Whamcloud - gitweb
2197e415d81b541b4e76e212c86e2ce2b0aebfe1
[fs/lustre-release.git] / lustre / lod / lod_lov.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License version 2 for more details.  A copy is
14  * included in the COPYING file that accompanied this code.
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2017, Intel Corporation.
27  */
28 /*
29  * lustre/lod/lod_lov.c
30  *
31  * A set of helpers to maintain Logical Object Volume (LOV)
32  * Extended Attribute (EA) and known OST targets
33  *
34  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35  */
36
37 #define DEBUG_SUBSYSTEM S_MDS
38
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
43
44 #include "lod_internal.h"
45
46 /**
47  * Increase reference count on the target table.
48  *
49  * Increase reference count on the target table usage to prevent racing with
50  * addition/deletion. Any function that expects the table to remain
51  * stationary must take a ref.
52  *
53  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
54  */
55 void lod_getref(struct lod_tgt_descs *ltd)
56 {
57         down_read(&ltd->ltd_rw_sem);
58         mutex_lock(&ltd->ltd_mutex);
59         ltd->ltd_refcount++;
60         mutex_unlock(&ltd->ltd_mutex);
61 }
62
63 /**
64  * Decrease reference count on the target table.
65  *
66  * Companion of lod_getref() to release a reference on the target table.
67  * If this is the last reference and the OST entry was scheduled for deletion,
68  * the descriptor is removed from the table.
69  *
70  * \param[in] lod       LOD device from which we release a reference
71  * \param[in] ltd       target table (lod_ost_descs or lod_mdt_descs)
72  */
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
74 {
75         mutex_lock(&ltd->ltd_mutex);
76         ltd->ltd_refcount--;
77         if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78                 struct lod_tgt_desc *tgt_desc, *tmp;
79                 struct list_head kill;
80                 unsigned int idx;
81
82                 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
83                        ltd->ltd_death_row);
84
85                 INIT_LIST_HEAD(&kill);
86
87                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
88                         tgt_desc = LTD_TGT(ltd, idx);
89                         LASSERT(tgt_desc);
90
91                         if (!tgt_desc->ltd_reap)
92                                 continue;
93
94                         list_add(&tgt_desc->ltd_kill, &kill);
95                         LTD_TGT(ltd, idx) = NULL;
96                         /*FIXME: only support ost pool for now */
97                         if (ltd == &lod->lod_ost_descs) {
98                                 lod_ost_pool_remove(&lod->lod_pool_info, idx);
99                                 if (tgt_desc->ltd_active)
100                                         lod->lod_desc.ld_active_tgt_count--;
101                         }
102                         ltd->ltd_tgtnr--;
103                         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
104                         ltd->ltd_death_row--;
105                 }
106                 mutex_unlock(&ltd->ltd_mutex);
107                 up_read(&ltd->ltd_rw_sem);
108
109                 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
110                         int rc;
111                         list_del(&tgt_desc->ltd_kill);
112                         if (ltd == &lod->lod_ost_descs) {
113                                 /* remove from QoS structures */
114                                 rc = qos_del_tgt(lod, tgt_desc);
115                                 if (rc)
116                                         CERROR("%s: qos_del_tgt(%s) failed:"
117                                                "rc = %d\n",
118                                                lod2obd(lod)->obd_name,
119                                               obd_uuid2str(&tgt_desc->ltd_uuid),
120                                                rc);
121                         }
122                         rc = obd_disconnect(tgt_desc->ltd_exp);
123                         if (rc)
124                                 CERROR("%s: failed to disconnect %s: rc = %d\n",
125                                        lod2obd(lod)->obd_name,
126                                        obd_uuid2str(&tgt_desc->ltd_uuid), rc);
127                         OBD_FREE_PTR(tgt_desc);
128                 }
129         } else {
130                 mutex_unlock(&ltd->ltd_mutex);
131                 up_read(&ltd->ltd_rw_sem);
132         }
133 }
134
135 /**
136  * Expand size of target table.
137  *
138  * When the target table is full, we have to extend the table. To do so,
139  * we allocate new memory with some reserve, move data from the old table
140  * to the new one and release memory consumed by the old table.
141  * Notice we take ltd_rw_sem exclusively to ensure atomic switch.
142  *
143  * \param[in] ltd               target table
144  * \param[in] newsize           new size of the table
145  *
146  * \retval                      0 on success
147  * \retval                      -ENOMEM if reallocation failed
148  */
149 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
150 {
151         struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
152         int           rc = 0;
153         ENTRY;
154
155         /* grab write reference on the lod. Relocating the array requires
156          * exclusive access */
157
158         down_write(&ltd->ltd_rw_sem);
159         if (newsize <= ltd->ltd_tgts_size)
160                 /* someone else has already resize the array */
161                 GOTO(out, rc = 0);
162
163         /* allocate new bitmap */
164         new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
165         if (!new_bitmap)
166                 GOTO(out, rc = -ENOMEM);
167
168         if (ltd->ltd_tgts_size > 0) {
169                 /* the bitmap already exists, we need
170                  * to copy data from old one */
171                 cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
172                 old_bitmap = ltd->ltd_tgt_bitmap;
173         }
174
175         ltd->ltd_tgts_size  = newsize;
176         ltd->ltd_tgt_bitmap = new_bitmap;
177
178         if (old_bitmap)
179                 CFS_FREE_BITMAP(old_bitmap);
180
181         CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
182
183         EXIT;
184 out:
185         up_write(&ltd->ltd_rw_sem);
186         return rc;
187 }
188
189 /**
190  * Connect LOD to a new OSP and add it to the target table.
191  *
192  * Connect to the OSP device passed, initialize all the internal
193  * structures related to the device and add it to the target table.
194  *
195  * \param[in] env               execution environment for this thread
196  * \param[in] lod               LOD device to be connected to the new OSP
197  * \param[in] osp               name of OSP device name to be added
198  * \param[in] index             index of the new target
199  * \param[in] gen               target's generation number
200  * \param[in] tgt_index         OSP's group
201  * \param[in] type              type of device (mdc or osc)
202  * \param[in] active            state of OSP: 0 - inactive, 1 - active
203  *
204  * \retval                      0 if added successfully
205  * \retval                      negative error number on failure
206  */
207 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
208                    char *osp, unsigned index, unsigned gen, int tgt_index,
209                    char *type, int active)
210 {
211         struct obd_connect_data *data = NULL;
212         struct obd_export       *exp = NULL;
213         struct obd_device       *obd;
214         struct lu_device        *lu_dev;
215         struct dt_device        *dt_dev;
216         int                      rc;
217         struct lod_tgt_desc     *tgt_desc;
218         struct lod_tgt_descs    *ltd;
219         struct lustre_cfg       *lcfg;
220         struct obd_uuid         obd_uuid;
221         bool                    for_ost;
222         bool lock = false;
223         bool connected = false;
224         ENTRY;
225
226         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
227
228         if (gen <= 0) {
229                 CERROR("request to add OBD %s with invalid generation: %d\n",
230                        osp, gen);
231                 RETURN(-EINVAL);
232         }
233
234         obd_str2uuid(&obd_uuid, osp);
235
236         obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
237                                 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
238         if (obd == NULL) {
239                 CERROR("can't find %s device\n", osp);
240                 RETURN(-EINVAL);
241         }
242
243         LASSERT(obd->obd_lu_dev != NULL);
244         LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
245
246         lu_dev = obd->obd_lu_dev;
247         dt_dev = lu2dt_dev(lu_dev);
248
249         OBD_ALLOC_PTR(data);
250         if (data == NULL)
251                 GOTO(out_cleanup, rc = -ENOMEM);
252
253         data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
254         data->ocd_version = LUSTRE_VERSION_CODE;
255         data->ocd_index = index;
256
257         if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
258                 for_ost = true;
259                 data->ocd_connect_flags |= OBD_CONNECT_AT |
260                                            OBD_CONNECT_FULL20 |
261                                            OBD_CONNECT_INDEX |
262 #ifdef HAVE_LRU_RESIZE_SUPPORT
263                                            OBD_CONNECT_LRU_RESIZE |
264 #endif
265                                            OBD_CONNECT_MDS |
266                                            OBD_CONNECT_REQPORTAL |
267                                            OBD_CONNECT_SKIP_ORPHAN |
268                                            OBD_CONNECT_FID |
269                                            OBD_CONNECT_LVB_TYPE |
270                                            OBD_CONNECT_VERSION |
271                                            OBD_CONNECT_PINGLESS |
272                                            OBD_CONNECT_LFSCK |
273                                            OBD_CONNECT_BULK_MBITS;
274
275                 data->ocd_group = tgt_index;
276                 ltd = &lod->lod_ost_descs;
277         } else {
278                 struct obd_import *imp = obd->u.cli.cl_import;
279
280                 for_ost = false;
281                 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
282                 data->ocd_connect_flags |= OBD_CONNECT_ACL |
283                                            OBD_CONNECT_IBITS |
284                                            OBD_CONNECT_MDS_MDS |
285                                            OBD_CONNECT_FID |
286                                            OBD_CONNECT_AT |
287                                            OBD_CONNECT_FULL20 |
288                                            OBD_CONNECT_LFSCK |
289                                            OBD_CONNECT_BULK_MBITS;
290                 spin_lock(&imp->imp_lock);
291                 imp->imp_server_timeout = 1;
292                 spin_unlock(&imp->imp_lock);
293                 imp->imp_client->cli_request_portal = OUT_PORTAL;
294                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
295                       obd->obd_name);
296                 ltd = &lod->lod_mdt_descs;
297         }
298
299         rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
300         OBD_FREE_PTR(data);
301         if (rc) {
302                 CERROR("%s: cannot connect to next dev %s (%d)\n",
303                        obd->obd_name, osp, rc);
304                 GOTO(out_cleanup, rc);
305         }
306         connected = true;
307
308         /* Allocate ost descriptor and fill it */
309         OBD_ALLOC_PTR(tgt_desc);
310         if (!tgt_desc)
311                 GOTO(out_cleanup, rc = -ENOMEM);
312
313         tgt_desc->ltd_tgt    = dt_dev;
314         tgt_desc->ltd_exp    = exp;
315         tgt_desc->ltd_uuid   = obd->u.cli.cl_target_uuid;
316         tgt_desc->ltd_gen    = gen;
317         tgt_desc->ltd_index  = index;
318         tgt_desc->ltd_active = active;
319
320         lod_getref(ltd);
321         if (index >= ltd->ltd_tgts_size) {
322                 /* we have to increase the size of the lod_osts array */
323                 __u32  newsize;
324
325                 newsize = max(ltd->ltd_tgts_size, (__u32)2);
326                 while (newsize < index + 1)
327                         newsize = newsize << 1;
328
329                 /* lod_bitmap_resize() needs lod_rw_sem
330                  * which we hold with th reference */
331                 lod_putref(lod, ltd);
332
333                 rc = ltd_bitmap_resize(ltd, newsize);
334                 if (rc)
335                         GOTO(out_desc, rc);
336
337                 lod_getref(ltd);
338         }
339
340         mutex_lock(&ltd->ltd_mutex);
341         lock = true;
342         if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
343                 CERROR("%s: device %d is registered already\n", obd->obd_name,
344                        index);
345                 GOTO(out_mutex, rc = -EEXIST);
346         }
347
348         if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
349                 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
350                 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
351                         CERROR("can't allocate index to add %s\n",
352                                obd->obd_name);
353                         GOTO(out_mutex, rc = -ENOMEM);
354                 }
355         }
356
357         if (for_ost) {
358                 /* pool and qos are not supported for MDS stack yet */
359                 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
360                                       lod->lod_osts_size);
361                 if (rc) {
362                         CERROR("%s: can't set up pool, failed with %d\n",
363                                obd->obd_name, rc);
364                         GOTO(out_mutex, rc);
365                 }
366
367                 rc = qos_add_tgt(lod, tgt_desc);
368                 if (rc) {
369                         CERROR("%s: qos_add_tgt failed with %d\n",
370                                 obd->obd_name, rc);
371                         GOTO(out_pool, rc);
372                 }
373
374                 /* The new OST is now a full citizen */
375                 if (index >= lod->lod_desc.ld_tgt_count)
376                         lod->lod_desc.ld_tgt_count = index + 1;
377                 if (active)
378                         lod->lod_desc.ld_active_tgt_count++;
379         }
380
381         LTD_TGT(ltd, index) = tgt_desc;
382         cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
383         ltd->ltd_tgtnr++;
384         mutex_unlock(&ltd->ltd_mutex);
385         lod_putref(lod, ltd);
386         lock = false;
387         if (lod->lod_recovery_completed)
388                 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
389
390         if (!for_ost && lod->lod_initialized) {
391                 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
392                 if (rc != 0) {
393                         CERROR("%s: cannot start llog on %s:rc = %d\n",
394                                lod2obd(lod)->obd_name, osp, rc);
395                         GOTO(out_ltd, rc);
396                 }
397         }
398
399         rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
400         if (rc != 0) {
401                 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
402                        "index = %u, rc = %d\n", osp, type, index, rc);
403                 GOTO(out_fini_llog, rc);
404         }
405         RETURN(rc);
406 out_fini_llog:
407         lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
408                           tgt_desc->ltd_recovery_thread);
409 out_ltd:
410         lod_getref(ltd);
411         mutex_lock(&ltd->ltd_mutex);
412         lock = true;
413         if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
414                 struct ptlrpc_thread *thread;
415
416                 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
417                 OBD_FREE_PTR(thread);
418         }
419         ltd->ltd_tgtnr--;
420         cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
421         LTD_TGT(ltd, index) = NULL;
422 out_pool:
423         lod_ost_pool_remove(&lod->lod_pool_info, index);
424 out_mutex:
425         if (lock) {
426                 mutex_unlock(&ltd->ltd_mutex);
427                 lod_putref(lod, ltd);
428         }
429 out_desc:
430         OBD_FREE_PTR(tgt_desc);
431 out_cleanup:
432         /* XXX OSP needs us to send down LCFG_CLEANUP because it uses
433          * objects from the MDT stack. See LU-7184. */
434         lcfg = &lod_env_info(env)->lti_lustre_cfg;
435         memset(lcfg, 0, sizeof(*lcfg));
436         lcfg->lcfg_version = LUSTRE_CFG_VERSION;
437         lcfg->lcfg_command = LCFG_CLEANUP;
438         lu_dev->ld_ops->ldo_process_config(env, lu_dev, lcfg);
439
440         if (connected)
441                 obd_disconnect(exp);
442
443         return rc;
444 }
445
446 /**
447  * Schedule target removal from the target table.
448  *
449  * Mark the device as dead. The device is not removed here because it may
450  * still be in use. The device will be removed in lod_putref() when the
451  * last reference is released.
452  *
453  * \param[in] env               execution environment for this thread
454  * \param[in] lod               LOD device the target table belongs to
455  * \param[in] ltd               target table
456  * \param[in] idx               index of the target
457  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
458  */
459 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
460                              struct lod_tgt_descs *ltd, unsigned idx,
461                              bool for_ost)
462 {
463         LASSERT(LTD_TGT(ltd, idx));
464
465         lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
466                          idx, for_ost);
467
468         if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
469                 struct ptlrpc_thread *thread;
470
471                 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
472                 OBD_FREE_PTR(thread);
473         }
474
475         if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
476                 LTD_TGT(ltd, idx)->ltd_reap = 1;
477                 ltd->ltd_death_row++;
478         }
479 }
480
481 /**
482  * Schedule removal of all the targets from the given target table.
483  *
484  * See more details in the description for __lod_del_device()
485  *
486  * \param[in] env               execution environment for this thread
487  * \param[in] lod               LOD device the target table belongs to
488  * \param[in] ltd               target table
489  * \param[in] for_ost           type of the target: MDT or OST
490  *
491  * \retval                      0 always
492  */
493 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
494                  struct lod_tgt_descs *ltd, bool for_ost)
495 {
496         unsigned int idx;
497
498         if (ltd->ltd_tgts_size <= 0)
499                 return 0;
500         lod_getref(ltd);
501         mutex_lock(&ltd->ltd_mutex);
502         cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
503                 __lod_del_device(env, lod, ltd, idx, for_ost);
504         mutex_unlock(&ltd->ltd_mutex);
505         lod_putref(lod, ltd);
506         CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
507         for (idx = 0; idx < TGT_PTRS; idx++) {
508                 if (ltd->ltd_tgt_idx[idx])
509                         OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
510         }
511         ltd->ltd_tgts_size = 0;
512         return 0;
513 }
514
515 /**
516  * Remove device by name.
517  *
518  * Remove a device identified by \a osp from the target table. Given
519  * the device can be in use, the real deletion happens in lod_putref().
520  *
521  * \param[in] env               execution environment for this thread
522  * \param[in] lod               LOD device to be connected to the new OSP
523  * \param[in] ltd               target table
524  * \param[in] osp               name of OSP device to be removed
525  * \param[in] idx               index of the target
526  * \param[in] gen               generation number, not used currently
527  * \param[in] for_ost           type of the target: 0 - MDT, 1 - OST
528  *
529  * \retval                      0 if the device was scheduled for removal
530  * \retval                      -EINVAL if no device was found
531  */
532 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
533                    struct lod_tgt_descs *ltd, char *osp, unsigned idx,
534                    unsigned gen, bool for_ost)
535 {
536         struct obd_device *obd;
537         int                rc = 0;
538         struct obd_uuid    uuid;
539         ENTRY;
540
541         CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
542
543         obd_str2uuid(&uuid, osp);
544
545         obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
546                                    &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
547         if (obd == NULL) {
548                 CERROR("can't find %s device\n", osp);
549                 RETURN(-EINVAL);
550         }
551
552         if (gen <= 0) {
553                 CERROR("%s: request to remove OBD %s with invalid generation %d"
554                        "\n", obd->obd_name, osp, gen);
555                 RETURN(-EINVAL);
556         }
557
558         obd_str2uuid(&uuid,  osp);
559
560         lod_getref(ltd);
561         mutex_lock(&ltd->ltd_mutex);
562         /* check that the index is allocated in the bitmap */
563         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
564             !LTD_TGT(ltd, idx)) {
565                 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
566                 GOTO(out, rc = -EINVAL);
567         }
568
569         /* check that the UUID matches */
570         if (!obd_uuid_equals(&uuid, &LTD_TGT(ltd, idx)->ltd_uuid)) {
571                 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
572                        obd->obd_name, obd_uuid2str(&LTD_TGT(ltd,idx)->ltd_uuid),
573                        idx, osp);
574                 GOTO(out, rc = -EINVAL);
575         }
576
577         __lod_del_device(env, lod, ltd, idx, for_ost);
578         EXIT;
579 out:
580         mutex_unlock(&ltd->ltd_mutex);
581         lod_putref(lod, ltd);
582         return(rc);
583 }
584
585 /**
586  * Resize per-thread storage to hold specified size.
587  *
588  * A helper function to resize per-thread temporary storage. This storage
589  * is used to process LOV/LVM EAs and may be quite large. We do not want to
590  * allocate/release it every time, so instead we put it into the env and
591  * reallocate on demand. The memory is released when the correspondent thread
592  * is finished.
593  *
594  * \param[in] info              LOD-specific storage in the environment
595  * \param[in] size              new size to grow the buffer to
596
597  * \retval                      0 on success, -ENOMEM if reallocation failed
598  */
599 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
600 {
601         __u32 round = size_roundup_power2(size);
602
603         LASSERT(round <=
604                 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
605         if (info->lti_ea_store) {
606                 LASSERT(info->lti_ea_store_size);
607                 LASSERT(info->lti_ea_store_size < round);
608                 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
609                        info->lti_ea_store_size, round);
610                 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
611                 info->lti_ea_store = NULL;
612                 info->lti_ea_store_size = 0;
613         }
614
615         OBD_ALLOC_LARGE(info->lti_ea_store, round);
616         if (info->lti_ea_store == NULL)
617                 RETURN(-ENOMEM);
618         info->lti_ea_store_size = round;
619
620         RETURN(0);
621 }
622
623 static void lod_free_comp_buffer(struct lod_layout_component *entries,
624                                  __u16 count, __u32 bufsize)
625 {
626         struct lod_layout_component *entry;
627         int i;
628
629         for (i = 0; i < count; i++) {
630                 entry = &entries[i];
631                 if (entry->llc_pool != NULL)
632                         lod_set_pool(&entry->llc_pool, NULL);
633                 if (entry->llc_ostlist.op_array)
634                         OBD_FREE(entry->llc_ostlist.op_array,
635                                  entry->llc_ostlist.op_size);
636                 LASSERT(entry->llc_stripe == NULL);
637                 LASSERT(entry->llc_stripes_allocated == 0);
638         }
639
640         if (bufsize != 0)
641                 OBD_FREE_LARGE(entries, bufsize);
642 }
643
644 void lod_free_def_comp_entries(struct lod_default_striping *lds)
645 {
646         lod_free_comp_buffer(lds->lds_def_comp_entries,
647                              lds->lds_def_comp_size_cnt,
648                              size_roundup_power2(
649                                      sizeof(*lds->lds_def_comp_entries) *
650                                      lds->lds_def_comp_size_cnt));
651         lds->lds_def_comp_entries = NULL;
652         lds->lds_def_comp_cnt = 0;
653         lds->lds_def_striping_is_composite = 0;
654         lds->lds_def_comp_size_cnt = 0;
655 }
656
657 /**
658  * Resize per-thread storage to hold default striping component entries
659  *
660  * A helper function to resize per-thread temporary storage. This storage
661  * is used to hold default LOV/LVM EAs and may be quite large. We do not want
662  * to allocate/release it every time, so instead we put it into the env and
663  * reallocate it on demand. The memory is released when the correspondent
664  * thread is finished.
665  *
666  * \param[in,out] lds           default striping
667  * \param[in] count             new component count to grow the buffer to
668
669  * \retval                      0 on success, -ENOMEM if reallocation failed
670  */
671 int lod_def_striping_comp_resize(struct lod_default_striping *lds, __u16 count)
672 {
673         struct lod_layout_component *entries;
674         __u32 new = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
675                                         count);
676         __u32 old = size_roundup_power2(sizeof(*lds->lds_def_comp_entries) *
677                                         lds->lds_def_comp_size_cnt);
678
679         if (new <= old)
680                 return 0;
681
682         OBD_ALLOC_LARGE(entries, new);
683         if (entries == NULL)
684                 return -ENOMEM;
685
686         if (lds->lds_def_comp_entries != NULL) {
687                 CDEBUG(D_INFO, "default striping component size %d is not "
688                        "enough, need %d\n", old, new);
689                 lod_free_def_comp_entries(lds);
690         }
691
692         lds->lds_def_comp_entries = entries;
693         lds->lds_def_comp_size_cnt = count;
694
695         RETURN(0);
696 }
697
698 void lod_free_comp_entries(struct lod_object *lo)
699 {
700         if (lo->ldo_mirrors) {
701                 OBD_FREE(lo->ldo_mirrors,
702                          sizeof(*lo->ldo_mirrors) * lo->ldo_mirror_count);
703                 lo->ldo_mirrors = NULL;
704                 lo->ldo_mirror_count = 0;
705         }
706         lod_free_comp_buffer(lo->ldo_comp_entries,
707                              lo->ldo_comp_cnt,
708                              sizeof(*lo->ldo_comp_entries) * lo->ldo_comp_cnt);
709         lo->ldo_comp_entries = NULL;
710         lo->ldo_comp_cnt = 0;
711         lo->ldo_is_composite = 0;
712 }
713
714 int lod_alloc_comp_entries(struct lod_object *lo,
715                            int mirror_count, int comp_count)
716 {
717         LASSERT(comp_count != 0);
718         LASSERT(lo->ldo_comp_cnt == 0 && lo->ldo_comp_entries == NULL);
719
720         if (mirror_count > 0) {
721                 OBD_ALLOC(lo->ldo_mirrors,
722                           sizeof(*lo->ldo_mirrors) * mirror_count);
723                 if (!lo->ldo_mirrors)
724                         return -ENOMEM;
725
726                 lo->ldo_mirror_count = mirror_count;
727         }
728
729         OBD_ALLOC_LARGE(lo->ldo_comp_entries,
730                         sizeof(*lo->ldo_comp_entries) * comp_count);
731         if (lo->ldo_comp_entries == NULL) {
732                 OBD_FREE(lo->ldo_mirrors,
733                          sizeof(*lo->ldo_mirrors) * mirror_count);
734                 lo->ldo_mirror_count = 0;
735                 return -ENOMEM;
736         }
737
738         lo->ldo_comp_cnt = comp_count;
739         return 0;
740 }
741
742 int lod_fill_mirrors(struct lod_object *lo)
743 {
744         struct lod_layout_component *lod_comp;
745         int mirror_idx = -1;
746         __u16 mirror_id = 0xffff;
747         int i;
748         ENTRY;
749
750         LASSERT(equi(!lo->ldo_is_composite, lo->ldo_mirror_count == 0));
751
752         if (!lo->ldo_is_composite)
753                 RETURN(0);
754
755         lod_comp = &lo->ldo_comp_entries[0];
756         for (i = 0; i < lo->ldo_comp_cnt; i++, lod_comp++) {
757                 int stale = !!(lod_comp->llc_flags & LCME_FL_STALE);
758                 int preferred = !!(lod_comp->llc_flags & LCME_FL_PREF_WR);
759
760                 if (mirror_id_of(lod_comp->llc_id) == mirror_id) {
761                         lo->ldo_mirrors[mirror_idx].lme_stale |= stale;
762                         lo->ldo_mirrors[mirror_idx].lme_primary |= preferred;
763                         lo->ldo_mirrors[mirror_idx].lme_end = i;
764                         continue;
765                 }
766
767                 /* new mirror */
768                 ++mirror_idx;
769                 if (mirror_idx >= lo->ldo_mirror_count)
770                         RETURN(-EINVAL);
771
772                 mirror_id = mirror_id_of(lod_comp->llc_id);
773
774                 lo->ldo_mirrors[mirror_idx].lme_id = mirror_id;
775                 lo->ldo_mirrors[mirror_idx].lme_stale = stale;
776                 lo->ldo_mirrors[mirror_idx].lme_primary = preferred;
777                 lo->ldo_mirrors[mirror_idx].lme_start = i;
778                 lo->ldo_mirrors[mirror_idx].lme_end = i;
779         }
780         if (mirror_idx != lo->ldo_mirror_count - 1)
781                 RETURN(-EINVAL);
782
783         RETURN(0);
784 }
785
786 /**
787  * Generate on-disk lov_mds_md structure for each layout component based on
788  * the information in lod_object->ldo_comp_entries[i].
789  *
790  * \param[in] env               execution environment for this thread
791  * \param[in] lo                LOD object
792  * \param[in] comp_idx          index of ldo_comp_entries
793  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
794  * \param[in|out] lmm_size      buffer size/lmm size
795  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
796  *                              the stripe info is from the default stripe
797  *                              template, which is collected in lod_ah_init(),
798  *                              either from parent object or root object; for
799  *                              file case, it's from the @lo object
800  *
801  * \retval                      0 if on disk structure is created successfully
802  * \retval                      negative error number on failure
803  */
804 static int lod_gen_component_ea(const struct lu_env *env,
805                                 struct lod_object *lo, int comp_idx,
806                                 struct lov_mds_md *lmm, int *lmm_size,
807                                 bool is_dir)
808 {
809         struct lod_thread_info  *info = lod_env_info(env);
810         const struct lu_fid     *fid  = lu_object_fid(&lo->ldo_obj.do_lu);
811         struct lod_device       *lod;
812         struct lov_ost_data_v1  *objs;
813         struct lod_layout_component *lod_comp;
814         __u32   magic;
815         __u16 stripe_count;
816         int     i, rc = 0;
817         ENTRY;
818
819         LASSERT(lo);
820         if (is_dir)
821                 lod_comp =
822                         &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
823         else
824                 lod_comp = &lo->ldo_comp_entries[comp_idx];
825
826         magic = lod_comp->llc_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
827         if (lod_comp->llc_pattern == 0) /* default striping */
828                 lod_comp->llc_pattern = LOV_PATTERN_RAID0;
829
830         lmm->lmm_magic = cpu_to_le32(magic);
831         lmm->lmm_pattern = cpu_to_le32(lod_comp->llc_pattern);
832         fid_to_lmm_oi(fid, &lmm->lmm_oi);
833         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
834                 lmm->lmm_oi.oi.oi_id++;
835         lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
836
837         lmm->lmm_stripe_size = cpu_to_le32(lod_comp->llc_stripe_size);
838         lmm->lmm_stripe_count = cpu_to_le16(lod_comp->llc_stripe_count);
839         /**
840          * for dir and uninstantiated component, lmm_layout_gen stores
841          * default stripe offset.
842          */
843         lmm->lmm_layout_gen =
844                 (is_dir || !lod_comp_inited(lod_comp)) ?
845                         cpu_to_le16(lod_comp->llc_stripe_offset) :
846                         cpu_to_le16(lod_comp->llc_layout_gen);
847
848         if (magic == LOV_MAGIC_V1) {
849                 objs = &lmm->lmm_objects[0];
850         } else {
851                 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
852                 size_t cplen = strlcpy(v3->lmm_pool_name,
853                                        lod_comp->llc_pool,
854                                        sizeof(v3->lmm_pool_name));
855                 if (cplen >= sizeof(v3->lmm_pool_name))
856                         RETURN(-E2BIG);
857                 objs = &v3->lmm_objects[0];
858         }
859         stripe_count = lod_comp_entry_stripe_count(lo, lod_comp, is_dir);
860         if (stripe_count == 0 && !is_dir &&
861             !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
862             !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
863                 RETURN(-E2BIG);
864
865         if (!is_dir && lo->ldo_is_composite)
866                 lod_comp_shrink_stripe_count(lod_comp, &stripe_count);
867
868         if (is_dir || lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
869                 GOTO(done, rc = 0);
870
871         /* generate ost_idx of this component stripe */
872         lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
873         for (i = 0; i < stripe_count; i++) {
874                 struct dt_object *object;
875                 __u32 ost_idx = (__u32)-1UL;
876                 int type = LU_SEQ_RANGE_OST;
877
878                 if (lod_comp->llc_stripe && lod_comp->llc_stripe[i]) {
879                         object = lod_comp->llc_stripe[i];
880                         /* instantiated component */
881                         info->lti_fid = *lu_object_fid(&object->do_lu);
882
883                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF) &&
884                             comp_idx == 0) {
885                                 if (cfs_fail_val == 0)
886                                         cfs_fail_val = info->lti_fid.f_oid;
887                                 else if (i == 0)
888                                         info->lti_fid.f_oid = cfs_fail_val;
889                         }
890
891                         rc = fid_to_ostid(&info->lti_fid, &info->lti_ostid);
892                         LASSERT(rc == 0);
893
894                         ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
895                         objs[i].l_ost_gen = cpu_to_le32(0);
896                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
897                                 rc = -ENOENT;
898                         else
899                                 rc = lod_fld_lookup(env, lod, &info->lti_fid,
900                                                     &ost_idx, &type);
901                         if (rc < 0) {
902                                 CERROR("%s: Can not locate "DFID": rc = %d\n",
903                                        lod2obd(lod)->obd_name,
904                                        PFID(&info->lti_fid), rc);
905                                 RETURN(rc);
906                         }
907                 } else if (lod_comp->llc_ostlist.op_array &&
908                            lod_comp->llc_ostlist.op_count) {
909                         /* user specified ost list */
910                         ost_idx = lod_comp->llc_ostlist.op_array[i];
911                 }
912                 /*
913                  * with un-instantiated or with no specified ost list
914                  * component, its l_ost_idx does not matter.
915                  */
916                 objs[i].l_ost_idx = cpu_to_le32(ost_idx);
917         }
918 done:
919         if (lmm_size != NULL)
920                 *lmm_size = lov_mds_md_size(stripe_count, magic);
921         RETURN(rc);
922 }
923
924 /**
925  * Generate on-disk lov_mds_md structure based on the information in
926  * the lod_object->ldo_comp_entries.
927  *
928  * \param[in] env               execution environment for this thread
929  * \param[in] lo                LOD object
930  * \param[in] lmm               buffer to cotain the on-disk lov_mds_md
931  * \param[in|out] lmm_size      buffer size/lmm size
932  * \param[in] is_dir            generate lov ea for dir or file? For dir case,
933  *                              the stripe info is from the default stripe
934  *                              template, which is collected in lod_ah_init(),
935  *                              either from parent object or root object; for
936  *                              file case, it's from the @lo object
937  *
938  * \retval                      0 if on disk structure is created successfully
939  * \retval                      negative error number on failure
940  */
941 int lod_generate_lovea(const struct lu_env *env, struct lod_object *lo,
942                        struct lov_mds_md *lmm, int *lmm_size, bool is_dir)
943 {
944         struct lov_comp_md_entry_v1 *lcme;
945         struct lov_comp_md_v1 *lcm;
946         struct lod_layout_component *comp_entries;
947         __u16 comp_cnt, mirror_cnt;
948         bool is_composite, is_foreign = false;
949         int i, rc = 0, offset;
950         ENTRY;
951
952         if (is_dir) {
953                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
954                 mirror_cnt = lo->ldo_def_striping->lds_def_mirror_cnt;
955                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
956                 is_composite =
957                         lo->ldo_def_striping->lds_def_striping_is_composite;
958         } else {
959                 comp_cnt = lo->ldo_comp_cnt;
960                 mirror_cnt = lo->ldo_mirror_count;
961                 comp_entries = lo->ldo_comp_entries;
962                 is_composite = lo->ldo_is_composite;
963                 is_foreign = lo->ldo_is_foreign;
964         }
965
966         LASSERT(lmm_size != NULL);
967
968         if (is_foreign) {
969                 struct lov_foreign_md *lfm;
970
971                 lfm = (struct lov_foreign_md *)lmm;
972                 memcpy(lfm, lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
973                 /* need to store little-endian */
974                 if (cpu_to_le32(LOV_MAGIC_FOREIGN) != LOV_MAGIC_FOREIGN) {
975                         __swab32s(&lfm->lfm_magic);
976                         __swab32s(&lfm->lfm_length);
977                         __swab32s(&lfm->lfm_type);
978                         __swab32s(&lfm->lfm_flags);
979                 }
980                 *lmm_size = lo->ldo_foreign_lov_size;
981                 RETURN(0);
982         }
983
984         LASSERT(comp_cnt != 0 && comp_entries != NULL);
985
986         if (!is_composite) {
987                 rc = lod_gen_component_ea(env, lo, 0, lmm, lmm_size, is_dir);
988                 RETURN(rc);
989         }
990
991         lcm = (struct lov_comp_md_v1 *)lmm;
992         memset(lcm, 0, sizeof(*lcm));
993
994         lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
995         lcm->lcm_entry_count = cpu_to_le16(comp_cnt);
996         lcm->lcm_mirror_count = cpu_to_le16(mirror_cnt - 1);
997         lcm->lcm_flags = cpu_to_le16(lo->ldo_flr_state);
998
999         offset = sizeof(*lcm) + sizeof(*lcme) * comp_cnt;
1000         LASSERT(offset % sizeof(__u64) == 0);
1001
1002         for (i = 0; i < comp_cnt; i++) {
1003                 struct lod_layout_component *lod_comp;
1004                 struct lov_mds_md *sub_md;
1005                 int size;
1006
1007                 lod_comp = &comp_entries[i];
1008                 lcme = &lcm->lcm_entries[i];
1009
1010                 LASSERT(ergo(!is_dir, lod_comp->llc_id != LCME_ID_INVAL));
1011                 lcme->lcme_id = cpu_to_le32(lod_comp->llc_id);
1012
1013                 /* component could be un-inistantiated */
1014                 lcme->lcme_flags = cpu_to_le32(lod_comp->llc_flags);
1015                 if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1016                         lcme->lcme_timestamp =
1017                                 cpu_to_le64(lod_comp->llc_timestamp);
1018                 lcme->lcme_extent.e_start =
1019                         cpu_to_le64(lod_comp->llc_extent.e_start);
1020                 lcme->lcme_extent.e_end =
1021                         cpu_to_le64(lod_comp->llc_extent.e_end);
1022                 lcme->lcme_offset = cpu_to_le32(offset);
1023
1024                 sub_md = (struct lov_mds_md *)((char *)lcm + offset);
1025                 rc = lod_gen_component_ea(env, lo, i, sub_md, &size, is_dir);
1026                 if (rc)
1027                         GOTO(out, rc);
1028                 lcme->lcme_size = cpu_to_le32(size);
1029                 offset += size;
1030                 LASSERTF((offset <= *lmm_size) && (offset % sizeof(__u64) == 0),
1031                          "offset:%d lmm_size:%d\n", offset, *lmm_size);
1032         }
1033         lcm->lcm_size = cpu_to_le32(offset);
1034         lcm->lcm_layout_gen = cpu_to_le32(is_dir ? 0 : lo->ldo_layout_gen);
1035
1036         lustre_print_user_md(D_LAYOUT, (struct lov_user_md *)lmm,
1037                              "generate lum");
1038 out:
1039         if (rc == 0)
1040                 *lmm_size = offset;
1041         RETURN(rc);
1042 }
1043
1044 /**
1045  * Get LOV EA.
1046  *
1047  * Fill lti_ea_store buffer in the environment with a value for the given
1048  * EA. The buffer is reallocated if the value doesn't fit.
1049  *
1050  * \param[in,out] env           execution environment for this thread
1051  *                              .lti_ea_store buffer is filled with EA's value
1052  * \param[in] lo                LOD object
1053  * \param[in] name              name of the EA
1054  *
1055  * \retval                      > 0 if EA is fetched successfully
1056  * \retval                      0 if EA is empty
1057  * \retval                      negative error number on failure
1058  */
1059 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
1060                const char *name)
1061 {
1062         struct lod_thread_info  *info = lod_env_info(env);
1063         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
1064         int                     rc;
1065         ENTRY;
1066
1067         LASSERT(info);
1068
1069         if (unlikely(info->lti_ea_store == NULL)) {
1070                 /* just to enter in allocation block below */
1071                 rc = -ERANGE;
1072         } else {
1073 repeat:
1074                 info->lti_buf.lb_buf = info->lti_ea_store;
1075                 info->lti_buf.lb_len = info->lti_ea_store_size;
1076                 rc = dt_xattr_get(env, next, &info->lti_buf, name);
1077         }
1078
1079         /* if object is not striped or inaccessible */
1080         if (rc == -ENODATA || rc == -ENOENT)
1081                 RETURN(0);
1082
1083         if (rc == -ERANGE) {
1084                 /* EA doesn't fit, reallocate new buffer */
1085                 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
1086                 if (rc == -ENODATA || rc == -ENOENT)
1087                         RETURN(0);
1088                 else if (rc < 0)
1089                         RETURN(rc);
1090
1091                 LASSERT(rc > 0);
1092                 rc = lod_ea_store_resize(info, rc);
1093                 if (rc)
1094                         RETURN(rc);
1095                 goto repeat;
1096         }
1097
1098         RETURN(rc);
1099 }
1100
1101 /**
1102  * Verify the target index is present in the current configuration.
1103  *
1104  * \param[in] md                LOD device where the target table is stored
1105  * \param[in] idx               target's index
1106  *
1107  * \retval                      0 if the index is present
1108  * \retval                      -EINVAL if not
1109  */
1110 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
1111 {
1112         if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
1113                      !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
1114                 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
1115                        md->lod_ost_descs.ltd_tgts_size);
1116                 return -EINVAL;
1117         }
1118
1119         if (unlikely(OST_TGT(md, idx) == NULL)) {
1120                 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
1121                        lod2obd(md)->obd_name, idx);
1122                 return -EINVAL;
1123         }
1124
1125         if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
1126                 CERROR("%s: invalid lod device, for idx: %d\n",
1127                        lod2obd(md)->obd_name , idx);
1128                 return -EINVAL;
1129         }
1130
1131         return 0;
1132 }
1133
1134 /**
1135  * Instantiate objects for stripes.
1136  *
1137  * Allocate and initialize LU-objects representing the stripes. The number
1138  * of the stripes (ldo_stripe_count) must be initialized already. The caller
1139  * must ensure nobody else is calling the function on the object at the same
1140  * time. FLDB service must be running to be able to map a FID to the targets
1141  * and find appropriate device representing that target.
1142  *
1143  * \param[in] env               execution environment for this thread
1144  * \param[in,out] lo            LOD object
1145  * \param[in] objs              an array of IDs to creates the objects from
1146  * \param[in] comp_idx          index of ldo_comp_entries
1147  *
1148  * \retval                      0 if the objects are instantiated successfully
1149  * \retval                      negative error number on failure
1150  */
1151 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
1152                            struct lov_ost_data_v1 *objs, int comp_idx)
1153 {
1154         struct lod_layout_component *lod_comp;
1155         struct lod_thread_info *info = lod_env_info(env);
1156         struct lod_device *md;
1157         struct lu_object *o, *n;
1158         struct lu_device *nd;
1159         struct dt_object **stripe = NULL;
1160         __u32 *ost_indices = NULL;
1161         int stripe_len;
1162         int i, rc = 0;
1163         __u32 idx;
1164         ENTRY;
1165
1166         LASSERT(lo != NULL);
1167         md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1168
1169         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1170         lod_comp = &lo->ldo_comp_entries[comp_idx];
1171
1172         LASSERT(lod_comp->llc_stripe == NULL);
1173         LASSERT(lod_comp->llc_stripe_count > 0);
1174         LASSERT(lod_comp->llc_stripe_size > 0);
1175
1176         stripe_len = lod_comp->llc_stripe_count;
1177         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
1178         if (stripe == NULL)
1179                 RETURN(-ENOMEM);
1180         OBD_ALLOC(ost_indices, sizeof(*ost_indices) * stripe_len);
1181         if (!ost_indices)
1182                 GOTO(out, rc = -ENOMEM);
1183
1184         for (i = 0; i < lod_comp->llc_stripe_count; i++) {
1185                 if (unlikely(lovea_slot_is_dummy(&objs[i])))
1186                         continue;
1187
1188                 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
1189                 idx = le32_to_cpu(objs[i].l_ost_idx);
1190                 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
1191                 if (rc != 0)
1192                         GOTO(out, rc);
1193                 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
1194                          PFID(&info->lti_fid));
1195                 lod_getref(&md->lod_ost_descs);
1196
1197                 rc = validate_lod_and_idx(md, idx);
1198                 if (unlikely(rc != 0)) {
1199                         lod_putref(md, &md->lod_ost_descs);
1200                         GOTO(out, rc);
1201                 }
1202
1203                 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
1204                 lod_putref(md, &md->lod_ost_descs);
1205
1206                 /* In the function below, .hs_keycmp resolves to
1207                  * u_obj_hop_keycmp() */
1208                 /* coverity[overrun-buffer-val] */
1209                 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
1210                 if (IS_ERR(o))
1211                         GOTO(out, rc = PTR_ERR(o));
1212
1213                 n = lu_object_locate(o->lo_header, nd->ld_type);
1214                 LASSERT(n);
1215
1216                 stripe[i] = container_of(n, struct dt_object, do_lu);
1217                 ost_indices[i] = idx;
1218         }
1219
1220 out:
1221         if (rc != 0) {
1222                 for (i = 0; i < stripe_len; i++)
1223                         if (stripe[i] != NULL)
1224                                 dt_object_put(env, stripe[i]);
1225
1226                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
1227                 lod_comp->llc_stripe_count = 0;
1228                 if (ost_indices)
1229                         OBD_FREE(ost_indices,
1230                                  sizeof(*ost_indices) * stripe_len);
1231         } else {
1232                 lod_comp->llc_stripe = stripe;
1233                 lod_comp->llc_ost_indices = ost_indices;
1234                 lod_comp->llc_stripes_allocated = stripe_len;
1235         }
1236
1237         RETURN(rc);
1238 }
1239
1240 /**
1241  * Instantiate objects for striping.
1242  *
1243  * Parse striping information in \a buf and instantiate the objects
1244  * representing the stripes.
1245  *
1246  * \param[in] env               execution environment for this thread
1247  * \param[in] lo                LOD object
1248  * \param[in] buf               buffer storing LOV EA to parse
1249  *
1250  * \retval                      0 if parsing and objects creation succeed
1251  * \retval                      negative error number on failure
1252  */
1253 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
1254                        const struct lu_buf *buf)
1255 {
1256         struct lov_mds_md_v1    *lmm;
1257         struct lov_comp_md_v1   *comp_v1 = NULL;
1258         struct lov_foreign_md   *foreign = NULL;
1259         struct lov_ost_data_v1  *objs;
1260         __u32   magic, pattern;
1261         int     i, j, rc = 0;
1262         __u16   comp_cnt;
1263         __u16   mirror_cnt = 0;
1264         ENTRY;
1265
1266         LASSERT(buf);
1267         LASSERT(buf->lb_buf);
1268         LASSERT(buf->lb_len);
1269         LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1270
1271         lmm = (struct lov_mds_md_v1 *)buf->lb_buf;
1272         magic = le32_to_cpu(lmm->lmm_magic);
1273
1274         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3 &&
1275             magic != LOV_MAGIC_COMP_V1 && magic != LOV_MAGIC_FOREIGN)
1276                 GOTO(out, rc = -EINVAL);
1277
1278         if (lo->ldo_is_foreign)
1279                 lod_free_foreign_lov(lo);
1280         else
1281                 lod_free_comp_entries(lo);
1282
1283         if (magic == LOV_MAGIC_COMP_V1) {
1284                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1285                 comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
1286                 if (comp_cnt == 0)
1287                         GOTO(out, rc = -EINVAL);
1288                 lo->ldo_layout_gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1289                 lo->ldo_is_composite = 1;
1290                 lo->ldo_flr_state = le16_to_cpu(comp_v1->lcm_flags) &
1291                                         LCM_FL_FLR_MASK;
1292                 mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
1293         } else if (magic == LOV_MAGIC_FOREIGN) {
1294                 size_t length;
1295
1296                 foreign = (struct lov_foreign_md *)buf->lb_buf;
1297                 length = offsetof(typeof(*foreign), lfm_value);
1298                 if (buf->lb_len < length ||
1299                     buf->lb_len < (length + le32_to_cpu(foreign->lfm_length))) {
1300                         CDEBUG(D_LAYOUT,
1301                                "buf len %zu too small for lov_foreign_md\n",
1302                                buf->lb_len);
1303                         GOTO(out, rc = -EINVAL);
1304                 }
1305
1306                 /* just cache foreign LOV EA raw */
1307                 rc = lod_alloc_foreign_lov(lo, length);
1308                 if (rc)
1309                         GOTO(out, rc);
1310                 memcpy(lo->ldo_foreign_lov, buf->lb_buf, length);
1311                 GOTO(out, rc);
1312         } else {
1313                 comp_cnt = 1;
1314                 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1315                 lo->ldo_is_composite = 0;
1316         }
1317
1318         rc = lod_alloc_comp_entries(lo, mirror_cnt, comp_cnt);
1319         if (rc)
1320                 GOTO(out, rc);
1321
1322         for (i = 0; i < comp_cnt; i++) {
1323                 struct lod_layout_component     *lod_comp;
1324                 struct lu_extent        *ext;
1325                 __u32   offs;
1326
1327                 lod_comp = &lo->ldo_comp_entries[i];
1328                 if (lo->ldo_is_composite) {
1329                         offs = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1330                         lmm = (struct lov_mds_md_v1 *)((char *)comp_v1 + offs);
1331                         magic = le32_to_cpu(lmm->lmm_magic);
1332
1333                         ext = &comp_v1->lcm_entries[i].lcme_extent;
1334                         lod_comp->llc_extent.e_start =
1335                                 le64_to_cpu(ext->e_start);
1336                         lod_comp->llc_extent.e_end = le64_to_cpu(ext->e_end);
1337                         lod_comp->llc_flags =
1338                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags);
1339                         if (lod_comp->llc_flags & LCME_FL_NOSYNC)
1340                                 lod_comp->llc_timestamp = le64_to_cpu(
1341                                         comp_v1->lcm_entries[i].lcme_timestamp);
1342                         lod_comp->llc_id =
1343                                 le32_to_cpu(comp_v1->lcm_entries[i].lcme_id);
1344                         if (lod_comp->llc_id == LCME_ID_INVAL)
1345                                 GOTO(out, rc = -EINVAL);
1346                 } else {
1347                         lod_comp_set_init(lod_comp);
1348                 }
1349
1350                 pattern = le32_to_cpu(lmm->lmm_pattern);
1351                 if (lov_pattern(pattern) != LOV_PATTERN_RAID0 &&
1352                     lov_pattern(pattern) != LOV_PATTERN_MDT)
1353                         GOTO(out, rc = -EINVAL);
1354
1355                 lod_comp->llc_pattern = pattern;
1356                 lod_comp->llc_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
1357                 lod_comp->llc_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1358                 lod_comp->llc_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
1359
1360                 if (magic == LOV_MAGIC_V3) {
1361                         struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *)lmm;
1362                         lod_set_pool(&lod_comp->llc_pool, v3->lmm_pool_name);
1363                         objs = &v3->lmm_objects[0];
1364                 } else {
1365                         lod_set_pool(&lod_comp->llc_pool, NULL);
1366                         objs = &lmm->lmm_objects[0];
1367                 }
1368
1369                 /**
1370                  * If uninstantiated template component has valid l_ost_idx,
1371                  * then user has specified ost list for this component.
1372                  */
1373                 if (!lod_comp_inited(lod_comp)) {
1374                         __u16 stripe_count;
1375
1376                         if (objs[0].l_ost_idx != (__u32)-1UL) {
1377                                 stripe_count = lod_comp_entry_stripe_count(
1378                                                         lo, lod_comp, false);
1379                                 if (stripe_count == 0 &&
1380                                     !(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1381                                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT))
1382                                         GOTO(out, rc = -E2BIG);
1383                                 /**
1384                                  * load the user specified ost list, when this
1385                                  * component is instantiated later, it will be
1386                                  * used in lod_alloc_ost_list().
1387                                  */
1388                                 lod_comp->llc_ostlist.op_count = stripe_count;
1389                                 lod_comp->llc_ostlist.op_size =
1390                                         stripe_count * sizeof(__u32);
1391                                 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
1392                                           lod_comp->llc_ostlist.op_size);
1393                                 if (!lod_comp->llc_ostlist.op_array)
1394                                         GOTO(out, rc = -ENOMEM);
1395
1396                                 for (j = 0; j < stripe_count; j++)
1397                                         lod_comp->llc_ostlist.op_array[j] =
1398                                                 le32_to_cpu(objs[j].l_ost_idx);
1399
1400                                 /**
1401                                  * this component OST objects starts from the
1402                                  * first ost_idx, lod_alloc_ost_list() will
1403                                  * check this.
1404                                  */
1405                                 lod_comp->llc_stripe_offset = objs[0].l_ost_idx;
1406                         } else {
1407                                 /**
1408                                  * for uninstantiated component,
1409                                  * lmm_layout_gen stores default stripe offset.
1410                                  */
1411                                 lod_comp->llc_stripe_offset =
1412                                                         lmm->lmm_layout_gen;
1413                         }
1414                 }
1415
1416                 /* skip un-instantiated component object initialization */
1417                 if (!lod_comp_inited(lod_comp))
1418                         continue;
1419
1420                 if (!(lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED) &&
1421                     !(lod_comp->llc_pattern & LOV_PATTERN_MDT)) {
1422                         rc = lod_initialize_objects(env, lo, objs, i);
1423                         if (rc)
1424                                 GOTO(out, rc);
1425                 }
1426         }
1427
1428         rc = lod_fill_mirrors(lo);
1429         if (rc)
1430                 GOTO(out, rc);
1431
1432 out:
1433         if (rc)
1434                 lod_striping_free_nolock(env, lo);
1435         RETURN(rc);
1436 }
1437
1438 /**
1439  * Check whether the striping (LOVEA for regular file, LMVEA for directory)
1440  * is already cached.
1441  *
1442  * \param[in] lo        LOD object
1443  *
1444  * \retval              True if the striping is cached, otherwise
1445  *                      return false.
1446  */
1447 static bool lod_striping_loaded(struct lod_object *lo)
1448 {
1449         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1450             lo->ldo_comp_cached)
1451                 return true;
1452
1453         if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1454                 if (lo->ldo_dir_stripe_loaded)
1455                         return true;
1456
1457                 /* Never load LMV stripe for slaves of striped dir */
1458                 if (lo->ldo_dir_slave_stripe)
1459                         return true;
1460         }
1461
1462         return false;
1463 }
1464
1465 /**
1466  * A generic function to initialize the stripe objects.
1467  *
1468  * A protected version of lod_striping_load_locked() - load the striping
1469  * information from storage, parse that and instantiate LU objects to
1470  * represent the stripes.  The LOD object \a lo supplies a pointer to the
1471  * next sub-object in the LU stack so we can lock it. Also use \a lo to
1472  * return an array of references to the newly instantiated objects.
1473  *
1474  * \param[in] env               execution environment for this thread
1475  * \param[in,out] lo            LOD object, where striping is stored and
1476  *                              which gets an array of references
1477  *
1478  * \retval                      0 if parsing and object creation succeed
1479  * \retval                      negative error number on failure
1480  **/
1481 int lod_striping_load(const struct lu_env *env, struct lod_object *lo)
1482 {
1483         struct lod_thread_info *info = lod_env_info(env);
1484         struct dt_object *next = dt_object_child(&lo->ldo_obj);
1485         struct lu_buf *buf = &info->lti_buf;
1486         int rc = 0;
1487
1488         ENTRY;
1489
1490         if (!dt_object_exists(next))
1491                 RETURN(0);
1492
1493         if (lod_striping_loaded(lo))
1494                 RETURN(0);
1495
1496         mutex_lock(&lo->ldo_layout_mutex);
1497         if (lod_striping_loaded(lo))
1498                 GOTO(unlock, rc = 0);
1499
1500         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr)) {
1501                 rc = lod_get_lov_ea(env, lo);
1502                 if (rc <= 0)
1503                         GOTO(unlock, rc);
1504
1505                 /*
1506                  * there is LOV EA (striping information) in this object
1507                  * let's parse it and create in-core objects for the stripes
1508                  */
1509                 buf->lb_buf = info->lti_ea_store;
1510                 buf->lb_len = info->lti_ea_store_size;
1511                 rc = lod_parse_striping(env, lo, buf);
1512                 if (rc == 0)
1513                         lo->ldo_comp_cached = 1;
1514         } else if (S_ISDIR(lod2lu_obj(lo)->lo_header->loh_attr)) {
1515                 rc = lod_get_lmv_ea(env, lo);
1516                 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) {
1517                         /* Let's set stripe_loaded to avoid further
1518                          * stripe loading especially for non-stripe directory,
1519                          * which can hurt performance. (See LU-9840)
1520                          */
1521                         if (rc == 0)
1522                                 lo->ldo_dir_stripe_loaded = 1;
1523                         GOTO(unlock, rc = rc > 0 ? -EINVAL : rc);
1524                 }
1525                 buf->lb_buf = info->lti_ea_store;
1526                 buf->lb_len = info->lti_ea_store_size;
1527                 if (rc == sizeof(struct lmv_mds_md_v1)) {
1528                         rc = lod_load_lmv_shards(env, lo, buf, true);
1529                         if (buf->lb_buf != info->lti_ea_store) {
1530                                 OBD_FREE_LARGE(info->lti_ea_store,
1531                                                info->lti_ea_store_size);
1532                                 info->lti_ea_store = buf->lb_buf;
1533                                 info->lti_ea_store_size = buf->lb_len;
1534                         }
1535
1536                         if (rc < 0)
1537                                 GOTO(unlock, rc);
1538                 }
1539
1540                 /*
1541                  * there is LMV EA (striping information) in this object
1542                  * let's parse it and create in-core objects for the stripes
1543                  */
1544                 rc = lod_parse_dir_striping(env, lo, buf);
1545                 if (rc == 0)
1546                         lo->ldo_dir_stripe_loaded = 1;
1547         }
1548         EXIT;
1549 unlock:
1550         mutex_unlock(&lo->ldo_layout_mutex);
1551
1552         return rc;
1553 }
1554
1555 int lod_striping_reload(const struct lu_env *env, struct lod_object *lo,
1556                          const struct lu_buf *buf)
1557 {
1558         int rc;
1559
1560         ENTRY;
1561
1562         mutex_lock(&lo->ldo_layout_mutex);
1563         lod_striping_free_nolock(env, lo);
1564         rc = lod_parse_striping(env, lo, buf);
1565         mutex_unlock(&lo->ldo_layout_mutex);
1566
1567         RETURN(rc);
1568 }
1569
1570 /**
1571  * Verify lov_user_md_v1/v3 striping.
1572  *
1573  * Check the validity of all fields including the magic, stripe size,
1574  * stripe count, stripe offset and that the pool is present.  Also check
1575  * that each target index points to an existing target. The additional
1576  * \a is_from_disk turns additional checks. In some cases zero fields
1577  * are allowed (like pattern=0).
1578  *
1579  * \param[in] d                 LOD device
1580  * \param[in] buf               buffer with LOV EA to verify
1581  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1582  *                              1 - from disk, do not allow
1583  *
1584  * \retval                      0 if the striping is valid
1585  * \retval                      -EINVAL if striping is invalid
1586  */
1587 static int lod_verify_v1v3(struct lod_device *d, const struct lu_buf *buf,
1588                            bool is_from_disk)
1589 {
1590         struct lov_user_md_v1   *lum;
1591         struct lov_user_md_v3   *lum3;
1592         struct pool_desc        *pool = NULL;
1593         __u32                    magic;
1594         __u32                    stripe_size;
1595         __u16                    stripe_count;
1596         __u16                    stripe_offset;
1597         size_t                   lum_size;
1598         int                      rc = 0;
1599         ENTRY;
1600
1601         lum = buf->lb_buf;
1602
1603         if (buf->lb_len < sizeof(*lum)) {
1604                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1605                        buf->lb_len);
1606                 GOTO(out, rc = -EINVAL);
1607         }
1608
1609         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1610         if (magic != LOV_USER_MAGIC_V1 &&
1611             magic != LOV_USER_MAGIC_V3 &&
1612             magic != LOV_USER_MAGIC_SPECIFIC) {
1613                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1614                        le32_to_cpu(lum->lmm_magic));
1615                 GOTO(out, rc = -EINVAL);
1616         }
1617
1618         /* the user uses "0" for default stripe pattern normally. */
1619         if (!is_from_disk && lum->lmm_pattern == LOV_PATTERN_NONE)
1620                 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1621
1622         if (!lov_pattern_supported(le32_to_cpu(lum->lmm_pattern))) {
1623                 CDEBUG(D_LAYOUT, "bad userland stripe pattern: %#x\n",
1624                        le32_to_cpu(lum->lmm_pattern));
1625                 GOTO(out, rc = -EINVAL);
1626         }
1627
1628         /* a released lum comes from creating orphan on hsm release,
1629          * doesn't make sense to verify it. */
1630         if (le32_to_cpu(lum->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1631                 GOTO(out, rc = 0);
1632
1633         /* 64kB is the largest common page size we see (ia64), and matches the
1634          * check in lfs */
1635         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1636         if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1637                 CDEBUG(D_LAYOUT, "stripe size %u not a multiple of %u\n",
1638                        stripe_size, LOV_MIN_STRIPE_SIZE);
1639                 GOTO(out, rc = -EINVAL);
1640         }
1641
1642         stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1643         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT &&
1644             lov_pattern(le32_to_cpu(lum->lmm_pattern)) != LOV_PATTERN_MDT) {
1645                 /* if offset is not within valid range [0, osts_size) */
1646                 if (stripe_offset >= d->lod_osts_size) {
1647                         CDEBUG(D_LAYOUT, "stripe offset %u >= bitmap size %u\n",
1648                                stripe_offset, d->lod_osts_size);
1649                         GOTO(out, rc = -EINVAL);
1650                 }
1651
1652                 /* if lmm_stripe_offset is *not* in bitmap */
1653                 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1654                         CDEBUG(D_LAYOUT, "stripe offset %u not in bitmap\n",
1655                                stripe_offset);
1656                         GOTO(out, rc = -EINVAL);
1657                 }
1658         }
1659
1660         if (magic == LOV_USER_MAGIC_V1)
1661                 lum_size = offsetof(struct lov_user_md_v1,
1662                                     lmm_objects[0]);
1663         else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC)
1664                 lum_size = offsetof(struct lov_user_md_v3,
1665                                     lmm_objects[0]);
1666         else
1667                 GOTO(out, rc = -EINVAL);
1668
1669         stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1670         if (buf->lb_len < lum_size) {
1671                 CDEBUG(D_LAYOUT, "invalid buf len %zu/%zu for lov_user_md with "
1672                        "magic %#x and stripe_count %u\n",
1673                        buf->lb_len, lum_size, magic, stripe_count);
1674                 GOTO(out, rc = -EINVAL);
1675         }
1676
1677         if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_USER_MAGIC_SPECIFIC))
1678                 goto out;
1679
1680         lum3 = buf->lb_buf;
1681         /* In the function below, .hs_keycmp resolves to
1682          * pool_hashkey_keycmp() */
1683         /* coverity[overrun-buffer-val] */
1684         pool = lod_find_pool(d, lum3->lmm_pool_name);
1685         if (pool == NULL)
1686                 goto out;
1687
1688         if (!is_from_disk && stripe_offset != LOV_OFFSET_DEFAULT) {
1689                 rc = lod_check_index_in_pool(stripe_offset, pool);
1690                 if (rc < 0)
1691                         GOTO(out, rc = -EINVAL);
1692         }
1693
1694         if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1695                 CDEBUG(D_LAYOUT, "stripe count %u > # OSTs %u in the pool\n",
1696                        stripe_count, pool_tgt_count(pool));
1697                 GOTO(out, rc = -EINVAL);
1698         }
1699
1700 out:
1701         if (pool != NULL)
1702                 lod_pool_putref(pool);
1703
1704         RETURN(rc);
1705 }
1706
1707 static inline
1708 struct lov_comp_md_entry_v1 *comp_entry_v1(struct lov_comp_md_v1 *comp, int i)
1709 {
1710         LASSERTF((le32_to_cpu(comp->lcm_magic) & ~LOV_MAGIC_DEFINED) ==
1711                  LOV_USER_MAGIC_COMP_V1, "Wrong magic %x\n",
1712                  le32_to_cpu(comp->lcm_magic));
1713         LASSERTF(i >= 0 && i < le16_to_cpu(comp->lcm_entry_count),
1714                  "bad index %d, max = %d\n",
1715                  i, le16_to_cpu(comp->lcm_entry_count));
1716
1717         return &comp->lcm_entries[i];
1718 }
1719
1720 #define for_each_comp_entry_v1(comp, entry) \
1721         for (entry = comp_entry_v1(comp, 0); \
1722              entry <= comp_entry_v1(comp, \
1723                                    le16_to_cpu(comp->lcm_entry_count) - 1); \
1724              entry++)
1725
1726 int lod_erase_dom_stripe(struct lov_comp_md_v1 *comp_v1)
1727 {
1728         struct lov_comp_md_entry_v1 *ent, *dom_ent;
1729         __u16 entries;
1730         __u32 dom_off, dom_size, comp_size;
1731         void *blob_src, *blob_dst;
1732         unsigned int blob_size, blob_shift;
1733
1734         entries = le16_to_cpu(comp_v1->lcm_entry_count) - 1;
1735         /* if file has only DoM stripe return just error */
1736         if (entries == 0)
1737                 return -EFBIG;
1738
1739         comp_size = le32_to_cpu(comp_v1->lcm_size);
1740         dom_ent = &comp_v1->lcm_entries[0];
1741         dom_off = le32_to_cpu(dom_ent->lcme_offset);
1742         dom_size = le32_to_cpu(dom_ent->lcme_size);
1743
1744         /* shift entries array first */
1745         comp_v1->lcm_entry_count = cpu_to_le16(entries);
1746         memmove(dom_ent, dom_ent + 1,
1747                 entries * sizeof(struct lov_comp_md_entry_v1));
1748
1749         /* now move blob of layouts */
1750         blob_dst = (void *)comp_v1 + dom_off - sizeof(*dom_ent);
1751         blob_src = (void *)comp_v1 + dom_off + dom_size;
1752         blob_size = (unsigned long)((void *)comp_v1 + comp_size - blob_src);
1753         blob_shift = sizeof(*dom_ent) + dom_size;
1754
1755         memmove(blob_dst, blob_src, blob_size);
1756
1757         for_each_comp_entry_v1(comp_v1, ent) {
1758                 __u32 off;
1759
1760                 off = le32_to_cpu(ent->lcme_offset);
1761                 ent->lcme_offset = cpu_to_le32(off - blob_shift);
1762         }
1763
1764         comp_v1->lcm_size = cpu_to_le32(comp_size - blob_shift);
1765
1766         /* notify a caller to re-check entry */
1767         return -ERESTART;
1768 }
1769
1770 int lod_fix_dom_stripe(struct lod_device *d, struct lov_comp_md_v1 *comp_v1)
1771 {
1772         struct lov_comp_md_entry_v1 *ent, *dom_ent;
1773         struct lu_extent *dom_ext, *ext;
1774         struct lov_user_md_v1 *lum;
1775         __u32 stripe_size;
1776         __u16 mid, dom_mid;
1777         int rc = 0;
1778
1779         dom_ent = &comp_v1->lcm_entries[0];
1780         dom_ext = &dom_ent->lcme_extent;
1781         dom_mid = mirror_id_of(le32_to_cpu(dom_ent->lcme_id));
1782         stripe_size = d->lod_dom_max_stripesize;
1783
1784         lum = (void *)comp_v1 + le32_to_cpu(dom_ent->lcme_offset);
1785         CDEBUG(D_LAYOUT, "DoM component size %u was bigger than MDT limit %u, "
1786                "new size is %u\n", le32_to_cpu(lum->lmm_stripe_size),
1787                d->lod_dom_max_stripesize, stripe_size);
1788         lum->lmm_stripe_size = cpu_to_le32(stripe_size);
1789
1790         for_each_comp_entry_v1(comp_v1, ent) {
1791                 if (ent == dom_ent)
1792                         continue;
1793
1794                 mid = mirror_id_of(le32_to_cpu(ent->lcme_id));
1795                 if (mid != dom_mid)
1796                         continue;
1797
1798                 ext = &ent->lcme_extent;
1799                 if (ext->e_start != dom_ext->e_end)
1800                         continue;
1801
1802                 /* Found next component after the DoM one with the same
1803                  * mirror_id and adjust its start with DoM component end.
1804                  *
1805                  * NOTE: we are considering here that there can be only one
1806                  * DoM component in a file, all replicas are located on OSTs
1807                  * always and don't need adjustment since use own layouts.
1808                  */
1809                 ext->e_start = cpu_to_le64(stripe_size);
1810                 break;
1811         }
1812
1813         if (stripe_size == 0) {
1814                 /* DoM component size is zero due to server setting,
1815                  * remove it from the layout */
1816                 rc = lod_erase_dom_stripe(comp_v1);
1817         } else {
1818                 /* Update DoM extent end finally */
1819                 dom_ext->e_end = cpu_to_le64(stripe_size);
1820         }
1821
1822         return rc;
1823 }
1824
1825 /**
1826  * Verify LOV striping.
1827  *
1828  * \param[in] d                 LOD device
1829  * \param[in] buf               buffer with LOV EA to verify
1830  * \param[in] is_from_disk      0 - from user, allow some fields to be 0
1831  *                              1 - from disk, do not allow
1832  * \param[in] start             extent start for composite layout
1833  *
1834  * \retval                      0 if the striping is valid
1835  * \retval                      -EINVAL if striping is invalid
1836  */
1837 int lod_verify_striping(struct lod_device *d, struct lod_object *lo,
1838                         const struct lu_buf *buf, bool is_from_disk)
1839 {
1840         struct lov_desc *desc = &d->lod_desc;
1841         struct lov_user_md_v1   *lum;
1842         struct lov_comp_md_v1   *comp_v1;
1843         struct lov_comp_md_entry_v1     *ent;
1844         struct lu_extent        *ext;
1845         struct lu_buf   tmp;
1846         __u64   prev_end = 0;
1847         __u32   stripe_size = 0;
1848         __u16   prev_mid = -1, mirror_id = -1;
1849         __u32   mirror_count;
1850         __u32   magic;
1851         int     rc = 0;
1852         ENTRY;
1853
1854         if (buf->lb_len < sizeof(lum->lmm_magic)) {
1855                 CDEBUG(D_LAYOUT, "invalid buf len %zu\n", buf->lb_len);
1856                 RETURN(-EINVAL);
1857         }
1858
1859         lum = buf->lb_buf;
1860
1861         magic = le32_to_cpu(lum->lmm_magic) & ~LOV_MAGIC_DEFINED;
1862         /* treat foreign LOV EA/object case first
1863          * XXX is it expected to try setting again a foreign?
1864          * XXX should we care about different current vs new layouts ?
1865          */
1866         if (unlikely(magic == LOV_USER_MAGIC_FOREIGN)) {
1867                 struct lov_foreign_md *lfm = buf->lb_buf;
1868
1869                 if (buf->lb_len < offsetof(typeof(*lfm), lfm_value)) {
1870                         CDEBUG(D_LAYOUT,
1871                                "buf len %zu < min lov_foreign_md size (%zu)\n",
1872                                buf->lb_len, offsetof(typeof(*lfm),
1873                                lfm_value));
1874                         RETURN(-EINVAL);
1875                 }
1876
1877                 if (foreign_size_le(lfm) > buf->lb_len) {
1878                         CDEBUG(D_LAYOUT,
1879                                "buf len %zu < this lov_foreign_md size (%zu)\n",
1880                                buf->lb_len, foreign_size_le(lfm));
1881                         RETURN(-EINVAL);
1882                 }
1883                 /* Don't do anything with foreign layouts */
1884                 RETURN(0);
1885         }
1886
1887         /* normal LOV/layout cases */
1888
1889         if (buf->lb_len < sizeof(*lum)) {
1890                 CDEBUG(D_LAYOUT, "buf len %zu too small for lov_user_md\n",
1891                        buf->lb_len);
1892                 RETURN(-EINVAL);
1893         }
1894
1895         if (magic != LOV_USER_MAGIC_V1 &&
1896             magic != LOV_USER_MAGIC_V3 &&
1897             magic != LOV_USER_MAGIC_SPECIFIC &&
1898             magic != LOV_USER_MAGIC_COMP_V1) {
1899                 CDEBUG(D_LAYOUT, "bad userland LOV MAGIC: %#x\n",
1900                        le32_to_cpu(lum->lmm_magic));
1901                 RETURN(-EINVAL);
1902         }
1903
1904         if (magic != LOV_USER_MAGIC_COMP_V1)
1905                 RETURN(lod_verify_v1v3(d, buf, is_from_disk));
1906
1907         /* magic == LOV_USER_MAGIC_COMP_V1 */
1908         comp_v1 = buf->lb_buf;
1909         if (buf->lb_len < le32_to_cpu(comp_v1->lcm_size)) {
1910                 CDEBUG(D_LAYOUT, "buf len %zu is less than %u\n",
1911                        buf->lb_len, le32_to_cpu(comp_v1->lcm_size));
1912                 RETURN(-EINVAL);
1913         }
1914
1915 recheck:
1916         mirror_count = 0;
1917         if (le16_to_cpu(comp_v1->lcm_entry_count) == 0) {
1918                 CDEBUG(D_LAYOUT, "entry count is zero\n");
1919                 RETURN(-EINVAL);
1920         }
1921
1922         if (S_ISREG(lod2lu_obj(lo)->lo_header->loh_attr) &&
1923             lo->ldo_comp_cnt > 0) {
1924                 /* could be called from lustre.lov.add */
1925                 __u32 cnt = lo->ldo_comp_cnt;
1926
1927                 ext = &lo->ldo_comp_entries[cnt - 1].llc_extent;
1928                 prev_end = ext->e_end;
1929
1930                 ++mirror_count;
1931         }
1932
1933         for_each_comp_entry_v1(comp_v1, ent) {
1934                 ext = &ent->lcme_extent;
1935
1936                 if (le64_to_cpu(ext->e_start) >= le64_to_cpu(ext->e_end)) {
1937                         CDEBUG(D_LAYOUT, "invalid extent "DEXT"\n",
1938                                le64_to_cpu(ext->e_start),
1939                                le64_to_cpu(ext->e_end));
1940                         RETURN(-EINVAL);
1941                 }
1942
1943                 if (is_from_disk) {
1944                         /* lcme_id contains valid value */
1945                         if (le32_to_cpu(ent->lcme_id) == 0 ||
1946                             le32_to_cpu(ent->lcme_id) > LCME_ID_MAX) {
1947                                 CDEBUG(D_LAYOUT, "invalid id %u\n",
1948                                        le32_to_cpu(ent->lcme_id));
1949                                 RETURN(-EINVAL);
1950                         }
1951
1952                         if (le16_to_cpu(comp_v1->lcm_mirror_count) > 0) {
1953                                 mirror_id = mirror_id_of(
1954                                                 le32_to_cpu(ent->lcme_id));
1955
1956                                 /* first component must start with 0 */
1957                                 if (mirror_id != prev_mid &&
1958                                     le64_to_cpu(ext->e_start) != 0) {
1959                                         CDEBUG(D_LAYOUT,
1960                                                "invalid start:%llu, expect:0\n",
1961                                                le64_to_cpu(ext->e_start));
1962                                         RETURN(-EINVAL);
1963                                 }
1964
1965                                 prev_mid = mirror_id;
1966                         }
1967                 }
1968
1969                 if (le64_to_cpu(ext->e_start) == 0) {
1970                         ++mirror_count;
1971                         prev_end = 0;
1972                 }
1973
1974                 /* the next must be adjacent with the previous one */
1975                 if (le64_to_cpu(ext->e_start) != prev_end) {
1976                         CDEBUG(D_LAYOUT,
1977                                "invalid start actual:%llu, expect:%llu\n",
1978                                le64_to_cpu(ext->e_start), prev_end);
1979                         RETURN(-EINVAL);
1980                 }
1981
1982                 tmp.lb_buf = (char *)comp_v1 + le32_to_cpu(ent->lcme_offset);
1983                 tmp.lb_len = le32_to_cpu(ent->lcme_size);
1984
1985                 /* Check DoM entry is always the first one */
1986                 lum = tmp.lb_buf;
1987                 if (lov_pattern(le32_to_cpu(lum->lmm_pattern)) ==
1988                     LOV_PATTERN_MDT) {
1989                         /* DoM component can be only the first stripe */
1990                         if (le64_to_cpu(ext->e_start) > 0) {
1991                                 CDEBUG(D_LAYOUT, "invalid DoM component "
1992                                        "with %llu extent start\n",
1993                                        le64_to_cpu(ext->e_start));
1994                                 RETURN(-EINVAL);
1995                         }
1996                         stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1997                         /* There is just one stripe on MDT and it must
1998                          * cover whole component size. */
1999                         if (stripe_size != le64_to_cpu(ext->e_end)) {
2000                                 CDEBUG(D_LAYOUT, "invalid DoM layout "
2001                                        "stripe size %u != %llu "
2002                                        "(component size)\n",
2003                                        stripe_size, prev_end);
2004                                 RETURN(-EINVAL);
2005                         }
2006                         /* Check stripe size againts per-MDT limit */
2007                         if (stripe_size > d->lod_dom_max_stripesize) {
2008                                 CDEBUG(D_LAYOUT, "DoM component size "
2009                                        "%u is bigger than MDT limit %u, check "
2010                                        "dom_max_stripesize parameter\n",
2011                                        stripe_size, d->lod_dom_max_stripesize);
2012                                 rc = lod_fix_dom_stripe(d, comp_v1);
2013                                 if (rc == -ERESTART) {
2014                                         /* DoM entry was removed, re-check
2015                                          * new layout from start */
2016                                         goto recheck;
2017                                 } else if (rc) {
2018                                         RETURN(rc);
2019                                 }
2020                         }
2021                 }
2022
2023                 prev_end = le64_to_cpu(ext->e_end);
2024
2025                 rc = lod_verify_v1v3(d, &tmp, is_from_disk);
2026                 if (rc)
2027                         RETURN(rc);
2028
2029                 if (prev_end == LUSTRE_EOF)
2030                         continue;
2031
2032                 /* extent end must be aligned with the stripe_size */
2033                 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
2034                 if (stripe_size == 0)
2035                         stripe_size = desc->ld_default_stripe_size;
2036                 if (stripe_size == 0 || (prev_end & (stripe_size - 1))) {
2037                         CDEBUG(D_LAYOUT, "stripe size isn't aligned, "
2038                                "stripe_sz: %u, [%llu, %llu)\n",
2039                                stripe_size, ext->e_start, prev_end);
2040                         RETURN(-EINVAL);
2041                 }
2042         }
2043
2044         /* make sure that the mirror_count is telling the truth */
2045         if (mirror_count != le16_to_cpu(comp_v1->lcm_mirror_count) + 1)
2046                 RETURN(-EINVAL);
2047
2048         RETURN(0);
2049 }
2050
2051 /**
2052  * set the default stripe size, if unset.
2053  *
2054  * \param[in,out] val   number of bytes per OST stripe
2055  *
2056  * The minimum stripe size is 64KB to ensure that a single stripe is an
2057  * even multiple of a client PAGE_SIZE (IA64, PPC, etc).  Otherwise, it
2058  * is difficult to split dirty pages across OSCs during writes.
2059  */
2060 void lod_fix_desc_stripe_size(__u64 *val)
2061 {
2062         if (*val < LOV_MIN_STRIPE_SIZE) {
2063                 if (*val != 0)
2064                         LCONSOLE_INFO("Increasing default stripe size to "
2065                                       "minimum value %u\n",
2066                                       LOV_DESC_STRIPE_SIZE_DEFAULT);
2067                 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
2068         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
2069                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
2070                 LCONSOLE_WARN("Changing default stripe size to %llu (a "
2071                               "multiple of %u)\n",
2072                               *val, LOV_MIN_STRIPE_SIZE);
2073         }
2074 }
2075
2076 /**
2077  * set the filesystem default number of stripes, if unset.
2078  *
2079  * \param[in,out] val   number of stripes
2080  *
2081  * A value of "0" means "use the system-wide default stripe count", which
2082  * has either been inherited by now, or falls back to 1 stripe per file.
2083  * A value of "-1" (0xffffffff) means "stripe over all available OSTs",
2084  * and is a valid value, so is left unchanged here.
2085  */
2086 void lod_fix_desc_stripe_count(__u32 *val)
2087 {
2088         if (*val == 0)
2089                 *val = 1;
2090 }
2091
2092 /**
2093  * set the filesystem default layout pattern
2094  *
2095  * \param[in,out] val   LOV_PATTERN_* layout
2096  *
2097  * A value of "0" means "use the system-wide default layout type", which
2098  * has either been inherited by now, or falls back to plain RAID0 striping.
2099  */
2100 void lod_fix_desc_pattern(__u32 *val)
2101 {
2102         /* from lov_setstripe */
2103         if ((*val != 0) && (*val != LOV_PATTERN_RAID0) &&
2104             (*val != LOV_PATTERN_MDT)) {
2105                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
2106                 *val = 0;
2107         }
2108 }
2109
2110 void lod_fix_desc_qos_maxage(__u32 *val)
2111 {
2112         /* fix qos_maxage */
2113         if (*val == 0)
2114                 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
2115 }
2116
2117 /**
2118  * Used to fix insane default striping.
2119  *
2120  * \param[in] desc      striping description
2121  */
2122 void lod_fix_desc(struct lov_desc *desc)
2123 {
2124         lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
2125         lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
2126         lod_fix_desc_pattern(&desc->ld_pattern);
2127         lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
2128 }
2129
2130 /**
2131  * Initialize the structures used to store pools and default striping.
2132  *
2133  * \param[in] lod       LOD device
2134  * \param[in] lcfg      configuration structure storing default striping.
2135  *
2136  * \retval              0 if initialization succeeds
2137  * \retval              negative error number on failure
2138  */
2139 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
2140 {
2141         struct obd_device          *obd;
2142         struct lov_desc            *desc;
2143         int                         rc;
2144         ENTRY;
2145
2146         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
2147         LASSERT(obd != NULL);
2148         obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
2149
2150         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2151                 CERROR("LOD setup requires a descriptor\n");
2152                 RETURN(-EINVAL);
2153         }
2154
2155         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
2156
2157         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
2158                 CERROR("descriptor size wrong: %d > %d\n",
2159                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
2160                 RETURN(-EINVAL);
2161         }
2162
2163         if (desc->ld_magic != LOV_DESC_MAGIC) {
2164                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
2165                         CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
2166                                obd->obd_name, desc);
2167                         lustre_swab_lov_desc(desc);
2168                 } else {
2169                         CERROR("%s: Bad lov desc magic: %#x\n",
2170                                obd->obd_name, desc->ld_magic);
2171                         RETURN(-EINVAL);
2172                 }
2173         }
2174
2175         lod_fix_desc(desc);
2176
2177         desc->ld_active_tgt_count = 0;
2178         lod->lod_desc = *desc;
2179
2180         lod->lod_sp_me = LUSTRE_SP_CLI;
2181
2182         /* Set up allocation policy (QoS and RR) */
2183         INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
2184         init_rwsem(&lod->lod_qos.lq_rw_sem);
2185         lod->lod_qos.lq_dirty = 1;
2186         lod->lod_qos.lq_rr.lqr_dirty = 1;
2187         lod->lod_qos.lq_reset = 1;
2188         /* Default priority is toward free space balance */
2189         lod->lod_qos.lq_prio_free = 232;
2190         /* Default threshold for rr (roughly 17%) */
2191         lod->lod_qos.lq_threshold_rr = 43;
2192
2193         /* Set up OST pool environment */
2194         lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
2195                                                    HASH_POOLS_MAX_BITS,
2196                                                    HASH_POOLS_BKT_BITS, 0,
2197                                                    CFS_HASH_MIN_THETA,
2198                                                    CFS_HASH_MAX_THETA,
2199                                                    &pool_hash_operations,
2200                                                    CFS_HASH_DEFAULT);
2201         if (lod->lod_pools_hash_body == NULL)
2202                 RETURN(-ENOMEM);
2203
2204         INIT_LIST_HEAD(&lod->lod_pool_list);
2205         lod->lod_pool_count = 0;
2206         rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
2207         if (rc)
2208                 GOTO(out_hash, rc);
2209         lod_qos_rr_init(&lod->lod_qos.lq_rr);
2210         rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
2211         if (rc)
2212                 GOTO(out_pool_info, rc);
2213
2214         RETURN(0);
2215
2216 out_pool_info:
2217         lod_ost_pool_free(&lod->lod_pool_info);
2218 out_hash:
2219         cfs_hash_putref(lod->lod_pools_hash_body);
2220
2221         return rc;
2222 }
2223
2224 /**
2225  * Release the structures describing the pools.
2226  *
2227  * \param[in] lod       LOD device from which we release the structures
2228  *
2229  * \retval              0 always
2230  */
2231 int lod_pools_fini(struct lod_device *lod)
2232 {
2233         struct obd_device   *obd = lod2obd(lod);
2234         struct pool_desc    *pool, *tmp;
2235         ENTRY;
2236
2237         list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
2238                 /* free pool structs */
2239                 CDEBUG(D_INFO, "delete pool %p\n", pool);
2240                 /* In the function below, .hs_keycmp resolves to
2241                  * pool_hashkey_keycmp() */
2242                 /* coverity[overrun-buffer-val] */
2243                 lod_pool_del(obd, pool->pool_name);
2244         }
2245
2246         cfs_hash_putref(lod->lod_pools_hash_body);
2247         lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
2248         lod_ost_pool_free(&lod->lod_pool_info);
2249
2250         RETURN(0);
2251 }