Whamcloud - gitweb
e099a6c906a9b4e52562194ccc178f605c5aeb4d
[fs/lustre-release.git] / lustre / lod / lod_lov.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License version 2 for more details.  A copy is
14  * included in the COPYING file that accompanied this code.
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2017, Intel Corporation.
27  */
28 /*
29  * lustre/lod/lod_lov.c
30  *
31  * A set of helpers to maintain Logical Object Volume (LOV)
32  * Extended Attribute (EA) and known OST targets
33  *
34  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35  */
36
37 #define DEBUG_SUBSYSTEM S_MDS
38
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
43
44 #include "lod_internal.h"
45
46 /**
47  * Increase reference count on the target table.
48  *
49  * Increase reference count on the target table usage to prevent racing with
50  * addition/deletion. Any function that expects the table to remain
51  * stationary must take a ref.
52  *
53  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
54  */
55 void lod_getref(struct lod_tgt_descs *ltd)
56 {
57         down_read(&ltd->ltd_rw_sem);
58         mutex_lock(&ltd->ltd_mutex);
59         ltd->ltd_refcount++;
60         mutex_unlock(&ltd->ltd_mutex);
61 }
62
63 /**
64  * Decrease reference count on the target table.
65  *
66  * Companion of lod_getref() to release a reference on the target table.
67  * If this is the last reference and the OST entry was scheduled for deletion,
68  * the descriptor is removed from the table.
69  *
70  * \param[in] lod       LOD device from which we release a reference
71  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
72  */
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
74 {
75         mutex_lock(&ltd->ltd_mutex);
76         ltd->ltd_refcount--;
77         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78                 struct lod_tgt_desc *tgt_desc, *tmp;
79                 struct list_head kill;
80                 unsigned int idx;
81
82                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
83                        ltd->ltd_death_row);
84
85                 INIT_LIST_HEAD(&kill);
86
87                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
88                         tgt_desc = LTD_TGT(ltd, idx);
89                         LASSERT(tgt_desc);
90
91                         if (!tgt_desc->ltd_reap)
92                                 continue;
93
94                         list_add(&tgt_desc->ltd_kill, &kill);
95                         LTD_TGT(ltd, idx) = NULL;
96                         /*FIXME: only support ost pool for now */
97                         if (ltd == &lod->lod_ost_descs) {
98                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
99                                 if (tgt_desc->ltd_active)
100                                         lod->lod_desc.ld_active_tgt_count--;
101                         }
102                         ltd->ltd_tgtnr--;
103                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
104                         ltd->ltd_death_row--;
105                 }
106                 mutex_unlock(&ltd->ltd_mutex);
107                 up_read(&ltd->ltd_rw_sem);
108
109                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
110                         int rc;
111                         list_del(&tgt_desc->ltd_kill);
112                         if (ltd == &lod->lod_ost_descs) {
113                                 /* remove from QoS structures */
114                                 rc = qos_del_tgt(lod, tgt_desc);
115                                 if (rc)
116                                         CERROR("%s: qos_del_tgt(%s) failed:"
117                                                "rc = %d\n",
118                                                lod2obd(lod)->obd_name,
119                                               obd_uuid2str(&tgt_desc->ltd_uuid),
120                                                rc);
121                         }
122                         rc = obd_disconnect(tgt_desc->ltd_exp);
123                         if (rc)
124                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
125                                        lod2obd(lod)->obd_name,
126                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
127                         OBD_FREE_PTR(tgt_desc);
128                 }
129         } else {
130                 mutex_unlock(&ltd->ltd_mutex);
131                 up_read(&ltd->ltd_rw_sem);
132         }
133 }
134
135 /**
136  * Expand size of target table.
137  *
138  * When the target table is full, we have to extend the table. To do so,
139  * we allocate new memory with some reserve, move data from the old table
140  * to the new one and release memory consumed by the old table.
141  * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
142  *
143  * \param[in] ltd               target table
144  * \param[in] newsize           new size of the table
145  *
146  * \retval                      0 on success
147  * \retval                      -ENOMEM if reallocation failed
148  */
149 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
150 {
151         struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
152         int           rc = 0;
153         ENTRY;
154
155         /* grab write reference on the lod. Relocating the array requires
156          * exclusive access */
157
158         down_write(&ltd->ltd_rw_sem);
159         if (newsize <= ltd->ltd_tgts_size)
160                 /* someone else has already resize the array */
161                 GOTO(out, rc = 0);
162
163         /* allocate new bitmap */
164         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
165         if (!new_bitmap)
166                 GOTO(out, rc = -ENOMEM);
167
168         if (ltd->ltd_tgts_size > 0) {
169                 /* the bitmap already exists, we need
170                  * to copy data from old one */
171                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
172                 old_bitmap = ltd->ltd_tgt_bitmap;
173         }
174
175         ltd->ltd_tgts_size  = newsize;
176         ltd->ltd_tgt_bitmap = new_bitmap;
177
178         if (old_bitmap)
179                 CFS_FREE_BITMAP(old_bitmap);
180
181         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
182
183         EXIT;
184 out:
185         up_write(&ltd->ltd_rw_sem);
186         return rc;
187 }
188
189 /**
190  * Connect LOD to a new OSP and add it to the target table.
191  *
192  * Connect to the OSP device passed, initialize all the internal
193  * structures related to the device and add it to the target table.
194  *
195  * \param[in] env               execution environment for this thread
196  * \param[in] lod               LOD device to be connected to the new OSP
197  * \param[in] osp               name of OSP device name to be added
198  * \param[in] index             index of the new target
199  * \param[in] gen               target's generation number
200  * \param[in] tgt_index         OSP's group
201  * \param[in] type              type of device (mdc or osc)
202  * \param[in] active            state of OSP: 0 - inactive, 1 - active
203  *
204  * \retval                      0 if added successfully
205  * \retval                      negative error number on failure
206  */
207 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
208                    char *osp, unsigned index, unsigned gen, int tgt_index,
209                    char *type, int active)
210 {
211         struct obd_connect_data *data = NULL;
212         struct obd_export       *exp = NULL;
213         struct obd_device       *obd;
214         struct lu_device        *lu_dev;
215         struct dt_device        *dt_dev;
216         int                      rc;
217         struct lod_tgt_desc     *tgt_desc;
218         struct lod_tgt_descs    *ltd;
219         struct lustre_cfg       *lcfg;
220         struct obd_uuid         obd_uuid;
221         bool                    for_ost;
222         bool lock = false;
223         ENTRY;
224
225         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
226
227         if (gen <= 0) {
228                 CERROR("request to add OBD %s with invalid generation: %d\n",
229                        osp, gen);
230                 RETURN(-EINVAL);
231         }
232
233         obd_str2uuid(&obd_uuid, osp);
234
235         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
236                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
237         if (obd == NULL) {
238                 CERROR("can't find %s device\n", osp);
239                 RETURN(-EINVAL);
240         }
241
242         LASSERT(obd->obd_lu_dev != NULL);
243         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
244
245         lu_dev = obd->obd_lu_dev;
246         dt_dev = lu2dt_dev(lu_dev);
247
248         OBD_ALLOC_PTR(data);
249         if (data == NULL)
250                 GOTO(out_cleanup, rc = -ENOMEM);
251
252         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
253         data->ocd_version = LUSTRE_VERSION_CODE;
254         data->ocd_index = index;
255
256         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
257                 for_ost = true;
258                 data->ocd_connect_flags |= OBD_CONNECT_AT |
259                                            OBD_CONNECT_FULL20 |
260                                            OBD_CONNECT_INDEX |
261 #ifdef HAVE_LRU_RESIZE_SUPPORT
262                                            OBD_CONNECT_LRU_RESIZE |
263 #endif
264                                            OBD_CONNECT_MDS |
265                                            OBD_CONNECT_REQPORTAL |
266                                            OBD_CONNECT_SKIP_ORPHAN |
267                                            OBD_CONNECT_FID |
268                                            OBD_CONNECT_LVB_TYPE |
269                                            OBD_CONNECT_VERSION |
270                                            OBD_CONNECT_PINGLESS |
271                                            OBD_CONNECT_LFSCK |
272                                            OBD_CONNECT_BULK_MBITS;
273
274                 data->ocd_group = tgt_index;
275                 ltd = &lod->lod_ost_descs;
276         } else {
277                 struct obd_import *imp = obd->u.cli.cl_import;
278
279                 for_ost = false;
280                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
281                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
282                                            OBD_CONNECT_IBITS |
283                                            OBD_CONNECT_MDS_MDS |
284                                            OBD_CONNECT_FID |
285                                            OBD_CONNECT_AT |
286                                            OBD_CONNECT_FULL20 |
287                                            OBD_CONNECT_LFSCK |
288                                            OBD_CONNECT_BULK_MBITS;
289                 spin_lock(&imp->imp_lock);
290                 imp->imp_server_timeout = 1;
291                 spin_unlock(&imp->imp_lock);
292                 imp->imp_client->cli_request_portal = OUT_PORTAL;
293                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
294                       obd->obd_name);
295                 ltd = &lod->lod_mdt_descs;
296         }
297
298         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
299         OBD_FREE_PTR(data);
300         if (rc) {
301                 CERROR("%s: cannot connect to next dev %s (%d)\n",
302                        obd->obd_name, osp, rc);
303                 GOTO(out_cleanup, rc);
304         }
305
306         /* Allocate ost descriptor and fill it */
307         OBD_ALLOC_PTR(tgt_desc);
308         if (!tgt_desc)
309                 GOTO(out_conn, rc = -ENOMEM);
310
311         tgt_desc->ltd_tgt    = dt_dev;
312         tgt_desc->ltd_exp    = exp;
313         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
314         tgt_desc->ltd_gen    = gen;
315         tgt_desc->ltd_index  = index;
316         tgt_desc->ltd_active = active;
317
318         lod_getref(ltd);
319         if (index >= ltd->ltd_tgts_size) {
320                 /* we have to increase the size of the lod_osts array */
321                 __u32  newsize;
322
323                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
324                 while (newsize < index + 1)
325                         newsize = newsize << 1;
326
327                 /* lod_bitmap_resize() needs lod_rw_sem
328                  * which we hold with th reference */
329                 lod_putref(lod, ltd);
330
331                 rc = ltd_bitmap_resize(ltd, newsize);
332                 if (rc)
333                         GOTO(out_desc, rc);
334
335                 lod_getref(ltd);
336         }
337
338         mutex_lock(&ltd->ltd_mutex);
339         lock = true;
340         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
341                 CERROR("%s: device %d is registered already\n", obd->obd_name,
342                        index);
343                 GOTO(out_mutex, rc = -EEXIST);
344         }
345
346         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
347                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
348                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
349                         CERROR("can't allocate index to add %s\n",
350                                obd->obd_name);
351                         GOTO(out_mutex, rc = -ENOMEM);
352                 }
353         }
354
355         if (for_ost) {
356                 /* pool and qos are not supported for MDS stack yet */
357                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
358                                       lod->lod_osts_size);
359                 if (rc) {
360                         CERROR("%s: can't set up pool, failed with %d\n",
361                                obd->obd_name, rc);
362                         GOTO(out_mutex, rc);
363                 }
364
365                 rc = qos_add_tgt(lod, tgt_desc);
366                 if (rc) {
367                         CERROR("%s: qos_add_tgt failed with %d\n",
368                                 obd->obd_name, rc);
369                         GOTO(out_pool, rc);
370                 }
371
372                 /* The new OST is now a full citizen */
373                 if (index >= lod->lod_desc.ld_tgt_count)
374                         lod->lod_desc.ld_tgt_count = index + 1;
375                 if (active)
376                         lod->lod_desc.ld_active_tgt_count++;
377         }
378
379         LTD_TGT(ltd, index) = tgt_desc;
380         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
381         ltd->ltd_tgtnr++;
382         mutex_unlock(&ltd->ltd_mutex);
383         lod_putref(lod, ltd);
384         lock = false;
385         if (lod->lod_recovery_completed)
386                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
387
388         if (!for_ost && lod->lod_initialized) {
389                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
390                 if (rc != 0) {
391                         CERROR("%s: cannot start llog on %s:rc = %d\n",
392                                lod2obd(lod)->obd_name, osp, rc);
393                         GOTO(out_ltd, rc);
394                 }
395         }
396
397         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
398         if (rc != 0) {
399                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
400                        "index = %u, rc = %d\n", osp, type, index, rc);
401                 GOTO(out_fini_llog, rc);
402         }
403         RETURN(rc);
404 out_fini_llog:
405         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
406                           tgt_desc->ltd_recovery_thread);
407 out_ltd:
408         lod_getref(ltd);
409         mutex_lock(&ltd->ltd_mutex);
410         lock = true;
411         if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
412                 struct ptlrpc_thread *thread;
413
414                 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
415                 OBD_FREE_PTR(thread);
416         }
417         ltd->ltd_tgtnr--;
418         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
419         LTD_TGT(ltd, index) = NULL;
420 out_pool:
421         lod_ost_pool_remove(&lod->lod_pool_info, index);
422 out_mutex:
423         if (lock) {
424                 mutex_unlock(&ltd->ltd_mutex);
425                 lod_putref(lod, ltd);
426         }
427 out_desc:
428         OBD_FREE_PTR(tgt_desc);
429 out_conn:
430         obd_disconnect(exp);
431 out_cleanup:
432         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
433          * objects from the MDT stack. See LU-7184. */
434         lcfg = &lod_env_info(env)->lti_lustre_cfg;
435         memset(lcfg, 0, sizeof(*lcfg));
436         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
437         lcfg->lcfg_command = LCFG_CLEANUP;
438         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
439
440         return rc;
441 }
442
443 /**
444  * Schedule target removal from the target table.
445  *
446  * Mark the device as dead. The device is not removed here because it may
447  * still be in use. The device will be removed in lod_putref() when the
448  * last reference is released.
449  *
450  * \param[in] env               execution environment for this thread
451  * \param[in] lod               LOD device the target table belongs to
452  * \param[in] ltd               target table
453  * \param[in] idx               index of the target
454  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
455  */
456 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
457                              struct lod_tgt_descs *ltd, unsigned idx,
458                              bool for_ost)
459 {
460         LASSERT(LTD_TGT(ltd, idx));
461
462         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
463                          idx, for_ost);
464
465         if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
466                 struct ptlrpc_thread *thread;
467
468                 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
469                 OBD_FREE_PTR(thread);
470         }
471
472         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
473                 LTD_TGT(ltd, idx)->ltd_reap = 1;
474                 ltd->ltd_death_row++;
475         }
476 }
477
478 /**
479  * Schedule removal of all the targets from the given target table.
480  *
481  * See more details in the description for __lod_del_device()
482  *
483  * \param[in] env               execution environment for this thread
484  * \param[in] lod               LOD device the target table belongs to
485  * \param[in] ltd               target table
486  * \param[in] for_ost           type of the target: MDT or OST
487  *
488  * \retval                      0 always
489  */
490 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
491                  struct lod_tgt_descs *ltd, bool for_ost)
492 {
493         unsigned int idx;
494
495         if (ltd->ltd_tgts_size <= 0)
496                 return 0;
497         lod_getref(ltd);
498         mutex_lock(&ltd->ltd_mutex);
499         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
500                 __lod_del_device(env, lod, ltd, idx, for_ost);
501         mutex_unlock(&ltd->ltd_mutex);
502         lod_putref(lod, ltd);
503         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
504         for (idx = 0; idx < TGT_PTRS; idx++) {
505                 if (ltd->ltd_tgt_idx[idx])
506                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
507         }
508         ltd->ltd_tgts_size = 0;
509         return 0;
510 }
511
512 /**
513  * Remove device by name.
514  *
515  * Remove a device identified by \a osp from the target table. Given
516  * the device can be in use, the real deletion happens in lod_putref().
517  *
518  * \param[in] env               execution environment for this thread
519  * \param[in] lod               LOD device to be connected to the new OSP
520  * \param[in] ltd               target table
521  * \param[in] osp               name of OSP device to be removed
522  * \param[in] idx               index of the target
523  * \param[in] gen               generation number, not used currently
524  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
525  *
526  * \retval                      0 if the device was scheduled for removal
527  * \retval                      -EINVAL if no device was found
528  */
529 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
530                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
531                    unsigned gen, bool for_ost)
532 {
533         struct obd_device *obd;
534         int                rc = 0;
535         struct obd_uuid    uuid;
536         ENTRY;
537
538         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
539
540         obd_str2uuid(&uuid, osp);
541
542         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
543                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
544         if (obd == NULL) {
545                 CERROR("can't find %s device\n", osp);
546                 RETURN(-EINVAL);
547         }
548
549         if (gen <= 0) {
550                 CERROR("%s: request to remove OBD %s with invalid generation %d"
551                        "\n", obd->obd_name, osp, gen);
552                 RETURN(-EINVAL);
553         }
554
555         obd_str2uuid(&uuid,  osp);
556
557         lod_getref(ltd);
558         mutex_lock(&ltd->ltd_mutex);
559         /* check that the index is allocated in the bitmap */
560         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
561             !LTD_TGT(ltd, idx)) {
562                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
563                 GOTO(out, rc = -EINVAL);
564         }
565
566         /* check that the UUID matches */
567         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
568                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
569                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
570                        idx, osp);
571                 GOTO(out, rc = -EINVAL);
572         }
573
574         __lod_del_device(env, lod, ltd, idx, for_ost);
575         EXIT;
576 out:
577         mutex_unlock(&ltd->ltd_mutex);
578         lod_putref(lod, ltd);
579         return(rc);
580 }
581
582 /**
583  * Resize per-thread storage to hold specified size.
584  *
585  * A helper function to resize per-thread temporary storage. This storage
586  * is used to process LOV/LVM EAs and may be quite large. We do not want to
587  * allocate/release it every time, so instead we put it into the env and
588  * reallocate on demand. The memory is released when the correspondent thread
589  * is finished.
590  *
591  * \param[in] info              LOD-specific storage in the environment
592  * \param[in] size              new size to grow the buffer to
593
594  * \retval                      0 on success, -ENOMEM if reallocation failed
595  */
596 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
597 {
598         __u32 round = size_roundup_power2(size);
599
600         LASSERT(round <=
601                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
602         if (info->lti_ea_store) {
603                 LASSERT(info->lti_ea_store_size);
604                 LASSERT(info->lti_ea_store_size < round);
605                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
606                        info->lti_ea_store_size, round);
607                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
608                 info->lti_ea_store = NULL;
609                 info->lti_ea_store_size = 0;
610         }
611
612         OBD_ALLOC_LARGE(info->lti_ea_store, round);
613         if (info->lti_ea_store == NULL)
614                 RETURN(-ENOMEM);
615         info->lti_ea_store_size = round;
616
617         RETURN(0);
618 }
619
620 static void lod_free_comp_buffer(struct lod_layout_component *entries,
621                                  __u16 count, __u32 bufsize)
622 {
623         struct lod_layout_component *entry;
624         int i;
625
626         for (i = 0; i < count; i++) {
627                 entry = &entries[i];
628                 if (entry->llc_pool != NULL)
629                         lod_set_pool(&entry->llc_pool, NULL);
630                 if (entry->llc_ostlist.op_array)
631                         OBD_FREE(entry->llc_ostlist.op_array,
632                                  entry->llc_ostlist.op_size);
633                 LASSERT(entry->llc_stripe == NULL);
634                 LASSERT(entry->llc_stripes_allocated == 0);
635         }
636
637         if (bufsize != 0)
638                 OBD_FREE_LARGE(entries, bufsize);
639 }
640
641 void lod_free_def_comp_entries(struct lod_default_striping *lds)
642 {
643         lod_free_comp_buffer(lds->lds_def_comp_entries,
644                              lds->lds_def_comp_size_cnt,
645                              size_roundup_power2(
646                                      sizeof(*lds->lds_def_comp_entries) *
647                                      lds->lds_def_comp_size_cnt));
648         lds->lds_def_comp_entries = NULL;
649         lds->lds_def_comp_cnt = 0;
650         lds->lds_def_striping_is_composite = 0;
651         lds->lds_def_comp_size_cnt = 0;
652 }
653
654 /**
655  * Resize per-thread storage to hold default striping component entries
656  *
657  * A helper function to resize per-thread temporary storage. This storage
658  * is used to hold default LOV/LVM EAs and may be quite large. We do not want
659  * to allocate/release it every time, so instead we put it into the env and
660  * reallocate it on demand. The memory is released when the correspondent
661  * thread is finished.
662  *
663  * \param[in,out] lds           default striping
664  * \param[in] count             new component count to grow the buffer to
665
666  * \retval                      0 on success, -ENOMEM if reallocation failed
667  */
668 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
669 {
670         struct lod_layout_component *entries;
671         __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
672                                         count);
673         __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
674                                         lds->lds_def_comp_size_cnt);
675
676         if (new <= old)
677                 return 0;
678
679         OBD_ALLOC_LARGE(entries, new);
680         if (entries == NULL)
681                 return -ENOMEM;
682
683         if (lds->lds_def_comp_entries != NULL) {
684                 CDEBUG(D_INFO, "default striping component size %d is not "
685                        "enough, need %d\n", old, new);
686                 lod_free_def_comp_entries(lds);
687         }
688
689         lds->lds_def_comp_entries = entries;
690         lds->lds_def_comp_size_cnt = count;
691
692         RETURN(0);
693 }
694
695 void lod_free_comp_entries(struct lod_object *lo)
696 {
697         if (lo->ldo_mirrors) {
698                 OBD_FREE(lo->ldo_mirrors,
699                          sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
700                 lo->ldo_mirrors = NULL;
701                 lo->ldo_mirror_count = 0;
702         }
703         lod_free_comp_buffer(lo->ldo_comp_entries,
704                              lo->ldo_comp_cnt,
705                              sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
706         lo->ldo_comp_entries = NULL;
707         lo->ldo_comp_cnt = 0;
708         lo->ldo_is_composite = 0;
709 }
710
711 int lod_alloc_comp_entries(struct lod_object *lo,
712                            int mirror_count, int comp_count)
713 {
714         LASSERT(comp_count != 0);
715         LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
716
717         if (mirror_count > 0) {
718                 OBD_ALLOC(lo->ldo_mirrors,
719                           sizeof(*lo->ldo_mirrors) * mirror_count);
720                 if (!lo->ldo_mirrors)
721                         return -ENOMEM;
722
723                 lo->ldo_mirror_count = mirror_count;
724         }
725
726         OBD_ALLOC_LARGE(lo->ldo_comp_entries,
727                         sizeof(*lo->ldo_comp_entries) * comp_count);
728         if (lo->ldo_comp_entries == NULL) {
729                 OBD_FREE(lo->ldo_mirrors,
730                          sizeof(*lo->ldo_mirrors) * mirror_count);
731                 lo->ldo_mirror_count = 0;
732                 return -ENOMEM;
733         }
734
735         lo->ldo_comp_cnt = comp_count;
736         return 0;
737 }
738
739 int lod_fill_mirrors(struct lod_object *lo)
740 {
741         struct lod_layout_component *lod_comp;
742         int mirror_idx = -1;
743         __u16 mirror_id = 0xffff;
744         int i;
745         ENTRY;
746
747         LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
748
749         if (!lo->ldo_is_composite)
750                 RETURN(0);
751
752         lod_comp = &lo->ldo_comp_entries[0];
753         for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
754                 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
755                 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
756
757                 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
758                         lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
759                         lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
760                         lo->ldo_mirrors[mirror_idx].lme_end = i;
761                         continue;
762                 }
763
764                 /* new mirror */
765                 ++mirror_idx;
766                 if (mirror_idx >= lo->ldo_mirror_count)
767                         RETURN(-EINVAL);
768
769                 mirror_id = mirror_id_of(lod_comp->llc_id);
770
771                 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
772                 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
773                 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
774                 lo->ldo_mirrors[mirror_idx].lme_start = i;
775                 lo->ldo_mirrors[mirror_idx].lme_end = i;
776         }
777         if (mirror_idx != lo->ldo_mirror_count - 1)
778                 RETURN(-EINVAL);
779
780         RETURN(0);
781 }
782
783 /**
784  * Generate on-disk lov_mds_md structure for each layout component based on
785  * the information in lod_object->ldo_comp_entries[i].
786  *
787  * \param[in] env               execution environment for this thread
788  * \param[in] lo                LOD object
789  * \param[in] comp_idx          index of ldo_comp_entries
790  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
791  * \param[in|out] lmm_size      buffer size/lmm size
792  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
793  *                              the stripe info is from the default stripe
794  *                              template, which is collected in lod_ah_init(),
795  *                              either from parent object or root object; for
796  *                              file case, it's from the @lo object
797  *
798  * \retval                      0 if on disk structure is created successfully
799  * \retval                      negative error number on failure
800  */
801 static int lod_gen_component_ea(const struct lu_env *env,
802                                 struct lod_object *lo, int comp_idx,
803                                 struct lov_mds_md *lmm, int *lmm_size,
804                                 bool is_dir)
805 {
806         struct lod_thread_info  *info = lod_env_info(env);
807         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
808         struct lod_device       *lod;
809         struct lov_ost_data_v1  *objs;
810         struct lod_layout_component *lod_comp;
811         __u32   magic;
812         __u16 stripe_count;
813         int     i, rc = 0;
814         ENTRY;
815
816         LASSERT(lo);
817         if (is_dir)
818                 lod_comp =
819                         &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
820         else
821                 lod_comp = &lo->ldo_comp_entries[comp_idx];
822
823         magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
824         if (lod_comp->llc_pattern == 0) /* default striping */
825                 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
826
827         lmm->lmm_magic = cpu_to_le32(magic);
828         lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
829         fid_to_lmm_oi(fid, &lmm->lmm_oi);
830         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
831                 lmm->lmm_oi.oi.oi_id++;
832         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
833
834         lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
835         lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
836         /**
837          * for dir and uninstantiated component, lmm_layout_gen stores
838          * default stripe offset.
839          */
840         lmm->lmm_layout_gen =
841                 (is_dir || !lod_comp_inited(lod_comp)) ?
842                         cpu_to_le16(lod_comp->llc_stripe_offset) :
843                         cpu_to_le16(lod_comp->llc_layout_gen);
844
845         if (magic == LOV_MAGIC_V1) {
846                 objs = &lmm->lmm_objects[0];
847         } else {
848                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
849                 size_t cplen = strlcpy(v3->lmm_pool_name,
850                                        lod_comp->llc_pool,
851                                        sizeof(v3->lmm_pool_name));
852                 if (cplen >= sizeof(v3->lmm_pool_name))
853                         RETURN(-E2BIG);
854                 objs = &v3->lmm_objects[0];
855         }
856         stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
857         if (!is_dir && lo->ldo_is_composite)
858                 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
859
860         if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
861                 GOTO(done, rc = 0);
862
863         /* generate ost_idx of this component stripe */
864         lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
865         for (i = 0; i < stripe_count; i++) {
866                 struct dt_object *object;
867                 __u32 ost_idx = (__u32)-1UL;
868                 int type = LU_SEQ_RANGE_OST;
869
870                 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
871                         object = lod_comp->llc_stripe[i];
872                         /* instantiated component */
873                         info->lti_fid = *lu_object_fid(&object->do_lu);
874
875                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
876                             comp_idx == 0) {
877                                 if (cfs_fail_val == 0)
878                                         cfs_fail_val = info->lti_fid.f_oid;
879                                 else if (i == 0)
880                                         info->lti_fid.f_oid = cfs_fail_val;
881                         }
882
883                         rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
884                         LASSERT(rc == 0);
885
886                         ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
887                         objs[i].l_ost_gen = cpu_to_le32(0);
888                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
889                                 rc = -ENOENT;
890                         else
891                                 rc = lod_fld_lookup(env, lod, &info->lti_fid,
892                                                     &ost_idx, &type);
893                         if (rc < 0) {
894                                 CERROR("%s: Can not locate "DFID": rc = %d\n",
895                                        lod2obd(lod)->obd_name,
896                                        PFID(&info->lti_fid), rc);
897                                 RETURN(rc);
898                         }
899                 } else if (lod_comp->llc_ostlist.op_array) {
900                         /* user specified ost list */
901                         ost_idx = lod_comp->llc_ostlist.op_array[i];
902                 }
903                 /*
904                  * with un-instantiated or with no specified ost list
905                  * component, its l_ost_idx does not matter.
906                  */
907                 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
908         }
909 done:
910         if (lmm_size != NULL)
911                 *lmm_size = lov_mds_md_size(stripe_count, magic);
912         RETURN(rc);
913 }
914
915 /**
916  * Generate on-disk lov_mds_md structure based on the information in
917  * the lod_object->ldo_comp_entries.
918  *
919  * \param[in] env               execution environment for this thread
920  * \param[in] lo                LOD object
921  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
922  * \param[in|out] lmm_size      buffer size/lmm size
923  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
924  *                              the stripe info is from the default stripe
925  *                              template, which is collected in lod_ah_init(),
926  *                              either from parent object or root object; for
927  *                              file case, it's from the @lo object
928  *
929  * \retval                      0 if on disk structure is created successfully
930  * \retval                      negative error number on failure
931  */
932 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
933                        struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
934 {
935         struct lov_comp_md_entry_v1 *lcme;
936         struct lov_comp_md_v1 *lcm;
937         struct lod_layout_component *comp_entries;
938         __u16 comp_cnt, mirror_cnt;
939         bool is_composite;
940         int i, rc = 0, offset;
941         ENTRY;
942
943         if (is_dir) {
944                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
945                 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
946                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
947                 is_composite =
948                         lo->ldo_def_striping->lds_def_striping_is_composite;
949         } else {
950                 comp_cnt = lo->ldo_comp_cnt;
951                 mirror_cnt = lo->ldo_mirror_count;
952                 comp_entries = lo->ldo_comp_entries;
953                 is_composite = lo->ldo_is_composite;
954         }
955
956         LASSERT(lmm_size != NULL);
957         LASSERT(comp_cnt != 0 && comp_entries != NULL);
958
959         if (!is_composite) {
960                 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
961                 RETURN(rc);
962         }
963
964         lcm = (struct lov_comp_md_v1 *)lmm;
965         memset(lcm, 0, sizeof(*lcm));
966
967         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
968         lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
969         lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
970         lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
971
972         offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
973         LASSERT(offset % sizeof(__u64) == 0);
974
975         for (i = 0; i < comp_cnt; i++) {
976                 struct lod_layout_component *lod_comp;
977                 struct lov_mds_md *sub_md;
978                 int size;
979
980                 lod_comp = &comp_entries[i];
981                 lcme = &lcm->lcm_entries[i];
982
983                 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
984                 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
985
986                 /* component could be un-inistantiated */
987                 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
988                 lcme->lcme_extent.e_start =
989                         cpu_to_le64(lod_comp->llc_extent.e_start);
990                 lcme->lcme_extent.e_end =
991                         cpu_to_le64(lod_comp->llc_extent.e_end);
992                 lcme->lcme_offset = cpu_to_le32(offset);
993
994                 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
995                 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
996                 if (rc)
997                         GOTO(out, rc);
998                 lcme->lcme_size = cpu_to_le32(size);
999                 offset += size;
1000                 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
1001                          "offset:%d lmm_size:%d\n", offset, *lmm_size);
1002         }
1003         lcm->lcm_size = cpu_to_le32(offset);
1004         lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
1005
1006         lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
1007                              "generate lum");
1008 out:
1009         if (rc == 0)
1010                 *lmm_size = offset;
1011         RETURN(rc);
1012 }
1013
1014 /**
1015  * Get LOV EA.
1016  *
1017  * Fill lti_ea_store buffer in the environment with a value for the given
1018  * EA. The buffer is reallocated if the value doesn't fit.
1019  *
1020  * \param[in,out] env           execution environment for this thread
1021  *                              .lti_ea_store buffer is filled with EA's value
1022  * \param[in] lo                LOD object
1023  * \param[in] name              name of the EA
1024  *
1025  * \retval                      > 0 if EA is fetched successfully
1026  * \retval                      0 if EA is empty
1027  * \retval                      negative error number on failure
1028  */
1029 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1030                const char *name)
1031 {
1032         struct lod_thread_info  *info = lod_env_info(env);
1033         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1034         int                     rc;
1035         ENTRY;
1036
1037         LASSERT(info);
1038
1039         if (unlikely(info->lti_ea_store == NULL)) {
1040                 /* just to enter in allocation block below */
1041                 rc = -ERANGE;
1042         } else {
1043 repeat:
1044                 info->lti_buf.lb_buf = info->lti_ea_store;
1045                 info->lti_buf.lb_len = info->lti_ea_store_size;
1046                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1047         }
1048
1049         /* if object is not striped or inaccessible */
1050         if (rc == -ENODATA || rc == -ENOENT)
1051                 RETURN(0);
1052
1053         if (rc == -ERANGE) {
1054                 /* EA doesn't fit, reallocate new buffer */
1055                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1056                 if (rc == -ENODATA || rc == -ENOENT)
1057                         RETURN(0);
1058                 else if (rc < 0)
1059                         RETURN(rc);
1060
1061                 LASSERT(rc > 0);
1062                 rc = lod_ea_store_resize(info, rc);
1063                 if (rc)
1064                         RETURN(rc);
1065                 goto repeat;
1066         }
1067
1068         RETURN(rc);
1069 }
1070
1071 /**
1072  * Verify the target index is present in the current configuration.
1073  *
1074  * \param[in] md                LOD device where the target table is stored
1075  * \param[in] idx               target's index
1076  *
1077  * \retval                      0 if the index is present
1078  * \retval                      -EINVAL if not
1079  */
1080 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1081 {
1082         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1083                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1084                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1085                        md->lod_ost_descs.ltd_tgts_size);
1086                 return -EINVAL;
1087         }
1088
1089         if (unlikely(OST_TGT(md, idx) == NULL)) {
1090                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1091                        lod2obd(md)->obd_name, idx);
1092                 return -EINVAL;
1093         }
1094
1095         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1096                 CERROR("%s: invalid lod device, for idx: %d\n",
1097                        lod2obd(md)->obd_name , idx);
1098                 return -EINVAL;
1099         }
1100
1101         return 0;
1102 }
1103
1104 /**
1105  * Instantiate objects for stripes.
1106  *
1107  * Allocate and initialize LU-objects representing the stripes. The number
1108  * of the stripes (ldo_stripe_count) must be initialized already. The caller
1109  * must ensure nobody else is calling the function on the object at the same
1110  * time. FLDB service must be running to be able to map a FID to the targets
1111  * and find appropriate device representing that target.
1112  *
1113  * \param[in] env               execution environment for this thread
1114  * \param[in,out] lo            LOD object
1115  * \param[in] objs              an array of IDs to creates the objects from
1116  * \param[in] comp_idx          index of ldo_comp_entries
1117  *
1118  * \retval                      0 if the objects are instantiated successfully
1119  * \retval                      negative error number on failure
1120  */
1121 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1122                            struct lov_ost_data_v1 *objs, int comp_idx)
1123 {
1124         struct lod_layout_component     *lod_comp;
1125         struct lod_thread_info  *info = lod_env_info(env);
1126         struct lod_device       *md;
1127         struct lu_object        *o, *n;
1128         struct lu_device        *nd;
1129         struct dt_object       **stripe;
1130         int                      stripe_len;
1131         int                      i, rc = 0;
1132         __u32                   idx;
1133         ENTRY;
1134
1135         LASSERT(lo != NULL);
1136         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1137
1138         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1139         lod_comp = &lo->ldo_comp_entries[comp_idx];
1140
1141         LASSERT(lod_comp->llc_stripe == NULL);
1142         LASSERT(lod_comp->llc_stripe_count > 0);
1143         LASSERT(lod_comp->llc_stripe_size > 0);
1144
1145         stripe_len = lod_comp->llc_stripe_count;
1146         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1147         if (stripe == NULL)
1148                 RETURN(-ENOMEM);
1149
1150         for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1151                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1152                         continue;
1153
1154                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1155                 idx = le32_to_cpu(objs[i].l_ost_idx);
1156                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1157                 if (rc != 0)
1158                         GOTO(out, rc);
1159                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1160                          PFID(&info->lti_fid));
1161                 lod_getref(&md->lod_ost_descs);
1162
1163                 rc = validate_lod_and_idx(md, idx);
1164                 if (unlikely(rc != 0)) {
1165                         lod_putref(md, &md->lod_ost_descs);
1166                         GOTO(out, rc);
1167                 }
1168
1169                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1170                 lod_putref(md, &md->lod_ost_descs);
1171
1172                 /* In the function below, .hs_keycmp resolves to
1173                  * u_obj_hop_keycmp() */
1174                 /* coverity[overrun-buffer-val] */
1175                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1176                 if (IS_ERR(o))
1177                         GOTO(out, rc = PTR_ERR(o));
1178
1179                 n = lu_object_locate(o->lo_header, nd->ld_type);
1180                 LASSERT(n);
1181
1182                 stripe[i] = container_of(n, struct dt_object, do_lu);
1183         }
1184
1185 out:
1186         if (rc != 0) {
1187                 for (i = 0; i < stripe_len; i++)
1188                         if (stripe[i] != NULL)
1189                                 dt_object_put(env, stripe[i]);
1190
1191                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1192                 lod_comp->llc_stripe_count = 0;
1193         } else {
1194                 lod_comp->llc_stripe = stripe;
1195                 lod_comp->llc_stripes_allocated = stripe_len;
1196         }
1197
1198         RETURN(rc);
1199 }
1200
1201 /**
1202  * Instantiate objects for striping.
1203  *
1204  * Parse striping information in \a buf and instantiate the objects
1205  * representing the stripes.
1206  *
1207  * \param[in] env               execution environment for this thread
1208  * \param[in] lo                LOD object
1209  * \param[in] buf               buffer storing LOV EA to parse
1210  *
1211  * \retval                      0 if parsing and objects creation succeed
1212  * \retval                      negative error number on failure
1213  */
1214 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1215                        const struct lu_buf *buf)
1216 {
1217         struct lov_mds_md_v1    *lmm;
1218         struct lov_comp_md_v1   *comp_v1 = NULL;
1219         struct lov_ost_data_v1  *objs;
1220         __u32   magic, pattern;
1221         int     i, j, rc = 0;
1222         __u16   comp_cnt;
1223         __u16   mirror_cnt = 0;
1224         ENTRY;
1225
1226         LASSERT(buf);
1227         LASSERT(buf->lb_buf);
1228         LASSERT(buf->lb_len);
1229
1230         lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1231         magic = le32_to_cpu(lmm->lmm_magic);
1232
1233         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1234             magic != LOV_MAGIC_COMP_V1)
1235                 GOTO(out, rc = -EINVAL);
1236
1237         lod_free_comp_entries(lo);
1238
1239         if (magic == LOV_MAGIC_COMP_V1) {
1240                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1241                 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1242                 if (comp_cnt == 0)
1243                         GOTO(out, rc = -EINVAL);
1244                 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1245                 lo->ldo_is_composite = 1;
1246                 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1247                                         LCM_FL_FLR_MASK;
1248                 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1249         } else {
1250                 comp_cnt = 1;
1251                 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1252                 lo->ldo_is_composite = 0;
1253         }
1254
1255         rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1256         if (rc)
1257                 GOTO(out, rc);
1258
1259         for (i = 0; i < comp_cnt; i++) {
1260                 struct lod_layout_component     *lod_comp;
1261                 struct lu_extent        *ext;
1262                 __u32   offs;
1263
1264                 lod_comp = &lo->ldo_comp_entries[i];
1265                 if (lo->ldo_is_composite) {
1266                         offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1267                         lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1268                         magic = le32_to_cpu(lmm->lmm_magic);
1269
1270                         ext = &comp_v1->lcm_entries[i].lcme_extent;
1271                         lod_comp->llc_extent.e_start =
1272                                 le64_to_cpu(ext->e_start);
1273                         lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1274                         lod_comp->llc_flags =
1275                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1276                         lod_comp->llc_id =
1277                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1278                         if (lod_comp->llc_id == LCME_ID_INVAL)
1279                                 GOTO(out, rc = -EINVAL);
1280                 } else {
1281                         lod_comp_set_init(lod_comp);
1282                 }
1283
1284                 pattern = le32_to_cpu(lmm->lmm_pattern);
1285                 if (lov_pattern(pattern) != LOV_PATTERN_RAID0 &&
1286                     lov_pattern(pattern) != LOV_PATTERN_MDT)
1287                         GOTO(out, rc = -EINVAL);
1288
1289                 lod_comp->llc_pattern = pattern;
1290                 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1291                 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1292                 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1293
1294                 if (magic == LOV_MAGIC_V3) {
1295                         struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1296                         lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1297                         objs = &v3->lmm_objects[0];
1298                 } else {
1299                         lod_set_pool(&lod_comp->llc_pool, NULL);
1300                         objs = &lmm->lmm_objects[0];
1301                 }
1302
1303                 /**
1304                  * If uninstantiated template component has valid l_ost_idx,
1305                  * then user has specified ost list for this component.
1306                  */
1307                 if (!lod_comp_inited(lod_comp)) {
1308                         if (objs[0].l_ost_idx != (__u32)-1UL) {
1309                                 /**
1310                                  * load the user specified ost list, when this
1311                                  * component is instantiated later, it will be
1312                                  * used in lod_alloc_ost_list().
1313                                  */
1314                                 lod_comp->llc_ostlist.op_count =
1315                                         lod_comp->llc_stripe_count;
1316                                 lod_comp->llc_ostlist.op_size =
1317                                         lod_comp->llc_stripe_count *
1318                                         sizeof(__u32);
1319                                 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1320                                           lod_comp->llc_ostlist.op_size);
1321                                 if (!lod_comp->llc_ostlist.op_array)
1322                                         GOTO(out, rc = -ENOMEM);
1323
1324                                 for (j = 0; j < lod_comp->llc_stripe_count; j++)
1325                                         lod_comp->llc_ostlist.op_array[j] =
1326                                                 le32_to_cpu(objs[j].l_ost_idx);
1327
1328                                 /**
1329                                  * this component OST objects starts from the
1330                                  * first ost_idx, lod_alloc_ost_list() will
1331                                  * check this.
1332                                  */
1333                                 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1334                         } else {
1335                                 /**
1336                                  * for uninstantiated component,
1337                                  * lmm_layout_gen stores default stripe offset.
1338                                  */
1339                                 lod_comp->llc_stripe_offset =
1340                                                         lmm->lmm_layout_gen;
1341                         }
1342                 }
1343
1344                 /* skip un-instantiated component object initialization */
1345                 if (!lod_comp_inited(lod_comp))
1346                         continue;
1347
1348                 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1349                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1350                         rc = lod_initialize_objects(env, lo, objs, i);
1351                         if (rc)
1352                                 GOTO(out, rc);
1353                 }
1354         }
1355
1356         rc = lod_fill_mirrors(lo);
1357         if (rc)
1358                 GOTO(out, rc);
1359
1360 out:
1361         if (rc)
1362                 lod_object_free_striping(env, lo);
1363         RETURN(rc);
1364 }
1365
1366 /**
1367  * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1368  * is already cached.
1369  *
1370  * \param[in] lo        LOD object
1371  *
1372  * \retval              True if the striping is cached, otherwise
1373  *                      return false.
1374  */
1375 static bool lod_striping_loaded(struct lod_object *lo)
1376 {
1377         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1378             lo->ldo_comp_cached)
1379                 return true;
1380
1381         if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1382                 if (lo->ldo_stripe != NULL || lo->ldo_dir_stripe_loaded)
1383                         return true;
1384
1385                 /* Never load LMV stripe for slaves of striped dir */
1386                 if (lo->ldo_dir_slave_stripe)
1387                         return true;
1388         }
1389
1390         return false;
1391 }
1392
1393 /**
1394  * Initialize the object representing the stripes.
1395  *
1396  * Unless the stripes are initialized already, fetch LOV (for regular
1397  * objects) or LMV (for directory objects) EA and call lod_parse_striping()
1398  * to instantiate the objects representing the stripes. Caller should
1399  * hold the dt_write_lock(next).
1400  *
1401  * \param[in] env               execution environment for this thread
1402  * \param[in,out] lo            LOD object
1403  *
1404  * \retval                      0 if parsing and object creation succeed
1405  * \retval                      negative error number on failure
1406  */
1407 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
1408 {
1409         struct lod_thread_info  *info = lod_env_info(env);
1410         struct lu_buf           *buf  = &info->lti_buf;
1411         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1412         int                      rc = 0;
1413         ENTRY;
1414
1415         if (!dt_object_exists(next))
1416                 GOTO(out, rc = 0);
1417
1418         if (lod_striping_loaded(lo))
1419                 GOTO(out, rc = 0);
1420
1421         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1422                 rc = lod_get_lov_ea(env, lo);
1423                 if (rc <= 0)
1424                         GOTO(out, rc);
1425                 /*
1426                  * there is LOV EA (striping information) in this object
1427                  * let's parse it and create in-core objects for the stripes
1428                  */
1429                 buf->lb_buf = info->lti_ea_store;
1430                 buf->lb_len = info->lti_ea_store_size;
1431                 rc = lod_parse_striping(env, lo, buf);
1432                 if (rc == 0)
1433                         lo->ldo_comp_cached = 1;
1434         } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1435                 rc = lod_get_lmv_ea(env, lo);
1436                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
1437                         /* Let's set stripe_loaded to avoid further
1438                          * stripe loading especially for non-stripe directory,
1439                          * which can hurt performance. (See LU-9840)
1440                          */
1441                         if (rc == 0)
1442                                 lo->ldo_dir_stripe_loaded = 1;
1443                         GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1444                 }
1445                 buf->lb_buf = info->lti_ea_store;
1446                 buf->lb_len = info->lti_ea_store_size;
1447                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1448                         rc = lod_load_lmv_shards(env, lo, buf, true);
1449                         if (buf->lb_buf != info->lti_ea_store) {
1450                                 OBD_FREE_LARGE(info->lti_ea_store,
1451                                                info->lti_ea_store_size);
1452                                 info->lti_ea_store = buf->lb_buf;
1453                                 info->lti_ea_store_size = buf->lb_len;
1454                         }
1455
1456                         if (rc < 0)
1457                                 GOTO(out, rc);
1458                 }
1459
1460                 /*
1461                  * there is LMV EA (striping information) in this object
1462                  * let's parse it and create in-core objects for the stripes
1463                  */
1464                 rc = lod_parse_dir_striping(env, lo, buf);
1465                 if (rc == 0)
1466                         lo->ldo_dir_stripe_loaded = 1;
1467         }
1468 out:
1469         RETURN(rc);
1470 }
1471
1472 /**
1473  * A generic function to initialize the stripe objects.
1474  *
1475  * A protected version of lod_load_striping_locked() - load the striping
1476  * information from storage, parse that and instantiate LU objects to
1477  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1478  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1479  * return an array of references to the newly instantiated objects.
1480  *
1481  * \param[in] env               execution environment for this thread
1482  * \param[in,out] lo            LOD object, where striping is stored and
1483  *                              which gets an array of references
1484  *
1485  * \retval                      0 if parsing and object creation succeed
1486  * \retval                      negative error number on failure
1487  **/
1488 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1489 {
1490         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1491         int                     rc;
1492
1493         if (!dt_object_exists(next))
1494                 return 0;
1495
1496         /* Check without locking first */
1497         if (lod_striping_loaded(lo))
1498                 return 0;
1499
1500         /* currently this code is supposed to be called from declaration
1501          * phase only, thus the object is not expected to be locked by caller */
1502         dt_write_lock(env, next, 0);
1503         rc = lod_load_striping_locked(env, lo);
1504         dt_write_unlock(env, next);
1505         return rc;
1506 }
1507
1508 /**
1509  * Verify lov_user_md_v1/v3 striping.
1510  *
1511  * Check the validity of all fields including the magic, stripe size,
1512  * stripe count, stripe offset and that the pool is present.  Also check
1513  * that each target index points to an existing target. The additional
1514  * \a is_from_disk turns additional checks. In some cases zero fields
1515  * are allowed (like pattern=0).
1516  *
1517  * \param[in] d                 LOD device
1518  * \param[in] buf               buffer with LOV EA to verify
1519  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1520  *                              1 - from disk, do not allow
1521  *
1522  * \retval                      0 if the striping is valid
1523  * \retval                      -EINVAL if striping is invalid
1524  */
1525 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1526                            bool is_from_disk)
1527 {
1528         struct lov_user_md_v1   *lum;
1529         struct lov_user_md_v3   *lum3;
1530         struct pool_desc        *pool = NULL;
1531         __u32                    magic;
1532         __u32                    stripe_size;
1533         __u16                    stripe_count;
1534         __u16                    stripe_offset;
1535         size_t                   lum_size;
1536         int                      rc = 0;
1537         ENTRY;
1538
1539         lum = buf->lb_buf;
1540
1541         if (buf->lb_len < sizeof(*lum)) {
1542                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1543                        buf->lb_len);
1544                 GOTO(out, rc = -EINVAL);
1545         }
1546
1547         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1548         if (magic != LOV_USER_MAGIC_V1 &&
1549             magic != LOV_USER_MAGIC_V3 &&
1550             magic != LOV_USER_MAGIC_SPECIFIC) {
1551                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1552                        le32_to_cpu(lum->lmm_magic));
1553                 GOTO(out, rc = -EINVAL);
1554         }
1555
1556         /* the user uses "0" for default stripe pattern normally. */
1557         if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1558                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1559
1560         if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1561                 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1562                        le32_to_cpu(lum->lmm_pattern));
1563                 GOTO(out, rc = -EINVAL);
1564         }
1565
1566         /* a released lum comes from creating orphan on hsm release,
1567          * doesn't make sense to verify it. */
1568         if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1569                 GOTO(out, rc = 0);
1570
1571         /* 64kB is the largest common page size we see (ia64), and matches the
1572          * check in lfs */
1573         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1574         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1575                 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1576                        stripe_size, LOV_MIN_STRIPE_SIZE);
1577                 GOTO(out, rc = -EINVAL);
1578         }
1579
1580         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1581         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1582             lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1583                 /* if offset is not within valid range [0, osts_size) */
1584                 if (stripe_offset >= d->lod_osts_size) {
1585                         CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1586                                stripe_offset, d->lod_osts_size);
1587                         GOTO(out, rc = -EINVAL);
1588                 }
1589
1590                 /* if lmm_stripe_offset is *not* in bitmap */
1591                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1592                         CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1593                                stripe_offset);
1594                         GOTO(out, rc = -EINVAL);
1595                 }
1596         }
1597
1598         if (magic == LOV_USER_MAGIC_V1)
1599                 lum_size = offsetof(struct lov_user_md_v1,
1600                                     lmm_objects[0]);
1601         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1602                 lum_size = offsetof(struct lov_user_md_v3,
1603                                     lmm_objects[0]);
1604         else
1605                 GOTO(out, rc = -EINVAL);
1606
1607         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1608         if (buf->lb_len < lum_size) {
1609                 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1610                        "magic %#x and stripe_count %u\n",
1611                        buf->lb_len, lum_size, magic, stripe_count);
1612                 GOTO(out, rc = -EINVAL);
1613         }
1614
1615         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1616                 goto out;
1617
1618         lum3 = buf->lb_buf;
1619         /* In the function below, .hs_keycmp resolves to
1620          * pool_hashkey_keycmp() */
1621         /* coverity[overrun-buffer-val] */
1622         pool = lod_find_pool(d, lum3->lmm_pool_name);
1623         if (pool == NULL)
1624                 goto out;
1625
1626         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1627                 rc = lod_check_index_in_pool(stripe_offset, pool);
1628                 if (rc < 0)
1629                         GOTO(out, rc = -EINVAL);
1630         }
1631
1632         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1633                 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1634                        stripe_count, pool_tgt_count(pool));
1635                 GOTO(out, rc = -EINVAL);
1636         }
1637
1638 out:
1639         if (pool != NULL)
1640                 lod_pool_putref(pool);
1641
1642         RETURN(rc);
1643 }
1644
1645 /**
1646  * Verify LOV striping.
1647  *
1648  * \param[in] d                 LOD device
1649  * \param[in] buf               buffer with LOV EA to verify
1650  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1651  *                              1 - from disk, do not allow
1652  * \param[in] start             extent start for composite layout
1653  *
1654  * \retval                      0 if the striping is valid
1655  * \retval                      -EINVAL if striping is invalid
1656  */
1657 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1658                         const struct lu_buf *buf, bool is_from_disk)
1659 {
1660         struct lov_desc *desc = &d->lod_desc;
1661         struct lov_user_md_v1   *lum;
1662         struct lov_comp_md_v1   *comp_v1;
1663         struct lov_comp_md_entry_v1     *ent;
1664         struct lu_extent        *ext;
1665         struct lu_buf   tmp;
1666         __u64   prev_end = 0;
1667         __u32   stripe_size = 0;
1668         __u16   prev_mid = -1, mirror_id = -1;
1669         __u32   mirror_count = 0;
1670         __u32   magic;
1671         int     rc = 0, i;
1672         ENTRY;
1673
1674         lum = buf->lb_buf;
1675
1676         if (buf->lb_len < sizeof(*lum)) {
1677                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1678                        buf->lb_len);
1679                 RETURN(-EINVAL);
1680         }
1681
1682         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1683         if (magic != LOV_USER_MAGIC_V1 &&
1684             magic != LOV_USER_MAGIC_V3 &&
1685             magic != LOV_USER_MAGIC_SPECIFIC &&
1686             magic != LOV_USER_MAGIC_COMP_V1) {
1687                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1688                        le32_to_cpu(lum->lmm_magic));
1689                 RETURN(-EINVAL);
1690         }
1691
1692         if (magic != LOV_USER_MAGIC_COMP_V1)
1693                 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1694
1695         /* magic == LOV_USER_MAGIC_COMP_V1 */
1696         comp_v1 = buf->lb_buf;
1697         if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1698                 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1699                        buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1700                 RETURN(-EINVAL);
1701         }
1702
1703         if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1704                 CDEBUG(D_LAYOUT, "entry count is zero\n");
1705                 RETURN(-EINVAL);
1706         }
1707
1708         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1709             lo->ldo_comp_cnt > 0) {
1710                 /* could be called from lustre.lov.add */
1711                 __u32 cnt = lo->ldo_comp_cnt;
1712
1713                 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1714                 prev_end = ext->e_end;
1715
1716                 ++mirror_count;
1717         }
1718
1719         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1720                 ent = &comp_v1->lcm_entries[i];
1721                 ext = &ent->lcme_extent;
1722
1723                 if (le64_to_cpu(ext->e_start) >= le64_to_cpu(ext->e_end)) {
1724                         CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1725                                le64_to_cpu(ext->e_start),
1726                                le64_to_cpu(ext->e_end));
1727                         RETURN(-EINVAL);
1728                 }
1729
1730                 if (is_from_disk) {
1731                         /* lcme_id contains valid value */
1732                         if (le32_to_cpu(ent->lcme_id) == 0 ||
1733                             le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1734                                 CDEBUG(D_LAYOUT, "invalid id %u\n",
1735                                        le32_to_cpu(ent->lcme_id));
1736                                 RETURN(-EINVAL);
1737                         }
1738
1739                         if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1740                                 mirror_id = mirror_id_of(
1741                                                 le32_to_cpu(ent->lcme_id));
1742
1743                                 /* first component must start with 0 */
1744                                 if (mirror_id != prev_mid &&
1745                                     le64_to_cpu(ext->e_start) != 0) {
1746                                         CDEBUG(D_LAYOUT,
1747                                                "invalid start:%llu, expect:0\n",
1748                                                le64_to_cpu(ext->e_start));
1749                                         RETURN(-EINVAL);
1750                                 }
1751
1752                                 prev_mid = mirror_id;
1753                         }
1754                 }
1755
1756                 if (le64_to_cpu(ext->e_start) == 0) {
1757                         ++mirror_count;
1758                         prev_end = 0;
1759                 }
1760
1761                 /* the next must be adjacent with the previous one */
1762                 if (le64_to_cpu(ext->e_start) != prev_end) {
1763                         CDEBUG(D_LAYOUT,
1764                                "invalid start actual:%llu, expect:%llu\n",
1765                                le64_to_cpu(ext->e_start), prev_end);
1766                         RETURN(-EINVAL);
1767                 }
1768
1769                 prev_end = le64_to_cpu(ext->e_end);
1770
1771                 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1772                 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1773
1774                 /* Check DoM entry is always the first one */
1775                 lum = tmp.lb_buf;
1776                 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1777                     LOV_PATTERN_MDT) {
1778                         /* DoM component can be only the first entry */
1779                         if (i > 0) {
1780                                 CDEBUG(D_LAYOUT, "invalid DoM layout "
1781                                        "entry found at %i index\n", i);
1782                                 RETURN(-EINVAL);
1783                         }
1784                         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1785                         /* There is just one stripe on MDT and it must
1786                          * cover whole component size. */
1787                         if (stripe_size != prev_end) {
1788                                 CDEBUG(D_LAYOUT, "invalid DoM layout "
1789                                        "stripe size %u != %llu "
1790                                        "(component size)\n",
1791                                        stripe_size, prev_end);
1792                                 RETURN(-EINVAL);
1793                         }
1794                         /* Check stripe size againts per-MDT limit */
1795                         if (stripe_size > d->lod_dom_max_stripesize) {
1796                                 CDEBUG(D_LAYOUT, "DoM component size "
1797                                        "%u is bigger than MDT limit %u, check "
1798                                        "dom_max_stripesize parameter\n",
1799                                        stripe_size, d->lod_dom_max_stripesize);
1800                                 RETURN(-EINVAL);
1801                         }
1802                 }
1803
1804                 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1805                 if (rc)
1806                         RETURN(rc);
1807
1808                 if (prev_end == LUSTRE_EOF)
1809                         continue;
1810
1811                 /* extent end must be aligned with the stripe_size */
1812                 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1813                 if (stripe_size == 0)
1814                         stripe_size = desc->ld_default_stripe_size;
1815                 if (stripe_size == 0 || (prev_end & (stripe_size - 1))) {
1816                         CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
1817                                "stripe_sz: %u, [%llu, %llu)\n",
1818                                stripe_size, ext->e_start, prev_end);
1819                         RETURN(-EINVAL);
1820                 }
1821         }
1822
1823         /* make sure that the mirror_count is telling the truth */
1824         if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
1825                 RETURN(-EINVAL);
1826
1827         RETURN(0);
1828 }
1829
1830 /**
1831  * set the default stripe size, if unset.
1832  *
1833  * \param[in,out] val   number of bytes per OST stripe
1834  *
1835  * The minimum stripe size is 64KB to ensure that a single stripe is an
1836  * even multiple of a client PAGE_SIZE (IA64, PPC, etc).  Otherwise, it
1837  * is difficult to split dirty pages across OSCs during writes.
1838  */
1839 void lod_fix_desc_stripe_size(__u64 *val)
1840 {
1841         if (*val < LOV_MIN_STRIPE_SIZE) {
1842                 if (*val != 0)
1843                         LCONSOLE_INFO("Increasing default stripe size to "
1844                                       "minimum value %u\n",
1845                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
1846                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1847         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1848                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1849                 LCONSOLE_WARN("Changing default stripe size to %llu (a "
1850                               "multiple of %u)\n",
1851                               *val, LOV_MIN_STRIPE_SIZE);
1852         }
1853 }
1854
1855 /**
1856  * set the filesystem default number of stripes, if unset.
1857  *
1858  * \param[in,out] val   number of stripes
1859  *
1860  * A value of "0" means "use the system-wide default stripe count", which
1861  * has either been inherited by now, or falls back to 1 stripe per file.
1862  * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
1863  * and is a valid value, so is left unchanged here.
1864  */
1865 void lod_fix_desc_stripe_count(__u32 *val)
1866 {
1867         if (*val == 0)
1868                 *val = 1;
1869 }
1870
1871 /**
1872  * set the filesystem default layout pattern
1873  *
1874  * \param[in,out] val   LOV_PATTERN_* layout
1875  *
1876  * A value of "0" means "use the system-wide default layout type", which
1877  * has either been inherited by now, or falls back to plain RAID0 striping.
1878  */
1879 void lod_fix_desc_pattern(__u32 *val)
1880 {
1881         /* from lov_setstripe */
1882         if ((*val != 0) && (*val != LOV_PATTERN_RAID0) &&
1883             (*val != LOV_PATTERN_MDT)) {
1884                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1885                 *val = 0;
1886         }
1887 }
1888
1889 void lod_fix_desc_qos_maxage(__u32 *val)
1890 {
1891         /* fix qos_maxage */
1892         if (*val == 0)
1893                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1894 }
1895
1896 /**
1897  * Used to fix insane default striping.
1898  *
1899  * \param[in] desc      striping description
1900  */
1901 void lod_fix_desc(struct lov_desc *desc)
1902 {
1903         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1904         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1905         lod_fix_desc_pattern(&desc->ld_pattern);
1906         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1907 }
1908
1909 /**
1910  * Initialize the structures used to store pools and default striping.
1911  *
1912  * \param[in] lod       LOD device
1913  * \param[in] lcfg      configuration structure storing default striping.
1914  *
1915  * \retval              0 if initialization succeeds
1916  * \retval              negative error number on failure
1917  */
1918 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1919 {
1920         struct obd_device          *obd;
1921         struct lov_desc            *desc;
1922         int                         rc;
1923         ENTRY;
1924
1925         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1926         LASSERT(obd != NULL);
1927         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1928
1929         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1930                 CERROR("LOD setup requires a descriptor\n");
1931                 RETURN(-EINVAL);
1932         }
1933
1934         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1935
1936         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1937                 CERROR("descriptor size wrong: %d > %d\n",
1938                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1939                 RETURN(-EINVAL);
1940         }
1941
1942         if (desc->ld_magic != LOV_DESC_MAGIC) {
1943                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1944                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1945                                obd->obd_name, desc);
1946                         lustre_swab_lov_desc(desc);
1947                 } else {
1948                         CERROR("%s: Bad lov desc magic: %#x\n",
1949                                obd->obd_name, desc->ld_magic);
1950                         RETURN(-EINVAL);
1951                 }
1952         }
1953
1954         lod_fix_desc(desc);
1955
1956         desc->ld_active_tgt_count = 0;
1957         lod->lod_desc = *desc;
1958
1959         lod->lod_sp_me = LUSTRE_SP_CLI;
1960
1961         /* Set up allocation policy (QoS and RR) */
1962         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1963         init_rwsem(&lod->lod_qos.lq_rw_sem);
1964         lod->lod_qos.lq_dirty = 1;
1965         lod->lod_qos.lq_rr.lqr_dirty = 1;
1966         lod->lod_qos.lq_reset = 1;
1967         /* Default priority is toward free space balance */
1968         lod->lod_qos.lq_prio_free = 232;
1969         /* Default threshold for rr (roughly 17%) */
1970         lod->lod_qos.lq_threshold_rr = 43;
1971
1972         /* Set up OST pool environment */
1973         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1974                                                    HASH_POOLS_MAX_BITS,
1975                                                    HASH_POOLS_BKT_BITS, 0,
1976                                                    CFS_HASH_MIN_THETA,
1977                                                    CFS_HASH_MAX_THETA,
1978                                                    &pool_hash_operations,
1979                                                    CFS_HASH_DEFAULT);
1980         if (lod->lod_pools_hash_body == NULL)
1981                 RETURN(-ENOMEM);
1982
1983         INIT_LIST_HEAD(&lod->lod_pool_list);
1984         lod->lod_pool_count = 0;
1985         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1986         if (rc)
1987                 GOTO(out_hash, rc);
1988         lod_qos_rr_init(&lod->lod_qos.lq_rr);
1989         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1990         if (rc)
1991                 GOTO(out_pool_info, rc);
1992
1993         RETURN(0);
1994
1995 out_pool_info:
1996         lod_ost_pool_free(&lod->lod_pool_info);
1997 out_hash:
1998         cfs_hash_putref(lod->lod_pools_hash_body);
1999
2000         return rc;
2001 }
2002
2003 /**
2004  * Release the structures describing the pools.
2005  *
2006  * \param[in] lod       LOD device from which we release the structures
2007  *
2008  * \retval              0 always
2009  */
2010 int lod_pools_fini(struct lod_device *lod)
2011 {
2012         struct obd_device   *obd = lod2obd(lod);
2013         struct pool_desc    *pool, *tmp;
2014         ENTRY;
2015
2016         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2017                 /* free pool structs */
2018                 CDEBUG(D_INFO, "delete pool %p\n", pool);
2019                 /* In the function below, .hs_keycmp resolves to
2020                  * pool_hashkey_keycmp() */
2021                 /* coverity[overrun-buffer-val] */
2022                 lod_pool_del(obd, pool->pool_name);
2023         }
2024
2025         cfs_hash_putref(lod->lod_pools_hash_body);
2026         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
2027         lod_ost_pool_free(&lod->lod_pool_info);
2028
2029         RETURN(0);
2030 }