Whamcloud - gitweb
c843e34fde047d33985fe42a31f351ea3229024e
[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 #include <lustre_lfsck.h>
50
51 static int ofd_export_stats_init(struct ofd_device *ofd,
52                                  struct obd_export *exp, void *client_nid)
53 {
54         struct obd_device       *obd = ofd_obd(ofd);
55         struct nid_stat         *stats;
56         int                      num_stats;
57         int                      rc, newnid = 0;
58
59         ENTRY;
60
61         LASSERT(obd->obd_uses_nid_stats);
62
63         if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
64                 /* Self-export gets no proc entry */
65                 RETURN(0);
66
67         rc = lprocfs_exp_setup(exp, client_nid, &newnid);
68         if (rc) {
69                 /* Mask error for already created
70                  * /proc entries */
71                 if (rc == -EALREADY)
72                         rc = 0;
73                 RETURN(rc);
74         }
75
76         if (newnid == 0)
77                 RETURN(0);
78
79         stats = exp->exp_nid_stats;
80         LASSERT(stats != NULL);
81
82         num_stats = NUM_OBD_STATS + LPROC_OFD_STATS_LAST;
83
84         stats->nid_stats = lprocfs_alloc_stats(num_stats,
85                                                LPROCFS_STATS_FLAG_NOPERCPU);
86         if (stats->nid_stats == NULL)
87                 return -ENOMEM;
88
89         lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
90         ofd_stats_counter_init(stats->nid_stats);
91         rc = lprocfs_register_stats(stats->nid_proc, "stats",
92                                     stats->nid_stats);
93         if (rc)
94                 GOTO(clean, rc);
95
96         rc = lprocfs_nid_ldlm_stats_init(stats);
97         if (rc) {
98                 lprocfs_free_stats(&stats->nid_stats);
99                 GOTO(clean, rc);
100         }
101
102         RETURN(0);
103 clean:
104         return rc;
105 }
106
107 static int ofd_parse_connect_data(const struct lu_env *env,
108                                   struct obd_export *exp,
109                                   struct obd_connect_data *data,
110                                   bool new_connection)
111 {
112         struct ofd_device                *ofd = ofd_exp(exp);
113         struct filter_export_data        *fed = &exp->exp_filter_data;
114
115         if (!data)
116                 RETURN(0);
117
118         CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
119                " ocd_version: %x ocd_grant: %d ocd_index: %u"
120                " ocd_group %u\n",
121                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
122                data->ocd_connect_flags, data->ocd_version,
123                data->ocd_grant, data->ocd_index, data->ocd_group);
124
125         if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
126                 CWARN("!!! This export (nid %s) used object group %d "
127                       "earlier; now it's trying to use group %d!  This could "
128                       "be a bug in the MDS. Please report to "
129                       "http://bugs.whamcloud.com/\n",
130                       obd_export_nid2str(exp), fed->fed_group,
131                       data->ocd_group);
132                 RETURN(-EPROTO);
133         }
134         fed->fed_group = data->ocd_group;
135
136         data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
137         data->ocd_version = LUSTRE_VERSION_CODE;
138
139         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
140         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
141                 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
142                        exp->exp_obd->obd_name, data->ocd_group);
143         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
144                 RETURN(-EPROTO);
145
146         if (ofd_grant_param_supp(exp)) {
147                 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
148                 /* ocd_{blocksize,inodespace} are log2 values */
149                 data->ocd_blocksize  = ofd->ofd_blockbits;
150                 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
151                 /* ocd_grant_extent is in 1K blocks */
152                 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
153         }
154
155         if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
156                 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
157                                                     new_connection);
158
159         if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
160                 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
161                 int                    index = lsd->lsd_osd_index;
162
163                 if (index != data->ocd_index) {
164                         LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
165                                            " %u doesn't match actual OST index"
166                                            " %u in last_rcvd file, bad "
167                                            "configuration?\n",
168                                            obd_export_nid2str(exp), index,
169                                            data->ocd_index);
170                         RETURN(-EBADF);
171                 }
172                 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
173                         /* this will only happen on the first connect */
174                         lsd->lsd_feature_compat |= OBD_COMPAT_OST;
175                         /* sync is not needed here as lut_client_add will
176                          * set exp_need_sync flag */
177                         tgt_server_data_update(env, &ofd->ofd_lut, 0);
178                 }
179         }
180         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
181                 data->ocd_brw_size = 65536;
182         } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
183                 data->ocd_brw_size = min(data->ocd_brw_size,
184                                          (__u32)DT_MAX_BRW_SIZE);
185                 if (data->ocd_brw_size == 0) {
186                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
187                                " ocd_version: %x ocd_grant: %d ocd_index: %u "
188                                "ocd_brw_size is unexpectedly zero, "
189                                "network data corruption?"
190                                "Refusing connection of this client\n",
191                                exp->exp_obd->obd_name,
192                                exp->exp_client_uuid.uuid,
193                                exp, data->ocd_connect_flags, data->ocd_version,
194                                data->ocd_grant, data->ocd_index);
195                         RETURN(-EPROTO);
196                 }
197         }
198
199         if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
200                 __u32 cksum_types = data->ocd_cksum_types;
201
202                 /* The client set in ocd_cksum_types the checksum types it
203                  * supports. We have to mask off the algorithms that we don't
204                  * support */
205                 data->ocd_cksum_types &= cksum_types_supported_server();
206
207                 if (unlikely(data->ocd_cksum_types == 0)) {
208                         CERROR("%s: Connect with checksum support but no "
209                                "ocd_cksum_types is set\n",
210                                exp->exp_obd->obd_name);
211                         RETURN(-EPROTO);
212                 }
213
214                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
215                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
216                        cksum_types, data->ocd_cksum_types);
217         } else {
218                 /* This client does not support OBD_CONNECT_CKSUM
219                  * fall back to CRC32 */
220                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
221                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
222                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
223         }
224
225         if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
226                 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
227
228         if (OCD_HAS_FLAG(data, PINGLESS)) {
229                 if (ptlrpc_pinger_suppress_pings()) {
230                         spin_lock(&exp->exp_obd->obd_dev_lock);
231                         list_del_init(&exp->exp_obd_chain_timed);
232                         spin_unlock(&exp->exp_obd->obd_dev_lock);
233                 } else {
234                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
235                 }
236         }
237
238         RETURN(0);
239 }
240
241 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
242                              struct obd_device *obd, struct obd_uuid *cluuid,
243                              struct obd_connect_data *data, void *localdata)
244 {
245         struct ofd_device       *ofd;
246         int                      rc;
247
248         ENTRY;
249
250         if (exp == NULL || obd == NULL || cluuid == NULL)
251                 RETURN(-EINVAL);
252
253         ofd = ofd_dev(obd->obd_lu_dev);
254
255         rc = ofd_parse_connect_data(env, exp, data, false);
256         if (rc == 0)
257                 ofd_export_stats_init(ofd, exp, localdata);
258
259         RETURN(rc);
260 }
261
262 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
263                            struct obd_device *obd, struct obd_uuid *cluuid,
264                            struct obd_connect_data *data, void *localdata)
265 {
266         struct obd_export       *exp;
267         struct ofd_device       *ofd;
268         struct lustre_handle     conn = { 0 };
269         int                      rc;
270         ENTRY;
271
272         if (_exp == NULL || obd == NULL || cluuid == NULL)
273                 RETURN(-EINVAL);
274
275         ofd = ofd_dev(obd->obd_lu_dev);
276
277         rc = class_connect(&conn, obd, cluuid);
278         if (rc)
279                 RETURN(rc);
280
281         exp = class_conn2export(&conn);
282         LASSERT(exp != NULL);
283
284         rc = ofd_parse_connect_data(env, exp, data, true);
285         if (rc)
286                 GOTO(out, rc);
287
288         if (obd->obd_replayable) {
289                 struct tg_export_data *ted = &exp->exp_target_data;
290
291                 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
292                        sizeof(ted->ted_lcd->lcd_uuid));
293                 rc = tgt_client_new(env, exp);
294                 if (rc != 0)
295                         GOTO(out, rc);
296                 ofd_export_stats_init(ofd, exp, localdata);
297         }
298
299         CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
300                data ? data->ocd_group : -1);
301
302 out:
303         if (rc != 0) {
304                 class_disconnect(exp);
305                 *_exp = NULL;
306         } else {
307                 *_exp = exp;
308         }
309         RETURN(rc);
310 }
311
312 static int ofd_obd_disconnect(struct obd_export *exp)
313 {
314         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
315         struct lu_env            env;
316         int                      rc;
317
318         ENTRY;
319
320         LASSERT(exp);
321         class_export_get(exp);
322
323         if (!(exp->exp_flags & OBD_OPT_FORCE))
324                 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
325
326         rc = server_disconnect_export(exp);
327
328         ofd_grant_discard(exp);
329
330         /* Do not erase record for recoverable client. */
331         if (exp->exp_obd->obd_replayable &&
332             (!exp->exp_obd->obd_fail || exp->exp_failed)) {
333                 rc = lu_env_init(&env, LCT_DT_THREAD);
334                 if (rc)
335                         GOTO(out, rc);
336
337                 tgt_client_del(&env, exp);
338                 lu_env_fini(&env);
339         }
340 out:
341         class_export_put(exp);
342         RETURN(rc);
343 }
344
345 static int ofd_init_export(struct obd_export *exp)
346 {
347         int rc;
348
349         spin_lock_init(&exp->exp_filter_data.fed_lock);
350         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
351         atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
352         spin_lock(&exp->exp_lock);
353         exp->exp_connecting = 1;
354         spin_unlock(&exp->exp_lock);
355
356         /* self-export doesn't need client data and ldlm initialization */
357         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
358                                      &exp->exp_client_uuid)))
359                 return 0;
360
361         rc = tgt_client_alloc(exp);
362         if (rc == 0)
363                 ldlm_init_export(exp);
364         if (rc)
365                 CERROR("%s: Can't initialize export: rc %d\n",
366                        exp->exp_obd->obd_name, rc);
367         return rc;
368 }
369
370 static int ofd_destroy_export(struct obd_export *exp)
371 {
372         struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
373
374         if (exp->exp_filter_data.fed_pending)
375                 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
376                        "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
377                        exp, exp->exp_filter_data.fed_pending);
378
379         target_destroy_export(exp);
380
381         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
382                                      &exp->exp_client_uuid)))
383                 return 0;
384
385         ldlm_destroy_export(exp);
386         tgt_client_free(exp);
387
388         ofd_fmd_cleanup(exp);
389
390         /*
391          * discard grants once we're sure no more
392          * interaction with the client is possible
393          */
394         ofd_grant_discard(exp);
395         ofd_fmd_cleanup(exp);
396
397         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
398                 if (ofd->ofd_tot_granted_clients > 0)
399                         ofd->ofd_tot_granted_clients --;
400         }
401
402         if (!(exp->exp_flags & OBD_OPT_FORCE))
403                 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
404
405         LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
406         return 0;
407 }
408
409 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
410 {
411         struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
412
413         CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
414         return ldev->ld_ops->ldo_recovery_complete(env, ldev);
415 }
416
417 int ofd_obd_postrecov(struct obd_device *obd)
418 {
419         struct lu_env            env;
420         struct lu_device        *ldev = obd->obd_lu_dev;
421         int                      rc;
422
423         ENTRY;
424
425         rc = lu_env_init(&env, LCT_DT_THREAD);
426         if (rc)
427                 RETURN(rc);
428         ofd_info_init(&env, obd->obd_self_export);
429
430         rc = ofd_postrecov(&env, ofd_dev(ldev));
431
432         lu_env_fini(&env);
433         RETURN(rc);
434 }
435
436 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
437                                   struct obd_device *obd, int initial)
438 {
439         struct filter_obd       *fo = &obd->u.filter;
440         struct sptlrpc_rule_set  tmp_rset;
441         int                      rc;
442
443         sptlrpc_rule_set_init(&tmp_rset);
444         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
445         if (rc) {
446                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
447                        obd->obd_name, rc);
448                 return rc;
449         }
450
451         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
452
453         write_lock(&fo->fo_sptlrpc_lock);
454         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
455         fo->fo_sptlrpc_rset = tmp_rset;
456         write_unlock(&fo->fo_sptlrpc_lock);
457
458         return 0;
459 }
460
461 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
462 {
463         int rc = 0;
464
465         ENTRY;
466
467         LCONSOLE_WARN("%s: received MDS connection from %s\n",
468                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
469         RETURN(rc);
470 }
471
472 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
473                               __u32 keylen, void *key, __u32 vallen, void *val,
474                               struct ptlrpc_request_set *set)
475 {
476         struct ofd_device       *ofd;
477         int                      rc = 0;
478
479         ENTRY;
480
481         if (exp->exp_obd == NULL) {
482                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
483                 RETURN(-EINVAL);
484         }
485
486         ofd = ofd_exp(exp);
487
488         if (KEY_IS(KEY_CAPA_KEY)) {
489                 rc = ofd_update_capa_key(ofd, val);
490                 if (rc)
491                         CERROR("%s: update capability key failed: rc = %d\n",
492                                exp->exp_obd->obd_name, rc);
493         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
494                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
495         } else if (KEY_IS(KEY_MDS_CONN)) {
496                 rc = ofd_set_mds_conn(exp, val);
497         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
498                 struct ost_body *body = val;
499
500                 ofd_info_init(env, exp);
501                 /** handle grant shrink, similar to a read request */
502                 ofd_grant_prepare_read(env, exp, &body->oa);
503         } else {
504                 CERROR("%s: Unsupported key %s\n",
505                        exp->exp_obd->obd_name, (char*)key);
506                 rc = -EOPNOTSUPP;
507         }
508
509         RETURN(rc);
510 }
511
512 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
513                         __u32 keylen, void *key, __u32 *vallen, void *val,
514                         struct lov_stripe_md *lsm)
515 {
516         struct ofd_device       *ofd;
517         int                     rc = 0;
518
519         ENTRY;
520
521         if (exp->exp_obd == NULL) {
522                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
523                 RETURN(-EINVAL);
524         }
525
526         /* Because ofd_get_info might be called from
527          * handle_request_in as well(see LU-3239), where env might
528          * not be initilaized correctly, and le_ses might be in
529          * an un-initialized state, so only refill le_ctx here */
530         rc = lu_env_refill((struct lu_env *)env);
531         if (rc != 0)
532                 RETURN(rc);
533
534         ofd = ofd_exp(exp);
535
536         if (KEY_IS(KEY_BLOCKSIZE)) {
537                 __u32 *blocksize = val;
538                 if (blocksize) {
539                         if (*vallen < sizeof(*blocksize))
540                                 GOTO(out, rc = -EOVERFLOW);
541                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
542                 }
543                 *vallen = sizeof(*blocksize);
544         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
545                 __u32 *blocksize_bits = val;
546                 if (blocksize_bits) {
547                         if (*vallen < sizeof(*blocksize_bits))
548                                 GOTO(out, rc = -EOVERFLOW);
549                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
550                 }
551                 *vallen = sizeof(*blocksize_bits);
552         } else if (KEY_IS(KEY_LAST_ID)) {
553                 obd_id *last_id = val;
554                 struct ofd_seq *oseq;
555
556                 if (val == NULL) {
557                         *vallen = sizeof(obd_id);
558                         GOTO(out, rc = 0);
559                 }
560                 ofd_info_init(env, exp);
561                 oseq = ofd_seq_load(env, ofd,
562                                     (obd_seq)exp->exp_filter_data.fed_group);
563                 LASSERT(!IS_ERR(oseq));
564                 if (last_id) {
565                         if (*vallen < sizeof(*last_id)) {
566                                 ofd_seq_put(env, oseq);
567                                 GOTO(out, rc = -EOVERFLOW);
568                         }
569                         *last_id = ofd_seq_last_oid(oseq);
570                 }
571                 ofd_seq_put(env, oseq);
572                 *vallen = sizeof(*last_id);
573         } else if (KEY_IS(KEY_FIEMAP)) {
574                 struct ofd_device               *ofd = ofd_exp(exp);
575                 struct ofd_object               *fo;
576                 struct ll_fiemap_info_key       *fm_key = key;
577                 struct lu_fid                    fid;
578
579                 if (val == NULL) {
580                         *vallen = fiemap_count_to_size(
581                                                fm_key->fiemap.fm_extent_count);
582                         GOTO(out, rc = 0);
583                 }
584
585                 rc = ostid_to_fid(&fid, &fm_key->oa.o_oi, 0);
586                 if (rc != 0)
587                         GOTO(out, rc);
588                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
589                        PFID(&fid));
590
591                 fo = ofd_object_find(env, ofd, &fid);
592                 if (IS_ERR(fo)) {
593                         CERROR("%s: error finding object "DFID"\n",
594                                exp->exp_obd->obd_name, PFID(&fid));
595                         rc = PTR_ERR(fo);
596                 } else {
597                         struct ll_user_fiemap *fiemap = val;
598
599                         ofd_read_lock(env, fo);
600                         if (ofd_object_exists(fo)) {
601                                 *fiemap = fm_key->fiemap;
602                                 rc = dt_fiemap_get(env,
603                                                    ofd_object_child(fo),
604                                                    fiemap);
605                         } else {
606                                 rc = -ENOENT;
607                         }
608                         ofd_read_unlock(env, fo);
609                         ofd_object_put(env, fo);
610                 }
611         } else if (KEY_IS(KEY_LAST_FID)) {
612                 struct ofd_device       *ofd = ofd_exp(exp);
613                 struct ofd_seq          *oseq;
614                 struct lu_fid           *fid = val;
615                 int                     rc;
616
617                 if (fid == NULL) {
618                         *vallen = sizeof(struct lu_fid);
619                         GOTO(out, rc = 0);
620                 }
621
622                 if (*vallen < sizeof(*fid))
623                         GOTO(out, rc = -EOVERFLOW);
624
625                 ofd_info_init(env, exp);
626
627                 fid_le_to_cpu(fid, fid);
628
629                 oseq = ofd_seq_load(env, ofd,
630                                     ostid_seq((struct ost_id *)fid));
631                 if (IS_ERR(oseq))
632                         GOTO(out, rc = PTR_ERR(oseq));
633
634                 rc = ostid_to_fid(fid, &oseq->os_oi,
635                              ofd->ofd_lut.lut_lsd.lsd_osd_index);
636                 if (rc != 0)
637                         GOTO(out_put, rc);
638
639                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
640                        PFID(fid));
641                 *vallen = sizeof(*fid);
642 out_put:
643                 ofd_seq_put(env, oseq);
644         } else {
645                 CERROR("Not supported key %s\n", (char*)key);
646                 rc = -EOPNOTSUPP;
647         }
648 out:
649         RETURN(rc);
650 }
651
652 /** helper function for statfs, also used by grant code */
653 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
654                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
655 {
656         int rc;
657
658         spin_lock(&ofd->ofd_osfs_lock);
659         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
660                 obd_size unstable;
661
662                 /* statfs data are too old, get up-to-date one.
663                  * we must be cautious here since multiple threads might be
664                  * willing to update statfs data concurrently and we must
665                  * grant that cached statfs data are always consistent */
666
667                 if (ofd->ofd_statfs_inflight == 0)
668                         /* clear inflight counter if no users, although it would
669                          * take a while to overflow this 64-bit counter ... */
670                         ofd->ofd_osfs_inflight = 0;
671                 /* notify ofd_grant_commit() that we want to track writes
672                  * completed as of now */
673                 ofd->ofd_statfs_inflight++;
674                 /* record value of inflight counter before running statfs to
675                  * compute the diff once statfs is completed */
676                 unstable = ofd->ofd_osfs_inflight;
677                 spin_unlock(&ofd->ofd_osfs_lock);
678
679                 /* statfs can sleep ... hopefully not for too long since we can
680                  * call it fairly often as space fills up */
681                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
682                 if (unlikely(rc))
683                         return rc;
684
685                 spin_lock(&ofd->ofd_grant_lock);
686                 spin_lock(&ofd->ofd_osfs_lock);
687                 /* calculate how much space was written while we released the
688                  * ofd_osfs_lock */
689                 unstable = ofd->ofd_osfs_inflight - unstable;
690                 ofd->ofd_osfs_unstable = 0;
691                 if (unstable) {
692                         /* some writes completed while we were running statfs
693                          * w/o the ofd_osfs_lock. Those ones got added to
694                          * the cached statfs data that we are about to crunch.
695                          * Take them into account in the new statfs data */
696                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
697                                                unstable >> ofd->ofd_blockbits);
698                         /* However, we don't really know if those writes got
699                          * accounted in the statfs call, so tell
700                          * ofd_grant_space_left() there is some uncertainty
701                          * on the accounting of those writes.
702                          * The purpose is to prevent spurious error messages in
703                          * ofd_grant_space_left() since those writes might be
704                          * accounted twice. */
705                         ofd->ofd_osfs_unstable += unstable;
706                 }
707                 /* similarly, there is some uncertainty on write requests
708                  * between prepare & commit */
709                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
710                 spin_unlock(&ofd->ofd_grant_lock);
711
712                 /* finally udpate cached statfs data */
713                 ofd->ofd_osfs = *osfs;
714                 ofd->ofd_osfs_age = cfs_time_current_64();
715
716                 ofd->ofd_statfs_inflight--; /* stop tracking */
717                 if (ofd->ofd_statfs_inflight == 0)
718                         ofd->ofd_osfs_inflight = 0;
719                 spin_unlock(&ofd->ofd_osfs_lock);
720
721                 if (from_cache)
722                         *from_cache = 0;
723         } else {
724                 /* use cached statfs data */
725                 *osfs = ofd->ofd_osfs;
726                 spin_unlock(&ofd->ofd_osfs_lock);
727                 if (from_cache)
728                         *from_cache = 1;
729         }
730         return 0;
731 }
732
733 int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
734                struct obd_statfs *osfs, __u64 max_age, __u32 flags)
735 {
736         struct obd_device       *obd = class_exp2obd(exp);
737         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
738         int                      rc;
739
740         ENTRY;
741
742         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
743         if (unlikely(rc))
744                 GOTO(out, rc);
745
746         /* at least try to account for cached pages.  its still racy and
747          * might be under-reporting if clients haven't announced their
748          * caches with brw recently */
749
750         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
751                " pending "LPU64" free "LPU64" avail "LPU64"\n",
752                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
753                osfs->os_bfree << ofd->ofd_blockbits,
754                osfs->os_bavail << ofd->ofd_blockbits);
755
756         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
757                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
758                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
759
760         /* The QoS code on the MDS does not care about space reserved for
761          * precreate, so take it out. */
762         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
763                 struct filter_export_data *fed;
764
765                 fed = &obd->obd_self_export->exp_filter_data;
766                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
767                                          fed->fed_grant >> ofd->ofd_blockbits);
768         }
769
770         ofd_grant_sanity_check(obd, __FUNCTION__);
771         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
772                LPU64" objects: "LPU64" free; state %x\n",
773                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
774                osfs->os_files, osfs->os_ffree, osfs->os_state);
775
776         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
777                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
778                 osfs->os_ffree = 0;
779
780         /* OS_STATE_READONLY can be set by OSD already */
781         if (ofd->ofd_raid_degraded)
782                 osfs->os_state |= OS_STATE_DEGRADED;
783
784         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
785                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
786                  * should not see a block size > page size, otherwise
787                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
788                  * block size which is the biggest block size known to work
789                  * with all client's page size. */
790                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
791                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
792                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
793                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
794         }
795
796         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
797                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
798                 osfs->os_bfree = osfs->os_bavail = 2;
799
800         EXIT;
801 out:
802         return rc;
803 }
804
805 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
806                 struct obd_info *oinfo, struct obd_trans_info *oti)
807 {
808         struct ofd_thread_info  *info;
809         struct ofd_device       *ofd = ofd_exp(exp);
810         struct ldlm_namespace   *ns = ofd->ofd_namespace;
811         struct ldlm_resource    *res;
812         struct ofd_object       *fo;
813         struct obdo             *oa = oinfo->oi_oa;
814         struct filter_fid       *ff = NULL;
815         int                      rc = 0;
816
817         ENTRY;
818
819         info = ofd_info_init(env, exp);
820         ofd_oti2info(info, oti);
821
822         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
823         if (rc != 0)
824                 RETURN(rc);
825
826         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
827         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
828                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
829         if (rc)
830                 GOTO(out, rc);
831
832         /* This would be very bad - accidentally truncating a file when
833          * changing the time or similar - bug 12203. */
834         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
835             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
836                 static char mdsinum[48];
837
838                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
839                         snprintf(mdsinum, sizeof(mdsinum) - 1,
840                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
841                                  oinfo->oi_oa->o_parent_oid, 0);
842                 else
843                         mdsinum[0] = '\0';
844
845                 CERROR("%s: setattr from %s trying to truncate object "DFID
846                        " %s\n", exp->exp_obd->obd_name,
847                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
848                 GOTO(out, rc = -EPERM);
849         }
850
851         fo = ofd_object_find(env, ofd, &info->fti_fid);
852         if (IS_ERR(fo)) {
853                 CERROR("%s: can't find object "DFID"\n",
854                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
855                 GOTO(out, rc = PTR_ERR(fo));
856         }
857
858         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
859         info->fti_attr.la_valid &= ~LA_TYPE;
860
861         if (oa->o_valid & OBD_MD_FLFID) {
862                 ff = &info->fti_mds_fid;
863                 ofd_prepare_fidea(ff, oa);
864         }
865
866         /* setting objects attributes (including owner/group) */
867         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
868         if (rc)
869                 GOTO(out_unlock, rc);
870
871         ofd_info2oti(info, oti);
872
873         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
874         EXIT;
875 out_unlock:
876         ofd_object_put(env, fo);
877 out:
878         if (rc == 0) {
879                 /* we do not call this before to avoid lu_object_find() in
880                  *  ->lvbo_update() holding another reference on the object.
881                  * otherwise concurrent destroy can make the object unavailable
882                  * for 2nd lu_object_find() waiting for the first reference
883                  * to go... deadlock! */
884                 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
885                 if (res != NULL) {
886                         ldlm_res_lvbo_update(res, NULL, 0);
887                         ldlm_resource_putref(res);
888                 }
889         }
890
891         return rc;
892 }
893
894 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
895                      struct obd_info *oinfo, struct obd_trans_info *oti,
896                      struct ptlrpc_request_set *rqset)
897 {
898         struct ofd_thread_info  *info;
899         struct ofd_device       *ofd = ofd_exp(exp);
900         struct ldlm_namespace   *ns = ofd->ofd_namespace;
901         struct ldlm_resource    *res;
902         struct ofd_object       *fo;
903         struct filter_fid       *ff = NULL;
904         int                      rc = 0;
905
906         ENTRY;
907
908         info = ofd_info_init(env, exp);
909         ofd_oti2info(info, oti);
910
911         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
912         if (rc != 0)
913                 RETURN(rc);
914         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
915
916         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
917                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
918                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
919                oinfo->oi_policy.l_extent.end);
920
921         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
922                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
923         if (rc)
924                 GOTO(out_env, rc);
925
926         fo = ofd_object_find(env, ofd, &info->fti_fid);
927         if (IS_ERR(fo)) {
928                 CERROR("%s: error finding object "DFID": rc = %ld\n",
929                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
930                        PTR_ERR(fo));
931                 GOTO(out_env, rc = PTR_ERR(fo));
932         }
933
934         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
935         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
936                 /* Truncate case */
937                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
938         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
939                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
940         }
941
942         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
943                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
944         info->fti_attr.la_valid &= ~LA_TYPE;
945         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
946         info->fti_attr.la_valid |= LA_SIZE;
947
948         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
949                 ff = &info->fti_mds_fid;
950                 ofd_prepare_fidea(ff, oinfo->oi_oa);
951         }
952
953         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
954                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
955                               ff);
956         if (rc)
957                 GOTO(out, rc);
958
959         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
960         if (res != NULL) {
961                 ldlm_res_lvbo_update(res, NULL, 0);
962                 ldlm_resource_putref(res);
963         }
964
965         oinfo->oi_oa->o_valid = OBD_MD_FLID;
966         /* Quota release needs uid/gid info */
967         rc = ofd_attr_get(env, fo, &info->fti_attr);
968         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
969                      OFD_VALID_FLAGS | LA_UID | LA_GID);
970         ofd_info2oti(info, oti);
971
972         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
973         EXIT;
974 out:
975         ofd_object_put(env, fo);
976 out_env:
977         return rc;
978 }
979
980 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
981                        const struct lu_fid *fid, int orphan)
982 {
983         struct ofd_thread_info  *info = ofd_info(env);
984         struct lustre_handle     lockh;
985         __u64                    flags = LDLM_FL_AST_DISCARD_DATA;
986         __u64                    rc = 0;
987         ldlm_policy_data_t       policy = {
988                                         .l_extent = { 0, OBD_OBJECT_EOF }
989                                  };
990         struct ofd_object       *fo;
991
992         ENTRY;
993
994         fo = ofd_object_find(env, ofd, fid);
995         if (IS_ERR(fo))
996                 RETURN(PTR_ERR(fo));
997         if (!ofd_object_exists(fo))
998                 GOTO(out, rc = -ENOENT);
999
1000         /* Tell the clients that the object is gone now and that they should
1001          * throw away any cached pages. */
1002         ost_fid_build_resid(fid, &info->fti_resid);
1003         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1004                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1005                                     ldlm_blocking_ast, ldlm_completion_ast,
1006                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1007
1008         /* We only care about the side-effects, just drop the lock. */
1009         if (rc == ELDLM_OK)
1010                 ldlm_lock_decref(&lockh, LCK_PW);
1011
1012         LASSERT(fo != NULL);
1013
1014         rc = ofd_object_destroy(env, fo, orphan);
1015         EXIT;
1016 out:
1017         ofd_object_put(env, fo);
1018         RETURN(rc);
1019 }
1020
1021 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1022                 struct obdo *oa, struct lov_stripe_md *md,
1023                 struct obd_trans_info *oti, struct obd_export *md_exp,
1024                 void *capa)
1025 {
1026         struct ofd_device       *ofd = ofd_exp(exp);
1027         struct ofd_thread_info  *info;
1028         obd_count                count;
1029         int                      rc = 0;
1030
1031         ENTRY;
1032
1033         info = ofd_info_init(env, exp);
1034         ofd_oti2info(info, oti);
1035
1036         if (!(oa->o_valid & OBD_MD_FLGROUP))
1037                 ostid_set_seq_mdt0(&oa->o_oi);
1038
1039         /* check that o_misc makes sense */
1040         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1041                 count = oa->o_misc;
1042         else
1043                 count = 1; /* default case - single destroy */
1044
1045         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1046                POSTID(&oa->o_oi), count);
1047         while (count > 0) {
1048                 int lrc;
1049
1050                 lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
1051                 if (lrc != 0) {
1052                         if (rc == 0)
1053                                 rc = lrc;
1054                         GOTO(out, rc);
1055                 }
1056                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1057                 if (lrc == -ENOENT) {
1058                         CDEBUG(D_INODE,
1059                                "%s: destroying non-existent object "DFID"\n",
1060                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1061                         /* rewrite rc with -ENOENT only if it is 0 */
1062                         if (rc == 0)
1063                                 rc = lrc;
1064                 } else if (lrc != 0) {
1065                         CERROR("%s: error destroying object "DFID": %d\n",
1066                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1067                                rc);
1068                         rc = lrc;
1069                 }
1070                 count--;
1071                 ostid_inc_id(&oa->o_oi);
1072         }
1073
1074         ofd_info2oti(info, oti);
1075 out:
1076         RETURN(rc);
1077 }
1078
1079 int ofd_orphans_destroy(const struct lu_env *env, struct obd_export *exp,
1080                         struct ofd_device *ofd, struct obdo *oa)
1081 {
1082         struct ofd_thread_info  *info = ofd_info(env);
1083         obd_id                   last;
1084         int                      skip_orphan;
1085         int                      rc = 0;
1086         struct ost_id            oi = oa->o_oi;
1087         __u64                    end_id = ostid_id(&oa->o_oi);
1088         struct ofd_seq          *oseq;
1089
1090         ENTRY;
1091
1092         oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
1093         if (oseq == NULL) {
1094                 CERROR("%s: Can not find seq for "DOSTID"\n",
1095                        ofd_name(ofd), POSTID(&oa->o_oi));
1096                 RETURN(-EINVAL);
1097         }
1098
1099         LASSERT(exp != NULL);
1100         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1101
1102         last = ofd_seq_last_oid(oseq);
1103         LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1104                  " to "DOSTID"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
1105                  end_id + 1, ostid_seq(&oa->o_oi), last);
1106
1107         for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
1108                           ostid_dec_id(&oi)) {
1109                 rc = ostid_to_fid(&info->fti_fid, &oi, 0);
1110                 if (rc != 0)
1111                         GOTO(out_put, rc);
1112                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1113                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1114                         CEMERG("%s: error destroying precreated id "DOSTID
1115                                ": rc = %d\n", ofd_name(ofd), POSTID(&oi), rc);
1116                 if (!skip_orphan) {
1117                         ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
1118                         /* update last_id on disk periodically so that if we
1119                          * restart * we don't need to re-scan all of the just
1120                          * deleted objects. */
1121                         if ((ostid_id(&oi) & 511) == 0)
1122                                 ofd_seq_last_oid_write(env, ofd, oseq);
1123                 }
1124         }
1125         CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1126                ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
1127         if (!skip_orphan) {
1128                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1129         } else {
1130                 /* don't reuse orphan object, return last used objid */
1131                 ostid_set_id(&oa->o_oi, last);
1132                 rc = 0;
1133         }
1134 out_put:
1135         ofd_seq_put(env, oseq);
1136         RETURN(rc);
1137 }
1138
1139 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1140                struct obdo *oa, struct lov_stripe_md **ea,
1141                struct obd_trans_info *oti)
1142 {
1143         struct ofd_device       *ofd = ofd_exp(exp);
1144         struct ofd_thread_info  *info;
1145         obd_seq                 seq = ostid_seq(&oa->o_oi);
1146         struct ofd_seq          *oseq;
1147         int                     rc = 0, diff;
1148         int                     sync_trans = 0;
1149
1150         ENTRY;
1151
1152         info = ofd_info_init(env, exp);
1153         ofd_oti2info(info, oti);
1154
1155         LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1156         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1157
1158         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1159
1160         oseq = ofd_seq_load(env, ofd, seq);
1161         if (IS_ERR(oseq)) {
1162                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1163                        ofd_name(ofd), seq, PTR_ERR(oseq));
1164                 RETURN(-EINVAL);
1165         }
1166
1167         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1168             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1169                 if (!ofd_obd(ofd)->obd_recovering ||
1170                     ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1171                         CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1172                                POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1173                         GOTO(out_nolock, rc = -EINVAL);
1174                 }
1175                 /* do nothing because we create objects during first write */
1176                 GOTO(out_nolock, rc = 0);
1177         }
1178         /* former ofd_handle_precreate */
1179         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1180             (oa->o_flags & OBD_FL_DELORPHAN)) {
1181                 /* destroy orphans */
1182                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1183                         CERROR("%s: dropping old orphan cleanup request\n",
1184                                ofd_name(ofd));
1185                         GOTO(out_nolock, rc = 0);
1186                 }
1187                 /* This causes inflight precreates to abort and drop lock */
1188                 oseq->os_destroys_in_progress = 1;
1189                 mutex_lock(&oseq->os_create_lock);
1190                 if (!oseq->os_destroys_in_progress) {
1191                         CERROR("%s:["LPU64"] destroys_in_progress already"
1192                                " cleared\n", exp->exp_obd->obd_name,
1193                                ostid_seq(&oa->o_oi));
1194                         ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1195                         GOTO(out, rc = 0);
1196                 }
1197                 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1198                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1199                         ofd_seq_last_oid(oseq), diff);
1200                 if (-diff > OST_MAX_PRECREATE) {
1201                         /* FIXME: should reset precreate_next_id on MDS */
1202                         rc = 0;
1203                 } else if (diff < 0) {
1204                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1205                         oseq->os_destroys_in_progress = 0;
1206                 } else {
1207                         /* XXX: Used by MDS for the first time! */
1208                         oseq->os_destroys_in_progress = 0;
1209                 }
1210         } else {
1211                 mutex_lock(&oseq->os_create_lock);
1212                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1213                         CERROR("%s: dropping old precreate request\n",
1214                                 ofd_obd(ofd)->obd_name);
1215                         GOTO(out, rc = 0);
1216                 }
1217                 /* only precreate if seq is 0, IDIF or normal and also o_id
1218                  * must be specfied */
1219                 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1220                      !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1221                      !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1222                                 ostid_id(&oa->o_oi) == 0) {
1223                         diff = 1; /* shouldn't we create this right now? */
1224                 } else {
1225                         diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1226                         /* Do sync create if the seq is about to used up */
1227                         if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1228                             fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1229                                 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1230                                         sync_trans = 1;
1231                         } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1232                                 if (unlikely(ostid_id(&oa->o_oi) >=
1233                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1234                                         sync_trans = 1;
1235                         } else {
1236                                 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1237                                        ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1238                                 GOTO(out, rc = -EINVAL);
1239                         }
1240                 }
1241         }
1242         if (diff > 0) {
1243                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1244                 obd_id           next_id;
1245                 int              created = 0;
1246                 int              count;
1247
1248                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1249                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1250                         /* don't enforce grant during orphan recovery */
1251                         rc = ofd_grant_create(env,
1252                                               ofd_obd(ofd)->obd_self_export,
1253                                               &diff);
1254                         if (rc) {
1255                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1256                                        "space for precreate (%d): rc = %d\n",
1257                                        ofd_name(ofd), diff, rc);
1258                                 diff = 0;
1259                         }
1260                 }
1261
1262                 /* This can happen if a new OST is formatted and installed
1263                  * in place of an old one at the same index.  Instead of
1264                  * precreating potentially millions of deleted old objects
1265                  * (possibly filling the OST), only precreate the last batch.
1266                  * LFSCK will eventually clean up any orphans. LU-14 */
1267                 if (diff > 5 * OST_MAX_PRECREATE) {
1268                         diff = OST_MAX_PRECREATE / 2;
1269                         LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1270                                       "larger than the LAST_ID "DOSTID", only "
1271                                       "precreating the last %u objects.\n",
1272                                       ofd_name(ofd), POSTID(&oa->o_oi),
1273                                       5 * OST_MAX_PRECREATE,
1274                                       POSTID(&oseq->os_oi), diff);
1275                         ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1276                 }
1277
1278                 while (diff > 0) {
1279                         next_id = ofd_seq_last_oid(oseq) + 1;
1280                         count = ofd_precreate_batch(ofd, diff);
1281
1282                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1283                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1284                                count, ostid_seq(&oa->o_oi), next_id);
1285
1286                         if (cfs_time_after(jiffies, enough_time)) {
1287                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1288                                               " created at a rate of %d/s\n",
1289                                               ofd_obd(ofd)->obd_name,
1290                                               created, diff + created,
1291                                               created / DISK_TIMEOUT);
1292                                 break;
1293                         }
1294
1295                         rc = ofd_precreate_objects(env, ofd, next_id,
1296                                                    oseq, count, sync_trans);
1297                         if (rc > 0) {
1298                                 created += rc;
1299                                 diff -= rc;
1300                         } else if (rc < 0) {
1301                                 break;
1302                         }
1303                 }
1304                 if (created > 0)
1305                         /* some objects got created, we can return
1306                          * them, even if last creation failed */
1307                         rc = 0;
1308                 else
1309                         CERROR("%s: unable to precreate: rc = %d\n",
1310                                ofd_name(ofd), rc);
1311
1312                 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1313                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1314
1315                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1316                     !(oa->o_flags & OBD_FL_DELORPHAN))
1317                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1318                                          rc);
1319         }
1320
1321         ofd_info2oti(info, oti);
1322 out:
1323         mutex_unlock(&oseq->os_create_lock);
1324 out_nolock:
1325         if (rc == 0 && ea != NULL) {
1326                 struct lov_stripe_md *lsm = *ea;
1327
1328                 lsm->lsm_oi = oa->o_oi;
1329         }
1330         ofd_seq_put(env, oseq);
1331         RETURN(rc);
1332 }
1333
1334 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1335                 struct obd_info *oinfo)
1336 {
1337         struct ofd_device       *ofd = ofd_exp(exp);
1338         struct ofd_thread_info  *info;
1339         struct ofd_object       *fo;
1340         int                      rc = 0;
1341
1342         ENTRY;
1343
1344         info = ofd_info_init(env, exp);
1345
1346         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1347         if (rc != 0)
1348                 GOTO(out, rc);
1349         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1350                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1351         if (rc)
1352                 GOTO(out, rc);
1353
1354         fo = ofd_object_find(env, ofd, &info->fti_fid);
1355         if (IS_ERR(fo))
1356                 GOTO(out, rc = PTR_ERR(fo));
1357         if (!ofd_object_exists(fo))
1358                 GOTO(out_put, rc = -ENOENT);
1359
1360         LASSERT(fo != NULL);
1361         rc = ofd_attr_get(env, fo, &info->fti_attr);
1362         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1363         if (rc == 0) {
1364                 __u64 curr_version;
1365
1366                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1367                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1368
1369                 /* Store object version in reply */
1370                 curr_version = dt_version_get(env, ofd_object_child(fo));
1371                 if ((__s64)curr_version != -EOPNOTSUPP) {
1372                         oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1373                         oinfo->oi_oa->o_data_version = curr_version;
1374                 }
1375         }
1376
1377 out_put:
1378         ofd_object_put(env, fo);
1379 out:
1380         RETURN(rc);
1381 }
1382
1383 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1384                     struct obd_info *oinfo, obd_size start, obd_size end,
1385                     struct ptlrpc_request_set *set)
1386 {
1387         struct ofd_device       *ofd = ofd_exp(exp);
1388         struct ofd_thread_info  *info;
1389         struct ofd_object       *fo;
1390         int                      rc = 0;
1391
1392         ENTRY;
1393
1394         /* if no objid is specified, it means "sync whole filesystem" */
1395         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1396                 rc = dt_sync(env, ofd->ofd_osd);
1397                 GOTO(out, rc);
1398         }
1399
1400         info = ofd_info_init(env, exp);
1401         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1402         if (rc != 0)
1403                 GOTO(out, rc);
1404
1405         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1406                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1407         if (rc)
1408                 GOTO(out, rc);
1409
1410         fo = ofd_object_find(env, ofd, &info->fti_fid);
1411         if (IS_ERR(fo)) {
1412                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1413                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1414                        PTR_ERR(fo));
1415                 GOTO(out, rc = PTR_ERR(fo));
1416         }
1417
1418         if (!ofd_object_exists(fo))
1419                 GOTO(put, rc = -ENOENT);
1420
1421         if (dt_version_get(env, ofd_object_child(fo)) >
1422             ofd_obd(ofd)->obd_last_committed) {
1423                 rc = dt_object_sync(env, ofd_object_child(fo));
1424                 if (rc)
1425                         GOTO(put, rc);
1426         }
1427
1428         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1429         rc = ofd_attr_get(env, fo, &info->fti_attr);
1430         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1431
1432         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1433         EXIT;
1434 put:
1435         ofd_object_put(env, fo);
1436 out:
1437         return rc;
1438 }
1439
1440 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1441                                    struct ofd_device *ofd, void *karg)
1442 {
1443         struct obd_ioctl_data *data = karg;
1444         struct lu_fid          fid;
1445         struct ofd_object     *fo;
1446         dt_obj_version_t       version;
1447         int                    rc = 0;
1448
1449         ENTRY;
1450
1451         if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1452                 GOTO(out, rc = -EINVAL);
1453
1454         if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1455                 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1456         } else if (data->ioc_inlbuf3 != NULL &&
1457                    data->ioc_inllen3 == sizeof(__u64) &&
1458                    data->ioc_inlbuf4 != NULL &&
1459                    data->ioc_inllen4 == sizeof(__u64)) {
1460                 struct ost_id ostid;
1461
1462                 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1463                 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1464                 rc = ostid_to_fid(&fid, &ostid, 0);
1465                 if (rc != 0)
1466                         GOTO(out, rc);
1467         } else {
1468                 GOTO(out, rc = -EINVAL);
1469         }
1470
1471         if (!fid_is_sane(&fid))
1472                 GOTO(out, rc = -EINVAL);
1473
1474         fo = ofd_object_find(env, ofd, &fid);
1475         if (IS_ERR(fo))
1476                 GOTO(out, rc = PTR_ERR(fo));
1477
1478         if (!ofd_object_exists(fo))
1479                 GOTO(out_fo, rc = -ENOENT);
1480
1481         if (lu_object_remote(&fo->ofo_obj.do_lu))
1482                 GOTO(out_fo, rc = -EREMOTE);
1483
1484         version = dt_version_get(env, ofd_object_child(fo));
1485         if (version == 0)
1486                 GOTO(out_fo, rc = -EIO);
1487
1488         *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1489
1490         EXIT;
1491 out_fo:
1492         ofd_object_put(env, fo);
1493 out:
1494         return rc;
1495 }
1496
1497 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1498                   void *karg, void *uarg)
1499 {
1500         struct lu_env            env;
1501         struct ofd_device       *ofd = ofd_exp(exp);
1502         struct obd_device       *obd = ofd_obd(ofd);
1503         int                      rc;
1504
1505         ENTRY;
1506
1507         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1508         rc = lu_env_init(&env, LCT_DT_THREAD);
1509         if (rc)
1510                 RETURN(rc);
1511
1512         switch (cmd) {
1513         case OBD_IOC_ABORT_RECOVERY:
1514                 CERROR("%s: aborting recovery\n", obd->obd_name);
1515                 target_stop_recovery_thread(obd);
1516                 break;
1517         case OBD_IOC_SYNC:
1518                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1519                 rc = dt_sync(&env, ofd->ofd_osd);
1520                 break;
1521         case OBD_IOC_SET_READONLY:
1522                 rc = dt_sync(&env, ofd->ofd_osd);
1523                 if (rc == 0)
1524                         rc = dt_ro(&env, ofd->ofd_osd);
1525                 break;
1526         case OBD_IOC_START_LFSCK: {
1527                 struct obd_ioctl_data *data = karg;
1528                 struct lfsck_start_param lsp;
1529
1530                 if (unlikely(data == NULL)) {
1531                         rc = -EINVAL;
1532                         break;
1533                 }
1534
1535                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1536                 lsp.lsp_namespace = ofd->ofd_namespace;
1537                 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1538                 break;
1539         }
1540         case OBD_IOC_STOP_LFSCK: {
1541                 rc = lfsck_stop(&env, ofd->ofd_osd, false);
1542                 break;
1543         }
1544         case OBD_IOC_GET_OBJ_VERSION:
1545                 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1546                 break;
1547         default:
1548                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1549                 rc = -ENOTTY;
1550         }
1551
1552         lu_env_fini(&env);
1553         RETURN(rc);
1554 }
1555
1556 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1557 {
1558         int rc = 0;
1559
1560         ENTRY;
1561
1562         switch(stage) {
1563         case OBD_CLEANUP_EARLY:
1564                 break;
1565         case OBD_CLEANUP_EXPORTS:
1566                 target_cleanup_recovery(obd);
1567                 break;
1568         }
1569         RETURN(rc);
1570 }
1571
1572 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1573 {
1574         ofd_fmd_expire(exp);
1575         return 0;
1576 }
1577
1578 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1579 {
1580         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1581         struct ofd_thread_info  *info;
1582         struct lu_env            env;
1583 #ifdef USE_HEALTH_CHECK_WRITE
1584         struct thandle          *th;
1585 #endif
1586         int                      rc = 0;
1587
1588         /* obd_proc_read_health pass NULL env, we need real one */
1589         rc = lu_env_init(&env, LCT_DT_THREAD);
1590         if (rc)
1591                 RETURN(rc);
1592
1593         info = ofd_info_init(&env, NULL);
1594         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1595         if (unlikely(rc))
1596                 GOTO(out, rc);
1597
1598         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1599                 GOTO(out, rc = -EROFS);
1600
1601 #ifdef USE_HEALTH_CHECK_WRITE
1602         OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1603         if (info->fti_buf.lb_buf == NULL)
1604                 GOTO(out, rc = -ENOMEM);
1605
1606         info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1607         info->fti_off = 0;
1608
1609         th = dt_trans_create(&env, ofd->ofd_osd);
1610         if (IS_ERR(th))
1611                 GOTO(out, rc = PTR_ERR(th));
1612
1613         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1614                                      info->fti_buf.lb_len, info->fti_off, th);
1615         if (rc == 0) {
1616                 th->th_sync = 1; /* sync IO is needed */
1617                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1618                 if (rc == 0)
1619                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1620                                              &info->fti_buf, &info->fti_off,
1621                                              th);
1622         }
1623         dt_trans_stop(&env, ofd->ofd_osd, th);
1624
1625         OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1626
1627         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1628 #endif
1629 out:
1630         lu_env_fini(&env);
1631         return !!rc;
1632 }
1633
1634 /*
1635  * Handle quota control requests to consult current usage/limit.
1636  *
1637  * \param obd - is the obd device associated with the ofd
1638  * \param exp - is the client's export
1639  * \param oqctl - is the obd_quotactl request to be processed
1640  */
1641 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1642                         struct obd_quotactl *oqctl)
1643 {
1644         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1645         struct lu_env       env;
1646         int                 rc;
1647         ENTRY;
1648
1649         /* report success for quota on/off for interoperability with current MDT
1650          * stack */
1651         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1652                 RETURN(0);
1653
1654         rc = lu_env_init(&env, LCT_DT_THREAD);
1655         if (rc)
1656                 RETURN(rc);
1657
1658         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1659         lu_env_fini(&env);
1660
1661         RETURN(rc);
1662 }
1663
1664 struct obd_ops ofd_obd_ops = {
1665         .o_owner                = THIS_MODULE,
1666         .o_connect              = ofd_obd_connect,
1667         .o_reconnect            = ofd_obd_reconnect,
1668         .o_disconnect           = ofd_obd_disconnect,
1669         .o_set_info_async       = ofd_set_info_async,
1670         .o_get_info             = ofd_get_info,
1671         .o_create               = ofd_create,
1672         .o_statfs               = ofd_statfs,
1673         .o_setattr              = ofd_setattr,
1674         .o_preprw               = ofd_preprw,
1675         .o_commitrw             = ofd_commitrw,
1676         .o_destroy              = ofd_destroy,
1677         .o_init_export          = ofd_init_export,
1678         .o_destroy_export       = ofd_destroy_export,
1679         .o_postrecov            = ofd_obd_postrecov,
1680         .o_punch                = ofd_punch,
1681         .o_getattr              = ofd_getattr,
1682         .o_sync                 = ofd_sync,
1683         .o_iocontrol            = ofd_iocontrol,
1684         .o_precleanup           = ofd_precleanup,
1685         .o_ping                 = ofd_ping,
1686         .o_health_check         = ofd_health_check,
1687         .o_quotactl             = ofd_quotactl,
1688 };