Whamcloud - gitweb
LU-7065 lod: Fix free of already added target description
[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, 2014, 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
43 #include "lod_internal.h"
44
45 /**
46  * Increase reference count on the target table.
47  *
48  * Increase reference count on the target table usage to prevent racing with
49  * addition/deletion. Any function that expects the table to remain
50  * stationary must take a ref.
51  *
52  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
53  */
54 void lod_getref(struct lod_tgt_descs *ltd)
55 {
56         down_read(&ltd->ltd_rw_sem);
57         mutex_lock(&ltd->ltd_mutex);
58         ltd->ltd_refcount++;
59         mutex_unlock(&ltd->ltd_mutex);
60 }
61
62 /**
63  * Decrease reference count on the target table.
64  *
65  * Companion of lod_getref() to release a reference on the target table.
66  * If this is the last reference and the OST entry was scheduled for deletion,
67  * the descriptor is removed from the table.
68  *
69  * \param[in] lod       LOD device from which we release a reference
70  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
71  */
72 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
73 {
74         mutex_lock(&ltd->ltd_mutex);
75         ltd->ltd_refcount--;
76         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
77                 struct lod_tgt_desc *tgt_desc, *tmp;
78                 struct list_head kill;
79                 unsigned int idx;
80
81                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
82                        ltd->ltd_death_row);
83
84                 INIT_LIST_HEAD(&kill);
85
86                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
87                         tgt_desc = LTD_TGT(ltd, idx);
88                         LASSERT(tgt_desc);
89
90                         if (!tgt_desc->ltd_reap)
91                                 continue;
92
93                         list_add(&tgt_desc->ltd_kill, &kill);
94                         LTD_TGT(ltd, idx) = NULL;
95                         /*FIXME: only support ost pool for now */
96                         if (ltd == &lod->lod_ost_descs) {
97                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
98                                 if (tgt_desc->ltd_active)
99                                         lod->lod_desc.ld_active_tgt_count--;
100                         }
101                         ltd->ltd_tgtnr--;
102                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
103                         ltd->ltd_death_row--;
104                 }
105                 mutex_unlock(&ltd->ltd_mutex);
106                 up_read(&ltd->ltd_rw_sem);
107
108                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
109                         int rc;
110                         list_del(&tgt_desc->ltd_kill);
111                         if (ltd == &lod->lod_ost_descs) {
112                                 /* remove from QoS structures */
113                                 rc = qos_del_tgt(lod, tgt_desc);
114                                 if (rc)
115                                         CERROR("%s: qos_del_tgt(%s) failed:"
116                                                "rc = %d\n",
117                                                lod2obd(lod)->obd_name,
118                                               obd_uuid2str(&tgt_desc->ltd_uuid),
119                                                rc);
120                         }
121                         rc = obd_disconnect(tgt_desc->ltd_exp);
122                         if (rc)
123                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
124                                        lod2obd(lod)->obd_name,
125                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
126                         OBD_FREE_PTR(tgt_desc);
127                 }
128         } else {
129                 mutex_unlock(&ltd->ltd_mutex);
130                 up_read(&ltd->ltd_rw_sem);
131         }
132 }
133
134 /**
135  * Expand size of target table.
136  *
137  * When the target table is full, we have to extend the table. To do so,
138  * we allocate new memory with some reserve, move data from the old table
139  * to the new one and release memory consumed by the old table.
140  * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
141  *
142  * \param[in] ltd               target table
143  * \param[in] newsize           new size of the table
144  *
145  * \retval                      0 on success
146  * \retval                      -ENOMEM if reallocation failed
147  */
148 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
149 {
150         cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
151         int           rc = 0;
152         ENTRY;
153
154         /* grab write reference on the lod. Relocating the array requires
155          * exclusive access */
156
157         down_write(&ltd->ltd_rw_sem);
158         if (newsize <= ltd->ltd_tgts_size)
159                 /* someone else has already resize the array */
160                 GOTO(out, rc = 0);
161
162         /* allocate new bitmap */
163         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
164         if (!new_bitmap)
165                 GOTO(out, rc = -ENOMEM);
166
167         if (ltd->ltd_tgts_size > 0) {
168                 /* the bitmap already exists, we need
169                  * to copy data from old one */
170                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
171                 old_bitmap = ltd->ltd_tgt_bitmap;
172         }
173
174         ltd->ltd_tgts_size  = newsize;
175         ltd->ltd_tgt_bitmap = new_bitmap;
176
177         if (old_bitmap)
178                 CFS_FREE_BITMAP(old_bitmap);
179
180         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
181
182         EXIT;
183 out:
184         up_write(&ltd->ltd_rw_sem);
185         return rc;
186 }
187
188 /**
189  * Connect LOD to a new OSP and add it to the target table.
190  *
191  * Connect to the OSP device passed, initialize all the internal
192  * structures related to the device and add it to the target table.
193  *
194  * \param[in] env               execution environment for this thread
195  * \param[in] lod               LOD device to be connected to the new OSP
196  * \param[in] osp               name of OSP device name to be added
197  * \param[in] index             index of the new target
198  * \param[in] gen               target's generation number
199  * \param[in] tgt_index         OSP's group
200  * \param[in] type              type of device (mdc or osc)
201  * \param[in] active            state of OSP: 0 - inactive, 1 - active
202  *
203  * \retval                      0 if added successfully
204  * \retval                      negative error number on failure
205  */
206 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
207                    char *osp, unsigned index, unsigned gen, int tgt_index,
208                    char *type, int active)
209 {
210         struct obd_connect_data *data = NULL;
211         struct obd_export       *exp = NULL;
212         struct obd_device       *obd;
213         struct lu_device        *ldev;
214         struct dt_device        *d;
215         int                      rc;
216         struct lod_tgt_desc     *tgt_desc;
217         struct lod_tgt_descs    *ltd;
218         struct obd_uuid         obd_uuid;
219         bool                    for_ost;
220         bool lock = false;
221         ENTRY;
222
223         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
224
225         if (gen <= 0) {
226                 CERROR("request to add OBD %s with invalid generation: %d\n",
227                        osp, gen);
228                 RETURN(-EINVAL);
229         }
230
231         obd_str2uuid(&obd_uuid, osp);
232
233         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
234                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
235         if (obd == NULL) {
236                 CERROR("can't find %s device\n", osp);
237                 RETURN(-EINVAL);
238         }
239
240         OBD_ALLOC_PTR(data);
241         if (data == NULL)
242                 RETURN(-ENOMEM);
243
244         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
245         data->ocd_version = LUSTRE_VERSION_CODE;
246         data->ocd_index = index;
247
248         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
249                 for_ost = true;
250                 data->ocd_connect_flags |= OBD_CONNECT_AT |
251                                            OBD_CONNECT_FULL20 |
252                                            OBD_CONNECT_INDEX |
253 #ifdef HAVE_LRU_RESIZE_SUPPORT
254                                            OBD_CONNECT_LRU_RESIZE |
255 #endif
256                                            OBD_CONNECT_MDS |
257                                            OBD_CONNECT_REQPORTAL |
258                                            OBD_CONNECT_SKIP_ORPHAN |
259                                            OBD_CONNECT_FID |
260                                            OBD_CONNECT_LVB_TYPE |
261                                            OBD_CONNECT_VERSION |
262                                            OBD_CONNECT_PINGLESS |
263                                            OBD_CONNECT_LFSCK |
264                                            OBD_CONNECT_BULK_MBITS;
265
266                 data->ocd_group = tgt_index;
267                 ltd = &lod->lod_ost_descs;
268         } else {
269                 struct obd_import *imp = obd->u.cli.cl_import;
270
271                 for_ost = false;
272                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
273                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
274                                            OBD_CONNECT_IBITS |
275                                            OBD_CONNECT_MDS_MDS |
276                                            OBD_CONNECT_FID |
277                                            OBD_CONNECT_AT |
278                                            OBD_CONNECT_FULL20 |
279                                            OBD_CONNECT_LFSCK |
280                                            OBD_CONNECT_BULK_MBITS;
281                 spin_lock(&imp->imp_lock);
282                 imp->imp_server_timeout = 1;
283                 spin_unlock(&imp->imp_lock);
284                 imp->imp_client->cli_request_portal = OUT_PORTAL;
285                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
286                       obd->obd_name);
287                 ltd = &lod->lod_mdt_descs;
288         }
289
290         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
291         OBD_FREE_PTR(data);
292         if (rc) {
293                 CERROR("%s: cannot connect to next dev %s (%d)\n",
294                        obd->obd_name, osp, rc);
295                 GOTO(out_free, rc);
296         }
297
298         LASSERT(obd->obd_lu_dev);
299         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
300
301         ldev = obd->obd_lu_dev;
302         d = lu2dt_dev(ldev);
303
304         /* Allocate ost descriptor and fill it */
305         OBD_ALLOC_PTR(tgt_desc);
306         if (!tgt_desc)
307                 GOTO(out_conn, rc = -ENOMEM);
308
309         tgt_desc->ltd_tgt    = d;
310         tgt_desc->ltd_exp    = exp;
311         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
312         tgt_desc->ltd_gen    = gen;
313         tgt_desc->ltd_index  = index;
314         tgt_desc->ltd_active = active;
315
316         lod_getref(ltd);
317         if (index >= ltd->ltd_tgts_size) {
318                 /* we have to increase the size of the lod_osts array */
319                 __u32  newsize;
320
321                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
322                 while (newsize < index + 1)
323                         newsize = newsize << 1;
324
325                 /* lod_bitmap_resize() needs lod_rw_sem
326                  * which we hold with th reference */
327                 lod_putref(lod, ltd);
328
329                 rc = ltd_bitmap_resize(ltd, newsize);
330                 if (rc)
331                         GOTO(out_desc, rc);
332
333                 lod_getref(ltd);
334         }
335
336         mutex_lock(&ltd->ltd_mutex);
337         lock = true;
338         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
339                 CERROR("%s: device %d is registered already\n", obd->obd_name,
340                        index);
341                 GOTO(out_mutex, rc = -EEXIST);
342         }
343
344         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
345                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
346                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
347                         CERROR("can't allocate index to add %s\n",
348                                obd->obd_name);
349                         GOTO(out_mutex, rc = -ENOMEM);
350                 }
351         }
352
353         if (for_ost) {
354                 /* pool and qos are not supported for MDS stack yet */
355                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
356                                       lod->lod_osts_size);
357                 if (rc) {
358                         CERROR("%s: can't set up pool, failed with %d\n",
359                                obd->obd_name, rc);
360                         GOTO(out_mutex, rc);
361                 }
362
363                 rc = qos_add_tgt(lod, tgt_desc);
364                 if (rc) {
365                         CERROR("%s: qos_add_tgt failed with %d\n",
366                                 obd->obd_name, rc);
367                         GOTO(out_pool, rc);
368                 }
369
370                 /* The new OST is now a full citizen */
371                 if (index >= lod->lod_desc.ld_tgt_count)
372                         lod->lod_desc.ld_tgt_count = index + 1;
373                 if (active)
374                         lod->lod_desc.ld_active_tgt_count++;
375         }
376
377         LTD_TGT(ltd, index) = tgt_desc;
378         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
379         ltd->ltd_tgtnr++;
380         mutex_unlock(&ltd->ltd_mutex);
381         lod_putref(lod, ltd);
382         lock = false;
383         if (lod->lod_recovery_completed)
384                 ldev->ld_ops->ldo_recovery_complete(env, ldev);
385
386         if (!for_ost && lod->lod_initialized) {
387                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
388                 if (rc != 0) {
389                         CERROR("%s: cannot start llog on %s:rc = %d\n",
390                                lod2obd(lod)->obd_name, osp, rc);
391                         GOTO(out_ltd, rc);
392                 }
393         }
394
395         rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost);
396         if (rc != 0) {
397                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
398                        "index = %u, rc = %d\n", osp, type, index, rc);
399                 GOTO(out_fini_llog, rc);
400         }
401         RETURN(rc);
402 out_fini_llog:
403         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
404                           tgt_desc->ltd_recovery_thread);
405 out_ltd:
406         lod_getref(ltd);
407         mutex_lock(&ltd->ltd_mutex);
408         lock = true;
409         if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
410                 struct ptlrpc_thread *thread;
411
412                 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
413                 OBD_FREE_PTR(thread);
414         }
415         ltd->ltd_tgtnr--;
416         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
417         LTD_TGT(ltd, index) = NULL;
418 out_pool:
419         lod_ost_pool_remove(&lod->lod_pool_info, index);
420 out_mutex:
421         if (lock) {
422                 mutex_unlock(&ltd->ltd_mutex);
423                 lod_putref(lod, ltd);
424         }
425 out_desc:
426         OBD_FREE_PTR(tgt_desc);
427 out_conn:
428         obd_disconnect(exp);
429 out_free:
430         return rc;
431 }
432
433 /**
434  * Schedule target removal from the target table.
435  *
436  * Mark the device as dead. The device is not removed here because it may
437  * still be in use. The device will be removed in lod_putref() when the
438  * last reference is released.
439  *
440  * \param[in] env               execution environment for this thread
441  * \param[in] lod               LOD device the target table belongs to
442  * \param[in] ltd               target table
443  * \param[in] idx               index of the target
444  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
445  */
446 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
447                              struct lod_tgt_descs *ltd, unsigned idx,
448                              bool for_ost)
449 {
450         LASSERT(LTD_TGT(ltd, idx));
451
452         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
453                          idx, for_ost);
454
455         if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
456                 struct ptlrpc_thread *thread;
457
458                 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
459                 OBD_FREE_PTR(thread);
460         }
461
462         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
463                 LTD_TGT(ltd, idx)->ltd_reap = 1;
464                 ltd->ltd_death_row++;
465         }
466 }
467
468 /**
469  * Schedule removal of all the targets from the given target table.
470  *
471  * See more details in the description for __lod_del_device()
472  *
473  * \param[in] env               execution environment for this thread
474  * \param[in] lod               LOD device the target table belongs to
475  * \param[in] ltd               target table
476  * \param[in] for_ost           type of the target: MDT or OST
477  *
478  * \retval                      0 always
479  */
480 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
481                  struct lod_tgt_descs *ltd, bool for_ost)
482 {
483         unsigned int idx;
484
485         if (ltd->ltd_tgts_size <= 0)
486                 return 0;
487         lod_getref(ltd);
488         mutex_lock(&ltd->ltd_mutex);
489         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
490                 __lod_del_device(env, lod, ltd, idx, for_ost);
491         mutex_unlock(&ltd->ltd_mutex);
492         lod_putref(lod, ltd);
493         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
494         for (idx = 0; idx < TGT_PTRS; idx++) {
495                 if (ltd->ltd_tgt_idx[idx])
496                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
497         }
498         ltd->ltd_tgts_size = 0;
499         return 0;
500 }
501
502 /**
503  * Remove device by name.
504  *
505  * Remove a device identified by \a osp from the target table. Given
506  * the device can be in use, the real deletion happens in lod_putref().
507  *
508  * \param[in] env               execution environment for this thread
509  * \param[in] lod               LOD device to be connected to the new OSP
510  * \param[in] ltd               target table
511  * \param[in] osp               name of OSP device to be removed
512  * \param[in] idx               index of the target
513  * \param[in] gen               generation number, not used currently
514  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
515  *
516  * \retval                      0 if the device was scheduled for removal
517  * \retval                      -EINVAL if no device was found
518  */
519 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
520                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
521                    unsigned gen, bool for_ost)
522 {
523         struct obd_device *obd;
524         int                rc = 0;
525         struct obd_uuid    uuid;
526         ENTRY;
527
528         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
529
530         obd_str2uuid(&uuid, osp);
531
532         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
533                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
534         if (obd == NULL) {
535                 CERROR("can't find %s device\n", osp);
536                 RETURN(-EINVAL);
537         }
538
539         if (gen <= 0) {
540                 CERROR("%s: request to remove OBD %s with invalid generation %d"
541                        "\n", obd->obd_name, osp, gen);
542                 RETURN(-EINVAL);
543         }
544
545         obd_str2uuid(&uuid,  osp);
546
547         lod_getref(ltd);
548         mutex_lock(&ltd->ltd_mutex);
549         /* check that the index is allocated in the bitmap */
550         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
551             !LTD_TGT(ltd, idx)) {
552                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
553                 GOTO(out, rc = -EINVAL);
554         }
555
556         /* check that the UUID matches */
557         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
558                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
559                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
560                        idx, osp);
561                 GOTO(out, rc = -EINVAL);
562         }
563
564         __lod_del_device(env, lod, ltd, idx, for_ost);
565         EXIT;
566 out:
567         mutex_unlock(&ltd->ltd_mutex);
568         lod_putref(lod, ltd);
569         return(rc);
570 }
571
572 /**
573  * Resize per-thread storage to hold specified size.
574  *
575  * A helper function to resize per-thread temporary storage. This storage
576  * is used to process LOV/LVM EAs and may be quite large. We do not want to
577  * allocate/release it every time, so instead we put it into the env and
578  * reallocate on demand. The memory is released when the correspondent thread
579  * is finished.
580  *
581  * \param[in] info              LOD-specific storage in the environment
582  * \param[in] size              new size to grow the buffer to
583
584  * \retval                      0 on success, -ENOMEM if reallocation failed
585  */
586 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
587 {
588         __u32 round = size_roundup_power2(size);
589
590         LASSERT(round <=
591                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
592         if (info->lti_ea_store) {
593                 LASSERT(info->lti_ea_store_size);
594                 LASSERT(info->lti_ea_store_size < round);
595                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
596                        info->lti_ea_store_size, round);
597                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
598                 info->lti_ea_store = NULL;
599                 info->lti_ea_store_size = 0;
600         }
601
602         OBD_ALLOC_LARGE(info->lti_ea_store, round);
603         if (info->lti_ea_store == NULL)
604                 RETURN(-ENOMEM);
605         info->lti_ea_store_size = round;
606         RETURN(0);
607 }
608
609 /**
610  * Make LOV EA for striped object.
611  *
612  * Generate striping information and store it in the LOV EA of the given
613  * object. The caller must ensure nobody else is calling the function
614  * against the object concurrently. The transaction must be started.
615  * FLDB service must be running as well; it's used to map FID to the target,
616  * which is stored in LOV EA.
617  *
618  * \param[in] env               execution environment for this thread
619  * \param[in] lo                LOD object
620  * \param[in] th                transaction handle
621  *
622  * \retval                      0 if LOV EA is stored successfully
623  * \retval                      negative error number on failure
624  */
625 int lod_generate_and_set_lovea(const struct lu_env *env,
626                                struct lod_object *lo, struct thandle *th)
627 {
628         struct lod_thread_info  *info = lod_env_info(env);
629         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
630         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
631         struct lov_mds_md_v1    *lmm;
632         struct lov_ost_data_v1  *objs;
633         __u32                    magic;
634         int                      i, rc;
635         size_t                   lmm_size;
636         ENTRY;
637
638         LASSERT(lo);
639
640         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
641         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
642         if (info->lti_ea_store_size < lmm_size) {
643                 rc = lod_ea_store_resize(info, lmm_size);
644                 if (rc)
645                         RETURN(rc);
646         }
647
648         if (lo->ldo_pattern == 0) /* default striping */
649                 lo->ldo_pattern = LOV_PATTERN_RAID0;
650
651         lmm = info->lti_ea_store;
652
653         lmm->lmm_magic = cpu_to_le32(magic);
654         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
655         fid_to_lmm_oi(fid, &lmm->lmm_oi);
656         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
657                 lmm->lmm_oi.oi.oi_id++;
658         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
659         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
660         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
661         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
662                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
663         lmm->lmm_layout_gen = 0;
664         if (magic == LOV_MAGIC_V1) {
665                 objs = &lmm->lmm_objects[0];
666         } else {
667                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
668                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
669                                 sizeof(v3->lmm_pool_name));
670                 if (cplen >= sizeof(v3->lmm_pool_name))
671                         RETURN(-E2BIG);
672                 objs = &v3->lmm_objects[0];
673         }
674
675         for (i = 0; i < lo->ldo_stripenr; i++) {
676                 struct lu_fid           *fid    = &info->lti_fid;
677                 struct lod_device       *lod;
678                 __u32                   index;
679                 int                     type    = LU_SEQ_RANGE_OST;
680
681                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
682                 LASSERT(lo->ldo_stripe[i]);
683
684                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
685                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
686                         if (cfs_fail_val == 0)
687                                 cfs_fail_val = fid->f_oid;
688                         else
689                                 fid->f_oid = cfs_fail_val;
690                 }
691
692                 rc = fid_to_ostid(fid, &info->lti_ostid);
693                 LASSERT(rc == 0);
694
695                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
696                 objs[i].l_ost_gen    = cpu_to_le32(0);
697                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
698                 if (rc < 0) {
699                         CERROR("%s: Can not locate "DFID": rc = %d\n",
700                                lod2obd(lod)->obd_name, PFID(fid), rc);
701                         lod_object_free_striping(env, lo);
702                         RETURN(rc);
703                 }
704                 objs[i].l_ost_idx = cpu_to_le32(index);
705         }
706
707         info->lti_buf.lb_buf = lmm;
708         info->lti_buf.lb_len = lmm_size;
709         rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
710                                       0, th);
711         if (rc < 0) {
712                 lod_object_free_striping(env, lo);
713                 RETURN(rc);
714         }
715
716         RETURN(rc);
717 }
718
719 /**
720  * Get LOV EA.
721  *
722  * Fill lti_ea_store buffer in the environment with a value for the given
723  * EA. The buffer is reallocated if the value doesn't fit.
724  *
725  * \param[in,out] env           execution environment for this thread
726  *                              .lti_ea_store buffer is filled with EA's value
727  * \param[in] lo                LOD object
728  * \param[in] name              name of the EA
729  *
730  * \retval                      0 if EA is fetched successfully
731  * \retval                      negative error number on failure
732  */
733 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
734                const char *name)
735 {
736         struct lod_thread_info  *info = lod_env_info(env);
737         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
738         int                     rc;
739         ENTRY;
740
741         LASSERT(info);
742
743         if (unlikely(info->lti_ea_store == NULL)) {
744                 /* just to enter in allocation block below */
745                 rc = -ERANGE;
746         } else {
747 repeat:
748                 info->lti_buf.lb_buf = info->lti_ea_store;
749                 info->lti_buf.lb_len = info->lti_ea_store_size;
750                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
751         }
752
753         /* if object is not striped or inaccessible */
754         if (rc == -ENODATA || rc == -ENOENT)
755                 RETURN(0);
756
757         if (rc == -ERANGE) {
758                 /* EA doesn't fit, reallocate new buffer */
759                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
760                 if (rc == -ENODATA || rc == -ENOENT)
761                         RETURN(0);
762                 else if (rc < 0)
763                         RETURN(rc);
764
765                 LASSERT(rc > 0);
766                 rc = lod_ea_store_resize(info, rc);
767                 if (rc)
768                         RETURN(rc);
769                 goto repeat;
770         }
771
772         RETURN(rc);
773 }
774
775 /**
776  * Verify the target index is present in the current configuration.
777  *
778  * \param[in] md                LOD device where the target table is stored
779  * \param[in] idx               target's index
780  *
781  * \retval                      0 if the index is present
782  * \retval                      -EINVAL if not
783  */
784 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
785 {
786         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
787                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
788                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
789                        md->lod_ost_descs.ltd_tgts_size);
790                 return -EINVAL;
791         }
792
793         if (unlikely(OST_TGT(md, idx) == NULL)) {
794                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
795                        lod2obd(md)->obd_name, idx);
796                 return -EINVAL;
797         }
798
799         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
800                 CERROR("%s: invalid lod device, for idx: %d\n",
801                        lod2obd(md)->obd_name , idx);
802                 return -EINVAL;
803         }
804
805         return 0;
806 }
807
808 /**
809  * Instantiate objects for stripes.
810  *
811  * Allocate and initialize LU-objects representing the stripes. The number
812  * of the stripes (ldo_stripenr) must be initialized already. The caller
813  * must ensure nobody else is calling the function on the object at the same
814  * time. FLDB service must be running to be able to map a FID to the targets
815  * and find appropriate device representing that target.
816  *
817  * \param[in] env               execution environment for this thread
818  * \param[in,out] lo            LOD object
819  * \param[in] objs              an array of IDs to creates the objects from
820  *
821  * \retval                      0 if the objects are instantiated successfully
822  * \retval                      negative error number on failure
823  */
824 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
825                            struct lov_ost_data_v1 *objs)
826 {
827         struct lod_thread_info  *info = lod_env_info(env);
828         struct lod_device       *md;
829         struct lu_object        *o, *n;
830         struct lu_device        *nd;
831         struct dt_object       **stripe;
832         int                      stripe_len;
833         int                      i, rc = 0;
834         __u32                   idx;
835         ENTRY;
836
837         LASSERT(lo != NULL);
838         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
839         LASSERT(lo->ldo_stripe == NULL);
840         LASSERT(lo->ldo_stripenr > 0);
841         LASSERT(lo->ldo_stripe_size > 0);
842
843         stripe_len = lo->ldo_stripenr;
844         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
845         if (stripe == NULL)
846                 RETURN(-ENOMEM);
847
848         for (i = 0; i < lo->ldo_stripenr; i++) {
849                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
850                         continue;
851
852                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
853                 idx = le32_to_cpu(objs[i].l_ost_idx);
854                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
855                 if (rc != 0)
856                         GOTO(out, rc);
857                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
858                          PFID(&info->lti_fid));
859                 lod_getref(&md->lod_ost_descs);
860
861                 rc = validate_lod_and_idx(md, idx);
862                 if (unlikely(rc != 0)) {
863                         lod_putref(md, &md->lod_ost_descs);
864                         GOTO(out, rc);
865                 }
866
867                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
868                 lod_putref(md, &md->lod_ost_descs);
869
870                 /* In the function below, .hs_keycmp resolves to
871                  * u_obj_hop_keycmp() */
872                 /* coverity[overrun-buffer-val] */
873                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
874                 if (IS_ERR(o))
875                         GOTO(out, rc = PTR_ERR(o));
876
877                 n = lu_object_locate(o->lo_header, nd->ld_type);
878                 LASSERT(n);
879
880                 stripe[i] = container_of(n, struct dt_object, do_lu);
881         }
882
883 out:
884         if (rc != 0) {
885                 for (i = 0; i < stripe_len; i++)
886                         if (stripe[i] != NULL)
887                                 lu_object_put(env, &stripe[i]->do_lu);
888
889                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
890                 lo->ldo_stripenr = 0;
891         } else {
892                 lo->ldo_stripe = stripe;
893                 lo->ldo_stripes_allocated = stripe_len;
894         }
895
896         RETURN(rc);
897 }
898
899 /**
900  * Instantiate objects for striping.
901  *
902  * Parse striping information in \a buf and instantiate the objects
903  * representing the stripes.
904  *
905  * \param[in] env               execution environment for this thread
906  * \param[in] lo                LOD object
907  * \param[in] buf               buffer storing LOV EA to parse
908  *
909  * \retval                      0 if parsing and objects creation succeed
910  * \retval                      negative error number on failure
911  */
912 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
913                        const struct lu_buf *buf)
914 {
915         struct lov_mds_md_v1    *lmm;
916         struct lov_ost_data_v1  *objs;
917         __u32                    magic;
918         __u32                    pattern;
919         int                      rc = 0;
920         ENTRY;
921
922         LASSERT(buf);
923         LASSERT(buf->lb_buf);
924         LASSERT(buf->lb_len);
925
926         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
927         magic = le32_to_cpu(lmm->lmm_magic);
928         pattern = le32_to_cpu(lmm->lmm_pattern);
929
930         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
931                 GOTO(out, rc = -EINVAL);
932         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
933                 GOTO(out, rc = -EINVAL);
934
935         lo->ldo_pattern = pattern;
936         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
937         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
938         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
939         /* released file stripenr fixup. */
940         if (pattern & LOV_PATTERN_F_RELEASED)
941                 lo->ldo_stripenr = 0;
942
943         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
944
945         if (magic == LOV_MAGIC_V3) {
946                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
947                 objs = &v3->lmm_objects[0];
948                 lod_object_set_pool(lo, v3->lmm_pool_name);
949         } else {
950                 objs = &lmm->lmm_objects[0];
951         }
952
953         if (lo->ldo_stripenr > 0)
954                 rc = lod_initialize_objects(env, lo, objs);
955
956 out:
957         RETURN(rc);
958 }
959
960 /**
961  * Initialize the object representing the stripes.
962  *
963  * Unless the stripes are initialized already, fetch LOV (for regular
964  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
965  * to instantiate the objects representing the stripes.
966  *
967  * \param[in] env               execution environment for this thread
968  * \param[in,out] lo            LOD object
969  *
970  * \retval                      0 if parsing and object creation succeed
971  * \retval                      negative error number on failure
972  */
973 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
974 {
975         struct lod_thread_info  *info = lod_env_info(env);
976         struct lu_buf           *buf  = &info->lti_buf;
977         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
978         int                      rc = 0;
979         ENTRY;
980
981         /* already initialized? */
982         if (lo->ldo_stripe != NULL)
983                 GOTO(out, rc = 0);
984
985         if (!dt_object_exists(next))
986                 GOTO(out, rc = 0);
987
988         /* Do not load stripe for slaves of striped dir */
989         if (lo->ldo_dir_slave_stripe)
990                 GOTO(out, rc = 0);
991
992         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
993                 rc = lod_get_lov_ea(env, lo);
994                 if (rc <= 0)
995                         GOTO(out, rc);
996                 /*
997                  * there is LOV EA (striping information) in this object
998                  * let's parse it and create in-core objects for the stripes
999                  */
1000                 buf->lb_buf = info->lti_ea_store;
1001                 buf->lb_len = info->lti_ea_store_size;
1002                 rc = lod_parse_striping(env, lo, buf);
1003         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
1004                 rc = lod_get_lmv_ea(env, lo);
1005                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1006                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1007
1008                 buf->lb_buf = info->lti_ea_store;
1009                 buf->lb_len = info->lti_ea_store_size;
1010                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1011                         rc = lod_load_lmv_shards(env, lo, buf, true);
1012                         if (buf->lb_buf != info->lti_ea_store) {
1013                                 OBD_FREE_LARGE(info->lti_ea_store,
1014                                                info->lti_ea_store_size);
1015                                 info->lti_ea_store = buf->lb_buf;
1016                                 info->lti_ea_store_size = buf->lb_len;
1017                         }
1018
1019                         if (rc < 0)
1020                                 GOTO(out, rc);
1021                 }
1022
1023                 /*
1024                  * there is LOV EA (striping information) in this object
1025                  * let's parse it and create in-core objects for the stripes
1026                  */
1027                 rc = lod_parse_dir_striping(env, lo, buf);
1028         }
1029
1030         if (rc == 0)
1031                 lo->ldo_striping_cached = 1;
1032 out:
1033         RETURN(rc);
1034 }
1035
1036 /**
1037  * A generic function to initialize the stripe objects.
1038  *
1039  * A protected version of lod_load_striping_locked() - load the striping
1040  * information from storage, parse that and instantiate LU objects to
1041  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1042  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1043  * return an array of references to the newly instantiated objects.
1044  *
1045  * \param[in] env               execution environment for this thread
1046  * \param[in,out] lo            LOD object, where striping is stored and
1047  *                              which gets an array of references
1048  *
1049  * \retval                      0 if parsing and object creation succeed
1050  * \retval                      negative error number on failure
1051  **/
1052 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1053 {
1054         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1055         int                     rc = 0;
1056
1057         /* currently this code is supposed to be called from declaration
1058          * phase only, thus the object is not expected to be locked by caller */
1059         dt_write_lock(env, next, 0);
1060         rc = lod_load_striping_locked(env, lo);
1061         dt_write_unlock(env, next);
1062         return rc;
1063 }
1064
1065 /**
1066  * Verify striping.
1067  *
1068  * Check the validity of all fields including the magic, stripe size,
1069  * stripe count, stripe offset and that the pool is present.  Also check
1070  * that each target index points to an existing target. The additional
1071  * \a is_from_disk turns additional checks. In some cases zero fields
1072  * are allowed (like pattern=0).
1073  *
1074  * \param[in] d                 LOD device
1075  * \param[in] buf               buffer with LOV EA to verify
1076  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1077  *                              1 - from disk, do not allow
1078  *
1079  * \retval                      0 if the striping is valid
1080  * \retval                      -EINVAL if striping is invalid
1081  */
1082 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1083                         bool is_from_disk)
1084 {
1085         struct lov_user_md_v1   *lum;
1086         struct lov_user_md_v3   *lum3;
1087         struct pool_desc        *pool = NULL;
1088         __u32                    magic;
1089         __u32                    stripe_size;
1090         __u16                    stripe_count;
1091         __u16                    stripe_offset;
1092         size_t                   lum_size;
1093         int                      rc = 0;
1094         ENTRY;
1095
1096         lum = buf->lb_buf;
1097
1098         LASSERT(sizeof(*lum) < sizeof(*lum3));
1099
1100         if (buf->lb_len < sizeof(*lum)) {
1101                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n",
1102                        buf->lb_len);
1103                 GOTO(out, rc = -EINVAL);
1104         }
1105
1106         magic = le32_to_cpu(lum->lmm_magic);
1107         if (magic != LOV_USER_MAGIC_V1 &&
1108             magic != LOV_USER_MAGIC_V3 &&
1109             magic != LOV_MAGIC_V1_DEF &&
1110             magic != LOV_MAGIC_V3_DEF) {
1111                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1112                 GOTO(out, rc = -EINVAL);
1113         }
1114
1115         /* the user uses "0" for default stripe pattern normally. */
1116         if (!is_from_disk && lum->lmm_pattern == 0)
1117                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1118
1119         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1120                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1121                        le32_to_cpu(lum->lmm_pattern));
1122                 GOTO(out, rc = -EINVAL);
1123         }
1124
1125         /* 64kB is the largest common page size we see (ia64), and matches the
1126          * check in lfs */
1127         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1128         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1129                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1130                        stripe_size, LOV_MIN_STRIPE_SIZE);
1131                 GOTO(out, rc = -EINVAL);
1132         }
1133
1134         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1135         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1136                 /* if offset is not within valid range [0, osts_size) */
1137                 if (stripe_offset >= d->lod_osts_size) {
1138                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1139                                stripe_offset, d->lod_osts_size);
1140                         GOTO(out, rc = -EINVAL);
1141                 }
1142
1143                 /* if lmm_stripe_offset is *not* in bitmap */
1144                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1145                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1146                                stripe_offset);
1147                         GOTO(out, rc = -EINVAL);
1148                 }
1149         }
1150
1151         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1152                 lum_size = offsetof(struct lov_user_md_v1,
1153                                     lmm_objects[0]);
1154         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1155                 lum_size = offsetof(struct lov_user_md_v3,
1156                                     lmm_objects[0]);
1157         else
1158                 GOTO(out, rc = -EINVAL);
1159
1160         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1161         if (buf->lb_len != lum_size) {
1162                 CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with "
1163                        "magic %#x and stripe_count %u\n",
1164                        buf->lb_len, magic, stripe_count);
1165                 GOTO(out, rc = -EINVAL);
1166         }
1167
1168         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1169                 goto out;
1170
1171         lum3 = buf->lb_buf;
1172         if (buf->lb_len < sizeof(*lum3)) {
1173                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n",
1174                        buf->lb_len);
1175                 GOTO(out, rc = -EINVAL);
1176         }
1177
1178         /* In the function below, .hs_keycmp resolves to
1179          * pool_hashkey_keycmp() */
1180         /* coverity[overrun-buffer-val] */
1181         pool = lod_find_pool(d, lum3->lmm_pool_name);
1182         if (pool == NULL)
1183                 goto out;
1184
1185         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1186                 rc = lod_check_index_in_pool(stripe_offset, pool);
1187                 if (rc < 0)
1188                         GOTO(out, rc = -EINVAL);
1189         }
1190
1191         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1192                 CDEBUG(D_IOCTL,
1193                        "stripe count %u > # OSTs %u in the pool\n",
1194                        stripe_count, pool_tgt_count(pool));
1195                 GOTO(out, rc = -EINVAL);
1196         }
1197
1198 out:
1199         if (pool != NULL)
1200                 lod_pool_putref(pool);
1201
1202         RETURN(rc);
1203 }
1204
1205 void lod_fix_desc_stripe_size(__u64 *val)
1206 {
1207         if (*val < LOV_MIN_STRIPE_SIZE) {
1208                 if (*val != 0)
1209                         LCONSOLE_INFO("Increasing default stripe size to "
1210                                       "minimum value %u\n",
1211                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1212                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1213         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1214                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1215                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1216                               "multiple of %u)\n",
1217                               *val, LOV_MIN_STRIPE_SIZE);
1218         }
1219 }
1220
1221 void lod_fix_desc_stripe_count(__u32 *val)
1222 {
1223         if (*val == 0)
1224                 *val = 1;
1225 }
1226
1227 void lod_fix_desc_pattern(__u32 *val)
1228 {
1229         /* from lov_setstripe */
1230         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1231                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1232                 *val = 0;
1233         }
1234 }
1235
1236 void lod_fix_desc_qos_maxage(__u32 *val)
1237 {
1238         /* fix qos_maxage */
1239         if (*val == 0)
1240                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1241 }
1242
1243 /**
1244  * Used to fix insane default striping.
1245  *
1246  * \param[in] desc      striping description
1247  */
1248 void lod_fix_desc(struct lov_desc *desc)
1249 {
1250         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1251         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1252         lod_fix_desc_pattern(&desc->ld_pattern);
1253         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1254 }
1255
1256 /**
1257  * Initialize the structures used to store pools and default striping.
1258  *
1259  * \param[in] lod       LOD device
1260  * \param[in] lcfg      configuration structure storing default striping.
1261  *
1262  * \retval              0 if initialization succeeds
1263  * \retval              negative error number on failure
1264  */
1265 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1266 {
1267         struct obd_device          *obd;
1268         struct lov_desc            *desc;
1269         int                         rc;
1270         ENTRY;
1271
1272         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1273         LASSERT(obd != NULL);
1274         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1275
1276         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1277                 CERROR("LOD setup requires a descriptor\n");
1278                 RETURN(-EINVAL);
1279         }
1280
1281         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1282
1283         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1284                 CERROR("descriptor size wrong: %d > %d\n",
1285                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1286                 RETURN(-EINVAL);
1287         }
1288
1289         if (desc->ld_magic != LOV_DESC_MAGIC) {
1290                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1291                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1292                                obd->obd_name, desc);
1293                         lustre_swab_lov_desc(desc);
1294                 } else {
1295                         CERROR("%s: Bad lov desc magic: %#x\n",
1296                                obd->obd_name, desc->ld_magic);
1297                         RETURN(-EINVAL);
1298                 }
1299         }
1300
1301         lod_fix_desc(desc);
1302
1303         desc->ld_active_tgt_count = 0;
1304         lod->lod_desc = *desc;
1305
1306         lod->lod_sp_me = LUSTRE_SP_CLI;
1307
1308         /* Set up allocation policy (QoS and RR) */
1309         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1310         init_rwsem(&lod->lod_qos.lq_rw_sem);
1311         lod->lod_qos.lq_dirty = 1;
1312         lod->lod_qos.lq_rr.lqr_dirty = 1;
1313         lod->lod_qos.lq_reset = 1;
1314         /* Default priority is toward free space balance */
1315         lod->lod_qos.lq_prio_free = 232;
1316         /* Default threshold for rr (roughly 17%) */
1317         lod->lod_qos.lq_threshold_rr = 43;
1318
1319         /* Set up OST pool environment */
1320         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1321                                                    HASH_POOLS_MAX_BITS,
1322                                                    HASH_POOLS_BKT_BITS, 0,
1323                                                    CFS_HASH_MIN_THETA,
1324                                                    CFS_HASH_MAX_THETA,
1325                                                    &pool_hash_operations,
1326                                                    CFS_HASH_DEFAULT);
1327         if (lod->lod_pools_hash_body == NULL)
1328                 RETURN(-ENOMEM);
1329
1330         INIT_LIST_HEAD(&lod->lod_pool_list);
1331         lod->lod_pool_count = 0;
1332         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1333         if (rc)
1334                 GOTO(out_hash, rc);
1335         lod_qos_rr_init(&lod->lod_qos.lq_rr);
1336         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1337         if (rc)
1338                 GOTO(out_pool_info, rc);
1339
1340         RETURN(0);
1341
1342 out_pool_info:
1343         lod_ost_pool_free(&lod->lod_pool_info);
1344 out_hash:
1345         cfs_hash_putref(lod->lod_pools_hash_body);
1346
1347         return rc;
1348 }
1349
1350 /**
1351  * Release the structures describing the pools.
1352  *
1353  * \param[in] lod       LOD device from which we release the structures
1354  *
1355  * \retval              0 always
1356  */
1357 int lod_pools_fini(struct lod_device *lod)
1358 {
1359         struct obd_device   *obd = lod2obd(lod);
1360         struct pool_desc    *pool, *tmp;
1361         ENTRY;
1362
1363         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1364                 /* free pool structs */
1365                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1366                 /* In the function below, .hs_keycmp resolves to
1367                  * pool_hashkey_keycmp() */
1368                 /* coverity[overrun-buffer-val] */
1369                 lod_pool_del(obd, pool->pool_name);
1370         }
1371
1372         cfs_hash_putref(lod->lod_pools_hash_body);
1373         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1374         lod_ost_pool_free(&lod->lod_pool_info);
1375
1376         RETURN(0);
1377 }