Whamcloud - gitweb
599d9f55814af3ba33e4236413d737057544d259
[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,
586                                   ofd->ofd_lut.lut_lsd.lsd_osd_index);
587                 if (rc != 0)
588                         GOTO(out, rc);
589                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
590                        PFID(&fid));
591
592                 fo = ofd_object_find(env, ofd, &fid);
593                 if (IS_ERR(fo)) {
594                         CERROR("%s: error finding object "DFID"\n",
595                                exp->exp_obd->obd_name, PFID(&fid));
596                         rc = PTR_ERR(fo);
597                 } else {
598                         struct ll_user_fiemap *fiemap = val;
599
600                         ofd_read_lock(env, fo);
601                         if (ofd_object_exists(fo)) {
602                                 *fiemap = fm_key->fiemap;
603                                 rc = dt_fiemap_get(env,
604                                                    ofd_object_child(fo),
605                                                    fiemap);
606                         } else {
607                                 rc = -ENOENT;
608                         }
609                         ofd_read_unlock(env, fo);
610                         ofd_object_put(env, fo);
611                 }
612         } else if (KEY_IS(KEY_LAST_FID)) {
613                 struct ofd_device       *ofd = ofd_exp(exp);
614                 struct ofd_seq          *oseq;
615                 struct lu_fid           *fid = val;
616                 int                     rc;
617
618                 if (fid == NULL) {
619                         *vallen = sizeof(struct lu_fid);
620                         GOTO(out, rc = 0);
621                 }
622
623                 if (*vallen < sizeof(*fid))
624                         GOTO(out, rc = -EOVERFLOW);
625
626                 ofd_info_init(env, exp);
627
628                 fid_le_to_cpu(fid, fid);
629
630                 oseq = ofd_seq_load(env, ofd,
631                                     ostid_seq((struct ost_id *)fid));
632                 if (IS_ERR(oseq))
633                         GOTO(out, rc = PTR_ERR(oseq));
634
635                 rc = ostid_to_fid(fid, &oseq->os_oi,
636                                   ofd->ofd_lut.lut_lsd.lsd_osd_index);
637                 if (rc != 0)
638                         GOTO(out_put, rc);
639
640                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
641                        PFID(fid));
642                 *vallen = sizeof(*fid);
643 out_put:
644                 ofd_seq_put(env, oseq);
645         } else {
646                 CERROR("Not supported key %s\n", (char*)key);
647                 rc = -EOPNOTSUPP;
648         }
649 out:
650         RETURN(rc);
651 }
652
653 /** helper function for statfs, also used by grant code */
654 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
655                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
656 {
657         int rc;
658
659         spin_lock(&ofd->ofd_osfs_lock);
660         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
661                 obd_size unstable;
662
663                 /* statfs data are too old, get up-to-date one.
664                  * we must be cautious here since multiple threads might be
665                  * willing to update statfs data concurrently and we must
666                  * grant that cached statfs data are always consistent */
667
668                 if (ofd->ofd_statfs_inflight == 0)
669                         /* clear inflight counter if no users, although it would
670                          * take a while to overflow this 64-bit counter ... */
671                         ofd->ofd_osfs_inflight = 0;
672                 /* notify ofd_grant_commit() that we want to track writes
673                  * completed as of now */
674                 ofd->ofd_statfs_inflight++;
675                 /* record value of inflight counter before running statfs to
676                  * compute the diff once statfs is completed */
677                 unstable = ofd->ofd_osfs_inflight;
678                 spin_unlock(&ofd->ofd_osfs_lock);
679
680                 /* statfs can sleep ... hopefully not for too long since we can
681                  * call it fairly often as space fills up */
682                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
683                 if (unlikely(rc))
684                         return rc;
685
686                 spin_lock(&ofd->ofd_grant_lock);
687                 spin_lock(&ofd->ofd_osfs_lock);
688                 /* calculate how much space was written while we released the
689                  * ofd_osfs_lock */
690                 unstable = ofd->ofd_osfs_inflight - unstable;
691                 ofd->ofd_osfs_unstable = 0;
692                 if (unstable) {
693                         /* some writes completed while we were running statfs
694                          * w/o the ofd_osfs_lock. Those ones got added to
695                          * the cached statfs data that we are about to crunch.
696                          * Take them into account in the new statfs data */
697                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
698                                                unstable >> ofd->ofd_blockbits);
699                         /* However, we don't really know if those writes got
700                          * accounted in the statfs call, so tell
701                          * ofd_grant_space_left() there is some uncertainty
702                          * on the accounting of those writes.
703                          * The purpose is to prevent spurious error messages in
704                          * ofd_grant_space_left() since those writes might be
705                          * accounted twice. */
706                         ofd->ofd_osfs_unstable += unstable;
707                 }
708                 /* similarly, there is some uncertainty on write requests
709                  * between prepare & commit */
710                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
711                 spin_unlock(&ofd->ofd_grant_lock);
712
713                 /* finally udpate cached statfs data */
714                 ofd->ofd_osfs = *osfs;
715                 ofd->ofd_osfs_age = cfs_time_current_64();
716
717                 ofd->ofd_statfs_inflight--; /* stop tracking */
718                 if (ofd->ofd_statfs_inflight == 0)
719                         ofd->ofd_osfs_inflight = 0;
720                 spin_unlock(&ofd->ofd_osfs_lock);
721
722                 if (from_cache)
723                         *from_cache = 0;
724         } else {
725                 /* use cached statfs data */
726                 *osfs = ofd->ofd_osfs;
727                 spin_unlock(&ofd->ofd_osfs_lock);
728                 if (from_cache)
729                         *from_cache = 1;
730         }
731         return 0;
732 }
733
734 int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
735                struct obd_statfs *osfs, __u64 max_age, __u32 flags)
736 {
737         struct obd_device       *obd = class_exp2obd(exp);
738         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
739         int                      rc;
740
741         ENTRY;
742
743         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
744         if (unlikely(rc))
745                 GOTO(out, rc);
746
747         /* at least try to account for cached pages.  its still racy and
748          * might be under-reporting if clients haven't announced their
749          * caches with brw recently */
750
751         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
752                " pending "LPU64" free "LPU64" avail "LPU64"\n",
753                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
754                osfs->os_bfree << ofd->ofd_blockbits,
755                osfs->os_bavail << ofd->ofd_blockbits);
756
757         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
758                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
759                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
760
761         /* The QoS code on the MDS does not care about space reserved for
762          * precreate, so take it out. */
763         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
764                 struct filter_export_data *fed;
765
766                 fed = &obd->obd_self_export->exp_filter_data;
767                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
768                                          fed->fed_grant >> ofd->ofd_blockbits);
769         }
770
771         ofd_grant_sanity_check(obd, __FUNCTION__);
772         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
773                LPU64" objects: "LPU64" free; state %x\n",
774                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
775                osfs->os_files, osfs->os_ffree, osfs->os_state);
776
777         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
778                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
779                 osfs->os_ffree = 0;
780
781         /* OS_STATE_READONLY can be set by OSD already */
782         if (ofd->ofd_raid_degraded)
783                 osfs->os_state |= OS_STATE_DEGRADED;
784
785         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
786                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
787                  * should not see a block size > page size, otherwise
788                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
789                  * block size which is the biggest block size known to work
790                  * with all client's page size. */
791                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
792                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
793                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
794                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
795         }
796
797         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
798                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
799                 osfs->os_bfree = osfs->os_bavail = 2;
800
801         EXIT;
802 out:
803         return rc;
804 }
805
806 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
807                 struct obd_info *oinfo, struct obd_trans_info *oti)
808 {
809         struct ofd_thread_info  *info;
810         struct ofd_device       *ofd = ofd_exp(exp);
811         struct ldlm_namespace   *ns = ofd->ofd_namespace;
812         struct ldlm_resource    *res;
813         struct ofd_object       *fo;
814         struct obdo             *oa = oinfo->oi_oa;
815         struct filter_fid       *ff = NULL;
816         int                      rc = 0;
817
818         ENTRY;
819
820         info = ofd_info_init(env, exp);
821         ofd_oti2info(info, oti);
822
823         info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
824         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
825         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
826                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
827         if (rc)
828                 GOTO(out, rc);
829
830         /* This would be very bad - accidentally truncating a file when
831          * changing the time or similar - bug 12203. */
832         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
833             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
834                 static char mdsinum[48];
835
836                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
837                         snprintf(mdsinum, sizeof(mdsinum) - 1,
838                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
839                                  oinfo->oi_oa->o_parent_oid, 0);
840                 else
841                         mdsinum[0] = '\0';
842
843                 CERROR("%s: setattr from %s trying to truncate object "DFID
844                        " %s\n", exp->exp_obd->obd_name,
845                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
846                 GOTO(out, rc = -EPERM);
847         }
848
849         fo = ofd_object_find(env, ofd, &info->fti_fid);
850         if (IS_ERR(fo)) {
851                 CERROR("%s: can't find object "DFID"\n",
852                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
853                 GOTO(out, rc = PTR_ERR(fo));
854         }
855
856         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
857         info->fti_attr.la_valid &= ~LA_TYPE;
858
859         if (oa->o_valid & OBD_MD_FLFID) {
860                 ff = &info->fti_mds_fid;
861                 ofd_prepare_fidea(ff, oa);
862         }
863
864         /* setting objects attributes (including owner/group) */
865         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
866         if (rc)
867                 GOTO(out_unlock, rc);
868
869         ofd_info2oti(info, oti);
870
871         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
872         EXIT;
873 out_unlock:
874         ofd_object_put(env, fo);
875 out:
876         if (rc == 0) {
877                 /* we do not call this before to avoid lu_object_find() in
878                  *  ->lvbo_update() holding another reference on the object.
879                  * otherwise concurrent destroy can make the object unavailable
880                  * for 2nd lu_object_find() waiting for the first reference
881                  * to go... deadlock! */
882                 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
883                 if (res != NULL) {
884                         ldlm_res_lvbo_update(res, NULL, 0);
885                         ldlm_resource_putref(res);
886                 }
887         }
888
889         return rc;
890 }
891
892 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
893                      struct obd_info *oinfo, struct obd_trans_info *oti,
894                      struct ptlrpc_request_set *rqset)
895 {
896         struct ofd_thread_info  *info;
897         struct ofd_device       *ofd = ofd_exp(exp);
898         struct ldlm_namespace   *ns = ofd->ofd_namespace;
899         struct ldlm_resource    *res;
900         struct ofd_object       *fo;
901         struct filter_fid       *ff = NULL;
902         int                      rc = 0;
903
904         ENTRY;
905
906         info = ofd_info_init(env, exp);
907         ofd_oti2info(info, oti);
908
909         info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
910         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
911
912         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
913                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
914                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
915                oinfo->oi_policy.l_extent.end);
916
917         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
918                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
919         if (rc)
920                 GOTO(out_env, rc);
921
922         fo = ofd_object_find(env, ofd, &info->fti_fid);
923         if (IS_ERR(fo)) {
924                 CERROR("%s: error finding object "DFID": rc = %ld\n",
925                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
926                        PTR_ERR(fo));
927                 GOTO(out_env, rc = PTR_ERR(fo));
928         }
929
930         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
931         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
932                 /* Truncate case */
933                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
934         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
935                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
936         }
937
938         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
939                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
940         info->fti_attr.la_valid &= ~LA_TYPE;
941         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
942         info->fti_attr.la_valid |= LA_SIZE;
943
944         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
945                 ff = &info->fti_mds_fid;
946                 ofd_prepare_fidea(ff, oinfo->oi_oa);
947         }
948
949         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
950                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
951                               ff);
952         if (rc)
953                 GOTO(out, rc);
954
955         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
956         if (res != NULL) {
957                 ldlm_res_lvbo_update(res, NULL, 0);
958                 ldlm_resource_putref(res);
959         }
960
961         oinfo->oi_oa->o_valid = OBD_MD_FLID;
962         /* Quota release needs uid/gid info */
963         rc = ofd_attr_get(env, fo, &info->fti_attr);
964         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
965                      OFD_VALID_FLAGS | LA_UID | LA_GID);
966         ofd_info2oti(info, oti);
967
968         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
969         EXIT;
970 out:
971         ofd_object_put(env, fo);
972 out_env:
973         return rc;
974 }
975
976 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
977                        const struct lu_fid *fid, int orphan)
978 {
979         struct ofd_thread_info  *info = ofd_info(env);
980         struct lustre_handle     lockh;
981         __u64                    flags = LDLM_FL_AST_DISCARD_DATA;
982         __u64                    rc = 0;
983         ldlm_policy_data_t       policy = {
984                                         .l_extent = { 0, OBD_OBJECT_EOF }
985                                  };
986         struct ofd_object       *fo;
987
988         ENTRY;
989
990         fo = ofd_object_find(env, ofd, fid);
991         if (IS_ERR(fo))
992                 RETURN(PTR_ERR(fo));
993         if (!ofd_object_exists(fo))
994                 GOTO(out, rc = -ENOENT);
995
996         /* Tell the clients that the object is gone now and that they should
997          * throw away any cached pages. */
998         ost_fid_build_resid(fid, &info->fti_resid);
999         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1000                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1001                                     ldlm_blocking_ast, ldlm_completion_ast,
1002                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1003
1004         /* We only care about the side-effects, just drop the lock. */
1005         if (rc == ELDLM_OK)
1006                 ldlm_lock_decref(&lockh, LCK_PW);
1007
1008         LASSERT(fo != NULL);
1009
1010         rc = ofd_object_destroy(env, fo, orphan);
1011         EXIT;
1012 out:
1013         ofd_object_put(env, fo);
1014         RETURN(rc);
1015 }
1016
1017 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1018                 struct obdo *oa, struct lov_stripe_md *md,
1019                 struct obd_trans_info *oti, struct obd_export *md_exp,
1020                 void *capa)
1021 {
1022         struct ofd_device       *ofd = ofd_exp(exp);
1023         struct ofd_thread_info  *info;
1024         struct lu_fid           *fid;
1025         obd_id                   oid;
1026         obd_count                count;
1027         int                      rc = 0;
1028
1029         ENTRY;
1030
1031         info = ofd_info_init(env, exp);
1032         ofd_oti2info(info, oti);
1033         fid = &info->fti_fid;
1034
1035         if (!(oa->o_valid & OBD_MD_FLGROUP))
1036                 ostid_set_seq_mdt0(&oa->o_oi);
1037
1038         *fid = oa->o_oi.oi_fid;
1039         oid = ostid_id(&oa->o_oi);
1040         LASSERT(oid != 0);
1041
1042         /* check that o_misc makes sense */
1043         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1044                 count = oa->o_misc;
1045         else
1046                 count = 1; /* default case - single destroy */
1047
1048         CDEBUG(D_INODE, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1049                POSTID(&oa->o_oi), count);
1050
1051         while (count > 0) {
1052                 int lrc;
1053
1054                 lrc = ofd_destroy_by_fid(env, ofd, fid, 0);
1055                 if (lrc == -ENOENT) {
1056                         CDEBUG(D_INODE,
1057                                "%s: destroying non-existent object "DFID"\n",
1058                                ofd_obd(ofd)->obd_name, PFID(fid));
1059                         /* rewrite rc with -ENOENT only if it is 0 */
1060                         if (rc == 0)
1061                                 rc = lrc;
1062                 } else if (lrc != 0) {
1063                         CERROR("%s: error destroying object "DFID": %d\n",
1064                                ofd_obd(ofd)->obd_name, PFID(fid),
1065                                rc);
1066                         rc = lrc;
1067                 }
1068
1069                 count--;
1070                 oid++;
1071                 lrc = fid_set_id(fid, oid);
1072                 if (unlikely(lrc != 0 && count > 0))
1073                         GOTO(out, rc = lrc);
1074         }
1075
1076         GOTO(out, rc);
1077
1078 out:
1079         ofd_info2oti(info, oti);
1080         fid_to_ostid(fid, &oa->o_oi);
1081         return rc;
1082 }
1083
1084 int ofd_orphans_destroy(const struct lu_env *env, struct obd_export *exp,
1085                         struct ofd_device *ofd, struct obdo *oa)
1086 {
1087         struct ofd_thread_info  *info   = ofd_info(env);
1088         struct lu_fid           *fid    = &info->fti_fid;
1089         struct ost_id           *oi     = &oa->o_oi;
1090         struct ofd_seq          *oseq;
1091         obd_seq                  seq    = ostid_seq(oi);
1092         obd_id                   end_id = ostid_id(oi);
1093         obd_id                   last;
1094         obd_id                   oid;
1095         int                      skip_orphan;
1096         int                      rc     = 0;
1097         ENTRY;
1098
1099         oseq = ofd_seq_get(ofd, seq);
1100         if (oseq == NULL) {
1101                 CERROR("%s: Can not find seq for "DOSTID"\n",
1102                        ofd_name(ofd), POSTID(oi));
1103                 RETURN(-EINVAL);
1104         }
1105
1106         *fid = oi->oi_fid;
1107         last = ofd_seq_last_oid(oseq);
1108         oid = last;
1109
1110         LASSERT(exp != NULL);
1111         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1112
1113         LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
1114                  " to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
1115
1116         while (oid > end_id) {
1117                 rc = fid_set_id(fid, oid);
1118                 if (unlikely(rc != 0))
1119                         GOTO(out_put, rc);
1120
1121                 rc = ofd_destroy_by_fid(env, ofd, fid, 1);
1122                 if (rc != 0 && rc != -ENOENT) /* this is pretty fatal... */
1123                         CEMERG("%s: error destroying precreated id "DFID
1124                                ": rc = %d\n", ofd_name(ofd), PFID(fid), rc);
1125
1126                 oid--;
1127                 if (!skip_orphan) {
1128                         ofd_seq_last_oid_set(oseq, oid);
1129                         /* update last_id on disk periodically so that if we
1130                          * restart * we don't need to re-scan all of the just
1131                          * deleted objects. */
1132                         if ((oid & 511) == 0)
1133                                 ofd_seq_last_oid_write(env, ofd, oseq);
1134                 }
1135         }
1136
1137         CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
1138                ofd_obd(ofd)->obd_name, seq, oid);
1139
1140         if (!skip_orphan) {
1141                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1142         } else {
1143                 /* don't reuse orphan object, return last used objid */
1144                 ostid_set_id(oi, last);
1145                 rc = 0;
1146         }
1147
1148         GOTO(out_put, rc);
1149
1150 out_put:
1151         ofd_seq_put(env, oseq);
1152         return rc;
1153 }
1154
1155 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1156                struct obdo *oa, struct lov_stripe_md **ea,
1157                struct obd_trans_info *oti)
1158 {
1159         struct ofd_device       *ofd = ofd_exp(exp);
1160         struct ofd_thread_info  *info;
1161         obd_seq                 seq = ostid_seq(&oa->o_oi);
1162         struct ofd_seq          *oseq;
1163         int                     rc = 0, diff;
1164         int                     sync_trans = 0;
1165
1166         ENTRY;
1167
1168         info = ofd_info_init(env, exp);
1169         ofd_oti2info(info, oti);
1170
1171         LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1172         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1173
1174         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1175
1176         oseq = ofd_seq_load(env, ofd, seq);
1177         if (IS_ERR(oseq)) {
1178                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1179                        ofd_name(ofd), seq, PTR_ERR(oseq));
1180                 RETURN(-EINVAL);
1181         }
1182
1183         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1184             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1185                 if (!ofd_obd(ofd)->obd_recovering ||
1186                     ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1187                         CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1188                                POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1189                         GOTO(out_nolock, rc = -EINVAL);
1190                 }
1191                 /* do nothing because we create objects during first write */
1192                 GOTO(out_nolock, rc = 0);
1193         }
1194         /* former ofd_handle_precreate */
1195         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1196             (oa->o_flags & OBD_FL_DELORPHAN)) {
1197                 /* destroy orphans */
1198                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1199                         CERROR("%s: dropping old orphan cleanup request\n",
1200                                ofd_name(ofd));
1201                         GOTO(out_nolock, rc = 0);
1202                 }
1203                 /* This causes inflight precreates to abort and drop lock */
1204                 oseq->os_destroys_in_progress = 1;
1205                 mutex_lock(&oseq->os_create_lock);
1206                 if (!oseq->os_destroys_in_progress) {
1207                         CERROR("%s:["LPU64"] destroys_in_progress already"
1208                                " cleared\n", exp->exp_obd->obd_name,
1209                                ostid_seq(&oa->o_oi));
1210                         ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1211                         GOTO(out, rc = 0);
1212                 }
1213                 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1214                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1215                         ofd_seq_last_oid(oseq), diff);
1216                 if (-diff > OST_MAX_PRECREATE) {
1217                         /* FIXME: should reset precreate_next_id on MDS */
1218                         rc = 0;
1219                 } else if (diff < 0) {
1220                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1221                         oseq->os_destroys_in_progress = 0;
1222                 } else {
1223                         /* XXX: Used by MDS for the first time! */
1224                         oseq->os_destroys_in_progress = 0;
1225                 }
1226         } else {
1227                 mutex_lock(&oseq->os_create_lock);
1228                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1229                         CERROR("%s: dropping old precreate request\n",
1230                                 ofd_obd(ofd)->obd_name);
1231                         GOTO(out, rc = 0);
1232                 }
1233                 /* only precreate if seq is 0, IDIF or normal and also o_id
1234                  * must be specfied */
1235                 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1236                      !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1237                      !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1238                                 ostid_id(&oa->o_oi) == 0) {
1239                         diff = 1; /* shouldn't we create this right now? */
1240                 } else {
1241                         diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1242                         /* Do sync create if the seq is about to used up */
1243                         if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1244                             fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1245                                 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1246                                         sync_trans = 1;
1247                         } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1248                                 if (unlikely(ostid_id(&oa->o_oi) >=
1249                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1250                                         sync_trans = 1;
1251                         } else {
1252                                 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1253                                        ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1254                                 GOTO(out, rc = -EINVAL);
1255                         }
1256                 }
1257         }
1258         if (diff > 0) {
1259                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1260                 obd_id           next_id;
1261                 int              created = 0;
1262                 int              count;
1263
1264                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1265                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1266                         /* don't enforce grant during orphan recovery */
1267                         rc = ofd_grant_create(env,
1268                                               ofd_obd(ofd)->obd_self_export,
1269                                               &diff);
1270                         if (rc) {
1271                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1272                                        "space for precreate (%d): rc = %d\n",
1273                                        ofd_name(ofd), diff, rc);
1274                                 diff = 0;
1275                         }
1276                 }
1277
1278                 /* This can happen if a new OST is formatted and installed
1279                  * in place of an old one at the same index.  Instead of
1280                  * precreating potentially millions of deleted old objects
1281                  * (possibly filling the OST), only precreate the last batch.
1282                  * LFSCK will eventually clean up any orphans. LU-14 */
1283                 if (diff > 5 * OST_MAX_PRECREATE) {
1284                         diff = OST_MAX_PRECREATE / 2;
1285                         LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
1286                                       "larger than the LAST_ID "DOSTID", only "
1287                                       "precreating the last %u objects.\n",
1288                                       ofd_name(ofd), POSTID(&oa->o_oi),
1289                                       5 * OST_MAX_PRECREATE,
1290                                       POSTID(&oseq->os_oi), diff);
1291                         ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
1292                 }
1293
1294                 while (diff > 0) {
1295                         next_id = ofd_seq_last_oid(oseq) + 1;
1296                         count = ofd_precreate_batch(ofd, diff);
1297
1298                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1299                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1300                                count, ostid_seq(&oa->o_oi), next_id);
1301
1302                         if (cfs_time_after(jiffies, enough_time)) {
1303                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1304                                               " created at a rate of %d/s\n",
1305                                               ofd_obd(ofd)->obd_name,
1306                                               created, diff + created,
1307                                               created / DISK_TIMEOUT);
1308                                 break;
1309                         }
1310
1311                         rc = ofd_precreate_objects(env, ofd, next_id,
1312                                                    oseq, count, sync_trans);
1313                         if (rc > 0) {
1314                                 created += rc;
1315                                 diff -= rc;
1316                         } else if (rc < 0) {
1317                                 break;
1318                         }
1319                 }
1320                 if (created > 0)
1321                         /* some objects got created, we can return
1322                          * them, even if last creation failed */
1323                         rc = 0;
1324                 else
1325                         CERROR("%s: unable to precreate: rc = %d\n",
1326                                ofd_name(ofd), rc);
1327
1328                 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1329                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1330
1331                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1332                     !(oa->o_flags & OBD_FL_DELORPHAN))
1333                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1334                                          rc);
1335         }
1336
1337         ofd_info2oti(info, oti);
1338 out:
1339         mutex_unlock(&oseq->os_create_lock);
1340 out_nolock:
1341         if (rc == 0 && ea != NULL) {
1342                 struct lov_stripe_md *lsm = *ea;
1343
1344                 lsm->lsm_oi = oa->o_oi;
1345         }
1346         ofd_seq_put(env, oseq);
1347         RETURN(rc);
1348 }
1349
1350 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1351                 struct obd_info *oinfo)
1352 {
1353         struct ofd_device       *ofd = ofd_exp(exp);
1354         struct ofd_thread_info  *info;
1355         struct ofd_object       *fo;
1356         int                      rc = 0;
1357
1358         ENTRY;
1359
1360         info = ofd_info_init(env, exp);
1361
1362         info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
1363         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1364                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1365         if (rc)
1366                 GOTO(out, rc);
1367
1368         fo = ofd_object_find(env, ofd, &info->fti_fid);
1369         if (IS_ERR(fo))
1370                 GOTO(out, rc = PTR_ERR(fo));
1371         if (!ofd_object_exists(fo))
1372                 GOTO(out_put, rc = -ENOENT);
1373
1374         LASSERT(fo != NULL);
1375         rc = ofd_attr_get(env, fo, &info->fti_attr);
1376         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1377         if (rc == 0) {
1378                 __u64 curr_version;
1379
1380                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1381                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1382
1383                 /* Store object version in reply */
1384                 curr_version = dt_version_get(env, ofd_object_child(fo));
1385                 if ((__s64)curr_version != -EOPNOTSUPP) {
1386                         oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1387                         oinfo->oi_oa->o_data_version = curr_version;
1388                 }
1389         }
1390
1391 out_put:
1392         ofd_object_put(env, fo);
1393 out:
1394         RETURN(rc);
1395 }
1396
1397 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1398                     struct obd_info *oinfo, obd_size start, obd_size end,
1399                     struct ptlrpc_request_set *set)
1400 {
1401         struct ofd_device       *ofd = ofd_exp(exp);
1402         struct ofd_thread_info  *info;
1403         struct ofd_object       *fo;
1404         int                      rc = 0;
1405
1406         ENTRY;
1407
1408         /* if no objid is specified, it means "sync whole filesystem" */
1409         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1410                 rc = dt_sync(env, ofd->ofd_osd);
1411                 GOTO(out, rc);
1412         }
1413
1414         info = ofd_info_init(env, exp);
1415         info->fti_fid = oinfo->oi_oa->o_oi.oi_fid;
1416         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1417                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1418         if (rc)
1419                 GOTO(out, rc);
1420
1421         fo = ofd_object_find(env, ofd, &info->fti_fid);
1422         if (IS_ERR(fo)) {
1423                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1424                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1425                        PTR_ERR(fo));
1426                 GOTO(out, rc = PTR_ERR(fo));
1427         }
1428
1429         if (!ofd_object_exists(fo))
1430                 GOTO(put, rc = -ENOENT);
1431
1432         if (dt_version_get(env, ofd_object_child(fo)) >
1433             ofd_obd(ofd)->obd_last_committed) {
1434                 rc = dt_object_sync(env, ofd_object_child(fo));
1435                 if (rc)
1436                         GOTO(put, rc);
1437         }
1438
1439         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1440         rc = ofd_attr_get(env, fo, &info->fti_attr);
1441         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1442
1443         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1444         EXIT;
1445 put:
1446         ofd_object_put(env, fo);
1447 out:
1448         return rc;
1449 }
1450
1451 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1452                                    struct ofd_device *ofd, void *karg)
1453 {
1454         struct obd_ioctl_data *data = karg;
1455         struct lu_fid          fid;
1456         struct ofd_object     *fo;
1457         dt_obj_version_t       version;
1458         int                    rc = 0;
1459
1460         ENTRY;
1461
1462         if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1463                 GOTO(out, rc = -EINVAL);
1464
1465         if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1466                 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1467         } else if (data->ioc_inlbuf3 != NULL &&
1468                    data->ioc_inllen3 == sizeof(__u64) &&
1469                    data->ioc_inlbuf4 != NULL &&
1470                    data->ioc_inllen4 == sizeof(__u64)) {
1471                 struct ost_id ostid;
1472
1473                 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1474                 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1475                 rc = ostid_to_fid(&fid, &ostid,
1476                                   ofd->ofd_lut.lut_lsd.lsd_osd_index);
1477                 if (rc != 0)
1478                         GOTO(out, rc);
1479         } else {
1480                 GOTO(out, rc = -EINVAL);
1481         }
1482
1483         if (!fid_is_sane(&fid))
1484                 GOTO(out, rc = -EINVAL);
1485
1486         fo = ofd_object_find(env, ofd, &fid);
1487         if (IS_ERR(fo))
1488                 GOTO(out, rc = PTR_ERR(fo));
1489
1490         if (!ofd_object_exists(fo))
1491                 GOTO(out_fo, rc = -ENOENT);
1492
1493         if (lu_object_remote(&fo->ofo_obj.do_lu))
1494                 GOTO(out_fo, rc = -EREMOTE);
1495
1496         version = dt_version_get(env, ofd_object_child(fo));
1497         if (version == 0)
1498                 GOTO(out_fo, rc = -EIO);
1499
1500         *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1501
1502         EXIT;
1503 out_fo:
1504         ofd_object_put(env, fo);
1505 out:
1506         return rc;
1507 }
1508
1509 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1510                   void *karg, void *uarg)
1511 {
1512         struct lu_env            env;
1513         struct ofd_device       *ofd = ofd_exp(exp);
1514         struct obd_device       *obd = ofd_obd(ofd);
1515         int                      rc;
1516
1517         ENTRY;
1518
1519         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1520         rc = lu_env_init(&env, LCT_DT_THREAD);
1521         if (rc)
1522                 RETURN(rc);
1523
1524         switch (cmd) {
1525         case OBD_IOC_ABORT_RECOVERY:
1526                 CERROR("%s: aborting recovery\n", obd->obd_name);
1527                 target_stop_recovery_thread(obd);
1528                 break;
1529         case OBD_IOC_SYNC:
1530                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1531                 rc = dt_sync(&env, ofd->ofd_osd);
1532                 break;
1533         case OBD_IOC_SET_READONLY:
1534                 rc = dt_sync(&env, ofd->ofd_osd);
1535                 if (rc == 0)
1536                         rc = dt_ro(&env, ofd->ofd_osd);
1537                 break;
1538         case OBD_IOC_START_LFSCK: {
1539                 struct obd_ioctl_data *data = karg;
1540                 struct lfsck_start_param lsp;
1541
1542                 if (unlikely(data == NULL)) {
1543                         rc = -EINVAL;
1544                         break;
1545                 }
1546
1547                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1548                 lsp.lsp_namespace = ofd->ofd_namespace;
1549                 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1550                 break;
1551         }
1552         case OBD_IOC_STOP_LFSCK: {
1553                 rc = lfsck_stop(&env, ofd->ofd_osd, false);
1554                 break;
1555         }
1556         case OBD_IOC_GET_OBJ_VERSION:
1557                 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1558                 break;
1559         default:
1560                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1561                 rc = -ENOTTY;
1562         }
1563
1564         lu_env_fini(&env);
1565         RETURN(rc);
1566 }
1567
1568 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1569 {
1570         int rc = 0;
1571
1572         ENTRY;
1573
1574         switch(stage) {
1575         case OBD_CLEANUP_EARLY:
1576                 break;
1577         case OBD_CLEANUP_EXPORTS:
1578                 target_cleanup_recovery(obd);
1579                 break;
1580         }
1581         RETURN(rc);
1582 }
1583
1584 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1585 {
1586         ofd_fmd_expire(exp);
1587         return 0;
1588 }
1589
1590 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1591 {
1592         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1593         struct ofd_thread_info  *info;
1594         struct lu_env            env;
1595 #ifdef USE_HEALTH_CHECK_WRITE
1596         struct thandle          *th;
1597 #endif
1598         int                      rc = 0;
1599
1600         /* obd_proc_read_health pass NULL env, we need real one */
1601         rc = lu_env_init(&env, LCT_DT_THREAD);
1602         if (rc)
1603                 RETURN(rc);
1604
1605         info = ofd_info_init(&env, NULL);
1606         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1607         if (unlikely(rc))
1608                 GOTO(out, rc);
1609
1610         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1611                 GOTO(out, rc = -EROFS);
1612
1613 #ifdef USE_HEALTH_CHECK_WRITE
1614         OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1615         if (info->fti_buf.lb_buf == NULL)
1616                 GOTO(out, rc = -ENOMEM);
1617
1618         info->fti_buf.lb_len = PAGE_CACHE_SIZE;
1619         info->fti_off = 0;
1620
1621         th = dt_trans_create(&env, ofd->ofd_osd);
1622         if (IS_ERR(th))
1623                 GOTO(out, rc = PTR_ERR(th));
1624
1625         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1626                                      info->fti_buf.lb_len, info->fti_off, th);
1627         if (rc == 0) {
1628                 th->th_sync = 1; /* sync IO is needed */
1629                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1630                 if (rc == 0)
1631                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1632                                              &info->fti_buf, &info->fti_off,
1633                                              th);
1634         }
1635         dt_trans_stop(&env, ofd->ofd_osd, th);
1636
1637         OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
1638
1639         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1640 #endif
1641 out:
1642         lu_env_fini(&env);
1643         return !!rc;
1644 }
1645
1646 /*
1647  * Handle quota control requests to consult current usage/limit.
1648  *
1649  * \param obd - is the obd device associated with the ofd
1650  * \param exp - is the client's export
1651  * \param oqctl - is the obd_quotactl request to be processed
1652  */
1653 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1654                         struct obd_quotactl *oqctl)
1655 {
1656         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1657         struct lu_env       env;
1658         int                 rc;
1659         ENTRY;
1660
1661         /* report success for quota on/off for interoperability with current MDT
1662          * stack */
1663         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1664                 RETURN(0);
1665
1666         rc = lu_env_init(&env, LCT_DT_THREAD);
1667         if (rc)
1668                 RETURN(rc);
1669
1670         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1671         lu_env_fini(&env);
1672
1673         RETURN(rc);
1674 }
1675
1676 struct obd_ops ofd_obd_ops = {
1677         .o_owner                = THIS_MODULE,
1678         .o_connect              = ofd_obd_connect,
1679         .o_reconnect            = ofd_obd_reconnect,
1680         .o_disconnect           = ofd_obd_disconnect,
1681         .o_set_info_async       = ofd_set_info_async,
1682         .o_get_info             = ofd_get_info,
1683         .o_create               = ofd_create,
1684         .o_statfs               = ofd_statfs,
1685         .o_setattr              = ofd_setattr,
1686         .o_preprw               = ofd_preprw,
1687         .o_commitrw             = ofd_commitrw,
1688         .o_destroy              = ofd_destroy,
1689         .o_init_export          = ofd_init_export,
1690         .o_destroy_export       = ofd_destroy_export,
1691         .o_postrecov            = ofd_obd_postrecov,
1692         .o_punch                = ofd_punch,
1693         .o_getattr              = ofd_getattr,
1694         .o_sync                 = ofd_sync,
1695         .o_iocontrol            = ofd_iocontrol,
1696         .o_precleanup           = ofd_precleanup,
1697         .o_ping                 = ofd_ping,
1698         .o_health_check         = ofd_health_check,
1699         .o_quotactl             = ofd_quotactl,
1700 };