Whamcloud - gitweb
ab139b75463717fecd38a58e05d8339de321bd19
[fs/lustre-release.git] / lustre / lod / lod_lov.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,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License version 2 for more details.  A copy is
14  * included in the COPYING 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.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2017, Intel Corporation.
27  */
28 /*
29  * lustre/lod/lod_lov.c
30  *
31  * A set of helpers to maintain Logical Object Volume (LOV)
32  * Extended Attribute (EA) and known OST targets
33  *
34  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35  */
36
37 #define DEBUG_SUBSYSTEM S_MDS
38
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
43
44 #include "lod_internal.h"
45
46 /**
47  * Increase reference count on the target table.
48  *
49  * Increase reference count on the target table usage to prevent racing with
50  * addition/deletion. Any function that expects the table to remain
51  * stationary must take a ref.
52  *
53  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
54  */
55 void lod_getref(struct lod_tgt_descs *ltd)
56 {
57         down_read(&ltd->ltd_rw_sem);
58         mutex_lock(&ltd->ltd_mutex);
59         ltd->ltd_refcount++;
60         mutex_unlock(&ltd->ltd_mutex);
61 }
62
63 /**
64  * Decrease reference count on the target table.
65  *
66  * Companion of lod_getref() to release a reference on the target table.
67  * If this is the last reference and the OST entry was scheduled for deletion,
68  * the descriptor is removed from the table.
69  *
70  * \param[in] lod       LOD device from which we release a reference
71  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
72  */
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
74 {
75         mutex_lock(&ltd->ltd_mutex);
76         ltd->ltd_refcount--;
77         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78                 struct lod_tgt_desc *tgt_desc, *tmp;
79                 LIST_HEAD(kill);
80
81                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
82                        ltd->ltd_death_row);
83
84                 ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
85                         LASSERT(tgt_desc);
86                         if (!tgt_desc->ltd_reap)
87                                 continue;
88
89                         list_add(&tgt_desc->ltd_kill, &kill);
90                         lu_tgt_pool_remove(&ltd->ltd_tgt_pool,
91                                            tgt_desc->ltd_index);
92                         ltd_del_tgt(ltd, tgt_desc);
93                         ltd->ltd_death_row--;
94                 }
95                 mutex_unlock(&ltd->ltd_mutex);
96                 up_read(&ltd->ltd_rw_sem);
97
98                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
99                         int rc;
100
101                         list_del(&tgt_desc->ltd_kill);
102                         rc = obd_disconnect(tgt_desc->ltd_exp);
103                         if (rc)
104                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
105                                        lod2obd(lod)->obd_name,
106                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
107                         OBD_FREE_PTR(tgt_desc);
108                 }
109         } else {
110                 mutex_unlock(&ltd->ltd_mutex);
111                 up_read(&ltd->ltd_rw_sem);
112         }
113 }
114
115 /**
116  * Connect LOD to a new OSP and add it to the target table.
117  *
118  * Connect to the OSP device passed, initialize all the internal
119  * structures related to the device and add it to the target table.
120  *
121  * \param[in] env               execution environment for this thread
122  * \param[in] lod               LOD device to be connected to the new OSP
123  * \param[in] osp               name of OSP device name to be added
124  * \param[in] index             index of the new target
125  * \param[in] gen               target's generation number
126  * \param[in] tgt_index         OSP's group
127  * \param[in] type              type of device (mdc or osc)
128  * \param[in] active            state of OSP: 0 - inactive, 1 - active
129  *
130  * \retval                      0 if added successfully
131  * \retval                      negative error number on failure
132  */
133 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
134                    char *osp, unsigned index, unsigned gen, int tgt_index,
135                    char *type, int active)
136 {
137         struct obd_connect_data *data = NULL;
138         struct obd_export       *exp = NULL;
139         struct obd_device       *obd;
140         struct lu_device        *lu_dev;
141         struct dt_device        *dt_dev;
142         int                      rc;
143         struct lod_tgt_desc     *tgt_desc;
144         struct lod_tgt_descs    *ltd;
145         struct lustre_cfg       *lcfg;
146         struct obd_uuid         obd_uuid;
147         bool                    for_ost;
148         bool connected = false;
149         ENTRY;
150
151         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
152
153         if (gen <= 0) {
154                 CERROR("request to add OBD %s with invalid generation: %d\n",
155                        osp, gen);
156                 RETURN(-EINVAL);
157         }
158
159         obd_str2uuid(&obd_uuid, osp);
160
161         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
162                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
163         if (obd == NULL) {
164                 CERROR("can't find %s device\n", osp);
165                 RETURN(-EINVAL);
166         }
167
168         LASSERT(obd->obd_lu_dev != NULL);
169         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
170
171         lu_dev = obd->obd_lu_dev;
172         dt_dev = lu2dt_dev(lu_dev);
173
174         OBD_ALLOC_PTR(data);
175         if (data == NULL)
176                 GOTO(out_cleanup, rc = -ENOMEM);
177
178         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
179         data->ocd_version = LUSTRE_VERSION_CODE;
180         data->ocd_index = index;
181
182         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
183                 for_ost = true;
184                 data->ocd_connect_flags |= OBD_CONNECT_AT |
185                                            OBD_CONNECT_FULL20 |
186                                            OBD_CONNECT_INDEX |
187 #ifdef HAVE_LRU_RESIZE_SUPPORT
188                                            OBD_CONNECT_LRU_RESIZE |
189 #endif
190                                            OBD_CONNECT_MDS |
191                                            OBD_CONNECT_REQPORTAL |
192                                            OBD_CONNECT_SKIP_ORPHAN |
193                                            OBD_CONNECT_FID |
194                                            OBD_CONNECT_LVB_TYPE |
195                                            OBD_CONNECT_VERSION |
196                                            OBD_CONNECT_PINGLESS |
197                                            OBD_CONNECT_LFSCK |
198                                            OBD_CONNECT_BULK_MBITS;
199
200                 data->ocd_group = tgt_index;
201                 ltd = &lod->lod_ost_descs;
202         } else {
203                 struct obd_import *imp = obd->u.cli.cl_import;
204
205                 for_ost = false;
206                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
207                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
208                                            OBD_CONNECT_IBITS |
209                                            OBD_CONNECT_MDS_MDS |
210                                            OBD_CONNECT_FID |
211                                            OBD_CONNECT_AT |
212                                            OBD_CONNECT_FULL20 |
213                                            OBD_CONNECT_LFSCK |
214                                            OBD_CONNECT_BULK_MBITS;
215                 spin_lock(&imp->imp_lock);
216                 imp->imp_server_timeout = 1;
217                 spin_unlock(&imp->imp_lock);
218                 imp->imp_client->cli_request_portal = OUT_PORTAL;
219                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
220                       obd->obd_name);
221                 ltd = &lod->lod_mdt_descs;
222         }
223
224         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
225         OBD_FREE_PTR(data);
226         if (rc) {
227                 CERROR("%s: cannot connect to next dev %s (%d)\n",
228                        obd->obd_name, osp, rc);
229                 GOTO(out_cleanup, rc);
230         }
231         connected = true;
232
233         /* Allocate ost descriptor and fill it */
234         OBD_ALLOC_PTR(tgt_desc);
235         if (!tgt_desc)
236                 GOTO(out_cleanup, rc = -ENOMEM);
237
238         tgt_desc->ltd_tgt    = dt_dev;
239         tgt_desc->ltd_exp    = exp;
240         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
241         tgt_desc->ltd_gen    = gen;
242         tgt_desc->ltd_index  = index;
243         tgt_desc->ltd_active = active;
244
245         down_write(&ltd->ltd_rw_sem);
246         mutex_lock(&ltd->ltd_mutex);
247         rc = ltd_add_tgt(ltd, tgt_desc);
248         if (rc)
249                 GOTO(out_mutex, rc);
250
251         rc = lu_qos_add_tgt(&ltd->ltd_qos, tgt_desc);
252         if (rc)
253                 GOTO(out_del_tgt, rc);
254
255         rc = lu_tgt_pool_add(&ltd->ltd_tgt_pool, index,
256                           ltd->ltd_lov_desc.ld_tgt_count);
257         if (rc) {
258                 CERROR("%s: can't set up pool, failed with %d\n",
259                        obd->obd_name, rc);
260                 GOTO(out_del_tgt, rc);
261         }
262
263         mutex_unlock(&ltd->ltd_mutex);
264         up_write(&ltd->ltd_rw_sem);
265
266         if (lod->lod_recovery_completed)
267                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
268
269         if (!for_ost && lod->lod_initialized) {
270                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
271                 if (rc != 0) {
272                         CERROR("%s: cannot start llog on %s:rc = %d\n",
273                                lod2obd(lod)->obd_name, osp, rc);
274                         GOTO(out_ltd, rc);
275                 }
276         }
277
278         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
279         if (rc != 0) {
280                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
281                        "index = %u, rc = %d\n", osp, type, index, rc);
282                 GOTO(out_fini_llog, rc);
283         }
284         RETURN(rc);
285 out_fini_llog:
286         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
287                           &tgt_desc->ltd_recovery_task);
288 out_ltd:
289         down_write(&ltd->ltd_rw_sem);
290         mutex_lock(&ltd->ltd_mutex);
291         lu_tgt_pool_remove(&ltd->ltd_tgt_pool, index);
292 out_del_tgt:
293         ltd_del_tgt(ltd, tgt_desc);
294 out_mutex:
295         mutex_unlock(&ltd->ltd_mutex);
296         up_write(&ltd->ltd_rw_sem);
297         OBD_FREE_PTR(tgt_desc);
298 out_cleanup:
299         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
300          * objects from the MDT stack. See LU-7184. */
301         lcfg = &lod_env_info(env)->lti_lustre_cfg;
302         memset(lcfg, 0, sizeof(*lcfg));
303         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
304         lcfg->lcfg_command = LCFG_CLEANUP;
305         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
306
307         if (connected)
308                 obd_disconnect(exp);
309
310         return rc;
311 }
312
313 /**
314  * Schedule target removal from the target table.
315  *
316  * Mark the device as dead. The device is not removed here because it may
317  * still be in use. The device will be removed in lod_putref() when the
318  * last reference is released.
319  *
320  * \param[in] env               execution environment for this thread
321  * \param[in] lod               LOD device the target table belongs to
322  * \param[in] ltd               target table
323  * \param[in] tgt               target
324  */
325 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
326                              struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
327 {
328         lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
329                          !ltd->ltd_is_mdt);
330
331         if (!tgt->ltd_reap) {
332                 tgt->ltd_reap = 1;
333                 ltd->ltd_death_row++;
334         }
335 }
336
337 /**
338  * Schedule removal of all the targets from the given target table.
339  *
340  * See more details in the description for __lod_del_device()
341  *
342  * \param[in] env               execution environment for this thread
343  * \param[in] lod               LOD device the target table belongs to
344  * \param[in] ltd               target table
345  *
346  * \retval                      0 always
347  */
348 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
349                  struct lod_tgt_descs *ltd)
350 {
351         struct lu_tgt_desc *tgt;
352
353         if (ltd->ltd_tgts_size <= 0)
354                 return 0;
355
356         lod_getref(ltd);
357         mutex_lock(&ltd->ltd_mutex);
358         ltd_foreach_tgt(ltd, tgt)
359                 __lod_del_device(env, lod, ltd, tgt);
360         mutex_unlock(&ltd->ltd_mutex);
361         lod_putref(lod, ltd);
362
363         lu_tgt_descs_fini(ltd);
364
365         return 0;
366 }
367
368 /**
369  * Remove device by name.
370  *
371  * Remove a device identified by \a osp from the target table. Given
372  * the device can be in use, the real deletion happens in lod_putref().
373  *
374  * \param[in] env               execution environment for this thread
375  * \param[in] lod               LOD device to be connected to the new OSP
376  * \param[in] ltd               target table
377  * \param[in] osp               name of OSP device to be removed
378  * \param[in] idx               index of the target
379  * \param[in] gen               generation number, not used currently
380  *
381  * \retval                      0 if the device was scheduled for removal
382  * \retval                      -EINVAL if no device was found
383  */
384 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
385                    struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
386                    unsigned int gen)
387 {
388         struct obd_device *obd;
389         struct lu_tgt_desc *tgt;
390         struct obd_uuid uuid;
391         int rc = 0;
392
393         ENTRY;
394
395         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
396
397         obd_str2uuid(&uuid, osp);
398
399         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
400                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
401         if (obd == NULL) {
402                 CERROR("can't find %s device\n", osp);
403                 RETURN(-EINVAL);
404         }
405
406         if (gen <= 0) {
407                 CERROR("%s: request to remove OBD %s with invalid generation %d"
408                        "\n", obd->obd_name, osp, gen);
409                 RETURN(-EINVAL);
410         }
411
412         obd_str2uuid(&uuid,  osp);
413
414         lod_getref(ltd);
415         mutex_lock(&ltd->ltd_mutex);
416         tgt = LTD_TGT(ltd, idx);
417         /* check that the index is allocated in the bitmap */
418         if (!test_bit(idx, ltd->ltd_tgt_bitmap) || !tgt) {
419                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
420                 GOTO(out, rc = -EINVAL);
421         }
422
423         /* check that the UUID matches */
424         if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
425                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
426                        obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
427                 GOTO(out, rc = -EINVAL);
428         }
429
430         __lod_del_device(env, lod, ltd, tgt);
431         EXIT;
432 out:
433         mutex_unlock(&ltd->ltd_mutex);
434         lod_putref(lod, ltd);
435         return(rc);
436 }
437
438 /**
439  * Resize per-thread storage to hold specified size.
440  *
441  * A helper function to resize per-thread temporary storage. This storage
442  * is used to process LOV/LVM EAs and may be quite large. We do not want to
443  * allocate/release it every time, so instead we put it into the env and
444  * reallocate on demand. The memory is released when the correspondent thread
445  * is finished.
446  *
447  * \param[in] info              LOD-specific storage in the environment
448  * \param[in] size              new size to grow the buffer to
449
450  * \retval                      0 on success, -ENOMEM if reallocation failed
451  */
452 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
453 {
454         __u32 round = size_roundup_power2(size);
455
456         if (info->lti_ea_store) {
457                 LASSERT(info->lti_ea_store_size);
458                 LASSERT(info->lti_ea_store_size < round);
459                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
460                        info->lti_ea_store_size, round);
461                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
462                 info->lti_ea_store = NULL;
463                 info->lti_ea_store_size = 0;
464         }
465
466         OBD_ALLOC_LARGE(info->lti_ea_store, round);
467         if (info->lti_ea_store == NULL)
468                 RETURN(-ENOMEM);
469         info->lti_ea_store_size = round;
470
471         RETURN(0);
472 }
473
474 static void lod_free_comp_buffer(struct lod_layout_component *entries,
475                                  __u16 count, __u32 bufsize)
476 {
477         struct lod_layout_component *entry;
478         int i;
479
480         for (i = 0; i < count; i++) {
481                 entry = &entries[i];
482                 if (entry->llc_pool != NULL)
483                         lod_set_pool(&entry->llc_pool, NULL);
484                 if (entry->llc_ostlist.op_array)
485                         OBD_FREE(entry->llc_ostlist.op_array,
486                                  entry->llc_ostlist.op_size);
487                 LASSERT(entry->llc_stripe == NULL);
488                 LASSERT(entry->llc_stripes_allocated == 0);
489         }
490
491         if (bufsize != 0)
492                 OBD_FREE_LARGE(entries, bufsize);
493 }
494
495 void lod_free_def_comp_entries(struct lod_default_striping *lds)
496 {
497         lod_free_comp_buffer(lds->lds_def_comp_entries,
498                              lds->lds_def_comp_size_cnt,
499                              size_roundup_power2(
500                                      sizeof(*lds->lds_def_comp_entries) *
501                                      lds->lds_def_comp_size_cnt));
502         lds->lds_def_comp_entries = NULL;
503         lds->lds_def_comp_cnt = 0;
504         lds->lds_def_striping_is_composite = 0;
505         lds->lds_def_comp_size_cnt = 0;
506 }
507
508 /**
509  * Resize per-thread storage to hold default striping component entries
510  *
511  * A helper function to resize per-thread temporary storage. This storage
512  * is used to hold default LOV/LVM EAs and may be quite large. We do not want
513  * to allocate/release it every time, so instead we put it into the env and
514  * reallocate it on demand. The memory is released when the correspondent
515  * thread is finished.
516  *
517  * \param[in,out] lds           default striping
518  * \param[in] count             new component count to grow the buffer to
519
520  * \retval                      0 on success, -ENOMEM if reallocation failed
521  */
522 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
523 {
524         struct lod_layout_component *entries;
525         __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
526                                         count);
527         __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
528                                         lds->lds_def_comp_size_cnt);
529
530         if (new <= old)
531                 return 0;
532
533         OBD_ALLOC_LARGE(entries, new);
534         if (entries == NULL)
535                 return -ENOMEM;
536
537         if (lds->lds_def_comp_entries != NULL) {
538                 CDEBUG(D_INFO, "default striping component size %d is not "
539                        "enough, need %d\n", old, new);
540                 lod_free_def_comp_entries(lds);
541         }
542
543         lds->lds_def_comp_entries = entries;
544         lds->lds_def_comp_size_cnt = count;
545
546         RETURN(0);
547 }
548
549 void lod_free_comp_entries(struct lod_object *lo)
550 {
551         if (lo->ldo_mirrors) {
552                 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
553                 lo->ldo_mirrors = NULL;
554                 lo->ldo_mirror_count = 0;
555         }
556         lod_free_comp_buffer(lo->ldo_comp_entries,
557                              lo->ldo_comp_cnt,
558                              sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
559         lo->ldo_comp_entries = NULL;
560         lo->ldo_comp_cnt = 0;
561         lo->ldo_is_composite = 0;
562 }
563
564 int lod_alloc_comp_entries(struct lod_object *lo,
565                            int mirror_count, int comp_count)
566 {
567         LASSERT(comp_count != 0);
568         LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
569
570         if (mirror_count > 0) {
571                 OBD_ALLOC_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
572                 if (!lo->ldo_mirrors)
573                         return -ENOMEM;
574
575                 lo->ldo_mirror_count = mirror_count;
576         }
577
578         OBD_ALLOC_LARGE(lo->ldo_comp_entries,
579                         sizeof(*lo->ldo_comp_entries) * comp_count);
580         if (lo->ldo_comp_entries == NULL) {
581                 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
582                 lo->ldo_mirror_count = 0;
583                 return -ENOMEM;
584         }
585
586         lo->ldo_comp_cnt = comp_count;
587         return 0;
588 }
589
590 int lod_fill_mirrors(struct lod_object *lo)
591 {
592         struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
593         struct lod_layout_component *lod_comp;
594         bool found_preferred = false;
595         int mirror_idx = -1;
596         __u16 mirror_id = 0xffff;
597         int i, pref;
598         ENTRY;
599
600         LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
601
602         if (!lo->ldo_is_composite)
603                 RETURN(0);
604
605         lod_comp = &lo->ldo_comp_entries[0];
606
607         for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
608                 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
609                 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
610                 int j;
611
612                 pref = 0;
613                 /* calculate component preference over all used OSTs */
614                 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
615                         int idx = lod_comp->llc_ost_indices[j];
616                         struct obd_statfs *osfs = &OST_TGT(lod,idx)->ltd_statfs;
617
618                         if (osfs->os_state & OS_STATFS_NONROT)
619                                 pref++;
620                 }
621
622                 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
623                         lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
624                         lo->ldo_mirrors[mirror_idx].lme_prefer |= preferred;
625                         lo->ldo_mirrors[mirror_idx].lme_preference += pref;
626                         lo->ldo_mirrors[mirror_idx].lme_end = i;
627                         continue;
628                 }
629
630                 if (mirror_idx >= 0 && preferred &&
631                     !lo->ldo_mirrors[mirror_idx].lme_stale)
632                         found_preferred = true;
633
634                 /* new mirror */
635                 ++mirror_idx;
636                 if (mirror_idx >= lo->ldo_mirror_count)
637                         RETURN(-EINVAL);
638
639                 mirror_id = mirror_id_of(lod_comp->llc_id);
640
641                 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
642                 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
643                 lo->ldo_mirrors[mirror_idx].lme_prefer = preferred;
644                 lo->ldo_mirrors[mirror_idx].lme_preference = pref;
645                 lo->ldo_mirrors[mirror_idx].lme_start = i;
646                 lo->ldo_mirrors[mirror_idx].lme_end = i;
647         }
648         if (mirror_idx != lo->ldo_mirror_count - 1)
649                 RETURN(-EINVAL);
650
651         if (!found_preferred && mirror_idx > 0) {
652                 int best = -1;
653
654                 /*
655                  * if no explicited preferred found, then find a mirror
656                  * with higher number of non-rotational OSTs
657                  * */
658                 pref = -1;
659                 for (i = 0; i <= mirror_idx; i++) {
660                         if (lo->ldo_mirrors[i].lme_stale)
661                                 continue;
662                         if (lo->ldo_mirrors[i].lme_preference > pref) {
663                                 pref = lo->ldo_mirrors[i].lme_preference;
664                                 best = i;
665                         }
666                 }
667
668                 LASSERT(best >= 0);
669                 lo->ldo_mirrors[best].lme_prefer = 1;
670         }
671
672         RETURN(0);
673 }
674
675 /**
676  * Generate on-disk lov_mds_md structure for each layout component based on
677  * the information in lod_object->ldo_comp_entries[i].
678  *
679  * \param[in] env               execution environment for this thread
680  * \param[in] lo                LOD object
681  * \param[in] comp_idx          index of ldo_comp_entries
682  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
683  * \param[in|out] lmm_size      buffer size/lmm size
684  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
685  *                              the stripe info is from the default stripe
686  *                              template, which is collected in lod_ah_init(),
687  *                              either from parent object or root object; for
688  *                              file case, it's from the @lo object
689  *
690  * \retval                      0 if on disk structure is created successfully
691  * \retval                      negative error number on failure
692  */
693 static int lod_gen_component_ea(const struct lu_env *env,
694                                 struct lod_object *lo, int comp_idx,
695                                 struct lov_mds_md *lmm, int *lmm_size,
696                                 bool is_dir)
697 {
698         struct lod_thread_info  *info = lod_env_info(env);
699         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
700         struct lod_device       *lod;
701         struct lov_ost_data_v1  *objs;
702         struct lod_layout_component *lod_comp;
703         __u32   magic;
704         __u16 stripe_count;
705         int     i, rc = 0;
706         ENTRY;
707
708         LASSERT(lo);
709         if (is_dir)
710                 lod_comp =
711                         &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
712         else
713                 lod_comp = &lo->ldo_comp_entries[comp_idx];
714
715         magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
716         if (lod_comp->llc_pattern == 0) /* default striping */
717                 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
718
719         lmm->lmm_magic = cpu_to_le32(magic);
720         lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
721         fid_to_lmm_oi(fid, &lmm->lmm_oi);
722         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
723                 lmm->lmm_oi.oi.oi_id++;
724         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
725
726         lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
727         lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
728         /**
729          * for dir and uninstantiated component, lmm_layout_gen stores
730          * default stripe offset.
731          */
732         lmm->lmm_layout_gen =
733                 (is_dir || !lod_comp_inited(lod_comp)) ?
734                         cpu_to_le16(lod_comp->llc_stripe_offset) :
735                         cpu_to_le16(lod_comp->llc_layout_gen);
736
737         if (magic == LOV_MAGIC_V1) {
738                 objs = &lmm->lmm_objects[0];
739         } else {
740                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
741                 size_t cplen = strlcpy(v3->lmm_pool_name,
742                                        lod_comp->llc_pool,
743                                        sizeof(v3->lmm_pool_name));
744                 if (cplen >= sizeof(v3->lmm_pool_name))
745                         RETURN(-E2BIG);
746                 objs = &v3->lmm_objects[0];
747         }
748         lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
749         stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
750         if (stripe_count == 0 && !is_dir &&
751             !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
752             !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
753                 /* Try again if all active targets are disconnected.
754                  * It is possible when MDS does failover. */
755                 if (!lod->lod_ost_active_count &&
756                     lod->lod_ost_count)
757                         RETURN(-EAGAIN);
758                 RETURN(-E2BIG);
759         }
760
761         if (!is_dir && lo->ldo_is_composite)
762                 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
763
764         if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
765                 GOTO(done, rc = 0);
766
767         /* generate ost_idx of this component stripe */
768         for (i = 0; i < stripe_count; i++) {
769                 struct dt_object *object;
770                 __u32 ost_idx = (__u32)-1UL;
771                 int type = LU_SEQ_RANGE_OST;
772
773                 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
774                         object = lod_comp->llc_stripe[i];
775                         /* instantiated component */
776                         info->lti_fid = *lu_object_fid(&object->do_lu);
777
778                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
779                             comp_idx == 0) {
780                                 if (cfs_fail_val == 0)
781                                         cfs_fail_val = info->lti_fid.f_oid;
782                                 else if (i == 0)
783                                         info->lti_fid.f_oid = cfs_fail_val;
784                         }
785
786                         rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
787                         LASSERT(rc == 0);
788
789                         ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
790                         objs[i].l_ost_gen = cpu_to_le32(0);
791                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
792                                 rc = -ENOENT;
793                         else
794                                 rc = lod_fld_lookup(env, lod, &info->lti_fid,
795                                                     &ost_idx, &type);
796                         if (rc < 0) {
797                                 CERROR("%s: Can not locate "DFID": rc = %d\n",
798                                        lod2obd(lod)->obd_name,
799                                        PFID(&info->lti_fid), rc);
800                                 RETURN(rc);
801                         }
802                 } else if (lod_comp->llc_ostlist.op_array &&
803                            lod_comp->llc_ostlist.op_count) {
804                         /* user specified ost list */
805                         ost_idx = lod_comp->llc_ostlist.op_array[i];
806                 }
807                 /*
808                  * with un-instantiated or with no specified ost list
809                  * component, its l_ost_idx does not matter.
810                  */
811                 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
812         }
813 done:
814         if (lmm_size != NULL)
815                 *lmm_size = lov_mds_md_size(stripe_count, magic);
816         RETURN(rc);
817 }
818
819 /**
820  * Generate on-disk lov_mds_md structure based on the information in
821  * the lod_object->ldo_comp_entries.
822  *
823  * \param[in] env               execution environment for this thread
824  * \param[in] lo                LOD object
825  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
826  * \param[in|out] lmm_size      buffer size/lmm size
827  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
828  *                              the stripe info is from the default stripe
829  *                              template, which is collected in lod_ah_init(),
830  *                              either from parent object or root object; for
831  *                              file case, it's from the @lo object
832  *
833  * \retval                      0 if on disk structure is created successfully
834  * \retval                      negative error number on failure
835  */
836 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
837                        struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
838 {
839         struct lov_comp_md_entry_v1 *lcme;
840         struct lov_comp_md_v1 *lcm;
841         struct lod_layout_component *comp_entries;
842         __u16 comp_cnt, mirror_cnt;
843         bool is_composite, is_foreign = false;
844         int i, rc = 0, offset;
845         ENTRY;
846
847         if (is_dir) {
848                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
849                 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
850                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
851                 is_composite =
852                         lo->ldo_def_striping->lds_def_striping_is_composite;
853         } else {
854                 comp_cnt = lo->ldo_comp_cnt;
855                 mirror_cnt = lo->ldo_mirror_count;
856                 comp_entries = lo->ldo_comp_entries;
857                 is_composite = lo->ldo_is_composite;
858                 is_foreign = lo->ldo_is_foreign;
859         }
860
861         LASSERT(lmm_size != NULL);
862
863         if (is_foreign) {
864                 struct lov_foreign_md *lfm;
865
866                 lfm = (struct lov_foreign_md *)lmm;
867                 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
868                 /* need to store little-endian */
869                 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
870                         __swab32s(&lfm->lfm_magic);
871                         __swab32s(&lfm->lfm_length);
872                         __swab32s(&lfm->lfm_type);
873                         __swab32s(&lfm->lfm_flags);
874                 }
875                 *lmm_size = lo->ldo_foreign_lov_size;
876                 RETURN(0);
877         }
878
879         LASSERT(comp_cnt != 0 && comp_entries != NULL);
880
881         if (!is_composite) {
882                 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
883                 RETURN(rc);
884         }
885
886         lcm = (struct lov_comp_md_v1 *)lmm;
887         memset(lcm, 0, sizeof(*lcm));
888
889         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
890         lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
891         lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
892         lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
893
894         offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
895         LASSERT(offset % sizeof(__u64) == 0);
896
897         for (i = 0; i < comp_cnt; i++) {
898                 struct lod_layout_component *lod_comp;
899                 struct lov_mds_md *sub_md;
900                 int size;
901
902                 lod_comp = &comp_entries[i];
903                 lcme = &lcm->lcm_entries[i];
904
905                 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
906                 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
907
908                 /* component could be un-inistantiated */
909                 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
910                 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
911                         lcme->lcme_timestamp =
912                                 cpu_to_le64(lod_comp->llc_timestamp);
913                 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
914                         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
915
916                 lcme->lcme_extent.e_start =
917                         cpu_to_le64(lod_comp->llc_extent.e_start);
918                 lcme->lcme_extent.e_end =
919                         cpu_to_le64(lod_comp->llc_extent.e_end);
920                 lcme->lcme_offset = cpu_to_le32(offset);
921
922                 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
923                 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
924                 if (rc)
925                         GOTO(out, rc);
926                 lcme->lcme_size = cpu_to_le32(size);
927                 offset += size;
928                 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
929                          "offset:%d lmm_size:%d\n", offset, *lmm_size);
930         }
931         lcm->lcm_size = cpu_to_le32(offset);
932         lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
933
934         lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
935                              "generate lum");
936 out:
937         if (rc == 0)
938                 *lmm_size = offset;
939         RETURN(rc);
940 }
941
942 /**
943  * Get LOV EA.
944  *
945  * Fill lti_ea_store buffer in the environment with a value for the given
946  * EA. The buffer is reallocated if the value doesn't fit.
947  *
948  * \param[in,out] env           execution environment for this thread
949  *                              .lti_ea_store buffer is filled with EA's value
950  * \param[in] lo                LOD object
951  * \param[in] name              name of the EA
952  *
953  * \retval                      > 0 if EA is fetched successfully
954  * \retval                      0 if EA is empty
955  * \retval                      negative error number on failure
956  */
957 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
958                const char *name)
959 {
960         struct lod_thread_info  *info = lod_env_info(env);
961         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
962         int                     rc;
963         ENTRY;
964
965         LASSERT(info);
966
967         if (unlikely(info->lti_ea_store == NULL)) {
968                 /* just to enter in allocation block below */
969                 rc = -ERANGE;
970         } else {
971 repeat:
972                 info->lti_buf.lb_buf = info->lti_ea_store;
973                 info->lti_buf.lb_len = info->lti_ea_store_size;
974                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
975         }
976
977         /* if object is not striped or inaccessible */
978         if (rc == -ENODATA || rc == -ENOENT)
979                 RETURN(0);
980
981         if (rc == -ERANGE) {
982                 /* EA doesn't fit, reallocate new buffer */
983                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
984                 if (rc == -ENODATA || rc == -ENOENT)
985                         RETURN(0);
986                 else if (rc < 0)
987                         RETURN(rc);
988
989                 LASSERT(rc > 0);
990                 rc = lod_ea_store_resize(info, rc);
991                 if (rc)
992                         RETURN(rc);
993                 goto repeat;
994         }
995
996         RETURN(rc);
997 }
998
999 /**
1000  * Verify the target index is present in the current configuration.
1001  *
1002  * \param[in] md                LOD device where the target table is stored
1003  * \param[in] idx               target's index
1004  *
1005  * \retval                      0 if the index is present
1006  * \retval                      -EINVAL if not
1007  */
1008 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1009 {
1010         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1011                      !test_bit(idx, md->lod_ost_bitmap))) {
1012                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1013                        md->lod_ost_descs.ltd_tgts_size);
1014                 return -EINVAL;
1015         }
1016
1017         if (unlikely(OST_TGT(md, idx) == NULL)) {
1018                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1019                        lod2obd(md)->obd_name, idx);
1020                 return -EINVAL;
1021         }
1022
1023         if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
1024                 CERROR("%s: invalid lod device, for idx: %d\n",
1025                        lod2obd(md)->obd_name , idx);
1026                 return -EINVAL;
1027         }
1028
1029         return 0;
1030 }
1031
1032 /**
1033  * Instantiate objects for stripes.
1034  *
1035  * Allocate and initialize LU-objects representing the stripes. The number
1036  * of the stripes (ldo_stripe_count) must be initialized already. The caller
1037  * must ensure nobody else is calling the function on the object at the same
1038  * time. FLDB service must be running to be able to map a FID to the targets
1039  * and find appropriate device representing that target.
1040  *
1041  * \param[in] env               execution environment for this thread
1042  * \param[in,out] lo            LOD object
1043  * \param[in] objs              an array of IDs to creates the objects from
1044  * \param[in] comp_idx          index of ldo_comp_entries
1045  *
1046  * \retval                      0 if the objects are instantiated successfully
1047  * \retval                      negative error number on failure
1048  */
1049 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1050                            struct lov_ost_data_v1 *objs, int comp_idx)
1051 {
1052         struct lod_layout_component *lod_comp;
1053         struct lod_thread_info *info = lod_env_info(env);
1054         struct lod_device *md;
1055         struct lu_object *o, *n;
1056         struct lu_device *nd;
1057         struct dt_object **stripe = NULL;
1058         __u32 *ost_indices = NULL;
1059         int stripe_len;
1060         int i, rc = 0;
1061         __u32 idx;
1062         ENTRY;
1063
1064         LASSERT(lo != NULL);
1065         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1066
1067         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1068         lod_comp = &lo->ldo_comp_entries[comp_idx];
1069
1070         LASSERT(lod_comp->llc_stripe == NULL);
1071         LASSERT(lod_comp->llc_stripe_count > 0);
1072         LASSERT(lod_comp->llc_stripe_size > 0);
1073
1074         stripe_len = lod_comp->llc_stripe_count;
1075         OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1076         if (stripe == NULL)
1077                 RETURN(-ENOMEM);
1078         OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1079         if (!ost_indices)
1080                 GOTO(out, rc = -ENOMEM);
1081
1082         for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1083                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1084                         continue;
1085
1086                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1087                 idx = le32_to_cpu(objs[i].l_ost_idx);
1088                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1089                 if (rc != 0)
1090                         GOTO(out, rc);
1091                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1092                          PFID(&info->lti_fid));
1093                 lod_getref(&md->lod_ost_descs);
1094
1095                 rc = validate_lod_and_idx(md, idx);
1096                 if (unlikely(rc != 0)) {
1097                         lod_putref(md, &md->lod_ost_descs);
1098                         GOTO(out, rc);
1099                 }
1100
1101                 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1102                 lod_putref(md, &md->lod_ost_descs);
1103
1104                 /* In the function below, .hs_keycmp resolves to
1105                  * u_obj_hop_keycmp() */
1106                 /* coverity[overrun-buffer-val] */
1107                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1108                 if (IS_ERR(o))
1109                         GOTO(out, rc = PTR_ERR(o));
1110
1111                 n = lu_object_locate(o->lo_header, nd->ld_type);
1112                 LASSERT(n);
1113
1114                 stripe[i] = container_of(n, struct dt_object, do_lu);
1115                 ost_indices[i] = idx;
1116         }
1117
1118 out:
1119         if (rc != 0) {
1120                 for (i = 0; i < stripe_len; i++)
1121                         if (stripe[i] != NULL)
1122                                 dt_object_put(env, stripe[i]);
1123
1124                 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1125                 lod_comp->llc_stripe_count = 0;
1126                 if (ost_indices)
1127                         OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1128         } else {
1129                 lod_comp->llc_stripe = stripe;
1130                 lod_comp->llc_ost_indices = ost_indices;
1131                 lod_comp->llc_stripes_allocated = stripe_len;
1132         }
1133
1134         RETURN(rc);
1135 }
1136
1137 /**
1138  * Instantiate objects for striping.
1139  *
1140  * Parse striping information in \a buf and instantiate the objects
1141  * representing the stripes.
1142  *
1143  * \param[in] env               execution environment for this thread
1144  * \param[in] lo                LOD object
1145  * \param[in] buf               buffer storing LOV EA to parse
1146  *
1147  * \retval                      0 if parsing and objects creation succeed
1148  * \retval                      negative error number on failure
1149  */
1150 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1151                        const struct lu_buf *buf)
1152 {
1153         struct lov_mds_md_v1 *lmm;
1154         struct lov_comp_md_v1 *comp_v1 = NULL;
1155         struct lov_foreign_md *foreign = NULL;
1156         struct lov_ost_data_v1 *objs;
1157         __u32 magic, pattern;
1158         __u16 mirror_cnt = 0;
1159         __u16 comp_cnt;
1160         int i, rc;
1161         ENTRY;
1162
1163         LASSERT(buf);
1164         LASSERT(buf->lb_buf);
1165         LASSERT(buf->lb_len);
1166         LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1167
1168         lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1169         magic = le32_to_cpu(lmm->lmm_magic);
1170
1171         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1172             magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1173             magic != LOV_MAGIC_SEL)
1174                 GOTO(out, rc = -EINVAL);
1175
1176         lod_striping_free_nolock(env, lo);
1177
1178         if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1179                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1180                 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1181                 if (comp_cnt == 0)
1182                         GOTO(out, rc = -EINVAL);
1183                 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1184                 lo->ldo_is_composite = 1;
1185                 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1186                 if (mirror_cnt > 1)
1187                         lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1188                                                         LCM_FL_FLR_MASK;
1189                 else
1190                         lo->ldo_flr_state = LCM_FL_NONE;
1191         } else if (magic == LOV_MAGIC_FOREIGN) {
1192                 size_t length;
1193
1194                 foreign = (struct lov_foreign_md *)buf->lb_buf;
1195                 length = offsetof(typeof(*foreign), lfm_value);
1196                 if (buf->lb_len < length ||
1197                     buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1198                         CDEBUG(D_LAYOUT,
1199                                "buf len %zu too small for lov_foreign_md\n",
1200                                buf->lb_len);
1201                         GOTO(out, rc = -EINVAL);
1202                 }
1203
1204                 /* just cache foreign LOV EA raw */
1205                 rc = lod_alloc_foreign_lov(lo, length);
1206                 if (rc)
1207                         GOTO(out, rc);
1208                 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1209                 GOTO(out, rc);
1210         } else {
1211                 comp_cnt = 1;
1212                 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1213                 lo->ldo_is_composite = 0;
1214         }
1215
1216         rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1217         if (rc)
1218                 GOTO(out, rc);
1219
1220         for (i = 0; i < comp_cnt; i++) {
1221                 struct lod_layout_component *lod_comp;
1222                 struct lu_extent *ext;
1223                 __u32 offs;
1224
1225                 lod_comp = &lo->ldo_comp_entries[i];
1226                 if (lo->ldo_is_composite) {
1227                         offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1228                         lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1229
1230                         ext = &comp_v1->lcm_entries[i].lcme_extent;
1231                         lod_comp->llc_extent.e_start =
1232                                 le64_to_cpu(ext->e_start);
1233                         if (lod_comp->llc_extent.e_start &
1234                             (LOV_MIN_STRIPE_SIZE - 1)) {
1235                                 CDEBUG(D_LAYOUT,
1236                                        "extent start %llu is not a multiple of min size %u\n",
1237                                        lod_comp->llc_extent.e_start,
1238                                        LOV_MIN_STRIPE_SIZE);
1239                                 GOTO(out, rc = -EINVAL);
1240                         }
1241
1242                         lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1243                         if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1244                             lod_comp->llc_extent.e_end &
1245                             (LOV_MIN_STRIPE_SIZE - 1)) {
1246                                 CDEBUG(D_LAYOUT,
1247                                        "extent end %llu is not a multiple of min size %u\n",
1248                                        lod_comp->llc_extent.e_end,
1249                                        LOV_MIN_STRIPE_SIZE);
1250                                 GOTO(out, rc = -EINVAL);
1251                         }
1252
1253                         lod_comp->llc_flags =
1254                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1255                         if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1256                                 lod_comp->llc_timestamp = le64_to_cpu(
1257                                         comp_v1->lcm_entries[i].lcme_timestamp);
1258                         lod_comp->llc_id =
1259                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1260                         if (lod_comp->llc_id == LCME_ID_INVAL)
1261                                 GOTO(out, rc = -EINVAL);
1262
1263                         if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1264                             comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1265                                 struct lod_device *d =
1266                                         lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1267
1268                                 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1269                                       lod2obd(d)->obd_name,
1270                                       lod_comp->llc_flags, lod_comp->llc_id, i,
1271                                       PFID(lod_object_fid(lo)),
1272                                       le32_to_cpu(comp_v1->lcm_magic));
1273                         }
1274                 } else {
1275                         lod_comp_set_init(lod_comp);
1276                 }
1277
1278                 pattern = le32_to_cpu(lmm->lmm_pattern);
1279                 if (!lov_pattern_supported(lov_pattern(pattern)))
1280                         GOTO(out, rc = -EINVAL);
1281
1282                 lod_comp->llc_pattern = pattern;
1283                 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1284                 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1285                 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1286
1287                 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1288                         struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1289
1290                         lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1291                         objs = &v3->lmm_objects[0];
1292                 } else {
1293                         lod_set_pool(&lod_comp->llc_pool, NULL);
1294                         objs = &lmm->lmm_objects[0];
1295                 }
1296
1297                 /**
1298                  * If uninstantiated template component has valid l_ost_idx,
1299                  * then user has specified ost list for this component.
1300                  */
1301                 if (!lod_comp_inited(lod_comp)) {
1302                         __u16 stripe_count;
1303
1304                         if (objs[0].l_ost_idx != (__u32)-1UL) {
1305                                 int j;
1306
1307                                 stripe_count = lod_comp_entry_stripe_count(
1308                                                         lo, i, false);
1309                                 if (stripe_count == 0 &&
1310                                     !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1311                                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1312                                         GOTO(out, rc = -E2BIG);
1313                                 /**
1314                                  * load the user specified ost list, when this
1315                                  * component is instantiated later, it will be
1316                                  * used in lod_alloc_ost_list().
1317                                  */
1318                                 lod_comp->llc_ostlist.op_count = stripe_count;
1319                                 lod_comp->llc_ostlist.op_size =
1320                                         stripe_count * sizeof(__u32);
1321                                 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1322                                           lod_comp->llc_ostlist.op_size);
1323                                 if (!lod_comp->llc_ostlist.op_array)
1324                                         GOTO(out, rc = -ENOMEM);
1325
1326                                 for (j = 0; j < stripe_count; j++)
1327                                         lod_comp->llc_ostlist.op_array[j] =
1328                                                 le32_to_cpu(objs[j].l_ost_idx);
1329
1330                                 /**
1331                                  * this component OST objects starts from the
1332                                  * first ost_idx, lod_alloc_ost_list() will
1333                                  * check this.
1334                                  */
1335                                 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1336                         } else {
1337                                 /**
1338                                  * for uninstantiated component,
1339                                  * lmm_layout_gen stores default stripe offset.
1340                                  */
1341                                 lod_comp->llc_stripe_offset =
1342                                                         lmm->lmm_layout_gen;
1343                         }
1344                 }
1345
1346                 /* skip un-instantiated component object initialization */
1347                 if (!lod_comp_inited(lod_comp))
1348                         continue;
1349
1350                 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1351                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1352                         rc = lod_initialize_objects(env, lo, objs, i);
1353                         if (rc)
1354                                 GOTO(out, rc);
1355                 }
1356         }
1357
1358         rc = lod_fill_mirrors(lo);
1359         if (rc)
1360                 GOTO(out, rc);
1361
1362 out:
1363         if (rc)
1364                 lod_striping_free_nolock(env, lo);
1365         RETURN(rc);
1366 }
1367
1368 /**
1369  * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1370  * is already cached.
1371  *
1372  * \param[in] lo        LOD object
1373  *
1374  * \retval              True if the striping is cached, otherwise
1375  *                      return false.
1376  */
1377 static bool lod_striping_loaded(struct lod_object *lo)
1378 {
1379         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1380             lo->ldo_comp_cached)
1381                 return true;
1382
1383         if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1384                 if (lo->ldo_dir_stripe_loaded)
1385                         return true;
1386
1387                 /* Never load LMV stripe for slaves of striped dir */
1388                 if (lo->ldo_dir_slave_stripe)
1389                         return true;
1390         }
1391
1392         return false;
1393 }
1394
1395 /**
1396  * A generic function to initialize the stripe objects.
1397  *
1398  * A protected version of lod_striping_load_locked() - load the striping
1399  * information from storage, parse that and instantiate LU objects to
1400  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1401  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1402  * return an array of references to the newly instantiated objects.
1403  *
1404  * \param[in] env               execution environment for this thread
1405  * \param[in,out] lo            LOD object, where striping is stored and
1406  *                              which gets an array of references
1407  *
1408  * \retval                      0 if parsing and object creation succeed
1409  * \retval                      negative error number on failure
1410  **/
1411 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1412 {
1413         struct lod_thread_info *info = lod_env_info(env);
1414         struct dt_object *next = dt_object_child(&lo->ldo_obj);
1415         struct lu_buf *buf = &info->lti_buf;
1416         int rc = 0;
1417
1418         ENTRY;
1419
1420         if (!dt_object_exists(next))
1421                 RETURN(0);
1422
1423         if (lod_striping_loaded(lo))
1424                 RETURN(0);
1425
1426         mutex_lock(&lo->ldo_layout_mutex);
1427         if (lod_striping_loaded(lo))
1428                 GOTO(unlock, rc = 0);
1429
1430         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1431                 rc = lod_get_lov_ea(env, lo);
1432                 if (rc <= 0)
1433                         GOTO(unlock, rc);
1434
1435                 /*
1436                  * there is LOV EA (striping information) in this object
1437                  * let's parse it and create in-core objects for the stripes
1438                  */
1439                 buf->lb_buf = info->lti_ea_store;
1440                 buf->lb_len = info->lti_ea_store_size;
1441                 rc = lod_parse_striping(env, lo, buf);
1442                 if (rc == 0)
1443                         lo->ldo_comp_cached = 1;
1444         } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1445                 rc = lod_get_lmv_ea(env, lo);
1446                 if (rc > sizeof(struct lmv_foreign_md)) {
1447                         struct lmv_foreign_md *lfm = info->lti_ea_store;
1448
1449                         if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1450                                 lo->ldo_foreign_lmv = info->lti_ea_store;
1451                                 lo->ldo_foreign_lmv_size =
1452                                         info->lti_ea_store_size;
1453                                 info->lti_ea_store = NULL;
1454                                 info->lti_ea_store_size = 0;
1455
1456                                 lo->ldo_dir_stripe_loaded = 1;
1457                                 lo->ldo_dir_is_foreign = 1;
1458                                 GOTO(unlock, rc = 0);
1459                         }
1460                 }
1461
1462                 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1463                         /* Let's set stripe_loaded to avoid further
1464                          * stripe loading especially for non-stripe directory,
1465                          * which can hurt performance. (See LU-9840)
1466                          */
1467                         if (rc == 0)
1468                                 lo->ldo_dir_stripe_loaded = 1;
1469                         GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1470                 }
1471                 buf->lb_buf = info->lti_ea_store;
1472                 buf->lb_len = info->lti_ea_store_size;
1473                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1474                         rc = lod_load_lmv_shards(env, lo, buf, true);
1475                         if (buf->lb_buf != info->lti_ea_store) {
1476                                 OBD_FREE_LARGE(info->lti_ea_store,
1477                                                info->lti_ea_store_size);
1478                                 info->lti_ea_store = buf->lb_buf;
1479                                 info->lti_ea_store_size = buf->lb_len;
1480                         }
1481
1482                         if (rc < 0)
1483                                 GOTO(unlock, rc);
1484                 }
1485
1486                 /*
1487                  * there is LMV EA (striping information) in this object
1488                  * let's parse it and create in-core objects for the stripes
1489                  */
1490                 rc = lod_parse_dir_striping(env, lo, buf);
1491                 if (rc == 0)
1492                         lo->ldo_dir_stripe_loaded = 1;
1493         }
1494         EXIT;
1495 unlock:
1496         mutex_unlock(&lo->ldo_layout_mutex);
1497
1498         return rc;
1499 }
1500
1501 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1502                          const struct lu_buf *buf)
1503 {
1504         int rc;
1505
1506         ENTRY;
1507
1508         mutex_lock(&lo->ldo_layout_mutex);
1509         rc = lod_parse_striping(env, lo, buf);
1510         mutex_unlock(&lo->ldo_layout_mutex);
1511
1512         RETURN(rc);
1513 }
1514
1515 /**
1516  * Verify lov_user_md_v1/v3 striping.
1517  *
1518  * Check the validity of all fields including the magic, stripe size,
1519  * stripe count, stripe offset and that the pool is present.  Also check
1520  * that each target index points to an existing target. The additional
1521  * \a is_from_disk turns additional checks. In some cases zero fields
1522  * are allowed (like pattern=0).
1523  *
1524  * \param[in] d                 LOD device
1525  * \param[in] buf               buffer with LOV EA to verify
1526  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1527  *                              1 - from disk, do not allow
1528  *
1529  * \retval                      0 if the striping is valid
1530  * \retval                      -EINVAL if striping is invalid
1531  */
1532 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1533                            bool is_from_disk)
1534 {
1535         struct lov_user_md_v1   *lum;
1536         struct lov_user_md_v3   *lum3;
1537         struct pool_desc        *pool = NULL;
1538         __u32                    magic;
1539         __u32                    stripe_size;
1540         __u16                    stripe_count;
1541         __u16                    stripe_offset;
1542         size_t                   lum_size;
1543         int                      rc = 0;
1544         ENTRY;
1545
1546         lum = buf->lb_buf;
1547
1548         if (buf->lb_len < sizeof(*lum)) {
1549                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1550                        buf->lb_len);
1551                 GOTO(out, rc = -EINVAL);
1552         }
1553
1554         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1555         if (magic != LOV_USER_MAGIC_V1 &&
1556             magic != LOV_USER_MAGIC_V3 &&
1557             magic != LOV_USER_MAGIC_SPECIFIC) {
1558                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1559                        le32_to_cpu(lum->lmm_magic));
1560                 GOTO(out, rc = -EINVAL);
1561         }
1562
1563         /* the user uses "0" for default stripe pattern normally. */
1564         if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1565                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1566
1567         if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1568                 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1569                        le32_to_cpu(lum->lmm_pattern));
1570                 GOTO(out, rc = -EINVAL);
1571         }
1572
1573         /* a released lum comes from creating orphan on hsm release,
1574          * doesn't make sense to verify it. */
1575         if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1576                 GOTO(out, rc = 0);
1577
1578         /* 64kB is the largest common page size we see (ia64), and matches the
1579          * check in lfs */
1580         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1581         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1582                 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1583                        stripe_size, LOV_MIN_STRIPE_SIZE);
1584                 GOTO(out, rc = -EINVAL);
1585         }
1586
1587         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1588         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1589             lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1590                 /* if offset is not within valid range [0, osts_size) */
1591                 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1592                         CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1593                                stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1594                         GOTO(out, rc = -EINVAL);
1595                 }
1596
1597                 /* if lmm_stripe_offset is *not* in bitmap */
1598                 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1599                         CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1600                                stripe_offset);
1601                         GOTO(out, rc = -EINVAL);
1602                 }
1603         }
1604
1605         if (magic == LOV_USER_MAGIC_V1)
1606                 lum_size = offsetof(struct lov_user_md_v1,
1607                                     lmm_objects[0]);
1608         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1609                 lum_size = offsetof(struct lov_user_md_v3,
1610                                     lmm_objects[0]);
1611         else
1612                 GOTO(out, rc = -EINVAL);
1613
1614         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1615         if (buf->lb_len < lum_size) {
1616                 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1617                        "magic %#x and stripe_count %u\n",
1618                        buf->lb_len, lum_size, magic, stripe_count);
1619                 GOTO(out, rc = -EINVAL);
1620         }
1621
1622         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1623                 goto out;
1624
1625         lum3 = buf->lb_buf;
1626         /* In the function below, .hs_keycmp resolves to
1627          * pool_hashkey_keycmp() */
1628         /* coverity[overrun-buffer-val] */
1629         pool = lod_find_pool(d, lum3->lmm_pool_name);
1630         if (pool == NULL)
1631                 goto out;
1632
1633         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1634                 rc = lod_check_index_in_pool(stripe_offset, pool);
1635                 if (rc < 0)
1636                         GOTO(out, rc = -EINVAL);
1637         }
1638
1639         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1640                 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1641                        stripe_count, pool_tgt_count(pool));
1642                 GOTO(out, rc = -EINVAL);
1643         }
1644
1645 out:
1646         if (pool != NULL)
1647                 lod_pool_putref(pool);
1648
1649         RETURN(rc);
1650 }
1651
1652 static inline
1653 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1654 {
1655         LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1656                  LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1657                  le32_to_cpu(comp->lcm_magic));
1658         LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1659                  "bad index %d, max = %d\n",
1660                  i, le16_to_cpu(comp->lcm_entry_count));
1661
1662         return &comp->lcm_entries[i];
1663 }
1664
1665 #define for_each_comp_entry_v1(comp, entry) \
1666         for (entry = comp_entry_v1(comp, 0); \
1667              entry <= comp_entry_v1(comp, \
1668                                    le16_to_cpu(comp->lcm_entry_count) - 1); \
1669              entry++)
1670
1671 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1672                          struct lov_comp_md_entry_v1 *dom_ent)
1673 {
1674         struct lov_comp_md_entry_v1 *ent;
1675         __u16 entries;
1676         __u32 dom_off, dom_size, comp_size, off;
1677         void *src, *dst;
1678         unsigned int size, shift;
1679
1680         entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1681         LASSERT(entries > 0);
1682         comp_v1->lcm_entry_count = cpu_to_le16(entries);
1683
1684         comp_size = le32_to_cpu(comp_v1->lcm_size);
1685         dom_off = le32_to_cpu(dom_ent->lcme_offset);
1686         dom_size = le32_to_cpu(dom_ent->lcme_size);
1687
1688         /* all entries offsets are shifted by entry size at least */
1689         shift = sizeof(*dom_ent);
1690         for_each_comp_entry_v1(comp_v1, ent) {
1691                 off = le32_to_cpu(ent->lcme_offset);
1692                 if (off == dom_off) {
1693                         /* Entry deletion creates two holes in layout data:
1694                          * - hole in entries array
1695                          * - hole in layout data at dom_off with dom_size
1696                          *
1697                          * First memmove is one entry shift from next entry
1698                          * start with size up to dom_off in blob
1699                          */
1700                         dst = (void *)ent;
1701                         src = (void *)(ent + 1);
1702                         size = (unsigned long)((void *)comp_v1 + dom_off - src);
1703                         memmove(dst, src, size);
1704                         /* take 'off' from just moved entry */
1705                         off = le32_to_cpu(ent->lcme_offset);
1706                         /* second memmove is blob tail after 'off' up to
1707                          * component end
1708                          */
1709                         dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1710                         src = (void *)comp_v1 + off;
1711                         size = (unsigned long)(comp_size - off);
1712                         memmove(dst, src, size);
1713                         /* all entries offsets after DoM entry are shifted by
1714                          * dom_size additionally
1715                          */
1716                         shift += dom_size;
1717                 }
1718                 ent->lcme_offset = cpu_to_le32(off - shift);
1719         }
1720         comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1721
1722         /* notify a caller to re-check entry */
1723         return -ERESTART;
1724 }
1725
1726 void lod_dom_stripesize_recalc(struct lod_device *d)
1727 {
1728         __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1729         __u32 max_size = d->lod_dom_stripesize_max_kb;
1730         __u32 def_size = d->lod_dom_stripesize_cur_kb;
1731
1732         /* use maximum allowed value if free space is above threshold */
1733         if (d->lod_lsfs_free_mb >= threshold_mb) {
1734                 def_size = max_size;
1735         } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1736                 def_size = 0;
1737         } else {
1738                 /* recalc threshold like it would be with def_size as max */
1739                 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1740                 if (d->lod_lsfs_free_mb < threshold_mb)
1741                         def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1742                 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1743                         def_size = max_t(unsigned int, def_size * 2,
1744                                          LOD_DOM_MIN_SIZE_KB);
1745         }
1746
1747         if (d->lod_dom_stripesize_cur_kb != def_size) {
1748                 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1749                        d->lod_dom_stripesize_cur_kb, def_size);
1750                 d->lod_dom_stripesize_cur_kb = def_size;
1751         }
1752 }
1753
1754 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1755                                       struct lod_device *d)
1756 {
1757         int rc;
1758
1759         /* set bfree as fraction of total space */
1760         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1761                 spin_lock(&d->lod_lsfs_lock);
1762                 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1763                                         min_t(int, cfs_fail_val, 100), 100);
1764                 GOTO(recalc, rc = 0);
1765         }
1766
1767         if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1768                 struct obd_statfs sfs;
1769
1770                 spin_lock(&d->lod_lsfs_lock);
1771                 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1772                         GOTO(unlock, rc = 0);
1773
1774                 d->lod_lsfs_age = ktime_get_seconds();
1775                 spin_unlock(&d->lod_lsfs_lock);
1776                 rc = dt_statfs(env, d->lod_child, &sfs);
1777                 if (rc) {
1778                         CDEBUG(D_LAYOUT,
1779                                "%s: failed to get OSD statfs: rc = %d\n",
1780                                lod2obd(d)->obd_name, rc);
1781                         GOTO(out, rc);
1782                 }
1783                 /* udpate local OSD cached statfs data */
1784                 spin_lock(&d->lod_lsfs_lock);
1785                 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1786                 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1787 recalc:
1788                 lod_dom_stripesize_recalc(d);
1789 unlock:
1790                 spin_unlock(&d->lod_lsfs_lock);
1791         }
1792 out:
1793         return d->lod_dom_stripesize_cur_kb << 10;
1794 }
1795
1796 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1797                               struct lov_comp_md_v1 *comp_v1,
1798                               struct lov_comp_md_entry_v1 *dom_ent,
1799                               __u32 stripe_size)
1800 {
1801         struct lov_comp_md_entry_v1 *ent;
1802         struct lu_extent *dom_ext, *ext;
1803         struct lov_user_md_v1 *lum;
1804         __u32 max_stripe_size;
1805         __u16 mid, dom_mid;
1806         int rc = 0;
1807         bool dom_next_entry = false;
1808
1809         dom_ext = &dom_ent->lcme_extent;
1810         dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1811         max_stripe_size = lod_dom_stripesize_limit(env, d);
1812
1813         /* Check stripe size againts current per-MDT limit */
1814         if (stripe_size <= max_stripe_size)
1815                 return 0;
1816
1817         lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1818         CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1819                stripe_size, max_stripe_size);
1820         lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1821
1822         /* In common case the DoM stripe is first entry in a mirror and
1823          * can be deleted only if it is not single entry in layout or
1824          * mirror, otherwise error should be returned.
1825          */
1826         for_each_comp_entry_v1(comp_v1, ent) {
1827                 if (ent == dom_ent)
1828                         continue;
1829
1830                 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1831                 if (mid != dom_mid)
1832                         continue;
1833
1834                 ext = &ent->lcme_extent;
1835                 if (ext->e_start != dom_ext->e_end)
1836                         continue;
1837
1838                 /* Found next component after the DoM one with the same
1839                  * mirror_id and adjust its start with DoM component end.
1840                  *
1841                  * NOTE: we are considering here that there can be only one
1842                  * DoM component in a file, all replicas are located on OSTs
1843                  * always and don't need adjustment since use own layouts.
1844                  */
1845                 ext->e_start = cpu_to_le64(max_stripe_size);
1846                 dom_next_entry = true;
1847                 break;
1848         }
1849
1850         if (max_stripe_size == 0) {
1851                 /* DoM component size is zero due to server setting, remove
1852                  * it from the layout but only if next component exists in
1853                  * the same mirror. That must be checked prior calling the
1854                  * lod_erase_dom_stripe().
1855                  */
1856                 if (!dom_next_entry)
1857                         return -EFBIG;
1858
1859                 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1860         } else {
1861                 /* Update DoM extent end finally */
1862                 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1863         }
1864
1865         return rc;
1866 }
1867
1868 /**
1869  * Verify LOV striping.
1870  *
1871  * \param[in] d                 LOD device
1872  * \param[in] buf               buffer with LOV EA to verify
1873  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1874  *                              1 - from disk, do not allow
1875  * \param[in] start             extent start for composite layout
1876  *
1877  * \retval                      0 if the striping is valid
1878  * \retval                      -EINVAL if striping is invalid
1879  */
1880 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1881                         struct lod_object *lo, const struct lu_buf *buf,
1882                         bool is_from_disk)
1883 {
1884         struct lov_user_md_v1   *lum;
1885         struct lov_comp_md_v1   *comp_v1;
1886         struct lov_comp_md_entry_v1     *ent;
1887         struct lu_extent        *ext;
1888         struct lu_buf   tmp;
1889         __u64   prev_end = 0;
1890         __u32   stripe_size = 0;
1891         __u16   prev_mid = -1, mirror_id = -1;
1892         __u32   mirror_count;
1893         __u32   magic;
1894         int     rc = 0;
1895         ENTRY;
1896
1897         if (buf->lb_len < sizeof(lum->lmm_magic)) {
1898                 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1899                 RETURN(-EINVAL);
1900         }
1901
1902         lum = buf->lb_buf;
1903
1904         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1905         /* treat foreign LOV EA/object case first
1906          * XXX is it expected to try setting again a foreign?
1907          * XXX should we care about different current vs new layouts ?
1908          */
1909         if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1910                 struct lov_foreign_md *lfm = buf->lb_buf;
1911
1912                 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1913                         CDEBUG(D_LAYOUT,
1914                                "buf len %zu < min lov_foreign_md size (%zu)\n",
1915                                buf->lb_len, offsetof(typeof(*lfm),
1916                                lfm_value));
1917                         RETURN(-EINVAL);
1918                 }
1919
1920                 if (foreign_size_le(lfm) > buf->lb_len) {
1921                         CDEBUG(D_LAYOUT,
1922                                "buf len %zu < this lov_foreign_md size (%zu)\n",
1923                                buf->lb_len, foreign_size_le(lfm));
1924                         RETURN(-EINVAL);
1925                 }
1926                 /* Don't do anything with foreign layouts */
1927                 RETURN(0);
1928         }
1929
1930         /* normal LOV/layout cases */
1931
1932         if (buf->lb_len < sizeof(*lum)) {
1933                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1934                        buf->lb_len);
1935                 RETURN(-EINVAL);
1936         }
1937
1938         if (magic != LOV_USER_MAGIC_V1 &&
1939             magic != LOV_USER_MAGIC_V3 &&
1940             magic != LOV_USER_MAGIC_SPECIFIC &&
1941             magic != LOV_USER_MAGIC_COMP_V1) {
1942                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1943                        le32_to_cpu(lum->lmm_magic));
1944                 RETURN(-EINVAL);
1945         }
1946
1947         if (magic != LOV_USER_MAGIC_COMP_V1)
1948                 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1949
1950         /* magic == LOV_USER_MAGIC_COMP_V1 */
1951         comp_v1 = buf->lb_buf;
1952         if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1953                 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1954                        buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1955                 RETURN(-EINVAL);
1956         }
1957
1958 recheck:
1959         mirror_count = 0;
1960         if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1961                 CDEBUG(D_LAYOUT, "entry count is zero\n");
1962                 RETURN(-EINVAL);
1963         }
1964
1965         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1966             lo->ldo_comp_cnt > 0) {
1967                 /* could be called from lustre.lov.add */
1968                 __u32 cnt = lo->ldo_comp_cnt;
1969
1970                 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1971                 prev_end = ext->e_end;
1972
1973                 ++mirror_count;
1974         }
1975
1976         for_each_comp_entry_v1(comp_v1, ent) {
1977                 ext = &ent->lcme_extent;
1978
1979                 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
1980                     le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
1981                     (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
1982                     le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
1983                         CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1984                                le64_to_cpu(ext->e_start),
1985                                le64_to_cpu(ext->e_end));
1986                         RETURN(-EINVAL);
1987                 }
1988
1989                 if (is_from_disk) {
1990                         /* lcme_id contains valid value */
1991                         if (le32_to_cpu(ent->lcme_id) == 0 ||
1992                             le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1993                                 CDEBUG(D_LAYOUT, "invalid id %u\n",
1994                                        le32_to_cpu(ent->lcme_id));
1995                                 RETURN(-EINVAL);
1996                         }
1997
1998                         if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1999                                 mirror_id = mirror_id_of(
2000                                                 le32_to_cpu(ent->lcme_id));
2001
2002                                 /* first component must start with 0 */
2003                                 if (mirror_id != prev_mid &&
2004                                     le64_to_cpu(ext->e_start) != 0) {
2005                                         CDEBUG(D_LAYOUT,
2006                                                "invalid start:%llu, expect:0\n",
2007                                                le64_to_cpu(ext->e_start));
2008                                         RETURN(-EINVAL);
2009                                 }
2010
2011                                 prev_mid = mirror_id;
2012                         }
2013                 }
2014
2015                 if (le64_to_cpu(ext->e_start) == 0) {
2016                         ++mirror_count;
2017                         prev_end = 0;
2018                 }
2019
2020                 /* the next must be adjacent with the previous one */
2021                 if (le64_to_cpu(ext->e_start) != prev_end) {
2022                         CDEBUG(D_LAYOUT,
2023                                "invalid start actual:%llu, expect:%llu\n",
2024                                le64_to_cpu(ext->e_start), prev_end);
2025                         RETURN(-EINVAL);
2026                 }
2027
2028                 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
2029                 tmp.lb_len = le32_to_cpu(ent->lcme_size);
2030
2031                 /* Check DoM entry is always the first one */
2032                 lum = tmp.lb_buf;
2033                 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
2034                     LOV_PATTERN_MDT) {
2035                         /* DoM component must be the first in a mirror */
2036                         if (le64_to_cpu(ext->e_start) > 0) {
2037                                 CDEBUG(D_LAYOUT, "invalid DoM component "
2038                                        "with %llu extent start\n",
2039                                        le64_to_cpu(ext->e_start));
2040                                 RETURN(-EINVAL);
2041                         }
2042                         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2043                         /* There is just one stripe on MDT and it must
2044                          * cover whole component size. */
2045                         if (stripe_size != le64_to_cpu(ext->e_end)) {
2046                                 CDEBUG(D_LAYOUT, "invalid DoM layout "
2047                                        "stripe size %u != %llu "
2048                                        "(component size)\n",
2049                                        stripe_size, prev_end);
2050                                 RETURN(-EINVAL);
2051                         }
2052                         /* Check and adjust stripe size by per-MDT limit */
2053                         rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2054                                                        stripe_size);
2055                         /* DoM entry was removed, re-check layout from start */
2056                         if (rc == -ERESTART)
2057                                 goto recheck;
2058                         else if (rc)
2059                                 RETURN(rc);
2060
2061                         /* Any stripe count is forbidden on DoM component */
2062                         if (lum->lmm_stripe_count) {
2063                                 CDEBUG(D_LAYOUT,
2064                                        "invalid DoM layout stripe count %u, must be 0\n",
2065                                        le16_to_cpu(lum->lmm_stripe_count));
2066                                 RETURN(-EINVAL);
2067                         }
2068
2069                         /* Any pool is forbidden on DoM component */
2070                         if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2071                                 struct lov_user_md_v3 *v3 = (void *)lum;
2072
2073                                 if (v3->lmm_pool_name[0] != '\0') {
2074                                         CDEBUG(D_LAYOUT,
2075                                                "DoM component cannot have pool assigned\n");
2076                                         RETURN(-EINVAL);
2077                                 }
2078                         }
2079                 }
2080
2081                 prev_end = le64_to_cpu(ext->e_end);
2082
2083                 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2084                 if (rc)
2085                         RETURN(rc);
2086
2087                 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2088                         continue;
2089
2090                 /* extent end must be aligned with the stripe_size */
2091                 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2092                 if (stripe_size && prev_end % stripe_size) {
2093                         CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2094                                "stripe_sz: %u, [%llu, %llu)\n",
2095                                stripe_size, ext->e_start, prev_end);
2096                         RETURN(-EINVAL);
2097                 }
2098         }
2099
2100         /* make sure that the mirror_count is telling the truth */
2101         if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2102                 RETURN(-EINVAL);
2103
2104         RETURN(0);
2105 }
2106
2107 /**
2108  * set the default stripe size, if unset.
2109  *
2110  * \param[in,out] val   number of bytes per OST stripe
2111  *
2112  * The minimum stripe size is 64KB to ensure that a single stripe is an
2113  * even multiple of a client PAGE_SIZE (IA64, PPC, etc).  Otherwise, it
2114  * is difficult to split dirty pages across OSCs during writes.
2115  */
2116 void lod_fix_desc_stripe_size(__u64 *val)
2117 {
2118         if (*val < LOV_MIN_STRIPE_SIZE) {
2119                 if (*val != 0)
2120                         LCONSOLE_INFO("Increasing default stripe size to "
2121                                       "minimum value %u\n",
2122                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
2123                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2124         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2125                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2126                 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2127                               "multiple of %u)\n",
2128                               *val, LOV_MIN_STRIPE_SIZE);
2129         }
2130 }
2131
2132 /**
2133  * set the filesystem default number of stripes, if unset.
2134  *
2135  * \param[in,out] val   number of stripes
2136  *
2137  * A value of "0" means "use the system-wide default stripe count", which
2138  * has either been inherited by now, or falls back to 1 stripe per file.
2139  * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2140  * and is a valid value, so is left unchanged here.
2141  */
2142 void lod_fix_desc_stripe_count(__u32 *val)
2143 {
2144         if (*val == 0)
2145                 *val = 1;
2146 }
2147
2148 /**
2149  * set the filesystem default layout pattern
2150  *
2151  * \param[in,out] val   LOV_PATTERN_* layout
2152  *
2153  * A value of "0" means "use the system-wide default layout type", which
2154  * has either been inherited by now, or falls back to plain RAID0 striping.
2155  */
2156 void lod_fix_desc_pattern(__u32 *val)
2157 {
2158         /* from lov_setstripe */
2159         if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2160                 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2161                 *val = 0;
2162         }
2163 }
2164
2165 void lod_fix_lmv_desc_pattern(__u32 *val)
2166 {
2167         if ((*val) && !lmv_is_known_hash_type(*val)) {
2168                 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2169                 *val = 0;
2170         }
2171 }
2172
2173 void lod_fix_desc_qos_maxage(__u32 *val)
2174 {
2175         /* fix qos_maxage */
2176         if (*val == 0)
2177                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2178 }
2179
2180 /**
2181  * Used to fix insane default striping.
2182  *
2183  * \param[in] desc      striping description
2184  */
2185 void lod_fix_desc(struct lov_desc *desc)
2186 {
2187         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2188         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2189         lod_fix_desc_pattern(&desc->ld_pattern);
2190         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2191 }
2192
2193 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2194 {
2195         desc->ld_active_tgt_count = 0;
2196         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2197         lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2198         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2199 }
2200
2201 /**
2202  * Initialize the structures used to store pools and default striping.
2203  *
2204  * \param[in] lod       LOD device
2205  * \param[in] lcfg      configuration structure storing default striping.
2206  *
2207  * \retval              0 if initialization succeeds
2208  * \retval              negative error number on failure
2209  */
2210 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2211 {
2212         struct obd_device          *obd;
2213         struct lov_desc            *desc;
2214         int                         rc;
2215         ENTRY;
2216
2217         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2218         LASSERT(obd != NULL);
2219         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2220
2221         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2222                 CERROR("LOD setup requires a descriptor\n");
2223                 RETURN(-EINVAL);
2224         }
2225
2226         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2227
2228         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2229                 CERROR("descriptor size wrong: %d > %d\n",
2230                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2231                 RETURN(-EINVAL);
2232         }
2233
2234         if (desc->ld_magic != LOV_DESC_MAGIC) {
2235                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2236                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2237                                obd->obd_name, desc);
2238                         lustre_swab_lov_desc(desc);
2239                 } else {
2240                         CERROR("%s: Bad lov desc magic: %#x\n",
2241                                obd->obd_name, desc->ld_magic);
2242                         RETURN(-EINVAL);
2243                 }
2244         }
2245
2246         lod_fix_desc(desc);
2247
2248         desc->ld_active_tgt_count = 0;
2249         lod->lod_ost_descs.ltd_lov_desc = *desc;
2250
2251         /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2252         lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2253
2254         lod->lod_sp_me = LUSTRE_SP_CLI;
2255
2256         /* Set up OST pool environment */
2257         lod->lod_pool_count = 0;
2258         rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2259         if (rc)
2260                 RETURN(-ENOMEM);
2261
2262         INIT_LIST_HEAD(&lod->lod_pool_list);
2263         lod->lod_pool_count = 0;
2264         rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2265         if (rc)
2266                 GOTO(out_hash, rc);
2267
2268         rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2269         if (rc)
2270                 GOTO(out_mdt_pool, rc);
2271
2272         rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2273         if (rc)
2274                 GOTO(out_mdt_rr_pool, rc);
2275
2276         rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2277         if (rc)
2278                 GOTO(out_ost_pool, rc);
2279
2280         RETURN(0);
2281
2282 out_ost_pool:
2283         lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2284 out_mdt_rr_pool:
2285         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2286 out_mdt_pool:
2287         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2288 out_hash:
2289         lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2290
2291         return rc;
2292 }
2293
2294 /**
2295  * Release the structures describing the pools.
2296  *
2297  * \param[in] lod       LOD device from which we release the structures
2298  *
2299  * \retval              0 always
2300  */
2301 int lod_pools_fini(struct lod_device *lod)
2302 {
2303         struct obd_device   *obd = lod2obd(lod);
2304         struct pool_desc    *pool, *tmp;
2305         ENTRY;
2306
2307         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2308                 /* free pool structs */
2309                 CDEBUG(D_INFO, "delete pool %p\n", pool);
2310                 /* In the function below, .hs_keycmp resolves to
2311                  * pool_hashkey_keycmp() */
2312                 /* coverity[overrun-buffer-val] */
2313                 lod_pool_del(obd, pool->pool_name);
2314         }
2315
2316         lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2317         lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2318         lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2319         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2320         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2321
2322         RETURN(0);
2323 }