Whamcloud - gitweb
b=22176 Add .sync_fs super block handler
[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         oa->o_flags |= OBD_FL_SHRINK_GRANT;
970         osc_update_next_shrink(cli);
971 }
972
973 /* Shrink the current grant, either from some large amount to enough for a
974  * full set of in-flight RPCs, or if we have already shrunk to that limit
975  * then to enough for a single RPC.  This avoids keeping more grant than
976  * needed, and avoids shrinking the grant piecemeal. */
977 static int osc_shrink_grant(struct client_obd *cli)
978 {
979         long target = (cli->cl_max_rpcs_in_flight + 1) *
980                       cli->cl_max_pages_per_rpc;
981
982         client_obd_list_lock(&cli->cl_loi_list_lock);
983         if (cli->cl_avail_grant <= target)
984                 target = cli->cl_max_pages_per_rpc;
985         client_obd_list_unlock(&cli->cl_loi_list_lock);
986
987         return osc_shrink_grant_to_target(cli, target);
988 }
989
990 int osc_shrink_grant_to_target(struct client_obd *cli, long target)
991 {
992         int    rc = 0;
993         struct ost_body     *body;
994         ENTRY;
995
996         client_obd_list_lock(&cli->cl_loi_list_lock);
997         /* Don't shrink if we are already above or below the desired limit
998          * We don't want to shrink below a single RPC, as that will negatively
999          * impact block allocation and long-term performance. */
1000         if (target < cli->cl_max_pages_per_rpc)
1001                 target = cli->cl_max_pages_per_rpc;
1002
1003         if (target >= cli->cl_avail_grant) {
1004                 client_obd_list_unlock(&cli->cl_loi_list_lock);
1005                 RETURN(0);
1006         }
1007         client_obd_list_unlock(&cli->cl_loi_list_lock);
1008
1009         OBD_ALLOC_PTR(body);
1010         if (!body)
1011                 RETURN(-ENOMEM);
1012
1013         osc_announce_cached(cli, &body->oa, 0);
1014
1015         client_obd_list_lock(&cli->cl_loi_list_lock);
1016         body->oa.o_grant = cli->cl_avail_grant - target;
1017         cli->cl_avail_grant = target;
1018         client_obd_list_unlock(&cli->cl_loi_list_lock);
1019         body->oa.o_flags |= OBD_FL_SHRINK_GRANT;
1020         osc_update_next_shrink(cli);
1021
1022         rc = osc_set_info_async(cli->cl_import->imp_obd->obd_self_export,
1023                                 sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK,
1024                                 sizeof(*body), body, NULL);
1025         if (rc != 0)
1026                 __osc_update_grant(cli, body->oa.o_grant);
1027         OBD_FREE_PTR(body);
1028         RETURN(rc);
1029 }
1030
1031 #define GRANT_SHRINK_LIMIT PTLRPC_MAX_BRW_SIZE
1032 static int osc_should_shrink_grant(struct client_obd *client)
1033 {
1034         cfs_time_t time = cfs_time_current();
1035         cfs_time_t next_shrink = client->cl_next_shrink_grant;
1036
1037         if ((client->cl_import->imp_connect_data.ocd_connect_flags &
1038              OBD_CONNECT_GRANT_SHRINK) == 0)
1039                 return 0;
1040
1041         if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
1042                 if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
1043                     client->cl_avail_grant > GRANT_SHRINK_LIMIT)
1044                         return 1;
1045                 else
1046                         osc_update_next_shrink(client);
1047         }
1048         return 0;
1049 }
1050
1051 static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
1052 {
1053         struct client_obd *client;
1054
1055         cfs_list_for_each_entry(client, &item->ti_obd_list,
1056                                 cl_grant_shrink_list) {
1057                 if (osc_should_shrink_grant(client))
1058                         osc_shrink_grant(client);
1059         }
1060         return 0;
1061 }
1062
1063 static int osc_add_shrink_grant(struct client_obd *client)
1064 {
1065         int rc;
1066
1067         rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval,
1068                                        TIMEOUT_GRANT,
1069                                        osc_grant_shrink_grant_cb, NULL,
1070                                        &client->cl_grant_shrink_list);
1071         if (rc) {
1072                 CERROR("add grant client %s error %d\n",
1073                         client->cl_import->imp_obd->obd_name, rc);
1074                 return rc;
1075         }
1076         CDEBUG(D_CACHE, "add grant client %s \n",
1077                client->cl_import->imp_obd->obd_name);
1078         osc_update_next_shrink(client);
1079         return 0;
1080 }
1081
1082 static int osc_del_shrink_grant(struct client_obd *client)
1083 {
1084         return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list,
1085                                          TIMEOUT_GRANT);
1086 }
1087
1088 static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
1089 {
1090         /*
1091          * ocd_grant is the total grant amount we're expect to hold: if we've
1092          * been evicted, it's the new avail_grant amount, cl_dirty will drop
1093          * to 0 as inflight RPCs fail out; otherwise, it's avail_grant + dirty.
1094          *
1095          * race is tolerable here: if we're evicted, but imp_state already
1096          * left EVICTED state, then cl_dirty must be 0 already.
1097          */
1098         client_obd_list_lock(&cli->cl_loi_list_lock);
1099         if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
1100                 cli->cl_avail_grant = ocd->ocd_grant;
1101         else
1102                 cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
1103
1104         if (cli->cl_avail_grant < 0) {
1105                 CWARN("%s: available grant < 0, the OSS is probably not running"
1106                       " with patch from bug20278 (%ld) \n",
1107                       cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant);
1108                 /* workaround for 1.6 servers which do not have 
1109                  * the patch from bug20278 */
1110                 cli->cl_avail_grant = ocd->ocd_grant;
1111         }
1112
1113         client_obd_list_unlock(&cli->cl_loi_list_lock);
1114
1115         CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld \n",
1116                cli->cl_import->imp_obd->obd_name,
1117                cli->cl_avail_grant, cli->cl_lost_grant);
1118
1119         if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
1120             cfs_list_empty(&cli->cl_grant_shrink_list))
1121                 osc_add_shrink_grant(cli);
1122 }
1123
1124 /* We assume that the reason this OSC got a short read is because it read
1125  * beyond the end of a stripe file; i.e. lustre is reading a sparse file
1126  * via the LOV, and it _knows_ it's reading inside the file, it's just that
1127  * this stripe never got written at or beyond this stripe offset yet. */
1128 static void handle_short_read(int nob_read, obd_count page_count,
1129                               struct brw_page **pga)
1130 {
1131         char *ptr;
1132         int i = 0;
1133
1134         /* skip bytes read OK */
1135         while (nob_read > 0) {
1136                 LASSERT (page_count > 0);
1137
1138                 if (pga[i]->count > nob_read) {
1139                         /* EOF inside this page */
1140                         ptr = cfs_kmap(pga[i]->pg) +
1141                                 (pga[i]->off & ~CFS_PAGE_MASK);
1142                         memset(ptr + nob_read, 0, pga[i]->count - nob_read);
1143                         cfs_kunmap(pga[i]->pg);
1144                         page_count--;
1145                         i++;
1146                         break;
1147                 }
1148
1149                 nob_read -= pga[i]->count;
1150                 page_count--;
1151                 i++;
1152         }
1153
1154         /* zero remaining pages */
1155         while (page_count-- > 0) {
1156                 ptr = cfs_kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
1157                 memset(ptr, 0, pga[i]->count);
1158                 cfs_kunmap(pga[i]->pg);
1159                 i++;
1160         }
1161 }
1162
1163 static int check_write_rcs(struct ptlrpc_request *req,
1164                            int requested_nob, int niocount,
1165                            obd_count page_count, struct brw_page **pga)
1166 {
1167         int     i;
1168         __u32   *remote_rcs;
1169
1170         remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS,
1171                                                   sizeof(*remote_rcs) *
1172                                                   niocount);
1173         if (remote_rcs == NULL) {
1174                 CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n");
1175                 return(-EPROTO);
1176         }
1177
1178         /* return error if any niobuf was in error */
1179         for (i = 0; i < niocount; i++) {
1180                 if (remote_rcs[i] < 0)
1181                         return(remote_rcs[i]);
1182
1183                 if (remote_rcs[i] != 0) {
1184                         CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
1185                                 i, remote_rcs[i], req);
1186                         return(-EPROTO);
1187                 }
1188         }
1189
1190         if (req->rq_bulk->bd_nob_transferred != requested_nob) {
1191                 CERROR("Unexpected # bytes transferred: %d (requested %d)\n",
1192                        req->rq_bulk->bd_nob_transferred, requested_nob);
1193                 return(-EPROTO);
1194         }
1195
1196         return (0);
1197 }
1198
1199 static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
1200 {
1201         if (p1->flag != p2->flag) {
1202                 unsigned mask = ~(OBD_BRW_FROM_GRANT|
1203                                   OBD_BRW_NOCACHE|OBD_BRW_SYNC|OBD_BRW_ASYNC);
1204
1205                 /* warn if we try to combine flags that we don't know to be
1206                  * safe to combine */
1207                 if ((p1->flag & mask) != (p2->flag & mask))
1208                         CERROR("is it ok to have flags 0x%x and 0x%x in the "
1209                                "same brw?\n", p1->flag, p2->flag);
1210                 return 0;
1211         }
1212
1213         return (p1->off + p1->count == p2->off);
1214 }
1215
1216 static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
1217                                    struct brw_page **pga, int opc,
1218                                    cksum_type_t cksum_type)
1219 {
1220         __u32 cksum;
1221         int i = 0;
1222
1223         LASSERT (pg_count > 0);
1224         cksum = init_checksum(cksum_type);
1225         while (nob > 0 && pg_count > 0) {
1226                 unsigned char *ptr = cfs_kmap(pga[i]->pg);
1227                 int off = pga[i]->off & ~CFS_PAGE_MASK;
1228                 int count = pga[i]->count > nob ? nob : pga[i]->count;
1229
1230                 /* corrupt the data before we compute the checksum, to
1231                  * simulate an OST->client data error */
1232                 if (i == 0 && opc == OST_READ &&
1233                     OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE))
1234                         memcpy(ptr + off, "bad1", min(4, nob));
1235                 cksum = compute_checksum(cksum, ptr + off, count, cksum_type);
1236                 cfs_kunmap(pga[i]->pg);
1237                 LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d checksum %x\n",
1238                                off, cksum);
1239
1240                 nob -= pga[i]->count;
1241                 pg_count--;
1242                 i++;
1243         }
1244         /* For sending we only compute the wrong checksum instead
1245          * of corrupting the data so it is still correct on a redo */
1246         if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
1247                 cksum++;
1248
1249         return cksum;
1250 }
1251
1252 static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
1253                                 struct lov_stripe_md *lsm, obd_count page_count,
1254                                 struct brw_page **pga,
1255                                 struct ptlrpc_request **reqp,
1256                                 struct obd_capa *ocapa, int reserve)
1257 {
1258         struct ptlrpc_request   *req;
1259         struct ptlrpc_bulk_desc *desc;
1260         struct ost_body         *body;
1261         struct obd_ioobj        *ioobj;
1262         struct niobuf_remote    *niobuf;
1263         int niocount, i, requested_nob, opc, rc;
1264         struct osc_brw_async_args *aa;
1265         struct req_capsule      *pill;
1266         struct brw_page *pg_prev;
1267
1268         ENTRY;
1269         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ))
1270                 RETURN(-ENOMEM); /* Recoverable */
1271         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2))
1272                 RETURN(-EINVAL); /* Fatal */
1273
1274         if ((cmd & OBD_BRW_WRITE) != 0) {
1275                 opc = OST_WRITE;
1276                 req = ptlrpc_request_alloc_pool(cli->cl_import,
1277                                                 cli->cl_import->imp_rq_pool,
1278                                                 &RQF_OST_BRW_WRITE);
1279         } else {
1280                 opc = OST_READ;
1281                 req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ);
1282         }
1283         if (req == NULL)
1284                 RETURN(-ENOMEM);
1285
1286         for (niocount = i = 1; i < page_count; i++) {
1287                 if (!can_merge_pages(pga[i - 1], pga[i]))
1288                         niocount++;
1289         }
1290
1291         pill = &req->rq_pill;
1292         req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT,
1293                              sizeof(*ioobj));
1294         req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT,
1295                              niocount * sizeof(*niobuf));
1296         osc_set_capa_size(req, &RMF_CAPA1, ocapa);
1297
1298         rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc);
1299         if (rc) {
1300                 ptlrpc_request_free(req);
1301                 RETURN(rc);
1302         }
1303         req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
1304         ptlrpc_at_set_req_timeout(req);
1305
1306         if (opc == OST_WRITE)
1307                 desc = ptlrpc_prep_bulk_imp(req, page_count,
1308                                             BULK_GET_SOURCE, OST_BULK_PORTAL);
1309         else
1310                 desc = ptlrpc_prep_bulk_imp(req, page_count,
1311                                             BULK_PUT_SINK, OST_BULK_PORTAL);
1312
1313         if (desc == NULL)
1314                 GOTO(out, rc = -ENOMEM);
1315         /* NB request now owns desc and will free it when it gets freed */
1316
1317         body = req_capsule_client_get(pill, &RMF_OST_BODY);
1318         ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ);
1319         niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
1320         LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
1321
1322         lustre_set_wire_obdo(&body->oa, oa);
1323
1324         obdo_to_ioobj(oa, ioobj);
1325         ioobj->ioo_bufcnt = niocount;
1326         osc_pack_capa(req, body, ocapa);
1327         LASSERT (page_count > 0);
1328         pg_prev = pga[0];
1329         for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
1330                 struct brw_page *pg = pga[i];
1331
1332                 LASSERT(pg->count > 0);
1333                 LASSERTF((pg->off & ~CFS_PAGE_MASK) + pg->count <= CFS_PAGE_SIZE,
1334                          "i: %d pg: %p off: "LPU64", count: %u\n", i, pg,
1335                          pg->off, pg->count);
1336 #ifdef __linux__
1337                 LASSERTF(i == 0 || pg->off > pg_prev->off,
1338                          "i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
1339                          " prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
1340                          i, page_count,
1341                          pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
1342                          pg_prev->pg, page_private(pg_prev->pg),
1343                          pg_prev->pg->index, pg_prev->off);
1344 #else
1345                 LASSERTF(i == 0 || pg->off > pg_prev->off,
1346                          "i %d p_c %u\n", i, page_count);
1347 #endif
1348                 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
1349                         (pg->flag & OBD_BRW_SRVLOCK));
1350
1351                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->off & ~CFS_PAGE_MASK,
1352                                       pg->count);
1353                 requested_nob += pg->count;
1354
1355                 if (i > 0 && can_merge_pages(pg_prev, pg)) {
1356                         niobuf--;
1357                         niobuf->len += pg->count;
1358                 } else {
1359                         niobuf->offset = pg->off;
1360                         niobuf->len    = pg->count;
1361                         niobuf->flags  = pg->flag;
1362                 }
1363                 pg_prev = pg;
1364         }
1365
1366         LASSERTF((void *)(niobuf - niocount) ==
1367                 req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE),
1368                 "want %p - real %p\n", req_capsule_client_get(&req->rq_pill,
1369                 &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount));
1370
1371         osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0);
1372         if (osc_should_shrink_grant(cli))
1373                 osc_shrink_grant_local(cli, &body->oa);
1374
1375         /* size[REQ_REC_OFF] still sizeof (*body) */
1376         if (opc == OST_WRITE) {
1377                 if (unlikely(cli->cl_checksum) &&
1378                     !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1379                         /* store cl_cksum_type in a local variable since
1380                          * it can be changed via lprocfs */
1381                         cksum_type_t cksum_type = cli->cl_cksum_type;
1382
1383                         if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) {
1384                                 oa->o_flags &= OBD_FL_LOCAL_MASK;
1385                                 body->oa.o_flags = 0;
1386                         }
1387                         body->oa.o_flags |= cksum_type_pack(cksum_type);
1388                         body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1389                         body->oa.o_cksum = osc_checksum_bulk(requested_nob,
1390                                                              page_count, pga,
1391                                                              OST_WRITE,
1392                                                              cksum_type);
1393                         CDEBUG(D_PAGE, "checksum at write origin: %x\n",
1394                                body->oa.o_cksum);
1395                         /* save this in 'oa', too, for later checking */
1396                         oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1397                         oa->o_flags |= cksum_type_pack(cksum_type);
1398                 } else {
1399                         /* clear out the checksum flag, in case this is a
1400                          * resend but cl_checksum is no longer set. b=11238 */
1401                         oa->o_valid &= ~OBD_MD_FLCKSUM;
1402                 }
1403                 oa->o_cksum = body->oa.o_cksum;
1404                 /* 1 RC per niobuf */
1405                 req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER,
1406                                      sizeof(__u32) * niocount);
1407         } else {
1408                 if (unlikely(cli->cl_checksum) &&
1409                     !sptlrpc_flavor_has_bulk(&req->rq_flvr)) {
1410                         if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0)
1411                                 body->oa.o_flags = 0;
1412                         body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type);
1413                         body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1414                 }
1415         }
1416         ptlrpc_request_set_replen(req);
1417
1418         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
1419         aa = ptlrpc_req_async_args(req);
1420         aa->aa_oa = oa;
1421         aa->aa_requested_nob = requested_nob;
1422         aa->aa_nio_count = niocount;
1423         aa->aa_page_count = page_count;
1424         aa->aa_resends = 0;
1425         aa->aa_ppga = pga;
1426         aa->aa_cli = cli;
1427         CFS_INIT_LIST_HEAD(&aa->aa_oaps);
1428         if (ocapa && reserve)
1429                 aa->aa_ocapa = capa_get(ocapa);
1430
1431         *reqp = req;
1432         RETURN(0);
1433
1434  out:
1435         ptlrpc_req_finished(req);
1436         RETURN(rc);
1437 }
1438
1439 static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
1440                                 __u32 client_cksum, __u32 server_cksum, int nob,
1441                                 obd_count page_count, struct brw_page **pga,
1442                                 cksum_type_t client_cksum_type)
1443 {
1444         __u32 new_cksum;
1445         char *msg;
1446         cksum_type_t cksum_type;
1447
1448         if (server_cksum == client_cksum) {
1449                 CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1450                 return 0;
1451         }
1452
1453         /* If this is mmaped file - it can be changed at any time */
1454         if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & OBD_FL_MMAP)
1455                 return 1;
1456
1457         if (oa->o_valid & OBD_MD_FLFLAGS)
1458                 cksum_type = cksum_type_unpack(oa->o_flags);
1459         else
1460                 cksum_type = OBD_CKSUM_CRC32;
1461
1462         new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE,
1463                                       cksum_type);
1464
1465         if (cksum_type != client_cksum_type)
1466                 msg = "the server did not use the checksum type specified in "
1467                       "the original request - likely a protocol problem";
1468         else if (new_cksum == server_cksum)
1469                 msg = "changed on the client after we checksummed it - "
1470                       "likely false positive due to mmap IO (bug 11742)";
1471         else if (new_cksum == client_cksum)
1472                 msg = "changed in transit before arrival at OST";
1473         else
1474                 msg = "changed in transit AND doesn't match the original - "
1475                       "likely false positive due to mmap IO (bug 11742)";
1476
1477         LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
1478                            " object "LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1479                            msg, libcfs_nid2str(peer->nid),
1480                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1481                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1482                            oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1483                            oa->o_id,
1484                            oa->o_valid & OBD_MD_FLGROUP ? oa->o_seq : (__u64)0,
1485                            pga[0]->off,
1486                            pga[page_count-1]->off + pga[page_count-1]->count - 1);
1487         CERROR("original client csum %x (type %x), server csum %x (type %x), "
1488                "client csum now %x\n", client_cksum, client_cksum_type,
1489                server_cksum, cksum_type, new_cksum);
1490         return 1;
1491 }
1492
1493 /* Note rc enters this function as number of bytes transferred */
1494 static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
1495 {
1496         struct osc_brw_async_args *aa = (void *)&req->rq_async_args;
1497         const lnet_process_id_t *peer =
1498                         &req->rq_import->imp_connection->c_peer;
1499         struct client_obd *cli = aa->aa_cli;
1500         struct ost_body *body;
1501         __u32 client_cksum = 0;
1502         ENTRY;
1503
1504         if (rc < 0 && rc != -EDQUOT) {
1505                 DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
1506                 RETURN(rc);
1507         }
1508
1509         LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
1510         body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1511         if (body == NULL) {
1512                 DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
1513                 RETURN(-EPROTO);
1514         }
1515
1516 #ifdef HAVE_QUOTA_SUPPORT
1517         /* set/clear over quota flag for a uid/gid */
1518         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
1519             body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) {
1520                 unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid };
1521
1522                 CDEBUG(D_QUOTA, "setdq for [%u %u] with valid "LPX64", flags %x\n",
1523                        body->oa.o_uid, body->oa.o_gid, body->oa.o_valid,
1524                        body->oa.o_flags);
1525                 lquota_setdq(quota_interface, cli, qid, body->oa.o_valid,
1526                              body->oa.o_flags);
1527         }
1528 #endif
1529
1530         osc_update_grant(cli, body);
1531
1532         if (rc < 0)
1533                 RETURN(rc);
1534
1535         if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
1536                 client_cksum = aa->aa_oa->o_cksum; /* save for later */
1537
1538         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
1539                 if (rc > 0) {
1540                         CERROR("Unexpected +ve rc %d\n", rc);
1541                         RETURN(-EPROTO);
1542                 }
1543                 LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob);
1544
1545                 if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
1546                         RETURN(-EAGAIN);
1547
1548                 if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
1549                     check_write_checksum(&body->oa, peer, client_cksum,
1550                                          body->oa.o_cksum, aa->aa_requested_nob,
1551                                          aa->aa_page_count, aa->aa_ppga,
1552                                          cksum_type_unpack(aa->aa_oa->o_flags)))
1553                         RETURN(-EAGAIN);
1554
1555                 rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
1556                                      aa->aa_page_count, aa->aa_ppga);
1557                 GOTO(out, rc);
1558         }
1559
1560         /* The rest of this function executes only for OST_READs */
1561
1562         /* if unwrap_bulk failed, return -EAGAIN to retry */
1563         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
1564         if (rc < 0)
1565                 GOTO(out, rc = -EAGAIN);
1566
1567         if (rc > aa->aa_requested_nob) {
1568                 CERROR("Unexpected rc %d (%d requested)\n", rc,
1569                        aa->aa_requested_nob);
1570                 RETURN(-EPROTO);
1571         }
1572
1573         if (rc != req->rq_bulk->bd_nob_transferred) {
1574                 CERROR ("Unexpected rc %d (%d transferred)\n",
1575                         rc, req->rq_bulk->bd_nob_transferred);
1576                 return (-EPROTO);
1577         }
1578
1579         if (rc < aa->aa_requested_nob)
1580                 handle_short_read(rc, aa->aa_page_count, aa->aa_ppga);
1581
1582         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1583                 static int cksum_counter;
1584                 __u32      server_cksum = body->oa.o_cksum;
1585                 char      *via;
1586                 char      *router;
1587                 cksum_type_t cksum_type;
1588
1589                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1590                         cksum_type = cksum_type_unpack(body->oa.o_flags);
1591                 else
1592                         cksum_type = OBD_CKSUM_CRC32;
1593                 client_cksum = osc_checksum_bulk(rc, aa->aa_page_count,
1594                                                  aa->aa_ppga, OST_READ,
1595                                                  cksum_type);
1596
1597                 if (peer->nid == req->rq_bulk->bd_sender) {
1598                         via = router = "";
1599                 } else {
1600                         via = " via ";
1601                         router = libcfs_nid2str(req->rq_bulk->bd_sender);
1602                 }
1603
1604                 if (server_cksum == ~0 && rc > 0) {
1605                         CERROR("Protocol error: server %s set the 'checksum' "
1606                                "bit, but didn't send a checksum.  Not fatal, "
1607                                "but please notify on http://bugzilla.lustre.org/\n",
1608                                libcfs_nid2str(peer->nid));
1609                 } else if (server_cksum != client_cksum) {
1610                         LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
1611                                            "%s%s%s inode "DFID" object "
1612                                            LPU64"/"LPU64" extent "
1613                                            "["LPU64"-"LPU64"]\n",
1614                                            req->rq_import->imp_obd->obd_name,
1615                                            libcfs_nid2str(peer->nid),
1616                                            via, router,
1617                                            body->oa.o_valid & OBD_MD_FLFID ?
1618                                                 body->oa.o_parent_seq : (__u64)0,
1619                                            body->oa.o_valid & OBD_MD_FLFID ?
1620                                                 body->oa.o_parent_oid : 0,
1621                                            body->oa.o_valid & OBD_MD_FLFID ?
1622                                                 body->oa.o_parent_ver : 0,
1623                                            body->oa.o_id,
1624                                            body->oa.o_valid & OBD_MD_FLGROUP ?
1625                                                 body->oa.o_seq : (__u64)0,
1626                                            aa->aa_ppga[0]->off,
1627                                            aa->aa_ppga[aa->aa_page_count-1]->off +
1628                                            aa->aa_ppga[aa->aa_page_count-1]->count -
1629                                                                         1);
1630                         CERROR("client %x, server %x, cksum_type %x\n",
1631                                client_cksum, server_cksum, cksum_type);
1632                         cksum_counter = 0;
1633                         aa->aa_oa->o_cksum = client_cksum;
1634                         rc = -EAGAIN;
1635                 } else {
1636                         cksum_counter++;
1637                         CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
1638                         rc = 0;
1639                 }
1640         } else if (unlikely(client_cksum)) {
1641                 static int cksum_missed;
1642
1643                 cksum_missed++;
1644                 if ((cksum_missed & (-cksum_missed)) == cksum_missed)
1645                         CERROR("Checksum %u requested from %s but not sent\n",
1646                                cksum_missed, libcfs_nid2str(peer->nid));
1647         } else {
1648                 rc = 0;
1649         }
1650 out:
1651         if (rc >= 0)
1652                 lustre_get_wire_obdo(aa->aa_oa, &body->oa);
1653
1654         RETURN(rc);
1655 }
1656
1657 static int osc_brw_internal(int cmd, struct obd_export *exp, struct obdo *oa,
1658                             struct lov_stripe_md *lsm,
1659                             obd_count page_count, struct brw_page **pga,
1660                             struct obd_capa *ocapa)
1661 {
1662         struct ptlrpc_request *req;
1663         int                    rc;
1664         cfs_waitq_t            waitq;
1665         int                    resends = 0;
1666         struct l_wait_info     lwi;
1667
1668         ENTRY;
1669
1670         cfs_waitq_init(&waitq);
1671
1672 restart_bulk:
1673         rc = osc_brw_prep_request(cmd, &exp->exp_obd->u.cli, oa, lsm,
1674                                   page_count, pga, &req, ocapa, 0);
1675         if (rc != 0)
1676                 return (rc);
1677
1678         rc = ptlrpc_queue_wait(req);
1679
1680         if (rc == -ETIMEDOUT && req->rq_resend) {
1681                 DEBUG_REQ(D_HA, req,  "BULK TIMEOUT");
1682                 ptlrpc_req_finished(req);
1683                 goto restart_bulk;
1684         }
1685
1686         rc = osc_brw_fini_request(req, rc);
1687
1688         ptlrpc_req_finished(req);
1689         if (osc_recoverable_error(rc)) {
1690                 resends++;
1691                 if (!osc_should_resend(resends, &exp->exp_obd->u.cli)) {
1692                         CERROR("too many resend retries, returning error\n");
1693                         RETURN(-EIO);
1694                 }
1695
1696                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
1697                 l_wait_event(waitq, 0, &lwi);
1698
1699                 goto restart_bulk;
1700         }
1701
1702         RETURN (rc);
1703 }
1704
1705 int osc_brw_redo_request(struct ptlrpc_request *request,
1706                          struct osc_brw_async_args *aa)
1707 {
1708         struct ptlrpc_request *new_req;
1709         struct ptlrpc_request_set *set = request->rq_set;
1710         struct osc_brw_async_args *new_aa;
1711         struct osc_async_page *oap;
1712         int rc = 0;
1713         ENTRY;
1714
1715         if (!osc_should_resend(aa->aa_resends, aa->aa_cli)) {
1716                 CERROR("too many resent retries, returning error\n");
1717                 RETURN(-EIO);
1718         }
1719
1720         DEBUG_REQ(D_ERROR, request, "redo for recoverable error");
1721
1722         rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) ==
1723                                         OST_WRITE ? OBD_BRW_WRITE :OBD_BRW_READ,
1724                                   aa->aa_cli, aa->aa_oa,
1725                                   NULL /* lsm unused by osc currently */,
1726                                   aa->aa_page_count, aa->aa_ppga,
1727                                   &new_req, aa->aa_ocapa, 0);
1728         if (rc)
1729                 RETURN(rc);
1730
1731         client_obd_list_lock(&aa->aa_cli->cl_loi_list_lock);
1732
1733         cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
1734                 if (oap->oap_request != NULL) {
1735                         LASSERTF(request == oap->oap_request,
1736                                  "request %p != oap_request %p\n",
1737                                  request, oap->oap_request);
1738                         if (oap->oap_interrupted) {
1739                                 client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
1740                                 ptlrpc_req_finished(new_req);
1741                                 RETURN(-EINTR);
1742                         }
1743                 }
1744         }
1745         /* New request takes over pga and oaps from old request.
1746          * Note that copying a list_head doesn't work, need to move it... */
1747         aa->aa_resends++;
1748         new_req->rq_interpret_reply = request->rq_interpret_reply;
1749         new_req->rq_async_args = request->rq_async_args;
1750         new_req->rq_sent = cfs_time_current_sec() + aa->aa_resends;
1751
1752         new_aa = ptlrpc_req_async_args(new_req);
1753
1754         CFS_INIT_LIST_HEAD(&new_aa->aa_oaps);
1755         cfs_list_splice(&aa->aa_oaps, &new_aa->aa_oaps);
1756         CFS_INIT_LIST_HEAD(&aa->aa_oaps);
1757
1758         cfs_list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
1759                 if (oap->oap_request) {
1760                         ptlrpc_req_finished(oap->oap_request);
1761                         oap->oap_request = ptlrpc_request_addref(new_req);
1762                 }
1763         }
1764
1765         new_aa->aa_ocapa = aa->aa_ocapa;
1766         aa->aa_ocapa = NULL;
1767
1768         /* use ptlrpc_set_add_req is safe because interpret functions work
1769          * in check_set context. only one way exist with access to request
1770          * from different thread got -EINTR - this way protected with
1771          * cl_loi_list_lock */
1772         ptlrpc_set_add_req(set, new_req);
1773
1774         client_obd_list_unlock(&aa->aa_cli->cl_loi_list_lock);
1775
1776         DEBUG_REQ(D_INFO, new_req, "new request");
1777         RETURN(0);
1778 }
1779
1780 /*
1781  * ugh, we want disk allocation on the target to happen in offset order.  we'll
1782  * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do
1783  * fine for our small page arrays and doesn't require allocation.  its an
1784  * insertion sort that swaps elements that are strides apart, shrinking the
1785  * stride down until its '1' and the array is sorted.
1786  */
1787 static void sort_brw_pages(struct brw_page **array, int num)
1788 {
1789         int stride, i, j;
1790         struct brw_page *tmp;
1791
1792         if (num == 1)
1793                 return;
1794         for (stride = 1; stride < num ; stride = (stride * 3) + 1)
1795                 ;
1796
1797         do {
1798                 stride /= 3;
1799                 for (i = stride ; i < num ; i++) {
1800                         tmp = array[i];
1801                         j = i;
1802                         while (j >= stride && array[j - stride]->off > tmp->off) {
1803                                 array[j] = array[j - stride];
1804                                 j -= stride;
1805                         }
1806                         array[j] = tmp;
1807                 }
1808         } while (stride > 1);
1809 }
1810
1811 static obd_count max_unfragmented_pages(struct brw_page **pg, obd_count pages)
1812 {
1813         int count = 1;
1814         int offset;
1815         int i = 0;
1816
1817         LASSERT (pages > 0);
1818         offset = pg[i]->off & ~CFS_PAGE_MASK;
1819
1820         for (;;) {
1821                 pages--;
1822                 if (pages == 0)         /* that's all */
1823                         return count;
1824
1825                 if (offset + pg[i]->count < CFS_PAGE_SIZE)
1826                         return count;   /* doesn't end on page boundary */
1827
1828                 i++;
1829                 offset = pg[i]->off & ~CFS_PAGE_MASK;
1830                 if (offset != 0)        /* doesn't start on page boundary */
1831                         return count;
1832
1833                 count++;
1834         }
1835 }
1836
1837 static struct brw_page **osc_build_ppga(struct brw_page *pga, obd_count count)
1838 {
1839         struct brw_page **ppga;
1840         int i;
1841
1842         OBD_ALLOC(ppga, sizeof(*ppga) * count);
1843         if (ppga == NULL)
1844                 return NULL;
1845
1846         for (i = 0; i < count; i++)
1847                 ppga[i] = pga + i;
1848         return ppga;
1849 }
1850
1851 static void osc_release_ppga(struct brw_page **ppga, obd_count count)
1852 {
1853         LASSERT(ppga != NULL);
1854         OBD_FREE(ppga, sizeof(*ppga) * count);
1855 }
1856
1857 static int osc_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1858                    obd_count page_count, struct brw_page *pga,
1859                    struct obd_trans_info *oti)
1860 {
1861         struct obdo *saved_oa = NULL;
1862         struct brw_page **ppga, **orig;
1863         struct obd_import *imp = class_exp2cliimp(exp);
1864         struct client_obd *cli;
1865         int rc, page_count_orig;
1866         ENTRY;
1867
1868         LASSERT((imp != NULL) && (imp->imp_obd != NULL));
1869         cli = &imp->imp_obd->u.cli;
1870
1871         if (cmd & OBD_BRW_CHECK) {
1872                 /* The caller just wants to know if there's a chance that this
1873                  * I/O can succeed */
1874
1875                 if (imp->imp_invalid)
1876                         RETURN(-EIO);
1877                 RETURN(0);
1878         }
1879
1880         /* test_brw with a failed create can trip this, maybe others. */
1881         LASSERT(cli->cl_max_pages_per_rpc);
1882
1883         rc = 0;
1884
1885         orig = ppga = osc_build_ppga(pga, page_count);
1886         if (ppga == NULL)
1887                 RETURN(-ENOMEM);
1888         page_count_orig = page_count;
1889
1890         sort_brw_pages(ppga, page_count);
1891         while (page_count) {
1892                 obd_count pages_per_brw;
1893
1894                 if (page_count > cli->cl_max_pages_per_rpc)
1895                         pages_per_brw = cli->cl_max_pages_per_rpc;
1896                 else
1897                         pages_per_brw = page_count;
1898
1899                 pages_per_brw = max_unfragmented_pages(ppga, pages_per_brw);
1900
1901                 if (saved_oa != NULL) {
1902                         /* restore previously saved oa */
1903                         *oinfo->oi_oa = *saved_oa;
1904                 } else if (page_count > pages_per_brw) {
1905                         /* save a copy of oa (brw will clobber it) */
1906                         OBDO_ALLOC(saved_oa);
1907                         if (saved_oa == NULL)
1908                                 GOTO(out, rc = -ENOMEM);
1909                         *saved_oa = *oinfo->oi_oa;
1910                 }
1911
1912                 rc = osc_brw_internal(cmd, exp, oinfo->oi_oa, oinfo->oi_md,
1913                                       pages_per_brw, ppga, oinfo->oi_capa);
1914
1915                 if (rc != 0)
1916                         break;
1917
1918                 page_count -= pages_per_brw;
1919                 ppga += pages_per_brw;
1920         }
1921
1922 out:
1923         osc_release_ppga(orig, page_count_orig);
1924
1925         if (saved_oa != NULL)
1926                 OBDO_FREE(saved_oa);
1927
1928         RETURN(rc);
1929 }
1930
1931 /* The companion to osc_enter_cache(), called when @oap is no longer part of
1932  * the dirty accounting.  Writeback completes or truncate happens before
1933  * writing starts.  Must be called with the loi lock held. */
1934 static void osc_exit_cache(struct client_obd *cli, struct osc_async_page *oap,
1935                            int sent)
1936 {
1937         osc_release_write_grant(cli, &oap->oap_brw_page, sent);
1938 }
1939
1940 static int lop_makes_syncfs_rpc(struct loi_oap_pages *lop)
1941 {
1942         struct osc_async_page *oap;
1943         ENTRY;
1944
1945         if (cfs_list_empty(&lop->lop_urgent))
1946                 RETURN(0);
1947
1948         oap = cfs_list_entry(lop->lop_urgent.next,
1949                              struct osc_async_page, oap_urgent_item);
1950
1951         if (oap->oap_async_flags & ASYNC_SYNCFS) {
1952                 CDEBUG(D_CACHE, "syncfs request forcing RPC\n");
1953                 RETURN(1);
1954         }
1955
1956         RETURN(0);
1957 }
1958
1959 /* This maintains the lists of pending pages to read/write for a given object
1960  * (lop).  This is used by osc_check_rpcs->osc_next_loi() and loi_list_maint()
1961  * to quickly find objects that are ready to send an RPC. */
1962 static int lop_makes_rpc(struct client_obd *cli, struct loi_oap_pages *lop,
1963                          int cmd)
1964 {
1965         int optimal;
1966         ENTRY;
1967
1968         if (lop->lop_num_pending == 0)
1969                 RETURN(0);
1970
1971         /* if we have an invalid import we want to drain the queued pages
1972          * by forcing them through rpcs that immediately fail and complete
1973          * the pages.  recovery relies on this to empty the queued pages
1974          * before canceling the locks and evicting down the llite pages */
1975         if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
1976                 RETURN(1);
1977
1978         /* stream rpcs in queue order as long as as there is an urgent page
1979          * queued.  this is our cheap solution for good batching in the case
1980          * where writepage marks some random page in the middle of the file
1981          * as urgent because of, say, memory pressure */
1982         if (!cfs_list_empty(&lop->lop_urgent)) {
1983                 CDEBUG(D_CACHE, "urgent request forcing RPC\n");
1984                 RETURN(1);
1985         }
1986         /* fire off rpcs when we have 'optimal' rpcs as tuned for the wire. */
1987         optimal = cli->cl_max_pages_per_rpc;
1988         if (cmd & OBD_BRW_WRITE) {
1989                 /* trigger a write rpc stream as long as there are dirtiers
1990                  * waiting for space.  as they're waiting, they're not going to
1991                  * create more pages to coalesce with what's waiting.. */
1992                 if (!cfs_list_empty(&cli->cl_cache_waiters)) {
1993                         CDEBUG(D_CACHE, "cache waiters forcing RPC\n");
1994                         RETURN(1);
1995                 }
1996                 /* +16 to avoid triggering rpcs that would want to include pages
1997                  * that are being queued but which can't be made ready until
1998                  * the queuer finishes with the page. this is a wart for
1999                  * llite::commit_write() */
2000                 optimal += 16;
2001         }
2002         if (lop->lop_num_pending >= optimal)
2003                 RETURN(1);
2004
2005         RETURN(0);
2006 }
2007
2008 static int lop_makes_hprpc(struct loi_oap_pages *lop)
2009 {
2010         struct osc_async_page *oap;
2011         ENTRY;
2012
2013         if (cfs_list_empty(&lop->lop_urgent))
2014                 RETURN(0);
2015
2016         oap = cfs_list_entry(lop->lop_urgent.next,
2017                          struct osc_async_page, oap_urgent_item);
2018
2019         if (oap->oap_async_flags & ASYNC_HP) {
2020                 CDEBUG(D_CACHE, "hp request forcing RPC\n");
2021                 RETURN(1);
2022         }
2023
2024         RETURN(0);
2025 }
2026
2027 static void on_list(cfs_list_t *item, cfs_list_t *list,
2028                     int should_be_on)
2029 {
2030         if (cfs_list_empty(item) && should_be_on)
2031                 cfs_list_add_tail(item, list);
2032         else if (!cfs_list_empty(item) && !should_be_on)
2033                 cfs_list_del_init(item);
2034 }
2035
2036 /* maintain the loi's cli list membership invariants so that osc_send_oap_rpc
2037  * can find pages to build into rpcs quickly */
2038 void loi_list_maint(struct client_obd *cli, struct lov_oinfo *loi)
2039 {
2040         if (lop_makes_hprpc(&loi->loi_write_lop) ||
2041             lop_makes_hprpc(&loi->loi_read_lop)) {
2042                 /* HP rpc */
2043                 on_list(&loi->loi_ready_item, &cli->cl_loi_ready_list, 0);
2044                 on_list(&loi->loi_hp_ready_item, &cli->cl_loi_hp_ready_list, 1);
2045         } else {
2046                 if (lop_makes_syncfs_rpc(&loi->loi_write_lop)) {
2047                         on_list(&loi->loi_sync_fs_item,
2048                                 &cli->cl_loi_sync_fs_list,
2049                                 loi->loi_write_lop.lop_num_pending);
2050                 } else {
2051                         on_list(&loi->loi_hp_ready_item,
2052                                 &cli->cl_loi_hp_ready_list, 0);
2053                         on_list(&loi->loi_ready_item, &cli->cl_loi_ready_list,
2054                                 lop_makes_rpc(cli, &loi->loi_write_lop,
2055                                               OBD_BRW_WRITE)||
2056                                 lop_makes_rpc(cli, &loi->loi_read_lop,
2057                                               OBD_BRW_READ));
2058                 }
2059         }
2060
2061         on_list(&loi->loi_write_item, &cli->cl_loi_write_list,
2062                 loi->loi_write_lop.lop_num_pending);
2063
2064         on_list(&loi->loi_read_item, &cli->cl_loi_read_list,
2065                 loi->loi_read_lop.lop_num_pending);
2066 }
2067
2068 static void lop_update_pending(struct client_obd *cli,
2069                                struct loi_oap_pages *lop, int cmd, int delta)
2070 {
2071         lop->lop_num_pending += delta;
2072         if (cmd & OBD_BRW_WRITE)
2073                 cli->cl_pending_w_pages += delta;
2074         else
2075                 cli->cl_pending_r_pages += delta;
2076 }
2077
2078 /**
2079  * this is called when a sync waiter receives an interruption.  Its job is to
2080  * get the caller woken as soon as possible.  If its page hasn't been put in an
2081  * rpc yet it can dequeue immediately.  Otherwise it has to mark the rpc as
2082  * desiring interruption which will forcefully complete the rpc once the rpc
2083  * has timed out.
2084  */
2085 int osc_oap_interrupted(const struct lu_env *env, struct osc_async_page *oap)
2086 {
2087         struct loi_oap_pages *lop;
2088         struct lov_oinfo *loi;
2089         int rc = -EBUSY;
2090         ENTRY;
2091
2092         LASSERT(!oap->oap_interrupted);
2093         oap->oap_interrupted = 1;
2094
2095         /* ok, it's been put in an rpc. only one oap gets a request reference */
2096         if (oap->oap_request != NULL) {
2097                 ptlrpc_mark_interrupted(oap->oap_request);
2098                 ptlrpcd_wake(oap->oap_request);
2099                 ptlrpc_req_finished(oap->oap_request);
2100                 oap->oap_request = NULL;
2101         }
2102
2103         /*
2104          * page completion may be called only if ->cpo_prep() method was
2105          * executed by osc_io_submit(), that also adds page the to pending list
2106          */
2107         if (!cfs_list_empty(&oap->oap_pending_item)) {
2108                 cfs_list_del_init(&oap->oap_pending_item);
2109                 cfs_list_del_init(&oap->oap_urgent_item);
2110
2111                 loi = oap->oap_loi;
2112                 lop = (oap->oap_cmd & OBD_BRW_WRITE) ?
2113                         &loi->loi_write_lop : &loi->loi_read_lop;
2114                 lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, -1);
2115                 loi_list_maint(oap->oap_cli, oap->oap_loi);
2116                 rc = oap->oap_caller_ops->ap_completion(env,
2117                                           oap->oap_caller_data,
2118                                           oap->oap_cmd, NULL, -EINTR);
2119         }
2120
2121         RETURN(rc);
2122 }
2123
2124 /* this is trying to propogate async writeback errors back up to the
2125  * application.  As an async write fails we record the error code for later if
2126  * the app does an fsync.  As long as errors persist we force future rpcs to be
2127  * sync so that the app can get a sync error and break the cycle of queueing
2128  * pages for which writeback will fail. */
2129 static void osc_process_ar(struct osc_async_rc *ar, __u64 xid,
2130                            int rc)
2131 {
2132         if (rc) {
2133                 if (!ar->ar_rc)
2134                         ar->ar_rc = rc;
2135
2136                 ar->ar_force_sync = 1;
2137                 ar->ar_min_xid = ptlrpc_sample_next_xid();
2138                 return;
2139
2140         }
2141
2142         if (ar->ar_force_sync && (xid >= ar->ar_min_xid))
2143                 ar->ar_force_sync = 0;
2144 }
2145
2146 void osc_oap_to_pending(struct osc_async_page *oap)
2147 {
2148         struct loi_oap_pages *lop;
2149
2150         if (oap->oap_cmd & OBD_BRW_WRITE)
2151                 lop = &oap->oap_loi->loi_write_lop;
2152         else
2153                 lop = &oap->oap_loi->loi_read_lop;
2154
2155         if (oap->oap_async_flags & ASYNC_HP)
2156                 cfs_list_add(&oap->oap_urgent_item, &lop->lop_urgent);
2157         else if (oap->oap_async_flags & ASYNC_URGENT)
2158                 cfs_list_add_tail(&oap->oap_urgent_item, &lop->lop_urgent);
2159         cfs_list_add_tail(&oap->oap_pending_item, &lop->lop_pending);
2160         lop_update_pending(oap->oap_cli, lop, oap->oap_cmd, 1);
2161 }
2162
2163 /* this must be called holding the loi list lock to give coverage to exit_cache,
2164  * async_flag maintenance, and oap_request */
2165 static void osc_ap_completion(const struct lu_env *env,
2166                               struct client_obd *cli, struct obdo *oa,
2167                               struct osc_async_page *oap, int sent, int rc)
2168 {
2169         __u64 xid = 0;
2170
2171         ENTRY;
2172         if (oap->oap_request != NULL) {
2173                 xid = ptlrpc_req_xid(oap->oap_request);
2174                 ptlrpc_req_finished(oap->oap_request);
2175                 oap->oap_request = NULL;
2176         }
2177
2178         cfs_spin_lock(&oap->oap_lock);
2179         oap->oap_async_flags = 0;
2180         cfs_spin_unlock(&oap->oap_lock);
2181         oap->oap_interrupted = 0;
2182
2183         if (oap->oap_cmd & OBD_BRW_WRITE) {
2184                 osc_process_ar(&cli->cl_ar, xid, rc);
2185                 osc_process_ar(&oap->oap_loi->loi_ar, xid, rc);
2186         }
2187
2188         if (rc == 0 && oa != NULL) {
2189                 if (oa->o_valid & OBD_MD_FLBLOCKS)
2190                         oap->oap_loi->loi_lvb.lvb_blocks = oa->o_blocks;
2191                 if (oa->o_valid & OBD_MD_FLMTIME)
2192                         oap->oap_loi->loi_lvb.lvb_mtime = oa->o_mtime;
2193                 if (oa->o_valid & OBD_MD_FLATIME)
2194                         oap->oap_loi->loi_lvb.lvb_atime = oa->o_atime;
2195                 if (oa->o_valid & OBD_MD_FLCTIME)
2196                         oap->oap_loi->loi_lvb.lvb_ctime = oa->o_ctime;
2197         }
2198
2199         rc = oap->oap_caller_ops->ap_completion(env, oap->oap_caller_data,
2200                                                 oap->oap_cmd, oa, rc);
2201
2202         /* ll_ap_completion (from llite) drops PG_locked. so, a new
2203          * I/O on the page could start, but OSC calls it under lock
2204          * and thus we can add oap back to pending safely */
2205         if (rc)
2206                 /* upper layer wants to leave the page on pending queue */
2207                 osc_oap_to_pending(oap);
2208         else
2209                 osc_exit_cache(cli, oap, sent);
2210         EXIT;
2211 }
2212
2213 static int brw_interpret(const struct lu_env *env,
2214                          struct ptlrpc_request *req, void *data, int rc)
2215 {
2216         struct osc_brw_async_args *aa = data;
2217         struct client_obd *cli;
2218         int async;
2219         ENTRY;
2220
2221         rc = osc_brw_fini_request(req, rc);
2222         CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
2223         if (osc_recoverable_error(rc)) {
2224                 /* Only retry once for mmaped files since the mmaped page
2225                  * might be modified at anytime. We have to retry at least
2226                  * once in case there WAS really a corruption of the page
2227                  * on the network, that was not caused by mmap() modifying
2228                  * the page. Bug11742 */
2229                 if ((rc == -EAGAIN) && (aa->aa_resends > 0) &&
2230                     aa->aa_oa->o_valid & OBD_MD_FLFLAGS &&
2231                     aa->aa_oa->o_flags & OBD_FL_MMAP) {
2232                         rc = 0;
2233                 } else {
2234                         rc = osc_brw_redo_request(req, aa);
2235                         if (rc == 0)
2236                                 RETURN(0);
2237                 }
2238         }
2239
2240         if (aa->aa_ocapa) {
2241                 capa_put(aa->aa_ocapa);
2242                 aa->aa_ocapa = NULL;
2243         }
2244
2245         cli = aa->aa_cli;
2246
2247         client_obd_list_lock(&cli->cl_loi_list_lock);
2248
2249         /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
2250          * is called so we know whether to go to sync BRWs or wait for more
2251          * RPCs to complete */
2252         if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE)
2253                 cli->cl_w_in_flight--;
2254         else
2255                 cli->cl_r_in_flight--;
2256
2257         async = cfs_list_empty(&aa->aa_oaps);
2258         if (!async) { /* from osc_send_oap_rpc() */
2259                 struct osc_async_page *oap, *tmp;
2260                 /* the caller may re-use the oap after the completion call so
2261                  * we need to clean it up a little */
2262                 cfs_list_for_each_entry_safe(oap, tmp, &aa->aa_oaps,
2263                                              oap_rpc_item) {
2264                         cfs_list_del_init(&oap->oap_rpc_item);
2265                         osc_ap_completion(env, cli, aa->aa_oa, oap, 1, rc);
2266                 }
2267                 OBDO_FREE(aa->aa_oa);
2268         } else { /* from async_internal() */
2269                 obd_count i;
2270                 for (i = 0; i < aa->aa_page_count; i++)
2271                         osc_release_write_grant(aa->aa_cli, aa->aa_ppga[i], 1);
2272         }
2273         osc_wake_cache_waiters(cli);
2274         osc_wake_sync_fs(cli);
2275         osc_check_rpcs(env, cli);
2276         client_obd_list_unlock(&cli->cl_loi_list_lock);
2277         if (!async)
2278                 cl_req_completion(env, aa->aa_clerq, rc);
2279         osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
2280
2281         RETURN(rc);
2282 }
2283
2284 static struct ptlrpc_request *osc_build_req(const struct lu_env *env,
2285                                             struct client_obd *cli,
2286                                             cfs_list_t *rpc_list,
2287                                             int page_count, int cmd)
2288 {
2289         struct ptlrpc_request *req;
2290         struct brw_page **pga = NULL;
2291         struct osc_brw_async_args *aa;
2292         struct obdo *oa = NULL;
2293         const struct obd_async_page_ops *ops = NULL;
2294         void *caller_data = NULL;
2295         struct osc_async_page *oap;
2296         struct osc_async_page *tmp;
2297         struct ost_body *body;
2298         struct cl_req *clerq = NULL;
2299         enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
2300         struct ldlm_lock *lock = NULL;
2301         struct cl_req_attr crattr;
2302         int i, rc, mpflag = 0;
2303
2304         ENTRY;
2305         LASSERT(!cfs_list_empty(rpc_list));
2306
2307         if (cmd & OBD_BRW_MEMALLOC)
2308                 mpflag = cfs_memory_pressure_get_and_set();
2309
2310         memset(&crattr, 0, sizeof crattr);
2311         OBD_ALLOC(pga, sizeof(*pga) * page_count);
2312         if (pga == NULL)
2313                 GOTO(out, req = ERR_PTR(-ENOMEM));
2314
2315         OBDO_ALLOC(oa);
2316         if (oa == NULL)
2317                 GOTO(out, req = ERR_PTR(-ENOMEM));
2318
2319         i = 0;
2320         cfs_list_for_each_entry(oap, rpc_list, oap_rpc_item) {
2321                 struct cl_page *page = osc_oap2cl_page(oap);
2322                 if (ops == NULL) {
2323                         ops = oap->oap_caller_ops;
2324                         caller_data = oap->oap_caller_data;
2325
2326                         clerq = cl_req_alloc(env, page, crt,
2327                                              1 /* only 1-object rpcs for
2328                                                 * now */);
2329                         if (IS_ERR(clerq))
2330                                 GOTO(out, req = (void *)clerq);
2331                         lock = oap->oap_ldlm_lock;
2332                 }
2333                 pga[i] = &oap->oap_brw_page;
2334                 pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
2335                 CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
2336                        pga[i]->pg, cfs_page_index(oap->oap_page), oap, pga[i]->flag);
2337                 i++;
2338                 cl_req_page_add(env, clerq, page);
2339         }
2340
2341         /* always get the data for the obdo for the rpc */
2342         LASSERT(ops != NULL);
2343         crattr.cra_oa = oa;
2344         crattr.cra_capa = NULL;
2345         cl_req_attr_set(env, clerq, &crattr, ~0ULL);
2346         if (lock) {
2347                 oa->o_handle = lock->l_remote_handle;
2348                 oa->o_valid |= OBD_MD_FLHANDLE;
2349         }
2350
2351         rc = cl_req_prep(env, clerq);
2352         if (rc != 0) {
2353                 CERROR("cl_req_prep failed: %d\n", rc);
2354                 GOTO(out, req = ERR_PTR(rc));
2355         }
2356
2357         sort_brw_pages(pga, page_count);
2358         rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
2359                                   pga, &req, crattr.cra_capa, 1);
2360         if (rc != 0) {
2361                 CERROR("prep_req failed: %d\n", rc);
2362                 GOTO(out, req = ERR_PTR(rc));
2363         }
2364
2365         if (cmd & OBD_BRW_MEMALLOC)
2366                 req->rq_memalloc = 1;
2367
2368         /* Need to update the timestamps after the request is built in case
2369          * we race with setattr (locally or in queue at OST).  If OST gets
2370          * later setattr before earlier BRW (as determined by the request xid),
2371          * the OST will not use BRW timestamps.  Sadly, there is no obvious
2372          * way to do this in a single call.  bug 10150 */
2373         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2374         cl_req_attr_set(env, clerq, &crattr,
2375                         OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
2376
2377         CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
2378         aa = ptlrpc_req_async_args(req);
2379         CFS_INIT_LIST_HEAD(&aa->aa_oaps);
2380         cfs_list_splice(rpc_list, &aa->aa_oaps);
2381         CFS_INIT_LIST_HEAD(rpc_list);
2382         aa->aa_clerq = clerq;
2383 out:
2384         if (cmd & OBD_BRW_MEMALLOC)
2385                 cfs_memory_pressure_restore(mpflag);
2386
2387         capa_put(crattr.cra_capa);
2388         if (IS_ERR(req)) {
2389                 if (oa)
2390                         OBDO_FREE(oa);
2391                 if (pga)
2392                         OBD_FREE(pga, sizeof(*pga) * page_count);
2393                 /* this should happen rarely and is pretty bad, it makes the
2394                  * pending list not follow the dirty order */
2395                 client_obd_list_lock(&cli->cl_loi_list_lock);
2396                 cfs_list_for_each_entry_safe(oap, tmp, rpc_list, oap_rpc_item) {
2397                         cfs_list_del_init(&oap->oap_rpc_item);
2398
2399                         /* queued sync pages can be torn down while the pages
2400                          * were between the pending list and the rpc */
2401                         if (oap->oap_interrupted) {
2402                                 CDEBUG(D_INODE, "oap %p interrupted\n", oap);
2403                                 osc_ap_completion(env, cli, NULL, oap, 0,
2404                                                   oap->oap_count);
2405                                 continue;
2406                         }
2407                         osc_ap_completion(env, cli, NULL, oap, 0, PTR_ERR(req));
2408                 }
2409                 if (clerq && !IS_ERR(clerq))
2410                         cl_req_completion(env, clerq, PTR_ERR(req));
2411         }
2412         RETURN(req);
2413 }
2414
2415 /**
2416  * prepare pages for ASYNC io and put pages in send queue.
2417  *
2418  * \param cmd OBD_BRW_* macroses
2419  * \param lop pending pages
2420  *
2421  * \return zero if no page added to send queue.
2422  * \return 1 if pages successfully added to send queue.
2423  * \return negative on errors.
2424  */
2425 static int
2426 osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli,
2427                  struct lov_oinfo *loi,
2428                  int cmd, struct loi_oap_pages *lop)
2429 {
2430         struct ptlrpc_request *req;
2431         obd_count page_count = 0;
2432         struct osc_async_page *oap = NULL, *tmp;
2433         struct osc_brw_async_args *aa;
2434         const struct obd_async_page_ops *ops;
2435         CFS_LIST_HEAD(rpc_list);
2436         CFS_LIST_HEAD(tmp_list);
2437         unsigned int ending_offset;
2438         unsigned  starting_offset = 0;
2439         int srvlock = 0, mem_tight = 0;
2440         struct cl_object *clob = NULL;
2441         ENTRY;
2442
2443         /* ASYNC_HP pages first. At present, when the lock the pages is
2444          * to be canceled, the pages covered by the lock will be sent out
2445          * with ASYNC_HP. We have to send out them as soon as possible. */
2446         cfs_list_for_each_entry_safe(oap, tmp, &lop->lop_urgent, oap_urgent_item) {
2447                 if (oap->oap_async_flags & ASYNC_HP)
2448                         cfs_list_move(&oap->oap_pending_item, &tmp_list);
2449                 else
2450                         cfs_list_move_tail(&oap->oap_pending_item, &tmp_list);
2451                 if (++page_count >= cli->cl_max_pages_per_rpc)
2452                         break;
2453         }
2454
2455         cfs_list_splice(&tmp_list, &lop->lop_pending);
2456         page_count = 0;
2457
2458         /* first we find the pages we're allowed to work with */
2459         cfs_list_for_each_entry_safe(oap, tmp, &lop->lop_pending,
2460                                      oap_pending_item) {
2461                 ops = oap->oap_caller_ops;
2462
2463                 LASSERTF(oap->oap_magic == OAP_MAGIC, "Bad oap magic: oap %p, "
2464                          "magic 0x%x\n", oap, oap->oap_magic);
2465
2466                 if (clob == NULL) {
2467                         /* pin object in memory, so that completion call-backs
2468                          * can be safely called under client_obd_list lock. */
2469                         clob = osc_oap2cl_page(oap)->cp_obj;
2470                         cl_object_get(clob);
2471                 }
2472
2473                 if (page_count != 0 &&
2474                     srvlock != !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK)) {
2475                         CDEBUG(D_PAGE, "SRVLOCK flag mismatch,"
2476                                " oap %p, page %p, srvlock %u\n",
2477                                oap, oap->oap_brw_page.pg, (unsigned)!srvlock);
2478                         break;
2479                 }
2480
2481                 /* If there is a gap at the start of this page, it can't merge
2482                  * with any previous page, so we'll hand the network a
2483                  * "fragmented" page array that it can't transfer in 1 RDMA */
2484                 if (page_count != 0 && oap->oap_page_off != 0)
2485                         break;
2486
2487                 /* in llite being 'ready' equates to the page being locked
2488                  * until completion unlocks it.  commit_write submits a page
2489                  * as not ready because its unlock will happen unconditionally
2490                  * as the call returns.  if we race with commit_write giving
2491                  * us that page we don't want to create a hole in the page
2492                  * stream, so we stop and leave the rpc to be fired by
2493                  * another dirtier or kupdated interval (the not ready page
2494                  * will still be on the dirty list).  we could call in
2495                  * at the end of ll_file_write to process the queue again. */
2496                 if (!(oap->oap_async_flags & ASYNC_READY)) {
2497                         int rc = ops->ap_make_ready(env, oap->oap_caller_data,
2498                                                     cmd);
2499                         if (rc < 0)
2500                                 CDEBUG(D_INODE, "oap %p page %p returned %d "
2501                                                 "instead of ready\n", oap,
2502                                                 oap->oap_page, rc);
2503                         switch (rc) {
2504                         case -EAGAIN:
2505                                 /* llite is telling us that the page is still
2506                                  * in commit_write and that we should try
2507                                  * and put it in an rpc again later.  we
2508                                  * break out of the loop so we don't create
2509                                  * a hole in the sequence of pages in the rpc
2510                                  * stream.*/
2511                                 oap = NULL;
2512                                 break;
2513                         case -EINTR:
2514                                 /* the io isn't needed.. tell the checks
2515                                  * below to complete the rpc with EINTR */
2516                                 cfs_spin_lock(&oap->oap_lock);
2517                                 oap->oap_async_flags |= ASYNC_COUNT_STABLE;
2518                                 cfs_spin_unlock(&oap->oap_lock);
2519                                 oap->oap_count = -EINTR;
2520                                 break;
2521                         case 0:
2522                                 cfs_spin_lock(&oap->oap_lock);
2523                                 oap->oap_async_flags |= ASYNC_READY;
2524                                 cfs_spin_unlock(&oap->oap_lock);
2525                                 break;
2526                         default:
2527                                 LASSERTF(0, "oap %p page %p returned %d "
2528                                             "from make_ready\n", oap,
2529                                             oap->oap_page, rc);
2530                                 break;
2531                         }
2532                 }
2533                 if (oap == NULL)
2534                         break;
2535                 /*
2536                  * Page submitted for IO has to be locked. Either by
2537                  * ->ap_make_ready() or by higher layers.
2538                  */
2539 #if defined(__KERNEL__) && defined(__linux__)
2540                 {
2541                         struct cl_page *page;
2542
2543                         page = osc_oap2cl_page(oap);
2544
2545                         if (page->cp_type == CPT_CACHEABLE &&
2546                             !(PageLocked(oap->oap_page) &&
2547                               (CheckWriteback(oap->oap_page, cmd)))) {
2548                                 CDEBUG(D_PAGE, "page %p lost wb %lx/%x\n",
2549                                        oap->oap_page,
2550                                        (long)oap->oap_page->flags,
2551                                        oap->oap_async_flags);
2552                                 LBUG();
2553                         }
2554                 }
2555 #endif
2556
2557                 /* take the page out of our book-keeping */
2558                 cfs_list_del_init(&oap->oap_pending_item);
2559                 lop_update_pending(cli, lop, cmd, -1);
2560                 cfs_list_del_init(&oap->oap_urgent_item);
2561
2562                 if (page_count == 0)
2563                         starting_offset = (oap->oap_obj_off+oap->oap_page_off) &
2564                                           (PTLRPC_MAX_BRW_SIZE - 1);
2565
2566                 /* ask the caller for the size of the io as the rpc leaves. */
2567                 if (!(oap->oap_async_flags & ASYNC_COUNT_STABLE)) {
2568                         oap->oap_count =
2569                                 ops->ap_refresh_count(env, oap->oap_caller_data,
2570                                                       cmd);
2571                         LASSERT(oap->oap_page_off + oap->oap_count <= CFS_PAGE_SIZE);
2572                 }
2573                 if (oap->oap_count <= 0) {
2574                         CDEBUG(D_CACHE, "oap %p count %d, completing\n", oap,
2575                                oap->oap_count);
2576                         osc_ap_completion(env, cli, NULL,
2577                                           oap, 0, oap->oap_count);
2578                         continue;
2579                 }
2580
2581                 /* now put the page back in our accounting */
2582                 cfs_list_add_tail(&oap->oap_rpc_item, &rpc_list);
2583                 if (oap->oap_brw_flags & OBD_BRW_MEMALLOC)
2584                         mem_tight = 1;
2585                 if (page_count == 0)
2586                         srvlock = !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK);
2587                 if (++page_count >= cli->cl_max_pages_per_rpc)
2588                         break;
2589
2590                 /* End on a PTLRPC_MAX_BRW_SIZE boundary.  We want full-sized
2591                  * RPCs aligned on PTLRPC_MAX_BRW_SIZE boundaries to help reads
2592                  * have the same alignment as the initial writes that allocated
2593                  * extents on the server. */
2594                 ending_offset = (oap->oap_obj_off + oap->oap_page_off +
2595                                  oap->oap_count) & (PTLRPC_MAX_BRW_SIZE - 1);
2596                 if (ending_offset == 0)
2597                         break;
2598
2599                 /* If there is a gap at the end of this page, it can't merge
2600                  * with any subsequent pages, so we'll hand the network a
2601                  * "fragmented" page array that it can't transfer in 1 RDMA */
2602                 if (oap->oap_page_off + oap->oap_count < CFS_PAGE_SIZE)
2603                         break;
2604         }
2605
2606         osc_wake_cache_waiters(cli);
2607         osc_wake_sync_fs(cli);
2608         loi_list_maint(cli, loi);
2609
2610         client_obd_list_unlock(&cli->cl_loi_list_lock);
2611
2612         if (clob != NULL)
2613                 cl_object_put(env, clob);
2614
2615         if (page_count == 0) {
2616                 client_obd_list_lock(&cli->cl_loi_list_lock);
2617                 RETURN(0);
2618         }
2619
2620         req = osc_build_req(env, cli, &rpc_list, page_count,
2621                             mem_tight ? (cmd | OBD_BRW_MEMALLOC) : cmd);
2622         if (IS_ERR(req)) {
2623                 LASSERT(cfs_list_empty(&rpc_list));
2624                 loi_list_maint(cli, loi);
2625                 RETURN(PTR_ERR(req));
2626         }
2627
2628         aa = ptlrpc_req_async_args(req);
2629
2630         if (cmd == OBD_BRW_READ) {
2631                 lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
2632                 lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight);
2633                 lprocfs_oh_tally_log2(&cli->cl_read_offset_hist,
2634                                       (starting_offset >> CFS_PAGE_SHIFT) + 1);
2635         } else {
2636                 lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count);
2637                 lprocfs_oh_tally(&cli->cl_write_rpc_hist,
2638                                  cli->cl_w_in_flight);
2639                 lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
2640                                       (starting_offset >> CFS_PAGE_SHIFT) + 1);
2641         }
2642         ptlrpc_lprocfs_brw(req, aa->aa_requested_nob);
2643
2644         client_obd_list_lock(&cli->cl_loi_list_lock);
2645
2646         if (cmd == OBD_BRW_READ)
2647                 cli->cl_r_in_flight++;
2648         else
2649                 cli->cl_w_in_flight++;
2650
2651         /* queued sync pages can be torn down while the pages
2652          * were between the pending list and the rpc */
2653         tmp = NULL;
2654         cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
2655                 /* only one oap gets a request reference */
2656                 if (tmp == NULL)
2657                         tmp = oap;
2658                 if (oap->oap_interrupted && !req->rq_intr) {
2659                         CDEBUG(D_INODE, "oap %p in req %p interrupted\n",
2660                                oap, req);
2661                         ptlrpc_mark_interrupted(req);
2662                 }
2663         }
2664         if (tmp != NULL)
2665                 tmp->oap_request = ptlrpc_request_addref(req);
2666
2667         DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight",
2668                   page_count, aa, cli->cl_r_in_flight, cli->cl_w_in_flight);
2669
2670         req->rq_interpret_reply = brw_interpret;
2671         ptlrpcd_add_req(req, PSCOPE_BRW);
2672         RETURN(1);
2673 }
2674
2675 #define LOI_DEBUG(LOI, STR, args...)                                     \
2676         CDEBUG(D_INODE, "loi ready %d wr %d:%d rd %d:%d " STR,           \
2677                !cfs_list_empty(&(LOI)->loi_ready_item) ||                \
2678                !cfs_list_empty(&(LOI)->loi_hp_ready_item),               \
2679                (LOI)->loi_write_lop.lop_num_pending,                     \
2680                !cfs_list_empty(&(LOI)->loi_write_lop.lop_urgent),        \
2681                (LOI)->loi_read_lop.lop_num_pending,                      \
2682                !cfs_list_empty(&(LOI)->loi_read_lop.lop_urgent),         \
2683                args)                                                     \
2684
2685 /* This is called by osc_check_rpcs() to find which objects have pages that
2686  * we could be sending.  These lists are maintained by lop_makes_rpc(). */
2687 struct lov_oinfo *osc_next_loi(struct client_obd *cli)
2688 {
2689         ENTRY;
2690
2691         /* First return objects that have blocked locks so that they
2692          * will be flushed quickly and other clients can get the lock,
2693          * then objects which have pages ready to be stuffed into RPCs */
2694         if (!cfs_list_empty(&cli->cl_loi_hp_ready_list))
2695                 RETURN(cfs_list_entry(cli->cl_loi_hp_ready_list.next,
2696                                       struct lov_oinfo, loi_hp_ready_item));
2697         if (!cfs_list_empty(&cli->cl_loi_ready_list))
2698                 RETURN(cfs_list_entry(cli->cl_loi_ready_list.next,
2699                                       struct lov_oinfo, loi_ready_item));
2700         if (!cfs_list_empty(&cli->cl_loi_sync_fs_list))
2701                 RETURN(cfs_list_entry(cli->cl_loi_sync_fs_list.next,
2702                                       struct lov_oinfo, loi_sync_fs_item));
2703
2704         /* then if we have cache waiters, return all objects with queued
2705          * writes.  This is especially important when many small files
2706          * have filled up the cache and not been fired into rpcs because
2707          * they don't pass the nr_pending/object threshhold */
2708         if (!cfs_list_empty(&cli->cl_cache_waiters) &&
2709             !cfs_list_empty(&cli->cl_loi_write_list))
2710                 RETURN(cfs_list_entry(cli->cl_loi_write_list.next,
2711                                       struct lov_oinfo, loi_write_item));
2712
2713         /* then return all queued objects when we have an invalid import
2714          * so that they get flushed */
2715         if (cli->cl_import == NULL || cli->cl_import->imp_invalid) {
2716                 if (!cfs_list_empty(&cli->cl_loi_write_list))
2717                         RETURN(cfs_list_entry(cli->cl_loi_write_list.next,
2718                                               struct lov_oinfo,
2719                                               loi_write_item));
2720                 if (!cfs_list_empty(&cli->cl_loi_read_list))
2721                         RETURN(cfs_list_entry(cli->cl_loi_read_list.next,
2722                                               struct lov_oinfo, loi_read_item));
2723         }
2724         RETURN(NULL);
2725 }
2726
2727 static int osc_max_rpc_in_flight(struct client_obd *cli, struct lov_oinfo *loi)
2728 {
2729         struct osc_async_page *oap;
2730         int hprpc = 0;
2731
2732         if (!cfs_list_empty(&loi->loi_write_lop.lop_urgent)) {
2733                 oap = cfs_list_entry(loi->loi_write_lop.lop_urgent.next,
2734                                      struct osc_async_page, oap_urgent_item);
2735                 hprpc = !!(oap->oap_async_flags & ASYNC_HP);
2736         }
2737
2738         if (!hprpc && !cfs_list_empty(&loi->loi_read_lop.lop_urgent)) {
2739                 oap = cfs_list_entry(loi->loi_read_lop.lop_urgent.next,
2740                                      struct osc_async_page, oap_urgent_item);
2741                 hprpc = !!(oap->oap_async_flags & ASYNC_HP);
2742         }
2743
2744         return rpcs_in_flight(cli) >= cli->cl_max_rpcs_in_flight + hprpc;
2745 }
2746
2747 /* called with the loi list lock held */
2748 void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli)
2749 {
2750         struct lov_oinfo *loi;
2751         int rc = 0, race_counter = 0;
2752         ENTRY;
2753
2754         while ((loi = osc_next_loi(cli)) != NULL) {
2755                 LOI_DEBUG(loi, "%lu in flight\n", rpcs_in_flight(cli));
2756
2757                 if (osc_max_rpc_in_flight(cli, loi))
2758                         break;
2759
2760                 /* attempt some read/write balancing by alternating between
2761                  * reads and writes in an object.  The makes_rpc checks here
2762                  * would be redundant if we were getting read/write work items
2763                  * instead of objects.  we don't want send_oap_rpc to drain a
2764                  * partial read pending queue when we're given this object to
2765                  * do io on writes while there are cache waiters */
2766                 if (lop_makes_rpc(cli, &loi->loi_write_lop, OBD_BRW_WRITE)) {
2767                         rc = osc_send_oap_rpc(env, cli, loi, OBD_BRW_WRITE,
2768                                               &loi->loi_write_lop);
2769                         if (rc < 0) {
2770                                 CERROR("Write request failed with %d\n", rc);
2771
2772                                 /* osc_send_oap_rpc failed, mostly because of
2773                                  * memory pressure.
2774                                  *
2775                                  * It can't break here, because if:
2776                                  *  - a page was submitted by osc_io_submit, so
2777                                  *    page locked;
2778                                  *  - no request in flight
2779                                  *  - no subsequent request
2780                                  * The system will be in live-lock state,
2781                                  * because there is no chance to call
2782                                  * osc_io_unplug() and osc_check_rpcs() any
2783                                  * more. pdflush can't help in this case,
2784                                  * because it might be blocked at grabbing
2785                                  * the page lock as we mentioned.
2786                                  *
2787                                  * Anyway, continue to drain pages. */
2788                                 /* break; */
2789                         }
2790
2791                         if (rc > 0)
2792                                 race_counter = 0;
2793                         else
2794                                 race_counter++;
2795                 }
2796                 if (lop_makes_rpc(cli, &loi->loi_read_lop, OBD_BRW_READ)) {
2797                         rc = osc_send_oap_rpc(env, cli, loi, OBD_BRW_READ,
2798                                               &loi->loi_read_lop);
2799                         if (rc < 0)
2800                                 CERROR("Read request failed with %d\n", rc);
2801
2802                         if (rc > 0)
2803                                 race_counter = 0;
2804                         else
2805                                 race_counter++;
2806                 }
2807
2808                 /* attempt some inter-object balancing by issuing rpcs
2809                  * for each object in turn */
2810                 if (!cfs_list_empty(&loi->loi_hp_ready_item))
2811                         cfs_list_del_init(&loi->loi_hp_ready_item);
2812                 if (!cfs_list_empty(&loi->loi_ready_item))
2813                         cfs_list_del_init(&loi->loi_ready_item);
2814                 if (!cfs_list_empty(&loi->loi_write_item))
2815                         cfs_list_del_init(&loi->loi_write_item);
2816                 if (!cfs_list_empty(&loi->loi_read_item))
2817                         cfs_list_del_init(&loi->loi_read_item);
2818                 if (!cfs_list_empty(&loi->loi_sync_fs_item))
2819                         cfs_list_del_init(&loi->loi_sync_fs_item);
2820
2821                 loi_list_maint(cli, loi);
2822
2823                 /* send_oap_rpc fails with 0 when make_ready tells it to
2824                  * back off.  llite's make_ready does this when it tries
2825                  * to lock a page queued for write that is already locked.
2826                  * we want to try sending rpcs from many objects, but we
2827                  * don't want to spin failing with 0.  */
2828                 if (race_counter == 10)
2829                         break;
2830         }
2831         EXIT;
2832 }
2833
2834 /* we're trying to queue a page in the osc so we're subject to the
2835  * 'cl_dirty_max' limit on the number of pages that can be queued in the osc.
2836  * If the osc's queued pages are already at that limit, then we want to sleep
2837  * until there is space in the osc's queue for us.  We also may be waiting for
2838  * write credits from the OST if there are RPCs in flight that may return some
2839  * before we fall back to sync writes.
2840  *
2841  * We need this know our allocation was granted in the presence of signals */
2842 static int ocw_granted(struct client_obd *cli, struct osc_cache_waiter *ocw)
2843 {
2844         int rc;
2845         ENTRY;
2846         client_obd_list_lock(&cli->cl_loi_list_lock);
2847         rc = cfs_list_empty(&ocw->ocw_entry) || rpcs_in_flight(cli) == 0;
2848         client_obd_list_unlock(&cli->cl_loi_list_lock);
2849         RETURN(rc);
2850 };
2851
2852 /**
2853  * Non-blocking version of osc_enter_cache() that consumes grant only when it
2854  * is available.
2855  */
2856 int osc_enter_cache_try(const struct lu_env *env,
2857                         struct client_obd *cli, struct lov_oinfo *loi,
2858                         struct osc_async_page *oap, int transient)
2859 {
2860         int has_grant;
2861
2862         has_grant = cli->cl_avail_grant >= CFS_PAGE_SIZE;
2863         if (has_grant) {
2864                 osc_consume_write_grant(cli, &oap->oap_brw_page);
2865                 if (transient) {
2866                         cli->cl_dirty_transit += CFS_PAGE_SIZE;
2867                         cfs_atomic_inc(&obd_dirty_transit_pages);
2868                         oap->oap_brw_flags |= OBD_BRW_NOCACHE;
2869                 }
2870         }
2871         return has_grant;
2872 }
2873
2874 /* Caller must hold loi_list_lock - we drop/regain it if we need to wait for
2875  * grant or cache space. */
2876 static int osc_enter_cache(const struct lu_env *env,
2877                            struct client_obd *cli, struct lov_oinfo *loi,
2878                            struct osc_async_page *oap)
2879 {
2880         struct osc_cache_waiter ocw;
2881         struct l_wait_info lwi = { 0 };
2882
2883         ENTRY;
2884
2885         CDEBUG(D_CACHE, "dirty: %ld/%d dirty_max: %ld/%d dropped: %lu "
2886                "grant: %lu\n", cli->cl_dirty, cfs_atomic_read(&obd_dirty_pages),
2887                cli->cl_dirty_max, obd_max_dirty_pages,
2888                cli->cl_lost_grant, cli->cl_avail_grant);
2889
2890         /* force the caller to try sync io.  this can jump the list
2891          * of queued writes and create a discontiguous rpc stream */
2892         if (cli->cl_dirty_max < CFS_PAGE_SIZE || cli->cl_ar.ar_force_sync ||
2893             loi->loi_ar.ar_force_sync)
2894                 RETURN(-EDQUOT);
2895
2896         /* Hopefully normal case - cache space and write credits available */
2897         if (cli->cl_dirty + CFS_PAGE_SIZE <= cli->cl_dirty_max &&
2898             cfs_atomic_read(&obd_dirty_pages) + 1 <= obd_max_dirty_pages &&
2899             osc_enter_cache_try(env, cli, loi, oap, 0))
2900                 RETURN(0);
2901
2902         /* It is safe to block as a cache waiter as long as there is grant
2903          * space available or the hope of additional grant being returned
2904          * when an in flight write completes.  Using the write back cache
2905          * if possible is preferable to sending the data synchronously
2906          * because write pages can then be merged in to large requests.
2907          * The addition of this cache waiter will causing pending write
2908          * pages to be sent immediately. */
2909         if (cli->cl_w_in_flight || cli->cl_avail_grant >= CFS_PAGE_SIZE) {
2910                 cfs_list_add_tail(&ocw.ocw_entry, &cli->cl_cache_waiters);
2911                 cfs_waitq_init(&ocw.ocw_waitq);
2912                 ocw.ocw_oap = oap;
2913                 ocw.ocw_rc = 0;
2914
2915                 loi_list_maint(cli, loi);
2916                 osc_check_rpcs(env, cli);
2917                 client_obd_list_unlock(&cli->cl_loi_list_lock);
2918
2919                 CDEBUG(D_CACHE, "sleeping for cache space\n");
2920                 l_wait_event(ocw.ocw_waitq, ocw_granted(cli, &ocw), &lwi);
2921
2922                 client_obd_list_lock(&cli->cl_loi_list_lock);
2923                 if (!cfs_list_empty(&ocw.ocw_entry)) {
2924                         cfs_list_del(&ocw.ocw_entry);
2925                         RETURN(-EINTR);
2926                 }
2927                 RETURN(ocw.ocw_rc);
2928         }
2929
2930         RETURN(-EDQUOT);
2931 }
2932
2933
2934 int osc_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
2935                         struct lov_oinfo *loi, cfs_page_t *page,
2936                         obd_off offset, const struct obd_async_page_ops *ops,
2937                         void *data, void **res, int nocache,
2938                         struct lustre_handle *lockh)
2939 {
2940         struct osc_async_page *oap;
2941
2942         ENTRY;
2943
2944         if (!page)
2945                 return cfs_size_round(sizeof(*oap));
2946
2947         oap = *res;
2948         oap->oap_magic = OAP_MAGIC;
2949         oap->oap_cli = &exp->exp_obd->u.cli;
2950         oap->oap_loi = loi;
2951
2952         oap->oap_caller_ops = ops;
2953         oap->oap_caller_data = data;
2954
2955         oap->oap_page = page;
2956         oap->oap_obj_off = offset;
2957         if (!client_is_remote(exp) &&
2958             cfs_capable(CFS_CAP_SYS_RESOURCE))
2959                 oap->oap_brw_flags = OBD_BRW_NOQUOTA;
2960
2961         LASSERT(!(offset & ~CFS_PAGE_MASK));
2962
2963         CFS_INIT_LIST_HEAD(&oap->oap_pending_item);
2964         CFS_INIT_LIST_HEAD(&oap->oap_urgent_item);
2965         CFS_INIT_LIST_HEAD(&oap->oap_rpc_item);
2966         CFS_INIT_LIST_HEAD(&oap->oap_page_list);
2967
2968         cfs_spin_lock_init(&oap->oap_lock);
2969         CDEBUG(D_CACHE, "oap %p page %p obj off "LPU64"\n", oap, page, offset);
2970         RETURN(0);
2971 }
2972
2973 struct osc_async_page *oap_from_cookie(void *cookie)
2974 {
2975         struct osc_async_page *oap = cookie;
2976         if (oap->oap_magic != OAP_MAGIC)
2977                 return ERR_PTR(-EINVAL);
2978         return oap;
2979 };
2980
2981 int osc_queue_async_io(const struct lu_env *env,
2982                        struct obd_export *exp, struct lov_stripe_md *lsm,
2983                        struct lov_oinfo *loi, void *cookie,
2984                        int cmd, obd_off off, int count,
2985                        obd_flag brw_flags, enum async_flags async_flags)
2986 {
2987         struct client_obd *cli = &exp->exp_obd->u.cli;
2988         struct osc_async_page *oap;
2989         int rc = 0;
2990         ENTRY;
2991
2992         oap = oap_from_cookie(cookie);
2993         if (IS_ERR(oap))
2994                 RETURN(PTR_ERR(oap));
2995
2996         if (cli->cl_import == NULL || cli->cl_import->imp_invalid)
2997                 RETURN(-EIO);
2998
2999         if (!cfs_list_empty(&oap->oap_pending_item) ||
3000             !cfs_list_empty(&oap->oap_urgent_item) ||
3001             !cfs_list_empty(&oap->oap_rpc_item))
3002                 RETURN(-EBUSY);
3003
3004         /* check if the file's owner/group is over quota */
3005         if ((cmd & OBD_BRW_WRITE) && !(cmd & OBD_BRW_NOQUOTA)) {
3006                 struct cl_object *obj;
3007                 struct cl_attr    attr; /* XXX put attr into thread info */
3008                 unsigned int qid[MAXQUOTAS];
3009
3010                 obj = cl_object_top(osc_oap2cl_page(oap)->cp_obj);
3011
3012                 cl_object_attr_lock(obj);
3013                 rc = cl_object_attr_get(env, obj, &attr);
3014                 cl_object_attr_unlock(obj);
3015
3016                 qid[USRQUOTA] = attr.cat_uid;
3017                 qid[GRPQUOTA] = attr.cat_gid;
3018                 if (rc == 0 &&
3019                     lquota_chkdq(quota_interface, cli, qid) == NO_QUOTA)
3020                         rc = -EDQUOT;
3021                 if (rc)
3022                         RETURN(rc);
3023         }
3024
3025         if (loi == NULL)
3026                 loi = lsm->lsm_oinfo[0];
3027
3028         client_obd_list_lock(&cli->cl_loi_list_lock);
3029
3030         LASSERT(off + count <= CFS_PAGE_SIZE);
3031         oap->oap_cmd = cmd;
3032         oap->oap_page_off = off;
3033         oap->oap_count = count;
3034         oap->oap_brw_flags = brw_flags;
3035         /* Give a hint to OST that requests are coming from kswapd - bug19529 */
3036         if (cfs_memory_pressure_get())
3037                 oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
3038         cfs_spin_lock(&oap->oap_lock);
3039         oap->oap_async_flags = async_flags;
3040         cfs_spin_unlock(&oap->oap_lock);
3041
3042         if (cmd & OBD_BRW_WRITE) {
3043                 rc = osc_enter_cache(env, cli, loi, oap);
3044                 if (rc) {
3045                         client_obd_list_unlock(&cli->cl_loi_list_lock);
3046                         RETURN(rc);
3047                 }
3048         }
3049
3050         osc_oap_to_pending(oap);
3051         loi_list_maint(cli, loi);
3052
3053         LOI_DEBUG(loi, "oap %p page %p added for cmd %d\n", oap, oap->oap_page,
3054                   cmd);
3055
3056         osc_check_rpcs(env, cli);
3057         client_obd_list_unlock(&cli->cl_loi_list_lock);
3058
3059         RETURN(0);
3060 }
3061
3062 /* aka (~was & now & flag), but this is more clear :) */
3063 #define SETTING(was, now, flag) (!(was & flag) && (now & flag))
3064
3065 int osc_set_async_flags_base(struct client_obd *cli,
3066                              struct lov_oinfo *loi, struct osc_async_page *oap,
3067                              obd_flag async_flags)
3068 {
3069         struct loi_oap_pages *lop;
3070         int flags = 0;
3071         ENTRY;
3072
3073         LASSERT(!cfs_list_empty(&oap->oap_pending_item));
3074
3075         if (oap->oap_cmd & OBD_BRW_WRITE) {
3076                 lop = &loi->loi_write_lop;
3077         } else {
3078                 lop = &loi->loi_read_lop;
3079         }
3080
3081         if ((oap->oap_async_flags & async_flags) == async_flags)
3082                 RETURN(0);
3083
3084         /* XXX: This introduces a tiny insignificant race for the case if this
3085          * loi already had other urgent items.
3086          */
3087         if (SETTING(oap->oap_async_flags, async_flags, ASYNC_SYNCFS) &&
3088             cfs_list_empty(&oap->oap_rpc_item) &&
3089             cfs_list_empty(&oap->oap_urgent_item)) {
3090                 cfs_list_add_tail(&oap->oap_urgent_item, &lop->lop_urgent);
3091                 flags |= ASYNC_SYNCFS;
3092                 cfs_spin_lock(&oap->oap_lock);
3093                 oap->oap_async_flags |= flags;
3094                 cfs_spin_unlock(&oap->oap_lock);
3095                 loi_list_maint(cli, loi);
3096                 RETURN(0);
3097         }
3098
3099         if (SETTING(oap->oap_async_flags, async_flags, ASYNC_READY))
3100                 flags |= ASYNC_READY;
3101
3102         if (SETTING(oap->oap_async_flags, async_flags, ASYNC_URGENT) &&
3103             cfs_list_empty(&oap->oap_rpc_item)) {
3104                 if (oap->oap_async_flags & ASYNC_HP)
3105                         cfs_list_add(&oap->oap_urgent_item, &lop->lop_urgent);
3106                 else
3107                         cfs_list_add_tail(&oap->oap_urgent_item,
3108                                           &lop->lop_urgent);
3109                 flags |= ASYNC_URGENT;
3110                 loi_list_maint(cli, loi);
3111         }
3112         cfs_spin_lock(&oap->oap_lock);
3113         oap->oap_async_flags |= flags;
3114         cfs_spin_unlock(&oap->oap_lock);
3115
3116         LOI_DEBUG(loi, "oap %p page %p has flags %x\n", oap, oap->oap_page,
3117                         oap->oap_async_flags);
3118         RETURN(0);
3119 }
3120
3121 int osc_teardown_async_page(struct obd_export *exp,
3122                             struct lov_stripe_md *lsm,
3123                             struct lov_oinfo *loi, void *cookie)
3124 {
3125         struct client_obd *cli = &exp->exp_obd->u.cli;
3126         struct loi_oap_pages *lop;
3127         struct osc_async_page *oap;
3128         int rc = 0;
3129         ENTRY;
3130
3131         oap = oap_from_cookie(cookie);
3132         if (IS_ERR(oap))
3133                 RETURN(PTR_ERR(oap));
3134
3135         if (loi == NULL)
3136                 loi = lsm->lsm_oinfo[0];
3137
3138         if (oap->oap_cmd & OBD_BRW_WRITE) {
3139                 lop = &loi->loi_write_lop;
3140         } else {
3141                 lop = &loi->loi_read_lop;
3142         }
3143
3144         client_obd_list_lock(&cli->cl_loi_list_lock);
3145
3146         if (!cfs_list_empty(&oap->oap_rpc_item))
3147                 GOTO(out, rc = -EBUSY);
3148
3149         osc_exit_cache(cli, oap, 0);
3150         osc_wake_cache_waiters(cli);
3151
3152         if (!cfs_list_empty(&oap->oap_urgent_item)) {
3153                 cfs_list_del_init(&oap->oap_urgent_item);
3154                 cfs_spin_lock(&oap->oap_lock);
3155                 oap->oap_async_flags &= ~(ASYNC_URGENT | ASYNC_HP |
3156                                           ASYNC_SYNCFS);
3157                 cfs_spin_unlock(&oap->oap_lock);
3158         }
3159         if (!cfs_list_empty(&oap->oap_pending_item)) {
3160                 cfs_list_del_init(&oap->oap_pending_item);
3161                 lop_update_pending(cli, lop, oap->oap_cmd, -1);
3162         }