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, 2015, 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 transno "LPD64"status %d",
94 req->rq_transno, req->rq_status);
96 reconstruct(env, dt, obj, reply, index);
99 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
100 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
104 static int out_create(struct tgt_session_info *tsi)
106 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
107 struct object_update *update = tti->tti_u.update.tti_update;
108 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
109 struct dt_object_format *dof = &tti->tti_u.update.tti_update_dof;
110 struct obdo *lobdo = &tti->tti_u.update.tti_obdo;
111 struct lu_attr *attr = &tti->tti_attr;
112 struct lu_fid *fid = NULL;
119 wobdo = object_update_param_get(update, 0, &size);
120 if (IS_ERR(wobdo) || size != sizeof(*wobdo)) {
121 CERROR("%s: obdo is NULL, invalid RPC: rc = %ld\n",
122 tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo));
123 RETURN(PTR_ERR(wobdo));
126 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
127 lustre_swab_obdo(wobdo);
128 lustre_get_wire_obdo(NULL, lobdo, wobdo);
129 la_from_obdo(attr, lobdo, lobdo->o_valid);
131 dof->dof_type = dt_mode_to_dft(attr->la_mode);
132 if (update->ou_params_count > 1) {
133 fid = object_update_param_get(update, 1, &size);
134 if (IS_ERR(fid) || size != sizeof(*fid)) {
135 CERROR("%s: invalid fid: rc = %ld\n",
136 tgt_name(tsi->tsi_tgt), PTR_ERR(fid));
137 RETURN(PTR_ERR(fid));
139 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
140 lustre_swab_lu_fid(fid);
141 if (!fid_is_sane(fid)) {
142 CERROR("%s: invalid fid "DFID": rc = %d\n",
143 tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO);
148 if (lu_object_exists(&obj->do_lu))
151 rc = out_tx_create(tsi->tsi_env, obj, attr, fid, dof,
152 &tti->tti_tea, tti->tti_tea.ta_handle,
153 tti->tti_u.update.tti_update_reply,
154 tti->tti_u.update.tti_update_reply_index);
159 static int out_attr_set(struct tgt_session_info *tsi)
161 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
162 struct object_update *update = tti->tti_u.update.tti_update;
163 struct lu_attr *attr = &tti->tti_attr;
164 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
165 struct obdo *lobdo = &tti->tti_u.update.tti_obdo;
172 wobdo = object_update_param_get(update, 0, &size);
173 if (IS_ERR(wobdo) || size != sizeof(*wobdo)) {
174 CERROR("%s: empty obdo in the update: rc = %ld\n",
175 tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo));
176 RETURN(PTR_ERR(wobdo));
182 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
183 lustre_swab_obdo(wobdo);
184 lustre_get_wire_obdo(NULL, lobdo, wobdo);
185 la_from_obdo(attr, lobdo, lobdo->o_valid);
187 rc = out_tx_attr_set(tsi->tsi_env, obj, attr, &tti->tti_tea,
188 tti->tti_tea.ta_handle,
189 tti->tti_u.update.tti_update_reply,
190 tti->tti_u.update.tti_update_reply_index);
195 static int out_attr_get(struct tgt_session_info *tsi)
197 const struct lu_env *env = tsi->tsi_env;
198 struct tgt_thread_info *tti = tgt_th_info(env);
199 struct object_update *update = tti->tti_u.update.tti_update;
200 struct obdo *obdo = &tti->tti_u.update.tti_obdo;
201 struct lu_attr *la = &tti->tti_attr;
202 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
203 int idx = tti->tti_u.update.tti_update_reply_index;
208 if (unlikely(update->ou_result_size < sizeof(*obdo)))
211 if (!lu_object_exists(&obj->do_lu)) {
212 /* Usually, this will be called when the master MDT try
213 * to init a remote object(see osp_object_init), so if
214 * the object does not exist on slave, we need set BANSHEE flag,
215 * so the object can be removed from the cache immediately */
216 set_bit(LU_OBJECT_HEARD_BANSHEE,
217 &obj->do_lu.lo_header->loh_flags);
221 dt_read_lock(env, obj, MOR_TGT_CHILD);
222 rc = dt_attr_get(env, obj, la);
224 GOTO(out_unlock, rc);
227 obdo_from_la(obdo, la, la->la_valid);
228 lustre_set_wire_obdo(NULL, obdo, obdo);
231 dt_read_unlock(env, obj);
233 CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
234 tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
237 object_update_result_insert(tti->tti_u.update.tti_update_reply, obdo,
238 sizeof(*obdo), idx, rc);
243 static int out_xattr_get(struct tgt_session_info *tsi)
245 const struct lu_env *env = tsi->tsi_env;
246 struct tgt_thread_info *tti = tgt_th_info(env);
247 struct object_update *update = tti->tti_u.update.tti_update;
248 struct lu_buf *lbuf = &tti->tti_buf;
249 struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
250 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
252 struct object_update_result *update_result;
253 int idx = tti->tti_u.update.tti_update_reply_index;
258 if (!lu_object_exists(&obj->do_lu)) {
259 set_bit(LU_OBJECT_HEARD_BANSHEE,
260 &obj->do_lu.lo_header->loh_flags);
264 name = object_update_param_get(update, 0, NULL);
266 CERROR("%s: empty name for xattr get: rc = %ld\n",
267 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
268 RETURN(PTR_ERR(name));
271 update_result = object_update_result_get(reply, 0, NULL);
272 if (update_result == NULL) {
273 CERROR("%s: empty name for xattr get: rc = %d\n",
274 tgt_name(tsi->tsi_tgt), -EPROTO);
278 lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size;
279 lbuf->lb_buf = update_result->our_data;
280 if (lbuf->lb_len == 0)
282 dt_read_lock(env, obj, MOR_TGT_CHILD);
283 rc = dt_xattr_get(env, obj, lbuf, name);
284 dt_read_unlock(env, obj);
287 CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n",
288 tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)),
289 name, (int)lbuf->lb_len);
294 object_update_result_insert(reply, lbuf->lb_buf, lbuf->lb_len, idx, rc);
298 static int out_index_lookup(struct tgt_session_info *tsi)
300 const struct lu_env *env = tsi->tsi_env;
301 struct tgt_thread_info *tti = tgt_th_info(env);
302 struct object_update *update = tti->tti_u.update.tti_update;
303 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
309 if (unlikely(update->ou_result_size < sizeof(tti->tti_fid1)))
312 if (!lu_object_exists(&obj->do_lu))
315 name = object_update_param_get(update, 0, NULL);
317 CERROR("%s: empty name for lookup: rc = %ld\n",
318 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
319 RETURN(PTR_ERR(name));
322 dt_read_lock(env, obj, MOR_TGT_CHILD);
323 if (!dt_try_as_dir(env, obj))
324 GOTO(out_unlock, rc = -ENOTDIR);
326 rc = dt_lookup(env, obj, (struct dt_rec *)&tti->tti_fid1,
327 (struct dt_key *)name);
330 GOTO(out_unlock, rc);
336 dt_read_unlock(env, obj);
338 CDEBUG(D_INFO, "lookup "DFID" %s get "DFID" rc %d\n",
339 PFID(lu_object_fid(&obj->do_lu)), name,
340 PFID(&tti->tti_fid1), rc);
342 CDEBUG(D_INFO, "%s: insert lookup reply %p index %d: rc = %d\n",
343 tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
346 object_update_result_insert(tti->tti_u.update.tti_update_reply,
347 &tti->tti_fid1, sizeof(tti->tti_fid1),
348 tti->tti_u.update.tti_update_reply_index, rc);
352 static int out_xattr_set(struct tgt_session_info *tsi)
354 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
355 struct object_update *update = tti->tti_u.update.tti_update;
356 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
357 struct lu_buf *lbuf = &tti->tti_buf;
367 name = object_update_param_get(update, 0, NULL);
369 CERROR("%s: empty name for xattr set: rc = %ld\n",
370 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
371 RETURN(PTR_ERR(name));
374 /* If buffer == NULL (-ENODATA), then it might mean delete xattr */
375 buf = object_update_param_get(update, 1, &buf_len);
376 if (IS_ERR(buf) && PTR_ERR(buf) != -ENODATA)
377 RETURN(PTR_ERR(buf));
380 lbuf->lb_len = buf_len;
382 tmp = object_update_param_get(update, 2, &size);
383 if (IS_ERR(tmp) || size != sizeof(*tmp)) {
384 CERROR("%s: emptry or wrong size %zu flag: rc = %ld\n",
385 tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp));
386 RETURN(PTR_ERR(tmp));
389 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
393 rc = out_tx_xattr_set(tsi->tsi_env, obj, lbuf, name, flag,
394 &tti->tti_tea, tti->tti_tea.ta_handle,
395 tti->tti_u.update.tti_update_reply,
396 tti->tti_u.update.tti_update_reply_index);
400 static int out_xattr_del(struct tgt_session_info *tsi)
402 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
403 struct object_update *update = tti->tti_u.update.tti_update;
404 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
409 name = object_update_param_get(update, 0, NULL);
411 CERROR("%s: empty name for xattr set: rc = %ld\n",
412 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
413 RETURN(PTR_ERR(name));
416 rc = out_tx_xattr_del(tsi->tsi_env, obj, name, &tti->tti_tea,
417 tti->tti_tea.ta_handle,
418 tti->tti_u.update.tti_update_reply,
419 tti->tti_u.update.tti_update_reply_index);
424 * increase ref of the object
426 static int out_ref_add(struct tgt_session_info *tsi)
428 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
429 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
434 rc = out_tx_ref_add(tsi->tsi_env, obj, &tti->tti_tea,
435 tti->tti_tea.ta_handle,
436 tti->tti_u.update.tti_update_reply,
437 tti->tti_u.update.tti_update_reply_index);
441 static int out_ref_del(struct tgt_session_info *tsi)
443 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
444 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
449 if (!lu_object_exists(&obj->do_lu))
452 rc = out_tx_ref_del(tsi->tsi_env, obj, &tti->tti_tea,
453 tti->tti_tea.ta_handle,
454 tti->tti_u.update.tti_update_reply,
455 tti->tti_u.update.tti_update_reply_index);
459 static int out_index_insert(struct tgt_session_info *tsi)
461 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
462 struct object_update *update = tti->tti_u.update.tti_update;
463 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
464 struct dt_insert_rec *rec = &tti->tti_rec;
472 name = object_update_param_get(update, 0, NULL);
474 CERROR("%s: empty name for index insert: rc = %ld\n",
475 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
476 RETURN(PTR_ERR(name));
479 fid = object_update_param_get(update, 1, &size);
480 if (IS_ERR(fid) || size != sizeof(*fid)) {
481 CERROR("%s: invalid fid: rc = %ld\n",
482 tgt_name(tsi->tsi_tgt), PTR_ERR(fid));
483 RETURN(PTR_ERR(fid));
486 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
487 lustre_swab_lu_fid(fid);
489 if (!fid_is_sane(fid)) {
490 CERROR("%s: invalid FID "DFID": rc = %d\n",
491 tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO);
495 ptype = object_update_param_get(update, 2, &size);
496 if (IS_ERR(ptype) || size != sizeof(*ptype)) {
497 CERROR("%s: invalid type for index insert: rc = %ld\n",
498 tgt_name(tsi->tsi_tgt), PTR_ERR(ptype));
499 RETURN(PTR_ERR(ptype));
502 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
506 rec->rec_type = *ptype;
508 rc = out_tx_index_insert(tsi->tsi_env, obj, (const struct dt_rec *)rec,
509 (const struct dt_key *)name, &tti->tti_tea,
510 tti->tti_tea.ta_handle,
511 tti->tti_u.update.tti_update_reply,
512 tti->tti_u.update.tti_update_reply_index);
516 static int out_index_delete(struct tgt_session_info *tsi)
518 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
519 struct object_update *update = tti->tti_u.update.tti_update;
520 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
524 if (!lu_object_exists(&obj->do_lu))
527 name = object_update_param_get(update, 0, NULL);
529 CERROR("%s: empty name for index delete: rc = %ld\n",
530 tgt_name(tsi->tsi_tgt), PTR_ERR(name));
531 RETURN(PTR_ERR(name));
534 rc = out_tx_index_delete(tsi->tsi_env, obj, (const struct dt_key *)name,
535 &tti->tti_tea, tti->tti_tea.ta_handle,
536 tti->tti_u.update.tti_update_reply,
537 tti->tti_u.update.tti_update_reply_index);
541 static int out_destroy(struct tgt_session_info *tsi)
543 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
544 struct object_update *update = tti->tti_u.update.tti_update;
545 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
550 fid = &update->ou_fid;
551 if (!fid_is_sane(fid)) {
552 CERROR("%s: invalid FID "DFID": rc = %d\n",
553 tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO);
557 if (!lu_object_exists(&obj->do_lu))
560 rc = out_tx_destroy(tsi->tsi_env, obj, &tti->tti_tea,
561 tti->tti_tea.ta_handle,
562 tti->tti_u.update.tti_update_reply,
563 tti->tti_u.update.tti_update_reply_index);
568 static int out_write(struct tgt_session_info *tsi)
570 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
571 struct object_update *update = tti->tti_u.update.tti_update;
572 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
573 struct lu_buf *lbuf = &tti->tti_buf;
582 buf = object_update_param_get(update, 0, &buf_len);
583 if (IS_ERR(buf) || buf_len == 0) {
584 CERROR("%s: empty buf for xattr set: rc = %ld\n",
585 tgt_name(tsi->tsi_tgt), PTR_ERR(buf));
586 RETURN(PTR_ERR(buf));
589 lbuf->lb_len = buf_len;
591 tmp = object_update_param_get(update, 1, &size);
592 if (IS_ERR(tmp) || size != sizeof(*tmp)) {
593 CERROR("%s: empty or wrong size %zu pos: rc = %ld\n",
594 tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp));
595 RETURN(PTR_ERR(tmp));
598 if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req))
602 rc = out_tx_write(tsi->tsi_env, obj, lbuf, pos,
603 &tti->tti_tea, tti->tti_tea.ta_handle,
604 tti->tti_u.update.tti_update_reply,
605 tti->tti_u.update.tti_update_reply_index);
609 static int out_read(struct tgt_session_info *tsi)
611 const struct lu_env *env = tsi->tsi_env;
612 struct tgt_thread_info *tti = tgt_th_info(env);
613 struct object_update *update = tti->tti_u.update.tti_update;
614 struct dt_object *obj = tti->tti_u.update.tti_dt_object;
615 struct object_update_reply *reply = tti->tti_u.update.tti_update_reply;
616 int index = tti->tti_u.update.tti_update_reply_index;
617 struct lu_rdbuf *rdbuf;
618 struct object_update_result *update_result;
619 struct out_read_reply *orr;
622 size_t total_size = 0;
629 update_result = object_update_result_get(reply, index, NULL);
630 LASSERT(update_result != NULL);
631 update_result->our_datalen = sizeof(*orr);
633 if (!lu_object_exists(&obj->do_lu))
634 GOTO(out, rc = -ENOENT);
636 tmp = object_update_param_get(update, 0, NULL);
638 CERROR("%s: empty size for read: rc = %ld\n",
639 tgt_name(tsi->tsi_tgt), PTR_ERR(tmp));
640 GOTO(out, rc = PTR_ERR(tmp));
642 size = le64_to_cpu(*(size_t *)(tmp));
644 tmp = object_update_param_get(update, 1, NULL);
646 CERROR("%s: empty pos for read: rc = %ld\n",
647 tgt_name(tsi->tsi_tgt), PTR_ERR(tmp));
648 GOTO(out, rc = PTR_ERR(tmp));
650 pos = le64_to_cpu(*(__u64 *)(tmp));
652 /* Put the offset into the begining of the buffer in reply */
653 orr = (struct out_read_reply *)update_result->our_data;
655 nbufs = (size + OUT_BULK_BUFFER_SIZE - 1) / OUT_BULK_BUFFER_SIZE;
656 OBD_ALLOC(rdbuf, sizeof(struct lu_rdbuf) +
657 nbufs * sizeof(rdbuf->rb_bufs[0]));
659 GOTO(out, rc = -ENOMEM);
663 for (i = 0; i < nbufs; i++) {
666 read_size = size > OUT_BULK_BUFFER_SIZE ?
667 OUT_BULK_BUFFER_SIZE : size;
668 OBD_ALLOC(rdbuf->rb_bufs[i].lb_buf, read_size);
669 if (rdbuf->rb_bufs[i].lb_buf == NULL)
670 GOTO(out_free, rc = -ENOMEM);
672 rdbuf->rb_bufs[i].lb_len = read_size;
673 dt_read_lock(env, obj, MOR_TGT_CHILD);
674 rc = dt_read(env, obj, &rdbuf->rb_bufs[i], &pos);
675 dt_read_unlock(env, obj);
677 total_size += rc < 0 ? 0 : rc;
685 /* send pages to client */
686 rc = tgt_send_buffer(tsi, rdbuf);
690 orr->orr_size = total_size;
691 orr->orr_offset = pos;
693 orr_cpu_to_le(orr, orr);
694 update_result->our_datalen += orr->orr_size;
696 for (i = 0; i < nbufs; i++) {
697 if (rdbuf->rb_bufs[i].lb_buf != NULL) {
698 OBD_FREE(rdbuf->rb_bufs[i].lb_buf,
699 rdbuf->rb_bufs[i].lb_len);
702 OBD_FREE(rdbuf, sizeof(struct lu_rdbuf) +
703 nbufs * sizeof(rdbuf->rb_bufs[0]));
705 /* Insert read buffer */
706 update_result->our_rc = ptlrpc_status_hton(rc);
707 reply->ourp_lens[index] = cfs_size_round(update_result->our_datalen +
708 sizeof(*update_result));
712 static int out_noop(struct tgt_session_info *tsi)
717 #define DEF_OUT_HNDL(opc, name, flags, fn) \
718 [opc - OUT_CREATE] = { \
728 static struct tgt_handler out_update_ops[] = {
729 DEF_OUT_HNDL(OUT_CREATE, "out_create", MUTABOR | HABEO_REFERO,
731 DEF_OUT_HNDL(OUT_DESTROY, "out_create", MUTABOR | HABEO_REFERO,
733 DEF_OUT_HNDL(OUT_REF_ADD, "out_ref_add", MUTABOR | HABEO_REFERO,
735 DEF_OUT_HNDL(OUT_REF_DEL, "out_ref_del", MUTABOR | HABEO_REFERO,
737 DEF_OUT_HNDL(OUT_ATTR_SET, "out_attr_set", MUTABOR | HABEO_REFERO,
739 DEF_OUT_HNDL(OUT_ATTR_GET, "out_attr_get", HABEO_REFERO,
741 DEF_OUT_HNDL(OUT_XATTR_SET, "out_xattr_set", MUTABOR | HABEO_REFERO,
743 DEF_OUT_HNDL(OUT_XATTR_DEL, "out_xattr_del", MUTABOR | HABEO_REFERO,
745 DEF_OUT_HNDL(OUT_XATTR_GET, "out_xattr_get", HABEO_REFERO,
747 DEF_OUT_HNDL(OUT_INDEX_LOOKUP, "out_index_lookup", HABEO_REFERO,
749 DEF_OUT_HNDL(OUT_INDEX_INSERT, "out_index_insert",
750 MUTABOR | HABEO_REFERO, out_index_insert),
751 DEF_OUT_HNDL(OUT_INDEX_DELETE, "out_index_delete",
752 MUTABOR | HABEO_REFERO, out_index_delete),
753 DEF_OUT_HNDL(OUT_WRITE, "out_write", MUTABOR | HABEO_REFERO, out_write),
754 DEF_OUT_HNDL(OUT_READ, "out_read", HABEO_REFERO, out_read),
755 DEF_OUT_HNDL(OUT_NOOP, "out_noop", HABEO_REFERO, out_noop),
758 static struct tgt_handler *out_handler_find(__u32 opc)
760 struct tgt_handler *h;
763 if (OUT_CREATE <= opc && opc < OUT_LAST) {
764 h = &out_update_ops[opc - OUT_CREATE];
765 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
768 h = NULL; /* unsupported opc */
773 static int out_tx_start(const struct lu_env *env, struct dt_device *dt,
774 struct thandle_exec_args *ta, struct obd_export *exp)
777 ta->ta_handle = dt_trans_create(env, dt);
778 if (IS_ERR(ta->ta_handle)) {
781 rc = PTR_ERR(ta->ta_handle);
782 ta->ta_handle = NULL;
783 CERROR("%s: start handle error: rc = %d\n", dt_obd_name(dt),
787 if (exp->exp_need_sync)
788 ta->ta_handle->th_sync = 1;
793 static int out_trans_start(const struct lu_env *env,
794 struct thandle_exec_args *ta)
796 return dt_trans_start(env, ta->ta_handle->th_dev, ta->ta_handle);
799 static int out_trans_stop(const struct lu_env *env,
800 struct thandle_exec_args *ta, int err)
805 ta->ta_handle->th_result = err;
806 rc = dt_trans_stop(env, ta->ta_handle->th_dev, ta->ta_handle);
807 for (i = 0; i < ta->ta_argno; i++) {
808 if (ta->ta_args[i]->object != NULL) {
809 struct dt_object *obj = ta->ta_args[i]->object;
811 /* If the object is being created during this
812 * transaction, we need to remove them from the
813 * cache immediately, because a few layers are
814 * missing in OUT handler, i.e. the object might
815 * not be initialized in all layers */
816 if (ta->ta_args[i]->exec_fn == out_tx_create_exec)
817 set_bit(LU_OBJECT_HEARD_BANSHEE,
818 &obj->do_lu.lo_header->loh_flags);
819 lu_object_put(env, &ta->ta_args[i]->object->do_lu);
820 ta->ta_args[i]->object = NULL;
823 ta->ta_handle = NULL;
829 static int out_tx_end(const struct lu_env *env, struct thandle_exec_args *ta,
832 struct tgt_session_info *tsi = tgt_ses_info(env);
838 if (ta->ta_handle == NULL)
841 if (declare_ret != 0 || ta->ta_argno == 0)
842 GOTO(stop, rc = declare_ret);
844 LASSERT(ta->ta_handle->th_dev != NULL);
845 rc = out_trans_start(env, ta);
846 if (unlikely(rc != 0))
849 for (i = 0; i < ta->ta_argno; i++) {
850 rc = ta->ta_args[i]->exec_fn(env, ta->ta_handle,
852 if (unlikely(rc != 0)) {
853 CDEBUG(D_INFO, "error during execution of #%u from"
854 " %s:%d: rc = %d\n", i, ta->ta_args[i]->file,
855 ta->ta_args[i]->line, rc);
857 if (ta->ta_args[i]->undo_fn != NULL)
858 ta->ta_args[i]->undo_fn(env,
862 CERROR("%s: undo for %s:%d: rc = %d\n",
863 dt_obd_name(ta->ta_handle->th_dev),
864 ta->ta_args[i]->file,
865 ta->ta_args[i]->line, -ENOTSUPP);
869 CDEBUG(D_INFO, "%s: executed %u/%u: rc = %d\n",
870 dt_obd_name(ta->ta_handle->th_dev), i, ta->ta_argno, rc);
873 /* Only fail for real updates, XXX right now llog updates will be
874 * ignore, whose updates count is usually 1, so failover test
875 * case will spot this FAIL_UPDATE_NET_REP precisely, and it will
876 * be removed after async update patch is landed. */
877 if (ta->ta_argno > 1)
878 tsi->tsi_reply_fail_id = OBD_FAIL_OUT_UPDATE_NET_REP;
881 rc1 = out_trans_stop(env, ta, rc);
885 ta->ta_handle = NULL;
892 * Object updates between Targets. Because all the updates has been
893 * dis-assemblied into object updates at sender side, so OUT will
894 * call OSD API directly to execute these updates.
896 * In DNE phase I all of the updates in the request need to be executed
897 * in one transaction, and the transaction has to be synchronously.
899 * Please refer to lustre/include/lustre/lustre_idl.h for req/reply
902 int out_handle(struct tgt_session_info *tsi)
904 const struct lu_env *env = tsi->tsi_env;
905 struct tgt_thread_info *tti = tgt_th_info(env);
906 struct thandle_exec_args *ta = &tti->tti_tea;
907 struct req_capsule *pill = tsi->tsi_pill;
908 struct dt_device *dt = tsi->tsi_tgt->lut_bottom;
909 struct out_update_header *ouh;
910 struct out_update_buffer *oub = NULL;
911 struct object_update *update;
912 struct object_update_reply *reply;
913 struct ptlrpc_bulk_desc *desc = NULL;
914 struct l_wait_info lwi;
916 int current_batchid = -1;
917 __u32 update_buf_count;
919 unsigned int reply_index = 0;
922 int ouh_size, reply_size;
926 req_capsule_set(pill, &RQF_OUT_UPDATE);
927 ouh_size = req_capsule_get_size(pill, &RMF_OUT_UPDATE_HEADER,
930 RETURN(err_serious(-EPROTO));
932 ouh = req_capsule_client_get(pill, &RMF_OUT_UPDATE_HEADER);
934 RETURN(err_serious(-EPROTO));
936 if (ouh->ouh_magic != OUT_UPDATE_HEADER_MAGIC) {
937 CERROR("%s: invalid update buffer magic %x expect %x: "
938 "rc = %d\n", tgt_name(tsi->tsi_tgt), ouh->ouh_magic,
939 UPDATE_REQUEST_MAGIC, -EPROTO);
940 RETURN(err_serious(-EPROTO));
943 update_buf_count = ouh->ouh_count;
944 if (update_buf_count == 0)
945 RETURN(err_serious(-EPROTO));
947 OBD_ALLOC(update_bufs, sizeof(*update_bufs) * update_buf_count);
948 if (update_bufs == NULL)
949 RETURN(err_serious(-ENOMEM));
951 if (ouh->ouh_inline_length > 0) {
952 update_bufs[0] = ouh->ouh_inline_data;
954 struct out_update_buffer *tmp;
956 oub = req_capsule_client_get(pill, &RMF_OUT_UPDATE_BUF);
958 GOTO(out_free, rc = err_serious(-EPROTO));
960 desc = ptlrpc_prep_bulk_exp(pill->rc_req, update_buf_count,
961 PTLRPC_BULK_OPS_COUNT,
962 PTLRPC_BULK_GET_SINK |
963 PTLRPC_BULK_BUF_KVEC,
965 &ptlrpc_bulk_kvec_ops);
967 GOTO(out_free, rc = err_serious(-ENOMEM));
970 for (i = 0; i < update_buf_count; i++, tmp++) {
971 if (tmp->oub_size >= OUT_MAXREQSIZE)
972 GOTO(out_free, rc = err_serious(-EPROTO));
974 OBD_ALLOC(update_bufs[i], tmp->oub_size);
975 if (update_bufs[i] == NULL)
976 GOTO(out_free, rc = err_serious(-ENOMEM));
978 desc->bd_frag_ops->add_iov_frag(desc, update_bufs[i],
982 pill->rc_req->rq_bulk_write = 1;
983 rc = sptlrpc_svc_prep_bulk(pill->rc_req, desc);
985 GOTO(out_free, rc = err_serious(rc));
987 rc = target_bulk_io(pill->rc_req->rq_export, desc, &lwi);
989 GOTO(out_free, rc = err_serious(rc));
991 /* validate the request and calculate the total update count and
995 for (i = 0; i < update_buf_count; i++) {
996 struct object_update_request *our;
999 our = update_bufs[i];
1000 if (ptlrpc_req_need_swab(pill->rc_req))
1001 lustre_swab_object_update_request(our);
1003 if (our->ourq_magic != UPDATE_REQUEST_MAGIC) {
1004 CERROR("%s: invalid update buffer magic %x"
1005 " expect %x: rc = %d\n",
1006 tgt_name(tsi->tsi_tgt), our->ourq_magic,
1007 UPDATE_REQUEST_MAGIC, -EPROTO);
1008 GOTO(out_free, rc = err_serious(-EPROTO));
1010 updates += our->ourq_count;
1012 /* need to calculate reply size */
1013 for (j = 0; j < our->ourq_count; j++) {
1014 update = object_update_request_get(our, j, NULL);
1016 GOTO(out, rc = err_serious(-EPROTO));
1017 if (ptlrpc_req_need_swab(pill->rc_req))
1018 lustre_swab_object_update(update);
1020 if (!fid_is_sane(&update->ou_fid)) {
1021 CERROR("%s: invalid FID "DFID": rc = %d\n",
1022 tgt_name(tsi->tsi_tgt),
1023 PFID(&update->ou_fid), -EPROTO);
1024 GOTO(out, rc = err_serious(-EPROTO));
1027 /* XXX: what ou_result_size can be considered safe? */
1029 reply_size += sizeof(reply->ourp_lens[0]);
1030 reply_size += sizeof(struct object_update_result);
1031 reply_size += update->ou_result_size;
1034 reply_size += sizeof(*reply);
1036 if (unlikely(reply_size > ouh->ouh_reply_size)) {
1037 CERROR("%s: too small reply buf %u for %u, need %u at least\n",
1038 tgt_name(tsi->tsi_tgt), ouh->ouh_reply_size,
1039 updates, reply_size);
1040 GOTO(out_free, rc = err_serious(-EPROTO));
1043 req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER,
1044 ouh->ouh_reply_size);
1045 rc = req_capsule_server_pack(pill);
1047 CERROR("%s: Can't pack response: rc = %d\n",
1048 tgt_name(tsi->tsi_tgt), rc);
1049 GOTO(out_free, rc = err_serious(-EPROTO));
1052 /* Prepare the update reply buffer */
1053 reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY);
1055 GOTO(out_free, rc = -EPROTO);
1056 reply->ourp_magic = UPDATE_REPLY_MAGIC;
1057 reply->ourp_count = updates;
1058 tti->tti_u.update.tti_update_reply = reply;
1059 tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi));
1061 /* Walk through updates in the request to execute them */
1062 for (i = 0; i < update_buf_count; i++) {
1063 struct tgt_handler *h;
1064 struct dt_object *dt_obj;
1066 struct object_update_request *our;
1069 our = update_bufs[i];
1070 update_count = our->ourq_count;
1071 for (j = 0; j < update_count; j++) {
1072 update = object_update_request_get(our, j, NULL);
1074 dt_obj = dt_locate(env, dt, &update->ou_fid);
1076 GOTO(out, rc = PTR_ERR(dt_obj));
1078 if (dt->dd_record_fid_accessed) {
1079 lfsck_pack_rfa(&tti->tti_lr,
1080 lu_object_fid(&dt_obj->do_lu),
1083 tgt_lfsck_in_notify(env, dt, &tti->tti_lr,
1087 tti->tti_u.update.tti_dt_object = dt_obj;
1088 tti->tti_u.update.tti_update = update;
1089 tti->tti_u.update.tti_update_reply_index = reply_index;
1091 h = out_handler_find(update->ou_type);
1092 if (unlikely(h == NULL)) {
1093 CERROR("%s: unsupported opc: 0x%x\n",
1094 tgt_name(tsi->tsi_tgt), update->ou_type);
1095 GOTO(next, rc = -ENOTSUPP);
1098 /* Check resend case only for modifying RPC */
1099 if (h->th_flags & MUTABOR) {
1100 struct ptlrpc_request *req = tgt_ses_req(tsi);
1102 if (out_check_resent(env, dt, dt_obj, req,
1103 out_reconstruct, reply,
1108 /* start transaction for modification RPC only */
1109 if (h->th_flags & MUTABOR && current_batchid == -1) {
1110 current_batchid = update->ou_batchid;
1111 rc = out_tx_start(env, dt, ta, tsi->tsi_exp);
1115 if (update->ou_flags & UPDATE_FL_SYNC)
1116 ta->ta_handle->th_sync = 1;
1119 /* Stop the current update transaction, if the update
1120 * has different batchid, or read-only update */
1121 if (((current_batchid != update->ou_batchid) ||
1122 !(h->th_flags & MUTABOR)) &&
1123 ta->ta_handle != NULL) {
1124 rc = out_tx_end(env, ta, rc);
1125 current_batchid = -1;
1129 /* start a new transaction if needed */
1130 if (h->th_flags & MUTABOR) {
1131 rc = out_tx_start(env, dt, ta,
1135 if (update->ou_flags & UPDATE_FL_SYNC)
1136 ta->ta_handle->th_sync = 1;
1137 current_batchid = update->ou_batchid;
1141 rc = h->th_act(tsi);
1144 lu_object_put(env, &dt_obj->do_lu);
1150 if (current_batchid != -1) {
1151 rc1 = out_tx_end(env, ta, rc);
1157 if (update_bufs != NULL) {
1159 for (i = 0; i < update_buf_count; i++, oub++) {
1160 if (update_bufs[i] != NULL)
1161 OBD_FREE(update_bufs[i], oub->oub_size);
1165 OBD_FREE(update_bufs, sizeof(*update_bufs) * update_buf_count);
1169 ptlrpc_free_bulk(desc);
1174 struct tgt_handler tgt_out_handlers[] = {
1175 TGT_UPDATE_HDL(MUTABOR, OUT_UPDATE, out_handle),
1177 EXPORT_SYMBOL(tgt_out_handlers);