X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=d607bba91ab5f88c91fe54c5655ee59d45475c09;hp=936554251f91759a568025f66d7dcffc2c42acdd;hb=6350af100c204a4db794e7472951ab72c3bc20ef;hpb=9f55850b884cac1c7bbde6d3b02764b712a2921f diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 9365542..d607bba 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -29,8 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011 Whamcloud, Inc. - * + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -40,32 +37,14 @@ * * Lustre Metadata Server (mdd) routines * - * Author: Wang Di + * Author: Wang Di */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS -#include -#ifdef HAVE_EXT4_LDISKFS -#include -#else -#include -#endif -#include #include -#include #include -#include -#ifdef HAVE_EXT4_LDISKFS -#include -#else -#include -#endif #include -#include #include #include "mdd_internal.h" @@ -78,24 +57,8 @@ static struct lu_name lname_dotdot = { sizeof(dotdot) - 1 }; -static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj, - const struct lu_name *lname, struct lu_fid* fid, - int mask); -static int mdd_declare_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct thandle *handle); -static int mdd_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *pfid, - const struct lu_name *lname, - struct thandle *handle, int first); -static int mdd_links_rename(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *oldpfid, - const struct lu_name *oldlname, - const struct lu_fid *newpfid, - const struct lu_name *newlname, - struct thandle *handle); +static int __mdd_lookup(const struct lu_env *, struct md_object *, + const struct lu_name *, struct lu_fid*, int); static int __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj, @@ -125,8 +88,8 @@ int mdd_lookup(const struct lu_env *env, RETURN(rc); } -static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj, - struct lu_fid *fid) +int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj, + struct lu_fid *fid) { return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0); } @@ -180,16 +143,17 @@ static int mdd_is_parent(const struct lu_env *env, GOTO(out, rc = 1); if (parent) mdd_object_put(env, parent); - parent = mdd_object_find(env, mdd, pfid); - - /* cross-ref parent */ - if (parent == NULL) { - if (pf != NULL) - *pf = *pfid; - GOTO(out, rc = -EREMOTE); - } else if (IS_ERR(parent)) - GOTO(out, rc = PTR_ERR(parent)); - p1 = parent; + + parent = mdd_object_find(env, mdd, pfid); + if (IS_ERR(parent)) { + GOTO(out, rc = PTR_ERR(parent)); + } else if (mdd_object_remote(parent)) { + /*FIXME: Because of the restriction of rename in Phase I. + * If the parent is remote, we just assumed lf is not the + * parent of P1 for now */ + GOTO(out, rc = 0); + } + p1 = parent; } EXIT; out: @@ -311,7 +275,7 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, int rc = 0; ENTRY; - if (cobj && mdd_object_exists(cobj)) + if (cobj && mdd_object_exists(cobj)) RETURN(-EEXIST); if (mdd_is_dead_obj(pobj)) @@ -321,7 +285,6 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, rc = mdd_permission_internal_locked(env, pobj, NULL, MAY_WRITE | MAY_EXEC, MOR_TGT_PARENT); - if (!rc && check_nlink) rc = __mdd_may_link(env, pobj); @@ -332,7 +295,7 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, * Check whether can unlink from the pobj in the case of "cobj == NULL". */ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, - const struct md_attr *ma) + const struct lu_attr *attr) { int rc; ENTRY; @@ -340,8 +303,8 @@ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, if (mdd_is_dead_obj(pobj)) RETURN(-ENOENT); - if ((ma->ma_attr.la_valid & LA_FLAGS) && - (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + if ((attr->la_valid & LA_FLAGS) && + (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) RETURN(-EPERM); rc = mdd_permission_internal_locked(env, pobj, NULL, @@ -361,31 +324,58 @@ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, * VTX feature has been checked already, no need check again. */ static inline int mdd_is_sticky(const struct lu_env *env, - struct mdd_object *pobj, - struct mdd_object *cobj) + struct mdd_object *pobj, + struct mdd_object *cobj) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - struct md_ucred *uc = md_ucred(env); - int rc; + struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; + struct lu_ucred *uc = lu_ucred_assert(env); + int rc; - if (pobj) { - rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA); - if (rc) - return rc; + if (pobj) { + rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA); + if (rc) + return rc; - if (!(tmp_la->la_mode & S_ISVTX) || - (tmp_la->la_uid == uc->mu_fsuid)) - return 0; - } + if (!(tmp_la->la_mode & S_ISVTX) || + (tmp_la->la_uid == uc->uc_fsuid)) + return 0; + } - rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA); - if (rc) - return rc; + rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA); + if (rc) + return rc; + + if (tmp_la->la_uid == uc->uc_fsuid) + return 0; + + return !md_capable(uc, CFS_CAP_FOWNER); +} + +static int mdd_may_delete_entry(const struct lu_env *env, + struct mdd_object *pobj, int check_perm) +{ + ENTRY; + + LASSERT(pobj != NULL); + if (!mdd_object_exists(pobj)) + RETURN(-ENOENT); - if (tmp_la->la_uid == uc->mu_fsuid) - return 0; + if (mdd_is_dead_obj(pobj)) + RETURN(-ENOENT); - return !mdd_capable(uc, CFS_CAP_FOWNER); + if (check_perm) { + int rc; + rc = mdd_permission_internal_locked(env, pobj, NULL, + MAY_WRITE | MAY_EXEC, + MOR_TGT_PARENT); + if (rc) + RETURN(rc); + } + + if (mdd_is_append(pobj)) + RETURN(-EPERM); + + RETURN(0); } /* @@ -393,61 +383,53 @@ static inline int mdd_is_sticky(const struct lu_env *env, * pobj maybe NULL */ int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj, - struct mdd_object *cobj, struct md_attr *ma, - int check_perm, int check_empty) + struct mdd_object *cobj, struct lu_attr *cattr, + struct lu_attr *src_attr, int check_perm, int check_empty) { int rc = 0; ENTRY; - LASSERT(cobj); + if (pobj) { + rc = mdd_may_delete_entry(env, pobj, check_perm); + if (rc != 0) + RETURN(rc); + } + + if (cobj == NULL) + RETURN(0); + if (!mdd_object_exists(cobj)) RETURN(-ENOENT); if (mdd_is_dead_obj(cobj)) RETURN(-ESTALE); - if (pobj) { - if (!mdd_object_exists(pobj)) - RETURN(-ENOENT); - - if (mdd_is_dead_obj(pobj)) - RETURN(-ENOENT); - - if (check_perm) { - rc = mdd_permission_internal_locked(env, pobj, NULL, - MAY_WRITE | MAY_EXEC, - MOR_TGT_PARENT); - if (rc) - RETURN(rc); - } - - if (mdd_is_append(pobj)) - RETURN(-EPERM); - } - if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) && - mdd_is_sticky(env, pobj, cobj)) + if (mdd_is_sticky(env, pobj, cobj)) RETURN(-EPERM); if (mdd_is_immutable(cobj) || mdd_is_append(cobj)) RETURN(-EPERM); - if ((ma->ma_attr.la_valid & LA_FLAGS) && - (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + if ((cattr->la_valid & LA_FLAGS) && + (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) RETURN(-EPERM); - if (S_ISDIR(ma->ma_attr.la_mode)) { - struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj); + /* additional check the rename case */ + if (src_attr) { + if (S_ISDIR(src_attr->la_mode)) { + struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj); - if (!S_ISDIR(mdd_object_type(cobj))) - RETURN(-ENOTDIR); + if (!S_ISDIR(cattr->la_mode)) + RETURN(-ENOTDIR); - if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid)) - RETURN(-EBUSY); - } else if (S_ISDIR(mdd_object_type(cobj))) - RETURN(-EISDIR); + if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid)) + RETURN(-EBUSY); + } else if (S_ISDIR(cattr->la_mode)) + RETURN(-EISDIR); + } - if (S_ISDIR(ma->ma_attr.la_mode) && check_empty) + if (S_ISDIR(cattr->la_mode) && check_empty) rc = mdd_dir_is_empty(env, cobj); RETURN(rc); @@ -494,43 +476,6 @@ int mdd_link_sanity_check(const struct lu_env *env, RETURN(rc); } -/** - * If subdir count is up to ddp_max_nlink, then enable MNLINK_OBJ flag and - * assign i_nlink to 1 which means the i_nlink for subdir count is incredible - * (maybe too large to be represented). It is a trick to break through the - * "i_nlink" limitation for subdir count. - */ -void __mdd_ref_add(const struct lu_env *env, struct mdd_object *obj, - struct thandle *handle) -{ - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - struct mdd_device *m = mdd_obj2mdd_dev(obj); - - if (!mdd_is_mnlink(obj)) { - if (S_ISDIR(mdd_object_type(obj))) { - if (mdd_la_get(env, obj, tmp_la, BYPASS_CAPA)) - return; - - if (tmp_la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) { - obj->mod_flags |= MNLINK_OBJ; - tmp_la->la_nlink = 1; - tmp_la->la_valid = LA_NLINK; - mdd_attr_set_internal(env, obj, tmp_la, handle, - 0); - return; - } - } - mdo_ref_add(env, obj, handle); - } -} - -void __mdd_ref_del(const struct lu_env *env, struct mdd_object *obj, - struct thandle *handle, int is_dot) -{ - if (!mdd_is_mnlink(obj) || is_dot) - mdo_ref_del(env, obj, handle); -} - static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj, const char *name, struct thandle *handle, struct lustre_capa *capa) @@ -550,27 +495,28 @@ static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object * } static int __mdd_index_insert_only(const struct lu_env *env, - struct mdd_object *pobj, - const struct lu_fid *lf, const char *name, - struct thandle *handle, - struct lustre_capa *capa) + struct mdd_object *pobj, + const struct lu_fid *lf, const char *name, + struct thandle *handle, + struct lustre_capa *capa) { - struct dt_object *next = mdd_object_child(pobj); - int rc; - ENTRY; - - if (dt_try_as_dir(env, next)) { - struct md_ucred *uc = md_ucred(env); - - rc = next->do_index_ops->dio_insert(env, next, - (struct dt_rec*)lf, - (const struct dt_key *)name, - handle, capa, uc->mu_cap & - CFS_CAP_SYS_RESOURCE_MASK); - } else { - rc = -ENOTDIR; - } - RETURN(rc); + struct dt_object *next = mdd_object_child(pobj); + int rc; + ENTRY; + + if (dt_try_as_dir(env, next)) { + struct lu_ucred *uc = lu_ucred_check(env); + int ignore_quota; + + ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1; + rc = next->do_index_ops->dio_insert(env, next, + (struct dt_rec*)lf, + (const struct dt_key *)name, + handle, capa, ignore_quota); + } else { + rc = -ENOTDIR; + } + RETURN(rc); } /* insert named index, add reference if isdir */ @@ -584,7 +530,7 @@ static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj, rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa); if (rc == 0 && is_dir) { mdd_write_lock(env, pobj, MOR_TGT_PARENT); - __mdd_ref_add(env, pobj, handle); + mdo_ref_add(env, pobj, handle); mdd_write_unlock(env, pobj); } RETURN(rc); @@ -600,138 +546,594 @@ static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj, rc = __mdd_index_delete_only(env, pobj, name, handle, capa); if (rc == 0 && is_dir) { - int is_dot = 0; - - if (name != NULL && name[0] == '.' && name[1] == 0) - is_dot = 1; mdd_write_lock(env, pobj, MOR_TGT_PARENT); - __mdd_ref_del(env, pobj, handle, is_dot); + mdo_ref_del(env, pobj, handle); mdd_write_unlock(env, pobj); } RETURN(rc); } -int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd, - int reclen, struct thandle *handle) +int mdd_declare_changelog_store(const struct lu_env *env, + struct mdd_device *mdd, + const struct lu_name *fname, + struct thandle *handle) { - int rc; + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + struct llog_changelog_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; + + /* Not recording */ + if (!(mdd->mdd_cl.mc_flags & CLM_ON)) + return 0; + + reclen = llog_data_len(sizeof(*rec) + + (fname != NULL ? fname->ln_namelen : 0)); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); + if (buf->lb_buf == NULL) + return -ENOMEM; + + rec = buf->lb_buf; + rec->cr_hdr.lrh_len = reclen; + rec->cr_hdr.lrh_type = CHANGELOG_REC; + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); + llog_ctxt_put(ctxt); + + return rc; +} + +static int mdd_declare_changelog_ext_store(const struct lu_env *env, + struct mdd_device *mdd, + const struct lu_name *tname, + const struct lu_name *sname, + struct thandle *handle) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + struct llog_changelog_ext_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; + + /* Not recording */ + if (!(mdd->mdd_cl.mc_flags & CLM_ON)) + return 0; + + reclen = llog_data_len(sizeof(*rec) + + (tname != NULL ? tname->ln_namelen : 0) + + (sname != NULL ? 1 + sname->ln_namelen : 0)); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); + if (buf->lb_buf == NULL) + return -ENOMEM; + + rec = buf->lb_buf; + rec->cr_hdr.lrh_len = reclen; + rec->cr_hdr.lrh_type = CHANGELOG_REC; + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); + llog_ctxt_put(ctxt); + + return rc; +} + +/** Add a changelog entry \a rec to the changelog llog + * \param mdd + * \param rec + * \param handle - currently ignored since llogs start their own transaction; + * this will hopefully be fixed in llog rewrite + * \retval 0 ok + */ +int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd, + struct llog_changelog_rec *rec, struct thandle *th) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + int rc; + + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); + rec->cr_hdr.lrh_type = CHANGELOG_REC; + rec->cr.cr_time = cl_time(); + + spin_lock(&mdd->mdd_cl.mc_lock); + /* NB: I suppose it's possible llog_add adds out of order wrt cr_index, + * but as long as the MDD transactions are ordered correctly for e.g. + * rename conflicts, I don't think this should matter. */ + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th); + llog_ctxt_put(ctxt); + if (rc > 0) + rc = 0; + return rc; +} - /* XXX: this is a temporary solution to declare llog changes - * will be fixed in 2.3 with new llog implementation */ +/** Add a changelog_ext entry \a rec to the changelog llog + * \param mdd + * \param rec + * \param handle - currently ignored since llogs start their own transaction; + * this will hopefully be fixed in llog rewrite + * \retval 0 ok + */ +int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd, + struct llog_changelog_ext_rec *rec, + struct thandle *th) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + int rc; + + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); + /* llog_lvfs_write_rec sets the llog tail len */ + rec->cr_hdr.lrh_type = CHANGELOG_REC; + rec->cr.cr_time = cl_time(); + + spin_lock(&mdd->mdd_cl.mc_lock); + /* NB: I suppose it's possible llog_add adds out of order wrt cr_index, + * but as long as the MDD transactions are ordered correctly for e.g. + * rename conflicts, I don't think this should matter. */ + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + /* nested journal transaction */ + rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th); + llog_ctxt_put(ctxt); + if (rc > 0) + rc = 0; + + return rc; +} - LASSERT(mdd->mdd_capa); +/** Store a namespace change changelog record + * If this fails, we must fail the whole transaction; we don't + * want the change to commit without the log entry. + * \param target - mdd_object of change + * \param parent - parent dir/object + * \param tname - target name string + * \param handle - transacion handle + */ +int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd, + enum changelog_rec_type type, unsigned flags, + struct mdd_object *target, struct mdd_object *parent, + const struct lu_name *tname, struct thandle *handle) +{ + struct llog_changelog_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; + ENTRY; + + /* Not recording */ + if (!(mdd->mdd_cl.mc_flags & CLM_ON)) + RETURN(0); + if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) + RETURN(0); + + LASSERT(target != NULL); + LASSERT(parent != NULL); + LASSERT(tname != NULL); + LASSERT(handle != NULL); + + reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + rec = buf->lb_buf; + + rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags); + rec->cr.cr_type = (__u32)type; + rec->cr.cr_tfid = *mdo2fid(target); + rec->cr.cr_pfid = *mdo2fid(parent); + rec->cr.cr_namelen = tname->ln_namelen; + memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen); + + target->mod_cltime = cfs_time_current_64(); + + rc = mdd_changelog_store(env, mdd, rec, handle); + if (rc < 0) { + CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", + rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid), + PFID(&rec->cr.cr_pfid)); + RETURN(-EFAULT); + } + + RETURN(0); +} - /* record itself */ - rc = dt_declare_record_write(env, mdd->mdd_capa, reclen, 0, handle); - if (rc) - return rc; - /* header will be updated as well */ - rc = dt_declare_record_write(env, mdd->mdd_capa, LLOG_CHUNK_SIZE, - 0, handle); - if (rc) - return rc; +/** Store a namespace change changelog record + * If this fails, we must fail the whole transaction; we don't + * want the change to commit without the log entry. + * \param target - mdd_object of change + * \param tpfid - target parent dir/object fid + * \param sfid - source object fid + * \param spfid - source parent fid + * \param tname - target name string + * \param sname - source name string + * \param handle - transacion handle + */ +static int mdd_changelog_ext_ns_store(const struct lu_env *env, + struct mdd_device *mdd, + enum changelog_rec_type type, + unsigned flags, + struct mdd_object *target, + const struct lu_fid *tpfid, + const struct lu_fid *sfid, + const struct lu_fid *spfid, + const struct lu_name *tname, + const struct lu_name *sname, + struct thandle *handle) +{ + struct llog_changelog_ext_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; + ENTRY; + + /* Not recording */ + if (!(mdd->mdd_cl.mc_flags & CLM_ON)) + RETURN(0); + if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) + RETURN(0); + + LASSERT(sfid != NULL); + LASSERT(tpfid != NULL); + LASSERT(tname != NULL); + LASSERT(handle != NULL); + + reclen = llog_data_len(sizeof(*rec) + + sname != NULL ? 1 + sname->ln_namelen : 0); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + rec = buf->lb_buf; + + rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags); + rec->cr.cr_type = (__u32)type; + rec->cr.cr_pfid = *tpfid; + rec->cr.cr_sfid = *sfid; + rec->cr.cr_spfid = *spfid; + rec->cr.cr_namelen = tname->ln_namelen; + memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen); + if (sname) { + rec->cr.cr_name[tname->ln_namelen] = '\0'; + memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name, + sname->ln_namelen); + rec->cr.cr_namelen += 1 + sname->ln_namelen; + } + + if (likely(target != NULL)) { + rec->cr.cr_tfid = *mdo2fid(target); + target->mod_cltime = cfs_time_current_64(); + } else { + fid_zero(&rec->cr.cr_tfid); + } + + rc = mdd_changelog_ext_store(env, mdd, rec, handle); + if (rc < 0) { + CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", + rc, type, tname->ln_name, PFID(sfid), PFID(tpfid)); + return -EFAULT; + } + + return 0; +} - /* also we should be able to create new plain log */ - rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle); - if (rc) - return rc; +static int __mdd_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata, + const struct lu_name *lname, + const struct lu_fid *pfid, + int first, int check) +{ + int rc; + + if (ldata->ld_leh == NULL) { + rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata); + if (rc) { + if (rc != -ENODATA) + return rc; + rc = linkea_data_new(ldata, + &mdd_env_info(env)->mti_link_buf); + if (rc) + return rc; + } + } + + if (check) { + rc = linkea_links_find(ldata, lname, pfid); + if (rc && rc != -ENOENT) + return rc; + if (rc == 0) + return -EEXIST; + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) { + struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2; + + *tfid = *pfid; + tfid->f_ver = ~0; + linkea_add_buf(ldata, lname, tfid); + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2)) + linkea_add_buf(ldata, lname, pfid); + + return linkea_add_buf(ldata, lname, pfid); +} - /* new record referencing new plain llog */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - sizeof(struct llog_logid_rec), 0, handle); - if (rc) - return rc; +static int __mdd_links_del(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata, + const struct lu_name *lname, + const struct lu_fid *pfid) +{ + int rc; - /* catalog's header will be updated as well */ - rc = dt_declare_record_write(env, mdd->mdd_capa, LLOG_CHUNK_SIZE, - 0, handle); + if (ldata->ld_leh == NULL) { + rc = mdd_links_read(env, mdd_obj, ldata); + if (rc) + return rc; + } - return rc; + rc = linkea_links_find(ldata, lname, pfid); + if (rc) + return rc; + + linkea_del_buf(ldata, lname); + return 0; } -int mdd_declare_changelog_store(const struct lu_env *env, - struct mdd_device *mdd, - const struct lu_name *fname, - struct thandle *handle) +static int mdd_linkea_prepare(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *oldpfid, + const struct lu_name *oldlname, + const struct lu_fid *newpfid, + const struct lu_name *newlname, + int first, int check, + struct linkea_data *ldata) { - int reclen; + int rc = 0; + int rc2 = 0; + ENTRY; + + if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) + 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 */ + RETURN(0); + } + + if (oldpfid != NULL) { + rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid); + if (rc) { + 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; + + RETURN(rc); +} - /* Not recording */ - if (!(mdd->mdd_cl.mc_flags & CLM_ON)) - return 0; +int mdd_links_rename(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *oldpfid, + const struct lu_name *oldlname, + const struct lu_fid *newpfid, + const struct lu_name *newlname, + struct thandle *handle, + struct linkea_data *ldata, + int first, int check) +{ + int rc2 = 0; + int rc = 0; + ENTRY; + + if (ldata == NULL) { + 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) + GOTO(out, rc); + } + + if (ldata->ld_reclen != 0) + rc = mdd_links_write(env, mdd_obj, ldata, handle); + EXIT; +out: + if (rc == 0) + rc = rc2; + if (rc) { + 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, + 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))); + 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))); + } + + if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + lu_buf_free(ldata->ld_buf); + + return rc; +} - /* we'll be writing payload + llog header */ - reclen = sizeof(struct llog_changelog_rec); - if (fname) - reclen += llog_data_len(fname->ln_namelen); +static inline int mdd_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct thandle *handle, + struct linkea_data *data, int first) +{ + return mdd_links_rename(env, mdd_obj, NULL, NULL, + pfid, lname, handle, data, first, 0); +} - return mdd_declare_llog_record(env, mdd, reclen, handle); +static inline int mdd_links_del(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct thandle *handle) +{ + return mdd_links_rename(env, mdd_obj, pfid, lname, + NULL, NULL, handle, NULL, 0, 0); } -/** Store a namespace change changelog record - * If this fails, we must fail the whole transaction; we don't - * want the change to commit without the log entry. - * \param target - mdd_object of change - * \param parent - parent dir/object - * \param tf - target lu_fid, overrides fid of \a target if this is non-null - * \param tname - target name string - * \param handle - transacion handle +/** 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_changelog_ns_store(const struct lu_env *env, - struct mdd_device *mdd, - enum changelog_rec_type type, - int flags, - struct mdd_object *target, - struct mdd_object *parent, - const struct lu_fid *tf, - const struct lu_name *tname, - struct thandle *handle) +int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj, + struct linkea_data *ldata) { - const struct lu_fid *tfid; - const struct lu_fid *tpfid = mdo2fid(parent); - struct llog_changelog_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; - ENTRY; + int rc; + + /* 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; + + if (!mdd_object_exists(mdd_obj)) + return -ENODATA; + + rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK, + BYPASS_CAPA); + 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, BYPASS_CAPA); + 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, BYPASS_CAPA); + } + if (rc < 0) + return rc; + + return linkea_init(ldata); +} - /* Not recording */ - if (!(mdd->mdd_cl.mc_flags & CLM_ON)) - RETURN(0); - if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) - RETURN(0); +/** 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) + */ +struct lu_buf *mdd_links_get(const struct lu_env *env, + struct mdd_object *mdd_obj) +{ + struct linkea_data ldata = { 0 }; + int rc; - LASSERT(parent != NULL); - LASSERT(tname != NULL); - LASSERT(handle != NULL); - - /* target */ - reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); - buf = mdd_buf_alloc(env, reclen); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); - rec = (struct llog_changelog_rec *)buf->lb_buf; - - rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags); - rec->cr.cr_type = (__u32)type; - tfid = tf ? tf : mdo2fid(target); - rec->cr.cr_tfid = *tfid; - rec->cr.cr_pfid = *tpfid; - rec->cr.cr_namelen = tname->ln_namelen; - memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen); - if (likely(target)) - target->mod_cltime = cfs_time_current_64(); - - rc = mdd_changelog_llog_write(mdd, rec, handle); - if (rc < 0) { - CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", - rc, type, tname->ln_name, PFID(tfid), PFID(tpfid)); - return -EFAULT; - } + rc = mdd_links_read(env, mdd_obj, &ldata); + return rc ? ERR_PTR(rc) : ldata.ld_buf; +} + +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); + return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle, + mdd_object_capa(env, mdd_obj)); +} + +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; + void *linkea; + + if (ldata != NULL && ldata->ld_lee != NULL) { + ea_len = ldata->ld_leh->leh_len; + linkea = ldata->ld_buf->lb_buf; + } else { + ea_len = DEFAULT_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); + return rc; +} + +static inline int mdd_declare_links_del(const struct lu_env *env, + struct mdd_object *c, + struct thandle *handle) +{ + int rc = 0; - return 0; + /* 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); + + return rc; } static int mdd_declare_link(const struct lu_env *env, @@ -739,7 +1141,9 @@ static int mdd_declare_link(const struct lu_env *env, struct mdd_object *p, struct mdd_object *c, const struct lu_name *name, - struct thandle *handle) + struct thandle *handle, + struct lu_attr *la, + struct linkea_data *data) { int rc; @@ -751,15 +1155,17 @@ static int mdd_declare_link(const struct lu_env *env, if (rc) return rc; - rc = mdo_declare_attr_set(env, p, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, p, la, handle); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, c, NULL, handle); + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, c, la, handle); if (rc) return rc; - rc = mdd_declare_links_add(env, c, handle); + rc = mdd_declare_links_add(env, c, handle, data); if (rc) return rc; @@ -779,39 +1185,21 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, struct mdd_device *mdd = mdo2mdd(src_obj); struct dynlock_handle *dlh; struct thandle *handle; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 }; -#endif + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; int rc; ENTRY; -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_tobj, &data); - quota_opc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qids); - /* get block quota for parent */ - lquota_chkquota(mds_quota_interface_ref, obd, exp, - qids, rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, data, 1); - } - } -#endif - handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); - rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle); + memset(ldata, 0, sizeof(*ldata)); + + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + + rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle, + la, ldata); if (rc) GOTO(stop, rc); @@ -828,65 +1216,67 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, if (rc) GOTO(out_unlock, rc); - rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), - name, handle, - mdd_object_capa(env, mdd_tobj)); - if (rc) - GOTO(out_unlock, rc); + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc) + GOTO(out_unlock, rc); - __mdd_ref_add(env, mdd_sobj, handle); - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), + name, handle, + mdd_object_capa(env, mdd_tobj)); + if (rc != 0) { + mdo_ref_del(env, mdd_sobj, handle); + GOTO(out_unlock, rc); + } la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); if (rc) GOTO(out_unlock, rc); la->la_valid = LA_CTIME; rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0); - if (rc == 0) { - mdd_links_add(env, mdd_sobj, - mdo2fid(mdd_tobj), lname, handle, 0); - } - + 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; out_unlock: mdd_write_unlock(env, mdd_sobj); mdd_pdo_write_unlock(env, mdd_tobj, dlh); out_trans: if (rc == 0) - rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, - mdd_tobj, NULL, lname, handle); + rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, + mdd_tobj, lname, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); + + if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + lu_buf_free(ldata->ld_buf); out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (quota_opc) { - lquota_pending_commit(mds_quota_interface_ref, obd, - qids, rec_pending, 1); - /* Trigger dqacq for the parent owner. If failed, - * the next call for lquota_chkquota will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc, - quota_opc); - } -#endif return rc; } int mdd_declare_finish_unlink(const struct lu_env *env, - struct mdd_object *obj, - struct md_attr *ma, - struct thandle *handle) + struct mdd_object *obj, + struct md_attr *ma, + struct thandle *handle) { - int rc; + int rc; - rc = orph_declare_index_insert(env, obj, handle); - if (rc) - return rc; + rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); + if (rc) + return rc; - return mdd_declare_object_kill(env, obj, ma, handle); + return mdo_declare_destroy(env, obj, handle); } /* caller should take a lock before calling */ @@ -894,17 +1284,15 @@ int mdd_finish_unlink(const struct lu_env *env, struct mdd_object *obj, struct md_attr *ma, struct thandle *th) { - int rc; - int reset = 1; + int rc = 0; + int is_dir = S_ISDIR(ma->ma_attr.la_mode); ENTRY; LASSERT(mdd_write_locked(env, obj) != 0); - /* read HSM flags, needed to set changelogs flags */ - ma->ma_need = MA_HSM | MA_INODE; - rc = mdd_attr_get_internal(env, obj, ma); - if (rc == 0 && ma->ma_attr.la_nlink == 0) { + if (ma->ma_attr.la_nlink == 0 || is_dir) { obj->mod_flags |= DEAD_OBJ; + /* add new orphan and the object * will be deleted during mdd_close() */ if (obj->mod_count) { @@ -921,14 +1309,9 @@ int mdd_finish_unlink(const struct lu_env *env, PFID(mdd_object_fid(obj)), obj->mod_count); } else { - rc = mdd_object_kill(env, obj, ma, th); - if (rc == 0) - reset = 0; + rc = mdo_destroy(env, obj, th); } - } - if (reset) - ma->ma_valid &= ~(MA_LOV | MA_COOKIE); RETURN(rc); } @@ -938,126 +1321,181 @@ int mdd_finish_unlink(const struct lu_env *env, * has mdd_write_lock on cobj already, but not on pobj yet */ int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj, - struct mdd_object *cobj, struct md_attr *ma) + struct mdd_object *cobj, struct lu_attr *cattr) { int rc; ENTRY; - rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1); + rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1); RETURN(rc); } static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd, - struct mdd_object *p, struct mdd_object *c, - const struct lu_name *name, struct md_attr *ma, - struct thandle *handle) + struct mdd_object *p, struct mdd_object *c, + const struct lu_name *name, struct md_attr *ma, + struct thandle *handle, int no_name) { + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; int rc; - rc = mdo_declare_index_delete(env, p, name->ln_name, handle); - if (rc) - return rc; + if (likely(no_name == 0)) { + rc = mdo_declare_index_delete(env, p, name->ln_name, handle); + if (rc) + return rc; + } rc = mdo_declare_ref_del(env, p, handle); if (rc) return rc; - rc = mdo_declare_ref_del(env, c, handle); - if (rc) - return rc; + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, p, la, handle); + if (rc) + return rc; - rc = mdo_declare_ref_del(env, c, handle); - if (rc) - return rc; + if (c != NULL) { + rc = mdo_declare_ref_del(env, c, handle); + if (rc) + return rc; - rc = mdo_declare_attr_set(env, p, NULL, handle); - if (rc) - return rc; + rc = mdo_declare_ref_del(env, c, handle); + if (rc) + return rc; - rc = mdo_declare_attr_set(env, c, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, c, la, handle); + if (rc) + return rc; - rc = mdd_declare_finish_unlink(env, c, ma, handle); - if (rc) - return rc; + rc = mdd_declare_finish_unlink(env, c, ma, handle); + if (rc) + return rc; - rc = mdd_declare_links_add(env, c, handle); - if (rc) - return rc; + rc = mdd_declare_links_del(env, c, handle); + if (rc != 0) + return rc; - rc = mdd_declare_changelog_store(env, mdd, name, handle); + /* FIXME: need changelog for remove entry */ + rc = mdd_declare_changelog_store(env, mdd, name, handle); + } - return rc; + return rc; } +/** + * Delete name entry and the object. + * Note: no_name == 1 means it only destory the object, i.e. name_entry + * does not exist for this object, and it could only happen during resending + * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname + * is also needed in this case(needed by changelog), so we have to add another + * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II, + * the ENOENT failure should be able to be fixed by redo mechanism. + */ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, - struct md_object *cobj, const struct lu_name *lname, - struct md_attr *ma) + struct md_object *cobj, const struct lu_name *lname, + struct md_attr *ma, int no_name) { const char *name = lname->ln_name; + struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr; struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; struct mdd_object *mdd_pobj = md2mdd_obj(pobj); - struct mdd_object *mdd_cobj = md2mdd_obj(cobj); + struct mdd_object *mdd_cobj = NULL; struct mdd_device *mdd = mdo2mdd(pobj); struct dynlock_handle *dlh; struct thandle *handle; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct mds_obd *mds = &obd->u.mds; - unsigned int qcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qpids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0; -#endif - int is_dir = S_ISDIR(ma->ma_attr.la_mode); - int rc; + int rc, is_dir = 0; ENTRY; - if (mdd_object_exists(mdd_cobj) <= 0) - RETURN(-ENOENT); - - handle = mdd_trans_create(env, mdd); + /* cobj == NULL means only delete name entry */ + if (likely(cobj != NULL)) { + mdd_cobj = md2mdd_obj(cobj); + if (mdd_object_exists(mdd_cobj) == 0) + RETURN(-ENOENT); + /* currently it is assume, it could only delete + * name entry of remote directory */ + is_dir = 1; + } + + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj, - lname, ma, handle); - if (rc) - GOTO(stop, rc); - - rc = mdd_trans_start(env, mdd, handle); - if (rc) - GOTO(stop, rc); - - dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); - - rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma); - if (rc) - GOTO(cleanup, rc); - - rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, - mdd_object_capa(env, mdd_pobj)); - if (rc) - GOTO(cleanup, rc); - - __mdd_ref_del(env, mdd_cobj, handle, 0); - if (is_dir) - /* unlink dot */ - __mdd_ref_del(env, mdd_cobj, handle, 1); - - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); - - if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) { + rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj, + lname, ma, handle, no_name); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); + if (dlh == NULL) + GOTO(stop, rc = -ENOMEM); + + if (likely(mdd_cobj != NULL)) { + mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); + + /* fetch cattr */ + rc = mdd_la_get(env, mdd_cobj, cattr, + mdd_object_capa(env, mdd_cobj)); + if (rc) + GOTO(cleanup, rc); + + is_dir = S_ISDIR(cattr->la_mode); + + } + + rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr); + if (rc) + GOTO(cleanup, rc); + + if (likely(no_name == 0)) { + rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, + mdd_object_capa(env, mdd_pobj)); + if (rc) + GOTO(cleanup, rc); + } + + if (likely(mdd_cobj != NULL)) { + rc = mdo_ref_del(env, mdd_cobj, handle); + if (rc != 0) { + __mdd_index_insert_only(env, mdd_pobj, + mdo2fid(mdd_cobj), + name, handle, + mdd_object_capa(env, mdd_pobj)); + GOTO(cleanup, rc); + } + + if (is_dir) + /* unlink dot */ + mdo_ref_del(env, mdd_cobj, handle); + + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_cobj, cattr, + mdd_object_capa(env, mdd_cobj)); + if (rc) + GOTO(cleanup, rc); + } + + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0); + if (rc) + GOTO(cleanup, rc); + + /* Enough for only unlink the entry */ + if (unlikely(mdd_cobj == NULL)) { + mdd_pdo_write_unlock(env, mdd_pobj, dlh); + GOTO(stop, rc); + } + + if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) { /* update ctime of an unlinked file only if it is still * opened or a link still exists */ la->la_valid = LA_CTIME; @@ -1066,505 +1504,74 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, GOTO(cleanup, rc); } + /* XXX: this transfer to ma will be removed with LOD/OSP */ + ma->ma_attr = *cattr; + ma->ma_valid |= MA_INODE; rc = mdd_finish_unlink(env, mdd_cobj, ma, handle); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && ma->ma_valid & MA_INODE && - ma->ma_attr.la_nlink == 0) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA); - if (!rc) { - mdd_quota_wrapper(la_tmp, qpids); - if (mdd_cobj->mod_count == 0) { - quota_opc = FSFILT_OP_UNLINK; - mdd_quota_wrapper(&ma->ma_attr, qcids); - } else { - quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT; - } - } - } -#endif - if (!is_dir) - /* old files may not have link ea; ignore errors */ - mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj), - lname, NULL, NULL, handle); + /* fetch updated nlink */ + if (rc == 0) + rc = mdd_la_get(env, mdd_cobj, cattr, + mdd_object_capa(env, mdd_cobj)); + + if (!is_dir) + /* old files may not have link ea; ignore errors */ + mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle); + + /* 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: mdd_write_unlock(env, mdd_cobj); mdd_pdo_write_unlock(env, mdd_pobj, dlh); -out_trans: if (rc == 0) { int cl_flags; - cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0; + cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0; if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_EXISTS)) cl_flags |= CLF_UNLINK_HSM_EXISTS; - rc = mdd_changelog_ns_store(env, mdd, - is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, - mdd_cobj, mdd_pobj, NULL, lname, handle); + rc = mdd_changelog_ns_store(env, mdd, + is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, + mdd_cobj, mdd_pobj, lname, handle); } stop: mdd_trans_stop(env, mdd, rc, handle); -#ifdef HAVE_QUOTA_SUPPORT - if (quota_opc) - /* Trigger dqrel on the owner of child and parent. If failed, - * the next call for lquota_chkquota will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc, - quota_opc); -#endif + return rc; } -/* has not lock on pobj yet */ -static int mdd_ni_sanity_check(const struct lu_env *env, - struct md_object *pobj, - const struct md_attr *ma) +/* + * The permission has been checked when obj created, no need check again. + */ +static int mdd_cd_sanity_check(const struct lu_env *env, + struct mdd_object *obj) { - struct mdd_object *obj = md2mdd_obj(pobj); - int rc; ENTRY; - if (ma->ma_attr_flags & MDS_PERM_BYPASS) - RETURN(0); + /* EEXIST check */ + if (!obj || mdd_is_dead_obj(obj)) + RETURN(-ENOENT); - rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode)); + RETURN(0); - RETURN(rc); } -/* - * Partial operation. - */ -static int mdd_name_insert(const struct lu_env *env, - struct md_object *pobj, - const struct lu_name *lname, - const struct lu_fid *fid, - const struct md_attr *ma) -{ - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_obj = md2mdd_obj(pobj); - struct mdd_device *mdd = mdo2mdd(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; - int is_dir = S_ISDIR(ma->ma_attr.la_mode); -#ifdef HAVE_QUOTA_SUPPORT - struct md_ucred *uc = md_ucred(env); - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 }; - cfs_cap_t save = uc->mu_cap; -#endif - int rc; - ENTRY; - - /* XXX: this code won't be used ever: - * DNE uses slightly different approach */ - LBUG(); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_obj, &data); - quota_opc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qids); - /* get block quota for parent */ - lquota_chkquota(mds_quota_interface_ref, obd, - exp, qids, rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, data, 1); - } - } else { - uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK; - } - } -#endif - handle = mdd_trans_create(env, mdo2mdd(pobj)); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); - - rc = mdd_trans_start(env, mdo2mdd(pobj), handle); - - dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - - rc = mdd_ni_sanity_check(env, pobj, ma); - if (rc) - GOTO(out_unlock, rc); - - rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir, - handle, BYPASS_CAPA); - if (rc) - GOTO(out_unlock, rc); - - /* - * For some case, no need update obj's ctime (LA_CTIME is not set), - * e.g. split_dir. - * For other cases, update obj's ctime (LA_CTIME is set), - * e.g. cmr_link. - */ - if (ma->ma_attr.la_valid & LA_CTIME) { - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la, - handle, 0); - } - EXIT; -out_unlock: - mdd_pdo_write_unlock(env, mdd_obj, dlh); -out_trans: - mdd_trans_stop(env, mdo2mdd(pobj), rc, handle); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (quota_opc) { - lquota_pending_commit(mds_quota_interface_ref, - obd, qids, rec_pending, 1); - /* Trigger dqacq for the parent owner. If failed, - * the next call for lquota_chkquota will process it*/ - lquota_adjust(mds_quota_interface_ref, obd, 0, qids, - rc, quota_opc); - } else { - uc->mu_cap = save; - } - } -#endif - return rc; -} - -/* has not lock on pobj yet */ -static int mdd_nr_sanity_check(const struct lu_env *env, - struct md_object *pobj, - const struct md_attr *ma) -{ - struct mdd_object *obj = md2mdd_obj(pobj); - int rc; - ENTRY; - - if (ma->ma_attr_flags & MDS_PERM_BYPASS) - RETURN(0); - - rc = mdd_may_unlink(env, obj, ma); - - RETURN(rc); -} - -/* - * Partial operation. - */ -static int mdd_name_remove(const struct lu_env *env, - struct md_object *pobj, - const struct lu_name *lname, - const struct md_attr *ma) -{ - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_obj = md2mdd_obj(pobj); - struct mdd_device *mdd = mdo2mdd(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; - int is_dir = S_ISDIR(ma->ma_attr.la_mode); -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct mds_obd *mds = &obd->u.mds; - unsigned int qids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0; -#endif - int rc; - ENTRY; - - /* XXX: this code won't be used ever: - * DNE uses slightly different approach */ - LBUG(); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA); - if (!rc) { - quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT; - mdd_quota_wrapper(la_tmp, qids); - } - } -#endif - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); - - rc = mdd_trans_start(env, mdd, handle); - - dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - - rc = mdd_nr_sanity_check(env, pobj, ma); - if (rc) - GOTO(out_unlock, rc); - - rc = __mdd_index_delete(env, mdd_obj, name, is_dir, - handle, BYPASS_CAPA); - if (rc) - GOTO(out_unlock, rc); - - /* - * For some case, no need update obj's ctime (LA_CTIME is not set), - * e.g. split_dir. - * For other cases, update obj's ctime (LA_CTIME is set), - * e.g. cmr_unlink. - */ - if (ma->ma_attr.la_valid & LA_CTIME) { - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la, - handle, 0); - } - EXIT; -out_unlock: - mdd_pdo_write_unlock(env, mdd_obj, dlh); -out_trans: - mdd_trans_stop(env, mdd, rc, handle); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - /* Trigger dqrel for the parent owner. - * If failed, the next call for lquota_chkquota will process it. */ - if (quota_opc) - lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc, - quota_opc); -#endif - return rc; -} - -/* - * tobj maybe NULL - * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet - */ -static int mdd_rt_sanity_check(const struct lu_env *env, - struct mdd_object *tgt_pobj, - struct mdd_object *tobj, - struct md_attr *ma) -{ - int rc; - ENTRY; - - if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS)) - RETURN(0); - - /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not - * exist. In fact, tobj must exist, otherwise the call trace will be: - * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert. - * When get here, tobj must be NOT NULL, the other case has been - * processed in cmr_rename_tgt before mdd_rename_tgt and enable - * MDS_PERM_BYPASS. - * So check may_delete, but not check nlink of tgt_pobj. */ - - rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1); - - RETURN(rc); -} - -/* Partial rename op on slave MDD */ -static int mdd_rename_tgt(const struct lu_env *env, - struct md_object *pobj, struct md_object *tobj, - const struct lu_fid *lf, const struct lu_name *lname, - struct md_attr *ma) -{ - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_tpobj = md2mdd_obj(pobj); - struct mdd_object *mdd_tobj = md2mdd_obj(tobj); - struct mdd_device *mdd = mdo2mdd(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qpids[MAXQUOTAS] = { 0, 0 }; - int quota_copc = 0, quota_popc = 0; - int rec_pending[MAXQUOTAS] = { 0, 0 }; -#endif - int rc; - ENTRY; - - /* XXX: this code won't be used ever: - * DNE uses slightly different approach */ - LBUG(); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && !tobj) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_tpobj, &data); - quota_popc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qpids); - /* get block quota for target parent */ - lquota_chkquota(mds_quota_interface_ref, obd, exp, - qpids, rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, data, 1); - } - } -#endif - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); - - rc = mdd_trans_start(env, mdd, handle); - - dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - if (tobj) - mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD); - - rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma); - if (rc) - GOTO(cleanup, rc); - - /* - * If rename_tgt is called then we should just re-insert name with - * correct fid, no need to dec/inc parent nlink if obj is dir. - */ - rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA); - if (rc) - GOTO(cleanup, rc); - - rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle, - BYPASS_CAPA); - if (rc) - GOTO(cleanup, rc); - - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); - - /* - * For tobj is remote case cmm layer has processed - * and pass NULL tobj to here. So when tobj is NOT NULL, - * it must be local one. - */ - if (tobj && mdd_object_exists(mdd_tobj)) { - __mdd_ref_del(env, mdd_tobj, handle, 0); - - /* Remove dot reference. */ - if (S_ISDIR(ma->ma_attr.la_mode)) - __mdd_ref_del(env, mdd_tobj, handle, 1); - - la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); - - rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); - if (rc) - GOTO(cleanup, rc); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && ma->ma_valid & MA_INODE && - ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) { - quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD; - mdd_quota_wrapper(&ma->ma_attr, qcids); - } -#endif - } - EXIT; -cleanup: - if (tobj) - mdd_write_unlock(env, mdd_tobj); - mdd_pdo_write_unlock(env, mdd_tpobj, dlh); -out_trans: - if (rc == 0) - /* Bare EXT record with no RENAME in front of it signifies - a partial slave op */ - rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj, - mdd_tpobj, NULL, lname, handle); - - mdd_trans_stop(env, mdd, rc, handle); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (quota_popc) - lquota_pending_commit(mds_quota_interface_ref, obd, - qpids, rec_pending, 1); - - if (quota_copc) - /* Trigger dqrel on the target owner of child. - * If failed, the next call for lquota_chkquota - * will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, - rc, quota_copc); - } -#endif - return rc; -} - -/* - * The permission has been checked when obj created, no need check again. - */ -static int mdd_cd_sanity_check(const struct lu_env *env, - struct mdd_object *obj) -{ - ENTRY; - - /* EEXIST check */ - if (!obj || mdd_is_dead_obj(obj)) - RETURN(-ENOENT); - - RETURN(0); - -} - -static int mdd_declare_create_data(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *obj, - int lmm_size, - struct thandle *handle) -{ - struct lu_buf *buf = &mdd_env_info(env)->mti_buf; - int rc; - - buf->lb_buf = NULL; - buf->lb_len = lmm_size; - rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV, - 0, handle); - if (rc) - return rc; - - rc = mdd_declare_lov_objid_update(env, mdd, handle); - - return rc; -} - -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); struct mdd_object *son = md2mdd_obj(cobj); - struct lu_attr *attr = &ma->ma_attr; - struct lov_mds_md *lmm = NULL; - int lmm_size = 0; struct thandle *handle; + const struct lu_buf *buf; + struct lu_attr *attr = &mdd_env_info(env)->mti_cattr; int rc; ENTRY; @@ -1574,55 +1581,65 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, if (!md_should_create(spec->sp_cr_flags)) RETURN(0); - lmm_size = ma->ma_lmm_size; - rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, - spec, attr); - if (rc) - RETURN(rc); + + /* + * there are following use cases for this function: + * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE + * striping can be specified or not + * 2) CMD? + */ + rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son)); + if (rc) + RETURN(rc); + + /* calling ->ah_make_hint() is used to transfer information from parent */ + mdd_object_make_hint(env, mdd_pobj, son, attr); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_free, rc = PTR_ERR(handle)); - rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle); - if (rc) - GOTO(stop, rc); - - rc = mdd_trans_start(env, mdd, handle); - if (rc) - GOTO(stop, rc); - /* * 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 %Lo, no_create %u\n", + spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen, + spec->sp_cr_flags, spec->no_create); - /* Replay creates has objects already */ -#if 0 - if (spec->no_create) { - CDEBUG(D_INFO, "we already have lov ea\n"); - rc = mdd_lov_set_md(env, mdd_pobj, son, - (struct lov_mds_md *)spec->u.sp_ea.eadata, - spec->u.sp_ea.eadatalen, handle, 0); - } else -#endif - /* No need mdd_lsm_sanity_check here */ - rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, - lmm_size, handle, 0); + if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) { + /* replay case or lfs setstripe */ + buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, + spec->u.sp_ea.eadatalen); + } else { + buf = &LU_BUF_NULL; + } - if (rc == 0) - rc = mdd_attr_get_internal_locked(env, son, ma); + rc = dt_declare_xattr_set(env, mdd_object_child(son), buf, + XATTR_NAME_LOV, 0, handle); + if (rc) + GOTO(stop, rc); - /* update lov_objid data, must be before transaction stop! */ - if (rc == 0) - mdd_lov_objid_update(mdd, lmm); + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV, + 0, handle, mdd_object_capa(env, son)); + + if (rc) + GOTO(stop, rc); + + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); out_free: - /* Finish mdd_lov_create() stuff. */ - mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec); - RETURN(rc); + RETURN(rc); } /* Get fid from name and parent */ @@ -1641,14 +1658,12 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, if (unlikely(mdd_is_dead_obj(mdd_obj))) RETURN(-ESTALE); - rc = mdd_object_exists(mdd_obj); - if (unlikely(rc == 0)) - RETURN(-ESTALE); - else if (unlikely(rc < 0)) { - CERROR("Object "DFID" locates on remote server\n", - PFID(mdo2fid(mdd_obj))); - RETURN(-EINVAL); - } + 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); + } /* The common filename length check. */ if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)) @@ -1675,30 +1690,49 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, RETURN(rc); } -int mdd_declare_object_initialize(const struct lu_env *env, - struct mdd_object *child, - struct md_attr *ma, - struct thandle *handle) +static int mdd_declare_object_initialize(const struct lu_env *env, + struct mdd_object *parent, + struct mdd_object *child, + struct lu_attr *attr, + struct thandle *handle, + struct linkea_data *ldata) { int rc; - - rc = mdo_declare_attr_set(env, child, &ma->ma_attr, handle); - if (rc == 0 && S_ISDIR(ma->ma_attr.la_mode)) { - rc = mdo_declare_index_insert(env, child, mdo2fid(child), - dot, handle); + 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)) { + rc = mdo_declare_index_insert(env, child, mdo2fid(child), + dot, handle); if (rc == 0) rc = mdo_declare_ref_add(env, child, handle); + + rc = mdo_declare_index_insert(env, child, mdo2fid(parent), + dotdot, handle); } - if (rc == 0) - mdd_declare_links_add(env, child, handle); - return rc; + if (rc == 0) + mdd_declare_links_add(env, child, handle, ldata); + + RETURN(rc); } -int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, - const struct lu_name *lname, struct mdd_object *child, - struct md_attr *ma, struct thandle *handle, - const struct md_op_spec *spec) +static int mdd_object_initialize(const struct lu_env *env, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct mdd_object *child, + struct lu_attr *attr, struct thandle *handle, + const struct md_op_spec *spec, + struct linkea_data *ldata) { int rc; ENTRY; @@ -1711,13 +1745,11 @@ int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, * (2) maybe, the child attributes should be set in OSD when creation. */ - rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0); - if (rc != 0) - RETURN(rc); - - if (S_ISDIR(ma->ma_attr.la_mode)) { + rc = mdd_attr_set_internal(env, child, attr, handle, 0); + /* arguments are supposed to stay the same */ + if (S_ISDIR(attr->la_mode)) { /* Add "." and ".." for newly created dir */ - __mdd_ref_add(env, child, handle); + mdo_ref_add(env, child, handle); rc = __mdd_index_insert_only(env, child, mdo2fid(child), dot, handle, BYPASS_CAPA); if (rc == 0) @@ -1725,27 +1757,27 @@ int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, dotdot, handle, BYPASS_CAPA); if (rc != 0) - __mdd_ref_del(env, child, handle, 1); + mdo_ref_del(env, child, handle); } - if (rc == 0) - mdd_links_add(env, child, pfid, lname, handle, 1); - RETURN(rc); + if (rc == 0) + mdd_links_add(env, child, pfid, lname, handle, ldata, 1); + + RETURN(rc); } /* has not lock on pobj yet */ static int mdd_create_sanity_check(const struct lu_env *env, struct md_object *pobj, + struct lu_attr *pattr, const struct lu_name *lname, - struct md_attr *ma, + struct lu_attr *cattr, struct md_op_spec *spec) { struct mdd_thread_info *info = mdd_env_info(env); - struct lu_attr *la = &info->mti_la; struct lu_fid *fid = &info->mti_fid; struct mdd_object *obj = md2mdd_obj(pobj); struct mdd_device *m = mdo2mdd(pobj); - int lookup = spec->sp_cr_lookup; int rc; ENTRY; @@ -1753,12 +1785,12 @@ static int mdd_create_sanity_check(const struct lu_env *env, if (mdd_is_dead_obj(obj)) RETURN(-ENOENT); - /* + /* * In some cases this lookup is not needed - we know before if name * exists or not because MDT performs lookup for it. * name length check is done in lookup. */ - if (lookup) { + if (spec->sp_cr_lookup) { /* * Check if the name already exist, though it will be checked in * _index_insert also, for avoiding rolling back if exists @@ -1774,26 +1806,22 @@ static int mdd_create_sanity_check(const struct lu_env *env, * EXEC permission have been checked * when lookup before create already. */ - rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE, - MOR_TGT_PARENT); + rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE, + MOR_TGT_PARENT); if (rc) RETURN(rc); } /* sgid check */ - rc = mdd_la_get(env, obj, la, BYPASS_CAPA); - if (rc != 0) - RETURN(rc); - - if (la->la_mode & S_ISGID) { - ma->ma_attr.la_gid = la->la_gid; - if (S_ISDIR(ma->ma_attr.la_mode)) { - ma->ma_attr.la_mode |= S_ISGID; - ma->ma_attr.la_valid |= LA_MODE; - } - } - - switch (ma->ma_attr.la_mode & S_IFMT) { + if (pattr->la_mode & S_ISGID) { + cattr->la_gid = pattr->la_gid; + if (S_ISDIR(cattr->la_mode)) { + cattr->la_mode |= S_ISGID; + cattr->la_valid |= LA_MODE; + } + } + + switch (cattr->la_mode & S_IFMT) { case S_IFLNK: { unsigned int symlen = strlen(spec->u.sp_symname) + 1; @@ -1817,111 +1845,160 @@ static int mdd_create_sanity_check(const struct lu_env *env, RETURN(rc); } -static int mdd_declare_create(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *p, - struct mdd_object *c, - const struct lu_name *name, - struct md_attr *ma, - int lmm_size, - struct thandle *handle, - const struct md_op_spec *spec) +static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, + struct mdd_object *p, struct mdd_object *c, + const struct lu_name *name, + struct lu_attr *attr, + struct thandle *handle, + const struct md_op_spec *spec, + struct linkea_data *ldata, + struct lu_buf *def_acl_buf, + struct lu_buf *acl_buf) { - struct lu_buf *buf = &mdd_env_info(env)->mti_buf; - int rc = 0; - - rc = mdd_declare_object_create_internal(env, p, c, ma, handle, spec); - if (rc) - GOTO(out, rc); - - /* if dir, then can inherit default ACl */ - buf->lb_buf = NULL; - buf->lb_len = lmm_size; - if (S_ISDIR(ma->ma_attr.la_mode)) { - rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_DEFAULT, - 0, handle); - if (rc == 0) - rc = mdo_declare_ref_add(env, p, handle); - } - if (rc) - GOTO(out, rc); + int rc; - rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_ACCESS, - 0, handle); + rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec); if (rc) GOTO(out, rc); - rc = mdd_declare_object_initialize(env, c, ma, handle); - if (rc) - GOTO(out, rc); - - rc = mdo_declare_index_insert(env, p, mdo2fid(c), - name->ln_name, handle); - if (rc) - GOTO(out, rc); - - rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, - 0, handle); - if (rc) - GOTO(out, rc); +#ifdef CONFIG_FS_POSIX_ACL + if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) { + /* if dir, then can inherit default ACl */ + rc = mdo_declare_xattr_set(env, c, def_acl_buf, + XATTR_NAME_ACL_DEFAULT, + 0, handle); + if (rc) + GOTO(out, rc); + } + + if (acl_buf->lb_len > 0) { + rc = mdo_declare_attr_set(env, c, attr, handle); + if (rc) + GOTO(out, rc); + + rc = mdo_declare_xattr_set(env, c, acl_buf, + XATTR_NAME_ACL_ACCESS, 0, handle); + if (rc) + GOTO(out, rc); + } +#endif - if (S_ISLNK(ma->ma_attr.la_mode)) { + if (S_ISDIR(attr->la_mode)) { + rc = mdo_declare_ref_add(env, p, handle); + if (rc) + GOTO(out, rc); + } + + rc = mdd_declare_object_initialize(env, p, c, attr, handle, ldata); + if (rc) + GOTO(out, rc); + + if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) + rc = orph_declare_index_insert(env, c, attr->la_mode, handle); + else + rc = mdo_declare_index_insert(env, p, mdo2fid(c), + name->ln_name, handle); + if (rc) + GOTO(out, rc); + + /* replay case, create LOV EA from client data */ + if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) { + 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, handle); + if (rc) + GOTO(out, rc); + } + + if (S_ISLNK(attr->la_mode)) { rc = dt_declare_record_write(env, mdd_object_child(c), strlen(spec->u.sp_symname), 0, handle); if (rc) GOTO(out, rc); } - rc = mdo_declare_attr_set(env, p, &ma->ma_attr, handle); - if (rc) - return rc; + + if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) { + rc = mdo_declare_attr_set(env, p, attr, handle); + if (rc) + return rc; + } rc = mdd_declare_changelog_store(env, mdd, name, handle); if (rc) return rc; - rc = mdd_declare_lov_objid_update(env, mdd, handle); - out: return rc; } +static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj, + struct lu_attr *la, struct lu_buf *def_acl_buf, + struct lu_buf *acl_buf) +{ + int rc; + ENTRY; + + if (S_ISLNK(la->la_mode)) { + acl_buf->lb_len = 0; + def_acl_buf->lb_len = 0; + RETURN(0); + } + + mdd_read_lock(env, pobj, MOR_TGT_PARENT); + rc = mdo_xattr_get(env, pobj, def_acl_buf, + XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA); + mdd_read_unlock(env, pobj); + if (rc > 0) { + /* If there are default ACL, fix mode/ACL by default ACL */ + def_acl_buf->lb_len = rc; + LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len); + memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc); + acl_buf->lb_len = rc; + rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode); + if (rc < 0) + RETURN(rc); + } else if (rc == -ENODATA || rc == -EOPNOTSUPP) { + /* If there are no default ACL, fix mode by mask */ + struct lu_ucred *uc = lu_ucred(env); + + /* The create triggered by MDT internal events, such as + * LFSCK reset, will not contain valid "uc". */ + if (unlikely(uc != NULL)) + la->la_mode &= ~uc->uc_umask; + rc = 0; + acl_buf->lb_len = 0; + def_acl_buf->lb_len = 0; + } + + RETURN(rc); +} /* * Create object and insert it into namespace. */ -static int mdd_create(const struct lu_env *env, - struct md_object *pobj, - const struct lu_name *lname, - struct md_object *child, - struct md_op_spec *spec, - struct md_attr* ma) +static int mdd_create(const struct lu_env *env, struct md_object *pobj, + const struct lu_name *lname, struct md_object *child, + struct md_op_spec *spec, struct md_attr* ma) { - struct mdd_thread_info *info = mdd_env_info(env); - struct lu_attr *la = &info->mti_la_for_fix; - struct md_attr *ma_acl = &info->mti_ma; - struct mdd_object *mdd_pobj = md2mdd_obj(pobj); - struct mdd_object *son = md2mdd_obj(child); - struct mdd_device *mdd = mdo2mdd(pobj); - struct lu_attr *attr = &ma->ma_attr; - struct lov_mds_md *lmm = NULL; - struct thandle *handle; - struct dynlock_handle *dlh; - const char *name = lname->ln_name; - int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0; - int got_def_acl = 0; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qpids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0, block_count = 0; - int inode_pending[MAXQUOTAS] = { 0, 0 }; - int block_pending[MAXQUOTAS] = { 0, 0 }; - int parent_pending[MAXQUOTAS] = { 0, 0 }; -#endif - ENTRY; + struct mdd_thread_info *info = mdd_env_info(env); + struct lu_attr *la = &info->mti_la_for_fix; + struct mdd_object *mdd_pobj = md2mdd_obj(pobj); + struct mdd_object *son = md2mdd_obj(child); + struct mdd_device *mdd = mdo2mdd(pobj); + struct lu_attr *attr = &ma->ma_attr; + struct thandle *handle; + struct lu_attr *pattr = &info->mti_pattr; + struct lu_buf acl_buf; + struct lu_buf def_acl_buf; + struct linkea_data *ldata = &info->mti_link_data; + struct dynlock_handle *dlh; + const char *name = lname->ln_name; + int rc, created = 0, initialized = 0, inserted = 0; + ENTRY; /* * Two operations have to be performed: @@ -1959,88 +2036,38 @@ static int mdd_create(const struct lu_env *env, * 2. insert (__mdd_index_insert(), lookup again) */ + rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA); + if (rc != 0) + RETURN(rc); + /* Sanity checks before big job. */ - rc = mdd_create_sanity_check(env, pobj, lname, ma, spec); + rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec); if (rc) RETURN(rc); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA); - if (!rc) { - int same = 0; - - quota_opc = FSFILT_OP_CREATE; - mdd_quota_wrapper(&ma->ma_attr, qcids); - mdd_quota_wrapper(la_tmp, qpids); - /* get file quota for child */ - lquota_chkquota(mds_quota_interface_ref, obd, exp, - qcids, inode_pending, 1, NULL, 0, NULL, - 0); - switch (ma->ma_attr.la_mode & S_IFMT) { - case S_IFLNK: - case S_IFDIR: - block_count = 2; - break; - case S_IFREG: - block_count = 1; - break; - } - if (qcids[USRQUOTA] == qpids[USRQUOTA] && - qcids[GRPQUOTA] == qpids[GRPQUOTA]) { - block_count += 1; - same = 1; - } - /* get block quota for child and parent */ - if (block_count) - lquota_chkquota(mds_quota_interface_ref, obd, - exp, qcids, block_pending, - block_count, NULL, - LQUOTA_FLAGS_BLK, NULL, 0); - if (!same) - lquota_chkquota(mds_quota_interface_ref, obd, - exp, qpids, parent_pending, 1, - NULL, LQUOTA_FLAGS_BLK, NULL, - 0); - } - } -#endif + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET)) + GOTO(out_free, rc = -EINPROGRESS); - /* - * No RPC inside the transaction, so OST objects should be created at - * first. - */ - if (S_ISREG(attr->la_mode)) { - lmm_size = ma->ma_lmm_size; - rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, - spec, attr); - if (rc) - GOTO(out_pending, rc); - } + acl_buf.lb_buf = info->mti_xattr_buf; + acl_buf.lb_len = sizeof(info->mti_xattr_buf); + def_acl_buf.lb_buf = info->mti_key; + def_acl_buf.lb_len = sizeof(info->mti_key); + rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf); + if (rc < 0) + GOTO(out_free, rc); - if (!S_ISLNK(attr->la_mode)) { - ma_acl->ma_acl_size = sizeof info->mti_xattr_buf; - ma_acl->ma_acl = info->mti_xattr_buf; - ma_acl->ma_need = MA_ACL_DEF; - ma_acl->ma_valid = 0; - - mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT); - rc = mdd_def_acl_get(env, mdd_pobj, ma_acl); - mdd_read_unlock(env, mdd_pobj); - if (rc) - GOTO(out_free, rc); - else if (ma_acl->ma_valid & MA_ACL_DEF) - got_def_acl = 1; - } + mdd_object_make_hint(env, mdd_pobj, son, attr); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_free, rc = PTR_ERR(handle)); - rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, ma, - lmm_size, handle, spec); + memset(ldata, 0, sizeof(*ldata)); + mdd_linkea_prepare(env, son, NULL, NULL, mdd_object_fid(mdd_pobj), + lname, 1, 0, ldata); + + rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr, + handle, spec, ldata, &def_acl_buf, &acl_buf); if (rc) GOTO(out_stop, rc); @@ -2048,71 +2075,87 @@ static int mdd_create(const struct lu_env *env, if (rc) GOTO(out_stop, rc); - dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); + dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); + if (dlh == NULL) + GOTO(out_trans, rc = -ENOMEM); - mdd_write_lock(env, son, MOR_TGT_CHILD); - rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec); - if (rc) { - mdd_write_unlock(env, son); - GOTO(cleanup, rc); - } + mdd_write_lock(env, son, MOR_TGT_CHILD); + rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec); + if (rc) { + mdd_write_unlock(env, son); + GOTO(cleanup, rc); + } - created = 1; + created = 1; #ifdef CONFIG_FS_POSIX_ACL - if (got_def_acl) { - struct lu_buf *acl_buf = &info->mti_buf; - acl_buf->lb_buf = ma_acl->ma_acl; - acl_buf->lb_len = ma_acl->ma_acl_size; - - rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle); - if (rc) { - mdd_write_unlock(env, son); - GOTO(cleanup, rc); - } else { - ma->ma_attr.la_valid |= LA_MODE; - } - } + if (def_acl_buf.lb_len > 0 && S_ISDIR(attr->la_mode)) { + /* set default acl */ + rc = mdo_xattr_set(env, son, &def_acl_buf, + XATTR_NAME_ACL_DEFAULT, 0, + handle, BYPASS_CAPA); + if (rc) { + mdd_write_unlock(env, son); + GOTO(cleanup, rc); + } + } + /* set its own acl */ + if (acl_buf.lb_len > 0) { + rc = mdo_xattr_set(env, son, &acl_buf, + XATTR_NAME_ACL_ACCESS, + 0, handle, BYPASS_CAPA); + if (rc) { + mdd_write_unlock(env, son); + GOTO(cleanup, rc); + } + } #endif - rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname, - son, ma, handle, spec); - mdd_write_unlock(env, son); - if (rc) - /* - * Object has no links, so it will be destroyed when last - * reference is released. (XXX not now.) - */ - GOTO(cleanup, rc); + rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname, + son, attr, handle, spec, ldata); - initialized = 1; + /* + * in case of replay we just set LOVEA provided by the client + * XXX: I think it would be interesting to try "old" way where + * MDT calls this xattr_set(LOV) in a different transaction. + * probably this way we code can be made better. + */ + if (rc == 0 && (spec->no_create || + (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) { + const struct lu_buf *buf; - rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son), - name, S_ISDIR(attr->la_mode), handle, - mdd_object_capa(env, mdd_pobj)); + buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, + spec->u.sp_ea.eadatalen); + rc = mdo_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle, + BYPASS_CAPA); + } - if (rc) - GOTO(cleanup, rc); + if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE) + rc = __mdd_orphan_add(env, son, handle); - inserted = 1; + mdd_write_unlock(env, son); - /* No need mdd_lsm_sanity_check here */ - rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0); - if (rc) { - CERROR("error on stripe info copy %d \n", rc); - GOTO(cleanup, rc); - } - if (lmm && lmm_size > 0) { - /* Set Lov here, do not get lmm again later */ - memcpy(ma->ma_lmm, lmm, lmm_size); - ma->ma_lmm_size = lmm_size; - ma->ma_valid |= MA_LOV; - } + if (rc != 0) + /* + * Object has no links, so it will be destroyed when last + * reference is released. (XXX not now.) + */ + GOTO(cleanup, rc); + + initialized = 1; + + if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) + rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son), + name, S_ISDIR(attr->la_mode), handle, + mdd_object_capa(env, mdd_pobj)); + + if (rc != 0) + GOTO(cleanup, rc); + + inserted = 1; if (S_ISLNK(attr->la_mode)) { - struct md_ucred *uc = md_ucred(env); + struct lu_ucred *uc = lu_ucred_assert(env); struct dt_object *dt = mdd_object_child(son); const char *target_name = spec->u.sp_symname; int sym_len = strlen(target_name); @@ -2120,10 +2163,10 @@ static int mdd_create(const struct lu_env *env, loff_t pos = 0; buf = mdd_buf_get_const(env, target_name, sym_len); - rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle, - mdd_object_capa(env, son), - uc->mu_cap & - CFS_CAP_SYS_RESOURCE_MASK); + rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle, + mdd_object_capa(env, son), + uc->uc_cap & + CFS_CAP_SYS_RESOURCE_MASK); if (rc == sym_len) rc = 0; @@ -2131,69 +2174,73 @@ static int mdd_create(const struct lu_env *env, GOTO(cleanup, rc = -EFAULT); } - *la = ma->ma_attr; - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); + /* volatile file creation does not update parent directory times */ + if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) + GOTO(cleanup, rc = 0); + + /* update parent directory mtime/ctime */ + *la = *attr; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0); + if (rc) + GOTO(cleanup, rc); - /* Return attr back. */ - rc = mdd_attr_get_internal_locked(env, son, ma); EXIT; cleanup: - if (rc && created) { - int rc2 = 0; - - if (inserted) { - rc2 = __mdd_index_delete(env, mdd_pobj, name, - S_ISDIR(attr->la_mode), - handle, BYPASS_CAPA); - if (rc2) - CERROR("error can not cleanup destroy %d\n", - rc2); - } - - if (rc2 == 0) { - mdd_write_lock(env, son, MOR_TGT_CHILD); - __mdd_ref_del(env, son, handle, 0); - if (initialized && S_ISDIR(attr->la_mode)) - __mdd_ref_del(env, son, handle, 1); - mdd_write_unlock(env, son); - } + if (rc != 0 && created != 0) { + int rc2; + + if (inserted != 0) { + if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) + rc2 = __mdd_orphan_del(env, son, handle); + else + rc2 = __mdd_index_delete(env, mdd_pobj, name, + S_ISDIR(attr->la_mode), + handle, BYPASS_CAPA); + if (rc2 != 0) + goto out_stop; + } + + mdd_write_lock(env, son, MOR_TGT_CHILD); + if (initialized != 0 && S_ISDIR(attr->la_mode)) { + /* Drop the reference, no need to delete "."/"..", + * because the object to be destroied directly. */ + rc2 = mdo_ref_del(env, son, handle); + if (rc2 != 0) { + mdd_write_unlock(env, son); + goto out_stop; + } + } + + rc2 = mdo_ref_del(env, son, handle); + if (rc2 != 0) { + mdd_write_unlock(env, son); + goto out_stop; + } + + mdo_destroy(env, son, handle); + mdd_write_unlock(env, son); } - /* update lov_objid data, must be before transaction stop! */ - if (rc == 0) - mdd_lov_objid_update(mdd, lmm); - mdd_pdo_write_unlock(env, mdd_pobj, dlh); out_trans: - if (rc == 0) - rc = mdd_changelog_ns_store(env, mdd, - S_ISDIR(attr->la_mode) ? CL_MKDIR : - S_ISREG(attr->la_mode) ? CL_CREATE : - S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD, - 0, son, mdd_pobj, NULL, lname, handle); + if (rc == 0 && fid_is_namespace_visible(mdo2fid(son))) + rc = mdd_changelog_ns_store(env, mdd, + S_ISDIR(attr->la_mode) ? CL_MKDIR : + S_ISREG(attr->la_mode) ? CL_CREATE : + S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD, + 0, son, mdd_pobj, lname, handle); out_stop: mdd_trans_stop(env, mdd, rc, handle); out_free: - /* finish lov_create stuff, free all temporary data */ - mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (quota_opc) { - lquota_pending_commit(mds_quota_interface_ref, obd, qcids, - inode_pending, 0); - lquota_pending_commit(mds_quota_interface_ref, obd, qcids, - block_pending, 1); - lquota_pending_commit(mds_quota_interface_ref, obd, qpids, - parent_pending, 1); - /* Trigger dqacq on the owner of child and parent. If failed, - * the next call for lquota_chkquota will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc, - quota_opc); - } -#endif + if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + lu_buf_free(ldata->ld_buf); + + /* The child object shouldn't be cached anymore */ + if (rc) + set_bit(LU_OBJECT_HEARD_BANSHEE, + &child->mo_lu.lo_header->loh_flags); return rc; } @@ -2244,54 +2291,57 @@ static int mdd_rename_sanity_check(const struct lu_env *env, struct mdd_object *tgt_pobj, struct mdd_object *sobj, struct mdd_object *tobj, - struct md_attr *ma) + struct lu_attr *so_attr, + struct lu_attr *tg_attr) { - int rc = 0; - ENTRY; - - if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS)) - RETURN(0); - - /* XXX: when get here, sobj must NOT be NULL, - * the other case has been processed in cml_rename - * before mdd_rename and enable MDS_PERM_BYPASS. */ - LASSERT(sobj); - - rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0); - if (rc) - RETURN(rc); - - /* XXX: when get here, "tobj == NULL" means tobj must - * NOT exist (neither on remote MDS, such case has been - * processed in cml_rename before mdd_rename and enable - * MDS_PERM_BYPASS). - * So check may_create, but not check may_unlink. */ - if (!tobj) - rc = mdd_may_create(env, tgt_pobj, NULL, - (src_pobj != tgt_pobj), 0); - else - rc = mdd_may_delete(env, tgt_pobj, tobj, ma, - (src_pobj != tgt_pobj), 1); - - if (!rc && !tobj && (src_pobj != tgt_pobj) && - S_ISDIR(ma->ma_attr.la_mode)) - rc = __mdd_may_link(env, tgt_pobj); - - RETURN(rc); + int rc = 0; + ENTRY; + + /* XXX: when get here, sobj must NOT be NULL, + * the other case has been processed in cld_rename + * before mdd_rename and enable MDS_PERM_BYPASS. */ + LASSERT(sobj); + + rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0); + if (rc) + RETURN(rc); + + /* XXX: when get here, "tobj == NULL" means tobj must + * NOT exist (neither on remote MDS, such case has been + * processed in cld_rename before mdd_rename and enable + * MDS_PERM_BYPASS). + * So check may_create, but not check may_unlink. */ + if (!tobj) + rc = mdd_may_create(env, tgt_pobj, NULL, + (src_pobj != tgt_pobj), 0); + else + rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr, + (src_pobj != tgt_pobj), 1); + + if (!rc && !tobj && (src_pobj != tgt_pobj) && + S_ISDIR(so_attr->la_mode)) + rc = __mdd_may_link(env, tgt_pobj); + + RETURN(rc); } static int mdd_declare_rename(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *mdd_spobj, - struct mdd_object *mdd_tpobj, - struct mdd_object *mdd_sobj, - struct mdd_object *mdd_tobj, - const struct lu_name *sname, - const struct lu_name *tname, - struct md_attr *ma, - struct thandle *handle) + struct mdd_device *mdd, + struct mdd_object *mdd_spobj, + 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, + struct md_attr *ma, + struct linkea_data *ldata, + struct thandle *handle) { - int rc; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + int rc; + + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; LASSERT(mdd_spobj); LASSERT(mdd_tpobj); @@ -2309,16 +2359,18 @@ static int mdd_declare_rename(const struct lu_env *env, rc = mdo_declare_ref_del(env, mdd_spobj, handle); if (rc) return rc; - - rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle); - if (rc) - return rc; - - rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj), - dotdot, handle); - if (rc) - return rc; - + if (mdd_spobj != mdd_tpobj) { + rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, + handle); + if (rc) + return rc; + + rc = mdo_declare_index_insert(env, mdd_sobj, + mdo2fid(mdd_tpobj), + dotdot, handle); + if (rc) + return rc; + } /* new target child can be directory, * counted by target dir's nlink */ rc = mdo_declare_ref_add(env, mdd_tpobj, handle); @@ -2327,20 +2379,23 @@ static int mdd_declare_rename(const struct lu_env *env, } - rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, mdd_spobj, la, handle); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle); - if (rc) - return rc; - mdd_declare_links_add(env, mdd_sobj, handle); - if (rc) - return rc; + rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, mdd_sobj, la, handle); + if (rc) + return rc; + + rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata); + if (rc) + return rc; /* new name */ rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj), @@ -2374,11 +2429,12 @@ static int mdd_declare_rename(const struct lu_env *env, return rc; } - rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle); + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, mdd_tobj, la, handle); if (rc) return rc; - mdd_declare_links_add(env, mdd_tobj, handle); + mdd_declare_links_del(env, mdd_tobj, handle); if (rc) return rc; @@ -2387,11 +2443,7 @@ static int mdd_declare_rename(const struct lu_env *env, return rc; } - rc = mdd_declare_changelog_store(env, mdd, tname, handle); - if (rc) - return rc; - - rc = mdd_declare_changelog_store(env, mdd, sname, handle); + rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle); if (rc) return rc; @@ -2405,75 +2457,61 @@ static int mdd_rename(const struct lu_env *env, struct md_object *tobj, const struct lu_name *ltname, struct md_attr *ma) { - const char *sname = lsname->ln_name; - const char *tname = ltname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */ - struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj); - struct mdd_device *mdd = mdo2mdd(src_pobj); - struct mdd_object *mdd_sobj = NULL; /* source object */ - struct mdd_object *mdd_tobj = NULL; - struct dynlock_handle *sdlh, *tdlh; - struct thandle *handle; - const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj); - const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj); - int is_dir; - int rc, rc2; - -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qspids[MAXQUOTAS] = { 0, 0 }; - unsigned int qtcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qtpids[MAXQUOTAS] = { 0, 0 }; - int quota_copc = 0, quota_popc = 0; - int rec_pending[MAXQUOTAS] = { 0, 0 }; -#endif - ENTRY; - - LASSERT(ma->ma_attr.la_mode & S_IFMT); - is_dir = S_ISDIR(ma->ma_attr.la_mode); + const char *sname = lsname->ln_name; + const char *tname = ltname->ln_name; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr; + struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr; + struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */ + struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj); + struct mdd_device *mdd = mdo2mdd(src_pobj); + struct mdd_object *mdd_sobj = NULL; /* source object */ + struct mdd_object *mdd_tobj = NULL; + struct dynlock_handle *sdlh = NULL, *tdlh = NULL; + struct thandle *handle; + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; + const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj); + const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj); + bool is_dir; + bool tobj_ref = 0; + bool tobj_locked = 0; + unsigned cl_flags = 0; + int rc, rc2; + ENTRY; if (tobj) mdd_tobj = md2mdd_obj(tobj); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA); - if (!rc) { - mdd_quota_wrapper(la_tmp, qspids); - if (!tobj) { - rc = mdd_la_get(env, mdd_tpobj, la_tmp, - BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_tpobj, &data); - quota_popc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qtpids); - /* get block quota for target parent */ - lquota_chkquota(mds_quota_interface_ref, - obd, exp, qtpids, - rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, - data, 1); - } - } - } - } -#endif mdd_sobj = mdd_object_find(env, mdd, lf); + rc = mdd_la_get(env, mdd_sobj, so_attr, + mdd_object_capa(env, mdd_sobj)); + if (rc) + GOTO(out_pending, rc); + + if (mdd_tobj) { + rc = mdd_la_get(env, mdd_tobj, tg_attr, + mdd_object_capa(env, mdd_tobj)); + if (rc) + GOTO(out_pending, rc); + } + + rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj, + mdd_tobj, so_attr, tg_attr); + if (rc) + GOTO(out_pending, rc); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); - rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, - mdd_tobj, lsname, ltname, ma, handle); - if (rc) - GOTO(stop, rc); + memset(ldata, 0, sizeof(*ldata)); + mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj), + ltname, 1, 0, ldata); + rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, + mdd_tobj, lsname, ltname, ma, ldata, handle); + if (rc) + GOTO(stop, rc); rc = mdd_trans_start(env, mdd, handle); if (rc) @@ -2504,10 +2542,7 @@ static int mdd_rename(const struct lu_env *env, if (sdlh == NULL || tdlh == NULL) GOTO(cleanup, rc = -ENOMEM); - rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, - mdd_sobj, mdd_tobj, ma); - if (rc) - GOTO(cleanup, rc); + is_dir = S_ISDIR(so_attr->la_mode); /* Remove source name from source directory */ rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle, @@ -2555,8 +2590,7 @@ static int mdd_rename(const struct lu_env *env, /* XXX: mdd_sobj must be local one if it is NOT NULL. */ if (mdd_sobj) { la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la, - handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0); if (rc) GOTO(fixup_tpobj, rc); } @@ -2568,57 +2602,91 @@ static int mdd_rename(const struct lu_env *env, */ if (tobj && mdd_object_exists(mdd_tobj)) { mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD); + tobj_locked = 1; if (mdd_is_dead_obj(mdd_tobj)) { - mdd_write_unlock(env, mdd_tobj); /* shld not be dead, something is wrong */ CERROR("tobj is dead, something is wrong\n"); rc = -EINVAL; goto cleanup; } - __mdd_ref_del(env, mdd_tobj, handle, 0); + mdo_ref_del(env, mdd_tobj, handle); /* Remove dot reference. */ - if (is_dir) - __mdd_ref_del(env, mdd_tobj, handle, 1); - - la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); - if (rc) - GOTO(fixup_tpobj, rc); - - rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); - mdd_write_unlock(env, mdd_tobj); - if (rc) - GOTO(fixup_tpobj, rc); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && ma->ma_valid & MA_INODE && - ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) { - quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD; - mdd_quota_wrapper(&ma->ma_attr, qtcids); - } -#endif + if (S_ISDIR(tg_attr->la_mode)) + mdo_ref_del(env, mdd_tobj, handle); + tobj_ref = 1; + + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_tobj, tg_attr, + mdd_object_capa(env, mdd_tobj)); + if (rc != 0) { + CERROR("%s: Failed to get nlink for tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + + la->la_valid = LA_CTIME; + rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); + if (rc != 0) { + CERROR("%s: Failed to set ctime for tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + + /* XXX: this transfer to ma will be removed with LOD/OSP */ + ma->ma_attr = *tg_attr; + ma->ma_valid |= MA_INODE; + rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); + if (rc != 0) { + CERROR("%s: Failed to unlink tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_tobj, tg_attr, + mdd_object_capa(env, mdd_tobj)); + if (rc != 0) { + CERROR("%s: Failed to get nlink for tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + /* XXX: this transfer to ma will be removed with LOD/OSP */ + ma->ma_attr = *tg_attr; + ma->ma_valid |= MA_INODE; + + if (so_attr->la_nlink == 0) + cl_flags |= CLF_RENAME_LAST; } la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0); if (rc) GOTO(fixup_tpobj, rc); if (mdd_spobj != mdd_tpobj) { la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, - handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_tpobj, la, + handle, 0); } - if (rc == 0 && mdd_sobj) { - mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); - rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname, - mdo2fid(mdd_tpobj), ltname, handle); + if (rc == 0 && mdd_sobj) { + mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); + rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname, + mdo2fid(mdd_tpobj), ltname, handle, NULL, + 0, 0); if (rc == -ENOENT) /* Old files might not have EA entry */ mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj), - lsname, handle, 0); + lsname, handle, NULL, 0); mdd_write_unlock(env, mdd_sobj); /* We don't fail the transaction if the link ea can't be updated -- fid2path will use alternate lookup method. */ @@ -2636,6 +2704,12 @@ fixup_tpobj: if (mdd_tobj && mdd_object_exists(mdd_tobj) && !mdd_is_dead_obj(mdd_tobj)) { + if (tobj_ref) { + mdo_ref_add(env, mdd_tobj, handle); + if (is_dir) + mdo_ref_add(env, mdd_tobj, handle); + } + rc2 = __mdd_index_insert(env, mdd_tpobj, mdo2fid(mdd_tobj), tname, is_dir, handle, @@ -2647,19 +2721,21 @@ fixup_tpobj: } fixup_spobj: - if (rc && is_dir && mdd_sobj) { - rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, - BYPASS_CAPA); + if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) { + rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, + BYPASS_CAPA); - if (rc2) - CWARN("sp obj dotdot delete error %d\n",rc2); + if (rc2) + CWARN("%s: sp obj dotdot delete error: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc2); - rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, - dotdot, handle, BYPASS_CAPA); - if (rc2) - CWARN("sp obj dotdot insert error %d\n",rc2); - } + rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, + dotdot, handle, BYPASS_CAPA); + if (rc2) + CWARN("%s: sp obj dotdot insert error: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc2); + } fixup_spobj2: if (rc) { @@ -2669,332 +2745,24 @@ fixup_spobj2: CWARN("sp obj fix error %d\n",rc2); } cleanup: + if (tobj_locked) + mdd_write_unlock(env, mdd_tobj); if (likely(tdlh) && sdlh != tdlh) mdd_pdo_write_unlock(env, mdd_tpobj, tdlh); if (likely(sdlh)) mdd_pdo_write_unlock(env, mdd_spobj, sdlh); cleanup_unlocked: if (rc == 0) - rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj, - mdd_spobj, lf, lsname, handle); - if (rc == 0) { - struct lu_fid zero_fid; - fid_zero(&zero_fid); - /* If the rename target exist, The CL_EXT record should save - * the target fid as tfid, otherwise, use zero fid. LU-543 */ - rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj, - mdd_tpobj, - mdd_tobj ? NULL : &zero_fid, - ltname, handle); - } + rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags, + mdd_tobj, tpobj_fid, lf, + spobj_fid, ltname, lsname, + handle); stop: mdd_trans_stop(env, mdd, rc, handle); - if (mdd_sobj) - mdd_object_put(env, mdd_sobj); out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (quota_popc) - lquota_pending_commit(mds_quota_interface_ref, obd, - qtpids, rec_pending, 1); - - if (quota_copc) { - /* Trigger dqrel on the source owner of parent. - * If failed, the next call for lquota_chkquota will - * process it. */ - lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc, - FSFILT_OP_UNLINK_PARTIAL_PARENT); - - /* Trigger dqrel on the target owner of child. - * If failed, the next call for lquota_chkquota - * will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qtcids, - qtpids, rc, quota_copc); - } - } -#endif - return rc; -} - -/** enable/disable storing of hardlink info */ -int mdd_linkea_enable = 1; -CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644, - "record hardlink info in EAs"); - -/** 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) - */ -struct lu_buf *mdd_links_get(const struct lu_env *env, - struct mdd_object *mdd_obj) -{ - struct lu_buf *buf; - struct lustre_capa *capa; - struct link_ea_header *leh; - int rc; - - /* First try a small buf */ - buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); - if (buf->lb_buf == NULL) - return ERR_PTR(-ENOMEM); - - capa = mdd_object_capa(env, mdd_obj); - rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa); - if (rc == -ERANGE) { - /* Buf was too small, figure out what we need. */ - mdd_buf_put(buf); - rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa); - if (rc < 0) - return ERR_PTR(rc); - buf = mdd_buf_alloc(env, rc); - if (buf->lb_buf == NULL) - return ERR_PTR(-ENOMEM); - rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa); - } - if (rc < 0) - return ERR_PTR(rc); - - leh = buf->lb_buf; - if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) { - leh->leh_magic = LINK_EA_MAGIC; - leh->leh_reccount = __swab32(leh->leh_reccount); - leh->leh_len = __swab64(leh->leh_len); - /* entries are swabbed by mdd_lee_unpack */ - } - if (leh->leh_magic != LINK_EA_MAGIC) - return ERR_PTR(-EINVAL); - if (leh->leh_reccount == 0) - return ERR_PTR(-ENODATA); - - return buf; -} - -/** Pack a link_ea_entry. - * All elements are stored as chars to avoid alignment issues. - * Numbers are always big-endian - * \retval record length - */ -static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname, - const struct lu_fid *pfid) -{ - struct lu_fid tmpfid; - int reclen; - - fid_cpu_to_be(&tmpfid, pfid); - memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid)); - memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen); - reclen = sizeof(struct link_ea_entry) + lname->ln_namelen; - - lee->lee_reclen[0] = (reclen >> 8) & 0xff; - lee->lee_reclen[1] = reclen & 0xff; - return reclen; -} - -void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen, - struct lu_name *lname, struct lu_fid *pfid) -{ - *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1]; - memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid)); - fid_be_to_cpu(pfid, pfid); - lname->ln_name = lee->lee_name; - lname->ln_namelen = *reclen - sizeof(struct link_ea_entry); -} - -/** Add a record to the end of link ea buf */ -static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf, - const struct lu_fid *pfid, - const struct lu_name *lname) -{ - struct link_ea_header *leh; - struct link_ea_entry *lee; - int reclen; - - if (lname == NULL || pfid == NULL) - return -EINVAL; - - /* Make sure our buf is big enough for the new one */ - leh = buf->lb_buf; - reclen = lname->ln_namelen + sizeof(struct link_ea_entry); - if (leh->leh_len + reclen > buf->lb_len) { - if (mdd_buf_grow(env, leh->leh_len + reclen) < 0) - return -ENOMEM; - } - - leh = buf->lb_buf; - lee = buf->lb_buf + leh->leh_len; - reclen = mdd_lee_pack(lee, lname, pfid); - leh->leh_len += reclen; - leh->leh_reccount++; - return 0; -} - -static int mdd_declare_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct thandle *handle) -{ - int rc; - - /* XXX: max size? */ - rc = mdo_declare_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, NULL, 4096), - XATTR_NAME_LINK, 0, handle); - - return rc; -} - -/* For pathologic linkers, we don't want to spend lots of time scanning the - * link ea. Limit ourseleves to something reasonable; links not in the EA - * can be looked up via (slower) parent lookup. - */ -#define LINKEA_MAX_COUNT 128 - -static int mdd_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *pfid, - const struct lu_name *lname, - struct thandle *handle, int first) -{ - struct lu_buf *buf; - struct link_ea_header *leh; - int rc; - ENTRY; - - if (!mdd_linkea_enable) - RETURN(0); - - buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj); - if (IS_ERR(buf)) { - rc = PTR_ERR(buf); - if (rc != -ENODATA) { - CERROR("link_ea read failed %d "DFID"\n", rc, - PFID(mdd_object_fid(mdd_obj))); - RETURN (rc); - } - /* empty EA; start one */ - buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); - leh = buf->lb_buf; - leh->leh_magic = LINK_EA_MAGIC; - leh->leh_len = sizeof(struct link_ea_header); - leh->leh_reccount = 0; - } - - leh = buf->lb_buf; - if (leh->leh_reccount > LINKEA_MAX_COUNT) - RETURN(-EOVERFLOW); - - rc = __mdd_links_add(env, buf, pfid, lname); - if (rc) - RETURN(rc); - - leh = buf->lb_buf; - rc = __mdd_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, buf->lb_buf, leh->leh_len), - XATTR_NAME_LINK, 0, handle); - if (rc) { - if (rc == -ENOSPC) - CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc, - PFID(mdd_object_fid(mdd_obj))); - else - CERROR("link_ea add failed %d "DFID"\n", rc, - PFID(mdd_object_fid(mdd_obj))); - } - - if (buf->lb_len > OBD_ALLOC_BIG) - /* if we vmalloced a large buffer drop it */ - mdd_buf_put(buf); - - RETURN (rc); -} - -static int mdd_links_rename(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *oldpfid, - const struct lu_name *oldlname, - const struct lu_fid *newpfid, - const struct lu_name *newlname, - struct thandle *handle) -{ - struct lu_buf *buf; - struct link_ea_header *leh; - struct link_ea_entry *lee; - struct lu_name *tmpname = &mdd_env_info(env)->mti_name; - struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid; - int reclen = 0; - int count; - int rc, rc2 = 0; - ENTRY; - - if (!mdd_linkea_enable) - RETURN(0); - - if (mdd_obj->mod_flags & DEAD_OBJ) - /* No more links, don't bother */ - RETURN(0); - - buf = mdd_links_get(env, mdd_obj); - if (IS_ERR(buf)) { - rc = PTR_ERR(buf); - if (rc == -ENODATA) - CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n", - rc, PFID(mdd_object_fid(mdd_obj))); - else - CERROR("link_ea read failed %d "DFID"\n", - rc, PFID(mdd_object_fid(mdd_obj))); - RETURN(rc); - } - leh = buf->lb_buf; - lee = (struct link_ea_entry *)(leh + 1); /* link #0 */ - - /* Find the old record */ - for(count = 0; count < leh->leh_reccount; count++) { - mdd_lee_unpack(lee, &reclen, tmpname, tmpfid); - if (tmpname->ln_namelen == oldlname->ln_namelen && - lu_fid_eq(tmpfid, oldpfid) && - (strncmp(tmpname->ln_name, oldlname->ln_name, - tmpname->ln_namelen) == 0)) - break; - lee = (struct link_ea_entry *)((char *)lee + reclen); - } - if ((count + 1) > leh->leh_reccount) { - CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n", - oldlname->ln_namelen, oldlname->ln_name); - GOTO(out, rc = -ENOENT); - } - - /* Remove the old record */ - leh->leh_reccount--; - leh->leh_len -= reclen; - memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len - - (char *)lee); - - /* If renaming, add the new record */ - if (newpfid != NULL) { - /* if the add fails, we still delete the out-of-date old link */ - rc2 = __mdd_links_add(env, buf, newpfid, newlname); - leh = buf->lb_buf; - } - - rc = __mdd_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, buf->lb_buf, leh->leh_len), - XATTR_NAME_LINK, 0, handle); - -out: - if (rc == 0) - rc = rc2; - if (rc) - CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n", - oldlname->ln_namelen, oldlname->ln_name, rc, - PFID(mdd_object_fid(mdd_obj))); - - if (buf->lb_len > OBD_ALLOC_BIG) - /* if we vmalloced a large buffer drop it */ - mdd_buf_put(buf); - - RETURN (rc); + mdd_object_put(env, mdd_sobj); + return rc; } const struct md_dir_operations mdd_dir_ops = { @@ -3004,8 +2772,5 @@ const struct md_dir_operations mdd_dir_ops = { .mdo_rename = mdd_rename, .mdo_link = mdd_link, .mdo_unlink = mdd_unlink, - .mdo_name_insert = mdd_name_insert, - .mdo_name_remove = mdd_name_remove, - .mdo_rename_tgt = mdd_rename_tgt, .mdo_create_data = mdd_create_data, };