X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=2618a7e7b87b1c352a1cd90e8dc3de52bdbb732e;hb=2deb4f149f4601f9128fd39efd4705573520f277;hp=c47a0b53b07cadac6891523a3bf0c7777eb98dd6;hpb=69fb67289f65d6ccf1e66837323daaed72bea928;p=fs%2Flustre-release.git diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c index c47a0b5..2618a7e 100644 --- a/lustre/mdd/mdd_device.c +++ b/lustre/mdd/mdd_device.c @@ -46,7 +46,11 @@ #define DEBUG_SUBSYSTEM S_MDS #include +#ifdef HAVE_EXT4_LDISKFS +#include +#else #include +#endif #include #include #include @@ -55,7 +59,11 @@ #include #include +#ifdef HAVE_EXT4_LDISKFS +#include +#else #include +#endif #include #include #include /* for changelogs */ @@ -84,8 +92,11 @@ static int mdd_device_init(const struct lu_env *env, struct lu_device *d, mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb; mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb; mdd->mdd_txn_cb.dtc_cookie = mdd; + mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD; CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage); mdd->mdd_atime_diff = MAX_ATIME_DIFF; + /* sync permission changes */ + mdd->mdd_sync_permission = 1; rc = mdd_procfs_init(mdd, name); RETURN(rc); @@ -115,8 +126,10 @@ static void mdd_device_shutdown(const struct lu_env *env, ENTRY; mdd_changelog_fini(env, m); dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb); - mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf); - mdd_object_put(env, m->mdd_dot_lustre); + if (m->mdd_dot_lustre_objs.mdd_obf) + mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf); + if (m->mdd_dot_lustre) + mdd_object_put(env, m->mdd_dot_lustre); if (m->mdd_obd_dev) mdd_fini_obd(env, m, cfg); orph_index_fini(env, m); @@ -132,49 +145,92 @@ static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr, struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; ENTRY; - if (!(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)) { - CERROR("log is not plain\n"); - RETURN(-EINVAL); - } - if (rec->cr_hdr.lrh_type != CHANGELOG_REC) { - CERROR("Not a changelog rec? %d\n", rec->cr_hdr.lrh_type); - RETURN(-EINVAL); - } + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC); - CDEBUG(D_INODE, + CDEBUG(D_INFO, "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n", - hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index, - rec->cr_type, rec->cr_namelen, rec->cr_name, + hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index, + rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name, llh->lgh_id.lgl_oid); - mdd->mdd_cl.mc_index = rec->cr_index; + mdd->mdd_cl.mc_index = rec->cr.cr_index; RETURN(LLOG_PROC_BREAK); } +static int changelog_user_init_cb(struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) +{ + struct mdd_device *mdd = (struct mdd_device *)data; + struct llog_changelog_user_rec *rec = + (struct llog_changelog_user_rec *)hdr; + ENTRY; + + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC); + + CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64 + " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index, + rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid); + + cfs_spin_lock(&mdd->mdd_cl.mc_user_lock); + mdd->mdd_cl.mc_lastuser = rec->cur_id; + cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); + + RETURN(LLOG_PROC_BREAK); +} + + static int mdd_changelog_llog_init(struct mdd_device *mdd) { struct obd_device *obd = mdd2obd_dev(mdd); struct llog_ctxt *ctxt; int rc; + /* Find last changelog entry number */ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); if (ctxt == NULL) { - CERROR("no context\n"); + CERROR("no changelog context\n"); return -EINVAL; } if (!ctxt->loc_handle) { - CERROR("no handle\n"); + llog_ctxt_put(ctxt); return -EINVAL; } + rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd); llog_ctxt_put(ctxt); - if (rc < 0) + if (rc < 0) { CERROR("changelog init failed: %d\n", rc); - else - rc = 0; /* llog_proc_break is ok */ + return rc; + } + CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n", + mdd->mdd_cl.mc_index); + + /* Find last changelog user id */ + ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT); + if (ctxt == NULL) { + CERROR("no changelog user context\n"); + return -EINVAL; + } + if (!ctxt->loc_handle) { + llog_ctxt_put(ctxt); + return -EINVAL; + } + + rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb, + mdd); + llog_ctxt_put(ctxt); + + if (rc < 0) { + CERROR("changelog user init failed: %d\n", rc); + return rc; + } - CDEBUG(D_INODE, "changelog_init index="LPU64"\n", mdd->mdd_cl.mc_index); + /* If we have registered users, assume we want changelogs on */ + if (mdd->mdd_cl.mc_lastuser > 0) + rc = mdd_changelog_on(mdd, 1); return rc; } @@ -184,17 +240,20 @@ static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) int rc; mdd->mdd_cl.mc_index = 0; - spin_lock_init(&mdd->mdd_cl.mc_lock); + cfs_spin_lock_init(&mdd->mdd_cl.mc_lock); cfs_waitq_init(&mdd->mdd_cl.mc_waitq); - mdd->mdd_cl.mc_starttime = cfs_time_current_64(); mdd->mdd_cl.mc_flags = 0; /* off by default */ - mdd->mdd_cl.mc_mask = CL_DEFMASK; + mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK; + cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock); + mdd->mdd_cl.mc_lastuser = 0; + rc = mdd_changelog_llog_init(mdd); if (rc) { CERROR("Changelog setup during init failed %d\n", rc); mdd->mdd_cl.mc_flags |= CLM_ERR; } + return rc; } @@ -203,6 +262,41 @@ static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd) mdd->mdd_cl.mc_flags = 0; } +/* Start / stop recording */ +int mdd_changelog_on(struct mdd_device *mdd, int on) +{ + int rc = 0; + + if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) { + LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name); + if (mdd->mdd_cl.mc_flags & CLM_ERR) { + CERROR("Changelogs cannot be enabled due to error " + "condition (see %s log).\n", + mdd2obd_dev(mdd)->obd_name); + rc = -ESRCH; + } else { + cfs_spin_lock(&mdd->mdd_cl.mc_lock); + mdd->mdd_cl.mc_flags |= CLM_ON; + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + rc = mdd_changelog_write_header(mdd, CLM_START); + } + } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) { + LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name); + rc = mdd_changelog_write_header(mdd, CLM_FINI); + cfs_spin_lock(&mdd->mdd_cl.mc_lock); + mdd->mdd_cl.mc_flags &= ~CLM_ON; + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + } + return rc; +} + +static __u64 cl_time(void) { + cfs_fs_time_t time; + + cfs_fs_time_current(&time); + return (((__u64)time.tv_sec) << 30) + time.tv_nsec; +} + /** Add a changelog entry \a rec to the changelog llog * \param mdd * \param rec @@ -218,19 +312,19 @@ int mdd_changelog_llog_write(struct mdd_device *mdd, struct llog_ctxt *ctxt; int rc; - if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0) + if ((mdd->mdd_cl.mc_mask & (1 << rec->cr.cr_type)) == 0) return 0; - rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen); + 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_time = cfs_time_current_64(); - spin_lock(&mdd->mdd_cl.mc_lock); + rec->cr.cr_time = cl_time(); + cfs_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_index = ++mdd->mdd_cl.mc_index; - spin_unlock(&mdd->mdd_cl.mc_lock); + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); if (ctxt == NULL) return -ENXIO; @@ -254,19 +348,41 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) { struct obd_device *obd = mdd2obd_dev(mdd); struct llog_ctxt *ctxt; + long long unsigned cur; int rc; ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); if (ctxt == NULL) return -ENXIO; - /* Some records purged; reset repeat-access time */ + cfs_spin_lock(&mdd->mdd_cl.mc_lock); + cur = (long long)mdd->mdd_cl.mc_index; + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + if (endrec > cur) + endrec = cur; + + /* purge to "0" is shorthand for everything */ + if (endrec == 0) + endrec = cur; + + /* If purging all records, write a header entry so we don't have an + empty catalog and we're sure to have a valid starting index next + time. In case of crash, we just restart with old log so we're + allright. */ + if (endrec == cur) { + rc = mdd_changelog_write_header(mdd, CLM_PURGE); + if (rc) + goto out; + } + + /* Some records were purged, so reset repeat-access time (so we + record new mtime update records, so users can see a file has been + changed since the last purge) */ mdd->mdd_cl.mc_starttime = cfs_time_current_64(); rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); - +out: llog_ctxt_put(ctxt); - return rc; } @@ -289,17 +405,17 @@ int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags) if (rec == NULL) RETURN(-ENOMEM); - rec->cr_flags = CLF_VERSION; - rec->cr_type = CL_MARK; - rec->cr_namelen = len; - memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen); + rec->cr.cr_flags = CLF_VERSION; + rec->cr.cr_type = CL_MARK; + rec->cr.cr_namelen = len; + memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen); /* Status and action flags */ - rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; + rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; rc = mdd_changelog_llog_write(mdd, rec, NULL); /* assume on or off event; reset repeat-access time */ - mdd->mdd_cl.mc_starttime = rec->cr_time; + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); OBD_FREE(rec, reclen); RETURN(rc); @@ -316,13 +432,13 @@ static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m) memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid)); mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child, - mdd_root_dir_name, mdd_dot_lustre_name, + mdd_root_dir_name, dot_lustre_name, fid, &dt_directory_features); /* .lustre dir may be already present */ if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) { rc = PTR_ERR(mdo); CERROR("creating obj [%s] fid = "DFID" rc = %d\n", - mdd_dot_lustre_name, PFID(fid), rc); + dot_lustre_name, PFID(fid), rc); RETURN(rc); } @@ -332,27 +448,89 @@ static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m) return 0; } -static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) + +static int dot_lustre_mdd_permission(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + struct md_attr *attr, int mask) +{ + if (mask & ~(MAY_READ | MAY_EXEC)) + return -EPERM; + else + return 0; +} + +static int dot_lustre_mdd_attr_get(const struct lu_env *env, + struct md_object *obj, struct md_attr *ma) { struct mdd_object *mdd_obj = md2mdd_obj(obj); return mdd_attr_get_internal_locked(env, mdd_obj, ma); } -static int dot_lustre_attr_set(const struct lu_env *env, struct md_object *obj, - const struct md_attr *ma) +static int dot_lustre_mdd_attr_set(const struct lu_env *env, + struct md_object *obj, + const struct md_attr *ma) { return -EPERM; } -static int dot_lustre_xattr_get(const struct lu_env *env, - struct md_object *obj, struct lu_buf *buf, - const char *name) +static int dot_lustre_mdd_xattr_get(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf, + const char *name) +{ + return 0; +} + +static int dot_lustre_mdd_xattr_list(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf) { return 0; } +static int dot_lustre_mdd_xattr_set(const struct lu_env *env, + struct md_object *obj, + const struct lu_buf *buf, const char *name, + int fl) +{ + return -EPERM; +} + +static int dot_lustre_mdd_xattr_del(const struct lu_env *env, + struct md_object *obj, + const char *name) +{ + return -EPERM; +} + +static int dot_lustre_mdd_readlink(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf) +{ + return 0; +} + +static int dot_lustre_mdd_object_create(const struct lu_env *env, + struct md_object *obj, + const struct md_op_spec *spec, + struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_ref_add(const struct lu_env *env, + struct md_object *obj, + const struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_ref_del(const struct lu_env *env, + struct md_object *obj, + struct md_attr *ma) +{ + return -EPERM; +} + static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj, int flags) { @@ -365,14 +543,8 @@ static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj, return 0; } -static int dot_lustre_path(const struct lu_env *env, struct md_object *obj, - char *path, int pathlen, __u64 recno, int *linkno) -{ - return -ENOSYS; -} - -static int dot_lustre_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) +static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj, + struct md_attr *ma) { struct mdd_object *mdd_obj = md2mdd_obj(obj); @@ -383,19 +555,58 @@ static int dot_lustre_close(const struct lu_env *env, struct md_object *obj, return 0; } +static int dot_lustre_mdd_object_sync(const struct lu_env *env, + struct md_object *obj) +{ + return -ENOSYS; +} + +static dt_obj_version_t dot_lustre_mdd_version_get(const struct lu_env *env, + struct md_object *obj) +{ + return 0; +} + +static void dot_lustre_mdd_version_set(const struct lu_env *env, + struct md_object *obj, + dt_obj_version_t version) +{ + return; +} + +static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj, + char *path, int pathlen, __u64 *recno, int *linkno) +{ + return -ENOSYS; +} + + static struct md_object_operations mdd_dot_lustre_obj_ops = { - .moo_attr_get = dot_lustre_attr_get, - .moo_attr_set = dot_lustre_attr_set, - .moo_xattr_get = dot_lustre_xattr_get, - .moo_open = dot_lustre_mdd_open, - .moo_close = dot_lustre_close, - .moo_readpage = mdd_readpage, - .moo_path = dot_lustre_path + .moo_permission = dot_lustre_mdd_permission, + .moo_attr_get = dot_lustre_mdd_attr_get, + .moo_attr_set = dot_lustre_mdd_attr_set, + .moo_xattr_get = dot_lustre_mdd_xattr_get, + .moo_xattr_list = dot_lustre_mdd_xattr_list, + .moo_xattr_set = dot_lustre_mdd_xattr_set, + .moo_xattr_del = dot_lustre_mdd_xattr_del, + .moo_readpage = mdd_readpage, + .moo_readlink = dot_lustre_mdd_readlink, + .moo_object_create = dot_lustre_mdd_object_create, + .moo_ref_add = dot_lustre_mdd_ref_add, + .moo_ref_del = dot_lustre_mdd_ref_del, + .moo_open = dot_lustre_mdd_open, + .moo_close = dot_lustre_mdd_close, + .moo_capa_get = mdd_capa_get, + .moo_object_sync = dot_lustre_mdd_object_sync, + .moo_version_get = dot_lustre_mdd_version_get, + .moo_version_set = dot_lustre_mdd_version_set, + .moo_path = dot_lustre_mdd_path, }; -static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p, - const struct lu_name *lname, struct lu_fid *f, - struct md_op_spec *spec) + +static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p, + const struct lu_name *lname, struct lu_fid *f, + struct md_op_spec *spec) { if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0) *f = LU_OBF_FID; @@ -405,45 +616,102 @@ static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p, return 0; } -static int dot_lustre_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 mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env, + struct md_object *obj, + mdl_mode_t mode) +{ + return MDL_MINMODE; +} + +static int dot_lustre_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) { return -EPERM; } -static int dot_lustre_rename(const struct lu_env *env, - struct md_object *src_pobj, - struct md_object *tgt_pobj, - const struct lu_fid *lf, - const struct lu_name *lsname, - struct md_object *tobj, - const struct lu_name *ltname, struct md_attr *ma) +static int dot_lustre_mdd_create_data(const struct lu_env *env, + struct md_object *p, + struct md_object *o, + const struct md_op_spec *spec, + struct md_attr *ma) { return -EPERM; } -static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj, - struct md_object *src_obj, - const struct lu_name *lname, struct md_attr *ma) +static int dot_lustre_mdd_rename(const struct lu_env *env, + struct md_object *src_pobj, + struct md_object *tgt_pobj, + const struct lu_fid *lf, + const struct lu_name *lsname, + struct md_object *tobj, + const struct lu_name *ltname, + struct md_attr *ma) { return -EPERM; } -static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj, - struct md_object *cobj, const struct lu_name *lname, - struct md_attr *ma) +static int dot_lustre_mdd_link(const struct lu_env *env, + struct md_object *tgt_obj, + struct md_object *src_obj, + const struct lu_name *lname, + struct md_attr *ma) { return -EPERM; } +static int dot_lustre_mdd_unlink(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + const struct lu_name *lname, + struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_name_insert(const struct lu_env *env, + struct md_object *obj, + const struct lu_name *lname, + const struct lu_fid *fid, + const struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_name_remove(const struct lu_env *env, + struct md_object *obj, + const struct lu_name *lname, + const struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_rename_tgt(const struct lu_env *env, + struct md_object *pobj, + struct md_object *tobj, + const struct lu_fid *fid, + const struct lu_name *lname, + struct md_attr *ma) +{ + return -EPERM; +} + + static struct md_dir_operations mdd_dot_lustre_dir_ops = { - .mdo_lookup = dot_lustre_lookup, - .mdo_create = dot_lustre_create, - .mdo_rename = dot_lustre_rename, - .mdo_link = dot_lustre_link, - .mdo_unlink = dot_lustre_unlink, + .mdo_is_subdir = mdd_is_subdir, + .mdo_lookup = dot_lustre_mdd_lookup, + .mdo_lock_mode = dot_lustre_mdd_lock_mode, + .mdo_create = dot_lustre_mdd_create, + .mdo_create_data = dot_lustre_mdd_create_data, + .mdo_rename = dot_lustre_mdd_rename, + .mdo_link = dot_lustre_mdd_link, + .mdo_unlink = dot_lustre_mdd_unlink, + .mdo_name_insert = dot_lustre_mdd_name_insert, + .mdo_name_remove = dot_lustre_mdd_name_remove, + .mdo_rename_tgt = dot_lustre_mdd_rename_tgt, }; static int obf_attr_get(const struct lu_env *env, struct md_object *obj, @@ -457,7 +725,8 @@ static int obf_attr_get(const struct lu_env *env, struct md_object *obj, /* "fid" is a virtual object and hence does not have any "real" * attributes. So we reuse attributes of .lustre for "fid" dir */ ma->ma_need |= MA_INODE; - rc = dot_lustre_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma); + rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, + ma); if (rc) return rc; ma->ma_valid |= MA_INODE; @@ -532,7 +801,7 @@ static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj, } static int obf_path(const struct lu_env *env, struct md_object *obj, - char *path, int pathlen, __u64 recno, int *linkno) + char *path, int pathlen, __u64 *recno, int *linkno) { return -ENOSYS; } @@ -563,8 +832,7 @@ static int obf_lookup(const struct lu_env *env, struct md_object *p, while (*name == '[') name++; - sscanf(name, SFID, &(f->f_seq), &(f->f_oid), - &(f->f_ver)); + sscanf(name, SFID, RFID(f)); if (!fid_is_sane(f)) { CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name, (__u64)1, 2, 0); @@ -665,7 +933,7 @@ static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m) return rc; dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name, - mdd_dot_lustre_name, fid); + dot_lustre_name, fid); if (IS_ERR(dt_dot_lustre)) { rc = PTR_ERR(dt_dot_lustre); GOTO(out, rc); @@ -742,7 +1010,7 @@ static int mdd_lov_set_nextid(const struct lu_env *env, ENTRY; LASSERT(mds->mds_lov_objids != NULL); - rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID), + rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID), KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count, mds->mds_lov_objids, NULL); @@ -786,7 +1054,7 @@ static int mdd_recovery_complete(const struct lu_env *env, } #endif /* Call that with obd_recovering = 1 just to update objids */ - obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ? + obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL); /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */ @@ -861,7 +1129,7 @@ static int mdd_root_get(const struct lu_env *env, * No permission check is needed. */ static int mdd_statfs(const struct lu_env *env, struct md_device *m, - struct kstatfs *sfs) + cfs_kstatfs_t *sfs) { struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); int rc; @@ -897,7 +1165,11 @@ static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m, int rc; ENTRY; + /* need barrier for mds_capa_keys access. */ + cfs_down_write(&mds->mds_notify_lock); mds->mds_capa_keys = keys; + cfs_up_write(&mds->mds_notify_lock); + rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode, timeout, alg, keys); RETURN(rc); @@ -907,16 +1179,26 @@ static int mdd_update_capa_key(const struct lu_env *env, struct md_device *m, struct lustre_capa_key *key) { + struct mds_capa_info info = { .uuid = NULL, .capa = key }; struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp; + struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp; int rc; ENTRY; rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY, - sizeof(*key), key, NULL); + sizeof(info), &info, NULL); RETURN(rc); } +static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m, + int idx, void **h) +{ + struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); + + *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx); + return (*h == NULL ? -ENOENT : 0); +} + static struct lu_device *mdd_device_alloc(const struct lu_env *env, struct lu_device_type *t, struct lustre_cfg *lcfg) @@ -945,7 +1227,7 @@ static struct lu_device *mdd_device_free(const struct lu_env *env, struct lu_device *next = &m->mdd_child->dd_lu_dev; ENTRY; - LASSERT(atomic_read(&lu->ld_ref) == 0); + LASSERT(cfs_atomic_read(&lu->ld_ref) == 0); md_device_fini(&m->mdd_md_dev); OBD_FREE_PTR(m); RETURN(next); @@ -992,8 +1274,248 @@ struct md_capainfo *md_capainfo(const struct lu_env *env) } EXPORT_SYMBOL(md_capainfo); +/* + * context key constructor/destructor: + * mdd_quota_key_init, mdd_quota_key_fini + */ +LU_KEY_INIT_FINI(mdd_quota, struct md_quota); + +struct lu_context_key mdd_quota_key = { + .lct_tags = LCT_SESSION, + .lct_init = mdd_quota_key_init, + .lct_fini = mdd_quota_key_fini +}; + +struct md_quota *md_quota(const struct lu_env *env) +{ + LASSERT(env->le_ses != NULL); + return lu_context_key_get(env->le_ses, &mdd_quota_key); +} +EXPORT_SYMBOL(md_quota); + +static int mdd_changelog_user_register(struct mdd_device *mdd, int *id) +{ + struct llog_ctxt *ctxt; + struct llog_changelog_user_rec *rec; + int rc; + ENTRY; + + ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT); + if (ctxt == NULL) + RETURN(-ENXIO); + + OBD_ALLOC_PTR(rec); + if (rec == NULL) { + llog_ctxt_put(ctxt); + RETURN(-ENOMEM); + } + + /* Assume we want it on since somebody registered */ + rc = mdd_changelog_on(mdd, 1); + if (rc) + GOTO(out, rc); + + rec->cur_hdr.lrh_len = sizeof(*rec); + rec->cur_hdr.lrh_type = CHANGELOG_USER_REC; + cfs_spin_lock(&mdd->mdd_cl.mc_user_lock); + if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) { + cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); + CERROR("Maximum number of changelog users exceeded!\n"); + GOTO(out, rc = -EOVERFLOW); + } + *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser; + rec->cur_endrec = mdd->mdd_cl.mc_index; + cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); + + rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0); + + CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id); +out: + OBD_FREE_PTR(rec); + llog_ctxt_put(ctxt); + RETURN(rc); +} + +struct mdd_changelog_user_data { + __u64 mcud_endrec; /**< purge record for this user */ + __u64 mcud_minrec; /**< lowest changelog recno still referenced */ + __u32 mcud_id; + __u32 mcud_minid; /**< user id with lowest rec reference */ + __u32 mcud_usercount; + int mcud_found:1; +}; +#define MCUD_UNREGISTER -1LL + +/** Two things: + * 1. Find the smallest record everyone is willing to purge + * 2. Update the last purgeable record for this user + */ +static int mdd_changelog_user_purge_cb(struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) +{ + struct llog_changelog_user_rec *rec; + struct mdd_changelog_user_data *mcud = + (struct mdd_changelog_user_data *)data; + int rc; + ENTRY; + + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + + rec = (struct llog_changelog_user_rec *)hdr; + + mcud->mcud_usercount++; + + /* If we have a new endrec for this id, use it for the following + min check instead of its old value */ + if (rec->cur_id == mcud->mcud_id) + rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec); + + /* Track the minimum referenced record */ + if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) { + mcud->mcud_minid = rec->cur_id; + mcud->mcud_minrec = rec->cur_endrec; + } + + if (rec->cur_id != mcud->mcud_id) + RETURN(0); + + /* Update this user's record */ + mcud->mcud_found = 1; + + /* Special case: unregister this user */ + if (mcud->mcud_endrec == MCUD_UNREGISTER) { + struct llog_cookie cookie; + cookie.lgc_lgl = llh->lgh_id; + cookie.lgc_index = hdr->lrh_index; + rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle, + 1, &cookie); + if (rc == 0) + mcud->mcud_usercount--; + RETURN(rc); + } + + /* Update the endrec */ + CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n", + mcud->mcud_id, rec->cur_endrec); + + /* hdr+1 is loc of data */ + hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail); + rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1), + hdr->lrh_index); + + RETURN(rc); +} + +static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, + long long endrec) +{ + struct mdd_changelog_user_data data; + struct llog_ctxt *ctxt; + int rc; + ENTRY; + + CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec); + + data.mcud_id = id; + data.mcud_minid = 0; + data.mcud_minrec = 0; + data.mcud_usercount = 0; + data.mcud_endrec = endrec; + cfs_spin_lock(&mdd->mdd_cl.mc_lock); + endrec = mdd->mdd_cl.mc_index; + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + if ((data.mcud_endrec == 0) || + ((data.mcud_endrec > endrec) && + (data.mcud_endrec != MCUD_UNREGISTER))) + data.mcud_endrec = endrec; + + ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT); + + rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb, + (void *)&data, 0, 0); + if ((rc >= 0) && (data.mcud_minrec > 0)) { + CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64 + ", referenced by "CHANGELOG_USER_PREFIX"%d\n", + data.mcud_minrec, data.mcud_minid); + rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec); + } else { + CWARN("Could not determine changelog records to purge; rc=%d\n", + rc); + } + + llog_ctxt_put(ctxt); + + if (!data.mcud_found) { + CWARN("No entry for user %d. Last changelog reference is " + LPD64" by changelog user %d\n", data.mcud_id, + data.mcud_minrec, data.mcud_minid); + rc = -ENOENT; + } + + if (!rc && data.mcud_usercount == 0) + /* No more users; turn changelogs off */ + rc = mdd_changelog_on(mdd, 0); + + RETURN (rc); +} + +/** mdd_iocontrol + * May be called remotely from mdt_iocontrol_handle or locally from + * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce + * an obd_ioctl_data format (but local ioctl handler does). + * \param cmd - ioc + * \param len - data len + * \param karg - ioctl data, in kernel space + */ +static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, + unsigned int cmd, int len, void *karg) +{ + struct mdd_device *mdd; + struct obd_ioctl_data *data = karg; + int rc; + ENTRY; + + mdd = lu2mdd_dev(&m->md_lu_dev); + + /* Doesn't use obd_ioctl_data */ + if (cmd == OBD_IOC_CHANGELOG_CLEAR) { + struct changelog_setinfo *cs = karg; + rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno); + RETURN(rc); + } + + /* Below ioctls use obd_ioctl_data */ + if (len != sizeof(*data)) { + CERROR("Bad ioctl size %d\n", len); + RETURN(-EINVAL); + } + if (data->ioc_version != OBD_IOCTL_VERSION) { + CERROR("Bad magic %x != %x\n", data->ioc_version, + OBD_IOCTL_VERSION); + RETURN(-EINVAL); + } + + switch (cmd) { + case OBD_IOC_CHANGELOG_REG: + rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1); + break; + case OBD_IOC_CHANGELOG_DEREG: + rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1, + MCUD_UNREGISTER); + break; + default: + rc = -EOPNOTSUPP; + } + + RETURN (rc); +} + /* type constructor/destructor: mdd_type_init, mdd_type_fini */ -LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key); +LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key, + &mdd_quota_key); const struct md_device_operations mdd_ops = { .mdo_statfs = mdd_statfs, @@ -1001,6 +1523,8 @@ const struct md_device_operations mdd_ops = { .mdo_maxsize_get = mdd_maxsize_get, .mdo_init_capa_ctxt = mdd_init_capa_ctxt, .mdo_update_capa_key= mdd_update_capa_key, + .mdo_llog_ctxt_get = mdd_llog_ctxt_get, + .mdo_iocontrol = mdd_iocontrol, #ifdef HAVE_QUOTA_SUPPORT .mdo_quota = { .mqo_notify = mdd_quota_notify,