4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2013, 2017, Intel Corporation.
25 * lustre/target/out_handler.c
27 * Object update handler between targets.
29 * Author: di.wang <di.wang@intel.com>
32 #define DEBUG_SUBSYSTEM S_CLASS
34 #include <llog_swab.h>
35 #include <lustre_obdo.h>
36 #include <lustre_swab.h>
37 #include <lustre_update.h>
38 #include <md_object.h>
39 #include <obd_class.h>
40 #include "tgt_internal.h"
42 static inline void orr_cpu_to_le(struct out_read_reply *orr_dst,
43 const struct out_read_reply *orr_src)
45 orr_dst->orr_size = cpu_to_le32(orr_src->orr_size);
46 orr_dst->orr_padding = cpu_to_le32(orr_src->orr_padding);
47 orr_dst->orr_offset = cpu_to_le64(orr_dst->orr_offset);
50 static void out_reconstruct(const struct lu_env *env, struct dt_device *dt,
51 struct dt_object *obj,
52 struct object_update_reply *reply,
55 CDEBUG(D_INFO, "%s: fork reply reply %p index %d: rc = %d\n",
56 dt_obd_name(dt), reply, index, 0);
58 object_update_result_insert(reply, NULL, 0, index, 0);
62 typedef void (*out_reconstruct_t)(const struct lu_env *env,
64 struct dt_object *obj,
65 struct object_update_reply *reply,
68 static inline int out_check_resent(const struct lu_env *env,
70 struct dt_object *obj,
71 struct ptlrpc_request *req,
72 out_reconstruct_t reconstruct,
73 struct object_update_reply *reply,
76 if (likely(!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)))
79 if (req_xid_is_last(req)) {
80 struct lsd_client_data *lcd;
82 /* XXX this does not support mulitple transactions yet, i.e.
83 * only 1 update RPC each time betwee MDTs */
84 lcd = req->rq_export->exp_target_data.ted_lcd;
86 req->rq_transno = lcd->lcd_last_transno;
87 req->rq_status = lcd->lcd_last_result;
88 if (req->rq_status != 0)
90 lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
91 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
93 DEBUG_REQ(D_RPCTRACE, req, "restoring resent RPC");
95 reconstruct(env, dt, obj, reply, index);
98 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have %lld)",
99 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
103 static int out_create(struct tgt_session_info *tsi)
105 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
106 struct object_update *update = tti->tti_u.update.tti_update;
107 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
108 struct dt_object_format *dof = &tti->tti_u.update.tti_update_dof;
109 struct obdo *lobdo = &tti->tti_u.update.tti_obdo;
110 struct lu_attr *attr = &tti->tti_attr;
111 struct lu_fid *fid = NULL;
118 wobdo = object_update_param_get(update, 0, &size);
119 if (IS_ERR(wobdo) || size != sizeof(*wobdo)) {
120 CERROR("%s: obdo is NULL, invalid RPC: rc = %ld\n",
121 tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo));
122 RETURN(PTR_ERR(wobdo));
125 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
126 lustre_swab_obdo(wobdo);
127 lustre_get_wire_obdo(NULL, lobdo, wobdo);
128 la_from_obdo(attr, lobdo, lobdo->o_valid);
130 dof->dof_type = dt_mode_to_dft(attr->la_mode);
131 if (update->ou_params_count > 1) {
132 fid = object_update_param_get(update, 1, &size);
133 if (IS_ERR(fid) || size != sizeof(*fid)) {
134 CERROR("%s: invalid fid: rc = %ld\n",
135 tgt_name(tsi->tsi_tgt), PTR_ERR(fid));
136 RETURN(PTR_ERR(fid));
138 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
139 lustre_swab_lu_fid(fid);
140 if (!fid_is_sane(fid)) {
141 CERROR("%s: invalid fid "DFID": rc = %d\n",
142 tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO);
147 if (lu_object_exists(&obj->do_lu))
150 rc = out_tx_create(tsi->tsi_env, obj, attr, fid, dof,
151 &tti->tti_tea, tti->tti_tea.ta_handle,
152 tti->tti_u.update.tti_update_reply,
153 tti->tti_u.update.tti_update_reply_index);
158 static int out_attr_set(struct tgt_session_info *tsi)
160 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
161 struct object_update *update = tti->tti_u.update.tti_update;
162 struct lu_attr *attr = &tti->tti_attr;
163 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
164 struct obdo *lobdo = &tti->tti_u.update.tti_obdo;
171 wobdo = object_update_param_get(update, 0, &size);
172 if (IS_ERR(wobdo) || size != sizeof(*wobdo)) {
173 CERROR("%s: empty obdo in the update: rc = %ld\n",
174 tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo));
175 RETURN(PTR_ERR(wobdo));
181 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
182 lustre_swab_obdo(wobdo);
183 lustre_get_wire_obdo(NULL, lobdo, wobdo);
184 la_from_obdo(attr, lobdo, lobdo->o_valid);
186 rc = out_tx_attr_set(tsi->tsi_env, obj, attr, &tti->tti_tea,
187 tti->tti_tea.ta_handle,
188 tti->tti_u.update.tti_update_reply,
189 tti->tti_u.update.tti_update_reply_index);
194 static int out_attr_get(struct tgt_session_info *tsi)
196 const struct lu_env *env = tsi->tsi_env;
197 struct tgt_thread_info *tti = tgt_th_info(env);
198 struct object_update *update = tti->tti_u.update.tti_update;
199 struct obdo *obdo = &tti->tti_u.update.tti_obdo;
200 struct lu_attr *la = &tti->tti_attr;
201 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
202 int idx = tti->tti_u.update.tti_update_reply_index;
207 if (unlikely(update->ou_result_size < sizeof(*obdo)))
210 if (!lu_object_exists(&obj->do_lu)) {
211 /* Usually, this will be called when the master MDT try
212 * to init a remote object(see osp_object_init), so if
213 * the object does not exist on slave, we need set BANSHEE flag,
214 * so the object can be removed from the cache immediately */
215 set_bit(LU_OBJECT_HEARD_BANSHEE,
216 &obj->do_lu.lo_header->loh_flags);
220 dt_read_lock(env, obj, MOR_TGT_CHILD);
221 rc = dt_attr_get(env, obj, la);
223 GOTO(out_unlock, rc);
226 obdo_from_la(obdo, la, la->la_valid);
229 dt_read_unlock(env, obj);
231 CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
232 tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
235 object_update_result_insert(tti->tti_u.update.tti_update_reply, obdo,
236 sizeof(*obdo), idx, rc);
241 static int out_xattr_get(struct tgt_session_info *tsi)
243 const struct lu_env *env = tsi->tsi_env;
244 struct tgt_thread_info *tti = tgt_th_info(env);
245 struct object_update *update = tti->tti_u.update.tti_update;
246 struct lu_buf *lbuf = &tti->tti_buf;
247 struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
248 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
250 struct object_update_result *update_result;
251 int idx = tti->tti_u.update.tti_update_reply_index;
256 if (!lu_object_exists(&obj->do_lu)) {
257 set_bit(LU_OBJECT_HEARD_BANSHEE,
258 &obj->do_lu.lo_header->loh_flags);
262 name = object_update_param_get(update, 0, NULL);
264 CERROR("%s: empty name for xattr get: rc = %ld\n",
265 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
266 RETURN(PTR_ERR(name));
269 update_result = object_update_result_get(reply, idx, NULL);
270 if (update_result == NULL) {
271 CERROR("%s: empty name for xattr get: rc = %d\n",
272 tgt_name(tsi->tsi_tgt), -EPROTO);
276 lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size;
277 if (lbuf->lb_len == 0)
280 lbuf->lb_buf = update_result->our_data;
282 dt_read_lock(env, obj, MOR_TGT_CHILD);
283 rc = dt_xattr_get(env, obj, lbuf, name);
284 dt_read_unlock(env, obj);
289 } else if (lbuf->lb_buf) {
292 CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n",
293 tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)),
299 object_update_result_insert(reply, lbuf->lb_buf, lbuf->lb_len, idx, rc);
303 static int out_xattr_list(struct tgt_session_info *tsi)
305 const struct lu_env *env = tsi->tsi_env;
306 struct tgt_thread_info *tti = tgt_th_info(env);
307 struct lu_buf *lbuf = &tti->tti_buf;
308 struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
309 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
310 struct object_update_result *update_result;
311 int idx = tti->tti_u.update.tti_update_reply_index;
316 if (!lu_object_exists(&obj->do_lu)) {
317 set_bit(LU_OBJECT_HEARD_BANSHEE,
318 &obj->do_lu.lo_header->loh_flags);
322 update_result = object_update_result_get(reply, 0, NULL);
323 if (!update_result) {
325 CERROR("%s: empty buf for xattr list: rc = %d\n",
326 tgt_name(tsi->tsi_tgt), rc);
330 lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size;
331 lbuf->lb_buf = update_result->our_data;
332 if (lbuf->lb_len == 0)
335 dt_read_lock(env, obj, MOR_TGT_CHILD);
336 rc = dt_xattr_list(env, obj, lbuf);
337 dt_read_unlock(env, obj);
342 } else if (lbuf->lb_buf) {
346 CDEBUG(D_INFO, "%s: "DFID" list xattr len %d\n",
347 tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), rc);
349 /* Since we directly use update_result->our_data as the lbuf->lb_buf,
350 * then use NULL for result_insert to avoid unnecessary memory copy. */
351 object_update_result_insert(reply, NULL, lbuf->lb_len, idx, rc);
356 static int out_index_lookup(struct tgt_session_info *tsi)
358 const struct lu_env *env = tsi->tsi_env;
359 struct tgt_thread_info *tti = tgt_th_info(env);
360 struct object_update *update = tti->tti_u.update.tti_update;
361 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
367 if (unlikely(update->ou_result_size < sizeof(tti->tti_fid1)))
370 if (!lu_object_exists(&obj->do_lu))
373 name = object_update_param_get(update, 0, NULL);
375 CERROR("%s: empty name for lookup: rc = %ld\n",
376 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
377 RETURN(PTR_ERR(name));
380 dt_read_lock(env, obj, MOR_TGT_CHILD);
381 if (!dt_try_as_dir(env, obj))
382 GOTO(out_unlock, rc = -ENOTDIR);
384 rc = dt_lookup(env, obj, (struct dt_rec *)&tti->tti_fid1,
385 (struct dt_key *)name);
388 GOTO(out_unlock, rc);
394 dt_read_unlock(env, obj);
396 CDEBUG(D_INFO, "lookup "DFID" %s get "DFID" rc %d\n",
397 PFID(lu_object_fid(&obj->do_lu)), name,
398 PFID(&tti->tti_fid1), rc);
400 CDEBUG(D_INFO, "%s: insert lookup reply %p index %d: rc = %d\n",
401 tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
404 object_update_result_insert(tti->tti_u.update.tti_update_reply,
405 &tti->tti_fid1, sizeof(tti->tti_fid1),
406 tti->tti_u.update.tti_update_reply_index, rc);
410 static int out_xattr_set(struct tgt_session_info *tsi)
412 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
413 struct object_update *update = tti->tti_u.update.tti_update;
414 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
415 struct lu_buf *lbuf = &tti->tti_buf;
425 name = object_update_param_get(update, 0, NULL);
427 CERROR("%s: empty name for xattr set: rc = %ld\n",
428 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
429 RETURN(PTR_ERR(name));
432 /* If buffer == NULL (-ENODATA), then it might mean delete xattr */
433 buf = object_update_param_get(update, 1, &buf_len);
434 if (IS_ERR(buf) && PTR_ERR(buf) != -ENODATA)
435 RETURN(PTR_ERR(buf));
438 lbuf->lb_len = buf_len;
440 tmp = object_update_param_get(update, 2, &size);
441 if (IS_ERR(tmp) || size != sizeof(*tmp)) {
442 CERROR("%s: emptry or wrong size %zu flag: rc = %ld\n",
443 tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp));
444 RETURN(PTR_ERR(tmp));
447 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
451 rc = out_tx_xattr_set(tsi->tsi_env, obj, lbuf, name, flag,
452 &tti->tti_tea, tti->tti_tea.ta_handle,
453 tti->tti_u.update.tti_update_reply,
454 tti->tti_u.update.tti_update_reply_index);
458 static int out_xattr_del(struct tgt_session_info *tsi)
460 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
461 struct object_update *update = tti->tti_u.update.tti_update;
462 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
467 name = object_update_param_get(update, 0, NULL);
469 CERROR("%s: empty name for xattr set: rc = %ld\n",
470 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
471 RETURN(PTR_ERR(name));
474 rc = out_tx_xattr_del(tsi->tsi_env, obj, name, &tti->tti_tea,
475 tti->tti_tea.ta_handle,
476 tti->tti_u.update.tti_update_reply,
477 tti->tti_u.update.tti_update_reply_index);
482 * increase ref of the object
484 static int out_ref_add(struct tgt_session_info *tsi)
486 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
487 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
492 rc = out_tx_ref_add(tsi->tsi_env, obj, &tti->tti_tea,
493 tti->tti_tea.ta_handle,
494 tti->tti_u.update.tti_update_reply,
495 tti->tti_u.update.tti_update_reply_index);
499 static int out_ref_del(struct tgt_session_info *tsi)
501 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
502 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
507 if (!lu_object_exists(&obj->do_lu))
510 rc = out_tx_ref_del(tsi->tsi_env, obj, &tti->tti_tea,
511 tti->tti_tea.ta_handle,
512 tti->tti_u.update.tti_update_reply,
513 tti->tti_u.update.tti_update_reply_index);
517 static int out_index_insert(struct tgt_session_info *tsi)
519 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
520 struct object_update *update = tti->tti_u.update.tti_update;
521 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
522 struct dt_insert_rec *rec = &tti->tti_rec;
530 name = object_update_param_get(update, 0, NULL);
532 CERROR("%s: empty name for index insert: rc = %ld\n",
533 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
534 RETURN(PTR_ERR(name));
537 fid = object_update_param_get(update, 1, &size);
538 if (IS_ERR(fid) || size != sizeof(*fid)) {
539 CERROR("%s: invalid fid: rc = %ld\n",
540 tgt_name(tsi->tsi_tgt), PTR_ERR(fid));
541 RETURN(PTR_ERR(fid));
544 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
545 lustre_swab_lu_fid(fid);
547 if (!fid_is_sane(fid)) {
548 CERROR("%s: invalid FID "DFID": rc = %d\n",
549 tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO);
553 ptype = object_update_param_get(update, 2, &size);
554 if (IS_ERR(ptype) || size != sizeof(*ptype)) {
555 CERROR("%s: invalid type for index insert: rc = %ld\n",
556 tgt_name(tsi->tsi_tgt), PTR_ERR(ptype));
557 RETURN(PTR_ERR(ptype));
560 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
564 rec->rec_type = *ptype;
566 rc = out_tx_index_insert(tsi->tsi_env, obj, (const struct dt_rec *)rec,
567 (const struct dt_key *)name, &tti->tti_tea,
568 tti->tti_tea.ta_handle,
569 tti->tti_u.update.tti_update_reply,
570 tti->tti_u.update.tti_update_reply_index);
572 CDEBUG(D_INFO, "%s: "DFID" index insert %s: rc = %d\n",
573 tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)),
579 static int out_index_delete(struct tgt_session_info *tsi)
581 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
582 struct object_update *update = tti->tti_u.update.tti_update;
583 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
587 if (!lu_object_exists(&obj->do_lu))
590 name = object_update_param_get(update, 0, NULL);
592 CERROR("%s: empty name for index delete: rc = %ld\n",
593 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
594 RETURN(PTR_ERR(name));
597 rc = out_tx_index_delete(tsi->tsi_env, obj, (const struct dt_key *)name,
598 &tti->tti_tea, tti->tti_tea.ta_handle,
599 tti->tti_u.update.tti_update_reply,
600 tti->tti_u.update.tti_update_reply_index);
604 static int out_destroy(struct tgt_session_info *tsi)
606 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
607 struct object_update *update = tti->tti_u.update.tti_update;
608 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
613 fid = &update->ou_fid;
614 if (!fid_is_sane(fid)) {
615 CERROR("%s: invalid FID "DFID": rc = %d\n",
616 tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO);
620 if (!lu_object_exists(&obj->do_lu))
623 rc = out_tx_destroy(tsi->tsi_env, obj, &tti->tti_tea,
624 tti->tti_tea.ta_handle,
625 tti->tti_u.update.tti_update_reply,
626 tti->tti_u.update.tti_update_reply_index);
631 static int out_write(struct tgt_session_info *tsi)
633 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
634 struct object_update *update = tti->tti_u.update.tti_update;
635 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
636 struct lu_buf *lbuf = &tti->tti_buf;
645 buf = object_update_param_get(update, 0, &buf_len);
646 if (IS_ERR(buf) || buf_len == 0) {
647 CERROR("%s: empty buf for xattr set: rc = %ld\n",
648 tgt_name(tsi->tsi_tgt), PTR_ERR(buf));
649 RETURN(PTR_ERR(buf));
652 lbuf->lb_len = buf_len;
654 tmp = object_update_param_get(update, 1, &size);
655 if (IS_ERR(tmp) || size != sizeof(*tmp)) {
656 CERROR("%s: empty or wrong size %zu pos: rc = %ld\n",
657 tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp));
658 RETURN(PTR_ERR(tmp));
661 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
665 rc = out_tx_write(tsi->tsi_env, obj, lbuf, pos,
666 &tti->tti_tea, tti->tti_tea.ta_handle,
667 tti->tti_u.update.tti_update_reply,
668 tti->tti_u.update.tti_update_reply_index);
672 static int out_read(struct tgt_session_info *tsi)
674 const struct lu_env *env = tsi->tsi_env;
675 struct tgt_thread_info *tti = tgt_th_info(env);
676 struct object_update *update = tti->tti_u.update.tti_update;
677 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
678 struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
679 int index = tti->tti_u.update.tti_update_reply_index;
680 struct lu_rdbuf *rdbuf;
681 struct object_update_result *update_result;
682 struct out_read_reply *orr;
685 size_t total_size = 0;
692 update_result = object_update_result_get(reply, index, NULL);
693 LASSERT(update_result != NULL);
694 update_result->our_datalen = sizeof(*orr);
696 if (!lu_object_exists(&obj->do_lu))
697 GOTO(out, rc = -ENOENT);
699 tmp = object_update_param_get(update, 0, NULL);
701 CERROR("%s: empty size for read: rc = %ld\n",
702 tgt_name(tsi->tsi_tgt), PTR_ERR(tmp));
703 GOTO(out, rc = PTR_ERR(tmp));
705 size = le64_to_cpu(*(size_t *)(tmp));
707 tmp = object_update_param_get(update, 1, NULL);
709 CERROR("%s: empty pos for read: rc = %ld\n",
710 tgt_name(tsi->tsi_tgt), PTR_ERR(tmp));
711 GOTO(out, rc = PTR_ERR(tmp));
713 pos = le64_to_cpu(*(__u64 *)(tmp));
715 /* Put the offset into the begining of the buffer in reply */
716 orr = (struct out_read_reply *)update_result->our_data;
718 nbufs = (size + OUT_BULK_BUFFER_SIZE - 1) / OUT_BULK_BUFFER_SIZE;
719 OBD_ALLOC(rdbuf, sizeof(struct lu_rdbuf) +
720 nbufs * sizeof(rdbuf->rb_bufs[0]));
722 GOTO(out, rc = -ENOMEM);
726 for (i = 0; i < nbufs; i++) {
729 read_size = size > OUT_BULK_BUFFER_SIZE ?
730 OUT_BULK_BUFFER_SIZE : size;
731 OBD_ALLOC(rdbuf->rb_bufs[i].lb_buf, read_size);
732 if (rdbuf->rb_bufs[i].lb_buf == NULL)
733 GOTO(out_free, rc = -ENOMEM);
735 rdbuf->rb_bufs[i].lb_len = read_size;
736 dt_read_lock(env, obj, MOR_TGT_CHILD);
737 rc = dt_read(env, obj, &rdbuf->rb_bufs[i], &pos);
738 dt_read_unlock(env, obj);
740 total_size += rc < 0 ? 0 : rc;
748 /* send pages to client */
749 rc = tgt_send_buffer(tsi, rdbuf);
753 orr->orr_size = total_size;
754 orr->orr_offset = pos;
756 orr_cpu_to_le(orr, orr);
757 update_result->our_datalen += orr->orr_size;
759 for (i = 0; i < nbufs; i++) {
760 if (rdbuf->rb_bufs[i].lb_buf != NULL) {
761 OBD_FREE(rdbuf->rb_bufs[i].lb_buf,
762 rdbuf->rb_bufs[i].lb_len);
765 OBD_FREE(rdbuf, sizeof(struct lu_rdbuf) +
766 nbufs * sizeof(rdbuf->rb_bufs[0]));
768 /* Insert read buffer */
769 update_result->our_rc = ptlrpc_status_hton(rc);
770 reply->ourp_lens[index] = cfs_size_round(update_result->our_datalen +
771 sizeof(*update_result));
775 static int out_noop(struct tgt_session_info *tsi)
780 #define DEF_OUT_HNDL(opc, name, flags, fn) \
781 [opc - OUT_CREATE] = { \
791 static struct tgt_handler out_update_ops[] = {
792 DEF_OUT_HNDL(OUT_CREATE, "out_create", IS_MUTABLE | HAS_REPLY,
794 DEF_OUT_HNDL(OUT_DESTROY, "out_create", IS_MUTABLE | HAS_REPLY,
796 DEF_OUT_HNDL(OUT_REF_ADD, "out_ref_add", IS_MUTABLE | HAS_REPLY,
798 DEF_OUT_HNDL(OUT_REF_DEL, "out_ref_del", IS_MUTABLE | HAS_REPLY,
800 DEF_OUT_HNDL(OUT_ATTR_SET, "out_attr_set", IS_MUTABLE | HAS_REPLY,
802 DEF_OUT_HNDL(OUT_ATTR_GET, "out_attr_get", HAS_REPLY,
804 DEF_OUT_HNDL(OUT_XATTR_SET, "out_xattr_set", IS_MUTABLE | HAS_REPLY,
806 DEF_OUT_HNDL(OUT_XATTR_DEL, "out_xattr_del", IS_MUTABLE | HAS_REPLY,
808 DEF_OUT_HNDL(OUT_XATTR_GET, "out_xattr_get", HAS_REPLY,
810 DEF_OUT_HNDL(OUT_INDEX_LOOKUP, "out_index_lookup", HAS_REPLY,
812 DEF_OUT_HNDL(OUT_INDEX_INSERT, "out_index_insert",
813 IS_MUTABLE | HAS_REPLY, out_index_insert),
814 DEF_OUT_HNDL(OUT_INDEX_DELETE, "out_index_delete",
815 IS_MUTABLE | HAS_REPLY, out_index_delete),
816 DEF_OUT_HNDL(OUT_WRITE, "out_write", IS_MUTABLE | HAS_REPLY, out_write),
817 DEF_OUT_HNDL(OUT_READ, "out_read", HAS_REPLY, out_read),
818 DEF_OUT_HNDL(OUT_NOOP, "out_noop", HAS_REPLY, out_noop),
819 DEF_OUT_HNDL(OUT_XATTR_LIST, "out_xattr_list", HAS_REPLY,
823 static struct tgt_handler *out_handler_find(__u32 opc)
825 struct tgt_handler *h;
828 if (OUT_CREATE <= opc && opc < OUT_LAST) {
829 h = &out_update_ops[opc - OUT_CREATE];
830 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
833 h = NULL; /* unsupported opc */
838 static int out_tx_start(const struct lu_env *env, struct dt_device *dt,
839 struct thandle_exec_args *ta, struct obd_export *exp)
842 ta->ta_handle = dt_trans_create(env, dt);
843 if (IS_ERR(ta->ta_handle)) {
846 rc = PTR_ERR(ta->ta_handle);
847 ta->ta_handle = NULL;
848 CERROR("%s: start handle error: rc = %d\n", dt_obd_name(dt),
852 if (exp->exp_need_sync)
853 ta->ta_handle->th_sync = 1;
858 static int out_trans_start(const struct lu_env *env,
859 struct thandle_exec_args *ta)
861 return dt_trans_start(env, ta->ta_handle->th_dev, ta->ta_handle);
864 static int out_trans_stop(const struct lu_env *env,
865 struct thandle_exec_args *ta, int err)
870 ta->ta_handle->th_result = err;
871 rc = dt_trans_stop(env, ta->ta_handle->th_dev, ta->ta_handle);
872 for (i = 0; i < ta->ta_argno; i++) {
873 if (ta->ta_args[i]->object != NULL) {
874 dt_object_put(env, ta->ta_args[i]->object);
875 ta->ta_args[i]->object = NULL;
878 ta->ta_handle = NULL;
884 static int out_tx_end(const struct lu_env *env, struct thandle_exec_args *ta,
887 struct tgt_session_info *tsi = tgt_ses_info(env);
893 if (ta->ta_handle == NULL)
896 if (declare_ret != 0 || ta->ta_argno == 0)
897 GOTO(stop, rc = declare_ret);
899 LASSERT(ta->ta_handle->th_dev != NULL);
900 rc = out_trans_start(env, ta);
901 if (unlikely(rc != 0))
904 for (i = 0; i < ta->ta_argno; i++) {
905 rc = ta->ta_args[i]->exec_fn(env, ta->ta_handle,
907 if (unlikely(rc != 0)) {
908 CDEBUG(D_INFO, "error during execution of #%u from"
909 " %s:%d: rc = %d\n", i, ta->ta_args[i]->file,
910 ta->ta_args[i]->line, rc);
912 if (ta->ta_args[i]->undo_fn != NULL)
913 ta->ta_args[i]->undo_fn(env,
917 CERROR("%s: undo for %s:%d: rc = %d\n",
918 dt_obd_name(ta->ta_handle->th_dev),
919 ta->ta_args[i]->file,
920 ta->ta_args[i]->line, -ENOTSUPP);
924 CDEBUG(D_INFO, "%s: executed %u/%u: rc = %d\n",
925 dt_obd_name(ta->ta_handle->th_dev), i, ta->ta_argno, rc);
928 /* Only fail for real updates, XXX right now llog updates will be
929 * ignore, whose updates count is usually 1, so failover test
930 * case will spot this FAIL_UPDATE_NET_REP precisely, and it will
931 * be removed after async update patch is landed. */
932 if (ta->ta_argno > 1)
933 tsi->tsi_reply_fail_id = OBD_FAIL_OUT_UPDATE_NET_REP;
936 rc1 = out_trans_stop(env, ta, rc);
940 ta->ta_handle = NULL;
947 * Object updates between Targets. Because all the updates has been
948 * dis-assemblied into object updates at sender side, so OUT will
949 * call OSD API directly to execute these updates.
951 * In DNE phase I all of the updates in the request need to be executed
952 * in one transaction, and the transaction has to be synchronously.
954 * Please refer to lustre/include/lustre/lustre_idl.h for req/reply
957 int out_handle(struct tgt_session_info *tsi)
959 const struct lu_env *env = tsi->tsi_env;
960 struct tgt_thread_info *tti = tgt_th_info(env);
961 struct thandle_exec_args *ta = &tti->tti_tea;
962 struct req_capsule *pill = tsi->tsi_pill;
963 struct dt_device *dt = tsi->tsi_tgt->lut_bottom;
964 struct out_update_header *ouh;
965 struct out_update_buffer *oub = NULL;
966 struct object_update *update;
967 struct object_update_reply *reply;
968 struct ptlrpc_bulk_desc *desc = NULL;
969 struct l_wait_info lwi;
971 int current_batchid = -1;
972 __u32 update_buf_count;
974 unsigned int reply_index = 0;
977 int ouh_size, reply_size;
981 req_capsule_set(pill, &RQF_OUT_UPDATE);
982 ouh_size = req_capsule_get_size(pill, &RMF_OUT_UPDATE_HEADER,
985 RETURN(err_serious(-EPROTO));
987 ouh = req_capsule_client_get(pill, &RMF_OUT_UPDATE_HEADER);
989 RETURN(err_serious(-EPROTO));
991 if (ouh->ouh_magic != OUT_UPDATE_HEADER_MAGIC) {
992 CERROR("%s: invalid update buffer magic %x expect %x: "
993 "rc = %d\n", tgt_name(tsi->tsi_tgt), ouh->ouh_magic,
994 UPDATE_REQUEST_MAGIC, -EPROTO);
995 RETURN(err_serious(-EPROTO));
998 update_buf_count = ouh->ouh_count;
999 if (update_buf_count == 0)
1000 RETURN(err_serious(-EPROTO));
1002 OBD_ALLOC(update_bufs, sizeof(*update_bufs) * update_buf_count);
1003 if (update_bufs == NULL)
1004 RETURN(err_serious(-ENOMEM));
1006 if (ouh->ouh_inline_length > 0) {
1007 update_bufs[0] = ouh->ouh_inline_data;
1009 struct out_update_buffer *tmp;
1011 oub = req_capsule_client_get(pill, &RMF_OUT_UPDATE_BUF);
1013 GOTO(out_free, rc = err_serious(-EPROTO));
1015 desc = ptlrpc_prep_bulk_exp(pill->rc_req, update_buf_count,
1016 PTLRPC_BULK_OPS_COUNT,
1017 PTLRPC_BULK_GET_SINK |
1018 PTLRPC_BULK_BUF_KVEC,
1020 &ptlrpc_bulk_kvec_ops);
1022 GOTO(out_free, rc = err_serious(-ENOMEM));
1025 for (i = 0; i < update_buf_count; i++, tmp++) {
1026 if (tmp->oub_size >= OUT_MAXREQSIZE)
1027 GOTO(out_free, rc = err_serious(-EPROTO));
1029 OBD_ALLOC_LARGE(update_bufs[i], tmp->oub_size);
1030 if (update_bufs[i] == NULL)
1031 GOTO(out_free, rc = err_serious(-ENOMEM));
1033 desc->bd_frag_ops->add_iov_frag(desc, update_bufs[i],
1037 pill->rc_req->rq_bulk_write = 1;
1038 rc = sptlrpc_svc_prep_bulk(pill->rc_req, desc);
1040 GOTO(out_free, rc = err_serious(rc));
1042 rc = target_bulk_io(pill->rc_req->rq_export, desc, &lwi);
1044 GOTO(out_free, rc = err_serious(rc));
1046 /* validate the request and calculate the total update count and
1047 * set it to reply */
1050 for (i = 0; i < update_buf_count; i++) {
1051 struct object_update_request *our;
1054 our = update_bufs[i];
1055 if (ptlrpc_req_need_swab(pill->rc_req))
1056 lustre_swab_object_update_request(our);
1058 if (our->ourq_magic != UPDATE_REQUEST_MAGIC) {
1059 CERROR("%s: invalid update buffer magic %x"
1060 " expect %x: rc = %d\n",
1061 tgt_name(tsi->tsi_tgt), our->ourq_magic,
1062 UPDATE_REQUEST_MAGIC, -EPROTO);
1063 GOTO(out_free, rc = err_serious(-EPROTO));
1065 updates += our->ourq_count;
1067 /* need to calculate reply size */
1068 for (j = 0; j < our->ourq_count; j++) {
1069 update = object_update_request_get(our, j, NULL);
1071 GOTO(out, rc = err_serious(-EPROTO));
1072 if (ptlrpc_req_need_swab(pill->rc_req))
1073 lustre_swab_object_update(update);
1075 if (!fid_is_sane(&update->ou_fid)) {
1076 CERROR("%s: invalid FID "DFID": rc = %d\n",
1077 tgt_name(tsi->tsi_tgt),
1078 PFID(&update->ou_fid), -EPROTO);
1079 GOTO(out, rc = err_serious(-EPROTO));
1082 /* XXX: what ou_result_size can be considered safe? */
1084 reply_size += sizeof(reply->ourp_lens[0]);
1085 reply_size += sizeof(struct object_update_result);
1086 reply_size += update->ou_result_size;
1089 reply_size += sizeof(*reply);
1091 if (unlikely(reply_size > ouh->ouh_reply_size)) {
1092 CERROR("%s: too small reply buf %u for %u, need %u at least\n",
1093 tgt_name(tsi->tsi_tgt), ouh->ouh_reply_size,
1094 updates, reply_size);
1095 GOTO(out_free, rc = err_serious(-EPROTO));
1098 req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER,
1099 ouh->ouh_reply_size);
1100 rc = req_capsule_server_pack(pill);
1102 CERROR("%s: Can't pack response: rc = %d\n",
1103 tgt_name(tsi->tsi_tgt), rc);
1104 GOTO(out_free, rc = err_serious(-EPROTO));
1107 /* Prepare the update reply buffer */
1108 reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY);
1110 GOTO(out_free, rc = -EPROTO);
1111 reply->ourp_magic = UPDATE_REPLY_MAGIC;
1112 reply->ourp_count = updates;
1113 tti->tti_u.update.tti_update_reply = reply;
1114 tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi));
1116 /* Walk through updates in the request to execute them */
1117 for (i = 0; i < update_buf_count; i++) {
1118 struct tgt_handler *h;
1119 struct dt_object *dt_obj;
1121 struct object_update_request *our;
1124 our = update_bufs[i];
1125 update_count = our->ourq_count;
1126 for (j = 0; j < update_count; j++) {
1127 struct lu_object_conf conf;
1129 update = object_update_request_get(our, j, NULL);
1130 if (update->ou_type == OUT_CREATE)
1131 conf.loc_flags = LOC_F_NEW;
1135 dt_obj = dt_locate_at(env, dt, &update->ou_fid,
1136 dt->dd_lu_dev.ld_site->ls_top_dev, &conf);
1138 GOTO(out, rc = PTR_ERR(dt_obj));
1140 if (dt->dd_record_fid_accessed) {
1141 struct lfsck_req_local *lrl = &tti->tti_lrl;
1144 lu_object_fid(&dt_obj->do_lu),
1147 tgt_lfsck_in_notify_local(env, dt, lrl, NULL);
1150 tti->tti_u.update.tti_dt_object = dt_obj;
1151 tti->tti_u.update.tti_update = update;
1152 tti->tti_u.update.tti_update_reply_index = reply_index;
1154 h = out_handler_find(update->ou_type);
1155 if (unlikely(h == NULL)) {
1156 CERROR("%s: unsupported opc: 0x%x\n",
1157 tgt_name(tsi->tsi_tgt), update->ou_type);
1158 GOTO(next, rc = -ENOTSUPP);
1161 /* Check resend case only for modifying RPC */
1162 if (h->th_flags & IS_MUTABLE) {
1163 struct ptlrpc_request *req = tgt_ses_req(tsi);
1165 if (out_check_resent(env, dt, dt_obj, req,
1166 out_reconstruct, reply,
1171 GOTO(next, rc = -EROFS);
1174 /* start transaction for modification RPC only */
1175 if (h->th_flags & IS_MUTABLE && current_batchid == -1) {
1176 current_batchid = update->ou_batchid;
1177 rc = out_tx_start(env, dt, ta, tsi->tsi_exp);
1181 if (update->ou_flags & UPDATE_FL_SYNC)
1182 ta->ta_handle->th_sync = 1;
1185 /* Stop the current update transaction, if the update
1186 * has different batchid, or read-only update */
1187 if (((current_batchid != update->ou_batchid) ||
1188 !(h->th_flags & IS_MUTABLE)) &&
1189 ta->ta_handle != NULL) {
1190 rc = out_tx_end(env, ta, rc);
1191 current_batchid = -1;
1195 /* start a new transaction if needed */
1196 if (h->th_flags & IS_MUTABLE) {
1197 rc = out_tx_start(env, dt, ta,
1201 if (update->ou_flags & UPDATE_FL_SYNC)
1202 ta->ta_handle->th_sync = 1;
1203 current_batchid = update->ou_batchid;
1207 rc = h->th_act(tsi);
1210 dt_object_put(env, dt_obj);
1216 if (current_batchid != -1) {
1217 rc1 = out_tx_end(env, ta, rc);
1223 if (update_bufs != NULL) {
1225 for (i = 0; i < update_buf_count; i++, oub++) {
1226 if (update_bufs[i] != NULL)
1227 OBD_FREE_LARGE(update_bufs[i],
1232 OBD_FREE(update_bufs, sizeof(*update_bufs) * update_buf_count);
1236 ptlrpc_free_bulk(desc);
1241 struct tgt_handler tgt_out_handlers[] = {
1242 TGT_UPDATE_HDL(IS_MUTABLE, OUT_UPDATE, out_handle),
1244 EXPORT_SYMBOL(tgt_out_handlers);