Whamcloud - gitweb
fix smfs_lookup, and add LASSERT to llog_lvfs_create
[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) 2001-2003 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/lprocfs_status.h>
40 #include "mdc_internal.h"
41
42 #define REQUEST_MINOR 244
43
44 static int mdc_cleanup(struct obd_device *obd, int flags);
45
46 extern int mds_queue_req(struct ptlrpc_request *);
47 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
48 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
49 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
50                           int level, int msg_flags)
51 {
52         struct ptlrpc_request *req;
53         struct mds_body *body;
54         int rc, size = sizeof(*body);
55         ENTRY;
56
57         req = ptlrpc_prep_req(imp, MDS_GETSTATUS, 1, &size, NULL);
58         if (!req)
59                 GOTO(out, rc = -ENOMEM);
60
61         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
62         req->rq_send_state = level;
63         req->rq_replen = lustre_msg_size(1, &size);
64
65         mdc_pack_req_body(req);
66         req->rq_reqmsg->flags |= msg_flags;
67         rc = ptlrpc_queue_wait(req);
68
69         if (!rc) {
70                 body = lustre_swab_repbuf (req, 0, sizeof (*body),
71                                            lustre_swab_mds_body);
72                 if (body == NULL) {
73                         CERROR ("Can't extract mds_body\n");
74                         GOTO (out, rc = -EPROTO);
75                 }
76
77                 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
78
79                 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
80                        ", last_xid="LPU64"\n",
81                        rootfid->id, req->rq_repmsg->last_committed,
82                        req->rq_repmsg->last_xid);
83         }
84
85         EXIT;
86  out:
87         ptlrpc_req_finished(req);
88         return rc;
89 }
90
91 /* This should be mdc_get_info("rootfid") */
92 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
93 {
94         return send_getstatus(class_exp2cliimp(exp), rootfid, LUSTRE_IMP_FULL,
95                               0);
96 }
97
98 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size, 
99                        struct ptlrpc_request *req)
100 {
101         struct mds_body *body;
102         void            *eadata;
103         int              rc;
104         int              size[2] = {sizeof(*body), 0};
105         int              bufcount = 1;
106         ENTRY;
107
108         /* request message already built */
109
110         if (ea_size != 0) {
111                 size[bufcount++] = ea_size;
112                 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
113                        ea_size);
114         }
115         req->rq_replen = lustre_msg_size(bufcount, size);
116
117         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
118         rc = ptlrpc_queue_wait(req);
119         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
120         if (rc != 0)
121                 RETURN (rc);
122
123         body = lustre_swab_repbuf (req, 0, sizeof (*body),
124                                    lustre_swab_mds_body);
125         if (body == NULL) {
126                 CERROR ("Can't unpack mds_body\n");
127                 RETURN (-EPROTO);
128         }
129
130         CDEBUG(D_NET, "mode: %o\n", body->mode);
131
132         LASSERT_REPSWAB (req, 1);
133         if (body->eadatasize != 0) {
134                 /* reply indicates presence of eadata; check it's there... */
135                 eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
136                 if (eadata == NULL) {
137                         CERROR ("Missing/short eadata\n");
138                         RETURN (-EPROTO);
139                 }
140         }
141
142         RETURN (0);
143 }
144
145 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
146                 unsigned long valid, unsigned int ea_size,
147                 struct ptlrpc_request **request)
148 {
149         struct ptlrpc_request *req;
150         struct mds_body *body;
151         int size = sizeof(*body);
152         int rc;
153         ENTRY;
154
155         /* XXX do we need to make another request here?  We just did a getattr
156          *     to do the lookup in the first place.
157          */
158         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_GETATTR, 1, &size,
159                               NULL);
160         if (!req)
161                 GOTO(out, rc = -ENOMEM);
162
163         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
164         memcpy(&body->fid1, fid, sizeof(*fid));
165         body->valid = valid;
166         body->eadatasize = ea_size;
167         mdc_pack_req_body(req);
168
169         rc = mdc_getattr_common(exp, ea_size, req);
170         if (rc != 0) {
171                 ptlrpc_req_finished (req);
172                 req = NULL;
173         }
174  out:
175         *request = req;
176         RETURN (rc);
177 }
178
179 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
180                      char *filename, int namelen, unsigned long valid,
181                      unsigned int ea_size, struct ptlrpc_request **request)
182 {
183         struct ptlrpc_request *req;
184         struct mds_body *body;
185         int rc, size[2] = {sizeof(*body), namelen};
186         ENTRY;
187
188         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_GETATTR_NAME, 2,
189                               size, NULL);
190         if (!req)
191                 GOTO(out, rc = -ENOMEM);
192
193         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
194         memcpy(&body->fid1, fid, sizeof(*fid));
195         body->valid = valid;
196         body->eadatasize = ea_size;
197         mdc_pack_req_body(req);
198
199         LASSERT (strnlen (filename, namelen) == namelen - 1);
200         memcpy(lustre_msg_buf(req->rq_reqmsg, 1, namelen), filename, namelen);
201
202         rc = mdc_getattr_common(exp, ea_size, req);
203         if (rc != 0) {
204                 ptlrpc_req_finished (req);
205                 req = NULL;
206         }
207  out:
208         *request = req;
209         RETURN(rc);
210 }
211
212 /* This should be called with both the request and the reply still packed. */
213 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
214                                 int repoff)
215 {
216         struct mds_rec_create *rec =
217                 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
218         struct mds_body *body =
219                 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
220
221         LASSERT (rec != NULL);
222         LASSERT (body != NULL);
223
224         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
225         DEBUG_REQ(D_HA, req, "storing generation %u for ino "LPU64,
226                   rec->cr_replayfid.generation, rec->cr_replayfid.id);
227 }
228
229 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
230                       struct obd_export *exp,
231                       struct lustre_md *md)
232 {
233         int rc = 0;
234         ENTRY;
235
236         LASSERT(md);
237         memset(md, 0, sizeof(*md));
238
239         md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
240         LASSERT (md->body != NULL);
241         LASSERT_REPSWABBED (req, offset);
242
243         if (md->body->valid & OBD_MD_FLEASIZE) {
244                 int lmmsize;
245                 struct lov_mds_md *lmm;
246
247                 LASSERT(S_ISREG(md->body->mode));
248
249                 if (md->body->eadatasize == 0) {
250                         CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
251                         RETURN(-EPROTO);
252                 }
253                 lmmsize = md->body->eadatasize;
254                 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1, lmmsize);
255                 LASSERT (lmm != NULL);
256                 LASSERT_REPSWABBED (req, offset + 1);
257
258                 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
259                 if (rc >= 0) {
260                         LASSERT (rc >= sizeof (*md->lsm));
261                         rc = 0;
262                 }
263         }
264         RETURN(rc);
265 }
266
267 static void mdc_commit_open(struct ptlrpc_request *req)
268 {
269         struct mdc_open_data *mod = req->rq_cb_data;
270         if (mod == NULL)
271                 return;
272
273         if (mod->mod_close_req != NULL)
274                 mod->mod_close_req->rq_cb_data = NULL;
275
276         if (mod->mod_och != NULL)
277                 mod->mod_och->och_mod = NULL;
278
279         OBD_FREE(mod, sizeof(*mod));
280         req->rq_cb_data = NULL;
281 }
282
283 static void mdc_replay_open(struct ptlrpc_request *req)
284 {
285         struct mdc_open_data *mod = req->rq_cb_data;
286         struct obd_client_handle *och;
287         struct ptlrpc_request *close_req;
288         struct lustre_handle old; 
289         struct mds_body *body;
290         ENTRY;
291
292         body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
293         LASSERT (body != NULL);
294
295         if (mod == NULL) {
296                 DEBUG_REQ(D_ERROR, req,
297                           "can't properly replay without open data");
298                 EXIT;
299                 return;
300         }
301
302         och = mod->mod_och;
303         if (och != NULL) {
304                 struct lustre_handle *file_fh; 
305                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
306                 file_fh = &och->och_fh;
307                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
308                        file_fh->cookie, body->handle.cookie);
309                 memcpy(&old, file_fh, sizeof(old));
310                 memcpy(file_fh, &body->handle, sizeof(*file_fh));
311         } 
312
313         close_req = mod->mod_close_req;
314         if (close_req != NULL) {
315                 struct mds_body *close_body;
316                 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
317                 close_body = lustre_msg_buf(close_req->rq_reqmsg, 0,
318                                             sizeof(*close_body));
319                 if (och != NULL)
320                         LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
321                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
322                 memcpy(&close_body->handle, &body->handle,
323                        sizeof(close_body->handle));
324         }
325
326         EXIT;
327 }
328
329 void mdc_set_open_replay_data(struct obd_client_handle *och,
330                               struct ptlrpc_request *open_req)
331 {
332         struct mdc_open_data *mod;
333         struct mds_rec_create *rec =
334                 lustre_msg_buf(open_req->rq_reqmsg, 2, sizeof(*rec));
335         struct mds_body *body =
336                 lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
337
338         LASSERT(rec != NULL);
339         /* outgoing messages always in my byte order */
340         LASSERT(body != NULL);
341         /* incoming message in my byte order (it's been swabbed) */
342         LASSERT_REPSWABBED(open_req, 1);
343
344         OBD_ALLOC(mod, sizeof(*mod));
345         if (mod == NULL) {
346                 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
347                 return;
348         }
349
350         och->och_mod = mod;
351         mod->mod_och = och;
352         mod->mod_open_req = open_req;
353
354         memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
355         open_req->rq_replay_cb = mdc_replay_open;
356         open_req->rq_commit_cb = mdc_commit_open;
357         open_req->rq_cb_data = mod;
358         DEBUG_REQ(D_HA, open_req, "set up replay data");
359 }
360
361 void mdc_clear_open_replay_data(struct obd_client_handle *och)
362 {
363         struct mdc_open_data *mod = och->och_mod;
364
365         /* Don't free the structure now (it happens in mdc_commit_open, after
366          * we're sure we won't need to fix up the close request in the future),
367          * but make sure that replay doesn't poke at the och, which is about to
368          * be freed. */
369         LASSERT(mod != (void *)0x5a5a5a5a);
370         if (mod != NULL)
371                 mod->mod_och = NULL;
372         och->och_mod = NULL;
373 }
374
375 static void mdc_commit_close(struct ptlrpc_request *req)
376 {
377         struct mdc_open_data *mod = req->rq_cb_data;
378         struct ptlrpc_request *open_req;
379         struct obd_import *imp = req->rq_import;
380
381         DEBUG_REQ(D_HA, req, "close req committed");
382         if (mod == NULL)
383                 return;
384
385         mod->mod_close_req = NULL;
386         req->rq_cb_data = NULL;
387         req->rq_commit_cb = NULL;
388
389         open_req = mod->mod_open_req;
390         LASSERT(open_req != NULL);
391         LASSERT(open_req != (void *)0x5a5a5a5a);
392
393         DEBUG_REQ(D_HA, open_req, "open req balanced");
394         LASSERT(open_req->rq_transno != 0);
395         LASSERT(open_req->rq_import == imp);
396
397         /* We no longer want to preserve this for transno-unconditional
398          * replay. */
399         spin_lock(&open_req->rq_lock);
400         open_req->rq_replay = 0;
401         spin_unlock(&open_req->rq_lock);
402 }
403
404 static int mdc_close_interpret(struct ptlrpc_request *req, void *data, int rc)
405 {
406         union ptlrpc_async_args *aa = data;
407         struct mdc_rpc_lock *rpc_lock = aa->pointer_arg[0];
408         struct obd_device *obd = aa->pointer_arg[1];
409
410         if (rpc_lock == NULL) {
411                 CERROR("called with NULL rpc_lock\n");
412         } else {
413                 mdc_put_rpc_lock(rpc_lock, NULL);
414                 LASSERTF(req->rq_async_args.pointer_arg[0] ==
415                          obd->u.cli.cl_rpc_lock, "%p != %p\n",
416                          req->rq_async_args.pointer_arg[0],
417                          obd->u.cli.cl_rpc_lock);
418                 aa->pointer_arg[0] = NULL;
419         }
420         wake_up(&req->rq_reply_waitq);
421         RETURN(rc);
422 }
423
424 /* We can't use ptlrpc_check_reply, because we don't want to wake up for
425  * anything but a reply or an error. */
426 static int mdc_close_check_reply(struct ptlrpc_request *req)
427 {
428         int rc = 0;
429         unsigned long flags;
430
431         spin_lock_irqsave(&req->rq_lock, flags);
432         if (PTLRPC_REQUEST_COMPLETE(req)) {
433                 rc = 1;
434         }
435         spin_unlock_irqrestore (&req->rq_lock, flags);
436         return rc;
437 }
438
439 static int go_back_to_sleep(void *unused)
440 {
441         return 0;
442 }
443
444 int mdc_close(struct obd_export *exp, struct obdo *obdo,
445               struct obd_client_handle *och, struct ptlrpc_request **request)
446 {
447         struct mds_body *body;
448         struct obd_device *obd = class_exp2obd(exp);
449         int reqsize = sizeof(*body);
450         int rc, repsize[3] = {sizeof(*body),
451                               obd->u.cli.cl_max_mds_easize,
452                               obd->u.cli.cl_max_mds_cookiesize};
453         struct ptlrpc_request *req;
454         struct mdc_open_data *mod;
455         struct l_wait_info lwi;
456         ENTRY;
457
458         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &reqsize,
459                               NULL);
460         if (req == NULL)
461                 GOTO(out, rc = -ENOMEM);
462
463         /* Ensure that this close's handle is fixed up during replay. */
464         LASSERT(och != NULL);
465         mod = och->och_mod;
466         if (likely(mod != NULL)) {
467                 mod->mod_close_req = req;
468                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open req %p",
469                           mod->mod_open_req);
470         } else {
471                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
472         }
473
474         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
475         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
476         memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
477         body->size = obdo->o_size;
478         body->blocks = obdo->o_blocks;
479         body->flags = obdo->o_flags;
480         body->valid = obdo->o_valid;
481
482         req->rq_replen = lustre_msg_size(3, repsize);
483         req->rq_commit_cb = mdc_commit_close;
484         LASSERT(req->rq_cb_data == NULL);
485         req->rq_cb_data = mod;
486
487         /* We hand a ref to the rpcd here, so we need another one of our own. */
488         ptlrpc_request_addref(req);
489
490         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
491         req->rq_interpret_reply = mdc_close_interpret;
492         req->rq_async_args.pointer_arg[0] = obd->u.cli.cl_rpc_lock;
493         req->rq_async_args.pointer_arg[1] = obd;
494         ptlrpcd_add_req(req);
495         lwi = LWI_TIMEOUT_INTR(MAX(req->rq_timeout * HZ, 1), go_back_to_sleep,
496                                NULL, NULL);
497         rc = l_wait_event(req->rq_reply_waitq, mdc_close_check_reply(req),
498                           &lwi);
499         if (req->rq_repmsg == NULL) {
500                 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
501                        req->rq_status);
502                 rc = req->rq_status;
503         } else if (rc == 0) {
504                 rc = req->rq_repmsg->status;
505                 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
506                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
507                                   "= %d", rc);
508                         if (rc > 0)
509                                 rc = -rc;
510                 } else if (mod == NULL) {
511                         CERROR("Unexpected: can't find mdc_open_data, but the "
512                                "close succeeded.  Please tell CFS.\n");
513                 }
514                 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
515                                         lustre_swab_mds_body)) {
516                         CERROR("Error unpacking mds_body\n");
517                         rc = -EPROTO;
518                 }
519         }
520         if (req->rq_async_args.pointer_arg[0] != NULL) {
521                 CERROR("returned without dropping rpc_lock: rc %d\n", rc);
522                 mdc_close_interpret(req, &req->rq_async_args, rc);
523                 portals_debug_dumplog();
524         }
525
526         EXIT;
527  out:
528         *request = req;
529         return rc;
530 }
531
532 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
533 {
534         struct ptlrpc_request *req;
535         struct mds_body *body;
536         int rc, size = sizeof(*body);
537         ENTRY;
538
539         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_DONE_WRITING, 1,
540                               &size, NULL);
541         if (req == NULL)
542                 RETURN(-ENOMEM);
543
544         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
545         mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
546         body->size = obdo->o_size;
547         body->blocks = obdo->o_blocks;
548         body->flags = obdo->o_flags;
549         body->valid = obdo->o_valid;
550 //        memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
551
552         req->rq_replen = lustre_msg_size(1, &size);
553
554         rc = ptlrpc_queue_wait(req);
555         ptlrpc_req_finished(req);
556         RETURN(rc);
557 }
558
559 int mdc_readpage(struct obd_export *exp, struct ll_fid *mdc_fid, __u64 offset,
560                  struct page *page, struct ptlrpc_request **request)
561 {
562         struct obd_import *imp = class_exp2cliimp(exp);
563         struct ptlrpc_request *req = NULL;
564         struct ptlrpc_bulk_desc *desc = NULL;
565         struct mds_body *body;
566         int rc, size = sizeof(*body);
567         ENTRY;
568
569         CDEBUG(D_INODE, "inode: %ld\n", (long)mdc_fid->id);
570
571         req = ptlrpc_prep_req(imp, MDS_READPAGE, 1, &size, NULL);
572         if (req == NULL)
573                 GOTO(out, rc = -ENOMEM);
574         /* XXX FIXME bug 249 */
575         req->rq_request_portal = MDS_READPAGE_PORTAL;
576
577         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
578         if (desc == NULL)
579                 GOTO(out, rc = -ENOMEM);
580         /* NB req now owns desc and will free it when it gets freed */
581
582         ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
583
584         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE, mdc_fid);
585
586         req->rq_replen = lustre_msg_size(1, &size);
587         rc = ptlrpc_queue_wait(req);
588
589         if (rc == 0) {
590                 body = lustre_swab_repbuf(req, 0, sizeof (*body),
591                                           lustre_swab_mds_body);
592                 if (body == NULL) {
593                         CERROR("Can't unpack mds_body\n");
594                         GOTO(out, rc = -EPROTO);
595                 }
596
597                 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
598                         CERROR ("Unexpected # bytes transferred: %d"
599                                 " (%ld expected)\n",
600                                 req->rq_bulk->bd_nob_transferred,
601                                 PAGE_CACHE_SIZE);
602                         GOTO (out, rc = -EPROTO);
603                 }
604         }
605
606         EXIT;
607  out:
608         *request = req;
609         return rc;
610 }
611
612 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
613                          void *karg, void *uarg)
614 {
615         struct obd_device *obd = exp->exp_obd;
616         struct obd_ioctl_data *data = karg;
617         struct obd_import *imp = obd->u.cli.cl_import;
618         struct llog_ctxt *ctxt;
619         int rc;
620         ENTRY;
621
622 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
623         MOD_INC_USE_COUNT;
624 #else
625         if (!try_module_get(THIS_MODULE)) {
626                 CERROR("Can't get module. Is it alive?");
627                 return -EINVAL;
628         }
629 #endif
630         switch (cmd) {
631         case OBD_IOC_CLIENT_RECOVER:
632                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
633                 if (rc < 0)
634                         GOTO(out, rc);
635                 GOTO(out, rc = 0);
636         case IOC_OSC_SET_ACTIVE:
637                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
638                 GOTO(out, rc);
639         case OBD_IOC_PARSE: {
640                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
641                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
642                 GOTO(out, rc);
643         }
644 #ifdef __KERNEL__
645         case OBD_IOC_LLOG_INFO:
646         case OBD_IOC_LLOG_PRINT: {
647                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
648                 rc = llog_ioctl(ctxt, cmd, data);
649                 
650                 GOTO(out, rc);
651         }
652 #endif
653         default:
654                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
655                 GOTO(out, rc = -ENOTTY);
656         }
657 out:
658 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
659         MOD_DEC_USE_COUNT;
660 #else
661         module_put(THIS_MODULE);
662 #endif
663
664         return rc;
665 }
666
667 int mdc_set_info(struct obd_export *exp, obd_count keylen,
668                  void *key, obd_count vallen, void *val)
669 {
670         int rc = -EINVAL;
671
672         if (keylen == strlen("initial_recov") &&
673             memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
674                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
675                 if (vallen != sizeof(int))
676                         RETURN(-EINVAL);
677                 imp->imp_initial_recov = *(int *)val;
678                 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
679                        exp->exp_obd->obd_name,
680                        imp->imp_initial_recov);
681                 RETURN(0);
682         }
683         
684         RETURN(rc);
685 }
686
687 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
688                       unsigned long max_age)
689 {
690         struct ptlrpc_request *req;
691         struct obd_statfs *msfs;
692         int rc, size = sizeof(*msfs);
693         ENTRY;
694
695         /* We could possibly pass max_age in the request (as an absolute
696          * timestamp or a "seconds.usec ago") so the target can avoid doing
697          * extra calls into the filesystem if that isn't necessary (e.g.
698          * during mount that would help a bit).  Having relative timestamps
699          * is not so great if request processing is slow, while absolute
700          * timestamps are not ideal because they need time synchronization. */
701         req = ptlrpc_prep_req(obd->u.cli.cl_import, MDS_STATFS, 0, NULL, NULL);
702         if (!req)
703                 RETURN(-ENOMEM);
704
705         req->rq_replen = lustre_msg_size(1, &size);
706
707         mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
708         rc = ptlrpc_queue_wait(req);
709         mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
710
711         if (rc)
712                 GOTO(out, rc);
713
714         msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),lustre_swab_obd_statfs);
715         if (msfs == NULL) {
716                 CERROR("Can't unpack obd_statfs\n");
717                 GOTO(out, rc = -EPROTO);
718         }
719
720         memcpy(osfs, msfs, sizeof (*msfs));
721         EXIT;
722 out:
723         ptlrpc_req_finished(req);
724
725         return rc;
726 }
727
728 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
729                    struct obd_client_handle *handle, int flag)
730 {
731         struct ptlrpc_request *req;
732         struct mds_body *body;
733         int rc, size = sizeof(*body);
734         ENTRY;
735
736         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_PIN, 1, &size, NULL);
737         if (req == NULL)
738                 RETURN(-ENOMEM);
739
740         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
741         mdc_pack_fid(&body->fid1, ino, gen, type);
742         body->flags = flag;
743
744         req->rq_replen = lustre_msg_size(1, &size);
745
746         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
747         rc = ptlrpc_queue_wait(req);
748         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
749         if (rc) {
750                 CERROR("pin failed: %d\n", rc);
751                 ptlrpc_req_finished(req);
752                 RETURN(rc);
753         }
754
755         body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
756         if (body == NULL) {
757                 ptlrpc_req_finished(req);
758                 RETURN(rc);
759         }
760
761         memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
762         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
763
764         OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
765         if (handle->och_mod == NULL) {
766                 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
767                 RETURN(-ENOMEM);
768         }
769         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
770
771         RETURN(rc);
772 }
773
774 static int mdc_unpin(struct obd_export *exp,
775                      struct obd_client_handle *handle, int flag)
776 {
777         struct ptlrpc_request *req;
778         struct mds_body *body;
779         int rc, size = sizeof(*body);
780         ENTRY;
781
782         if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
783                 RETURN(0);
784
785         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_CLOSE, 1, &size, NULL);
786         if (req == NULL)
787                 RETURN(-ENOMEM);
788
789         body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
790         memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
791         body->flags = flag;
792
793         req->rq_replen = lustre_msg_size(0, NULL);
794         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
795         rc = ptlrpc_queue_wait(req);
796         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
797
798         if (rc != 0)
799                 CERROR("unpin failed: %d\n", rc);
800
801         ptlrpc_req_finished(req);
802         ptlrpc_req_finished(handle->och_mod->mod_open_req);
803         OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
804         RETURN(rc);
805 }
806
807 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
808              struct ptlrpc_request **request)
809 {
810         struct ptlrpc_request *req;
811         struct mds_body *body;
812         int size = sizeof(*body);
813         int rc;
814         ENTRY;
815
816         req = ptlrpc_prep_req(class_exp2cliimp(exp), MDS_SYNC, 1,&size,NULL);
817         if (!req)
818                 RETURN(rc = -ENOMEM);
819
820         if (fid) {
821                 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
822                 memcpy(&body->fid1, fid, sizeof(*fid));
823                 mdc_pack_req_body(req);
824         }
825
826         req->rq_replen = lustre_msg_size(1, &size);
827
828         rc = ptlrpc_queue_wait(req);
829         if (rc || request == NULL)
830                 ptlrpc_req_finished(req);
831         else
832                 *request = req;
833
834         RETURN(rc);
835 }
836
837 static int mdc_import_event(struct obd_device *obd,
838                             struct obd_import *imp, 
839                             enum obd_import_event event)
840 {
841         int rc = 0;
842
843         LASSERT(imp->imp_obd == obd);
844
845         switch (event) {
846         case IMP_EVENT_DISCON: {
847                 break;
848         }
849         case IMP_EVENT_INACTIVE: {
850                 if (obd->obd_observer)
851                         rc = obd_notify(obd->obd_observer, obd, 0);
852                 break;
853         }
854         case IMP_EVENT_INVALIDATE: {
855                 struct ldlm_namespace *ns = obd->obd_namespace;
856
857                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
858
859                 break;
860         }
861         case IMP_EVENT_ACTIVE: {
862                 if (obd->obd_observer)
863                         rc = obd_notify(obd->obd_observer, obd, 1);
864                 break;
865         }
866         default:
867                 CERROR("Unknown import event %d\n", event);
868                 LBUG();
869         }
870         RETURN(rc);
871 }
872
873 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
874 {
875         struct client_obd *cli = &obd->u.cli;
876         struct lprocfs_static_vars lvars;
877         int rc;
878         ENTRY;
879
880         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
881         if (!cli->cl_rpc_lock)
882                 RETURN(-ENOMEM);
883         mdc_init_rpc_lock(cli->cl_rpc_lock);
884
885         ptlrpcd_addref();
886
887         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
888         if (!cli->cl_setattr_lock)
889                 GOTO(err_rpc_lock, rc = -ENOMEM);
890         mdc_init_rpc_lock(cli->cl_setattr_lock);
891
892         rc = client_obd_setup(obd, len, buf);
893         if (rc)
894                 GOTO(err_setattr_lock, rc);
895         lprocfs_init_vars(mdc, &lvars);
896         lprocfs_obd_setup(obd, lvars.obd_vars);
897
898         rc = obd_llog_init(obd, obd, 0, NULL);
899         if (rc) {
900                 mdc_cleanup(obd, 0);
901                 CERROR("failed to setup llogging subsystems\n");
902         }
903
904         RETURN(rc);
905
906 err_setattr_lock:
907         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
908 err_rpc_lock:
909         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
910         ptlrpcd_decref();
911         RETURN(rc);
912 }
913
914 int mdc_init_ea_size(struct obd_device *obd, char *lov_name)
915 {
916         struct client_obd *cli = &obd->u.cli;
917         struct obd_device *lov_obd;
918         struct obd_export *exp;
919         struct lustre_handle conn;
920         struct lov_desc desc;
921         int valsize;
922         int rc;
923
924         lov_obd = class_name2obd(lov_name);
925         if (!lov_obd) {
926                 CERROR("MDC cannot locate LOV %s!\n", lov_name);
927                 RETURN(-ENOTCONN);
928         }
929
930         rc = obd_connect(&conn, lov_obd, &obd->obd_uuid);
931         if (rc) {
932                 CERROR("MDC failed connect to LOV %s (%d)\n", lov_name, rc);
933                 RETURN(rc);
934         }
935         exp = class_conn2export(&conn);
936
937         valsize = sizeof(desc);
938         rc = obd_get_info(exp, strlen("lovdesc") + 1, "lovdesc",
939                           &valsize, &desc);
940         if (rc == 0) {
941                 cli->cl_max_mds_easize = obd_size_diskmd(exp, NULL);
942                 cli->cl_max_mds_cookiesize = desc.ld_tgt_count *
943                         sizeof(struct llog_cookie);
944         }
945         obd_disconnect(exp, 0);
946         RETURN(rc);
947 }
948
949 static int mdc_precleanup(struct obd_device *obd, int flags)
950 {
951         int rc = 0;
952
953         rc = obd_llog_finish(obd, 0);
954         if (rc != 0)
955                 CERROR("failed to cleanup llogging subsystems\n");
956
957         RETURN(rc);
958 }
959
960 static int mdc_cleanup(struct obd_device *obd, int flags)
961 {
962         struct client_obd *cli = &obd->u.cli;
963
964         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
965         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
966
967         lprocfs_obd_cleanup(obd);
968         ptlrpcd_decref();
969
970         return client_obd_cleanup(obd, flags);
971 }
972
973
974 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
975                          int count, struct llog_catid *logid)
976 {
977         struct llog_ctxt *ctxt;
978         int rc;
979         ENTRY;
980
981         rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
982                         &llog_client_ops);
983         if (rc == 0) {
984                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
985                 ctxt->loc_imp = obd->u.cli.cl_import;
986         }
987
988         RETURN(rc);
989 }
990
991 static int mdc_llog_finish(struct obd_device *obd, int count)
992 {
993         int rc;
994         ENTRY;
995
996         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
997         RETURN(rc);
998 }
999
1000 struct obd_ops mdc_obd_ops = {
1001         .o_owner        = THIS_MODULE,
1002         .o_setup        = mdc_setup,
1003         .o_precleanup   = mdc_precleanup,
1004         .o_cleanup      = mdc_cleanup,
1005         .o_connect      = client_connect_import,
1006         .o_disconnect   = client_disconnect_export,
1007         .o_iocontrol    = mdc_iocontrol,
1008         .o_set_info     = mdc_set_info,
1009         .o_statfs       = mdc_statfs,
1010         .o_pin          = mdc_pin,
1011         .o_unpin        = mdc_unpin,
1012         .o_import_event = mdc_import_event,
1013         .o_llog_init    = mdc_llog_init,
1014         .o_llog_finish  = mdc_llog_finish,
1015 };
1016
1017 int __init mdc_init(void)
1018 {
1019         struct lprocfs_static_vars lvars;
1020         lprocfs_init_vars(mdc, &lvars);
1021         return class_register_type(&mdc_obd_ops, lvars.module_vars,
1022                                    LUSTRE_MDC_NAME);
1023 }
1024
1025 #ifdef __KERNEL__
1026 static void /*__exit*/ mdc_exit(void)
1027 {
1028         class_unregister_type(LUSTRE_MDC_NAME);
1029 }
1030
1031 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1032 MODULE_DESCRIPTION("Lustre Metadata Client");
1033 MODULE_LICENSE("GPL");
1034
1035 EXPORT_SYMBOL(mdc_req2lustre_md);
1036 EXPORT_SYMBOL(mdc_change_cbdata);
1037 EXPORT_SYMBOL(mdc_getstatus);
1038 EXPORT_SYMBOL(mdc_getattr);
1039 EXPORT_SYMBOL(mdc_getattr_name);
1040 EXPORT_SYMBOL(mdc_create);
1041 EXPORT_SYMBOL(mdc_unlink);
1042 EXPORT_SYMBOL(mdc_rename);
1043 EXPORT_SYMBOL(mdc_link);
1044 EXPORT_SYMBOL(mdc_readpage);
1045 EXPORT_SYMBOL(mdc_setattr);
1046 EXPORT_SYMBOL(mdc_close);
1047 EXPORT_SYMBOL(mdc_done_writing);
1048 EXPORT_SYMBOL(mdc_sync);
1049 EXPORT_SYMBOL(mdc_set_open_replay_data);
1050 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1051 EXPORT_SYMBOL(mdc_store_inode_generation);
1052 EXPORT_SYMBOL(mdc_init_ea_size);
1053
1054 module_init(mdc_init);
1055 module_exit(mdc_exit);
1056 #endif