Whamcloud - gitweb
bea81b413b20cbfce3945f0ffc06e3a48075f7e0
[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, 2015, 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        *lu_dev;
214         struct dt_device        *dt_dev;
215         int                      rc;
216         struct lod_tgt_desc     *tgt_desc;
217         struct lod_tgt_descs    *ltd;
218         struct lustre_cfg       *lcfg;
219         struct obd_uuid         obd_uuid;
220         bool                    for_ost;
221         bool lock = false;
222         ENTRY;
223
224         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
225
226         if (gen <= 0) {
227                 CERROR("request to add OBD %s with invalid generation: %d\n",
228                        osp, gen);
229                 RETURN(-EINVAL);
230         }
231
232         obd_str2uuid(&obd_uuid, osp);
233
234         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
235                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
236         if (obd == NULL) {
237                 CERROR("can't find %s device\n", osp);
238                 RETURN(-EINVAL);
239         }
240
241         LASSERT(obd->obd_lu_dev != NULL);
242         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
243
244         lu_dev = obd->obd_lu_dev;
245         dt_dev = lu2dt_dev(lu_dev);
246
247         OBD_ALLOC_PTR(data);
248         if (data == NULL)
249                 GOTO(out_cleanup, rc = -ENOMEM);
250
251         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
252         data->ocd_version = LUSTRE_VERSION_CODE;
253         data->ocd_index = index;
254
255         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
256                 for_ost = true;
257                 data->ocd_connect_flags |= OBD_CONNECT_AT |
258                                            OBD_CONNECT_FULL20 |
259                                            OBD_CONNECT_INDEX |
260 #ifdef HAVE_LRU_RESIZE_SUPPORT
261                                            OBD_CONNECT_LRU_RESIZE |
262 #endif
263                                            OBD_CONNECT_MDS |
264                                            OBD_CONNECT_REQPORTAL |
265                                            OBD_CONNECT_SKIP_ORPHAN |
266                                            OBD_CONNECT_FID |
267                                            OBD_CONNECT_LVB_TYPE |
268                                            OBD_CONNECT_VERSION |
269                                            OBD_CONNECT_PINGLESS |
270                                            OBD_CONNECT_LFSCK |
271                                            OBD_CONNECT_BULK_MBITS;
272
273                 data->ocd_group = tgt_index;
274                 ltd = &lod->lod_ost_descs;
275         } else {
276                 struct obd_import *imp = obd->u.cli.cl_import;
277
278                 for_ost = false;
279                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
280                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
281                                            OBD_CONNECT_IBITS |
282                                            OBD_CONNECT_MDS_MDS |
283                                            OBD_CONNECT_FID |
284                                            OBD_CONNECT_AT |
285                                            OBD_CONNECT_FULL20 |
286                                            OBD_CONNECT_LFSCK |
287                                            OBD_CONNECT_BULK_MBITS;
288                 spin_lock(&imp->imp_lock);
289                 imp->imp_server_timeout = 1;
290                 spin_unlock(&imp->imp_lock);
291                 imp->imp_client->cli_request_portal = OUT_PORTAL;
292                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
293                       obd->obd_name);
294                 ltd = &lod->lod_mdt_descs;
295         }
296
297         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
298         OBD_FREE_PTR(data);
299         if (rc) {
300                 CERROR("%s: cannot connect to next dev %s (%d)\n",
301                        obd->obd_name, osp, rc);
302                 GOTO(out_cleanup, rc);
303         }
304
305         /* Allocate ost descriptor and fill it */
306         OBD_ALLOC_PTR(tgt_desc);
307         if (!tgt_desc)
308                 GOTO(out_conn, rc = -ENOMEM);
309
310         tgt_desc->ltd_tgt    = dt_dev;
311         tgt_desc->ltd_exp    = exp;
312         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
313         tgt_desc->ltd_gen    = gen;
314         tgt_desc->ltd_index  = index;
315         tgt_desc->ltd_active = active;
316
317         lod_getref(ltd);
318         if (index >= ltd->ltd_tgts_size) {
319                 /* we have to increase the size of the lod_osts array */
320                 __u32  newsize;
321
322                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
323                 while (newsize < index + 1)
324                         newsize = newsize << 1;
325
326                 /* lod_bitmap_resize() needs lod_rw_sem
327                  * which we hold with th reference */
328                 lod_putref(lod, ltd);
329
330                 rc = ltd_bitmap_resize(ltd, newsize);
331                 if (rc)
332                         GOTO(out_desc, rc);
333
334                 lod_getref(ltd);
335         }
336
337         mutex_lock(&ltd->ltd_mutex);
338         lock = true;
339         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
340                 CERROR("%s: device %d is registered already\n", obd->obd_name,
341                        index);
342                 GOTO(out_mutex, rc = -EEXIST);
343         }
344
345         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
346                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
347                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
348                         CERROR("can't allocate index to add %s\n",
349                                obd->obd_name);
350                         GOTO(out_mutex, rc = -ENOMEM);
351                 }
352         }
353
354         if (for_ost) {
355                 /* pool and qos are not supported for MDS stack yet */
356                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
357                                       lod->lod_osts_size);
358                 if (rc) {
359                         CERROR("%s: can't set up pool, failed with %d\n",
360                                obd->obd_name, rc);
361                         GOTO(out_mutex, rc);
362                 }
363
364                 rc = qos_add_tgt(lod, tgt_desc);
365                 if (rc) {
366                         CERROR("%s: qos_add_tgt failed with %d\n",
367                                 obd->obd_name, rc);
368                         GOTO(out_pool, rc);
369                 }
370
371                 /* The new OST is now a full citizen */
372                 if (index >= lod->lod_desc.ld_tgt_count)
373                         lod->lod_desc.ld_tgt_count = index + 1;
374                 if (active)
375                         lod->lod_desc.ld_active_tgt_count++;
376         }
377
378         LTD_TGT(ltd, index) = tgt_desc;
379         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
380         ltd->ltd_tgtnr++;
381         mutex_unlock(&ltd->ltd_mutex);
382         lod_putref(lod, ltd);
383         lock = false;
384         if (lod->lod_recovery_completed)
385                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
386
387         if (!for_ost && lod->lod_initialized) {
388                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
389                 if (rc != 0) {
390                         CERROR("%s: cannot start llog on %s:rc = %d\n",
391                                lod2obd(lod)->obd_name, osp, rc);
392                         GOTO(out_ltd, rc);
393                 }
394         }
395
396         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
397         if (rc != 0) {
398                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
399                        "index = %u, rc = %d\n", osp, type, index, rc);
400                 GOTO(out_fini_llog, rc);
401         }
402         RETURN(rc);
403 out_fini_llog:
404         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
405                           tgt_desc->ltd_recovery_thread);
406 out_ltd:
407         lod_getref(ltd);
408         mutex_lock(&ltd->ltd_mutex);
409         lock = true;
410         if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
411                 struct ptlrpc_thread *thread;
412
413                 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
414                 OBD_FREE_PTR(thread);
415         }
416         ltd->ltd_tgtnr--;
417         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
418         LTD_TGT(ltd, index) = NULL;
419 out_pool:
420         lod_ost_pool_remove(&lod->lod_pool_info, index);
421 out_mutex:
422         if (lock) {
423                 mutex_unlock(&ltd->ltd_mutex);
424                 lod_putref(lod, ltd);
425         }
426 out_desc:
427         OBD_FREE_PTR(tgt_desc);
428 out_conn:
429         obd_disconnect(exp);
430 out_cleanup:
431         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
432          * objects from the MDT stack. See LU-7184. */
433         lcfg = &lod_env_info(env)->lti_lustre_cfg;
434         memset(lcfg, 0, sizeof(*lcfg));
435         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
436         lcfg->lcfg_command = LCFG_CLEANUP;
437         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
438
439         return rc;
440 }
441
442 /**
443  * Schedule target removal from the target table.
444  *
445  * Mark the device as dead. The device is not removed here because it may
446  * still be in use. The device will be removed in lod_putref() when the
447  * last reference is released.
448  *
449  * \param[in] env               execution environment for this thread
450  * \param[in] lod               LOD device the target table belongs to
451  * \param[in] ltd               target table
452  * \param[in] idx               index of the target
453  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
454  */
455 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
456                              struct lod_tgt_descs *ltd, unsigned idx,
457                              bool for_ost)
458 {
459         LASSERT(LTD_TGT(ltd, idx));
460
461         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
462                          idx, for_ost);
463
464         if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
465                 struct ptlrpc_thread *thread;
466
467                 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
468                 OBD_FREE_PTR(thread);
469         }
470
471         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
472                 LTD_TGT(ltd, idx)->ltd_reap = 1;
473                 ltd->ltd_death_row++;
474         }
475 }
476
477 /**
478  * Schedule removal of all the targets from the given target table.
479  *
480  * See more details in the description for __lod_del_device()
481  *
482  * \param[in] env               execution environment for this thread
483  * \param[in] lod               LOD device the target table belongs to
484  * \param[in] ltd               target table
485  * \param[in] for_ost           type of the target: MDT or OST
486  *
487  * \retval                      0 always
488  */
489 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
490                  struct lod_tgt_descs *ltd, bool for_ost)
491 {
492         unsigned int idx;
493
494         if (ltd->ltd_tgts_size <= 0)
495                 return 0;
496         lod_getref(ltd);
497         mutex_lock(&ltd->ltd_mutex);
498         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
499                 __lod_del_device(env, lod, ltd, idx, for_ost);
500         mutex_unlock(&ltd->ltd_mutex);
501         lod_putref(lod, ltd);
502         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
503         for (idx = 0; idx < TGT_PTRS; idx++) {
504                 if (ltd->ltd_tgt_idx[idx])
505                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
506         }
507         ltd->ltd_tgts_size = 0;
508         return 0;
509 }
510
511 /**
512  * Remove device by name.
513  *
514  * Remove a device identified by \a osp from the target table. Given
515  * the device can be in use, the real deletion happens in lod_putref().
516  *
517  * \param[in] env               execution environment for this thread
518  * \param[in] lod               LOD device to be connected to the new OSP
519  * \param[in] ltd               target table
520  * \param[in] osp               name of OSP device to be removed
521  * \param[in] idx               index of the target
522  * \param[in] gen               generation number, not used currently
523  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
524  *
525  * \retval                      0 if the device was scheduled for removal
526  * \retval                      -EINVAL if no device was found
527  */
528 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
529                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
530                    unsigned gen, bool for_ost)
531 {
532         struct obd_device *obd;
533         int                rc = 0;
534         struct obd_uuid    uuid;
535         ENTRY;
536
537         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
538
539         obd_str2uuid(&uuid, osp);
540
541         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
542                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
543         if (obd == NULL) {
544                 CERROR("can't find %s device\n", osp);
545                 RETURN(-EINVAL);
546         }
547
548         if (gen <= 0) {
549                 CERROR("%s: request to remove OBD %s with invalid generation %d"
550                        "\n", obd->obd_name, osp, gen);
551                 RETURN(-EINVAL);
552         }
553
554         obd_str2uuid(&uuid,  osp);
555
556         lod_getref(ltd);
557         mutex_lock(&ltd->ltd_mutex);
558         /* check that the index is allocated in the bitmap */
559         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
560             !LTD_TGT(ltd, idx)) {
561                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
562                 GOTO(out, rc = -EINVAL);
563         }
564
565         /* check that the UUID matches */
566         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
567                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
568                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
569                        idx, osp);
570                 GOTO(out, rc = -EINVAL);
571         }
572
573         __lod_del_device(env, lod, ltd, idx, for_ost);
574         EXIT;
575 out:
576         mutex_unlock(&ltd->ltd_mutex);
577         lod_putref(lod, ltd);
578         return(rc);
579 }
580
581 /**
582  * Resize per-thread storage to hold specified size.
583  *
584  * A helper function to resize per-thread temporary storage. This storage
585  * is used to process LOV/LVM EAs and may be quite large. We do not want to
586  * allocate/release it every time, so instead we put it into the env and
587  * reallocate on demand. The memory is released when the correspondent thread
588  * is finished.
589  *
590  * \param[in] info              LOD-specific storage in the environment
591  * \param[in] size              new size to grow the buffer to
592
593  * \retval                      0 on success, -ENOMEM if reallocation failed
594  */
595 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
596 {
597         __u32 round = size_roundup_power2(size);
598
599         LASSERT(round <=
600                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
601         if (info->lti_ea_store) {
602                 LASSERT(info->lti_ea_store_size);
603                 LASSERT(info->lti_ea_store_size < round);
604                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
605                        info->lti_ea_store_size, round);
606                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
607                 info->lti_ea_store = NULL;
608                 info->lti_ea_store_size = 0;
609         }
610
611         OBD_ALLOC_LARGE(info->lti_ea_store, round);
612         if (info->lti_ea_store == NULL)
613                 RETURN(-ENOMEM);
614         info->lti_ea_store_size = round;
615         RETURN(0);
616 }
617
618 /**
619  * Make LOV EA for striped object.
620  *
621  * Generate striping information and store it in the LOV EA of the given
622  * object. The caller must ensure nobody else is calling the function
623  * against the object concurrently. The transaction must be started.
624  * FLDB service must be running as well; it's used to map FID to the target,
625  * which is stored in LOV EA.
626  *
627  * \param[in] env               execution environment for this thread
628  * \param[in] lo                LOD object
629  * \param[in] th                transaction handle
630  *
631  * \retval                      0 if LOV EA is stored successfully
632  * \retval                      negative error number on failure
633  */
634 int lod_generate_and_set_lovea(const struct lu_env *env,
635                                struct lod_object *lo, struct thandle *th)
636 {
637         struct lod_thread_info  *info = lod_env_info(env);
638         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
639         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
640         struct lov_mds_md_v1    *lmm;
641         struct lov_ost_data_v1  *objs;
642         __u32                    magic;
643         int                      i, rc;
644         size_t                   lmm_size;
645         ENTRY;
646
647         LASSERT(lo);
648
649         magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
650         lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
651         if (info->lti_ea_store_size < lmm_size) {
652                 rc = lod_ea_store_resize(info, lmm_size);
653                 if (rc)
654                         RETURN(rc);
655         }
656
657         if (lo->ldo_pattern == 0) /* default striping */
658                 lo->ldo_pattern = LOV_PATTERN_RAID0;
659
660         lmm = info->lti_ea_store;
661
662         lmm->lmm_magic = cpu_to_le32(magic);
663         lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
664         fid_to_lmm_oi(fid, &lmm->lmm_oi);
665         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
666                 lmm->lmm_oi.oi.oi_id++;
667         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
668         lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
669         lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
670         if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
671                 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
672         lmm->lmm_layout_gen = 0;
673         if (magic == LOV_MAGIC_V1) {
674                 objs = &lmm->lmm_objects[0];
675         } else {
676                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
677                 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
678                                 sizeof(v3->lmm_pool_name));
679                 if (cplen >= sizeof(v3->lmm_pool_name))
680                         RETURN(-E2BIG);
681                 objs = &v3->lmm_objects[0];
682         }
683
684         for (i = 0; i < lo->ldo_stripenr; i++) {
685                 struct lu_fid           *fid    = &info->lti_fid;
686                 struct lod_device       *lod;
687                 __u32                   index;
688                 int                     type    = LU_SEQ_RANGE_OST;
689
690                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
691                 LASSERT(lo->ldo_stripe[i]);
692
693                 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
694                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
695                         if (cfs_fail_val == 0)
696                                 cfs_fail_val = fid->f_oid;
697                         else
698                                 fid->f_oid = cfs_fail_val;
699                 }
700
701                 rc = fid_to_ostid(fid, &info->lti_ostid);
702                 LASSERT(rc == 0);
703
704                 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
705                 objs[i].l_ost_gen    = cpu_to_le32(0);
706                 rc = lod_fld_lookup(env, lod, fid, &index, &type);
707                 if (rc < 0) {
708                         CERROR("%s: Can not locate "DFID": rc = %d\n",
709                                lod2obd(lod)->obd_name, PFID(fid), rc);
710                         lod_object_free_striping(env, lo);
711                         RETURN(rc);
712                 }
713                 objs[i].l_ost_idx = cpu_to_le32(index);
714         }
715
716         info->lti_buf.lb_buf = lmm;
717         info->lti_buf.lb_len = lmm_size;
718         rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
719                                       0, th);
720         if (rc < 0) {
721                 lod_object_free_striping(env, lo);
722                 RETURN(rc);
723         }
724
725         RETURN(rc);
726 }
727
728 /**
729  * Get LOV EA.
730  *
731  * Fill lti_ea_store buffer in the environment with a value for the given
732  * EA. The buffer is reallocated if the value doesn't fit.
733  *
734  * \param[in,out] env           execution environment for this thread
735  *                              .lti_ea_store buffer is filled with EA's value
736  * \param[in] lo                LOD object
737  * \param[in] name              name of the EA
738  *
739  * \retval                      0 if EA is fetched successfully
740  * \retval                      negative error number on failure
741  */
742 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
743                const char *name)
744 {
745         struct lod_thread_info  *info = lod_env_info(env);
746         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
747         int                     rc;
748         ENTRY;
749
750         LASSERT(info);
751
752         if (unlikely(info->lti_ea_store == NULL)) {
753                 /* just to enter in allocation block below */
754                 rc = -ERANGE;
755         } else {
756 repeat:
757                 info->lti_buf.lb_buf = info->lti_ea_store;
758                 info->lti_buf.lb_len = info->lti_ea_store_size;
759                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
760         }
761
762         /* if object is not striped or inaccessible */
763         if (rc == -ENODATA || rc == -ENOENT)
764                 RETURN(0);
765
766         if (rc == -ERANGE) {
767                 /* EA doesn't fit, reallocate new buffer */
768                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
769                 if (rc == -ENODATA || rc == -ENOENT)
770                         RETURN(0);
771                 else if (rc < 0)
772                         RETURN(rc);
773
774                 LASSERT(rc > 0);
775                 rc = lod_ea_store_resize(info, rc);
776                 if (rc)
777                         RETURN(rc);
778                 goto repeat;
779         }
780
781         RETURN(rc);
782 }
783
784 /**
785  * Verify the target index is present in the current configuration.
786  *
787  * \param[in] md                LOD device where the target table is stored
788  * \param[in] idx               target's index
789  *
790  * \retval                      0 if the index is present
791  * \retval                      -EINVAL if not
792  */
793 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
794 {
795         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
796                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
797                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
798                        md->lod_ost_descs.ltd_tgts_size);
799                 return -EINVAL;
800         }
801
802         if (unlikely(OST_TGT(md, idx) == NULL)) {
803                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
804                        lod2obd(md)->obd_name, idx);
805                 return -EINVAL;
806         }
807
808         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
809                 CERROR("%s: invalid lod device, for idx: %d\n",
810                        lod2obd(md)->obd_name , idx);
811                 return -EINVAL;
812         }
813
814         return 0;
815 }
816
817 /**
818  * Instantiate objects for stripes.
819  *
820  * Allocate and initialize LU-objects representing the stripes. The number
821  * of the stripes (ldo_stripenr) must be initialized already. The caller
822  * must ensure nobody else is calling the function on the object at the same
823  * time. FLDB service must be running to be able to map a FID to the targets
824  * and find appropriate device representing that target.
825  *
826  * \param[in] env               execution environment for this thread
827  * \param[in,out] lo            LOD object
828  * \param[in] objs              an array of IDs to creates the objects from
829  *
830  * \retval                      0 if the objects are instantiated successfully
831  * \retval                      negative error number on failure
832  */
833 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
834                            struct lov_ost_data_v1 *objs)
835 {
836         struct lod_thread_info  *info = lod_env_info(env);
837         struct lod_device       *md;
838         struct lu_object        *o, *n;
839         struct lu_device        *nd;
840         struct dt_object       **stripe;
841         int                      stripe_len;
842         int                      i, rc = 0;
843         __u32                   idx;
844         ENTRY;
845
846         LASSERT(lo != NULL);
847         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
848         LASSERT(lo->ldo_stripe == NULL);
849         LASSERT(lo->ldo_stripenr > 0);
850         LASSERT(lo->ldo_stripe_size > 0);
851
852         stripe_len = lo->ldo_stripenr;
853         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
854         if (stripe == NULL)
855                 RETURN(-ENOMEM);
856
857         for (i = 0; i < lo->ldo_stripenr; i++) {
858                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
859                         continue;
860
861                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
862                 idx = le32_to_cpu(objs[i].l_ost_idx);
863                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
864                 if (rc != 0)
865                         GOTO(out, rc);
866                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
867                          PFID(&info->lti_fid));
868                 lod_getref(&md->lod_ost_descs);
869
870                 rc = validate_lod_and_idx(md, idx);
871                 if (unlikely(rc != 0)) {
872                         lod_putref(md, &md->lod_ost_descs);
873                         GOTO(out, rc);
874                 }
875
876                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
877                 lod_putref(md, &md->lod_ost_descs);
878
879                 /* In the function below, .hs_keycmp resolves to
880                  * u_obj_hop_keycmp() */
881                 /* coverity[overrun-buffer-val] */
882                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
883                 if (IS_ERR(o))
884                         GOTO(out, rc = PTR_ERR(o));
885
886                 n = lu_object_locate(o->lo_header, nd->ld_type);
887                 LASSERT(n);
888
889                 stripe[i] = container_of(n, struct dt_object, do_lu);
890         }
891
892 out:
893         if (rc != 0) {
894                 for (i = 0; i < stripe_len; i++)
895                         if (stripe[i] != NULL)
896                                 lu_object_put(env, &stripe[i]->do_lu);
897
898                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
899                 lo->ldo_stripenr = 0;
900         } else {
901                 lo->ldo_stripe = stripe;
902                 lo->ldo_stripes_allocated = stripe_len;
903         }
904
905         RETURN(rc);
906 }
907
908 /**
909  * Instantiate objects for striping.
910  *
911  * Parse striping information in \a buf and instantiate the objects
912  * representing the stripes.
913  *
914  * \param[in] env               execution environment for this thread
915  * \param[in] lo                LOD object
916  * \param[in] buf               buffer storing LOV EA to parse
917  *
918  * \retval                      0 if parsing and objects creation succeed
919  * \retval                      negative error number on failure
920  */
921 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
922                        const struct lu_buf *buf)
923 {
924         struct lov_mds_md_v1    *lmm;
925         struct lov_ost_data_v1  *objs;
926         __u32                    magic;
927         __u32                    pattern;
928         int                      rc = 0;
929         ENTRY;
930
931         LASSERT(buf);
932         LASSERT(buf->lb_buf);
933         LASSERT(buf->lb_len);
934
935         lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
936         magic = le32_to_cpu(lmm->lmm_magic);
937         pattern = le32_to_cpu(lmm->lmm_pattern);
938
939         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
940                 GOTO(out, rc = -EINVAL);
941         if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
942                 GOTO(out, rc = -EINVAL);
943
944         lo->ldo_pattern = pattern;
945         lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
946         lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
947         lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
948         /* released file stripenr fixup. */
949         if (pattern & LOV_PATTERN_F_RELEASED)
950                 lo->ldo_stripenr = 0;
951
952         LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
953
954         if (magic == LOV_MAGIC_V3) {
955                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
956                 objs = &v3->lmm_objects[0];
957                 lod_object_set_pool(lo, v3->lmm_pool_name);
958         } else {
959                 objs = &lmm->lmm_objects[0];
960         }
961
962         if (lo->ldo_stripenr > 0)
963                 rc = lod_initialize_objects(env, lo, objs);
964
965 out:
966         RETURN(rc);
967 }
968
969 /**
970  * Initialize the object representing the stripes.
971  *
972  * Unless the stripes are initialized already, fetch LOV (for regular
973  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
974  * to instantiate the objects representing the stripes.
975  *
976  * \param[in] env               execution environment for this thread
977  * \param[in,out] lo            LOD object
978  *
979  * \retval                      0 if parsing and object creation succeed
980  * \retval                      negative error number on failure
981  */
982 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
983 {
984         struct lod_thread_info  *info = lod_env_info(env);
985         struct lu_buf           *buf  = &info->lti_buf;
986         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
987         int                      rc = 0;
988         ENTRY;
989
990         /* already initialized? */
991         if (lo->ldo_stripe != NULL)
992                 GOTO(out, rc = 0);
993
994         if (!dt_object_exists(next))
995                 GOTO(out, rc = 0);
996
997         /* Do not load stripe for slaves of striped dir */
998         if (lo->ldo_dir_slave_stripe)
999                 GOTO(out, rc = 0);
1000
1001         if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
1002                 rc = lod_get_lov_ea(env, lo);
1003                 if (rc <= 0)
1004                         GOTO(out, rc);
1005                 /*
1006                  * there is LOV EA (striping information) in this object
1007                  * let's parse it and create in-core objects for the stripes
1008                  */
1009                 buf->lb_buf = info->lti_ea_store;
1010                 buf->lb_len = info->lti_ea_store_size;
1011                 rc = lod_parse_striping(env, lo, buf);
1012         } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
1013                 rc = lod_get_lmv_ea(env, lo);
1014                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1015                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1016
1017                 buf->lb_buf = info->lti_ea_store;
1018                 buf->lb_len = info->lti_ea_store_size;
1019                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1020                         rc = lod_load_lmv_shards(env, lo, buf, true);
1021                         if (buf->lb_buf != info->lti_ea_store) {
1022                                 OBD_FREE_LARGE(info->lti_ea_store,
1023                                                info->lti_ea_store_size);
1024                                 info->lti_ea_store = buf->lb_buf;
1025                                 info->lti_ea_store_size = buf->lb_len;
1026                         }
1027
1028                         if (rc < 0)
1029                                 GOTO(out, rc);
1030                 }
1031
1032                 /*
1033                  * there is LOV EA (striping information) in this object
1034                  * let's parse it and create in-core objects for the stripes
1035                  */
1036                 rc = lod_parse_dir_striping(env, lo, buf);
1037         }
1038
1039         if (rc == 0)
1040                 lo->ldo_striping_cached = 1;
1041 out:
1042         RETURN(rc);
1043 }
1044
1045 /**
1046  * A generic function to initialize the stripe objects.
1047  *
1048  * A protected version of lod_load_striping_locked() - load the striping
1049  * information from storage, parse that and instantiate LU objects to
1050  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1051  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1052  * return an array of references to the newly instantiated objects.
1053  *
1054  * \param[in] env               execution environment for this thread
1055  * \param[in,out] lo            LOD object, where striping is stored and
1056  *                              which gets an array of references
1057  *
1058  * \retval                      0 if parsing and object creation succeed
1059  * \retval                      negative error number on failure
1060  **/
1061 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1062 {
1063         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1064         int                     rc = 0;
1065
1066         /* currently this code is supposed to be called from declaration
1067          * phase only, thus the object is not expected to be locked by caller */
1068         dt_write_lock(env, next, 0);
1069         rc = lod_load_striping_locked(env, lo);
1070         dt_write_unlock(env, next);
1071         return rc;
1072 }
1073
1074 /**
1075  * Verify striping.
1076  *
1077  * Check the validity of all fields including the magic, stripe size,
1078  * stripe count, stripe offset and that the pool is present.  Also check
1079  * that each target index points to an existing target. The additional
1080  * \a is_from_disk turns additional checks. In some cases zero fields
1081  * are allowed (like pattern=0).
1082  *
1083  * \param[in] d                 LOD device
1084  * \param[in] buf               buffer with LOV EA to verify
1085  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1086  *                              1 - from disk, do not allow
1087  *
1088  * \retval                      0 if the striping is valid
1089  * \retval                      -EINVAL if striping is invalid
1090  */
1091 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1092                         bool is_from_disk)
1093 {
1094         struct lov_user_md_v1   *lum;
1095         struct lov_user_md_v3   *lum3;
1096         struct pool_desc        *pool = NULL;
1097         __u32                    magic;
1098         __u32                    stripe_size;
1099         __u16                    stripe_count;
1100         __u16                    stripe_offset;
1101         size_t                   lum_size;
1102         int                      rc = 0;
1103         ENTRY;
1104
1105         lum = buf->lb_buf;
1106
1107         LASSERT(sizeof(*lum) < sizeof(*lum3));
1108
1109         if (buf->lb_len < sizeof(*lum)) {
1110                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n",
1111                        buf->lb_len);
1112                 GOTO(out, rc = -EINVAL);
1113         }
1114
1115         magic = le32_to_cpu(lum->lmm_magic);
1116         if (magic != LOV_USER_MAGIC_V1 &&
1117             magic != LOV_USER_MAGIC_V3 &&
1118             magic != LOV_MAGIC_V1_DEF &&
1119             magic != LOV_MAGIC_V3_DEF) {
1120                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1121                 GOTO(out, rc = -EINVAL);
1122         }
1123
1124         /* the user uses "0" for default stripe pattern normally. */
1125         if (!is_from_disk && lum->lmm_pattern == 0)
1126                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1127
1128         if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1129                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1130                        le32_to_cpu(lum->lmm_pattern));
1131                 GOTO(out, rc = -EINVAL);
1132         }
1133
1134         /* 64kB is the largest common page size we see (ia64), and matches the
1135          * check in lfs */
1136         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1137         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1138                 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1139                        stripe_size, LOV_MIN_STRIPE_SIZE);
1140                 GOTO(out, rc = -EINVAL);
1141         }
1142
1143         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1144         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1145                 /* if offset is not within valid range [0, osts_size) */
1146                 if (stripe_offset >= d->lod_osts_size) {
1147                         CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1148                                stripe_offset, d->lod_osts_size);
1149                         GOTO(out, rc = -EINVAL);
1150                 }
1151
1152                 /* if lmm_stripe_offset is *not* in bitmap */
1153                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1154                         CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1155                                stripe_offset);
1156                         GOTO(out, rc = -EINVAL);
1157                 }
1158         }
1159
1160         if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1161                 lum_size = offsetof(struct lov_user_md_v1,
1162                                     lmm_objects[0]);
1163         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1164                 lum_size = offsetof(struct lov_user_md_v3,
1165                                     lmm_objects[0]);
1166         else
1167                 GOTO(out, rc = -EINVAL);
1168
1169         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1170         if (buf->lb_len != lum_size) {
1171                 CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with "
1172                        "magic %#x and stripe_count %u\n",
1173                        buf->lb_len, magic, stripe_count);
1174                 GOTO(out, rc = -EINVAL);
1175         }
1176
1177         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1178                 goto out;
1179
1180         lum3 = buf->lb_buf;
1181         if (buf->lb_len < sizeof(*lum3)) {
1182                 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n",
1183                        buf->lb_len);
1184                 GOTO(out, rc = -EINVAL);
1185         }
1186
1187         /* In the function below, .hs_keycmp resolves to
1188          * pool_hashkey_keycmp() */
1189         /* coverity[overrun-buffer-val] */
1190         pool = lod_find_pool(d, lum3->lmm_pool_name);
1191         if (pool == NULL)
1192                 goto out;
1193
1194         if (stripe_offset != LOV_OFFSET_DEFAULT) {
1195                 rc = lod_check_index_in_pool(stripe_offset, pool);
1196                 if (rc < 0)
1197                         GOTO(out, rc = -EINVAL);
1198         }
1199
1200         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1201                 CDEBUG(D_IOCTL,
1202                        "stripe count %u > # OSTs %u in the pool\n",
1203                        stripe_count, pool_tgt_count(pool));
1204                 GOTO(out, rc = -EINVAL);
1205         }
1206
1207 out:
1208         if (pool != NULL)
1209                 lod_pool_putref(pool);
1210
1211         RETURN(rc);
1212 }
1213
1214 void lod_fix_desc_stripe_size(__u64 *val)
1215 {
1216         if (*val < LOV_MIN_STRIPE_SIZE) {
1217                 if (*val != 0)
1218                         LCONSOLE_INFO("Increasing default stripe size to "
1219                                       "minimum value %u\n",
1220                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1221                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1222         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1223                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1224                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1225                               "multiple of %u)\n",
1226                               *val, LOV_MIN_STRIPE_SIZE);
1227         }
1228 }
1229
1230 void lod_fix_desc_stripe_count(__u32 *val)
1231 {
1232         if (*val == 0)
1233                 *val = 1;
1234 }
1235
1236 void lod_fix_desc_pattern(__u32 *val)
1237 {
1238         /* from lov_setstripe */
1239         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1240                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1241                 *val = 0;
1242         }
1243 }
1244
1245 void lod_fix_desc_qos_maxage(__u32 *val)
1246 {
1247         /* fix qos_maxage */
1248         if (*val == 0)
1249                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1250 }
1251
1252 /**
1253  * Used to fix insane default striping.
1254  *
1255  * \param[in] desc      striping description
1256  */
1257 void lod_fix_desc(struct lov_desc *desc)
1258 {
1259         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1260         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1261         lod_fix_desc_pattern(&desc->ld_pattern);
1262         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1263 }
1264
1265 /**
1266  * Initialize the structures used to store pools and default striping.
1267  *
1268  * \param[in] lod       LOD device
1269  * \param[in] lcfg      configuration structure storing default striping.
1270  *
1271  * \retval              0 if initialization succeeds
1272  * \retval              negative error number on failure
1273  */
1274 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1275 {
1276         struct obd_device          *obd;
1277         struct lov_desc            *desc;
1278         int                         rc;
1279         ENTRY;
1280
1281         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1282         LASSERT(obd != NULL);
1283         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1284
1285         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1286                 CERROR("LOD setup requires a descriptor\n");
1287                 RETURN(-EINVAL);
1288         }
1289
1290         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1291
1292         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1293                 CERROR("descriptor size wrong: %d > %d\n",
1294                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1295                 RETURN(-EINVAL);
1296         }
1297
1298         if (desc->ld_magic != LOV_DESC_MAGIC) {
1299                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1300                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1301                                obd->obd_name, desc);
1302                         lustre_swab_lov_desc(desc);
1303                 } else {
1304                         CERROR("%s: Bad lov desc magic: %#x\n",
1305                                obd->obd_name, desc->ld_magic);
1306                         RETURN(-EINVAL);
1307                 }
1308         }
1309
1310         lod_fix_desc(desc);
1311
1312         desc->ld_active_tgt_count = 0;
1313         lod->lod_desc = *desc;
1314
1315         lod->lod_sp_me = LUSTRE_SP_CLI;
1316
1317         /* Set up allocation policy (QoS and RR) */
1318         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1319         init_rwsem(&lod->lod_qos.lq_rw_sem);
1320         lod->lod_qos.lq_dirty = 1;
1321         lod->lod_qos.lq_rr.lqr_dirty = 1;
1322         lod->lod_qos.lq_reset = 1;
1323         /* Default priority is toward free space balance */
1324         lod->lod_qos.lq_prio_free = 232;
1325         /* Default threshold for rr (roughly 17%) */
1326         lod->lod_qos.lq_threshold_rr = 43;
1327
1328         /* Set up OST pool environment */
1329         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1330                                                    HASH_POOLS_MAX_BITS,
1331                                                    HASH_POOLS_BKT_BITS, 0,
1332                                                    CFS_HASH_MIN_THETA,
1333                                                    CFS_HASH_MAX_THETA,
1334                                                    &pool_hash_operations,
1335                                                    CFS_HASH_DEFAULT);
1336         if (lod->lod_pools_hash_body == NULL)
1337                 RETURN(-ENOMEM);
1338
1339         INIT_LIST_HEAD(&lod->lod_pool_list);
1340         lod->lod_pool_count = 0;
1341         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1342         if (rc)
1343                 GOTO(out_hash, rc);
1344         lod_qos_rr_init(&lod->lod_qos.lq_rr);
1345         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1346         if (rc)
1347                 GOTO(out_pool_info, rc);
1348
1349         RETURN(0);
1350
1351 out_pool_info:
1352         lod_ost_pool_free(&lod->lod_pool_info);
1353 out_hash:
1354         cfs_hash_putref(lod->lod_pools_hash_body);
1355
1356         return rc;
1357 }
1358
1359 /**
1360  * Release the structures describing the pools.
1361  *
1362  * \param[in] lod       LOD device from which we release the structures
1363  *
1364  * \retval              0 always
1365  */
1366 int lod_pools_fini(struct lod_device *lod)
1367 {
1368         struct obd_device   *obd = lod2obd(lod);
1369         struct pool_desc    *pool, *tmp;
1370         ENTRY;
1371
1372         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1373                 /* free pool structs */
1374                 CDEBUG(D_INFO, "delete pool %p\n", pool);
1375                 /* In the function below, .hs_keycmp resolves to
1376                  * pool_hashkey_keycmp() */
1377                 /* coverity[overrun-buffer-val] */
1378                 lod_pool_del(obd, pool->pool_name);
1379         }
1380
1381         cfs_hash_putref(lod->lod_pools_hash_body);
1382         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1383         lod_ost_pool_free(&lod->lod_pool_info);
1384
1385         RETURN(0);
1386 }