Whamcloud - gitweb
17f088c3114124057cbbb5d41e78e7fe26ed4621
[fs/lustre-release.git] / lustre / osc / osc_request.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 only,
10  * as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful, but
13  * WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * General Public License version 2 for more details (a copy is included
16  * in the LICENSE file that accompanied this code).
17  *
18  * You should have received a copy of the GNU General Public License
19  * version 2 along with this program; If not, see
20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
21  *
22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23  * CA 95054 USA or visit www.sun.com if you need additional information or
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30  * Use is subject to license terms.
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 #ifndef EXPORT_SYMTAB
38 # define EXPORT_SYMTAB
39 #endif
40 #define DEBUG_SUBSYSTEM S_OSC
41
42 #include <libcfs/libcfs.h>
43
44 #ifndef __KERNEL__
45 # include <liblustre.h>
46 #endif
47
48 #include <lustre_dlm.h>
49 #include <lustre_net.h>
50 #include <lustre/lustre_user.h>
51 #include <obd_cksum.h>
52 #include <obd_ost.h>
53 #include <obd_lov.h>
54
55 #ifdef  __CYGWIN__
56 # include <ctype.h>
57 #endif
58
59 #include <lustre_ha.h>
60 #include <lprocfs_status.h>
61 #include <lustre_log.h>
62 #include <lustre_debug.h>
63 #include <lustre_param.h>
64 #include "osc_internal.h"
65
66 static quota_interface_t *quota_interface = NULL;
67 extern quota_interface_t osc_quota_interface;
68
69 static void osc_release_ppga(struct brw_page **ppga, obd_count count);
70 static int brw_interpret(const struct lu_env *env,
71                          struct ptlrpc_request *req, void *data, int rc);
72 int osc_cleanup(struct obd_device *obd);
73
74 /* Pack OSC object metadata for disk storage (LE byte order). */
75 static int osc_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
76                       struct lov_stripe_md *lsm)
77 {
78         int lmm_size;
79         ENTRY;
80
81         lmm_size = sizeof(**lmmp);
82         if (!lmmp)
83                 RETURN(lmm_size);
84
85         if (*lmmp && !lsm) {
86                 OBD_FREE(*lmmp, lmm_size);
87                 *lmmp = NULL;
88                 RETURN(0);
89         }
90
91         if (!*lmmp) {
92                 OBD_ALLOC(*lmmp, lmm_size);
93                 if (!*lmmp)
94                         RETURN(-ENOMEM);
95         }
96
97         if (lsm) {
98                 LASSERT(lsm->lsm_object_id);
99                 LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
100                 (*lmmp)->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
101                 (*lmmp)->lmm_object_seq = cpu_to_le64(lsm->lsm_object_seq);
102         }
103
104         RETURN(lmm_size);
105 }
106
107 /* Unpack OSC object metadata from disk storage (LE byte order). */
108 static int osc_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
109                         struct lov_mds_md *lmm, int lmm_bytes)
110 {
111         int lsm_size;
112         ENTRY;
113
114         if (lmm != NULL) {
115                 if (lmm_bytes < sizeof (*lmm)) {
116                         CERROR("lov_mds_md too small: %d, need %d\n",
117                                lmm_bytes, (int)sizeof(*lmm));
118                         RETURN(-EINVAL);
119                 }
120                 /* XXX LOV_MAGIC etc check? */
121
122                 if (lmm->lmm_object_id == 0) {
123                         CERROR("lov_mds_md: zero lmm_object_id\n");
124                         RETURN(-EINVAL);
125                 }
126         }
127
128         lsm_size = lov_stripe_md_size(1);
129         if (lsmp == NULL)
130                 RETURN(lsm_size);
131
132         if (*lsmp != NULL && lmm == NULL) {
133                 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
134                 OBD_FREE(*lsmp, lsm_size);
135                 *lsmp = NULL;
136                 RETURN(0);
137         }
138
139         if (*lsmp == NULL) {
140                 OBD_ALLOC(*lsmp, lsm_size);
141                 if (*lsmp == NULL)
142                         RETURN(-ENOMEM);
143                 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
144                 if ((*lsmp)->lsm_oinfo[0] == NULL) {
145                         OBD_FREE(*lsmp, lsm_size);
146                         RETURN(-ENOMEM);
147                 }
148                 loi_init((*lsmp)->lsm_oinfo[0]);
149         }
150
151         if (lmm != NULL) {
152                 /* XXX zero *lsmp? */
153                 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
154                 (*lsmp)->lsm_object_seq = le64_to_cpu (lmm->lmm_object_seq);
155                 LASSERT((*lsmp)->lsm_object_id);
156                 LASSERT_SEQ_IS_MDT((*lsmp)->lsm_object_seq);
157         }
158
159         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
160
161         RETURN(lsm_size);
162 }
163
164 static inline void osc_pack_capa(struct ptlrpc_request *req,
165                                  struct ost_body *body, void *capa)
166 {
167         struct obd_capa *oc = (struct obd_capa *)capa;
168         struct lustre_capa *c;
169
170         if (!capa)
171                 return;
172
173         c = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
174         LASSERT(c);
175         capa_cpy(c, oc);
176         body->oa.o_valid |= OBD_MD_FLOSSCAPA;
177         DEBUG_CAPA(D_SEC, c, "pack");
178 }
179
180 static inline void osc_pack_req_body(struct ptlrpc_request *req,
181                                      struct obd_info *oinfo)
182 {
183         struct ost_body *body;
184
185         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
186         LASSERT(body);
187
188         lustre_set_wire_obdo(&body->oa, oinfo->oi_oa);
189         osc_pack_capa(req, body, oinfo->oi_capa);
190 }
191
192 static inline void osc_set_capa_size(struct ptlrpc_request *req,
193                                      const struct req_msg_field *field,
194                                      struct obd_capa *oc)
195 {
196         if (oc == NULL)
197                 req_capsule_set_size(&req->rq_pill, field, RCL_CLIENT, 0);
198         else
199                 /* it is already calculated as sizeof struct obd_capa */
200                 ;
201 }
202
203 static int osc_getattr_interpret(const struct lu_env *env,
204                                  struct ptlrpc_request *req,
205                                  struct osc_async_args *aa, int rc)
206 {
207         struct ost_body *body;
208         ENTRY;
209
210         if (rc != 0)
211                 GOTO(out, rc);
212
213         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
214         if (body) {
215                 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
216                 lustre_get_wire_obdo(aa->aa_oi->oi_oa, &body->oa);
217
218                 /* This should really be sent by the OST */
219                 aa->aa_oi->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
220                 aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
221         } else {
222                 CDEBUG(D_INFO, "can't unpack ost_body\n");
223                 rc = -EPROTO;
224                 aa->aa_oi->oi_oa->o_valid = 0;
225         }
226 out:
227         rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
228         RETURN(rc);
229 }
230
231 static int osc_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
232                              struct ptlrpc_request_set *set)
233 {
234         struct ptlrpc_request *req;
235         struct osc_async_args *aa;
236         int                    rc;
237         ENTRY;
238
239         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
240         if (req == NULL)
241                 RETURN(-ENOMEM);
242
243         osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
244         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
245         if (rc) {
246                 ptlrpc_request_free(req);
247                 RETURN(rc);
248         }
249
250         osc_pack_req_body(req, oinfo);
251
252         ptlrpc_request_set_replen(req);
253         req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_getattr_interpret;
254
255         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
256         aa = ptlrpc_req_async_args(req);
257         aa->aa_oi = oinfo;
258
259         ptlrpc_set_add_req(set, req);
260         RETURN(0);
261 }
262
263 static int osc_getattr(struct obd_export *exp, struct obd_info *oinfo)
264 {
265         struct ptlrpc_request *req;
266         struct ost_body       *body;
267         int                    rc;
268         ENTRY;
269
270         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR);
271         if (req == NULL)
272                 RETURN(-ENOMEM);
273
274         osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
275         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR);
276         if (rc) {
277                 ptlrpc_request_free(req);
278                 RETURN(rc);
279         }
280
281         osc_pack_req_body(req, oinfo);
282
283         ptlrpc_request_set_replen(req);
284
285         rc = ptlrpc_queue_wait(req);
286         if (rc)
287                 GOTO(out, rc);
288
289         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
290         if (body == NULL)
291                 GOTO(out, rc = -EPROTO);
292
293         CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
294         lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
295
296         /* This should really be sent by the OST */
297         oinfo->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
298         oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
299
300         EXIT;
301  out:
302         ptlrpc_req_finished(req);
303         return rc;
304 }
305
306 static int osc_setattr(struct obd_export *exp, struct obd_info *oinfo,
307                        struct obd_trans_info *oti)
308 {
309         struct ptlrpc_request *req;
310         struct ost_body       *body;
311         int                    rc;
312         ENTRY;
313
314         LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP);
315
316         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
317         if (req == NULL)
318                 RETURN(-ENOMEM);
319
320         osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
321         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
322         if (rc) {
323                 ptlrpc_request_free(req);
324                 RETURN(rc);
325         }
326
327         osc_pack_req_body(req, oinfo);
328
329         ptlrpc_request_set_replen(req);
330
331         rc = ptlrpc_queue_wait(req);
332         if (rc)
333                 GOTO(out, rc);
334
335         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
336         if (body == NULL)
337                 GOTO(out, rc = -EPROTO);
338
339         lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
340
341         EXIT;
342 out:
343         ptlrpc_req_finished(req);
344         RETURN(rc);
345 }
346
347 static int osc_setattr_interpret(const struct lu_env *env,
348                                  struct ptlrpc_request *req,
349                                  struct osc_setattr_args *sa, int rc)
350 {
351         struct ost_body *body;
352         ENTRY;
353
354         if (rc != 0)
355                 GOTO(out, rc);
356
357         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
358         if (body == NULL)
359                 GOTO(out, rc = -EPROTO);
360
361         lustre_get_wire_obdo(sa->sa_oa, &body->oa);
362 out:
363         rc = sa->sa_upcall(sa->sa_cookie, rc);
364         RETURN(rc);
365 }
366
367 int osc_setattr_async_base(struct obd_export *exp, struct obd_info *oinfo,
368                            struct obd_trans_info *oti,
369                            obd_enqueue_update_f upcall, void *cookie,
370                            struct ptlrpc_request_set *rqset)
371 {
372         struct ptlrpc_request   *req;
373         struct osc_setattr_args *sa;
374         int                      rc;
375         ENTRY;
376
377         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR);
378         if (req == NULL)
379                 RETURN(-ENOMEM);
380
381         osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
382         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR);
383         if (rc) {
384                 ptlrpc_request_free(req);
385                 RETURN(rc);
386         }
387
388         if (oti && oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
389                 oinfo->oi_oa->o_lcookie = *oti->oti_logcookies;
390
391         osc_pack_req_body(req, oinfo);
392
393         ptlrpc_request_set_replen(req);
394
395         /* do mds to ost setattr asynchronously */
396         if (!rqset) {
397                 /* Do not wait for response. */
398                 ptlrpcd_add_req(req, PSCOPE_OTHER);
399         } else {
400                 req->rq_interpret_reply =
401                         (ptlrpc_interpterer_t)osc_setattr_interpret;
402
403                 CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
404                 sa = ptlrpc_req_async_args(req);
405                 sa->sa_oa = oinfo->oi_oa;
406                 sa->sa_upcall = upcall;
407                 sa->sa_cookie = cookie;
408
409                 if (rqset == PTLRPCD_SET)
410                         ptlrpcd_add_req(req, PSCOPE_OTHER);
411                 else
412                         ptlrpc_set_add_req(rqset, req);
413         }
414
415         RETURN(0);
416 }
417
418 static int osc_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
419                              struct obd_trans_info *oti,
420                              struct ptlrpc_request_set *rqset)
421 {
422         return osc_setattr_async_base(exp, oinfo, oti,
423                                       oinfo->oi_cb_up, oinfo, rqset);
424 }
425
426 int osc_real_create(struct obd_export *exp, struct obdo *oa,
427                     struct lov_stripe_md **ea, struct obd_trans_info *oti)
428 {
429         struct ptlrpc_request *req;
430         struct ost_body       *body;
431         struct lov_stripe_md  *lsm;
432         int                    rc;
433         ENTRY;
434
435         LASSERT(oa);
436         LASSERT(ea);
437
438         lsm = *ea;
439         if (!lsm) {
440                 rc = obd_alloc_memmd(exp, &lsm);
441                 if (rc < 0)
442                         RETURN(rc);
443         }
444
445         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
446         if (req == NULL)
447                 GOTO(out, rc = -ENOMEM);
448
449         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE);
450         if (rc) {
451                 ptlrpc_request_free(req);
452                 GOTO(out, rc);
453         }
454
455         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
456         LASSERT(body);
457         lustre_set_wire_obdo(&body->oa, oa);
458
459         ptlrpc_request_set_replen(req);
460
461         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
462             oa->o_flags == OBD_FL_DELORPHAN) {
463                 DEBUG_REQ(D_HA, req,
464                           "delorphan from OST integration");
465                 /* Don't resend the delorphan req */
466                 req->rq_no_resend = req->rq_no_delay = 1;
467         }
468
469         rc = ptlrpc_queue_wait(req);
470         if (rc)
471                 GOTO(out_req, rc);
472
473         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
474         if (body == NULL)
475                 GOTO(out_req, rc = -EPROTO);
476
477         lustre_get_wire_obdo(oa, &body->oa);
478
479         /* This should really be sent by the OST */
480         oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
481         oa->o_valid |= OBD_MD_FLBLKSZ;
482
483         /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
484          * have valid lsm_oinfo data structs, so don't go touching that.
485          * This needs to be fixed in a big way.
486          */
487         lsm->lsm_object_id = oa->o_id;
488         lsm->lsm_object_seq = oa->o_seq;
489         *ea = lsm;
490
491         if (oti != NULL) {
492                 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
493
494                 if (oa->o_valid & OBD_MD_FLCOOKIE) {
495                         if (!oti->oti_logcookies)
496                                 oti_alloc_cookies(oti, 1);
497                         *oti->oti_logcookies = oa->o_lcookie;
498                 }
499         }
500
501         CDEBUG(D_HA, "transno: "LPD64"\n",
502                lustre_msg_get_transno(req->rq_repmsg));
503 out_req:
504         ptlrpc_req_finished(req);
505 out:
506         if (rc && !*ea)
507                 obd_free_memmd(exp, &lsm);
508         RETURN(rc);
509 }
510
511 int osc_punch_base(struct obd_export *exp, struct obd_info *oinfo,
512                    obd_enqueue_update_f upcall, void *cookie,
513                    struct ptlrpc_request_set *rqset)
514 {
515         struct ptlrpc_request   *req;
516         struct osc_setattr_args *sa;
517         struct ost_body         *body;
518         int                      rc;
519         ENTRY;
520
521         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_PUNCH);
522         if (req == NULL)
523                 RETURN(-ENOMEM);
524
525         osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa);
526         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH);
527         if (rc) {
528                 ptlrpc_request_free(req);
529                 RETURN(rc);
530         }
531         req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
532         ptlrpc_at_set_req_timeout(req);
533
534         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
535         LASSERT(body);
536         lustre_set_wire_obdo(&body->oa, oinfo->oi_oa);
537         osc_pack_capa(req, body, oinfo->oi_capa);
538
539         ptlrpc_request_set_replen(req);
540
541
542         req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
543         CLASSERT (sizeof(*sa) <= sizeof(req->rq_async_args));
544         sa = ptlrpc_req_async_args(req);
545         sa->sa_oa     = oinfo->oi_oa;
546         sa->sa_upcall = upcall;
547         sa->sa_cookie = cookie;
548         if (rqset == PTLRPCD_SET)
549                 ptlrpcd_add_req(req, PSCOPE_OTHER);
550         else
551                 ptlrpc_set_add_req(rqset, req);
552
553         RETURN(0);
554 }
555
556 static int osc_punch(struct obd_export *exp, struct obd_info *oinfo,
557                      struct obd_trans_info *oti,
558                      struct ptlrpc_request_set *rqset)
559 {
560         oinfo->oi_oa->o_size   = oinfo->oi_policy.l_extent.start;
561         oinfo->oi_oa->o_blocks = oinfo->oi_policy.l_extent.end;
562         oinfo->oi_oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
563         return osc_punch_base(exp, oinfo,
564                               oinfo->oi_cb_up, oinfo, rqset);
565 }
566
567 static int osc_sync(struct obd_export *exp, struct obdo *oa,
568                     struct lov_stripe_md *md, obd_size start, obd_size end,
569                     void *capa)
570 {
571         struct ptlrpc_request *req;
572         struct ost_body       *body;
573         int                    rc;
574         ENTRY;
575
576         if (!oa) {
577                 CDEBUG(D_INFO, "oa NULL\n");
578                 RETURN(-EINVAL);
579         }
580
581         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC);
582         if (req == NULL)
583                 RETURN(-ENOMEM);
584
585         osc_set_capa_size(req, &RMF_CAPA1, capa);
586         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC);
587         if (rc) {
588                 ptlrpc_request_free(req);
589                 RETURN(rc);
590         }
591
592         /* overload the size and blocks fields in the oa with start/end */
593         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
594         LASSERT(body);
595         lustre_set_wire_obdo(&body->oa, oa);
596         body->oa.o_size = start;
597         body->oa.o_blocks = end;
598         body->oa.o_valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
599         osc_pack_capa(req, body, capa);
600
601         ptlrpc_request_set_replen(req);
602
603         rc = ptlrpc_queue_wait(req);
604         if (rc)
605                 GOTO(out, rc);
606
607         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
608         if (body == NULL)
609                 GOTO(out, rc = -EPROTO);
610
611         lustre_get_wire_obdo(oa, &body->oa);
612
613         EXIT;
614  out:
615         ptlrpc_req_finished(req);
616         return rc;
617 }
618
619 /* Find and cancel locally locks matched by @mode in the resource found by
620  * @objid. Found locks are added into @cancel list. Returns the amount of
621  * locks added to @cancels list. */
622 static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
623                                    cfs_list_t *cancels,
624                                    ldlm_mode_t mode, int lock_flags)
625 {
626         struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
627         struct ldlm_res_id res_id;
628         struct ldlm_resource *res;
629         int count;
630         ENTRY;
631
632         osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
633         res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
634         if (res == NULL)
635                 RETURN(0);
636
637         LDLM_RESOURCE_ADDREF(res);
638         count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
639                                            lock_flags, 0, NULL);
640         LDLM_RESOURCE_DELREF(res);
641         ldlm_resource_putref(res);
642         RETURN(count);
643 }
644
645 static int osc_destroy_interpret(const struct lu_env *env,
646                                  struct ptlrpc_request *req, void *data,
647                                  int rc)
648 {
649         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
650
651         cfs_atomic_dec(&cli->cl_destroy_in_flight);
652         cfs_waitq_signal(&cli->cl_destroy_waitq);
653         return 0;
654 }
655
656 static int osc_can_send_destroy(struct client_obd *cli)
657 {
658         if (cfs_atomic_inc_return(&cli->cl_destroy_in_flight) <=
659             cli->cl_max_rpcs_in_flight) {
660                 /* The destroy request can be sent */
661                 return 1;
662         }
663         if (cfs_atomic_dec_return(&cli->cl_destroy_in_flight) <
664             cli->cl_max_rpcs_in_flight) {
665                 /*
666                  * The counter has been modified between the two atomic
667                  * operations.
668                  */
669                 cfs_waitq_signal(&cli->cl_destroy_waitq);
670         }
671         return 0;
672 }
673
674 /* Destroy requests can be async always on the client, and we don't even really
675  * care about the return code since the client cannot do anything at all about
676  * a destroy failure.
677  * When the MDS is unlinking a filename, it saves the file objects into a
678  * recovery llog, and these object records are cancelled when the OST reports
679  * they were destroyed and sync'd to disk (i.e. transaction committed).
680  * If the client dies, or the OST is down when the object should be destroyed,
681  * the records are not cancelled, and when the OST reconnects to the MDS next,
682  * it will retrieve the llog unlink logs and then sends the log cancellation
683  * cookies to the MDS after committing destroy transactions. */
684 static int osc_destroy(struct obd_export *exp, struct obdo *oa,
685                        struct lov_stripe_md *ea, struct obd_trans_info *oti,
686                        struct obd_export *md_export, void *capa)
687 {
688         struct client_obd     *cli = &exp->exp_obd->u.cli;
689         struct ptlrpc_request *req;
690         struct ost_body       *body;
691         CFS_LIST_HEAD(cancels);
692         int rc, count;
693         ENTRY;
694
695         if (!oa) {
696                 CDEBUG(D_INFO, "oa NULL\n");
697                 RETURN(-EINVAL);
698         }
699
700         count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW,
701                                         LDLM_FL_DISCARD_DATA);
702
703         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY);
704         if (req == NULL) {
705                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
706                 RETURN(-ENOMEM);
707         }
708
709         osc_set_capa_size(req, &RMF_CAPA1, (struct obd_capa *)capa);
710         rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
711                                0, &cancels, count);
712         if (rc) {
713                 ptlrpc_request_free(req);
714                 RETURN(rc);
715         }
716
717         req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
718         ptlrpc_at_set_req_timeout(req);
719
720         if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
721                 oa->o_lcookie = *oti->oti_logcookies;
722         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
723         LASSERT(body);
724         lustre_set_wire_obdo(&body->oa, oa);
725
726         osc_pack_capa(req, body, (struct obd_capa *)capa);
727         ptlrpc_request_set_replen(req);
728
729         /* don't throttle destroy RPCs for the MDT */
730         if (!(cli->cl_import->imp_connect_flags_orig & OBD_CONNECT_MDS)) {
731                 req->rq_interpret_reply = osc_destroy_interpret;
732                 if (!osc_can_send_destroy(cli)) {
733                         struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP,
734                                                           NULL);
735
736                         /*
737                          * Wait until the number of on-going destroy RPCs drops
738                          * under max_rpc_in_flight
739                          */
740                         l_wait_event_exclusive(cli->cl_destroy_waitq,
741                                                osc_can_send_destroy(cli), &lwi);
742                 }
743         }
744
745         /* Do not wait for response */
746         ptlrpcd_add_req(req, PSCOPE_OTHER);
747         RETURN(0);
748 }
749
750 static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
751                                 long writing_bytes)
752 {
753         obd_flag bits = OBD_MD_FLBLOCKS|OBD_MD_FLGRANT;
754
755         LASSERT(!(oa->o_valid & bits));
756
757         oa->o_valid |= bits;
758         client_obd_list_lock(&cli->cl_loi_list_lock);
759         oa->o_dirty = cli->cl_dirty;
760         if (cli->cl_dirty - cli->cl_dirty_transit > cli->cl_dirty_max) {
761                 CERROR("dirty %lu - %lu > dirty_max %lu\n",
762                        cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
763                 oa->o_undirty = 0;
764         } else if (cfs_atomic_read(&obd_dirty_pages) -
765                    cfs_atomic_read(&obd_dirty_transit_pages) >
766                    obd_max_dirty_pages + 1){
767                 /* The cfs_atomic_read() allowing the cfs_atomic_inc() are
768                  * not covered by a lock thus they may safely race and trip
769                  * this CERROR() unless we add in a small fudge factor (+1). */
770                 CERROR("dirty %d - %d > system dirty_max %d\n",
771                        cfs_atomic_read(&obd_dirty_pages),
772                        cfs_atomic_read(&obd_dirty_transit_pages),
773                        obd_max_dirty_pages);
774                 oa->o_undirty = 0;
775         } else if (cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff) {
776                 CERROR("dirty %lu - dirty_max %lu too big???\n",
777                        cli->cl_dirty, cli->cl_dirty_max);
778                 oa->o_undirty = 0;
779         } else {
780                 long max_in_flight = (cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT)*
781                                 (cli->cl_max_rpcs_in_flight + 1);
782                 oa->o_undirty = max(cli->cl_dirty_max, max_in_flight);
783         }
784         oa->o_grant = cli->cl_avail_grant;
785         oa->o_dropped = cli->cl_lost_grant;
786         cli->cl_lost_grant = 0;
787         client_obd_list_unlock(&cli->cl_loi_list_lock);
788         CDEBUG(D_CACHE,"dirty: "LPU64" undirty: %u dropped %u grant: "LPU64"\n",
789                oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
790
791 }
792
793 static void osc_update_next_shrink(struct client_obd *cli)
794 {
795         cli->cl_next_shrink_grant =
796                 cfs_time_shift(cli->cl_grant_shrink_interval);
797         CDEBUG(D_CACHE, "next time %ld to shrink grant \n",
798                cli->cl_next_shrink_grant);
799 }
800
801 /* caller must hold loi_list_lock */
802 static void osc_consume_write_grant(struct client_obd *cli,
803                                     struct brw_page *pga)
804 {
805         LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
806         LASSERT(!(pga->flag & OBD_BRW_FROM_GRANT));
807         cfs_atomic_inc(&obd_dirty_pages);
808         cli->cl_dirty += CFS_PAGE_SIZE;
809         cli->cl_avail_grant -= CFS_PAGE_SIZE;
810         pga->flag |= OBD_BRW_FROM_GRANT;
811         CDEBUG(D_CACHE, "using %lu grant credits for brw %p page %p\n",
812                CFS_PAGE_SIZE, pga, pga->pg);
813         LASSERT(cli->cl_avail_grant >= 0);
814         osc_update_next_shrink(cli);
815 }
816
817 /* the companion to osc_consume_write_grant, called when a brw has completed.
818  * must be called with the loi lock held. */
819 static void osc_release_write_grant(struct client_obd *cli,
820                                     struct brw_page *pga, int sent)
821 {
822         int blocksize = cli->cl_import->imp_obd->obd_osfs.os_bsize ? : 4096;
823         ENTRY;
824
825         LASSERT_SPIN_LOCKED(&cli->cl_loi_list_lock.lock);
826         if (!(pga->flag & OBD_BRW_FROM_GRANT)) {
827                 EXIT;
828                 return;
829         }
830
831         pga->flag &= ~OBD_BRW_FROM_GRANT;
832         cfs_atomic_dec(&obd_dirty_pages);
833         cli->cl_dirty -= CFS_PAGE_SIZE;
834         if (pga->flag & OBD_BRW_NOCACHE) {
835                 pga->flag &= ~OBD_BRW_NOCACHE;
836                 cfs_atomic_dec(&obd_dirty_transit_pages);
837                 cli->cl_dirty_transit -= CFS_PAGE_SIZE;
838         }
839         if (!sent) {
840                 cli->cl_lost_grant += CFS_PAGE_SIZE;
841                 CDEBUG(D_CACHE, "lost grant: %lu avail grant: %lu dirty: %lu\n",
842                        cli->cl_lost_grant, cli->cl_avail_grant, cli->cl_dirty);
843         } else if (CFS_PAGE_SIZE != blocksize && pga->count != CFS_PAGE_SIZE) {
844                 /* For short writes we shouldn't count parts of pages that
845                  * span a whole block on the OST side, or our accounting goes
846                  * wrong.  Should match the code in filter_grant_check. */
847                 int offset = pga->off & ~CFS_PAGE_MASK;
848                 int count = pga->count + (offset & (blocksize - 1));
849                 int end = (offset + pga->count) & (blocksize - 1);
850                 if (end)
851                         count += blocksize - end;
852
853                 cli->cl_lost_grant += CFS_PAGE_SIZE - count;
854                 CDEBUG(D_CACHE, "lost %lu grant: %lu avail: %lu dirty: %lu\n",
855                        CFS_PAGE_SIZE - count, cli->cl_lost_grant,
856                        cli->cl_avail_grant, cli->cl_dirty);
857         }
858
859         EXIT;
860 }
861
862 static unsigned long rpcs_in_flight(struct client_obd *cli)
863 {
864         return cli->cl_r_in_flight + cli->cl_w_in_flight;
865 }
866
867 int osc_wake_sync_fs(struct client_obd *cli)
868 {
869         int rc = 0;
870         ENTRY;
871         if (cfs_list_empty(&cli->cl_loi_sync_fs_list) &&
872             cli->cl_sf_wait.started) {
873                 cli->cl_sf_wait.sfw_upcall(cli->cl_sf_wait.sfw_oi, rc);
874                 cli->cl_sf_wait.started = 0;
875                 CDEBUG(D_CACHE, "sync_fs_loi list is empty\n");
876         }
877         RETURN(rc);
878 }
879
880 /* caller must hold loi_list_lock */
881 void osc_wake_cache_waiters(struct client_obd *cli)
882 {
883         cfs_list_t *l, *tmp;
884         struct osc_cache_waiter *ocw;
885
886         ENTRY;
887         cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
888                 /* if we can't dirty more, we must wait until some is written */
889                 if ((cli->cl_dirty + CFS_PAGE_SIZE > cli->cl_dirty_max) ||
890                    (cfs_atomic_read(&obd_dirty_pages) + 1 >
891                     obd_max_dirty_pages)) {
892                         CDEBUG(D_CACHE, "no dirty room: dirty: %ld "
893                                "osc max %ld, sys max %d\n", cli->cl_dirty,
894                                cli->cl_dirty_max, obd_max_dirty_pages);
895                         return;
896                 }
897
898                 /* if still dirty cache but no grant wait for pending RPCs that
899                  * may yet return us some grant before doing sync writes */
900                 if (cli->cl_w_in_flight && cli->cl_avail_grant < CFS_PAGE_SIZE) {
901                         CDEBUG(D_CACHE, "%u BRW writes in flight, no grant\n",
902                                cli->cl_w_in_flight);
903                         return;
904                 }
905
906                 ocw = cfs_list_entry(l, struct osc_cache_waiter, ocw_entry);
907                 cfs_list_del_init(&ocw->ocw_entry);
908                 if (cli->cl_avail_grant < CFS_PAGE_SIZE) {
909                         /* no more RPCs in flight to return grant, do sync IO */
910                         ocw->ocw_rc = -EDQUOT;
911                         CDEBUG(D_INODE, "wake oap %p for sync\n", ocw->ocw_oap);
912                 } else {
913                         osc_consume_write_grant(cli,
914                                                 &ocw->ocw_oap->oap_brw_page);
915                 }
916
917                 cfs_waitq_signal(&ocw->ocw_waitq);
918         }
919
920         EXIT;
921 }
922
923 static void __osc_update_grant(struct client_obd *cli, obd_size grant)
924 {
925         client_obd_list_lock(&cli->cl_loi_list_lock);
926         cli->cl_avail_grant += grant;
927         client_obd_list_unlock(&cli->cl_loi_list_lock);
928 }
929
930 static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
931 {
932         if (body->oa.o_valid & OBD_MD_FLGRANT) {
933                 CDEBUG(D_CACHE, "got "LPU64" extra grant\n", body->oa.o_grant);
934                 __osc_update_grant(cli, body->oa.o_grant);
935         }
936 }
937
938 static int osc_set_info_async(struct obd_export *exp, obd_count keylen,
939                               void *key, obd_count vallen, void *val,
940                               struct ptlrpc_request_set *set);
941
942 static int osc_shrink_grant_interpret(const struct lu_env *env,
943                                       struct ptlrpc_request *req,
944                                       void *aa, int rc)
945 {
946         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
947         struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
948         struct ost_body *body;
949
950         if (rc != 0) {
951                 __osc_update_grant(cli, oa->o_grant);
952                 GOTO(out, rc);
953         }
954
955         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
956         LASSERT(body);
957         osc_update_grant(cli, body);
958 out:
959         OBDO_FREE(oa);
960         return rc;
961 }
962
963 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
964 {
965         client_obd_list_lock(&cli->cl_loi_list_lock);
966         oa->o_grant = cli->cl_avail_grant / 4;
967         cli->cl_avail_grant -= oa->o_grant;
968         client_obd_list_unlock(&cli->cl_loi_list_lock);
969         if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
970                 oa->o_valid |= OBD_MD_FLFLAGS;
971                 oa->o_flags = 0;
972         }
973         oa->o_flags |= OBD_FL_SHRINK_GRANT;
974         osc_update_next_shrink(cli);
975 }
976
977 /* Shrink the current grant, either from some large amount to enough for a
978  * full set of in-flight RPCs, or if we have already shrunk to that limit
979  * then to enough for a single RPC.  This avoids keeping more grant than
980  * needed, and avoids shrinking the grant piecemeal. */
981 static int osc_shrink_grant(struct client_obd *cli)
982 {
983         long target = (cli->cl_max_rpcs_in_flight + 1) *
984                       cli->cl_max_pages_per_rpc;
985
986         client_obd_list_lock(&cli->cl_loi_list_lock);
987         if (cli->cl_avail_grant <= target)
988                 target = cli->cl_max_pages_per_rpc;
989         client_obd_list_unlock(&cli->cl_loi_list_lock);
990
991         return osc_shrink_grant_to_target(cli, target);
992 }
993
994 int osc_shrink_grant_to_target(struct client_obd *cli, long target)
995 {
996         int    rc = 0;
997         struct ost_body     *body;
998         ENTRY;
999
1000         client_obd_list_lock(&cli->cl_loi_list_lock);
1001         /* Don't shrink if we are already above or below the desired limit
1002          * We don't want to shrink below a single RPC, as that will negatively
1003          * impact block allocation and long-term performance. */
1004         if (target < cli->cl_max_pages_per_rpc)
1005                 target = cli->cl_max_pages_per_rpc;
1006
1007         if (target >= cli->cl_avail_grant) {
1008                 client_obd_list_unlock(&cli->cl_loi_list_lock);
1009                 RETURN(0);
1010         }
1011         client_obd_list_unlock(&cli->cl_loi_list_lock);
1012
1013         OBD_ALLOC_PTR(body);
1014         if (!body)
1015                 RETURN(-ENOMEM);
1016
1017         osc_announce_cached(cli, &body->oa, 0);
1018
1019         client_obd_list_lock(&cli->cl_loi_list_lock);
1020         body->oa.o_grant = cli->cl_avail_grant - target;
1021         cli->cl_avail_grant = target;
1022         client_obd_list_unlock(&cli->cl_loi_list_lock);
1023         if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
1024                 body->oa.o_valid |= OBD_MD_FLFLAGS;
1025                 body->oa.o_flags = 0;
1026         }
1027         body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
1028         osc_update_next_shrink(cli);
1029
1030         rc = osc_set_info_async(cli->cl_import->imp_obd->obd_self_export,
1031                                 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
1032                                 sizeof(*body), body, NULL);
1033         if (rc != 0)
1034                 __osc_update_grant(cli, body->oa.o_grant);
1035         OBD_FREE_PTR(body);
1036         RETURN(rc);
1037 }
1038
1039 #define GRANT_SHRINK_LIMIT PTLRPC_MAX_BRW_SIZE
1040 static int osc_should_shrink_grant(struct client_obd *client)
1041 {
1042         cfs_time_t time = cfs_time_current();
1043         cfs_time_t next_shrink = client->cl_next_shrink_grant;
1044
1045         if ((client->cl_import->imp_connect_data.ocd_connect_flags &
1046              OBD_CONNECT_GRANT_SHRINK) == 0)
1047                 return 0;
1048
1049         if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
1050                 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
1051                     client->cl_avail_grant > GRANT_SHRINK_LIMIT)
1052                         return 1;
1053                 else
1054                         osc_update_next_shrink(client);
1055         }
1056         return 0;
1057 }
1058
1059 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
1060 {
1061         struct client_obd *client;
1062
1063         cfs_list_for_each_entry(client, &item->ti_obd_list,
1064                                 cl_grant_shrink_list) {
1065                 if (osc_should_shrink_grant(client))
1066                         osc_shrink_grant(client);
1067         }
1068         return 0;
1069 }
1070
1071 static int osc_add_shrink_grant(struct client_obd *client)
1072 {
1073         int rc;
1074
1075         rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
1076                                        TIMEOUT_GRANT,
1077                                        osc_grant_shrink_grant_cb, NULL,
1078                                        &client->cl_grant_shrink_list);
1079         if (rc) {
1080                 CERROR("add grant client %s error %d\n",
1081                         client->cl_import->imp_obd->obd_name, rc);
1082                 return rc;
1083         }
1084         CDEBUG(D_CACHE, "add grant client %s \n",
1085                client->cl_import->imp_obd->obd_name);
1086         osc_update_next_shrink(client);
1087         return 0;
1088 }
1089
1090 static int osc_del_shrink_grant(struct client_obd *client)
1091 {
1092         return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
1093                                          TIMEOUT_GRANT);
1094 }
1095
1096 static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
1097 {
1098         /*
1099          * ocd_grant is the total grant amount we're expect to hold: if we've
1100          * been evicted, it's the new avail_grant amount, cl_dirty will drop
1101          * to 0 as inflight RPCs fail out; otherwise, it's avail_grant + dirty.
1102          *
1103          * race is tolerable here: if we're evicted, but imp_state already
1104          * left EVICTED state, then cl_dirty must be 0 already.
1105          */
1106         client_obd_list_lock(&cli->cl_loi_list_lock);
1107         if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
1108                 cli->cl_avail_grant = ocd->ocd_grant;
1109         else
1110                 cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
1111
1112         if (cli->cl_avail_grant < 0) {
1113                 CWARN("%s: available grant < 0, the OSS is probably not running"
1114                       " with patch from bug20278 (%ld) \n",
1115                       cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant);
1116                 /* workaround for 1.6 servers which do not have 
1117                  * the patch from bug20278 */
1118                 cli->cl_avail_grant = ocd->ocd_grant;
1119         }
1120
1121         client_obd_list_unlock(&cli->cl_loi_list_lock);
1122
1123         CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld \n",
1124                cli->cl_import->imp_obd->obd_name,
1125                cli->cl_avail_grant, cli->cl_lost_grant);
1126
1127         if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
1128             cfs_list_empty(&cli->cl_grant_shrink_list))
1129                 osc_add_shrink_grant(cli);
1130 }
1131
1132 /* We assume that the reason this OSC got a short read is because it read
1133  * beyond the end of a stripe file; i.e. lustre is reading a sparse file
1134  * via the LOV, and it _knows_ it's reading inside the file, it's just that
1135  * this stripe never got written at or beyond this stripe offset yet. */
1136 static void handle_short_read(int nob_read, obd_count page_count,
1137                               struct brw_page **pga)
1138 {
1139         char *ptr;
1140         int i = 0;
1141
1142         /* skip bytes read OK */
1143         while (nob_read > 0) {
1144                 LASSERT (page_count > 0);
1145
1146                 if (pga[i]->count > nob_read) {
1147                         /* EOF inside this page */
1148                         ptr = cfs_kmap(pga[i]->pg) +
1149                                 (pga[i]->off & ~CFS_PAGE_MASK);
1150                         memset(ptr + nob_read, 0, pga[i]->count - nob_read);
1151                         cfs_kunmap(pga[i]->pg);
1152                         page_count--;
1153                         i++;
1154                         break;
1155                 }
1156
1157                 nob_read -= pga[i]->count;
1158                 page_count--;
1159                 i++;
1160         }
1161
1162         /* zero remaining pages */
1163         while (page_count-- > 0) {
1164                 ptr = cfs_kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
1165                 memset(ptr, 0, pga[i]->count);
1166                 cfs_kunmap(pga[i]->pg);
1167                 i++;
1168         }
1169 }
1170
1171 static int check_write_rcs(struct ptlrpc_request *req,
1172                            int requested_nob, int niocount,
1173                            obd_count page_count, struct brw_page **pga)
1174 {
1175         int     i;
1176         __u32   *remote_rcs;
1177
1178         remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
1179                                                   sizeof(*remote_rcs) *
1180                                                   niocount);
1181         if (remote_rcs == NULL) {
1182                 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1183                 return(-EPROTO);
1184         }
1185
1186         /* return error if any niobuf was in error */
1187         for (i = 0; i < niocount; i++) {
1188                 if (remote_rcs[i] < 0)
1189                         return(remote_rcs[i]);
1190
1191                 if (remote_rcs[i] != 0) {
1192                         CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1193                                 i, remote_rcs[i], req);
1194                         return(-EPROTO);
1195                 }
1196         }
1197
1198         if (req->rq_bulk->bd_nob_transferred != requested_nob) {
1199                 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1200                        req->rq_bulk->bd_nob_transferred, requested_nob);
1201                 return(-EPROTO);
1202         }
1203
1204         return (0);
1205 }
1206
1207 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1208 {
1209         if (p1->flag != p2->flag) {
1210                 unsigned mask = ~(OBD_BRW_FROM_GRANT|
1211                                   OBD_BRW_NOCACHE|OBD_BRW_SYNC|OBD_BRW_ASYNC);
1212
1213                 /* warn if we try to combine flags that we don't know to be
1214                  * safe to combine */
1215                 if ((p1->flag & mask) != (p2->flag & mask))
1216                         CERROR("is it ok to have flags 0x%x and 0x%x in the "
1217                                "same brw?\n", p1->flag, p2->flag);
1218                 return 0;
1219         }
1220
1221         return (p1->off + p1->count == p2->off);
1222 }
1223
1224 static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
1225                                    struct brw_page **pga, int opc,
1226                                    cksum_type_t cksum_type)
1227 {
1228         __u32 cksum;
1229         int i = 0;
1230
1231         LASSERT (pg_count > 0);
1232         cksum = init_checksum(cksum_type);
1233         while (nob > 0 && pg_count > 0) {
1234                 unsigned char *ptr = cfs_kmap(pga[i]->pg);
1235                 int off = pga[i]->off & ~CFS_PAGE_MASK;
1236                 int count = pga[i]->count > nob ? nob : pga[i]->count;
1237
1238                 /* corrupt the data before we compute the checksum, to
1239                  * simulate an OST->client data error */
1240                 if (i == 0 && opc == OST_READ &&
1241                     OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE))
1242                         memcpy(ptr + off, "bad1", min(4, nob));
1243                 cksum = compute_checksum(cksum, ptr + off, count, cksum_type);
1244                 cfs_kunmap(pga[i]->pg);
1245                 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d checksum %x\n",
1246                                off, cksum);
1247
1248                 nob -= pga[i]->count;
1249                 pg_count--;
1250                 i++;
1251         }
1252         /* For sending we only compute the wrong checksum instead
1253          * of corrupting the data so it is still correct on a redo */
1254         if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1255                 cksum++;
1256
1257         return cksum;
1258 }
1259
1260 static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
1261                                 struct lov_stripe_md *lsm, obd_count page_count,
1262                                 struct brw_page **pga,
1263                                 struct ptlrpc_request **reqp,
1264                                 struct obd_capa *ocapa, int reserve)
1265 {
1266         struct ptlrpc_request   *req;
1267         struct ptlrpc_bulk_desc *desc;
1268         struct ost_body         *body;
1269         struct obd_ioobj        *ioobj;
1270         struct niobuf_remote    *niobuf;
1271         int niocount, i, requested_nob, opc, rc;
1272         struct osc_brw_async_args *aa;
1273         struct req_capsule      *pill;
1274         struct brw_page *pg_prev;
1275
1276         ENTRY;
1277         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1278                 RETURN(-ENOMEM); /* Recoverable */
1279         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1280                 RETURN(-EINVAL); /* Fatal */
1281
1282         if ((cmd & OBD_BRW_WRITE) != 0) {
1283                 opc = OST_WRITE;
1284                 req = ptlrpc_request_alloc_pool(cli->cl_import,
1285                                                 cli->cl_import->imp_rq_pool,
1286                                                 &RQF_OST_BRW_WRITE);
1287         } else {
1288                 opc = OST_READ;
1289                 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1290         }
1291         if (req == NULL)
1292                 RETURN(-ENOMEM);
1293
1294         for (niocount = i = 1; i < page_count; i++) {
1295                 if (!can_merge_pages(pga[i - 1], pga[i]))
1296                         niocount++;
1297         }
1298
1299         pill = &req->rq_pill;
1300         req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1301                              sizeof(*ioobj));
1302         req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1303                              niocount * sizeof(*niobuf));
1304         osc_set_capa_size(req, &RMF_CAPA1, ocapa);
1305
1306         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1307         if (rc) {
1308                 ptlrpc_request_free(req);
1309                 RETURN(rc);
1310         }
1311         req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
1312         ptlrpc_at_set_req_timeout(req);
1313
1314         if (opc == OST_WRITE)
1315                 desc = ptlrpc_prep_bulk_imp(req, page_count,
1316                                             BULK_GET_SOURCE, OST_BULK_PORTAL);
1317         else
1318                 desc = ptlrpc_prep_bulk_imp(req, page_count,
1319                                             BULK_PUT_SINK, OST_BULK_PORTAL);
1320
1321         if (desc == NULL)
1322                 GOTO(out, rc = -ENOMEM);
1323         /* NB request now owns desc and will free it when it gets freed */
1324
1325         body = req_capsule_client_get(pill, &RMF_OST_BODY);
1326         ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1327         niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1328         LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1329
1330         lustre_set_wire_obdo(&body->oa, oa);
1331
1332         obdo_to_ioobj(oa, ioobj);
1333         ioobj->ioo_bufcnt = niocount;
1334         osc_pack_capa(req, body, ocapa);
1335         LASSERT (page_count > 0);
1336         pg_prev = pga[0];
1337         for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1338                 struct brw_page *pg = pga[i];
1339
1340                 LASSERT(pg->count > 0);
1341                 LASSERTF((pg->off & ~CFS_PAGE_MASK) + pg->count <= CFS_PAGE_SIZE,
1342                          "i: %d pg: %p off: "LPU64", count: %u\n", i, pg,
1343                          pg->off, pg->count);
1344 #ifdef __linux__
1345                 LASSERTF(i == 0 || pg->off > pg_prev->off,
1346                          "i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
1347                          " prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
1348                          i, page_count,
1349                          pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1350                          pg_prev->pg, page_private(pg_prev->pg),
1351                          pg_prev->pg->index, pg_prev->off);
1352 #else
1353                 LASSERTF(i == 0 || pg->off > pg_prev->off,
1354                          "i %d p_c %u\n", i, page_count);
1355 #endif
1356                 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1357                         (pg->flag & OBD_BRW_SRVLOCK));
1358
1359                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->off & ~CFS_PAGE_MASK,
1360                                       pg->count);
1361                 requested_nob += pg->count;
1362
1363                 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1364                         niobuf--;
1365                         niobuf->len += pg->count;
1366                 } else {
1367                         niobuf->offset = pg->off;
1368                         niobuf->len    = pg->count;
1369                         niobuf->flags  = pg->flag;
1370                 }
1371                 pg_prev = pg;
1372         }
1373
1374         LASSERTF((void *)(niobuf - niocount) ==
1375                 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1376                 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1377                 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1378
1379         osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1380         if (osc_should_shrink_grant(cli))
1381                 osc_shrink_grant_local(cli, &body->oa);
1382
1383         /* size[REQ_REC_OFF] still sizeof (*body) */
1384         if (opc == OST_WRITE) {
1385                 if (unlikely(cli->cl_checksum) &&
1386                     !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1387                         /* store cl_cksum_type in a local variable since
1388                          * it can be changed via lprocfs */
1389                         cksum_type_t cksum_type = cli->cl_cksum_type;
1390
1391                         if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1392                                 oa->o_flags &= OBD_FL_LOCAL_MASK;
1393                                 body->oa.o_flags = 0;
1394                         }
1395                         body->oa.o_flags |= cksum_type_pack(cksum_type);
1396                         body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1397                         body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1398                                                              page_count, pga,
1399                                                              OST_WRITE,
1400                                                              cksum_type);
1401                         CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1402                                body->oa.o_cksum);
1403                         /* save this in 'oa', too, for later checking */
1404                         oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1405                         oa->o_flags |= cksum_type_pack(cksum_type);
1406                 } else {
1407                         /* clear out the checksum flag, in case this is a
1408                          * resend but cl_checksum is no longer set. b=11238 */
1409                         oa->o_valid &= ~OBD_MD_FLCKSUM;
1410                 }
1411                 oa->o_cksum = body->oa.o_cksum;
1412                 /* 1 RC per niobuf */
1413                 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1414                                      sizeof(__u32) * niocount);
1415         } else {
1416                 if (unlikely(cli->cl_checksum) &&
1417                     !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1418                         if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1419                                 body->oa.o_flags = 0;
1420                         body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1421                         body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1422                 }
1423         }
1424         ptlrpc_request_set_replen(req);
1425
1426         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1427         aa = ptlrpc_req_async_args(req);
1428         aa->aa_oa = oa;
1429         aa->aa_requested_nob = requested_nob;
1430         aa->aa_nio_count = niocount;
1431         aa->aa_page_count = page_count;
1432         aa->aa_resends = 0;
1433         aa->aa_ppga = pga;
1434         aa->aa_cli = cli;
1435         CFS_INIT_LIST_HEAD(&aa->aa_oaps);
1436         if (ocapa && reserve)
1437                 aa->aa_ocapa = capa_get(ocapa);
1438
1439         *reqp = req;
1440         RETURN(0);
1441
1442  out:
1443         ptlrpc_req_finished(req);
1444         RETURN(rc);
1445 }
1446
1447 static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
1448                                 __u32 client_cksum, __u32 server_cksum, int nob,
1449                                 obd_count page_count, struct brw_page **pga,
1450                                 cksum_type_t client_cksum_type)
1451 {
1452         __u32 new_cksum;
1453         char *msg;
1454         cksum_type_t cksum_type;
1455
1456         if (server_cksum == client_cksum) {
1457                 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1458                 return 0;
1459         }
1460
1461         /* If this is mmaped file - it can be changed at any time */
1462         if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & OBD_FL_MMAP)
1463                 return 1;
1464
1465         if (oa->o_valid & OBD_MD_FLFLAGS)
1466                 cksum_type = cksum_type_unpack(oa->o_flags);
1467         else
1468                 cksum_type = OBD_CKSUM_CRC32;
1469
1470         new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE,
1471                                       cksum_type);
1472
1473         if (cksum_type != client_cksum_type)
1474                 msg = "the server did not use the checksum type specified in "
1475                       "the original request - likely a protocol problem";
1476         else if (new_cksum == server_cksum)
1477                 msg = "changed on the client after we checksummed it - "
1478                       "likely false positive due to mmap IO (bug 11742)";
1479         else if (new_cksum == client_cksum)
1480                 msg = "changed in transit before arrival at OST";
1481         else
1482                 msg = "changed in transit AND doesn't match the original - "
1483                       "likely false positive due to mmap IO (bug 11742)";
1484
1485         LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
1486                            " object "LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1487                            msg, libcfs_nid2str(peer->nid),
1488                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1489                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1490                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1491                            oa->o_id,
1492                            oa->o_valid & OBD_MD_FLGROUP ? oa->o_seq : (__u64)0,
1493                            pga[0]->off,
1494                            pga[page_count-1]->off + pga[page_count-1]->count - 1);
1495         CERROR("original client csum %x (type %x), server csum %x (type %x), "
1496                "client csum now %x\n", client_cksum, client_cksum_type,
1497                server_cksum, cksum_type, new_cksum);
1498         return 1;
1499 }
1500
1501 /* Note rc enters this function as number of bytes transferred */
1502 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1503 {
1504         struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1505         const lnet_process_id_t *peer =
1506                         &req->rq_import->imp_connection->c_peer;
1507         struct client_obd *cli = aa->aa_cli;
1508         struct ost_body *body;
1509         __u32 client_cksum = 0;
1510         ENTRY;
1511
1512         if (rc < 0 && rc != -EDQUOT) {
1513                 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
1514                 RETURN(rc);
1515         }
1516
1517         LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1518         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1519         if (body == NULL) {
1520                 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
1521                 RETURN(-EPROTO);
1522         }
1523
1524 #ifdef HAVE_QUOTA_SUPPORT
1525         /* set/clear over quota flag for a uid/gid */
1526         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1527             body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) {
1528                 unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid };
1529
1530                 CDEBUG(D_QUOTA, "setdq for [%u %u] with valid "LPX64", flags %x\n",
1531                        body->oa.o_uid, body->oa.o_gid, body->oa.o_valid,
1532                        body->oa.o_flags);
1533                 lquota_setdq(quota_interface, cli, qid, body->oa.o_valid,
1534                              body->oa.o_flags);
1535         }
1536 #endif
1537
1538         osc_update_grant(cli, body);
1539
1540         if (rc < 0)
1541                 RETURN(rc);
1542
1543         if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1544                 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1545
1546         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1547                 if (rc > 0) {
1548                         CERROR("Unexpected +ve rc %d\n", rc);
1549                         RETURN(-EPROTO);
1550                 }
1551                 LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob);
1552
1553                 if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1554                         RETURN(-EAGAIN);
1555
1556                 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1557                     check_write_checksum(&body->oa, peer, client_cksum,
1558                                          body->oa.o_cksum, aa->aa_requested_nob,
1559                                          aa->aa_page_count, aa->aa_ppga,
1560                                          cksum_type_unpack(aa->aa_oa->o_flags)))
1561                         RETURN(-EAGAIN);
1562
1563                 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1564                                      aa->aa_page_count, aa->aa_ppga);
1565                 GOTO(out, rc);
1566         }
1567
1568         /* The rest of this function executes only for OST_READs */
1569
1570         /* if unwrap_bulk failed, return -EAGAIN to retry */
1571         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1572         if (rc < 0)
1573                 GOTO(out, rc = -EAGAIN);
1574
1575         if (rc > aa->aa_requested_nob) {
1576                 CERROR("Unexpected rc %d (%d requested)\n", rc,
1577                        aa->aa_requested_nob);
1578                 RETURN(-EPROTO);
1579         }
1580
1581         if (rc != req->rq_bulk->bd_nob_transferred) {
1582                 CERROR ("Unexpected rc %d (%d transferred)\n",
1583                         rc, req->rq_bulk->bd_nob_transferred);
1584                 return (-EPROTO);
1585         }
1586
1587         if (rc < aa->aa_requested_nob)
1588                 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1589
1590         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1591                 static int cksum_counter;
1592                 __u32      server_cksum = body->oa.o_cksum;
1593                 char      *via;
1594                 char      *router;
1595                 cksum_type_t cksum_type;
1596
1597                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1598                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1599                 else
1600                         cksum_type = OBD_CKSUM_CRC32;
1601                 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1602                                                  aa->aa_ppga, OST_READ,
1603                                                  cksum_type);
1604
1605                 if (peer->nid == req->rq_bulk->bd_sender) {
1606                         via = router = "";
1607                 } else {
1608                         via = " via ";
1609                         router = libcfs_nid2str(req->rq_bulk->bd_sender);
1610                 }
1611
1612                 if (server_cksum == ~0 && rc > 0) {
1613                         CERROR("Protocol error: server %s set the 'checksum' "
1614                                "bit, but didn't send a checksum.  Not fatal, "
1615                                "but please notify on http://bugzilla.lustre.org/\n",
1616                                libcfs_nid2str(peer->nid));
1617                 } else if (server_cksum != client_cksum) {
1618                         LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1619                                            "%s%s%s inode "DFID" object "
1620                                            LPU64"/"LPU64" extent "
1621                                            "["LPU64"-"LPU64"]\n",
1622                                            req->rq_import->imp_obd->obd_name,
1623                                            libcfs_nid2str(peer->nid),
1624                                            via, router,
1625                                            body->oa.o_valid & OBD_MD_FLFID ?
1626                                                 body->oa.o_parent_seq : (__u64)0,
1627                                            body->oa.o_valid & OBD_MD_FLFID ?
1628                                                 body->oa.o_parent_oid : 0,
1629                                            body->oa.o_valid & OBD_MD_FLFID ?
1630                                                 body->oa.o_parent_ver : 0,
1631                                            body->oa.o_id,
1632                                            body->oa.o_valid & OBD_MD_FLGROUP ?
1633                                                 body->oa.o_seq : (__u64)0,
1634                                            aa->aa_ppga[0]->off,
1635                                            aa->aa_ppga[aa->aa_page_count-1]->off +
1636                                            aa->aa_ppga[aa->aa_page_count-1]->count -
1637                                                                         1);
1638                         CERROR("client %x, server %x, cksum_type %x\n",
1639                                client_cksum, server_cksum, cksum_type);
1640                         cksum_counter = 0;
1641                         aa->aa_oa->o_cksum = client_cksum;
1642                         rc = -EAGAIN;
1643                 } else {
1644                         cksum_counter++;
1645                         CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1646                         rc = 0;
1647                 }
1648         } else if (unlikely(client_cksum)) {
1649                 static int cksum_missed;
1650
1651                 cksum_missed++;
1652                 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1653                         CERROR("Checksum %u requested from %s but not sent\n",
1654                                cksum_missed, libcfs_nid2str(peer->nid));
1655         } else {
1656                 rc = 0;
1657         }
1658 out:
1659         if (rc >= 0)
1660                 lustre_get_wire_obdo(aa->aa_oa, &body->oa);
1661
1662         RETURN(rc);
1663 }
1664
1665 static int osc_brw_internal(int cmd, struct obd_export *exp, struct obdo *oa,
1666                             struct lov_stripe_md *lsm,
1667                             obd_count page_count, struct brw_page **pga,
1668                             struct obd_capa *ocapa)
1669 {
1670         struct ptlrpc_request *req;
1671         int                    rc;
1672         cfs_waitq_t            waitq;
1673         int                    resends = 0;
1674         struct l_wait_info     lwi;
1675
1676         ENTRY;
1677
1678         cfs_waitq_init(&waitq);
1679
1680 restart_bulk:
1681         rc = osc_brw_prep_request(cmd, &exp->exp_obd->u.cli, oa, lsm,
1682                                   page_count, pga, &req, ocapa, 0);
1683         if (rc != 0)
1684                 return (rc);
1685
1686         rc = ptlrpc_queue_wait(req);
1687
1688         if (rc == -ETIMEDOUT && req->rq_resend) {
1689                 DEBUG_REQ(D_HA, req,  "BULK TIMEOUT");
1690                 ptlrpc_req_finished(req);
1691                 goto restart_bulk;
1692         }
1693
1694         rc = osc_brw_fini_request(req, rc);
1695
1696         ptlrpc_req_finished(req);
1697         if (osc_recoverable_error(rc)) {
1698                 resends++;
1699                 if (!osc_should_resend(resends, &exp->exp_obd->u.cli)) {
1700                         CERROR("too many resend retries, returning error\n");
1701                         RETURN(-EIO);
1702                 }
1703
1704                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
1705                 l_wait_event(waitq, 0, &lwi);
1706
1707                 goto restart_bulk;
1708         }
1709
1710         RETURN (rc);
1711 }
1712
1713 int osc_brw_redo_request(struct ptlrpc_request *request,
1714                          struct osc_brw_async_args *aa)
1715 {
1716         struct ptlrpc_request *new_req;
1717         struct ptlrpc_request_set *set = request->rq_set;
1718         struct osc_brw_async_args *new_aa;
1719         struct osc_async_page *oap;
1720         int rc = 0;
1721         ENTRY;
1722
1723         if (!osc_should_resend(aa->aa_resends, aa->aa_cli)) {
1724                 CERROR("too many resent retries, returning error\n");
1725                 RETURN(-EIO);
1726         }
1727
1728         DEBUG_REQ(D_ERROR, request, "redo for recoverable error");
1729
1730         rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1731                                         OST_WRITE ? OBD_BRW_WRITE :OBD_BRW_READ,
1732                                   aa->aa_cli, aa->aa_oa,
1733                                   NULL /* lsm unused by osc currently */,
1734                                   aa->aa_page_count, aa->aa_ppga,
1735                                   &new_req, aa->aa_ocapa, 0);
1736         if (rc)
1737                 RETURN(rc);
1738
1739         client_obd_list_lock(&aa->aa_cli->cl_loi_list_lock);
1740
1741         cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1742                 if (oap->oap_request != NULL) {
1743                         LASSERTF(request == oap->oap_request,
1744                                  "request %p != oap_request %p\n",
1745                                  request, oap->oap_request);
1746                         if (oap->oap_interrupted) {
1747                                 client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
1748                                 ptlrpc_req_finished(new_req);
1749                                 RETURN(-EINTR);
1750                         }
1751                 }
1752         }
1753         /* New request takes over pga and oaps from old request.
1754          * Note that copying a list_head doesn't work, need to move it... */
1755         aa->aa_resends++;
1756         new_req->rq_interpret_reply = request->rq_interpret_reply;
1757         new_req->rq_async_args = request->rq_async_args;
1758         new_req->rq_sent = cfs_time_current_sec() + aa->aa_resends;
1759
1760         new_aa = ptlrpc_req_async_args(new_req);
1761
1762         CFS_INIT_LIST_HEAD(&new_aa->aa_oaps);
1763         cfs_list_splice(&aa->aa_oaps, &new_aa->aa_oaps);
1764         CFS_INIT_LIST_HEAD(&aa->aa_oaps);
1765
1766         cfs_list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1767                 if (oap->oap_request) {
1768                         ptlrpc_req_finished(oap->oap_request);
1769                         oap->oap_request = ptlrpc_request_addref(new_req);
1770                 }
1771         }
1772
1773         new_aa->aa_ocapa = aa->aa_ocapa;
1774         aa->aa_ocapa = NULL;
1775
1776         /* use ptlrpc_set_add_req is safe because interpret functions work
1777          * in check_set context. only one way exist with access to request
1778          * from different thread got -EINTR - this way protected with
1779          * cl_loi_list_lock */
1780         ptlrpc_set_add_req(set, new_req);
1781
1782         client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
1783
1784         DEBUG_REQ(D_INFO, new_req, "new request");
1785         RETURN(0);
1786 }
1787
1788 /*
1789  * ugh, we want disk allocation on the target to happen in offset order.  we'll
1790  * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1791  * fine for our small page arrays and doesn't require allocation.  its an
1792  * insertion sort that swaps elements that are strides apart, shrinking the
1793  * stride down until its '1' and the array is sorted.
1794  */
1795 static void sort_brw_pages(struct brw_page **array, int num)
1796 {
1797         int stride, i, j;
1798         struct brw_page *tmp;
1799
1800         if (num == 1)
1801                 return;
1802         for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1803                 ;
1804
1805         do {
1806                 stride /= 3;
1807                 for (i = stride ; i < num ; i++) {
1808                         tmp = array[i];
1809                         j = i;
1810                         while (j >= stride && array[j - stride]->off > tmp->off) {
1811                                 array[j] = array[j - stride];
1812                                 j -= stride;
1813                         }
1814                         array[j] = tmp;
1815                 }
1816         } while (stride > 1);
1817 }
1818
1819 static obd_count max_unfragmented_pages(struct brw_page **pg, obd_count pages)
1820 {
1821         int count = 1;
1822         int offset;
1823         int i = 0;
1824
1825         LASSERT (pages > 0);
1826         offset = pg[i]->off & ~CFS_PAGE_MASK;
1827
1828         for (;;) {
1829                 pages--;
1830                 if (pages == 0)         /* that's all */
1831                         return count;
1832
1833                 if (offset + pg[i]->count < CFS_PAGE_SIZE)
1834                         return count;   /* doesn't end on page boundary */
1835
1836                 i++;
1837                 offset = pg[i]->off & ~CFS_PAGE_MASK;
1838                 if (offset != 0)        /* doesn't start on page boundary */
1839                         return count;
1840
1841                 count++;
1842         }
1843 }
1844
1845 static struct brw_page **osc_build_ppga(struct brw_page *pga, obd_count count)
1846 {
1847         struct brw_page **ppga;
1848         int i;
1849
1850         OBD_ALLOC(ppga, sizeof(*ppga) * count);
1851         if (ppga == NULL)
1852                 return NULL;
1853
1854         for (i = 0; i < count; i++)
1855                 ppga[i] = pga + i;
1856         return ppga;
1857 }
1858
1859 static void osc_release_ppga(struct brw_page **ppga, obd_count count)
1860 {
1861         LASSERT(ppga != NULL);
1862         OBD_FREE(ppga, sizeof(*ppga) * count);
1863 }
1864
1865 static int osc_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1866                    obd_count page_count, struct brw_page *pga,
1867                    struct obd_trans_info *oti)
1868 {
1869         struct obdo *saved_oa = NULL;
1870         struct brw_page **ppga, **orig;
1871         struct obd_import *imp = class_exp2cliimp(exp);
1872         struct client_obd *cli;
1873         int rc, page_count_orig;
1874         ENTRY;
1875
1876         LASSERT((imp != NULL) && (imp->imp_obd != NULL));
1877         cli = &imp->imp_obd->u.cli;
1878
1879         if (cmd & OBD_BRW_CHECK) {
1880                 /* The caller just wants to know if there's a chance that this
1881                  * I/O can succeed */
1882
1883                 if (imp->imp_invalid)
1884                         RETURN(-EIO);
1885                 RETURN(0);
1886         }
1887
1888         /* test_brw with a failed create can trip this, maybe others. */
1889         LASSERT(cli->cl_max_pages_per_rpc);
1890
1891         rc = 0;
1892
1893         orig = ppga = osc_build_ppga(pga, page_count);
1894         if (ppga == NULL)
1895                 RETURN(-ENOMEM);
1896         page_count_orig = page_count;
1897
1898         sort_brw_pages(ppga, page_count);
1899         while (page_count) {
1900                 obd_count pages_per_brw;
1901
1902                 if (page_count > cli->cl_max_pages_per_rpc)
1903                         pages_per_brw = cli->cl_max_pages_per_rpc;
1904                 else
1905                         pages_per_brw = page_count;
1906
1907                 pages_per_brw = max_unfragmented_pages(ppga, pages_per_brw);
1908
1909                 if (saved_oa != NULL) {
1910                         /* restore previously saved oa */
1911                         *oinfo->oi_oa = *saved_oa;
1912                 } else if (page_count > pages_per_brw) {
1913                         /* save a copy of oa (brw will clobber it) */
1914                         OBDO_ALLOC(saved_oa);
1915                         if (saved_oa == NULL)
1916                                 GOTO(out, rc = -ENOMEM);
1917                         *saved_oa = *oinfo->oi_oa;
1918                 }
1919
1920                 rc = osc_brw_internal(cmd, exp, oinfo->oi_oa, oinfo->oi_md,
1921                                       pages_per_brw, ppga, oinfo->oi_capa);
1922
1923                 if (rc != 0)
1924                         break;
1925
1926                 page_count -= pages_per_brw;
1927                 ppga += pages_per_brw;
1928         }
1929
1930 out:
1931         osc_release_ppga(orig, page_count_orig);
1932
1933         if (saved_oa != NULL)
1934                 OBDO_FREE(saved_oa);
1935
1936         RETURN(rc);
1937 }
1938
1939 /* The companion to osc_enter_cache(), called when @oap is no longer part of
1940  * the dirty accounting.  Writeback completes or truncate happens before
1941  * writing starts.  Must be called with the loi lock held. */
1942 static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap,
1943                            int sent)
1944 {
1945         osc_release_write_grant(cli, &oap->oap_brw_page, sent);
1946 }
1947
1948 static int lop_makes_syncfs_rpc(struct loi_oap_pages *lop)
1949 {
1950         struct osc_async_page *oap;
1951         ENTRY;
1952
1953         if (cfs_list_empty(&lop->lop_urgent))
1954                 RETURN(0);
1955
1956         oap = cfs_list_entry(lop->lop_urgent.next,
1957                              struct osc_async_page, oap_urgent_item);
1958
1959         if (oap->oap_async_flags & ASYNC_SYNCFS) {
1960                 CDEBUG(D_CACHE, "syncfs request forcing RPC\n");
1961                 RETURN(1);
1962         }
1963
1964         RETURN(0);
1965 }
1966
1967 /* This maintains the lists of pending pages to read/write for a given object
1968  * (lop).  This is used by osc_check_rpcs->osc_next_loi() and loi_list_maint()
1969  * to quickly find objects that are ready to send an RPC. */
1970 static int lop_makes_rpc(struct client_obd *cli, struct loi_oap_pages *lop,
1971                          int cmd)
1972 {
1973         int optimal;
1974         ENTRY;
1975
1976         if (lop->lop_num_pending == 0)
1977                 RETURN(0);
1978
1979         /* if we have an invalid import we want to drain the queued pages
1980          * by forcing them through rpcs that immediately fail and complete
1981          * the pages.  recovery relies on this to empty the queued pages
1982          * before canceling the locks and evicting down the llite pages */
1983         if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
1984                 RETURN(1);
1985
1986         /* stream rpcs in queue order as long as as there is an urgent page
1987          * queued.  this is our cheap solution for good batching in the case
1988          * where writepage marks some random page in the middle of the file
1989          * as urgent because of, say, memory pressure */
1990         if (!cfs_list_empty(&lop->lop_urgent)) {
1991                 CDEBUG(D_CACHE, "urgent request forcing RPC\n");
1992                 RETURN(1);
1993         }
1994         /* fire off rpcs when we have 'optimal' rpcs as tuned for the wire. */
1995         optimal = cli->cl_max_pages_per_rpc;
1996         if (cmd & OBD_BRW_WRITE) {
1997                 /* trigger a write rpc stream as long as there are dirtiers
1998                  * waiting for space.  as they're waiting, they're not going to
1999                  * create more pages to coalesce with what's waiting.. */
2000                 if (!cfs_list_empty(&cli->cl_cache_waiters)) {
2001                         CDEBUG(D_CACHE, "cache waiters forcing RPC\n");
2002                         RETURN(1);
2003                 }
2004                 /* +16 to avoid triggering rpcs that would want to include pages
2005                  * that are being queued but which can't be made ready until
2006                  * the queuer finishes with the page. this is a wart for
2007                  * llite::commit_write() */
2008                 optimal += 16;
2009         }
2010         if (lop->lop_num_pending >= optimal)
2011                 RETURN(1);
2012
2013         RETURN(0);
2014 }
2015
2016 static int lop_makes_hprpc(struct loi_oap_pages *lop)
2017 {
2018         struct osc_async_page *oap;
2019         ENTRY;
2020
2021         if (cfs_list_empty(&lop->lop_urgent))
2022                 RETURN(0);
2023
2024         oap = cfs_list_entry(lop->lop_urgent.next,
2025                          struct osc_async_page, oap_urgent_item);
2026
2027         if (oap->oap_async_flags & ASYNC_HP) {
2028                 CDEBUG(D_CACHE, "hp request forcing RPC\n");
2029                 RETURN(1);
2030         }
2031
2032         RETURN(0);
2033 }
2034
2035 static void on_list(cfs_list_t *item, cfs_list_t *list,
2036                     int should_be_on)
2037 {
2038         if (cfs_list_empty(item) && should_be_on)
2039                 cfs_list_add_tail(item, list);
2040         else if (!cfs_list_empty(item) && !should_be_on)
2041                 cfs_list_del_init(item);
2042 }
2043
2044 /* maintain the loi's cli list membership invariants so that osc_send_oap_rpc
2045  * can find pages to build into rpcs quickly */
2046 void loi_list_maint(struct client_obd *cli, struct lov_oinfo *loi)
2047 {
2048         if (lop_makes_hprpc(&loi->loi_write_lop) ||
2049             lop_makes_hprpc(&loi->loi_read_lop)) {
2050                 /* HP rpc */
2051                 on_list(&loi->loi_ready_item, &cli->cl_loi_ready_list, 0);
2052                 on_list(&loi->loi_hp_ready_item, &cli->cl_loi_hp_ready_list, 1);
2053         } else {
2054                 if (lop_makes_syncfs_rpc(&loi->loi_write_lop)) {
2055                         on_list(&loi->loi_sync_fs_item,
2056                                 &cli->cl_loi_sync_fs_list,
2057                                 loi->loi_write_lop.lop_num_pending);
2058                 } else {
2059                         on_list(&loi->loi_hp_ready_item,
2060                                 &cli->cl_loi_hp_ready_list, 0);
2061                         on_list(&loi->loi_ready_item, &cli->cl_loi_ready_list,
2062                                 lop_makes_rpc(cli, &loi->loi_write_lop,
2063                                               OBD_BRW_WRITE)||
2064                                 lop_makes_rpc(cli, &loi->loi_read_lop,
2065                                               OBD_BRW_READ));
2066                 }
2067         }
2068
2069         on_list(&loi->loi_write_item, &cli->cl_loi_write_list,
2070                 loi->loi_write_lop.lop_num_pending);
2071
2072         on_list(&loi->loi_read_item, &cli->cl_loi_read_list,
2073                 loi->loi_read_lop.lop_num_pending);
2074 }
2075
2076 static void lop_update_pending(struct client_obd *cli,
2077                                struct loi_oap_pages *lop, int cmd, int delta)
2078 {
2079         lop->lop_num_pending += delta;
2080         if (cmd & OBD_BRW_WRITE)
2081                 cli->cl_pending_w_pages += delta;
2082         else
2083                 cli->cl_pending_r_pages += delta;
2084 }
2085
2086 /**
2087  * this is called when a sync waiter receives an interruption.  Its job is to
2088  * get the caller woken as soon as possible.  If its page hasn't been put in an
2089  * rpc yet it can dequeue immediately.  Otherwise it has to mark the rpc as
2090  * desiring interruption which will forcefully complete the rpc once the rpc
2091  * has timed out.
2092  */
2093 int osc_oap_interrupted(const struct lu_env *env, struct osc_async_page *oap)
2094 {
2095         struct loi_oap_pages *lop;
2096         struct lov_oinfo *loi;
2097         int rc = -EBUSY;
2098         ENTRY;
2099
2100         LASSERT(!oap->oap_interrupted);
2101         oap->oap_interrupted = 1;
2102
2103         /* ok, it's been put in an rpc. only one oap gets a request reference */
2104         if (oap->oap_request != NULL) {
2105                 ptlrpc_mark_interrupted(oap->oap_request);
2106                 ptlrpcd_wake(oap->oap_request);
2107                 ptlrpc_req_finished(oap->oap_request);
2108                 oap->oap_request = NULL;
2109         }
2110
2111         /*
2112          * page completion may be called only if ->cpo_prep() method was
2113          * executed by osc_io_submit(), that also adds page the to pending list
2114          */
2115         if (!cfs_list_empty(&oap->oap_pending_item)) {
2116                 cfs_list_del_init(&oap->oap_pending_item);
2117                 cfs_list_del_init(&oap->oap_urgent_item);
2118
2119                 loi = oap->oap_loi;
2120                 lop = (oap->oap_cmd & OBD_BRW_WRITE) ?
2121                         &loi->loi_write_lop : &loi->loi_read_lop;
2122                 lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, -1);
2123                 loi_list_maint(oap->oap_cli, oap->oap_loi);
2124                 rc = oap->oap_caller_ops->ap_completion(env,
2125                                           oap->oap_caller_data,
2126                                           oap->oap_cmd, NULL, -EINTR);
2127         }
2128
2129         RETURN(rc);
2130 }
2131
2132 /* this is trying to propogate async writeback errors back up to the
2133  * application.  As an async write fails we record the error code for later if
2134  * the app does an fsync.  As long as errors persist we force future rpcs to be
2135  * sync so that the app can get a sync error and break the cycle of queueing
2136  * pages for which writeback will fail. */
2137 static void osc_process_ar(struct osc_async_rc *ar, __u64 xid,
2138                            int rc)
2139 {
2140         if (rc) {
2141                 if (!ar->ar_rc)
2142                         ar->ar_rc = rc;
2143
2144                 ar->ar_force_sync = 1;
2145                 ar->ar_min_xid = ptlrpc_sample_next_xid();
2146                 return;
2147
2148         }
2149
2150         if (ar->ar_force_sync && (xid >= ar->ar_min_xid))
2151                 ar->ar_force_sync = 0;
2152 }
2153
2154 void osc_oap_to_pending(struct osc_async_page *oap)
2155 {
2156         struct loi_oap_pages *lop;
2157
2158         if (oap->oap_cmd & OBD_BRW_WRITE)
2159                 lop = &oap->oap_loi->loi_write_lop;
2160         else
2161                 lop = &oap->oap_loi->loi_read_lop;
2162
2163         if (oap->oap_async_flags & ASYNC_HP)
2164                 cfs_list_add(&oap->oap_urgent_item, &lop->lop_urgent);
2165         else if (oap->oap_async_flags & ASYNC_URGENT)
2166                 cfs_list_add_tail(&oap->oap_urgent_item, &lop->lop_urgent);
2167         cfs_list_add_tail(&oap->oap_pending_item, &lop->lop_pending);
2168         lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, 1);
2169 }
2170
2171 /* this must be called holding the loi list lock to give coverage to exit_cache,
2172  * async_flag maintenance, and oap_request */
2173 static void osc_ap_completion(const struct lu_env *env,
2174                               struct client_obd *cli, struct obdo *oa,
2175                               struct osc_async_page *oap, int sent, int rc)
2176 {
2177         __u64 xid = 0;
2178
2179         ENTRY;
2180         if (oap->oap_request != NULL) {
2181                 xid = ptlrpc_req_xid(oap->oap_request);
2182                 ptlrpc_req_finished(oap->oap_request);
2183                 oap->oap_request = NULL;
2184         }
2185
2186         cfs_spin_lock(&oap->oap_lock);
2187         oap->oap_async_flags = 0;
2188         cfs_spin_unlock(&oap->oap_lock);
2189         oap->oap_interrupted = 0;
2190
2191         if (oap->oap_cmd & OBD_BRW_WRITE) {
2192                 osc_process_ar(&cli->cl_ar, xid, rc);
2193                 osc_process_ar(&oap->oap_loi->loi_ar, xid, rc);
2194         }
2195
2196         if (rc == 0 && oa != NULL) {
2197                 if (oa->o_valid & OBD_MD_FLBLOCKS)
2198                         oap->oap_loi->loi_lvb.lvb_blocks = oa->o_blocks;
2199                 if (oa->o_valid & OBD_MD_FLMTIME)
2200                         oap->oap_loi->loi_lvb.lvb_mtime = oa->o_mtime;
2201                 if (oa->o_valid & OBD_MD_FLATIME)
2202                         oap->oap_loi->loi_lvb.lvb_atime = oa->o_atime;
2203                 if (oa->o_valid & OBD_MD_FLCTIME)
2204                         oap->oap_loi->loi_lvb.lvb_ctime = oa->o_ctime;
2205         }
2206
2207         rc = oap->oap_caller_ops->ap_completion(env, oap->oap_caller_data,
2208                                                 oap->oap_cmd, oa, rc);
2209
2210         /* ll_ap_completion (from llite) drops PG_locked. so, a new
2211          * I/O on the page could start, but OSC calls it under lock
2212          * and thus we can add oap back to pending safely */
2213         if (rc)
2214                 /* upper layer wants to leave the page on pending queue */
2215                 osc_oap_to_pending(oap);
2216         else
2217                 osc_exit_cache(cli, oap, sent);
2218         EXIT;
2219 }
2220
2221 static int brw_interpret(const struct lu_env *env,
2222                          struct ptlrpc_request *req, void *data, int rc)
2223 {
2224         struct osc_brw_async_args *aa = data;
2225         struct client_obd *cli;
2226         int async;
2227         ENTRY;
2228
2229         rc = osc_brw_fini_request(req, rc);
2230         CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
2231         if (osc_recoverable_error(rc)) {
2232                 /* Only retry once for mmaped files since the mmaped page
2233                  * might be modified at anytime. We have to retry at least
2234                  * once in case there WAS really a corruption of the page
2235                  * on the network, that was not caused by mmap() modifying
2236                  * the page. Bug11742 */
2237                 if ((rc == -EAGAIN) && (aa->aa_resends > 0) &&
2238                     aa->aa_oa->o_valid & OBD_MD_FLFLAGS &&
2239                     aa->aa_oa->o_flags & OBD_FL_MMAP) {
2240                         rc = 0;
2241                 } else {
2242                         rc = osc_brw_redo_request(req, aa);
2243                         if (rc == 0)
2244                                 RETURN(0);
2245                 }
2246         }
2247
2248         if (aa->aa_ocapa) {
2249                 capa_put(aa->aa_ocapa);
2250                 aa->aa_ocapa = NULL;
2251         }
2252
2253         cli = aa->aa_cli;
2254
2255         client_obd_list_lock(&cli->cl_loi_list_lock);
2256
2257         /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
2258          * is called so we know whether to go to sync BRWs or wait for more
2259          * RPCs to complete */
2260         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
2261                 cli->cl_w_in_flight--;
2262         else
2263                 cli->cl_r_in_flight--;
2264
2265         async = cfs_list_empty(&aa->aa_oaps);
2266         if (!async) { /* from osc_send_oap_rpc() */
2267                 struct osc_async_page *oap, *tmp;
2268                 /* the caller may re-use the oap after the completion call so
2269                  * we need to clean it up a little */
2270                 cfs_list_for_each_entry_safe(oap, tmp, &aa->aa_oaps,
2271                                              oap_rpc_item) {
2272                         cfs_list_del_init(&oap->oap_rpc_item);
2273                         osc_ap_completion(env, cli, aa->aa_oa, oap, 1, rc);
2274                 }
2275                 OBDO_FREE(aa->aa_oa);
2276         } else { /* from async_internal() */
2277                 obd_count i;
2278                 for (i = 0; i < aa->aa_page_count; i++)
2279                         osc_release_write_grant(aa->aa_cli, aa->aa_ppga[i], 1);
2280         }
2281         osc_wake_cache_waiters(cli);
2282         osc_wake_sync_fs(cli);
2283         osc_check_rpcs(env, cli);
2284         client_obd_list_unlock(&cli->cl_loi_list_lock);
2285         if (!async)
2286                 cl_req_completion(env, aa->aa_clerq, rc);
2287         osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
2288
2289         RETURN(rc);
2290 }
2291
2292 static struct ptlrpc_request *osc_build_req(const struct lu_env *env,
2293                                             struct client_obd *cli,
2294                                             cfs_list_t *rpc_list,
2295                                             int page_count, int cmd)
2296 {
2297         struct ptlrpc_request *req;
2298         struct brw_page **pga = NULL;
2299         struct osc_brw_async_args *aa;
2300         struct obdo *oa = NULL;
2301         const struct obd_async_page_ops *ops = NULL;
2302         void *caller_data = NULL;
2303         struct osc_async_page *oap;
2304         struct osc_async_page *tmp;
2305         struct ost_body *body;
2306         struct cl_req *clerq = NULL;
2307         enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
2308         struct ldlm_lock *lock = NULL;
2309         struct cl_req_attr crattr;
2310         int i, rc, mpflag = 0;
2311
2312         ENTRY;
2313         LASSERT(!cfs_list_empty(rpc_list));
2314
2315         if (cmd & OBD_BRW_MEMALLOC)
2316                 mpflag = cfs_memory_pressure_get_and_set();
2317
2318         memset(&crattr, 0, sizeof crattr);
2319         OBD_ALLOC(pga, sizeof(*pga) * page_count);
2320         if (pga == NULL)
2321                 GOTO(out, req = ERR_PTR(-ENOMEM));
2322
2323         OBDO_ALLOC(oa);
2324         if (oa == NULL)
2325                 GOTO(out, req = ERR_PTR(-ENOMEM));
2326
2327         i = 0;
2328         cfs_list_for_each_entry(oap, rpc_list, oap_rpc_item) {
2329                 struct cl_page *page = osc_oap2cl_page(oap);
2330                 if (ops == NULL) {
2331                         ops = oap->oap_caller_ops;
2332                         caller_data = oap->oap_caller_data;
2333
2334                         clerq = cl_req_alloc(env, page, crt,
2335                                              1 /* only 1-object rpcs for
2336                                                 * now */);
2337                         if (IS_ERR(clerq))
2338                                 GOTO(out, req = (void *)clerq);
2339                         lock = oap->oap_ldlm_lock;
2340                 }
2341                 pga[i] = &oap->oap_brw_page;
2342                 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
2343                 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
2344                        pga[i]->pg, cfs_page_index(oap->oap_page), oap, pga[i]->flag);
2345                 i++;
2346                 cl_req_page_add(env, clerq, page);
2347         }
2348
2349         /* always get the data for the obdo for the rpc */
2350         LASSERT(ops != NULL);
2351         crattr.cra_oa = oa;
2352         crattr.cra_capa = NULL;
2353         cl_req_attr_set(env, clerq, &crattr, ~0ULL);
2354         if (lock) {
2355                 oa->o_handle = lock->l_remote_handle;
2356                 oa->o_valid |= OBD_MD_FLHANDLE;
2357         }
2358
2359         rc = cl_req_prep(env, clerq);
2360         if (rc != 0) {
2361                 CERROR("cl_req_prep failed: %d\n", rc);
2362                 GOTO(out, req = ERR_PTR(rc));
2363         }
2364
2365         sort_brw_pages(pga, page_count);
2366         rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
2367                                   pga, &req, crattr.cra_capa, 1);
2368         if (rc != 0) {
2369                 CERROR("prep_req failed: %d\n", rc);
2370                 GOTO(out, req = ERR_PTR(rc));
2371         }
2372
2373         if (cmd & OBD_BRW_MEMALLOC)
2374                 req->rq_memalloc = 1;
2375
2376         /* Need to update the timestamps after the request is built in case
2377          * we race with setattr (locally or in queue at OST).  If OST gets
2378          * later setattr before earlier BRW (as determined by the request xid),
2379          * the OST will not use BRW timestamps.  Sadly, there is no obvious
2380          * way to do this in a single call.  bug 10150 */
2381         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2382         cl_req_attr_set(env, clerq, &crattr,
2383                         OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
2384
2385         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2386         aa = ptlrpc_req_async_args(req);
2387         CFS_INIT_LIST_HEAD(&aa->aa_oaps);
2388         cfs_list_splice(rpc_list, &aa->aa_oaps);
2389         CFS_INIT_LIST_HEAD(rpc_list);
2390         aa->aa_clerq = clerq;
2391 out:
2392         if (cmd & OBD_BRW_MEMALLOC)
2393                 cfs_memory_pressure_restore(mpflag);
2394
2395         capa_put(crattr.cra_capa);
2396         if (IS_ERR(req)) {
2397                 if (oa)
2398                         OBDO_FREE(oa);
2399                 if (pga)
2400                         OBD_FREE(pga, sizeof(*pga) * page_count);
2401                 /* this should happen rarely and is pretty bad, it makes the
2402                  * pending list not follow the dirty order */
2403                 client_obd_list_lock(&cli->cl_loi_list_lock);
2404                 cfs_list_for_each_entry_safe(oap, tmp, rpc_list, oap_rpc_item) {
2405                         cfs_list_del_init(&oap->oap_rpc_item);
2406
2407                         /* queued sync pages can be torn down while the pages
2408                          * were between the pending list and the rpc */
2409                         if (oap->oap_interrupted) {
2410                                 CDEBUG(D_INODE, "oap %p interrupted\n", oap);
2411                                 osc_ap_completion(env, cli, NULL, oap, 0,
2412                                                   oap->oap_count);
2413                                 continue;
2414                         }
2415                         osc_ap_completion(env, cli, NULL, oap, 0, PTR_ERR(req));
2416                 }
2417                 if (clerq && !IS_ERR(clerq))
2418                         cl_req_completion(env, clerq, PTR_ERR(req));
2419         }
2420         RETURN(req);
2421 }
2422
2423 /**
2424  * prepare pages for ASYNC io and put pages in send queue.
2425  *
2426  * \param cmd OBD_BRW_* macroses
2427  * \param lop pending pages
2428  *
2429  * \return zero if no page added to send queue.
2430  * \return 1 if pages successfully added to send queue.
2431  * \return negative on errors.
2432  */
2433 static int
2434 osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli,
2435                  struct lov_oinfo *loi,
2436                  int cmd, struct loi_oap_pages *lop)
2437 {
2438         struct ptlrpc_request *req;
2439         obd_count page_count = 0;
2440         struct osc_async_page *oap = NULL, *tmp;
2441         struct osc_brw_async_args *aa;
2442         const struct obd_async_page_ops *ops;
2443         CFS_LIST_HEAD(rpc_list);
2444         CFS_LIST_HEAD(tmp_list);
2445         unsigned int ending_offset;
2446         unsigned  starting_offset = 0;
2447         int srvlock = 0, mem_tight = 0;
2448         struct cl_object *clob = NULL;
2449         ENTRY;
2450
2451         /* ASYNC_HP pages first. At present, when the lock the pages is
2452          * to be canceled, the pages covered by the lock will be sent out
2453          * with ASYNC_HP. We have to send out them as soon as possible. */
2454         cfs_list_for_each_entry_safe(oap, tmp, &lop->lop_urgent, oap_urgent_item) {
2455                 if (oap->oap_async_flags & ASYNC_HP)
2456                         cfs_list_move(&oap->oap_pending_item, &tmp_list);
2457                 else
2458                         cfs_list_move_tail(&oap->oap_pending_item, &tmp_list);
2459                 if (++page_count >= cli->cl_max_pages_per_rpc)
2460                         break;
2461         }
2462
2463         cfs_list_splice(&tmp_list, &lop->lop_pending);
2464         page_count = 0;
2465
2466         /* first we find the pages we're allowed to work with */
2467         cfs_list_for_each_entry_safe(oap, tmp, &lop->lop_pending,
2468                                      oap_pending_item) {
2469                 ops = oap->oap_caller_ops;
2470
2471                 LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, "
2472                          "magic 0x%x\n", oap, oap->oap_magic);
2473
2474                 if (clob == NULL) {
2475                         /* pin object in memory, so that completion call-backs
2476                          * can be safely called under client_obd_list lock. */
2477                         clob = osc_oap2cl_page(oap)->cp_obj;
2478                         cl_object_get(clob);
2479                 }
2480
2481                 if (page_count != 0 &&
2482                     srvlock != !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK)) {
2483                         CDEBUG(D_PAGE, "SRVLOCK flag mismatch,"
2484                                " oap %p, page %p, srvlock %u\n",
2485                                oap, oap->oap_brw_page.pg, (unsigned)!srvlock);
2486                         break;
2487                 }
2488
2489                 /* If there is a gap at the start of this page, it can't merge
2490                  * with any previous page, so we'll hand the network a
2491                  * "fragmented" page array that it can't transfer in 1 RDMA */
2492                 if (page_count != 0 && oap->oap_page_off != 0)
2493                         break;
2494
2495                 /* in llite being 'ready' equates to the page being locked
2496                  * until completion unlocks it.  commit_write submits a page
2497                  * as not ready because its unlock will happen unconditionally
2498                  * as the call returns.  if we race with commit_write giving
2499                  * us that page we don't want to create a hole in the page
2500                  * stream, so we stop and leave the rpc to be fired by
2501                  * another dirtier or kupdated interval (the not ready page
2502                  * will still be on the dirty list).  we could call in
2503                  * at the end of ll_file_write to process the queue again. */
2504                 if (!(oap->oap_async_flags & ASYNC_READY)) {
2505                         int rc = ops->ap_make_ready(env, oap->oap_caller_data,
2506                                                     cmd);
2507                         if (rc < 0)
2508                                 CDEBUG(D_INODE, "oap %p page %p returned %d "
2509                                                 "instead of ready\n", oap,
2510                                                 oap->oap_page, rc);
2511                         switch (rc) {
2512                         case -EAGAIN:
2513                                 /* llite is telling us that the page is still
2514                                  * in commit_write and that we should try
2515                                  * and put it in an rpc again later.  we
2516                                  * break out of the loop so we don't create
2517                                  * a hole in the sequence of pages in the rpc
2518                                  * stream.*/
2519                                 oap = NULL;
2520                                 break;
2521                         case -EINTR:
2522                                 /* the io isn't needed.. tell the checks
2523                                  * below to complete the rpc with EINTR */
2524                                 cfs_spin_lock(&oap->oap_lock);
2525                                 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
2526                                 cfs_spin_unlock(&oap->oap_lock);
2527                                 oap->oap_count = -EINTR;
2528                                 break;
2529                         case 0:
2530                                 cfs_spin_lock(&oap->oap_lock);
2531                                 oap->oap_async_flags |= ASYNC_READY;
2532                                 cfs_spin_unlock(&oap->oap_lock);
2533                                 break;
2534                         default:
2535                                 LASSERTF(0, "oap %p page %p returned %d "
2536                                             "from make_ready\n", oap,
2537                                             oap->oap_page, rc);
2538                                 break;
2539                         }
2540                 }
2541                 if (oap == NULL)
2542                         break;
2543                 /*
2544                  * Page submitted for IO has to be locked. Either by
2545                  * ->ap_make_ready() or by higher layers.
2546                  */
2547 #if defined(__KERNEL__) && defined(__linux__)
2548                 {
2549                         struct cl_page *page;
2550
2551                         page = osc_oap2cl_page(oap);
2552
2553                         if (page->cp_type == CPT_CACHEABLE &&
2554                             !(PageLocked(oap->oap_page) &&
2555                               (CheckWriteback(oap->oap_page, cmd)))) {
2556                                 CDEBUG(D_PAGE, "page %p lost wb %lx/%x\n",
2557                                        oap->oap_page,
2558                                        (long)oap->oap_page->flags,
2559                                        oap->oap_async_flags);
2560                                 LBUG();
2561                         }
2562                 }
2563 #endif
2564
2565                 /* take the page out of our book-keeping */
2566                 cfs_list_del_init(&oap->oap_pending_item);
2567                 lop_update_pending(cli, lop, cmd, -1);
2568                 cfs_list_del_init(&oap->oap_urgent_item);
2569
2570                 if (page_count == 0)
2571                         starting_offset = (oap->oap_obj_off+oap->oap_page_off) &
2572                                           (PTLRPC_MAX_BRW_SIZE - 1);
2573
2574                 /* ask the caller for the size of the io as the rpc leaves. */
2575                 if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
2576                         oap->oap_count =
2577                                 ops->ap_refresh_count(env, oap->oap_caller_data,
2578                                                       cmd);
2579                         LASSERT(oap->oap_page_off + oap->oap_count <= CFS_PAGE_SIZE);
2580                 }
2581                 if (oap->oap_count <= 0) {
2582                         CDEBUG(D_CACHE, "oap %p count %d, completing\n", oap,
2583                                oap->oap_count);
2584                         osc_ap_completion(env, cli, NULL,
2585                                           oap, 0, oap->oap_count);
2586                         continue;
2587                 }
2588
2589                 /* now put the page back in our accounting */
2590                 cfs_list_add_tail(&oap->oap_rpc_item, &rpc_list);
2591                 if (oap->oap_brw_flags & OBD_BRW_MEMALLOC)
2592                         mem_tight = 1;
2593                 if (page_count == 0)
2594                         srvlock = !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK);
2595                 if (++page_count >= cli->cl_max_pages_per_rpc)
2596                         break;
2597
2598                 /* End on a PTLRPC_MAX_BRW_SIZE boundary.  We want full-sized
2599                  * RPCs aligned on PTLRPC_MAX_BRW_SIZE boundaries to help reads
2600                  * have the same alignment as the initial writes that allocated
2601                  * extents on the server. */
2602                 ending_offset = (oap->oap_obj_off + oap->oap_page_off +
2603                                  oap->oap_count) & (PTLRPC_MAX_BRW_SIZE - 1);
2604                 if (ending_offset == 0)
2605                         break;
2606
2607                 /* If there is a gap at the end of this page, it can't merge
2608                  * with any subsequent pages, so we'll hand the network a
2609                  * "fragmented" page array that it can't transfer in 1 RDMA */
2610                 if (oap->oap_page_off + oap->oap_count < CFS_PAGE_SIZE)
2611                         break;
2612         }
2613
2614         osc_wake_cache_waiters(cli);
2615         osc_wake_sync_fs(cli);
2616         loi_list_maint(cli, loi);
2617
2618         client_obd_list_unlock(&cli->cl_loi_list_lock);
2619
2620         if (clob != NULL)
2621                 cl_object_put(env, clob);
2622
2623         if (page_count == 0) {
2624                 client_obd_list_lock(&cli->cl_loi_list_lock);
2625                 RETURN(0);
2626         }
2627
2628         req = osc_build_req(env, cli, &rpc_list, page_count,
2629                             mem_tight ? (cmd | OBD_BRW_MEMALLOC) : cmd);
2630         if (IS_ERR(req)) {
2631                 LASSERT(cfs_list_empty(&rpc_list));
2632                 loi_list_maint(cli, loi);
2633                 RETURN(PTR_ERR(req));
2634         }
2635
2636         aa = ptlrpc_req_async_args(req);
2637
2638         if (cmd == OBD_BRW_READ) {
2639                 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2640                 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2641                 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2642                                       (starting_offset >> CFS_PAGE_SHIFT) + 1);
2643         } else {
2644                 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2645                 lprocfs_oh_tally(&cli->cl_write_rpc_hist,
2646                                  cli->cl_w_in_flight);
2647                 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2648                                       (starting_offset >> CFS_PAGE_SHIFT) + 1);
2649         }
2650         ptlrpc_lprocfs_brw(req, aa->aa_requested_nob);
2651
2652         client_obd_list_lock(&cli->cl_loi_list_lock);
2653
2654         if (cmd == OBD_BRW_READ)
2655                 cli->cl_r_in_flight++;
2656         else
2657                 cli->cl_w_in_flight++;
2658
2659         /* queued sync pages can be torn down while the pages
2660          * were between the pending list and the rpc */
2661         tmp = NULL;
2662         cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
2663                 /* only one oap gets a request reference */
2664                 if (tmp == NULL)
2665                         tmp = oap;
2666                 if (oap->oap_interrupted && !req->rq_intr) {
2667                         CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
2668                                oap, req);
2669                         ptlrpc_mark_interrupted(req);
2670                 }
2671         }
2672         if (tmp != NULL)
2673                 tmp->oap_request = ptlrpc_request_addref(req);
2674
2675         DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight",
2676                   page_count, aa, cli->cl_r_in_flight, cli->cl_w_in_flight);
2677
2678         req->rq_interpret_reply = brw_interpret;
2679         ptlrpcd_add_req(req, PSCOPE_BRW);
2680         RETURN(1);
2681 }
2682
2683 #define LOI_DEBUG(LOI, STR, args...)                                     \
2684         CDEBUG(D_INODE, "loi ready %d wr %d:%d rd %d:%d " STR,           \
2685                !cfs_list_empty(&(LOI)->loi_ready_item) ||                \
2686                !cfs_list_empty(&(LOI)->loi_hp_ready_item),               \
2687                (LOI)->loi_write_lop.lop_num_pending,                     \
2688                !cfs_list_empty(&(LOI)->loi_write_lop.lop_urgent),        \
2689                (LOI)->loi_read_lop.lop_num_pending,                      \
2690                !cfs_list_empty(&(LOI)->loi_read_lop.lop_urgent),         \
2691                args)                                                     \
2692
2693 /* This is called by osc_check_rpcs() to find which objects have pages that
2694  * we could be sending.  These lists are maintained by lop_makes_rpc(). */
2695 struct lov_oinfo *osc_next_loi(struct client_obd *cli)
2696 {
2697         ENTRY;
2698
2699         /* First return objects that have blocked locks so that they
2700          * will be flushed quickly and other clients can get the lock,
2701          * then objects which have pages ready to be stuffed into RPCs */
2702         if (!cfs_list_empty(&cli->cl_loi_hp_ready_list))
2703                 RETURN(cfs_list_entry(cli->cl_loi_hp_ready_list.next,
2704                                       struct lov_oinfo, loi_hp_ready_item));
2705         if (!cfs_list_empty(&cli->cl_loi_ready_list))
2706                 RETURN(cfs_list_entry(cli->cl_loi_ready_list.next,
2707                                       struct lov_oinfo, loi_ready_item));
2708         if (!cfs_list_empty(&cli->cl_loi_sync_fs_list))
2709                 RETURN(cfs_list_entry(cli->cl_loi_sync_fs_list.next,
2710                                       struct lov_oinfo, loi_sync_fs_item));
2711
2712         /* then if we have cache waiters, return all objects with queued
2713          * writes.  This is especially important when many small files
2714          * have filled up the cache and not been fired into rpcs because
2715          * they don't pass the nr_pending/object threshhold */
2716         if (!cfs_list_empty(&cli->cl_cache_waiters) &&
2717             !cfs_list_empty(&cli->cl_loi_write_list))
2718                 RETURN(cfs_list_entry(cli->cl_loi_write_list.next,
2719                                       struct lov_oinfo, loi_write_item));
2720
2721         /* then return all queued objects when we have an invalid import
2722          * so that they get flushed */
2723         if (cli->cl_import == NULL || cli->cl_import->imp_invalid) {
2724                 if (!cfs_list_empty(&cli->cl_loi_write_list))
2725                         RETURN(cfs_list_entry(cli->cl_loi_write_list.next,
2726                                               struct lov_oinfo,
2727                                               loi_write_item));
2728                 if (!cfs_list_empty(&cli->cl_loi_read_list))
2729                         RETURN(cfs_list_entry(cli->cl_loi_read_list.next,
2730                                               struct lov_oinfo, loi_read_item));
2731         }
2732         RETURN(NULL);
2733 }
2734
2735 static int osc_max_rpc_in_flight(struct client_obd *cli, struct lov_oinfo *loi)
2736 {
2737         struct osc_async_page *oap;
2738         int hprpc = 0;
2739
2740         if (!cfs_list_empty(&loi->loi_write_lop.lop_urgent)) {
2741                 oap = cfs_list_entry(loi->loi_write_lop.lop_urgent.next,
2742                                      struct osc_async_page, oap_urgent_item);
2743                 hprpc = !!(oap->oap_async_flags & ASYNC_HP);
2744         }
2745
2746         if (!hprpc && !cfs_list_empty(&loi->loi_read_lop.lop_urgent)) {
2747                 oap = cfs_list_entry(loi->loi_read_lop.lop_urgent.next,
2748                                      struct osc_async_page, oap_urgent_item);
2749                 hprpc = !!(oap->oap_async_flags & ASYNC_HP);
2750         }
2751
2752         return rpcs_in_flight(cli) >= cli->cl_max_rpcs_in_flight + hprpc;
2753 }
2754
2755 /* called with the loi list lock held */
2756 void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli)
2757 {
2758         struct lov_oinfo *loi;
2759         int rc = 0, race_counter = 0;
2760         ENTRY;
2761
2762         while ((loi = osc_next_loi(cli)) != NULL) {
2763                 LOI_DEBUG(loi, "%lu in flight\n", rpcs_in_flight(cli));
2764
2765                 if (osc_max_rpc_in_flight(cli, loi))
2766                         break;
2767
2768                 /* attempt some read/write balancing by alternating between
2769                  * reads and writes in an object.  The makes_rpc checks here
2770                  * would be redundant if we were getting read/write work items
2771                  * instead of objects.  we don't want send_oap_rpc to drain a
2772                  * partial read pending queue when we're given this object to
2773                  * do io on writes while there are cache waiters */
2774                 if (lop_makes_rpc(cli, &loi->loi_write_lop, OBD_BRW_WRITE)) {
2775                         rc = osc_send_oap_rpc(env, cli, loi, OBD_BRW_WRITE,
2776                                               &loi->loi_write_lop);
2777                         if (rc < 0) {
2778                                 CERROR("Write request failed with %d\n", rc);
2779
2780                                 /* osc_send_oap_rpc failed, mostly because of
2781                                  * memory pressure.
2782                                  *
2783                                  * It can't break here, because if:
2784                                  *  - a page was submitted by osc_io_submit, so
2785                                  *    page locked;
2786                                  *  - no request in flight
2787                                  *  - no subsequent request
2788                                  * The system will be in live-lock state,
2789                                  * because there is no chance to call
2790                                  * osc_io_unplug() and osc_check_rpcs() any
2791                                  * more. pdflush can't help in this case,
2792                                  * because it might be blocked at grabbing
2793                                  * the page lock as we mentioned.
2794                                  *
2795                                  * Anyway, continue to drain pages. */
2796                                 /* break; */
2797                         }
2798
2799                         if (rc > 0)
2800                                 race_counter = 0;
2801                         else
2802                                 race_counter++;
2803                 }
2804                 if (lop_makes_rpc(cli, &loi->loi_read_lop, OBD_BRW_READ)) {
2805                         rc = osc_send_oap_rpc(env, cli, loi, OBD_BRW_READ,
2806                                               &loi->loi_read_lop);
2807                         if (rc < 0)
2808                                 CERROR("Read request failed with %d\n", rc);
2809
2810                         if (rc > 0)
2811                                 race_counter = 0;
2812                         else
2813                                 race_counter++;
2814                 }
2815
2816                 /* attempt some inter-object balancing by issuing rpcs
2817                  * for each object in turn */
2818                 if (!cfs_list_empty(&loi->loi_hp_ready_item))
2819                         cfs_list_del_init(&loi->loi_hp_ready_item);
2820                 if (!cfs_list_empty(&loi->loi_ready_item))
2821                         cfs_list_del_init(&loi->loi_ready_item);
2822                 if (!cfs_list_empty(&loi->loi_write_item))
2823                         cfs_list_del_init(&loi->loi_write_item);
2824                 if (!cfs_list_empty(&loi->loi_read_item))
2825                         cfs_list_del_init(&loi->loi_read_item);
2826                 if (!cfs_list_empty(&loi->loi_sync_fs_item))
2827                         cfs_list_del_init(&loi->loi_sync_fs_item);
2828
2829                 loi_list_maint(cli, loi);
2830
2831                 /* send_oap_rpc fails with 0 when make_ready tells it to
2832                  * back off.  llite's make_ready does this when it tries
2833                  * to lock a page queued for write that is already locked.
2834                  * we want to try sending rpcs from many objects, but we
2835                  * don't want to spin failing with 0.  */
2836                 if (race_counter == 10)
2837                         break;
2838         }
2839         EXIT;
2840 }
2841
2842 /* we're trying to queue a page in the osc so we're subject to the
2843  * 'cl_dirty_max' limit on the number of pages that can be queued in the osc.
2844  * If the osc's queued pages are already at that limit, then we want to sleep
2845  * until there is space in the osc's queue for us.  We also may be waiting for
2846  * write credits from the OST if there are RPCs in flight that may return some
2847  * before we fall back to sync writes.
2848  *
2849  * We need this know our allocation was granted in the presence of signals */
2850 static int ocw_granted(struct client_obd *cli, struct osc_cache_waiter *ocw)
2851 {
2852         int rc;
2853         ENTRY;
2854         client_obd_list_lock(&cli->cl_loi_list_lock);
2855         rc = cfs_list_empty(&ocw->ocw_entry) || rpcs_in_flight(cli) == 0;
2856         client_obd_list_unlock(&cli->cl_loi_list_lock);
2857         RETURN(rc);
2858 };
2859
2860 /**
2861  * Non-blocking version of osc_enter_cache() that consumes grant only when it
2862  * is available.
2863  */
2864 int osc_enter_cache_try(const struct lu_env *env,
2865                         struct client_obd *cli, struct lov_oinfo *loi,
2866                         struct osc_async_page *oap, int transient)
2867 {
2868         int has_grant;
2869
2870         has_grant = cli->cl_avail_grant >= CFS_PAGE_SIZE;
2871         if (has_grant) {
2872                 osc_consume_write_grant(cli, &oap->oap_brw_page);
2873                 if (transient) {
2874                         cli->cl_dirty_transit += CFS_PAGE_SIZE;
2875                         cfs_atomic_inc(&obd_dirty_transit_pages);
2876                         oap->oap_brw_flags |= OBD_BRW_NOCACHE;
2877                 }
2878         }
2879         return has_grant;
2880 }
2881
2882 /* Caller must hold loi_list_lock - we drop/regain it if we need to wait for
2883  * grant or cache space. */
2884 static int osc_enter_cache(const struct lu_env *env,
2885                            struct client_obd *cli, struct lov_oinfo *loi,
2886                            struct osc_async_page *oap)
2887 {
2888         struct osc_cache_waiter ocw;
2889         struct l_wait_info lwi = { 0 };
2890
2891         ENTRY;
2892
2893         CDEBUG(D_CACHE, "dirty: %ld/%d dirty_max: %ld/%d dropped: %lu "
2894                "grant: %lu\n", cli->cl_dirty, cfs_atomic_read(&obd_dirty_pages),
2895                cli->cl_dirty_max, obd_max_dirty_pages,
2896                cli->cl_lost_grant, cli->cl_avail_grant);
2897
2898         /* force the caller to try sync io.  this can jump the list
2899          * of queued writes and create a discontiguous rpc stream */
2900         if (cli->cl_dirty_max < CFS_PAGE_SIZE || cli->cl_ar.ar_force_sync ||
2901             loi->loi_ar.ar_force_sync)
2902                 RETURN(-EDQUOT);
2903
2904         /* Hopefully normal case - cache space and write credits available */
2905         if (cli->cl_dirty + CFS_PAGE_SIZE <= cli->cl_dirty_max &&
2906             cfs_atomic_read(&obd_dirty_pages) + 1 <= obd_max_dirty_pages &&
2907             osc_enter_cache_try(env, cli, loi, oap, 0))
2908                 RETURN(0);
2909
2910         /* It is safe to block as a cache waiter as long as there is grant
2911          * space available or the hope of additional grant being returned
2912          * when an in flight write completes.  Using the write back cache
2913          * if possible is preferable to sending the data synchronously
2914          * because write pages can then be merged in to large requests.
2915          * The addition of this cache waiter will causing pending write
2916          * pages to be sent immediately. */
2917         if (cli->cl_w_in_flight || cli->cl_avail_grant >= CFS_PAGE_SIZE) {
2918                 cfs_list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters);
2919                 cfs_waitq_init(&ocw.ocw_waitq);
2920                 ocw.ocw_oap = oap;
2921                 ocw.ocw_rc = 0;
2922
2923                 loi_list_maint(cli, loi);
2924                 osc_check_rpcs(env, cli);
2925                 client_obd_list_unlock(&cli->cl_loi_list_lock);
2926
2927                 CDEBUG(D_CACHE, "sleeping for cache space\n");
2928                 l_wait_event(ocw.ocw_waitq, ocw_granted(cli, &ocw), &lwi);
2929
2930                 client_obd_list_lock(&cli->cl_loi_list_lock);
2931                 if (!cfs_list_empty(&ocw.ocw_entry)) {
2932                         cfs_list_del(&ocw.ocw_entry);
2933                         RETURN(-EINTR);
2934                 }
2935                 RETURN(ocw.ocw_rc);
2936         }
2937
2938         RETURN(-EDQUOT);
2939 }
2940
2941
2942 int osc_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
2943                         struct lov_oinfo *loi, cfs_page_t *page,
2944                         obd_off offset, const struct obd_async_page_ops *ops,
2945                         void *data, void **res, int nocache,
2946                         struct lustre_handle *lockh)
2947 {
2948         struct osc_async_page *oap;
2949
2950         ENTRY;
2951
2952         if (!page)
2953                 return cfs_size_round(sizeof(*oap));
2954
2955         oap = *res;
2956         oap->oap_magic = OAP_MAGIC;
2957         oap->oap_cli = &exp->exp_obd->u.cli;
2958         oap->oap_loi = loi;
2959
2960         oap->oap_caller_ops = ops;
2961         oap->oap_caller_data = data;
2962
2963         oap->oap_page = page;
2964         oap->oap_obj_off = offset;
2965         if (!client_is_remote(exp) &&
2966             cfs_capable(CFS_CAP_SYS_RESOURCE))
2967                 oap->oap_brw_flags = OBD_BRW_NOQUOTA;
2968
2969         LASSERT(!(offset & ~CFS_PAGE_MASK));
2970
2971         CFS_INIT_LIST_HEAD(&oap->oap_pending_item);
2972         CFS_INIT_LIST_HEAD(&oap->oap_urgent_item);
2973         CFS_INIT_LIST_HEAD(&oap->oap_rpc_item);
2974         CFS_INIT_LIST_HEAD(&oap->oap_page_list);
2975
2976         cfs_spin_lock_init(&oap->oap_lock);
2977         CDEBUG(D_CACHE, "oap %p page %p obj off "LPU64"\n", oap, page, offset);
2978         RETURN(0);
2979 }
2980
2981 struct osc_async_page *oap_from_cookie(void *cookie)
2982 {
2983         struct osc_async_page *oap = cookie;
2984         if (oap->oap_magic != OAP_MAGIC)
2985                 return ERR_PTR(-EINVAL);
2986         return oap;
2987 };
2988
2989 int osc_queue_async_io(const struct lu_env *env,
2990                        struct obd_export *exp, struct lov_stripe_md *lsm,
2991                        struct lov_oinfo *loi, void *cookie,
2992                        int cmd, obd_off off, int count,
2993                        obd_flag brw_flags, enum async_flags async_flags)
2994 {
2995         struct client_obd *cli = &exp->exp_obd->u.cli;
2996         struct osc_async_page *oap;
2997         int rc = 0;
2998         ENTRY;
2999
3000         oap = oap_from_cookie(cookie);
3001         if (IS_ERR(oap))
3002                 RETURN(PTR_ERR(oap));
3003
3004         if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
3005                 RETURN(-EIO);
3006
3007         if (!cfs_list_empty(&oap->oap_pending_item) ||
3008             !cfs_list_empty(&oap->oap_urgent_item) ||
3009             !cfs_list_empty(&oap->oap_rpc_item))
3010                 RETURN(-EBUSY);
3011
3012         /* check if the file's owner/group is over quota */
3013         if ((cmd & OBD_BRW_WRITE) && !(cmd & OBD_BRW_NOQUOTA)) {
3014                 struct cl_object *obj;
3015                 struct cl_attr    attr; /* XXX put attr into thread info */
3016                 unsigned int qid[MAXQUOTAS];
3017
3018                 obj = cl_object_top(osc_oap2cl_page(oap)->cp_obj);
3019
3020                 cl_object_attr_lock(obj);
3021                 rc = cl_object_attr_get(env, obj, &attr);
3022                 cl_object_attr_unlock(obj);
3023
3024                 qid[USRQUOTA] = attr.cat_uid;
3025                 qid[GRPQUOTA] = attr.cat_gid;
3026                 if (rc == 0 &&
3027                     lquota_chkdq(quota_interface, cli, qid) == NO_QUOTA)
3028                         rc = -EDQUOT;
3029                 if (rc)
3030                         RETURN(rc);
3031         }
3032
3033         if (loi == NULL)
3034                 loi = lsm->lsm_oinfo[0];
3035
3036         client_obd_list_lock(&cli->cl_loi_list_lock);
3037
3038         LASSERT(off + count <= CFS_PAGE_SIZE);
3039         oap->oap_cmd = cmd;
3040         oap->oap_page_off = off;
3041         oap->oap_count = count;
3042         oap->oap_brw_flags = brw_flags;
3043         /* Give a hint to OST that requests are coming from kswapd - bug19529 */
3044         if (cfs_memory_pressure_get())
3045                 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
3046         cfs_spin_lock(&oap->oap_lock);
3047         oap->oap_async_flags = async_flags;
3048         cfs_spin_unlock(&oap->oap_lock);
3049
3050         if (cmd & OBD_BRW_WRITE) {
3051                 rc = osc_enter_cache(env, cli, loi, oap);
3052                 if (rc) {
3053                         client_obd_list_unlock(&cli->cl_loi_list_lock);
3054                         RETURN(rc);
3055                 }
3056         }
3057
3058         osc_oap_to_pending(oap);
3059         loi_list_maint(cli, loi);
3060
3061         LOI_DEBUG(loi, "oap %p page %p added for cmd %d\n", oap, oap->oap_page,
3062                   cmd);
3063
3064         osc_check_rpcs(env, cli);
3065         client_obd_list_unlock(&cli->cl_loi_list_lock);
3066
3067         RETURN(0);
3068 }
3069
3070 /* aka (~was & now & flag), but this is more clear :) */
3071 #define SETTING(was, now, flag) (!(was & flag) && (now & flag))
3072
3073 int osc_set_async_flags_base(struct client_obd *cli,
3074                              struct lov_oinfo *loi, struct osc_async_page *oap,
3075                              obd_flag async_flags)
3076 {
3077         struct loi_oap_pages *lop;
3078         int flags = 0;
3079         ENTRY;
3080
3081         LASSERT(!cfs_list_empty(&oap->oap_pending_item));
3082
3083         if (oap->oap_cmd & OBD_BRW_WRITE) {
3084                 lop = &loi->loi_write_lop;
3085         } else {
3086                 lop = &loi->loi_read_lop;
3087         }
3088
3089         if ((oap->oap_async_flags & async_flags) == async_flags)
3090                 RETURN(0);
3091
3092         /* XXX: This introduces a tiny insignificant race for the case if this
3093          * loi already had other urgent items.
3094          */
3095         if (SETTING(oap->oap_async_flags, async_flags, ASYNC_SYNCFS) &&
3096             cfs_list_empty(&oap->oap_rpc_item) &&
3097             cfs_list_empty(&oap->oap_urgent_item)) {
3098                 cfs_list_add_tail(&oap->oap_urgent_item, &lop->lop_urgent);
3099                 flags |= ASYNC_SYNCFS;
3100                 cfs_spin_lock(&oap->oap_lock);
3101                 oap->oap_async_flags |= flags;
3102                 cfs_spin_unlock(&oap->oap_lock);
3103                 loi_list_maint(cli, loi);
3104                 RETURN(0);
3105         }
3106
3107         if (SETTING(oap->oap_async_flags, async_flags, ASYNC_READY))
3108                 flags |= ASYNC_READY;
3109
3110         if (SETTING(oap->oap_async_flags, async_flags, ASYNC_URGENT) &&
3111             cfs_list_empty(&oap->oap_rpc_item)) {
3112                 if (oap->oap_async_flags & ASYNC_HP)
3113                         cfs_list_add(&oap->oap_urgent_item, &lop->lop_urgent);
3114                 else
3115                         cfs_list_add_tail(&oap->oap_urgent_item,
3116                                           &lop->lop_urgent);
3117                 flags |= ASYNC_URGENT;
3118                 loi_list_maint(cli, loi);
3119         }
3120         cfs_spin_lock(&oap->oap_lock);
3121         oap->oap_async_flags |= flags;
3122         cfs_spin_unlock(&oap->oap_lock);
3123
3124         LOI_DEBUG(loi, "oap %p page %p has flags %x\n", oap, oap->oap_page,
3125                         oap->oap_async_flags);
3126         RETURN(0);
3127 }
3128
3129 int osc_teardown_async_page(struct obd_export *exp,
3130                             struct lov_stripe_md *lsm,
3131                             struct lov_oinfo *loi, void *cookie)
3132 {
3133         struct client_obd *cli = &exp->exp_obd->u.cli;
3134         struct loi_oap_pages *lop;
3135         struct osc_async_page *oap;
3136         int rc = 0;
3137         ENTRY;
3138
3139         oap = oap_from_cookie(cookie);
3140         if (IS_ERR(oap))
3141                 RETURN(PTR_ERR(oap));
3142
3143         if (loi == NULL)
3144                 loi = lsm->lsm_oinfo[0];
3145
3146         if (oap->oap_cmd & OBD_BRW_WRITE) {
3147                 lop = &loi->loi_write_lop;
3148         } else {
3149                 lop = &loi->loi_read_lop;
3150         }
3151
3152         client_obd_list_lock(&cli->cl_loi_list_lock);
3153
3154         if (!cfs_list_empty(&oap->oap_rpc_item))
3155                 GOTO(out, rc = -EBUSY);
3156
3157         osc_exit_cache(cli, oap, 0);
3158         osc_wake_cache_waiters(cli);
3159
3160         if (!cfs_list_empty(&oap->oap_urgent_item)) {
3161                 cfs_list_del_init(&oap->oap_urgent_item);
3162                 cfs_spin_lock(&oap->oap_lock);
3163                 oap->oap_async_flags &= ~(ASYNC_URGENT | ASYNC_HP |
3164                                           ASYNC_SYNCFS);
3165                 cfs_spin_unlock(&oap->oap_lock);
3166         }
3167         if (!cfs_list_empty(&oap->oap_pending_item)) {
3168                 cfs_list_del_init(&oap->oap_pending_item);
3169                 lop_update_pending(cli, lop, oap->oap_cmd, -1);
3170         }
3171         loi_list_maint(cli, loi);
3172         LOI_DEBUG(loi, "oap %p page %p torn down\n", oap, oap->oap_page);
3173 out:
3174         client_obd_list_unlock(&cli->cl_loi_list_lock);
3175         RETURN(rc);
3176 }
3177
3178 static void osc_set_lock_data_with_check(struct ldlm_lock *lock,
3179                                          struct ldlm_enqueue_info *einfo,
3180                                          int flags)
3181 {
3182         void *data = einfo->ei_cbdata;
3183
3184         LASSERT(lock != NULL);
3185         LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl);
3186         LASSERT(lock->l_resource->lr_type == einfo->ei_type);
3187         LASSERT(lock->l_completion_ast == einfo->ei_cb_cp);
3188         LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
3189
3190         lock_res_and_lock(lock);
3191         cfs_spin_lock(&osc_ast_guard);
3192         LASSERT(lock->l_ast_data == NULL || lock->l_ast_data == data);
3193         lock->l_ast_data = data;
3194         cfs_spin_unlock(&osc_ast_guard);
3195         unlock_res_and_lock(lock);
3196 }
3197
3198 static void osc_set_data_with_check(struct lustre_handle *lockh,
3199                                     struct ldlm_enqueue_info *einfo,
3200                                     int flags)
3201 {
3202         struct ldlm_lock *lock = ldlm_handle2lock(lockh);
3203
3204         if (lock != NULL) {
3205                 osc_set_lock_data_with_check(lock, einfo, flags);
3206                 LDLM_LOCK_PUT(lock);
3207         } else
3208                 CERROR("lockh %p, data %p - client evicted?\n",
3209                        lockh, einfo->ei_cbdata);
3210 }
3211
3212 static int osc_change_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
3213                              ldlm_iterator_t replace, void *data)
3214 {
3215         struct ldlm_res_id res_id;
3216         struct obd_device *obd = class_exp2obd(exp);
3217
3218         osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_seq, &res_id);
3219         ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
3220         return 0;
3221 }
3222
3223 /* find any ldlm lock of the inode in osc
3224  * return 0    not find
3225  *        1    find one
3226  *      < 0    error */
3227 static int osc_find_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
3228                            ldlm_iterator_t replace, void *data)
3229 {
3230         struct ldlm_res_id res_id;
3231         struct obd_device *obd = class_exp2obd(exp);
3232         int rc = 0;
3233
3234         osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_seq, &res_id);
3235         rc = ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
3236         if (rc == LDLM_ITER_STOP)
3237                 return(1);
3238         if (rc == LDLM_ITER_CONTINUE)
3239                 return(0);
3240         return(rc);
3241 }
3242
3243 static int osc_enqueue_fini(struct ptlrpc_request *req, struct ost_lvb *lvb,
3244                             obd_enqueue_update_f upcall, void *cookie,
3245                             int *flags, int rc)
3246 {
3247         int intent = *flags & LDLM_FL_HAS_INTENT;
3248         ENTRY;
3249
3250         if (intent) {
3251                 /* The request was created before ldlm_cli_enqueue call. */
3252                 if (rc == ELDLM_LOCK_ABORTED) {
3253                         struct ldlm_reply *rep;
3254                         rep = req_capsule_server_get(&req->rq_pill,
3255                                                      &RMF_DLM_REP);
3256
3257                         LASSERT(rep != NULL);
3258                         if (rep->lock_policy_res1)
3259                                 rc = rep->lock_policy_res1;
3260                 }
3261         }
3262
3263         if ((intent && rc == ELDLM_LOCK_ABORTED) || !rc) {
3264                 *flags |= LDLM_FL_LVB_READY;
3265                 CDEBUG(D_INODE,"got kms "LPU64" blocks "LPU64" mtime "LPU64"\n",
3266                        lvb->lvb_size, lvb->lvb_blocks, lvb->lvb_mtime);
3267         }
3268
3269         /* Call the update callback. */
3270         rc = (*upcall)(cookie, rc);
3271         RETURN(rc);
3272 }
3273
3274 static int osc_enqueue_interpret(const struct lu_env *env,
3275                                  struct ptlrpc_request *req,
3276                                  struct osc_enqueue_args *aa, int rc)
3277 {
3278         struct ldlm_lock *lock;
3279         struct lustre_handle handle;
3280         __u32 mode;
3281
3282         /* Make a local copy of a lock handle and a mode, because aa->oa_*
3283          * might be freed anytime after lock upcall has been called. */
3284         lustre_handle_copy(&handle, aa->oa_lockh);
3285         mode = aa->oa_ei->ei_mode;
3286
3287         /* ldlm_cli_enqueue is holding a reference on the lock, so it must
3288          * be valid. */
3289         lock = ldlm_handle2lock(&handle);
3290
3291         /* Take an additional reference so that a blocking AST that
3292          * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
3293          * to arrive after an upcall has been executed by
3294          * osc_enqueue_fini(). */
3295         ldlm_lock_addref(&handle, mode);
3296
3297         /* Let CP AST to grant the lock first. */
3298         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
3299
3300         /* Complete obtaining the lock procedure. */
3301         rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1,
3302                                    mode, aa->oa_flags, aa->oa_lvb,
3303                                    sizeof(*aa->oa_lvb), &handle, rc);
3304         /* Complete osc stuff. */
3305         rc = osc_enqueue_fini(req, aa->oa_lvb,
3306                               aa->oa_upcall, aa->oa_cookie, aa->oa_flags, rc);
3307
3308         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
3309
3310         /* Release the lock for async request. */
3311         if (lustre_handle_is_used(&handle) && rc == ELDLM_OK)
3312                 /*
3313                  * Releases a reference taken by ldlm_cli_enqueue(), if it is
3314                  * not already released by
3315                  * ldlm_cli_enqueue_fini()->failed_lock_cleanup()
3316                  */
3317                 ldlm_lock_decref(&handle, mode);
3318
3319         LASSERTF(lock != NULL, "lockh %p, req %p, aa %p - client evicted?\n",
3320                  aa->oa_lockh, req, aa);
3321         ldlm_lock_decref(&handle, mode);
3322         LDLM_LOCK_PUT(lock);
3323         return rc;
3324 }
3325
3326 void osc_update_enqueue(struct lustre_handle *lov_lockhp,
3327                         struct lov_oinfo *loi, int flags,
3328                         struct ost_lvb *lvb, __u32 mode, int rc)
3329 {
3330         if (rc == ELDLM_OK) {
3331                 struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
3332                 __u64 tmp;
3333
3334                 LASSERT(lock != NULL);
3335                 loi->loi_lvb = *lvb;
3336                 tmp = loi->loi_lvb.lvb_size;
3337                 /* Extend KMS up to the end of this lock and no further
3338                  * A lock on [x,y] means a KMS of up to y + 1 bytes! */
3339                 if (tmp > lock->l_policy_data.l_extent.end)
3340                         tmp = lock->l_policy_data.l_extent.end + 1;
3341                 if (tmp >= loi->loi_kms) {
3342                         LDLM_DEBUG(lock, "lock acquired, setting rss="LPU64
3343                                    ", kms="LPU64, loi->loi_lvb.lvb_size, tmp);
3344                         loi_kms_set(loi, tmp);
3345                 } else {
3346                         LDLM_DEBUG(lock, "lock acquired, setting rss="
3347                                    LPU64"; leaving kms="LPU64", end="LPU64,
3348                                    loi->loi_lvb.lvb_size, loi->loi_kms,
3349                                    lock->l_policy_data.l_extent.end);
3350                 }
3351                 ldlm_lock_allow_match(lock);
3352                 LDLM_LOCK_PUT(lock);
3353         } else if (rc == ELDLM_LOCK_ABORTED && (flags & LDLM_FL_HAS_INTENT)) {
3354                 loi->loi_lvb = *lvb;
3355                 CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
3356                        " kms="LPU64"\n", loi->loi_lvb.lvb_size, loi->loi_kms);
3357                 rc = ELDLM_OK;
3358         }
3359 }
3360 EXPORT_SYMBOL(osc_update_enqueue);
3361
3362 struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
3363
3364 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
3365  * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
3366  * other synchronous requests, however keeping some locks and trying to obtain
3367  * others may take a considerable amount of time in a case of ost failure; and
3368  * when other sync requests do not get released lock from a client, the client
3369  * is excluded from the cluster -- such scenarious make the life difficult, so
3370  * release locks just after they are obtained. */
3371 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
3372                      int *flags, ldlm_policy_data_t *policy,
3373                      struct ost_lvb *lvb, int kms_valid,
3374                      obd_enqueue_update_f upcall, void *cookie,
3375                      struct ldlm_enqueue_info *einfo,
3376                      struct lustre_handle *lockh,
3377                      struct ptlrpc_request_set *rqset, int async)
3378 {
3379         struct obd_device *obd = exp->exp_obd;
3380         struct ptlrpc_request *req = NULL;
3381         int intent = *flags & LDLM_FL_HAS_INTENT;
3382         ldlm_mode_t mode;
3383         int rc;
3384         ENTRY;
3385
3386         /* Filesystem lock extents are extended to page boundaries so that
3387          * dealing with the page cache is a little smoother.  */
3388         policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
3389         policy->l_extent.end |= ~CFS_PAGE_MASK;
3390
3391         /*
3392          * kms is not valid when either object is completely fresh (so that no
3393          * locks are cached), or object was evicted. In the latter case cached
3394          * lock cannot be used, because it would prime inode state with
3395          * potentially stale LVB.
3396          */
3397         if (!kms_valid)
3398                 goto no_match;
3399
3400         /* Next, search for already existing extent locks that will cover us */
3401         /* If we're trying to read, we also search for an existing PW lock.  The
3402          * VFS and page cache already protect us locally, so lots of readers/
3403          * writers can share a single PW lock.
3404          *
3405          * There are problems with conversion deadlocks, so instead of
3406          * converting a read lock to a write lock, we'll just enqueue a new
3407          * one.
3408          *
3409          * At some point we should cancel the read lock instead of making them
3410          * send us a blocking callback, but there are problems with canceling
3411          * locks out from other users right now, too. */
3412         mode = einfo->ei_mode;
3413         if (einfo->ei_mode == LCK_PR)
3414                 mode |= LCK_PW;
3415         mode = ldlm_lock_match(obd->obd_namespace,
3416                                *flags | LDLM_FL_LVB_READY, res_id,
3417                                einfo->ei_type, policy, mode, lockh, 0);
3418         if (mode) {
3419                 struct ldlm_lock *matched = ldlm_handle2lock(lockh);
3420
3421                 if (matched->l_ast_data == NULL ||
3422                     matched->l_ast_data == einfo->ei_cbdata) {
3423                         /* addref the lock only if not async requests and PW
3424                          * lock is matched whereas we asked for PR. */
3425                         if (!rqset && einfo->ei_mode != mode)
3426                                 ldlm_lock_addref(lockh, LCK_PR);
3427                         osc_set_lock_data_with_check(matched, einfo, *flags);
3428                         if (intent) {
3429                                 /* I would like to be able to ASSERT here that
3430                                  * rss <= kms, but I can't, for reasons which
3431                                  * are explained in lov_enqueue() */
3432                         }
3433
3434                         /* We already have a lock, and it's referenced */
3435                         (*upcall)(cookie, ELDLM_OK);
3436
3437                         /* For async requests, decref the lock. */
3438                         if (einfo->ei_mode != mode)
3439                                 ldlm_lock_decref(lockh, LCK_PW);
3440                         else if (rqset)
3441                                 ldlm_lock_decref(lockh, einfo->ei_mode);
3442                         LDLM_LOCK_PUT(matched);
3443                         RETURN(ELDLM_OK);
3444                 } else
3445                         ldlm_lock_decref(lockh, mode);
3446                 LDLM_LOCK_PUT(matched);
3447         }
3448
3449  no_match:
3450         if (intent) {
3451                 CFS_LIST_HEAD(cancels);
3452                 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
3453                                            &RQF_LDLM_ENQUEUE_LVB);
3454                 if (req == NULL)
3455                         RETURN(-ENOMEM);
3456
3457                 rc = ldlm_prep_enqueue_req(exp, req, &cancels, 0);
3458                 if (rc) {
3459                         ptlrpc_request_free(req);
3460                         RETURN(rc);
3461                 }
3462
3463                 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
3464                                      sizeof *lvb);
3465                 ptlrpc_request_set_replen(req);
3466         }
3467
3468         /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */
3469         *flags &= ~LDLM_FL_BLOCK_GRANTED;
3470
3471         rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
3472                               sizeof(*lvb), lockh, async);
3473         if (rqset) {
3474                 if (!rc) {
3475                         struct osc_enqueue_args *aa;
3476                         CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
3477                         aa = ptlrpc_req_async_args(req);
3478                         aa->oa_ei = einfo;
3479                         aa->oa_exp = exp;
3480                         aa->oa_flags  = flags;
3481                         aa->oa_upcall = upcall;
3482                         aa->oa_cookie = cookie;
3483                         aa->oa_lvb    = lvb;
3484                         aa->oa_lockh  = lockh;
3485
3486                         req->rq_interpret_reply =
3487                                 (ptlrpc_interpterer_t)osc_enqueue_interpret;
3488                         if (rqset == PTLRPCD_SET)
3489                                 ptlrpcd_add_req(req, PSCOPE_OTHER);
3490                         else
3491                                 ptlrpc_set_add_req(rqset, req);
3492                 } else if (intent) {
3493                         ptlrpc_req_finished(req);
3494                 }
3495                 RETURN(rc);
3496         }
3497
3498         rc = osc_enqueue_fini(req, lvb, upcall, cookie, flags, rc);
3499         if (intent)
3500                 ptlrpc_req_finished(req);
3501
3502         RETURN(rc);
3503 }
3504
3505 static int osc_enqueue(struct obd_export *exp, struct obd_info *oinfo,
3506                        struct ldlm_enqueue_info *einfo,
3507                        struct ptlrpc_request_set *rqset)
3508 {
3509         struct ldlm_res_id res_id;
3510         int rc;
3511         ENTRY;
3512
3513         osc_build_res_name(oinfo->oi_md->lsm_object_id,
3514                            oinfo->oi_md->lsm_object_seq, &res_id);
3515
3516         rc = osc_enqueue_base(exp, &res_id, &oinfo->oi_flags, &oinfo->oi_policy,
3517                               &oinfo->oi_md->lsm_oinfo[0]->loi_lvb,
3518                               oinfo->oi_md->lsm_oinfo[0]->loi_kms_valid,
3519                               oinfo->oi_cb_up, oinfo, einfo, oinfo->oi_lockh,
3520                               rqset, rqset != NULL);
3521         RETURN(rc);
3522 }
3523
3524 int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
3525                    __u32 type, ldlm_policy_data_t *policy, __u32 mode,
3526                    int *flags, void *data, struct lustre_handle *lockh,
3527                    int unref)
3528 {
3529         struct obd_device *obd = exp->exp_obd;
3530         int lflags = *flags;
3531         ldlm_mode_t rc;
3532         ENTRY;
3533
3534         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
3535                 RETURN(-EIO);
3536
3537         /* Filesystem lock extents are extended to page boundaries so that
3538          * dealing with the page cache is a little smoother */
3539         policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK;
3540         policy->l_extent.end |= ~CFS_PAGE_MASK;
3541
3542         /* Next, search for already existing extent locks that will cover us */
3543         /* If we're trying to read, we also search for an existing PW lock.  The
3544          * VFS and page cache already protect us locally, so lots of readers/
3545          * writers can share a single PW lock. */
3546         rc = mode;
3547         if (mode == LCK_PR)
3548                 rc |= LCK_PW;
3549         rc = ldlm_lock_match(obd->obd_namespace, lflags,
3550                              res_id, type, policy, rc, lockh, unref);
3551         if (rc) {
3552                 if (data != NULL)
3553                         osc_set_data_with_check(lockh, data, lflags);
3554                 if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) {
3555                         ldlm_lock_addref(lockh, LCK_PR);
3556                         ldlm_lock_decref(lockh, LCK_PW);
3557                 }
3558                 RETURN(rc);
3559         }
3560         RETURN(rc);
3561 }
3562
3563 int osc_cancel_base(struct lustre_handle *lockh, __u32 mode)
3564 {
3565         ENTRY;
3566
3567         if (unlikely(mode == LCK_GROUP))
3568                 ldlm_lock_decref_and_cancel(lockh, mode);
3569         else
3570                 ldlm_lock_decref(lockh, mode);
3571
3572         RETURN(0);
3573 }
3574
3575 static int osc_cancel(struct obd_export *exp, struct lov_stripe_md *md,
3576                       __u32 mode, struct lustre_handle *lockh)
3577 {
3578         ENTRY;
3579         RETURN(osc_cancel_base(lockh, mode));
3580 }
3581
3582 static int osc_cancel_unused(struct obd_export *exp,
3583                              struct lov_stripe_md *lsm,
3584                              ldlm_cancel_flags_t flags,
3585                              void *opaque)
3586 {
3587         struct obd_device *obd = class_exp2obd(exp);
3588         struct ldlm_res_id res_id, *resp = NULL;
3589
3590         if (lsm != NULL) {
3591                 resp = osc_build_res_name(lsm->lsm_object_id,
3592                                           lsm->lsm_object_seq, &res_id);
3593         }
3594
3595         return ldlm_cli_cancel_unused(obd->obd_namespace, resp, flags, opaque);
3596 }
3597
3598 static int osc_statfs_interpret(const struct lu_env *env,
3599                                 struct ptlrpc_request *req,
3600                                 struct osc_async_args *aa, int rc)
3601 {
3602         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
3603         struct obd_statfs *msfs;
3604         __u64 used;
3605         ENTRY;
3606
3607         if (rc == -EBADR)
3608                 /* The request has in fact never been sent
3609                  * due to issues at a higher level (LOV).
3610                  * Exit immediately since the caller is
3611                  * aware of the problem and takes care
3612                  * of the clean up */
3613                  RETURN(rc);
3614
3615         if ((rc == -ENOTCONN || rc == -EAGAIN) &&
3616             (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
3617                 GOTO(out, rc = 0);
3618
3619         if (rc != 0)
3620                 GOTO(out, rc);
3621
3622         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
3623         if (msfs == NULL) {
3624                 GOTO(out, rc = -EPROTO);
3625         }
3626
3627         /* Reinitialize the RDONLY and DEGRADED flags at the client
3628          * on each statfs, so they don't stay set permanently. */
3629         cfs_spin_lock(&cli->cl_oscc.oscc_lock);
3630
3631         if (unlikely(msfs->os_state & OS_STATE_DEGRADED))
3632                 cli->cl_oscc.oscc_flags |= OSCC_FLAG_DEGRADED;
3633         else if (unlikely(cli->cl_oscc.oscc_flags & OSCC_FLAG_DEGRADED))
3634                 cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_DEGRADED;
3635
3636         if (unlikely(msfs->os_state & OS_STATE_READONLY))
3637                 cli->cl_oscc.oscc_flags |= OSCC_FLAG_RDONLY;
3638         else if (unlikely(cli->cl_oscc.oscc_flags & OSCC_FLAG_RDONLY))
3639                 cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_RDONLY;
3640
3641         /* Add a bit of hysteresis so this flag isn't continually flapping,
3642          * and ensure that new files don't get extremely fragmented due to
3643          * only a small amount of available space in the filesystem.
3644          * We want t