Whamcloud - gitweb
- in order to prevent flood during renewing capa for deleted inodes:
[fs/lustre-release.git] / lustre / mdc / mdc_request.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Copyright (C) 2001-2004 Cluster File Systems, Inc.
5  *
6  *   This file is part of Lustre, http://www.sf.net/projects/lustre/
7  *
8  *   Lustre is free software; you can redistribute it and/or
9  *   modify it under the terms of version 2 of the GNU General Public
10  *   License as published by the Free Software Foundation.
11  *
12  *   Lustre is distributed in the hope that it will be useful,
13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *   GNU General Public License for more details.
16  *
17  *   You should have received a copy of the GNU General Public License
18  *   along with Lustre; if not, write to the Free Software
19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20  */
21
22 #ifndef EXPORT_SYMTAB
23 # define EXPORT_SYMTAB
24 #endif
25 #define DEBUG_SUBSYSTEM S_MDC
26
27 #ifdef __KERNEL__
28 # include <linux/module.h>
29 # include <linux/pagemap.h>
30 # include <linux/miscdevice.h>
31 # include <linux/init.h>
32 #else
33 # include <liblustre.h>
34 #endif
35
36 #include <linux/obd_class.h>
37 #include <linux/lustre_mds.h>
38 #include <linux/lustre_dlm.h>
39 #include <linux/lustre_sec.h>
40 #include <linux/lprocfs_status.h>
41 #include <linux/lustre_acl.h>
42 #include <linux/lustre_lite.h>
43 #include <linux/lustre_gs.h>
44 #include "mdc_internal.h"
45
46 #define REQUEST_MINOR 244
47
48 static int mdc_cleanup(struct obd_device *obd, int flags);
49
50 extern int mds_queue_req(struct ptlrpc_request *);
51 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
52 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
53 static int send_getstatus(struct obd_import *imp, struct lustre_id *rootid,
54                           int level, int msg_flags)
55 {
56         struct ptlrpc_request *req;
57         struct mds_body *body;
58         int rc, size[2] = {0, sizeof(*body)};
59         ENTRY;
60
61         //size[0] = lustre_secdesc_size();
62
63         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS,
64                               2, size, NULL);
65         if (!req)
66                 GOTO(out, rc = -ENOMEM);
67
68         //lustre_pack_secdesc(req, size[0]);
69
70         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
71         req->rq_send_state = level;
72         req->rq_replen = lustre_msg_size(1, &size[1]);
73
74         req->rq_reqmsg->flags |= msg_flags;
75         rc = ptlrpc_queue_wait(req);
76
77         if (!rc) {
78                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
79                                            lustre_swab_mds_body);
80                 if (body == NULL) {
81                         CERROR ("Can't extract mds_body\n");
82                         GOTO (out, rc = -EPROTO);
83                 }
84
85                 memcpy(rootid, &body->id1, sizeof(*rootid));
86
87                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
88                        ", last_xid="LPU64"\n", rootid->li_stc.u.e3s.l3s_ino,
89                        req->rq_repmsg->last_committed, req->rq_repmsg->last_xid);
90         }
91
92         EXIT;
93  out:
94         ptlrpc_req_finished(req);
95         return rc;
96 }
97
98 /* This should be mdc_get_info("rootid") */
99 int mdc_getstatus(struct obd_export *exp, struct lustre_id *rootid)
100 {
101         return send_getstatus(class_exp2cliimp(exp), rootid,
102                               LUSTRE_IMP_FULL, 0);
103 }
104
105 int
106 mdc_interpret_getattr(struct ptlrpc_request *req, void *unused, int rc)
107 {
108         struct mds_body *body = NULL;
109         struct obd_capa *ocapa;
110         struct lustre_capa *capa = NULL;
111         unsigned long expiry;
112         ENTRY;
113
114         if (rc == 0)
115                 rc = req->rq_status;
116
117         if (rc) {
118                 DEBUG_REQ(rc == -ENOENT ? D_INFO : D_ERROR , req,
119                           "async getattr failed: rc = %d", rc);
120                 RETURN(rc);
121         }
122
123         body = lustre_swab_repbuf(req, 0, sizeof (*body), lustre_swab_mds_body);
124         if (body == NULL) {
125                 CERROR ("Can't unpack mds_body\n");
126                 RETURN(-EPROTO);
127         }
128
129         if (!(body->valid & OBD_MD_CAPA)) {
130                 CDEBUG(D_INFO, "MDS has disabled capability\n");
131                 RETURN(0);
132         }
133
134         capa = lustre_swab_repbuf(req, 1, sizeof(*capa),
135                                   lustre_swab_lustre_capa);
136         if (capa == NULL && rc != 0) {
137                 CERROR ("Can't unpack lustre_capa\n");
138                 RETURN(-EPROTO);
139         }
140
141         ocapa = capa_renew(capa, CLIENT_CAPA);
142         if (!ocapa)
143                 RETURN(-ENOENT);
144
145         spin_lock(&capa_lock);
146         expiry = expiry_to_jiffies(capa->lc_expiry - capa_pre_expiry(capa));
147         CDEBUG(D_INFO, "expiry %lu vs timer %lu, base %p\n",
148                expiry, ll_capa_timer.expires, ll_capa_timer.base);
149         if (time_before(expiry, ll_capa_timer.expires) ||
150             !timer_pending(&ll_capa_timer)) {
151                 mod_timer(&ll_capa_timer, expiry);
152                 CDEBUG(D_INFO, "ll_capa_timer new expiry: %lu\n", expiry);
153         }
154         spin_unlock(&capa_lock);
155
156         RETURN(rc);
157 }
158
159 int mdc_getattr_async(struct obd_export *exp, struct ptlrpc_request *req)
160 {
161         int repsize[2] = {sizeof(struct mds_body), sizeof(struct lustre_capa)};
162         ENTRY;
163
164         req->rq_replen = lustre_msg_size(2, repsize);
165         req->rq_interpret_reply = mdc_interpret_getattr;
166         ptlrpcd_add_req(req);
167
168         RETURN (0);
169 }
170
171 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
172                        struct ptlrpc_request *req)
173 {
174         struct mds_body *body, *reqbody;
175         void            *eadata;
176         int              rc;
177         int              repsize[2] = {sizeof(*body)};
178         int              bufcount = 1;
179         ENTRY;
180
181         /* request message already built */
182
183         if (ea_size != 0) {
184                 repsize[bufcount++] = ea_size;
185                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
186                        ea_size);
187         }
188
189         reqbody = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*reqbody));
190         LASSERT(!(reqbody->valid & OBD_MD_FLACL));
191
192         if (reqbody->valid & OBD_MD_FLKEY) {
193                 repsize[bufcount++] = 5;
194                 repsize[bufcount++] = sizeof(struct lustre_key);
195         } else if (reqbody->valid & OBD_MD_CAPA) {
196                 LASSERT(ea_size == 0);
197                 repsize[bufcount++] = sizeof(struct lustre_capa);
198         }
199
200         req->rq_replen = lustre_msg_size(bufcount, repsize);
201
202         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
203         rc = ptlrpc_queue_wait(req);
204         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
205         if (rc != 0)
206                 RETURN (rc);
207
208         body = lustre_swab_repbuf (req, 0, sizeof (*body),
209                                    lustre_swab_mds_body);
210         if (body == NULL) {
211                 CERROR ("Can't unpack mds_body\n");
212                 RETURN (-EPROTO);
213         }
214
215         CDEBUG(D_NET, "mode: %o\n", body->mode);
216
217         LASSERT_REPSWAB (req, 1);
218
219         /* Skip the check if getxattr/listxattr are called with no buffers */
220         if ((reqbody->eadatasize != 0) &&
221             !(reqbody->valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLIST))) {
222                 /* reply indicates presence of eadata; check it's there... */
223                 eadata = lustre_msg_buf (req->rq_repmsg, 1,
224                                          body->eadatasize);
225                 if (eadata == NULL) {
226                         CERROR ("Missing/short eadata\n");
227                         RETURN (-EPROTO);
228                 }
229         }
230
231         RETURN (0);
232 }
233
234 static int mdc_cancel_unused(struct obd_export *exp,
235                              struct lov_stripe_md *lsm, 
236                              int flags, void *opaque)
237 {
238         struct obd_device *obd = class_exp2obd(exp);
239
240         ENTRY;
241         RETURN(ldlm_cli_cancel_unused(obd->obd_namespace,
242                                       NULL, flags, opaque));
243 }
244
245 int mdc_getattr(struct obd_export *exp, struct lustre_id *id,
246                 __u64 valid, const char *xattr_name,
247                 const void *xattr_data, unsigned int xattr_datalen,
248                 unsigned int ea_size, struct obd_capa *ocapa,
249                 struct ptlrpc_request **request)
250 {
251         struct ptlrpc_request *req;
252         struct mds_body *body;
253         int xattr_namelen = xattr_name ? strlen(xattr_name) + 1 : 0;
254         int size[4] = {0, sizeof(*body)};
255         int bufcount = 2;
256         int rc;
257         ENTRY;
258
259         size[0] = lustre_secdesc_size();
260
261         if (valid & OBD_MD_FLXATTR) {
262                 size[bufcount++] = xattr_namelen;
263
264                 if (xattr_datalen > 0) {
265                         LASSERT(xattr_data);
266                         size[bufcount++] = xattr_datalen;
267                 }
268         } else if (valid & OBD_MD_CAPA) {
269                 LASSERT(valid  == OBD_MD_CAPA);
270                 LASSERT(ocapa);
271                 size[bufcount++] = sizeof(*ocapa);
272         } else {
273                 LASSERT(!xattr_data && !xattr_datalen);
274         }
275
276         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
277                               MDS_GETATTR, bufcount, size, NULL);
278         if (!req)
279                 GOTO(out, rc = -ENOMEM);
280
281         lustre_pack_secdesc(req, size[0]);
282
283         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
284         memcpy(&body->id1, id, sizeof(*id));
285         body->valid = valid;
286         body->eadatasize = ea_size;
287
288         if (valid & OBD_MD_FLXATTR) {
289                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, xattr_namelen),
290                        xattr_name, xattr_namelen);
291                 if (xattr_datalen)
292                         memcpy(lustre_msg_buf(req->rq_reqmsg, 3, xattr_datalen),
293                                xattr_data, xattr_datalen);
294         }
295
296         if (valid & OBD_MD_CAPA) {
297                 /* renew capability */
298                 memcpy(&body->handle, &ocapa->c_handle, sizeof(body->handle));
299                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, sizeof(ocapa->c_capa)),
300                        &ocapa->c_capa, sizeof(ocapa->c_capa));
301
302                 rc = mdc_getattr_async(exp, req);
303                 req = NULL;     /* ptlrpcd will finish request */
304         } else {
305                 rc = mdc_getattr_common(exp, ea_size, req);
306                 if (rc != 0) {
307                         ptlrpc_req_finished (req);
308                         req = NULL;
309                 }
310         }
311  out:
312         *request = req;
313         RETURN (rc);
314 }
315
316 int mdc_access_check(struct obd_export *exp, struct lustre_id *id,
317                      struct ptlrpc_request **request)
318
319 {
320         struct ptlrpc_request *req;
321         struct mds_body *body;
322         int size[2] = {0, sizeof(*body)};
323         int rc;
324         ENTRY;
325
326         size[0] = lustre_secdesc_size();
327         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
328                               MDS_ACCESS_CHECK, 2, size, NULL);
329         if (!req)
330                 GOTO(out, rc = -ENOMEM);
331
332         lustre_pack_secdesc(req, size[0]);
333         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
334         memcpy(&body->id1, id, sizeof(*id));
335
336         size[0] = sizeof(*body);
337         size[1] = sizeof(struct mds_remote_perm);
338         req->rq_replen = lustre_msg_size(2, size);
339
340         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
341         rc = ptlrpc_queue_wait(req);
342         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
343         if (rc != 0) {
344                 ptlrpc_req_finished (req);
345                 req = NULL;
346         } else {
347                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
348                                            lustre_swab_mds_body);
349                 if (body == NULL) {
350                         CERROR ("Can't unpack mds_body\n");
351                         RETURN (-EPROTO);
352                 }
353         }
354
355  out:
356         *request = req;
357         RETURN (rc);
358 }
359
360 int mdc_getattr_lock(struct obd_export *exp, struct lustre_id *id,
361                      char *filename, int namelen, __u64 valid,
362                      unsigned int ea_size, struct ptlrpc_request **request)
363 {
364         struct ptlrpc_request *req;
365         struct mds_body *body;
366         int rc, size[3] = {0, sizeof(*body), namelen};
367         ENTRY;
368
369         size[0] = lustre_secdesc_size();
370
371         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
372                               MDS_GETATTR_LOCK, 3, size, NULL);
373         if (!req)
374                 GOTO(out, rc = -ENOMEM);
375
376         lustre_pack_secdesc(req, size[0]);
377
378         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
379         memcpy(&body->id1, id, sizeof(*id));
380         body->valid = valid;
381         body->eadatasize = ea_size;
382
383         if (filename != NULL) {
384                 LASSERT (strnlen (filename, namelen) == namelen - 1);
385                 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, namelen),
386                        filename, namelen);
387         } else {
388                 LASSERT(namelen == 1);
389         }
390
391         rc = mdc_getattr_common(exp, ea_size, req);
392         if (rc != 0) {
393                 ptlrpc_req_finished (req);
394                 req = NULL;
395         }
396  out:
397         *request = req;
398         RETURN(rc);
399 }
400
401 /* This should be called with both the request and the reply still packed. */
402 int mdc_store_inode_generation(struct obd_export *exp,
403                                struct ptlrpc_request *req,
404                                int reqoff, int repoff)
405 {
406         struct mds_rec_create *rec =
407                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
408         struct mds_body *body =
409                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
410
411         LASSERT (rec != NULL);
412         LASSERT (body != NULL);
413
414         memcpy(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid));
415         DEBUG_REQ(D_HA, req, "storing generation for ino "DLID4,
416                   OLID4(&rec->cr_replayid));
417         return 0;
418 }
419
420 static
421 int mdc_unpack_acl(struct obd_export *exp_lmv, struct ptlrpc_request *req, 
422                    unsigned int offset, struct lustre_md *md)
423 {
424         struct posix_acl *acl;
425         struct mds_remote_perm *perm;
426         int    size, rc;
427         void  *buf;
428         ENTRY;
429  
430         if (!(md->body->valid & OBD_MD_FLACL))
431                 RETURN(0);
432
433         if (md->body->valid & OBD_MD_FLRMTACL) {
434                 offset++; /* first 'size' is not used */
435
436                 buf = lustre_swab_repbuf(req, offset, sizeof(*perm),
437                                          lustre_swab_remote_perm);
438                 if (buf == NULL) {
439                         CERROR("Can't unpack remote perm\n");
440                         RETURN(-EFAULT);
441                 }
442
443                 OBD_ALLOC(perm, sizeof(*perm));
444                 if (!perm)
445                         RETURN(-ENOMEM);
446                 memcpy(perm, buf, sizeof(*perm));
447                 md->remote_perm = perm;
448         } else {
449                 size = le32_to_cpu(*(__u32 *) lustre_msg_buf(
450                                    req->rq_repmsg, offset, 4));
451                 buf = lustre_msg_buf(req->rq_repmsg, offset + 1, size);
452
453                 acl = posix_acl_from_xattr(buf, size);
454                 if (IS_ERR(acl)) {
455                         rc = PTR_ERR(acl);
456                         CERROR("convert xattr to acl failed: %d\n", rc);
457                         RETURN(rc);
458                 } else if (acl) {
459                         rc = posix_acl_valid(acl);
460                         if (rc) {
461                                 CERROR("acl valid error: %d\n", rc);
462                                 posix_acl_release(acl);
463                                 RETURN(rc);
464                         }
465                 }
466
467                 md->posix_acl = acl;
468         }
469
470         RETURN(0);
471 }
472
473 static int mdc_unpack_gskey(struct obd_export *exp_lmv, struct ptlrpc_request *req, 
474                             unsigned int *offset, struct lustre_md *md)
475
476         int key_off = 0, rc = 0, size = 0;
477         void *buf;
478         
479         key_off = *offset;
480         if (md->body->valid & OBD_MD_FLKEY) {
481                 size = le32_to_cpu(*(__u32 *) lustre_msg_buf(req->rq_repmsg, 
482                                    key_off++, 4));
483                 buf = lustre_msg_buf(req->rq_repmsg, key_off++, size);
484                 
485                 CDEBUG(D_INFO, "buf %p key_off %d size %d \n", 
486                        buf, key_off, size);
487                 md->key = (struct lustre_key *)buf; 
488                 *offset = key_off; 
489         } else {
490                 *offset += 2;
491         } 
492         RETURN(rc);
493 }
494
495 int mdc_req2lustre_md(struct obd_export *exp_lmv, struct ptlrpc_request *req, 
496                       unsigned int offset, struct obd_export *exp_lov, 
497                       struct lustre_md *md)
498 {
499         struct lov_mds_md *lmm;
500         int rc = 0, reply_off;
501         ENTRY;
502
503         LASSERT(md != NULL);
504         memset(md, 0, sizeof(*md));
505
506         md->body = lustre_msg_buf(req->rq_repmsg, offset,
507                                   sizeof(*md->body));
508         if (!md->body)
509                 RETURN(-ENOMEM);
510
511         LASSERT_REPSWABBED(req, offset);
512
513         if (!(md->body->valid & OBD_MD_FLEASIZE) &&
514             !(md->body->valid & OBD_MD_FLDIREA))
515                 RETURN(0);
516
517         if (S_ISREG(md->body->mode)) {
518                 if (md->body->eadatasize == 0) {
519                         CERROR("invalid EA size (0) is detected\n");
520                         RETURN(-EPROTO);
521                 }
522
523                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
524                                      md->body->eadatasize);
525                 if (!lmm)
526                         RETURN(-EINVAL);
527
528                 LASSERT(exp_lov != NULL);
529                 
530                 rc = obd_unpackmd(exp_lov, &md->lsm, lmm,
531                                   md->body->eadatasize);
532                 if (rc > 0) {
533                         LASSERT(rc >= sizeof(*md->lsm));
534                         rc = 0;
535                 }
536         } else if (S_ISDIR(md->body->mode)) {
537                 /* dir can be non-splitted */
538                 if (md->body->eadatasize == 0)
539                         RETURN(0);
540
541                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
542                                      md->body->eadatasize);
543                 if (!lmm)
544                         RETURN(-EINVAL);
545
546                 if (md->body->valid & OBD_MD_MEA) {
547                         LASSERT(exp_lmv != NULL);
548                 
549                         rc = obd_unpackmd(exp_lmv, (void *)&md->mea,
550                                           lmm, md->body->eadatasize);
551                         if (rc > 0) {
552                                 LASSERT(rc >= sizeof(*md->mea));
553                                 rc = 0;
554                         }
555                 }
556         } else {
557                 LASSERT(S_ISCHR(md->body->mode) ||
558                         S_ISBLK(md->body->mode) ||
559                         S_ISFIFO(md->body->mode)||
560                         S_ISLNK(md->body->mode) ||
561                         S_ISSOCK(md->body->mode));
562         }
563
564         /* if anything wrong when unpacking md, we don't check acl
565          * stuff, for simplicity
566          */
567         if (rc)
568                 RETURN(rc);
569
570         reply_off = (md->body->valid & OBD_MD_FLEASIZE) ?
571                                 (offset + 2) : (offset + 1);
572         rc = mdc_unpack_acl(exp_lmv, req, reply_off, md);
573         if (rc) {
574                 CERROR("upack acl error %d \n", rc);
575                 RETURN(rc);
576         }
577         reply_off += 2;
578         
579         rc = mdc_unpack_gskey(exp_lmv, req, &reply_off, md);
580         if (rc)
581                 RETURN(rc);
582
583         RETURN(rc);
584 }
585
586 static void mdc_commit_open(struct ptlrpc_request *req)
587 {
588         struct mdc_open_data *mod = req->rq_cb_data;
589         if (mod == NULL)
590                 return;
591
592         if (mod->mod_close_req != NULL)
593                 mod->mod_close_req->rq_cb_data = NULL;
594
595         if (mod->mod_och != NULL)
596                 mod->mod_och->och_mod = NULL;
597
598         OBD_FREE(mod, sizeof(*mod));
599         req->rq_cb_data = NULL;
600         LASSERT(atomic_read(&req->rq_refcount) > 1);
601         ptlrpc_req_finished(req);
602 }
603
604 static void mdc_replay_open(struct ptlrpc_request *req)
605 {
606         struct mdc_open_data *mod = req->rq_cb_data;
607         struct obd_client_handle *och;
608         struct ptlrpc_request *close_req;
609         struct lustre_handle old;
610         struct mds_body *body;
611         ENTRY;
612
613         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
614         LASSERT (body != NULL);
615
616         if (mod == NULL) {
617                 DEBUG_REQ(D_ERROR, req,
618                           "can't properly replay without open data");
619                 EXIT;
620                 return;
621         }
622
623         och = mod->mod_och;
624         if (och != NULL) {
625                 struct lustre_handle *file_fh;
626                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
627                 file_fh = &och->och_fh;
628                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
629                        file_fh->cookie, body->handle.cookie);
630                 memcpy(&old, file_fh, sizeof(old));
631                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
632         }
633
634         close_req = mod->mod_close_req;
635         if (close_req != NULL) {
636                 struct mds_body *close_body;
637                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
638                 close_body = lustre_msg_buf(close_req->rq_reqmsg,
639                                             MDS_REQ_REC_OFF,
640                                             sizeof(*close_body));
641                 if (och != NULL)
642                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
643                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
644                 memcpy(&close_body->handle, &body->handle,
645                        sizeof(close_body->handle));
646         }
647
648         EXIT;
649 }
650
651 int mdc_set_open_replay_data(struct obd_export *exp,
652                              struct obd_client_handle *och,
653                              struct ptlrpc_request *open_req)
654 {
655         struct mdc_open_data *mod;
656         struct mds_rec_create *rec;
657         struct mds_body *body;
658
659         rec = lustre_msg_buf(open_req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
660                              sizeof(*rec));
661         body = lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
662
663         LASSERT(rec != NULL);
664         /* outgoing messages always in my byte order */
665         LASSERT(body != NULL);
666         /* incoming message in my byte order (it's been swabbed) */
667         LASSERT_REPSWABBED(open_req, 1);
668
669         OBD_ALLOC(mod, sizeof(*mod));
670         if (mod == NULL) {
671                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
672                 return 0;
673         }
674
675         och->och_mod = mod;
676         mod->mod_och = och;
677         mod->mod_open_req = ptlrpc_request_addref(open_req);
678
679         if (memcmp(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid)))
680                 CDEBUG(D_ERROR, DLID4" != "DLID4"\n", OLID4(&rec->cr_replayid),
681                        OLID4(&body->id1));
682         LASSERT(id_ino(&rec->cr_replayid) != 0);
683         memcpy(&rec->cr_replayid, &body->id1, sizeof rec->cr_replayid);
684         memcpy(&rec->cr_ioepoch, &body->io_epoch, sizeof rec->cr_ioepoch);
685         open_req->rq_replay_cb = mdc_replay_open;
686         open_req->rq_commit_cb = mdc_commit_open;
687         open_req->rq_cb_data = mod;
688         DEBUG_REQ(D_HA, open_req, "set up replay data");
689         return 0;
690 }
691
692 int mdc_clear_open_replay_data(struct obd_export *exp,
693                                struct obd_client_handle *och)
694 {
695         struct mdc_open_data *mod = och->och_mod;
696
697         /* Don't free the structure now (it happens in mdc_commit_open, after
698          * we're sure we won't need to fix up the close request in the future),
699          * but make sure that replay doesn't poke at the och, which is about to
700          * be freed. */
701         LASSERT(mod != LP_POISON);
702         if (mod != NULL)
703                 mod->mod_och = NULL;
704         och->och_mod = NULL;
705         return 0;
706 }
707
708 static void mdc_commit_close(struct ptlrpc_request *req)
709 {
710         struct mdc_open_data *mod = req->rq_cb_data;
711         struct obd_import *imp = req->rq_import;
712         struct ptlrpc_request *open_req;
713
714         DEBUG_REQ(D_HA, req, "close req committed");
715         if (mod == NULL)
716                 return;
717
718         mod->mod_close_req = NULL;
719         req->rq_cb_data = NULL;
720         req->rq_commit_cb = NULL;
721
722         open_req = mod->mod_open_req;
723         LASSERT(open_req != NULL);
724         LASSERT(open_req != LP_POISON);
725         LASSERT(open_req->rq_type != LI_POISON);
726
727         DEBUG_REQ(D_HA, open_req, "open req balanced");
728         if (open_req->rq_transno == 0) {
729                 DEBUG_REQ(D_ERROR, open_req, "BUG 3892  open");
730                 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
731                 LASSERTF(open_req->rq_transno != 0, "BUG 3892\n");
732         }
733         LASSERT(open_req->rq_import == imp);
734
735         /* We no longer want to preserve this for transno-unconditional
736          * replay. */
737         spin_lock(&open_req->rq_lock);
738         open_req->rq_replay = 0;
739         spin_unlock(&open_req->rq_lock);
740 }
741
742 int mdc_close(struct obd_export *exp, struct mdc_op_data *op_data,
743               struct obd_client_handle *och, struct ptlrpc_request **request)
744 {
745         struct obd_device *obd = class_exp2obd(exp);
746         struct obd_import *imp = class_exp2cliimp(exp);
747         int reqsize[3] = {0, sizeof(struct mds_body),
748                           obd->u.cli.cl_max_mds_cookiesize};
749         int rc, repsize[3] = {sizeof(struct mds_body),
750                               obd->u.cli.cl_max_mds_easize,
751                               obd->u.cli.cl_max_mds_cookiesize};
752         struct ptlrpc_request *req;
753         struct mdc_open_data *mod;
754         ENTRY;
755
756         if (imp->imp_connection == NULL) {
757                 CERROR("request on not connected import %s\n",
758                         imp->imp_obd->obd_name);
759                 RETURN(-EIO);
760         }
761
762         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
763                               MDS_CLOSE, 3, reqsize, NULL);
764         if (req == NULL)
765                 GOTO(out, rc = -ENOMEM);
766         //req->rq_request_portal = MDS_CLOSE_PORTAL;
767
768         /* ensure that this close's handle is fixed up during replay. */
769         LASSERT(och != NULL);
770         mod = och->och_mod;
771         if (likely(mod != NULL)) {
772                 mod->mod_close_req = req;
773                 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
774                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
775         } else {
776                 CDEBUG(D_HA, "couldn't find open req; "
777                        "expecting close error\n");
778         }
779
780         mdc_close_pack(req, 1, op_data, och);
781
782         req->rq_replen = lustre_msg_size(3, repsize);
783         req->rq_commit_cb = mdc_commit_close;
784         LASSERT(req->rq_cb_data == NULL);
785         req->rq_cb_data = mod;
786
787         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
788         rc = ptlrpc_queue_wait(req);
789         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
790
791         if (req->rq_repmsg == NULL) {
792                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
793                        req->rq_status);
794                 if (rc == 0)
795                         rc = req->rq_status ? req->rq_status : -EIO;
796         } else if (rc == 0) {
797                 rc = req->rq_repmsg->status;
798                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
799                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, "
800                                   "err = %d", rc);
801                         if (rc > 0)
802                                 rc = -rc;
803                 } else {
804                         if (mod == NULL)
805                                 CERROR("Unexpected: can't find mdc_open_data, but "
806                                        "close succeeded. Please tell CFS.\n");
807                         if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
808                                                 lustre_swab_mds_body))
809                         {
810                                 CERROR("Error unpacking mds_body\n");
811                                 rc = -EPROTO;
812                         }
813                 }
814         }
815
816         EXIT;
817  out:
818         *request = req;
819         return rc;
820 }
821
822 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
823 {
824         struct ptlrpc_request *req;
825         struct mds_body *body;
826         int rc, size[2] = {0, sizeof(*body)};
827         ENTRY;
828
829         size[0] = lustre_secdesc_size();
830
831         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
832                               MDS_DONE_WRITING, 2, size, NULL);
833         if (req == NULL)
834                 RETURN(-ENOMEM);
835
836         lustre_pack_secdesc(req, size[0]);
837
838         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, 
839                               sizeof(*body));
840         
841         mdc_pack_id(&body->id1, obdo->o_id, 0, obdo->o_mode, 
842                     obdo->o_mds, obdo->o_fid);
843         
844         body->size = obdo->o_size;
845         body->blocks = obdo->o_blocks;
846         body->flags = obdo->o_flags;
847         body->valid = obdo->o_valid;
848
849         req->rq_replen = lustre_msg_size(1, &size[1]);
850
851         rc = ptlrpc_queue_wait(req);
852         ptlrpc_req_finished(req);
853         RETURN(rc);
854 }
855
856 int mdc_readpage(struct obd_export *exp,
857                  struct lustre_id *id,
858                  __u64 offset, struct page *page,
859                  struct ptlrpc_request **request)
860 {
861         struct obd_import *imp = class_exp2cliimp(exp);
862         struct ptlrpc_request *req = NULL;
863         struct ptlrpc_bulk_desc *desc = NULL;
864         struct mds_body *body;
865         int rc, size[2] = {0, sizeof(*body)};
866         ENTRY;
867
868         CDEBUG(D_INODE, "inode: %ld\n", (long)id->li_stc.u.e3s.l3s_ino);
869
870         size[0] = lustre_secdesc_size();
871
872         req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
873                               2, size, NULL);
874         if (req == NULL)
875                 GOTO(out, rc = -ENOMEM);
876         /* XXX FIXME bug 249 */
877         req->rq_request_portal = MDS_READPAGE_PORTAL;
878
879         lustre_pack_secdesc(req, size[0]);
880
881         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
882         if (desc == NULL)
883                 GOTO(out, rc = -ENOMEM);
884         /* NB req now owns desc and will free it when it gets freed */
885
886         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
887         mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, id);
888
889         req->rq_replen = lustre_msg_size(1, &size[1]);
890         rc = ptlrpc_queue_wait(req);
891
892         if (rc == 0) {
893                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
894                                           lustre_swab_mds_body);
895                 if (body == NULL) {
896                         CERROR("Can't unpack mds_body\n");
897                         GOTO(out, rc = -EPROTO);
898                 }
899
900                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
901                         CERROR ("Unexpected # bytes transferred: %d"
902                                 " (%ld expected)\n",
903                                 req->rq_bulk->bd_nob_transferred,
904                                 PAGE_CACHE_SIZE);
905                         GOTO (out, rc = -EPROTO);
906                 }
907         }
908
909         EXIT;
910  out:
911         *request = req;
912         return rc;
913 }
914
915 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
916                          void *karg, void *uarg)
917 {
918         struct obd_device *obd = exp->exp_obd;
919         struct obd_ioctl_data *data = karg;
920         struct obd_import *imp = obd->u.cli.cl_import;
921         struct llog_ctxt *ctxt;
922         int rc;
923         ENTRY;
924
925 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
926         MOD_INC_USE_COUNT;
927 #else
928         if (!try_module_get(THIS_MODULE)) {
929                 CERROR("Can't get module. Is it alive?");
930                 return -EINVAL;
931         }
932 #endif
933         switch (cmd) {
934         case OBD_IOC_CLIENT_RECOVER:
935                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
936                 if (rc < 0)
937                         GOTO(out, rc);
938                 GOTO(out, rc = 0);
939         case IOC_OSC_SET_ACTIVE:
940                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
941                 GOTO(out, rc);
942         case IOC_OSC_CTL_RECOVERY:
943                 rc = ptlrpc_import_control_recovery(imp, data->ioc_offset);
944                 GOTO(out, rc);
945         case OBD_IOC_PARSE: {
946                 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
947                                         LLOG_CONFIG_REPL_CTXT);
948                 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
949                 GOTO(out, rc);
950         }
951 #ifdef __KERNEL__
952         case OBD_IOC_LLOG_INFO:
953         case OBD_IOC_LLOG_PRINT: {
954                 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
955                 rc = llog_ioctl(ctxt, cmd, data);
956
957                 GOTO(out, rc);
958         }
959 #endif
960         default:
961                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
962                 GOTO(out, rc = -ENOTTY);
963         }
964 out:
965 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
966         MOD_DEC_USE_COUNT;
967 #else
968         module_put(THIS_MODULE);
969 #endif
970
971         return rc;
972 }
973
974 int mdc_set_info(struct obd_export *exp, obd_count keylen,
975                  void *key, obd_count vallen, void *val)
976 {
977         int rc = -EINVAL;
978
979         if (keylen == strlen("initial_recov") &&
980             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
981                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
982                 if (vallen != sizeof(int))
983                         RETURN(-EINVAL);
984                 imp->imp_initial_recov = *(int *)val;
985                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
986                        exp->exp_obd->obd_name,
987                        imp->imp_initial_recov);
988                 RETURN(0);
989         } else if ((keylen >= strlen("crypto_type")) && 
990                     strcmp(key, "crypto_type") == 0) {
991                 struct ptlrpc_request *req;
992                 char *bufs[2] = {key, val};
993                 int rc, size[2] = {keylen, vallen};
994
995                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
996                                       OST_SET_INFO, 2, size, bufs);
997                 if (req == NULL)
998                         RETURN(-ENOMEM);
999
1000                 req->rq_replen = lustre_msg_size(0, NULL);
1001                 rc = ptlrpc_queue_wait(req);
1002                 ptlrpc_req_finished(req);
1003                 RETURN(rc);
1004         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1005                 struct obd_import *imp = class_exp2cliimp(exp);
1006                 imp->imp_server_timeout = 1;
1007                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1008                 RETURN(0);
1009         } else if (keylen == strlen("sec") &&
1010                    memcmp(key, "sec", keylen) == 0) {
1011                 struct client_obd *cli = &exp->exp_obd->u.cli;
1012
1013                 cli->cl_sec_flavor = ptlrpcs_name2flavor(val);
1014                 if (cli->cl_sec_flavor == PTLRPCS_FLVR_INVALID) {
1015                         CERROR("unrecognized security type %s\n", (char*) val);
1016                         RETURN(-EINVAL);
1017                 }
1018
1019                 RETURN(0);
1020         } else if (keylen == strlen("sec_flags") &&
1021                    memcmp(key, "sec_flags", keylen) == 0) {
1022                 struct client_obd *cli = &exp->exp_obd->u.cli;
1023
1024                 cli->cl_sec_flags = *((unsigned long *) val);
1025                 RETURN(0);
1026         } else if (keylen == strlen("flush_cred") &&
1027                    memcmp(key, "flush_cred", keylen) == 0) {
1028                 struct client_obd *cli = &exp->exp_obd->u.cli;
1029
1030                 if (cli->cl_import)
1031                         ptlrpcs_import_flush_current_creds(cli->cl_import);
1032                 RETURN(0);
1033         } else if (keylen == strlen("async") && memcmp(key, "async", keylen) == 0) {
1034                 struct client_obd *cl = &exp->exp_obd->u.cli;
1035                 if (vallen != sizeof(int))
1036                         RETURN(-EINVAL);
1037                 cl->cl_async = *(int *)val;
1038                 CDEBUG(D_HA, "%s: set async = %d\n",
1039                        exp->exp_obd->obd_name, cl->cl_async);
1040                 RETURN(0);
1041         } else if (keylen == strlen("setext") && memcmp(key, "setext", keylen) == 0) {
1042                 struct ptlrpc_request *req;
1043                 char *bufs[2] = {key, val};
1044                 int rc, size[2] = {keylen, vallen};
1045
1046                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1047                                       OST_SET_INFO, 2, size, bufs);
1048                 if (req == NULL)
1049                         RETURN(-ENOMEM);
1050
1051                 req->rq_replen = lustre_msg_size(0, NULL);
1052                 rc = ptlrpc_queue_wait(req);
1053                 ptlrpc_req_finished(req);
1054                 RETURN(rc);
1055         } else if (keylen == 5 && strcmp(key, "audit") == 0) {
1056                 struct ptlrpc_request *req;
1057                 char *bufs[2] = {key, val};
1058                 int rc, size[2] = {keylen, vallen};
1059
1060                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1061                                       OST_SET_INFO, 2, size, bufs);
1062                 if (req == NULL)
1063                         RETURN(-ENOMEM);
1064
1065                 req->rq_replen = lustre_msg_size(0, NULL);
1066                 lustre_swab_reqbuf(req, 1, sizeof(struct audit_attr_msg),
1067                                    lustre_swab_audit_attr);
1068                 rc = ptlrpc_queue_wait(req);
1069                 ptlrpc_req_finished(req);
1070
1071                 RETURN(rc);
1072         } else if (keylen == strlen("ids") && memcmp(key, "ids", keylen) == 0) {
1073                 struct ptlrpc_request *req;
1074                 struct lustre_id *ids = (struct lustre_id *)val;
1075                 char *bufs[3] = {key, (char *)ids, (char *)(ids + 1)};
1076                 int rc, size[3] = {keylen, sizeof(struct lustre_id), 
1077                                    sizeof(struct lustre_id)};
1078
1079                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1080                                       OST_SET_INFO, 3, size, bufs);
1081                 if (req == NULL)
1082                         RETURN(-ENOMEM);
1083
1084                 req->rq_replen = lustre_msg_size(0, NULL);
1085                 rc = ptlrpc_queue_wait(req);
1086                 ptlrpc_req_finished(req);
1087                 RETURN(rc);
1088         }
1089         RETURN(rc);
1090 }
1091
1092 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1093                       unsigned long max_age)
1094 {
1095         struct obd_statfs *msfs;
1096         struct ptlrpc_request *req;
1097         int rc, size = sizeof(*msfs);
1098         ENTRY;
1099
1100         /* We could possibly pass max_age in the request (as an absolute
1101          * timestamp or a "seconds.usec ago") so the target can avoid doing
1102          * extra calls into the filesystem if that isn't necessary (e.g.
1103          * during mount that would help a bit).  Having relative timestamps
1104          * is not so great if request processing is slow, while absolute
1105          * timestamps are not ideal because they need time synchronization. */
1106         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1107                               MDS_STATFS, 0, NULL, NULL);
1108         if (!req)
1109                 RETURN(-ENOMEM);
1110
1111         req->rq_replen = lustre_msg_size(1, &size);
1112
1113         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
1114         rc = ptlrpc_queue_wait(req);
1115         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
1116
1117         if (rc) {
1118                 /* this can be LMV fake import, whcih is not connected. */
1119                 if (!req->rq_import->imp_connection)
1120                         memset(osfs, 0, sizeof(*osfs));
1121                 GOTO(out, rc);
1122         }
1123
1124         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),
1125                                   lustre_swab_obd_statfs);
1126         if (msfs == NULL) {
1127                 CERROR("Can't unpack obd_statfs\n");
1128                 GOTO(out, rc = -EPROTO);
1129         }
1130
1131         memcpy(osfs, msfs, sizeof (*msfs));
1132         EXIT;
1133 out:
1134         ptlrpc_req_finished(req);
1135         return rc;
1136 }
1137
1138 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
1139                    struct obd_client_handle *handle, int flag)
1140 {
1141         struct ptlrpc_request *req;
1142         struct mds_body *body;
1143         int rc, size[2] = {0, sizeof(*body)};
1144         ENTRY;
1145
1146         //size[0] = lustre_secdesc_size();
1147
1148         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1149                               MDS_PIN, 2, size, NULL);
1150         if (req == NULL)
1151                 RETURN(-ENOMEM);
1152
1153         //lustre_pack_secdesc(req, size[0]);
1154
1155         body = lustre_msg_buf(req->rq_reqmsg, 
1156                               MDS_REQ_REC_OFF, sizeof(*body));
1157
1158         /* FIXME-UMKA: here should be also mdsnum and fid. */
1159         mdc_pack_id(&body->id1, ino, gen, type, 0, 0);
1160         body->flags = flag;
1161
1162         req->rq_replen = lustre_msg_size(1, &size[1]);
1163
1164         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1165         rc = ptlrpc_queue_wait(req);
1166         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1167         if (rc) {
1168                 CERROR("pin failed: %d\n", rc);
1169                 ptlrpc_req_finished(req);
1170                 RETURN(rc);
1171         }
1172
1173         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
1174         if (body == NULL) {
1175                 ptlrpc_req_finished(req);
1176                 RETURN(rc);
1177         }
1178
1179         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1180         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1181
1182         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1183         if (handle->och_mod == NULL) {
1184                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1185                 RETURN(-ENOMEM);
1186         }
1187         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1188
1189         RETURN(rc);
1190 }
1191
1192 static int mdc_unpin(struct obd_export *exp,
1193                      struct obd_client_handle *handle, int flag)
1194 {
1195         struct ptlrpc_request *req;
1196         struct mds_body *body;
1197         int rc, size[2] = {0, sizeof(*body)};
1198         ENTRY;
1199
1200         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1201                 RETURN(0);
1202
1203         //size[0] = lustre_secdesc_size();
1204
1205         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1206                               MDS_CLOSE, 2, size, NULL);
1207         if (req == NULL)
1208                 RETURN(-ENOMEM);
1209
1210         //lustre_pack_secdesc(req, size[0]);
1211
1212         body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
1213         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1214         body->flags = flag;
1215
1216         req->rq_replen = lustre_msg_size(0, NULL);
1217         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1218         rc = ptlrpc_queue_wait(req);
1219         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1220
1221         if (rc != 0)
1222                 CERROR("unpin failed: %d\n", rc);
1223
1224         ptlrpc_req_finished(req);
1225         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1226         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1227         RETURN(rc);
1228 }
1229
1230 int mdc_sync(struct obd_export *exp, struct lustre_id *id,
1231              struct ptlrpc_request **request)
1232 {
1233         struct ptlrpc_request *req;
1234         struct mds_body *body;
1235         int size[2] = {0, sizeof(*body)};
1236         int rc;
1237         ENTRY;
1238
1239         //size[0] = lustre_secdesc_size();
1240
1241         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1242                               MDS_SYNC, 2, size, NULL);
1243         if (!req)
1244                 RETURN(rc = -ENOMEM);
1245
1246         //lustre_pack_secdesc(req, size[0]);
1247
1248         if (id) {
1249                 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
1250                                       sizeof (*body));
1251                 memcpy(&body->id1, id, sizeof(*id));
1252         }
1253
1254         req->rq_replen = lustre_msg_size(1, &size[1]);
1255
1256         rc = ptlrpc_queue_wait(req);
1257         if (rc || request == NULL)
1258                 ptlrpc_req_finished(req);
1259         else
1260                 *request = req;
1261
1262         RETURN(rc);
1263 }
1264
1265 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1266                             enum obd_import_event event)
1267 {
1268         int rc = 0;
1269
1270         LASSERT(imp->imp_obd == obd);
1271
1272         switch (event) {
1273         case IMP_EVENT_DISCON: {
1274                 break;
1275         }
1276         case IMP_EVENT_INACTIVE: {
1277                 if (obd->obd_observer)
1278                         rc = obd_notify(obd->obd_observer, obd, 0, 0);
1279                 break;
1280         }
1281         case IMP_EVENT_INVALIDATE: {
1282                 struct ldlm_namespace *ns = obd->obd_namespace;
1283
1284                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1285
1286                 break;
1287         }
1288         case IMP_EVENT_ACTIVE: {
1289                 if (obd->obd_observer)
1290                         rc = obd_notify(obd->obd_observer, obd, 1, 0);
1291                 break;
1292         }
1293         default:
1294                 CERROR("Unknown import event %d\n", event);
1295                 LBUG();
1296         }
1297         RETURN(rc);
1298 }
1299
1300 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1301 {
1302         struct lprocfs_static_vars lvars;
1303
1304         lprocfs_init_vars(mdc, &lvars);
1305         return lprocfs_obd_attach(dev, lvars.obd_vars);
1306 }
1307
1308 static int mdc_detach(struct obd_device *dev)
1309 {
1310         return lprocfs_obd_detach(dev);
1311 }
1312
1313 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1314 {
1315         struct client_obd *cli = &obd->u.cli;
1316         int rc;
1317         ENTRY;
1318
1319         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1320         if (!cli->cl_rpc_lock)
1321                 RETURN(-ENOMEM);
1322         mdc_init_rpc_lock(cli->cl_rpc_lock);
1323
1324         ptlrpcd_addref();
1325
1326         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1327         if (!cli->cl_setattr_lock)
1328                 GOTO(err_rpc_lock, rc = -ENOMEM);
1329         mdc_init_rpc_lock(cli->cl_setattr_lock);
1330
1331         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1332         if (!cli->cl_close_lock)
1333                 GOTO(err_setattr_lock, rc = -ENOMEM);
1334         mdc_init_rpc_lock(cli->cl_close_lock);
1335
1336         rc = client_obd_setup(obd, len, buf);
1337         if (rc)
1338                 GOTO(err_close_lock, rc);
1339
1340         rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1341         if (rc) {
1342                 mdc_cleanup(obd, 0);
1343                 CERROR("failed to setup llogging subsystems\n");
1344         }
1345
1346         RETURN(rc);
1347
1348 err_close_lock:
1349         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1350 err_setattr_lock:
1351         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1352 err_rpc_lock:
1353         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1354         ptlrpcd_decref();
1355         RETURN(rc);
1356 }
1357
1358 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1359 {
1360         struct obd_device *obd = exp->exp_obd;
1361         struct client_obd *cli = &obd->u.cli;
1362         ENTRY;
1363
1364         if (cli->cl_max_mds_easize < easize)
1365                 cli->cl_max_mds_easize = easize;
1366         if (cli->cl_max_mds_cookiesize < cookiesize)
1367                 cli->cl_max_mds_cookiesize = cookiesize;
1368         RETURN(0);
1369 }
1370
1371 static int mdc_precleanup(struct obd_device *obd, int flags)
1372 {
1373         int rc = 0;
1374         
1375         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1376         if (rc != 0)
1377                 CERROR("failed to cleanup llogging subsystems\n");
1378
1379         RETURN(rc);
1380 }
1381
1382 static int mdc_cleanup(struct obd_device *obd, int flags)
1383 {
1384         struct client_obd *cli = &obd->u.cli;
1385
1386         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1387         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1388         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1389
1390         ptlrpcd_decref();
1391
1392         return client_obd_cleanup(obd, flags);
1393 }
1394
1395
1396 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs, 
1397                          struct obd_device *tgt, int count,
1398                          struct llog_catid *logid)
1399 {
1400         struct llog_ctxt *ctxt;
1401         int rc;
1402         ENTRY;
1403
1404         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1405                             &llog_client_ops);
1406         if (rc == 0) {
1407                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1408                 ctxt->loc_imp = obd->u.cli.cl_import;
1409         }
1410
1411         RETURN(rc);
1412 }
1413
1414 static int mdc_llog_finish(struct obd_device *obd,
1415                            struct obd_llogs *llogs, int count)
1416 {
1417         int rc;
1418         ENTRY;
1419
1420         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1421         RETURN(rc);
1422 }
1423
1424 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1425                                            struct lustre_id *id)
1426 {
1427        ENTRY;
1428        RETURN(exp->exp_obd);
1429 }
1430
1431 static int mdc_get_info(struct obd_export *exp, __u32 keylen,
1432                         void *key, __u32 *valsize, void *val)
1433 {
1434         struct ptlrpc_request *req;
1435         char *bufs[1] = {key};
1436         int rc = 0;
1437         ENTRY;
1438         
1439         if (!valsize || !val)
1440                 RETURN(-EFAULT);
1441
1442         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
1443                 struct obd_import *imp;
1444                 struct obd_connect_data *data;
1445
1446                 imp = class_exp2cliimp(exp);
1447                 if (!imp) {
1448                         LBUG();
1449                         RETURN(-EINVAL);
1450                 }
1451
1452                 if (imp->imp_state != LUSTRE_IMP_FULL) {
1453                         CERROR("import state not full\n");
1454                         RETURN(-EINVAL);
1455                 }
1456
1457                 data = &imp->imp_connect_data;
1458                 if (data->ocd_connect_flags & OBD_CONNECT_REMOTE) {
1459                         *((int *)val) = 1;
1460                         RETURN(0);
1461                 } else if (data->ocd_connect_flags & OBD_CONNECT_LOCAL) {
1462                         *((int *)val) = 0;
1463                         RETURN(0);
1464                 }
1465                 CERROR("no remote flag set?\n");
1466                 RETURN(-EINVAL);
1467         }
1468
1469         if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
1470             (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
1471             (keylen < strlen("lovdesc") || strcmp(key, "lovdesc") != 0) &&
1472             (keylen < strlen("getext") || strcmp(key, "getext") != 0) &&
1473             (keylen < strlen("rootid") || strcmp(key, "rootid") != 0) &&
1474             (keylen < strlen("auditid") || strcmp(key, "auditid") != 0))
1475                 RETURN(-EPROTO);
1476                 
1477         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1478                               OST_GET_INFO, 1, (int *)&keylen, bufs);
1479         if (req == NULL)
1480                 RETURN(-ENOMEM);
1481
1482         req->rq_replen = lustre_msg_size(1, (int *)valsize);
1483         rc = ptlrpc_queue_wait(req);
1484         if (rc)
1485                 GOTO(out_req, rc);
1486
1487         if ((keylen >= strlen("rootid") && !strcmp(key, "rootid")) ||
1488             (keylen >= strlen("auditid") && !strcmp(key, "auditid"))) {
1489                 struct lustre_id *reply;
1490                 
1491                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1492                                            lustre_swab_lustre_id);
1493                 if (reply == NULL) {
1494                         CERROR("Can't unpack %s\n", (char *)key);
1495                         GOTO(out_req, rc = -EPROTO);
1496                 }
1497
1498                 *(struct lustre_id *)val = *reply;
1499         } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
1500                 struct lov_desc *reply;
1501                 
1502                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1503                                            lustre_swab_lov_desc);
1504                 if (reply == NULL) {
1505                         CERROR("Can't unpack %s\n", (char *)key);
1506                         GOTO(out_req, rc = -EPROTO);
1507                 }
1508
1509                 *(struct lov_desc *)val = *reply;
1510                 RETURN(0);
1511         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
1512                 struct fid_extent *reply;
1513                 
1514                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1515                                            lustre_swab_fid_extent);
1516                 if (reply == NULL) {
1517                         CERROR("Can't unpack %s\n", (char *)key);
1518                         GOTO(out_req, rc = -EPROTO);
1519                 }
1520
1521                 *(struct fid_extent *)val = *reply;
1522                 RETURN(0);
1523         } else {
1524                 __u32 *reply;
1525                 
1526                 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1527                                            lustre_swab_generic_32s);
1528                 if (reply == NULL) {
1529                         CERROR("Can't unpack %s\n", (char *)key);
1530                         GOTO(out_req, rc = -EPROTO);
1531                 }
1532                 *((__u32 *)val) = *reply;
1533         }
1534 out_req:
1535         ptlrpc_req_finished(req);
1536         RETURN(rc);
1537 }
1538
1539 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1540                    void *acl, int acl_size, struct lov_stripe_md **ea,
1541                    struct obd_trans_info *oti)
1542 {
1543         struct ptlrpc_request *request;
1544         struct ost_body *body;
1545         char *acl_buf;
1546         int rc, size[2] = { sizeof(*body), acl_size };
1547         ENTRY;
1548
1549         LASSERT(oa);
1550
1551         request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1552                                   OST_CREATE, 2, size, NULL);
1553         if (!request)
1554                 GOTO(out_req, rc = -ENOMEM);
1555
1556         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1557         memcpy(&body->oa, oa, sizeof(body->oa));
1558
1559         if (acl_size) {
1560                 acl_buf = lustre_msg_buf(request->rq_reqmsg, 1, acl_size);
1561                 memcpy(acl_buf, acl, acl_size);
1562         }
1563
1564         request->rq_replen = lustre_msg_size(1, size);
1565         rc = ptlrpc_queue_wait(request);
1566         if (rc)
1567                 GOTO(out_req, rc);
1568
1569         body = lustre_swab_repbuf(request, 0, sizeof(*body),
1570                                   lustre_swab_ost_body);
1571         if (body == NULL) {
1572                 CERROR ("can't unpack ost_body\n");
1573                 GOTO (out_req, rc = -EPROTO);
1574         }
1575
1576         memcpy(oa, &body->oa, sizeof(*oa));
1577
1578         /* store ino/generation for recovery */
1579         body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1580         body->oa.o_id = oa->o_id;
1581         body->oa.o_generation = oa->o_generation;
1582         body->oa.o_fid = oa->o_fid;
1583         body->oa.o_mds = oa->o_mds;
1584
1585         CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1586         EXIT;
1587 out_req:
1588         ptlrpc_req_finished(request);
1589         return rc;
1590 }
1591
1592 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1593             struct lov_stripe_md *ea, obd_count oa_bufs,
1594             struct brw_page *pgarr, struct obd_trans_info *oti)
1595 {
1596         struct ptlrpc_bulk_desc *desc;
1597         struct niobuf_remote *niobuf;
1598         struct ptlrpc_request *req;
1599         struct obd_ioobj *ioobj;
1600         struct ost_body *body;
1601         int err, opc, i;
1602         int size[3];
1603
1604         opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1605         
1606         size[0] = sizeof(*body);
1607         size[1] = sizeof(*ioobj);
1608         size[2] = oa_bufs * sizeof(*niobuf);
1609
1610         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1611                               3, size, NULL);
1612         LASSERT(req != NULL);
1613
1614         if (opc == OST_WRITE)
1615                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1616                                             OST_BULK_PORTAL);
1617         else
1618                 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1619                                             OST_BULK_PORTAL);
1620         LASSERT(desc != NULL);
1621
1622         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1623         ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1624         niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1625
1626         memcpy(&body->oa, oa, sizeof(*oa));
1627         obdo_to_ioobj(oa, ioobj);
1628         ioobj->ioo_bufcnt = oa_bufs;
1629
1630         for (i = 0; i < oa_bufs; i++, niobuf++) {
1631                 struct brw_page *pg = &pgarr[i];
1632
1633                 LASSERT(pg->count > 0);
1634                 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1635
1636                 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1637                                       pg->count);
1638
1639                 niobuf->offset = pg->disk_offset;
1640                 niobuf->len = pg->count;
1641                 niobuf->flags = pg->flag;
1642         }
1643
1644         /* size[0] still sizeof (*body) */
1645         if (opc == OST_WRITE) {
1646                 /* 1 RC per niobuf */
1647                 size[1] = sizeof(__u32) * oa_bufs;
1648                 req->rq_replen = lustre_msg_size(2, size);
1649         } else {
1650                 /* 1 RC for the whole I/O */
1651                 req->rq_replen = lustre_msg_size(1, size);
1652         }
1653         err = ptlrpc_queue_wait(req);
1654         LASSERT(err == 0);
1655
1656         ptlrpc_req_finished(req);
1657         return 0;
1658 }
1659
1660 static int mdc_valid_attrs(struct obd_export *exp,
1661                            struct lustre_id *id)
1662 {
1663         struct ldlm_res_id res_id = { .name = {0} };
1664         struct obd_device *obd = exp->exp_obd;
1665         struct lustre_handle lockh;
1666         ldlm_policy_data_t policy;
1667         int flags;
1668         ENTRY;
1669
1670         res_id.name[0] = id_fid(id);
1671         res_id.name[1] = id_group(id);
1672         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1673
1674         CDEBUG(D_INFO, "trying to match res "LPU64"\n",
1675                res_id.name[0]);
1676
1677         /* FIXME use LDLM_FL_TEST_LOCK instead */
1678         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1679         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1680                             LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1681                 ldlm_lock_decref(&lockh, LCK_PR);
1682                 RETURN(1);
1683         }
1684
1685         if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1686                             LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1687                 ldlm_lock_decref(&lockh, LCK_PW);
1688                 RETURN(1);
1689         }
1690         RETURN(0);
1691 }
1692
1693 static int mdc_change_cbdata_name(struct obd_export *exp,
1694                                   struct lustre_id *pid,
1695                                   char *name, int len,
1696                                   struct lustre_id *cid,
1697                                   ldlm_iterator_t it, void *data)
1698 {
1699         int rc;
1700         rc = mdc_change_cbdata(exp, cid, it, data);
1701         RETURN(rc);
1702 }
1703
1704 struct obd_ops mdc_obd_ops = {
1705         .o_owner         = THIS_MODULE,
1706         .o_attach        = mdc_attach,
1707         .o_detach        = mdc_detach,
1708         .o_setup         = mdc_setup,
1709         .o_precleanup    = mdc_precleanup,
1710         .o_cleanup       = mdc_cleanup,
1711         .o_add_conn      = client_import_add_conn,
1712         .o_del_conn      = client_import_del_conn,
1713         .o_connect       = client_connect_import,
1714         .o_disconnect    = client_disconnect_export,
1715         .o_iocontrol     = mdc_iocontrol,
1716         .o_packmd        = mdc_packmd,
1717         .o_unpackmd      = mdc_unpackmd,
1718         .o_statfs        = mdc_statfs,
1719         .o_pin           = mdc_pin,
1720         .o_unpin         = mdc_unpin,
1721         .o_import_event  = mdc_import_event,
1722         .o_llog_init     = mdc_llog_init,
1723         .o_llog_finish   = mdc_llog_finish,
1724         .o_create        = mdc_obj_create,
1725         .o_set_info      = mdc_set_info,
1726         .o_get_info      = mdc_get_info,
1727         .o_brw           = mdc_brw,
1728         .o_cancel_unused = mdc_cancel_unused,
1729         .o_init_ea_size  = mdc_init_ea_size,
1730 };
1731
1732 struct md_ops mdc_md_ops = {
1733         .m_getstatus     = mdc_getstatus,
1734         .m_getattr       = mdc_getattr,
1735         .m_close         = mdc_close,
1736         .m_create        = mdc_create,
1737         .m_done_writing  = mdc_done_writing,
1738         .m_enqueue       = mdc_enqueue,
1739         .m_getattr_lock  = mdc_getattr_lock,
1740         .m_intent_lock   = mdc_intent_lock,
1741         .m_link          = mdc_link,
1742         .m_rename        = mdc_rename,
1743         .m_setattr       = mdc_setattr,
1744         .m_sync          = mdc_sync,
1745         .m_readpage      = mdc_readpage,
1746         .m_unlink        = mdc_unlink,
1747         .m_valid_attrs   = mdc_valid_attrs,
1748         .m_req2lustre_md = mdc_req2lustre_md,
1749         .m_set_open_replay_data   = mdc_set_open_replay_data,
1750         .m_clear_open_replay_data = mdc_clear_open_replay_data,
1751         .m_store_inode_generation = mdc_store_inode_generation,
1752         .m_set_lock_data = mdc_set_lock_data,
1753         .m_get_real_obd  = mdc_get_real_obd,
1754         .m_change_cbdata_name = mdc_change_cbdata_name,
1755         .m_change_cbdata = mdc_change_cbdata,
1756         .m_access_check  = mdc_access_check,
1757 };
1758
1759 int __init mdc_init(void)
1760 {
1761         struct lprocfs_static_vars lvars;
1762         
1763         lprocfs_init_vars(mdc, &lvars);
1764         return class_register_type(&mdc_obd_ops, &mdc_md_ops,
1765                                    lvars.module_vars, OBD_MDC_DEVICENAME);
1766 }
1767
1768 #ifdef __KERNEL__
1769 static void /*__exit*/ mdc_exit(void)
1770 {
1771         class_unregister_type(OBD_MDC_DEVICENAME);
1772 }
1773
1774 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1775 MODULE_DESCRIPTION("Lustre Metadata Client");
1776 MODULE_LICENSE("GPL");
1777
1778 EXPORT_SYMBOL(mdc_req2lustre_md);
1779 EXPORT_SYMBOL(mdc_change_cbdata);
1780 EXPORT_SYMBOL(mdc_getstatus);
1781 EXPORT_SYMBOL(mdc_getattr);
1782 EXPORT_SYMBOL(mdc_getattr_lock);
1783 EXPORT_SYMBOL(mdc_create);
1784 EXPORT_SYMBOL(mdc_unlink);
1785 EXPORT_SYMBOL(mdc_rename);
1786 EXPORT_SYMBOL(mdc_link);
1787 EXPORT_SYMBOL(mdc_readpage);
1788 EXPORT_SYMBOL(mdc_setattr);
1789 EXPORT_SYMBOL(mdc_close);
1790 EXPORT_SYMBOL(mdc_done_writing);
1791 EXPORT_SYMBOL(mdc_sync);
1792 EXPORT_SYMBOL(mdc_set_open_replay_data);
1793 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1794 EXPORT_SYMBOL(mdc_store_inode_generation);
1795
1796 module_init(mdc_init);
1797 module_exit(mdc_exit);
1798 #endif