Whamcloud - gitweb
LU-2193 ofd: check object exists before orphan destroy
[fs/lustre-release.git] / lustre / ofd / ofd_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2012, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/ofd/ofd_obd.c
37  *
38  * Author: Andreas Dilger <adilger@whamcloud.com>
39  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40  * Author: Mike Pershin <tappro@whamcloud.com>
41  */
42
43 #define DEBUG_SUBSYSTEM S_FILTER
44
45 #include <lustre/lustre_idl.h>
46 #include "ofd_internal.h"
47 #include <obd_cksum.h>
48 #include <lustre_quota.h>
49
50 static int ofd_export_stats_init(struct ofd_device *ofd,
51                                  struct obd_export *exp, void *client_nid)
52 {
53         struct obd_device       *obd = ofd_obd(ofd);
54         struct nid_stat         *stats;
55         int                      num_stats;
56         int                      rc, newnid = 0;
57
58         ENTRY;
59
60         if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
61                 /* Self-export gets no proc entry */
62                 RETURN(0);
63
64         rc = lprocfs_exp_setup(exp, client_nid, &newnid);
65         if (rc) {
66                 /* Mask error for already created
67                  * /proc entries */
68                 if (rc == -EALREADY)
69                         rc = 0;
70                 RETURN(rc);
71         }
72
73         if (newnid == 0)
74                 RETURN(0);
75
76         stats = exp->exp_nid_stats;
77         LASSERT(stats != NULL);
78
79         num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
80                      LPROC_OFD_LAST - 1;
81
82         stats->nid_stats = lprocfs_alloc_stats(num_stats,
83                                                LPROCFS_STATS_FLAG_NOPERCPU);
84         if (stats->nid_stats == NULL)
85                 return -ENOMEM;
86
87         lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
88         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
89                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
90         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
91                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
92
93         rc = lprocfs_register_stats(stats->nid_proc, "stats",
94                                     stats->nid_stats);
95         if (rc)
96                 GOTO(clean, rc);
97
98         rc = lprocfs_nid_ldlm_stats_init(stats);
99         if (rc) {
100                 lprocfs_free_stats(&stats->nid_stats);
101                 GOTO(clean, rc);
102         }
103
104         RETURN(0);
105 clean:
106         return rc;
107 }
108
109 static int ofd_parse_connect_data(const struct lu_env *env,
110                                   struct obd_export *exp,
111                                   struct obd_connect_data *data,
112                                   bool new_connection)
113 {
114         struct ofd_device                *ofd = ofd_exp(exp);
115         struct filter_export_data        *fed = &exp->exp_filter_data;
116
117         if (!data)
118                 RETURN(0);
119
120         CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
121                " ocd_version: %x ocd_grant: %d ocd_index: %u"
122                " ocd_group %u\n",
123                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
124                data->ocd_connect_flags, data->ocd_version,
125                data->ocd_grant, data->ocd_index, data->ocd_group);
126
127         if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
128                 CWARN("!!! This export (nid %s) used object group %d "
129                       "earlier; now it's trying to use group %d!  This could "
130                       "be a bug in the MDS. Please report to "
131                       "http://bugs.whamcloud.com/\n",
132                       obd_export_nid2str(exp), fed->fed_group,
133                       data->ocd_group);
134                 RETURN(-EPROTO);
135         }
136         fed->fed_group = data->ocd_group;
137
138         data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
139         exp->exp_connect_data = *data;
140         data->ocd_version = LUSTRE_VERSION_CODE;
141
142         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
143         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
144                 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
145                        exp->exp_obd->obd_name, data->ocd_group);
146         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
147                 RETURN(-EPROTO);
148
149         if (ofd_grant_param_supp(exp)) {
150                 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
151                 /* ocd_{blocksize,inodespace} are log2 values */
152                 data->ocd_blocksize  = ofd->ofd_blockbits;
153                 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
154                 /* ocd_grant_extent is in 1K blocks */
155                 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
156         }
157
158         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
159                 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
160                                                     new_connection);
161
162         if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
163                 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
164                 int                    index = lsd->lsd_osd_index;
165
166                 if (index != data->ocd_index) {
167                         LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
168                                            " %u doesn't match actual OST index"
169                                            " %u in last_rcvd file, bad "
170                                            "configuration?\n",
171                                            obd_export_nid2str(exp), index,
172                                            data->ocd_index);
173                         RETURN(-EBADF);
174                 }
175                 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
176                         /* this will only happen on the first connect */
177                         lsd->lsd_feature_compat |= OBD_COMPAT_OST;
178                         /* sync is not needed here as lut_client_add will
179                          * set exp_need_sync flag */
180                         tgt_server_data_update(env, &ofd->ofd_lut, 0);
181                 }
182         }
183         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
184                 data->ocd_brw_size = 65536;
185         } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
186                 data->ocd_brw_size = min(data->ocd_brw_size,
187                                          (__u32)DT_MAX_BRW_SIZE);
188                 if (data->ocd_brw_size == 0) {
189                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
190                                " ocd_version: %x ocd_grant: %d ocd_index: %u "
191                                "ocd_brw_size is unexpectedly zero, "
192                                "network data corruption?"
193                                "Refusing connection of this client\n",
194                                exp->exp_obd->obd_name,
195                                exp->exp_client_uuid.uuid,
196                                exp, data->ocd_connect_flags, data->ocd_version,
197                                data->ocd_grant, data->ocd_index);
198                         RETURN(-EPROTO);
199                 }
200         }
201
202         if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
203                 __u32 cksum_types = data->ocd_cksum_types;
204
205                 /* The client set in ocd_cksum_types the checksum types it
206                  * supports. We have to mask off the algorithms that we don't
207                  * support */
208                 data->ocd_cksum_types &= cksum_types_supported_server();
209
210                 if (unlikely(data->ocd_cksum_types == 0)) {
211                         CERROR("%s: Connect with checksum support but no "
212                                "ocd_cksum_types is set\n",
213                                exp->exp_obd->obd_name);
214                         RETURN(-EPROTO);
215                 }
216
217                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
218                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
219                        cksum_types, data->ocd_cksum_types);
220         } else {
221                 /* This client does not support OBD_CONNECT_CKSUM
222                  * fall back to CRC32 */
223                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
224                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
225                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
226         }
227
228         if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
229                 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
230
231         if (OCD_HAS_FLAG(data, PINGLESS)) {
232                 if (ptlrpc_pinger_suppress_pings()) {
233                         spin_lock(&exp->exp_obd->obd_dev_lock);
234                         list_del_init(&exp->exp_obd_chain_timed);
235                         spin_unlock(&exp->exp_obd->obd_dev_lock);
236                 } else {
237                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
238                 }
239         }
240
241         RETURN(0);
242 }
243
244 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
245                              struct obd_device *obd, struct obd_uuid *cluuid,
246                              struct obd_connect_data *data, void *localdata)
247 {
248         struct ofd_device       *ofd;
249         int                      rc;
250
251         ENTRY;
252
253         if (exp == NULL || obd == NULL || cluuid == NULL)
254                 RETURN(-EINVAL);
255
256         ofd = ofd_dev(obd->obd_lu_dev);
257
258         rc = lu_env_refill((struct lu_env *)env);
259         if (rc != 0) {
260                 CERROR("Failure to refill session: '%d'\n", rc);
261                 RETURN(rc);
262         }
263
264         ofd_info_init(env, exp);
265         rc = ofd_parse_connect_data(env, exp, data, false);
266         if (rc == 0)
267                 ofd_export_stats_init(ofd, exp, localdata);
268
269         RETURN(rc);
270 }
271
272 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
273                            struct obd_device *obd, struct obd_uuid *cluuid,
274                            struct obd_connect_data *data, void *localdata)
275 {
276         struct obd_export       *exp;
277         struct ofd_device       *ofd;
278         struct lustre_handle     conn = { 0 };
279         int                      rc;
280         ENTRY;
281
282         if (_exp == NULL || obd == NULL || cluuid == NULL)
283                 RETURN(-EINVAL);
284
285         ofd = ofd_dev(obd->obd_lu_dev);
286
287         rc = class_connect(&conn, obd, cluuid);
288         if (rc)
289                 RETURN(rc);
290
291         exp = class_conn2export(&conn);
292         LASSERT(exp != NULL);
293
294         rc = lu_env_refill((struct lu_env *)env);
295         if (rc != 0) {
296                 CERROR("Failure to refill session: '%d'\n", rc);
297                 GOTO(out, rc);
298         }
299
300         ofd_info_init(env, exp);
301
302         rc = ofd_parse_connect_data(env, exp, data, true);
303         if (rc)
304                 GOTO(out, rc);
305
306         if (obd->obd_replayable) {
307                 struct tg_export_data *ted = &exp->exp_target_data;
308
309                 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
310                        sizeof(ted->ted_lcd->lcd_uuid));
311                 rc = tgt_client_new(env, exp);
312                 if (rc != 0)
313                         GOTO(out, rc);
314                 ofd_export_stats_init(ofd, exp, localdata);
315         }
316
317         CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
318                data->ocd_group);
319
320 out:
321         if (rc != 0) {
322                 class_disconnect(exp);
323                 *_exp = NULL;
324         } else {
325                 *_exp = exp;
326         }
327         RETURN(rc);
328 }
329
330 static int ofd_obd_disconnect(struct obd_export *exp)
331 {
332         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
333         struct lu_env            env;
334         int                      rc;
335
336         ENTRY;
337
338         LASSERT(exp);
339         class_export_get(exp);
340
341         if (!(exp->exp_flags & OBD_OPT_FORCE))
342                 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
343
344         rc = server_disconnect_export(exp);
345
346         ofd_grant_discard(exp);
347
348         rc = lu_env_init(&env, LCT_DT_THREAD);
349         if (rc)
350                 RETURN(rc);
351
352         /* Do not erase record for recoverable client. */
353         if (exp->exp_obd->obd_replayable &&
354             (!exp->exp_obd->obd_fail || exp->exp_failed))
355                 tgt_client_del(&env, exp);
356         lu_env_fini(&env);
357
358         class_export_put(exp);
359         RETURN(rc);
360 }
361
362 static int ofd_init_export(struct obd_export *exp)
363 {
364         int rc;
365
366         spin_lock_init(&exp->exp_filter_data.fed_lock);
367         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
368         spin_lock(&exp->exp_lock);
369         exp->exp_connecting = 1;
370         spin_unlock(&exp->exp_lock);
371
372         /* self-export doesn't need client data and ldlm initialization */
373         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
374                                      &exp->exp_client_uuid)))
375                 return 0;
376
377         rc = tgt_client_alloc(exp);
378         if (rc == 0)
379                 ldlm_init_export(exp);
380         if (rc)
381                 CERROR("%s: Can't initialize export: rc %d\n",
382                        exp->exp_obd->obd_name, rc);
383         return rc;
384 }
385
386 static int ofd_destroy_export(struct obd_export *exp)
387 {
388         struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
389
390         if (exp->exp_filter_data.fed_pending)
391                 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
392                        "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
393                        exp, exp->exp_filter_data.fed_pending);
394
395         target_destroy_export(exp);
396
397         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
398                                      &exp->exp_client_uuid)))
399                 return 0;
400
401         ldlm_destroy_export(exp);
402         tgt_client_free(exp);
403
404         ofd_fmd_cleanup(exp);
405
406         /*
407          * discard grants once we're sure no more
408          * interaction with the client is possible
409          */
410         ofd_grant_discard(exp);
411         ofd_fmd_cleanup(exp);
412
413         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
414                 if (ofd->ofd_tot_granted_clients > 0)
415                         ofd->ofd_tot_granted_clients --;
416         }
417
418         if (!(exp->exp_flags & OBD_OPT_FORCE))
419                 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
420
421         LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
422         return 0;
423 }
424
425 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
426 {
427         struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
428
429         CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
430         return ldev->ld_ops->ldo_recovery_complete(env, ldev);
431 }
432
433 int ofd_obd_postrecov(struct obd_device *obd)
434 {
435         struct lu_env            env;
436         struct lu_device        *ldev = obd->obd_lu_dev;
437         int                      rc;
438
439         ENTRY;
440
441         rc = lu_env_init(&env, LCT_DT_THREAD);
442         if (rc)
443                 RETURN(rc);
444         ofd_info_init(&env, obd->obd_self_export);
445
446         rc = ofd_postrecov(&env, ofd_dev(ldev));
447
448         lu_env_fini(&env);
449         RETURN(rc);
450 }
451
452 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
453                                   struct obd_device *obd, int initial)
454 {
455         struct filter_obd       *fo = &obd->u.filter;
456         struct sptlrpc_rule_set  tmp_rset;
457         int                      rc;
458
459         sptlrpc_rule_set_init(&tmp_rset);
460         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
461         if (rc) {
462                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
463                        obd->obd_name, rc);
464                 return rc;
465         }
466
467         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
468
469         write_lock(&fo->fo_sptlrpc_lock);
470         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
471         fo->fo_sptlrpc_rset = tmp_rset;
472         write_unlock(&fo->fo_sptlrpc_lock);
473
474         return 0;
475 }
476
477 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
478 {
479         int rc = 0;
480
481         ENTRY;
482
483         LCONSOLE_WARN("%s: received MDS connection from %s\n",
484                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
485         RETURN(rc);
486 }
487
488 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
489                               __u32 keylen, void *key, __u32 vallen, void *val,
490                               struct ptlrpc_request_set *set)
491 {
492         struct ofd_device       *ofd;
493         int                      rc = 0;
494
495         ENTRY;
496
497         if (exp->exp_obd == NULL) {
498                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
499                 RETURN(-EINVAL);
500         }
501
502         ofd = ofd_exp(exp);
503
504         if (KEY_IS(KEY_CAPA_KEY)) {
505                 rc = ofd_update_capa_key(ofd, val);
506                 if (rc)
507                         CERROR("%s: update capability key failed: rc = %d\n",
508                                exp->exp_obd->obd_name, rc);
509         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
510                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
511         } else if (KEY_IS(KEY_MDS_CONN)) {
512                 rc = ofd_set_mds_conn(exp, val);
513         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
514                 struct ost_body *body = val;
515
516                 ofd_info_init(env, exp);
517                 /** handle grant shrink, similar to a read request */
518                 ofd_grant_prepare_read(env, exp, &body->oa);
519         } else {
520                 CERROR("%s: Unsupported key %s\n",
521                        exp->exp_obd->obd_name, (char*)key);
522                 rc = -EOPNOTSUPP;
523         }
524
525         RETURN(rc);
526 }
527
528 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
529                         __u32 keylen, void *key, __u32 *vallen, void *val,
530                         struct lov_stripe_md *lsm)
531 {
532         struct ofd_device       *ofd;
533         int                     rc = 0;
534
535         ENTRY;
536
537         if (exp->exp_obd == NULL) {
538                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
539                 RETURN(-EINVAL);
540         }
541
542         /* Because ofd_get_info might be called from
543          * handle_request_in as well(see LU-3239), where env might
544          * not be initilaized correctly, and le_ses might be in
545          * an un-initialized state, so only refill le_ctx here */
546         rc = lu_env_refill((struct lu_env *)env);
547         if (rc != 0)
548                 RETURN(rc);
549
550         ofd = ofd_exp(exp);
551
552         if (KEY_IS(KEY_BLOCKSIZE)) {
553                 __u32 *blocksize = val;
554                 if (blocksize) {
555                         if (*vallen < sizeof(*blocksize))
556                                 GOTO(out, rc = -EOVERFLOW);
557                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
558                 }
559                 *vallen = sizeof(*blocksize);
560         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
561                 __u32 *blocksize_bits = val;
562                 if (blocksize_bits) {
563                         if (*vallen < sizeof(*blocksize_bits))
564                                 GOTO(out, rc = -EOVERFLOW);
565                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
566                 }
567                 *vallen = sizeof(*blocksize_bits);
568         } else if (KEY_IS(KEY_LAST_ID)) {
569                 obd_id *last_id = val;
570                 struct ofd_seq *oseq;
571
572                 if (val == NULL) {
573                         *vallen = sizeof(obd_id);
574                         GOTO(out, rc = 0);
575                 }
576                 ofd_info_init(env, exp);
577                 oseq = ofd_seq_load(env, ofd,
578                                     (obd_seq)exp->exp_filter_data.fed_group);
579                 LASSERT(!IS_ERR(oseq));
580                 if (last_id) {
581                         if (*vallen < sizeof(*last_id)) {
582                                 ofd_seq_put(env, oseq);
583                                 GOTO(out, rc = -EOVERFLOW);
584                         }
585                         *last_id = ofd_seq_last_oid(oseq);
586                 }
587                 ofd_seq_put(env, oseq);
588                 *vallen = sizeof(*last_id);
589         } else if (KEY_IS(KEY_FIEMAP)) {
590                 struct ofd_thread_info          *info;
591                 struct ofd_device               *ofd = ofd_exp(exp);
592                 struct ofd_object               *fo;
593                 struct ll_fiemap_info_key       *fm_key = key;
594
595                 if (val == NULL) {
596                         *vallen = fiemap_count_to_size(
597                                                fm_key->fiemap.fm_extent_count);
598                         GOTO(out, rc = 0);
599                 }
600
601                 info = ofd_info_init(env, exp);
602                 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0);
603                 if (rc != 0)
604                         GOTO(out, rc);
605                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
606                        PFID(&info->fti_fid));
607
608                 fo = ofd_object_find(env, ofd, &info->fti_fid);
609                 if (IS_ERR(fo)) {
610                         CERROR("%s: error finding object "DFID"\n",
611                                exp->exp_obd->obd_name, PFID(&info->fti_fid));
612                         rc = PTR_ERR(fo);
613                 } else {
614                         struct ll_user_fiemap *fiemap = val;
615
616                         ofd_read_lock(env, fo);
617                         if (ofd_object_exists(fo)) {
618                                 *fiemap = fm_key->fiemap;
619                                 rc = dt_fiemap_get(env,
620                                                    ofd_object_child(fo),
621                                                    fiemap);
622                         } else {
623                                 rc = -ENOENT;
624                         }
625                         ofd_read_unlock(env, fo);
626                         ofd_object_put(env, fo);
627                 }
628         } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
629                 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
630                 *vallen = sizeof(__u32);
631         } else if (KEY_IS(KEY_LAST_FID)) {
632                 struct ofd_device       *ofd = ofd_exp(exp);
633                 struct ofd_seq          *oseq;
634                 struct lu_fid           *fid = val;
635                 int                     rc;
636
637                 if (fid == NULL) {
638                         *vallen = sizeof(struct lu_fid);
639                         GOTO(out, rc = 0);
640                 }
641
642                 if (*vallen < sizeof(*fid))
643                         GOTO(out, rc = -EOVERFLOW);
644
645                 ofd_info_init(env, exp);
646
647                 fid_le_to_cpu(fid, fid);
648
649                 oseq = ofd_seq_load(env, ofd,
650                                     ostid_seq((struct ost_id *)fid));
651                 if (IS_ERR(oseq))
652                         GOTO(out, rc = PTR_ERR(oseq));
653
654                 rc = ostid_to_fid(fid, &oseq->os_oi,
655                              ofd->ofd_lut.lut_lsd.lsd_osd_index);
656                 if (rc != 0)
657                         GOTO(out_put, rc);
658
659                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
660                        PFID(fid));
661                 *vallen = sizeof(*fid);
662 out_put:
663                 ofd_seq_put(env, oseq);
664         } else {
665                 CERROR("Not supported key %s\n", (char*)key);
666                 rc = -EOPNOTSUPP;
667         }
668 out:
669         RETURN(rc);
670 }
671
672 /** helper function for statfs, also used by grant code */
673 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
674                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
675 {
676         int rc;
677
678         spin_lock(&ofd->ofd_osfs_lock);
679         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
680                 obd_size unstable;
681
682                 /* statfs data are too old, get up-to-date one.
683                  * we must be cautious here since multiple threads might be
684                  * willing to update statfs data concurrently and we must
685                  * grant that cached statfs data are always consistent */
686
687                 if (ofd->ofd_statfs_inflight == 0)
688                         /* clear inflight counter if no users, although it would
689                          * take a while to overflow this 64-bit counter ... */
690                         ofd->ofd_osfs_inflight = 0;
691                 /* notify ofd_grant_commit() that we want to track writes
692                  * completed as of now */
693                 ofd->ofd_statfs_inflight++;
694                 /* record value of inflight counter before running statfs to
695                  * compute the diff once statfs is completed */
696                 unstable = ofd->ofd_osfs_inflight;
697                 spin_unlock(&ofd->ofd_osfs_lock);
698
699                 /* statfs can sleep ... hopefully not for too long since we can
700                  * call it fairly often as space fills up */
701                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
702                 if (unlikely(rc))
703                         return rc;
704
705                 spin_lock(&ofd->ofd_grant_lock);
706                 spin_lock(&ofd->ofd_osfs_lock);
707                 /* calculate how much space was written while we released the
708                  * ofd_osfs_lock */
709                 unstable = ofd->ofd_osfs_inflight - unstable;
710                 ofd->ofd_osfs_unstable = 0;
711                 if (unstable) {
712                         /* some writes completed while we were running statfs
713                          * w/o the ofd_osfs_lock. Those ones got added to
714                          * the cached statfs data that we are about to crunch.
715                          * Take them into account in the new statfs data */
716                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
717                                                unstable >> ofd->ofd_blockbits);
718                         /* However, we don't really know if those writes got
719                          * accounted in the statfs call, so tell
720                          * ofd_grant_space_left() there is some uncertainty
721                          * on the accounting of those writes.
722                          * The purpose is to prevent spurious error messages in
723                          * ofd_grant_space_left() since those writes might be
724                          * accounted twice. */
725                         ofd->ofd_osfs_unstable += unstable;
726                 }
727                 /* similarly, there is some uncertainty on write requests
728                  * between prepare & commit */
729                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
730                 spin_unlock(&ofd->ofd_grant_lock);
731
732                 /* finally udpate cached statfs data */
733                 ofd->ofd_osfs = *osfs;
734                 ofd->ofd_osfs_age = cfs_time_current_64();
735
736                 ofd->ofd_statfs_inflight--; /* stop tracking */
737                 if (ofd->ofd_statfs_inflight == 0)
738                         ofd->ofd_osfs_inflight = 0;
739                 spin_unlock(&ofd->ofd_osfs_lock);
740
741                 if (from_cache)
742                         *from_cache = 0;
743         } else {
744                 /* use cached statfs data */
745                 *osfs = ofd->ofd_osfs;
746                 spin_unlock(&ofd->ofd_osfs_lock);
747                 if (from_cache)
748                         *from_cache = 1;
749         }
750         return 0;
751 }
752
753 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
754                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
755 {
756         struct obd_device       *obd = class_exp2obd(exp);
757         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
758         int                      rc;
759
760         ENTRY;
761
762         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
763         if (unlikely(rc))
764                 GOTO(out, rc);
765
766         /* at least try to account for cached pages.  its still racy and
767          * might be under-reporting if clients haven't announced their
768          * caches with brw recently */
769
770         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
771                " pending "LPU64" free "LPU64" avail "LPU64"\n",
772                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
773                osfs->os_bfree << ofd->ofd_blockbits,
774                osfs->os_bavail << ofd->ofd_blockbits);
775
776         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
777                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
778                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
779
780         /* The QoS code on the MDS does not care about space reserved for
781          * precreate, so take it out. */
782         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
783                 struct filter_export_data *fed;
784
785                 fed = &obd->obd_self_export->exp_filter_data;
786                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
787                                          fed->fed_grant >> ofd->ofd_blockbits);
788         }
789
790         ofd_grant_sanity_check(obd, __FUNCTION__);
791         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
792                LPU64" objects: "LPU64" free; state %x\n",
793                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
794                osfs->os_files, osfs->os_ffree, osfs->os_state);
795
796         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
797                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
798                 osfs->os_ffree = 0;
799
800         /* OS_STATE_READONLY can be set by OSD already */
801         if (ofd->ofd_raid_degraded)
802                 osfs->os_state |= OS_STATE_DEGRADED;
803
804         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
805                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
806                  * should not see a block size > page size, otherwise
807                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
808                  * block size which is the biggest block size known to work
809                  * with all client's page size. */
810                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
811                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
812                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
813                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
814         }
815
816         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
817                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
818                 osfs->os_bfree = osfs->os_bavail = 2;
819
820         EXIT;
821 out:
822         return rc;
823 }
824
825 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
826                 struct obd_info *oinfo, struct obd_trans_info *oti)
827 {
828         struct ofd_thread_info  *info;
829         struct ofd_device       *ofd = ofd_exp(exp);
830         struct ldlm_namespace   *ns = ofd->ofd_namespace;
831         struct ldlm_resource    *res;
832         struct ofd_object       *fo;
833         struct obdo             *oa = oinfo->oi_oa;
834         struct filter_fid       *ff = NULL;
835         int                      rc = 0;
836
837         ENTRY;
838
839         info = ofd_info_init(env, exp);
840         ofd_oti2info(info, oti);
841
842         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
843         if (rc != 0)
844                 RETURN(rc);
845
846         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
847         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
848                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
849         if (rc)
850                 GOTO(out, rc);
851
852         /* This would be very bad - accidentally truncating a file when
853          * changing the time or similar - bug 12203. */
854         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
855             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
856                 static char mdsinum[48];
857
858                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
859                         snprintf(mdsinum, sizeof(mdsinum) - 1,
860                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
861                                  oinfo->oi_oa->o_parent_oid, 0);
862                 else
863                         mdsinum[0] = '\0';
864
865                 CERROR("%s: setattr from %s trying to truncate object "DFID
866                        " %s\n", exp->exp_obd->obd_name,
867                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
868                 GOTO(out, rc = -EPERM);
869         }
870
871         fo = ofd_object_find(env, ofd, &info->fti_fid);
872         if (IS_ERR(fo)) {
873                 CERROR("%s: can't find object "DFID"\n",
874                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
875                 GOTO(out, rc = PTR_ERR(fo));
876         }
877
878         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
879         info->fti_attr.la_valid &= ~LA_TYPE;
880
881         if (oa->o_valid & OBD_MD_FLFID) {
882                 ff = &info->fti_mds_fid;
883                 ofd_prepare_fidea(ff, oa);
884         }
885
886         /* setting objects attributes (including owner/group) */
887         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
888         if (rc)
889                 GOTO(out_unlock, rc);
890
891         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
892         if (res != NULL) {
893                 ldlm_res_lvbo_update(res, NULL, 0);
894                 ldlm_resource_putref(res);
895         }
896
897         oinfo->oi_oa->o_valid = OBD_MD_FLID;
898
899         /* Quota release needs uid/gid info */
900         rc = ofd_attr_get(env, fo, &info->fti_attr);
901         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
902                      OFD_VALID_FLAGS | LA_UID | LA_GID);
903         ofd_info2oti(info, oti);
904
905         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
906         EXIT;
907 out_unlock:
908         ofd_object_put(env, fo);
909 out:
910         return rc;
911 }
912
913 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
914                      struct obd_info *oinfo, struct obd_trans_info *oti,
915                      struct ptlrpc_request_set *rqset)
916 {
917         struct ofd_thread_info  *info;
918         struct ofd_device       *ofd = ofd_exp(exp);
919         struct ldlm_namespace   *ns = ofd->ofd_namespace;
920         struct ldlm_resource    *res;
921         struct ofd_object       *fo;
922         struct filter_fid       *ff = NULL;
923         int                      rc = 0;
924
925         ENTRY;
926
927         info = ofd_info_init(env, exp);
928         ofd_oti2info(info, oti);
929
930         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
931         if (rc != 0)
932                 RETURN(rc);
933         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
934
935         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
936                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
937                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
938                oinfo->oi_policy.l_extent.end);
939
940         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
941                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
942         if (rc)
943                 GOTO(out_env, rc);
944
945         fo = ofd_object_find(env, ofd, &info->fti_fid);
946         if (IS_ERR(fo)) {
947                 CERROR("%s: error finding object "DFID": rc = %ld\n",
948                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
949                        PTR_ERR(fo));
950                 GOTO(out_env, rc = PTR_ERR(fo));
951         }
952
953         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
954         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
955                 /* Truncate case */
956                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
957         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
958                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
959         }
960
961         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
962                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
963         info->fti_attr.la_valid &= ~LA_TYPE;
964         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
965         info->fti_attr.la_valid |= LA_SIZE;
966
967         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
968                 ff = &info->fti_mds_fid;
969                 ofd_prepare_fidea(ff, oinfo->oi_oa);
970         }
971
972         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
973                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
974                               ff);
975         if (rc)
976                 GOTO(out, rc);
977
978         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
979         if (res != NULL) {
980                 ldlm_res_lvbo_update(res, NULL, 0);
981                 ldlm_resource_putref(res);
982         }
983
984         oinfo->oi_oa->o_valid = OBD_MD_FLID;
985         /* Quota release needs uid/gid info */
986         rc = ofd_attr_get(env, fo, &info->fti_attr);
987         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
988                      OFD_VALID_FLAGS | LA_UID | LA_GID);
989         ofd_info2oti(info, oti);
990
991         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
992         EXIT;
993 out:
994         ofd_object_put(env, fo);
995 out_env:
996         return rc;
997 }
998
999 static int ofd_destroy_by_fid(const struct lu_env *env,
1000                               struct ofd_device *ofd,
1001                               const struct lu_fid *fid, int orphan)
1002 {
1003         struct ofd_thread_info  *info = ofd_info(env);
1004         struct lustre_handle     lockh;
1005         __u64                    flags = LDLM_AST_DISCARD_DATA, rc = 0;
1006         ldlm_policy_data_t       policy = {
1007                                         .l_extent = { 0, OBD_OBJECT_EOF }
1008                                  };
1009         struct ofd_object       *fo;
1010
1011         ENTRY;
1012
1013         fo = ofd_object_find(env, ofd, fid);
1014         if (IS_ERR(fo))
1015                 RETURN(PTR_ERR(fo));
1016         if (!ofd_object_exists(fo))
1017                 GOTO(out, rc = -ENOENT);
1018
1019         /* Tell the clients that the object is gone now and that they should
1020          * throw away any cached pages. */
1021         ost_fid_build_resid(fid, &info->fti_resid);
1022         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1023                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1024                                     ldlm_blocking_ast, ldlm_completion_ast,
1025                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1026
1027         /* We only care about the side-effects, just drop the lock. */
1028         if (rc == ELDLM_OK)
1029                 ldlm_lock_decref(&lockh, LCK_PW);
1030
1031         LASSERT(fo != NULL);
1032
1033         rc = ofd_object_destroy(env, fo, orphan);
1034         EXIT;
1035 out:
1036         ofd_object_put(env, fo);
1037         RETURN(rc);
1038 }
1039
1040 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1041                 struct obdo *oa, struct lov_stripe_md *md,
1042                 struct obd_trans_info *oti, struct obd_export *md_exp,
1043                 void *capa)
1044 {
1045         struct ofd_device       *ofd = ofd_exp(exp);
1046         struct ofd_thread_info  *info;
1047         obd_count                count;
1048         int                      rc = 0;
1049
1050         ENTRY;
1051
1052         info = ofd_info_init(env, exp);
1053         ofd_oti2info(info, oti);
1054
1055         if (!(oa->o_valid & OBD_MD_FLGROUP))
1056                 ostid_set_seq_mdt0(&oa->o_oi);
1057
1058         /* check that o_misc makes sense */
1059         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1060                 count = oa->o_misc;
1061         else
1062                 count = 1; /* default case - single destroy */
1063
1064         /**
1065          * There can be sequence of objects to destroy. Therefore this request
1066          * may have multiple transaction involved in. It is OK, we need only
1067          * the highest used transno to be reported back in reply but not for
1068          * replays, they must report their transno
1069          */
1070         if (info->fti_transno == 0) /* not replay */
1071                 info->fti_mult_trans = 1;
1072
1073         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1074                POSTID(&oa->o_oi), count);
1075         while (count > 0) {
1076                 int lrc;
1077
1078                 lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
1079                 if (lrc != 0) {
1080                         if (rc == 0)
1081                                 rc = lrc;
1082                         GOTO(out, rc);
1083                 }
1084                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1085                 if (lrc == -ENOENT) {
1086                         CDEBUG(D_INODE,
1087                                "%s: destroying non-existent object "DFID"\n",
1088                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1089                         /* rewrite rc with -ENOENT only if it is 0 */
1090                         if (rc == 0)
1091                                 rc = lrc;
1092                 } else if (lrc != 0) {
1093                         CERROR("%s: error destroying object "DFID": %d\n",
1094                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1095                                rc);
1096                         rc = lrc;
1097                 }
1098                 count--;
1099                 ostid_inc_id(&oa->o_oi);
1100         }
1101
1102         /* if we have transaction then there were some deletions, we don't
1103          * need to return ENOENT in that case because it will not wait
1104          * for commit of these deletions. The ENOENT must be returned only
1105          * if there were no transations.
1106          */
1107         if (rc == -ENOENT) {
1108                 if (info->fti_transno != 0)
1109                         rc = 0;
1110         } else if (rc != 0) {
1111                 /*
1112                  * If we have at least one transaction then llog record
1113                  * on server will be removed upon commit, so for rc != 0
1114                  * we return no transno and llog record will be reprocessed.
1115                  */
1116                 info->fti_transno = 0;
1117         }
1118         ofd_info2oti(info, oti);
1119 out:
1120         RETURN(rc);
1121 }
1122
1123 static int ofd_orphans_destroy(const struct lu_env *env,
1124                                struct obd_export *exp, struct ofd_device *ofd,
1125                                struct obdo *oa)
1126 {
1127         struct ofd_thread_info  *info = ofd_info(env);
1128         obd_id                   last;
1129         int                      skip_orphan;
1130         int                      rc = 0;
1131         struct ost_id            oi = oa->o_oi;
1132         __u64                    end_id = ostid_id(&oa->o_oi);
1133         struct ofd_seq          *oseq;
1134
1135         ENTRY;
1136
1137         oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
1138         if (oseq == NULL) {
1139                 CERROR("%s: Can not find seq for "DOSTID"\n",
1140                        ofd_name(ofd), POSTID(&oa->o_oi));
1141                 RETURN(-EINVAL);
1142         }
1143
1144         LASSERT(exp != NULL);
1145         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1146
1147         last = ofd_seq_last_oid(oseq);
1148         LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1149                       " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
1150                       end_id + 1, last);
1151
1152         for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
1153                           ostid_dec_id(&oi)) {
1154                 rc = ostid_to_fid(&info->fti_fid, &oi, 0);
1155                 if (rc != 0)
1156                         GOTO(out_put, rc);
1157                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1158                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1159                         CEMERG("error destroying precreated id "DOSTID": %d\n",
1160                                POSTID(&oi), rc);
1161                 if (!skip_orphan) {
1162                         ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
1163                         /* update last_id on disk periodically so that if we
1164                          * restart * we don't need to re-scan all of the just
1165                          * deleted objects. */
1166                         if ((ostid_id(&oi) & 511) == 0)
1167                                 ofd_seq_last_oid_write(env, ofd, oseq);
1168                 }
1169         }
1170         CDEBUG(D_HA, "%s: after destroy: set last_objids"DOSTID"\n",
1171                ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
1172         if (!skip_orphan) {
1173                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1174         } else {
1175                 /* don't reuse orphan object, return last used objid */
1176                 ostid_set_id(&oa->o_oi, last);
1177                 rc = 0;
1178         }
1179 out_put:
1180         ofd_seq_put(env, oseq);
1181         RETURN(rc);
1182 }
1183
1184 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1185                struct obdo *oa, struct lov_stripe_md **ea,
1186                struct obd_trans_info *oti)
1187 {
1188         struct ofd_device       *ofd = ofd_exp(exp);
1189         struct ofd_thread_info  *info;
1190         obd_seq                 seq = ostid_seq(&oa->o_oi);
1191         struct ofd_seq          *oseq;
1192         int                     rc = 0, diff;
1193         int                     sync_trans = 0;
1194
1195         ENTRY;
1196
1197         info = ofd_info_init(env, exp);
1198         ofd_oti2info(info, oti);
1199
1200         LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1201         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1202
1203         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1204
1205         oseq = ofd_seq_load(env, ofd, seq);
1206         if (IS_ERR(oseq)) {
1207                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1208                        ofd_name(ofd), seq, PTR_ERR(oseq));
1209                 RETURN(-EINVAL);
1210         }
1211
1212         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1213             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1214                 if (!ofd_obd(ofd)->obd_recovering ||
1215                     ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1216                         CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1217                                POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1218                         GOTO(out_nolock, rc = -EINVAL);
1219                 }
1220                 /* do nothing because we create objects during first write */
1221                 GOTO(out_nolock, rc = 0);
1222         }
1223         /* former ofd_handle_precreate */
1224         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1225             (oa->o_flags & OBD_FL_DELORPHAN)) {
1226                 /* destroy orphans */
1227                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1228                         CERROR("%s: dropping old orphan cleanup request\n",
1229                                ofd_name(ofd));
1230                         GOTO(out_nolock, rc = 0);
1231                 }
1232                 /* This causes inflight precreates to abort and drop lock */
1233                 oseq->os_destroys_in_progress = 1;
1234                 mutex_lock(&oseq->os_create_lock);
1235                 if (!oseq->os_destroys_in_progress) {
1236                         CERROR("%s:["LPU64"] destroys_in_progress already"
1237                                " cleared\n", exp->exp_obd->obd_name,
1238                                ostid_seq(&oa->o_oi));
1239                         ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1240                         GOTO(out, rc = 0);
1241                 }
1242                 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1243                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1244                         ofd_seq_last_oid(oseq), diff);
1245                 if (-diff > OST_MAX_PRECREATE) {
1246                         /* FIXME: should reset precreate_next_id on MDS */
1247                         rc = 0;
1248                 } else if (diff < 0) {
1249                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1250                         oseq->os_destroys_in_progress = 0;
1251                 } else {
1252                         /* XXX: Used by MDS for the first time! */
1253                         oseq->os_destroys_in_progress = 0;
1254                 }
1255         } else {
1256                 mutex_lock(&oseq->os_create_lock);
1257                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1258                         CERROR("%s: dropping old precreate request\n",
1259                                 ofd_obd(ofd)->obd_name);
1260                         GOTO(out, rc = 0);
1261                 }
1262                 /* only precreate if seq is 0, IDIF or normal and also o_id
1263                  * must be specfied */
1264                 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1265                      !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1266                      !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1267                                 ostid_id(&oa->o_oi) == 0) {
1268                         diff = 1; /* shouldn't we create this right now? */
1269                 } else {
1270                         diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1271                         /* Do sync create if the seq is about to used up */
1272                         if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1273                             fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1274                                 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1275                                         sync_trans = 1;
1276                         } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1277                                 if (unlikely(ostid_id(&oa->o_oi) >=
1278                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1279                                         sync_trans = 1;
1280                         } else {
1281                                 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1282                                        ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1283                                 GOTO(out, rc = -EINVAL);
1284                         }
1285                 }
1286         }
1287         if (diff > 0) {
1288                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1289                 obd_id           next_id;
1290                 int              created = 0;
1291                 int              count;
1292
1293                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1294                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1295                         /* don't enforce grant during orphan recovery */
1296                         rc = ofd_grant_create(env,
1297                                               ofd_obd(ofd)->obd_self_export,
1298                                               &diff);
1299                         if (rc) {
1300                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1301                                        "space for precreate (%d): rc = %d\n",
1302                                        ofd_name(ofd), diff, rc);
1303                                 diff = 0;
1304                         }
1305                 }
1306
1307                 while (diff > 0) {
1308                         next_id = ofd_seq_last_oid(oseq) + 1;
1309                         count = ofd_precreate_batch(ofd, diff);
1310
1311                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1312                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1313                                count, ostid_seq(&oa->o_oi), next_id);
1314
1315                         if (cfs_time_after(jiffies, enough_time)) {
1316                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1317                                               " created at a rate of %d/s\n",
1318                                               ofd_obd(ofd)->obd_name,
1319                                               created, diff + created,
1320                                               created / DISK_TIMEOUT);
1321                                 break;
1322                         }
1323
1324                         rc = ofd_precreate_objects(env, ofd, next_id,
1325                                                    oseq, count, sync_trans);
1326                         if (rc > 0) {
1327                                 created += rc;
1328                                 diff -= rc;
1329                         } else if (rc < 0) {
1330                                 break;
1331                         }
1332                 }
1333                 if (created > 0)
1334                         /* some objects got created, we can return
1335                          * them, even if last creation failed */
1336                         rc = 0;
1337                 else
1338                         CERROR("%s: unable to precreate: rc = %d\n",
1339                                ofd_name(ofd), rc);
1340
1341                 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1342                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1343
1344                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1345                     !(oa->o_flags & OBD_FL_DELORPHAN))
1346                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1347                                          rc);
1348         }
1349
1350         ofd_info2oti(info, oti);
1351 out:
1352         mutex_unlock(&oseq->os_create_lock);
1353 out_nolock:
1354         if (rc == 0 && ea != NULL) {
1355                 struct lov_stripe_md *lsm = *ea;
1356
1357                 lsm->lsm_oi = oa->o_oi;
1358         }
1359         ofd_seq_put(env, oseq);
1360         RETURN(rc);
1361 }
1362
1363 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1364                 struct obd_info *oinfo)
1365 {
1366         struct ofd_device       *ofd = ofd_exp(exp);
1367         struct ofd_thread_info  *info;
1368         struct ofd_object       *fo;
1369         __u64                    curr_version;
1370         int                      rc = 0;
1371
1372         ENTRY;
1373
1374         info = ofd_info_init(env, exp);
1375
1376         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1377         if (rc != 0)
1378                 GOTO(out, rc);
1379         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1380                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1381         if (rc)
1382                 GOTO(out, rc);
1383
1384         fo = ofd_object_find(env, ofd, &info->fti_fid);
1385         if (IS_ERR(fo))
1386                 GOTO(out, rc = PTR_ERR(fo));
1387         LASSERT(fo != NULL);
1388         rc = ofd_attr_get(env, fo, &info->fti_attr);
1389         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1390         if (rc == 0)
1391                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1392                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1393
1394         /* Store object version in reply */
1395         curr_version = dt_version_get(env, ofd_object_child(fo));
1396         if ((__s64)curr_version != -EOPNOTSUPP) {
1397                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1398                 oinfo->oi_oa->o_data_version = curr_version;
1399         }
1400         ofd_object_put(env, fo);
1401 out:
1402         RETURN(rc);
1403 }
1404
1405 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1406                     struct obd_info *oinfo, obd_size start, obd_size end,
1407                     struct ptlrpc_request_set *set)
1408 {
1409         struct ofd_device       *ofd = ofd_exp(exp);
1410         struct ofd_thread_info  *info;
1411         struct ofd_object       *fo;
1412         int                      rc = 0;
1413
1414         ENTRY;
1415
1416         /* if no objid is specified, it means "sync whole filesystem" */
1417         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1418                 rc = dt_sync(env, ofd->ofd_osd);
1419                 GOTO(out, rc);
1420         }
1421
1422         info = ofd_info_init(env, exp);
1423         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1424         if (rc != 0)
1425                 GOTO(out, rc);
1426
1427         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1428                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1429         if (rc)
1430                 GOTO(out, rc);
1431
1432         fo = ofd_object_find(env, ofd, &info->fti_fid);
1433         if (IS_ERR(fo)) {
1434                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1435                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1436                        PTR_ERR(fo));
1437                 GOTO(out, rc = PTR_ERR(fo));
1438         }
1439
1440         if (!ofd_object_exists(fo))
1441                 GOTO(put, rc = -ENOENT);
1442
1443         if (dt_version_get(env, ofd_object_child(fo)) >
1444             ofd_obd(ofd)->obd_last_committed) {
1445                 rc = dt_object_sync(env, ofd_object_child(fo));
1446                 if (rc)
1447                         GOTO(put, rc);
1448         }
1449
1450         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1451         rc = ofd_attr_get(env, fo, &info->fti_attr);
1452         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1453
1454         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1455         EXIT;
1456 put:
1457         ofd_object_put(env, fo);
1458 out:
1459         return rc;
1460 }
1461
1462 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1463                                    struct ofd_device *ofd, void *karg)
1464 {
1465         struct obd_ioctl_data *data = karg;
1466         struct lu_fid          fid;
1467         struct ofd_object     *fo;
1468         dt_obj_version_t       version;
1469         int                    rc = 0;
1470
1471         ENTRY;
1472
1473         if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1474                 GOTO(out, rc = -EINVAL);
1475
1476         if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1477                 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1478         } else if (data->ioc_inlbuf3 != NULL &&
1479                    data->ioc_inllen3 == sizeof(__u64) &&
1480                    data->ioc_inlbuf4 != NULL &&
1481                    data->ioc_inllen4 == sizeof(__u64)) {
1482                 struct ost_id ostid;
1483
1484                 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1485                 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1486                 rc = ostid_to_fid(&fid, &ostid, 0);
1487                 if (rc != 0)
1488                         GOTO(out, rc);
1489         } else {
1490                 GOTO(out, rc = -EINVAL);
1491         }
1492
1493         if (!fid_is_sane(&fid))
1494                 GOTO(out, rc = -EINVAL);
1495
1496         fo = ofd_object_find(env, ofd, &fid);
1497         if (IS_ERR(fo))
1498                 GOTO(out, rc = PTR_ERR(fo));
1499
1500         if (!ofd_object_exists(fo))
1501                 GOTO(out_fo, rc = -ENOENT);
1502
1503         if (lu_object_remote(&fo->ofo_obj.do_lu))
1504                 GOTO(out_fo, rc = -EREMOTE);
1505
1506         version = dt_version_get(env, ofd_object_child(fo));
1507         if (version == 0)
1508                 GOTO(out_fo, rc = -EIO);
1509
1510         *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1511
1512         EXIT;
1513 out_fo:
1514         ofd_object_put(env, fo);
1515 out:
1516         return rc;
1517 }
1518
1519 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1520                   void *karg, void *uarg)
1521 {
1522         struct lu_env            env;
1523         struct ofd_device       *ofd = ofd_exp(exp);
1524         struct obd_device       *obd = ofd_obd(ofd);
1525         int                      rc;
1526
1527         ENTRY;
1528
1529         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1530         rc = lu_env_init(&env, LCT_DT_THREAD);
1531         if (rc)
1532                 RETURN(rc);
1533
1534         switch (cmd) {
1535         case OBD_IOC_ABORT_RECOVERY:
1536                 CERROR("%s: aborting recovery\n", obd->obd_name);
1537                 target_stop_recovery_thread(obd);
1538                 break;
1539         case OBD_IOC_SYNC:
1540                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1541                 rc = dt_sync(&env, ofd->ofd_osd);
1542                 break;
1543         case OBD_IOC_SET_READONLY:
1544                 rc = dt_sync(&env, ofd->ofd_osd);
1545                 if (rc == 0)
1546                         rc = dt_ro(&env, ofd->ofd_osd);
1547                 break;
1548         case OBD_IOC_GET_OBJ_VERSION:
1549                 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1550                 break;
1551         default:
1552                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1553                 rc = -ENOTTY;
1554         }
1555
1556         lu_env_fini(&env);
1557         RETURN(rc);
1558 }
1559
1560 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1561 {
1562         int rc = 0;
1563
1564         ENTRY;
1565
1566         switch(stage) {
1567         case OBD_CLEANUP_EARLY:
1568                 break;
1569         case OBD_CLEANUP_EXPORTS:
1570                 target_cleanup_recovery(obd);
1571                 break;
1572         }
1573         RETURN(rc);
1574 }
1575
1576 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1577 {
1578         ofd_fmd_expire(exp);
1579         return 0;
1580 }
1581
1582 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1583 {
1584         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1585         struct ofd_thread_info  *info;
1586         struct lu_env            env;
1587 #ifdef USE_HEALTH_CHECK_WRITE
1588         struct thandle          *th;
1589 #endif
1590         int                      rc = 0;
1591
1592         /* obd_proc_read_health pass NULL env, we need real one */
1593         rc = lu_env_init(&env, LCT_DT_THREAD);
1594         if (rc)
1595                 RETURN(rc);
1596
1597         info = ofd_info_init(&env, NULL);
1598         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1599         if (unlikely(rc))
1600                 GOTO(out, rc);
1601
1602         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1603                 GOTO(out, rc = -EROFS);
1604
1605 #ifdef USE_HEALTH_CHECK_WRITE
1606         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1607         if (info->fti_buf.lb_buf == NULL)
1608                 GOTO(out, rc = -ENOMEM);
1609
1610         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1611         info->fti_off = 0;
1612
1613         th = dt_trans_create(&env, ofd->ofd_osd);
1614         if (IS_ERR(th))
1615                 GOTO(out, rc = PTR_ERR(th));
1616
1617         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1618                                      info->fti_buf.lb_len, info->fti_off, th);
1619         if (rc == 0) {
1620                 th->th_sync = 1; /* sync IO is needed */
1621                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1622                 if (rc == 0)
1623                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1624                                              &info->fti_buf, &info->fti_off,
1625                                              th);
1626         }
1627         dt_trans_stop(&env, ofd->ofd_osd, th);
1628
1629         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1630
1631         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1632 #endif
1633 out:
1634         lu_env_fini(&env);
1635         return !!rc;
1636 }
1637
1638 /*
1639  * Handle quota control requests to consult current usage/limit.
1640  *
1641  * \param obd - is the obd device associated with the ofd
1642  * \param exp - is the client's export
1643  * \param oqctl - is the obd_quotactl request to be processed
1644  */
1645 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1646                         struct obd_quotactl *oqctl)
1647 {
1648         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1649         struct lu_env       env;
1650         int                 rc;
1651         ENTRY;
1652
1653         /* report success for quota on/off for interoperability with current MDT
1654          * stack */
1655         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1656                 RETURN(0);
1657
1658         rc = lu_env_init(&env, LCT_DT_THREAD);
1659         if (rc)
1660                 RETURN(rc);
1661
1662         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1663         lu_env_fini(&env);
1664
1665         RETURN(rc);
1666 }
1667
1668 struct obd_ops ofd_obd_ops = {
1669         .o_owner                = THIS_MODULE,
1670         .o_connect              = ofd_obd_connect,
1671         .o_reconnect            = ofd_obd_reconnect,
1672         .o_disconnect           = ofd_obd_disconnect,
1673         .o_set_info_async       = ofd_set_info_async,
1674         .o_get_info             = ofd_get_info,
1675         .o_create               = ofd_create,
1676         .o_statfs               = ofd_statfs,
1677         .o_setattr              = ofd_setattr,
1678         .o_preprw               = ofd_preprw,
1679         .o_commitrw             = ofd_commitrw,
1680         .o_destroy              = ofd_destroy,
1681         .o_init_export          = ofd_init_export,
1682         .o_destroy_export       = ofd_destroy_export,
1683         .o_postrecov            = ofd_obd_postrecov,
1684         .o_punch                = ofd_punch,
1685         .o_getattr              = ofd_getattr,
1686         .o_sync                 = ofd_sync,
1687         .o_iocontrol            = ofd_iocontrol,
1688         .o_precleanup           = ofd_precleanup,
1689         .o_ping                 = ofd_ping,
1690         .o_health_check         = ofd_health_check,
1691         .o_quotactl             = ofd_quotactl,
1692 };