@ int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); - down(&obj->oo_guard); + down_read(&obj->oo_guard); __osd_xattr_declare_set(env, obj, buf->lb_len, name, oh); - up(&obj->oo_guard); + up_read(&obj->oo_guard); RETURN(0); } +int __osd_sa_attr_init(const struct lu_env *env, struct osd_object *obj, + struct osd_thandle *oh) +{ + sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk; + struct osa_attr *osa = &osd_oti_get(env)->oti_osa; + struct lu_buf *lb = &osd_oti_get(env)->oti_xattr_lbuf; + struct osd_device *osd = osd_obj2dev(obj); + uint64_t crtime[2], gen; + timestruc_t now; + size_t size; + int rc, cnt; + + obj->oo_late_xattr = 0; + obj->oo_late_attr_set = 0; + + gen = dmu_tx_get_txg(oh->ot_tx); + gethrestime(&now); + ZFS_TIME_ENCODE(&now, crtime); + + osa->atime[0] = obj->oo_attr.la_atime; + osa->ctime[0] = obj->oo_attr.la_ctime; + osa->mtime[0] = obj->oo_attr.la_mtime; + osa->mode = obj->oo_attr.la_mode; + osa->uid = obj->oo_attr.la_uid; + osa->gid = obj->oo_attr.la_gid; + osa->rdev = obj->oo_attr.la_rdev; + osa->nlink = obj->oo_attr.la_nlink; + osa->flags = attrs_fs2zfs(obj->oo_attr.la_flags); + osa->size = obj->oo_attr.la_size; +#ifdef ZFS_PROJINHERIT + if (osd->od_projectused_dn) { + if (obj->oo_attr.la_valid & LA_PROJID) + osa->projid = obj->oo_attr.la_projid; + else + osa->projid = ZFS_DEFAULT_PROJID; + osa->flags |= ZFS_PROJID; + obj->oo_with_projid = 1; + } +#endif + + cnt = 0; + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, + &obj->oo_parent, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8); +#ifdef ZFS_PROJINHERIT + if (osd->od_projectused_dn) + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL, + &osa->projid, 8); +#endif + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8); + LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk)); + + /* Update the SA for additions, modifications, and removals. */ + rc = -nvlist_size(obj->oo_sa_xattr, &size, NV_ENCODE_XDR); + if (rc) + return rc; + + lu_buf_check_and_alloc(lb, size); + if (lb->lb_buf == NULL) { + CERROR("%s: can't allocate buffer for xattr update\n", + osd->od_svname); + return -ENOMEM; + } + + rc = -nvlist_pack(obj->oo_sa_xattr, (char **)&lb->lb_buf, &size, + NV_ENCODE_XDR, KM_SLEEP); + if (rc) + return rc; + + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd), NULL, lb->lb_buf, size); + + rc = -sa_replace_all_by_template(obj->oo_sa_hdl, bulk, cnt, oh->ot_tx); + + return rc; +} + +int __osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj, + struct osd_thandle *oh) +{ + struct lu_buf *lb = &osd_oti_get(env)->oti_xattr_lbuf; + struct osd_device *osd = osd_obj2dev(obj); + char *dxattr; + size_t size; + int rc; + + obj->oo_late_xattr = 0; + + /* Update the SA for additions, modifications, and removals. */ + rc = -nvlist_size(obj->oo_sa_xattr, &size, NV_ENCODE_XDR); + if (rc) + return rc; + + lu_buf_check_and_alloc(lb, size); + if (lb->lb_buf == NULL) { + CERROR("%s: can't allocate buffer for xattr update\n", + osd->od_svname); + return -ENOMEM; + } + + dxattr = lb->lb_buf; + rc = -nvlist_pack(obj->oo_sa_xattr, &dxattr, &size, + NV_ENCODE_XDR, KM_SLEEP); + if (rc) + return rc; + LASSERT(dxattr == lb->lb_buf); + + sa_update(obj->oo_sa_hdl, SA_ZPL_DXATTR(osd), dxattr, size, oh->ot_tx); + + return 0; +} + /* * Set an extended attribute. * This transaction must have called udmu_xattr_declare_set() first. @@ -353,43 +475,25 @@ int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, * * No locking is done here. */ -static int -__osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj, - struct osd_thandle *oh) +int __osd_sa_xattr_schedule_update(const struct lu_env *env, + struct osd_object *obj, + struct osd_thandle *oh) { - struct osd_device *osd = osd_obj2dev(obj); - udmu_objset_t *uos = &osd->od_objset; - char *dxattr; - size_t sa_size; - int rc; - ENTRY; LASSERT(obj->oo_sa_hdl); LASSERT(obj->oo_sa_xattr); - /* Update the SA for additions, modifications, and removals. */ - rc = -nvlist_size(obj->oo_sa_xattr, &sa_size, NV_ENCODE_XDR); - if (rc) - return rc; + /* schedule batched SA update in osd_object_sa_dirty_rele() */ + obj->oo_late_xattr = 1; + osd_object_sa_dirty_add(obj, oh); - dxattr = sa_spill_alloc(KM_SLEEP); - if (dxattr == NULL) - RETURN(-ENOMEM); - - rc = -nvlist_pack(obj->oo_sa_xattr, &dxattr, &sa_size, - NV_ENCODE_XDR, KM_SLEEP); - if (rc) - GOTO(out_free, rc); + RETURN(0); - rc = osd_object_sa_update(obj, SA_ZPL_DXATTR(uos), dxattr, sa_size, oh); -out_free: - sa_spill_free(dxattr); - RETURN(rc); } int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj, - const struct lu_buf *buf, const char *name, int fl, - struct osd_thandle *oh) + const struct lu_buf *buf, const char *name, int fl, + struct osd_thandle *oh) { uchar_t *nv_value; size_t size; @@ -397,12 +501,9 @@ int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj, int rc; int too_big = 0; - LASSERT(obj->oo_sa_hdl); - if (obj->oo_sa_xattr == NULL) { - rc = __osd_xattr_cache(env, obj); - if (rc) - return rc; - } + rc = __osd_xattr_cache(obj); + if (rc) + return rc; LASSERT(obj->oo_sa_xattr); /* Limited to 32k to keep nvpair memory allocations small */ @@ -431,7 +532,7 @@ int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj, DATA_TYPE_BYTE_ARRAY); if (rc < 0) return rc; - rc = __osd_sa_xattr_update(env, obj, oh); + rc = __osd_sa_xattr_schedule_update(env, obj, oh); return rc == 0 ? -EFBIG : rc; } } else if (rc == -ENOENT) { @@ -445,14 +546,14 @@ int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj, /* Ensure xattr doesn't exist in ZAP */ if (obj->oo_xattr != ZFS_NO_OBJECT) { - udmu_objset_t *uos = &osd_obj2dev(obj)->od_objset; - uint64_t xa_data_obj; - rc = -zap_lookup(uos->os, obj->oo_xattr, - name, 8, 1, &xa_data_obj); + struct osd_device *osd = osd_obj2dev(obj); + uint64_t objid; + rc = -zap_lookup(osd->od_os, obj->oo_xattr, + name, 8, 1, &objid); if (rc == 0) { - rc = __osd_object_free(uos, xa_data_obj, oh->ot_tx); + rc = -dmu_object_free(osd->od_os, objid, oh->ot_tx); if (rc == 0) - zap_remove(uos->os, obj->oo_xattr, + zap_remove(osd->od_os, obj->oo_xattr, name, oh->ot_tx); } } @@ -462,7 +563,13 @@ int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj, if (rc) return rc; - rc = __osd_sa_xattr_update(env, obj, oh); + /* batch updates only for just created dnodes where we + * used to set number of EAs in a single transaction */ + if (obj->oo_dn->dn_allocated_txg == oh->ot_tx->tx_txg) + rc = __osd_sa_xattr_schedule_update(env, obj, oh); + else + rc = __osd_sa_xattr_update(env, obj, oh); + return rc; } @@ -472,9 +579,8 @@ __osd_xattr_set(const struct lu_env *env, struct osd_object *obj, struct osd_thandle *oh) { struct osd_device *osd = osd_obj2dev(obj); - udmu_objset_t *uos = &osd->od_objset; - dmu_buf_t *xa_zap_db = NULL; - dmu_buf_t *xa_data_db = NULL; + dnode_t *xa_zap_dn = NULL; + dnode_t *xa_data_dn = NULL; uint64_t xa_data_obj; sa_handle_t *sa_hdl = NULL; dmu_tx_t *tx = oh->ot_tx; @@ -488,20 +594,19 @@ __osd_xattr_set(const struct lu_env *env, struct osd_object *obj, la->la_valid = LA_MODE; la->la_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; - rc = __osd_zap_create(env, uos, &xa_zap_db, tx, la, - obj->oo_db->db_object, FTAG, 0); + rc = __osd_zap_create(env, osd, &xa_zap_dn, tx, la, 0, 0); if (rc) return rc; - obj->oo_xattr = xa_zap_db->db_object; - rc = osd_object_sa_update(obj, SA_ZPL_XATTR(uos), + obj->oo_xattr = xa_zap_dn->dn_object; + rc = osd_object_sa_update(obj, SA_ZPL_XATTR(osd), &obj->oo_xattr, 8, oh); if (rc) goto out; } - rc = -zap_lookup(uos->os, obj->oo_xattr, name, sizeof(uint64_t), 1, - &xa_data_obj); + rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1, + &xa_data_obj); if (rc == 0) { if (fl & LU_XATTR_CREATE) { rc = -EEXIST; @@ -511,22 +616,21 @@ __osd_xattr_set(const struct lu_env *env, struct osd_object *obj, * Entry already exists. * We'll truncate the existing object. */ - rc = __osd_obj2dbuf(env, uos->os, xa_data_obj, - &xa_data_db, FTAG); + rc = __osd_obj2dnode(osd->od_os, xa_data_obj, &xa_data_dn); if (rc) goto out; - rc = -sa_handle_get(uos->os, xa_data_obj, NULL, + rc = -sa_handle_get(osd->od_os, xa_data_obj, NULL, SA_HDL_PRIVATE, &sa_hdl); if (rc) goto out; - rc = -sa_lookup(sa_hdl, SA_ZPL_SIZE(uos), &size, 8); + rc = -sa_lookup(sa_hdl, SA_ZPL_SIZE(osd), &size, 8); if (rc) goto out_sa; - rc = -dmu_free_range(uos->os, xa_data_db->db_object, - 0, DMU_OBJECT_END, tx); + rc = -dmu_free_range(osd->od_os, xa_data_dn->dn_object, + 0, DMU_OBJECT_END, tx); if (rc) goto out_sa; } else if (rc == -ENOENT) { @@ -544,18 +648,17 @@ __osd_xattr_set(const struct lu_env *env, struct osd_object *obj, la->la_valid = LA_MODE; la->la_mode = S_IFREG | S_IRUGO | S_IWUSR; - rc = __osd_object_create(env, uos, &xa_data_db, tx, la, - obj->oo_xattr, FTAG); + rc = __osd_object_create(env, obj, &xa_data_dn, tx, la); if (rc) goto out; - xa_data_obj = xa_data_db->db_object; + xa_data_obj = xa_data_dn->dn_object; - rc = -sa_handle_get(uos->os, xa_data_obj, NULL, + rc = -sa_handle_get(osd->od_os, xa_data_obj, NULL, SA_HDL_PRIVATE, &sa_hdl); if (rc) goto out; - rc = -zap_add(uos->os, obj->oo_xattr, name, sizeof(uint64_t), + rc = -zap_add(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1, &xa_data_obj, tx); if (rc) goto out_sa; @@ -565,25 +668,25 @@ __osd_xattr_set(const struct lu_env *env, struct osd_object *obj, } /* Finally write the xattr value */ - dmu_write(uos->os, xa_data_obj, 0, buf->lb_len, buf->lb_buf, tx); + dmu_write(osd->od_os, xa_data_obj, 0, buf->lb_len, buf->lb_buf, tx); size = buf->lb_len; - rc = -sa_update(sa_hdl, SA_ZPL_SIZE(uos), &size, 8, tx); + rc = -sa_update(sa_hdl, SA_ZPL_SIZE(osd), &size, 8, tx); out_sa: sa_handle_destroy(sa_hdl); out: - if (xa_data_db != NULL) - dmu_buf_rele(xa_data_db, FTAG); - if (xa_zap_db != NULL) - dmu_buf_rele(xa_zap_db, FTAG); + if (xa_data_dn != NULL) + osd_dnode_rele(xa_data_dn); + if (xa_zap_dn != NULL) + osd_dnode_rele(xa_zap_dn); return rc; } int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int fl, - struct thandle *handle, struct lustre_capa *capa) + struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thandle *oh; @@ -592,21 +695,19 @@ int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, LASSERT(handle != NULL); LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); - LASSERT(obj->oo_db); if (!osd_obj2dev(obj)->od_posix_acl && - (strcmp(name, POSIX_ACL_XATTR_ACCESS) == 0 || - strcmp(name, POSIX_ACL_XATTR_DEFAULT) == 0)) + (strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS) == 0 || + strcmp(name, XATTR_NAME_POSIX_ACL_DEFAULT) == 0)) RETURN(-EOPNOTSUPP); oh = container_of0(handle, struct osd_thandle, ot_super); - down(&obj->oo_guard); + down_write(&obj->oo_guard); CDEBUG(D_INODE, "Setting xattr %s with size %d\n", name, (int)buf->lb_len); - rc = osd_xattr_set_internal(env, obj, buf, name, fl, oh, capa); - up(&obj->oo_guard); + rc = osd_xattr_set_internal(env, obj, buf, name, fl, oh); + up_write(&obj->oo_guard); RETURN(rc); } @@ -616,7 +717,6 @@ __osd_xattr_declare_del(const struct lu_env *env, struct osd_object *obj, const char *name, struct osd_thandle *oh) { struct osd_device *osd = osd_obj2dev(obj); - udmu_objset_t *uos = &osd->od_objset; dmu_tx_t *tx = oh->ot_tx; uint64_t xa_data_obj; int rc; @@ -627,7 +727,7 @@ __osd_xattr_declare_del(const struct lu_env *env, struct osd_object *obj, if (obj->oo_xattr == ZFS_NO_OBJECT) return; - rc = -zap_lookup(uos->os, obj->oo_xattr, name, 8, 1, &xa_data_obj); + rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, 8, 1, &xa_data_obj); if (rc == 0) { /* * Entry exists. @@ -649,52 +749,52 @@ __osd_xattr_declare_del(const struct lu_env *env, struct osd_object *obj, } int osd_declare_xattr_del(const struct lu_env *env, struct dt_object *dt, - const char *name, struct thandle *handle) + const char *name, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thandle *oh; ENTRY; LASSERT(handle != NULL); - LASSERT(dt_object_exists(dt)); LASSERT(osd_invariant(obj)); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_tx != NULL); - LASSERT(obj->oo_db != NULL); + LASSERT(obj->oo_dn != NULL); - down(&obj->oo_guard); - __osd_xattr_declare_del(env, obj, name, oh); - up(&obj->oo_guard); + down_read(&obj->oo_guard); + if (likely(dt_object_exists(&obj->oo_dt) && !obj->oo_destroyed)) + __osd_xattr_declare_del(env, obj, name, oh); + up_read(&obj->oo_guard); RETURN(0); } -int __osd_sa_xattr_del(const struct lu_env *env, struct osd_object *obj, - const char *name, struct osd_thandle *oh) +static int __osd_sa_xattr_del(const struct lu_env *env, struct osd_object *obj, + const char *name, struct osd_thandle *oh) { int rc; - if (obj->oo_sa_xattr == NULL) { - rc = __osd_xattr_cache(env, obj); - if (rc) - return rc; - } + rc = __osd_xattr_cache(obj); + if (rc) + return rc; rc = -nvlist_remove(obj->oo_sa_xattr, name, DATA_TYPE_BYTE_ARRAY); if (rc == 0) - rc = __osd_sa_xattr_update(env, obj, oh); + rc = __osd_sa_xattr_schedule_update(env, obj, oh); return rc; } -int __osd_xattr_del(const struct lu_env *env, struct osd_object *obj, - const char *name, struct osd_thandle *oh) +static int __osd_xattr_del(const struct lu_env *env, struct osd_object *obj, + const char *name, struct osd_thandle *oh) { struct osd_device *osd = osd_obj2dev(obj); - udmu_objset_t *uos = &osd->od_objset; uint64_t xa_data_obj; int rc; + if (unlikely(!dt_object_exists(&obj->oo_dt) || obj->oo_destroyed)) + return -ENOENT; + /* try remove xattr from SA at first */ rc = __osd_sa_xattr_del(env, obj, name, oh); if (rc != -ENOENT) @@ -703,7 +803,7 @@ int __osd_xattr_del(const struct lu_env *env, struct osd_object *obj, if (obj->oo_xattr == ZFS_NO_OBJECT) return 0; - rc = -zap_lookup(uos->os, obj->oo_xattr, name, sizeof(uint64_t), 1, + rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1, &xa_data_obj); if (rc == -ENOENT) { rc = 0; @@ -712,19 +812,18 @@ int __osd_xattr_del(const struct lu_env *env, struct osd_object *obj, * Entry exists. * We'll delete the existing object and ZAP entry. */ - rc = __osd_object_free(uos, xa_data_obj, oh->ot_tx); + rc = -dmu_object_free(osd->od_os, xa_data_obj, oh->ot_tx); if (rc) return rc; - rc = -zap_remove(uos->os, obj->oo_xattr, name, oh->ot_tx); + rc = -zap_remove(osd->od_os, obj->oo_xattr, name, oh->ot_tx); } return rc; } int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, - const char *name, struct thandle *handle, - struct lustre_capa *capa) + const char *name, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thandle *oh; @@ -732,58 +831,140 @@ int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, ENTRY; LASSERT(handle != NULL); - LASSERT(obj->oo_db != NULL); + LASSERT(obj->oo_dn != NULL); LASSERT(osd_invariant(obj)); LASSERT(dt_object_exists(dt)); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_tx != NULL); if (!osd_obj2dev(obj)->od_posix_acl && - (strcmp(name, POSIX_ACL_XATTR_ACCESS) == 0 || - strcmp(name, POSIX_ACL_XATTR_DEFAULT) == 0)) + (strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS) == 0 || + strcmp(name, XATTR_NAME_POSIX_ACL_DEFAULT) == 0)) RETURN(-EOPNOTSUPP); - down(&obj->oo_guard); + down_write(&obj->oo_guard); rc = __osd_xattr_del(env, obj, name, oh); - up(&obj->oo_guard); + up_write(&obj->oo_guard); RETURN(rc); } +void osd_declare_xattrs_destroy(const struct lu_env *env, + struct osd_object *obj, struct osd_thandle *oh) +{ + struct osd_device *osd = osd_obj2dev(obj); + zap_attribute_t *za = &osd_oti_get(env)->oti_za; + uint64_t oid = obj->oo_xattr, xid; + dmu_tx_t *tx = oh->ot_tx; + zap_cursor_t *zc; + int rc; + + if (oid == ZFS_NO_OBJECT) + return; /* Nothing to do for SA xattrs */ + + /* Declare to free the ZAP holding xattrs */ + dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END); + + rc = osd_zap_cursor_init(&zc, osd->od_os, oid, 0); + if (rc) + goto out; + + while (zap_cursor_retrieve(zc, za) == 0) { + LASSERT(za->za_num_integers == 1); + LASSERT(za->za_integer_length == sizeof(uint64_t)); + + rc = -zap_lookup(osd->od_os, oid, za->za_name, + sizeof(uint64_t), 1, &xid); + if (rc) { + CERROR("%s: xattr %s lookup failed: rc = %d\n", + osd->od_svname, za->za_name, rc); + break; + } + dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END); + + zap_cursor_advance(zc); + } + + osd_zap_cursor_fini(zc); +out: + if (rc && tx->tx_err == 0) + tx->tx_err = -rc; +} + +int osd_xattrs_destroy(const struct lu_env *env, + struct osd_object *obj, struct osd_thandle *oh) +{ + struct osd_device *osd = osd_obj2dev(obj); + dmu_tx_t *tx = oh->ot_tx; + zap_attribute_t *za = &osd_oti_get(env)->oti_za; + zap_cursor_t *zc; + uint64_t xid; + int rc; + + /* The transaction must have been assigned to a transaction group. */ + LASSERT(tx->tx_txg != 0); + + if (obj->oo_xattr == ZFS_NO_OBJECT) + return 0; /* Nothing to do for SA xattrs */ + + /* Free the ZAP holding the xattrs */ + rc = osd_zap_cursor_init(&zc, osd->od_os, obj->oo_xattr, 0); + if (rc) + return rc; + + while (zap_cursor_retrieve(zc, za) == 0) { + LASSERT(za->za_num_integers == 1); + LASSERT(za->za_integer_length == sizeof(uint64_t)); + + rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name, + sizeof(uint64_t), 1, &xid); + if (rc) { + CERROR("%s: lookup xattr %s failed: rc = %d\n", + osd->od_svname, za->za_name, rc); + } else { + rc = -dmu_object_free(osd->od_os, xid, tx); + if (rc) + CERROR("%s: free xattr %s failed: rc = %d\n", + osd->od_svname, za->za_name, rc); + } + zap_cursor_advance(zc); + } + osd_zap_cursor_fini(zc); + + rc = -dmu_object_free(osd->od_os, obj->oo_xattr, tx); + if (rc) + CERROR("%s: free xattr %llu failed: rc = %d\n", + osd->od_svname, obj->oo_xattr, rc); + + return rc; +} + static int osd_sa_xattr_list(const struct lu_env *env, struct osd_object *obj, - struct lu_buf *lb) + const struct lu_buf *lb) { nvpair_t *nvp = NULL; - int len, counted = 0, remain = lb->lb_len; + int len, counted = 0; int rc = 0; - if (obj->oo_sa_xattr == NULL) { - rc = __osd_xattr_cache(env, obj); - if (rc) - return rc; - } - - LASSERT(obj->oo_sa_xattr); + rc = __osd_xattr_cache(obj); + if (rc) + return rc; while ((nvp = nvlist_next_nvpair(obj->oo_sa_xattr, nvp)) != NULL) { const char *name = nvpair_name(nvp); if (!osd_obj2dev(obj)->od_posix_acl && - (strcmp(name, POSIX_ACL_XATTR_ACCESS) == 0 || - strcmp(name, POSIX_ACL_XATTR_DEFAULT) == 0)) + (strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS) == 0 || + strcmp(name, XATTR_NAME_POSIX_ACL_DEFAULT) == 0)) continue; - len = strlen(nvpair_name(nvp)); + len = strlen(name); if (lb->lb_buf != NULL) { - if (len + 1 > remain) + if (counted + len + 1 > lb->lb_len) return -ERANGE; - memcpy(lb->lb_buf, name, len); - lb->lb_buf += len; - *((char *)lb->lb_buf) = '\0'; - lb->lb_buf++; - remain -= len + 1; + memcpy(lb->lb_buf + counted, name, len + 1); } counted += len + 1; } @@ -791,55 +972,49 @@ osd_sa_xattr_list(const struct lu_env *env, struct osd_object *obj, } int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *lb, struct lustre_capa *capa) + const struct lu_buf *lb) { - struct osd_thread_info *oti = osd_oti_get(env); struct osd_object *obj = osd_dt_obj(dt); struct osd_device *osd = osd_obj2dev(obj); - udmu_objset_t *uos = &osd->od_objset; + zap_attribute_t *za = &osd_oti_get(env)->oti_za; zap_cursor_t *zc; - int rc, counted = 0, remain = lb->lb_len; + int rc, counted; ENTRY; - LASSERT(obj->oo_db != NULL); + LASSERT(obj->oo_dn != NULL); LASSERT(osd_invariant(obj)); LASSERT(dt_object_exists(dt)); - down(&obj->oo_guard); + down_read(&obj->oo_guard); rc = osd_sa_xattr_list(env, obj, lb); if (rc < 0) GOTO(out, rc); + counted = rc; - remain -= counted; /* continue with dnode xattr if any */ if (obj->oo_xattr == ZFS_NO_OBJECT) GOTO(out, rc = counted); - rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0); + rc = osd_zap_cursor_init(&zc, osd->od_os, obj->oo_xattr, 0); if (rc) GOTO(out, rc); - while ((rc = -udmu_zap_cursor_retrieve_key(env, zc, oti->oti_key, - MAXNAMELEN)) == 0) { + while ((rc = -zap_cursor_retrieve(zc, za)) == 0) { if (!osd_obj2dev(obj)->od_posix_acl && - (strcmp(oti->oti_key, POSIX_ACL_XATTR_ACCESS) == 0 || - strcmp(oti->oti_key, POSIX_ACL_XATTR_DEFAULT) == 0)) { + (strcmp(za->za_name, XATTR_NAME_POSIX_ACL_ACCESS) == 0 || + strcmp(za->za_name, XATTR_NAME_POSIX_ACL_DEFAULT) == 0)) { zap_cursor_advance(zc); continue; } - rc = strlen(oti->oti_key); + rc = strlen(za->za_name); if (lb->lb_buf != NULL) { - if (rc + 1 > remain) - RETURN(-ERANGE); - - memcpy(lb->lb_buf, oti->oti_key, rc); - lb->lb_buf += rc; - *((char *)lb->lb_buf) = '\0'; - lb->lb_buf++; - remain -= rc + 1; + if (counted + rc + 1 > lb->lb_len) + GOTO(out_fini, rc = -ERANGE); + + memcpy(lb->lb_buf + counted, za->za_name, rc + 1); } counted += rc + 1; @@ -852,11 +1027,9 @@ int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, rc = counted; out_fini: - udmu_zap_cursor_fini(zc); + osd_zap_cursor_fini(zc); out: - up(&obj->oo_guard); + up_read(&obj->oo_guard); RETURN(rc); } - -