Whamcloud - gitweb
9dc00ce0846f4266c20021b453406c55ad0c0af3
[fs/lustre-release.git] / lustre / mdc / mdc_request.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36
37 #define DEBUG_SUBSYSTEM S_MDC
38
39 #ifdef __KERNEL__
40 # include <linux/module.h>
41 # include <linux/pagemap.h>
42 # include <linux/miscdevice.h>
43 # include <linux/init.h>
44 # include <linux/utsname.h>
45 #else
46 # include <liblustre.h>
47 #endif
48
49 #include <lustre_acl.h>
50 #include <lustre_ioctl.h>
51 #include <obd_class.h>
52 #include <lustre_lmv.h>
53 #include <lustre_fid.h>
54 #include <lprocfs_status.h>
55 #include <lustre_param.h>
56 #include <lustre_log.h>
57 #include <cl_object.h>
58 #include <lclient.h>
59
60 #include "mdc_internal.h"
61
62 #define REQUEST_MINOR 244
63
64 struct mdc_renew_capa_args {
65         struct obd_capa        *ra_oc;
66         renew_capa_cb_t         ra_cb;
67 };
68
69 static int mdc_cleanup(struct obd_device *obd);
70
71 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
72                     const struct req_msg_field *field, struct obd_capa **oc)
73 {
74         struct lustre_capa *capa;
75         struct obd_capa *c;
76         ENTRY;
77
78         /* swabbed already in mdc_enqueue */
79         capa = req_capsule_server_get(&req->rq_pill, field);
80         if (capa == NULL)
81                 RETURN(-EPROTO);
82
83         c = alloc_capa(CAPA_SITE_CLIENT);
84         if (IS_ERR(c)) {
85                 CDEBUG(D_INFO, "alloc capa failed!\n");
86                 RETURN(PTR_ERR(c));
87         } else {
88                 c->c_capa = *capa;
89                 *oc = c;
90                 RETURN(0);
91         }
92 }
93
94 static inline int mdc_queue_wait(struct ptlrpc_request *req)
95 {
96         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
97         int rc;
98
99         /* obd_get_request_slot() ensures that this client has no more
100          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
101          * against an MDT. */
102         rc = obd_get_request_slot(cli);
103         if (rc != 0)
104                 return rc;
105
106         rc = ptlrpc_queue_wait(req);
107         obd_put_request_slot(cli);
108
109         return rc;
110 }
111
112 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
113 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
114 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
115                           struct obd_capa **pc, int level, int msg_flags)
116 {
117         struct ptlrpc_request *req;
118         struct mdt_body       *body;
119         int                    rc;
120         ENTRY;
121
122         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
123                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
124         if (req == NULL)
125                 RETURN(-ENOMEM);
126
127         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
128         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
129         req->rq_send_state = level;
130
131         ptlrpc_request_set_replen(req);
132
133         rc = ptlrpc_queue_wait(req);
134         if (rc)
135                 GOTO(out, rc);
136
137         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
138         if (body == NULL)
139                 GOTO(out, rc = -EPROTO);
140
141         if (body->valid & OBD_MD_FLMDSCAPA) {
142                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
143                 if (rc)
144                         GOTO(out, rc);
145         }
146
147         *rootfid = body->fid1;
148         CDEBUG(D_NET,
149                "root fid="DFID", last_committed="LPU64"\n",
150                PFID(rootfid),
151                lustre_msg_get_last_committed(req->rq_repmsg));
152         EXIT;
153 out:
154         ptlrpc_req_finished(req);
155         return rc;
156 }
157
158 /* This should be mdc_get_info("rootfid") */
159 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
160                   struct obd_capa **pc)
161 {
162         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
163                               LUSTRE_IMP_FULL, 0);
164 }
165
166 /*
167  * This function now is known to always saying that it will receive 4 buffers
168  * from server. Even for cases when acl_size and md_size is zero, RPC header
169  * will contain 4 fields and RPC itself will contain zero size fields. This is
170  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
171  * and thus zero, it shrinks it, making zero size. The same story about
172  * md_size. And this is course of problem when client waits for smaller number
173  * of fields. This issue will be fixed later when client gets aware of RPC
174  * layouts.  --umka
175  */
176 static int mdc_getattr_common(struct obd_export *exp,
177                               struct ptlrpc_request *req)
178 {
179         struct req_capsule *pill = &req->rq_pill;
180         struct mdt_body    *body;
181         void               *eadata;
182         int                 rc;
183         ENTRY;
184
185         /* Request message already built. */
186         rc = ptlrpc_queue_wait(req);
187         if (rc != 0)
188                 RETURN(rc);
189
190         /* sanity check for the reply */
191         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
192         if (body == NULL)
193                 RETURN(-EPROTO);
194
195         CDEBUG(D_NET, "mode: %o\n", body->mode);
196
197         mdc_update_max_ea_from_body(exp, body);
198         if (body->eadatasize != 0) {
199                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
200                                                       body->eadatasize);
201                 if (eadata == NULL)
202                         RETURN(-EPROTO);
203         }
204
205         if (body->valid & OBD_MD_FLRMTPERM) {
206                 struct mdt_remote_perm *perm;
207
208                 LASSERT(client_is_remote(exp));
209                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
210                                                 lustre_swab_mdt_remote_perm);
211                 if (perm == NULL)
212                         RETURN(-EPROTO);
213         }
214
215         if (body->valid & OBD_MD_FLMDSCAPA) {
216                 struct lustre_capa *capa;
217                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
218                 if (capa == NULL)
219                         RETURN(-EPROTO);
220         }
221
222         RETURN(0);
223 }
224
225 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
226                 struct ptlrpc_request **request)
227 {
228         struct ptlrpc_request *req;
229         int                    rc;
230         ENTRY;
231
232         /* Single MDS without an LMV case */
233         if (op_data->op_flags & MF_GET_MDT_IDX) {
234                 op_data->op_mds = 0;
235                 RETURN(0);
236         }
237         *request = NULL;
238         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
239         if (req == NULL)
240                 RETURN(-ENOMEM);
241
242         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
243
244         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
245         if (rc) {
246                 ptlrpc_request_free(req);
247                 RETURN(rc);
248         }
249
250         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
251                       op_data->op_valid, op_data->op_mode, -1, 0);
252
253         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
254                              op_data->op_mode);
255         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
256                 LASSERT(client_is_remote(exp));
257                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
258                                      sizeof(struct mdt_remote_perm));
259         }
260         ptlrpc_request_set_replen(req);
261
262         rc = mdc_getattr_common(exp, req);
263         if (rc)
264                 ptlrpc_req_finished(req);
265         else
266                 *request = req;
267         RETURN(rc);
268 }
269
270 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
271                      struct ptlrpc_request **request)
272 {
273         struct ptlrpc_request *req;
274         int                    rc;
275         ENTRY;
276
277         *request = NULL;
278         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
279                                    &RQF_MDS_GETATTR_NAME);
280         if (req == NULL)
281                 RETURN(-ENOMEM);
282
283         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
284         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
285                              op_data->op_namelen + 1);
286
287         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
288         if (rc) {
289                 ptlrpc_request_free(req);
290                 RETURN(rc);
291         }
292
293         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
294                       op_data->op_valid, op_data->op_mode,
295                       op_data->op_suppgids[0], 0);
296
297         if (op_data->op_name) {
298                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
299                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
300                                 op_data->op_namelen);
301                 memcpy(name, op_data->op_name, op_data->op_namelen);
302         }
303
304         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
305                              op_data->op_mode);
306         ptlrpc_request_set_replen(req);
307
308         rc = mdc_getattr_common(exp, req);
309         if (rc)
310                 ptlrpc_req_finished(req);
311         else
312                 *request = req;
313         RETURN(rc);
314 }
315
316 static int mdc_is_subdir(struct obd_export *exp,
317                          const struct lu_fid *pfid,
318                          const struct lu_fid *cfid,
319                          struct ptlrpc_request **request)
320 {
321         struct ptlrpc_request  *req;
322         int                     rc;
323
324         ENTRY;
325
326         *request = NULL;
327         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
328                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
329                                         MDS_IS_SUBDIR);
330         if (req == NULL)
331                 RETURN(-ENOMEM);
332
333         mdc_is_subdir_pack(req, pfid, cfid, 0);
334         ptlrpc_request_set_replen(req);
335
336         rc = ptlrpc_queue_wait(req);
337         if (rc && rc != -EREMOTE)
338                 ptlrpc_req_finished(req);
339         else
340                 *request = req;
341         RETURN(rc);
342 }
343
344 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
345                             const struct lu_fid *fid,
346                             struct obd_capa *oc, int opcode, obd_valid valid,
347                             const char *xattr_name, const char *input,
348                             int input_size, int output_size, int flags,
349                             __u32 suppgid, struct ptlrpc_request **request)
350 {
351         struct ptlrpc_request *req;
352         int   xattr_namelen = 0;
353         char *tmp;
354         int   rc;
355         ENTRY;
356
357         *request = NULL;
358         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
359         if (req == NULL)
360                 RETURN(-ENOMEM);
361
362         mdc_set_capa_size(req, &RMF_CAPA1, oc);
363         if (xattr_name) {
364                 xattr_namelen = strlen(xattr_name) + 1;
365                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
366                                      xattr_namelen);
367         }
368         if (input_size) {
369                 LASSERT(input);
370                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
371                                      input_size);
372         }
373
374         /* Flush local XATTR locks to get rid of a possible cancel RPC */
375         if (opcode == MDS_REINT && fid_is_sane(fid) &&
376             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
377                 struct list_head cancels = LIST_HEAD_INIT(cancels);
378                 int count;
379
380                 /* Without that packing would fail */
381                 if (input_size == 0)
382                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
383                                              RCL_CLIENT, 0);
384
385                 count = mdc_resource_get_unused(exp, fid,
386                                                 &cancels, LCK_EX,
387                                                 MDS_INODELOCK_XATTR);
388
389                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
390                 if (rc) {
391                         ptlrpc_request_free(req);
392                         RETURN(rc);
393                 }
394         } else {
395                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
396                 if (rc) {
397                         ptlrpc_request_free(req);
398                         RETURN(rc);
399                 }
400         }
401
402         if (opcode == MDS_REINT) {
403                 struct mdt_rec_setxattr *rec;
404
405                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
406                          sizeof(struct mdt_rec_reint));
407                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
408                 rec->sx_opcode = REINT_SETXATTR;
409                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
410                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
411                 rec->sx_cap    = cfs_curproc_cap_pack();
412                 rec->sx_suppgid1 = suppgid;
413                 rec->sx_suppgid2 = -1;
414                 rec->sx_fid    = *fid;
415                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
416                 rec->sx_time   = cfs_time_current_sec();
417                 rec->sx_size   = output_size;
418                 rec->sx_flags  = flags;
419
420                 mdc_pack_capa(req, &RMF_CAPA1, oc);
421         } else {
422                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
423         }
424
425         if (xattr_name) {
426                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
427                 memcpy(tmp, xattr_name, xattr_namelen);
428         }
429         if (input_size) {
430                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
431                 memcpy(tmp, input, input_size);
432         }
433
434         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
435                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
436                                      RCL_SERVER, output_size);
437         ptlrpc_request_set_replen(req);
438
439         /* make rpc */
440         if (opcode == MDS_REINT)
441                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
442
443         rc = ptlrpc_queue_wait(req);
444
445         if (opcode == MDS_REINT)
446                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
447
448         if (rc)
449                 ptlrpc_req_finished(req);
450         else
451                 *request = req;
452         RETURN(rc);
453 }
454
455 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
456                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
457                  const char *input, int input_size, int output_size,
458                  int flags, __u32 suppgid, struct ptlrpc_request **request)
459 {
460         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
461                                 fid, oc, MDS_REINT, valid, xattr_name,
462                                 input, input_size, output_size, flags,
463                                 suppgid, request);
464 }
465
466 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
467                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
468                  const char *input, int input_size, int output_size,
469                  int flags, struct ptlrpc_request **request)
470 {
471         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
472                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
473                                 input, input_size, output_size, flags,
474                                 -1, request);
475 }
476
477 #ifdef CONFIG_FS_POSIX_ACL
478 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
479 {
480         struct req_capsule     *pill = &req->rq_pill;
481         struct mdt_body        *body = md->body;
482         struct posix_acl       *acl;
483         void                   *buf;
484         int                     rc;
485         ENTRY;
486
487         if (!body->aclsize)
488                 RETURN(0);
489
490         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
491
492         if (!buf)
493                 RETURN(-EPROTO);
494
495         acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
496         if (IS_ERR(acl)) {
497                 rc = PTR_ERR(acl);
498                 CERROR("convert xattr to acl: %d\n", rc);
499                 RETURN(rc);
500         }
501
502         rc = posix_acl_valid(acl);
503         if (rc) {
504                 CERROR("validate acl: %d\n", rc);
505                 posix_acl_release(acl);
506                 RETURN(rc);
507         }
508
509         md->posix_acl = acl;
510         RETURN(0);
511 }
512 #else
513 #define mdc_unpack_acl(req, md) 0
514 #endif
515
516 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
517                       struct obd_export *dt_exp, struct obd_export *md_exp,
518                       struct lustre_md *md)
519 {
520         struct req_capsule *pill = &req->rq_pill;
521         int rc;
522         ENTRY;
523
524         LASSERT(md);
525         memset(md, 0, sizeof(*md));
526
527         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
528         LASSERT(md->body != NULL);
529
530         if (md->body->valid & OBD_MD_FLEASIZE) {
531                 int lmmsize;
532                 struct lov_mds_md *lmm;
533
534                 if (!S_ISREG(md->body->mode)) {
535                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
536                                "regular file, but is not\n");
537                         GOTO(out, rc = -EPROTO);
538                 }
539
540                 if (md->body->eadatasize == 0) {
541                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
542                                "but eadatasize 0\n");
543                         GOTO(out, rc = -EPROTO);
544                 }
545                 lmmsize = md->body->eadatasize;
546                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
547                 if (!lmm)
548                         GOTO(out, rc = -EPROTO);
549
550                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
551                 if (rc < 0)
552                         GOTO(out, rc);
553
554                 if (rc < sizeof(*md->lsm)) {
555                         CDEBUG(D_INFO, "lsm size too small: "
556                                "rc < sizeof (*md->lsm) (%d < %d)\n",
557                                rc, (int)sizeof(*md->lsm));
558                         GOTO(out, rc = -EPROTO);
559                 }
560
561         } else if (md->body->valid & OBD_MD_FLDIREA) {
562                 int lmvsize;
563                 struct lov_mds_md *lmv;
564
565                 if(!S_ISDIR(md->body->mode)) {
566                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
567                                "directory, but is not\n");
568                         GOTO(out, rc = -EPROTO);
569                 }
570
571                 if (md->body->eadatasize == 0) {
572                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
573                                "but eadatasize 0\n");
574                         RETURN(-EPROTO);
575                 }
576                 if (md->body->valid & OBD_MD_MEA) {
577                         lmvsize = md->body->eadatasize;
578                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
579                                                            lmvsize);
580                         if (!lmv)
581                                 GOTO(out, rc = -EPROTO);
582
583                         rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
584                                           lmvsize);
585                         if (rc < 0)
586                                 GOTO(out, rc);
587
588                         if (rc < sizeof(*md->lmv)) {
589                                 CDEBUG(D_INFO, "size too small:  "
590                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
591                                         rc, (int)sizeof(*md->lmv));
592                                 GOTO(out, rc = -EPROTO);
593                         }
594                 }
595         }
596         rc = 0;
597
598         if (md->body->valid & OBD_MD_FLRMTPERM) {
599                 /* remote permission */
600                 LASSERT(client_is_remote(exp));
601                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
602                                                 lustre_swab_mdt_remote_perm);
603                 if (!md->remote_perm)
604                         GOTO(out, rc = -EPROTO);
605         }
606         else if (md->body->valid & OBD_MD_FLACL) {
607                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
608                  * only when aclsize != 0 there's an actual segment for ACL
609                  * in reply buffer.
610                  */
611                 if (md->body->aclsize) {
612                         rc = mdc_unpack_acl(req, md);
613                         if (rc)
614                                 GOTO(out, rc);
615 #ifdef CONFIG_FS_POSIX_ACL
616                 } else {
617                         md->posix_acl = NULL;
618 #endif
619                 }
620         }
621         if (md->body->valid & OBD_MD_FLMDSCAPA) {
622                 struct obd_capa *oc = NULL;
623
624                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
625                 if (rc)
626                         GOTO(out, rc);
627                 md->mds_capa = oc;
628         }
629
630         if (md->body->valid & OBD_MD_FLOSSCAPA) {
631                 struct obd_capa *oc = NULL;
632
633                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
634                 if (rc)
635                         GOTO(out, rc);
636                 md->oss_capa = oc;
637         }
638
639         EXIT;
640 out:
641         if (rc) {
642                 if (md->oss_capa) {
643                         capa_put(md->oss_capa);
644                         md->oss_capa = NULL;
645                 }
646                 if (md->mds_capa) {
647                         capa_put(md->mds_capa);
648                         md->mds_capa = NULL;
649                 }
650 #ifdef CONFIG_FS_POSIX_ACL
651                 posix_acl_release(md->posix_acl);
652 #endif
653                 if (md->lsm)
654                         obd_free_memmd(dt_exp, &md->lsm);
655         }
656         return rc;
657 }
658
659 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
660 {
661         ENTRY;
662         RETURN(0);
663 }
664
665 /**
666  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
667  * RPC chains.
668  */
669 void mdc_replay_open(struct ptlrpc_request *req)
670 {
671         struct md_open_data *mod = req->rq_cb_data;
672         struct ptlrpc_request *close_req;
673         struct obd_client_handle *och;
674         struct lustre_handle old;
675         struct mdt_body *body;
676         ENTRY;
677
678         if (mod == NULL) {
679                 DEBUG_REQ(D_ERROR, req,
680                           "Can't properly replay without open data.");
681                 EXIT;
682                 return;
683         }
684
685         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
686         LASSERT(body != NULL);
687
688         och = mod->mod_och;
689         if (och != NULL) {
690                 struct lustre_handle *file_fh;
691
692                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
693
694                 file_fh = &och->och_fh;
695                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
696                        file_fh->cookie, body->handle.cookie);
697                 old = *file_fh;
698                 *file_fh = body->handle;
699         }
700         close_req = mod->mod_close_req;
701         if (close_req != NULL) {
702                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
703                 struct mdt_ioepoch *epoch;
704
705                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
706                 epoch = req_capsule_client_get(&close_req->rq_pill,
707                                                &RMF_MDT_EPOCH);
708                 LASSERT(epoch);
709
710                 if (och != NULL)
711                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
712                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
713                 epoch->handle = body->handle;
714         }
715         EXIT;
716 }
717
718 void mdc_commit_open(struct ptlrpc_request *req)
719 {
720         struct md_open_data *mod = req->rq_cb_data;
721         if (mod == NULL)
722                 return;
723
724         /**
725          * No need to touch md_open_data::mod_och, it holds a reference on
726          * \var mod and will zero references to each other, \var mod will be
727          * freed after that when md_open_data::mod_och will put the reference.
728          */
729
730         /**
731          * Do not let open request to disappear as it still may be needed
732          * for close rpc to happen (it may happen on evict only, otherwise
733          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
734          * called), just mark this rpc as committed to distinguish these 2
735          * cases, see mdc_close() for details. The open request reference will
736          * be put along with freeing \var mod.
737          */
738         ptlrpc_request_addref(req);
739         spin_lock(&req->rq_lock);
740         req->rq_committed = 1;
741         spin_unlock(&req->rq_lock);
742         req->rq_cb_data = NULL;
743         obd_mod_put(mod);
744 }
745
746 int mdc_set_open_replay_data(struct obd_export *exp,
747                              struct obd_client_handle *och,
748                              struct lookup_intent *it)
749 {
750         struct md_open_data     *mod;
751         struct mdt_rec_create   *rec;
752         struct mdt_body         *body;
753         struct ptlrpc_request   *open_req = it->d.lustre.it_data;
754         struct obd_import       *imp = open_req->rq_import;
755         ENTRY;
756
757         if (!open_req->rq_replay)
758                 RETURN(0);
759
760         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
761         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
762         LASSERT(rec != NULL);
763         /* Incoming message in my byte order (it's been swabbed). */
764         /* Outgoing messages always in my byte order. */
765         LASSERT(body != NULL);
766
767         /* Only if the import is replayable, we set replay_open data */
768         if (och && imp->imp_replayable) {
769                 mod = obd_mod_alloc();
770                 if (mod == NULL) {
771                         DEBUG_REQ(D_ERROR, open_req,
772                                   "Can't allocate md_open_data");
773                         RETURN(0);
774                 }
775
776                 /**
777                  * Take a reference on \var mod, to be freed on mdc_close().
778                  * It protects \var mod from being freed on eviction (commit
779                  * callback is called despite rq_replay flag).
780                  * Another reference for \var och.
781                  */
782                 obd_mod_get(mod);
783                 obd_mod_get(mod);
784
785                 spin_lock(&open_req->rq_lock);
786                 och->och_mod = mod;
787                 mod->mod_och = och;
788                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
789                                      it_disposition(it, DISP_OPEN_STRIPE);
790                 mod->mod_open_req = open_req;
791                 open_req->rq_cb_data = mod;
792                 open_req->rq_commit_cb = mdc_commit_open;
793                 spin_unlock(&open_req->rq_lock);
794         }
795
796         rec->cr_fid2 = body->fid1;
797         rec->cr_ioepoch = body->ioepoch;
798         rec->cr_old_handle.cookie = body->handle.cookie;
799         open_req->rq_replay_cb = mdc_replay_open;
800         if (!fid_is_sane(&body->fid1)) {
801                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
802                           "insane fid");
803                 LBUG();
804         }
805
806         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
807         RETURN(0);
808 }
809
810 static void mdc_free_open(struct md_open_data *mod)
811 {
812         int committed = 0;
813
814         if (mod->mod_is_create == 0 &&
815             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
816                 committed = 1;
817
818         LASSERT(mod->mod_open_req->rq_replay == 0);
819
820         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
821
822         ptlrpc_request_committed(mod->mod_open_req, committed);
823         if (mod->mod_close_req)
824                 ptlrpc_request_committed(mod->mod_close_req, committed);
825 }
826
827 int mdc_clear_open_replay_data(struct obd_export *exp,
828                                struct obd_client_handle *och)
829 {
830         struct md_open_data *mod = och->och_mod;
831         ENTRY;
832
833         /**
834          * It is possible to not have \var mod in a case of eviction between
835          * lookup and ll_file_open().
836          **/
837         if (mod == NULL)
838                 RETURN(0);
839
840         LASSERT(mod != LP_POISON);
841         LASSERT(mod->mod_open_req != NULL);
842         mdc_free_open(mod);
843
844         mod->mod_och = NULL;
845         och->och_mod = NULL;
846         obd_mod_put(mod);
847
848         RETURN(0);
849 }
850
851 /* Prepares the request for the replay by the given reply */
852 static void mdc_close_handle_reply(struct ptlrpc_request *req,
853                                    struct md_op_data *op_data, int rc) {
854         struct mdt_body  *repbody;
855         struct mdt_ioepoch *epoch;
856
857         if (req && rc == -EAGAIN) {
858                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
859                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
860
861                 epoch->flags |= MF_SOM_AU;
862                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
863                         op_data->op_flags |= MF_GETATTR_LOCK;
864         }
865 }
866
867 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
868               struct md_open_data *mod, struct ptlrpc_request **request)
869 {
870         struct obd_device     *obd = class_exp2obd(exp);
871         struct ptlrpc_request *req;
872         struct req_format     *req_fmt;
873         int                    rc;
874         int                    saved_rc = 0;
875         ENTRY;
876
877         req_fmt = &RQF_MDS_CLOSE;
878         if (op_data->op_bias & MDS_HSM_RELEASE) {
879                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
880
881                 /* allocate a FID for volatile file */
882                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
883                 if (rc < 0) {
884                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
885                                obd->obd_name, PFID(&op_data->op_fid1), rc);
886                         /* save the errcode and proceed to close */
887                         saved_rc = rc;
888                 }
889         }
890
891         *request = NULL;
892         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
893         if (req == NULL)
894                 RETURN(-ENOMEM);
895
896         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
897
898         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
899         if (rc) {
900                 ptlrpc_request_free(req);
901                 RETURN(rc);
902         }
903
904         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
905          * portal whose threads are not taking any DLM locks and are therefore
906          * always progressing */
907         req->rq_request_portal = MDS_READPAGE_PORTAL;
908         ptlrpc_at_set_req_timeout(req);
909
910         /* Ensure that this close's handle is fixed up during replay. */
911         if (likely(mod != NULL)) {
912                 LASSERTF(mod->mod_open_req != NULL &&
913                          mod->mod_open_req->rq_type != LI_POISON,
914                          "POISONED open %p!\n", mod->mod_open_req);
915
916                 mod->mod_close_req = req;
917
918                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
919                 /* We no longer want to preserve this open for replay even
920                  * though the open was committed. b=3632, b=3633 */
921                 spin_lock(&mod->mod_open_req->rq_lock);
922                 mod->mod_open_req->rq_replay = 0;
923                 spin_unlock(&mod->mod_open_req->rq_lock);
924         } else {
925                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
926         }
927
928         mdc_close_pack(req, op_data);
929
930         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
931                              obd->u.cli.cl_default_mds_easize);
932         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
933                              obd->u.cli.cl_default_mds_cookiesize);
934
935         ptlrpc_request_set_replen(req);
936
937         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
938         rc = ptlrpc_queue_wait(req);
939         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
940
941         if (req->rq_repmsg == NULL) {
942                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
943                        req->rq_status);
944                 if (rc == 0)
945                         rc = req->rq_status ?: -EIO;
946         } else if (rc == 0 || rc == -EAGAIN) {
947                 struct mdt_body *body;
948
949                 rc = lustre_msg_get_status(req->rq_repmsg);
950                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
951                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
952                                   "= %d", rc);
953                         if (rc > 0)
954                                 rc = -rc;
955                 }
956                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
957                 if (body == NULL)
958                         rc = -EPROTO;
959         } else if (rc == -ESTALE) {
960                 /**
961                  * it can be allowed error after 3633 if open was committed and
962                  * server failed before close was sent. Let's check if mod
963                  * exists and return no error in that case
964                  */
965                 if (mod) {
966                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
967                         LASSERT(mod->mod_open_req != NULL);
968                         if (mod->mod_open_req->rq_committed)
969                                 rc = 0;
970                 }
971         }
972
973         if (mod) {
974                 if (rc != 0)
975                         mod->mod_close_req = NULL;
976                 /* Since now, mod is accessed through open_req only,
977                  * thus close req does not keep a reference on mod anymore. */
978                 obd_mod_put(mod);
979         }
980         *request = req;
981         mdc_close_handle_reply(req, op_data, rc);
982         RETURN(rc < 0 ? rc : saved_rc);
983 }
984
985 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
986                      struct md_open_data *mod)
987 {
988         struct obd_device     *obd = class_exp2obd(exp);
989         struct ptlrpc_request *req;
990         int                    rc;
991         ENTRY;
992
993         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
994                                    &RQF_MDS_DONE_WRITING);
995         if (req == NULL)
996                 RETURN(-ENOMEM);
997
998         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
999         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
1000         if (rc) {
1001                 ptlrpc_request_free(req);
1002                 RETURN(rc);
1003         }
1004
1005         if (mod != NULL) {
1006                 LASSERTF(mod->mod_open_req != NULL &&
1007                          mod->mod_open_req->rq_type != LI_POISON,
1008                          "POISONED setattr %p!\n", mod->mod_open_req);
1009
1010                 mod->mod_close_req = req;
1011                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
1012                 /* We no longer want to preserve this setattr for replay even
1013                  * though the open was committed. b=3632, b=3633 */
1014                 spin_lock(&mod->mod_open_req->rq_lock);
1015                 mod->mod_open_req->rq_replay = 0;
1016                 spin_unlock(&mod->mod_open_req->rq_lock);
1017         }
1018
1019         mdc_close_pack(req, op_data);
1020         ptlrpc_request_set_replen(req);
1021
1022         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1023         rc = ptlrpc_queue_wait(req);
1024         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1025
1026         if (rc == -ESTALE) {
1027                 /**
1028                  * it can be allowed error after 3633 if open or setattr were
1029                  * committed and server failed before close was sent.
1030                  * Let's check if mod exists and return no error in that case
1031                  */
1032                 if (mod) {
1033                         LASSERT(mod->mod_open_req != NULL);
1034                         if (mod->mod_open_req->rq_committed)
1035                                 rc = 0;
1036                 }
1037         }
1038
1039         if (mod) {
1040                 if (rc != 0)
1041                         mod->mod_close_req = NULL;
1042                 LASSERT(mod->mod_open_req != NULL);
1043                 mdc_free_open(mod);
1044
1045                 /* Since now, mod is accessed through setattr req only,
1046                  * thus DW req does not keep a reference on mod anymore. */
1047                 obd_mod_put(mod);
1048         }
1049
1050         mdc_close_handle_reply(req, op_data, rc);
1051         ptlrpc_req_finished(req);
1052         RETURN(rc);
1053 }
1054
1055 #ifdef HAVE_SPLIT_SUPPORT
1056 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
1057                  const struct page *page, int offset)
1058 {
1059         struct ptlrpc_request   *req;
1060         struct ptlrpc_bulk_desc *desc;
1061         int                      rc;
1062         ENTRY;
1063
1064         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
1065         if (req == NULL)
1066                 RETURN(-ENOMEM);
1067
1068         /* FIXME: capa doesn't support split yet */
1069         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
1070
1071         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
1072         if (rc) {
1073                 ptlrpc_request_free(req);
1074                 RETURN(rc);
1075         }
1076
1077         req->rq_request_portal = MDS_READPAGE_PORTAL;
1078         ptlrpc_at_set_req_timeout(req);
1079
1080         desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
1081         if (desc == NULL)
1082                 GOTO(out, rc = -ENOMEM);
1083
1084         /* NB req now owns desc and will free it when it gets freed. */
1085         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1086         mdc_readdir_pack(req, 0, offset, fid, NULL);
1087
1088         ptlrpc_request_set_replen(req);
1089         rc = ptlrpc_queue_wait(req);
1090         if (rc)
1091                 GOTO(out, rc);
1092
1093         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1094 out:
1095         ptlrpc_req_finished(req);
1096         return rc;
1097 }
1098 EXPORT_SYMBOL(mdc_sendpage);
1099 #endif
1100
1101 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1102                        __u64 offset, struct obd_capa *oc,
1103                        struct page **pages, int npages,
1104                        struct ptlrpc_request **request)
1105 {
1106         struct ptlrpc_request   *req;
1107         struct ptlrpc_bulk_desc *desc;
1108         int                      i;
1109         wait_queue_head_t        waitq;
1110         int                      resends = 0;
1111         struct l_wait_info       lwi;
1112         int                      rc;
1113         ENTRY;
1114
1115         *request = NULL;
1116         init_waitqueue_head(&waitq);
1117
1118 restart_bulk:
1119         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1120         if (req == NULL)
1121                 RETURN(-ENOMEM);
1122
1123         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1124
1125         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1126         if (rc) {
1127                 ptlrpc_request_free(req);
1128                 RETURN(rc);
1129         }
1130
1131         req->rq_request_portal = MDS_READPAGE_PORTAL;
1132         ptlrpc_at_set_req_timeout(req);
1133
1134         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
1135                                     MDS_BULK_PORTAL);
1136         if (desc == NULL) {
1137                 ptlrpc_request_free(req);
1138                 RETURN(-ENOMEM);
1139         }
1140
1141         /* NB req now owns desc and will free it when it gets freed */
1142         for (i = 0; i < npages; i++)
1143                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1144
1145         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
1146
1147         ptlrpc_request_set_replen(req);
1148         rc = ptlrpc_queue_wait(req);
1149         if (rc) {
1150                 ptlrpc_req_finished(req);
1151                 if (rc != -ETIMEDOUT)
1152                         RETURN(rc);
1153
1154                 resends++;
1155                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1156                         CERROR("%s: too many resend retries: rc = %d\n",
1157                                exp->exp_obd->obd_name, -EIO);
1158                         RETURN(-EIO);
1159                 }
1160                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1161                                        NULL);
1162                 l_wait_event(waitq, 0, &lwi);
1163
1164                 goto restart_bulk;
1165         }
1166
1167         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1168                                           req->rq_bulk->bd_nob_transferred);
1169         if (rc < 0) {
1170                 ptlrpc_req_finished(req);
1171                 RETURN(rc);
1172         }
1173
1174         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1175                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1176                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1177                        PAGE_CACHE_SIZE * npages);
1178                 ptlrpc_req_finished(req);
1179                 RETURN(-EPROTO);
1180         }
1181
1182         *request = req;
1183         RETURN(0);
1184 }
1185
1186 #ifdef __KERNEL__
1187 static void mdc_release_page(struct page *page, int remove)
1188 {
1189         if (remove) {
1190                 lock_page(page);
1191                 if (likely(page->mapping != NULL))
1192                         truncate_complete_page(page->mapping, page);
1193                 unlock_page(page);
1194         }
1195         page_cache_release(page);
1196 }
1197
1198 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1199                                     __u64 *start, __u64 *end, int hash64)
1200 {
1201         /*
1202          * Complement of hash is used as an index so that
1203          * radix_tree_gang_lookup() can be used to find a page with starting
1204          * hash _smaller_ than one we are looking for.
1205          */
1206         unsigned long offset = hash_x_index(*hash, hash64);
1207         struct page *page;
1208         int found;
1209
1210         spin_lock_irq(&mapping->tree_lock);
1211         found = radix_tree_gang_lookup(&mapping->page_tree,
1212                                        (void **)&page, offset, 1);
1213         if (found > 0) {
1214                 struct lu_dirpage *dp;
1215
1216                 page_cache_get(page);
1217                 spin_unlock_irq(&mapping->tree_lock);
1218                 /*
1219                  * In contrast to find_lock_page() we are sure that directory
1220                  * page cannot be truncated (while DLM lock is held) and,
1221                  * hence, can avoid restart.
1222                  *
1223                  * In fact, page cannot be locked here at all, because
1224                  * mdc_read_page_remote does synchronous io.
1225                  */
1226                 wait_on_page_locked(page);
1227                 if (PageUptodate(page)) {
1228                         dp = kmap(page);
1229                         if (BITS_PER_LONG == 32 && hash64) {
1230                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1231                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1232                                 *hash  = *hash >> 32;
1233                         } else {
1234                                 *start = le64_to_cpu(dp->ldp_hash_start);
1235                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1236                         }
1237                         if (unlikely(*start == 1 && *hash == 0))
1238                                 *hash = *start;
1239                         else
1240                                 LASSERTF(*start <= *hash, "start = "LPX64
1241                                          ",end = "LPX64",hash = "LPX64"\n",
1242                                          *start, *end, *hash);
1243                         CDEBUG(D_VFSTRACE, "page%lu [%llu %llu], hash"LPU64"\n",
1244                                offset, *start, *end, *hash);
1245                         if (*hash > *end) {
1246                                 kunmap(page);
1247                                 mdc_release_page(page, 0);
1248                                 page = NULL;
1249                         } else if (*end != *start && *hash == *end) {
1250                                 /*
1251                                  * upon hash collision, remove this page,
1252                                  * otherwise put page reference, and
1253                                  * ll_get_dir_page() will issue RPC to fetch
1254                                  * the page we want.
1255                                  */
1256                                 kunmap(page);
1257                                 mdc_release_page(page,
1258                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1259                                 page = NULL;
1260                         }
1261                 } else {
1262                         page_cache_release(page);
1263                         page = ERR_PTR(-EIO);
1264                 }
1265         } else {
1266                 spin_unlock_irq(&mapping->tree_lock);
1267                 page = NULL;
1268         }
1269         return page;
1270 }
1271
1272 /*
1273  * Adjust a set of pages, each page containing an array of lu_dirpages,
1274  * so that each page can be used as a single logical lu_dirpage.
1275  *
1276  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1277  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1278  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1279  * value is used as a cookie to request the next lu_dirpage in a
1280  * directory listing that spans multiple pages (two in this example):
1281  *   ________
1282  *  |        |
1283  * .|--------v-------   -----.
1284  * |s|e|f|p|ent|ent| ... |ent|
1285  * '--|--------------   -----'   Each CFS_PAGE contains a single
1286  *    '------.                   lu_dirpage.
1287  * .---------v-------   -----.
1288  * |s|e|f|p|ent| 0 | ... | 0 |
1289  * '-----------------   -----'
1290  *
1291  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1292  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1293  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1294  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1295  * after it in the same CFS_PAGE (arrows simplified for brevity, but
1296  * in general e0==s1, e1==s2, etc.):
1297  *
1298  * .--------------------   -----.
1299  * |s0|e0|f0|p|ent|ent| ... |ent|
1300  * |---v----------------   -----|
1301  * |s1|e1|f1|p|ent|ent| ... |ent|
1302  * |---v----------------   -----|  Here, each CFS_PAGE contains
1303  *             ...                 multiple lu_dirpages.
1304  * |---v----------------   -----|
1305  * |s'|e'|f'|p|ent|ent| ... |ent|
1306  * '---|----------------   -----'
1307  *     v
1308  * .----------------------------.
1309  * |        next CFS_PAGE       |
1310  *
1311  * This structure is transformed into a single logical lu_dirpage as follows:
1312  *
1313  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1314  *   labeled 'next CFS_PAGE'.
1315  *
1316  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1317  *   a hash collision with the next page exists.
1318  *
1319  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1320  *   to the first entry of the next lu_dirpage.
1321  */
1322 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1323 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1324 {
1325         int i;
1326
1327         for (i = 0; i < cfs_pgs; i++) {
1328                 struct lu_dirpage       *dp = kmap(pages[i]);
1329                 struct lu_dirpage       *first = dp;
1330                 struct lu_dirent        *end_dirent = NULL;
1331                 struct lu_dirent        *ent;
1332                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1333                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1334
1335                 while (--lu_pgs > 0) {
1336                         ent = lu_dirent_start(dp);
1337                         for (end_dirent = ent; ent != NULL;
1338                              end_dirent = ent, ent = lu_dirent_next(ent));
1339
1340                         /* Advance dp to next lu_dirpage. */
1341                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1342
1343                         /* Check if we've reached the end of the CFS_PAGE. */
1344                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1345                                 break;
1346
1347                         /* Save the hash and flags of this lu_dirpage. */
1348                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1349                         flags = le32_to_cpu(dp->ldp_flags);
1350
1351                         /* Check if lu_dirpage contains no entries. */
1352                         if (end_dirent == NULL)
1353                                 break;
1354
1355                         /* Enlarge the end entry lde_reclen from 0 to
1356                          * first entry of next lu_dirpage. */
1357                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1358                         end_dirent->lde_reclen =
1359                                 cpu_to_le16((char *)(dp->ldp_entries) -
1360                                             (char *)end_dirent);
1361                 }
1362
1363                 first->ldp_hash_end = hash_end;
1364                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1365                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1366
1367                 kunmap(pages[i]);
1368         }
1369         LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1370 }
1371 #else
1372 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1373 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1374
1375 /* parameters for readdir page */
1376 struct readpage_param {
1377         struct md_op_data       *rp_mod;
1378         __u64                   rp_off;
1379         int                     rp_hash64;
1380         struct obd_export       *rp_exp;
1381         struct md_callback      *rp_cb;
1382 };
1383
1384 /**
1385  * Read pages from server.
1386  *
1387  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1388  * a header lu_dirpage which describes the start/end hash, and whether this
1389  * page is empty (contains no dir entry) or hash collide with next page.
1390  * After client receives reply, several pages will be integrated into dir page
1391  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1392  * lu_dirpage for this integrated page will be adjusted.
1393  **/
1394 static int mdc_read_page_remote(void *data, struct page *page0)
1395 {
1396         struct readpage_param   *rp = data;
1397         struct page             **page_pool;
1398         struct page             *page;
1399         struct lu_dirpage       *dp;
1400         int                     rd_pgs = 0; /* number of pages read actually */
1401         int                     npages;
1402         struct md_op_data       *op_data = rp->rp_mod;
1403         struct ptlrpc_request   *req;
1404         int                     max_pages = op_data->op_max_pages;
1405         struct inode            *inode;
1406         struct lu_fid           *fid;
1407         int                     i;
1408         int                     rc;
1409         ENTRY;
1410
1411         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1412         if (op_data->op_mea1 != NULL) {
1413                 __u32 index = op_data->op_stripe_offset;
1414
1415                 inode = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1416                 fid = &op_data->op_mea1->lsm_md_oinfo[index].lmo_fid;
1417         } else {
1418                 inode = op_data->op_data;
1419                 fid = &op_data->op_fid1;
1420         }
1421         LASSERT(inode != NULL);
1422
1423         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1424         if (page_pool != NULL) {
1425                 page_pool[0] = page0;
1426         } else {
1427                 page_pool = &page0;
1428                 max_pages = 1;
1429         }
1430
1431         for (npages = 1; npages < max_pages; npages++) {
1432                 page = page_cache_alloc_cold(inode->i_mapping);
1433                 if (page == NULL)
1434                         break;
1435                 page_pool[npages] = page;
1436         }
1437
1438         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1439                          page_pool, npages, &req);
1440         if (rc == 0) {
1441                 int lu_pgs;
1442
1443                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1444                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1445                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1446                                                         LU_PAGE_SHIFT;
1447                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1448
1449                 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1450                        op_data->op_npages);
1451
1452                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1453
1454                 SetPageUptodate(page0);
1455         }
1456
1457         unlock_page(page0);
1458         ptlrpc_req_finished(req);
1459         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1460         for (i = 1; i < npages; i++) {
1461                 unsigned long   offset;
1462                 __u64           hash;
1463                 int ret;
1464
1465                 page = page_pool[i];
1466
1467                 if (rc < 0 || i >= rd_pgs) {
1468                         page_cache_release(page);
1469                         continue;
1470                 }
1471
1472                 SetPageUptodate(page);
1473
1474                 dp = kmap(page);
1475                 hash = le64_to_cpu(dp->ldp_hash_start);
1476                 kunmap(page);
1477
1478                 offset = hash_x_index(hash, rp->rp_hash64);
1479
1480                 prefetchw(&page->flags);
1481                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1482                                             GFP_KERNEL);
1483                 if (ret == 0)
1484                         unlock_page(page);
1485                 else
1486                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1487                                " rc = %d\n", offset, ret);
1488                 page_cache_release(page);
1489         }
1490
1491         if (page_pool != &page0)
1492                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1493
1494         RETURN(rc);
1495 }
1496
1497 /**
1498  * Read dir page from cache first, if it can not find it, read it from
1499  * server and add into the cache.
1500  */
1501 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1502                          struct md_callback *cb_op, struct page **ppage)
1503 {
1504         struct lookup_intent    it = { .it_op = IT_READDIR };
1505         struct page             *page;
1506         struct inode            *dir = NULL;
1507         struct address_space    *mapping;
1508         struct lu_dirpage       *dp;
1509         __u64                   start = 0;
1510         __u64                   end = 0;
1511         struct lustre_handle    lockh;
1512         struct ptlrpc_request   *enq_req = NULL;
1513         struct readpage_param   rp_param;
1514         int rc;
1515
1516         ENTRY;
1517
1518         *ppage = NULL;
1519
1520         if (op_data->op_mea1 != NULL) {
1521                 __u32 index = op_data->op_stripe_offset;
1522
1523                 dir = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1524         } else {
1525                 dir = op_data->op_data;
1526         }
1527         LASSERT(dir != NULL);
1528
1529         mapping = dir->i_mapping;
1530
1531         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1532                              cb_op->md_blocking_ast, 0);
1533         if (enq_req != NULL)
1534                 ptlrpc_req_finished(enq_req);
1535
1536         if (rc < 0) {
1537                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1538                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1539                 RETURN(rc);
1540         }
1541
1542         rc = 0;
1543         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1544
1545         rp_param.rp_off = op_data->op_hash_offset;
1546         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1547         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1548                                rp_param.rp_hash64);
1549         if (IS_ERR(page)) {
1550                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1551                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1552                        rp_param.rp_off, PTR_ERR(page));
1553                 GOTO(out_unlock, rc = PTR_ERR(page));
1554         } else if (page != NULL) {
1555                 /*
1556                  * XXX nikita: not entirely correct handling of a corner case:
1557                  * suppose hash chain of entries with hash value HASH crosses
1558                  * border between pages P0 and P1. First both P0 and P1 are
1559                  * cached, seekdir() is called for some entry from the P0 part
1560                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1561                  * happens and finds P1, as it starts with matching hash
1562                  * value. Remaining entries from P0 part of the chain are
1563                  * skipped. (Is that really a bug?)
1564                  *
1565                  * Possible solutions: 0. don't cache P1 is such case, handle
1566                  * it as an "overflow" page. 1. invalidate all pages at
1567                  * once. 2. use HASH|1 as an index for P1.
1568                  */
1569                 GOTO(hash_collision, page);
1570         }
1571
1572         rp_param.rp_exp = exp;
1573         rp_param.rp_mod = op_data;
1574         page = read_cache_page(mapping,
1575                                hash_x_index(rp_param.rp_off,
1576                                             rp_param.rp_hash64),
1577                                mdc_read_page_remote, &rp_param);
1578         if (IS_ERR(page)) {
1579                 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1580                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1581                        rp_param.rp_off, PTR_ERR(page));
1582                 GOTO(out_unlock, rc = PTR_ERR(page));
1583         }
1584
1585         wait_on_page_locked(page);
1586         (void)kmap(page);
1587         if (!PageUptodate(page)) {
1588                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1589                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1590                        rp_param.rp_off, -5);
1591                 goto fail;
1592         }
1593         if (!PageChecked(page))
1594                 SetPageChecked(page);
1595         if (PageError(page)) {
1596                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1597                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1598                        rp_param.rp_off, -5);
1599                 goto fail;
1600         }
1601
1602 hash_collision:
1603         dp = page_address(page);
1604         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1605                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1606                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1607                 rp_param.rp_off = op_data->op_hash_offset >> 32;
1608         } else {
1609                 start = le64_to_cpu(dp->ldp_hash_start);
1610                 end   = le64_to_cpu(dp->ldp_hash_end);
1611                 rp_param.rp_off = op_data->op_hash_offset;
1612         }
1613         if (end == start) {
1614                 LASSERT(start == rp_param.rp_off);
1615                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1616 #if BITS_PER_LONG == 32
1617                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1618                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1619                       le64_to_cpu(dp->ldp_hash_end), op_data->op_hash_offset);
1620 #endif
1621
1622                 /*
1623                  * Fetch whole overflow chain...
1624                  *
1625                  * XXX not yet.
1626                  */
1627                 goto fail;
1628         }
1629         *ppage = page;
1630 out_unlock:
1631         lockh.cookie = it.d.lustre.it_lock_handle;
1632         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1633         it.d.lustre.it_lock_handle = 0;
1634         return rc;
1635 fail:
1636         kunmap(page);
1637         mdc_release_page(page, 1);
1638         rc = -EIO;
1639         goto out_unlock;
1640 }
1641
1642 /**
1643  * Read one directory entry from the cache.
1644  */
1645 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1646                    struct md_callback *cb_op, struct lu_dirent **entp,
1647                    struct page **ppage)
1648 {
1649         struct page             *page = NULL;
1650         struct lu_dirpage       *dp;
1651         struct lu_dirent        *ent;
1652         int                     rc = 0;
1653         ENTRY;
1654
1655         CDEBUG(D_INFO, DFID "offset = "LPU64"\n", PFID(&op_data->op_fid1),
1656                op_data->op_hash_offset);
1657
1658         *ppage = NULL;
1659         *entp = NULL;
1660
1661         if (op_data->op_hash_offset == MDS_DIR_END_OFF)
1662                 RETURN(0);
1663
1664         rc = mdc_read_page(exp, op_data, cb_op, &page);
1665         if (rc != 0)
1666                 RETURN(rc);
1667
1668         dp = page_address(page);
1669         for (ent = lu_dirent_start(dp); ent != NULL;
1670              ent = lu_dirent_next(ent)) {
1671                 /* Skip dummy entry */
1672                 if (le16_to_cpu(ent->lde_namelen) == 0)
1673                         continue;
1674
1675                 if (le64_to_cpu(ent->lde_hash) > op_data->op_hash_offset)
1676                         break;
1677         }
1678
1679         /* If it can not find entry in current page, try next page. */
1680         if (ent == NULL) {
1681                 __u64 orig_offset = op_data->op_hash_offset;
1682
1683                 if (le64_to_cpu(dp->ldp_hash_end) == MDS_DIR_END_OFF) {
1684                         mdc_release_page(page, 0);
1685                         RETURN(0);
1686                 }
1687
1688                 op_data->op_hash_offset = le64_to_cpu(dp->ldp_hash_end);
1689                 mdc_release_page(page,
1690                                  le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1691                 rc = mdc_read_page(exp, op_data, cb_op, &page);
1692                 if (rc != 0)
1693                         RETURN(rc);
1694
1695                 if (page != NULL) {
1696                         dp = page_address(page);
1697                         ent = lu_dirent_start(dp);
1698                 }
1699
1700                 op_data->op_hash_offset = orig_offset;
1701         }
1702
1703         *ppage = page;
1704         *entp = ent;
1705
1706         RETURN(rc);
1707 }
1708
1709 #else /* __KERNEL__ */
1710
1711 static struct page
1712 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1713                       const __u64 hash, struct obd_capa *oc)
1714 {
1715         struct ptlrpc_request *req = NULL;
1716         struct page *page;
1717         int rc;
1718
1719         OBD_PAGE_ALLOC(page, 0);
1720         if (page == NULL)
1721                 return ERR_PTR(-ENOMEM);
1722
1723         rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1724         if (req != NULL)
1725                 ptlrpc_req_finished(req);
1726
1727         if (unlikely(rc)) {
1728                 OBD_PAGE_FREE(page);
1729                 return ERR_PTR(rc);
1730         }
1731         return page;
1732 }
1733
1734
1735 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1736                         struct md_callback *cb_op,
1737                         struct page **ppage)
1738 {
1739         struct page *page;
1740         struct lmv_oinfo *lmo;
1741         int rc = 0;
1742
1743         /* No local cache for liblustre, always read entry remotely */
1744         lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1745         page = mdc_read_page_remote(exp, lmo, op_data->op_hash_offset,
1746                                     op_data->op_capa1);
1747         if (IS_ERR(page))
1748                 return PTR_ERR(page);
1749
1750         *ppage = page;
1751
1752         return rc;
1753 }
1754
1755 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1756                    struct md_callback *cb_op, struct lu_dirent **entp,
1757                    struct page **ppage)
1758 {
1759         struct page             *page = NULL;
1760         struct lu_dirpage       *dp;
1761         struct lu_dirent        *ent;
1762         int                     rc;
1763         ENTRY;
1764
1765         rc = mdc_read_page(exp, op_data, cb_op, &page);
1766         if (rc != 0)
1767                 RETURN(rc);
1768
1769         dp = page_address(page);
1770         if (le64_to_cpu(dp->ldp_hash_end) < op_data->op_hash_offset)
1771                 GOTO(out, *entp = NULL);
1772
1773         for (ent = lu_dirent_start(dp); ent != NULL;
1774              ent = lu_dirent_next(ent))
1775                 if (le64_to_cpu(ent->lde_hash) >= op_data->op_hash_offset)
1776                         break;
1777         *entp = ent;
1778 out:
1779
1780         OBD_PAGE_FREE(page);
1781         RETURN(rc);
1782 }
1783
1784 #endif
1785
1786 static int mdc_statfs(const struct lu_env *env,
1787                       struct obd_export *exp, struct obd_statfs *osfs,
1788                       __u64 max_age, __u32 flags)
1789 {
1790         struct obd_device     *obd = class_exp2obd(exp);
1791         struct ptlrpc_request *req;
1792         struct obd_statfs     *msfs;
1793         struct obd_import     *imp = NULL;
1794         int                    rc;
1795         ENTRY;
1796
1797         /*
1798          * Since the request might also come from lprocfs, so we need
1799          * sync this with client_disconnect_export Bug15684
1800          */
1801         down_read(&obd->u.cli.cl_sem);
1802         if (obd->u.cli.cl_import)
1803                 imp = class_import_get(obd->u.cli.cl_import);
1804         up_read(&obd->u.cli.cl_sem);
1805         if (!imp)
1806                 RETURN(-ENODEV);
1807
1808         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1809                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1810         if (req == NULL)
1811                 GOTO(output, rc = -ENOMEM);
1812
1813         ptlrpc_request_set_replen(req);
1814
1815         if (flags & OBD_STATFS_NODELAY) {
1816                 /* procfs requests not want stay in wait for avoid deadlock */
1817                 req->rq_no_resend = 1;
1818                 req->rq_no_delay = 1;
1819         }
1820
1821         rc = ptlrpc_queue_wait(req);
1822         if (rc) {
1823                 /* check connection error first */
1824                 if (imp->imp_connect_error)
1825                         rc = imp->imp_connect_error;
1826                 GOTO(out, rc);
1827         }
1828
1829         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1830         if (msfs == NULL)
1831                 GOTO(out, rc = -EPROTO);
1832
1833         *osfs = *msfs;
1834         EXIT;
1835 out:
1836         ptlrpc_req_finished(req);
1837 output:
1838         class_import_put(imp);
1839         return rc;
1840 }
1841
1842 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1843 {
1844         __u32 keylen, vallen;
1845         void *key;
1846         int rc;
1847
1848         if (gf->gf_pathlen > PATH_MAX)
1849                 RETURN(-ENAMETOOLONG);
1850         if (gf->gf_pathlen < 2)
1851                 RETURN(-EOVERFLOW);
1852
1853         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1854         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1855         OBD_ALLOC(key, keylen);
1856         if (key == NULL)
1857                 RETURN(-ENOMEM);
1858         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1859         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1860
1861         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1862                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1863
1864         if (!fid_is_sane(&gf->gf_fid))
1865                 GOTO(out, rc = -EINVAL);
1866
1867         /* Val is struct getinfo_fid2path result plus path */
1868         vallen = sizeof(*gf) + gf->gf_pathlen;
1869
1870         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1871         if (rc != 0 && rc != -EREMOTE)
1872                 GOTO(out, rc);
1873
1874         if (vallen <= sizeof(*gf))
1875                 GOTO(out, rc = -EPROTO);
1876         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1877                 GOTO(out, rc = -EOVERFLOW);
1878
1879         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1880                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1881
1882 out:
1883         OBD_FREE(key, keylen);
1884         return rc;
1885 }
1886
1887 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1888                                 struct hsm_progress_kernel *hpk)
1889 {
1890         struct obd_import               *imp = class_exp2cliimp(exp);
1891         struct hsm_progress_kernel      *req_hpk;
1892         struct ptlrpc_request           *req;
1893         int                              rc;
1894         ENTRY;
1895
1896         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1897                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1898         if (req == NULL)
1899                 GOTO(out, rc = -ENOMEM);
1900
1901         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1902
1903         /* Copy hsm_progress struct */
1904         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1905         if (req_hpk == NULL)
1906                 GOTO(out, rc = -EPROTO);
1907
1908         *req_hpk = *hpk;
1909         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1910
1911         ptlrpc_request_set_replen(req);
1912
1913         rc = mdc_queue_wait(req);
1914         GOTO(out, rc);
1915 out:
1916         ptlrpc_req_finished(req);
1917         return rc;
1918 }
1919
1920 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1921 {
1922         __u32                   *archive_mask;
1923         struct ptlrpc_request   *req;
1924         int                      rc;
1925         ENTRY;
1926
1927         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1928                                         LUSTRE_MDS_VERSION,
1929                                         MDS_HSM_CT_REGISTER);
1930         if (req == NULL)
1931                 GOTO(out, rc = -ENOMEM);
1932
1933         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1934
1935         /* Copy hsm_progress struct */
1936         archive_mask = req_capsule_client_get(&req->rq_pill,
1937                                               &RMF_MDS_HSM_ARCHIVE);
1938         if (archive_mask == NULL)
1939                 GOTO(out, rc = -EPROTO);
1940
1941         *archive_mask = archives;
1942
1943         ptlrpc_request_set_replen(req);
1944
1945         rc = mdc_queue_wait(req);
1946         GOTO(out, rc);
1947 out:
1948         ptlrpc_req_finished(req);
1949         return rc;
1950 }
1951
1952 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1953                                       struct md_op_data *op_data)
1954 {
1955         struct hsm_current_action       *hca = op_data->op_data;
1956         struct hsm_current_action       *req_hca;
1957         struct ptlrpc_request           *req;
1958         int                              rc;
1959         ENTRY;
1960
1961         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1962                                    &RQF_MDS_HSM_ACTION);
1963         if (req == NULL)
1964                 RETURN(-ENOMEM);
1965
1966         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1967
1968         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1969         if (rc) {
1970                 ptlrpc_request_free(req);
1971                 RETURN(rc);
1972         }
1973
1974         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1975                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1976
1977         ptlrpc_request_set_replen(req);
1978
1979         rc = mdc_queue_wait(req);
1980         if (rc)
1981                 GOTO(out, rc);
1982
1983         req_hca = req_capsule_server_get(&req->rq_pill,
1984                                          &RMF_MDS_HSM_CURRENT_ACTION);
1985         if (req_hca == NULL)
1986                 GOTO(out, rc = -EPROTO);
1987
1988         *hca = *req_hca;
1989
1990         EXIT;
1991 out:
1992         ptlrpc_req_finished(req);
1993         return rc;
1994 }
1995
1996 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1997 {
1998         struct ptlrpc_request   *req;
1999         int                      rc;
2000         ENTRY;
2001
2002         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
2003                                         LUSTRE_MDS_VERSION,
2004                                         MDS_HSM_CT_UNREGISTER);
2005         if (req == NULL)
2006                 GOTO(out, rc = -ENOMEM);
2007
2008         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2009
2010         ptlrpc_request_set_replen(req);
2011
2012         rc = mdc_queue_wait(req);
2013         GOTO(out, rc);
2014 out:
2015         ptlrpc_req_finished(req);
2016         return rc;
2017 }
2018
2019 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
2020                                  struct md_op_data *op_data)
2021 {
2022         struct hsm_user_state   *hus = op_data->op_data;
2023         struct hsm_user_state   *req_hus;
2024         struct ptlrpc_request   *req;
2025         int                      rc;
2026         ENTRY;
2027
2028         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2029                                    &RQF_MDS_HSM_STATE_GET);
2030         if (req == NULL)
2031                 RETURN(-ENOMEM);
2032
2033         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2034
2035         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
2036         if (rc != 0) {
2037                 ptlrpc_request_free(req);
2038                 RETURN(rc);
2039         }
2040
2041         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2042                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2043
2044         ptlrpc_request_set_replen(req);
2045
2046         rc = mdc_queue_wait(req);
2047         if (rc)
2048                 GOTO(out, rc);
2049
2050         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
2051         if (req_hus == NULL)
2052                 GOTO(out, rc = -EPROTO);
2053
2054         *hus = *req_hus;
2055
2056         EXIT;
2057 out:
2058         ptlrpc_req_finished(req);
2059         return rc;
2060 }
2061
2062 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
2063                                  struct md_op_data *op_data)
2064 {
2065         struct hsm_state_set    *hss = op_data->op_data;
2066         struct hsm_state_set    *req_hss;
2067         struct ptlrpc_request   *req;
2068         int                      rc;
2069         ENTRY;
2070
2071         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2072                                    &RQF_MDS_HSM_STATE_SET);
2073         if (req == NULL)
2074                 RETURN(-ENOMEM);
2075
2076         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2077
2078         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
2079         if (rc) {
2080                 ptlrpc_request_free(req);
2081                 RETURN(rc);
2082         }
2083
2084         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2085                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2086
2087         /* Copy states */
2088         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
2089         if (req_hss == NULL)
2090                 GOTO(out, rc = -EPROTO);
2091         *req_hss = *hss;
2092
2093         ptlrpc_request_set_replen(req);
2094
2095         rc = mdc_queue_wait(req);
2096         GOTO(out, rc);
2097
2098         EXIT;
2099 out:
2100         ptlrpc_req_finished(req);
2101         return rc;
2102 }
2103
2104 static int mdc_ioc_hsm_request(struct obd_export *exp,
2105                                struct hsm_user_request *hur)
2106 {
2107         struct obd_import       *imp = class_exp2cliimp(exp);
2108         struct ptlrpc_request   *req;
2109         struct hsm_request      *req_hr;
2110         struct hsm_user_item    *req_hui;
2111         char                    *req_opaque;
2112         int                      rc;
2113         ENTRY;
2114
2115         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2116         if (req == NULL)
2117                 GOTO(out, rc = -ENOMEM);
2118
2119         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2120                              hur->hur_request.hr_itemcount
2121                              * sizeof(struct hsm_user_item));
2122         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2123                              hur->hur_request.hr_data_len);
2124
2125         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2126         if (rc) {
2127                 ptlrpc_request_free(req);
2128                 RETURN(rc);
2129         }
2130
2131         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2132
2133         /* Copy hsm_request struct */
2134         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2135         if (req_hr == NULL)
2136                 GOTO(out, rc = -EPROTO);
2137         *req_hr = hur->hur_request;
2138
2139         /* Copy hsm_user_item structs */
2140         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2141         if (req_hui == NULL)
2142                 GOTO(out, rc = -EPROTO);
2143         memcpy(req_hui, hur->hur_user_item,
2144                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2145
2146         /* Copy opaque field */
2147         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2148         if (req_opaque == NULL)
2149                 GOTO(out, rc = -EPROTO);
2150         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2151
2152         ptlrpc_request_set_replen(req);
2153
2154         rc = mdc_queue_wait(req);
2155         GOTO(out, rc);
2156
2157 out:
2158         ptlrpc_req_finished(req);
2159         return rc;
2160 }
2161
2162 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2163 {
2164         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2165
2166         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2167
2168         lh->kuc_magic = KUC_MAGIC;
2169         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2170         lh->kuc_flags = flags;
2171         lh->kuc_msgtype = CL_RECORD;
2172         lh->kuc_msglen = len;
2173         return lh;
2174 }
2175
2176 #define D_CHANGELOG 0
2177
2178 struct changelog_show {
2179         __u64           cs_startrec;
2180         __u32           cs_flags;
2181         struct file     *cs_fp;
2182         char            *cs_buf;
2183         struct obd_device *cs_obd;
2184 };
2185
2186 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2187                              struct llog_rec_hdr *hdr, void *data)
2188 {
2189         struct changelog_show *cs = data;
2190         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2191         struct kuc_hdr *lh;
2192         int len, rc;
2193         ENTRY;
2194
2195         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2196                 rc = -EINVAL;
2197                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2198                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2199                        rec->cr.cr_type, rc);
2200                 RETURN(rc);
2201         }
2202
2203         if (rec->cr.cr_index < cs->cs_startrec) {
2204                 /* Skip entries earlier than what we are interested in */
2205                 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2206                        rec->cr.cr_index, cs->cs_startrec);
2207                 RETURN(0);
2208         }
2209
2210         CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2211                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2212                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2213                 rec->cr.cr_flags & CLF_FLAGMASK,
2214                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2215                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2216
2217         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2218
2219         /* Set up the message */
2220         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2221         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2222
2223         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2224         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2225
2226         RETURN(rc);
2227 }
2228
2229 static int mdc_changelog_send_thread(void *csdata)
2230 {
2231         struct changelog_show *cs = csdata;
2232         struct llog_ctxt *ctxt = NULL;
2233         struct llog_handle *llh = NULL;
2234         struct kuc_hdr *kuch;
2235         int rc;
2236
2237         CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2238                cs->cs_fp, cs->cs_startrec);
2239
2240         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2241         if (cs->cs_buf == NULL)
2242                 GOTO(out, rc = -ENOMEM);
2243
2244         /* Set up the remote catalog handle */
2245         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2246         if (ctxt == NULL)
2247                 GOTO(out, rc = -ENOENT);
2248         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2249                        LLOG_OPEN_EXISTS);
2250         if (rc) {
2251                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2252                        cs->cs_obd->obd_name, rc);
2253                 GOTO(out, rc);
2254         }
2255         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2256         if (rc) {
2257                 CERROR("llog_init_handle failed %d\n", rc);
2258                 GOTO(out, rc);
2259         }
2260
2261         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2262
2263         /* Send EOF no matter what our result */
2264         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2265                                       cs->cs_flags))) {
2266                 kuch->kuc_msgtype = CL_EOF;
2267                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2268         }
2269
2270 out:
2271         fput(cs->cs_fp);
2272         if (llh)
2273                 llog_cat_close(NULL, llh);
2274         if (ctxt)
2275                 llog_ctxt_put(ctxt);
2276         if (cs->cs_buf)
2277                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2278         OBD_FREE_PTR(cs);
2279         return rc;
2280 }
2281
2282 static int mdc_ioc_changelog_send(struct obd_device *obd,
2283                                   struct ioc_changelog *icc)
2284 {
2285         struct changelog_show *cs;
2286         struct task_struct *task;
2287         int rc;
2288
2289         /* Freed in mdc_changelog_send_thread */
2290         OBD_ALLOC_PTR(cs);
2291         if (!cs)
2292                 return -ENOMEM;
2293
2294         cs->cs_obd = obd;
2295         cs->cs_startrec = icc->icc_recno;
2296         /* matching fput in mdc_changelog_send_thread */
2297         cs->cs_fp = fget(icc->icc_id);
2298         cs->cs_flags = icc->icc_flags;
2299
2300         /*
2301          * New thread because we should return to user app before
2302          * writing into our pipe
2303          */
2304         task = kthread_run(mdc_changelog_send_thread, cs,
2305                            "mdc_clg_send_thread");
2306         if (IS_ERR(task)) {
2307                 rc = PTR_ERR(task);
2308                 CERROR("%s: cannot start changelog thread: rc = %d\n",
2309                        obd->obd_name, rc);
2310                 OBD_FREE_PTR(cs);
2311         } else {
2312                 rc = 0;
2313                 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
2314                        obd->obd_name);
2315         }
2316
2317         return rc;
2318 }
2319
2320 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2321                                 struct lustre_kernelcomm *lk);
2322
2323 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2324                           struct obd_quotactl *oqctl)
2325 {
2326         struct client_obd       *cli = &exp->exp_obd->u.cli;
2327         struct ptlrpc_request   *req;
2328         struct obd_quotactl     *body;
2329         int                      rc;
2330         ENTRY;
2331
2332         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2333                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2334                                         MDS_QUOTACHECK);
2335         if (req == NULL)
2336                 RETURN(-ENOMEM);
2337
2338         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2339         *body = *oqctl;
2340
2341         ptlrpc_request_set_replen(req);
2342
2343         /* the next poll will find -ENODATA, that means quotacheck is
2344          * going on */
2345         cli->cl_qchk_stat = -ENODATA;
2346         rc = ptlrpc_queue_wait(req);
2347         if (rc)
2348                 cli->cl_qchk_stat = rc;
2349         ptlrpc_req_finished(req);
2350         RETURN(rc);
2351 }
2352
2353 static int mdc_quota_poll_check(struct obd_export *exp,
2354                                 struct if_quotacheck *qchk)
2355 {
2356         struct client_obd *cli = &exp->exp_obd->u.cli;
2357         int rc;
2358         ENTRY;
2359
2360         qchk->obd_uuid = cli->cl_target_uuid;
2361         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2362
2363         rc = cli->cl_qchk_stat;
2364         /* the client is not the previous one */
2365         if (rc == CL_NOT_QUOTACHECKED)
2366                 rc = -EINTR;
2367         RETURN(rc);
2368 }
2369
2370 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2371                         struct obd_quotactl *oqctl)
2372 {
2373         struct ptlrpc_request   *req;
2374         struct obd_quotactl     *oqc;
2375         int                      rc;
2376         ENTRY;
2377
2378         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2379                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2380                                         MDS_QUOTACTL);
2381         if (req == NULL)
2382                 RETURN(-ENOMEM);
2383
2384         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2385         *oqc = *oqctl;
2386
2387         ptlrpc_request_set_replen(req);
2388         ptlrpc_at_set_req_timeout(req);
2389         req->rq_no_resend = 1;
2390
2391         rc = ptlrpc_queue_wait(req);
2392         if (rc)
2393                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2394
2395         if (req->rq_repmsg &&
2396             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2397                 *oqctl = *oqc;
2398         } else if (!rc) {
2399                 CERROR ("Can't unpack obd_quotactl\n");
2400                 rc = -EPROTO;
2401         }
2402         ptlrpc_req_finished(req);
2403
2404         RETURN(rc);
2405 }
2406
2407 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2408                                 struct md_op_data *op_data)
2409 {
2410         struct list_head cancels = LIST_HEAD_INIT(cancels);
2411         struct ptlrpc_request   *req;
2412         int                      rc, count;
2413         struct mdc_swap_layouts *msl, *payload;
2414         ENTRY;
2415
2416         msl = op_data->op_data;
2417
2418         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2419          * first thing it will do is to cancel the 2 layout
2420          * locks hold by this client.
2421          * So the client must cancel its layout locks on the 2 fids
2422          * with the request RPC to avoid extra RPC round trips
2423          */
2424         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2425                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2426                                         MDS_INODELOCK_XATTR);
2427         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2428                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2429                                          MDS_INODELOCK_XATTR);
2430
2431         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2432                                    &RQF_MDS_SWAP_LAYOUTS);
2433         if (req == NULL) {
2434                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2435                 RETURN(-ENOMEM);
2436         }
2437
2438         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2439         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2440
2441         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2442         if (rc) {
2443                 ptlrpc_request_free(req);
2444                 RETURN(rc);
2445         }
2446
2447         mdc_swap_layouts_pack(req, op_data);
2448
2449         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2450         LASSERT(payload);
2451
2452         *payload = *msl;
2453
2454         ptlrpc_request_set_replen(req);
2455
2456         rc = ptlrpc_queue_wait(req);
2457         if (rc)
2458                 GOTO(out, rc);
2459         EXIT;
2460
2461 out:
2462         ptlrpc_req_finished(req);
2463         return rc;
2464 }
2465
2466 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2467                          void *karg, void *uarg)
2468 {
2469         struct obd_device *obd = exp->exp_obd;
2470         struct obd_ioctl_data *data = karg;
2471         struct obd_import *imp = obd->u.cli.cl_import;
2472         int rc;
2473         ENTRY;
2474
2475         if (!try_module_get(THIS_MODULE)) {
2476                 CERROR("Can't get module. Is it alive?");
2477                 return -EINVAL;
2478         }
2479         switch (cmd) {
2480         case OBD_IOC_CHANGELOG_SEND:
2481                 rc = mdc_ioc_changelog_send(obd, karg);
2482                 GOTO(out, rc);
2483         case OBD_IOC_CHANGELOG_CLEAR: {
2484                 struct ioc_changelog *icc = karg;
2485                 struct changelog_setinfo cs =
2486                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2487                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2488                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2489                                         NULL);
2490                 GOTO(out, rc);
2491         }
2492         case OBD_IOC_FID2PATH:
2493                 rc = mdc_ioc_fid2path(exp, karg);
2494                 GOTO(out, rc);
2495         case LL_IOC_HSM_CT_START:
2496                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2497                 /* ignore if it was already registered on this MDS. */
2498                 if (rc == -EEXIST)
2499                         rc = 0;
2500                 GOTO(out, rc);
2501         case LL_IOC_HSM_PROGRESS:
2502                 rc = mdc_ioc_hsm_progress(exp, karg);
2503                 GOTO(out, rc);
2504         case LL_IOC_HSM_STATE_GET:
2505                 rc = mdc_ioc_hsm_state_get(exp, karg);
2506                 GOTO(out, rc);
2507         case LL_IOC_HSM_STATE_SET:
2508                 rc = mdc_ioc_hsm_state_set(exp, karg);
2509                 GOTO(out, rc);
2510         case LL_IOC_HSM_ACTION:
2511                 rc = mdc_ioc_hsm_current_action(exp, karg);
2512                 GOTO(out, rc);
2513         case LL_IOC_HSM_REQUEST:
2514                 rc = mdc_ioc_hsm_request(exp, karg);
2515                 GOTO(out, rc);
2516         case OBD_IOC_CLIENT_RECOVER:
2517                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2518                 if (rc < 0)
2519                         GOTO(out, rc);
2520                 GOTO(out, rc = 0);
2521         case IOC_OSC_SET_ACTIVE:
2522                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2523                 GOTO(out, rc);
2524         case OBD_IOC_POLL_QUOTACHECK:
2525                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2526                 GOTO(out, rc);
2527         case OBD_IOC_PING_TARGET:
2528                 rc = ptlrpc_obd_ping(obd);
2529                 GOTO(out, rc);
2530         /*
2531          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2532          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2533          * there'd be no LMV layer thus we might be called here. Eventually
2534          * this code should be removed.
2535          * bz20731, LU-592.
2536          */
2537         case IOC_OBD_STATFS: {
2538                 struct obd_statfs stat_buf = {0};
2539
2540                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2541                         GOTO(out, rc = -ENODEV);
2542
2543                 /* copy UUID */
2544                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2545                                  min((int)data->ioc_plen2,
2546                                      (int)sizeof(struct obd_uuid))))
2547                         GOTO(out, rc = -EFAULT);
2548
2549                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2550                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2551                                 0);
2552                 if (rc != 0)
2553                         GOTO(out, rc);
2554
2555                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2556                                      min((int) data->ioc_plen1,
2557                                          (int) sizeof(stat_buf))))
2558                         GOTO(out, rc = -EFAULT);
2559
2560                 GOTO(out, rc = 0);
2561         }
2562         case OBD_IOC_QUOTACTL: {
2563                 struct if_quotactl *qctl = karg;
2564                 struct obd_quotactl *oqctl;
2565
2566                 OBD_ALLOC_PTR(oqctl);
2567                 if (oqctl == NULL)
2568                         GOTO(out, rc = -ENOMEM);
2569
2570                 QCTL_COPY(oqctl, qctl);
2571                 rc = obd_quotactl(exp, oqctl);
2572                 if (rc == 0) {
2573                         QCTL_COPY(qctl, oqctl);
2574                         qctl->qc_valid = QC_MDTIDX;
2575                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2576                 }
2577
2578                 OBD_FREE_PTR(oqctl);
2579                 GOTO(out, rc);
2580         }
2581         case LL_IOC_GET_CONNECT_FLAGS:
2582                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2583                                  sizeof(*exp_connect_flags_ptr(exp))))
2584                         GOTO(out, rc = -EFAULT);
2585
2586                 GOTO(out, rc = 0);
2587         case LL_IOC_LOV_SWAP_LAYOUTS:
2588                 rc = mdc_ioc_swap_layouts(exp, karg);
2589                 GOTO(out, rc);
2590         default:
2591                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2592                 GOTO(out, rc = -ENOTTY);
2593         }
2594 out:
2595         module_put(THIS_MODULE);
2596
2597         return rc;
2598 }
2599
2600 int mdc_get_info_rpc(struct obd_export *exp,
2601                      obd_count keylen, void *key,
2602                      int vallen, void *val)
2603 {
2604         struct obd_import      *imp = class_exp2cliimp(exp);
2605         struct ptlrpc_request  *req;
2606         char                   *tmp;
2607         int                     rc = -EINVAL;
2608         ENTRY;
2609
2610         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2611         if (req == NULL)
2612                 RETURN(-ENOMEM);
2613
2614         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2615                              RCL_CLIENT, keylen);
2616         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2617                              RCL_CLIENT, sizeof(__u32));
2618
2619         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2620         if (rc) {
2621                 ptlrpc_request_free(req);
2622                 RETURN(rc);
2623         }
2624
2625         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2626         memcpy(tmp, key, keylen);
2627         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2628         memcpy(tmp, &vallen, sizeof(__u32));
2629
2630         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2631                              RCL_SERVER, vallen);
2632         ptlrpc_request_set_replen(req);
2633
2634         rc = ptlrpc_queue_wait(req);
2635         /* -EREMOTE means the get_info result is partial, and it needs to
2636          * continue on another MDT, see fid2path part in lmv_iocontrol */
2637         if (rc == 0 || rc == -EREMOTE) {
2638                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2639                 memcpy(val, tmp, vallen);
2640                 if (ptlrpc_rep_need_swab(req)) {
2641                         if (KEY_IS(KEY_FID2PATH))
2642                                 lustre_swab_fid2path(val);
2643                 }
2644         }
2645         ptlrpc_req_finished(req);
2646
2647         RETURN(rc);
2648 }
2649
2650 static void lustre_swab_hai(struct hsm_action_item *h)
2651 {
2652         __swab32s(&h->hai_len);
2653         __swab32s(&h->hai_action);
2654         lustre_swab_lu_fid(&h->hai_fid);
2655         lustre_swab_lu_fid(&h->hai_dfid);
2656         __swab64s(&h->hai_cookie);
2657         __swab64s(&h->hai_extent.offset);
2658         __swab64s(&h->hai_extent.length);
2659         __swab64s(&h->hai_gid);
2660 }
2661
2662 static void lustre_swab_hal(struct hsm_action_list *h)
2663 {
2664         struct hsm_action_item  *hai;
2665         int                      i;
2666
2667         __swab32s(&h->hal_version);
2668         __swab32s(&h->hal_count);
2669         __swab32s(&h->hal_archive_id);
2670         __swab64s(&h->hal_flags);
2671         hai = hai_first(h);
2672         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2673                 lustre_swab_hai(hai);
2674 }
2675
2676 static void lustre_swab_kuch(struct kuc_hdr *l)
2677 {
2678         __swab16s(&l->kuc_magic);
2679         /* __u8 l->kuc_transport */
2680         __swab16s(&l->kuc_msgtype);
2681         __swab16s(&l->kuc_msglen);
2682 }
2683
2684 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2685                                 struct lustre_kernelcomm *lk)
2686 {
2687         struct obd_import  *imp = class_exp2cliimp(exp);
2688         __u32               archive = lk->lk_data;
2689         int                 rc = 0;
2690
2691         if (lk->lk_group != KUC_GRP_HSM) {
2692                 CERROR("Bad copytool group %d\n", lk->lk_group);
2693                 return -EINVAL;
2694         }
2695
2696         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2697                lk->lk_uid, lk->lk_group, lk->lk_flags);
2698
2699         if (lk->lk_flags & LK_FLG_STOP) {
2700                 /* Unregister with the coordinator */
2701                 rc = mdc_ioc_hsm_ct_unregister(imp);
2702         } else {
2703                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2704         }
2705
2706         return rc;
2707 }
2708
2709 /**
2710  * Send a message to any listening copytools
2711  * @param val KUC message (kuc_hdr + hsm_action_list)
2712  * @param len total length of message
2713  */
2714 static int mdc_hsm_copytool_send(int len, void *val)
2715 {
2716         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2717         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2718         int                      rc;
2719         ENTRY;
2720
2721         if (len < sizeof(*lh) + sizeof(*hal)) {
2722                 CERROR("Short HSM message %d < %d\n", len,
2723                        (int) (sizeof(*lh) + sizeof(*hal)));
2724                 RETURN(-EPROTO);
2725         }
2726         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2727                 lustre_swab_kuch(lh);
2728                 lustre_swab_hal(hal);
2729         } else if (lh->kuc_magic != KUC_MAGIC) {
2730                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2731                 RETURN(-EPROTO);
2732         }
2733
2734         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2735                "on %s\n",
2736                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2737                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2738
2739         /* Broadcast to HSM listeners */
2740         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2741
2742         RETURN(rc);
2743 }
2744
2745 /**
2746  * callback function passed to kuc for re-registering each HSM copytool
2747  * running on MDC, after MDT shutdown/recovery.
2748  * @param data copytool registration data
2749  * @param cb_arg callback argument (obd_import)
2750  */
2751 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2752 {
2753         struct kkuc_ct_data     *kcd = data;
2754         struct obd_import       *imp = (struct obd_import *)cb_arg;
2755         int                      rc;
2756
2757         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2758                 return -EPROTO;
2759
2760         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2761                 return 0;
2762
2763         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2764                imp->imp_obd->obd_name, kcd->kcd_archive);
2765         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2766
2767         /* ignore error if the copytool is already registered */
2768         return (rc == -EEXIST) ? 0 : rc;
2769 }
2770
2771 /**
2772  * Re-establish all kuc contexts with MDT
2773  * after MDT shutdown/recovery.
2774  */
2775 static int mdc_kuc_reregister(struct obd_import *imp)
2776 {
2777         /* re-register HSM agents */
2778         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2779                                          (void *)imp);
2780 }
2781
2782 int mdc_set_info_async(const struct lu_env *env,
2783                        struct obd_export *exp,
2784                        obd_count keylen, void *key,
2785                        obd_count vallen, void *val,
2786                        struct ptlrpc_request_set *set)
2787 {
2788         struct obd_import       *imp = class_exp2cliimp(exp);
2789         int                      rc;
2790         ENTRY;
2791
2792         if (KEY_IS(KEY_READ_ONLY)) {
2793                 if (vallen != sizeof(int))
2794                         RETURN(-EINVAL);
2795
2796                 spin_lock(&imp->imp_lock);
2797                 if (*((int *)val)) {
2798                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2799                         imp->imp_connect_data.ocd_connect_flags |=
2800                                                         OBD_CONNECT_RDONLY;
2801                 } else {
2802                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2803                         imp->imp_connect_data.ocd_connect_flags &=
2804                                                         ~OBD_CONNECT_RDONLY;
2805                 }
2806                 spin_unlock(&imp->imp_lock);
2807
2808                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2809                                        keylen, key, vallen, val, set);
2810                 RETURN(rc);
2811         }
2812         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2813                 sptlrpc_conf_client_adapt(exp->exp_obd);
2814                 RETURN(0);
2815         }
2816         if (KEY_IS(KEY_FLUSH_CTX)) {
2817                 sptlrpc_import_flush_my_ctx(imp);
2818                 RETURN(0);
2819         }
2820         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2821                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2822                                        keylen, key, vallen, val, set);
2823                 RETURN(rc);
2824         }
2825         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2826                 rc = mdc_hsm_copytool_send(vallen, val);
2827                 RETURN(rc);
2828         }
2829
2830         CERROR("Unknown key %s\n", (char *)key);
2831         RETURN(-EINVAL);
2832 }
2833
2834 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2835                  __u32 keylen, void *key, __u32 *vallen, void *val,
2836                  struct lov_stripe_md *lsm)
2837 {
2838         int rc = -EINVAL;
2839
2840         if (KEY_IS(KEY_MAX_EASIZE)) {
2841                 int mdsize, *max_easize;
2842
2843                 if (*vallen != sizeof(int))
2844                         RETURN(-EINVAL);
2845                 mdsize = *(int *)val;
2846                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2847                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2848                 max_easize = val;
2849                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2850                 RETURN(0);
2851         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2852                 int *default_easize;
2853
2854                 if (*vallen != sizeof(int))
2855                         RETURN(-EINVAL);
2856                 default_easize = val;
2857                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2858                 RETURN(0);
2859         } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2860                 int mdsize, *max_cookiesize;
2861
2862                 if (*vallen != sizeof(int))
2863                         RETURN(-EINVAL);
2864                 mdsize = *(int *)val;
2865                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2866                         exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2867                 max_cookiesize = val;
2868                 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2869                 RETURN(0);
2870         } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2871                 int *default_cookiesize;
2872
2873                 if (*vallen != sizeof(int))
2874                         RETURN(-EINVAL);
2875                 default_cookiesize = val;
2876                 *default_cookiesize =
2877                         exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2878                 RETURN(0);
2879         } else if (KEY_IS(KEY_CONN_DATA)) {
2880                 struct obd_import *imp = class_exp2cliimp(exp);
2881                 struct obd_connect_data *data = val;
2882
2883                 if (*vallen != sizeof(*data))
2884                         RETURN(-EINVAL);
2885
2886                 *data = imp->imp_connect_data;
2887                 RETURN(0);
2888         } else if (KEY_IS(KEY_TGT_COUNT)) {
2889                 *((int *)val) = 1;
2890                 RETURN(0);
2891         }
2892
2893         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2894
2895         RETURN(rc);
2896 }
2897
2898 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2899                    struct obd_capa *oc, struct obd_client_handle *handle,
2900                    int flags)
2901 {
2902         struct ptlrpc_request *req;
2903         struct mdt_body       *body;
2904         int                    rc;
2905         ENTRY;
2906
2907         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2908         if (req == NULL)
2909                 RETURN(-ENOMEM);
2910
2911         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2912
2913         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2914         if (rc) {
2915                 ptlrpc_request_free(req);
2916                 RETURN(rc);
2917         }
2918
2919         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2920
2921         ptlrpc_request_set_replen(req);
2922
2923         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2924         rc = ptlrpc_queue_wait(req);
2925         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2926         if (rc) {
2927                 CERROR("Pin failed: %d\n", rc);
2928                 GOTO(err_out, rc);
2929         }
2930
2931         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2932         if (body == NULL)
2933                 GOTO(err_out, rc = -EPROTO);
2934
2935         handle->och_fh = body->handle;
2936         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2937
2938         handle->och_mod = obd_mod_alloc();
2939         if (handle->och_mod == NULL) {
2940                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2941                 GOTO(err_out, rc = -ENOMEM);
2942         }
2943         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2944
2945         RETURN(0);
2946
2947 err_out:
2948         ptlrpc_req_finished(req);
2949         RETURN(rc);
2950 }
2951
2952 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2953                      int flag)
2954 {
2955         struct ptlrpc_request *req;
2956         struct mdt_body       *body;
2957         int                    rc;
2958         ENTRY;
2959
2960         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2961                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
2962         if (req == NULL)
2963                 RETURN(-ENOMEM);
2964
2965         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2966         body->handle = handle->och_fh;
2967         body->flags = flag;
2968
2969         ptlrpc_request_set_replen(req);
2970
2971         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2972         rc = ptlrpc_queue_wait(req);
2973         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2974
2975         if (rc != 0)
2976                 CERROR("Unpin failed: %d\n", rc);
2977
2978         ptlrpc_req_finished(req);
2979         ptlrpc_req_finished(handle->och_mod->mod_open_req);
2980
2981         obd_mod_put(handle->och_mod);
2982         RETURN(rc);
2983 }
2984
2985 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2986               struct obd_capa *oc, struct ptlrpc_request **request)
2987 {
2988         struct ptlrpc_request *req;
2989         int                    rc;
2990         ENTRY;
2991
2992         *request = NULL;
2993         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2994         if (req == NULL)
2995                 RETURN(-ENOMEM);
2996
2997         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2998
2999         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
3000         if (rc) {
3001                 ptlrpc_request_free(req);
3002                 RETURN(rc);
3003         }
3004
3005         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
3006
3007         ptlrpc_request_set_replen(req);
3008
3009         rc = ptlrpc_queue_wait(req);
3010         if (rc)
3011                 ptlrpc_req_finished(req);
3012         else
3013                 *request = req;
3014         RETURN(rc);
3015 }
3016
3017 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
3018                             enum obd_import_event event)
3019 {
3020         int rc = 0;
3021
3022         LASSERT(imp->imp_obd == obd);
3023
3024         switch (event) {
3025         case IMP_EVENT_DISCON: {
3026 #if 0
3027                 /* XXX Pass event up to OBDs stack. used only for FLD now */
3028                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
3029 #endif
3030                 break;
3031         }
3032         case IMP_EVENT_INACTIVE: {
3033                 struct client_obd *cli = &obd->u.cli;
3034                 /*
3035                  * Flush current sequence to make client obtain new one
3036                  * from server in case of disconnect/reconnect.
3037                  */
3038                 if (cli->cl_seq != NULL)
3039                         seq_client_flush(cli->cl_seq);
3040
3041                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
3042                 break;
3043         }
3044         case IMP_EVENT_INVALIDATE: {
3045                 struct ldlm_namespace *ns = obd->obd_namespace;
3046
3047                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3048
3049                 break;
3050         }
3051         case IMP_EVENT_ACTIVE:
3052                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
3053                 /* redo the kuc registration after reconnecting */
3054                 if (rc == 0)
3055                         rc = mdc_kuc_reregister(imp);
3056                 break;
3057         case IMP_EVENT_OCD:
3058                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
3059                 break;
3060         case IMP_EVENT_DEACTIVATE:
3061         case IMP_EVENT_ACTIVATE:
3062                 break;
3063         default:
3064                 CERROR("Unknown import event %x\n", event);
3065                 LBUG();
3066         }
3067         RETURN(rc);
3068 }
3069
3070 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
3071                   struct lu_fid *fid, struct md_op_data *op_data)
3072 {
3073         struct client_obd *cli = &exp->exp_obd->u.cli;
3074         struct lu_client_seq *seq = cli->cl_seq;
3075         ENTRY;
3076         RETURN(seq_client_alloc_fid(env, seq, fid));
3077 }
3078
3079 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
3080         struct client_obd *cli = &exp->exp_obd->u.cli;
3081         return &cli->cl_target_uuid;
3082 }
3083
3084 /**
3085  * Determine whether the lock can be canceled before replaying it during
3086  * recovery, non zero value will be return if the lock can be canceled,
3087  * or zero returned for not
3088  */
3089 static int mdc_cancel_weight(struct ldlm_lock *lock)
3090 {
3091         if (lock->l_resource->lr_type != LDLM_IBITS)
3092                 RETURN(0);
3093
3094         /* FIXME: if we ever get into a situation where there are too many
3095          * opened files with open locks on a single node, then we really
3096          * should replay these open locks to reget it */
3097         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
3098                 RETURN(0);
3099
3100         RETURN(1);
3101 }
3102
3103 static int mdc_resource_inode_free(struct ldlm_resource *res)
3104 {
3105         if (res->lr_lvb_inode)
3106                 res->lr_lvb_inode = NULL;
3107
3108         return 0;
3109 }
3110
3111 struct ldlm_valblock_ops inode_lvbo = {
3112         .lvbo_free = mdc_resource_inode_free
3113 };
3114
3115 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
3116 {
3117         struct client_obd               *cli = &obd->u.cli;
3118         int                             rc;
3119         ENTRY;
3120
3121         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3122         if (!cli->cl_rpc_lock)
3123                 RETURN(-ENOMEM);
3124         mdc_init_rpc_lock(cli->cl_rpc_lock);
3125
3126         rc = ptlrpcd_addref();
3127         if (rc < 0)
3128                 GOTO(err_rpc_lock, rc);
3129
3130         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3131         if (!cli->cl_close_lock)
3132                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
3133         mdc_init_rpc_lock(cli->cl_close_lock);
3134
3135         rc = client_obd_setup(obd, cfg);
3136         if (rc)
3137                 GOTO(err_close_lock, rc);
3138 #ifdef LPROCFS
3139         obd->obd_vars = lprocfs_mdc_obd_vars;
3140         lprocfs_seq_obd_setup(obd);
3141         lprocfs_alloc_md_stats(obd, 0);
3142 #endif
3143         sptlrpc_lprocfs_cliobd_attach(obd);
3144         ptlrpc_lprocfs_register_obd(obd);
3145
3146         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
3147
3148         obd->obd_namespace->ns_lvbo = &inode_lvbo;
3149
3150         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
3151         if (rc) {
3152                 mdc_cleanup(obd);
3153                 CERROR("failed to setup llogging subsystems\n");
3154         }
3155
3156         RETURN(rc);
3157
3158 err_close_lock:
3159         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3160 err_ptlrpcd_decref:
3161         ptlrpcd_decref();
3162 err_rpc_lock:
3163         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3164         RETURN(rc);
3165 }
3166
3167 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
3168  * us to make MDS RPCs with large enough reply buffers to hold a default
3169  * sized EA and cookie without having to calculate this (via a call into the
3170  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
3171  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
3172  * a large number of stripes is possible.  If a larger reply buffer is
3173  * required it will be reallocated in the ptlrpc layer due to overflow.
3174  */
3175 static int mdc_init_ea_size(struct obd_export *exp, int easize,
3176                             int def_easize, int cookiesize, int def_cookiesize)
3177 {
3178         struct obd_device *obd = exp->exp_obd;
3179         struct client_obd *cli = &obd->u.cli;
3180         ENTRY;
3181
3182         if (cli->cl_max_mds_easize < easize)
3183                 cli->cl_max_mds_easize = easize;
3184
3185         if (cli->cl_default_mds_easize < def_easize)
3186                 cli->cl_default_mds_easize = def_easize;
3187
3188         if (cli->cl_max_mds_cookiesize < cookiesize)
3189                 cli->cl_max_mds_cookiesize = cookiesize;
3190
3191         if (cli->cl_default_mds_cookiesize < def_cookiesize)
3192                 cli->cl_default_mds_cookiesize = def_cookiesize;
3193
3194         RETURN(0);
3195 }
3196
3197 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3198 {
3199         int rc = 0;
3200         ENTRY;
3201
3202         switch (stage) {
3203         case OBD_CLEANUP_EARLY:
3204                 break;
3205         case OBD_CLEANUP_EXPORTS:
3206                 /* Failsafe, ok if racy */
3207                 if (obd->obd_type->typ_refcnt <= 1)
3208                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
3209
3210                 obd_cleanup_client_import(obd);
3211                 ptlrpc_lprocfs_unregister_obd(obd);
3212                 lprocfs_obd_cleanup(obd);
3213                 lprocfs_free_md_stats(obd);
3214
3215                 rc = obd_llog_finish(obd, 0);
3216                 if (rc != 0)
3217                         CERROR("failed to cleanup llogging subsystems\n");
3218                 break;
3219         }
3220         RETURN(rc);
3221 }
3222
3223 static int mdc_cleanup(struct obd_device *obd)
3224 {
3225         struct client_obd *cli = &obd->u.cli;
3226
3227         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3228         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3229
3230         ptlrpcd_decref();
3231
3232         return client_obd_cleanup(obd);
3233 }
3234
3235
3236 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3237                          struct obd_device *tgt, int *index)
3238 {
3239         struct llog_ctxt        *ctxt;
3240         int                      rc;
3241
3242         ENTRY;
3243
3244         LASSERT(olg == &obd->obd_olg);
3245
3246         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3247                         &llog_client_ops);
3248         if (rc)
3249                 RETURN(rc);
3250
3251         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3252         llog_initiator_connect(ctxt);
3253         llog_ctxt_put(ctxt);
3254
3255         RETURN(0);
3256 }
3257
3258 static int mdc_llog_finish(struct obd_device *obd, int count)
3259 {
3260         struct llog_ctxt *ctxt;
3261
3262         ENTRY;
3263
3264         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3265         if (ctxt)
3266                 llog_cleanup(NULL, ctxt);
3267
3268         RETURN(0);
3269 }
3270
3271 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3272 {
3273         struct lustre_cfg *lcfg = buf;
3274         int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3275                                               lcfg, obd);
3276         return (rc > 0 ? 0: rc);
3277 }
3278
3279
3280 /* get remote permission for current user on fid */
3281 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3282                         struct obd_capa *oc, __u32 suppgid,
3283                         struct ptlrpc_request **request)
3284 {
3285         struct ptlrpc_request  *req;
3286         int                    rc;
3287         ENTRY;
3288
3289         LASSERT(client_is_remote(exp));
3290