projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging: lnet: ptllnd_wire.h: checkpatch cleanup: whitespace neatening
[deliverable/linux.git] / drivers / staging / lustre / lustre / osc / osc_request.c
CommitLineData
d7e09d03
PT		 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, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37#define DEBUG_SUBSYSTEM S_OSC
38
39#include <linux/libcfs/libcfs.h>
40
41
42#include <lustre_dlm.h>
43#include <lustre_net.h>
44#include <lustre/lustre_user.h>
45#include <obd_cksum.h>
46#include <obd_ost.h>
d7e09d03 47
d7e09d03
PT		 48#include <lustre_ha.h>
49#include <lprocfs_status.h>
50#include <lustre_log.h>
51#include <lustre_debug.h>
52#include <lustre_param.h>
53#include <lustre_fid.h>
54#include "osc_internal.h"
55#include "osc_cl_internal.h"
56
57static void osc_release_ppga(struct brw_page **ppga, obd_count count);
58static int brw_interpret(const struct lu_env *env,
59 struct ptlrpc_request *req, void *data, int rc);
60int osc_cleanup(struct obd_device *obd);
61
62/* Pack OSC object metadata for disk storage (LE byte order). */
63static int osc_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
64 struct lov_stripe_md *lsm)
65{
66 int lmm_size;
d7e09d03
PT		 67
68 lmm_size = sizeof(**lmmp);
69 if (lmmp == NULL)
0a3bdb00 70 return lmm_size;
d7e09d03
PT		 71
72 if (*lmmp != NULL && lsm == NULL) {
73 OBD_FREE(*lmmp, lmm_size);
74 *lmmp = NULL;
0a3bdb00 75 return 0;
d7e09d03 76 } else if (unlikely(lsm != NULL && ostid_id(&lsm->lsm_oi) == 0)) {
0a3bdb00 77 return -EBADF;
d7e09d03
PT		 78 }
79
80 if (*lmmp == NULL) {
81 OBD_ALLOC(*lmmp, lmm_size);
82 if (*lmmp == NULL)
0a3bdb00 83 return -ENOMEM;
d7e09d03
PT		 84 }
85
86 if (lsm)
87 ostid_cpu_to_le(&lsm->lsm_oi, &(*lmmp)->lmm_oi);
88
0a3bdb00 89 return lmm_size;
d7e09d03
PT		 90}
91
92/* Unpack OSC object metadata from disk storage (LE byte order). */
93static int osc_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
94 struct lov_mds_md *lmm, int lmm_bytes)
95{
96 int lsm_size;
97 struct obd_import *imp = class_exp2cliimp(exp);
d7e09d03
PT		 98
99 if (lmm != NULL) {
100 if (lmm_bytes < sizeof(*lmm)) {
101 CERROR("%s: lov_mds_md too small: %d, need %d\n",
102 exp->exp_obd->obd_name, lmm_bytes,
103 (int)sizeof(*lmm));
0a3bdb00 104 return -EINVAL;
d7e09d03
PT		 105 }
106 /* XXX LOV_MAGIC etc check? */
107
108 if (unlikely(ostid_id(&lmm->lmm_oi) == 0)) {
109 CERROR("%s: zero lmm_object_id: rc = %d\n",
110 exp->exp_obd->obd_name, -EINVAL);
0a3bdb00 111 return -EINVAL;
d7e09d03
PT		 112 }
113 }
114
115 lsm_size = lov_stripe_md_size(1);
116 if (lsmp == NULL)
0a3bdb00 117 return lsm_size;
d7e09d03
PT		 118
119 if (*lsmp != NULL && lmm == NULL) {
120 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
121 OBD_FREE(*lsmp, lsm_size);
122 *lsmp = NULL;
0a3bdb00 123 return 0;
d7e09d03
PT		 124 }
125
126 if (*lsmp == NULL) {
127 OBD_ALLOC(*lsmp, lsm_size);
128 if (unlikely(*lsmp == NULL))
0a3bdb00 129 return -ENOMEM;
d7e09d03
PT		 130 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
131 if (unlikely((*lsmp)->lsm_oinfo[0] == NULL)) {
132 OBD_FREE(*lsmp, lsm_size);
0a3bdb00 133 return -ENOMEM;
d7e09d03
PT		 134 }
135 loi_init((*lsmp)->lsm_oinfo[0]);
136 } else if (unlikely(ostid_id(&(*lsmp)->lsm_oi) == 0)) {
0a3bdb00 137 return -EBADF;
d7e09d03
PT		 138 }
139
140 if (lmm != NULL)
141 /* XXX zero *lsmp? */
142 ostid_le_to_cpu(&lmm->lmm_oi, &(*lsmp)->lsm_oi);
143
144 if (imp != NULL &&
145 (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES))
146 (*lsmp)->lsm_maxbytes = imp->imp_connect_data.ocd_maxbytes;
147 else
148 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
149
0a3bdb00 150 return lsm_size;
d7e09d03
PT		 151}
152
153static inline void osc_pack_capa(struct ptlrpc_request *req,
154 struct ost_body *body, void *capa)
155{
156 struct obd_capa *oc = (struct obd_capa *)capa;
157 struct lustre_capa *c;
158
159 if (!capa)
160 return;
161
162 c = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
163 LASSERT(c);
164 capa_cpy(c, oc);
165 body->oa.o_valid |= OBD_MD_FLOSSCAPA;
166 DEBUG_CAPA(D_SEC, c, "pack");
167}
168
169static inline void osc_pack_req_body(struct ptlrpc_request *req,
170 struct obd_info *oinfo)
171{
172 struct ost_body *body;
173
174 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
175 LASSERT(body);
176
3b2f75fd 177 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
178 oinfo->oi_oa);
d7e09d03
PT		 179 osc_pack_capa(req, body, oinfo->oi_capa);
180}
181
182static inline void osc_set_capa_size(struct ptlrpc_request *req,
183 const struct req_msg_field *field,
184 struct obd_capa *oc)
185{
186 if (oc == NULL)
187 req_capsule_set_size(&req->rq_pill, field, RCL_CLIENT, 0);
188 else
189 /* it is already calculated as sizeof struct obd_capa */
190 ;
191}
192
193static int osc_getattr_interpret(const struct lu_env *env,
194 struct ptlrpc_request *req,
195 struct osc_async_args *aa, int rc)
196{
197 struct ost_body *body;
d7e09d03
PT		 198
199 if (rc != 0)
200 GOTO(out, rc);
201
202 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
203 if (body) {
204 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
3b2f75fd 205 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
206 aa->aa_oi->oi_oa, &body->oa);
d7e09d03
PT		 207
208 /* This should really be sent by the OST */
209 aa->aa_oi->oi_oa->o_blksize = DT_MAX_BRW_SIZE;
210 aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
211 } else {
212 CDEBUG(D_INFO, "can't unpack ost_body\n");
213 rc = -EPROTO;
214 aa->aa_oi->oi_oa->o_valid = 0;
215 }
216out:
217 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
0a3bdb00 218 return rc;
d7e09d03
PT		 219}
220
221static int osc_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
222 struct ptlrpc_request_set *set)
223{
224 struct ptlrpc_request *req;
225 struct osc_async_args *aa;
226 int rc;
d7e09d03
PT		 227
228 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
229 if (req == NULL)
0a3bdb00 230 return -ENOMEM;
d7e09d03
PT		 231
232 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
233 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
234 if (rc) {
235 ptlrpc_request_free(req);
0a3bdb00 236 return rc;
d7e09d03
PT		 237 }
238
239 osc_pack_req_body(req, oinfo);
240
241 ptlrpc_request_set_replen(req);
242 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_getattr_interpret;
243
244 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
245 aa = ptlrpc_req_async_args(req);
246 aa->aa_oi = oinfo;
247
248 ptlrpc_set_add_req(set, req);
0a3bdb00 249 return 0;
d7e09d03
PT		 250}
251
252static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
253 struct obd_info *oinfo)
254{
255 struct ptlrpc_request *req;
256 struct ost_body *body;
257 int rc;
d7e09d03
PT		 258
259 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
260 if (req == NULL)
0a3bdb00 261 return -ENOMEM;
d7e09d03
PT		 262
263 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
264 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
265 if (rc) {
266 ptlrpc_request_free(req);
0a3bdb00 267 return rc;
d7e09d03
PT		 268 }
269
270 osc_pack_req_body(req, oinfo);
271
272 ptlrpc_request_set_replen(req);
273
274 rc = ptlrpc_queue_wait(req);
275 if (rc)
276 GOTO(out, rc);
277
278 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
279 if (body == NULL)
280 GOTO(out, rc = -EPROTO);
281
282 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
3b2f75fd 283 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
284 &body->oa);
d7e09d03
PT		 285
286 oinfo->oi_oa->o_blksize = cli_brw_size(exp->exp_obd);
287 oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
288
d7e09d03
PT		 289 out:
290 ptlrpc_req_finished(req);
291 return rc;
292}
293
294static int osc_setattr(const struct lu_env *env, struct obd_export *exp,
295 struct obd_info *oinfo, struct obd_trans_info *oti)
296{
297 struct ptlrpc_request *req;
298 struct ost_body *body;
299 int rc;
d7e09d03
PT		 300
301 LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP);
302
303 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
304 if (req == NULL)
0a3bdb00 305 return -ENOMEM;
d7e09d03
PT		 306
307 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
308 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
309 if (rc) {
310 ptlrpc_request_free(req);
0a3bdb00 311 return rc;
d7e09d03
PT		 312 }
313
314 osc_pack_req_body(req, oinfo);
315
316 ptlrpc_request_set_replen(req);
317
318 rc = ptlrpc_queue_wait(req);
319 if (rc)
320 GOTO(out, rc);
321
322 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
323 if (body == NULL)
324 GOTO(out, rc = -EPROTO);
325
3b2f75fd 326 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
327 &body->oa);
d7e09d03 328
d7e09d03
PT		 329out:
330 ptlrpc_req_finished(req);
0a3bdb00 331 return rc;
d7e09d03
PT		 332}
333
334static int osc_setattr_interpret(const struct lu_env *env,
335 struct ptlrpc_request *req,
336 struct osc_setattr_args *sa, int rc)
337{
338 struct ost_body *body;
d7e09d03
PT		 339
340 if (rc != 0)
341 GOTO(out, rc);
342
343 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
344 if (body == NULL)
345 GOTO(out, rc = -EPROTO);
346
3b2f75fd 347 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
348 &body->oa);
d7e09d03
PT		 349out:
350 rc = sa->sa_upcall(sa->sa_cookie, rc);
0a3bdb00 351 return rc;
d7e09d03
PT		 352}
353
354int osc_setattr_async_base(struct obd_export *exp, struct obd_info *oinfo,
355 struct obd_trans_info *oti,
356 obd_enqueue_update_f upcall, void *cookie,
357 struct ptlrpc_request_set *rqset)
358{
359 struct ptlrpc_request *req;
360 struct osc_setattr_args *sa;
361 int rc;
d7e09d03
PT		 362
363 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
364 if (req == NULL)
0a3bdb00 365 return -ENOMEM;
d7e09d03
PT		 366
367 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
368 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
369 if (rc) {
370 ptlrpc_request_free(req);
0a3bdb00 371 return rc;
d7e09d03
PT		 372 }
373
374 if (oti && oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
375 oinfo->oi_oa->o_lcookie = *oti->oti_logcookies;
376
377 osc_pack_req_body(req, oinfo);
378
379 ptlrpc_request_set_replen(req);
380
381 /* do mds to ost setattr asynchronously */
382 if (!rqset) {
383 /* Do not wait for response. */
384 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
385 } else {
386 req->rq_interpret_reply =
387 (ptlrpc_interpterer_t)osc_setattr_interpret;
388
389 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
390 sa = ptlrpc_req_async_args(req);
391 sa->sa_oa = oinfo->oi_oa;
392 sa->sa_upcall = upcall;
393 sa->sa_cookie = cookie;
394
395 if (rqset == PTLRPCD_SET)
396 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
397 else
398 ptlrpc_set_add_req(rqset, req);
399 }
400
0a3bdb00 401 return 0;
d7e09d03
PT		 402}
403
404static int osc_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
405 struct obd_trans_info *oti,
406 struct ptlrpc_request_set *rqset)
407{
408 return osc_setattr_async_base(exp, oinfo, oti,
409 oinfo->oi_cb_up, oinfo, rqset);
410}
411
412int osc_real_create(struct obd_export *exp, struct obdo *oa,
413 struct lov_stripe_md **ea, struct obd_trans_info *oti)
414{
415 struct ptlrpc_request *req;
416 struct ost_body *body;
417 struct lov_stripe_md *lsm;
418 int rc;
d7e09d03
PT		 419
420 LASSERT(oa);
421 LASSERT(ea);
422
423 lsm = *ea;
424 if (!lsm) {
425 rc = obd_alloc_memmd(exp, &lsm);
426 if (rc < 0)
0a3bdb00 427 return rc;
d7e09d03
PT		 428 }
429
430 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
431 if (req == NULL)
432 GOTO(out, rc = -ENOMEM);
433
434 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
435 if (rc) {
436 ptlrpc_request_free(req);
437 GOTO(out, rc);
438 }
439
440 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
441 LASSERT(body);
3b2f75fd 442
443 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
d7e09d03
PT		 444
445 ptlrpc_request_set_replen(req);
446
447 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
448 oa->o_flags == OBD_FL_DELORPHAN) {
449 DEBUG_REQ(D_HA, req,
450 "delorphan from OST integration");
451 /* Don't resend the delorphan req */
452 req->rq_no_resend = req->rq_no_delay = 1;
453 }
454
455 rc = ptlrpc_queue_wait(req);
456 if (rc)
457 GOTO(out_req, rc);
458
459 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
460 if (body == NULL)
461 GOTO(out_req, rc = -EPROTO);
462
3b2f75fd 463 CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
464 lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
d7e09d03
PT		 465
466 oa->o_blksize = cli_brw_size(exp->exp_obd);
467 oa->o_valid |= OBD_MD_FLBLKSZ;
468
469 /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
470 * have valid lsm_oinfo data structs, so don't go touching that.
471 * This needs to be fixed in a big way.
472 */
473 lsm->lsm_oi = oa->o_oi;
474 *ea = lsm;
475
476 if (oti != NULL) {
477 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
478
479 if (oa->o_valid & OBD_MD_FLCOOKIE) {
480 if (!oti->oti_logcookies)
481 oti_alloc_cookies(oti, 1);
482 *oti->oti_logcookies = oa->o_lcookie;
483 }
484 }
485
486 CDEBUG(D_HA, "transno: "LPD64"\n",
487 lustre_msg_get_transno(req->rq_repmsg));
488out_req:
489 ptlrpc_req_finished(req);
490out:
491 if (rc && !*ea)
492 obd_free_memmd(exp, &lsm);
0a3bdb00 493 return rc;
d7e09d03
PT		 494}
495
496int osc_punch_base(struct obd_export *exp, struct obd_info *oinfo,
497 obd_enqueue_update_f upcall, void *cookie,
498 struct ptlrpc_request_set *rqset)
499{
500 struct ptlrpc_request *req;
501 struct osc_setattr_args *sa;
502 struct ost_body *body;
503 int rc;
d7e09d03
PT		 504
505 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_PUNCH);
506 if (req == NULL)
0a3bdb00 507 return -ENOMEM;
d7e09d03
PT		 508
509 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
510 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
511 if (rc) {
512 ptlrpc_request_free(req);
0a3bdb00 513 return rc;
d7e09d03
PT		 514 }
515 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
516 ptlrpc_at_set_req_timeout(req);
517
518 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
519 LASSERT(body);
3b2f75fd 520 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
521 oinfo->oi_oa);
d7e09d03
PT		 522 osc_pack_capa(req, body, oinfo->oi_capa);
523
524 ptlrpc_request_set_replen(req);
525
526 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
527 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
528 sa = ptlrpc_req_async_args(req);
529 sa->sa_oa = oinfo->oi_oa;
530 sa->sa_upcall = upcall;
531 sa->sa_cookie = cookie;
532 if (rqset == PTLRPCD_SET)
533 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
534 else
535 ptlrpc_set_add_req(rqset, req);
536
0a3bdb00 537 return 0;
d7e09d03
PT		 538}
539
540static int osc_punch(const struct lu_env *env, struct obd_export *exp,
541 struct obd_info *oinfo, struct obd_trans_info *oti,
542 struct ptlrpc_request_set *rqset)
543{
544 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
545 oinfo->oi_oa->o_blocks = oinfo->oi_policy.l_extent.end;
546 oinfo->oi_oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
547 return osc_punch_base(exp, oinfo,
548 oinfo->oi_cb_up, oinfo, rqset);
549}
550
551static int osc_sync_interpret(const struct lu_env *env,
552 struct ptlrpc_request *req,
553 void *arg, int rc)
554{
555 struct osc_fsync_args *fa = arg;
556 struct ost_body *body;
d7e09d03
PT		 557
558 if (rc)
559 GOTO(out, rc);
560
561 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
562 if (body == NULL) {
563 CERROR ("can't unpack ost_body\n");
564 GOTO(out, rc = -EPROTO);
565 }
566
567 *fa->fa_oi->oi_oa = body->oa;
568out:
569 rc = fa->fa_upcall(fa->fa_cookie, rc);
0a3bdb00 570 return rc;
d7e09d03
PT		 571}
572
573int osc_sync_base(struct obd_export *exp, struct obd_info *oinfo,
574 obd_enqueue_update_f upcall, void *cookie,
575 struct ptlrpc_request_set *rqset)
576{
577 struct ptlrpc_request *req;
578 struct ost_body *body;
579 struct osc_fsync_args *fa;
580 int rc;
d7e09d03
PT		 581
582 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
583 if (req == NULL)
0a3bdb00 584 return -ENOMEM;
d7e09d03
PT		 585
586 osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
587 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
588 if (rc) {
589 ptlrpc_request_free(req);
0a3bdb00 590 return rc;
d7e09d03
PT		 591 }
592
593 /* overload the size and blocks fields in the oa with start/end */
594 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
595 LASSERT(body);
3b2f75fd 596 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
597 oinfo->oi_oa);
d7e09d03
PT		 598 osc_pack_capa(req, body, oinfo->oi_capa);
599
600 ptlrpc_request_set_replen(req);
601 req->rq_interpret_reply = osc_sync_interpret;
602
603 CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args));
604 fa = ptlrpc_req_async_args(req);
605 fa->fa_oi = oinfo;
606 fa->fa_upcall = upcall;
607 fa->fa_cookie = cookie;
608
609 if (rqset == PTLRPCD_SET)
610 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
611 else
612 ptlrpc_set_add_req(rqset, req);
613
0a3bdb00 614 return 0;
d7e09d03
PT		 615}
616
617static int osc_sync(const struct lu_env *env, struct obd_export *exp,
618 struct obd_info *oinfo, obd_size start, obd_size end,
619 struct ptlrpc_request_set *set)
620{
d7e09d03
PT		 621 if (!oinfo->oi_oa) {
622 CDEBUG(D_INFO, "oa NULL\n");
0a3bdb00 623 return -EINVAL;
d7e09d03
PT		 624 }
625
626 oinfo->oi_oa->o_size = start;
627 oinfo->oi_oa->o_blocks = end;
628 oinfo->oi_oa->o_valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
629
0a3bdb00 630 return osc_sync_base(exp, oinfo, oinfo->oi_cb_up, oinfo, set);
d7e09d03
PT		 631}
632
633/* Find and cancel locally locks matched by @mode in the resource found by
634 * @objid. Found locks are added into @cancel list. Returns the amount of
635 * locks added to @cancels list. */
636static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
637 struct list_head *cancels,
875332d4 638 ldlm_mode_t mode, __u64 lock_flags)
d7e09d03
PT		 639{
640 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
641 struct ldlm_res_id res_id;
642 struct ldlm_resource *res;
643 int count;
d7e09d03
PT		 644
645 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
646 * export) but disabled through procfs (flag in NS).
647 *
648 * This distinguishes from a case when ELC is not supported originally,
649 * when we still want to cancel locks in advance and just cancel them
650 * locally, without sending any RPC. */
651 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
0a3bdb00 652 return 0;
d7e09d03
PT		 653
654 ostid_build_res_name(&oa->o_oi, &res_id);
655 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
656 if (res == NULL)
0a3bdb00 657 return 0;
d7e09d03
PT		 658
659 LDLM_RESOURCE_ADDREF(res);
660 count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
661 lock_flags, 0, NULL);
662 LDLM_RESOURCE_DELREF(res);
663 ldlm_resource_putref(res);
0a3bdb00 664 return count;
d7e09d03
PT		 665}
666
667static int osc_destroy_interpret(const struct lu_env *env,
668 struct ptlrpc_request *req, void *data,
669 int rc)
670{
671 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
672
673 atomic_dec(&cli->cl_destroy_in_flight);
674 wake_up(&cli->cl_destroy_waitq);
675 return 0;
676}
677
678static int osc_can_send_destroy(struct client_obd *cli)
679{
680 if (atomic_inc_return(&cli->cl_destroy_in_flight) <=
681 cli->cl_max_rpcs_in_flight) {
682 /* The destroy request can be sent */
683 return 1;
684 }
685 if (atomic_dec_return(&cli->cl_destroy_in_flight) <
686 cli->cl_max_rpcs_in_flight) {
687 /*
688 * The counter has been modified between the two atomic
689 * operations.
690 */
691 wake_up(&cli->cl_destroy_waitq);
692 }
693 return 0;
694}
695
696int osc_create(const struct lu_env *env, struct obd_export *exp,
697 struct obdo *oa, struct lov_stripe_md **ea,
698 struct obd_trans_info *oti)
699{
700 int rc = 0;
d7e09d03
PT		 701
702 LASSERT(oa);
703 LASSERT(ea);
704 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
705
706 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
707 oa->o_flags == OBD_FL_RECREATE_OBJS) {
0a3bdb00 708 return osc_real_create(exp, oa, ea, oti);
d7e09d03
PT		 709 }
710
711 if (!fid_seq_is_mdt(ostid_seq(&oa->o_oi)))
0a3bdb00 712 return osc_real_create(exp, oa, ea, oti);
d7e09d03
PT		 713
714 /* we should not get here anymore */
715 LBUG();
716
0a3bdb00 717 return rc;
d7e09d03
PT		 718}
719
720/* Destroy requests can be async always on the client, and we don't even really
721 * care about the return code since the client cannot do anything at all about
722 * a destroy failure.
723 * When the MDS is unlinking a filename, it saves the file objects into a
724 * recovery llog, and these object records are cancelled when the OST reports
725 * they were destroyed and sync'd to disk (i.e. transaction committed).
726 * If the client dies, or the OST is down when the object should be destroyed,
727 * the records are not cancelled, and when the OST reconnects to the MDS next,
728 * it will retrieve the llog unlink logs and then sends the log cancellation
729 * cookies to the MDS after committing destroy transactions. */
730static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
731 struct obdo *oa, struct lov_stripe_md *ea,
732 struct obd_trans_info *oti, struct obd_export *md_export,
733 void *capa)
734{
735 struct client_obd *cli = &exp->exp_obd->u.cli;
736 struct ptlrpc_request *req;
737 struct ost_body *body;
738 LIST_HEAD(cancels);
739 int rc, count;
d7e09d03
PT		 740
741 if (!oa) {
742 CDEBUG(D_INFO, "oa NULL\n");
0a3bdb00 743 return -EINVAL;
d7e09d03
PT		 744 }
745
746 count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
747 LDLM_FL_DISCARD_DATA);
748
749 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
750 if (req == NULL) {
751 ldlm_lock_list_put(&cancels, l_bl_ast, count);
0a3bdb00 752 return -ENOMEM;
d7e09d03
PT		 753 }
754
755 osc_set_capa_size(req, &RMF_CAPA1, (struct obd_capa *)capa);
756 rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
757 0, &cancels, count);
758 if (rc) {
759 ptlrpc_request_free(req);
0a3bdb00 760 return rc;
d7e09d03
PT		 761 }
762
763 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
764 ptlrpc_at_set_req_timeout(req);
765
766 if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
767 oa->o_lcookie = *oti->oti_logcookies;
768 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
769 LASSERT(body);
3b2f75fd 770 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
d7e09d03
PT		 771
772 osc_pack_capa(req, body, (struct obd_capa *)capa);
773 ptlrpc_request_set_replen(req);
774
11d66e89 775 /* If osc_destroy is for destroying the unlink orphan,
d7e09d03
PT		 776 * sent from MDT to OST, which should not be blocked here,
777 * because the process might be triggered by ptlrpcd, and
778 * it is not good to block ptlrpcd thread (b=16006)*/
779 if (!(oa->o_flags & OBD_FL_DELORPHAN)) {
780 req->rq_interpret_reply = osc_destroy_interpret;
781 if (!osc_can_send_destroy(cli)) {
782 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP,
783 NULL);
784
785 /*
786 * Wait until the number of on-going destroy RPCs drops
787 * under max_rpc_in_flight
788 */
789 l_wait_event_exclusive(cli->cl_destroy_waitq,
790 osc_can_send_destroy(cli), &lwi);
791 }
792 }
793
794 /* Do not wait for response */
795 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
0a3bdb00 796 return 0;
d7e09d03
PT		 797}
798
799static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
800 long writing_bytes)
801{
802 obd_flag bits = OBD_MD_FLBLOCKS|OBD_MD_FLGRANT;
803
804 LASSERT(!(oa->o_valid & bits));
805
806 oa->o_valid |= bits;
807 client_obd_list_lock(&cli->cl_loi_list_lock);
808 oa->o_dirty = cli->cl_dirty;
809 if (unlikely(cli->cl_dirty - cli->cl_dirty_transit >
810 cli->cl_dirty_max)) {
811 CERROR("dirty %lu - %lu > dirty_max %lu\n",
812 cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
813 oa->o_undirty = 0;
c52f69c5 814 } else if (unlikely(atomic_read(&obd_dirty_pages) -
d7e09d03
PT		 815 atomic_read(&obd_dirty_transit_pages) >
816 (long)(obd_max_dirty_pages + 1))) {
817 /* The atomic_read() allowing the atomic_inc() are
818 * not covered by a lock thus they may safely race and trip
819 * this CERROR() unless we add in a small fudge factor (+1). */
c52f69c5 820 CERROR("dirty %d - %d > system dirty_max %d\n",
d7e09d03
PT		 821 atomic_read(&obd_dirty_pages),
822 atomic_read(&obd_dirty_transit_pages),
823 obd_max_dirty_pages);
824 oa->o_undirty = 0;
825 } else if (unlikely(cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff)) {
826 CERROR("dirty %lu - dirty_max %lu too big???\n",
827 cli->cl_dirty, cli->cl_dirty_max);
828 oa->o_undirty = 0;
829 } else {
830 long max_in_flight = (cli->cl_max_pages_per_rpc <<
831 PAGE_CACHE_SHIFT)*
832 (cli->cl_max_rpcs_in_flight + 1);
833 oa->o_undirty = max(cli->cl_dirty_max, max_in_flight);
834 }
835 oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
836 oa->o_dropped = cli->cl_lost_grant;
837 cli->cl_lost_grant = 0;
838 client_obd_list_unlock(&cli->cl_loi_list_lock);
839 CDEBUG(D_CACHE,"dirty: "LPU64" undirty: %u dropped %u grant: "LPU64"\n",
840 oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
841
842}
843
844void osc_update_next_shrink(struct client_obd *cli)
845{
846 cli->cl_next_shrink_grant =
847 cfs_time_shift(cli->cl_grant_shrink_interval);
848 CDEBUG(D_CACHE, "next time %ld to shrink grant \n",
849 cli->cl_next_shrink_grant);
850}
851
852static void __osc_update_grant(struct client_obd *cli, obd_size grant)
853{
854 client_obd_list_lock(&cli->cl_loi_list_lock);
855 cli->cl_avail_grant += grant;
856 client_obd_list_unlock(&cli->cl_loi_list_lock);
857}
858
859static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
860{
861 if (body->oa.o_valid & OBD_MD_FLGRANT) {
862 CDEBUG(D_CACHE, "got "LPU64" extra grant\n", body->oa.o_grant);
863 __osc_update_grant(cli, body->oa.o_grant);
864 }
865}
866
867static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
868 obd_count keylen, void *key, obd_count vallen,
869 void *val, struct ptlrpc_request_set *set);
870
871static int osc_shrink_grant_interpret(const struct lu_env *env,
872 struct ptlrpc_request *req,
873 void *aa, int rc)
874{
875 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
876 struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
877 struct ost_body *body;
878
879 if (rc != 0) {
880 __osc_update_grant(cli, oa->o_grant);
881 GOTO(out, rc);
882 }
883
884 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
885 LASSERT(body);
886 osc_update_grant(cli, body);
887out:
888 OBDO_FREE(oa);
889 return rc;
890}
891
892static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
893{
894 client_obd_list_lock(&cli->cl_loi_list_lock);
895 oa->o_grant = cli->cl_avail_grant / 4;
896 cli->cl_avail_grant -= oa->o_grant;
897 client_obd_list_unlock(&cli->cl_loi_list_lock);
898 if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
899 oa->o_valid |= OBD_MD_FLFLAGS;
900 oa->o_flags = 0;
901 }
902 oa->o_flags |= OBD_FL_SHRINK_GRANT;
903 osc_update_next_shrink(cli);
904}
905
906/* Shrink the current grant, either from some large amount to enough for a
907 * full set of in-flight RPCs, or if we have already shrunk to that limit
908 * then to enough for a single RPC. This avoids keeping more grant than
909 * needed, and avoids shrinking the grant piecemeal. */
910static int osc_shrink_grant(struct client_obd *cli)
911{
912 __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
913 (cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT);
914
915 client_obd_list_lock(&cli->cl_loi_list_lock);
916 if (cli->cl_avail_grant <= target_bytes)
917 target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
918 client_obd_list_unlock(&cli->cl_loi_list_lock);
919
920 return osc_shrink_grant_to_target(cli, target_bytes);
921}
922
923int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
924{
925 int rc = 0;
926 struct ost_body *body;
d7e09d03
PT		 927
928 client_obd_list_lock(&cli->cl_loi_list_lock);
929 /* Don't shrink if we are already above or below the desired limit
930 * We don't want to shrink below a single RPC, as that will negatively
931 * impact block allocation and long-term performance. */
932 if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT)
933 target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
934
935 if (target_bytes >= cli->cl_avail_grant) {
936 client_obd_list_unlock(&cli->cl_loi_list_lock);
0a3bdb00 937 return 0;
d7e09d03
PT		 938 }
939 client_obd_list_unlock(&cli->cl_loi_list_lock);
940
941 OBD_ALLOC_PTR(body);
942 if (!body)
0a3bdb00 943 return -ENOMEM;
d7e09d03
PT		 944
945 osc_announce_cached(cli, &body->oa, 0);
946
947 client_obd_list_lock(&cli->cl_loi_list_lock);
948 body->oa.o_grant = cli->cl_avail_grant - target_bytes;
949 cli->cl_avail_grant = target_bytes;
950 client_obd_list_unlock(&cli->cl_loi_list_lock);
951 if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
952 body->oa.o_valid |= OBD_MD_FLFLAGS;
953 body->oa.o_flags = 0;
954 }
955 body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
956 osc_update_next_shrink(cli);
957
958 rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export,
959 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
960 sizeof(*body), body, NULL);
961 if (rc != 0)
962 __osc_update_grant(cli, body->oa.o_grant);
963 OBD_FREE_PTR(body);
0a3bdb00 964 return rc;
d7e09d03
PT		 965}
966
967static int osc_should_shrink_grant(struct client_obd *client)
968{
969 cfs_time_t time = cfs_time_current();
970 cfs_time_t next_shrink = client->cl_next_shrink_grant;
971
972 if ((client->cl_import->imp_connect_data.ocd_connect_flags &
973 OBD_CONNECT_GRANT_SHRINK) == 0)
974 return 0;
975
976 if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
977 /* Get the current RPC size directly, instead of going via:
978 * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
979 * Keep comment here so that it can be found by searching. */
980 int brw_size = client->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
981
982 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
983 client->cl_avail_grant > brw_size)
984 return 1;
985 else
986 osc_update_next_shrink(client);
987 }
988 return 0;
989}
990
991static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
992{
993 struct client_obd *client;
994
995 list_for_each_entry(client, &item->ti_obd_list,
996 cl_grant_shrink_list) {
997 if (osc_should_shrink_grant(client))
998 osc_shrink_grant(client);
999 }
1000 return 0;
1001}
1002
1003static int osc_add_shrink_grant(struct client_obd *client)
1004{
1005 int rc;
1006
1007 rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
1008 TIMEOUT_GRANT,
1009 osc_grant_shrink_grant_cb, NULL,
1010 &client->cl_grant_shrink_list);
1011 if (rc) {
1012 CERROR("add grant client %s error %d\n",
1013 client->cl_import->imp_obd->obd_name, rc);
1014 return rc;
1015 }
1016 CDEBUG(D_CACHE, "add grant client %s \n",
1017 client->cl_import->imp_obd->obd_name);
1018 osc_update_next_shrink(client);
1019 return 0;
1020}
1021
1022static int osc_del_shrink_grant(struct client_obd *client)
1023{
1024 return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
1025 TIMEOUT_GRANT);
1026}
1027
1028static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
1029{
1030 /*
1031 * ocd_grant is the total grant amount we're expect to hold: if we've
1032 * been evicted, it's the new avail_grant amount, cl_dirty will drop
1033 * to 0 as inflight RPCs fail out; otherwise, it's avail_grant + dirty.
1034 *
1035 * race is tolerable here: if we're evicted, but imp_state already
1036 * left EVICTED state, then cl_dirty must be 0 already.
1037 */
1038 client_obd_list_lock(&cli->cl_loi_list_lock);
1039 if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
1040 cli->cl_avail_grant = ocd->ocd_grant;
1041 else
1042 cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
1043
1044 if (cli->cl_avail_grant < 0) {
1045 CWARN("%s: available grant < 0: avail/ocd/dirty %ld/%u/%ld\n",
1046 cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant,
1047 ocd->ocd_grant, cli->cl_dirty);
1048 /* workaround for servers which do not have the patch from
1049 * LU-2679 */
1050 cli->cl_avail_grant = ocd->ocd_grant;
1051 }
1052
1053 /* determine the appropriate chunk size used by osc_extent. */
1054 cli->cl_chunkbits = max_t(int, PAGE_CACHE_SHIFT, ocd->ocd_blocksize);
1055 client_obd_list_unlock(&cli->cl_loi_list_lock);
1056
1057 CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
1058 "chunk bits: %d.\n", cli->cl_import->imp_obd->obd_name,
1059 cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits);
1060
1061 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
1062 list_empty(&cli->cl_grant_shrink_list))
1063 osc_add_shrink_grant(cli);
1064}
1065
1066/* We assume that the reason this OSC got a short read is because it read
1067 * beyond the end of a stripe file; i.e. lustre is reading a sparse file
1068 * via the LOV, and it _knows_ it's reading inside the file, it's just that
1069 * this stripe never got written at or beyond this stripe offset yet. */
1070static void handle_short_read(int nob_read, obd_count page_count,
1071 struct brw_page **pga)
1072{
1073 char *ptr;
1074 int i = 0;
1075
1076 /* skip bytes read OK */
1077 while (nob_read > 0) {
1078 LASSERT (page_count > 0);
1079
1080 if (pga[i]->count > nob_read) {
1081 /* EOF inside this page */
1082 ptr = kmap(pga[i]->pg) +
1083 (pga[i]->off & ~CFS_PAGE_MASK);
1084 memset(ptr + nob_read, 0, pga[i]->count - nob_read);
1085 kunmap(pga[i]->pg);
1086 page_count--;
1087 i++;
1088 break;
1089 }
1090
1091 nob_read -= pga[i]->count;
1092 page_count--;
1093 i++;
1094 }
1095
1096 /* zero remaining pages */
1097 while (page_count-- > 0) {
1098 ptr = kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
1099 memset(ptr, 0, pga[i]->count);
1100 kunmap(pga[i]->pg);
1101 i++;
1102 }
1103}
1104
1105static int check_write_rcs(struct ptlrpc_request *req,
1106 int requested_nob, int niocount,
1107 obd_count page_count, struct brw_page **pga)
1108{
1109 int i;
1110 __u32 *remote_rcs;
1111
1112 remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
1113 sizeof(*remote_rcs) *
1114 niocount);
1115 if (remote_rcs == NULL) {
1116 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1117 return(-EPROTO);
1118 }
1119
1120 /* return error if any niobuf was in error */
1121 for (i = 0; i < niocount; i++) {
1122 if ((int)remote_rcs[i] < 0)
1123 return(remote_rcs[i]);
1124
1125 if (remote_rcs[i] != 0) {
1126 CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1127 i, remote_rcs[i], req);
1128 return(-EPROTO);
1129 }
1130 }
1131
1132 if (req->rq_bulk->bd_nob_transferred != requested_nob) {
1133 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1134 req->rq_bulk->bd_nob_transferred, requested_nob);
1135 return(-EPROTO);
1136 }
1137
1138 return (0);
1139}
1140
1141static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1142{
1143 if (p1->flag != p2->flag) {
1144 unsigned mask = ~(OBD_BRW_FROM_GRANT| OBD_BRW_NOCACHE|
1145 OBD_BRW_SYNC|OBD_BRW_ASYNC|OBD_BRW_NOQUOTA);
1146
1147 /* warn if we try to combine flags that we don't know to be
1148 * safe to combine */
1149 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
1150 CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
1151 "report this at http://bugs.whamcloud.com/\n",
1152 p1->flag, p2->flag);
1153 }
1154 return 0;
1155 }
1156
1157 return (p1->off + p1->count == p2->off);
1158}
1159
1160static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
1161 struct brw_page **pga, int opc,
1162 cksum_type_t cksum_type)
1163{
1164 __u32 cksum;
1165 int i = 0;
1166 struct cfs_crypto_hash_desc *hdesc;
1167 unsigned int bufsize;
1168 int err;
1169 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1170
1171 LASSERT(pg_count > 0);
1172
1173 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1174 if (IS_ERR(hdesc)) {
1175 CERROR("Unable to initialize checksum hash %s\n",
1176 cfs_crypto_hash_name(cfs_alg));
1177 return PTR_ERR(hdesc);
1178 }
1179
1180 while (nob > 0 && pg_count > 0) {
1181 int count = pga[i]->count > nob ? nob : pga[i]->count;
1182
1183 /* corrupt the data before we compute the checksum, to
1184 * simulate an OST->client data error */
1185 if (i == 0 && opc == OST_READ &&
1186 OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
1187 unsigned char *ptr = kmap(pga[i]->pg);
1188 int off = pga[i]->off & ~CFS_PAGE_MASK;
1189 memcpy(ptr + off, "bad1", min(4, nob));
1190 kunmap(pga[i]->pg);
1191 }
1192 cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
1193 pga[i]->off & ~CFS_PAGE_MASK,
1194 count);
aa3bee0d
GKH		 1195 CDEBUG(D_PAGE,
1196 "page %p map %p index %lu flags %lx count %u priv %0lx: off %d\n",
1197 pga[i]->pg, pga[i]->pg->mapping, pga[i]->pg->index,
1198 (long)pga[i]->pg->flags, page_count(pga[i]->pg),
1199 page_private(pga[i]->pg),
1200 (int)(pga[i]->off & ~CFS_PAGE_MASK));
d7e09d03
PT		 1201
1202 nob -= pga[i]->count;
1203 pg_count--;
1204 i++;
1205 }
1206
1207 bufsize = 4;
1208 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1209
1210 if (err)
1211 cfs_crypto_hash_final(hdesc, NULL, NULL);
1212
1213 /* For sending we only compute the wrong checksum instead
1214 * of corrupting the data so it is still correct on a redo */
1215 if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1216 cksum++;
1217
1218 return cksum;
1219}
1220
1221static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
1222 struct lov_stripe_md *lsm, obd_count page_count,
1223 struct brw_page **pga,
1224 struct ptlrpc_request **reqp,
1225 struct obd_capa *ocapa, int reserve,
1226 int resend)
1227{
1228 struct ptlrpc_request *req;
1229 struct ptlrpc_bulk_desc *desc;
1230 struct ost_body *body;
1231 struct obd_ioobj *ioobj;
1232 struct niobuf_remote *niobuf;
1233 int niocount, i, requested_nob, opc, rc;
1234 struct osc_brw_async_args *aa;
1235 struct req_capsule *pill;
1236 struct brw_page *pg_prev;
1237
d7e09d03 1238 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
0a3bdb00 1239 return -ENOMEM; /* Recoverable */
d7e09d03 1240 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
0a3bdb00 1241 return -EINVAL; /* Fatal */
d7e09d03
PT		 1242
1243 if ((cmd & OBD_BRW_WRITE) != 0) {
1244 opc = OST_WRITE;
1245 req = ptlrpc_request_alloc_pool(cli->cl_import,
1246 cli->cl_import->imp_rq_pool,
1247 &RQF_OST_BRW_WRITE);
1248 } else {
1249 opc = OST_READ;
1250 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1251 }
1252 if (req == NULL)
0a3bdb00 1253 return -ENOMEM;
d7e09d03
PT		 1254
1255 for (niocount = i = 1; i < page_count; i++) {
1256 if (!can_merge_pages(pga[i - 1], pga[i]))
1257 niocount++;
1258 }
1259
1260 pill = &req->rq_pill;
1261 req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1262 sizeof(*ioobj));
1263 req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1264 niocount * sizeof(*niobuf));
1265 osc_set_capa_size(req, &RMF_CAPA1, ocapa);
1266
1267 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1268 if (rc) {
1269 ptlrpc_request_free(req);
0a3bdb00 1270 return rc;
d7e09d03
PT		 1271 }
1272 req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
1273 ptlrpc_at_set_req_timeout(req);
1274 /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own
1275 * retry logic */
1276 req->rq_no_retry_einprogress = 1;
1277
1278 desc = ptlrpc_prep_bulk_imp(req, page_count,
1279 cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
1280 opc == OST_WRITE ? BULK_GET_SOURCE : BULK_PUT_SINK,
1281 OST_BULK_PORTAL);
1282
1283 if (desc == NULL)
1284 GOTO(out, rc = -ENOMEM);
1285 /* NB request now owns desc and will free it when it gets freed */
1286
1287 body = req_capsule_client_get(pill, &RMF_OST_BODY);
1288 ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1289 niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1290 LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1291
3b2f75fd 1292 lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
d7e09d03
PT		 1293
1294 obdo_to_ioobj(oa, ioobj);
1295 ioobj->ioo_bufcnt = niocount;
1296 /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
1297 * that might be send for this request. The actual number is decided
1298 * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
1299 * "max - 1" for old client compatibility sending "0", and also so the
1300 * the actual maximum is a power-of-two number, not one less. LU-1431 */
1301 ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
1302 osc_pack_capa(req, body, ocapa);
1303 LASSERT(page_count > 0);
1304 pg_prev = pga[0];
1305 for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1306 struct brw_page *pg = pga[i];
1307 int poff = pg->off & ~CFS_PAGE_MASK;
1308
1309 LASSERT(pg->count > 0);
1310 /* make sure there is no gap in the middle of page array */
1311 LASSERTF(page_count == 1 ||
1312 (ergo(i == 0, poff + pg->count == PAGE_CACHE_SIZE) &&
1313 ergo(i > 0 && i < page_count - 1,
1314 poff == 0 && pg->count == PAGE_CACHE_SIZE) &&
1315 ergo(i == page_count - 1, poff == 0)),
1316 "i: %d/%d pg: %p off: "LPU64", count: %u\n",
1317 i, page_count, pg, pg->off, pg->count);
1318 LASSERTF(i == 0 || pg->off > pg_prev->off,
1319 "i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
1320 " prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
1321 i, page_count,
1322 pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1323 pg_prev->pg, page_private(pg_prev->pg),
1324 pg_prev->pg->index, pg_prev->off);
1325 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1326 (pg->flag & OBD_BRW_SRVLOCK));
1327
1328 ptlrpc_prep_bulk_page_pin(desc, pg->pg, poff, pg->count);
1329 requested_nob += pg->count;
1330
1331 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1332 niobuf--;
1333 niobuf->len += pg->count;
1334 } else {
1335 niobuf->offset = pg->off;
1336 niobuf->len = pg->count;
1337 niobuf->flags = pg->flag;
1338 }
1339 pg_prev = pg;
1340 }
1341
1342 LASSERTF((void *)(niobuf - niocount) ==
1343 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1344 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1345 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1346
1347 osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1348 if (resend) {
1349 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1350 body->oa.o_valid |= OBD_MD_FLFLAGS;
1351 body->oa.o_flags = 0;
1352 }
1353 body->oa.o_flags |= OBD_FL_RECOV_RESEND;
1354 }
1355
1356 if (osc_should_shrink_grant(cli))
1357 osc_shrink_grant_local(cli, &body->oa);
1358
1359 /* size[REQ_REC_OFF] still sizeof (*body) */
1360 if (opc == OST_WRITE) {
1361 if (cli->cl_checksum &&
1362 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1363 /* store cl_cksum_type in a local variable since
1364 * it can be changed via lprocfs */
1365 cksum_type_t cksum_type = cli->cl_cksum_type;
1366
1367 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1368 oa->o_flags &= OBD_FL_LOCAL_MASK;
1369 body->oa.o_flags = 0;
1370 }
1371 body->oa.o_flags |= cksum_type_pack(cksum_type);
1372 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1373 body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1374 page_count, pga,
1375 OST_WRITE,
1376 cksum_type);
1377 CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1378 body->oa.o_cksum);
1379 /* save this in 'oa', too, for later checking */
1380 oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1381 oa->o_flags |= cksum_type_pack(cksum_type);
1382 } else {
1383 /* clear out the checksum flag, in case this is a
1384 * resend but cl_checksum is no longer set. b=11238 */
1385 oa->o_valid &= ~OBD_MD_FLCKSUM;
1386 }
1387 oa->o_cksum = body->oa.o_cksum;
1388 /* 1 RC per niobuf */
1389 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1390 sizeof(__u32) * niocount);
1391 } else {
1392 if (cli->cl_checksum &&
1393 !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1394 if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1395 body->oa.o_flags = 0;
1396 body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1397 body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1398 }
1399 }
1400 ptlrpc_request_set_replen(req);
1401
1402 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1403 aa = ptlrpc_req_async_args(req);
1404 aa->aa_oa = oa;
1405 aa->aa_requested_nob = requested_nob;
1406 aa->aa_nio_count = niocount;
1407 aa->aa_page_count = page_count;
1408 aa->aa_resends = 0;
1409 aa->aa_ppga = pga;
1410 aa->aa_cli = cli;
1411 INIT_LIST_HEAD(&aa->aa_oaps);
1412 if (ocapa && reserve)
1413 aa->aa_ocapa = capa_get(ocapa);
1414
1415 *reqp = req;
0a3bdb00 1416 return 0;
d7e09d03
PT		 1417
1418 out:
1419 ptlrpc_req_finished(req);
0a3bdb00 1420 return rc;
d7e09d03
PT		 1421}
1422
1423static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
1424 __u32 client_cksum, __u32 server_cksum, int nob,
1425 obd_count page_count, struct brw_page **pga,
1426 cksum_type_t client_cksum_type)
1427{
1428 __u32 new_cksum;
1429 char *msg;
1430 cksum_type_t cksum_type;
1431
1432 if (server_cksum == client_cksum) {
1433 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1434 return 0;
1435 }
1436
1437 cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1438 oa->o_flags : 0);
1439 new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE,
1440 cksum_type);
1441
1442 if (cksum_type != client_cksum_type)
1443 msg = "the server did not use the checksum type specified in "
1444 "the original request - likely a protocol problem";
1445 else if (new_cksum == server_cksum)
1446 msg = "changed on the client after we checksummed it - "
1447 "likely false positive due to mmap IO (bug 11742)";
1448 else if (new_cksum == client_cksum)
1449 msg = "changed in transit before arrival at OST";
1450 else
1451 msg = "changed in transit AND doesn't match the original - "
1452 "likely false positive due to mmap IO (bug 11742)";
1453
1454 LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
1455 " object "DOSTID" extent ["LPU64"-"LPU64"]\n",
1456 msg, libcfs_nid2str(peer->nid),
1457 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1458 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1459 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1460 POSTID(&oa->o_oi), pga[0]->off,
1461 pga[page_count-1]->off + pga[page_count-1]->count - 1);
1462 CERROR("original client csum %x (type %x), server csum %x (type %x), "
1463 "client csum now %x\n", client_cksum, client_cksum_type,
1464 server_cksum, cksum_type, new_cksum);
1465 return 1;
1466}
1467
1468/* Note rc enters this function as number of bytes transferred */
1469static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1470{
1471 struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1472 const lnet_process_id_t *peer =
1473 &req->rq_import->imp_connection->c_peer;
1474 struct client_obd *cli = aa->aa_cli;
1475 struct ost_body *body;
1476 __u32 client_cksum = 0;
d7e09d03
PT		 1477
1478 if (rc < 0 && rc != -EDQUOT) {
1479 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
0a3bdb00 1480 return rc;
d7e09d03
PT		 1481 }
1482
1483 LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1484 body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1485 if (body == NULL) {
1486 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
0a3bdb00 1487 return -EPROTO;
d7e09d03
PT		 1488 }
1489
1490 /* set/clear over quota flag for a uid/gid */
1491 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1492 body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) {
1493 unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid };
1494
1495 CDEBUG(D_QUOTA, "setdq for [%u %u] with valid "LPX64", flags %x\n",
1496 body->oa.o_uid, body->oa.o_gid, body->oa.o_valid,
1497 body->oa.o_flags);
1498 osc_quota_setdq(cli, qid, body->oa.o_valid, body->oa.o_flags);
1499 }
1500
1501 osc_update_grant(cli, body);
1502
1503 if (rc < 0)
0a3bdb00 1504 return rc;
d7e09d03
PT		 1505
1506 if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1507 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1508
1509 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1510 if (rc > 0) {
1511 CERROR("Unexpected +ve rc %d\n", rc);
0a3bdb00 1512 return -EPROTO;
d7e09d03
PT		 1513 }
1514 LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob);
1515
1516 if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
0a3bdb00 1517 return -EAGAIN;
d7e09d03
PT		 1518
1519 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1520 check_write_checksum(&body->oa, peer, client_cksum,
1521 body->oa.o_cksum, aa->aa_requested_nob,
1522 aa->aa_page_count, aa->aa_ppga,
1523 cksum_type_unpack(aa->aa_oa->o_flags)))
0a3bdb00 1524 return -EAGAIN;
d7e09d03
PT		 1525
1526 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1527 aa->aa_page_count, aa->aa_ppga);
1528 GOTO(out, rc);
1529 }
1530
1531 /* The rest of this function executes only for OST_READs */
1532
1533 /* if unwrap_bulk failed, return -EAGAIN to retry */
1534 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1535 if (rc < 0)
1536 GOTO(out, rc = -EAGAIN);
1537
1538 if (rc > aa->aa_requested_nob) {
1539 CERROR("Unexpected rc %d (%d requested)\n", rc,
1540 aa->aa_requested_nob);
0a3bdb00 1541 return -EPROTO;
d7e09d03
PT		 1542 }
1543
1544 if (rc != req->rq_bulk->bd_nob_transferred) {
1545 CERROR ("Unexpected rc %d (%d transferred)\n",
1546 rc, req->rq_bulk->bd_nob_transferred);
1547 return (-EPROTO);
1548 }
1549
1550 if (rc < aa->aa_requested_nob)
1551 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1552
1553 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1554 static int cksum_counter;
1555 __u32 server_cksum = body->oa.o_cksum;
1556 char *via;
1557 char *router;
1558 cksum_type_t cksum_type;
1559
1560 cksum_type = cksum_type_unpack(body->oa.o_valid &OBD_MD_FLFLAGS?
1561 body->oa.o_flags : 0);
1562 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1563 aa->aa_ppga, OST_READ,
1564 cksum_type);
1565
1566 if (peer->nid == req->rq_bulk->bd_sender) {
1567 via = router = "";
1568 } else {
1569 via = " via ";
1570 router = libcfs_nid2str(req->rq_bulk->bd_sender);
1571 }
1572
a2ff0f97 1573 if (server_cksum != client_cksum) {
d7e09d03
PT		 1574 LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1575 "%s%s%s inode "DFID" object "DOSTID
1576 " extent ["LPU64"-"LPU64"]\n",
1577 req->rq_import->imp_obd->obd_name,
1578 libcfs_nid2str(peer->nid),
1579 via, router,
1580 body->oa.o_valid & OBD_MD_FLFID ?
1581 body->oa.o_parent_seq : (__u64)0,
1582 body->oa.o_valid & OBD_MD_FLFID ?
1583 body->oa.o_parent_oid : 0,
1584 body->oa.o_valid & OBD_MD_FLFID ?
1585 body->oa.o_parent_ver : 0,
1586 POSTID(&body->oa.o_oi),
1587 aa->aa_ppga[0]->off,
1588 aa->aa_ppga[aa->aa_page_count-1]->off +
1589 aa->aa_ppga[aa->aa_page_count-1]->count -
1590 1);
1591 CERROR("client %x, server %x, cksum_type %x\n",
1592 client_cksum, server_cksum, cksum_type);
1593 cksum_counter = 0;
1594 aa->aa_oa->o_cksum = client_cksum;
1595 rc = -EAGAIN;
1596 } else {
1597 cksum_counter++;
1598 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1599 rc = 0;
1600 }
1601 } else if (unlikely(client_cksum)) {
1602 static int cksum_missed;
1603
1604 cksum_missed++;
1605 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1606 CERROR("Checksum %u requested from %s but not sent\n",
1607 cksum_missed, libcfs_nid2str(peer->nid));
1608 } else {
1609 rc = 0;
1610 }
1611out:
1612 if (rc >= 0)
3b2f75fd 1613 lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
1614 aa->aa_oa, &body->oa);
d7e09d03 1615
0a3bdb00 1616 return rc;
d7e09d03
PT		 1617}
1618
1619static int osc_brw_internal(int cmd, struct obd_export *exp, struct obdo *oa,
1620 struct lov_stripe_md *lsm,
1621 obd_count page_count, struct brw_page **pga,
1622 struct obd_capa *ocapa)
1623{
1624 struct ptlrpc_request *req;
1625 int rc;
1626 wait_queue_head_t waitq;
1627 int generation, resends = 0;
1628 struct l_wait_info lwi;
1629
d7e09d03
PT		 1630 init_waitqueue_head(&waitq);
1631 generation = exp->exp_obd->u.cli.cl_import->imp_generation;
1632
1633restart_bulk:
1634 rc = osc_brw_prep_request(cmd, &exp->exp_obd->u.cli, oa, lsm,
1635 page_count, pga, &req, ocapa, 0, resends);
1636 if (rc != 0)
1637 return (rc);
1638
1639 if (resends) {
1640 req->rq_generation_set = 1;
1641 req->rq_import_generation = generation;
1642 req->rq_sent = cfs_time_current_sec() + resends;
1643 }
1644
1645 rc = ptlrpc_queue_wait(req);
1646
1647 if (rc == -ETIMEDOUT && req->rq_resend) {
1648 DEBUG_REQ(D_HA, req, "BULK TIMEOUT");
1649 ptlrpc_req_finished(req);
1650 goto restart_bulk;
1651 }
1652
1653 rc = osc_brw_fini_request(req, rc);
1654
1655 ptlrpc_req_finished(req);
1656 /* When server return -EINPROGRESS, client should always retry
1657 * regardless of the number of times the bulk was resent already.*/
1658 if (osc_recoverable_error(rc)) {
1659 resends++;
1660 if (rc != -EINPROGRESS &&
1661 !client_should_resend(resends, &exp->exp_obd->u.cli)) {
1662 CERROR("%s: too many resend retries for object: "
1663 ""DOSTID", rc = %d.\n", exp->exp_obd->obd_name,
1664 POSTID(&oa->o_oi), rc);
1665 goto out;
1666 }
1667 if (generation !=
1668 exp->exp_obd->u.cli.cl_import->imp_generation) {
1669 CDEBUG(D_HA, "%s: resend cross eviction for object: "
1670 ""DOSTID", rc = %d.\n", exp->exp_obd->obd_name,
1671 POSTID(&oa->o_oi), rc);
1672 goto out;
1673 }
1674
1675 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1676 NULL);
1677 l_wait_event(waitq, 0, &lwi);
1678
1679 goto restart_bulk;
1680 }
1681out:
1682 if (rc == -EAGAIN || rc == -EINPROGRESS)
1683 rc = -EIO;
0a3bdb00 1684 return rc;
d7e09d03
PT		 1685}
1686
1687static int osc_brw_redo_request(struct ptlrpc_request *request,
1688 struct osc_brw_async_args *aa, int rc)
1689{
1690 struct ptlrpc_request *new_req;
1691 struct osc_brw_async_args *new_aa;
1692 struct osc_async_page *oap;
d7e09d03
PT		 1693
1694 DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
1695 "redo for recoverable error %d", rc);
1696
1697 rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1698 OST_WRITE ? OBD_BRW_WRITE :OBD_BRW_READ,
1699 aa->aa_cli, aa->aa_oa,
1700 NULL /* lsm unused by osc currently */,
1701 aa->aa_page_count, aa->aa_ppga,
1702 &new_req, aa->aa_ocapa, 0, 1);
1703 if (rc)
0a3bdb00 1704 return rc;
d7e09d03
PT		 1705
1706 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1707 if (oap->oap_request != NULL) {
1708 LASSERTF(request == oap->oap_request,
1709 "request %p != oap_request %p\n",
1710 request, oap->oap_request);
1711 if (oap->oap_interrupted) {
1712 ptlrpc_req_finished(new_req);
0a3bdb00 1713 return -EINTR;
d7e09d03
PT		 1714 }
1715 }
1716 }
1717 /* New request takes over pga and oaps from old request.
1718 * Note that copying a list_head doesn't work, need to move it... */
1719 aa->aa_resends++;
1720 new_req->rq_interpret_reply = request->rq_interpret_reply;
1721 new_req->rq_async_args = request->rq_async_args;
d7e09d03
PT		 1722 /* cap resend delay to the current request timeout, this is similar to
1723 * what ptlrpc does (see after_reply()) */
1724 if (aa->aa_resends > new_req->rq_timeout)
1725 new_req->rq_sent = cfs_time_current_sec() + new_req->rq_timeout;
1726 else
1727 new_req->rq_sent = cfs_time_current_sec() + aa->aa_resends;
1728 new_req->rq_generation_set = 1;
1729 new_req->rq_import_generation = request->rq_import_generation;
1730
1731 new_aa = ptlrpc_req_async_args(new_req);
1732
1733 INIT_LIST_HEAD(&new_aa->aa_oaps);
1734 list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
1735 INIT_LIST_HEAD(&new_aa->aa_exts);
1736 list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
1737 new_aa->aa_resends = aa->aa_resends;
1738
1739 list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1740 if (oap->oap_request) {
1741 ptlrpc_req_finished(oap->oap_request);
1742 oap->oap_request = ptlrpc_request_addref(new_req);
1743 }
1744 }
1745
1746 new_aa->aa_ocapa = aa->aa_ocapa;
1747 aa->aa_ocapa = NULL;
1748
1749 /* XXX: This code will run into problem if we're going to support
1750 * to add a series of BRW RPCs into a self-defined ptlrpc_request_set
1751 * and wait for all of them to be finished. We should inherit request
1752 * set from old request. */
1753 ptlrpcd_add_req(new_req, PDL_POLICY_SAME, -1);
1754
1755 DEBUG_REQ(D_INFO, new_req, "new request");
0a3bdb00 1756 return 0;
d7e09d03
PT		 1757}
1758
1759/*
1760 * ugh, we want disk allocation on the target to happen in offset order. we'll
1761 * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1762 * fine for our small page arrays and doesn't require allocation. its an
1763 * insertion sort that swaps elements that are strides apart, shrinking the
1764 * stride down until its '1' and the array is sorted.
1765 */
1766static void sort_brw_pages(struct brw_page **array, int num)
1767{
1768 int stride, i, j;
1769 struct brw_page *tmp;
1770
1771 if (num == 1)
1772 return;
1773 for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1774 ;
1775
1776 do {
1777 stride /= 3;
1778 for (i = stride ; i < num ; i++) {
1779 tmp = array[i];
1780 j = i;
1781 while (j >= stride && array[j - stride]->off > tmp->off) {
1782 array[j] = array[j - stride];
1783 j -= stride;
1784 }
1785 array[j] = tmp;
1786 }
1787 } while (stride > 1);
1788}
1789
1790static obd_count max_unfragmented_pages(struct brw_page **pg, obd_count pages)
1791{
1792 int count = 1;
1793 int offset;
1794 int i = 0;
1795
1796 LASSERT (pages > 0);
1797 offset = pg[i]->off & ~CFS_PAGE_MASK;
1798
1799 for (;;) {
1800 pages--;
1801 if (pages == 0) /* that's all */
1802 return count;
1803
1804 if (offset + pg[i]->count < PAGE_CACHE_SIZE)
1805 return count; /* doesn't end on page boundary */
1806
1807 i++;
1808 offset = pg[i]->off & ~CFS_PAGE_MASK;
1809 if (offset != 0) /* doesn't start on page boundary */
1810 return count;
1811
1812 count++;
1813 }
1814}
1815
1816static struct brw_page **osc_build_ppga(struct brw_page *pga, obd_count count)
1817{
1818 struct brw_page **ppga;
1819 int i;
1820
1821 OBD_ALLOC(ppga, sizeof(*ppga) * count);
1822 if (ppga == NULL)
1823 return NULL;
1824
1825 for (i = 0; i < count; i++)
1826 ppga[i] = pga + i;
1827 return ppga;
1828}
1829
1830static void osc_release_ppga(struct brw_page **ppga, obd_count count)
1831{
1832 LASSERT(ppga != NULL);
1833 OBD_FREE(ppga, sizeof(*ppga) * count);
1834}
1835
1836static int osc_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1837 obd_count page_count, struct brw_page *pga,
1838 struct obd_trans_info *oti)
1839{
1840 struct obdo *saved_oa = NULL;
1841 struct brw_page **ppga, **orig;
1842 struct obd_import *imp = class_exp2cliimp(exp);
1843 struct client_obd *cli;
1844 int rc, page_count_orig;
d7e09d03
PT		 1845
1846 LASSERT((imp != NULL) && (imp->imp_obd != NULL));
1847 cli = &imp->imp_obd->u.cli;
1848
1849 if (cmd & OBD_BRW_CHECK) {
1850 /* The caller just wants to know if there's a chance that this
1851 * I/O can succeed */
1852
1853 if (imp->imp_invalid)
0a3bdb00
GKH		 1854 return -EIO;
1855 return 0;
d7e09d03
PT		 1856 }
1857
1858 /* test_brw with a failed create can trip this, maybe others. */
1859 LASSERT(cli->cl_max_pages_per_rpc);
1860
1861 rc = 0;
1862
1863 orig = ppga = osc_build_ppga(pga, page_count);
1864 if (ppga == NULL)
0a3bdb00 1865 return -ENOMEM;
d7e09d03
PT		 1866 page_count_orig = page_count;
1867
1868 sort_brw_pages(ppga, page_count);
1869 while (page_count) {
1870 obd_count pages_per_brw;
1871
1872 if (page_count > cli->cl_max_pages_per_rpc)
1873 pages_per_brw = cli->cl_max_pages_per_rpc;
1874 else
1875 pages_per_brw = page_count;
1876
1877 pages_per_brw = max_unfragmented_pages(ppga, pages_per_brw);
1878
1879 if (saved_oa != NULL) {
1880 /* restore previously saved oa */
1881 *oinfo->oi_oa = *saved_oa;
1882 } else if (page_count > pages_per_brw) {
1883 /* save a copy of oa (brw will clobber it) */
1884 OBDO_ALLOC(saved_oa);
1885 if (saved_oa == NULL)
1886 GOTO(out, rc = -ENOMEM);
1887 *saved_oa = *oinfo->oi_oa;
1888 }
1889
1890 rc = osc_brw_internal(cmd, exp, oinfo->oi_oa, oinfo->oi_md,
1891 pages_per_brw, ppga, oinfo->oi_capa);
1892
1893 if (rc != 0)
1894 break;
1895
1896 page_count -= pages_per_brw;
1897 ppga += pages_per_brw;
1898 }
1899
1900out:
1901 osc_release_ppga(orig, page_count_orig);
1902
1903 if (saved_oa != NULL)
1904 OBDO_FREE(saved_oa);
1905
0a3bdb00 1906 return rc;
d7e09d03
PT		 1907}
1908
1909static int brw_interpret(const struct lu_env *env,
1910 struct ptlrpc_request *req, void *data, int rc)
1911{
1912 struct osc_brw_async_args *aa = data;
1913 struct osc_extent *ext;
1914 struct osc_extent *tmp;
1915 struct cl_object *obj = NULL;
1916 struct client_obd *cli = aa->aa_cli;
d7e09d03
PT		 1917
1918 rc = osc_brw_fini_request(req, rc);
1919 CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
1920 /* When server return -EINPROGRESS, client should always retry
1921 * regardless of the number of times the bulk was resent already. */
1922 if (osc_recoverable_error(rc)) {
1923 if (req->rq_import_generation !=
1924 req->rq_import->imp_generation) {
1925 CDEBUG(D_HA, "%s: resend cross eviction for object: "
1926 ""DOSTID", rc = %d.\n",
1927 req->rq_import->imp_obd->obd_name,
1928 POSTID(&aa->aa_oa->o_oi), rc);
1929 } else if (rc == -EINPROGRESS ||
1930 client_should_resend(aa->aa_resends, aa->aa_cli)) {
1931 rc = osc_brw_redo_request(req, aa, rc);
1932 } else {
1933 CERROR("%s: too many resent retries for object: "
1934 ""LPU64":"LPU64", rc = %d.\n",
1935 req->rq_import->imp_obd->obd_name,
1936 POSTID(&aa->aa_oa->o_oi), rc);
1937 }
1938
1939 if (rc == 0)
0a3bdb00 1940 return 0;
d7e09d03
PT		 1941 else if (rc == -EAGAIN || rc == -EINPROGRESS)
1942 rc = -EIO;
1943 }
1944
1945 if (aa->aa_ocapa) {
1946 capa_put(aa->aa_ocapa);
1947 aa->aa_ocapa = NULL;
1948 }
1949
1950 list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
1951 if (obj == NULL && rc == 0) {
1952 obj = osc2cl(ext->oe_obj);
1953 cl_object_get(obj);
1954 }
1955
1956 list_del_init(&ext->oe_link);
1957 osc_extent_finish(env, ext, 1, rc);
1958 }
1959 LASSERT(list_empty(&aa->aa_exts));
1960 LASSERT(list_empty(&aa->aa_oaps));
1961
1962 if (obj != NULL) {
1963 struct obdo *oa = aa->aa_oa;
1964 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1965 unsigned long valid = 0;
1966
1967 LASSERT(rc == 0);
1968 if (oa->o_valid & OBD_MD_FLBLOCKS) {
1969 attr->cat_blocks = oa->o_blocks;
1970 valid |= CAT_BLOCKS;
1971 }
1972 if (oa->o_valid & OBD_MD_FLMTIME) {
1973 attr->cat_mtime = oa->o_mtime;
1974 valid |= CAT_MTIME;
1975 }
1976 if (oa->o_valid & OBD_MD_FLATIME) {
1977 attr->cat_atime = oa->o_atime;
1978 valid |= CAT_ATIME;
1979 }
1980 if (oa->o_valid & OBD_MD_FLCTIME) {
1981 attr->cat_ctime = oa->o_ctime;
1982 valid |= CAT_CTIME;
1983 }
1984 if (valid != 0) {
1985 cl_object_attr_lock(obj);
1986 cl_object_attr_set(env, obj, attr, valid);
1987 cl_object_attr_unlock(obj);
1988 }
1989 cl_object_put(env, obj);
1990 }
1991 OBDO_FREE(aa->aa_oa);
1992
1993 cl_req_completion(env, aa->aa_clerq, rc < 0 ? rc :
1994 req->rq_bulk->bd_nob_transferred);
1995 osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
1996 ptlrpc_lprocfs_brw(req, req->rq_bulk->bd_nob_transferred);
1997
1998 client_obd_list_lock(&cli->cl_loi_list_lock);
1999 /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
2000 * is called so we know whether to go to sync BRWs or wait for more
2001 * RPCs to complete */
2002 if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
2003 cli->cl_w_in_flight--;
2004 else
2005 cli->cl_r_in_flight--;
2006 osc_wake_cache_waiters(cli);
2007 client_obd_list_unlock(&cli->cl_loi_list_lock);
2008
2009 osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
0a3bdb00 2010 return rc;
d7e09d03
PT		 2011}
2012
d7e09d03
PT		 2013/**
2014 * Build an RPC by the list of extent @ext_list. The caller must ensure
2015 * that the total pages in this list are NOT over max pages per RPC.
2016 * Extents in the list must be in OES_RPC state.
2017 */
2018int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
2019 struct list_head *ext_list, int cmd, pdl_policy_t pol)
2020{
cad6fafa
BJ		 2021 struct ptlrpc_request *req = NULL;
2022 struct osc_extent *ext;
2023 struct brw_page **pga = NULL;
2024 struct osc_brw_async_args *aa = NULL;
2025 struct obdo *oa = NULL;
2026 struct osc_async_page *oap;
2027 struct osc_async_page *tmp;
2028 struct cl_req *clerq = NULL;
2029 enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE :
2030 CRT_READ;
2031 struct ldlm_lock *lock = NULL;
2032 struct cl_req_attr *crattr = NULL;
2033 obd_off starting_offset = OBD_OBJECT_EOF;
2034 obd_off ending_offset = 0;
2035 int mpflag = 0;
2036 int mem_tight = 0;
2037 int page_count = 0;
2038 int i;
2039 int rc;
d7e09d03 2040 LIST_HEAD(rpc_list);
d7e09d03 2041
d7e09d03
PT		 2042 LASSERT(!list_empty(ext_list));
2043
2044 /* add pages into rpc_list to build BRW rpc */
2045 list_for_each_entry(ext, ext_list, oe_link) {
2046 LASSERT(ext->oe_state == OES_RPC);
2047 mem_tight |= ext->oe_memalloc;
2048 list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
2049 ++page_count;
2050 list_add_tail(&oap->oap_rpc_item, &rpc_list);
2051 if (starting_offset > oap->oap_obj_off)
2052 starting_offset = oap->oap_obj_off;
2053 else
2054 LASSERT(oap->oap_page_off == 0);
2055 if (ending_offset < oap->oap_obj_off + oap->oap_count)
2056 ending_offset = oap->oap_obj_off +
2057 oap->oap_count;
2058 else
2059 LASSERT(oap->oap_page_off + oap->oap_count ==
2060 PAGE_CACHE_SIZE);
2061 }
2062 }
2063
2064 if (mem_tight)
2065 mpflag = cfs_memory_pressure_get_and_set();
2066
cad6fafa
BJ		 2067 OBD_ALLOC(crattr, sizeof(*crattr));
2068 if (crattr == NULL)
2069 GOTO(out, rc = -ENOMEM);
2070
d7e09d03
PT		 2071 OBD_ALLOC(pga, sizeof(*pga) * page_count);
2072 if (pga == NULL)
2073 GOTO(out, rc = -ENOMEM);
2074
2075 OBDO_ALLOC(oa);
2076 if (oa == NULL)
2077 GOTO(out, rc = -ENOMEM);
2078
2079 i = 0;
2080 list_for_each_entry(oap, &rpc_list, oap_rpc_item) {
2081 struct cl_page *page = oap2cl_page(oap);
2082 if (clerq == NULL) {
2083 clerq = cl_req_alloc(env, page, crt,
cad6fafa 2084 1 /* only 1-object rpcs for now */);
d7e09d03
PT		 2085 if (IS_ERR(clerq))
2086 GOTO(out, rc = PTR_ERR(clerq));
2087 lock = oap->oap_ldlm_lock;
2088 }
2089 if (mem_tight)
2090 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
2091 pga[i] = &oap->oap_brw_page;
2092 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
2093 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
cad6fafa
BJ		 2094 pga[i]->pg, page_index(oap->oap_page), oap,
2095 pga[i]->flag);
d7e09d03
PT		 2096 i++;
2097 cl_req_page_add(env, clerq, page);
2098 }
2099
2100 /* always get the data for the obdo for the rpc */
2101 LASSERT(clerq != NULL);
cad6fafa
BJ		 2102 crattr->cra_oa = oa;
2103 cl_req_attr_set(env, clerq, crattr, ~0ULL);
d7e09d03
PT		 2104 if (lock) {
2105 oa->o_handle = lock->l_remote_handle;
2106 oa->o_valid |= OBD_MD_FLHANDLE;
2107 }
2108
2109 rc = cl_req_prep(env, clerq);
2110 if (rc != 0) {
2111 CERROR("cl_req_prep failed: %d\n", rc);
2112 GOTO(out, rc);
2113 }
2114
2115 sort_brw_pages(pga, page_count);
2116 rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
cad6fafa 2117 pga, &req, crattr->cra_capa, 1, 0);
d7e09d03
PT		 2118 if (rc != 0) {
2119 CERROR("prep_req failed: %d\n", rc);
2120 GOTO(out, rc);
2121 }
2122
d7e09d03
PT		 2123 req->rq_interpret_reply = brw_interpret;
2124
2125 if (mem_tight != 0)
2126 req->rq_memalloc = 1;
2127
2128 /* Need to update the timestamps after the request is built in case
2129 * we race with setattr (locally or in queue at OST). If OST gets
2130 * later setattr before earlier BRW (as determined by the request xid),
2131 * the OST will not use BRW timestamps. Sadly, there is no obvious
2132 * way to do this in a single call. bug 10150 */
cad6fafa 2133 cl_req_attr_set(env, clerq, crattr,
d7e09d03
PT		 2134 OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
2135
cad6fafa 2136 lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
d7e09d03
PT		 2137
2138 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2139 aa = ptlrpc_req_async_args(req);
2140 INIT_LIST_HEAD(&aa->aa_oaps);
2141 list_splice_init(&rpc_list, &aa->aa_oaps);
2142 INIT_LIST_HEAD(&aa->aa_exts);
2143 list_splice_init(ext_list, &aa->aa_exts);
2144 aa->aa_clerq = clerq;
2145
2146 /* queued sync pages can be torn down while the pages
2147 * were between the pending list and the rpc */
2148 tmp = NULL;
2149 list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
2150 /* only one oap gets a request reference */
2151 if (tmp == NULL)
2152 tmp = oap;
2153 if (oap->oap_interrupted && !req->rq_intr) {
2154 CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
2155 oap, req);
2156 ptlrpc_mark_interrupted(req);
2157 }
2158 }
2159 if (tmp != NULL)
2160 tmp->oap_request = ptlrpc_request_addref(req);
2161
2162 client_obd_list_lock(&cli->cl_loi_list_lock);
2163 starting_offset >>= PAGE_CACHE_SHIFT;
2164 if (cmd == OBD_BRW_READ) {
2165 cli->cl_r_in_flight++;
2166 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2167 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2168 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2169 starting_offset + 1);
2170 } else {
2171 cli->cl_w_in_flight++;
2172 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2173 lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight);
2174 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2175 starting_offset + 1);
2176 }
2177 client_obd_list_unlock(&cli->cl_loi_list_lock);
2178
2179 DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight",
2180 page_count, aa, cli->cl_r_in_flight,
2181 cli->cl_w_in_flight);
2182
2183 /* XXX: Maybe the caller can check the RPC bulk descriptor to
2184 * see which CPU/NUMA node the majority of pages were allocated
2185 * on, and try to assign the async RPC to the CPU core
2186 * (PDL_POLICY_PREFERRED) to reduce cross-CPU memory traffic.
2187 *
2188 * But on the other hand, we expect that multiple ptlrpcd
2189 * threads and the initial write sponsor can run in parallel,
2190 * especially when data checksum is enabled, which is CPU-bound
2191 * operation and single ptlrpcd thread cannot process in time.
2192 * So more ptlrpcd threads sharing BRW load
2193 * (with PDL_POLICY_ROUND) seems better.
2194 */
2195 ptlrpcd_add_req(req, pol, -1);
2196 rc = 0;
d7e09d03
PT		 2197
2198out:
2199 if (mem_tight != 0)
2200 cfs_memory_pressure_restore(mpflag);
2201
cad6fafa
BJ		 2202 if (crattr != NULL) {
2203 capa_put(crattr->cra_capa);
2204 OBD_FREE(crattr, sizeof(*crattr));
2205 }
2206
d7e09d03
PT		 2207 if (rc != 0) {
2208 LASSERT(req == NULL);
2209
2210 if (oa)
2211 OBDO_FREE(oa);
2212 if (pga)
2213 OBD_FREE(pga, sizeof(*pga) * page_count);
2214 /* this should happen rarely and is pretty bad, it makes the
2215 * pending list not follow the dirty order */
2216 while (!list_empty(ext_list)) {
2217 ext = list_entry(ext_list->next, struct osc_extent,
2218 oe_link);
2219 list_del_init(&ext->oe_link);
2220 osc_extent_finish(env, ext, 0, rc);
2221 }
2222 if (clerq && !IS_ERR(clerq))
2223 cl_req_completion(env, clerq, rc);
2224 }
0a3bdb00 2225 return rc;
d7e09d03
PT		 2226}
2227
2228static int osc_set_lock_data_with_check(struct ldlm_lock *lock,
2229 struct ldlm_enqueue_info *einfo)
2230{
2231 void *data = einfo->ei_cbdata;
2232 int set = 0;
2233
2234 LASSERT(lock != NULL);
2235 LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl);
2236 LASSERT(lock->l_resource->lr_type == einfo->ei_type);
2237 LASSERT(lock->l_completion_ast == einfo->ei_cb_cp);
2238 LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
2239
2240 lock_res_and_lock(lock);
2241 spin_lock(&osc_ast_guard);
2242
2243 if (lock->l_ast_data == NULL)
2244 lock->l_ast_data = data;
2245 if (lock->l_ast_data == data)
2246 set = 1;
2247
2248 spin_unlock(&osc_ast_guard);
2249 unlock_res_and_lock(lock);
2250
2251 return set;
2252}
2253
2254static int osc_set_data_with_check(struct lustre_handle *lockh,
2255 struct ldlm_enqueue_info *einfo)
2256{
2257 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2258 int set = 0;
2259
2260 if (lock != NULL) {
2261 set = osc_set_lock_data_with_check(lock, einfo);
2262 LDLM_LOCK_PUT(lock);
2263 } else
2264 CERROR("lockh %p, data %p - client evicted?\n",
2265 lockh, einfo->ei_cbdata);
2266 return set;
2267}
2268
2269static int osc_change_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
2270 ldlm_iterator_t replace, void *data)
2271{
2272 struct ldlm_res_id res_id;
2273 struct obd_device *obd = class_exp2obd(exp);
2274
2275 ostid_build_res_name(&lsm->lsm_oi, &res_id);
2276 ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
2277 return 0;
2278}
2279
2280/* find any ldlm lock of the inode in osc
2281 * return 0 not find
2282 * 1 find one
2283 * < 0 error */
2284static int osc_find_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
2285 ldlm_iterator_t replace, void *data)
2286{
2287 struct ldlm_res_id res_id;
2288 struct obd_device *obd = class_exp2obd(exp);
2289 int rc = 0;
2290
2291 ostid_build_res_name(&lsm->lsm_oi, &res_id);
2292 rc = ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
2293 if (rc == LDLM_ITER_STOP)
2294 return(1);
2295 if (rc == LDLM_ITER_CONTINUE)
2296 return(0);
2297 return(rc);
2298}
2299
2300static int osc_enqueue_fini(struct ptlrpc_request *req, struct ost_lvb *lvb,
2301 obd_enqueue_update_f upcall, void *cookie,
2302 __u64 *flags, int agl, int rc)
2303{
2304 int intent = *flags & LDLM_FL_HAS_INTENT;
d7e09d03
PT		 2305
2306 if (intent) {
2307 /* The request was created before ldlm_cli_enqueue call. */
2308 if (rc == ELDLM_LOCK_ABORTED) {
2309 struct ldlm_reply *rep;
2310 rep = req_capsule_server_get(&req->rq_pill,
2311 &RMF_DLM_REP);
2312
2313 LASSERT(rep != NULL);
2d58de78
LW		 2314 rep->lock_policy_res1 =
2315 ptlrpc_status_ntoh(rep->lock_policy_res1);
d7e09d03
PT		 2316 if (rep->lock_policy_res1)
2317 rc = rep->lock_policy_res1;
2318 }
2319 }
2320
2321 if ((intent != 0 && rc == ELDLM_LOCK_ABORTED && agl == 0) ||
2322 (rc == 0)) {
2323 *flags |= LDLM_FL_LVB_READY;
2324 CDEBUG(D_INODE,"got kms "LPU64" blocks "LPU64" mtime "LPU64"\n",
2325 lvb->lvb_size, lvb->lvb_blocks, lvb->lvb_mtime);
2326 }
2327
2328 /* Call the update callback. */
2329 rc = (*upcall)(cookie, rc);
0a3bdb00 2330 return rc;
d7e09d03
PT		 2331}
2332
2333static int osc_enqueue_interpret(const struct lu_env *env,
2334 struct ptlrpc_request *req,
2335 struct osc_enqueue_args *aa, int rc)
2336{
2337 struct ldlm_lock *lock;
2338 struct lustre_handle handle;
2339 __u32 mode;
2340 struct ost_lvb *lvb;
2341 __u32 lvb_len;
2342 __u64 *flags = aa->oa_flags;
2343
2344 /* Make a local copy of a lock handle and a mode, because aa->oa_*
2345 * might be freed anytime after lock upcall has been called. */
2346 lustre_handle_copy(&handle, aa->oa_lockh);
2347 mode = aa->oa_ei->ei_mode;
2348
2349 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
2350 * be valid. */
2351 lock = ldlm_handle2lock(&handle);
2352
2353 /* Take an additional reference so that a blocking AST that
2354 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
2355 * to arrive after an upcall has been executed by
2356 * osc_enqueue_fini(). */
2357 ldlm_lock_addref(&handle, mode);
2358
2359 /* Let CP AST to grant the lock first. */
2360 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
2361
2362 if (aa->oa_agl && rc == ELDLM_LOCK_ABORTED) {
2363 lvb = NULL;
2364 lvb_len = 0;
2365 } else {
2366 lvb = aa->oa_lvb;
2367 lvb_len = sizeof(*aa->oa_lvb);
2368 }
2369
2370 /* Complete obtaining the lock procedure. */
2371 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1,
2372 mode, flags, lvb, lvb_len, &handle, rc);
2373 /* Complete osc stuff. */
2374 rc = osc_enqueue_fini(req, aa->oa_lvb, aa->oa_upcall, aa->oa_cookie,
2375 flags, aa->oa_agl, rc);
2376
2377 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
2378
2379 /* Release the lock for async request. */
2380 if (lustre_handle_is_used(&handle) && rc == ELDLM_OK)
2381 /*
2382 * Releases a reference taken by ldlm_cli_enqueue(), if it is
2383 * not already released by
2384 * ldlm_cli_enqueue_fini()->failed_lock_cleanup()
2385 */
2386 ldlm_lock_decref(&handle, mode);
2387
2388 LASSERTF(lock != NULL, "lockh %p, req %p, aa %p - client evicted?\n",
2389 aa->oa_lockh, req, aa);
2390 ldlm_lock_decref(&handle, mode);
2391 LDLM_LOCK_PUT(lock);
2392 return rc;
2393}
2394
2395void osc_update_enqueue(struct lustre_handle *lov_lockhp,
875332d4 2396 struct lov_oinfo *loi, __u64 flags,
d7e09d03
PT		 2397 struct ost_lvb *lvb, __u32 mode, int rc)
2398{
2399 struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
2400
2401 if (rc == ELDLM_OK) {
2402 __u64 tmp;
2403
2404 LASSERT(lock != NULL);
2405 loi->loi_lvb = *lvb;
2406 tmp = loi->loi_lvb.lvb_size;
2407 /* Extend KMS up to the end of this lock and no further
2408 * A lock on [x,y] means a KMS of up to y + 1 bytes! */
2409 if (tmp > lock->l_policy_data.l_extent.end)
2410 tmp = lock->l_policy_data.l_extent.end + 1;
2411 if (tmp >= loi->loi_kms) {
2412 LDLM_DEBUG(lock, "lock acquired, setting rss="LPU64
2413 ", kms="LPU64, loi->loi_lvb.lvb_size, tmp);
2414 loi_kms_set(loi, tmp);
2415 } else {
2416 LDLM_DEBUG(lock, "lock acquired, setting rss="
2417 LPU64"; leaving kms="LPU64", end="LPU64,
2418 loi->loi_lvb.lvb_size, loi->loi_kms,
2419 lock->l_policy_data.l_extent.end);
2420 }
2421 ldlm_lock_allow_match(lock);
2422 } else if (rc == ELDLM_LOCK_ABORTED && (flags & LDLM_FL_HAS_INTENT)) {
2423 LASSERT(lock != NULL);
2424 loi->loi_lvb = *lvb;
2425 ldlm_lock_allow_match(lock);
2426 CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
2427 " kms="LPU64"\n", loi->loi_lvb.lvb_size, loi->loi_kms);
2428 rc = ELDLM_OK;
2429 }
2430
2431 if (lock != NULL) {
2432 if (rc != ELDLM_OK)
2433 ldlm_lock_fail_match(lock);
2434
2435 LDLM_LOCK_PUT(lock);
2436 }
2437}
2438EXPORT_SYMBOL(osc_update_enqueue);
2439
2440struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
2441
2442/* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
2443 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
2444 * other synchronous requests, however keeping some locks and trying to obtain
2445 * others may take a considerable amount of time in a case of ost failure; and
2446 * when other sync requests do not get released lock from a client, the client
2447 * is excluded from the cluster -- such scenarious make the life difficult, so
2448 * release locks just after they are obtained. */
2449int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2450 __u64 *flags, ldlm_policy_data_t *policy,
2451 struct ost_lvb *lvb, int kms_valid,
2452 obd_enqueue_update_f upcall, void *cookie,
2453 struct ldlm_enqueue_info *einfo,
2454 struct lustre_handle *lockh,
2455 struct ptlrpc_request_set *rqset, int async, int agl)
2456{
2457 struct obd_device *obd = exp->exp_obd;
2458 struct ptlrpc_request *req = NULL;
2459 int intent = *flags & LDLM_FL_HAS_INTENT;
875332d4 2460 __u64 match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY);
d7e09d03
PT		 2461 ldlm_mode_t mode;
2462 int rc;
d7e09d03
PT		 2463
2464 /* Filesystem lock extents are extended to page boundaries so that
2465 * dealing with the page cache is a little smoother. */
2466 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
2467 policy->l_extent.end |= ~CFS_PAGE_MASK;
2468
2469 /*
2470 * kms is not valid when either object is completely fresh (so that no
2471 * locks are cached), or object was evicted. In the latter case cached
2472 * lock cannot be used, because it would prime inode state with
2473 * potentially stale LVB.
2474 */
2475 if (!kms_valid)
2476 goto no_match;
2477
2478 /* Next, search for already existing extent locks that will cover us */
2479 /* If we're trying to read, we also search for an existing PW lock. The
2480 * VFS and page cache already protect us locally, so lots of readers/
2481 * writers can share a single PW lock.
2482 *
2483 * There are problems with conversion deadlocks, so instead of
2484 * converting a read lock to a write lock, we'll just enqueue a new
2485 * one.
2486 *
2487 * At some point we should cancel the read lock instead of making them
2488 * send us a blocking callback, but there are problems with canceling
2489 * locks out from other users right now, too. */
2490 mode = einfo->ei_mode;
2491 if (einfo->ei_mode == LCK_PR)
2492 mode |= LCK_PW;
2493 mode = ldlm_lock_match(obd->obd_namespace, *flags | match_lvb, res_id,
2494 einfo->ei_type, policy, mode, lockh, 0);
2495 if (mode) {
2496 struct ldlm_lock *matched = ldlm_handle2lock(lockh);
2497
2498 if ((agl != 0) && !(matched->l_flags & LDLM_FL_LVB_READY)) {
2499 /* For AGL, if enqueue RPC is sent but the lock is not
2500 * granted, then skip to process this strpe.
2501 * Return -ECANCELED to tell the caller. */
2502 ldlm_lock_decref(lockh, mode);
2503 LDLM_LOCK_PUT(matched);
0a3bdb00 2504 return -ECANCELED;
d7e09d03
PT		 2505 } else if (osc_set_lock_data_with_check(matched, einfo)) {
2506 *flags |= LDLM_FL_LVB_READY;
2507 /* addref the lock only if not async requests and PW
2508 * lock is matched whereas we asked for PR. */
2509 if (!rqset && einfo->ei_mode != mode)
2510 ldlm_lock_addref(lockh, LCK_PR);
2511 if (intent) {
2512 /* I would like to be able to ASSERT here that
2513 * rss <= kms, but I can't, for reasons which
2514 * are explained in lov_enqueue() */
2515 }
2516
2517 /* We already have a lock, and it's referenced.
2518 *
2519 * At this point, the cl_lock::cll_state is CLS_QUEUING,
2520 * AGL upcall may change it to CLS_HELD directly. */
2521 (*upcall)(cookie, ELDLM_OK);
2522
2523 if (einfo->ei_mode != mode)
2524 ldlm_lock_decref(lockh, LCK_PW);
2525 else if (rqset)
2526 /* For async requests, decref the lock. */
2527 ldlm_lock_decref(lockh, einfo->ei_mode);
2528 LDLM_LOCK_PUT(matched);
0a3bdb00 2529 return ELDLM_OK;
d7e09d03
PT		 2530 } else {
2531 ldlm_lock_decref(lockh, mode);
2532 LDLM_LOCK_PUT(matched);
2533 }
2534 }
2535
2536 no_match:
2537 if (intent) {
2538 LIST_HEAD(cancels);
2539 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2540 &RQF_LDLM_ENQUEUE_LVB);
2541 if (req == NULL)
0a3bdb00 2542 return -ENOMEM;
d7e09d03
PT		 2543
2544 rc = ldlm_prep_enqueue_req(exp, req, &cancels, 0);
2545 if (rc) {
2546 ptlrpc_request_free(req);
0a3bdb00 2547 return rc;
d7e09d03
PT		 2548 }
2549
2550 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
ec83e611 2551 sizeof(*lvb));
d7e09d03
PT		 2552 ptlrpc_request_set_replen(req);
2553 }
2554
2555 /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
2556 *flags &= ~LDLM_FL_BLOCK_GRANTED;
2557
2558 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
2559 sizeof(*lvb), LVB_T_OST, lockh, async);
2560 if (rqset) {
2561 if (!rc) {
2562 struct osc_enqueue_args *aa;
2563 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
2564 aa = ptlrpc_req_async_args(req);
2565 aa->oa_ei = einfo;
2566 aa->oa_exp = exp;
2567 aa->oa_flags = flags;
2568 aa->oa_upcall = upcall;
2569 aa->oa_cookie = cookie;
2570 aa->oa_lvb = lvb;
2571 aa->oa_lockh = lockh;
2572 aa->oa_agl = !!agl;
2573
2574 req->rq_interpret_reply =
2575 (ptlrpc_interpterer_t)osc_enqueue_interpret;
2576 if (rqset == PTLRPCD_SET)
2577 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
2578 else
2579 ptlrpc_set_add_req(rqset, req);
2580 } else if (intent) {
2581 ptlrpc_req_finished(req);
2582 }
0a3bdb00 2583 return rc;
d7e09d03
PT		 2584 }
2585
2586 rc = osc_enqueue_fini(req, lvb, upcall, cookie, flags, agl, rc);
2587 if (intent)
2588 ptlrpc_req_finished(req);
2589
0a3bdb00 2590 return rc;
d7e09d03
PT		 2591}
2592
2593static int osc_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2594 struct ldlm_enqueue_info *einfo,
2595 struct ptlrpc_request_set *rqset)
2596{
2597 struct ldlm_res_id res_id;
2598 int rc;
d7e09d03
PT		 2599
2600 ostid_build_res_name(&oinfo->oi_md->lsm_oi, &res_id);
2601 rc = osc_enqueue_base(exp, &res_id, &oinfo->oi_flags, &oinfo->oi_policy,
2602 &oinfo->oi_md->lsm_oinfo[0]->loi_lvb,
2603 oinfo->oi_md->lsm_oinfo[0]->loi_kms_valid,
2604 oinfo->oi_cb_up, oinfo, einfo, oinfo->oi_lockh,
2605 rqset, rqset != NULL, 0);
0a3bdb00 2606 return rc;
d7e09d03
PT		 2607}
2608
2609int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
2610 __u32 type, ldlm_policy_data_t *policy, __u32 mode,
875332d4 2611 __u64 *flags, void *data, struct lustre_handle *lockh,
d7e09d03
PT		 2612 int unref)
2613{
2614 struct obd_device *obd = exp->exp_obd;
875332d4 2615 __u64 lflags = *flags;
d7e09d03 2616 ldlm_mode_t rc;
d7e09d03
PT		 2617
2618 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
0a3bdb00 2619 return -EIO;
d7e09d03
PT		 2620
2621 /* Filesystem lock extents are extended to page boundaries so that
2622 * dealing with the page cache is a little smoother */
2623 policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
2624 policy->l_extent.end |= ~CFS_PAGE_MASK;
2625
2626 /* Next, search for already existing extent locks that will cover us */
2627 /* If we're trying to read, we also search for an existing PW lock. The
2628 * VFS and page cache already protect us locally, so lots of readers/
2629 * writers can share a single PW lock. */
2630 rc = mode;
2631 if (mode == LCK_PR)
2632 rc |= LCK_PW;
2633 rc = ldlm_lock_match(obd->obd_namespace, lflags,
2634 res_id, type, policy, rc, lockh, unref);
2635 if (rc) {
2636 if (data != NULL) {
2637 if (!osc_set_data_with_check(lockh, data)) {
2638 if (!(lflags & LDLM_FL_TEST_LOCK))
2639 ldlm_lock_decref(lockh, rc);
0a3bdb00 2640 return 0;
d7e09d03
PT		 2641 }
2642 }
2643 if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) {
2644 ldlm_lock_addref(lockh, LCK_PR);
2645 ldlm_lock_decref(lockh, LCK_PW);
2646 }
0a3bdb00 2647 return rc;
d7e09d03 2648 }
0a3bdb00 2649 return rc;
d7e09d03
PT		 2650}
2651
2652int osc_cancel_base(struct lustre_handle *lockh, __u32 mode)
2653{
d7e09d03
PT		 2654 if (unlikely(mode == LCK_GROUP))
2655 ldlm_lock_decref_and_cancel(lockh, mode);
2656 else
2657 ldlm_lock_decref(lockh, mode);
2658
0a3bdb00 2659 return 0;
d7e09d03
PT		 2660}
2661
2662static int osc_cancel(struct obd_export *exp, struct lov_stripe_md *md,
2663 __u32 mode, struct lustre_handle *lockh)
2664{
0a3bdb00 2665 return osc_cancel_base(lockh, mode);
d7e09d03
PT		 2666}
2667
2668static int osc_cancel_unused(struct obd_export *exp,
2669 struct lov_stripe_md *lsm,
2670 ldlm_cancel_flags_t flags,
2671 void *opaque)
2672{
2673 struct obd_device *obd = class_exp2obd(exp);
2674 struct ldlm_res_id res_id, *resp = NULL;
2675
2676 if (lsm != NULL) {
2677 ostid_build_res_name(&lsm->lsm_oi, &res_id);
2678 resp = &res_id;
2679 }
2680
2681 return ldlm_cli_cancel_unused(obd->obd_namespace, resp, flags, opaque);
2682}
2683
2684static int osc_statfs_interpret(const struct lu_env *env,
2685 struct ptlrpc_request *req,
2686 struct osc_async_args *aa, int rc)
2687{
2688 struct obd_statfs *msfs;
d7e09d03
PT		 2689
2690 if (rc == -EBADR)
2691 /* The request has in fact never been sent
2692 * due to issues at a higher level (LOV).
2693 * Exit immediately since the caller is
2694 * aware of the problem and takes care
2695 * of the clean up */
0a3bdb00 2696 return rc;
d7e09d03
PT		 2697
2698 if ((rc == -ENOTCONN || rc == -EAGAIN) &&
2699 (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
2700 GOTO(out, rc = 0);
2701
2702 if (rc != 0)
2703 GOTO(out, rc);
2704
2705 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2706 if (msfs == NULL) {
2707 GOTO(out, rc = -EPROTO);
2708 }
2709
2710 *aa->aa_oi->oi_osfs = *msfs;
2711out:
2712 rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
0a3bdb00 2713 return rc;
d7e09d03
PT		 2714}
2715
2716static int osc_statfs_async(struct obd_export *exp,
2717 struct obd_info *oinfo, __u64 max_age,
2718 struct ptlrpc_request_set *rqset)
2719{
2720 struct obd_device *obd = class_exp2obd(exp);
2721 struct ptlrpc_request *req;
2722 struct osc_async_args *aa;
2723 int rc;
d7e09d03
PT		 2724
2725 /* We could possibly pass max_age in the request (as an absolute
2726 * timestamp or a "seconds.usec ago") so the target can avoid doing
2727 * extra calls into the filesystem if that isn't necessary (e.g.
2728 * during mount that would help a bit). Having relative timestamps
2729 * is not so great if request processing is slow, while absolute
2730 * timestamps are not ideal because they need time synchronization. */
2731 req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS);
2732 if (req == NULL)
0a3bdb00 2733 return -ENOMEM;
d7e09d03
PT		 2734
2735 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2736 if (rc) {
2737 ptlrpc_request_free(req);
0a3bdb00 2738 return rc;
d7e09d03
PT		 2739 }
2740 ptlrpc_request_set_replen(req);
2741 req->rq_request_portal = OST_CREATE_PORTAL;
2742 ptlrpc_at_set_req_timeout(req);
2743
2744 if (oinfo->oi_flags & OBD_STATFS_NODELAY) {
2745 /* procfs requests not want stat in wait for avoid deadlock */
2746 req->rq_no_resend = 1;
2747 req->rq_no_delay = 1;
2748 }
2749
2750 req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
2751 CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
2752 aa = ptlrpc_req_async_args(req);
2753 aa->aa_oi = oinfo;
2754
2755 ptlrpc_set_add_req(rqset, req);
0a3bdb00 2756 return 0;
d7e09d03
PT		 2757}
2758
2759static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
2760 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
2761{
2762 struct obd_device *obd = class_exp2obd(exp);
2763 struct obd_statfs *msfs;
2764 struct ptlrpc_request *req;
2765 struct obd_import *imp = NULL;
2766 int rc;
d7e09d03
PT		 2767
2768 /*Since the request might also come from lprocfs, so we need
2769 *sync this with client_disconnect_export Bug15684*/
2770 down_read(&obd->u.cli.cl_sem);
2771 if (obd->u.cli.cl_import)
2772 imp = class_import_get(obd->u.cli.cl_import);
2773 up_read(&obd->u.cli.cl_sem);
2774 if (!imp)
0a3bdb00 2775 return -ENODEV;
d7e09d03
PT		 2776
2777 /* We could possibly pass max_age in the request (as an absolute
2778 * timestamp or a "seconds.usec ago") so the target can avoid doing
2779 * extra calls into the filesystem if that isn't necessary (e.g.
2780 * during mount that would help a bit). Having relative timestamps
2781 * is not so great if request processing is slow, while absolute
2782 * timestamps are not ideal because they need time synchronization. */
2783 req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
2784
2785 class_import_put(imp);
2786
2787 if (req == NULL)
0a3bdb00 2788 return -ENOMEM;
d7e09d03
PT		 2789
2790 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
2791 if (rc) {
2792 ptlrpc_request_free(req);
0a3bdb00 2793 return rc;
d7e09d03
PT		 2794 }
2795 ptlrpc_request_set_replen(req);
2796 req->rq_request_portal = OST_CREATE_PORTAL;
2797 ptlrpc_at_set_req_timeout(req);
2798
2799 if (flags & OBD_STATFS_NODELAY) {
2800 /* procfs requests not want stat in wait for avoid deadlock */
2801 req->rq_no_resend = 1;
2802 req->rq_no_delay = 1;
2803 }
2804
2805 rc = ptlrpc_queue_wait(req);
2806 if (rc)
2807 GOTO(out, rc);
2808
2809 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
2810 if (msfs == NULL) {
2811 GOTO(out, rc = -EPROTO);
2812 }
2813
2814 *osfs = *msfs;
2815
d7e09d03
PT		 2816 out:
2817 ptlrpc_req_finished(req);
2818 return rc;
2819}
2820
2821/* Retrieve object striping information.
2822 *
2823 * @lmmu is a pointer to an in-core struct with lmm_ost_count indicating
2824 * the maximum number of OST indices which will fit in the user buffer.
2825 * lmm_magic must be LOV_MAGIC (we only use 1 slot here).
2826 */
2827static int osc_getstripe(struct lov_stripe_md *lsm, struct lov_user_md *lump)
2828{
2829 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
2830 struct lov_user_md_v3 lum, *lumk;
2831 struct lov_user_ost_data_v1 *lmm_objects;
2832 int rc = 0, lum_size;
d7e09d03
PT		 2833
2834 if (!lsm)
0a3bdb00 2835 return -ENODATA;
d7e09d03
PT		 2836
2837 /* we only need the header part from user space to get lmm_magic and
2838 * lmm_stripe_count, (the header part is common to v1 and v3) */
2839 lum_size = sizeof(struct lov_user_md_v1);
2840 if (copy_from_user(&lum, lump, lum_size))
0a3bdb00 2841 return -EFAULT;
d7e09d03
PT		 2842
2843 if ((lum.lmm_magic != LOV_USER_MAGIC_V1) &&
2844 (lum.lmm_magic != LOV_USER_MAGIC_V3))
0a3bdb00 2845 return -EINVAL;
d7e09d03
PT		 2846
2847 /* lov_user_md_vX and lov_mds_md_vX must have the same size */
2848 LASSERT(sizeof(struct lov_user_md_v1) == sizeof(struct lov_mds_md_v1));
2849 LASSERT(sizeof(struct lov_user_md_v3) == sizeof(struct lov_mds_md_v3));
2850 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lumk->lmm_objects[0]));
2851
2852 /* we can use lov_mds_md_size() to compute lum_size
2853 * because lov_user_md_vX and lov_mds_md_vX have the same size */
2854 if (lum.lmm_stripe_count > 0) {
2855 lum_size = lov_mds_md_size(lum.lmm_stripe_count, lum.lmm_magic);
2856 OBD_ALLOC(lumk, lum_size);
2857 if (!lumk)
0a3bdb00 2858 return -ENOMEM;
d7e09d03
PT		 2859
2860 if (lum.lmm_magic == LOV_USER_MAGIC_V1)
2861 lmm_objects =
2862 &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]);
2863 else
2864 lmm_objects = &(lumk->lmm_objects[0]);
2865 lmm_objects->l_ost_oi = lsm->lsm_oi;
2866 } else {
2867 lum_size = lov_mds_md_size(0, lum.lmm_magic);
2868 lumk = &lum;
2869 }
2870
2871 lumk->lmm_oi = lsm->lsm_oi;
2872 lumk->lmm_stripe_count = 1;
2873
2874 if (copy_to_user(lump, lumk, lum_size))
2875 rc = -EFAULT;
2876
2877 if (lumk != &lum)
2878 OBD_FREE(lumk, lum_size);
2879
0a3bdb00 2880 return rc;
d7e09d03
PT		 2881}
2882
2883
2884static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2885 void *karg, void *uarg)
2886{
2887 struct obd_device *obd = exp->exp_obd;
2888 struct obd_ioctl_data *data = karg;
2889 int err = 0;
d7e09d03
PT		 2890
2891 if (!try_module_get(THIS_MODULE)) {
2892 CERROR("Can't get module. Is it alive?");
2893 return -EINVAL;
2894 }
2895 switch (cmd) {
2896 case OBD_IOC_LOV_GET_CONFIG: {
2897 char *buf;
2898 struct lov_desc *desc;
2899 struct obd_uuid uuid;
2900
2901 buf = NULL;
2902 len = 0;
2903 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2904 GOTO(out, err = -EINVAL);
2905
2906 data = (struct obd_ioctl_data *)buf;
2907
2908 if (sizeof(*desc) > data->ioc_inllen1) {
2909 obd_ioctl_freedata(buf, len);
2910 GOTO(out, err = -EINVAL);
2911 }
2912
2913 if (data->ioc_inllen2 < sizeof(uuid)) {
2914 obd_ioctl_freedata(buf, len);
2915 GOTO(out, err = -EINVAL);
2916 }
2917
2918 desc = (struct lov_desc *)data->ioc_inlbuf1;
2919 desc->ld_tgt_count = 1;
2920 desc->ld_active_tgt_count = 1;
2921 desc->ld_default_stripe_count = 1;
2922 desc->ld_default_stripe_size = 0;
2923 desc->ld_default_stripe_offset = 0;
2924 desc->ld_pattern = 0;
2925 memcpy(&desc->ld_uuid, &obd->obd_uuid, sizeof(uuid));
2926
2927 memcpy(data->ioc_inlbuf2, &obd->obd_uuid, sizeof(uuid));
2928
2929 err = copy_to_user((void *)uarg, buf, len);
2930 if (err)
2931 err = -EFAULT;
2932 obd_ioctl_freedata(buf, len);
2933 GOTO(out, err);
2934 }
2935 case LL_IOC_LOV_SETSTRIPE:
2936 err = obd_alloc_memmd(exp, karg);
2937 if (err > 0)
2938 err = 0;
2939 GOTO(out, err);
2940 case LL_IOC_LOV_GETSTRIPE:
2941 err = osc_getstripe(karg, uarg);
2942 GOTO(out, err);
2943 case OBD_IOC_CLIENT_RECOVER:
2944 err = ptlrpc_recover_import(obd->u.cli.cl_import,
2945 data->ioc_inlbuf1, 0);
2946 if (err > 0)
2947 err = 0;
2948 GOTO(out, err);
2949 case IOC_OSC_SET_ACTIVE:
2950 err = ptlrpc_set_import_active(obd->u.cli.cl_import,
2951 data->ioc_offset);
2952 GOTO(out, err);
2953 case OBD_IOC_POLL_QUOTACHECK:
2954 err = osc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2955 GOTO(out, err);
2956 case OBD_IOC_PING_TARGET:
2957 err = ptlrpc_obd_ping(obd);
2958 GOTO(out, err);
2959 default:
2960 CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
2961 cmd, current_comm());
2962 GOTO(out, err = -ENOTTY);
2963 }
2964out:
2965 module_put(THIS_MODULE);
2966 return err;
2967}
2968
2969static int osc_get_info(const struct lu_env *env, struct obd_export *exp,
2970 obd_count keylen, void *key, __u32 *vallen, void *val,
2971 struct lov_stripe_md *lsm)
2972{
d7e09d03 2973 if (!vallen || !val)
0a3bdb00 2974 return -EFAULT;
d7e09d03
PT		 2975
2976 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2977 __u32 *stripe = val;
2978 *vallen = sizeof(*stripe);
2979 *stripe = 0;
0a3bdb00 2980 return 0;
d7e09d03
PT		 2981 } else if (KEY_IS(KEY_LAST_ID)) {
2982 struct ptlrpc_request *req;
2983 obd_id *reply;
2984 char *tmp;
2985 int rc;
2986
2987 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2988 &RQF_OST_GET_INFO_LAST_ID);
2989 if (req == NULL)
0a3bdb00 2990 return -ENOMEM;
d7e09d03
PT		 2991
2992 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
2993 RCL_CLIENT, keylen);
2994 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
2995 if (rc) {
2996 ptlrpc_request_free(req);
0a3bdb00 2997 return rc;
d7e09d03
PT		 2998 }
2999
3000 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
3001 memcpy(tmp, key, keylen);
3002
3003 req->rq_no_delay = req->rq_no_resend = 1;
3004 ptlrpc_request_set_replen(req);
3005 rc = ptlrpc_queue_wait(req);
3006 if (rc)
3007 GOTO(out, rc);
3008
3009 reply = req_capsule_server_get(&req->rq_pill, &RMF_OBD_ID);
3010 if (reply == NULL)
3011 GOTO(out, rc = -EPROTO);
3012
3013 *((obd_id *)val) = *reply;
3014 out:
3015 ptlrpc_req_finished(req);
0a3bdb00 3016 return rc;
d7e09d03 3017 } else if (KEY_IS(KEY_FIEMAP)) {
9d865439
AB		 3018 struct ll_fiemap_info_key *fm_key =
3019 (struct ll_fiemap_info_key *)key;
3020 struct ldlm_res_id res_id;
3021 ldlm_policy_data_t policy;
3022 struct lustre_handle lockh;
3023 ldlm_mode_t mode = 0;
3024 struct ptlrpc_request *req;
3025 struct ll_user_fiemap *reply;
3026 char *tmp;
3027 int rc;
3028
3029 if (!(fm_key->fiemap.fm_flags & FIEMAP_FLAG_SYNC))
3030 goto skip_locking;
3031
3032 policy.l_extent.start = fm_key->fiemap.fm_start &
3033 CFS_PAGE_MASK;
3034
3035 if (OBD_OBJECT_EOF - fm_key->fiemap.fm_length <=
3036 fm_key->fiemap.fm_start + PAGE_CACHE_SIZE - 1)
3037 policy.l_extent.end = OBD_OBJECT_EOF;
3038 else
3039 policy.l_extent.end = (fm_key->fiemap.fm_start +
3040 fm_key->fiemap.fm_length +
3041 PAGE_CACHE_SIZE - 1) & CFS_PAGE_MASK;
3042
3043 ostid_build_res_name(&fm_key->oa.o_oi, &res_id);
3044 mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
3045 LDLM_FL_BLOCK_GRANTED |
3046 LDLM_FL_LVB_READY,
3047 &res_id, LDLM_EXTENT, &policy,
3048 LCK_PR | LCK_PW, &lockh, 0);
3049 if (mode) { /* lock is cached on client */
3050 if (mode != LCK_PR) {
3051 ldlm_lock_addref(&lockh, LCK_PR);
3052 ldlm_lock_decref(&lockh, LCK_PW);
3053 }
3054 } else { /* no cached lock, needs acquire lock on server side */
3055 fm_key->oa.o_valid |= OBD_MD_FLFLAGS;
3056 fm_key->oa.o_flags |= OBD_FL_SRVLOCK;
3057 }
d7e09d03 3058
9d865439 3059skip_locking:
d7e09d03
PT		 3060 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
3061 &RQF_OST_GET_INFO_FIEMAP);
3062 if (req == NULL)
9d865439 3063 GOTO(drop_lock, rc = -ENOMEM);
d7e09d03
PT		 3064
3065 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY,
3066 RCL_CLIENT, keylen);
3067 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL,
3068 RCL_CLIENT, *vallen);
3069 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL,
3070 RCL_SERVER, *vallen);
3071
3072 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
3073 if (rc) {
3074 ptlrpc_request_free(req);
9d865439 3075 GOTO(drop_lock, rc);
d7e09d03
PT		 3076 }
3077
3078 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
3079 memcpy(tmp, key, keylen);
3080 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
3081 memcpy(tmp, val, *vallen);
3082
3083 ptlrpc_request_set_replen(req);
3084 rc = ptlrpc_queue_wait(req);
3085 if (rc)
9d865439 3086 GOTO(fini_req, rc);
d7e09d03
PT		 3087
3088 reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
3089 if (reply == NULL)
9d865439 3090 GOTO(fini_req, rc = -EPROTO);
d7e09d03
PT		 3091
3092 memcpy(val, reply, *vallen);
9d865439 3093fini_req:
d7e09d03 3094 ptlrpc_req_finished(req);
9d865439
AB		 3095drop_lock:
3096 if (mode)
3097 ldlm_lock_decref(&lockh, LCK_PR);
0a3bdb00 3098 return rc;
d7e09d03
PT		 3099 }
3100
0a3bdb00 3101 return -EINVAL;
d7e09d03
PT		 3102}
3103
3104static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
3105 obd_count keylen, void *key, obd_count vallen,
3106 void *val, struct ptlrpc_request_set *set)
3107{
3108 struct ptlrpc_request *req;
3109 struct obd_device *obd = exp->exp_obd;
3110 struct obd_import *imp = class_exp2cliimp(exp);
3111 char *tmp;
3112 int rc;
d7e09d03
PT		 3113
3114 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
3115
3116 if (KEY_IS(KEY_CHECKSUM)) {
3117 if (vallen != sizeof(int))
0a3bdb00 3118 return -EINVAL;
d7e09d03 3119 exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0;
0a3bdb00 3120 return 0;
d7e09d03
PT		 3121 }
3122
3123 if (KEY_IS(KEY_SPTLRPC_CONF)) {
3124 sptlrpc_conf_client_adapt(obd);
0a3bdb00 3125 return 0;
d7e09d03
PT		 3126 }
3127
3128 if (KEY_IS(KEY_FLUSH_CTX)) {
3129 sptlrpc_import_flush_my_ctx(imp);
0a3bdb00 3130 return 0;
d7e09d03
PT		 3131 }
3132
3133 if (KEY_IS(KEY_CACHE_SET)) {
3134 struct client_obd *cli = &obd->u.cli;
3135
3136 LASSERT(cli->cl_cache == NULL); /* only once */
3137 cli->cl_cache = (struct cl_client_cache *)val;
3138 atomic_inc(&cli->cl_cache->ccc_users);
3139 cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
3140
3141 /* add this osc into entity list */
3142 LASSERT(list_empty(&cli->cl_lru_osc));
3143 spin_lock(&cli->cl_cache->ccc_lru_lock);
3144 list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
3145 spin_unlock(&cli->cl_cache->ccc_lru_lock);
3146
0a3bdb00 3147 return 0;
d7e09d03
PT		 3148 }
3149
3150 if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
3151 struct client_obd *cli = &obd->u.cli;
3152 int nr = atomic_read(&cli->cl_lru_in_list) >> 1;
3153 int target = *(int *)val;
3154
3155 nr = osc_lru_shrink(cli, min(nr, target));
3156 *(int *)val -= nr;
0a3bdb00 3157 return 0;
d7e09d03
PT		 3158 }
3159
3160 if (!set && !KEY_IS(KEY_GRANT_SHRINK))
0a3bdb00 3161 return -EINVAL;
d7e09d03
PT		 3162
3163 /* We pass all other commands directly to OST. Since nobody calls osc
3164 methods directly and everybody is supposed to go through LOV, we
3165 assume lov checked invalid values for us.
3166 The only recognised values so far are evict_by_nid and mds_conn.
3167 Even if something bad goes through, we'd get a -EINVAL from OST
3168 anyway. */
3169
3170 req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
3171 &RQF_OST_SET_GRANT_INFO :
3172 &RQF_OBD_SET_INFO);
3173 if (req == NULL)
0a3bdb00 3174 return -ENOMEM;
d7e09d03
PT		 3175
3176 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
3177 RCL_CLIENT, keylen);
3178 if (!KEY_IS(KEY_GRANT_SHRINK))
3179 req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
3180 RCL_CLIENT, vallen);
3181 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
3182 if (rc) {
3183 ptlrpc_request_free(req);
0a3bdb00 3184 return rc;
d7e09d03
PT		 3185 }
3186
3187 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
3188 memcpy(tmp, key, keylen);
3189 tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
3190 &RMF_OST_BODY :
3191 &RMF_SETINFO_VAL);
3192 memcpy(tmp, val, vallen);
3193
3194 if (KEY_IS(KEY_GRANT_SHRINK)) {
3195 struct osc_grant_args *aa;
3196 struct obdo *oa;
3197
3198 CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
3199 aa = ptlrpc_req_async_args(req);
3200 OBDO_ALLOC(oa);
3201 if (!oa) {
3202 ptlrpc_req_finished(req);
0a3bdb00 3203 return -ENOMEM;
d7e09d03
PT		 3204 }
3205 *oa = ((struct ost_body *)val)->oa;
3206 aa->aa_oa = oa;
3207 req->rq_interpret_reply = osc_shrink_grant_interpret;
3208 }
3209
3210 ptlrpc_request_set_replen(req);
3211 if (!KEY_IS(KEY_GRANT_SHRINK)) {
3212 LASSERT(set != NULL);
3213 ptlrpc_set_add_req(set, req);
3214 ptlrpc_check_set(NULL, set);
3215 } else
3216 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
3217
0a3bdb00 3218 return 0;
d7e09d03
PT		 3219}
3220
3221
3222static int osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3223 struct obd_device *disk_obd, int *index)
3224{
3225 /* this code is not supposed to be used with LOD/OSP
3226 * to be removed soon */
3227 LBUG();
3228 return 0;
3229}
3230
3231static int osc_llog_finish(struct obd_device *obd, int count)
3232{
3233 struct llog_ctxt *ctxt;
3234
d7e09d03
PT		 3235 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
3236 if (ctxt) {
3237 llog_cat_close(NULL, ctxt->loc_handle);
3238 llog_cleanup(NULL, ctxt);
3239 }
3240
3241 ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
3242 if (ctxt)
3243 llog_cleanup(NULL, ctxt);
0a3bdb00 3244 return 0;
d7e09d03
PT		 3245}
3246
3247static int osc_reconnect(const struct lu_env *env,
3248 struct obd_export *exp, struct obd_device *obd,
3249 struct obd_uuid *cluuid,
3250 struct obd_connect_data *data,
3251 void *localdata)
3252{
3253 struct client_obd *cli = &obd->u.cli;
3254
3255 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
3256 long lost_grant;
3257
3258 client_obd_list_lock(&cli->cl_loi_list_lock);
3259 data->ocd_grant = (cli->cl_avail_grant + cli->cl_dirty) ?:
3260 2 * cli_brw_size(obd);
3261 lost_grant = cli->cl_lost_grant;
3262 cli->cl_lost_grant = 0;
3263 client_obd_list_unlock(&cli->cl_loi_list_lock);
3264
3265 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d"
3266 " ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
3267 data->ocd_version, data->ocd_grant, lost_grant);
3268 }
3269
0a3bdb00 3270 return 0;
d7e09d03
PT		 3271}
3272
3273static int osc_disconnect(struct obd_export *exp)
3274{
3275 struct obd_device *obd = class_exp2obd(exp);
3276 struct llog_ctxt *ctxt;
3277 int rc;
3278
3279 ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
3280 if (ctxt) {
3281 if (obd->u.cli.cl_conn_count == 1) {
3282 /* Flush any remaining cancel messages out to the
3283 * target */
3284 llog_sync(ctxt, exp, 0);
3285 }
3286 llog_ctxt_put(ctxt);
3287 } else {
3288 CDEBUG(D_HA, "No LLOG_SIZE_REPL_CTXT found in obd %p\n",
3289 obd);
3290 }
3291
3292 rc = client_disconnect_export(exp);
3293 /**
3294 * Initially we put del_shrink_grant before disconnect_export, but it
3295 * causes the following problem if setup (connect) and cleanup
3296 * (disconnect) are tangled together.
3297 * connect p1 disconnect p2
3298 * ptlrpc_connect_import
3299 * ............... class_manual_cleanup
3300 * osc_disconnect
3301 * del_shrink_grant
3302 * ptlrpc_connect_interrupt
3303 * init_grant_shrink
3304 * add this client to shrink list
3305 * cleanup_osc
3306 * Bang! pinger trigger the shrink.
3307 * So the osc should be disconnected from the shrink list, after we
3308 * are sure the import has been destroyed. BUG18662
3309 */
3310 if (obd->u.cli.cl_import == NULL)
3311 osc_del_shrink_grant(&obd->u.cli);
3312 return rc;
3313}
3314
3315static int osc_import_event(struct obd_device *obd,
3316 struct obd_import *imp,
3317 enum obd_import_event event)
3318{
3319 struct client_obd *cli;
3320 int rc = 0;
3321
d7e09d03
PT		 3322 LASSERT(imp->imp_obd == obd);
3323
3324 switch (event) {
3325 case IMP_EVENT_DISCON: {
3326 cli = &obd->u.cli;
3327 client_obd_list_lock(&cli->cl_loi_list_lock);
3328 cli->cl_avail_grant = 0;
3329 cli->cl_lost_grant = 0;
3330 client_obd_list_unlock(&cli->cl_loi_list_lock);
3331 break;
3332 }
3333 case IMP_EVENT_INACTIVE: {
3334 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
3335 break;
3336 }
3337 case IMP_EVENT_INVALIDATE: {
3338 struct ldlm_namespace *ns = obd->obd_namespace;
3339 struct lu_env *env;
3340 int refcheck;
3341
3342 env = cl_env_get(&refcheck);
3343 if (!IS_ERR(env)) {
3344 /* Reset grants */
3345 cli = &obd->u.cli;
3346 /* all pages go to failing rpcs due to the invalid
3347 * import */
3348 osc_io_unplug(env, cli, NULL, PDL_POLICY_ROUND);
3349
3350 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3351 cl_env_put(env, &refcheck);
3352 } else
3353 rc = PTR_ERR(env);
3354 break;
3355 }
3356 case IMP_EVENT_ACTIVE: {
3357 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
3358 break;
3359 }
3360 case IMP_EVENT_OCD: {
3361 struct obd_connect_data *ocd = &imp->imp_connect_data;
3362
3363 if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT)
3364 osc_init_grant(&obd->u.cli, ocd);
3365
3366 /* See bug 7198 */
3367 if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL)
3368 imp->imp_client->cli_request_portal =OST_REQUEST_PORTAL;
3369
3370 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
3371 break;
3372 }
3373 case IMP_EVENT_DEACTIVATE: {
3374 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE, NULL);
3375 break;
3376 }
3377 case IMP_EVENT_ACTIVATE: {
3378 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE, NULL);
3379 break;
3380 }
3381 default:
3382 CERROR("Unknown import event %d\n", event);
3383 LBUG();
3384 }
0a3bdb00 3385 return rc;
d7e09d03
PT		 3386}
3387
3388/**
3389 * Determine whether the lock can be canceled before replaying the lock
3390 * during recovery, see bug16774 for detailed information.
3391 *
3392 * \retval zero the lock can't be canceled
3393 * \retval other ok to cancel
3394 */
3395static int osc_cancel_for_recovery(struct ldlm_lock *lock)
3396{
3397 check_res_locked(lock->l_resource);
3398
3399 /*
3400 * Cancel all unused extent lock in granted mode LCK_PR or LCK_CR.
3401 *
3402 * XXX as a future improvement, we can also cancel unused write lock
3403 * if it doesn't have dirty data and active mmaps.
3404 */
3405 if (lock->l_resource->lr_type == LDLM_EXTENT &&
3406 (lock->l_granted_mode == LCK_PR ||
3407 lock->l_granted_mode == LCK_CR) &&
3408 (osc_dlm_lock_pageref(lock) == 0))
0a3bdb00 3409 return 1;
d7e09d03 3410
0a3bdb00 3411 return 0;
d7e09d03
PT		 3412}
3413
3414static int brw_queue_work(const struct lu_env *env, void *data)
3415{
3416 struct client_obd *cli = data;
3417
3418 CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli);
3419
3420 osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
0a3bdb00 3421 return 0;
d7e09d03
PT		 3422}
3423
3424int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
3425{
3426 struct lprocfs_static_vars lvars = { 0 };
3427 struct client_obd *cli = &obd->u.cli;
3428 void *handler;
3429 int rc;
d7e09d03
PT		 3430
3431 rc = ptlrpcd_addref();
3432 if (rc)
0a3bdb00 3433 return rc;
d7e09d03
PT		 3434
3435 rc = client_obd_setup(obd, lcfg);
3436 if (rc)
3437 GOTO(out_ptlrpcd, rc);
3438
3439 handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
3440 if (IS_ERR(handler))
3441 GOTO(out_client_setup, rc = PTR_ERR(handler));
3442 cli->cl_writeback_work = handler;
3443
3444 rc = osc_quota_setup(obd);
3445 if (rc)
3446 GOTO(out_ptlrpcd_work, rc);
3447
3448 cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
3449 lprocfs_osc_init_vars(&lvars);
3450 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
3451 lproc_osc_attach_seqstat(obd);
3452 sptlrpc_lprocfs_cliobd_attach(obd);
3453 ptlrpc_lprocfs_register_obd(obd);
3454 }
3455
3456 /* We need to allocate a few requests more, because
3457 * brw_interpret tries to create new requests before freeing
3458 * previous ones, Ideally we want to have 2x max_rpcs_in_flight
3459 * reserved, but I'm afraid that might be too much wasted RAM
3460 * in fact, so 2 is just my guess and still should work. */
3461 cli->cl_import->imp_rq_pool =
3462 ptlrpc_init_rq_pool(cli->cl_max_rpcs_in_flight + 2,
3463 OST_MAXREQSIZE,
3464 ptlrpc_add_rqs_to_pool);
3465
3466 INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
3467 ns_register_cancel(obd->obd_namespace, osc_cancel_for_recovery);
0a3bdb00 3468 return rc;
d7e09d03
PT		 3469
3470out_ptlrpcd_work:
3471 ptlrpcd_destroy_work(handler);
3472out_client_setup:
3473 client_obd_cleanup(obd);
3474out_ptlrpcd:
3475 ptlrpcd_decref();
0a3bdb00 3476 return rc;
d7e09d03
PT		 3477}
3478
3479static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3480{
3481 int rc = 0;
d7e09d03
PT		 3482
3483 switch (stage) {
3484 case OBD_CLEANUP_EARLY: {
3485 struct obd_import *imp;
3486 imp = obd->u.cli.cl_import;
3487 CDEBUG(D_HA, "Deactivating import %s\n", obd->obd_name);
3488 /* ptlrpc_abort_inflight to stop an mds_lov_synchronize */
3489 ptlrpc_deactivate_import(imp);
3490 spin_lock(&imp->imp_lock);
3491 imp->imp_pingable = 0;
3492 spin_unlock(&imp->imp_lock);
3493 break;
3494 }
3495 case OBD_CLEANUP_EXPORTS: {
3496 struct client_obd *cli = &obd->u.cli;
3497 /* LU-464
3498 * for echo client, export may be on zombie list, wait for
3499 * zombie thread to cull it, because cli.cl_import will be
3500 * cleared in client_disconnect_export():
3501 * class_export_destroy() -> obd_cleanup() ->
3502 * echo_device_free() -> echo_client_cleanup() ->
3503 * obd_disconnect() -> osc_disconnect() ->
3504 * client_disconnect_export()
3505 */
3506 obd_zombie_barrier();
3507 if (cli->cl_writeback_work) {
3508 ptlrpcd_destroy_work(cli->cl_writeback_work);
3509 cli->cl_writeback_work = NULL;
3510 }
3511 obd_cleanup_client_import(obd);
3512 ptlrpc_lprocfs_unregister_obd(obd);
3513 lprocfs_obd_cleanup(obd);
3514 rc = obd_llog_finish(obd, 0);
3515 if (rc != 0)
3516 CERROR("failed to cleanup llogging subsystems\n");
3517 break;
3518 }
3519 }
0a3bdb00 3520 return rc;
d7e09d03
PT		 3521}
3522
3523int osc_cleanup(struct obd_device *obd)
3524{
3525 struct client_obd *cli = &obd->u.cli;
3526 int rc;
3527
d7e09d03
PT		 3528 /* lru cleanup */
3529 if (cli->cl_cache != NULL) {
3530 LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
3531 spin_lock(&cli->cl_cache->ccc_lru_lock);
3532 list_del_init(&cli->cl_lru_osc);
3533 spin_unlock(&cli->cl_cache->ccc_lru_lock);
3534 cli->cl_lru_left = NULL;
3535 atomic_dec(&cli->cl_cache->ccc_users);
3536 cli->cl_cache = NULL;
3537 }
3538
3539 /* free memory of osc quota cache */
3540 osc_quota_cleanup(obd);
3541
3542 rc = client_obd_cleanup(obd);
3543
3544 ptlrpcd_decref();
0a3bdb00 3545 return rc;
d7e09d03
PT		 3546}
3547
3548int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
3549{
3550 struct lprocfs_static_vars lvars = { 0 };
3551 int rc = 0;
3552
3553 lprocfs_osc_init_vars(&lvars);
3554
3555 switch (lcfg->lcfg_command) {
3556 default:
3557 rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars,
3558 lcfg, obd);
3559 if (rc > 0)
3560 rc = 0;
3561 break;
3562 }
3563
3564 return(rc);
3565}
3566
3567static int osc_process_config(struct obd_device *obd, obd_count len, void *buf)
3568{
3569 return osc_process_config_base(obd, buf);
3570}
3571
3572struct obd_ops osc_obd_ops = {
3573 .o_owner = THIS_MODULE,
3574 .o_setup = osc_setup,
3575 .o_precleanup = osc_precleanup,
3576 .o_cleanup = osc_cleanup,
3577 .o_add_conn = client_import_add_conn,
3578 .o_del_conn = client_import_del_conn,
3579 .o_connect = client_connect_import,
3580 .o_reconnect = osc_reconnect,
3581 .o_disconnect = osc_disconnect,
3582 .o_statfs = osc_statfs,
3583 .o_statfs_async = osc_statfs_async,
3584 .o_packmd = osc_packmd,
3585 .o_unpackmd = osc_unpackmd,
3586 .o_create = osc_create,
3587 .o_destroy = osc_destroy,
3588 .o_getattr = osc_getattr,
3589 .o_getattr_async = osc_getattr_async,
3590 .o_setattr = osc_setattr,
3591 .o_setattr_async = osc_setattr_async,
3592 .o_brw = osc_brw,
3593 .o_punch = osc_punch,
3594 .o_sync = osc_sync,
3595 .o_enqueue = osc_enqueue,
3596 .o_change_cbdata = osc_change_cbdata,
3597 .o_find_cbdata = osc_find_cbdata,
3598 .o_cancel = osc_cancel,
3599 .o_cancel_unused = osc_cancel_unused,
3600 .o_iocontrol = osc_iocontrol,
3601 .o_get_info = osc_get_info,
3602 .o_set_info_async = osc_set_info_async,
3603 .o_import_event = osc_import_event,
3604 .o_llog_init = osc_llog_init,
3605 .o_llog_finish = osc_llog_finish,
3606 .o_process_config = osc_process_config,
3607 .o_quotactl = osc_quotactl,
3608 .o_quotacheck = osc_quotacheck,
3609};
3610
3611extern struct lu_kmem_descr osc_caches[];
3612extern spinlock_t osc_ast_guard;
3613extern struct lock_class_key osc_ast_guard_class;
3614
3615int __init osc_init(void)
3616{
3617 struct lprocfs_static_vars lvars = { 0 };
3618 int rc;
d7e09d03
PT		 3619
3620 /* print an address of _any_ initialized kernel symbol from this
3621 * module, to allow debugging with gdb that doesn't support data
3622 * symbols from modules.*/
3623 CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
3624
3625 rc = lu_kmem_init(osc_caches);
a55e0f44 3626 if (rc)
0a3bdb00 3627 return rc;
d7e09d03
PT		 3628
3629 lprocfs_osc_init_vars(&lvars);
3630
3631 rc = class_register_type(&osc_obd_ops, NULL, lvars.module_vars,
3632 LUSTRE_OSC_NAME, &osc_device_type);
3633 if (rc) {
3634 lu_kmem_fini(osc_caches);
0a3bdb00 3635 return rc;
d7e09d03
PT		 3636 }
3637
3638 spin_lock_init(&osc_ast_guard);
3639 lockdep_set_class(&osc_ast_guard, &osc_ast_guard_class);
3640
0a3bdb00 3641 return rc;
d7e09d03
PT		 3642}
3643
3644static void /*__exit*/ osc_exit(void)
3645{
3646 class_unregister_type(LUSTRE_OSC_NAME);
3647 lu_kmem_fini(osc_caches);
3648}
3649
3650MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3651MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)");
3652MODULE_LICENSE("GPL");
6960736c 3653MODULE_VERSION(LUSTRE_VERSION_STRING);
d7e09d03 3654
6960736c
GKH		 3655module_init(osc_init);
3656module_exit(osc_exit);
Linux kernel - Deliverables
This page took 0.359072 seconds and 5 git commands to generate.