X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=07fe0790c001f76d9452b71a66b94afb577e4c80;hb=2e92c57d71384d50f06baf5b6591d2809c8288b2;hp=67b95a618d9ee1ac4b0e69684faadd75e0c0c969;hpb=af46754e1e9c3e66928e13b2301890cfa94ac059;p=fs%2Flustre-release.git diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 67b95a6..07fe079 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -15,11 +15,7 @@ * * 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 */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2014, Intel Corporation. + * Copyright (c) 2011, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -53,8 +49,8 @@ static const char dot[] = "."; static const char dotdot[] = ".."; static struct lu_name lname_dotdot = { - (char *) dotdot, - sizeof(dotdot) - 1 + .ln_name = (char *) dotdot, + .ln_namelen = sizeof(dotdot) - 1, }; static inline int @@ -85,11 +81,12 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, if (unlikely(mdd_is_dead_obj(mdd_obj))) RETURN(-ESTALE); + if (!mdd_object_exists(mdd_obj)) + RETURN(-ESTALE); + if (mdd_object_remote(mdd_obj)) { CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n", mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj))); - } else if (!mdd_object_exists(mdd_obj)) { - RETURN(-ESTALE); } rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask, @@ -123,6 +120,77 @@ int mdd_lookup(const struct lu_env *env, RETURN(rc); } +/** Read the link EA into a temp buffer. + * Uses the mdd_thread_info::mti_big_buf since it is generally large. + * A pointer to the buffer is stored in \a ldata::ld_buf. + * + * \retval 0 or error + */ +static int __mdd_links_read(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata) +{ + int rc; + + if (!mdd_object_exists(mdd_obj)) + return -ENODATA; + + /* First try a small buf */ + LASSERT(env != NULL); + ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf, + PAGE_SIZE); + if (ldata->ld_buf->lb_buf == NULL) + return -ENOMEM; + + rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK); + if (rc == -ERANGE) { + /* Buf was too small, figure out what we need. */ + lu_buf_free(ldata->ld_buf); + rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, + XATTR_NAME_LINK); + if (rc < 0) + return rc; + ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc); + if (ldata->ld_buf->lb_buf == NULL) + return -ENOMEM; + rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, + XATTR_NAME_LINK); + } + if (rc < 0) { + lu_buf_free(ldata->ld_buf); + ldata->ld_buf = NULL; + return rc; + } + + return linkea_init(ldata); +} + +static int mdd_links_read(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata) +{ + int rc; + + rc = __mdd_links_read(env, mdd_obj, ldata); + if (!rc) + rc = linkea_init(ldata); + + return rc; +} + +static int mdd_links_read_with_rec(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata) +{ + int rc; + + rc = __mdd_links_read(env, mdd_obj, ldata); + if (!rc) + rc = linkea_init_with_rec(ldata); + + return rc; +} + /** * Get parent FID of the directory * @@ -157,7 +225,7 @@ static inline int mdd_parent_fid(const struct lu_env *env, GOTO(lookup, rc = 0); ldata.ld_buf = buf; - rc = mdd_links_read(env, obj, &ldata); + rc = mdd_links_read_with_rec(env, obj, &ldata); if (rc != 0) GOTO(lookup, rc); @@ -231,6 +299,10 @@ static int mdd_is_parent(const struct lu_env *env, parent = mdd_object_find(env, mdd, pfid); if (IS_ERR(parent)) GOTO(out, rc = PTR_ERR(parent)); + + if (!mdd_object_exists(parent)) + GOTO(out, rc = -EINVAL); + p1 = parent; } EXIT; @@ -375,8 +447,6 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, const struct lu_attr *pattr, struct mdd_object *cobj, bool check_perm) { - struct mdd_thread_info *info = mdd_env_info(env); - struct lu_buf *xbuf; int rc = 0; ENTRY; @@ -386,18 +456,6 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, if (mdd_is_dead_obj(pobj)) RETURN(-ENOENT); - /* If the parent is a sub-stripe, check whether it is dead */ - xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key)); - rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV); - if (unlikely(rc > 0)) { - struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf; - - if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE && - le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD) - RETURN(-ESTALE); - } - rc = 0; - if (check_perm) rc = mdd_permission_internal_locked(env, pobj, pattr, MAY_WRITE | MAY_EXEC, @@ -975,41 +1033,32 @@ static int mdd_linkea_prepare(const struct lu_env *env, struct linkea_data *ldata) { int rc = 0; - int rc2 = 0; ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) - return 0; + RETURN(0); LASSERT(oldpfid != NULL || newpfid != NULL); - if (mdd_obj->mod_flags & DEAD_OBJ) { - /* Prevent linkea to be updated which is NOT necessary. */ - ldata->ld_reclen = 0; - /* No more links, don't bother */ + if (mdd_obj->mod_flags & DEAD_OBJ) + /* Unnecessary to update linkEA for dead object. */ RETURN(0); - } if (oldpfid != NULL) { rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid); if (rc) { - if ((check == 1) || - (rc != -ENODATA && rc != -ENOENT)) + if ((check == 1) || (rc != -ENODATA && rc != -ENOENT)) RETURN(rc); + /* No changes done. */ rc = 0; } } /* If renaming, add the new record */ - if (newpfid != NULL) { - /* even if the add fails, we still delete the out-of-date - * old link */ - rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid, - first, check); - } - - rc = rc != 0 ? rc : rc2; + if (newpfid != NULL) + rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid, + first, check); RETURN(rc); } @@ -1031,37 +1080,34 @@ int mdd_links_rename(const struct lu_env *env, ldata = &mdd_env_info(env)->mti_link_data; memset(ldata, 0, sizeof(*ldata)); rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname, - newpfid, newlname, first, check, - ldata); - if (rc != 0) + newpfid, newlname, first, check, ldata); + if (rc) GOTO(out, rc); } - if (ldata->ld_reclen != 0) + if (!(mdd_obj->mod_flags & DEAD_OBJ)) rc = mdd_links_write(env, mdd_obj, ldata, handle); - EXIT; + + GOTO(out, rc); + out: if (rc != 0) { - int error = 1; - if (rc == -EOVERFLOW || rc == -ENOSPC) - error = 0; - if (oldpfid == NULL) - CDEBUG(error ? D_ERROR : D_OTHER, - "link_ea add '%.*s' failed %d "DFID"\n", - newlname->ln_namelen, newlname->ln_name, + if (newlname == NULL) + CERROR("link_ea add failed %d "DFID"\n", rc, PFID(mdd_object_fid(mdd_obj))); + else if (oldpfid == NULL) + CERROR("link_ea add '%.*s' failed %d "DFID"\n", + newlname->ln_namelen, newlname->ln_name, rc, + PFID(mdd_object_fid(mdd_obj))); else if (newpfid == NULL) - CDEBUG(error ? D_ERROR : D_OTHER, - "link_ea del '%.*s' failed %d "DFID"\n", - oldlname->ln_namelen, oldlname->ln_name, - rc, PFID(mdd_object_fid(mdd_obj))); + CERROR("link_ea del '%.*s' failed %d "DFID"\n", + oldlname->ln_namelen, oldlname->ln_name, rc, + PFID(mdd_object_fid(mdd_obj))); else - CDEBUG(error ? D_ERROR : D_OTHER, - "link_ea rename '%.*s'->'%.*s' failed %d " - DFID"\n", - oldlname->ln_namelen, oldlname->ln_name, - newlname->ln_namelen, newlname->ln_name, - rc, PFID(mdd_object_fid(mdd_obj))); + CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID + "\n", oldlname->ln_namelen, oldlname->ln_name, + newlname->ln_namelen, newlname->ln_name, rc, + PFID(mdd_object_fid(mdd_obj))); } if (is_vmalloc_addr(ldata->ld_buf)) @@ -1093,50 +1139,6 @@ static inline int mdd_links_del(const struct lu_env *env, } /** Read the link EA into a temp buffer. - * Uses the mdd_thread_info::mti_big_buf since it is generally large. - * A pointer to the buffer is stored in \a ldata::ld_buf. - * - * \retval 0 or error - */ -int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj, - struct linkea_data *ldata) -{ - int rc; - - if (!mdd_object_exists(mdd_obj)) - return -ENODATA; - - /* First try a small buf */ - LASSERT(env != NULL); - ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf, - PAGE_CACHE_SIZE); - if (ldata->ld_buf->lb_buf == NULL) - return -ENOMEM; - - rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK); - if (rc == -ERANGE) { - /* Buf was too small, figure out what we need. */ - lu_buf_free(ldata->ld_buf); - rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, - XATTR_NAME_LINK); - if (rc < 0) - return rc; - ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc); - if (ldata->ld_buf->lb_buf == NULL) - return -ENOMEM; - rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, - XATTR_NAME_LINK); - } - if (rc < 0) { - lu_buf_free(ldata->ld_buf); - ldata->ld_buf = NULL; - return rc; - } - - return linkea_init(ldata); -} - -/** Read the link EA into a temp buffer. * Uses the name_buf since it is generally large. * \retval IS_ERR err * \retval ptr to \a lu_buf (always \a mti_big_buf) @@ -1154,40 +1156,33 @@ struct lu_buf *mdd_links_get(const struct lu_env *env, int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj, struct linkea_data *ldata, struct thandle *handle) { - const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf, - ldata->ld_leh->leh_len); + const struct lu_buf *buf; int rc; + if (ldata == NULL || ldata->ld_buf == NULL || + ldata->ld_leh == NULL) + return 0; + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA)) return 0; +again: + buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf, + ldata->ld_leh->leh_len); rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle); - if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) && - mdd_object_remote(mdd_obj) == 0) { - struct lfsck_request *lr = &mdd_env_info(env)->mti_lr; - struct thandle *sub_th; - - /* XXX: If the linkEA is overflow, then we need to notify the - * namespace LFSCK to skip "nlink" attribute verification - * on this object to avoid the "nlink" to be shrinked by - * wrong. It may be not good an interaction with LFSCK - * like this. We will consider to replace it with other - * mechanism in future. LU-5802. */ - lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK, - LFSCK_TYPE_NAMESPACE); - - sub_th = thandle_get_sub_by_dt(env, handle, - mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom); - lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom, - lr, sub_th); + if (unlikely(rc == -ENOSPC)) { + rc = linkea_overflow_shrink(ldata); + if (likely(rc > 0)) + goto again; } return rc; } -int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, - struct thandle *handle, struct linkea_data *ldata, - enum mdd_links_add_overflow overflow) +static int mdd_declare_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct thandle *handle, + struct linkea_data *ldata) { int rc; int ea_len; @@ -1197,36 +1192,13 @@ int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, ea_len = ldata->ld_leh->leh_len; linkea = ldata->ld_buf->lb_buf; } else { - ea_len = DEFAULT_LINKEA_SIZE; + ea_len = MAX_LINKEA_SIZE; linkea = NULL; } - /* XXX: max size? */ rc = mdo_declare_xattr_set(env, mdd_obj, mdd_buf_get_const(env, linkea, ea_len), XATTR_NAME_LINK, 0, handle); - if (rc != 0) - return rc; - - if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) { - struct lfsck_request *lr = &mdd_env_info(env)->mti_lr; - struct thandle *sub_th; - - /* XXX: If the linkEA is overflow, then we need to notify the - * namespace LFSCK to skip "nlink" attribute verification - * on this object to avoid the "nlink" to be shrinked by - * wrong. It may be not good an interaction with LFSCK - * like this. We will consider to replace it with other - * mechanism in future. LU-5802. */ - lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE, - LFSCK_TYPE_NAMESPACE); - - sub_th = thandle_get_sub_by_dt(env, handle, - mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom); - rc = lfsck_in_notify(env, - mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom, - lr, sub_th); - } return rc; } @@ -1240,7 +1212,7 @@ static inline int mdd_declare_links_del(const struct lu_env *env, /* For directory, the linkEA will be removed together * with the object. */ if (!S_ISDIR(mdd_object_type(c))) - rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE); + rc = mdd_declare_links_add(env, c, handle, NULL); return rc; } @@ -1254,9 +1226,13 @@ static int mdd_declare_link(const struct lu_env *env, struct lu_attr *la, struct linkea_data *data) { + struct lu_fid tfid = *mdo2fid(c); int rc; - rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c), + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) + tfid.f_oid = cfs_fail_val; + + rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c), name->ln_name, handle); if (rc != 0) return rc; @@ -1265,12 +1241,6 @@ static int mdd_declare_link(const struct lu_env *env, if (rc != 0) return rc; - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) { - rc = mdo_declare_ref_add(env, c, handle); - if (rc != 0) - return rc; - } - la->la_valid = LA_CTIME | LA_MTIME; rc = mdo_declare_attr_set(env, p, la, handle); if (rc != 0) @@ -1281,8 +1251,7 @@ static int mdd_declare_link(const struct lu_env *env, if (rc != 0) return rc; - rc = mdd_declare_links_add(env, c, handle, data, - S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE); + rc = mdd_declare_links_add(env, c, handle, data); if (rc != 0) return rc; @@ -1301,8 +1270,9 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, struct mdd_object *mdd_sobj = md2mdd_obj(src_obj); struct lu_attr *cattr = MDD_ENV_VAR(env, cattr); struct lu_attr *tattr = MDD_ENV_VAR(env, tattr); - struct mdd_device *mdd = mdo2mdd(src_obj); - struct thandle *handle; + struct mdd_device *mdd = mdo2mdd(src_obj); + struct thandle *handle; + struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2; struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; int rc; ENTRY; @@ -1315,6 +1285,15 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, if (rc != 0) RETURN(rc); + /* + * If we are using project inheritance, we only allow hard link + * creation in our tree when the project IDs are the same; + * otherwise the tree quota mechanism could be circumvented. + */ + if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) && + (tattr->la_projid != cattr->la_projid)) + RETURN(-EXDEV); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); @@ -1324,6 +1303,14 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, LASSERT(ma->ma_attr.la_valid & LA_CTIME); la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + /* Note: even this function will change ldata, but it comes from + * thread_info, which is completely temporary and only seen in + * this function, so we do not need reset ldata once it fails.*/ + rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj), + lname, 0, 0, ldata); + if (rc != 0) + GOTO(stop, rc); + rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle, la, ldata); if (rc) @@ -1345,25 +1332,12 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, GOTO(out_unlock, rc); } - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) { - rc = mdo_ref_add(env, mdd_sobj, handle); - if (rc != 0) - GOTO(out_unlock, rc); - } - - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) { - struct lu_fid tfid = *mdo2fid(mdd_sobj); - - tfid.f_oid++; - rc = __mdd_index_insert_only(env, mdd_tobj, &tfid, - mdd_object_type(mdd_sobj), - name, handle); - } else { - rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), - mdd_object_type(mdd_sobj), - name, handle); - } + *tfid = *mdo2fid(mdd_sobj); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) + tfid->f_oid = cfs_fail_val; + rc = __mdd_index_insert_only(env, mdd_tobj, tfid, + mdd_object_type(mdd_sobj), name, handle); if (rc != 0) { mdo_ref_del(env, mdd_sobj, handle); GOTO(out_unlock, rc); @@ -1376,49 +1350,40 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, la->la_valid = LA_CTIME; rc = mdd_update_time(env, mdd_sobj, cattr, la, handle); - if (rc == 0) { - rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, - mdo2fid(mdd_tobj), lname, 0, 0, - ldata); - if (rc == 0) - mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), - lname, handle, ldata, 0); - /* The failure of links_add should not cause the link - * failure, reset rc here */ - rc = 0; - } - EXIT; + if (rc == 0) + /* Note: The failure of links_add should not cause the + * link failure, so do not check return value. */ + mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), + lname, handle, ldata, 0); + + EXIT; out_unlock: - mdd_write_unlock(env, mdd_sobj); - if (rc == 0) + mdd_write_unlock(env, mdd_sobj); + if (rc == 0) rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, mdo2fid(mdd_tobj), NULL, NULL, lname, NULL, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); - + rc = mdd_trans_stop(env, mdd, rc, handle); if (is_vmalloc_addr(ldata->ld_buf)) /* if we vmalloced a large buffer drop it */ lu_buf_free(ldata->ld_buf); out_pending: - return rc; + return rc; } -static int mdd_mark_dead_object(const struct lu_env *env, +static int mdd_mark_orphan_object(const struct lu_env *env, struct mdd_object *obj, struct thandle *handle, bool declare) { struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start); int rc; - if (!declare) - obj->mod_flags |= DEAD_OBJ; - if (!S_ISDIR(mdd_object_type(obj))) return 0; attr->la_valid = LA_FLAGS; - attr->la_flags = LUSTRE_SLAVE_DEAD_FL; + attr->la_flags = LUSTRE_ORPHAN_FL; if (declare) rc = mdo_declare_attr_set(env, obj, attr, handle); @@ -1434,15 +1399,18 @@ static int mdd_declare_finish_unlink(const struct lu_env *env, { int rc; - rc = mdd_mark_dead_object(env, obj, handle, true); + /* Sigh, we do not know if the unlink object will become orphan in + * declare phase, but fortunately the flags here does not matter + * in current declare implementation */ + rc = mdd_mark_orphan_object(env, obj, handle, true); if (rc != 0) return rc; - rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); + rc = mdo_declare_destroy(env, obj, handle); if (rc != 0) return rc; - rc = mdo_declare_destroy(env, obj, handle); + rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); if (rc != 0) return rc; @@ -1457,34 +1425,37 @@ int mdd_finish_unlink(const struct lu_env *env, struct thandle *th) { int rc = 0; - int is_dir = S_ISDIR(ma->ma_attr.la_mode); - ENTRY; + int is_dir = S_ISDIR(ma->ma_attr.la_mode); + ENTRY; - LASSERT(mdd_write_locked(env, obj) != 0); + LASSERT(mdd_write_locked(env, obj) != 0); if (ma->ma_attr.la_nlink == 0 || is_dir) { - rc = mdd_mark_dead_object(env, obj, th, false); - if (rc != 0) - RETURN(rc); - - /* add new orphan and the object - * will be deleted during mdd_close() */ - if (obj->mod_count) { - rc = __mdd_orphan_add(env, obj, th); - if (rc == 0) - CDEBUG(D_HA, "Object "DFID" is inserted into " - "orphan list, open count = %d\n", - PFID(mdd_object_fid(obj)), - obj->mod_count); - else - CERROR("Object "DFID" fail to be an orphan, " - "open count = %d, maybe cause failed " - "open replay\n", - PFID(mdd_object_fid(obj)), - obj->mod_count); - } else { + /* add new orphan and the object + * will be deleted during mdd_close() */ + obj->mod_flags |= DEAD_OBJ; + if (obj->mod_count) { + rc = __mdd_orphan_add(env, obj, th); + if (rc == 0) + CDEBUG(D_HA, "Object "DFID" is inserted into " + "orphan list, open count = %d\n", + PFID(mdd_object_fid(obj)), + obj->mod_count); + else + CERROR("Object "DFID" fail to be an orphan, " + "open count = %d, maybe cause failed " + "open replay\n", + PFID(mdd_object_fid(obj)), + obj->mod_count); + + /* mark object as an orphan here, not + * before __mdd_orphan_add() as racing + * mdd_la_get() may propagate ORPHAN_OBJ + * causing the asserition */ + rc = mdd_mark_orphan_object(env, obj, th, false); + } else { rc = mdo_destroy(env, obj, th); - } + } } else if (!is_dir) { /* old files may not have link ea; ignore errors */ mdd_links_del(env, obj, mdo2fid(pobj), lname, th); @@ -1590,7 +1561,7 @@ static bool mdd_hsm_archive_exists(const struct lu_env *env, if (rc < 0) RETURN(false); - ma->ma_valid = MA_HSM; + ma->ma_valid |= MA_HSM; } if (ma->ma_hsm.mh_flags & HS_EXISTS) RETURN(true); @@ -1618,7 +1589,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, struct mdd_object *mdd_cobj = NULL; struct mdd_device *mdd = mdo2mdd(pobj); struct thandle *handle; - int rc, is_dir = 0; + int rc, is_dir = 0, cl_flags = 0; ENTRY; /* cobj == NULL means only delete name entry */ @@ -1639,6 +1610,11 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, RETURN(rc); is_dir = S_ISDIR(cattr->la_mode); + /* search for an existing archive. + * we should check ahead as the object + * can be destroyed in this transaction */ + if (mdd_hsm_archive_exists(env, mdd_cobj, ma)) + cl_flags |= CLF_UNLINK_HSM_EXISTS; } rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr); @@ -1716,30 +1692,33 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, ma->ma_attr = *cattr; ma->ma_valid |= MA_INODE; rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle); + if (rc != 0) + GOTO(cleanup, rc); /* fetch updated nlink */ - if (rc == 0) - rc = mdd_la_get(env, mdd_cobj, cattr); + rc = mdd_la_get(env, mdd_cobj, cattr); + /* if object is removed then we can't get its attrs, + * use last get */ + if (rc == -ENOENT) { + cattr->la_nlink = 0; + rc = 0; + } - /* if object is removed then we can't get its attrs, use last get */ if (cattr->la_nlink == 0) { ma->ma_attr = *cattr; ma->ma_valid |= MA_INODE; } + EXIT; cleanup: if (likely(mdd_cobj != NULL)) mdd_write_unlock(env, mdd_cobj); if (rc == 0) { - int cl_flags = 0; - - if (cattr->la_nlink == 0) { + if (cattr->la_nlink == 0) cl_flags |= CLF_UNLINK_LAST; - /* search for an existing archive */ - if (mdd_hsm_archive_exists(env, mdd_cobj, ma)) - cl_flags |= CLF_UNLINK_HSM_EXISTS; - } + else + cl_flags &= ~CLF_UNLINK_HSM_EXISTS; rc = mdd_changelog_ns_store(env, mdd, is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, @@ -1748,7 +1727,7 @@ cleanup: } stop: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); return rc; } @@ -1768,9 +1747,11 @@ static int mdd_cd_sanity_check(const struct lu_env *env, RETURN(0); } -static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, - struct md_object *cobj, const struct md_op_spec *spec, - struct md_attr *ma) +static int mdd_create_data(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + const struct md_op_spec *spec, + struct md_attr *ma) { struct mdd_device *mdd = mdo2mdd(cobj); struct mdd_object *mdd_pobj = md2mdd_obj(pobj); @@ -1802,15 +1783,15 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, /* calling ->ah_make_hint() is used to transfer information from parent */ mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint); - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_free, rc = PTR_ERR(handle)); + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + GOTO(out_free, rc = PTR_ERR(handle)); - /* - * XXX: Setting the lov ea is not locked but setting the attr is locked? - * Should this be fixed? - */ - CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n", + /* + * XXX: Setting the lov ea is not locked but setting the attr is locked? + * Should this be fixed? + */ + CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n", spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen, spec->sp_cr_flags, spec->no_create); @@ -1843,7 +1824,8 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); + out_free: RETURN(rc); } @@ -1851,24 +1833,15 @@ out_free: static int mdd_declare_object_initialize(const struct lu_env *env, struct mdd_object *parent, struct mdd_object *child, - struct lu_attr *attr, + const struct lu_attr *attr, struct thandle *handle) { int rc; ENTRY; - /* - * inode mode has been set in creation time, and it's based on umask, - * la_mode and acl, don't set here again! (which will go wrong - * because below function doesn't consider umask). - * I'd suggest set all object attributes in creation time, see above. - */ LASSERT(attr->la_valid & (LA_MODE | LA_TYPE)); - attr->la_valid &= ~(LA_MODE | LA_TYPE); - rc = mdo_declare_attr_set(env, child, attr, handle); - attr->la_valid |= LA_MODE | LA_TYPE; - if (rc != 0 || !S_ISDIR(attr->la_mode)) - RETURN(rc); + if (!S_ISDIR(attr->la_mode)) + RETURN(0); rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR, dot, handle); @@ -1941,9 +1914,9 @@ static int mdd_create_sanity_check(const struct lu_env *env, int rc; ENTRY; - /* EEXIST check */ - if (mdd_is_dead_obj(obj)) - RETURN(-ENOENT); + /* EEXIST check */ + if (mdd_is_dead_obj(obj)) + RETURN(-ENOENT); /* * In some cases this lookup is not needed - we know before if name @@ -1978,19 +1951,29 @@ static int mdd_create_sanity_check(const struct lu_env *env, } } + /* Inherit project ID from parent directory */ + if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) { + cattr->la_projid = pattr->la_projid; + if (S_ISDIR(cattr->la_mode)) { + cattr->la_flags |= LUSTRE_PROJINHERIT_FL; + cattr->la_valid |= LA_FLAGS; + } + cattr->la_valid |= LA_PROJID; + } + rc = mdd_name_check(m, lname); if (rc < 0) RETURN(rc); switch (cattr->la_mode & S_IFMT) { - case S_IFLNK: { - unsigned int symlen = strlen(spec->u.sp_symname) + 1; + case S_IFLNK: { + unsigned int symlen = strlen(spec->u.sp_symname) + 1; - if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift)) - RETURN(-ENAMETOOLONG); - else - RETURN(0); - } + if (symlen > m->mdd_dt_conf.ddp_symlink_max) + RETURN(-ENAMETOOLONG); + else + RETURN(0); + } case S_IFDIR: case S_IFREG: case S_IFCHR: @@ -2016,6 +1999,7 @@ static int mdd_declare_object_create(const struct lu_env *env, struct lu_buf *acl_buf, struct dt_allocation_hint *hint) { + const struct lu_buf *buf; int rc; rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec, @@ -2051,8 +2035,6 @@ static int mdd_declare_object_create(const struct lu_env *env, /* replay case, create LOV EA from client data */ if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) { - const struct lu_buf *buf; - buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen); rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0, @@ -2072,6 +2054,16 @@ static int mdd_declare_object_create(const struct lu_env *env, if (rc) GOTO(out, rc); } + + if (spec->sp_cr_file_secctx_name != NULL) { + buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx, + spec->sp_cr_file_secctx_size); + rc = mdo_declare_xattr_set(env, c, buf, + spec->sp_cr_file_secctx_name, 0, + handle); + if (rc < 0) + GOTO(out, rc); + } out: return rc; } @@ -2112,7 +2104,7 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, if (rc != 0) return rc; - rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE); + rc = mdd_declare_links_add(env, c, handle, ldata); if (rc) return rc; @@ -2182,6 +2174,7 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, struct dt_allocation_hint *hint, struct thandle *handle) { + const struct lu_buf *buf; int rc; mdd_write_lock(env, son, MOR_TGT_CHILD); @@ -2217,8 +2210,6 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, if (spec->no_create || (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) || S_ISDIR(attr->la_mode)) { - const struct lu_buf *buf; - buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen); rc = mdo_xattr_set(env, son, buf, @@ -2268,6 +2259,15 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, GOTO(err_initlized, rc = -EFAULT); } + if (spec->sp_cr_file_secctx_name != NULL) { + buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx, + spec->sp_cr_file_secctx_size); + rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name, + 0, handle); + if (rc < 0) + GOTO(err_initlized, rc); + } + err_initlized: if (unlikely(rc != 0)) { int rc2; @@ -2289,6 +2289,48 @@ unlock: RETURN(rc); } +static int mdd_index_delete(const struct lu_env *env, + struct mdd_object *mdd_pobj, + struct lu_attr *cattr, + const struct lu_name *lname) +{ + struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj); + struct thandle *handle; + int rc; + ENTRY; + + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); + + rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, + handle); + if (rc != 0) + GOTO(stop, rc); + + if (S_ISDIR(cattr->la_mode)) { + rc = mdo_declare_ref_del(env, mdd_pobj, handle); + if (rc != 0) + GOTO(stop, rc); + } + + /* Since this will only be used in the error handler path, + * Let's set the thandle to be local and not mess the transno */ + handle->th_local = 1; + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name, + S_ISDIR(cattr->la_mode), handle); + if (rc) + GOTO(stop, rc); +stop: + rc = mdd_trans_stop(env, mdd, rc, handle); + + RETURN(rc); +} + /* * Create object and insert it into namespace. */ @@ -2427,8 +2469,6 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, EXIT; err_insert: if (rc != 0) { - int rc2; - if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) rc2 = __mdd_orphan_del(env, son, handle); else @@ -2476,8 +2516,16 @@ out_volatile: NULL, handle); out_stop: rc2 = mdd_trans_stop(env, mdd, rc, handle); - if (rc == 0) + if (rc == 0) { + /* If creation fails, it is most likely due to the remote update + * failure, because local transaction will mostly succeed at + * this stage. There is no easy way to rollback all of previous + * updates, so let's remove the object from namespace, and + * LFSCK should handle the orphan object. */ + if (rc2 < 0 && !mdd_object_remote(mdd_pobj)) + mdd_index_delete(env, mdd_pobj, attr, lname); rc = rc2; + } out_free: if (is_vmalloc_addr(ldata->ld_buf)) /* if we vmalloced a large buffer drop it */ @@ -2552,6 +2600,17 @@ static int mdd_rename_sanity_check(const struct lu_env *env, * before mdd_rename and enable MDS_PERM_BYPASS. */ LASSERT(sobj); + /* + * If we are using project inheritance, we only allow renames + * into our tree when the project IDs are the same; otherwise + * tree quota mechanism would be circumvented. + */ + if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) && + tpattr->la_projid != cattr->la_projid) || + ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) && + (pattr->la_projid != tpattr->la_projid))) + RETURN(-EXDEV); + rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0); if (rc) RETURN(rc); @@ -2580,8 +2639,8 @@ static int mdd_declare_rename(const struct lu_env *env, struct mdd_object *mdd_tpobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - const struct lu_name *tname, const struct lu_name *sname, + const struct lu_name *tname, struct md_attr *ma, struct linkea_data *ldata, struct thandle *handle) @@ -2642,8 +2701,7 @@ static int mdd_declare_rename(const struct lu_env *env, if (rc) return rc; - rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata, - S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE); + rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata); if (rc) return rc; @@ -2744,6 +2802,11 @@ static int mdd_rename(const struct lu_env *env, rc = mdd_la_get(env, mdd_tobj, tattr); if (rc) GOTO(out_pending, rc); + /* search for an existing archive. + * we should check ahead as the object + * can be destroyed in this transaction */ + if (mdd_hsm_archive_exists(env, mdd_tobj, ma)) + cl_flags |= CLF_RENAME_LAST_EXISTS; } rc = mdd_la_get(env, mdd_tpobj, tpattr); @@ -2769,8 +2832,12 @@ static int mdd_rename(const struct lu_env *env, GOTO(out_pending, rc = PTR_ERR(handle)); memset(ldata, 0, sizeof(*ldata)); - mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname, - mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata); + rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), + lsname, mdd_object_fid(mdd_tpobj), ltname, + 1, 0, ldata); + if (rc) + GOTO(stop, rc); + rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, mdd_tobj, lsname, ltname, ma, ldata, handle); if (rc) @@ -2891,7 +2958,12 @@ static int mdd_rename(const struct lu_env *env, /* fetch updated nlink */ rc = mdd_la_get(env, mdd_tobj, tattr); - if (rc != 0) { + if (rc == -ENOENT) { + /* the object got removed, let's + * return the latest known attributes */ + tattr->la_nlink = 0; + rc = 0; + } else if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", mdd2obd_dev(mdd)->obd_name, @@ -2902,11 +2974,10 @@ static int mdd_rename(const struct lu_env *env, ma->ma_attr = *tattr; ma->ma_valid |= MA_INODE; - if (tattr->la_nlink == 0) { + if (tattr->la_nlink == 0) cl_flags |= CLF_RENAME_LAST; - if (mdd_hsm_archive_exists(env, mdd_tobj, ma)) - cl_flags |= CLF_RENAME_LAST_EXISTS; - } + else + cl_flags &= ~CLF_RENAME_LAST_EXISTS; } la->la_valid = LA_CTIME | LA_MTIME; @@ -2980,7 +3051,7 @@ cleanup: ltname, lsname, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); out_pending: mdd_object_put(env, mdd_sobj); @@ -2993,6 +3064,7 @@ out_pending: **/ static int mdd_linkea_update_child_internal(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *newparent, struct mdd_object *child, const char *name, int namelen, struct thandle *handle, @@ -3029,7 +3101,7 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, &lname, &fid); if (strncmp(lname.ln_name, name, namelen) != 0 || - lu_fid_eq(&fid, mdd_object_fid(parent))) { + !lu_fid_eq(&fid, mdd_object_fid(parent))) { ldata.ld_lee = (struct link_ea_entry *) ((char *)ldata.ld_lee + ldata.ld_reclen); @@ -3039,13 +3111,12 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n", mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)), lname.ln_namelen, lname.ln_name, - PFID(mdd_object_fid(parent))); + PFID(mdd_object_fid(newparent))); /* update to the new parent fid */ linkea_entry_pack(ldata.ld_lee, &lname, - mdd_object_fid(parent)); + mdd_object_fid(newparent)); if (declare) - rc = mdd_declare_links_add(env, child, handle, &ldata, - MLAO_IGNORE); + rc = mdd_declare_links_add(env, child, handle, &ldata); else rc = mdd_links_write(env, child, &ldata, handle); break; @@ -3055,21 +3126,25 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, static int mdd_linkea_declare_update_child(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *newparent, struct mdd_object *child, const char *name, int namelen, struct thandle *handle) { - return mdd_linkea_update_child_internal(env, parent, child, name, + return mdd_linkea_update_child_internal(env, parent, newparent, + child, name, namelen, handle, true); } static int mdd_linkea_update_child(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *newparent, struct mdd_object *child, const char *name, int namelen, struct thandle *handle) { - return mdd_linkea_update_child_internal(env, parent, child, name, + return mdd_linkea_update_child_internal(env, parent, newparent, + child, name, namelen, handle, false); } @@ -3078,25 +3153,20 @@ static int mdd_update_linkea_internal(const struct lu_env *env, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, const struct lu_name *child_name, + struct linkea_data *ldata, struct thandle *handle, int declare) { struct mdd_thread_info *info = mdd_env_info(env); - struct linkea_data *ldata = &info->mti_link_data; int count; int rc = 0; ENTRY; - rc = mdd_links_read(env, mdd_sobj, ldata); - if (rc != 0) { - if (rc == -ENOENT || rc == -ENODATA) - rc = 0; - RETURN(rc); - } + LASSERT(ldata->ld_buf != NULL); + LASSERT(ldata->ld_leh != NULL); /* If it is mulitple links file, we need update the name entry for * all parent */ - LASSERT(ldata->ld_leh != NULL); ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1); for (count = 0; count < ldata->ld_leh->leh_reccount; count++) { struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); @@ -3106,8 +3176,6 @@ static int mdd_update_linkea_internal(const struct lu_env *env, linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname, &fid); - ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + - ldata->ld_reclen); pobj = mdd_object_find(env, mdd, &fid); if (IS_ERR(pobj)) { CWARN("%s: cannot find obj "DFID": rc = %ld\n", @@ -3119,7 +3187,7 @@ static int mdd_update_linkea_internal(const struct lu_env *env, if (!mdd_object_exists(pobj)) { CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n", mdd2obd_dev(mdd)->obd_name, PFID(&fid)); - GOTO(next_put, rc); + goto next_put; } if (pobj == mdd_pobj && @@ -3129,7 +3197,7 @@ static int mdd_update_linkea_internal(const struct lu_env *env, CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n", mdd2obd_dev(mdd)->obd_name, child_name->ln_name, PFID(&fid)); - GOTO(next_put, rc); + goto next_put; } CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n", @@ -3159,15 +3227,42 @@ static int mdd_update_linkea_internal(const struct lu_env *env, if (rc) GOTO(next_put, rc); } else { - rc = __mdd_index_delete(env, pobj, lname.ln_name, - 0, handle); - if (rc) + char *tmp_name = info->mti_key; + + if (lname.ln_namelen >= sizeof(info->mti_key)) { + /* lnamelen is too big(> NAME_MAX + 16), + * something wrong about this linkea, let's + * skip it */ + CWARN("%s: the name %.*s is too long under " + DFID"\n", mdd2obd_dev(mdd)->obd_name, + lname.ln_namelen, lname.ln_name, + PFID(&fid)); + goto next_put; + } + + /* Note: lname might be without \0 at the end, see + * linkea_entry_unpack(), let's add extra \0 by + * snprintf */ + snprintf(tmp_name, sizeof(info->mti_key), "%.*s", + lname.ln_namelen, lname.ln_name); + lname.ln_name = tmp_name; + + /* Let's check if this linkEA still valid, before + * it might be packed into the RPC buffer. */ + rc = mdd_lookup(env, &pobj->mod_obj, &lname, + &info->mti_fid, NULL); + if (rc < 0 || !lu_fid_eq(&info->mti_fid, + mdd_object_fid(mdd_sobj))) + GOTO(next_put, rc == -ENOENT ? 0 : rc); + + rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle); + if (rc != 0) GOTO(next_put, rc); rc = __mdd_index_insert(env, pobj, mdd_object_fid(mdd_tobj), mdd_object_type(mdd_tobj), - lname.ln_name, handle); + tmp_name, handle); if (rc != 0) GOTO(next_put, rc); @@ -3185,6 +3280,9 @@ next_put: mdd_object_put(env, pobj); if (rc != 0) break; + + ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + + ldata->ld_reclen); } RETURN(rc); @@ -3228,8 +3326,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, xname = list_xbuf.lb_buf; while (rem > 0) { xlen = strnlen(xname, rem - 1) + 1; - if (strcmp(XATTR_NAME_LINK, xname) == 0 || - strcmp(XATTR_NAME_LMA, xname) == 0 || + if (strcmp(XATTR_NAME_LMA, xname) == 0 || strcmp(XATTR_NAME_LMV, xname) == 0) goto next; @@ -3272,14 +3369,25 @@ static int mdd_migrate_xattrs(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); +again: rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle); if (rc == -EEXIST) GOTO(stop_trans, rc = 0); + if (unlikely(rc == -ENOSPC && + strcmp(xname, XATTR_NAME_LINK) == 0)) { + rc = linkea_overflow_shrink( + (struct linkea_data *)(xbuf.lb_buf)); + if (likely(rc > 0)) { + xbuf.lb_len = rc; + goto again; + } + } + if (rc != 0) GOTO(stop_trans, rc); stop_trans: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); if (rc != 0) GOTO(out, rc); next: @@ -3325,9 +3433,8 @@ static int mdd_declare_migrate_create(const struct lu_env *env, buf, 0, handle); if (rc != 0) return rc; - } else if (S_ISDIR(la->la_mode)) { - rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata, - MLAO_IGNORE); + } else if (S_ISDIR(la->la_mode) && ldata != NULL) { + rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata); if (rc != 0) return rc; } @@ -3397,7 +3504,7 @@ static int mdd_migrate_create(const struct lu_env *env, RETURN(rc); } spec->u.sp_symname = link_buf.lb_buf; - } else if S_ISREG(la->la_mode) { + } else if (S_ISREG(la->la_mode)) { /* retrieve lov of the old object */ rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf); if (rc != 0 && rc != -ENODATA) @@ -3408,8 +3515,13 @@ static int mdd_migrate_create(const struct lu_env *env, spec->sp_cr_flags |= MDS_OPEN_HAS_EA; } } else if (S_ISDIR(la->la_mode)) { - rc = mdd_links_read(env, mdd_sobj, ldata); - if (rc < 0 && rc != -ENODATA) + rc = mdd_links_read_with_rec(env, mdd_sobj, ldata); + if (rc == -ENODATA) { + /* ignore the non-linkEA error */ + ldata = NULL; + rc = 0; + } + if (rc < 0) RETURN(rc); } @@ -3443,13 +3555,16 @@ static int mdd_migrate_create(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); + /* don't set nlink from the original object */ + la->la_valid &= ~LA_NLINK; + /* create the target object */ rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL, hint, handle); if (rc != 0) GOTO(stop_trans, rc); - if (S_ISDIR(la->la_mode)) { + if (S_ISDIR(la->la_mode) && ldata != NULL) { rc = mdd_links_write(env, mdd_tobj, ldata, handle); if (rc != 0) GOTO(stop_trans, rc); @@ -3474,7 +3589,7 @@ static int mdd_migrate_create(const struct lu_env *env, rc = mdo_attr_set(env, mdd_sobj, la_flag, handle); stop_trans: if (handle != NULL) - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); out_free: if (lmm_buf.lb_buf != NULL) OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len); @@ -3491,9 +3606,9 @@ static int mdd_migrate_entries(const struct lu_env *env, struct thandle *handle; struct dt_it *it; const struct dt_it_ops *iops; - int rc; int result; struct lu_dirent *ent; + int rc; ENTRY; OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1); @@ -3526,7 +3641,6 @@ static int mdd_migrate_entries(const struct lu_env *env, struct mdd_object *child; char *name = mdd_env_info(env)->mti_key; int len; - int recsize; int is_dir; bool target_exist = false; @@ -3544,7 +3658,6 @@ static int mdd_migrate_entries(const struct lu_env *env, } fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid); - recsize = le16_to_cpu(ent->lde_reclen); /* Insert new fid with target name into target dir */ if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') || @@ -3556,6 +3669,7 @@ static int mdd_migrate_entries(const struct lu_env *env, if (IS_ERR(child)) GOTO(out, rc = PTR_ERR(child)); + mdd_write_lock(env, child, MOR_SRC_CHILD); is_dir = S_ISDIR(mdd_object_type(child)); snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name); @@ -3572,7 +3686,7 @@ static int mdd_migrate_entries(const struct lu_env *env, handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) - GOTO(out, rc = PTR_ERR(handle)); + GOTO(out_put, rc = PTR_ERR(handle)); /* Note: this transaction is part of migration, and it is not * the last step of migration, so we set th_local = 1 to avoid @@ -3615,7 +3729,7 @@ static int mdd_migrate_entries(const struct lu_env *env, GOTO(out_put, rc); } - rc = mdd_linkea_declare_update_child(env, mdd_tobj, + rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj, child, name, strlen(name), handle); @@ -3654,12 +3768,14 @@ static int mdd_migrate_entries(const struct lu_env *env, GOTO(out_put, rc); } - rc = mdd_linkea_update_child(env, mdd_tobj, child, name, + rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj, + child, name, strlen(name), handle); out_put: + mdd_write_unlock(env, child); mdd_object_put(env, child); - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); if (rc != 0) GOTO(out, rc); next: @@ -3682,22 +3798,24 @@ static int mdd_declare_update_linkea(const struct lu_env *env, struct mdd_object *mdd_pobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - struct thandle *handle, - const struct lu_name *child_name) + const struct lu_name *child_name, + struct linkea_data *ldata, + struct thandle *handle) { return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj, - child_name, handle, 1); + child_name, ldata, handle, 1); } static int mdd_update_linkea(const struct lu_env *env, struct mdd_object *mdd_pobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - struct thandle *handle, - const struct lu_name *child_name) + const struct lu_name *child_name, + struct linkea_data *ldata, + struct thandle *handle) { return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj, - child_name, handle, 0); + child_name, ldata, handle, 0); } static int mdd_declare_migrate_update_name(const struct lu_env *env, @@ -3707,10 +3825,12 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, const struct lu_name *lname, struct lu_attr *la, struct lu_attr *parent_la, + struct linkea_data *ldata, struct thandle *handle) { - struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); - int rc; + struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); + struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); + int rc; /* Revert IMMUTABLE flag */ la_flag->la_valid = LA_FLAGS; @@ -3724,16 +3844,25 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, if (rc != 0) return rc; - rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj, - mdd_tobj, handle, lname); - if (rc != 0) - return rc; + if (ldata->ld_buf != NULL) { + rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj, + mdd_tobj, lname, ldata, handle); + if (rc != 0) + return rc; + } if (S_ISREG(mdd_object_type(mdd_sobj))) { rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV, handle); if (rc != 0) return rc; + + handle->th_complex = 1; + rc = mdo_declare_xattr_set(env, mdd_tobj, NULL, + XATTR_NAME_FID, + LU_XATTR_REPLACE, handle); + if (rc < 0) + return rc; } if (S_ISDIR(mdd_object_type(mdd_sobj))) { @@ -3749,7 +3878,7 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, if (rc != 0) return rc; - rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE); + rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL); if (rc != 0) return rc; @@ -3780,6 +3909,10 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, return rc; rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle); + if (rc != 0) + return rc; + + rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle); return rc; } @@ -3795,6 +3928,7 @@ static int mdd_migrate_update_name(const struct lu_env *env, struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr); struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; struct thandle *handle; int is_dir = S_ISDIR(mdd_object_type(mdd_sobj)); const char *name = lname->ln_name; @@ -3810,12 +3944,18 @@ static int mdd_migrate_update_name(const struct lu_env *env, if (rc != 0) RETURN(rc); + ldata->ld_buf = NULL; + rc = mdd_links_read(env, mdd_sobj, ldata); + if (rc != 0 && rc != -ENOENT && rc != -ENODATA) + RETURN(rc); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, - lname, so_attr, p_la, handle); + lname, so_attr, p_la, ldata, + handle); if (rc != 0) { /* If the migration can not be fit in one transaction, just * leave it in the original MDT */ @@ -3846,10 +3986,18 @@ static int mdd_migrate_update_name(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); - rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj, - handle, lname); - if (rc != 0) - GOTO(stop_trans, rc); + if (ldata->ld_buf != NULL) { + rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj, + lname, ldata, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + /* linkea update might decrease the source object + * nlink, let's get the attr again after ref_del */ + rc = mdd_la_get(env, mdd_sobj, so_attr); + if (rc != 0) + GOTO(stop_trans, rc); + } if (S_ISREG(so_attr->la_mode)) { if (so_attr->la_nlink == 1) { @@ -3857,6 +4005,12 @@ static int mdd_migrate_update_name(const struct lu_env *env, handle); if (rc != 0 && rc != -ENODATA) GOTO(stop_trans, rc); + + rc = mdo_xattr_set(env, mdd_tobj, NULL, + XATTR_NAME_FID, + LU_XATTR_REPLACE, handle); + if (rc < 0) + GOTO(stop_trans, rc); } } @@ -3866,12 +4020,17 @@ static int mdd_migrate_update_name(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); - rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle, - NULL, 1); + mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD); + + mdd_sobj->mod_flags |= DEAD_OBJ; + rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false); if (rc != 0) - GOTO(stop_trans, rc); + GOTO(out_unlock, rc); + + rc = __mdd_orphan_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); - mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); mdo_ref_del(env, mdd_sobj, handle); if (is_dir) mdo_ref_del(env, mdd_sobj, handle); @@ -3883,23 +4042,51 @@ static int mdd_migrate_update_name(const struct lu_env *env, ma->ma_attr = *so_attr; ma->ma_valid |= MA_INODE; - rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle); - if (rc != 0) - GOTO(out_unlock, rc); rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0); if (rc != 0) GOTO(out_unlock, rc); + rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj, + mdo2fid(mdd_pobj), mdo2fid(mdd_sobj), + mdo2fid(mdd_pobj), lname, lname, handle); + if (rc != 0) { + CWARN("%s: changelog for migrate %s "DFID + "under "DFID" failed: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, lname->ln_name, + PFID(mdd_object_fid(mdd_sobj)), + PFID(mdd_object_fid(mdd_pobj)), rc); + /* Sigh, there are no easy way to migrate back the object, so + * let's reset the result to 0 for now XXX */ + rc = 0; + } out_unlock: mdd_write_unlock(env, mdd_sobj); stop_trans: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); RETURN(rc); } +static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd, + const struct lu_fid *fid, __u32 *mdt_index) +{ + struct lu_seq_range *range = &mdd_env_info(env)->mti_range; + struct seq_server_site *ss; + int rc; + + ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site; + + range->lsr_flags = LU_SEQ_RANGE_MDT; + rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range); + if (rc != 0) + return rc; + + *mdt_index = range->lsr_index; + + return 0; +} /** * Check whether we should migrate the file/dir * return val @@ -3916,11 +4103,12 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, { struct mdd_thread_info *info = mdd_env_info(env); struct linkea_data *ldata = &info->mti_link_data; + struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj); int mgr_easize; struct lu_buf *mgr_buf; int count; int rc; - + __u64 mdt_index; ENTRY; mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1); @@ -3962,46 +4150,46 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, if (rc != 0) { /* For multiple links files, if there are no linkEA data at all, * means the file might be created before linkEA is enabled, and - * all all of its links should not be migrated yet, otherwise - * it should have some linkEA there */ + * all of its links should not be migrated yet, otherwise it + * should have some linkEA there */ if (rc == -ENOENT || rc == -ENODATA) RETURN(1); RETURN(rc); } - /* If it is mulitple links file, we need update the name entry for - * all parent */ + mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id; + /* If there are still links locally, then the file will not be + * migrated. */ LASSERT(ldata->ld_leh != NULL); + + /* If the linkEA is overflow, then means there are some unknown name + * entries under unknown parents, that will prevent the migration. */ + if (unlikely(ldata->ld_leh->leh_overflow_time)) + RETURN(1); + ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1); for (count = 0; count < ldata->ld_leh->leh_reccount; count++) { - struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj); - struct mdd_object *lpobj; struct lu_name lname; struct lu_fid fid; + __u32 parent_mdt_index; linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname, &fid); ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + ldata->ld_reclen); - lpobj = mdd_object_find(env, mdd, &fid); - if (IS_ERR(lpobj)) { - CWARN("%s: cannot find obj "DFID": rc = %ld\n", - mdd2obd_dev(mdd)->obd_name, PFID(&fid), - PTR_ERR(lpobj)); - continue; - } - if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) { - CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n", - PFID(&fid), lname.ln_namelen, lname.ln_name); - mdd_object_put(env, lpobj); + rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index); + if (rc != 0) + RETURN(rc); + + /* Migrate the object only if none of its parents are on the + * current MDT. */ + if (parent_mdt_index != mdt_index) continue; - } CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n", PFID(mdd_object_fid(sobj)), lname.ln_namelen, lname.ln_name, PFID(&fid)); - mdd_object_put(env, lpobj); rc = 1; break; } @@ -4016,17 +4204,19 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, struct mdd_object *mdd_pobj = md2mdd_obj(pobj); struct mdd_device *mdd = mdo2mdd(pobj); struct mdd_object *mdd_sobj = md2mdd_obj(sobj); - struct mdd_object *mdd_tobj = NULL; + struct mdd_object *mdd_tobj = md2mdd_obj(tobj); struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr); struct lu_attr *pattr = MDD_ENV_VAR(env, pattr); + bool created = false; int rc; ENTRY; /* If the file will being migrated, it will check whether * the file is being opened by someone else right now */ mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD); - if (mdd_sobj->mod_count >= 1) { - CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n", + if (mdd_sobj->mod_count > 0) { + CDEBUG(D_OTHER, + "%s: "DFID"%s is already opened count %d: rc = %d\n", mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_sobj)), lname->ln_name, mdd_sobj->mod_count, -EBUSY); @@ -4072,14 +4262,15 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, /* step 1: Check whether the orphan object has been created, and create * orphan object on the remote MDT if needed */ - mdd_tobj = md2mdd_obj(tobj); if (!mdd_object_exists(mdd_tobj)) { rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, lname, so_attr); if (rc != 0) GOTO(put, rc); + created = true; } + LASSERT(mdd_object_exists(mdd_tobj)); /* step 2: migrate xattr */ rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj); if (rc != 0) @@ -4097,11 +4288,16 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val); } + LASSERT(mdd_object_exists(mdd_tobj)); /* step 4: update name entry to the new object */ rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname, ma); if (rc != 0) GOTO(put, rc); + + /* newly created target was not locked, don't cache its attributes */ + if (created) + mdd_invalidate(env, tobj); put: RETURN(rc); }