X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftarget%2Fout_handler.c;h=ab902aaf2a3765b5a1ca538998583fd17aafd7a8;hp=5a00dfb3d6578b086b7dc28cfab3a8ab348c6419;hb=713174908cb8e5e3ceadd3ca1cb42a88b200e576;hpb=cb7da2d4c0241f52dec5e05c57847ad6cd0dbb83 diff --git a/lustre/target/out_handler.c b/lustre/target/out_handler.c index 5a00dfb..ab902aa 100644 --- a/lustre/target/out_handler.c +++ b/lustre/target/out_handler.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2013, 2014, Intel Corporation. + * Copyright (c) 2013, 2015, Intel Corporation. * * lustre/target/out_handler.c * @@ -31,81 +31,13 @@ #define DEBUG_SUBSYSTEM S_CLASS -#include +#include +#include +#include +#include #include +#include #include "tgt_internal.h" -#include - -static int tx_extend_args(struct thandle_exec_args *ta, int new_alloc_ta) -{ - struct tx_arg **new_ta; - int i; - int rc = 0; - - if (ta->ta_alloc_args >= new_alloc_ta) - return 0; - - OBD_ALLOC(new_ta, sizeof(*new_ta) * new_alloc_ta); - if (new_ta == NULL) - return -ENOMEM; - - for (i = 0; i < new_alloc_ta; i++) { - if (i < ta->ta_alloc_args) { - /* copy the old args to new one */ - new_ta[i] = ta->ta_args[i]; - } else { - OBD_ALLOC_PTR(new_ta[i]); - if (new_ta[i] == NULL) - GOTO(out, rc = -ENOMEM); - } - } - - /* free the old args */ - if (ta->ta_args != NULL) - OBD_FREE(ta->ta_args, sizeof(ta->ta_args[0]) * - ta->ta_alloc_args); - - ta->ta_args = new_ta; - ta->ta_alloc_args = new_alloc_ta; -out: - if (rc != 0) { - for (i = 0; i < new_alloc_ta; i++) { - if (new_ta[i] != NULL) - OBD_FREE_PTR(new_ta[i]); - } - OBD_FREE(new_ta, sizeof(*new_ta) * new_alloc_ta); - } - return rc; -} - -#define TX_ALLOC_STEP 8 -static struct tx_arg *tx_add_exec(struct thandle_exec_args *ta, - tx_exec_func_t func, tx_exec_func_t undo, - const char *file, int line) -{ - int rc; - int i; - - LASSERT(ta != NULL); - LASSERT(func != NULL); - - if (ta->ta_argno + 1 >= ta->ta_alloc_args) { - rc = tx_extend_args(ta, ta->ta_alloc_args + TX_ALLOC_STEP); - if (rc != 0) - return ERR_PTR(rc); - } - - i = ta->ta_argno; - - ta->ta_argno++; - - ta->ta_args[i]->exec_fn = func; - ta->ta_args[i]->undo_fn = undo; - ta->ta_args[i]->file = file; - ta->ta_args[i]->line = line; - - return ta->ta_args[i]; -} static void out_reconstruct(const struct lu_env *env, struct dt_device *dt, struct dt_object *obj, @@ -137,6 +69,22 @@ static inline int out_check_resent(const struct lu_env *env, return 0; if (req_xid_is_last(req)) { + struct lsd_client_data *lcd; + + /* XXX this does not support mulitple transactions yet, i.e. + * only 1 update RPC each time betwee MDTs */ + lcd = req->rq_export->exp_target_data.ted_lcd; + + req->rq_transno = lcd->lcd_last_transno; + req->rq_status = lcd->lcd_last_result; + if (req->rq_status != 0) + req->rq_transno = 0; + lustre_msg_set_transno(req->rq_repmsg, req->rq_transno); + lustre_msg_set_status(req->rq_repmsg, req->rq_status); + + DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"status %d", + req->rq_transno, req->rq_status); + reconstruct(env, dt, obj, reply, index); return 1; } @@ -145,101 +93,6 @@ static inline int out_check_resent(const struct lu_env *env, return 0; } -static int out_obj_destroy(const struct lu_env *env, struct dt_object *dt_obj, - struct thandle *th) -{ - int rc; - - CDEBUG(D_INFO, "%s: destroy "DFID"\n", dt_obd_name(th->th_dev), - PFID(lu_object_fid(&dt_obj->do_lu))); - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_destroy(env, dt_obj, th); - dt_write_unlock(env, dt_obj); - - return rc; -} - -/** - * All of the xxx_undo will be used once execution failed, - * But because all of the required resource has been reserved in - * declare phase, i.e. if declare succeed, it should make sure - * the following executing phase succeed in anyway, so these undo - * should be useless for most of the time in Phase I - */ -static int out_tx_create_undo(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - int rc; - - rc = out_obj_destroy(env, arg->object, th); - if (rc != 0) - CERROR("%s: undo failure, we are doomed!: rc = %d\n", - dt_obd_name(th->th_dev), rc); - return rc; -} - -static int out_tx_create_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - CDEBUG(D_OTHER, "%s: create "DFID": dof %u, mode %o\n", - dt_obd_name(th->th_dev), - PFID(lu_object_fid(&arg->object->do_lu)), - arg->u.create.dof.dof_type, - arg->u.create.attr.la_mode & S_IFMT); - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_create(env, dt_obj, &arg->u.create.attr, - &arg->u.create.hint, &arg->u.create.dof, th); - - dt_write_unlock(env, dt_obj); - - CDEBUG(D_INFO, "%s: insert create reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int __out_tx_create(const struct lu_env *env, struct dt_object *obj, - struct lu_attr *attr, struct lu_fid *parent_fid, - struct dt_object_format *dof, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_create(env, obj, attr, NULL, dof, - ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_create_exec, out_tx_create_undo, file, - line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - /* release the object in out_trans_stop */ - lu_object_get(&obj->do_lu); - arg->object = obj; - arg->u.create.attr = *attr; - if (parent_fid != NULL) - arg->u.create.fid = *parent_fid; - memset(&arg->u.create.hint, 0, sizeof(arg->u.create.hint)); - arg->u.create.dof = *dof; - arg->reply = reply; - arg->index = index; - - return 0; -} - static int out_create(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -256,10 +109,10 @@ static int out_create(struct tgt_session_info *tsi) ENTRY; wobdo = object_update_param_get(update, 0, &size); - if (wobdo == NULL || IS_ERR(wobdo) || size != sizeof(*wobdo)) { - CERROR("%s: obdo is NULL, invalid RPC: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(wobdo) || size != sizeof(*wobdo)) { + CERROR("%s: obdo is NULL, invalid RPC: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo)); + RETURN(PTR_ERR(wobdo)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -270,17 +123,17 @@ static int out_create(struct tgt_session_info *tsi) dof->dof_type = dt_mode_to_dft(attr->la_mode); if (update->ou_params_count > 1) { fid = object_update_param_get(update, 1, &size); - if (fid == NULL || IS_ERR(fid) || size != sizeof(*fid)) { - CERROR("%s: invalid fid: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(fid) || size != sizeof(*fid)) { + CERROR("%s: invalid fid: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(fid)); + RETURN(PTR_ERR(fid)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) lustre_swab_lu_fid(fid); if (!fid_is_sane(fid)) { CERROR("%s: invalid fid "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } } @@ -288,72 +141,13 @@ static int out_create(struct tgt_session_info *tsi) RETURN(-EEXIST); rc = out_tx_create(tsi->tsi_env, obj, attr, fid, dof, - &tti->tti_tea, + &tti->tti_tea, tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_tx_attr_set_undo(const struct lu_env *env, - struct thandle *th, struct tx_arg *arg) -{ - CERROR("%s: attr set undo "DFID" unimplemented yet!: rc = %d\n", - dt_obd_name(th->th_dev), - PFID(lu_object_fid(&arg->object->do_lu)), -ENOTSUPP); - - return -ENOTSUPP; -} - -static int out_tx_attr_set_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - CDEBUG(D_OTHER, "%s: attr set "DFID"\n", dt_obd_name(th->th_dev), - PFID(lu_object_fid(&dt_obj->do_lu))); - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_attr_set(env, dt_obj, &arg->u.attr_set.attr, th); - dt_write_unlock(env, dt_obj); - - CDEBUG(D_INFO, "%s: insert attr_set reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int __out_tx_attr_set(const struct lu_env *env, - struct dt_object *dt_obj, - const struct lu_attr *attr, - struct thandle_exec_args *th, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(th->ta_handle != NULL); - rc = dt_declare_attr_set(env, dt_obj, attr, th->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(th, out_tx_attr_set_exec, out_tx_attr_set_undo, - file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->u.attr_set.attr = *attr; - arg->reply = reply; - arg->index = index; - return 0; -} - static int out_attr_set(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -368,10 +162,10 @@ static int out_attr_set(struct tgt_session_info *tsi) ENTRY; wobdo = object_update_param_get(update, 0, &size); - if (wobdo == NULL || IS_ERR(wobdo) || size != sizeof(*wobdo)) { - CERROR("%s: empty obdo in the update: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(wobdo) || size != sizeof(*wobdo)) { + CERROR("%s: empty obdo in the update: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo)); + RETURN(PTR_ERR(wobdo)); } attr->la_valid = 0; @@ -383,6 +177,7 @@ static int out_attr_set(struct tgt_session_info *tsi) la_from_obdo(attr, lobdo, lobdo->o_valid); rc = out_tx_attr_set(tsi->tsi_env, obj, attr, &tti->tti_tea, + tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); @@ -393,6 +188,7 @@ static int out_attr_get(struct tgt_session_info *tsi) { const struct lu_env *env = tsi->tsi_env; struct tgt_thread_info *tti = tgt_th_info(env); + struct object_update *update = tti->tti_u.update.tti_update; struct obdo *obdo = &tti->tti_u.update.tti_obdo; struct lu_attr *la = &tti->tti_attr; struct dt_object *obj = tti->tti_u.update.tti_dt_object; @@ -401,6 +197,9 @@ static int out_attr_get(struct tgt_session_info *tsi) ENTRY; + if (unlikely(update->ou_result_size < sizeof(*obdo))) + return -EPROTO; + if (!lu_object_exists(&obj->do_lu)) { /* Usually, this will be called when the master MDT try * to init a remote object(see osp_object_init), so if @@ -455,32 +254,28 @@ static int out_xattr_get(struct tgt_session_info *tsi) } name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for xattr get: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for xattr get: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } update_result = object_update_result_get(reply, 0, NULL); if (update_result == NULL) { CERROR("%s: empty name for xattr get: rc = %d\n", tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } + lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size; lbuf->lb_buf = update_result->our_data; - lbuf->lb_len = OUT_UPDATE_REPLY_SIZE - - cfs_size_round((unsigned long)update_result->our_data - - (unsigned long)update_result); + if (lbuf->lb_len == 0) + lbuf->lb_buf = 0; dt_read_lock(env, obj, MOR_TGT_CHILD); rc = dt_xattr_get(env, obj, lbuf, name); dt_read_unlock(env, obj); - if (rc < 0) { + if (rc < 0) lbuf->lb_len = 0; - GOTO(out, rc); - } - lbuf->lb_len = rc; - rc = 0; CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n", tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), name, (int)lbuf->lb_len); @@ -489,7 +284,7 @@ static int out_xattr_get(struct tgt_session_info *tsi) out: object_update_result_insert(reply, lbuf->lb_buf, lbuf->lb_len, idx, rc); - RETURN(rc); + RETURN(0); } static int out_index_lookup(struct tgt_session_info *tsi) @@ -503,14 +298,17 @@ static int out_index_lookup(struct tgt_session_info *tsi) ENTRY; + if (unlikely(update->ou_result_size < sizeof(tti->tti_fid1))) + return -EPROTO; + if (!lu_object_exists(&obj->do_lu)) RETURN(-ENOENT); name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for lookup: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for lookup: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } dt_read_lock(env, obj, MOR_TGT_CHILD); @@ -543,105 +341,6 @@ out_unlock: RETURN(rc); } -static int out_tx_xattr_set_exec(const struct lu_env *env, - struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - CDEBUG(D_INFO, "%s: set xattr buf %p name %s flag %d\n", - dt_obd_name(th->th_dev), arg->u.xattr_set.buf.lb_buf, - arg->u.xattr_set.name, arg->u.xattr_set.flags); - - if (!lu_object_exists(&dt_obj->do_lu)) - GOTO(out, rc = -ENOENT); - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_xattr_set(env, dt_obj, &arg->u.xattr_set.buf, - arg->u.xattr_set.name, arg->u.xattr_set.flags, th); - /** - * Ignore errors if this is LINK EA - **/ - if (unlikely(rc != 0 && - strcmp(arg->u.xattr_set.name, XATTR_NAME_LINK) == 0)) { - /* XXX: If the linkEA is overflow, then we need to notify the - * namespace LFSCK to skip "nlink" attribute verification - * on this object to avoid the "nlink" to be shrinked by - * wrong. It may be not good an interaction with LFSCK - * like this. We will consider to replace it with other - * mechanism in future. LU-5802. */ - if (rc == -ENOSPC) { - struct lfsck_request *lr = &tgt_th_info(env)->tti_lr; - - lfsck_pack_rfa(lr, lu_object_fid(&dt_obj->do_lu), - LE_SKIP_NLINK, LFSCK_TYPE_NAMESPACE); - tgt_lfsck_in_notify(env, - tgt_ses_info(env)->tsi_tgt->lut_bottom, lr, th); - } - - rc = 0; - } - dt_write_unlock(env, dt_obj); - -out: - CDEBUG(D_INFO, "%s: insert xattr set reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int __out_tx_xattr_set(const struct lu_env *env, - struct dt_object *dt_obj, - const struct lu_buf *buf, - const char *name, int flags, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_xattr_set(env, dt_obj, buf, name, flags, ta->ta_handle); - if (rc != 0) - return rc; - - if (strcmp(name, XATTR_NAME_LINK) == 0) { - struct lfsck_request *lr = &tgt_th_info(env)->tti_lr; - - /* XXX: If the linkEA is overflow, then we need to notify the - * namespace LFSCK to skip "nlink" attribute verification - * on this object to avoid the "nlink" to be shrinked by - * wrong. It may be not good an interaction with LFSCK - * like this. We will consider to replace it with other - * mechanism in future. LU-5802. */ - lfsck_pack_rfa(lr, lu_object_fid(&dt_obj->do_lu), - LE_SKIP_NLINK_DECLARE, LFSCK_TYPE_NAMESPACE); - rc = tgt_lfsck_in_notify(env, - tgt_ses_info(env)->tsi_tgt->lut_bottom, - lr, ta->ta_handle); - if (rc != 0) - return rc; - } - - arg = tx_add_exec(ta, out_tx_xattr_set_exec, NULL, file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->u.xattr_set.name = name; - arg->u.xattr_set.flags = flags; - arg->u.xattr_set.buf = *buf; - arg->reply = reply; - arg->index = index; - arg->u.xattr_set.csum = 0; - return 0; -} - static int out_xattr_set(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -658,24 +357,25 @@ static int out_xattr_set(struct tgt_session_info *tsi) ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for xattr set: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for xattr set: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } + /* If buffer == NULL (-ENODATA), then it might mean delete xattr */ buf = object_update_param_get(update, 1, &buf_len); - if (IS_ERR(buf)) - RETURN(err_serious(-EPROTO)); + if (IS_ERR(buf) && PTR_ERR(buf) != -ENODATA) + RETURN(PTR_ERR(buf)); lbuf->lb_buf = buf; lbuf->lb_len = buf_len; tmp = object_update_param_get(update, 2, &size); - if (tmp == NULL || size != sizeof(*tmp)) { - CERROR("%s: emptry or wrong size %zu flag: rc = %d\n", - tgt_name(tsi->tsi_tgt), size, -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(tmp) || size != sizeof(*tmp)) { + CERROR("%s: emptry or wrong size %zu flag: rc = %ld\n", + tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp)); + RETURN(PTR_ERR(tmp)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -683,62 +383,12 @@ static int out_xattr_set(struct tgt_session_info *tsi) flag = *tmp; rc = out_tx_xattr_set(tsi->tsi_env, obj, lbuf, name, flag, - &tti->tti_tea, + &tti->tti_tea, tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_tx_xattr_del_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - CDEBUG(D_INFO, "%s: del xattr name '%s' on "DFID"\n", - dt_obd_name(th->th_dev), arg->u.xattr_set.name, - PFID(lu_object_fid(&dt_obj->do_lu))); - - if (!lu_object_exists(&dt_obj->do_lu)) - GOTO(out, rc = -ENOENT); - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_xattr_del(env, dt_obj, arg->u.xattr_set.name, th); - dt_write_unlock(env, dt_obj); -out: - CDEBUG(D_INFO, "%s: insert xattr del reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int __out_tx_xattr_del(const struct lu_env *env, - struct dt_object *dt_obj, const char *name, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - rc = dt_declare_xattr_del(env, dt_obj, name, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_xattr_del_exec, NULL, file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->u.xattr_set.name = name; - arg->reply = reply; - arg->index = index; - return 0; -} - static int out_xattr_del(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -749,91 +399,19 @@ static int out_xattr_del(struct tgt_session_info *tsi) ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for xattr set: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for xattr set: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } rc = out_tx_xattr_del(tsi->tsi_env, obj, name, &tti->tti_tea, + tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_obj_ref_add(const struct lu_env *env, - struct dt_object *dt_obj, - struct thandle *th) -{ - int rc; - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_ref_add(env, dt_obj, th); - dt_write_unlock(env, dt_obj); - - return rc; -} - -static int out_obj_ref_del(const struct lu_env *env, - struct dt_object *dt_obj, - struct thandle *th) -{ - int rc; - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_ref_del(env, dt_obj, th); - dt_write_unlock(env, dt_obj); - - return rc; -} - -static int out_tx_ref_add_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - rc = out_obj_ref_add(env, dt_obj, th); - - CDEBUG(D_INFO, "%s: insert ref_add reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - return rc; -} - -static int out_tx_ref_add_undo(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - return out_obj_ref_del(env, arg->object, th); -} - -static int __out_tx_ref_add(const struct lu_env *env, - struct dt_object *dt_obj, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_ref_add(env, dt_obj, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_ref_add_exec, out_tx_ref_add_undo, file, - line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - return 0; -} - /** * increase ref of the object **/ @@ -846,59 +424,12 @@ static int out_ref_add(struct tgt_session_info *tsi) ENTRY; rc = out_tx_ref_add(tsi->tsi_env, obj, &tti->tti_tea, + tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_tx_ref_del_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - rc = out_obj_ref_del(env, dt_obj, th); - - CDEBUG(D_INFO, "%s: insert ref_del reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, 0); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int out_tx_ref_del_undo(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - return out_obj_ref_add(env, arg->object, th); -} - -static int __out_tx_ref_del(const struct lu_env *env, - struct dt_object *dt_obj, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_ref_del(env, dt_obj, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_ref_del_exec, out_tx_ref_del_undo, file, - line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - return 0; -} - static int out_ref_del(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -911,118 +442,12 @@ static int out_ref_del(struct tgt_session_info *tsi) RETURN(-ENOENT); rc = out_tx_ref_del(tsi->tsi_env, obj, &tti->tti_tea, + tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_obj_index_insert(const struct lu_env *env, - struct dt_object *dt_obj, - const struct dt_rec *rec, - const struct dt_key *key, - struct thandle *th) -{ - int rc; - - CDEBUG(D_INFO, "%s: index insert "DFID" name: %s fid "DFID", type %u\n", - dt_obd_name(th->th_dev), PFID(lu_object_fid(&dt_obj->do_lu)), - (char *)key, PFID(((struct dt_insert_rec *)rec)->rec_fid), - ((struct dt_insert_rec *)rec)->rec_type); - - if (dt_try_as_dir(env, dt_obj) == 0) - return -ENOTDIR; - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_insert(env, dt_obj, rec, key, th, 0); - dt_write_unlock(env, dt_obj); - - return rc; -} - -static int out_obj_index_delete(const struct lu_env *env, - struct dt_object *dt_obj, - const struct dt_key *key, - struct thandle *th) -{ - int rc; - - CDEBUG(D_INFO, "%s: index delete "DFID" name: %s\n", - dt_obd_name(th->th_dev), PFID(lu_object_fid(&dt_obj->do_lu)), - (char *)key); - - if (dt_try_as_dir(env, dt_obj) == 0) - return -ENOTDIR; - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_delete(env, dt_obj, key, th); - dt_write_unlock(env, dt_obj); - - return rc; -} - -static int out_tx_index_insert_exec(const struct lu_env *env, - struct thandle *th, struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - if (unlikely(!dt_object_exists(dt_obj))) - RETURN(-ESTALE); - - rc = out_obj_index_insert(env, dt_obj, - (const struct dt_rec *)&arg->u.insert.rec, - arg->u.insert.key, th); - - CDEBUG(D_INFO, "%s: insert idx insert reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int out_tx_index_insert_undo(const struct lu_env *env, - struct thandle *th, struct tx_arg *arg) -{ - return out_obj_index_delete(env, arg->object, arg->u.insert.key, th); -} - -static int __out_tx_index_insert(const struct lu_env *env, - struct dt_object *dt_obj, - const struct dt_rec *rec, - const struct dt_key *key, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - if (dt_try_as_dir(env, dt_obj) == 0) { - rc = -ENOTDIR; - return rc; - } - - rc = dt_declare_insert(env, dt_obj, rec, key, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_index_insert_exec, - out_tx_index_insert_undo, file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - arg->u.insert.rec = *(const struct dt_insert_rec *)rec; - arg->u.insert.key = key; - - return 0; -} - static int out_index_insert(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -1037,17 +462,17 @@ static int out_index_insert(struct tgt_session_info *tsi) ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for index insert: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for index insert: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } fid = object_update_param_get(update, 1, &size); - if (fid == NULL || IS_ERR(fid) || size != sizeof(*fid)) { - CERROR("%s: invalid fid: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(fid) || size != sizeof(*fid)) { + CERROR("%s: invalid fid: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(fid)); + RETURN(PTR_ERR(fid)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -1056,14 +481,14 @@ static int out_index_insert(struct tgt_session_info *tsi) if (!fid_is_sane(fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } ptype = object_update_param_get(update, 2, &size); - if (ptype == NULL || IS_ERR(ptype) || size != sizeof(*ptype)) { - CERROR("%s: invalid type for index insert: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(ptype) || size != sizeof(*ptype)) { + CERROR("%s: invalid type for index insert: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(ptype)); + RETURN(PTR_ERR(ptype)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -1074,69 +499,12 @@ static int out_index_insert(struct tgt_session_info *tsi) rc = out_tx_index_insert(tsi->tsi_env, obj, (const struct dt_rec *)rec, (const struct dt_key *)name, &tti->tti_tea, + tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_tx_index_delete_exec(const struct lu_env *env, - struct thandle *th, - struct tx_arg *arg) -{ - int rc; - - rc = out_obj_index_delete(env, arg->object, arg->u.insert.key, th); - - CDEBUG(D_INFO, "%s: delete idx insert reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc; -} - -static int out_tx_index_delete_undo(const struct lu_env *env, - struct thandle *th, - struct tx_arg *arg) -{ - CERROR("%s: Oops, can not rollback index_delete yet: rc = %d\n", - dt_obd_name(th->th_dev), -ENOTSUPP); - return -ENOTSUPP; -} - -static int __out_tx_index_delete(const struct lu_env *env, - struct dt_object *dt_obj, - const struct dt_key *key, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - if (dt_try_as_dir(env, dt_obj) == 0) { - rc = -ENOTDIR; - return rc; - } - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_delete(env, dt_obj, key, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_index_delete_exec, - out_tx_index_delete_undo, file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - arg->u.insert.key = key; - return 0; -} - static int out_index_delete(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -1149,68 +517,19 @@ static int out_index_delete(struct tgt_session_info *tsi) RETURN(-ENOENT); name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for index delete: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for index delete: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } rc = out_tx_index_delete(tsi->tsi_env, obj, (const struct dt_key *)name, - &tti->tti_tea, + &tti->tti_tea, tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_tx_destroy_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - rc = out_obj_destroy(env, dt_obj, th); - - CDEBUG(D_INFO, "%s: insert destroy reply %p index %d: rc = %d\n", - dt_obd_name(th->th_dev), arg->reply, arg->index, rc); - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - RETURN(rc); -} - -static int out_tx_destroy_undo(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - CERROR("%s: not support destroy undo yet!: rc = %d\n", - dt_obd_name(th->th_dev), -ENOTSUPP); - return -ENOTSUPP; -} - -static int __out_tx_destroy(const struct lu_env *env, struct dt_object *dt_obj, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_destroy(env, dt_obj, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_destroy_exec, out_tx_destroy_undo, - file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - return 0; -} - static int out_destroy(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -1224,66 +543,20 @@ static int out_destroy(struct tgt_session_info *tsi) if (!fid_is_sane(fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } if (!lu_object_exists(&obj->do_lu)) RETURN(-ENOENT); rc = out_tx_destroy(tsi->tsi_env, obj, &tti->tti_tea, + tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } -static int out_tx_write_exec(const struct lu_env *env, struct thandle *th, - struct tx_arg *arg) -{ - struct dt_object *dt_obj = arg->object; - int rc; - - dt_write_lock(env, dt_obj, MOR_TGT_CHILD); - rc = dt_record_write(env, dt_obj, &arg->u.write.buf, - &arg->u.write.pos, th); - dt_write_unlock(env, dt_obj); - - if (rc == 0) - rc = arg->u.write.buf.lb_len; - - object_update_result_insert(arg->reply, NULL, 0, arg->index, rc); - - return rc > 0 ? 0 : rc; -} - -static int __out_tx_write(const struct lu_env *env, - struct dt_object *dt_obj, - const struct lu_buf *buf, - loff_t pos, struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, const char *file, int line) -{ - struct tx_arg *arg; - int rc; - - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_record_write(env, dt_obj, buf, pos, ta->ta_handle); - if (rc != 0) - return rc; - - arg = tx_add_exec(ta, out_tx_write_exec, NULL, file, line); - if (IS_ERR(arg)) - return PTR_ERR(arg); - - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->u.write.buf = *buf; - arg->u.write.pos = pos; - arg->reply = reply; - arg->index = index; - return 0; -} - static int out_write(struct tgt_session_info *tsi) { struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); @@ -1299,19 +572,19 @@ static int out_write(struct tgt_session_info *tsi) ENTRY; buf = object_update_param_get(update, 0, &buf_len); - if (buf == NULL || IS_ERR(buf) || buf_len == 0) { - CERROR("%s: empty buf for xattr set: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(buf) || buf_len == 0) { + CERROR("%s: empty buf for xattr set: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(buf)); + RETURN(PTR_ERR(buf)); } lbuf->lb_buf = buf; lbuf->lb_len = buf_len; tmp = object_update_param_get(update, 1, &size); - if (tmp == NULL || IS_ERR(tmp) || size != sizeof(*tmp)) { - CERROR("%s: empty or wrong size %zu pos: rc = %d\n", - tgt_name(tsi->tsi_tgt), size, -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(tmp) || size != sizeof(*tmp)) { + CERROR("%s: empty or wrong size %zu pos: rc = %ld\n", + tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp)); + RETURN(PTR_ERR(tmp)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -1319,12 +592,120 @@ static int out_write(struct tgt_session_info *tsi) pos = *tmp; rc = out_tx_write(tsi->tsi_env, obj, lbuf, pos, - &tti->tti_tea, + &tti->tti_tea, tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); RETURN(rc); } +static int out_read(struct tgt_session_info *tsi) +{ + const struct lu_env *env = tsi->tsi_env; + struct tgt_thread_info *tti = tgt_th_info(env); + struct object_update *update = tti->tti_u.update.tti_update; + struct dt_object *obj = tti->tti_u.update.tti_dt_object; + struct object_update_reply *reply = tti->tti_u.update.tti_update_reply; + int index = tti->tti_u.update.tti_update_reply_index; + struct lu_rdbuf *rdbuf; + struct object_update_result *update_result; + struct out_read_reply *orr; + void *tmp; + size_t size; + size_t total_size = 0; + __u64 pos; + unsigned int i; + unsigned int nbufs; + int rc = 0; + ENTRY; + + update_result = object_update_result_get(reply, index, NULL); + LASSERT(update_result != NULL); + update_result->our_datalen = sizeof(*orr); + + if (!lu_object_exists(&obj->do_lu)) + GOTO(out, rc = -ENOENT); + + tmp = object_update_param_get(update, 0, NULL); + if (IS_ERR(tmp)) { + CERROR("%s: empty size for read: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(tmp)); + GOTO(out, rc = PTR_ERR(tmp)); + } + size = le64_to_cpu(*(size_t *)(tmp)); + + tmp = object_update_param_get(update, 1, NULL); + if (IS_ERR(tmp)) { + CERROR("%s: empty pos for read: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(tmp)); + GOTO(out, rc = PTR_ERR(tmp)); + } + pos = le64_to_cpu(*(__u64 *)(tmp)); + + /* Put the offset into the begining of the buffer in reply */ + orr = (struct out_read_reply *)update_result->our_data; + + nbufs = (size + OUT_BULK_BUFFER_SIZE - 1) / OUT_BULK_BUFFER_SIZE; + OBD_ALLOC(rdbuf, sizeof(struct lu_rdbuf) + + nbufs * sizeof(rdbuf->rb_bufs[0])); + if (rdbuf == NULL) + GOTO(out, rc = -ENOMEM); + + rdbuf->rb_nbufs = 0; + total_size = 0; + for (i = 0; i < nbufs; i++) { + __u32 read_size; + + read_size = size > OUT_BULK_BUFFER_SIZE ? + OUT_BULK_BUFFER_SIZE : size; + OBD_ALLOC(rdbuf->rb_bufs[i].lb_buf, read_size); + if (rdbuf->rb_bufs[i].lb_buf == NULL) + GOTO(out_free, rc = -ENOMEM); + + rdbuf->rb_bufs[i].lb_len = read_size; + dt_read_lock(env, obj, MOR_TGT_CHILD); + rc = dt_read(env, obj, &rdbuf->rb_bufs[i], &pos); + dt_read_unlock(env, obj); + + total_size += rc < 0 ? 0 : rc; + if (rc <= 0) + break; + + rdbuf->rb_nbufs++; + size -= read_size; + } + + /* send pages to client */ + rc = tgt_send_buffer(tsi, rdbuf); + if (rc < 0) + GOTO(out_free, rc); + + orr->orr_size = total_size; + orr->orr_offset = pos; + + orr_cpu_to_le(orr, orr); + update_result->our_datalen += orr->orr_size; +out_free: + for (i = 0; i < nbufs; i++) { + if (rdbuf->rb_bufs[i].lb_buf != NULL) { + OBD_FREE(rdbuf->rb_bufs[i].lb_buf, + rdbuf->rb_bufs[i].lb_len); + } + } + OBD_FREE(rdbuf, sizeof(struct lu_rdbuf) + + nbufs * sizeof(rdbuf->rb_bufs[0])); +out: + /* Insert read buffer */ + update_result->our_rc = ptlrpc_status_hton(rc); + reply->ourp_lens[index] = cfs_size_round(update_result->our_datalen + + sizeof(*update_result)); + RETURN(rc); +} + +static int out_noop(struct tgt_session_info *tsi) +{ + return 0; +} + #define DEF_OUT_HNDL(opc, name, flags, fn) \ [opc - OUT_CREATE] = { \ .th_name = name, \ @@ -1362,6 +743,8 @@ static struct tgt_handler out_update_ops[] = { DEF_OUT_HNDL(OUT_INDEX_DELETE, "out_index_delete", MUTABOR | HABEO_REFERO, out_index_delete), DEF_OUT_HNDL(OUT_WRITE, "out_write", MUTABOR | HABEO_REFERO, out_write), + DEF_OUT_HNDL(OUT_READ, "out_read", HABEO_REFERO, out_read), + DEF_OUT_HNDL(OUT_NOOP, "out_noop", HABEO_REFERO, out_noop), }; static struct tgt_handler *out_handler_find(__u32 opc) @@ -1479,8 +862,13 @@ static int out_tx_end(const struct lu_env *env, struct thandle_exec_args *ta, dt_obd_name(ta->ta_handle->th_dev), i, ta->ta_argno, rc); } - /* Only fail for real update */ - tsi->tsi_reply_fail_id = OBD_FAIL_OUT_UPDATE_NET_REP; + /* Only fail for real updates, XXX right now llog updates will be + * ignore, whose updates count is usually 1, so failover test + * case will spot this FAIL_UPDATE_NET_REP precisely, and it will + * be removed after async update patch is landed. */ + if (ta->ta_argno > 1) + tsi->tsi_reply_fail_id = OBD_FAIL_OUT_UPDATE_NET_REP; + stop: rc1 = out_trans_stop(env, ta, rc); if (rc == 0) @@ -1510,147 +898,245 @@ int out_handle(struct tgt_session_info *tsi) struct thandle_exec_args *ta = &tti->tti_tea; struct req_capsule *pill = tsi->tsi_pill; struct dt_device *dt = tsi->tsi_tgt->lut_bottom; - struct object_update_request *ureq; + struct out_update_header *ouh; + struct out_update_buffer *oub = NULL; struct object_update *update; struct object_update_reply *reply; - int bufsize; - int count; - int current_batchid = -1; - int i; - int rc = 0; - int rc1 = 0; - + struct ptlrpc_bulk_desc *desc = NULL; + struct l_wait_info lwi; + void **update_bufs; + int current_batchid = -1; + __u32 update_buf_count; + unsigned int i; + unsigned int reply_index = 0; + int rc = 0; + int rc1 = 0; + int ouh_size, reply_size; + int updates; ENTRY; req_capsule_set(pill, &RQF_OUT_UPDATE); - ureq = req_capsule_client_get(pill, &RMF_OUT_UPDATE); - if (ureq == NULL) { - CERROR("%s: No buf!: rc = %d\n", tgt_name(tsi->tsi_tgt), - -EPROTO); + ouh_size = req_capsule_get_size(pill, &RMF_OUT_UPDATE_HEADER, + RCL_CLIENT); + if (ouh_size <= 0) RETURN(err_serious(-EPROTO)); - } - bufsize = req_capsule_get_size(pill, &RMF_OUT_UPDATE, RCL_CLIENT); - if (bufsize != object_update_request_size(ureq)) { - CERROR("%s: invalid bufsize %d: rc = %d\n", - tgt_name(tsi->tsi_tgt), bufsize, -EPROTO); + ouh = req_capsule_client_get(pill, &RMF_OUT_UPDATE_HEADER); + if (ouh == NULL) RETURN(err_serious(-EPROTO)); - } - if (ureq->ourq_magic != UPDATE_REQUEST_MAGIC) { + if (ouh->ouh_magic != OUT_UPDATE_HEADER_MAGIC) { CERROR("%s: invalid update buffer magic %x expect %x: " - "rc = %d\n", tgt_name(tsi->tsi_tgt), ureq->ourq_magic, + "rc = %d\n", tgt_name(tsi->tsi_tgt), ouh->ouh_magic, UPDATE_REQUEST_MAGIC, -EPROTO); RETURN(err_serious(-EPROTO)); } - count = ureq->ourq_count; - if (count <= 0) { - CERROR("%s: empty update: rc = %d\n", tgt_name(tsi->tsi_tgt), - -EPROTO); + update_buf_count = ouh->ouh_count; + if (update_buf_count == 0) RETURN(err_serious(-EPROTO)); + + OBD_ALLOC(update_bufs, sizeof(*update_bufs) * update_buf_count); + if (update_bufs == NULL) + RETURN(err_serious(-ENOMEM)); + + if (ouh->ouh_inline_length > 0) { + update_bufs[0] = ouh->ouh_inline_data; + } else { + struct out_update_buffer *tmp; + + oub = req_capsule_client_get(pill, &RMF_OUT_UPDATE_BUF); + if (oub == NULL) + GOTO(out_free, rc = err_serious(-EPROTO)); + + desc = ptlrpc_prep_bulk_exp(pill->rc_req, update_buf_count, + PTLRPC_BULK_OPS_COUNT, + PTLRPC_BULK_GET_SINK | + PTLRPC_BULK_BUF_KVEC, + MDS_BULK_PORTAL, + &ptlrpc_bulk_kvec_ops); + if (desc == NULL) + GOTO(out_free, rc = err_serious(-ENOMEM)); + + tmp = oub; + for (i = 0; i < update_buf_count; i++, tmp++) { + if (tmp->oub_size >= OUT_MAXREQSIZE) + GOTO(out_free, rc = err_serious(-EPROTO)); + + OBD_ALLOC(update_bufs[i], tmp->oub_size); + if (update_bufs[i] == NULL) + GOTO(out_free, rc = err_serious(-ENOMEM)); + + desc->bd_frag_ops->add_iov_frag(desc, update_bufs[i], + tmp->oub_size); + } + + pill->rc_req->rq_bulk_write = 1; + rc = sptlrpc_svc_prep_bulk(pill->rc_req, desc); + if (rc != 0) + GOTO(out_free, rc = err_serious(rc)); + + rc = target_bulk_io(pill->rc_req->rq_export, desc, &lwi); + if (rc < 0) + GOTO(out_free, rc = err_serious(rc)); + } + /* validate the request and calculate the total update count and + * set it to reply */ + reply_size = 0; + updates = 0; + for (i = 0; i < update_buf_count; i++) { + struct object_update_request *our; + int j; + + our = update_bufs[i]; + if (ptlrpc_req_need_swab(pill->rc_req)) + lustre_swab_object_update_request(our); + + if (our->ourq_magic != UPDATE_REQUEST_MAGIC) { + CERROR("%s: invalid update buffer magic %x" + " expect %x: rc = %d\n", + tgt_name(tsi->tsi_tgt), our->ourq_magic, + UPDATE_REQUEST_MAGIC, -EPROTO); + GOTO(out_free, rc = err_serious(-EPROTO)); + } + updates += our->ourq_count; + + /* need to calculate reply size */ + for (j = 0; j < our->ourq_count; j++) { + update = object_update_request_get(our, j, NULL); + if (update == NULL) + GOTO(out, rc = err_serious(-EPROTO)); + if (ptlrpc_req_need_swab(pill->rc_req)) + lustre_swab_object_update(update); + + if (!fid_is_sane(&update->ou_fid)) { + CERROR("%s: invalid FID "DFID": rc = %d\n", + tgt_name(tsi->tsi_tgt), + PFID(&update->ou_fid), -EPROTO); + GOTO(out, rc = err_serious(-EPROTO)); + } + + /* XXX: what ou_result_size can be considered safe? */ + + reply_size += sizeof(reply->ourp_lens[0]); + reply_size += sizeof(struct object_update_result); + reply_size += update->ou_result_size; + } + } + reply_size += sizeof(*reply); + + if (unlikely(reply_size > ouh->ouh_reply_size)) { + CERROR("%s: too small reply buf %u for %u, need %u at least\n", + tgt_name(tsi->tsi_tgt), ouh->ouh_reply_size, + updates, reply_size); + GOTO(out_free, rc = err_serious(-EPROTO)); } req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER, - OUT_UPDATE_REPLY_SIZE); + ouh->ouh_reply_size); rc = req_capsule_server_pack(pill); if (rc != 0) { CERROR("%s: Can't pack response: rc = %d\n", tgt_name(tsi->tsi_tgt), rc); - RETURN(rc); + GOTO(out_free, rc = err_serious(-EPROTO)); } /* Prepare the update reply buffer */ reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY); if (reply == NULL) - RETURN(err_serious(-EPROTO)); - object_update_reply_init(reply, count); + GOTO(out_free, rc = -EPROTO); + reply->ourp_magic = UPDATE_REPLY_MAGIC; + reply->ourp_count = updates; tti->tti_u.update.tti_update_reply = reply; tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi)); - - /* Walk through updates in the request to execute them synchronously */ - for (i = 0; i < count; i++) { + + /* Walk through updates in the request to execute them */ + for (i = 0; i < update_buf_count; i++) { struct tgt_handler *h; struct dt_object *dt_obj; + int update_count; + struct object_update_request *our; + int j; + + our = update_bufs[i]; + update_count = our->ourq_count; + for (j = 0; j < update_count; j++) { + update = object_update_request_get(our, j, NULL); + + dt_obj = dt_locate(env, dt, &update->ou_fid); + if (IS_ERR(dt_obj)) + GOTO(out, rc = PTR_ERR(dt_obj)); + + if (dt->dd_record_fid_accessed) { + lfsck_pack_rfa(&tti->tti_lr, + lu_object_fid(&dt_obj->do_lu), + LE_FID_ACCESSED, + LFSCK_TYPE_LAYOUT); + tgt_lfsck_in_notify(env, dt, &tti->tti_lr, + NULL); + } - update = object_update_request_get(ureq, i, NULL); - if (update == NULL) - GOTO(out, rc = -EPROTO); - - if (ptlrpc_req_need_swab(pill->rc_req)) - lustre_swab_object_update(update); - - if (!fid_is_sane(&update->ou_fid)) { - CERROR("%s: invalid FID "DFID": rc = %d\n", - tgt_name(tsi->tsi_tgt), PFID(&update->ou_fid), - -EPROTO); - GOTO(out, rc = err_serious(-EPROTO)); - } - - dt_obj = dt_locate(env, dt, &update->ou_fid); - if (IS_ERR(dt_obj)) - GOTO(out, rc = PTR_ERR(dt_obj)); - - if (dt->dd_record_fid_accessed) { - lfsck_pack_rfa(&tti->tti_lr, - lu_object_fid(&dt_obj->do_lu), - LE_FID_ACCESSED, - LFSCK_TYPE_LAYOUT); - tgt_lfsck_in_notify(env, dt, &tti->tti_lr, NULL); - } - - tti->tti_u.update.tti_dt_object = dt_obj; - tti->tti_u.update.tti_update = update; - tti->tti_u.update.tti_update_reply_index = i; + tti->tti_u.update.tti_dt_object = dt_obj; + tti->tti_u.update.tti_update = update; + tti->tti_u.update.tti_update_reply_index = reply_index; - h = out_handler_find(update->ou_type); - if (unlikely(h == NULL)) { - CERROR("%s: unsupported opc: 0x%x\n", - tgt_name(tsi->tsi_tgt), update->ou_type); - GOTO(next, rc = -ENOTSUPP); - } + h = out_handler_find(update->ou_type); + if (unlikely(h == NULL)) { + CERROR("%s: unsupported opc: 0x%x\n", + tgt_name(tsi->tsi_tgt), update->ou_type); + GOTO(next, rc = -ENOTSUPP); + } - /* Check resend case only for modifying RPC */ - if (h->th_flags & MUTABOR) { - struct ptlrpc_request *req = tgt_ses_req(tsi); + /* Check resend case only for modifying RPC */ + if (h->th_flags & MUTABOR) { + struct ptlrpc_request *req = tgt_ses_req(tsi); - if (out_check_resent(env, dt, dt_obj, req, - out_reconstruct, reply, i)) - GOTO(next, rc = 0); - } + if (out_check_resent(env, dt, dt_obj, req, + out_reconstruct, reply, + reply_index)) + GOTO(next, rc = 0); + } - /* start transaction for modification RPC only */ - if (h->th_flags & MUTABOR && current_batchid == -1) { - current_batchid = update->ou_batchid; - rc = out_tx_start(env, dt, ta, tsi->tsi_exp); - if (rc != 0) - GOTO(next, rc); - } + /* start transaction for modification RPC only */ + if (h->th_flags & MUTABOR && current_batchid == -1) { + current_batchid = update->ou_batchid; + rc = out_tx_start(env, dt, ta, tsi->tsi_exp); + if (rc != 0) + GOTO(next, rc); - /* Stop the current update transaction, if the update has - * different batchid, or read-only update */ - if (((current_batchid != update->ou_batchid) || - !(h->th_flags & MUTABOR)) && ta->ta_handle != NULL) { - rc = out_tx_end(env, ta, rc); - current_batchid = -1; - if (rc != 0) - GOTO(next, rc); + if (update->ou_flags & UPDATE_FL_SYNC) + ta->ta_handle->th_sync = 1; + } - /* start a new transaction if needed */ - if (h->th_flags & MUTABOR) { - rc = out_tx_start(env, dt, ta, tsi->tsi_exp); + /* Stop the current update transaction, if the update + * has different batchid, or read-only update */ + if (((current_batchid != update->ou_batchid) || + !(h->th_flags & MUTABOR)) && + ta->ta_handle != NULL) { + rc = out_tx_end(env, ta, rc); + current_batchid = -1; if (rc != 0) GOTO(next, rc); - current_batchid = update->ou_batchid; + /* start a new transaction if needed */ + if (h->th_flags & MUTABOR) { + rc = out_tx_start(env, dt, ta, + tsi->tsi_exp); + if (rc != 0) + GOTO(next, rc); + if (update->ou_flags & UPDATE_FL_SYNC) + ta->ta_handle->th_sync = 1; + current_batchid = update->ou_batchid; + } } - } - rc = h->th_act(tsi); + rc = h->th_act(tsi); next: - lu_object_put(env, &dt_obj->do_lu); - if (rc < 0) - GOTO(out, rc); + reply_index++; + lu_object_put(env, &dt_obj->do_lu); + if (rc < 0) + GOTO(out, rc); + } } out: if (current_batchid != -1) { @@ -1659,6 +1145,21 @@ out: rc = rc1; } +out_free: + if (update_bufs != NULL) { + if (oub != NULL) { + for (i = 0; i < update_buf_count; i++, oub++) { + if (update_bufs[i] != NULL) + OBD_FREE(update_bufs[i], oub->oub_size); + } + } + + OBD_FREE(update_bufs, sizeof(*update_bufs) * update_buf_count); + } + + if (desc != NULL) + ptlrpc_free_bulk(desc); + RETURN(rc); }