Whamcloud - gitweb
LU-10318 dom: support DATA_VERSION IO type
[fs/lustre-release.git] / lustre / target / tgt_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; if not, write to the
18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19  * Boston, MA 021110-1307, USA
20  *
21  * GPL HEADER END
22  */
23 /*
24  * Copyright (c) 2013, 2017, Intel Corporation.
25  */
26 /*
27  * lustre/target/tgt_handler.c
28  *
29  * Lustre Unified Target request handler code
30  *
31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
32  * Author: Mikhail Pershin <mike.pershin@intel.com>
33  */
34
35 #define DEBUG_SUBSYSTEM S_CLASS
36
37 #include <linux/user_namespace.h>
38 #ifdef HAVE_UIDGID_HEADER
39 # include <linux/uidgid.h>
40 #endif
41
42 #include <obd.h>
43 #include <obd_class.h>
44 #include <obd_cksum.h>
45 #include <lustre_lfsck.h>
46 #include <lustre_nodemap.h>
47 #include <lustre_acl.h>
48
49 #include "tgt_internal.h"
50
51 char *tgt_name(struct lu_target *tgt)
52 {
53         LASSERT(tgt->lut_obd != NULL);
54         return tgt->lut_obd->obd_name;
55 }
56 EXPORT_SYMBOL(tgt_name);
57
58 /*
59  * Generic code handling requests that have struct mdt_body passed in:
60  *
61  *  - extract mdt_body from request and save it in @tsi, if present;
62  *
63  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
64  *  @tsi;
65  *
66  *  - if HABEO_CORPUS flag is set for this request type check whether object
67  *  actually exists on storage (lu_object_exists()).
68  *
69  */
70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
71 {
72         const struct mdt_body   *body;
73         struct lu_object        *obj;
74         struct req_capsule      *pill = tsi->tsi_pill;
75         int                      rc;
76
77         ENTRY;
78
79         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
80         if (body == NULL)
81                 RETURN(-EFAULT);
82
83         tsi->tsi_mdt_body = body;
84
85         if (!(body->mbo_valid & OBD_MD_FLID))
86                 RETURN(0);
87
88         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
89          * in any case in pre-2.5 clients. Fix that here if needed */
90         if (unlikely(fid_is_zero(&body->mbo_fid1)))
91                 RETURN(0);
92
93         if (!fid_is_sane(&body->mbo_fid1)) {
94                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
95                        PFID(&body->mbo_fid1));
96                 RETURN(-EINVAL);
97         }
98
99         obj = lu_object_find(tsi->tsi_env,
100                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
101                              &body->mbo_fid1, NULL);
102         if (!IS_ERR(obj)) {
103                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
104                         lu_object_put(tsi->tsi_env, obj);
105                         rc = -ENOENT;
106                 } else {
107                         tsi->tsi_corpus = obj;
108                         rc = 0;
109                 }
110         } else {
111                 rc = PTR_ERR(obj);
112         }
113
114         tsi->tsi_fid = body->mbo_fid1;
115
116         RETURN(rc);
117 }
118
119 /**
120  * Validate oa from client.
121  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
122  * req are valid.
123  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
124  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
125  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
126  *       whether this is oi_fid or real ostid. So it will check
127  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
128  *    c. Old FID-disable osc will send IDIF.
129  *    d. new FID-enable osc/osp will send normal FID.
130  *
131  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
132  * be used for LAST_ID file, and only being accessed inside OST now.
133  */
134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
135 {
136         struct ost_id   *oi     = &oa->o_oi;
137         u64              seq    = ostid_seq(oi);
138         u64              id     = ostid_id(oi);
139         int              rc;
140         ENTRY;
141
142         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
143                      fid_seq_is_echo(seq))) {
144                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
145                  * during create, and we will reset this to 1, since this
146                  * oi_id is basically useless in the following create process,
147                  * but oi_id == 0 will make it difficult to tell whether it is
148                  * real FID or ost_id. */
149                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
150                 oi->oi_fid.f_oid = id ?: 1;
151                 oi->oi_fid.f_ver = 0;
152         } else {
153                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
154
155                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
156                         GOTO(out, rc = -EPROTO);
157
158                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
159                  * all of the requests are required to setup FLGROUP */
160                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
161                         ostid_set_seq_mdt0(oi);
162                         oa->o_valid |= OBD_MD_FLGROUP;
163                         seq = ostid_seq(oi);
164                 }
165
166                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
167                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
168                         GOTO(out, rc = -EPROTO);
169
170                 rc = ostid_to_fid(&tti->tti_fid1, oi,
171                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
172                 if (unlikely(rc != 0))
173                         GOTO(out, rc);
174
175                 oi->oi_fid = tti->tti_fid1;
176         }
177
178         RETURN(0);
179
180 out:
181         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
182                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
183                seq, id, rc);
184         return rc;
185 }
186 EXPORT_SYMBOL(tgt_validate_obdo);
187
188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
189 {
190         unsigned                 max_brw;
191         struct niobuf_remote    *rnb;
192         struct obd_ioobj        *ioo;
193         int                      obj_count;
194
195         ENTRY;
196
197         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
198         if (ioo == NULL)
199                 RETURN(-EPROTO);
200
201         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
202         if (rnb == NULL)
203                 RETURN(-EPROTO);
204
205         max_brw = ioobj_max_brw_get(ioo);
206         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
207                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
208                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
209                        obd_export_nid2str(tsi->tsi_exp), max_brw,
210                        POSTID(oi), -EPROTO);
211                 RETURN(-EPROTO);
212         }
213         ioo->ioo_oid = *oi;
214
215         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
216                                         RCL_CLIENT) / sizeof(*ioo);
217         if (obj_count == 0) {
218                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
219                 RETURN(-EPROTO);
220         } else if (obj_count > 1) {
221                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
222                        obj_count);
223                 RETURN(-EPROTO);
224         }
225
226         if (ioo->ioo_bufcnt == 0) {
227                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
228                 RETURN(-EPROTO);
229         }
230
231         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
232                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
233                           "bulk has too many pages (%d)",
234                           ioo->ioo_bufcnt);
235                 RETURN(-EPROTO);
236         }
237
238         RETURN(0);
239 }
240
241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
242 {
243         struct ost_body         *body;
244         struct req_capsule      *pill = tsi->tsi_pill;
245         struct lu_nodemap       *nodemap;
246         int                      rc;
247
248         ENTRY;
249
250         body = req_capsule_client_get(pill, &RMF_OST_BODY);
251         if (body == NULL)
252                 RETURN(-EFAULT);
253
254         rc = tgt_validate_obdo(tsi, &body->oa);
255         if (rc)
256                 RETURN(rc);
257
258         nodemap = nodemap_get_from_exp(tsi->tsi_exp);
259         if (IS_ERR(nodemap))
260                 RETURN(PTR_ERR(nodemap));
261
262         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
263                                         NODEMAP_CLIENT_TO_FS,
264                                         body->oa.o_uid);
265         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
266                                         NODEMAP_CLIENT_TO_FS,
267                                         body->oa.o_gid);
268         nodemap_putref(nodemap);
269
270         tsi->tsi_ost_body = body;
271         tsi->tsi_fid = body->oa.o_oi.oi_fid;
272
273         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
274                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
275                 if (rc < 0)
276                         RETURN(rc);
277         }
278
279         if (!(body->oa.o_valid & OBD_MD_FLID)) {
280                 if (flags & HABEO_CORPUS) {
281                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
282                                "but OID/FID is mandatory with HABEO_CORPUS\n",
283                                tgt_name(tsi->tsi_tgt));
284                         RETURN(-EPROTO);
285                 } else {
286                         RETURN(0);
287                 }
288         }
289
290         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
291
292         /*
293          * OST doesn't get object in advance for further use to prevent
294          * situations with nested object_find which is potential deadlock.
295          */
296         tsi->tsi_corpus = NULL;
297         RETURN(rc);
298 }
299
300 /*
301  * Do necessary preprocessing according to handler ->th_flags.
302  */
303 static int tgt_request_preprocess(struct tgt_session_info *tsi,
304                                   struct tgt_handler *h,
305                                   struct ptlrpc_request *req)
306 {
307         struct req_capsule      *pill = tsi->tsi_pill;
308         __u32                    flags = h->th_flags;
309         int                      rc = 0;
310
311         ENTRY;
312
313         if (tsi->tsi_preprocessed)
314                 RETURN(0);
315
316         LASSERT(h->th_act != NULL);
317         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
318         LASSERT(current->journal_info == NULL);
319
320         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
321                      h->th_fmt != NULL));
322         if (h->th_fmt != NULL) {
323                 req_capsule_set(pill, h->th_fmt);
324                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
325                         rc = tgt_mdt_body_unpack(tsi, flags);
326                         if (rc < 0)
327                                 RETURN(rc);
328                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
329                                                  RCL_CLIENT)) {
330                         rc = tgt_ost_body_unpack(tsi, flags);
331                         if (rc < 0)
332                                 RETURN(rc);
333                 }
334         }
335
336         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
337                 RETURN(-EROFS);
338
339         if (flags & HABEO_CLAVIS) {
340                 struct ldlm_request *dlm_req;
341
342                 LASSERT(h->th_fmt != NULL);
343
344                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
345                 if (dlm_req != NULL) {
346                         union ldlm_wire_policy_data *policy =
347                                         &dlm_req->lock_desc.l_policy_data;
348
349                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
350                                      LDLM_IBITS &&
351                                      (policy->l_inodebits.bits |
352                                       policy->l_inodebits.try_bits) == 0)) {
353                                 /*
354                                  * Lock without inodebits makes no sense and
355                                  * will oops later in ldlm. If client miss to
356                                  * set such bits, do not trigger ASSERTION.
357                                  *
358                                  * For liblustre flock case, it maybe zero.
359                                  */
360                                 rc = -EPROTO;
361                         } else {
362                                 tsi->tsi_dlm_req = dlm_req;
363                         }
364                 } else {
365                         rc = -EFAULT;
366                 }
367         }
368         tsi->tsi_preprocessed = 1;
369         RETURN(rc);
370 }
371
372 /*
373  * Invoke handler for this request opc. Also do necessary preprocessing
374  * (according to handler ->th_flags), and post-processing (setting of
375  * ->last_{xid,committed}).
376  */
377 static int tgt_handle_request0(struct tgt_session_info *tsi,
378                                struct tgt_handler *h,
379                                struct ptlrpc_request *req)
380 {
381         int      serious = 0;
382         int      rc;
383         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
384
385         ENTRY;
386
387
388         /* When dealing with sec context requests, no export is associated yet,
389          * because these requests are sent before *_CONNECT requests.
390          * A NULL req->rq_export means the normal *_common_slice handlers will
391          * not be called, because there is no reference to the target.
392          * So deal with them by hand and jump directly to target_send_reply().
393          */
394         switch (opc) {
395         case SEC_CTX_INIT:
396         case SEC_CTX_INIT_CONT:
397         case SEC_CTX_FINI:
398                 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
399                 GOTO(out, rc = 0);
400         }
401
402         /*
403          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
404          * to put same checks into handlers like mdt_close(), mdt_reint(),
405          * etc., without talking to mdt authors first. Checking same thing
406          * there again is useless and returning 0 error without packing reply
407          * is buggy! Handlers either pack reply or return error.
408          *
409          * We return 0 here and do not send any reply in order to emulate
410          * network failure. Do not send any reply in case any of NET related
411          * fail_id has occured.
412          */
413         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
414                 RETURN(0);
415         if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
416                      OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
417                 RETURN(0);
418
419         rc = tgt_request_preprocess(tsi, h, req);
420         /* pack reply if reply format is fixed */
421         if (rc == 0 && h->th_flags & HABEO_REFERO) {
422                 /* Pack reply */
423                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
424                                           RCL_SERVER))
425                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
426                                              RCL_SERVER,
427                                              tsi->tsi_mdt_body->mbo_eadatasize);
428                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
429                                           RCL_SERVER))
430                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
431                                              RCL_SERVER, 0);
432                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_ACL, RCL_SERVER))
433                         req_capsule_set_size(tsi->tsi_pill,
434                                              &RMF_ACL, RCL_SERVER,
435                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
436
437                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_SHORT_IO,
438                                           RCL_SERVER)) {
439                         struct niobuf_remote *remote_nb =
440                                 req_capsule_client_get(tsi->tsi_pill,
441                                                        &RMF_NIOBUF_REMOTE);
442                         struct ost_body *body = tsi->tsi_ost_body;
443
444                         req_capsule_set_size(tsi->tsi_pill, &RMF_SHORT_IO,
445                                          RCL_SERVER,
446                                          (body->oa.o_flags & OBD_FL_SHORT_IO) ?
447                                          remote_nb[0].rnb_len : 0);
448                 }
449
450                 rc = req_capsule_server_pack(tsi->tsi_pill);
451         }
452
453         if (likely(rc == 0)) {
454                 /*
455                  * Process request, there can be two types of rc:
456                  * 1) errors with msg unpack/pack, other failures outside the
457                  * operation itself. This is counted as serious errors;
458                  * 2) errors during fs operation, should be placed in rq_status
459                  * only
460                  */
461                 rc = h->th_act(tsi);
462                 if (!is_serious(rc) &&
463                     !req->rq_no_reply && req->rq_reply_state == NULL) {
464                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
465                                   "pack reply and returned 0 error\n",
466                                   tgt_name(tsi->tsi_tgt), h->th_name);
467                         LBUG();
468                 }
469                 serious = is_serious(rc);
470                 rc = clear_serious(rc);
471         } else {
472                 serious = 1;
473         }
474
475         req->rq_status = rc;
476
477         /*
478          * ELDLM_* codes which > 0 should be in rq_status only as well as
479          * all non-serious errors.
480          */
481         if (rc > 0 || !serious)
482                 rc = 0;
483
484         LASSERT(current->journal_info == NULL);
485
486         if (likely(rc == 0 && req->rq_export))
487                 target_committed_to_req(req);
488
489 out:
490         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
491         RETURN(0);
492 }
493
494 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
495                                        struct obd_device *obd, int *process)
496 {
497         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
498         case MDS_DISCONNECT:
499         case OST_DISCONNECT:
500         case OBD_IDX_READ:
501                 *process = 1;
502                 RETURN(0);
503         case MDS_CLOSE:
504         case MDS_SYNC: /* used in unmounting */
505         case OBD_PING:
506         case MDS_REINT:
507         case OUT_UPDATE:
508         case SEQ_QUERY:
509         case FLD_QUERY:
510         case FLD_READ:
511         case LDLM_ENQUEUE:
512         case OST_CREATE:
513         case OST_DESTROY:
514         case OST_PUNCH:
515         case OST_SETATTR:
516         case OST_SYNC:
517         case OST_WRITE:
518         case MDS_HSM_PROGRESS:
519         case MDS_HSM_STATE_SET:
520         case MDS_HSM_REQUEST:
521                 *process = target_queue_recovery_request(req, obd);
522                 RETURN(0);
523
524         default:
525                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
526                 *process = -EAGAIN;
527                 RETURN(0);
528         }
529 }
530
531 /*
532  * Handle recovery. Return:
533  *        +1: continue request processing;
534  *       -ve: abort immediately with the given error code;
535  *         0: send reply with error code in req->rq_status;
536  */
537 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
538 {
539         ENTRY;
540
541         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
542         case MDS_CONNECT:
543         case OST_CONNECT:
544         case MGS_CONNECT:
545         case SEC_CTX_INIT:
546         case SEC_CTX_INIT_CONT:
547         case SEC_CTX_FINI:
548                 RETURN(+1);
549         }
550
551         if (!req->rq_export->exp_obd->obd_replayable)
552                 RETURN(+1);
553
554         /* sanity check: if the xid matches, the request must be marked as a
555          * resent or replayed */
556         if (req_can_reconstruct(req, NULL)) {
557                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
558                       (MSG_RESENT | MSG_REPLAY))) {
559                         DEBUG_REQ(D_WARNING, req, "rq_xid %llu matches "
560                                   "saved xid, expected REPLAY or RESENT flag "
561                                   "(%x)", req->rq_xid,
562                                   lustre_msg_get_flags(req->rq_reqmsg));
563                         req->rq_status = -ENOTCONN;
564                         RETURN(-ENOTCONN);
565                 }
566         }
567         /* else: note the opposite is not always true; a RESENT req after a
568          * failover will usually not match the last_xid, since it was likely
569          * never committed. A REPLAYed request will almost never match the
570          * last xid, however it could for a committed, but still retained,
571          * open. */
572
573         /* Check for aborted recovery... */
574         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
575                 int rc;
576                 int should_process;
577
578                 DEBUG_REQ(D_INFO, req, "Got new replay");
579                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
580                                                  &should_process);
581                 if (rc != 0 || !should_process)
582                         RETURN(rc);
583                 else if (should_process < 0) {
584                         req->rq_status = should_process;
585                         rc = ptlrpc_error(req);
586                         RETURN(rc);
587                 }
588         }
589         RETURN(+1);
590 }
591
592 /* Initial check for request, it is validation mostly */
593 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
594 {
595         struct tgt_handler      *h;
596         struct tgt_opc_slice    *s;
597         struct lu_target        *tgt;
598         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
599
600         ENTRY;
601
602         tgt = class_exp2tgt(req->rq_export);
603         if (unlikely(tgt == NULL)) {
604                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
605                           class_exp2obd(req->rq_export)->obd_name);
606                 RETURN(ERR_PTR(-EINVAL));
607         }
608
609         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
610                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
611                         break;
612
613         /* opcode was not found in slice */
614         if (unlikely(s->tos_hs == NULL)) {
615                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
616                        opc);
617                 RETURN(ERR_PTR(-ENOTSUPP));
618         }
619
620         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
621         h = s->tos_hs + (opc - s->tos_opc_start);
622         if (unlikely(h->th_opc == 0)) {
623                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
624                 RETURN(ERR_PTR(-ENOTSUPP));
625         }
626
627         RETURN(h);
628 }
629
630 static int process_req_last_xid(struct ptlrpc_request *req)
631 {
632         __u64   last_xid;
633         ENTRY;
634
635         /* check request's xid is consistent with export's last_xid */
636         last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
637         if (last_xid > req->rq_export->exp_last_xid)
638                 req->rq_export->exp_last_xid = last_xid;
639
640         if (req->rq_xid == 0 ||
641             (req->rq_xid <= req->rq_export->exp_last_xid)) {
642                 DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
643                           "last_xid %llx\n", req->rq_xid,
644                           req->rq_export->exp_last_xid);
645                 /* Some request is allowed to be sent during replay,
646                  * such as OUT update requests, FLD requests, so it
647                  * is possible that replay requests has smaller XID
648                  * than the exp_last_xid.
649                  *
650                  * Some non-replay requests may have smaller XID as
651                  * well:
652                  *
653                  * - Client send a no_resend RPC, like statfs;
654                  * - The RPC timedout (or some other error) on client,
655                  *   then it's removed from the unreplied list;
656                  * - Client send some other request to bump the
657                  *   exp_last_xid on server;
658                  * - The former RPC got chance to be processed;
659                  */
660                 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
661                         RETURN(-EPROTO);
662         }
663
664         /* try to release in-memory reply data */
665         if (tgt_is_multimodrpcs_client(req->rq_export)) {
666                 tgt_handle_received_xid(req->rq_export,
667                                 lustre_msg_get_last_xid(req->rq_reqmsg));
668                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
669                       (MSG_RESENT | MSG_REPLAY)))
670                         tgt_handle_tag(req->rq_export,
671                                        lustre_msg_get_tag(req->rq_reqmsg));
672         }
673         RETURN(0);
674 }
675
676 int tgt_request_handle(struct ptlrpc_request *req)
677 {
678         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
679
680         struct lustre_msg       *msg = req->rq_reqmsg;
681         struct tgt_handler      *h;
682         struct lu_target        *tgt;
683         int                      request_fail_id = 0;
684         __u32                    opc = lustre_msg_get_opc(msg);
685         struct obd_device       *obd;
686         int                      rc;
687         bool                     is_connect = false;
688         ENTRY;
689
690         /* Refill the context, to make sure all thread keys are allocated */
691         lu_env_refill(req->rq_svc_thread->t_env);
692
693         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
694         tsi->tsi_pill = &req->rq_pill;
695         tsi->tsi_env = req->rq_svc_thread->t_env;
696
697         /* if request has export then get handlers slice from corresponding
698          * target, otherwise that should be connect operation */
699         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
700             opc == MGS_CONNECT) {
701                 is_connect = true;
702                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
703                 rc = target_handle_connect(req);
704                 if (rc != 0) {
705                         rc = ptlrpc_error(req);
706                         GOTO(out, rc);
707                 }
708                 /* recovery-small test 18c asks to drop connect reply */
709                 if (unlikely(opc == OST_CONNECT &&
710                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
711                         GOTO(out, rc = 0);
712         }
713
714         if (unlikely(!class_connected_export(req->rq_export))) {
715                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
716                     opc == SEC_CTX_FINI) {
717                         /* sec context initialization has to be handled
718                          * by hand in tgt_handle_request0() */
719                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
720                         h = NULL;
721                         GOTO(handle_recov, rc = 0);
722                 }
723                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
724                        opc, libcfs_id2str(req->rq_peer));
725                 req->rq_status = -ENOTCONN;
726                 rc = ptlrpc_error(req);
727                 GOTO(out, rc);
728         }
729
730         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
731         tsi->tsi_exp = req->rq_export;
732         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
733                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
734         else
735                 tsi->tsi_jobid = NULL;
736
737         if (tgt == NULL) {
738                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
739                           class_exp2obd(req->rq_export)->obd_name);
740                 req->rq_status = -EINVAL;
741                 rc = ptlrpc_error(req);
742                 GOTO(out, rc);
743         }
744
745         /* Skip last_xid processing for the recovery thread, otherwise, the
746          * last_xid on same request could be processed twice: first time when
747          * processing the incoming request, second time when the request is
748          * being processed by recovery thread. */
749         obd = class_exp2obd(req->rq_export);
750         if (is_connect) {
751                 /* reset the exp_last_xid on each connection. */
752                 req->rq_export->exp_last_xid = 0;
753         } else if (obd->obd_recovery_data.trd_processing_task !=
754                    current_pid()) {
755                 rc = process_req_last_xid(req);
756                 if (rc) {
757                         req->rq_status = rc;
758                         rc = ptlrpc_error(req);
759                         GOTO(out, rc);
760                 }
761         }
762
763         request_fail_id = tgt->lut_request_fail_id;
764         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
765
766         h = tgt_handler_find_check(req);
767         if (IS_ERR(h)) {
768                 req->rq_status = PTR_ERR(h);
769                 rc = ptlrpc_error(req);
770                 GOTO(out, rc);
771         }
772
773         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
774                  h->th_opc, opc);
775
776         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
777                 GOTO(out, rc = 0);
778
779         rc = lustre_msg_check_version(msg, h->th_version);
780         if (unlikely(rc)) {
781                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
782                           " %08x, expecting %08x\n", tgt_name(tgt),
783                           lustre_msg_get_version(msg), h->th_version);
784                 req->rq_status = -EINVAL;
785                 rc = ptlrpc_error(req);
786                 GOTO(out, rc);
787         }
788
789 handle_recov:
790         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
791         if (likely(rc == 1)) {
792                 rc = tgt_handle_request0(tsi, h, req);
793                 if (rc)
794                         GOTO(out, rc);
795         }
796         EXIT;
797 out:
798         req_capsule_fini(tsi->tsi_pill);
799         if (tsi->tsi_corpus != NULL) {
800                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
801                 tsi->tsi_corpus = NULL;
802         }
803         return rc;
804 }
805 EXPORT_SYMBOL(tgt_request_handle);
806
807 /** Assign high priority operations to the request if needed. */
808 int tgt_hpreq_handler(struct ptlrpc_request *req)
809 {
810         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
811         struct tgt_handler      *h;
812         int                      rc;
813
814         ENTRY;
815
816         if (req->rq_export == NULL)
817                 RETURN(0);
818
819         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
820         tsi->tsi_pill = &req->rq_pill;
821         tsi->tsi_env = req->rq_svc_thread->t_env;
822         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
823         tsi->tsi_exp = req->rq_export;
824
825         h = tgt_handler_find_check(req);
826         if (IS_ERR(h)) {
827                 rc = PTR_ERR(h);
828                 RETURN(rc);
829         }
830
831         rc = tgt_request_preprocess(tsi, h, req);
832         if (unlikely(rc != 0))
833                 RETURN(rc);
834
835         if (h->th_hp != NULL)
836                 h->th_hp(tsi);
837         RETURN(0);
838 }
839 EXPORT_SYMBOL(tgt_hpreq_handler);
840
841 void tgt_counter_incr(struct obd_export *exp, int opcode)
842 {
843         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
844         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
845                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
846 }
847 EXPORT_SYMBOL(tgt_counter_incr);
848
849 /*
850  * Unified target generic handlers.
851  */
852
853 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
854 {
855         struct lu_target        *tgt = class_exp2tgt(exp);
856         struct sptlrpc_flavor    flvr;
857         int                      rc = 0;
858
859         LASSERT(tgt);
860         LASSERT(tgt->lut_obd);
861         LASSERT(tgt->lut_slice);
862
863         /* always allow ECHO client */
864         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
865                             LUSTRE_ECHO_NAME) == 0)) {
866                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
867                 return 0;
868         }
869
870         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
871                 read_lock(&tgt->lut_sptlrpc_lock);
872                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
873                                              req->rq_sp_from,
874                                              req->rq_peer.nid,
875                                              &flvr);
876                 read_unlock(&tgt->lut_sptlrpc_lock);
877
878                 spin_lock(&exp->exp_lock);
879                 exp->exp_sp_peer = req->rq_sp_from;
880                 exp->exp_flvr = flvr;
881
882                 /* when on mgs, if no restriction is set, or if client
883                  * is loopback, allow any flavor */
884                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
885                            LUSTRE_MGS_NAME) == 0) &&
886                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
887                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
888                       == LOLND))
889                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
890
891                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
892                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
893                         CERROR("%s: unauthorized rpc flavor %x from %s, "
894                                "expect %x\n", tgt_name(tgt),
895                                req->rq_flvr.sf_rpc,
896                                libcfs_nid2str(req->rq_peer.nid),
897                                exp->exp_flvr.sf_rpc);
898                         rc = -EACCES;
899                 }
900                 spin_unlock(&exp->exp_lock);
901         } else {
902                 if (exp->exp_sp_peer != req->rq_sp_from) {
903                         CERROR("%s: RPC source %s doesn't match %s\n",
904                                tgt_name(tgt),
905                                sptlrpc_part2name(req->rq_sp_from),
906                                sptlrpc_part2name(exp->exp_sp_peer));
907                         rc = -EACCES;
908                 } else {
909                         rc = sptlrpc_target_export_check(exp, req);
910                 }
911         }
912
913         return rc;
914 }
915
916 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt)
917 {
918         struct sptlrpc_rule_set  tmp_rset;
919         int                      rc;
920
921         if (unlikely(tgt == NULL)) {
922                 CERROR("No target passed");
923                 return -EINVAL;
924         }
925
926         sptlrpc_rule_set_init(&tmp_rset);
927         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset);
928         if (rc) {
929                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
930                        tgt_name(tgt), rc);
931                 return rc;
932         }
933
934         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
935
936         write_lock(&tgt->lut_sptlrpc_lock);
937         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
938         tgt->lut_sptlrpc_rset = tmp_rset;
939         write_unlock(&tgt->lut_sptlrpc_lock);
940
941         return 0;
942 }
943 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
944
945 int tgt_connect(struct tgt_session_info *tsi)
946 {
947         struct ptlrpc_request   *req = tgt_ses_req(tsi);
948         struct obd_connect_data *reply;
949         int                      rc;
950
951         ENTRY;
952
953         /* XXX: better to call this check right after getting new export but
954          * before last_rcvd slot allocation to avoid server load upon insecure
955          * connects. This is to be fixed after unifiyng all targets.
956          */
957         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
958         if (rc)
959                 GOTO(out, rc);
960
961         /* To avoid exposing partially initialized connection flags, changes up
962          * to this point have been staged in reply->ocd_connect_flags. Now that
963          * connection handling has completed successfully, atomically update
964          * the connect flags in the shared export data structure. LU-1623 */
965         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
966         spin_lock(&tsi->tsi_exp->exp_lock);
967         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
968         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
969         spin_unlock(&tsi->tsi_exp->exp_lock);
970
971         RETURN(0);
972 out:
973         obd_disconnect(class_export_get(tsi->tsi_exp));
974         return rc;
975 }
976 EXPORT_SYMBOL(tgt_connect);
977
978 int tgt_disconnect(struct tgt_session_info *tsi)
979 {
980         int rc;
981
982         ENTRY;
983
984         rc = target_handle_disconnect(tgt_ses_req(tsi));
985         if (rc)
986                 RETURN(err_serious(rc));
987
988         RETURN(rc);
989 }
990 EXPORT_SYMBOL(tgt_disconnect);
991
992 /*
993  * Unified target OBD handlers
994  */
995 int tgt_obd_ping(struct tgt_session_info *tsi)
996 {
997         int rc;
998
999         ENTRY;
1000
1001         rc = target_handle_ping(tgt_ses_req(tsi));
1002         if (rc)
1003                 RETURN(err_serious(rc));
1004
1005         RETURN(rc);
1006 }
1007 EXPORT_SYMBOL(tgt_obd_ping);
1008
1009 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1010 {
1011         return err_serious(-EOPNOTSUPP);
1012 }
1013
1014 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1015 {
1016         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1017         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1018         struct obd_export       *exp = req->rq_export;
1019         struct ptlrpc_bulk_desc *desc;
1020         struct l_wait_info      *lwi = &tti->tti_u.update.tti_wait_info;
1021         int                      i;
1022         int                      rc;
1023
1024         ENTRY;
1025
1026         desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1027                                   PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1028                                     MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1029         if (desc == NULL)
1030                 RETURN(-ENOMEM);
1031
1032         for (i = 0; i < rdbuf->rb_nbufs; i++)
1033                 desc->bd_frag_ops->add_iov_frag(desc,
1034                                         rdbuf->rb_bufs[i].lb_buf,
1035                                         rdbuf->rb_bufs[i].lb_len);
1036
1037         rc = target_bulk_io(exp, desc, lwi);
1038         ptlrpc_free_bulk(desc);
1039         RETURN(rc);
1040 }
1041 EXPORT_SYMBOL(tgt_send_buffer);
1042
1043 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1044 {
1045         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1046         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1047         struct obd_export       *exp = req->rq_export;
1048         struct ptlrpc_bulk_desc *desc;
1049         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1050         int                      tmpcount;
1051         int                      tmpsize;
1052         int                      i;
1053         int                      rc;
1054
1055         ENTRY;
1056
1057         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1058                                     PTLRPC_BULK_PUT_SOURCE |
1059                                         PTLRPC_BULK_BUF_KIOV,
1060                                     MDS_BULK_PORTAL,
1061                                     &ptlrpc_bulk_kiov_pin_ops);
1062         if (desc == NULL)
1063                 RETURN(-ENOMEM);
1064
1065         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1066                 /* old client requires reply size in it's PAGE_SIZE,
1067                  * which is rdpg->rp_count */
1068                 nob = rdpg->rp_count;
1069
1070         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1071              i++, tmpcount -= tmpsize) {
1072                 tmpsize = min_t(int, tmpcount, PAGE_SIZE);
1073                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1074                                                  tmpsize);
1075         }
1076
1077         LASSERT(desc->bd_nob == nob);
1078         rc = target_bulk_io(exp, desc, lwi);
1079         ptlrpc_free_bulk(desc);
1080         RETURN(rc);
1081 }
1082 EXPORT_SYMBOL(tgt_sendpage);
1083
1084 /*
1085  * OBD_IDX_READ handler
1086  */
1087 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1088 {
1089         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1090         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1091         struct idx_info         *req_ii, *rep_ii;
1092         int                      rc, i;
1093
1094         ENTRY;
1095
1096         memset(rdpg, 0, sizeof(*rdpg));
1097         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1098
1099         /* extract idx_info buffer from request & reply */
1100         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1101         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1102                 RETURN(err_serious(-EPROTO));
1103
1104         rc = req_capsule_server_pack(tsi->tsi_pill);
1105         if (rc)
1106                 RETURN(err_serious(rc));
1107
1108         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1109         if (rep_ii == NULL)
1110                 RETURN(err_serious(-EFAULT));
1111         rep_ii->ii_magic = IDX_INFO_MAGIC;
1112
1113         /* extract hash to start with */
1114         rdpg->rp_hash = req_ii->ii_hash_start;
1115
1116         /* extract requested attributes */
1117         rdpg->rp_attrs = req_ii->ii_attrs;
1118
1119         /* check that fid packed in request is valid and supported */
1120         if (!fid_is_sane(&req_ii->ii_fid))
1121                 RETURN(-EINVAL);
1122         rep_ii->ii_fid = req_ii->ii_fid;
1123
1124         /* copy flags */
1125         rep_ii->ii_flags = req_ii->ii_flags;
1126
1127         /* compute number of pages to allocate, ii_count is the number of 4KB
1128          * containers */
1129         if (req_ii->ii_count <= 0)
1130                 GOTO(out, rc = -EFAULT);
1131         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1132                                exp_max_brw_size(tsi->tsi_exp));
1133         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> PAGE_SHIFT;
1134
1135         /* allocate pages to store the containers */
1136         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1137         if (rdpg->rp_pages == NULL)
1138                 GOTO(out, rc = -ENOMEM);
1139         for (i = 0; i < rdpg->rp_npages; i++) {
1140                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1141                 if (rdpg->rp_pages[i] == NULL)
1142                         GOTO(out, rc = -ENOMEM);
1143         }
1144
1145         /* populate pages with key/record pairs */
1146         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1147         if (rc < 0)
1148                 GOTO(out, rc);
1149
1150         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1151                  "asked %d > %d\n", rc, rdpg->rp_count);
1152
1153         /* send pages to client */
1154         rc = tgt_sendpage(tsi, rdpg, rc);
1155         if (rc)
1156                 GOTO(out, rc);
1157         EXIT;
1158 out:
1159         if (rdpg->rp_pages) {
1160                 for (i = 0; i < rdpg->rp_npages; i++)
1161                         if (rdpg->rp_pages[i])
1162                                 __free_page(rdpg->rp_pages[i]);
1163                 OBD_FREE(rdpg->rp_pages,
1164                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1165         }
1166         return rc;
1167 }
1168
1169 struct tgt_handler tgt_obd_handlers[] = {
1170 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1171 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1172 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1173 };
1174 EXPORT_SYMBOL(tgt_obd_handlers);
1175
1176 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1177              struct dt_object *obj, __u64 start, __u64 end)
1178 {
1179         int rc = 0;
1180
1181         ENTRY;
1182
1183         /* if no objid is specified, it means "sync whole filesystem" */
1184         if (obj == NULL) {
1185                 rc = dt_sync(env, tgt->lut_bottom);
1186         } else if (dt_version_get(env, obj) >
1187                    tgt->lut_obd->obd_last_committed) {
1188                 rc = dt_object_sync(env, obj, start, end);
1189         }
1190         atomic_inc(&tgt->lut_sync_count);
1191
1192         RETURN(rc);
1193 }
1194 EXPORT_SYMBOL(tgt_sync);
1195 /*
1196  * Unified target DLM handlers.
1197  */
1198
1199 /**
1200  * Unified target BAST
1201  *
1202  * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1203  * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1204  * but if it's cross-MDT lock, because remote object version is not set, a
1205  * filesystem sync is needed.
1206  *
1207  * \param lock server side lock
1208  * \param desc lock desc
1209  * \param data ldlm_cb_set_arg
1210  * \param flag  indicates whether this cancelling or blocking callback
1211  * \retval      0 on success
1212  * \retval      negative number on error
1213  */
1214 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1215                             void *data, int flag)
1216 {
1217         struct lu_env            env;
1218         struct lu_target        *tgt;
1219         struct dt_object        *obj = NULL;
1220         struct lu_fid            fid;
1221         int                      rc = 0;
1222
1223         ENTRY;
1224
1225         tgt = class_exp2tgt(lock->l_export);
1226
1227         if (unlikely(tgt == NULL)) {
1228                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1229                        class_exp2obd(lock->l_export)->obd_name);
1230                 RETURN(-EINVAL);
1231         }
1232
1233         if (flag == LDLM_CB_CANCELING &&
1234             (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1235             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1236              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1237               ldlm_is_cbpending(lock))) &&
1238             ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1239              lock->l_resource->lr_type == LDLM_EXTENT)) {
1240                 __u64 start = 0;
1241                 __u64 end = OBD_OBJECT_EOF;
1242
1243                 rc = lu_env_init(&env, LCT_DT_THREAD);
1244                 if (unlikely(rc != 0))
1245                         RETURN(rc);
1246
1247                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1248                                    tgt->lut_lsd.lsd_osd_index);
1249
1250                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1251                         obj = dt_locate(&env, tgt->lut_bottom, &fid);
1252                         if (IS_ERR(obj))
1253                                 GOTO(err_env, rc = PTR_ERR(obj));
1254
1255                         if (!dt_object_exists(obj))
1256                                 GOTO(err_put, rc = -ENOENT);
1257
1258                         start = lock->l_policy_data.l_extent.start;
1259                         end = lock->l_policy_data.l_extent.end;
1260                 }
1261
1262                 rc = tgt_sync(&env, tgt, obj, start, end);
1263                 if (rc < 0) {
1264                         CERROR("%s: syncing "DFID" (%llu-%llu) on lock "
1265                                "cancel: rc = %d\n",
1266                                tgt_name(tgt), PFID(&fid),
1267                                lock->l_policy_data.l_extent.start,
1268                                lock->l_policy_data.l_extent.end, rc);
1269                 }
1270 err_put:
1271                 if (obj != NULL)
1272                         dt_object_put(&env, obj);
1273 err_env:
1274                 lu_env_fini(&env);
1275         }
1276
1277         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1278         RETURN(rc);
1279 }
1280
1281 static struct ldlm_callback_suite tgt_dlm_cbs = {
1282         .lcs_completion = ldlm_server_completion_ast,
1283         .lcs_blocking   = tgt_blocking_ast,
1284         .lcs_glimpse    = ldlm_server_glimpse_ast
1285 };
1286
1287 int tgt_enqueue(struct tgt_session_info *tsi)
1288 {
1289         struct ptlrpc_request *req = tgt_ses_req(tsi);
1290         int rc;
1291
1292         ENTRY;
1293         /*
1294          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1295          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1296          */
1297         LASSERT(tsi->tsi_dlm_req != NULL);
1298         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1299                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1300         if (rc)
1301                 RETURN(err_serious(rc));
1302
1303         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1304         case LUT_FAIL_MDT:
1305                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1306                 break;
1307         case LUT_FAIL_OST:
1308                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1309                 break;
1310         case LUT_FAIL_MGT:
1311                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1312                 break;
1313         default:
1314                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1315                 break;
1316         }
1317         RETURN(req->rq_status);
1318 }
1319 EXPORT_SYMBOL(tgt_enqueue);
1320
1321 int tgt_convert(struct tgt_session_info *tsi)
1322 {
1323         struct ptlrpc_request *req = tgt_ses_req(tsi);
1324         int rc;
1325
1326         ENTRY;
1327         LASSERT(tsi->tsi_dlm_req);
1328         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1329         if (rc)
1330                 RETURN(err_serious(rc));
1331
1332         RETURN(req->rq_status);
1333 }
1334
1335 int tgt_bl_callback(struct tgt_session_info *tsi)
1336 {
1337         return err_serious(-EOPNOTSUPP);
1338 }
1339
1340 int tgt_cp_callback(struct tgt_session_info *tsi)
1341 {
1342         return err_serious(-EOPNOTSUPP);
1343 }
1344
1345 /* generic LDLM target handler */
1346 struct tgt_handler tgt_dlm_handlers[] = {
1347 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1348 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1349 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1350 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1351 };
1352 EXPORT_SYMBOL(tgt_dlm_handlers);
1353
1354 /*
1355  * Unified target LLOG handlers.
1356  */
1357 int tgt_llog_open(struct tgt_session_info *tsi)
1358 {
1359         int rc;
1360
1361         ENTRY;
1362
1363         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1364
1365         RETURN(rc);
1366 }
1367 EXPORT_SYMBOL(tgt_llog_open);
1368
1369 int tgt_llog_close(struct tgt_session_info *tsi)
1370 {
1371         int rc;
1372
1373         ENTRY;
1374
1375         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1376
1377         RETURN(rc);
1378 }
1379 EXPORT_SYMBOL(tgt_llog_close);
1380
1381
1382 int tgt_llog_destroy(struct tgt_session_info *tsi)
1383 {
1384         int rc;
1385
1386         ENTRY;
1387
1388         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1389
1390         RETURN(rc);
1391 }
1392
1393 int tgt_llog_read_header(struct tgt_session_info *tsi)
1394 {
1395         int rc;
1396
1397         ENTRY;
1398
1399         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1400
1401         RETURN(rc);
1402 }
1403 EXPORT_SYMBOL(tgt_llog_read_header);
1404
1405 int tgt_llog_next_block(struct tgt_session_info *tsi)
1406 {
1407         int rc;
1408
1409         ENTRY;
1410
1411         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1412
1413         RETURN(rc);
1414 }
1415 EXPORT_SYMBOL(tgt_llog_next_block);
1416
1417 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1418 {
1419         int rc;
1420
1421         ENTRY;
1422
1423         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1424
1425         RETURN(rc);
1426 }
1427 EXPORT_SYMBOL(tgt_llog_prev_block);
1428
1429 /* generic llog target handler */
1430 struct tgt_handler tgt_llog_handlers[] = {
1431 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1432 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1433 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1434 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1435 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1436 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1437 };
1438 EXPORT_SYMBOL(tgt_llog_handlers);
1439
1440 /*
1441  * sec context handlers
1442  */
1443 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1444 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1445 {
1446         return 0;
1447 }
1448
1449 struct tgt_handler tgt_sec_ctx_handlers[] = {
1450 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT,           tgt_sec_ctx_handle),
1451 TGT_SEC_HDL_VAR(0,      SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
1452 TGT_SEC_HDL_VAR(0,      SEC_CTX_FINI,           tgt_sec_ctx_handle),
1453 };
1454 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1455
1456 int (*tgt_lfsck_in_notify_local)(const struct lu_env *env,
1457                                  struct dt_device *key,
1458                                  struct lfsck_req_local *lrl,
1459                                  struct thandle *th) = NULL;
1460
1461 void tgt_register_lfsck_in_notify_local(int (*notify)(const struct lu_env *,
1462                                                       struct dt_device *,
1463                                                       struct lfsck_req_local *,
1464                                                       struct thandle *))
1465 {
1466         tgt_lfsck_in_notify_local = notify;
1467 }
1468 EXPORT_SYMBOL(tgt_register_lfsck_in_notify_local);
1469
1470 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1471                            struct dt_device *key,
1472                            struct lfsck_request *lr) = NULL;
1473
1474 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1475                                                 struct dt_device *,
1476                                                 struct lfsck_request *))
1477 {
1478         tgt_lfsck_in_notify = notify;
1479 }
1480 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1481
1482 static int (*tgt_lfsck_query)(const struct lu_env *env,
1483                               struct dt_device *key,
1484                               struct lfsck_request *req,
1485                               struct lfsck_reply *rep,
1486                               struct lfsck_query *que) = NULL;
1487
1488 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1489                                            struct dt_device *,
1490                                            struct lfsck_request *,
1491                                            struct lfsck_reply *,
1492                                            struct lfsck_query *))
1493 {
1494         tgt_lfsck_query = query;
1495 }
1496 EXPORT_SYMBOL(tgt_register_lfsck_query);
1497
1498 /* LFSCK request handlers */
1499 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1500 {
1501         const struct lu_env     *env = tsi->tsi_env;
1502         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1503         struct lfsck_request    *lr;
1504         int                      rc;
1505         ENTRY;
1506
1507         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1508         if (lr == NULL)
1509                 RETURN(-EPROTO);
1510
1511         rc = tgt_lfsck_in_notify(env, key, lr);
1512
1513         RETURN(rc);
1514 }
1515
1516 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1517 {
1518         struct lfsck_request    *request;
1519         struct lfsck_reply      *reply;
1520         int                      rc;
1521         ENTRY;
1522
1523         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1524         if (request == NULL)
1525                 RETURN(-EPROTO);
1526
1527         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1528         if (reply == NULL)
1529                 RETURN(-ENOMEM);
1530
1531         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1532                              request, reply, NULL);
1533
1534         RETURN(rc < 0 ? rc : 0);
1535 }
1536
1537 struct tgt_handler tgt_lfsck_handlers[] = {
1538 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1539 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1540 };
1541 EXPORT_SYMBOL(tgt_lfsck_handlers);
1542
1543 /*
1544  * initialize per-thread page pool (bug 5137).
1545  */
1546 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1547 {
1548         struct tgt_thread_big_cache *tbc;
1549
1550         ENTRY;
1551
1552         LASSERT(thread != NULL);
1553         LASSERT(thread->t_data == NULL);
1554
1555         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1556         if (tbc == NULL)
1557                 RETURN(-ENOMEM);
1558         thread->t_data = tbc;
1559         RETURN(0);
1560 }
1561 EXPORT_SYMBOL(tgt_io_thread_init);
1562
1563 /*
1564  * free per-thread pool created by tgt_thread_init().
1565  */
1566 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1567 {
1568         struct tgt_thread_big_cache *tbc;
1569
1570         ENTRY;
1571
1572         LASSERT(thread != NULL);
1573
1574         /*
1575          * be prepared to handle partially-initialized pools (because this is
1576          * called from ost_io_thread_init() for cleanup.
1577          */
1578         tbc = thread->t_data;
1579         if (tbc != NULL) {
1580                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1581                 thread->t_data = NULL;
1582         }
1583         EXIT;
1584 }
1585 EXPORT_SYMBOL(tgt_io_thread_done);
1586
1587 /**
1588  * Helper function for getting Data-on-MDT file server DLM lock
1589  * if asked by client.
1590  */
1591 int tgt_mdt_data_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1592                       struct lustre_handle *lh, int mode, __u64 *flags)
1593 {
1594         union ldlm_policy_data policy = {
1595                 .l_inodebits.bits = MDS_INODELOCK_DOM,
1596         };
1597         int rc;
1598
1599         ENTRY;
1600
1601         LASSERT(lh != NULL);
1602         LASSERT(ns != NULL);
1603         LASSERT(!lustre_handle_is_used(lh));
1604
1605         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, &policy, mode,
1606                                     flags, ldlm_blocking_ast,
1607                                     ldlm_completion_ast, ldlm_glimpse_ast,
1608                                     NULL, 0, LVB_T_NONE, NULL, lh);
1609
1610         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1611 }
1612 EXPORT_SYMBOL(tgt_mdt_data_lock);
1613
1614 void tgt_mdt_data_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1615 {
1616         LASSERT(lustre_handle_is_used(lh));
1617         ldlm_lock_decref(lh, mode);
1618 }
1619 EXPORT_SYMBOL(tgt_mdt_data_unlock);
1620
1621 /**
1622  * Helper function for getting server side [start, start+count] DLM lock
1623  * if asked by client.
1624  */
1625 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1626                     __u64 start, __u64 end, struct lustre_handle *lh,
1627                     int mode, __u64 *flags)
1628 {
1629         union ldlm_policy_data policy;
1630         int rc;
1631
1632         ENTRY;
1633
1634         LASSERT(lh != NULL);
1635         LASSERT(ns != NULL);
1636         LASSERT(!lustre_handle_is_used(lh));
1637
1638         policy.l_extent.gid = 0;
1639         policy.l_extent.start = start & PAGE_MASK;
1640
1641         /*
1642          * If ->o_blocks is EOF it means "lock till the end of the file".
1643          * Otherwise, it's size of an extent or hole being punched (in bytes).
1644          */
1645         if (end == OBD_OBJECT_EOF || end < start)
1646                 policy.l_extent.end = OBD_OBJECT_EOF;
1647         else
1648                 policy.l_extent.end = end | ~PAGE_MASK;
1649
1650         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1651                                     flags, ldlm_blocking_ast,
1652                                     ldlm_completion_ast, ldlm_glimpse_ast,
1653                                     NULL, 0, LVB_T_NONE, NULL, lh);
1654         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1655 }
1656 EXPORT_SYMBOL(tgt_extent_lock);
1657
1658 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1659 {
1660         LASSERT(lustre_handle_is_used(lh));
1661         ldlm_lock_decref(lh, mode);
1662 }
1663 EXPORT_SYMBOL(tgt_extent_unlock);
1664
1665 static int tgt_brw_lock(struct obd_export *exp, struct ldlm_res_id *res_id,
1666                         struct obd_ioobj *obj, struct niobuf_remote *nb,
1667                         struct lustre_handle *lh, enum ldlm_mode mode)
1668 {
1669         struct ldlm_namespace   *ns = exp->exp_obd->obd_namespace;
1670         __u64                    flags = 0;
1671         int                      nrbufs = obj->ioo_bufcnt;
1672         int                      i;
1673         int                      rc;
1674
1675         ENTRY;
1676
1677         LASSERT(mode == LCK_PR || mode == LCK_PW);
1678         LASSERT(!lustre_handle_is_used(lh));
1679
1680         if (ns->ns_obd->obd_recovering)
1681                 RETURN(0);
1682
1683         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1684                 RETURN(0);
1685
1686         for (i = 1; i < nrbufs; i++)
1687                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1688                         RETURN(-EFAULT);
1689
1690         /* MDT IO for data-on-mdt */
1691         if (exp->exp_connect_data.ocd_connect_flags & OBD_CONNECT_IBITS)
1692                 rc = tgt_mdt_data_lock(ns, res_id, lh, mode, &flags);
1693         else
1694                 rc = tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1695                                      nb[nrbufs - 1].rnb_offset +
1696                                      nb[nrbufs - 1].rnb_len - 1,
1697                                      lh, mode, &flags);
1698         RETURN(rc);
1699 }
1700
1701 static void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1702                            struct lustre_handle *lh, enum ldlm_mode mode)
1703 {
1704         ENTRY;
1705
1706         LASSERT(mode == LCK_PR || mode == LCK_PW);
1707         LASSERT((obj->ioo_bufcnt > 0 &&
1708                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1709                 lustre_handle_is_used(lh));
1710
1711         if (lustre_handle_is_used(lh))
1712                 tgt_extent_unlock(lh, mode);
1713         EXIT;
1714 }
1715 static __u32 tgt_checksum_niobuf(struct lu_target *tgt,
1716                                  struct niobuf_local *local_nb, int npages,
1717                                  int opc, enum cksum_types cksum_type)
1718 {
1719         struct cfs_crypto_hash_desc     *hdesc;
1720         unsigned int                    bufsize;
1721         int                             i, err;
1722         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1723         __u32                           cksum;
1724
1725         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1726         if (IS_ERR(hdesc)) {
1727                 CERROR("%s: unable to initialize checksum hash %s\n",
1728                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1729                 return PTR_ERR(hdesc);
1730         }
1731
1732         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1733         for (i = 0; i < npages; i++) {
1734                 /* corrupt the data before we compute the checksum, to
1735                  * simulate a client->OST data error */
1736                 if (i == 0 && opc == OST_WRITE &&
1737                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1738                         int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
1739                         int len = local_nb[i].lnb_len;
1740                         struct page *np = tgt_page_to_corrupt;
1741
1742                         if (np) {
1743                                 char *ptr = ll_kmap_atomic(local_nb[i].lnb_page,
1744                                                         KM_USER0);
1745                                 char *ptr2 = page_address(np);
1746
1747                                 memcpy(ptr2 + off, ptr + off, len);
1748                                 memcpy(ptr2 + off, "bad3", min(4, len));
1749                                 ll_kunmap_atomic(ptr, KM_USER0);
1750
1751                                 /* LU-8376 to preserve original index for
1752                                  * display in dump_all_bulk_pages() */
1753                                 np->index = i;
1754
1755                                 cfs_crypto_hash_update_page(hdesc, np, off,
1756                                                             len);
1757                                 continue;
1758                         } else {
1759                                 CERROR("%s: can't alloc page for corruption\n",
1760                                        tgt_name(tgt));
1761                         }
1762                 }
1763                 cfs_crypto_hash_update_page(hdesc, local_nb[i].lnb_page,
1764                                   local_nb[i].lnb_page_offset & ~PAGE_MASK,
1765                                   local_nb[i].lnb_len);
1766
1767                  /* corrupt the data after we compute the checksum, to
1768                  * simulate an OST->client data error */
1769                 if (i == 0 && opc == OST_READ &&
1770                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1771                         int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
1772                         int len = local_nb[i].lnb_len;
1773                         struct page *np = tgt_page_to_corrupt;
1774
1775                         if (np) {
1776                                 char *ptr = ll_kmap_atomic(local_nb[i].lnb_page,
1777                                                         KM_USER0);
1778                                 char *ptr2 = page_address(np);
1779
1780                                 memcpy(ptr2 + off, ptr + off, len);
1781                                 memcpy(ptr2 + off, "bad4", min(4, len));
1782                                 ll_kunmap_atomic(ptr, KM_USER0);
1783
1784                                 /* LU-8376 to preserve original index for
1785                                  * display in dump_all_bulk_pages() */
1786                                 np->index = i;
1787
1788                                 cfs_crypto_hash_update_page(hdesc, np, off,
1789                                                             len);
1790                                 continue;
1791                         } else {
1792                                 CERROR("%s: can't alloc page for corruption\n",
1793                                        tgt_name(tgt));
1794                         }
1795                 }
1796         }
1797
1798         bufsize = sizeof(cksum);
1799         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1800
1801         return cksum;
1802 }
1803
1804 char dbgcksum_file_name[PATH_MAX];
1805
1806 static void dump_all_bulk_pages(struct obdo *oa, int count,
1807                                 struct niobuf_local *local_nb,
1808                                 __u32 server_cksum, __u32 client_cksum)
1809 {
1810         struct file *filp;
1811         int rc, i;
1812         unsigned int len;
1813         char *buf;
1814         mm_segment_t oldfs;
1815
1816         /* will only keep dump of pages on first error for the same range in
1817          * file/fid, not during the resends/retries. */
1818         snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name),
1819                  "%s-checksum_dump-ost-"DFID":[%llu-%llu]-%x-%x",
1820                  (strncmp(libcfs_debug_file_path_arr, "NONE", 4) != 0 ?
1821                   libcfs_debug_file_path_arr :
1822                   LIBCFS_DEBUG_FILE_PATH_DEFAULT),
1823                  oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1824                  oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1825                  oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1826                  local_nb[0].lnb_file_offset,
1827                  local_nb[count-1].lnb_file_offset +
1828                  local_nb[count-1].lnb_len - 1, client_cksum, server_cksum);
1829         filp = filp_open(dbgcksum_file_name,
1830                          O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600);
1831         if (IS_ERR(filp)) {
1832                 rc = PTR_ERR(filp);
1833                 if (rc == -EEXIST)
1834                         CDEBUG(D_INFO, "%s: can't open to dump pages with "
1835                                "checksum error: rc = %d\n", dbgcksum_file_name,
1836                                rc);
1837                 else
1838                         CERROR("%s: can't open to dump pages with checksum "
1839                                "error: rc = %d\n", dbgcksum_file_name, rc);
1840                 return;
1841         }
1842
1843         oldfs = get_fs();
1844         set_fs(KERNEL_DS);
1845         for (i = 0; i < count; i++) {
1846                 len = local_nb[i].lnb_len;
1847                 buf = kmap(local_nb[i].lnb_page);
1848                 while (len != 0) {
1849                         rc = vfs_write(filp, (__force const char __user *)buf,
1850                                        len, &filp->f_pos);
1851                         if (rc < 0) {
1852                                 CERROR("%s: wanted to write %u but got %d "
1853                                        "error\n", dbgcksum_file_name, len, rc);
1854                                 break;
1855                         }
1856                         len -= rc;
1857                         buf += rc;
1858                         CDEBUG(D_INFO, "%s: wrote %d bytes\n",
1859                                dbgcksum_file_name, rc);
1860                 }
1861                 kunmap(local_nb[i].lnb_page);
1862         }
1863         set_fs(oldfs);
1864
1865         rc = ll_vfs_fsync_range(filp, 0, LLONG_MAX, 1);
1866         if (rc)
1867                 CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
1868         filp_close(filp, NULL);
1869         return;
1870 }
1871
1872 static int check_read_checksum(struct niobuf_local *local_nb, int npages,
1873                                struct obd_export *exp, struct obdo *oa,
1874                                const struct lnet_process_id *peer,
1875                                __u32 client_cksum, __u32 server_cksum,
1876                                enum cksum_types server_cksum_type)
1877 {
1878         char *msg;
1879         enum cksum_types cksum_type;
1880         loff_t start, end;
1881
1882         /* unlikely to happen and only if resend does not occur due to cksum
1883          * control failure on Client */
1884         if (unlikely(server_cksum == client_cksum)) {
1885                 CDEBUG(D_PAGE, "checksum %x confirmed upon retry\n",
1886                        client_cksum);
1887                 return 0;
1888         }
1889
1890         if (exp->exp_obd->obd_checksum_dump)
1891                 dump_all_bulk_pages(oa, npages, local_nb, server_cksum,
1892                                     client_cksum);
1893
1894         cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1895                                        oa->o_flags : 0);
1896
1897         if (cksum_type != server_cksum_type)
1898                 msg = "the server may have not used the checksum type specified"
1899                       " in the original request - likely a protocol problem";
1900         else
1901                 msg = "should have changed on the client or in transit";
1902
1903         start = local_nb[0].lnb_file_offset;
1904         end = local_nb[npages-1].lnb_file_offset +
1905                                         local_nb[npages-1].lnb_len - 1;
1906
1907         LCONSOLE_ERROR_MSG(0x132, "%s: BAD READ CHECKSUM: %s: from %s inode "
1908                 DFID " object "DOSTID" extent [%llu-%llu], client returned csum"
1909                 " %x (type %x), server csum %x (type %x)\n",
1910                 exp->exp_obd->obd_name,
1911                 msg, libcfs_nid2str(peer->nid),
1912                 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : 0ULL,
1913                 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1914                 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1915                 POSTID(&oa->o_oi),
1916                 start, end, client_cksum, cksum_type, server_cksum,
1917                 server_cksum_type);
1918
1919         return 1;
1920 }
1921
1922 static int tgt_pages2shortio(struct niobuf_local *local, int npages,
1923                              unsigned char *buf, int size)
1924 {
1925         int     i, off, len, copied = size;
1926         char    *ptr;
1927
1928         for (i = 0; i < npages; i++) {
1929                 off = local[i].lnb_page_offset & ~PAGE_MASK;
1930                 len = local[i].lnb_len;
1931
1932                 CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n",
1933                        i, off, len, size);
1934                 if (len > size)
1935                         return -EINVAL;
1936
1937                 ptr = ll_kmap_atomic(local[i].lnb_page, KM_USER0);
1938                 memcpy(buf + off, ptr, len);
1939                 ll_kunmap_atomic(ptr, KM_USER0);
1940                 buf += len;
1941                 size -= len;
1942         }
1943         return copied - size;
1944 }
1945
1946 int tgt_brw_read(struct tgt_session_info *tsi)
1947 {
1948         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1949         struct ptlrpc_bulk_desc *desc = NULL;
1950         struct obd_export       *exp = tsi->tsi_exp;
1951         struct niobuf_remote    *remote_nb;
1952         struct niobuf_local     *local_nb;
1953         struct obd_ioobj        *ioo;
1954         struct ost_body         *body, *repbody;
1955         struct l_wait_info       lwi;
1956         struct lustre_handle     lockh = { 0 };
1957         int                      npages, nob = 0, rc, i, no_reply = 0,
1958                                  npages_read;
1959         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1960
1961         ENTRY;
1962
1963         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL &&
1964             ptlrpc_req2svc(req)->srv_req_portal != MDS_IO_PORTAL) {
1965                 CERROR("%s: deny read request from %s to portal %u\n",
1966                        tgt_name(tsi->tsi_tgt),
1967                        obd_export_nid2str(req->rq_export),
1968                        ptlrpc_req2svc(req)->srv_req_portal);
1969                 RETURN(-EPROTO);
1970         }
1971
1972         req->rq_bulk_read = 1;
1973
1974         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1975                 RETURN(-EIO);
1976
1977         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1978                          cfs_fail_val : (obd_timeout + 1) / 4);
1979
1980         /* Check if there is eviction in progress, and if so, wait for it to
1981          * finish */
1982         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1983                 /* We do not care how long it takes */
1984                 lwi = LWI_INTR(NULL, NULL);
1985                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1986                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1987                          &lwi);
1988         }
1989
1990         /* There must be big cache in current thread to process this request
1991          * if it is NULL then something went wrong and it wasn't allocated,
1992          * report -ENOMEM in that case */
1993         if (tbc == NULL)
1994                 RETURN(-ENOMEM);
1995
1996         body = tsi->tsi_ost_body;
1997         LASSERT(body != NULL);
1998
1999         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2000         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2001
2002         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2003         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2004
2005         local_nb = tbc->local;
2006
2007         rc = tgt_brw_lock(exp, &tsi->tsi_resid, ioo, remote_nb, &lockh,
2008                           LCK_PR);
2009         if (rc != 0)
2010                 RETURN(rc);
2011
2012         /*
2013          * If getting the lock took more time than
2014          * client was willing to wait, drop it. b=11330
2015          */
2016         if (ktime_get_real_seconds() > req->rq_deadline ||
2017             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2018                 no_reply = 1;
2019                 CERROR("Dropping timed-out read from %s because locking object " DOSTID " took %lld seconds (limit was %lld).\n",
2020                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
2021                        ktime_get_real_seconds() - req->rq_arrival_time.tv_sec,
2022                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2023                 GOTO(out_lock, rc = -ETIMEDOUT);
2024         }
2025
2026         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2027         repbody->oa = body->oa;
2028
2029         npages = PTLRPC_MAX_BRW_PAGES;
2030         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
2031                         ioo, remote_nb, &npages, local_nb);
2032         if (rc != 0)
2033                 GOTO(out_lock, rc);
2034
2035         if (body->oa.o_flags & OBD_FL_SHORT_IO) {
2036                 desc = NULL;
2037         } else {
2038                 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2039                                             PTLRPC_BULK_PUT_SOURCE |
2040                                                 PTLRPC_BULK_BUF_KIOV,
2041                                             OST_BULK_PORTAL,
2042                                             &ptlrpc_bulk_kiov_nopin_ops);
2043                 if (desc == NULL)
2044                         GOTO(out_commitrw, rc = -ENOMEM);
2045         }
2046
2047         nob = 0;
2048         npages_read = npages;
2049         for (i = 0; i < npages; i++) {
2050                 int page_rc = local_nb[i].lnb_rc;
2051
2052                 if (page_rc < 0) {
2053                         rc = page_rc;
2054                         npages_read = i;
2055                         break;
2056                 }
2057
2058                 nob += page_rc;
2059                 if (page_rc != 0 && desc != NULL) { /* some data! */
2060                         LASSERT(local_nb[i].lnb_page != NULL);
2061                         desc->bd_frag_ops->add_kiov_frag
2062                           (desc, local_nb[i].lnb_page,
2063                            local_nb[i].lnb_page_offset & ~PAGE_MASK,
2064                            page_rc);
2065                 }
2066
2067                 if (page_rc != local_nb[i].lnb_len) { /* short read */
2068                         local_nb[i].lnb_len = page_rc;
2069                         npages_read = i + (page_rc != 0 ? 1 : 0);
2070                         /* All subsequent pages should be 0 */
2071                         while (++i < npages)
2072                                 LASSERT(local_nb[i].lnb_rc == 0);
2073                         break;
2074                 }
2075         }
2076         if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
2077             nob != cfs_fail_val)
2078                 rc = -E2BIG;
2079
2080         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
2081                 enum cksum_types cksum_type =
2082                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
2083                                           body->oa.o_flags : 0);
2084
2085                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
2086                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2087                 repbody->oa.o_cksum = tgt_checksum_niobuf(tsi->tsi_tgt,
2088                                                          local_nb, npages_read,
2089                                                          OST_READ, cksum_type);
2090                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
2091                        repbody->oa.o_cksum);
2092
2093                 /* if a resend it could be for a cksum error, so check Server
2094                  * cksum with returned Client cksum (this should even cover
2095                  * zero-cksum case) */
2096                 if ((body->oa.o_valid & OBD_MD_FLFLAGS) &&
2097                     (body->oa.o_flags & OBD_FL_RECOV_RESEND))
2098                         check_read_checksum(local_nb, npages_read, exp,
2099                                             &body->oa, &req->rq_peer,
2100                                             body->oa.o_cksum,
2101                                             repbody->oa.o_cksum, cksum_type);
2102         } else {
2103                 repbody->oa.o_valid = 0;
2104         }
2105         /* We're finishing using body->oa as an input variable */
2106
2107         /* Check if client was evicted while we were doing i/o before touching
2108          * network */
2109         if (rc == 0) {
2110                 if (body->oa.o_flags & OBD_FL_SHORT_IO) {
2111                         unsigned char *short_io_buf;
2112                         int short_io_size;
2113
2114                         short_io_buf = req_capsule_server_get(&req->rq_pill,
2115                                                               &RMF_SHORT_IO);
2116                         short_io_size = req_capsule_get_size(&req->rq_pill,
2117                                                              &RMF_SHORT_IO,
2118                                                              RCL_SERVER);
2119                         rc = tgt_pages2shortio(local_nb, npages_read,
2120                                                short_io_buf, short_io_size);
2121                         if (rc >= 0)
2122                                 req_capsule_shrink(&req->rq_pill,
2123                                                    &RMF_SHORT_IO, rc,
2124                                                    RCL_SERVER);
2125                         rc = rc > 0 ? 0 : rc;
2126                 } else if (!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) {
2127                         rc = target_bulk_io(exp, desc, &lwi);
2128                 }
2129                 no_reply = rc != 0;
2130         } else {
2131                 if (body->oa.o_flags & OBD_FL_SHORT_IO)
2132                         req_capsule_shrink(&req->rq_pill, &RMF_SHORT_IO, 0,
2133                                            RCL_SERVER);
2134         }
2135
2136 out_commitrw:
2137         /* Must commit after prep above in all cases */
2138         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
2139                           remote_nb, npages, local_nb, rc);
2140 out_lock:
2141         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
2142
2143         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
2144                 ptlrpc_free_bulk(desc);
2145
2146         LASSERT(rc <= 0);
2147         if (rc == 0) {
2148                 rc = nob;
2149                 ptlrpc_lprocfs_brw(req, nob);
2150         } else if (no_reply) {
2151                 req->rq_no_reply = 1;
2152                 /* reply out callback would free */
2153                 ptlrpc_req_drop_rs(req);
2154                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
2155                               "client will retry: rc %d\n",
2156                               exp->exp_obd->obd_name,
2157                               obd_uuid2str(&exp->exp_client_uuid),
2158                               obd_export_nid2str(exp), rc);
2159         }
2160         /* send a bulk after reply to simulate a network delay or reordering
2161          * by a router - Note that !desc implies short io, so there is no bulk
2162          * to reorder. */
2163         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) &&
2164             desc) {
2165                 wait_queue_head_t        waitq;
2166                 struct l_wait_info       lwi1;
2167
2168                 CDEBUG(D_INFO, "reorder BULK\n");
2169                 init_waitqueue_head(&waitq);
2170
2171                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
2172                 l_wait_event(waitq, 0, &lwi1);
2173                 target_bulk_io(exp, desc, &lwi);
2174                 ptlrpc_free_bulk(desc);
2175         }
2176
2177         RETURN(rc);
2178 }
2179 EXPORT_SYMBOL(tgt_brw_read);
2180
2181 static int tgt_shortio2pages(struct niobuf_local *local, int npages,
2182                              unsigned char *buf, int size)
2183 {
2184         int     i, off, len;
2185         char    *ptr;
2186
2187         for (i = 0; i < npages; i++) {
2188                 off = local[i].lnb_page_offset & ~PAGE_MASK;
2189                 len = local[i].lnb_len;
2190
2191                 if (len == 0)
2192                         continue;
2193
2194                 CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n",
2195                        i, off, len, size);
2196                 ptr = ll_kmap_atomic(local[i].lnb_page, KM_USER0);
2197                 if (ptr == NULL)
2198                         return -EINVAL;
2199                 memcpy(ptr + off, buf, len < size ? len : size);
2200                 ll_kunmap_atomic(ptr, KM_USER0);
2201                 buf += len;
2202                 size -= len;
2203         }
2204         return 0;
2205 }
2206
2207 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2208                               struct ptlrpc_bulk_desc *desc,
2209                               struct niobuf_local *local_nb, int npages,
2210                               u32 client_cksum, u32 server_cksum,
2211                               bool mmap)
2212 {
2213         struct obd_export *exp = req->rq_export;
2214         struct ost_body *body;
2215         char *router = "";
2216         char *via = "";
2217
2218         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2219         LASSERT(body != NULL);
2220
2221         if (desc && req->rq_peer.nid != desc->bd_sender) {
2222                 via = " via ";
2223                 router = libcfs_nid2str(desc->bd_sender);
2224         }
2225
2226         if (exp->exp_obd->obd_checksum_dump)
2227                 dump_all_bulk_pages(&body->oa, npages, local_nb, server_cksum,
2228                                     client_cksum);
2229
2230         if (mmap) {
2231                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2232                              client_cksum, server_cksum);
2233                 return;
2234         }
2235
2236         LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: from %s%s%s inode "
2237                            DFID" object "DOSTID" extent [%llu-%llu"
2238                            "]: client csum %x, server csum %x\n",
2239                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2240                            via, router,
2241                            body->oa.o_valid & OBD_MD_FLFID ?
2242                            body->oa.o_parent_seq : (__u64)0,
2243                            body->oa.o_valid & OBD_MD_FLFID ?
2244                            body->oa.o_parent_oid : 0,
2245                            body->oa.o_valid & OBD_MD_FLFID ?
2246                            body->oa.o_parent_ver : 0,
2247                            POSTID(&body->oa.o_oi),
2248                            local_nb[0].lnb_file_offset,
2249                            local_nb[npages-1].lnb_file_offset +
2250                            local_nb[npages - 1].lnb_len - 1,
2251                            client_cksum, server_cksum);
2252 }
2253
2254 int tgt_brw_write(struct tgt_session_info *tsi)
2255 {
2256         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2257         struct ptlrpc_bulk_desc *desc = NULL;
2258         struct obd_export       *exp = req->rq_export;
2259         struct niobuf_remote    *remote_nb;
2260         struct niobuf_local     *local_nb;
2261         struct obd_ioobj        *ioo;
2262         struct ost_body         *body, *repbody;
2263         struct l_wait_info       lwi;
2264         struct lustre_handle     lockh = {0};
2265         __u32                   *rcs;
2266         int                      objcount, niocount, npages;
2267         int                      rc, i, j;
2268         enum cksum_types cksum_type = OBD_CKSUM_CRC32;
2269         bool                     no_reply = false, mmap;
2270         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2271         bool wait_sync = false;
2272
2273         ENTRY;
2274
2275         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL &&
2276             ptlrpc_req2svc(req)->srv_req_portal != MDS_IO_PORTAL) {
2277                 CERROR("%s: deny write request from %s to portal %u\n",
2278                        tgt_name(tsi->tsi_tgt),
2279                        obd_export_nid2str(req->rq_export),
2280                        ptlrpc_req2svc(req)->srv_req_portal);
2281                 RETURN(err_serious(-EPROTO));
2282         }
2283
2284         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2285                 RETURN(err_serious(-ENOSPC));
2286         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2287                 RETURN(err_serious(-EROFS));
2288
2289         req->rq_bulk_write = 1;
2290
2291         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2292                 RETURN(err_serious(-EIO));
2293         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2294                 RETURN(err_serious(-EFAULT));
2295
2296         /* pause before transaction has been started */
2297         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2298                          cfs_fail_val : (obd_timeout + 1) / 4);
2299
2300         /* There must be big cache in current thread to process this request
2301          * if it is NULL then something went wrong and it wasn't allocated,
2302          * report -ENOMEM in that case */
2303         if (tbc == NULL)
2304                 RETURN(-ENOMEM);
2305
2306         body = tsi->tsi_ost_body;
2307         LASSERT(body != NULL);
2308
2309         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2310         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2311
2312         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2313                                         RCL_CLIENT) / sizeof(*ioo);
2314
2315         for (niocount = i = 0; i < objcount; i++)
2316                 niocount += ioo[i].ioo_bufcnt;
2317
2318         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2319         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2320         if (niocount != req_capsule_get_size(&req->rq_pill,
2321                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2322                         sizeof(*remote_nb))
2323                 RETURN(err_serious(-EPROTO));
2324
2325         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2326             ptlrpc_connection_is_local(exp->exp_connection))
2327                 memory_pressure_set();
2328
2329         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2330                              niocount * sizeof(*rcs));
2331         rc = req_capsule_server_pack(&req->rq_pill);
2332         if (rc != 0)
2333                 GOTO(out, rc = err_serious(rc));
2334
2335         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2336         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2337
2338         local_nb = tbc->local;
2339
2340         rc = tgt_brw_lock(exp, &tsi->tsi_resid, ioo, remote_nb, &lockh,
2341                           LCK_PW);
2342         if (rc != 0)
2343                 GOTO(out, rc);
2344
2345         /*
2346          * If getting the lock took more time than
2347          * client was willing to wait, drop it. b=11330
2348          */
2349         if (ktime_get_real_seconds() > req->rq_deadline ||
2350             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2351                 no_reply = true;
2352                 CERROR("%s: Dropping timed-out write from %s because locking object " DOSTID " took %lld seconds (limit was %lld).\n",
2353                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2354                        POSTID(&ioo->ioo_oid),
2355                        ktime_get_real_seconds() - req->rq_arrival_time.tv_sec,
2356                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2357                 GOTO(out_lock, rc = -ETIMEDOUT);
2358         }
2359
2360         /* Because we already sync grant info with client when reconnect,
2361          * grant info will be cleared for resent req, then fed_grant and
2362          * total_grant will not be modified in following preprw_write */
2363         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2364                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2365                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2366         }
2367
2368         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2369         if (repbody == NULL)
2370                 GOTO(out_lock, rc = -ENOMEM);
2371         repbody->oa = body->oa;
2372
2373         npages = PTLRPC_MAX_BRW_PAGES;
2374         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2375                         objcount, ioo, remote_nb, &npages, local_nb);
2376         if (rc < 0)
2377                 GOTO(out_lock, rc);
2378         if (body->oa.o_flags & OBD_FL_SHORT_IO) {
2379                 int short_io_size;
2380                 unsigned char *short_io_buf;
2381
2382                 short_io_size = req_capsule_get_size(&req->rq_pill,
2383                                                      &RMF_SHORT_IO,
2384                                                      RCL_CLIENT);
2385                 short_io_buf = req_capsule_client_get(&req->rq_pill,
2386                                                       &RMF_SHORT_IO);
2387                 CDEBUG(D_INFO, "Client use short io for data transfer,"
2388                                " size = %d\n", short_io_size);
2389
2390                 /* Copy short io buf to pages */
2391                 rc = tgt_shortio2pages(local_nb, npages, short_io_buf,
2392                                        short_io_size);
2393                 desc = NULL;
2394         } else {
2395                 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2396                                             PTLRPC_BULK_GET_SINK |
2397                                             PTLRPC_BULK_BUF_KIOV,
2398                                             OST_BULK_PORTAL,
2399                                             &ptlrpc_bulk_kiov_nopin_ops);
2400                 if (desc == NULL)
2401                         GOTO(skip_transfer, rc = -ENOMEM);
2402
2403                 /* NB Having prepped, we must commit... */
2404                 for (i = 0; i < npages; i++)
2405                         desc->bd_frag_ops->add_kiov_frag(desc,
2406                                         local_nb[i].lnb_page,
2407                                         local_nb[i].lnb_page_offset & ~PAGE_MASK,
2408                                         local_nb[i].lnb_len);
2409
2410                 rc = sptlrpc_svc_prep_bulk(req, desc);
2411                 if (rc != 0)
2412                         GOTO(skip_transfer, rc);
2413
2414                 rc = target_bulk_io(exp, desc, &lwi);
2415         }
2416
2417         no_reply = rc != 0;
2418
2419 skip_transfer:
2420         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2421                 static int cksum_counter;
2422
2423                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2424                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2425
2426                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2427                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2428                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2429                 repbody->oa.o_cksum = tgt_checksum_niobuf(tsi->tsi_tgt,
2430                                                           local_nb, npages,
2431                                                           OST_WRITE,
2432                                                           cksum_type);
2433                 cksum_counter++;
2434
2435                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2436                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2437                                 body->oa.o_flags & OBD_FL_MMAP);
2438
2439                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2440                                           body->oa.o_cksum,
2441                                           repbody->oa.o_cksum, mmap);
2442                         cksum_counter = 0;
2443                 } else if ((cksum_counter & (-cksum_counter)) ==
2444                            cksum_counter) {
2445                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2446                                cksum_counter, libcfs_id2str(req->rq_peer),
2447                                repbody->oa.o_cksum);
2448                 }
2449         }
2450
2451         /* Must commit after prep above in all cases */
2452         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2453                           objcount, ioo, remote_nb, npages, local_nb, rc);
2454         if (rc == -ENOTCONN)
2455                 /* quota acquire process has been given up because
2456                  * either the client has been evicted or the client
2457                  * has timed out the request already */
2458                 no_reply = true;
2459
2460         for (i = 0; i < niocount; i++) {
2461                 if (!(local_nb[i].lnb_flags & OBD_BRW_ASYNC)) {
2462                         wait_sync = true;
2463                         break;
2464                 }
2465         }
2466         /*
2467          * Disable sending mtime back to the client. If the client locked the
2468          * whole object, then it has already updated the mtime on its side,
2469          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2470          */
2471         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2472
2473         if (rc == 0) {
2474                 int nob = 0;
2475
2476                 /* set per-requested niobuf return codes */
2477                 for (i = j = 0; i < niocount; i++) {
2478                         int len = remote_nb[i].rnb_len;
2479
2480                         nob += len;
2481                         rcs[i] = 0;
2482                         do {
2483                                 LASSERT(j < npages);
2484                                 if (local_nb[j].lnb_rc < 0)
2485                                         rcs[i] = local_nb[j].lnb_rc;
2486                                 len -= local_nb[j].lnb_len;
2487                                 j++;
2488                         } while (len > 0);
2489                         LASSERT(len == 0);
2490                 }
2491                 LASSERT(j == npages);
2492                 ptlrpc_lprocfs_brw(req, nob);
2493         }
2494 out_lock:
2495         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2496         if (desc)
2497                 ptlrpc_free_bulk(desc);
2498 out:
2499         if (unlikely(no_reply || (exp->exp_obd->obd_no_transno && wait_sync))) {
2500                 req->rq_no_reply = 1;
2501                 /* reply out callback would free */
2502                 ptlrpc_req_drop_rs(req);
2503                 if (!exp->exp_obd->obd_no_transno)
2504                         LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s),"
2505                                       " client will retry: rc = %d\n",
2506                                       exp->exp_obd->obd_name,
2507                                       obd_uuid2str(&exp->exp_client_uuid),
2508                                       obd_export_nid2str(exp), rc);
2509         }
2510         memory_pressure_clr();
2511         RETURN(rc);
2512 }
2513 EXPORT_SYMBOL(tgt_brw_write);
2514
2515 /* Check if request can be reconstructed from saved reply data
2516  * A copy of the reply data is returned in @trd if the pointer is not NULL
2517  */
2518 bool req_can_reconstruct(struct ptlrpc_request *req,
2519                          struct tg_reply_data *trd)
2520 {
2521         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2522         struct lsd_client_data *lcd = ted->ted_lcd;
2523         bool found;
2524
2525         if (tgt_is_multimodrpcs_client(req->rq_export))
2526                 return tgt_lookup_reply(req, trd);
2527
2528         mutex_lock(&ted->ted_lcd_lock);
2529         found = req->rq_xid == lcd->lcd_last_xid ||
2530                 req->rq_xid == lcd->lcd_last_close_xid;
2531
2532         if (found && trd != NULL) {
2533                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2534                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2535                         trd->trd_reply.lrd_transno =
2536                                                 lcd->lcd_last_close_transno;
2537                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2538                 } else {
2539                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2540                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2541                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2542                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2543                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2544                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2545                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2546                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2547                 }
2548         }
2549         mutex_unlock(&ted->ted_lcd_lock);
2550
2551         return found;
2552 }
2553 EXPORT_SYMBOL(req_can_reconstruct);
2554