Whamcloud - gitweb
2f082a2abc830094a4367fc74869be510374c08a
[fs/lustre-release.git] / lustre / ofd / ofd_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2012, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/ofd/ofd_obd.c
37  *
38  * Author: Andreas Dilger <adilger@whamcloud.com>
39  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
40  * Author: Mike Pershin <tappro@whamcloud.com>
41  */
42
43 #define DEBUG_SUBSYSTEM S_FILTER
44
45 #include <lustre/lustre_idl.h>
46 #include "ofd_internal.h"
47 #include <obd_cksum.h>
48 #include <lustre_quota.h>
49
50 static int ofd_export_stats_init(struct ofd_device *ofd,
51                                  struct obd_export *exp, void *client_nid)
52 {
53         struct obd_device       *obd = ofd_obd(ofd);
54         struct nid_stat         *stats;
55         int                      num_stats;
56         int                      rc, newnid = 0;
57
58         ENTRY;
59
60         if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
61                 /* Self-export gets no proc entry */
62                 RETURN(0);
63
64         rc = lprocfs_exp_setup(exp, client_nid, &newnid);
65         if (rc) {
66                 /* Mask error for already created
67                  * /proc entries */
68                 if (rc == -EALREADY)
69                         rc = 0;
70                 RETURN(rc);
71         }
72
73         if (newnid == 0)
74                 RETURN(0);
75
76         stats = exp->exp_nid_stats;
77         LASSERT(stats != NULL);
78
79         num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
80                      LPROC_OFD_LAST - 1;
81
82         stats->nid_stats = lprocfs_alloc_stats(num_stats,
83                                                LPROCFS_STATS_FLAG_NOPERCPU);
84         if (stats->nid_stats == NULL)
85                 return -ENOMEM;
86
87         lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats);
88         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES,
89                              LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
90         lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES,
91                              LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
92
93         rc = lprocfs_register_stats(stats->nid_proc, "stats",
94                                     stats->nid_stats);
95         if (rc)
96                 GOTO(clean, rc);
97
98         rc = lprocfs_nid_ldlm_stats_init(stats);
99         if (rc) {
100                 lprocfs_free_stats(&stats->nid_stats);
101                 GOTO(clean, rc);
102         }
103
104         RETURN(0);
105 clean:
106         return rc;
107 }
108
109 static int ofd_parse_connect_data(const struct lu_env *env,
110                                   struct obd_export *exp,
111                                   struct obd_connect_data *data,
112                                   bool new_connection)
113 {
114         struct ofd_device                *ofd = ofd_exp(exp);
115         struct filter_export_data        *fed = &exp->exp_filter_data;
116
117         if (!data)
118                 RETURN(0);
119
120         CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: "LPX64
121                " ocd_version: %x ocd_grant: %d ocd_index: %u"
122                " ocd_group %u\n",
123                exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
124                data->ocd_connect_flags, data->ocd_version,
125                data->ocd_grant, data->ocd_index, data->ocd_group);
126
127         if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
128                 CWARN("!!! This export (nid %s) used object group %d "
129                       "earlier; now it's trying to use group %d!  This could "
130                       "be a bug in the MDS. Please report to "
131                       "http://bugs.whamcloud.com/\n",
132                       obd_export_nid2str(exp), fed->fed_group,
133                       data->ocd_group);
134                 RETURN(-EPROTO);
135         }
136         fed->fed_group = data->ocd_group;
137
138         data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
139         exp->exp_connect_data = *data;
140         data->ocd_version = LUSTRE_VERSION_CODE;
141
142         /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
143         if (data->ocd_connect_flags & OBD_CONNECT_MDS)
144                 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
145                        exp->exp_obd->obd_name, data->ocd_group);
146         else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
147                 RETURN(-EPROTO);
148
149         if (ofd_grant_param_supp(exp)) {
150                 exp->exp_filter_data.fed_pagesize = data->ocd_blocksize;
151                 /* ocd_{blocksize,inodespace} are log2 values */
152                 data->ocd_blocksize  = ofd->ofd_blockbits;
153                 data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace;
154                 /* ocd_grant_extent is in 1K blocks */
155                 data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
156         }
157
158         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
159                 data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
160                                                     new_connection);
161
162         if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
163                 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
164                 int                    index = lsd->lsd_osd_index;
165
166                 if (index != data->ocd_index) {
167                         LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
168                                            " %u doesn't match actual OST index"
169                                            " %u in last_rcvd file, bad "
170                                            "configuration?\n",
171                                            obd_export_nid2str(exp), index,
172                                            data->ocd_index);
173                         RETURN(-EBADF);
174                 }
175                 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
176                         /* this will only happen on the first connect */
177                         lsd->lsd_feature_compat |= OBD_COMPAT_OST;
178                         /* sync is not needed here as lut_client_add will
179                          * set exp_need_sync flag */
180                         tgt_server_data_update(env, &ofd->ofd_lut, 0);
181                 }
182         }
183         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
184                 data->ocd_brw_size = 65536;
185         } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
186                 data->ocd_brw_size = min(data->ocd_brw_size,
187                                          (__u32)DT_MAX_BRW_SIZE);
188                 if (data->ocd_brw_size == 0) {
189                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
190                                " ocd_version: %x ocd_grant: %d ocd_index: %u "
191                                "ocd_brw_size is unexpectedly zero, "
192                                "network data corruption?"
193                                "Refusing connection of this client\n",
194                                exp->exp_obd->obd_name,
195                                exp->exp_client_uuid.uuid,
196                                exp, data->ocd_connect_flags, data->ocd_version,
197                                data->ocd_grant, data->ocd_index);
198                         RETURN(-EPROTO);
199                 }
200         }
201
202         if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
203                 __u32 cksum_types = data->ocd_cksum_types;
204
205                 /* The client set in ocd_cksum_types the checksum types it
206                  * supports. We have to mask off the algorithms that we don't
207                  * support */
208                 data->ocd_cksum_types &= cksum_types_supported_server();
209
210                 if (unlikely(data->ocd_cksum_types == 0)) {
211                         CERROR("%s: Connect with checksum support but no "
212                                "ocd_cksum_types is set\n",
213                                exp->exp_obd->obd_name);
214                         RETURN(-EPROTO);
215                 }
216
217                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
218                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
219                        cksum_types, data->ocd_cksum_types);
220         } else {
221                 /* This client does not support OBD_CONNECT_CKSUM
222                  * fall back to CRC32 */
223                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
224                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
225                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
226         }
227
228         if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
229                 data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes;
230
231         if (OCD_HAS_FLAG(data, PINGLESS)) {
232                 if (ptlrpc_pinger_suppress_pings()) {
233                         spin_lock(&exp->exp_obd->obd_dev_lock);
234                         list_del_init(&exp->exp_obd_chain_timed);
235                         spin_unlock(&exp->exp_obd->obd_dev_lock);
236                 } else {
237                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
238                 }
239         }
240
241         RETURN(0);
242 }
243
244 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
245                              struct obd_device *obd, struct obd_uuid *cluuid,
246                              struct obd_connect_data *data, void *localdata)
247 {
248         struct ofd_device       *ofd;
249         int                      rc;
250
251         ENTRY;
252
253         if (exp == NULL || obd == NULL || cluuid == NULL)
254                 RETURN(-EINVAL);
255
256         ofd = ofd_dev(obd->obd_lu_dev);
257
258         rc = lu_env_refill((struct lu_env *)env);
259         if (rc != 0) {
260                 CERROR("Failure to refill session: '%d'\n", rc);
261                 RETURN(rc);
262         }
263
264         ofd_info_init(env, exp);
265         rc = ofd_parse_connect_data(env, exp, data, false);
266         if (rc == 0)
267                 ofd_export_stats_init(ofd, exp, localdata);
268
269         RETURN(rc);
270 }
271
272 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
273                            struct obd_device *obd, struct obd_uuid *cluuid,
274                            struct obd_connect_data *data, void *localdata)
275 {
276         struct obd_export       *exp;
277         struct ofd_device       *ofd;
278         struct lustre_handle     conn = { 0 };
279         int                      rc;
280         ENTRY;
281
282         if (_exp == NULL || obd == NULL || cluuid == NULL)
283                 RETURN(-EINVAL);
284
285         ofd = ofd_dev(obd->obd_lu_dev);
286
287         rc = class_connect(&conn, obd, cluuid);
288         if (rc)
289                 RETURN(rc);
290
291         exp = class_conn2export(&conn);
292         LASSERT(exp != NULL);
293
294         rc = lu_env_refill((struct lu_env *)env);
295         if (rc != 0) {
296                 CERROR("Failure to refill session: '%d'\n", rc);
297                 GOTO(out, rc);
298         }
299
300         ofd_info_init(env, exp);
301
302         rc = ofd_parse_connect_data(env, exp, data, true);
303         if (rc)
304                 GOTO(out, rc);
305
306         if (obd->obd_replayable) {
307                 struct tg_export_data *ted = &exp->exp_target_data;
308
309                 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
310                        sizeof(ted->ted_lcd->lcd_uuid));
311                 rc = tgt_client_new(env, exp);
312                 if (rc != 0)
313                         GOTO(out, rc);
314                 ofd_export_stats_init(ofd, exp, localdata);
315         }
316
317         CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
318                data->ocd_group);
319
320 out:
321         if (rc != 0) {
322                 class_disconnect(exp);
323                 *_exp = NULL;
324         } else {
325                 *_exp = exp;
326         }
327         RETURN(rc);
328 }
329
330 static int ofd_obd_disconnect(struct obd_export *exp)
331 {
332         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
333         struct lu_env            env;
334         int                      rc;
335
336         ENTRY;
337
338         LASSERT(exp);
339         class_export_get(exp);
340
341         if (!(exp->exp_flags & OBD_OPT_FORCE))
342                 ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__);
343
344         rc = server_disconnect_export(exp);
345
346         ofd_grant_discard(exp);
347
348         rc = lu_env_init(&env, LCT_DT_THREAD);
349         if (rc)
350                 RETURN(rc);
351
352         /* Do not erase record for recoverable client. */
353         if (exp->exp_obd->obd_replayable &&
354             (!exp->exp_obd->obd_fail || exp->exp_failed))
355                 tgt_client_del(&env, exp);
356         lu_env_fini(&env);
357
358         class_export_put(exp);
359         RETURN(rc);
360 }
361
362 static int ofd_init_export(struct obd_export *exp)
363 {
364         int rc;
365
366         spin_lock_init(&exp->exp_filter_data.fed_lock);
367         CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
368         spin_lock(&exp->exp_lock);
369         exp->exp_connecting = 1;
370         spin_unlock(&exp->exp_lock);
371
372         /* self-export doesn't need client data and ldlm initialization */
373         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
374                                      &exp->exp_client_uuid)))
375                 return 0;
376
377         rc = tgt_client_alloc(exp);
378         if (rc == 0)
379                 ldlm_init_export(exp);
380         if (rc)
381                 CERROR("%s: Can't initialize export: rc %d\n",
382                        exp->exp_obd->obd_name, rc);
383         return rc;
384 }
385
386 static int ofd_destroy_export(struct obd_export *exp)
387 {
388         struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
389
390         if (exp->exp_filter_data.fed_pending)
391                 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
392                        "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
393                        exp, exp->exp_filter_data.fed_pending);
394
395         target_destroy_export(exp);
396
397         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
398                                      &exp->exp_client_uuid)))
399                 return 0;
400
401         ldlm_destroy_export(exp);
402         tgt_client_free(exp);
403
404         ofd_fmd_cleanup(exp);
405
406         /*
407          * discard grants once we're sure no more
408          * interaction with the client is possible
409          */
410         ofd_grant_discard(exp);
411         ofd_fmd_cleanup(exp);
412
413         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) {
414                 if (ofd->ofd_tot_granted_clients > 0)
415                         ofd->ofd_tot_granted_clients --;
416         }
417
418         if (!(exp->exp_flags & OBD_OPT_FORCE))
419                 ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__);
420
421         LASSERT(cfs_list_empty(&exp->exp_filter_data.fed_mod_list));
422         return 0;
423 }
424
425 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
426 {
427         struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
428
429         CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name);
430         return ldev->ld_ops->ldo_recovery_complete(env, ldev);
431 }
432
433 int ofd_obd_postrecov(struct obd_device *obd)
434 {
435         struct lu_env            env;
436         struct lu_device        *ldev = obd->obd_lu_dev;
437         int                      rc;
438
439         ENTRY;
440
441         rc = lu_env_init(&env, LCT_DT_THREAD);
442         if (rc)
443                 RETURN(rc);
444         ofd_info_init(&env, obd->obd_self_export);
445
446         rc = ofd_postrecov(&env, ofd_dev(ldev));
447
448         lu_env_fini(&env);
449         RETURN(rc);
450 }
451
452 static int ofd_adapt_sptlrpc_conf(const struct lu_env *env,
453                                   struct obd_device *obd, int initial)
454 {
455         struct filter_obd       *fo = &obd->u.filter;
456         struct sptlrpc_rule_set  tmp_rset;
457         int                      rc;
458
459         sptlrpc_rule_set_init(&tmp_rset);
460         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
461         if (rc) {
462                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
463                        obd->obd_name, rc);
464                 return rc;
465         }
466
467         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
468
469         write_lock(&fo->fo_sptlrpc_lock);
470         sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset);
471         fo->fo_sptlrpc_rset = tmp_rset;
472         write_unlock(&fo->fo_sptlrpc_lock);
473
474         return 0;
475 }
476
477 static int ofd_set_mds_conn(struct obd_export *exp, void *val)
478 {
479         int rc = 0;
480
481         ENTRY;
482
483         LCONSOLE_WARN("%s: received MDS connection from %s\n",
484                       exp->exp_obd->obd_name, obd_export_nid2str(exp));
485         RETURN(rc);
486 }
487
488 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
489                               __u32 keylen, void *key, __u32 vallen, void *val,
490                               struct ptlrpc_request_set *set)
491 {
492         struct ofd_device       *ofd;
493         int                      rc = 0;
494
495         ENTRY;
496
497         if (exp->exp_obd == NULL) {
498                 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
499                 RETURN(-EINVAL);
500         }
501
502         ofd = ofd_exp(exp);
503
504         if (KEY_IS(KEY_CAPA_KEY)) {
505                 rc = ofd_update_capa_key(ofd, val);
506                 if (rc)
507                         CERROR("%s: update capability key failed: rc = %d\n",
508                                exp->exp_obd->obd_name, rc);
509         } else if (KEY_IS(KEY_SPTLRPC_CONF)) {
510                 ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0);
511         } else if (KEY_IS(KEY_MDS_CONN)) {
512                 rc = ofd_set_mds_conn(exp, val);
513         } else if (KEY_IS(KEY_GRANT_SHRINK)) {
514                 struct ost_body *body = val;
515
516                 ofd_info_init(env, exp);
517                 /** handle grant shrink, similar to a read request */
518                 ofd_grant_prepare_read(env, exp, &body->oa);
519         } else {
520                 CERROR("%s: Unsupported key %s\n",
521                        exp->exp_obd->obd_name, (char*)key);
522                 rc = -EOPNOTSUPP;
523         }
524
525         RETURN(rc);
526 }
527
528 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
529                         __u32 keylen, void *key, __u32 *vallen, void *val,
530                         struct lov_stripe_md *lsm)
531 {
532         struct ofd_device       *ofd;
533         int                      rc = 0;
534
535         ENTRY;
536
537         if (exp->exp_obd == NULL) {
538                 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
539                 RETURN(-EINVAL);
540         }
541
542         ofd = ofd_exp(exp);
543
544         if (KEY_IS(KEY_BLOCKSIZE)) {
545                 __u32 *blocksize = val;
546                 if (blocksize) {
547                         if (*vallen < sizeof(*blocksize))
548                                 RETURN(-EOVERFLOW);
549                         *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
550                 }
551                 *vallen = sizeof(*blocksize);
552         } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) {
553                 __u32 *blocksize_bits = val;
554                 if (blocksize_bits) {
555                         if (*vallen < sizeof(*blocksize_bits))
556                                 RETURN(-EOVERFLOW);
557                         *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
558                 }
559                 *vallen = sizeof(*blocksize_bits);
560         } else if (KEY_IS(KEY_LAST_ID)) {
561                 obd_id *last_id = val;
562                 struct ofd_seq *oseq;
563
564                 if (val == NULL) {
565                         *vallen = sizeof(obd_id);
566                         RETURN(0);
567                 }
568                 ofd_info_init(env, exp);
569                 oseq = ofd_seq_load(env, ofd,
570                                     (obd_seq)exp->exp_filter_data.fed_group);
571                 LASSERT(!IS_ERR(oseq));
572                 if (last_id) {
573                         if (*vallen < sizeof(*last_id)) {
574                                 ofd_seq_put(env, oseq);
575                                 RETURN(-EOVERFLOW);
576                         }
577                         *last_id = ofd_seq_last_oid(oseq);
578                 }
579                 ofd_seq_put(env, oseq);
580                 *vallen = sizeof(*last_id);
581         } else if (KEY_IS(KEY_FIEMAP)) {
582                 struct ofd_thread_info          *info;
583                 struct ofd_device               *ofd = ofd_exp(exp);
584                 struct ofd_object               *fo;
585                 struct ll_fiemap_info_key       *fm_key = key;
586
587                 if (val == NULL) {
588                         *vallen = fiemap_count_to_size(
589                                                fm_key->fiemap.fm_extent_count);
590                         RETURN(0);
591                 }
592
593                 info = ofd_info_init(env, exp);
594                 rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0);
595                 if (rc != 0)
596                         RETURN(rc);
597                 CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
598                        PFID(&info->fti_fid));
599
600                 fo = ofd_object_find(env, ofd, &info->fti_fid);
601                 if (IS_ERR(fo)) {
602                         CERROR("%s: error finding object "DFID"\n",
603                                exp->exp_obd->obd_name, PFID(&info->fti_fid));
604                         rc = PTR_ERR(fo);
605                 } else {
606                         struct ll_user_fiemap *fiemap = val;
607
608                         ofd_read_lock(env, fo);
609                         if (ofd_object_exists(fo)) {
610                                 *fiemap = fm_key->fiemap;
611                                 rc = dt_fiemap_get(env,
612                                                    ofd_object_child(fo),
613                                                    fiemap);
614                         } else {
615                                 rc = -ENOENT;
616                         }
617                         ofd_read_unlock(env, fo);
618                         ofd_object_put(env, fo);
619                 }
620         } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) {
621                 *((__u32 *) val) = ofd->ofd_sync_lock_cancel;
622                 *vallen = sizeof(__u32);
623         } else if (KEY_IS(KEY_LAST_FID)) {
624                 struct lu_env           env;
625                 struct ofd_device       *ofd = ofd_exp(exp);
626                 struct ofd_seq          *oseq;
627                 struct lu_fid           *fid = val;
628                 int                     rc;
629
630                 if (fid == NULL) {
631                         *vallen = sizeof(struct lu_fid);
632                         RETURN(0);
633                 }
634
635                 if (*vallen < sizeof(*fid))
636                         RETURN(-EOVERFLOW);
637
638                 rc = lu_env_init(&env, LCT_DT_THREAD);
639                 if (rc != 0)
640                         RETURN(rc);
641                 ofd_info_init(&env, exp);
642
643                 fid_le_to_cpu(fid, fid);
644
645                 oseq = ofd_seq_load(&env, ofd, ostid_seq((struct ost_id *)fid));
646                 if (IS_ERR(oseq))
647                         GOTO(out_fini, rc = PTR_ERR(oseq));
648
649                 rc = ostid_to_fid(fid, &oseq->os_oi,
650                              ofd->ofd_lut.lut_lsd.lsd_osd_index);
651                 if (rc != 0)
652                         GOTO(out_put, rc);
653
654                 CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd),
655                        PFID(fid));
656                 *vallen = sizeof(*fid);
657 out_put:
658                 ofd_seq_put(&env, oseq);
659 out_fini:
660                 lu_env_fini(&env);
661         } else {
662                 CERROR("Not supported key %s\n", (char*)key);
663                 rc = -EOPNOTSUPP;
664         }
665
666         RETURN(rc);
667 }
668
669 /** helper function for statfs, also used by grant code */
670 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
671                         struct obd_statfs *osfs, __u64 max_age, int *from_cache)
672 {
673         int rc;
674
675         spin_lock(&ofd->ofd_osfs_lock);
676         if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) {
677                 obd_size unstable;
678
679                 /* statfs data are too old, get up-to-date one.
680                  * we must be cautious here since multiple threads might be
681                  * willing to update statfs data concurrently and we must
682                  * grant that cached statfs data are always consistent */
683
684                 if (ofd->ofd_statfs_inflight == 0)
685                         /* clear inflight counter if no users, although it would
686                          * take a while to overflow this 64-bit counter ... */
687                         ofd->ofd_osfs_inflight = 0;
688                 /* notify ofd_grant_commit() that we want to track writes
689                  * completed as of now */
690                 ofd->ofd_statfs_inflight++;
691                 /* record value of inflight counter before running statfs to
692                  * compute the diff once statfs is completed */
693                 unstable = ofd->ofd_osfs_inflight;
694                 spin_unlock(&ofd->ofd_osfs_lock);
695
696                 /* statfs can sleep ... hopefully not for too long since we can
697                  * call it fairly often as space fills up */
698                 rc = dt_statfs(env, ofd->ofd_osd, osfs);
699                 if (unlikely(rc))
700                         return rc;
701
702                 spin_lock(&ofd->ofd_grant_lock);
703                 spin_lock(&ofd->ofd_osfs_lock);
704                 /* calculate how much space was written while we released the
705                  * ofd_osfs_lock */
706                 unstable = ofd->ofd_osfs_inflight - unstable;
707                 ofd->ofd_osfs_unstable = 0;
708                 if (unstable) {
709                         /* some writes completed while we were running statfs
710                          * w/o the ofd_osfs_lock. Those ones got added to
711                          * the cached statfs data that we are about to crunch.
712                          * Take them into account in the new statfs data */
713                         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
714                                                unstable >> ofd->ofd_blockbits);
715                         /* However, we don't really know if those writes got
716                          * accounted in the statfs call, so tell
717                          * ofd_grant_space_left() there is some uncertainty
718                          * on the accounting of those writes.
719                          * The purpose is to prevent spurious error messages in
720                          * ofd_grant_space_left() since those writes might be
721                          * accounted twice. */
722                         ofd->ofd_osfs_unstable += unstable;
723                 }
724                 /* similarly, there is some uncertainty on write requests
725                  * between prepare & commit */
726                 ofd->ofd_osfs_unstable += ofd->ofd_tot_pending;
727                 spin_unlock(&ofd->ofd_grant_lock);
728
729                 /* finally udpate cached statfs data */
730                 ofd->ofd_osfs = *osfs;
731                 ofd->ofd_osfs_age = cfs_time_current_64();
732
733                 ofd->ofd_statfs_inflight--; /* stop tracking */
734                 if (ofd->ofd_statfs_inflight == 0)
735                         ofd->ofd_osfs_inflight = 0;
736                 spin_unlock(&ofd->ofd_osfs_lock);
737
738                 if (from_cache)
739                         *from_cache = 0;
740         } else {
741                 /* use cached statfs data */
742                 *osfs = ofd->ofd_osfs;
743                 spin_unlock(&ofd->ofd_osfs_lock);
744                 if (from_cache)
745                         *from_cache = 1;
746         }
747         return 0;
748 }
749
750 static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
751                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
752 {
753         struct obd_device       *obd = class_exp2obd(exp);
754         struct ofd_device       *ofd = ofd_dev(exp->exp_obd->obd_lu_dev);
755         int                      rc;
756
757         ENTRY;
758
759         rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL);
760         if (unlikely(rc))
761                 GOTO(out, rc);
762
763         /* at least try to account for cached pages.  its still racy and
764          * might be under-reporting if clients haven't announced their
765          * caches with brw recently */
766
767         CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
768                " pending "LPU64" free "LPU64" avail "LPU64"\n",
769                ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending,
770                osfs->os_bfree << ofd->ofd_blockbits,
771                osfs->os_bavail << ofd->ofd_blockbits);
772
773         osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
774                                  ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending +
775                                    osfs->os_bsize - 1) >> ofd->ofd_blockbits));
776
777         /* The QoS code on the MDS does not care about space reserved for
778          * precreate, so take it out. */
779         if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
780                 struct filter_export_data *fed;
781
782                 fed = &obd->obd_self_export->exp_filter_data;
783                 osfs->os_bavail -= min_t(obd_size, osfs->os_bavail,
784                                          fed->fed_grant >> ofd->ofd_blockbits);
785         }
786
787         ofd_grant_sanity_check(obd, __FUNCTION__);
788         CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; "
789                LPU64" objects: "LPU64" free; state %x\n",
790                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
791                osfs->os_files, osfs->os_ffree, osfs->os_state);
792
793         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
794                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
795                 osfs->os_ffree = 0;
796
797         /* OS_STATE_READONLY can be set by OSD already */
798         if (ofd->ofd_raid_degraded)
799                 osfs->os_state |= OS_STATE_DEGRADED;
800
801         if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) {
802                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
803                  * should not see a block size > page size, otherwise
804                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
805                  * block size which is the biggest block size known to work
806                  * with all client's page size. */
807                 osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
808                 osfs->os_bfree  <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
809                 osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT;
810                 osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
811         }
812
813         if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
814                                  ofd->ofd_lut.lut_lsd.lsd_osd_index))
815                 osfs->os_bfree = osfs->os_bavail = 2;
816
817         EXIT;
818 out:
819         return rc;
820 }
821
822 int ofd_setattr(const struct lu_env *env, struct obd_export *exp,
823                 struct obd_info *oinfo, struct obd_trans_info *oti)
824 {
825         struct ofd_thread_info  *info;
826         struct ofd_device       *ofd = ofd_exp(exp);
827         struct ldlm_namespace   *ns = ofd->ofd_namespace;
828         struct ldlm_resource    *res;
829         struct ofd_object       *fo;
830         struct obdo             *oa = oinfo->oi_oa;
831         struct filter_fid       *ff = NULL;
832         int                      rc = 0;
833
834         ENTRY;
835
836         info = ofd_info_init(env, exp);
837         ofd_oti2info(info, oti);
838
839         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
840         if (rc != 0)
841                 RETURN(rc);
842
843         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
844         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi),
845                            oinfo_capa(oinfo), CAPA_OPC_META_WRITE);
846         if (rc)
847                 GOTO(out, rc);
848
849         /* This would be very bad - accidentally truncating a file when
850          * changing the time or similar - bug 12203. */
851         if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE &&
852             oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) {
853                 static char mdsinum[48];
854
855                 if (oinfo->oi_oa->o_valid & OBD_MD_FLFID)
856                         snprintf(mdsinum, sizeof(mdsinum) - 1,
857                                  "of parent "DFID, oinfo->oi_oa->o_parent_seq,
858                                  oinfo->oi_oa->o_parent_oid, 0);
859                 else
860                         mdsinum[0] = '\0';
861
862                 CERROR("%s: setattr from %s trying to truncate object "DFID
863                        " %s\n", exp->exp_obd->obd_name,
864                        obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum);
865                 GOTO(out, rc = -EPERM);
866         }
867
868         fo = ofd_object_find(env, ofd, &info->fti_fid);
869         if (IS_ERR(fo)) {
870                 CERROR("%s: can't find object "DFID"\n",
871                        exp->exp_obd->obd_name, PFID(&info->fti_fid));
872                 GOTO(out, rc = PTR_ERR(fo));
873         }
874
875         la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid);
876         info->fti_attr.la_valid &= ~LA_TYPE;
877
878         if (oa->o_valid & OBD_MD_FLFID) {
879                 ff = &info->fti_mds_fid;
880                 ofd_prepare_fidea(ff, oa);
881         }
882
883         /* setting objects attributes (including owner/group) */
884         rc = ofd_attr_set(env, fo, &info->fti_attr, ff);
885         if (rc)
886                 GOTO(out_unlock, rc);
887
888         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
889         if (res != NULL) {
890                 ldlm_res_lvbo_update(res, NULL, 0);
891                 ldlm_resource_putref(res);
892         }
893
894         oinfo->oi_oa->o_valid = OBD_MD_FLID;
895
896         /* Quota release needs uid/gid info */
897         rc = ofd_attr_get(env, fo, &info->fti_attr);
898         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
899                      OFD_VALID_FLAGS | LA_UID | LA_GID);
900         ofd_info2oti(info, oti);
901
902         ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
903         EXIT;
904 out_unlock:
905         ofd_object_put(env, fo);
906 out:
907         return rc;
908 }
909
910 static int ofd_punch(const struct lu_env *env, struct obd_export *exp,
911                      struct obd_info *oinfo, struct obd_trans_info *oti,
912                      struct ptlrpc_request_set *rqset)
913 {
914         struct ofd_thread_info  *info;
915         struct ofd_device       *ofd = ofd_exp(exp);
916         struct ldlm_namespace   *ns = ofd->ofd_namespace;
917         struct ldlm_resource    *res;
918         struct ofd_object       *fo;
919         struct filter_fid       *ff = NULL;
920         int                      rc = 0;
921
922         ENTRY;
923
924         info = ofd_info_init(env, exp);
925         ofd_oti2info(info, oti);
926
927         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
928         if (rc != 0)
929                 RETURN(rc);
930         ost_fid_build_resid(&info->fti_fid, &info->fti_resid);
931
932         CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64
933                ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid),
934                oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start,
935                oinfo->oi_policy.l_extent.end);
936
937         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
938                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
939         if (rc)
940                 GOTO(out_env, rc);
941
942         fo = ofd_object_find(env, ofd, &info->fti_fid);
943         if (IS_ERR(fo)) {
944                 CERROR("%s: error finding object "DFID": rc = %ld\n",
945                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
946                        PTR_ERR(fo));
947                 GOTO(out_env, rc = PTR_ERR(fo));
948         }
949
950         LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF);
951         if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) {
952                 /* Truncate case */
953                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
954         } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) {
955                 oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end;
956         }
957
958         la_from_obdo(&info->fti_attr, oinfo->oi_oa,
959                      OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
960         info->fti_attr.la_valid &= ~LA_TYPE;
961         info->fti_attr.la_size = oinfo->oi_policy.l_extent.start;
962         info->fti_attr.la_valid |= LA_SIZE;
963
964         if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) {
965                 ff = &info->fti_mds_fid;
966                 ofd_prepare_fidea(ff, oinfo->oi_oa);
967         }
968
969         rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start,
970                               oinfo->oi_policy.l_extent.end, &info->fti_attr,
971                               ff);
972         if (rc)
973                 GOTO(out, rc);
974
975         res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
976         if (res != NULL) {
977                 ldlm_res_lvbo_update(res, NULL, 0);
978                 ldlm_resource_putref(res);
979         }
980
981         oinfo->oi_oa->o_valid = OBD_MD_FLID;
982         /* Quota release needs uid/gid info */
983         rc = ofd_attr_get(env, fo, &info->fti_attr);
984         obdo_from_la(oinfo->oi_oa, &info->fti_attr,
985                      OFD_VALID_FLAGS | LA_UID | LA_GID);
986         ofd_info2oti(info, oti);
987
988         ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1);
989         EXIT;
990 out:
991         ofd_object_put(env, fo);
992 out_env:
993         return rc;
994 }
995
996 static int ofd_destroy_by_fid(const struct lu_env *env,
997                               struct ofd_device *ofd,
998                               const struct lu_fid *fid, int orphan)
999 {
1000         struct ofd_thread_info  *info = ofd_info(env);
1001         struct lustre_handle     lockh;
1002         __u64                    flags = LDLM_AST_DISCARD_DATA, rc = 0;
1003         ldlm_policy_data_t       policy = {
1004                                         .l_extent = { 0, OBD_OBJECT_EOF }
1005                                  };
1006         struct ofd_object       *fo;
1007
1008         ENTRY;
1009
1010         fo = ofd_object_find(env, ofd, fid);
1011         if (IS_ERR(fo))
1012                 RETURN(PTR_ERR(fo));
1013
1014         /* Tell the clients that the object is gone now and that they should
1015          * throw away any cached pages. */
1016         ost_fid_build_resid(fid, &info->fti_resid);
1017         rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1018                                     LDLM_EXTENT, &policy, LCK_PW, &flags,
1019                                     ldlm_blocking_ast, ldlm_completion_ast,
1020                                     NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1021
1022         /* We only care about the side-effects, just drop the lock. */
1023         if (rc == ELDLM_OK)
1024                 ldlm_lock_decref(&lockh, LCK_PW);
1025
1026         LASSERT(fo != NULL);
1027
1028         rc = ofd_object_destroy(env, fo, orphan);
1029
1030         ofd_object_put(env, fo);
1031         RETURN(rc);
1032 }
1033
1034 int ofd_destroy(const struct lu_env *env, struct obd_export *exp,
1035                 struct obdo *oa, struct lov_stripe_md *md,
1036                 struct obd_trans_info *oti, struct obd_export *md_exp,
1037                 void *capa)
1038 {
1039         struct ofd_device       *ofd = ofd_exp(exp);
1040         struct ofd_thread_info  *info;
1041         obd_count                count;
1042         int                      rc = 0;
1043
1044         ENTRY;
1045
1046         info = ofd_info_init(env, exp);
1047         ofd_oti2info(info, oti);
1048
1049         if (!(oa->o_valid & OBD_MD_FLGROUP))
1050                 ostid_set_seq_mdt0(&oa->o_oi);
1051
1052         /* check that o_misc makes sense */
1053         if (oa->o_valid & OBD_MD_FLOBJCOUNT)
1054                 count = oa->o_misc;
1055         else
1056                 count = 1; /* default case - single destroy */
1057
1058         /**
1059          * There can be sequence of objects to destroy. Therefore this request
1060          * may have multiple transaction involved in. It is OK, we need only
1061          * the highest used transno to be reported back in reply but not for
1062          * replays, they must report their transno
1063          */
1064         if (info->fti_transno == 0) /* not replay */
1065                 info->fti_mult_trans = 1;
1066
1067         CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd),
1068                POSTID(&oa->o_oi), count);
1069         while (count > 0) {
1070                 int lrc;
1071
1072                 lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0);
1073                 if (lrc != 0) {
1074                         if (rc == 0)
1075                                 rc = lrc;
1076                         GOTO(out, rc);
1077                 }
1078                 lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0);
1079                 if (lrc == -ENOENT) {
1080                         CDEBUG(D_INODE,
1081                                "%s: destroying non-existent object "DFID"\n",
1082                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid));
1083                         /* rewrite rc with -ENOENT only if it is 0 */
1084                         if (rc == 0)
1085                                 rc = lrc;
1086                 } else if (lrc != 0) {
1087                         CERROR("%s: error destroying object "DFID": %d\n",
1088                                ofd_obd(ofd)->obd_name, PFID(&info->fti_fid),
1089                                rc);
1090                         rc = lrc;
1091                 }
1092                 count--;
1093                 ostid_inc_id(&oa->o_oi);
1094         }
1095
1096         /* if we have transaction then there were some deletions, we don't
1097          * need to return ENOENT in that case because it will not wait
1098          * for commit of these deletions. The ENOENT must be returned only
1099          * if there were no transations.
1100          */
1101         if (rc == -ENOENT) {
1102                 if (info->fti_transno != 0)
1103                         rc = 0;
1104         } else if (rc != 0) {
1105                 /*
1106                  * If we have at least one transaction then llog record
1107                  * on server will be removed upon commit, so for rc != 0
1108                  * we return no transno and llog record will be reprocessed.
1109                  */
1110                 info->fti_transno = 0;
1111         }
1112         ofd_info2oti(info, oti);
1113 out:
1114         RETURN(rc);
1115 }
1116
1117 static int ofd_orphans_destroy(const struct lu_env *env,
1118                                struct obd_export *exp, struct ofd_device *ofd,
1119                                struct obdo *oa)
1120 {
1121         struct ofd_thread_info  *info = ofd_info(env);
1122         obd_id                   last;
1123         int                      skip_orphan;
1124         int                      rc = 0;
1125         struct ost_id            oi = oa->o_oi;
1126         __u64                    end_id = ostid_id(&oa->o_oi);
1127         struct ofd_seq          *oseq;
1128
1129         ENTRY;
1130
1131         oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi));
1132         if (oseq == NULL) {
1133                 CERROR("%s: Can not find seq for "DOSTID"\n",
1134                        ofd_name(ofd), POSTID(&oa->o_oi));
1135                 RETURN(-EINVAL);
1136         }
1137
1138         LASSERT(exp != NULL);
1139         skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
1140
1141         last = ofd_seq_last_oid(oseq);
1142         LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
1143                       " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
1144                       end_id + 1, last);
1145
1146         for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
1147                           ostid_dec_id(&oi)) {
1148                 rc = ostid_to_fid(&info->fti_fid, &oi, 0);
1149                 if (rc != 0)
1150                         GOTO(out_put, rc);
1151                 rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
1152                 if (rc && rc != -ENOENT) /* this is pretty fatal... */
1153                         CEMERG("error destroying precreated id "DOSTID": %d\n",
1154                                POSTID(&oi), rc);
1155                 if (!skip_orphan) {
1156                         ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
1157                         /* update last_id on disk periodically so that if we
1158                          * restart * we don't need to re-scan all of the just
1159                          * deleted objects. */
1160                         if ((ostid_id(&oi) & 511) == 0)
1161                                 ofd_seq_last_oid_write(env, ofd, oseq);
1162                 }
1163         }
1164         CDEBUG(D_HA, "%s: after destroy: set last_objids"DOSTID"\n",
1165                ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
1166         if (!skip_orphan) {
1167                 rc = ofd_seq_last_oid_write(env, ofd, oseq);
1168         } else {
1169                 /* don't reuse orphan object, return last used objid */
1170                 ostid_set_id(&oa->o_oi, last);
1171                 rc = 0;
1172         }
1173 out_put:
1174         ofd_seq_put(env, oseq);
1175         RETURN(rc);
1176 }
1177
1178 int ofd_create(const struct lu_env *env, struct obd_export *exp,
1179                struct obdo *oa, struct lov_stripe_md **ea,
1180                struct obd_trans_info *oti)
1181 {
1182         struct ofd_device       *ofd = ofd_exp(exp);
1183         struct ofd_thread_info  *info;
1184         obd_seq                 seq = ostid_seq(&oa->o_oi);
1185         struct ofd_seq          *oseq;
1186         int                     rc = 0, diff;
1187         int                     sync_trans = 0;
1188
1189         ENTRY;
1190
1191         info = ofd_info_init(env, exp);
1192         ofd_oti2info(info, oti);
1193
1194         LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0);
1195         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1196
1197         CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1198
1199         oseq = ofd_seq_load(env, ofd, seq);
1200         if (IS_ERR(oseq)) {
1201                 CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n",
1202                        ofd_name(ofd), seq, PTR_ERR(oseq));
1203                 RETURN(-EINVAL);
1204         }
1205
1206         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1207             (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1208                 if (!ofd_obd(ofd)->obd_recovering ||
1209                     ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) {
1210                         CERROR("recreate objid "DOSTID" > last id "LPU64"\n",
1211                                POSTID(&oa->o_oi), ofd_seq_last_oid(oseq));
1212                         GOTO(out_nolock, rc = -EINVAL);
1213                 }
1214                 /* do nothing because we create objects during first write */
1215                 GOTO(out_nolock, rc = 0);
1216         }
1217         /* former ofd_handle_precreate */
1218         if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1219             (oa->o_flags & OBD_FL_DELORPHAN)) {
1220                 /* destroy orphans */
1221                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1222                         CERROR("%s: dropping old orphan cleanup request\n",
1223                                ofd_name(ofd));
1224                         GOTO(out_nolock, rc = 0);
1225                 }
1226                 /* This causes inflight precreates to abort and drop lock */
1227                 oseq->os_destroys_in_progress = 1;
1228                 mutex_lock(&oseq->os_create_lock);
1229                 if (!oseq->os_destroys_in_progress) {
1230                         CERROR("%s:["LPU64"] destroys_in_progress already"
1231                                " cleared\n", exp->exp_obd->obd_name,
1232                                ostid_seq(&oa->o_oi));
1233                         ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1234                         GOTO(out, rc = 0);
1235                 }
1236                 diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1237                 CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
1238                         ofd_seq_last_oid(oseq), diff);
1239                 if (-diff > OST_MAX_PRECREATE) {
1240                         /* FIXME: should reset precreate_next_id on MDS */
1241                         rc = 0;
1242                 } else if (diff < 0) {
1243                         rc = ofd_orphans_destroy(env, exp, ofd, oa);
1244                         oseq->os_destroys_in_progress = 0;
1245                 } else {
1246                         /* XXX: Used by MDS for the first time! */
1247                         oseq->os_destroys_in_progress = 0;
1248                 }
1249         } else {
1250                 mutex_lock(&oseq->os_create_lock);
1251                 if (oti->oti_conn_cnt < exp->exp_conn_cnt) {
1252                         CERROR("%s: dropping old precreate request\n",
1253                                 ofd_obd(ofd)->obd_name);
1254                         GOTO(out, rc = 0);
1255                 }
1256                 /* only precreate if seq is 0, IDIF or normal and also o_id
1257                  * must be specfied */
1258                 if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) &&
1259                      !fid_seq_is_norm(ostid_seq(&oa->o_oi)) &&
1260                      !fid_seq_is_idif(ostid_seq(&oa->o_oi))) ||
1261                                 ostid_id(&oa->o_oi) == 0) {
1262                         diff = 1; /* shouldn't we create this right now? */
1263                 } else {
1264                         diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq);
1265                         /* Do sync create if the seq is about to used up */
1266                         if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
1267                             fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) {
1268                                 if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1))
1269                                         sync_trans = 1;
1270                         } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) {
1271                                 if (unlikely(ostid_id(&oa->o_oi) >=
1272                                              LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
1273                                         sync_trans = 1;
1274                         } else {
1275                                 CERROR("%s : invalid o_seq "DOSTID": rc = %d\n",
1276                                        ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL);
1277                                 GOTO(out, rc = -EINVAL);
1278                         }
1279                 }
1280         }
1281         if (diff > 0) {
1282                 cfs_time_t       enough_time = cfs_time_shift(DISK_TIMEOUT);
1283                 obd_id           next_id;
1284                 int              created = 0;
1285                 int              count;
1286
1287                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1288                     !(oa->o_flags & OBD_FL_DELORPHAN)) {
1289                         /* don't enforce grant during orphan recovery */
1290                         rc = ofd_grant_create(env,
1291                                               ofd_obd(ofd)->obd_self_export,
1292                                               &diff);
1293                         if (rc) {
1294                                 CDEBUG(D_HA, "%s: failed to acquire grant "
1295                                        "space for precreate (%d): rc = %d\n",
1296                                        ofd_name(ofd), diff, rc);
1297                                 diff = 0;
1298                         }
1299                 }
1300
1301                 while (diff > 0) {
1302                         next_id = ofd_seq_last_oid(oseq) + 1;
1303                         count = ofd_precreate_batch(ofd, diff);
1304
1305                         CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64
1306                                " at "LPU64"\n", ofd_obd(ofd)->obd_name,
1307                                count, ostid_seq(&oa->o_oi), next_id);
1308
1309                         if (cfs_time_after(jiffies, enough_time)) {
1310                                 LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
1311                                               " created at a rate of %d/s\n",
1312                                               ofd_obd(ofd)->obd_name,
1313                                               created, diff + created,
1314                                               created / DISK_TIMEOUT);
1315                                 break;
1316                         }
1317
1318                         rc = ofd_precreate_objects(env, ofd, next_id,
1319                                                    oseq, count, sync_trans);
1320                         if (rc > 0) {
1321                                 created += rc;
1322                                 diff -= rc;
1323                         } else if (rc < 0) {
1324                                 break;
1325                         }
1326                 }
1327                 if (created > 0)
1328                         /* some objects got created, we can return
1329                          * them, even if last creation failed */
1330                         rc = 0;
1331                 else
1332                         CERROR("%s: unable to precreate: rc = %d\n",
1333                                ofd_name(ofd), rc);
1334
1335                 ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1336                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1337
1338                 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
1339                     !(oa->o_flags & OBD_FL_DELORPHAN))
1340                         ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export,
1341                                          rc);
1342         }
1343
1344         ofd_info2oti(info, oti);
1345 out:
1346         mutex_unlock(&oseq->os_create_lock);
1347 out_nolock:
1348         if (rc == 0 && ea != NULL) {
1349                 struct lov_stripe_md *lsm = *ea;
1350
1351                 lsm->lsm_oi = oa->o_oi;
1352         }
1353         ofd_seq_put(env, oseq);
1354         RETURN(rc);
1355 }
1356
1357 int ofd_getattr(const struct lu_env *env, struct obd_export *exp,
1358                 struct obd_info *oinfo)
1359 {
1360         struct ofd_device       *ofd = ofd_exp(exp);
1361         struct ofd_thread_info  *info;
1362         struct ofd_object       *fo;
1363         __u64                    curr_version;
1364         int                      rc = 0;
1365
1366         ENTRY;
1367
1368         info = ofd_info_init(env, exp);
1369
1370         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1371         if (rc != 0)
1372                 GOTO(out, rc);
1373         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1374                            oinfo_capa(oinfo), CAPA_OPC_META_READ);
1375         if (rc)
1376                 GOTO(out, rc);
1377
1378         fo = ofd_object_find(env, ofd, &info->fti_fid);
1379         if (IS_ERR(fo))
1380                 GOTO(out, rc = PTR_ERR(fo));
1381         LASSERT(fo != NULL);
1382         rc = ofd_attr_get(env, fo, &info->fti_attr);
1383         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1384         if (rc == 0)
1385                 obdo_from_la(oinfo->oi_oa, &info->fti_attr,
1386                              OFD_VALID_FLAGS | LA_UID | LA_GID);
1387
1388         /* Store object version in reply */
1389         curr_version = dt_version_get(env, ofd_object_child(fo));
1390         if ((__s64)curr_version != -EOPNOTSUPP) {
1391                 oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
1392                 oinfo->oi_oa->o_data_version = curr_version;
1393         }
1394         ofd_object_put(env, fo);
1395 out:
1396         RETURN(rc);
1397 }
1398
1399 static int ofd_sync(const struct lu_env *env, struct obd_export *exp,
1400                     struct obd_info *oinfo, obd_size start, obd_size end,
1401                     struct ptlrpc_request_set *set)
1402 {
1403         struct ofd_device       *ofd = ofd_exp(exp);
1404         struct ofd_thread_info  *info;
1405         struct ofd_object       *fo;
1406         int                      rc = 0;
1407
1408         ENTRY;
1409
1410         /* if no objid is specified, it means "sync whole filesystem" */
1411         if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
1412                 rc = dt_sync(env, ofd->ofd_osd);
1413                 GOTO(out, rc);
1414         }
1415
1416         info = ofd_info_init(env, exp);
1417         rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0);
1418         if (rc != 0)
1419                 GOTO(out, rc);
1420
1421         rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi),
1422                            oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC);
1423         if (rc)
1424                 GOTO(out, rc);
1425
1426         fo = ofd_object_find(env, ofd, &info->fti_fid);
1427         if (IS_ERR(fo)) {
1428                 CERROR("%s: error finding object "DFID": rc = %ld\n",
1429                        exp->exp_obd->obd_name, PFID(&info->fti_fid),
1430                        PTR_ERR(fo));
1431                 GOTO(out, rc = PTR_ERR(fo));
1432         }
1433
1434         if (!ofd_object_exists(fo))
1435                 GOTO(put, rc = -ENOENT);
1436
1437         if (dt_version_get(env, ofd_object_child(fo)) >
1438             ofd_obd(ofd)->obd_last_committed) {
1439                 rc = dt_object_sync(env, ofd_object_child(fo));
1440                 if (rc)
1441                         GOTO(put, rc);
1442         }
1443
1444         oinfo->oi_oa->o_valid = OBD_MD_FLID;
1445         rc = ofd_attr_get(env, fo, &info->fti_attr);
1446         obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS);
1447
1448         ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1);
1449         EXIT;
1450 put:
1451         ofd_object_put(env, fo);
1452 out:
1453         return rc;
1454 }
1455
1456 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1457                                    struct ofd_device *ofd, void *karg)
1458 {
1459         struct obd_ioctl_data *data = karg;
1460         struct lu_fid          fid;
1461         struct ofd_object     *fo;
1462         dt_obj_version_t       version;
1463         int                    rc = 0;
1464
1465         ENTRY;
1466
1467         if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1468                 GOTO(out, rc = -EINVAL);
1469
1470         if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1471                 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1472         } else if (data->ioc_inlbuf3 != NULL &&
1473                    data->ioc_inllen3 == sizeof(__u64) &&
1474                    data->ioc_inlbuf4 != NULL &&
1475                    data->ioc_inllen4 == sizeof(__u64)) {
1476                 struct ost_id ostid;
1477
1478                 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1479                 ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1480                 rc = ostid_to_fid(&fid, &ostid, 0);
1481                 if (rc != 0)
1482                         GOTO(out, rc);
1483         } else {
1484                 GOTO(out, rc = -EINVAL);
1485         }
1486
1487         if (!fid_is_sane(&fid))
1488                 GOTO(out, rc = -EINVAL);
1489
1490         fo = ofd_object_find(env, ofd, &fid);
1491         if (IS_ERR(fo))
1492                 GOTO(out, rc = PTR_ERR(fo));
1493
1494         if (!ofd_object_exists(fo))
1495                 GOTO(out_fo, rc = -ENOENT);
1496
1497         if (lu_object_remote(&fo->ofo_obj.do_lu))
1498                 GOTO(out_fo, rc = -EREMOTE);
1499
1500         version = dt_version_get(env, ofd_object_child(fo));
1501         if (version == 0)
1502                 GOTO(out_fo, rc = -EIO);
1503
1504         *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1505
1506         EXIT;
1507 out_fo:
1508         ofd_object_put(env, fo);
1509 out:
1510         return rc;
1511 }
1512
1513 int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1514                   void *karg, void *uarg)
1515 {
1516         struct lu_env            env;
1517         struct ofd_device       *ofd = ofd_exp(exp);
1518         struct obd_device       *obd = ofd_obd(ofd);
1519         int                      rc;
1520
1521         ENTRY;
1522
1523         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1524         rc = lu_env_init(&env, LCT_DT_THREAD);
1525         if (rc)
1526                 RETURN(rc);
1527
1528         switch (cmd) {
1529         case OBD_IOC_ABORT_RECOVERY:
1530                 CERROR("%s: aborting recovery\n", obd->obd_name);
1531                 target_stop_recovery_thread(obd);
1532                 break;
1533         case OBD_IOC_SYNC:
1534                 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1535                 rc = dt_sync(&env, ofd->ofd_osd);
1536                 break;
1537         case OBD_IOC_SET_READONLY:
1538                 rc = dt_sync(&env, ofd->ofd_osd);
1539                 if (rc == 0)
1540                         rc = dt_ro(&env, ofd->ofd_osd);
1541                 break;
1542         case OBD_IOC_GET_OBJ_VERSION:
1543                 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1544                 break;
1545         default:
1546                 CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd);
1547                 rc = -ENOTTY;
1548         }
1549
1550         lu_env_fini(&env);
1551         RETURN(rc);
1552 }
1553
1554 static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1555 {
1556         int rc = 0;
1557
1558         ENTRY;
1559
1560         switch(stage) {
1561         case OBD_CLEANUP_EARLY:
1562                 break;
1563         case OBD_CLEANUP_EXPORTS:
1564                 target_cleanup_recovery(obd);
1565                 break;
1566         }
1567         RETURN(rc);
1568 }
1569
1570 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1571 {
1572         ofd_fmd_expire(exp);
1573         return 0;
1574 }
1575
1576 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1577 {
1578         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
1579         struct ofd_thread_info  *info;
1580         struct lu_env            env;
1581 #ifdef USE_HEALTH_CHECK_WRITE
1582         struct thandle          *th;
1583 #endif
1584         int                      rc = 0;
1585
1586         /* obd_proc_read_health pass NULL env, we need real one */
1587         rc = lu_env_init(&env, LCT_DT_THREAD);
1588         if (rc)
1589                 RETURN(rc);
1590
1591         info = ofd_info_init(&env, NULL);
1592         rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1593         if (unlikely(rc))
1594                 GOTO(out, rc);
1595
1596         if (info->fti_u.osfs.os_state == OS_STATE_READONLY)
1597                 GOTO(out, rc = -EROFS);
1598
1599 #ifdef USE_HEALTH_CHECK_WRITE
1600         OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1601         if (info->fti_buf.lb_buf == NULL)
1602                 GOTO(out, rc = -ENOMEM);
1603
1604         info->fti_buf.lb_len = CFS_PAGE_SIZE;
1605         info->fti_off = 0;
1606
1607         th = dt_trans_create(&env, ofd->ofd_osd);
1608         if (IS_ERR(th))
1609                 GOTO(out, rc = PTR_ERR(th));
1610
1611         rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1612                                      info->fti_buf.lb_len, info->fti_off, th);
1613         if (rc == 0) {
1614                 th->th_sync = 1; /* sync IO is needed */
1615                 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1616                 if (rc == 0)
1617                         rc = dt_record_write(&env, ofd->ofd_health_check_file,
1618                                              &info->fti_buf, &info->fti_off,
1619                                              th);
1620         }
1621         dt_trans_stop(&env, ofd->ofd_osd, th);
1622
1623         OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
1624
1625         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1626 #endif
1627 out:
1628         lu_env_fini(&env);
1629         return !!rc;
1630 }
1631
1632 /*
1633  * Handle quota control requests to consult current usage/limit.
1634  *
1635  * \param obd - is the obd device associated with the ofd
1636  * \param exp - is the client's export
1637  * \param oqctl - is the obd_quotactl request to be processed
1638  */
1639 static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp,
1640                         struct obd_quotactl *oqctl)
1641 {
1642         struct ofd_device  *ofd = ofd_dev(obd->obd_lu_dev);
1643         struct lu_env       env;
1644         int                 rc;
1645         ENTRY;
1646
1647         /* report success for quota on/off for interoperability with current MDT
1648          * stack */
1649         if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF)
1650                 RETURN(0);
1651
1652         rc = lu_env_init(&env, LCT_DT_THREAD);
1653         if (rc)
1654                 RETURN(rc);
1655
1656         rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl);
1657         lu_env_fini(&env);
1658
1659         RETURN(rc);
1660 }
1661
1662 struct obd_ops ofd_obd_ops = {
1663         .o_owner                = THIS_MODULE,
1664         .o_connect              = ofd_obd_connect,
1665         .o_reconnect            = ofd_obd_reconnect,
1666         .o_disconnect           = ofd_obd_disconnect,
1667         .o_set_info_async       = ofd_set_info_async,
1668         .o_get_info             = ofd_get_info,
1669         .o_create               = ofd_create,
1670         .o_statfs               = ofd_statfs,
1671         .o_setattr              = ofd_setattr,
1672         .o_preprw               = ofd_preprw,
1673         .o_commitrw             = ofd_commitrw,
1674         .o_destroy              = ofd_destroy,
1675         .o_init_export          = ofd_init_export,
1676         .o_destroy_export       = ofd_destroy_export,
1677         .o_postrecov            = ofd_obd_postrecov,
1678         .o_punch                = ofd_punch,
1679         .o_getattr              = ofd_getattr,
1680         .o_sync                 = ofd_sync,
1681         .o_iocontrol            = ofd_iocontrol,
1682         .o_precleanup           = ofd_precleanup,
1683         .o_ping                 = ofd_ping,
1684         .o_health_check         = ofd_health_check,
1685         .o_quotactl             = ofd_quotactl,
1686 };