Whamcloud - gitweb
LU-11023 quota: quota pools for OSTs
[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                         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 = 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         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 (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) || !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(lo->ldo_mirrors,
553                          sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
554                 lo->ldo_mirrors = NULL;
555                 lo->ldo_mirror_count = 0;
556         }
557         lod_free_comp_buffer(lo->ldo_comp_entries,
558                              lo->ldo_comp_cnt,
559                              sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
560         lo->ldo_comp_entries = NULL;
561         lo->ldo_comp_cnt = 0;
562         lo->ldo_is_composite = 0;
563 }
564
565 int lod_alloc_comp_entries(struct lod_object *lo,
566                            int mirror_count, int comp_count)
567 {
568         LASSERT(comp_count != 0);
569         LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
570
571         if (mirror_count > 0) {
572                 OBD_ALLOC(lo->ldo_mirrors,
573                           sizeof(*lo->ldo_mirrors) * mirror_count);
574                 if (!lo->ldo_mirrors)
575                         return -ENOMEM;
576
577                 lo->ldo_mirror_count = mirror_count;
578         }
579
580         OBD_ALLOC_LARGE(lo->ldo_comp_entries,
581                         sizeof(*lo->ldo_comp_entries) * comp_count);
582         if (lo->ldo_comp_entries == NULL) {
583                 OBD_FREE(lo->ldo_mirrors,
584                          sizeof(*lo->ldo_mirrors) * mirror_count);
585                 lo->ldo_mirror_count = 0;
586                 return -ENOMEM;
587         }
588
589         lo->ldo_comp_cnt = comp_count;
590         return 0;
591 }
592
593 int lod_fill_mirrors(struct lod_object *lo)
594 {
595         struct lod_layout_component *lod_comp;
596         int mirror_idx = -1;
597         __u16 mirror_id = 0xffff;
598         int i;
599         ENTRY;
600
601         LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
602
603         if (!lo->ldo_is_composite)
604                 RETURN(0);
605
606         lod_comp = &lo->ldo_comp_entries[0];
607         for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
608                 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
609                 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
610
611                 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
612                         lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
613                         lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
614                         lo->ldo_mirrors[mirror_idx].lme_end = i;
615                         continue;
616                 }
617
618                 /* new mirror */
619                 ++mirror_idx;
620                 if (mirror_idx >= lo->ldo_mirror_count)
621                         RETURN(-EINVAL);
622
623                 mirror_id = mirror_id_of(lod_comp->llc_id);
624
625                 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
626                 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
627                 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
628                 lo->ldo_mirrors[mirror_idx].lme_start = i;
629                 lo->ldo_mirrors[mirror_idx].lme_end = i;
630         }
631         if (mirror_idx != lo->ldo_mirror_count - 1)
632                 RETURN(-EINVAL);
633
634         RETURN(0);
635 }
636
637 /**
638  * Generate on-disk lov_mds_md structure for each layout component based on
639  * the information in lod_object->ldo_comp_entries[i].
640  *
641  * \param[in] env               execution environment for this thread
642  * \param[in] lo                LOD object
643  * \param[in] comp_idx          index of ldo_comp_entries
644  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
645  * \param[in|out] lmm_size      buffer size/lmm size
646  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
647  *                              the stripe info is from the default stripe
648  *                              template, which is collected in lod_ah_init(),
649  *                              either from parent object or root object; for
650  *                              file case, it's from the @lo object
651  *
652  * \retval                      0 if on disk structure is created successfully
653  * \retval                      negative error number on failure
654  */
655 static int lod_gen_component_ea(const struct lu_env *env,
656                                 struct lod_object *lo, int comp_idx,
657                                 struct lov_mds_md *lmm, int *lmm_size,
658                                 bool is_dir)
659 {
660         struct lod_thread_info  *info = lod_env_info(env);
661         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
662         struct lod_device       *lod;
663         struct lov_ost_data_v1  *objs;
664         struct lod_layout_component *lod_comp;
665         __u32   magic;
666         __u16 stripe_count;
667         int     i, rc = 0;
668         ENTRY;
669
670         LASSERT(lo);
671         if (is_dir)
672                 lod_comp =
673                         &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
674         else
675                 lod_comp = &lo->ldo_comp_entries[comp_idx];
676
677         magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
678         if (lod_comp->llc_pattern == 0) /* default striping */
679                 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
680
681         lmm->lmm_magic = cpu_to_le32(magic);
682         lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
683         fid_to_lmm_oi(fid, &lmm->lmm_oi);
684         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
685                 lmm->lmm_oi.oi.oi_id++;
686         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
687
688         lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
689         lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
690         /**
691          * for dir and uninstantiated component, lmm_layout_gen stores
692          * default stripe offset.
693          */
694         lmm->lmm_layout_gen =
695                 (is_dir || !lod_comp_inited(lod_comp)) ?
696                         cpu_to_le16(lod_comp->llc_stripe_offset) :
697                         cpu_to_le16(lod_comp->llc_layout_gen);
698
699         if (magic == LOV_MAGIC_V1) {
700                 objs = &lmm->lmm_objects[0];
701         } else {
702                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
703                 size_t cplen = strlcpy(v3->lmm_pool_name,
704                                        lod_comp->llc_pool,
705                                        sizeof(v3->lmm_pool_name));
706                 if (cplen >= sizeof(v3->lmm_pool_name))
707                         RETURN(-E2BIG);
708                 objs = &v3->lmm_objects[0];
709         }
710         stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
711         if (stripe_count == 0 && !is_dir &&
712             !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
713             !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
714                 RETURN(-E2BIG);
715
716         if (!is_dir && lo->ldo_is_composite)
717                 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
718
719         if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
720                 GOTO(done, rc = 0);
721
722         /* generate ost_idx of this component stripe */
723         lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
724         for (i = 0; i < stripe_count; i++) {
725                 struct dt_object *object;
726                 __u32 ost_idx = (__u32)-1UL;
727                 int type = LU_SEQ_RANGE_OST;
728
729                 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
730                         object = lod_comp->llc_stripe[i];
731                         /* instantiated component */
732                         info->lti_fid = *lu_object_fid(&object->do_lu);
733
734                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
735                             comp_idx == 0) {
736                                 if (cfs_fail_val == 0)
737                                         cfs_fail_val = info->lti_fid.f_oid;
738                                 else if (i == 0)
739                                         info->lti_fid.f_oid = cfs_fail_val;
740                         }
741
742                         rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
743                         LASSERT(rc == 0);
744
745                         ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
746                         objs[i].l_ost_gen = cpu_to_le32(0);
747                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
748                                 rc = -ENOENT;
749                         else
750                                 rc = lod_fld_lookup(env, lod, &info->lti_fid,
751                                                     &ost_idx, &type);
752                         if (rc < 0) {
753                                 CERROR("%s: Can not locate "DFID": rc = %d\n",
754                                        lod2obd(lod)->obd_name,
755                                        PFID(&info->lti_fid), rc);
756                                 RETURN(rc);
757                         }
758                 } else if (lod_comp->llc_ostlist.op_array &&
759                            lod_comp->llc_ostlist.op_count) {
760                         /* user specified ost list */
761                         ost_idx = lod_comp->llc_ostlist.op_array[i];
762                 }
763                 /*
764                  * with un-instantiated or with no specified ost list
765                  * component, its l_ost_idx does not matter.
766                  */
767                 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
768         }
769 done:
770         if (lmm_size != NULL)
771                 *lmm_size = lov_mds_md_size(stripe_count, magic);
772         RETURN(rc);
773 }
774
775 /**
776  * Generate on-disk lov_mds_md structure based on the information in
777  * the lod_object->ldo_comp_entries.
778  *
779  * \param[in] env               execution environment for this thread
780  * \param[in] lo                LOD object
781  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
782  * \param[in|out] lmm_size      buffer size/lmm size
783  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
784  *                              the stripe info is from the default stripe
785  *                              template, which is collected in lod_ah_init(),
786  *                              either from parent object or root object; for
787  *                              file case, it's from the @lo object
788  *
789  * \retval                      0 if on disk structure is created successfully
790  * \retval                      negative error number on failure
791  */
792 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
793                        struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
794 {
795         struct lov_comp_md_entry_v1 *lcme;
796         struct lov_comp_md_v1 *lcm;
797         struct lod_layout_component *comp_entries;
798         __u16 comp_cnt, mirror_cnt;
799         bool is_composite, is_foreign = false;
800         int i, rc = 0, offset;
801         ENTRY;
802
803         if (is_dir) {
804                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
805                 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
806                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
807                 is_composite =
808                         lo->ldo_def_striping->lds_def_striping_is_composite;
809         } else {
810                 comp_cnt = lo->ldo_comp_cnt;
811                 mirror_cnt = lo->ldo_mirror_count;
812                 comp_entries = lo->ldo_comp_entries;
813                 is_composite = lo->ldo_is_composite;
814                 is_foreign = lo->ldo_is_foreign;
815         }
816
817         LASSERT(lmm_size != NULL);
818
819         if (is_foreign) {
820                 struct lov_foreign_md *lfm;
821
822                 lfm = (struct lov_foreign_md *)lmm;
823                 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
824                 /* need to store little-endian */
825                 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
826                         __swab32s(&lfm->lfm_magic);
827                         __swab32s(&lfm->lfm_length);
828                         __swab32s(&lfm->lfm_type);
829                         __swab32s(&lfm->lfm_flags);
830                 }
831                 *lmm_size = lo->ldo_foreign_lov_size;
832                 RETURN(0);
833         }
834
835         LASSERT(comp_cnt != 0 && comp_entries != NULL);
836
837         if (!is_composite) {
838                 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
839                 RETURN(rc);
840         }
841
842         lcm = (struct lov_comp_md_v1 *)lmm;
843         memset(lcm, 0, sizeof(*lcm));
844
845         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
846         lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
847         lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
848         lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
849
850         offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
851         LASSERT(offset % sizeof(__u64) == 0);
852
853         for (i = 0; i < comp_cnt; i++) {
854                 struct lod_layout_component *lod_comp;
855                 struct lov_mds_md *sub_md;
856                 int size;
857
858                 lod_comp = &comp_entries[i];
859                 lcme = &lcm->lcm_entries[i];
860
861                 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
862                 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
863
864                 /* component could be un-inistantiated */
865                 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
866                 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
867                         lcme->lcme_timestamp =
868                                 cpu_to_le64(lod_comp->llc_timestamp);
869                 if (lod_comp->llc_flags & LCME_FL_EXTENSION && !is_dir)
870                         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
871
872                 lcme->lcme_extent.e_start =
873                         cpu_to_le64(lod_comp->llc_extent.e_start);
874                 lcme->lcme_extent.e_end =
875                         cpu_to_le64(lod_comp->llc_extent.e_end);
876                 lcme->lcme_offset = cpu_to_le32(offset);
877
878                 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
879                 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
880                 if (rc)
881                         GOTO(out, rc);
882                 lcme->lcme_size = cpu_to_le32(size);
883                 offset += size;
884                 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
885                          "offset:%d lmm_size:%d\n", offset, *lmm_size);
886         }
887         lcm->lcm_size = cpu_to_le32(offset);
888         lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
889
890         lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
891                              "generate lum");
892 out:
893         if (rc == 0)
894                 *lmm_size = offset;
895         RETURN(rc);
896 }
897
898 /**
899  * Get LOV EA.
900  *
901  * Fill lti_ea_store buffer in the environment with a value for the given
902  * EA. The buffer is reallocated if the value doesn't fit.
903  *
904  * \param[in,out] env           execution environment for this thread
905  *                              .lti_ea_store buffer is filled with EA's value
906  * \param[in] lo                LOD object
907  * \param[in] name              name of the EA
908  *
909  * \retval                      > 0 if EA is fetched successfully
910  * \retval                      0 if EA is empty
911  * \retval                      negative error number on failure
912  */
913 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
914                const char *name)
915 {
916         struct lod_thread_info  *info = lod_env_info(env);
917         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
918         int                     rc;
919         ENTRY;
920
921         LASSERT(info);
922
923         if (unlikely(info->lti_ea_store == NULL)) {
924                 /* just to enter in allocation block below */
925                 rc = -ERANGE;
926         } else {
927 repeat:
928                 info->lti_buf.lb_buf = info->lti_ea_store;
929                 info->lti_buf.lb_len = info->lti_ea_store_size;
930                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
931         }
932
933         /* if object is not striped or inaccessible */
934         if (rc == -ENODATA || rc == -ENOENT)
935                 RETURN(0);
936
937         if (rc == -ERANGE) {
938                 /* EA doesn't fit, reallocate new buffer */
939                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
940                 if (rc == -ENODATA || rc == -ENOENT)
941                         RETURN(0);
942                 else if (rc < 0)
943                         RETURN(rc);
944
945                 LASSERT(rc > 0);
946                 rc = lod_ea_store_resize(info, rc);
947                 if (rc)
948                         RETURN(rc);
949                 goto repeat;
950         }
951
952         RETURN(rc);
953 }
954
955 /**
956  * Verify the target index is present in the current configuration.
957  *
958  * \param[in] md                LOD device where the target table is stored
959  * \param[in] idx               target's index
960  *
961  * \retval                      0 if the index is present
962  * \retval                      -EINVAL if not
963  */
964 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
965 {
966         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
967                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
968                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
969                        md->lod_ost_descs.ltd_tgts_size);
970                 return -EINVAL;
971         }
972
973         if (unlikely(OST_TGT(md, idx) == NULL)) {
974                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
975                        lod2obd(md)->obd_name, idx);
976                 return -EINVAL;
977         }
978
979         if (unlikely(OST_TGT(md, idx)->ltd_tgt == NULL)) {
980                 CERROR("%s: invalid lod device, for idx: %d\n",
981                        lod2obd(md)->obd_name , idx);
982                 return -EINVAL;
983         }
984
985         return 0;
986 }
987
988 /**
989  * Instantiate objects for stripes.
990  *
991  * Allocate and initialize LU-objects representing the stripes. The number
992  * of the stripes (ldo_stripe_count) must be initialized already. The caller
993  * must ensure nobody else is calling the function on the object at the same
994  * time. FLDB service must be running to be able to map a FID to the targets
995  * and find appropriate device representing that target.
996  *
997  * \param[in] env               execution environment for this thread
998  * \param[in,out] lo            LOD object
999  * \param[in] objs              an array of IDs to creates the objects from
1000  * \param[in] comp_idx          index of ldo_comp_entries
1001  *
1002  * \retval                      0 if the objects are instantiated successfully
1003  * \retval                      negative error number on failure
1004  */
1005 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1006                            struct lov_ost_data_v1 *objs, int comp_idx)
1007 {
1008         struct lod_layout_component *lod_comp;
1009         struct lod_thread_info *info = lod_env_info(env);
1010         struct lod_device *md;
1011         struct lu_object *o, *n;
1012         struct lu_device *nd;
1013         struct dt_object **stripe = NULL;
1014         __u32 *ost_indices = NULL;
1015         int stripe_len;
1016         int i, rc = 0;
1017         __u32 idx;
1018         ENTRY;
1019
1020         LASSERT(lo != NULL);
1021         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1022
1023         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1024         lod_comp = &lo->ldo_comp_entries[comp_idx];
1025
1026         LASSERT(lod_comp->llc_stripe == NULL);
1027         LASSERT(lod_comp->llc_stripe_count > 0);
1028         LASSERT(lod_comp->llc_stripe_size > 0);
1029
1030         stripe_len = lod_comp->llc_stripe_count;
1031         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1032         if (stripe == NULL)
1033                 RETURN(-ENOMEM);
1034         OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1035         if (!ost_indices)
1036                 GOTO(out, rc = -ENOMEM);
1037
1038         for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1039                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1040                         continue;
1041
1042                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1043                 idx = le32_to_cpu(objs[i].l_ost_idx);
1044                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1045                 if (rc != 0)
1046                         GOTO(out, rc);
1047                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1048                          PFID(&info->lti_fid));
1049                 lod_getref(&md->lod_ost_descs);
1050
1051                 rc = validate_lod_and_idx(md, idx);
1052                 if (unlikely(rc != 0)) {
1053                         lod_putref(md, &md->lod_ost_descs);
1054                         GOTO(out, rc);
1055                 }
1056
1057                 nd = &OST_TGT(md, idx)->ltd_tgt->dd_lu_dev;
1058                 lod_putref(md, &md->lod_ost_descs);
1059
1060                 /* In the function below, .hs_keycmp resolves to
1061                  * u_obj_hop_keycmp() */
1062                 /* coverity[overrun-buffer-val] */
1063                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1064                 if (IS_ERR(o))
1065                         GOTO(out, rc = PTR_ERR(o));
1066
1067                 n = lu_object_locate(o->lo_header, nd->ld_type);
1068                 LASSERT(n);
1069
1070                 stripe[i] = container_of(n, struct dt_object, do_lu);
1071                 ost_indices[i] = idx;
1072         }
1073
1074 out:
1075         if (rc != 0) {
1076                 for (i = 0; i < stripe_len; i++)
1077                         if (stripe[i] != NULL)
1078                                 dt_object_put(env, stripe[i]);
1079
1080                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1081                 lod_comp->llc_stripe_count = 0;
1082                 if (ost_indices)
1083                         OBD_FREE(ost_indices,
1084                                  sizeof(*ost_indices) * stripe_len);
1085         } else {
1086                 lod_comp->llc_stripe = stripe;
1087                 lod_comp->llc_ost_indices = ost_indices;
1088                 lod_comp->llc_stripes_allocated = stripe_len;
1089         }
1090
1091         RETURN(rc);
1092 }
1093
1094 /**
1095  * Instantiate objects for striping.
1096  *
1097  * Parse striping information in \a buf and instantiate the objects
1098  * representing the stripes.
1099  *
1100  * \param[in] env               execution environment for this thread
1101  * \param[in] lo                LOD object
1102  * \param[in] buf               buffer storing LOV EA to parse
1103  *
1104  * \retval                      0 if parsing and objects creation succeed
1105  * \retval                      negative error number on failure
1106  */
1107 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1108                        const struct lu_buf *buf)
1109 {
1110         struct lov_mds_md_v1 *lmm;
1111         struct lov_comp_md_v1 *comp_v1 = NULL;
1112         struct lov_foreign_md *foreign = NULL;
1113         struct lov_ost_data_v1 *objs;
1114         __u32 magic, pattern;
1115         __u16 mirror_cnt = 0;
1116         __u16 comp_cnt;
1117         int i, rc;
1118         ENTRY;
1119
1120         LASSERT(buf);
1121         LASSERT(buf->lb_buf);
1122         LASSERT(buf->lb_len);
1123         LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1124
1125         lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1126         magic = le32_to_cpu(lmm->lmm_magic);
1127
1128         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1129             magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN &&
1130             magic != LOV_MAGIC_SEL)
1131                 GOTO(out, rc = -EINVAL);
1132
1133         if (lo->ldo_is_foreign)
1134                 lod_free_foreign_lov(lo);
1135         else
1136                 lod_free_comp_entries(lo);
1137
1138         if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1139                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1140                 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1141                 if (comp_cnt == 0)
1142                         GOTO(out, rc = -EINVAL);
1143                 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1144                 lo->ldo_is_composite = 1;
1145                 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1146                                         LCM_FL_FLR_MASK;
1147                 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1148         } else if (magic == LOV_MAGIC_FOREIGN) {
1149                 size_t length;
1150
1151                 foreign = (struct lov_foreign_md *)buf->lb_buf;
1152                 length = offsetof(typeof(*foreign), lfm_value);
1153                 if (buf->lb_len < length ||
1154                     buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1155                         CDEBUG(D_LAYOUT,
1156                                "buf len %zu too small for lov_foreign_md\n",
1157                                buf->lb_len);
1158                         GOTO(out, rc = -EINVAL);
1159                 }
1160
1161                 /* just cache foreign LOV EA raw */
1162                 rc = lod_alloc_foreign_lov(lo, length);
1163                 if (rc)
1164                         GOTO(out, rc);
1165                 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1166                 GOTO(out, rc);
1167         } else {
1168                 comp_cnt = 1;
1169                 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1170                 lo->ldo_is_composite = 0;
1171         }
1172
1173         rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1174         if (rc)
1175                 GOTO(out, rc);
1176
1177         for (i = 0; i < comp_cnt; i++) {
1178                 struct lod_layout_component *lod_comp;
1179                 struct lu_extent *ext;
1180                 __u32 offs;
1181
1182                 lod_comp = &lo->ldo_comp_entries[i];
1183                 if (lo->ldo_is_composite) {
1184                         offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1185                         lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1186
1187                         ext = &comp_v1->lcm_entries[i].lcme_extent;
1188                         lod_comp->llc_extent.e_start =
1189                                 le64_to_cpu(ext->e_start);
1190                         lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1191                         lod_comp->llc_flags =
1192                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1193                         if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1194                                 lod_comp->llc_timestamp = le64_to_cpu(
1195                                         comp_v1->lcm_entries[i].lcme_timestamp);
1196                         lod_comp->llc_id =
1197                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1198                         if (lod_comp->llc_id == LCME_ID_INVAL)
1199                                 GOTO(out, rc = -EINVAL);
1200
1201                         if ((lod_comp->llc_flags & LCME_FL_EXTENSION) &&
1202                             comp_v1->lcm_magic != cpu_to_le32(LOV_MAGIC_SEL)) {
1203                                 struct lod_device *d =
1204                                         lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1205
1206                                 CWARN("%s: EXTENSION flags=%x set on component[%u]=%x of non-SEL file "DFID" with magic=%#08x\n",
1207                                       lod2obd(d)->obd_name,
1208                                       lod_comp->llc_flags, lod_comp->llc_id, i,
1209                                       PFID(lod_object_fid(lo)),
1210                                       le32_to_cpu(comp_v1->lcm_magic));
1211                         }
1212                 } else {
1213                         lod_comp_set_init(lod_comp);
1214                 }
1215
1216                 pattern = le32_to_cpu(lmm->lmm_pattern);
1217                 if (!lov_pattern_supported(lov_pattern(pattern)))
1218                         GOTO(out, rc = -EINVAL);
1219
1220                 lod_comp->llc_pattern = pattern;
1221                 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1222                 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1223                 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1224
1225                 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1226                         struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1227
1228                         lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1229                         objs = &v3->lmm_objects[0];
1230                 } else {
1231                         lod_set_pool(&lod_comp->llc_pool, NULL);
1232                         objs = &lmm->lmm_objects[0];
1233                 }
1234
1235                 /**
1236                  * If uninstantiated template component has valid l_ost_idx,
1237                  * then user has specified ost list for this component.
1238                  */
1239                 if (!lod_comp_inited(lod_comp)) {
1240                         __u16 stripe_count;
1241
1242                         if (objs[0].l_ost_idx != (__u32)-1UL) {
1243                                 int j;
1244
1245                                 stripe_count = lod_comp_entry_stripe_count(
1246                                                         lo, lod_comp, false);
1247                                 if (stripe_count == 0 &&
1248                                     !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1249                                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1250                                         GOTO(out, rc = -E2BIG);
1251                                 /**
1252                                  * load the user specified ost list, when this
1253                                  * component is instantiated later, it will be
1254                                  * used in lod_alloc_ost_list().
1255                                  */
1256                                 lod_comp->llc_ostlist.op_count = stripe_count;
1257                                 lod_comp->llc_ostlist.op_size =
1258                                         stripe_count * sizeof(__u32);
1259                                 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1260                                           lod_comp->llc_ostlist.op_size);
1261                                 if (!lod_comp->llc_ostlist.op_array)
1262                                         GOTO(out, rc = -ENOMEM);
1263
1264                                 for (j = 0; j < stripe_count; j++)
1265                                         lod_comp->llc_ostlist.op_array[j] =
1266                                                 le32_to_cpu(objs[j].l_ost_idx);
1267
1268                                 /**
1269                                  * this component OST objects starts from the
1270                                  * first ost_idx, lod_alloc_ost_list() will
1271                                  * check this.
1272                                  */
1273                                 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1274                         } else {
1275                                 /**
1276                                  * for uninstantiated component,
1277                                  * lmm_layout_gen stores default stripe offset.
1278                                  */
1279                                 lod_comp->llc_stripe_offset =
1280                                                         lmm->lmm_layout_gen;
1281                         }
1282                 }
1283
1284                 /* skip un-instantiated component object initialization */
1285                 if (!lod_comp_inited(lod_comp))
1286                         continue;
1287
1288                 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1289                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1290                         rc = lod_initialize_objects(env, lo, objs, i);
1291                         if (rc)
1292                                 GOTO(out, rc);
1293                 }
1294         }
1295
1296         rc = lod_fill_mirrors(lo);
1297         if (rc)
1298                 GOTO(out, rc);
1299
1300 out:
1301         if (rc)
1302                 lod_striping_free_nolock(env, lo);
1303         RETURN(rc);
1304 }
1305
1306 /**
1307  * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1308  * is already cached.
1309  *
1310  * \param[in] lo        LOD object
1311  *
1312  * \retval              True if the striping is cached, otherwise
1313  *                      return false.
1314  */
1315 static bool lod_striping_loaded(struct lod_object *lo)
1316 {
1317         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1318             lo->ldo_comp_cached)
1319                 return true;
1320
1321         if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1322                 if (lo->ldo_dir_stripe_loaded)
1323                         return true;
1324
1325                 /* Never load LMV stripe for slaves of striped dir */
1326                 if (lo->ldo_dir_slave_stripe)
1327                         return true;
1328         }
1329
1330         return false;
1331 }
1332
1333 /**
1334  * A generic function to initialize the stripe objects.
1335  *
1336  * A protected version of lod_striping_load_locked() - load the striping
1337  * information from storage, parse that and instantiate LU objects to
1338  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1339  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1340  * return an array of references to the newly instantiated objects.
1341  *
1342  * \param[in] env               execution environment for this thread
1343  * \param[in,out] lo            LOD object, where striping is stored and
1344  *                              which gets an array of references
1345  *
1346  * \retval                      0 if parsing and object creation succeed
1347  * \retval                      negative error number on failure
1348  **/
1349 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1350 {
1351         struct lod_thread_info *info = lod_env_info(env);
1352         struct dt_object *next = dt_object_child(&lo->ldo_obj);
1353         struct lu_buf *buf = &info->lti_buf;
1354         int rc = 0;
1355
1356         ENTRY;
1357
1358         if (!dt_object_exists(next))
1359                 RETURN(0);
1360
1361         if (lod_striping_loaded(lo))
1362                 RETURN(0);
1363
1364         mutex_lock(&lo->ldo_layout_mutex);
1365         if (lod_striping_loaded(lo))
1366                 GOTO(unlock, rc = 0);
1367
1368         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1369                 rc = lod_get_lov_ea(env, lo);
1370                 if (rc <= 0)
1371                         GOTO(unlock, rc);
1372
1373                 /*
1374                  * there is LOV EA (striping information) in this object
1375                  * let's parse it and create in-core objects for the stripes
1376                  */
1377                 buf->lb_buf = info->lti_ea_store;
1378                 buf->lb_len = info->lti_ea_store_size;
1379                 rc = lod_parse_striping(env, lo, buf);
1380                 if (rc == 0)
1381                         lo->ldo_comp_cached = 1;
1382         } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1383                 rc = lod_get_lmv_ea(env, lo);
1384                 if (rc > sizeof(struct lmv_foreign_md)) {
1385                         struct lmv_foreign_md *lfm = info->lti_ea_store;
1386
1387                         if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
1388                                 lo->ldo_foreign_lmv = info->lti_ea_store;
1389                                 lo->ldo_foreign_lmv_size =
1390                                         info->lti_ea_store_size;
1391                                 info->lti_ea_store = NULL;
1392                                 info->lti_ea_store_size = 0;
1393
1394                                 lo->ldo_dir_stripe_loaded = 1;
1395                                 lo->ldo_dir_is_foreign = 1;
1396                                 GOTO(unlock, rc = 0);
1397                         }
1398                 }
1399
1400                 if (rc < (int)sizeof(struct lmv_mds_md_v1)) {
1401                         /* Let's set stripe_loaded to avoid further
1402                          * stripe loading especially for non-stripe directory,
1403                          * which can hurt performance. (See LU-9840)
1404                          */
1405                         if (rc == 0)
1406                                 lo->ldo_dir_stripe_loaded = 1;
1407                         GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1408                 }
1409                 buf->lb_buf = info->lti_ea_store;
1410                 buf->lb_len = info->lti_ea_store_size;
1411                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1412                         rc = lod_load_lmv_shards(env, lo, buf, true);
1413                         if (buf->lb_buf != info->lti_ea_store) {
1414                                 OBD_FREE_LARGE(info->lti_ea_store,
1415                                                info->lti_ea_store_size);
1416                                 info->lti_ea_store = buf->lb_buf;
1417                                 info->lti_ea_store_size = buf->lb_len;
1418                         }
1419
1420                         if (rc < 0)
1421                                 GOTO(unlock, rc);
1422                 }
1423
1424                 /*
1425                  * there is LMV EA (striping information) in this object
1426                  * let's parse it and create in-core objects for the stripes
1427                  */
1428                 rc = lod_parse_dir_striping(env, lo, buf);
1429                 if (rc == 0)
1430                         lo->ldo_dir_stripe_loaded = 1;
1431         }
1432         EXIT;
1433 unlock:
1434         mutex_unlock(&lo->ldo_layout_mutex);
1435
1436         return rc;
1437 }
1438
1439 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1440                          const struct lu_buf *buf)
1441 {
1442         int rc;
1443
1444         ENTRY;
1445
1446         mutex_lock(&lo->ldo_layout_mutex);
1447         lod_striping_free_nolock(env, lo);
1448         rc = lod_parse_striping(env, lo, buf);
1449         mutex_unlock(&lo->ldo_layout_mutex);
1450
1451         RETURN(rc);
1452 }
1453
1454 /**
1455  * Verify lov_user_md_v1/v3 striping.
1456  *
1457  * Check the validity of all fields including the magic, stripe size,
1458  * stripe count, stripe offset and that the pool is present.  Also check
1459  * that each target index points to an existing target. The additional
1460  * \a is_from_disk turns additional checks. In some cases zero fields
1461  * are allowed (like pattern=0).
1462  *
1463  * \param[in] d                 LOD device
1464  * \param[in] buf               buffer with LOV EA to verify
1465  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1466  *                              1 - from disk, do not allow
1467  *
1468  * \retval                      0 if the striping is valid
1469  * \retval                      -EINVAL if striping is invalid
1470  */
1471 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1472                            bool is_from_disk)
1473 {
1474         struct lov_user_md_v1   *lum;
1475         struct lov_user_md_v3   *lum3;
1476         struct pool_desc        *pool = NULL;
1477         __u32                    magic;
1478         __u32                    stripe_size;
1479         __u16                    stripe_count;
1480         __u16                    stripe_offset;
1481         size_t                   lum_size;
1482         int                      rc = 0;
1483         ENTRY;
1484
1485         lum = buf->lb_buf;
1486
1487         if (buf->lb_len < sizeof(*lum)) {
1488                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1489                        buf->lb_len);
1490                 GOTO(out, rc = -EINVAL);
1491         }
1492
1493         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1494         if (magic != LOV_USER_MAGIC_V1 &&
1495             magic != LOV_USER_MAGIC_V3 &&
1496             magic != LOV_USER_MAGIC_SPECIFIC) {
1497                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1498                        le32_to_cpu(lum->lmm_magic));
1499                 GOTO(out, rc = -EINVAL);
1500         }
1501
1502         /* the user uses "0" for default stripe pattern normally. */
1503         if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1504                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1505
1506         if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1507                 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1508                        le32_to_cpu(lum->lmm_pattern));
1509                 GOTO(out, rc = -EINVAL);
1510         }
1511
1512         /* a released lum comes from creating orphan on hsm release,
1513          * doesn't make sense to verify it. */
1514         if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1515                 GOTO(out, rc = 0);
1516
1517         /* 64kB is the largest common page size we see (ia64), and matches the
1518          * check in lfs */
1519         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1520         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1521                 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1522                        stripe_size, LOV_MIN_STRIPE_SIZE);
1523                 GOTO(out, rc = -EINVAL);
1524         }
1525
1526         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1527         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1528             lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1529                 /* if offset is not within valid range [0, osts_size) */
1530                 if (stripe_offset >= d->lod_ost_descs.ltd_tgts_size) {
1531                         CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1532                                stripe_offset, d->lod_ost_descs.ltd_tgts_size);
1533                         GOTO(out, rc = -EINVAL);
1534                 }
1535
1536                 /* if lmm_stripe_offset is *not* in bitmap */
1537                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1538                         CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1539                                stripe_offset);
1540                         GOTO(out, rc = -EINVAL);
1541                 }
1542         }
1543
1544         if (magic == LOV_USER_MAGIC_V1)
1545                 lum_size = offsetof(struct lov_user_md_v1,
1546                                     lmm_objects[0]);
1547         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1548                 lum_size = offsetof(struct lov_user_md_v3,
1549                                     lmm_objects[0]);
1550         else
1551                 GOTO(out, rc = -EINVAL);
1552
1553         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1554         if (buf->lb_len < lum_size) {
1555                 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1556                        "magic %#x and stripe_count %u\n",
1557                        buf->lb_len, lum_size, magic, stripe_count);
1558                 GOTO(out, rc = -EINVAL);
1559         }
1560
1561         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1562                 goto out;
1563
1564         lum3 = buf->lb_buf;
1565         /* In the function below, .hs_keycmp resolves to
1566          * pool_hashkey_keycmp() */
1567         /* coverity[overrun-buffer-val] */
1568         pool = lod_find_pool(d, lum3->lmm_pool_name);
1569         if (pool == NULL)
1570                 goto out;
1571
1572         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1573                 rc = lod_check_index_in_pool(stripe_offset, pool);
1574                 if (rc < 0)
1575                         GOTO(out, rc = -EINVAL);
1576         }
1577
1578         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1579                 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1580                        stripe_count, pool_tgt_count(pool));
1581                 GOTO(out, rc = -EINVAL);
1582         }
1583
1584 out:
1585         if (pool != NULL)
1586                 lod_pool_putref(pool);
1587
1588         RETURN(rc);
1589 }
1590
1591 static inline
1592 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1593 {
1594         LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1595                  LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1596                  le32_to_cpu(comp->lcm_magic));
1597         LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1598                  "bad index %d, max = %d\n",
1599                  i, le16_to_cpu(comp->lcm_entry_count));
1600
1601         return &comp->lcm_entries[i];
1602 }
1603
1604 #define for_each_comp_entry_v1(comp, entry) \
1605         for (entry = comp_entry_v1(comp, 0); \
1606              entry <= comp_entry_v1(comp, \
1607                                    le16_to_cpu(comp->lcm_entry_count) - 1); \
1608              entry++)
1609
1610 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1,
1611                          struct lov_comp_md_entry_v1 *dom_ent)
1612 {
1613         struct lov_comp_md_entry_v1 *ent;
1614         __u16 entries;
1615         __u32 dom_off, dom_size, comp_size;
1616         void *blob_src, *blob_dst;
1617         unsigned int blob_size, blob_shift;
1618
1619         entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1620         /* if file has only DoM stripe return just error */
1621         if (entries == 0)
1622                 return -EFBIG;
1623
1624         comp_size = le32_to_cpu(comp_v1->lcm_size);
1625         dom_off = le32_to_cpu(dom_ent->lcme_offset);
1626         dom_size = le32_to_cpu(dom_ent->lcme_size);
1627
1628         /* shift entries array first */
1629         comp_v1->lcm_entry_count = cpu_to_le16(entries);
1630         memmove(dom_ent, dom_ent + 1,
1631                 entries * sizeof(struct lov_comp_md_entry_v1));
1632
1633         /* now move blob of layouts */
1634         blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1635         blob_src = (void *)comp_v1 + dom_off + dom_size;
1636         blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1637         blob_shift = sizeof(*dom_ent) + dom_size;
1638
1639         memmove(blob_dst, blob_src, blob_size);
1640
1641         for_each_comp_entry_v1(comp_v1, ent) {
1642                 __u32 off;
1643
1644                 off = le32_to_cpu(ent->lcme_offset);
1645                 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1646         }
1647
1648         comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1649
1650         /* notify a caller to re-check entry */
1651         return -ERESTART;
1652 }
1653
1654 void lod_dom_stripesize_recalc(struct lod_device *d)
1655 {
1656         __u64 threshold_mb = d->lod_dom_threshold_free_mb;
1657         __u32 max_size = d->lod_dom_stripesize_max_kb;
1658         __u32 def_size = d->lod_dom_stripesize_cur_kb;
1659
1660         /* use maximum allowed value if free space is above threshold */
1661         if (d->lod_lsfs_free_mb >= threshold_mb) {
1662                 def_size = max_size;
1663         } else if (!d->lod_lsfs_free_mb || max_size <= LOD_DOM_MIN_SIZE_KB) {
1664                 def_size = 0;
1665         } else {
1666                 /* recalc threshold like it would be with def_size as max */
1667                 threshold_mb = mult_frac(threshold_mb, def_size, max_size);
1668                 if (d->lod_lsfs_free_mb < threshold_mb)
1669                         def_size = rounddown(def_size / 2, LOD_DOM_MIN_SIZE_KB);
1670                 else if (d->lod_lsfs_free_mb > threshold_mb * 2)
1671                         def_size = max_t(unsigned int, def_size * 2,
1672                                          LOD_DOM_MIN_SIZE_KB);
1673         }
1674
1675         if (d->lod_dom_stripesize_cur_kb != def_size) {
1676                 CDEBUG(D_LAYOUT, "Change default DOM stripe size %d->%d\n",
1677                        d->lod_dom_stripesize_cur_kb, def_size);
1678                 d->lod_dom_stripesize_cur_kb = def_size;
1679         }
1680 }
1681
1682 static __u32 lod_dom_stripesize_limit(const struct lu_env *env,
1683                                       struct lod_device *d)
1684 {
1685         int rc;
1686
1687         /* set bfree as fraction of total space */
1688         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_SPOOF)) {
1689                 spin_lock(&d->lod_lsfs_lock);
1690                 d->lod_lsfs_free_mb = mult_frac(d->lod_lsfs_total_mb,
1691                                         min_t(int, cfs_fail_val, 100), 100);
1692                 GOTO(recalc, rc = 0);
1693         }
1694
1695         if (d->lod_lsfs_age < ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE) {
1696                 struct obd_statfs sfs;
1697
1698                 spin_lock(&d->lod_lsfs_lock);
1699                 if (d->lod_lsfs_age > ktime_get_seconds() - LOD_DOM_SFS_MAX_AGE)
1700                         GOTO(unlock, rc = 0);
1701
1702                 d->lod_lsfs_age = ktime_get_seconds();
1703                 spin_unlock(&d->lod_lsfs_lock);
1704                 rc = dt_statfs(env, d->lod_child, &sfs);
1705                 if (rc) {
1706                         CDEBUG(D_LAYOUT,
1707                                "%s: failed to get OSD statfs: rc = %d\n",
1708                                lod2obd(d)->obd_name, rc);
1709                         GOTO(out, rc);
1710                 }
1711                 /* udpate local OSD cached statfs data */
1712                 spin_lock(&d->lod_lsfs_lock);
1713                 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1714                 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1715 recalc:
1716                 lod_dom_stripesize_recalc(d);
1717 unlock:
1718                 spin_unlock(&d->lod_lsfs_lock);
1719         }
1720 out:
1721         return d->lod_dom_stripesize_cur_kb << 10;
1722 }
1723
1724 int lod_dom_stripesize_choose(const struct lu_env *env, struct lod_device *d,
1725                               struct lov_comp_md_v1 *comp_v1,
1726                               struct lov_comp_md_entry_v1 *dom_ent,
1727                               __u32 stripe_size)
1728 {
1729         struct lov_comp_md_entry_v1 *ent;
1730         struct lu_extent *dom_ext, *ext;
1731         struct lov_user_md_v1 *lum;
1732         __u32 max_stripe_size;
1733         __u16 mid, dom_mid;
1734         int rc = 0;
1735
1736         dom_ext = &dom_ent->lcme_extent;
1737         dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1738         max_stripe_size = lod_dom_stripesize_limit(env, d);
1739
1740         /* Check stripe size againts current per-MDT limit */
1741         if (stripe_size <= max_stripe_size)
1742                 return 0;
1743
1744         lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1745         CDEBUG(D_LAYOUT, "overwrite DoM component size %u with MDT limit %u\n",
1746                stripe_size, max_stripe_size);
1747         lum->lmm_stripe_size = cpu_to_le32(max_stripe_size);
1748
1749         for_each_comp_entry_v1(comp_v1, ent) {
1750                 if (ent == dom_ent)
1751                         continue;
1752
1753                 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1754                 if (mid != dom_mid)
1755                         continue;
1756
1757                 ext = &ent->lcme_extent;
1758                 if (ext->e_start != dom_ext->e_end)
1759                         continue;
1760
1761                 /* Found next component after the DoM one with the same
1762                  * mirror_id and adjust its start with DoM component end.
1763                  *
1764                  * NOTE: we are considering here that there can be only one
1765                  * DoM component in a file, all replicas are located on OSTs
1766                  * always and don't need adjustment since use own layouts.
1767                  */
1768                 ext->e_start = cpu_to_le64(max_stripe_size);
1769                 break;
1770         }
1771
1772         if (max_stripe_size == 0) {
1773                 /* DoM component size is zero due to server setting,
1774                  * remove it from the layout */
1775                 rc = lod_erase_dom_stripe(comp_v1, dom_ent);
1776         } else {
1777                 /* Update DoM extent end finally */
1778                 dom_ext->e_end = cpu_to_le64(max_stripe_size);
1779         }
1780
1781         return rc;
1782 }
1783
1784 /**
1785  * Verify LOV striping.
1786  *
1787  * \param[in] d                 LOD device
1788  * \param[in] buf               buffer with LOV EA to verify
1789  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1790  *                              1 - from disk, do not allow
1791  * \param[in] start             extent start for composite layout
1792  *
1793  * \retval                      0 if the striping is valid
1794  * \retval                      -EINVAL if striping is invalid
1795  */
1796 int lod_verify_striping(const struct lu_env *env, struct lod_device *d,
1797                         struct lod_object *lo, const struct lu_buf *buf,
1798                         bool is_from_disk)
1799 {
1800         struct lov_user_md_v1   *lum;
1801         struct lov_comp_md_v1   *comp_v1;
1802         struct lov_comp_md_entry_v1     *ent;
1803         struct lu_extent        *ext;
1804         struct lu_buf   tmp;
1805         __u64   prev_end = 0;
1806         __u32   stripe_size = 0;
1807         __u16   prev_mid = -1, mirror_id = -1;
1808         __u32   mirror_count;
1809         __u32   magic;
1810         int     rc = 0;
1811         ENTRY;
1812
1813         if (buf->lb_len < sizeof(lum->lmm_magic)) {
1814                 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1815                 RETURN(-EINVAL);
1816         }
1817
1818         lum = buf->lb_buf;
1819
1820         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1821         /* treat foreign LOV EA/object case first
1822          * XXX is it expected to try setting again a foreign?
1823          * XXX should we care about different current vs new layouts ?
1824          */
1825         if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1826                 struct lov_foreign_md *lfm = buf->lb_buf;
1827
1828                 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1829                         CDEBUG(D_LAYOUT,
1830                                "buf len %zu < min lov_foreign_md size (%zu)\n",
1831                                buf->lb_len, offsetof(typeof(*lfm),
1832                                lfm_value));
1833                         RETURN(-EINVAL);
1834                 }
1835
1836                 if (foreign_size_le(lfm) > buf->lb_len) {
1837                         CDEBUG(D_LAYOUT,
1838                                "buf len %zu < this lov_foreign_md size (%zu)\n",
1839                                buf->lb_len, foreign_size_le(lfm));
1840                         RETURN(-EINVAL);
1841                 }
1842                 /* Don't do anything with foreign layouts */
1843                 RETURN(0);
1844         }
1845
1846         /* normal LOV/layout cases */
1847
1848         if (buf->lb_len < sizeof(*lum)) {
1849                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1850                        buf->lb_len);
1851                 RETURN(-EINVAL);
1852         }
1853
1854         if (magic != LOV_USER_MAGIC_V1 &&
1855             magic != LOV_USER_MAGIC_V3 &&
1856             magic != LOV_USER_MAGIC_SPECIFIC &&
1857             magic != LOV_USER_MAGIC_COMP_V1) {
1858                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1859                        le32_to_cpu(lum->lmm_magic));
1860                 RETURN(-EINVAL);
1861         }
1862
1863         if (magic != LOV_USER_MAGIC_COMP_V1)
1864                 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1865
1866         /* magic == LOV_USER_MAGIC_COMP_V1 */
1867         comp_v1 = buf->lb_buf;
1868         if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1869                 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1870                        buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1871                 RETURN(-EINVAL);
1872         }
1873
1874 recheck:
1875         mirror_count = 0;
1876         if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1877                 CDEBUG(D_LAYOUT, "entry count is zero\n");
1878                 RETURN(-EINVAL);
1879         }
1880
1881         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1882             lo->ldo_comp_cnt > 0) {
1883                 /* could be called from lustre.lov.add */
1884                 __u32 cnt = lo->ldo_comp_cnt;
1885
1886                 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1887                 prev_end = ext->e_end;
1888
1889                 ++mirror_count;
1890         }
1891
1892         for_each_comp_entry_v1(comp_v1, ent) {
1893                 ext = &ent->lcme_extent;
1894
1895                 if (le64_to_cpu(ext->e_start) > le64_to_cpu(ext->e_end)) {
1896                         CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1897                                le64_to_cpu(ext->e_start),
1898                                le64_to_cpu(ext->e_end));
1899                         RETURN(-EINVAL);
1900                 }
1901
1902                 if (is_from_disk) {
1903                         /* lcme_id contains valid value */
1904                         if (le32_to_cpu(ent->lcme_id) == 0 ||
1905                             le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1906                                 CDEBUG(D_LAYOUT, "invalid id %u\n",
1907                                        le32_to_cpu(ent->lcme_id));
1908                                 RETURN(-EINVAL);
1909                         }
1910
1911                         if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1912                                 mirror_id = mirror_id_of(
1913                                                 le32_to_cpu(ent->lcme_id));
1914
1915                                 /* first component must start with 0 */
1916                                 if (mirror_id != prev_mid &&
1917                                     le64_to_cpu(ext->e_start) != 0) {
1918                                         CDEBUG(D_LAYOUT,
1919                                                "invalid start:%llu, expect:0\n",
1920                                                le64_to_cpu(ext->e_start));
1921                                         RETURN(-EINVAL);
1922                                 }
1923
1924                                 prev_mid = mirror_id;
1925                         }
1926                 }
1927
1928                 if (le64_to_cpu(ext->e_start) == 0) {
1929                         ++mirror_count;
1930                         prev_end = 0;
1931                 }
1932
1933                 /* the next must be adjacent with the previous one */
1934                 if (le64_to_cpu(ext->e_start) != prev_end) {
1935                         CDEBUG(D_LAYOUT,
1936                                "invalid start actual:%llu, expect:%llu\n",
1937                                le64_to_cpu(ext->e_start), prev_end);
1938                         RETURN(-EINVAL);
1939                 }
1940
1941                 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1942                 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1943
1944                 /* Check DoM entry is always the first one */
1945                 lum = tmp.lb_buf;
1946                 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1947                     LOV_PATTERN_MDT) {
1948                         /* DoM component must be the first in a mirror */
1949                         if (le64_to_cpu(ext->e_start) > 0) {
1950                                 CDEBUG(D_LAYOUT, "invalid DoM component "
1951                                        "with %llu extent start\n",
1952                                        le64_to_cpu(ext->e_start));
1953                                 RETURN(-EINVAL);
1954                         }
1955                         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1956                         /* There is just one stripe on MDT and it must
1957                          * cover whole component size. */
1958                         if (stripe_size != le64_to_cpu(ext->e_end)) {
1959                                 CDEBUG(D_LAYOUT, "invalid DoM layout "
1960                                        "stripe size %u != %llu "
1961                                        "(component size)\n",
1962                                        stripe_size, prev_end);
1963                                 RETURN(-EINVAL);
1964                         }
1965                         /* Check and adjust stripe size by per-MDT limit */
1966                         rc = lod_dom_stripesize_choose(env, d, comp_v1, ent,
1967                                                        stripe_size);
1968                         /* DoM entry was removed, re-check layout from start */
1969                         if (rc == -ERESTART)
1970                                 goto recheck;
1971                         else if (rc)
1972                                 RETURN(rc);
1973
1974                         /* Any stripe count is forbidden on DoM component */
1975                         if (lum->lmm_stripe_count) {
1976                                 CDEBUG(D_LAYOUT,
1977                                        "invalid DoM layout stripe count %u, must be 0\n",
1978                                        le16_to_cpu(lum->lmm_stripe_count));
1979                                 RETURN(-EINVAL);
1980                         }
1981
1982                         /* Any pool is forbidden on DoM component */
1983                         if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
1984                                 struct lov_user_md_v3 *v3 = (void *)lum;
1985
1986                                 if (v3->lmm_pool_name[0] != '\0') {
1987                                         CDEBUG(D_LAYOUT,
1988                                                "DoM component cannot have pool assigned\n");
1989                                         RETURN(-EINVAL);
1990                                 }
1991                         }
1992                 }
1993
1994                 prev_end = le64_to_cpu(ext->e_end);
1995
1996                 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
1997                 if (rc)
1998                         RETURN(rc);
1999
2000                 if (prev_end == LUSTRE_EOF)
2001                         continue;
2002
2003                 /* extent end must be aligned with the stripe_size */
2004                 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2005                 if (stripe_size && prev_end % stripe_size) {
2006                         CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2007                                "stripe_sz: %u, [%llu, %llu)\n",
2008                                stripe_size, ext->e_start, prev_end);
2009                         RETURN(-EINVAL);
2010                 }
2011         }
2012
2013         /* make sure that the mirror_count is telling the truth */
2014         if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2015                 RETURN(-EINVAL);
2016
2017         RETURN(0);
2018 }
2019
2020 /**
2021  * set the default stripe size, if unset.
2022  *
2023  * \param[in,out] val   number of bytes per OST stripe
2024  *
2025  * The minimum stripe size is 64KB to ensure that a single stripe is an
2026  * even multiple of a client PAGE_SIZE (IA64, PPC, etc).  Otherwise, it
2027  * is difficult to split dirty pages across OSCs during writes.
2028  */
2029 void lod_fix_desc_stripe_size(__u64 *val)
2030 {
2031         if (*val < LOV_MIN_STRIPE_SIZE) {
2032                 if (*val != 0)
2033                         LCONSOLE_INFO("Increasing default stripe size to "
2034                                       "minimum value %u\n",
2035                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
2036                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2037         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2038                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2039                 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2040                               "multiple of %u)\n",
2041                               *val, LOV_MIN_STRIPE_SIZE);
2042         }
2043 }
2044
2045 /**
2046  * set the filesystem default number of stripes, if unset.
2047  *
2048  * \param[in,out] val   number of stripes
2049  *
2050  * A value of "0" means "use the system-wide default stripe count", which
2051  * has either been inherited by now, or falls back to 1 stripe per file.
2052  * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2053  * and is a valid value, so is left unchanged here.
2054  */
2055 void lod_fix_desc_stripe_count(__u32 *val)
2056 {
2057         if (*val == 0)
2058                 *val = 1;
2059 }
2060
2061 /**
2062  * set the filesystem default layout pattern
2063  *
2064  * \param[in,out] val   LOV_PATTERN_* layout
2065  *
2066  * A value of "0" means "use the system-wide default layout type", which
2067  * has either been inherited by now, or falls back to plain RAID0 striping.
2068  */
2069 void lod_fix_desc_pattern(__u32 *val)
2070 {
2071         /* from lov_setstripe */
2072         if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
2073                 LCONSOLE_WARN("lod: Unknown stripe pattern: %#x\n", *val);
2074                 *val = 0;
2075         }
2076 }
2077
2078 void lod_fix_lmv_desc_pattern(__u32 *val)
2079 {
2080         if ((*val) && !lmv_is_known_hash_type(*val)) {
2081                 LCONSOLE_WARN("lod: Unknown md stripe pattern: %#x\n", *val);
2082                 *val = 0;
2083         }
2084 }
2085
2086 void lod_fix_desc_qos_maxage(__u32 *val)
2087 {
2088         /* fix qos_maxage */
2089         if (*val == 0)
2090                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2091 }
2092
2093 /**
2094  * Used to fix insane default striping.
2095  *
2096  * \param[in] desc      striping description
2097  */
2098 void lod_fix_desc(struct lov_desc *desc)
2099 {
2100         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2101         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2102         lod_fix_desc_pattern(&desc->ld_pattern);
2103         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2104 }
2105
2106 static void lod_fix_lmv_desc(struct lmv_desc *desc)
2107 {
2108         desc->ld_active_tgt_count = 0;
2109         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2110         lod_fix_lmv_desc_pattern(&desc->ld_pattern);
2111         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2112 }
2113
2114 /**
2115  * Initialize the structures used to store pools and default striping.
2116  *
2117  * \param[in] lod       LOD device
2118  * \param[in] lcfg      configuration structure storing default striping.
2119  *
2120  * \retval              0 if initialization succeeds
2121  * \retval              negative error number on failure
2122  */
2123 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2124 {
2125         struct obd_device          *obd;
2126         struct lov_desc            *desc;
2127         int                         rc;
2128         ENTRY;
2129
2130         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2131         LASSERT(obd != NULL);
2132         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2133
2134         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2135                 CERROR("LOD setup requires a descriptor\n");
2136                 RETURN(-EINVAL);
2137         }
2138
2139         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2140
2141         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2142                 CERROR("descriptor size wrong: %d > %d\n",
2143                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2144                 RETURN(-EINVAL);
2145         }
2146
2147         if (desc->ld_magic != LOV_DESC_MAGIC) {
2148                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2149                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2150                                obd->obd_name, desc);
2151                         lustre_swab_lov_desc(desc);
2152                 } else {
2153                         CERROR("%s: Bad lov desc magic: %#x\n",
2154                                obd->obd_name, desc->ld_magic);
2155                         RETURN(-EINVAL);
2156                 }
2157         }
2158
2159         lod_fix_desc(desc);
2160
2161         desc->ld_active_tgt_count = 0;
2162         lod->lod_ost_descs.ltd_lov_desc = *desc;
2163
2164         /* NB: config doesn't contain lmv_desc, alter it via sysfs. */
2165         lod_fix_lmv_desc(&lod->lod_mdt_descs.ltd_lmv_desc);
2166
2167         lod->lod_sp_me = LUSTRE_SP_CLI;
2168
2169         /* Set up OST pool environment */
2170         lod->lod_pool_count = 0;
2171         rc = lod_pool_hash_init(&lod->lod_pools_hash_body);
2172         if (rc)
2173                 RETURN(-ENOMEM);
2174
2175         INIT_LIST_HEAD(&lod->lod_pool_list);
2176         lod->lod_pool_count = 0;
2177         rc = tgt_pool_init(&lod->lod_mdt_descs.ltd_tgt_pool, 0);
2178         if (rc)
2179                 GOTO(out_hash, rc);
2180
2181         rc = tgt_pool_init(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool, 0);
2182         if (rc)
2183                 GOTO(out_mdt_pool, rc);
2184
2185         rc = tgt_pool_init(&lod->lod_ost_descs.ltd_tgt_pool, 0);
2186         if (rc)
2187                 GOTO(out_mdt_rr_pool, rc);
2188
2189         rc = tgt_pool_init(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool, 0);
2190         if (rc)
2191                 GOTO(out_ost_pool, rc);
2192
2193         RETURN(0);
2194
2195 out_ost_pool:
2196         tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2197 out_mdt_rr_pool:
2198         tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2199 out_mdt_pool:
2200         tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2201 out_hash:
2202         lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2203
2204         return rc;
2205 }
2206
2207 /**
2208  * Release the structures describing the pools.
2209  *
2210  * \param[in] lod       LOD device from which we release the structures
2211  *
2212  * \retval              0 always
2213  */
2214 int lod_pools_fini(struct lod_device *lod)
2215 {
2216         struct obd_device   *obd = lod2obd(lod);
2217         struct pool_desc    *pool, *tmp;
2218         ENTRY;
2219
2220         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2221                 /* free pool structs */
2222                 CDEBUG(D_INFO, "delete pool %p\n", pool);
2223                 /* In the function below, .hs_keycmp resolves to
2224                  * pool_hashkey_keycmp() */
2225                 /* coverity[overrun-buffer-val] */
2226                 lod_pool_del(obd, pool->pool_name);
2227         }
2228
2229         lod_pool_hash_destroy(&lod->lod_pools_hash_body);
2230         tgt_pool_free(&lod->lod_ost_descs.ltd_qos.lq_rr.lqr_pool);
2231         tgt_pool_free(&lod->lod_ost_descs.ltd_tgt_pool);
2232         tgt_pool_free(&lod->lod_mdt_descs.ltd_qos.lq_rr.lqr_pool);
2233         tgt_pool_free(&lod->lod_mdt_descs.ltd_tgt_pool);
2234
2235         RETURN(0);
2236 }