X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftarget%2Fout_handler.c;h=6f5dd686351e777e6884e0ec9f7cf98fb7bae892;hp=40e5ce33308bf3aec50bb8f85793e06788774476;hb=86fda8611930a6f9ed73cc5e7813c4809a96ab6d;hpb=1a937b36bb212eb720d504f124d9b135975892d6 diff --git a/lustre/target/out_handler.c b/lustre/target/out_handler.c index 40e5ce3..6f5dd68 100644 --- a/lustre/target/out_handler.c +++ b/lustre/target/out_handler.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2013, Intel Corporation. + * Copyright (c) 2013, 2017, Intel Corporation. * * lustre/target/out_handler.c * @@ -31,80 +31,20 @@ #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) +static inline void orr_cpu_to_le(struct out_read_reply *orr_dst, + const struct out_read_reply *orr_src) { - 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, - 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]; + orr_dst->orr_size = cpu_to_le32(orr_src->orr_size); + orr_dst->orr_padding = cpu_to_le32(orr_src->orr_padding); + orr_dst->orr_offset = cpu_to_le64(orr_dst->orr_offset); } static void out_reconstruct(const struct lu_env *env, struct dt_device *dt, @@ -116,7 +56,6 @@ static void out_reconstruct(const struct lu_env *env, struct dt_device *dt, dt_obd_name(dt), reply, index, 0); object_update_result_insert(reply, NULL, 0, index, 0); - return; } typedef void (*out_reconstruct_t)(const struct lu_env *env, @@ -137,106 +76,26 @@ static inline int out_check_resent(const struct lu_env *env, return 0; if (req_xid_is_last(req)) { - reconstruct(env, dt, obj, reply, index); - return 1; - } - DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")", - req->rq_export->exp_target_data.ted_lcd->lcd_last_xid); - 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 - */ -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; -} - -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; + struct lsd_client_data *lcd; - 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); + /* 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; - 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); + 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); - 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, 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; + DEBUG_REQ(D_RPCTRACE, req, "restoring resent RPC"); + reconstruct(env, dt, obj, reply, index); + return 1; + } + DEBUG_REQ(D_HA, req, "no reply for RESENT req (have %lld)", + req->rq_export->exp_target_data.ted_lcd->lcd_last_xid); return 0; } @@ -250,16 +109,16 @@ static int out_create(struct tgt_session_info *tsi) struct lu_attr *attr = &tti->tti_attr; struct lu_fid *fid = NULL; struct obdo *wobdo; - int size; + size_t size; int rc; ENTRY; wobdo = object_update_param_get(update, 0, &size); - if (wobdo == NULL || 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)) @@ -269,20 +128,18 @@ 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) { - int size; - fid = object_update_param_get(update, 1, &size); - if (fid == NULL || 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); } } @@ -290,72 +147,13 @@ 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, NULL); - 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, 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); @@ -364,16 +162,16 @@ static int out_attr_set(struct tgt_session_info *tsi) struct dt_object *obj = tti->tti_u.update.tti_dt_object; struct obdo *lobdo = &tti->tti_u.update.tti_obdo; struct obdo *wobdo; - int size; + size_t size; int rc; ENTRY; wobdo = object_update_param_get(update, 0, &size); - if (wobdo == NULL || 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; @@ -385,6 +183,7 @@ 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); @@ -395,6 +194,7 @@ 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; @@ -403,6 +203,9 @@ 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 @@ -413,50 +216,13 @@ static int out_attr_get(struct tgt_session_info *tsi) RETURN(-ENOENT); } - dt_read_lock(env, obj, MOR_TGT_CHILD); - rc = dt_attr_get(env, obj, la, NULL); + dt_read_lock(env, obj, DT_TGT_CHILD); + rc = dt_attr_get(env, obj, la); if (rc) GOTO(out_unlock, rc); - /* - * If it is a directory, we will also check whether the - * directory is empty. - * la_flags = 0 : Empty. - * = 1 : Not empty. - */ - la->la_flags = 0; - if (S_ISDIR(la->la_mode)) { - struct dt_it *it; - const struct dt_it_ops *iops; - - if (!dt_try_as_dir(env, obj)) - GOTO(out_unlock, rc = -ENOTDIR); - - iops = &obj->do_index_ops->dio_it; - it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA); - if (!IS_ERR(it)) { - int result; - result = iops->get(env, it, (const void *)""); - if (result > 0) { - int i; - for (result = 0, i = 0; result == 0 && i < 3; - ++i) - result = iops->next(env, it); - if (result == 0) - la->la_flags = 1; - } else if (result == 0) - /* - * Huh? Index contains no zero key? - */ - rc = -EIO; - - iops->put(env, it); - iops->fini(env, it); - } - } obdo->o_valid = 0; obdo_from_la(obdo, la, la->la_valid); - lustre_set_wire_obdo(NULL, obdo, obdo); out_unlock: dt_read_unlock(env, obj); @@ -493,45 +259,97 @@ static int out_xattr_get(struct tgt_session_info *tsi) } name = object_update_param_get(update, 0, NULL); - if (name == NULL) { - 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); + update_result = object_update_result_get(reply, idx, 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_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); - dt_read_lock(env, obj, MOR_TGT_CHILD); - rc = dt_xattr_get(env, obj, lbuf, name, NULL); + lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size; + if (lbuf->lb_len == 0) + lbuf->lb_buf = NULL; + else + lbuf->lb_buf = update_result->our_data; + + dt_read_lock(env, obj, DT_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); + if (unlikely(!rc)) + rc = -ENODATA; + } else if (lbuf->lb_buf) { + lbuf->lb_len = rc; } - if (rc == 0) { - lbuf->lb_len = 0; - GOTO(out, rc = -ENOENT); - } - 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); + name, rc); GOTO(out, rc); out: object_update_result_insert(reply, lbuf->lb_buf, lbuf->lb_len, idx, rc); - RETURN(rc); + RETURN(0); +} + +static int out_xattr_list(struct tgt_session_info *tsi) +{ + const struct lu_env *env = tsi->tsi_env; + struct tgt_thread_info *tti = tgt_th_info(env); + struct lu_buf *lbuf = &tti->tti_buf; + struct object_update_reply *reply = tti->tti_u.update.tti_update_reply; + struct dt_object *obj = tti->tti_u.update.tti_dt_object; + struct object_update_result *update_result; + int idx = tti->tti_u.update.tti_update_reply_index; + int rc; + + ENTRY; + + if (!lu_object_exists(&obj->do_lu)) { + set_bit(LU_OBJECT_HEARD_BANSHEE, + &obj->do_lu.lo_header->loh_flags); + RETURN(-ENOENT); + } + + update_result = object_update_result_get(reply, 0, NULL); + if (!update_result) { + rc = -EPROTO; + CERROR("%s: empty buf for xattr list: rc = %d\n", + tgt_name(tsi->tsi_tgt), rc); + RETURN(rc); + } + + lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size; + lbuf->lb_buf = update_result->our_data; + if (lbuf->lb_len == 0) + lbuf->lb_buf = 0; + + dt_read_lock(env, obj, DT_TGT_CHILD); + rc = dt_xattr_list(env, obj, lbuf); + dt_read_unlock(env, obj); + if (rc <= 0) { + lbuf->lb_len = 0; + if (unlikely(!rc)) + rc = -ENODATA; + } else if (lbuf->lb_buf) { + lbuf->lb_len = rc; + } + + CDEBUG(D_INFO, "%s: "DFID" list xattr len %d\n", + tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), rc); + + /* Since we directly use update_result->our_data as the lbuf->lb_buf, + * then use NULL for result_insert to avoid unnecessary memory copy. */ + object_update_result_insert(reply, NULL, lbuf->lb_len, idx, rc); + + RETURN(0); } static int out_index_lookup(struct tgt_session_info *tsi) @@ -545,22 +363,25 @@ 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) { - 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); + dt_read_lock(env, obj, DT_TGT_CHILD); if (!dt_try_as_dir(env, obj)) GOTO(out_unlock, rc = -ENOTDIR); rc = dt_lookup(env, obj, (struct dt_rec *)&tti->tti_fid1, - (struct dt_key *)name, NULL); + (struct dt_key *)name); if (rc < 0) GOTO(out_unlock, rc); @@ -585,70 +406,6 @@ 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, NULL); - dt_write_unlock(env, dt_obj); - /** - * Ignore errors if this is LINK EA - **/ - if (unlikely(rc && !strcmp(arg->u.xattr_set.name, XATTR_NAME_LINK))) - rc = 0; -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, 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; - - 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); @@ -658,34 +415,32 @@ static int out_xattr_set(struct tgt_session_info *tsi) char *name; char *buf; __u32 *tmp; - int buf_len = 0; + size_t buf_len = 0; int flag; - int size = 0; + size_t size = 0; int rc; ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL) { - 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 (buf == NULL || 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) && 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 %d 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)) @@ -693,83 +448,33 @@ 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_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) +static int out_xattr_del(struct tgt_session_info *tsi) { - 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, 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; + struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); + struct object_update *update = tti->tti_u.update.tti_update; + struct dt_object *obj = tti->tti_u.update.tti_dt_object; + char *name; + int rc; + ENTRY; - 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); + name = object_update_param_get(update, 0, NULL); + 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)); + } - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - return 0; + 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); } /** @@ -784,59 +489,12 @@ 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, 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); @@ -849,137 +507,37 @@ 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"\n", - dt_obd_name(th->th_dev), PFID(lu_object_fid(&dt_obj->do_lu)), - (char *)key, PFID((struct lu_fid *)rec)); - - 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, NULL, 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, NULL); - 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; - - rc = out_obj_index_insert(env, dt_obj, 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, - char *name, struct lu_fid *fid, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, 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, (struct dt_rec *)fid, - (struct dt_key *)name, 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 = (struct dt_rec *)fid; - arg->u.insert.key = (struct dt_key *)name; - - return 0; -} - static int out_index_insert(struct tgt_session_info *tsi) { - struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); - struct object_update *update = tti->tti_u.update.tti_update; - struct dt_object *obj = tti->tti_u.update.tti_dt_object; - struct lu_fid *fid; - char *name; - int rc = 0; - int size; - + struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); + struct object_update *update = tti->tti_u.update.tti_update; + struct dt_object *obj = tti->tti_u.update.tti_dt_object; + struct dt_insert_rec *rec = &tti->tti_rec; + struct lu_fid *fid; + char *name; + __u32 *ptype; + int rc = 0; + size_t size; ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL) { - 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 || 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)) @@ -988,72 +546,33 @@ 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); } - rc = out_tx_index_insert(tsi->tsi_env, obj, name, fid, - &tti->tti_tea, - 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: 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_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; -} + ptype = object_update_param_get(update, 2, &size); + 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)); + } -static int __out_tx_index_delete(const struct lu_env *env, - struct dt_object *dt_obj, char *name, - struct thandle_exec_args *ta, - struct object_update_reply *reply, - int index, char *file, int line) -{ - struct tx_arg *arg; - int rc; + if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) + __swab32s(ptype); - if (dt_try_as_dir(env, dt_obj) == 0) { - rc = -ENOTDIR; - return rc; - } + rec->rec_fid = fid; + rec->rec_type = *ptype; - LASSERT(ta->ta_handle != NULL); - rc = dt_declare_delete(env, dt_obj, (struct dt_key *)name, - ta->ta_handle); - if (rc != 0) - return rc; + 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); - 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); + CDEBUG(D_INFO, "%s: "DFID" index insert %s: rc = %d\n", + tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), + name, rc); - lu_object_get(&dt_obj->do_lu); - arg->object = dt_obj; - arg->reply = reply; - arg->index = index; - arg->u.insert.key = (struct dt_key *)name; - return 0; + RETURN(rc); } static int out_index_delete(struct tgt_session_info *tsi) @@ -1068,67 +587,19 @@ static int out_index_delete(struct tgt_session_info *tsi) RETURN(-ENOENT); name = object_update_param_get(update, 0, NULL); - if (name == NULL) { - 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, name, &tti->tti_tea, + rc = out_tx_index_delete(tsi->tsi_env, obj, (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_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, 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); @@ -1142,66 +613,20 @@ 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, 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); @@ -1210,26 +635,26 @@ static int out_write(struct tgt_session_info *tsi) struct lu_buf *lbuf = &tti->tti_buf; char *buf; __u64 *tmp; - int size = 0; - int buf_len = 0; + size_t size = 0; + size_t buf_len = 0; loff_t pos; int rc; ENTRY; buf = object_update_param_get(update, 0, &buf_len); - if (buf == NULL || 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 || size != sizeof(*tmp)) { - CERROR("%s: empty or wrong size %d 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)) @@ -1237,12 +662,119 @@ 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(*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, DT_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(*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, \ @@ -1255,32 +787,38 @@ static int out_write(struct tgt_session_info *tsi) } static struct tgt_handler out_update_ops[] = { - DEF_OUT_HNDL(OUT_CREATE, "out_create", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_CREATE, "out_create", IS_MUTABLE | HAS_REPLY, out_create), - DEF_OUT_HNDL(OUT_DESTROY, "out_create", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_DESTROY, "out_create", IS_MUTABLE | HAS_REPLY, out_destroy), - DEF_OUT_HNDL(OUT_REF_ADD, "out_ref_add", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_REF_ADD, "out_ref_add", IS_MUTABLE | HAS_REPLY, out_ref_add), - DEF_OUT_HNDL(OUT_REF_DEL, "out_ref_del", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_REF_DEL, "out_ref_del", IS_MUTABLE | HAS_REPLY, out_ref_del), - DEF_OUT_HNDL(OUT_ATTR_SET, "out_attr_set", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_ATTR_SET, "out_attr_set", IS_MUTABLE | HAS_REPLY, out_attr_set), - DEF_OUT_HNDL(OUT_ATTR_GET, "out_attr_get", HABEO_REFERO, + DEF_OUT_HNDL(OUT_ATTR_GET, "out_attr_get", HAS_REPLY, out_attr_get), - DEF_OUT_HNDL(OUT_XATTR_SET, "out_xattr_set", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_XATTR_SET, "out_xattr_set", IS_MUTABLE | HAS_REPLY, out_xattr_set), - DEF_OUT_HNDL(OUT_XATTR_GET, "out_xattr_get", HABEO_REFERO, + DEF_OUT_HNDL(OUT_XATTR_DEL, "out_xattr_del", IS_MUTABLE | HAS_REPLY, + out_xattr_del), + DEF_OUT_HNDL(OUT_XATTR_GET, "out_xattr_get", HAS_REPLY, out_xattr_get), - DEF_OUT_HNDL(OUT_INDEX_LOOKUP, "out_index_lookup", HABEO_REFERO, + DEF_OUT_HNDL(OUT_INDEX_LOOKUP, "out_index_lookup", HAS_REPLY, out_index_lookup), DEF_OUT_HNDL(OUT_INDEX_INSERT, "out_index_insert", - MUTABOR | HABEO_REFERO, out_index_insert), + IS_MUTABLE | HAS_REPLY, out_index_insert), 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), + IS_MUTABLE | HAS_REPLY, out_index_delete), + DEF_OUT_HNDL(OUT_WRITE, "out_write", IS_MUTABLE | HAS_REPLY, out_write), + DEF_OUT_HNDL(OUT_READ, "out_read", HAS_REPLY, out_read), + DEF_OUT_HNDL(OUT_NOOP, "out_noop", HAS_REPLY, out_noop), + DEF_OUT_HNDL(OUT_XATTR_LIST, "out_xattr_list", HAS_REPLY, + out_xattr_list), }; -struct tgt_handler *out_handler_find(__u32 opc) +static struct tgt_handler *out_handler_find(__u32 opc) { struct tgt_handler *h; @@ -1331,17 +869,7 @@ static int out_trans_stop(const struct lu_env *env, rc = dt_trans_stop(env, ta->ta_handle->th_dev, ta->ta_handle); for (i = 0; i < ta->ta_argno; i++) { if (ta->ta_args[i]->object != NULL) { - struct dt_object *obj = ta->ta_args[i]->object; - - /* If the object is being created during this - * transaction, we need to remove them from the - * cache immediately, because a few layers are - * missing in OUT handler, i.e. the object might - * not be initialized in all layers */ - if (ta->ta_args[i]->exec_fn == out_tx_create_exec) - set_bit(LU_OBJECT_HEARD_BANSHEE, - &obj->do_lu.lo_header->loh_flags); - lu_object_put(env, &ta->ta_args[i]->object->do_lu); + dt_object_put(env, ta->ta_args[i]->object); ta->ta_args[i]->object = NULL; } } @@ -1351,8 +879,8 @@ static int out_trans_stop(const struct lu_env *env, return rc; } -int out_tx_end(const struct lu_env *env, struct thandle_exec_args *ta, - int declare_ret) +static int out_tx_end(const struct lu_env *env, struct thandle_exec_args *ta, + int declare_ret) { struct tgt_session_info *tsi = tgt_ses_info(env); int i; @@ -1395,8 +923,13 @@ 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) @@ -1426,145 +959,261 @@ 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; + 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; + int page_count = 0; + + oub = req_capsule_client_get(pill, &RMF_OUT_UPDATE_BUF); + if (oub == NULL) + GOTO(out_free, rc = err_serious(-EPROTO)); + + for (i = 0; i < update_buf_count; i++) + /* First *and* last might be partial pages, hence +1 */ + page_count += DIV_ROUND_UP(oub[i].oub_size, + PAGE_SIZE) + 1; + + desc = ptlrpc_prep_bulk_exp(pill->rc_req, page_count, + PTLRPC_BULK_OPS_COUNT, + PTLRPC_BULK_GET_SINK | + PTLRPC_BULK_BUF_KIOV, + MDS_BULK_PORTAL, + &ptlrpc_bulk_kiov_nopin_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_LARGE(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); + 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++) { + struct lu_object_conf conf; + + update = object_update_request_get(our, j, NULL); + if (update->ou_type == OUT_CREATE) + conf.loc_flags = LOC_F_NEW; + else + conf.loc_flags = 0; + + dt_obj = dt_locate_at(env, dt, &update->ou_fid, + dt->dd_lu_dev.ld_site->ls_top_dev, &conf); + if (IS_ERR(dt_obj)) + GOTO(out, rc = PTR_ERR(dt_obj)); + + if (dt->dd_record_fid_accessed) { + struct lfsck_req_local *lrl = &tti->tti_lrl; + + lfsck_pack_rfa(lrl, + lu_object_fid(&dt_obj->do_lu), + LEL_FID_ACCESSED, + LFSCK_TYPE_LAYOUT); + tgt_lfsck_in_notify_local(env, dt, lrl, 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)); + 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; - if (dt->dd_record_fid_accessed) { - lfsck_pack_rfa(&tti->tti_lr, - lu_object_fid(&dt_obj->do_lu)); - tgt_lfsck_in_notify(env, dt, &tti->tti_lr); - } + 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); + } - 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; + /* Check resend case only for modifying RPC */ + if (h->th_flags & IS_MUTABLE) { + struct ptlrpc_request *req = tgt_ses_req(tsi); - 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); - } + if (out_check_resent(env, dt, dt_obj, req, + out_reconstruct, reply, + reply_index)) + GOTO(next, rc = 0); - /* Check resend case only for modifying RPC */ - if (h->th_flags & MUTABOR) { - struct ptlrpc_request *req = tgt_ses_req(tsi); + if (dt->dd_rdonly) + GOTO(next, rc = -EROFS); + } - if (out_check_resent(env, dt, dt_obj, req, - out_reconstruct, reply, i)) - GOTO(next, rc = 0); - } + /* start transaction for modification RPC only */ + if (h->th_flags & IS_MUTABLE && 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); - } + if (update->ou_flags & UPDATE_FL_SYNC) + ta->ta_handle->th_sync = 1; + } - /* 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); - - /* 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 & IS_MUTABLE)) && + 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 & IS_MUTABLE) { + 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++; + dt_object_put(env, dt_obj); + if (rc < 0) + GOTO(out, rc); + } } out: if (current_batchid != -1) { @@ -1573,11 +1222,27 @@ 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_LARGE(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); } struct tgt_handler tgt_out_handlers[] = { -TGT_UPDATE_HDL(MUTABOR, OUT_UPDATE, out_handle), +TGT_UPDATE_HDL(IS_MUTABLE, OUT_UPDATE, out_handle), }; EXPORT_SYMBOL(tgt_out_handlers);