Whamcloud - gitweb
LU-13717 sec: filename encryption
[fs/lustre-release.git] / lustre / lod / lod_lov.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License version 2 for more details.  A copy is
14  * included in the COPYING file that accompanied this code.
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2017, Intel Corporation.
27  */
28 /*
29  * lustre/lod/lod_lov.c
30  *
31  * A set of helpers to maintain Logical Object Volume (LOV)
32  * Extended Attribute (EA) and known OST targets
33  *
34  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35  */
36
37 #define DEBUG_SUBSYSTEM S_MDS
38
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
43
44 #include "lod_internal.h"
45
46 /**
47  * Increase reference count on the target table.
48  *
49  * Increase reference count on the target table usage to prevent racing with
50  * addition/deletion. Any function that expects the table to remain
51  * stationary must take a ref.
52  *
53  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
54  */
55 void lod_getref(struct lod_tgt_descs *ltd)
56 {
57         down_read(&ltd->ltd_rw_sem);
58         mutex_lock(&ltd->ltd_mutex);
59         ltd->ltd_refcount++;
60         mutex_unlock(&ltd->ltd_mutex);
61 }
62
63 /**
64  * Decrease reference count on the target table.
65  *
66  * Companion of lod_getref() to release a reference on the target table.
67  * If this is the last reference and the OST entry was scheduled for deletion,
68  * the descriptor is removed from the table.
69  *
70  * \param[in] lod       LOD device from which we release a reference
71  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
72  */
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
74 {
75         mutex_lock(&ltd->ltd_mutex);
76         ltd->ltd_refcount--;
77         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78                 struct lod_tgt_desc *tgt_desc, *tmp;
79                 LIST_HEAD(kill);
80
81                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
82                        ltd->ltd_death_row);
83
84                 ltd_foreach_tgt_safe(ltd, tgt_desc, tmp) {
85                         LASSERT(tgt_desc);
86                         if (!tgt_desc->ltd_reap)
87                                 continue;
88
89                         list_add(&tgt_desc->ltd_kill, &kill);
90                         lu_tgt_pool_remove(&ltd->ltd_tgt_pool,
91                                            tgt_desc->ltd_index);
92                         ltd_del_tgt(ltd, tgt_desc);
93                         ltd->ltd_death_row--;
94                 }
95                 mutex_unlock(&ltd->ltd_mutex);
96                 up_read(&ltd->ltd_rw_sem);
97
98                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
99                         int rc;
100
101                         list_del(&tgt_desc->ltd_kill);
102                         rc = obd_disconnect(tgt_desc->ltd_exp);
103                         if (rc)
104                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
105                                        lod2obd(lod)->obd_name,
106                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
107                         OBD_FREE_PTR(tgt_desc);
108                 }
109         } else {
110                 mutex_unlock(&ltd->ltd_mutex);
111                 up_read(&ltd->ltd_rw_sem);
112         }
113 }
114
115 /**
116  * Connect LOD to a new OSP and add it to the target table.
117  *
118  * Connect to the OSP device passed, initialize all the internal
119  * structures related to the device and add it to the target table.
120  *
121  * \param[in] env               execution environment for this thread
122  * \param[in] lod               LOD device to be connected to the new OSP
123  * \param[in] osp               name of OSP device name to be added
124  * \param[in] index             index of the new target
125  * \param[in] gen               target's generation number
126  * \param[in] tgt_index         OSP's group
127  * \param[in] type              type of device (mdc or osc)
128  * \param[in] active            state of OSP: 0 - inactive, 1 - active
129  *
130  * \retval                      0 if added successfully
131  * \retval                      negative error number on failure
132  */
133 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
134                    char *osp, unsigned index, unsigned gen, int tgt_index,
135                    char *type, int active)
136 {
137         struct obd_connect_data *data = NULL;
138         struct obd_export       *exp = NULL;
139         struct obd_device       *obd;
140         struct lu_device        *lu_dev;
141         struct dt_device        *dt_dev;
142         int                      rc;
143         struct lod_tgt_desc     *tgt_desc;
144         struct lod_tgt_descs    *ltd;
145         struct lustre_cfg       *lcfg;
146         struct obd_uuid         obd_uuid;
147         bool                    for_ost;
148         bool connected = false;
149         ENTRY;
150
151         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
152
153         if (gen <= 0) {
154                 CERROR("request to add OBD %s with invalid generation: %d\n",
155                        osp, gen);
156                 RETURN(-EINVAL);
157         }
158
159         obd_str2uuid(&obd_uuid, osp);
160
161         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
162                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
163         if (obd == NULL) {
164                 CERROR("can't find %s device\n", osp);
165                 RETURN(-EINVAL);
166         }
167
168         LASSERT(obd->obd_lu_dev != NULL);
169         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
170
171         lu_dev = obd->obd_lu_dev;
172         dt_dev = lu2dt_dev(lu_dev);
173
174         OBD_ALLOC_PTR(data);
175         if (data == NULL)
176                 GOTO(out_cleanup, rc = -ENOMEM);
177
178         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
179         data->ocd_version = LUSTRE_VERSION_CODE;
180         data->ocd_index = index;
181
182         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
183                 for_ost = true;
184                 data->ocd_connect_flags |= OBD_CONNECT_AT |
185                                            OBD_CONNECT_FULL20 |
186                                            OBD_CONNECT_INDEX |
187 #ifdef HAVE_LRU_RESIZE_SUPPORT
188                                            OBD_CONNECT_LRU_RESIZE |
189 #endif
190                                            OBD_CONNECT_MDS |
191                                            OBD_CONNECT_REQPORTAL |
192                                            OBD_CONNECT_SKIP_ORPHAN |
193                                            OBD_CONNECT_FID |
194                                            OBD_CONNECT_LVB_TYPE |
195                                            OBD_CONNECT_VERSION |
196                                            OBD_CONNECT_PINGLESS |
197                                            OBD_CONNECT_LFSCK |
198                                            OBD_CONNECT_BULK_MBITS;
199
200                 data->ocd_group = tgt_index;
201                 ltd = &lod->lod_ost_descs;
202         } else {
203                 struct obd_import *imp = obd->u.cli.cl_import;
204
205                 for_ost = false;
206                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
207                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
208                                            OBD_CONNECT_IBITS |
209                                            OBD_CONNECT_MDS_MDS |
210                                            OBD_CONNECT_FID |
211                                            OBD_CONNECT_AT |
212                                            OBD_CONNECT_FULL20 |
213                                            OBD_CONNECT_LFSCK |
214                                            OBD_CONNECT_BULK_MBITS;
215                 spin_lock(&imp->imp_lock);
216                 imp->imp_server_timeout = 1;
217                 spin_unlock(&imp->imp_lock);
218                 imp->imp_client->cli_request_portal = OUT_PORTAL;
219                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
220                       obd->obd_name);
221                 ltd = &lod->lod_mdt_descs;
222         }
223
224         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
225         OBD_FREE_PTR(data);
226         if (rc) {
227                 CERROR("%s: cannot connect to next dev %s (%d)\n",
228                        obd->obd_name, osp, rc);
229                 GOTO(out_cleanup, rc);
230         }
231         connected = true;
232
233         /* Allocate ost descriptor and fill it */
234         OBD_ALLOC_PTR(tgt_desc);
235         if (!tgt_desc)
236                 GOTO(out_cleanup, rc = -ENOMEM);
237
238         tgt_desc->ltd_tgt    = dt_dev;
239         tgt_desc->ltd_exp    = exp;
240         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
241         tgt_desc->ltd_gen    = gen;
242         tgt_desc->ltd_index  = index;
243         tgt_desc->ltd_active = active;
244
245         down_write(&ltd->ltd_rw_sem);
246         mutex_lock(&ltd->ltd_mutex);
247         rc = ltd_add_tgt(ltd, tgt_desc);
248         if (rc)
249                 GOTO(out_mutex, rc);
250
251         rc = lu_qos_add_tgt(&ltd->ltd_qos, tgt_desc);
252         if (rc)
253                 GOTO(out_del_tgt, rc);
254
255         rc = lu_tgt_pool_add(&ltd->ltd_tgt_pool, index,
256                           ltd->ltd_lov_desc.ld_tgt_count);
257         if (rc) {
258                 CERROR("%s: can't set up pool, failed with %d\n",
259                        obd->obd_name, rc);
260                 GOTO(out_del_tgt, rc);
261         }
262
263         mutex_unlock(&ltd->ltd_mutex);
264         up_write(&ltd->ltd_rw_sem);
265
266         if (lod->lod_recovery_completed)
267                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
268
269         if (!for_ost && lod->lod_initialized) {
270                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
271                 if (rc != 0) {
272                         CERROR("%s: cannot start llog on %s:rc = %d\n",
273                                lod2obd(lod)->obd_name, osp, rc);
274                         GOTO(out_ltd, rc);
275                 }
276         }
277
278         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
279         if (rc != 0) {
280                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
281                        "index = %u, rc = %d\n", osp, type, index, rc);
282                 GOTO(out_fini_llog, rc);
283         }
284         RETURN(rc);
285 out_fini_llog:
286         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
287                           &tgt_desc->ltd_recovery_task);
288 out_ltd:
289         down_write(&ltd->ltd_rw_sem);
290         mutex_lock(&ltd->ltd_mutex);
291         lu_tgt_pool_remove(&ltd->ltd_tgt_pool, index);
292 out_del_tgt:
293         ltd_del_tgt(ltd, tgt_desc);
294 out_mutex:
295         mutex_unlock(&ltd->ltd_mutex);
296         up_write(&ltd->ltd_rw_sem);
297         OBD_FREE_PTR(tgt_desc);
298 out_cleanup:
299         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
300          * objects from the MDT stack. See LU-7184. */
301         lcfg = &lod_env_info(env)->lti_lustre_cfg;
302         memset(lcfg, 0, sizeof(*lcfg));
303         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
304         lcfg->lcfg_command = LCFG_CLEANUP;
305         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
306
307         if (connected)
308                 obd_disconnect(exp);
309
310         return rc;
311 }
312
313 /**
314  * Schedule target removal from the target table.
315  *
316  * Mark the device as dead. The device is not removed here because it may
317  * still be in use. The device will be removed in lod_putref() when the
318  * last reference is released.
319  *
320  * \param[in] env               execution environment for this thread
321  * \param[in] lod               LOD device the target table belongs to
322  * \param[in] ltd               target table
323  * \param[in] tgt               target
324  */
325 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
326                              struct lod_tgt_descs *ltd, struct lu_tgt_desc *tgt)
327 {
328         lfsck_del_target(env, lod->lod_child, tgt->ltd_tgt, tgt->ltd_index,
329                          !ltd->ltd_is_mdt);
330
331         if (!tgt->ltd_reap) {
332                 tgt->ltd_reap = 1;
333                 ltd->ltd_death_row++;
334         }
335 }
336
337 /**
338  * Schedule removal of all the targets from the given target table.
339  *
340  * See more details in the description for __lod_del_device()
341  *
342  * \param[in] env               execution environment for this thread
343  * \param[in] lod               LOD device the target table belongs to
344  * \param[in] ltd               target table
345  *
346  * \retval                      0 always
347  */
348 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
349                  struct lod_tgt_descs *ltd)
350 {
351         struct lu_tgt_desc *tgt;
352
353         if (ltd->ltd_tgts_size <= 0)
354                 return 0;
355
356         lod_getref(ltd);
357         mutex_lock(&ltd->ltd_mutex);
358         ltd_foreach_tgt(ltd, tgt)
359                 __lod_del_device(env, lod, ltd, tgt);
360         mutex_unlock(&ltd->ltd_mutex);
361         lod_putref(lod, ltd);
362
363         lu_tgt_descs_fini(ltd);
364
365         return 0;
366 }
367
368 /**
369  * Remove device by name.
370  *
371  * Remove a device identified by \a osp from the target table. Given
372  * the device can be in use, the real deletion happens in lod_putref().
373  *
374  * \param[in] env               execution environment for this thread
375  * \param[in] lod               LOD device to be connected to the new OSP
376  * \param[in] ltd               target table
377  * \param[in] osp               name of OSP device to be removed
378  * \param[in] idx               index of the target
379  * \param[in] gen               generation number, not used currently
380  *
381  * \retval                      0 if the device was scheduled for removal
382  * \retval                      -EINVAL if no device was found
383  */
384 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
385                    struct lod_tgt_descs *ltd, char *osp, unsigned int idx,
386                    unsigned int gen)
387 {
388         struct obd_device *obd;
389         struct lu_tgt_desc *tgt;
390         struct obd_uuid uuid;
391         int rc = 0;
392
393         ENTRY;
394
395         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
396
397         obd_str2uuid(&uuid, osp);
398
399         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
400                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
401         if (obd == NULL) {
402                 CERROR("can't find %s device\n", osp);
403                 RETURN(-EINVAL);
404         }
405
406         if (gen <= 0) {
407                 CERROR("%s: request to remove OBD %s with invalid generation %d"
408                        "\n", obd->obd_name, osp, gen);
409                 RETURN(-EINVAL);
410         }
411
412         obd_str2uuid(&uuid,  osp);
413
414         lod_getref(ltd);
415         mutex_lock(&ltd->ltd_mutex);
416         tgt = LTD_TGT(ltd, idx);
417         /* check that the index is allocated in the bitmap */
418         if (!test_bit(idx, ltd->ltd_tgt_bitmap) || !tgt) {
419                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
420                 GOTO(out, rc = -EINVAL);
421         }
422
423         /* check that the UUID matches */
424         if (!obd_uuid_equals(&uuid, &tgt->ltd_uuid)) {
425                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
426                        obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), idx, osp);
427                 GOTO(out, rc = -EINVAL);
428         }
429
430         __lod_del_device(env, lod, ltd, tgt);
431         EXIT;
432 out:
433         mutex_unlock(&ltd->ltd_mutex);
434         lod_putref(lod, ltd);
435         return(rc);
436 }
437
438 /**
439  * Resize per-thread storage to hold specified size.
440  *
441  * A helper function to resize per-thread temporary storage. This storage
442  * is used to process LOV/LVM EAs and may be quite large. We do not want to
443  * allocate/release it every time, so instead we put it into the env and
444  * reallocate on demand. The memory is released when the correspondent thread
445  * is finished.
446  *
447  * \param[in] info              LOD-specific storage in the environment
448  * \param[in] size              new size to grow the buffer to
449
450  * \retval                      0 on success, -ENOMEM if reallocation failed
451  */
452 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
453 {
454         __u32 round = size_roundup_power2(size);
455
456         if (info->lti_ea_store) {
457                 LASSERT(info->lti_ea_store_size);
458                 LASSERT(info->lti_ea_store_size < round);
459                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
460                        info->lti_ea_store_size, round);
461                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
462                 info->lti_ea_store = NULL;
463                 info->lti_ea_store_size = 0;
464         }
465
466         OBD_ALLOC_LARGE(info->lti_ea_store, round);
467         if (info->lti_ea_store == NULL)
468                 RETURN(-ENOMEM);
469         info->lti_ea_store_size = round;
470
471         RETURN(0);
472 }
473
474 static void lod_free_comp_buffer(struct lod_layout_component *entries,
475                                  __u16 count, __u32 bufsize)
476 {
477         struct lod_layout_component *entry;
478         int i;
479
480         for (i = 0; i < count; i++) {
481                 entry = &entries[i];
482                 if (entry->llc_pool != NULL)
483                         lod_set_pool(&entry->llc_pool, NULL);
484                 if (entry->llc_ostlist.op_array)
485                         OBD_FREE(entry->llc_ostlist.op_array,
486                                  entry->llc_ostlist.op_size);
487                 LASSERT(entry->llc_stripe == NULL);
488                 LASSERT(entry->llc_stripes_allocated == 0);
489         }
490
491         if (bufsize != 0)
492                 OBD_FREE_LARGE(entries, bufsize);
493 }
494
495 void lod_free_def_comp_entries(struct lod_default_striping *lds)
496 {
497         lod_free_comp_buffer(lds->lds_def_comp_entries,
498                              lds->lds_def_comp_size_cnt,
499                              size_roundup_power2(
500                                      sizeof(*lds->lds_def_comp_entries) *
501                                      lds->lds_def_comp_size_cnt));
502         lds->lds_def_comp_entries = NULL;
503         lds->lds_def_comp_cnt = 0;
504         lds->lds_def_striping_is_composite = 0;
505         lds->lds_def_comp_size_cnt = 0;
506 }
507
508 /**
509  * Resize per-thread storage to hold default striping component entries
510  *
511  * A helper function to resize per-thread temporary storage. This storage
512  * is used to hold default LOV/LVM EAs and may be quite large. We do not want
513  * to allocate/release it every time, so instead we put it into the env and
514  * reallocate it on demand. The memory is released when the correspondent
515  * thread is finished.
516  *
517  * \param[in,out] lds           default striping
518  * \param[in] count             new component count to grow the buffer to
519
520  * \retval                      0 on success, -ENOMEM if reallocation failed
521  */
522 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
523 {
524         struct lod_layout_component *entries;
525         __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
526                                         count);
527         __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
528                                         lds->lds_def_comp_size_cnt);
529
530         if (new <= old)
531                 return 0;
532
533         OBD_ALLOC_LARGE(entries, new);
534         if (entries == NULL)
535                 return -ENOMEM;
536
537         if (lds->lds_def_comp_entries != NULL) {
538                 CDEBUG(D_INFO, "default striping component size %d is not "
539                        "enough, need %d\n", old, new);
540                 lod_free_def_comp_entries(lds);
541         }
542
543         lds->lds_def_comp_entries = entries;
544         lds->lds_def_comp_size_cnt = count;
545
546         RETURN(0);
547 }
548
549 void lod_free_comp_entries(struct lod_object *lo)
550 {
551         if (lo->ldo_mirrors) {
552                 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
553                 lo->ldo_mirrors = NULL;
554                 lo->ldo_mirror_count = 0;
555         }
556         lod_free_comp_buffer(lo->ldo_comp_entries,
557                              lo->ldo_comp_cnt,
558                              sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
559         lo->ldo_comp_entries = NULL;
560         lo->ldo_comp_cnt = 0;
561         lo->ldo_is_composite = 0;
562 }
563
564 int lod_alloc_comp_entries(struct lod_object *lo,
565                            int mirror_count, int comp_count)
566 {
567         LASSERT(comp_count != 0);
568         LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
569
570         if (mirror_count > 0) {
571                 OBD_ALLOC_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
572                 if (!lo->ldo_mirrors)
573                         return -ENOMEM;
574
575                 lo->ldo_mirror_count = mirror_count;
576         }
577
578         OBD_ALLOC_LARGE(lo->ldo_comp_entries,
579                         sizeof(*lo->ldo_comp_entries) * comp_count);
580         if (lo->ldo_comp_entries == NULL) {
581                 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, mirror_count);
582                 lo->ldo_mirror_count = 0;
583                 return -ENOMEM;
584         }
585
586         lo->ldo_comp_cnt = comp_count;
587         return 0;
588 }
589
590 int lod_fill_mirrors(struct lod_object *lo)
591 {
592         struct lod_layout_component *lod_comp;
593         int mirror_idx = -1;
594         __u16 mirror_id = 0xffff;
595         int i;
596         ENTRY;
597
598         LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
599
600         if (!lo->ldo_is_composite)
601                 RETURN(0);
602
603         lod_comp = &lo->ldo_comp_entries[0];
604         for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
605                 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
606                 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
607
608                 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
609                         lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
610                         lo->ldo_mirrors[mirror_idx].lme_prefer |= preferred;
611                         lo->ldo_mirrors[mirror_idx].lme_end = i;
612                         continue;
613                 }
614
615                 /* new mirror */
616                 ++mirror_idx;
617                 if (mirror_idx >= lo->ldo_mirror_count)
618                         RETURN(-EINVAL);
619
620                 mirror_id = mirror_id_of(lod_comp->llc_id);
621
622                 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
623                 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
624                 lo->ldo_mirrors[mirror_idx].lme_prefer = preferred;
625                 lo->ldo_mirrors[mirror_idx].lme_start = i;
626                 lo->ldo_mirrors[mirror_idx].lme_end = i;
627         }
628         if (mirror_idx != lo->ldo_mirror_count - 1)
629                 RETURN(-EINVAL);
630
631         RETURN(0);
632 }
633
634 /**
635  * Generate on-disk lov_mds_md structure for each layout component based on
636  * the information in lod_object->ldo_comp_entries[i].
637  *
638  * \param[in] env               execution environment for this thread
639  * \param[in] lo                LOD object
640  * \param[in] comp_idx          index of ldo_comp_entries
641  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
642  * \param[in|out] lmm_size      buffer size/lmm size
643  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
644  *                              the stripe info is from the default stripe
645  *                              template, which is collected in lod_ah_init(),
646  *                              either from parent object or root object; for
647  *                              file case, it's from the @lo object
648  *
649  * \retval                      0 if on disk structure is created successfully
650  * \retval                      negative error number on failure
651  */
652 static int lod_gen_component_ea(const struct lu_env *env,
653                                 struct lod_object *lo, int comp_idx,
654                                 struct lov_mds_md *lmm, int *lmm_size,
655                                 bool is_dir)
656 {
657         struct lod_thread_info  *info = lod_env_info(env);
658         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
659         struct lod_device       *lod;
660         struct lov_ost_data_v1  *objs;
661         struct lod_layout_component *lod_comp;
662         __u32   magic;
663         __u16 stripe_count;
664         int     i, rc = 0;
665         ENTRY;
666
667         LASSERT(lo);
668         if (is_dir)
669                 lod_comp =
670                         &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
671         else
672                 lod_comp = &lo->ldo_comp_entries[comp_idx];
673
674         magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
675         if (lod_comp->llc_pattern == 0) /* default striping */
676                 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
677
678         lmm->lmm_magic = cpu_to_le32(magic);
679         lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
680         fid_to_lmm_oi(fid, &lmm->lmm_oi);
681         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
682                 lmm->lmm_oi.oi.oi_id++;
683         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
684
685         lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
686         lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
687         /**
688          * for dir and uninstantiated component, lmm_layout_gen stores
689          * default stripe offset.
690          */
691         lmm->lmm_layout_gen =
692                 (is_dir || !lod_comp_inited(lod_comp)) ?
693                         cpu_to_le16(lod_comp->llc_stripe_offset) :
694                         cpu_to_le16(lod_comp->llc_layout_gen);
695
696         if (magic == LOV_MAGIC_V1) {
697                 objs = &lmm->lmm_objects[0];
698         } else {
699                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
700                 size_t cplen = strlcpy(v3->lmm_pool_name,
701                                        lod_comp->llc_pool,
702                                        sizeof(v3->lmm_pool_name));
703                 if (cplen >= sizeof(v3->lmm_pool_name))
704                         RETURN(-E2BIG);
705                 objs = &v3->lmm_objects[0];
706         }
707         stripe_count = lod_comp_entry_stripe_count(lo, comp_idx, is_dir);
708         if (stripe_count == 0 && !is_dir &&
709             !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
710             !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
711                 RETURN(-E2BIG);
712
713         if (!is_dir && lo->ldo_is_composite)
714                 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
715
716         if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
717                 GOTO(done, rc = 0);
718
719         /* generate ost_idx of this component stripe */
720         lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
721         for (i = 0; i < stripe_count; i++) {
722                 struct dt_object *object;
723                 __u32 ost_idx = (__u32)-1UL;
724                 int type = LU_SEQ_RANGE_OST;
725
726                 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
727                         object = lod_comp->llc_stripe[i];
728                         /* instantiated component */
729                         info->lti_fid = *lu_object_fid(&object->do_lu);
730
731                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
732                             comp_idx == 0) {
733                                 if (cfs_fail_val == 0)
734                                         cfs_fail_val = info->lti_fid.f_oid;
735                                 else if (i == 0)
736                                         info->lti_fid.f_oid = cfs_fail_val;
737                         }
738
739                         rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
740                         LASSERT(rc == 0);
741
742                         ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
743                         objs[i].l_ost_gen = cpu_to_le32(0);
744                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
745                                 rc = -ENOENT;
746                         else
747                                 rc = lod_fld_lookup(env, lod, &info->lti_fid,
748                                                     &ost_idx, &type);
749                         if (rc < 0) {
750                                 CERROR("%s: Can not locate "DFID": rc = %d\n",
751                                        lod2obd(lod)->obd_name,
752                                        PFID(&info->lti_fid), rc);
753                                 RETURN(rc);
754                         }
755                 } else if (lod_comp->llc_ostlist.op_array &&
756                            lod_comp->llc_ostlist.op_count) {
757                         /* user specified ost list */
758                         ost_idx = lod_comp->llc_ostlist.op_array[i];
759                 }
760                 /*
761                  * with un-instantiated or with no specified ost list
762                  * component, its l_ost_idx does not matter.
763                  */
764                 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
765         }
766 done:
767         if (lmm_size != NULL)
768                 *lmm_size = lov_mds_md_size(stripe_count, magic);
769         RETURN(rc);
770 }
771
772 /**
773  * Generate on-disk lov_mds_md structure based on the information in
774  * the lod_object->ldo_comp_entries.
775  *
776  * \param[in] env               execution environment for this thread
777  * \param[in] lo                LOD object
778  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
779  * \param[in|out] lmm_size      buffer size/lmm size
780  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
781  *                              the stripe info is from the default stripe
782  *                              template, which is collected in lod_ah_init(),
783  *                              either from parent object or root object; for
784  *                              file case, it's from the @lo object
785  *
786  * \retval                      0 if on disk structure is created successfully
787  * \retval                      negative error number on failure
788  */
789 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
790                        struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
791 {
792         struct lov_comp_md_entry_v1 *lcme;
793         struct lov_comp_md_v1 *lcm;
794         struct lod_layout_component *comp_entries;
795         __u16 comp_cnt, mirror_cnt;
796         bool is_composite, is_foreign = false;
797         int i, rc = 0, offset;
798         ENTRY;
799
800         if (is_dir) {
801                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
802                 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
803                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
804                 is_composite =
805                         lo->ldo_def_striping->lds_def_striping_is_composite;
806         } else {
807                 comp_cnt = lo->ldo_comp_cnt;
808                 mirror_cnt = lo->ldo_mirror_count;
809                 comp_entries = lo->ldo_comp_entries;
810                 is_composite = lo->ldo_is_composite;
811                 is_foreign = lo->ldo_is_foreign;
812         }
813
814         LASSERT(lmm_size != NULL);
815
816         if (is_foreign) {
817                 struct lov_foreign_md *lfm;
818
819                 lfm = (struct lov_foreign_md *)lmm;
820                 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
821                 /* need to store little-endian */
822                 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
823                         __swab32s(&lfm->lfm_magic);
824                         __swab32s(&lfm->lfm_length);
825                         __swab32s(&lfm->lfm_type);
826                         __swab32s(&lfm->lfm_flags);
827                 }
828                 *lmm_size = lo->ldo_foreign_lov_size;
829                 RETURN(0);
830         }
831
832         LASSERT(comp_cnt != 0 && comp_entries != NULL);
833
834         if (!is_composite) {
835                 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
836                 RETURN(rc);
837         }
838
839         lcm = (struct lov_comp_md_v1 *)lmm;
840         memset(lcm, 0, sizeof(*lcm));
841
842         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
843         lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
844         lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
845         lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
846
847         offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
848         LASSERT(offset % sizeof(__u64) == 0);
849
850         for (i = 0; i < comp_cnt; i++) {
851                 struct lod_layout_component *lod_comp;
852                 struct lov_mds_md *sub_md;
853                 int size;
854
855                 lod_comp = &comp_entries[i];
856                 lcme = &lcm->lcm_entries[i];
857
858                 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
859                 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
860
861                 /* component could be un-inistantiated */
862                 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
863                 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
864                         lcme->lcme_timestamp =
865                                 cpu_to_le64(lod_comp->llc_timestamp);
866                 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
867                         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
868
869                 lcme->lcme_extent.e_start =
870                         cpu_to_le64(lod_comp->llc_extent.e_start);
871                 lcme->lcme_extent.e_end =
872                         cpu_to_le64(lod_comp->llc_extent.e_end);
873                 lcme->lcme_offset = cpu_to_le32(offset);
874
875                 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
876                 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
877                 if (rc)
878                         GOTO(out, rc);
879                 lcme->lcme_size = cpu_to_le32(size);
880                 offset += size;
881                 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
882                          "offset:%d lmm_size:%d\n", offset, *lmm_size);
883         }
884         lcm->lcm_size = cpu_to_le32(offset);
885         lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
886
887         lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
888                              "generate lum");
889 out:
890         if (rc == 0)
891                 *lmm_size = offset;
892         RETURN(rc);
893 }
894
895 /**
896  * Get LOV EA.
897  *
898  * Fill lti_ea_store buffer in the environment with a value for the given
899  * EA. The buffer is reallocated if the value doesn't fit.
900  *
901  * \param[in,out] env           execution environment for this thread
902  *                              .lti_ea_store buffer is filled with EA's value
903  * \param[in] lo                LOD object
904  * \param[in] name              name of the EA
905  *
906  * \retval                      > 0 if EA is fetched successfully
907  * \retval                      0 if EA is empty
908  * \retval                      negative error number on failure
909  */
910 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
911                const char *name)
912 {
913         struct lod_thread_info  *info = lod_env_info(env);
914         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
915         int                     rc;
916         ENTRY;
917
918         LASSERT(info);
919
920         if (unlikely(info->lti_ea_store == NULL)) {
921                 /* just to enter in allocation block below */
922                 rc = -ERANGE;
923         } else {
924 repeat:
925                 info->lti_buf.lb_buf = info->lti_ea_store;
926                 info->lti_buf.lb_len = info->lti_ea_store_size;
927                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
928         }
929
930         /* if object is not striped or inaccessible */
931         if (rc == -ENODATA || rc == -ENOENT)
932                 RETURN(0);
933
934         if (rc == -ERANGE) {
935                 /* EA doesn't fit, reallocate new buffer */
936                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
937                 if (rc == -ENODATA || rc == -ENOENT)
938                         RETURN(0);
939                 else if (rc < 0)
940                         RETURN(rc);
941
942                 LASSERT(rc > 0);
943                 rc = lod_ea_store_resize(info, rc);
944                 if (rc)
945                         RETURN(rc);
946                 goto repeat;
947         }
948
949         RETURN(rc);
950 }
951
952 /**
953  * Verify the target index is present in the current configuration.
954  *
955  * \param[in] md                LOD device where the target table is stored
956  * \param[in] idx               target's index
957  *
958  * \retval                      0 if the index is present
959  * \retval                      -EINVAL if not
960  */
961 int validate_lod_and_idx(struct lod_device *md, __u32 idx)
962 {
963         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
964                      !test_bit(idx, md->lod_ost_bitmap))) {
965                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
966                        md->lod_ost_descs.ltd_tgts_size);
967                 return -EINVAL;
968         }
969
970         if (unlikely(OST_TGT(md, idx) == NULL)) {
971                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
972                        lod2obd(md)->obd_name, idx);
973                 return -EINVAL;
974         }
975
976         if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
977                 CERROR("%s: invalid lod device, for idx: %d\n",
978                        lod2obd(md)->obd_name , idx);
979                 return -EINVAL;
980         }
981
982         return 0;
983 }
984
985 /**
986  * Instantiate objects for stripes.
987  *
988  * Allocate and initialize LU-objects representing the stripes. The number
989  * of the stripes (ldo_stripe_count) must be initialized already. The caller
990  * must ensure nobody else is calling the function on the object at the same
991  * time. FLDB service must be running to be able to map a FID to the targets
992  * and find appropriate device representing that target.
993  *
994  * \param[in] env               execution environment for this thread
995  * \param[in,out] lo            LOD object
996  * \param[in] objs              an array of IDs to creates the objects from
997  * \param[in] comp_idx          index of ldo_comp_entries
998  *
999  * \retval                      0 if the objects are instantiated successfully
1000  * \retval                      negative error number on failure
1001  */
1002 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1003                            struct lov_ost_data_v1 *objs, int comp_idx)
1004 {
1005         struct lod_layout_component *lod_comp;
1006         struct lod_thread_info *info = lod_env_info(env);
1007         struct lod_device *md;
1008         struct lu_object *o, *n;
1009         struct lu_device *nd;
1010         struct dt_object **stripe = NULL;
1011         __u32 *ost_indices = NULL;
1012         int stripe_len;
1013         int i, rc = 0;
1014         __u32 idx;
1015         ENTRY;
1016
1017         LASSERT(lo != NULL);
1018         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1019
1020         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1021         lod_comp = &lo->ldo_comp_entries[comp_idx];
1022
1023         LASSERT(lod_comp->llc_stripe == NULL);
1024         LASSERT(lod_comp->llc_stripe_count > 0);
1025         LASSERT(lod_comp->llc_stripe_size > 0);
1026
1027         stripe_len = lod_comp->llc_stripe_count;
1028         OBD_ALLOC_PTR_ARRAY(stripe, stripe_len);
1029         if (stripe == NULL)
1030                 RETURN(-ENOMEM);
1031         OBD_ALLOC_PTR_ARRAY(ost_indices, stripe_len);
1032         if (!ost_indices)
1033                 GOTO(out, rc = -ENOMEM);
1034
1035         for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1036                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1037                         continue;
1038
1039                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1040                 idx = le32_to_cpu(objs[i].l_ost_idx);
1041                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1042                 if (rc != 0)
1043                         GOTO(out, rc);
1044                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1045                          PFID(&info->lti_fid));
1046                 lod_getref(&md->lod_ost_descs);
1047
1048                 rc = validate_lod_and_idx(md, idx);
1049                 if (unlikely(rc != 0)) {
1050                         lod_putref(md, &md->lod_ost_descs);
1051                         GOTO(out, rc);
1052                 }
1053
1054                 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1055                 lod_putref(md, &md->lod_ost_descs);
1056
1057                 /* In the function below, .hs_keycmp resolves to
1058                  * u_obj_hop_keycmp() */
1059                 /* coverity[overrun-buffer-val] */
1060                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1061                 if (IS_ERR(o))
1062                         GOTO(out, rc = PTR_ERR(o));
1063
1064                 n = lu_object_locate(o->lo_header, nd->ld_type);
1065                 LASSERT(n);
1066
1067                 stripe[i] = container_of(n, struct dt_object, do_lu);
1068                 ost_indices[i] = idx;
1069         }
1070
1071 out:
1072         if (rc != 0) {
1073                 for (i = 0; i < stripe_len; i++)
1074                         if (stripe[i] != NULL)
1075                                 dt_object_put(env, stripe[i]);
1076
1077                 OBD_FREE_PTR_ARRAY(stripe, stripe_len);
1078                 lod_comp->llc_stripe_count = 0;
1079                 if (ost_indices)
1080                         OBD_FREE_PTR_ARRAY(ost_indices, stripe_len);
1081         } else {
1082                 lod_comp->llc_stripe = stripe;
1083                 lod_comp->llc_ost_indices = ost_indices;
1084                 lod_comp->llc_stripes_allocated = stripe_len;
1085         }
1086
1087         RETURN(rc);
1088 }
1089
1090 /**
1091  * Instantiate objects for striping.
1092  *
1093  * Parse striping information in \a buf and instantiate the objects
1094  * representing the stripes.
1095  *
1096  * \param[in] env               execution environment for this thread
1097  * \param[in] lo                LOD object
1098  * \param[in] buf               buffer storing LOV EA to parse
1099  *
1100  * \retval                      0 if parsing and objects creation succeed
1101  * \retval                      negative error number on failure
1102  */
1103 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1104                        const struct lu_buf *buf)
1105 {
1106         struct lov_mds_md_v1 *lmm;
1107         struct lov_comp_md_v1 *comp_v1 = NULL;
1108         struct lov_foreign_md *foreign = NULL;
1109         struct lov_ost_data_v1 *objs;
1110         __u32 magic, pattern;
1111         __u16 mirror_cnt = 0;
1112         __u16 comp_cnt;
1113         int i, rc;
1114         ENTRY;
1115
1116         LASSERT(buf);
1117         LASSERT(buf->lb_buf);
1118         LASSERT(buf->lb_len);
1119         LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1120
1121         lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1122         magic = le32_to_cpu(lmm->lmm_magic);
1123
1124         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1125             magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1126             magic != LOV_MAGIC_SEL)
1127                 GOTO(out, rc = -EINVAL);
1128
1129         lod_striping_free_nolock(env, lo);
1130
1131         if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1132                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1133                 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1134                 if (comp_cnt == 0)
1135                         GOTO(out, rc = -EINVAL);
1136                 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1137                 lo->ldo_is_composite = 1;
1138                 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1139                 if (mirror_cnt > 1)
1140                         lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1141                                                         LCM_FL_FLR_MASK;
1142                 else
1143                         lo->ldo_flr_state = LCM_FL_NONE;
1144         } else if (magic == LOV_MAGIC_FOREIGN) {
1145                 size_t length;
1146
1147                 foreign = (struct lov_foreign_md *)buf->lb_buf;
1148                 length = offsetof(typeof(*foreign), lfm_value);
1149                 if (buf->lb_len < length ||
1150                     buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1151                         CDEBUG(D_LAYOUT,
1152                                "buf len %zu too small for lov_foreign_md\n",
1153                                buf->lb_len);
1154                         GOTO(out, rc = -EINVAL);
1155                 }
1156
1157                 /* just cache foreign LOV EA raw */
1158                 rc = lod_alloc_foreign_lov(lo, length);
1159                 if (rc)
1160                         GOTO(out, rc);
1161                 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1162                 GOTO(out, rc);
1163         } else {
1164                 comp_cnt = 1;
1165                 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1166                 lo->ldo_is_composite = 0;
1167         }
1168
1169         rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1170         if (rc)
1171                 GOTO(out, rc);
1172
1173         for (i = 0; i < comp_cnt; i++) {
1174                 struct lod_layout_component *lod_comp;
1175                 struct lu_extent *ext;
1176                 __u32 offs;
1177
1178                 lod_comp = &lo->ldo_comp_entries[i];
1179                 if (lo->ldo_is_composite) {
1180                         offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1181                         lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1182
1183                         ext = &comp_v1->lcm_entries[i].lcme_extent;
1184                         lod_comp->llc_extent.e_start =
1185                                 le64_to_cpu(ext->e_start);
1186                         if (lod_comp->llc_extent.e_start &
1187                             (LOV_MIN_STRIPE_SIZE - 1)) {
1188                                 CDEBUG(D_LAYOUT,
1189                                        "extent start %llu is not a multiple of min size %u\n",
1190                                        lod_comp->llc_extent.e_start,
1191                                        LOV_MIN_STRIPE_SIZE);
1192                                 GOTO(out, rc = -EINVAL);
1193                         }
1194
1195                         lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1196                         if (lod_comp->llc_extent.e_end != LUSTRE_EOF &&
1197                             lod_comp->llc_extent.e_end &
1198                             (LOV_MIN_STRIPE_SIZE - 1)) {
1199                                 CDEBUG(D_LAYOUT,
1200                                        "extent end %llu is not a multiple of min size %u\n",
1201                                        lod_comp->llc_extent.e_end,
1202                                        LOV_MIN_STRIPE_SIZE);
1203                                 GOTO(out, rc = -EINVAL);
1204                         }
1205
1206                         lod_comp->llc_flags =
1207                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1208                         if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1209                                 lod_comp->llc_timestamp = le64_to_cpu(
1210                                         comp_v1->lcm_entries[i].lcme_timestamp);
1211                         lod_comp->llc_id =
1212                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1213                         if (lod_comp->llc_id == LCME_ID_INVAL)
1214                                 GOTO(out, rc = -EINVAL);
1215
1216                         if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1217                             comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1218                                 struct lod_device *d =
1219                                         lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1220
1221                                 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1222                                       lod2obd(d)->obd_name,
1223                                       lod_comp->llc_flags, lod_comp->llc_id, i,
1224                                       PFID(lod_object_fid(lo)),
1225                                       le32_to_cpu(comp_v1->lcm_magic));
1226                         }
1227                 } else {
1228                         lod_comp_set_init(lod_comp);
1229                 }
1230
1231                 pattern = le32_to_cpu(lmm->lmm_pattern);
1232                 if (!lov_pattern_supported(lov_pattern(pattern)))
1233                         GOTO(out, rc = -EINVAL);
1234
1235                 lod_comp->llc_pattern = pattern;
1236                 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1237                 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1238                 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1239
1240                 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1241                         struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1242
1243                         lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1244                         objs = &v3->lmm_objects[0];
1245                 } else {
1246                         lod_set_pool(&lod_comp->llc_pool, NULL);
1247                         objs = &lmm->lmm_objects[0];
1248                 }
1249
1250                 /**
1251                  * If uninstantiated template component has valid l_ost_idx,
1252                  * then user has specified ost list for this component.
1253                  */
1254                 if (!lod_comp_inited(lod_comp)) {
1255                         __u16 stripe_count;
1256
1257                         if (objs[0].l_ost_idx != (__u32)-1UL) {
1258                                 int j;
1259
1260                                 stripe_count = lod_comp_entry_stripe_count(
1261                                                         lo, i, false);
1262                                 if (stripe_count == 0 &&
1263                                     !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1264                                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1265                                         GOTO(out, rc = -E2BIG);
1266                                 /**
1267                                  * load the user specified ost list, when this
1268                                  * component is instantiated later, it will be
1269                                  * used in lod_alloc_ost_list().
1270                                  */
1271                                 lod_comp->llc_ostlist.op_count = stripe_count;
1272                                 lod_comp->llc_ostlist.op_size =
1273                                         stripe_count * sizeof(__u32);
1274                                 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1275                                           lod_comp->llc_ostlist.op_size);
1276                                 if (!lod_comp->llc_ostlist.op_array)
1277                                         GOTO(out, rc = -ENOMEM);
1278
1279                                 for (j = 0; j < stripe_count; j++)
1280                                         lod_comp->llc_ostlist.op_array[j] =
1281                                                 le32_to_cpu(objs[j].l_ost_idx);
1282
1283                                 /**
1284                                  * this component OST objects starts from the
1285                                  * first ost_idx, lod_alloc_ost_list() will
1286                                  * check this.
1287                                  */
1288                                 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1289                         } else {
1290                                 /**
1291                                  * for uninstantiated component,
1292                                  * lmm_layout_gen stores default stripe offset.
1293                                  */
1294                                 lod_comp->llc_stripe_offset =
1295                                                         lmm->lmm_layout_gen;
1296                         }
1297                 }
1298
1299                 /* skip un-instantiated component object initialization */
1300                 if (!lod_comp_inited(lod_comp))
1301                         continue;
1302
1303                 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1304                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1305                         rc = lod_initialize_objects(env, lo, objs, i);
1306                         if (rc)
1307                                 GOTO(out, rc);
1308                 }
1309         }
1310
1311         rc = lod_fill_mirrors(lo);
1312         if (rc)
1313                 GOTO(out, rc);
1314
1315 out:
1316         if (rc)
1317                 lod_striping_free_nolock(env, lo);
1318         RETURN(rc);
1319 }
1320
1321 /**
1322  * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1323  * is already cached.
1324  *
1325  * \param[in] lo        LOD object
1326  *
1327  * \retval              True if the striping is cached, otherwise
1328  *                      return false.
1329  */
1330 static bool lod_striping_loaded(struct lod_object *lo)
1331 {
1332         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1333             lo->ldo_comp_cached)
1334                 return true;
1335
1336         if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1337                 if (lo->ldo_dir_stripe_loaded)
1338                         return true;
1339
1340                 /* Never load LMV stripe for slaves of striped dir */
1341                 if (lo->ldo_dir_slave_stripe)
1342                         return true;
1343         }
1344
1345         return false;
1346 }
1347
1348 /**
1349  * A generic function to initialize the stripe objects.
1350  *
1351  * A protected version of lod_striping_load_locked() - load the striping
1352  * information from storage, parse that and instantiate LU objects to
1353  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1354  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1355  * return an array of references to the newly instantiated objects.
1356  *
1357  * \param[in] env               execution environment for this thread
1358  * \param[in,out] lo            LOD object, where striping is stored and
1359  *                              which gets an array of references
1360  *
1361  * \retval                      0 if parsing and object creation succeed
1362  * \retval                      negative error number on failure
1363  **/
1364 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1365 {
1366         struct lod_thread_info *info = lod_env_info(env);
1367         struct dt_object *next = dt_object_child(&lo->ldo_obj);
1368         struct lu_buf *buf = &info->lti_buf;
1369         int rc = 0;
1370
1371         ENTRY;
1372
1373         if (!dt_object_exists(next))
1374                 RETURN(0);
1375
1376         if (lod_striping_loaded(lo))
1377                 RETURN(0);
1378
1379         mutex_lock(&lo->ldo_layout_mutex);
1380         if (lod_striping_loaded(lo))
1381                 GOTO(unlock, rc = 0);
1382
1383         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1384                 rc = lod_get_lov_ea(env, lo);
1385                 if (rc <= 0)
1386                         GOTO(unlock, rc);
1387
1388                 /*
1389                  * there is LOV EA (striping information) in this object
1390                  * let's parse it and create in-core objects for the stripes
1391                  */
1392                 buf->lb_buf = info->lti_ea_store;
1393                 buf->lb_len = info->lti_ea_store_size;
1394                 rc = lod_parse_striping(env, lo, buf);
1395                 if (rc == 0)
1396                         lo->ldo_comp_cached = 1;
1397         } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1398                 rc = lod_get_lmv_ea(env, lo);
1399                 if (rc > sizeof(struct lmv_foreign_md)) {
1400                         struct lmv_foreign_md *lfm = info->lti_ea_store;
1401
1402                         if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1403                                 lo->ldo_foreign_lmv = info->lti_ea_store;
1404                                 lo->ldo_foreign_lmv_size =
1405                                         info->lti_ea_store_size;
1406                                 info->lti_ea_store = NULL;
1407                                 info->lti_ea_store_size = 0;
1408
1409                                 lo->ldo_dir_stripe_loaded = 1;
1410                                 lo->ldo_dir_is_foreign = 1;
1411                                 GOTO(unlock, rc = 0);
1412                         }
1413                 }
1414
1415                 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1416                         /* Let's set stripe_loaded to avoid further
1417                          * stripe loading especially for non-stripe directory,
1418                          * which can hurt performance. (See LU-9840)
1419                          */
1420                         if (rc == 0)
1421                                 lo->ldo_dir_stripe_loaded = 1;
1422                         GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1423                 }
1424                 buf->lb_buf = info->lti_ea_store;
1425                 buf->lb_len = info->lti_ea_store_size;
1426                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1427                         rc = lod_load_lmv_shards(env, lo, buf, true);
1428                         if (buf->lb_buf != info->lti_ea_store) {
1429                                 OBD_FREE_LARGE(info->lti_ea_store,
1430                                                info->lti_ea_store_size);
1431                                 info->lti_ea_store = buf->lb_buf;
1432                                 info->lti_ea_store_size = buf->lb_len;
1433                         }
1434
1435                         if (rc < 0)
1436                                 GOTO(unlock, rc);
1437                 }
1438
1439                 /*
1440                  * there is LMV EA (striping information) in this object
1441                  * let's parse it and create in-core objects for the stripes
1442                  */
1443                 rc = lod_parse_dir_striping(env, lo, buf);
1444                 if (rc == 0)
1445                         lo->ldo_dir_stripe_loaded = 1;
1446         }
1447         EXIT;
1448 unlock:
1449         mutex_unlock(&lo->ldo_layout_mutex);
1450
1451         return rc;
1452 }
1453
1454 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1455                          const struct lu_buf *buf)
1456 {
1457         int rc;
1458
1459         ENTRY;
1460
1461         mutex_lock(&lo->ldo_layout_mutex);
1462         rc = lod_parse_striping(env, lo, buf);
1463         mutex_unlock(&lo->ldo_layout_mutex);
1464
1465         RETURN(rc);
1466 }
1467
1468 /**
1469  * Verify lov_user_md_v1/v3 striping.
1470  *
1471  * Check the validity of all fields including the magic, stripe size,
1472  * stripe count, stripe offset and that the pool is present.  Also check
1473  * that each target index points to an existing target. The additional
1474  * \a is_from_disk turns additional checks. In some cases zero fields
1475  * are allowed (like pattern=0).
1476  *
1477  * \param[in] d                 LOD device
1478  * \param[in] buf               buffer with LOV EA to verify
1479  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1480  *                              1 - from disk, do not allow
1481  *
1482  * \retval                      0 if the striping is valid
1483  * \retval                      -EINVAL if striping is invalid
1484  */
1485 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1486                            bool is_from_disk)
1487 {
1488         struct lov_user_md_v1   *lum;
1489         struct lov_user_md_v3   *lum3;
1490         struct pool_desc        *pool = NULL;
1491         __u32                    magic;
1492         __u32                    stripe_size;
1493         __u16                    stripe_count;
1494         __u16                    stripe_offset;
1495         size_t                   lum_size;
1496         int                      rc = 0;
1497         ENTRY;
1498
1499         lum = buf->lb_buf;
1500
1501         if (buf->lb_len < sizeof(*lum)) {
1502                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1503                        buf->lb_len);
1504                 GOTO(out, rc = -EINVAL);
1505         }
1506
1507         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1508         if (magic != LOV_USER_MAGIC_V1 &&
1509             magic != LOV_USER_MAGIC_V3 &&
1510             magic != LOV_USER_MAGIC_SPECIFIC) {
1511                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1512                        le32_to_cpu(lum->lmm_magic));
1513                 GOTO(out, rc = -EINVAL);
1514         }
1515
1516         /* the user uses "0" for default stripe pattern normally. */
1517         if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1518                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1519
1520         if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1521                 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1522                        le32_to_cpu(lum->lmm_pattern));
1523                 GOTO(out, rc = -EINVAL);
1524         }
1525
1526         /* a released lum comes from creating orphan on hsm release,
1527          * doesn't make sense to verify it. */
1528         if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1529                 GOTO(out, rc = 0);
1530
1531         /* 64kB is the largest common page size we see (ia64), and matches the
1532          * check in lfs */
1533         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1534         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1535                 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1536                        stripe_size, LOV_MIN_STRIPE_SIZE);
1537                 GOTO(out, rc = -EINVAL);
1538         }
1539
1540         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1541         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1542             lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1543                 /* if offset is not within valid range [0, osts_size) */
1544                 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1545                         CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1546                                stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1547                         GOTO(out, rc = -EINVAL);
1548                 }
1549
1550                 /* if lmm_stripe_offset is *not* in bitmap */
1551                 if (!test_bit(stripe_offset, d->lod_ost_bitmap)) {
1552                         CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1553                                stripe_offset);
1554                         GOTO(out, rc = -EINVAL);
1555                 }
1556         }
1557
1558         if (magic == LOV_USER_MAGIC_V1)
1559                 lum_size = offsetof(struct lov_user_md_v1,
1560                                     lmm_objects[0]);
1561         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1562                 lum_size = offsetof(struct lov_user_md_v3,
1563                                     lmm_objects[0]);
1564         else
1565                 GOTO(out, rc = -EINVAL);
1566
1567         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1568         if (buf->lb_len < lum_size) {
1569                 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1570                        "magic %#x and stripe_count %u\n",
1571                        buf->lb_len, lum_size, magic, stripe_count);
1572                 GOTO(out, rc = -EINVAL);
1573         }
1574
1575         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1576                 goto out;
1577
1578         lum3 = buf->lb_buf;
1579         /* In the function below, .hs_keycmp resolves to
1580          * pool_hashkey_keycmp() */
1581         /* coverity[overrun-buffer-val] */
1582         pool = lod_find_pool(d, lum3->lmm_pool_name);
1583         if (pool == NULL)
1584                 goto out;
1585
1586         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1587                 rc = lod_check_index_in_pool(stripe_offset, pool);
1588                 if (rc < 0)
1589                         GOTO(out, rc = -EINVAL);
1590         }
1591
1592         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1593                 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1594                        stripe_count, pool_tgt_count(pool));
1595                 GOTO(out, rc = -EINVAL);
1596         }
1597
1598 out:
1599         if (pool != NULL)
1600                 lod_pool_putref(pool);
1601
1602         RETURN(rc);
1603 }
1604
1605 static inline
1606 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1607 {
1608         LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1609                  LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1610                  le32_to_cpu(comp->lcm_magic));
1611         LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1612                  "bad index %d, max = %d\n",
1613                  i, le16_to_cpu(comp->lcm_entry_count));
1614
1615         return &comp->lcm_entries[i];
1616 }
1617
1618 #define for_each_comp_entry_v1(comp, entry) \
1619         for (entry = comp_entry_v1(comp, 0); \
1620              entry <= comp_entry_v1(comp, \
1621                                    le16_to_cpu(comp->lcm_entry_count) - 1); \
1622              entry++)
1623
1624 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1625                          struct lov_comp_md_entry_v1 *dom_ent)
1626 {
1627         struct lov_comp_md_entry_v1 *ent;
1628         __u16 entries;
1629         __u32 dom_off, dom_size, comp_size, off;
1630         void *src, *dst;
1631         unsigned int size, shift;
1632
1633         entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1634         LASSERT(entries > 0);
1635         comp_v1->lcm_entry_count = cpu_to_le16(entries);
1636
1637         comp_size = le32_to_cpu(comp_v1->lcm_size);
1638         dom_off = le32_to_cpu(dom_ent->lcme_offset);
1639         dom_size = le32_to_cpu(dom_ent->lcme_size);
1640
1641         /* all entries offsets are shifted by entry size at least */
1642         shift = sizeof(*dom_ent);
1643         for_each_comp_entry_v1(comp_v1, ent) {
1644                 off = le32_to_cpu(ent->lcme_offset);
1645                 if (off == dom_off) {
1646                         /* Entry deletion creates two holes in layout data:
1647                          * - hole in entries array
1648                          * - hole in layout data at dom_off with dom_size
1649                          *
1650                          * First memmove is one entry shift from next entry
1651                          * start with size up to dom_off in blob
1652                          */
1653                         dst = (void *)ent;
1654                         src = (void *)(ent + 1);
1655                         size = (unsigned long)((void *)comp_v1 + dom_off - src);
1656                         memmove(dst, src, size);
1657                         /* take 'off' from just moved entry */
1658                         off = le32_to_cpu(ent->lcme_offset);
1659                         /* second memmove is blob tail after 'off' up to
1660                          * component end
1661                          */
1662                         dst = (void *)comp_v1 + dom_off - sizeof(*ent);
1663                         src = (void *)comp_v1 + off;
1664                         size = (unsigned long)(comp_size - off);
1665                         memmove(dst, src, size);
1666                         /* all entries offsets after DoM entry are shifted by
1667                          * dom_size additionally
1668                          */
1669                         shift += dom_size;
1670                 }
1671                 ent->lcme_offset = cpu_to_le32(off - shift);
1672         }
1673         comp_v1->lcm_size = cpu_to_le32(comp_size - shift);
1674
1675         /* notify a caller to re-check entry */
1676         return -ERESTART;
1677 }
1678
1679 void lod_dom_stripesize_recalc(struct lod_device *d)
1680 {
1681         __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1682         __u32 max_size = d->lod_dom_stripesize_max_kb;
1683         __u32 def_size = d->lod_dom_stripesize_cur_kb;
1684
1685         /* use maximum allowed value if free space is above threshold */
1686         if (d->lod_lsfs_free_mb >= threshold_mb) {
1687                 def_size = max_size;
1688         } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1689                 def_size = 0;
1690         } else {
1691                 /* recalc threshold like it would be with def_size as max */
1692                 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1693                 if (d->lod_lsfs_free_mb < threshold_mb)
1694                         def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1695                 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1696                         def_size = max_t(unsigned int, def_size * 2,
1697                                          LOD_DOM_MIN_SIZE_KB);
1698         }
1699
1700         if (d->lod_dom_stripesize_cur_kb != def_size) {
1701                 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1702                        d->lod_dom_stripesize_cur_kb, def_size);
1703                 d->lod_dom_stripesize_cur_kb = def_size;
1704         }
1705 }
1706
1707 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1708                                       struct lod_device *d)
1709 {
1710         int rc;
1711
1712         /* set bfree as fraction of total space */
1713         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1714                 spin_lock(&d->lod_lsfs_lock);
1715                 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1716                                         min_t(int, cfs_fail_val, 100), 100);
1717                 GOTO(recalc, rc = 0);
1718         }
1719
1720         if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1721                 struct obd_statfs sfs;
1722
1723                 spin_lock(&d->lod_lsfs_lock);
1724                 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1725                         GOTO(unlock, rc = 0);
1726
1727                 d->lod_lsfs_age = ktime_get_seconds();
1728                 spin_unlock(&d->lod_lsfs_lock);
1729                 rc = dt_statfs(env, d->lod_child, &sfs);
1730                 if (rc) {
1731                         CDEBUG(D_LAYOUT,
1732                                "%s: failed to get OSD statfs: rc = %d\n",
1733                                lod2obd(d)->obd_name, rc);
1734                         GOTO(out, rc);
1735                 }
1736                 /* udpate local OSD cached statfs data */
1737                 spin_lock(&d->lod_lsfs_lock);
1738                 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1739                 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1740 recalc:
1741                 lod_dom_stripesize_recalc(d);
1742 unlock:
1743                 spin_unlock(&d->lod_lsfs_lock);
1744         }
1745 out:
1746         return d->lod_dom_stripesize_cur_kb << 10;
1747 }
1748
1749 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1750                               struct lov_comp_md_v1 *comp_v1,
1751                               struct lov_comp_md_entry_v1 *dom_ent,
1752                               __u32 stripe_size)
1753 {
1754         struct lov_comp_md_entry_v1 *ent;
1755         struct lu_extent *dom_ext, *ext;
1756         struct lov_user_md_v1 *lum;
1757         __u32 max_stripe_size;
1758         __u16 mid, dom_mid;
1759         int rc = 0;
1760         bool dom_next_entry = false;
1761
1762         dom_ext = &dom_ent->lcme_extent;
1763         dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1764         max_stripe_size = lod_dom_stripesize_limit(env, d);
1765
1766         /* Check stripe size againts current per-MDT limit */
1767         if (stripe_size <= max_stripe_size)
1768                 return 0;
1769
1770         lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1771         CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1772                stripe_size, max_stripe_size);
1773         lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1774
1775         /* In common case the DoM stripe is first entry in a mirror and
1776          * can be deleted only if it is not single entry in layout or
1777          * mirror, otherwise error should be returned.
1778          */
1779         for_each_comp_entry_v1(comp_v1, ent) {
1780                 if (ent == dom_ent)
1781                         continue;
1782
1783                 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1784                 if (mid != dom_mid)
1785                         continue;
1786
1787                 ext = &ent->lcme_extent;
1788                 if (ext->e_start != dom_ext->e_end)
1789                         continue;
1790
1791                 /* Found next component after the DoM one with the same
1792                  * mirror_id and adjust its start with DoM component end.
1793                  *
1794                  * NOTE: we are considering here that there can be only one
1795                  * DoM component in a file, all replicas are located on OSTs
1796                  * always and don't need adjustment since use own layouts.
1797                  */
1798                 ext->e_start = cpu_to_le64(max_stripe_size);
1799                 dom_next_entry = true;
1800                 break;
1801         }
1802
1803         if (max_stripe_size == 0) {
1804                 /* DoM component size is zero due to server setting, remove
1805                  * it from the layout but only if next component exists in
1806                  * the same mirror. That must be checked prior calling the
1807                  * lod_erase_dom_stripe().
1808                  */
1809                 if (!dom_next_entry)
1810                         return -EFBIG;
1811
1812                 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1813         } else {
1814                 /* Update DoM extent end finally */
1815                 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1816         }
1817
1818         return rc;
1819 }
1820
1821 /**
1822  * Verify LOV striping.
1823  *
1824  * \param[in] d                 LOD device
1825  * \param[in] buf               buffer with LOV EA to verify
1826  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1827  *                              1 - from disk, do not allow
1828  * \param[in] start             extent start for composite layout
1829  *
1830  * \retval                      0 if the striping is valid
1831  * \retval                      -EINVAL if striping is invalid
1832  */
1833 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1834                         struct lod_object *lo, const struct lu_buf *buf,
1835                         bool is_from_disk)
1836 {
1837         struct lov_user_md_v1   *lum;
1838         struct lov_comp_md_v1   *comp_v1;
1839         struct lov_comp_md_entry_v1     *ent;
1840         struct lu_extent        *ext;
1841         struct lu_buf   tmp;
1842         __u64   prev_end = 0;
1843         __u32   stripe_size = 0;
1844         __u16   prev_mid = -1, mirror_id = -1;
1845         __u32   mirror_count;
1846         __u32   magic;
1847         int     rc = 0;
1848         ENTRY;
1849
1850         if (buf->lb_len < sizeof(lum->lmm_magic)) {
1851                 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1852                 RETURN(-EINVAL);
1853         }
1854
1855         lum = buf->lb_buf;
1856
1857         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1858         /* treat foreign LOV EA/object case first
1859          * XXX is it expected to try setting again a foreign?
1860          * XXX should we care about different current vs new layouts ?
1861          */
1862         if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1863                 struct lov_foreign_md *lfm = buf->lb_buf;
1864
1865                 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1866                         CDEBUG(D_LAYOUT,
1867                                "buf len %zu < min lov_foreign_md size (%zu)\n",
1868                                buf->lb_len, offsetof(typeof(*lfm),
1869                                lfm_value));
1870                         RETURN(-EINVAL);
1871                 }
1872
1873                 if (foreign_size_le(lfm) > buf->lb_len) {
1874                         CDEBUG(D_LAYOUT,
1875                                "buf len %zu < this lov_foreign_md size (%zu)\n",
1876                                buf->lb_len, foreign_size_le(lfm));
1877                         RETURN(-EINVAL);
1878                 }
1879                 /* Don't do anything with foreign layouts */
1880                 RETURN(0);
1881         }
1882
1883         /* normal LOV/layout cases */
1884
1885         if (buf->lb_len < sizeof(*lum)) {
1886                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1887                        buf->lb_len);
1888                 RETURN(-EINVAL);
1889         }
1890
1891         if (magic != LOV_USER_MAGIC_V1 &&
1892             magic != LOV_USER_MAGIC_V3 &&
1893             magic != LOV_USER_MAGIC_SPECIFIC &&
1894             magic != LOV_USER_MAGIC_COMP_V1) {
1895                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1896                        le32_to_cpu(lum->lmm_magic));
1897                 RETURN(-EINVAL);
1898         }
1899
1900         if (magic != LOV_USER_MAGIC_COMP_V1)
1901                 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1902
1903         /* magic == LOV_USER_MAGIC_COMP_V1 */
1904         comp_v1 = buf->lb_buf;
1905         if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1906                 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1907                        buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1908                 RETURN(-EINVAL);
1909         }
1910
1911 recheck:
1912         mirror_count = 0;
1913         if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1914                 CDEBUG(D_LAYOUT, "entry count is zero\n");
1915                 RETURN(-EINVAL);
1916         }
1917
1918         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1919             lo->ldo_comp_cnt > 0) {
1920                 /* could be called from lustre.lov.add */
1921                 __u32 cnt = lo->ldo_comp_cnt;
1922
1923                 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1924                 prev_end = ext->e_end;
1925
1926                 ++mirror_count;
1927         }
1928
1929         for_each_comp_entry_v1(comp_v1, ent) {
1930                 ext = &ent->lcme_extent;
1931
1932                 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end) ||
1933                     le64_to_cpu(ext->e_start) & (LOV_MIN_STRIPE_SIZE - 1) ||
1934                     (le64_to_cpu(ext->e_end) != LUSTRE_EOF &&
1935                     le64_to_cpu(ext->e_end) & (LOV_MIN_STRIPE_SIZE - 1))) {
1936                         CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1937                                le64_to_cpu(ext->e_start),
1938                                le64_to_cpu(ext->e_end));
1939                         RETURN(-EINVAL);
1940                 }
1941
1942                 if (is_from_disk) {
1943                         /* lcme_id contains valid value */
1944                         if (le32_to_cpu(ent->lcme_id) == 0 ||
1945                             le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1946                                 CDEBUG(D_LAYOUT, "invalid id %u\n",
1947                                        le32_to_cpu(ent->lcme_id));
1948                                 RETURN(-EINVAL);
1949                         }
1950
1951                         if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1952                                 mirror_id = mirror_id_of(
1953                                                 le32_to_cpu(ent->lcme_id));
1954
1955                                 /* first component must start with 0 */
1956                                 if (mirror_id != prev_mid &&
1957                                     le64_to_cpu(ext->e_start) != 0) {
1958                                         CDEBUG(D_LAYOUT,
1959                                                "invalid start:%llu, expect:0\n",
1960                                                le64_to_cpu(ext->e_start));
1961                                         RETURN(-EINVAL);
1962                                 }
1963
1964                                 prev_mid = mirror_id;
1965                         }
1966                 }
1967
1968                 if (le64_to_cpu(ext->e_start) == 0) {
1969                         ++mirror_count;
1970                         prev_end = 0;
1971                 }
1972
1973                 /* the next must be adjacent with the previous one */
1974                 if (le64_to_cpu(ext->e_start) != prev_end) {
1975                         CDEBUG(D_LAYOUT,
1976                                "invalid start actual:%llu, expect:%llu\n",
1977                                le64_to_cpu(ext->e_start), prev_end);
1978                         RETURN(-EINVAL);
1979                 }
1980
1981                 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1982                 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1983
1984                 /* Check DoM entry is always the first one */
1985                 lum = tmp.lb_buf;
1986                 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1987                     LOV_PATTERN_MDT) {
1988                         /* DoM component must be the first in a mirror */
1989                         if (le64_to_cpu(ext->e_start) > 0) {
1990                                 CDEBUG(D_LAYOUT, "invalid DoM component "
1991                                        "with %llu extent start\n",
1992                                        le64_to_cpu(ext->e_start));
1993                                 RETURN(-EINVAL);
1994                         }
1995                         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1996                         /* There is just one stripe on MDT and it must
1997                          * cover whole component size. */
1998                         if (stripe_size != le64_to_cpu(ext->e_end)) {
1999                                 CDEBUG(D_LAYOUT, "invalid DoM layout "
2000                                        "stripe size %u != %llu "
2001                                        "(component size)\n",
2002                                        stripe_size, prev_end);
2003                                 RETURN(-EINVAL);
2004                         }
2005                         /* Check and adjust stripe size by per-MDT limit */
2006                         rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
2007                                                        stripe_size);
2008                         /* DoM entry was removed, re-check layout from start */
2009                         if (rc == -ERESTART)
2010                                 goto recheck;
2011                         else if (rc)
2012                                 RETURN(rc);
2013
2014                         /* Any stripe count is forbidden on DoM component */
2015                         if (lum->lmm_stripe_count) {
2016                                 CDEBUG(D_LAYOUT,
2017                                        "invalid DoM layout stripe count %u, must be 0\n",
2018                                        le16_to_cpu(lum->lmm_stripe_count));
2019                                 RETURN(-EINVAL);
2020                         }
2021
2022                         /* Any pool is forbidden on DoM component */
2023                         if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2024                                 struct lov_user_md_v3 *v3 = (void *)lum;
2025
2026                                 if (v3->lmm_pool_name[0] != '\0') {
2027                                         CDEBUG(D_LAYOUT,
2028                                                "DoM component cannot have pool assigned\n");
2029                                         RETURN(-EINVAL);
2030                                 }
2031                         }
2032                 }
2033
2034                 prev_end = le64_to_cpu(ext->e_end);
2035
2036                 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2037                 if (rc)
2038                         RETURN(rc);
2039
2040                 if (prev_end == LUSTRE_EOF || ext->e_start == prev_end)
2041                         continue;
2042
2043                 /* extent end must be aligned with the stripe_size */
2044                 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2045                 if (stripe_size && prev_end % stripe_size) {
2046                         CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2047                                "stripe_sz: %u, [%llu, %llu)\n",
2048                                stripe_size, ext->e_start, prev_end);
2049                         RETURN(-EINVAL);
2050                 }
2051         }
2052
2053         /* make sure that the mirror_count is telling the truth */
2054         if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2055                 RETURN(-EINVAL);
2056
2057         RETURN(0);
2058 }
2059
2060 /**
2061  * set the default stripe size, if unset.
2062  *
2063  * \param[in,out] val   number of bytes per OST stripe
2064  *
2065  * The minimum stripe size is 64KB to ensure that a single stripe is an
2066  * even multiple of a client PAGE_SIZE (IA64, PPC, etc).  Otherwise, it
2067  * is difficult to split dirty pages across OSCs during writes.
2068  */
2069 void lod_fix_desc_stripe_size(__u64 *val)
2070 {
2071         if (*val < LOV_MIN_STRIPE_SIZE) {
2072                 if (*val != 0)
2073                         LCONSOLE_INFO("Increasing default stripe size to "
2074                                       "minimum value %u\n",
2075                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
2076                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2077         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2078                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2079                 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2080                               "multiple of %u)\n",
2081                               *val, LOV_MIN_STRIPE_SIZE);
2082         }
2083 }
2084
2085 /**
2086  * set the filesystem default number of stripes, if unset.
2087  *
2088  * \param[in,out] val   number of stripes
2089  *
2090  * A value of "0" means "use the system-wide default stripe count", which
2091  * has either been inherited by now, or falls back to 1 stripe per file.
2092  * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2093  * and is a valid value, so is left unchanged here.
2094  */
2095 void lod_fix_desc_stripe_count(__u32 *val)
2096 {
2097         if (*val == 0)
2098                 *val = 1;
2099 }
2100
2101 /**
2102  * set the filesystem default layout pattern
2103  *
2104  * \param[in,out] val   LOV_PATTERN_* layout
2105  *
2106  * A value of "0" means "use the system-wide default layout type", which
2107  * has either been inherited by now, or falls back to plain RAID0 striping.
2108  */
2109 void lod_fix_desc_pattern(__u32 *val)
2110 {
2111         /* from lov_setstripe */
2112         if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2113                 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2114                 *val = 0;
2115         }
2116 }
2117
2118 void lod_fix_lmv_desc_pattern(__u32 *val)
2119 {
2120         if ((*val) && !lmv_is_known_hash_type(*val)) {
2121                 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2122                 *val = 0;
2123         }
2124 }
2125
2126 void lod_fix_desc_qos_maxage(__u32 *val)
2127 {
2128         /* fix qos_maxage */
2129         if (*val == 0)
2130                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2131 }
2132
2133 /**
2134  * Used to fix insane default striping.
2135  *
2136  * \param[in] desc      striping description
2137  */
2138 void lod_fix_desc(struct lov_desc *desc)
2139 {
2140         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2141         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2142         lod_fix_desc_pattern(&desc->ld_pattern);
2143         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2144 }
2145
2146 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2147 {
2148         desc->ld_active_tgt_count = 0;
2149         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2150         lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2151         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2152 }
2153
2154 /**
2155  * Initialize the structures used to store pools and default striping.
2156  *
2157  * \param[in] lod       LOD device
2158  * \param[in] lcfg      configuration structure storing default striping.
2159  *
2160  * \retval              0 if initialization succeeds
2161  * \retval              negative error number on failure
2162  */
2163 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2164 {
2165         struct obd_device          *obd;
2166         struct lov_desc            *desc;
2167         int                         rc;
2168         ENTRY;
2169
2170         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2171         LASSERT(obd != NULL);
2172         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2173
2174         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2175                 CERROR("LOD setup requires a descriptor\n");
2176                 RETURN(-EINVAL);
2177         }
2178
2179         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2180
2181         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2182                 CERROR("descriptor size wrong: %d > %d\n",
2183                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2184                 RETURN(-EINVAL);
2185         }
2186
2187         if (desc->ld_magic != LOV_DESC_MAGIC) {
2188                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2189                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2190                                obd->obd_name, desc);
2191                         lustre_swab_lov_desc(desc);
2192                 } else {
2193                         CERROR("%s: Bad lov desc magic: %#x\n",
2194                                obd->obd_name, desc->ld_magic);
2195                         RETURN(-EINVAL);
2196                 }
2197         }
2198
2199         lod_fix_desc(desc);
2200
2201         desc->ld_active_tgt_count = 0;
2202         lod->lod_ost_descs.ltd_lov_desc = *desc;
2203
2204         /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2205         lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2206
2207         lod->lod_sp_me = LUSTRE_SP_CLI;
2208
2209         /* Set up OST pool environment */
2210         lod->lod_pool_count = 0;
2211         rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2212         if (rc)
2213                 RETURN(-ENOMEM);
2214
2215         INIT_LIST_HEAD(&lod->lod_pool_list);
2216         lod->lod_pool_count = 0;
2217         rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2218         if (rc)
2219                 GOTO(out_hash, rc);
2220
2221         rc = lu_tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2222         if (rc)
2223                 GOTO(out_mdt_pool, rc);
2224
2225         rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2226         if (rc)
2227                 GOTO(out_mdt_rr_pool, rc);
2228
2229         rc = lu_tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2230         if (rc)
2231                 GOTO(out_ost_pool, rc);
2232
2233         RETURN(0);
2234
2235 out_ost_pool:
2236         lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2237 out_mdt_rr_pool:
2238         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2239 out_mdt_pool:
2240         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2241 out_hash:
2242         lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2243
2244         return rc;
2245 }
2246
2247 /**
2248  * Release the structures describing the pools.
2249  *
2250  * \param[in] lod       LOD device from which we release the structures
2251  *
2252  * \retval              0 always
2253  */
2254 int lod_pools_fini(struct lod_device *lod)
2255 {
2256         struct obd_device   *obd = lod2obd(lod);
2257         struct pool_desc    *pool, *tmp;
2258         ENTRY;
2259
2260         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2261                 /* free pool structs */
2262                 CDEBUG(D_INFO, "delete pool %p\n", pool);
2263                 /* In the function below, .hs_keycmp resolves to
2264                  * pool_hashkey_keycmp() */
2265                 /* coverity[overrun-buffer-val] */
2266                 lod_pool_del(obd, pool->pool_name);
2267         }
2268
2269         lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2270         lu_tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2271         lu_tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2272         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2273         lu_tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2274
2275         RETURN(0);
2276 }