4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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.
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
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * lustre/lod/lod_lov.c
31 * A set of helpers to maintain Logical Object Volume (LOV)
32 * Extended Attribute (EA) and known OST targets
34 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 #define DEBUG_SUBSYSTEM S_MDS
39 #include <obd_class.h>
40 #include <lustre_lfsck.h>
41 #include <lustre_lmv.h>
42 #include <lustre_swab.h>
44 #include "lod_internal.h"
47 * Increase reference count on the target table.
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.
53 * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs)
55 void lod_getref(struct lod_tgt_descs *ltd)
57 down_read(<d->ltd_rw_sem);
58 mutex_lock(<d->ltd_mutex);
60 mutex_unlock(<d->ltd_mutex);
64 * Decrease reference count on the target table.
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.
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)
73 void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
75 mutex_lock(<d->ltd_mutex);
77 if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
78 struct lod_tgt_desc *tgt_desc, *tmp;
79 struct list_head kill;
82 CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
85 INIT_LIST_HEAD(&kill);
87 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
88 tgt_desc = LTD_TGT(ltd, idx);
91 if (!tgt_desc->ltd_reap)
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--;
103 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
104 ltd->ltd_death_row--;
106 mutex_unlock(<d->ltd_mutex);
107 up_read(<d->ltd_rw_sem);
109 list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
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);
116 CERROR("%s: qos_del_tgt(%s) failed:"
118 lod2obd(lod)->obd_name,
119 obd_uuid2str(&tgt_desc->ltd_uuid),
122 rc = obd_disconnect(tgt_desc->ltd_exp);
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);
130 mutex_unlock(<d->ltd_mutex);
131 up_read(<d->ltd_rw_sem);
136 * Expand size of target table.
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.
143 * \param[in] ltd target table
144 * \param[in] newsize new size of the table
146 * \retval 0 on success
147 * \retval -ENOMEM if reallocation failed
149 static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
151 struct cfs_bitmap *new_bitmap, *old_bitmap = NULL;
155 /* grab write reference on the lod. Relocating the array requires
156 * exclusive access */
158 down_write(<d->ltd_rw_sem);
159 if (newsize <= ltd->ltd_tgts_size)
160 /* someone else has already resize the array */
163 /* allocate new bitmap */
164 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
166 GOTO(out, rc = -ENOMEM);
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;
175 ltd->ltd_tgts_size = newsize;
176 ltd->ltd_tgt_bitmap = new_bitmap;
179 CFS_FREE_BITMAP(old_bitmap);
181 CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
185 up_write(<d->ltd_rw_sem);
190 * Connect LOD to a new OSP and add it to the target table.
192 * Connect to the OSP device passed, initialize all the internal
193 * structures related to the device and add it to the target table.
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
204 * \retval 0 if added successfully
205 * \retval negative error number on failure
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)
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;
217 struct lod_tgt_desc *tgt_desc;
218 struct lod_tgt_descs *ltd;
219 struct lustre_cfg *lcfg;
220 struct obd_uuid obd_uuid;
225 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
228 CERROR("request to add OBD %s with invalid generation: %d\n",
233 obd_str2uuid(&obd_uuid, osp);
235 obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
236 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
238 CERROR("can't find %s device\n", osp);
242 LASSERT(obd->obd_lu_dev != NULL);
243 LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
245 lu_dev = obd->obd_lu_dev;
246 dt_dev = lu2dt_dev(lu_dev);
250 GOTO(out_cleanup, rc = -ENOMEM);
252 data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
253 data->ocd_version = LUSTRE_VERSION_CODE;
254 data->ocd_index = index;
256 if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
258 data->ocd_connect_flags |= OBD_CONNECT_AT |
261 #ifdef HAVE_LRU_RESIZE_SUPPORT
262 OBD_CONNECT_LRU_RESIZE |
265 OBD_CONNECT_REQPORTAL |
266 OBD_CONNECT_SKIP_ORPHAN |
268 OBD_CONNECT_LVB_TYPE |
269 OBD_CONNECT_VERSION |
270 OBD_CONNECT_PINGLESS |
272 OBD_CONNECT_BULK_MBITS;
274 data->ocd_group = tgt_index;
275 ltd = &lod->lod_ost_descs;
277 struct obd_import *imp = obd->u.cli.cl_import;
280 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
281 data->ocd_connect_flags |= OBD_CONNECT_ACL |
283 OBD_CONNECT_MDS_MDS |
288 OBD_CONNECT_BULK_MBITS;
289 spin_lock(&imp->imp_lock);
290 imp->imp_server_timeout = 1;
291 spin_unlock(&imp->imp_lock);
292 imp->imp_client->cli_request_portal = OUT_PORTAL;
293 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
295 ltd = &lod->lod_mdt_descs;
298 rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
301 CERROR("%s: cannot connect to next dev %s (%d)\n",
302 obd->obd_name, osp, rc);
303 GOTO(out_cleanup, rc);
306 /* Allocate ost descriptor and fill it */
307 OBD_ALLOC_PTR(tgt_desc);
309 GOTO(out_conn, rc = -ENOMEM);
311 tgt_desc->ltd_tgt = dt_dev;
312 tgt_desc->ltd_exp = exp;
313 tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
314 tgt_desc->ltd_gen = gen;
315 tgt_desc->ltd_index = index;
316 tgt_desc->ltd_active = active;
319 if (index >= ltd->ltd_tgts_size) {
320 /* we have to increase the size of the lod_osts array */
323 newsize = max(ltd->ltd_tgts_size, (__u32)2);
324 while (newsize < index + 1)
325 newsize = newsize << 1;
327 /* lod_bitmap_resize() needs lod_rw_sem
328 * which we hold with th reference */
329 lod_putref(lod, ltd);
331 rc = ltd_bitmap_resize(ltd, newsize);
338 mutex_lock(<d->ltd_mutex);
340 if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
341 CERROR("%s: device %d is registered already\n", obd->obd_name,
343 GOTO(out_mutex, rc = -EEXIST);
346 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
347 OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
348 if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
349 CERROR("can't allocate index to add %s\n",
351 GOTO(out_mutex, rc = -ENOMEM);
356 /* pool and qos are not supported for MDS stack yet */
357 rc = lod_ost_pool_add(&lod->lod_pool_info, index,
360 CERROR("%s: can't set up pool, failed with %d\n",
365 rc = qos_add_tgt(lod, tgt_desc);
367 CERROR("%s: qos_add_tgt failed with %d\n",
372 /* The new OST is now a full citizen */
373 if (index >= lod->lod_desc.ld_tgt_count)
374 lod->lod_desc.ld_tgt_count = index + 1;
376 lod->lod_desc.ld_active_tgt_count++;
379 LTD_TGT(ltd, index) = tgt_desc;
380 cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
382 mutex_unlock(<d->ltd_mutex);
383 lod_putref(lod, ltd);
385 if (lod->lod_recovery_completed)
386 lu_dev->ld_ops->ldo_recovery_complete(env, lu_dev);
388 if (!for_ost && lod->lod_initialized) {
389 rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt);
391 CERROR("%s: cannot start llog on %s:rc = %d\n",
392 lod2obd(lod)->obd_name, osp, rc);
397 rc = lfsck_add_target(env, lod->lod_child, dt_dev, exp, index, for_ost);
399 CERROR("Fail to add LFSCK target: name = %s, type = %s, "
400 "index = %u, rc = %d\n", osp, type, index, rc);
401 GOTO(out_fini_llog, rc);
405 lod_sub_fini_llog(env, tgt_desc->ltd_tgt,
406 tgt_desc->ltd_recovery_thread);
409 mutex_lock(<d->ltd_mutex);
411 if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) {
412 struct ptlrpc_thread *thread;
414 thread = LTD_TGT(ltd, index)->ltd_recovery_thread;
415 OBD_FREE_PTR(thread);
418 cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index);
419 LTD_TGT(ltd, index) = NULL;
421 lod_ost_pool_remove(&lod->lod_pool_info, index);
424 mutex_unlock(<d->ltd_mutex);
425 lod_putref(lod, ltd);
428 OBD_FREE_PTR(tgt_desc);
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);
444 * Schedule target removal from the target table.
446 * Mark the device as dead. The device is not removed here because it may
447 * still be in use. The device will be removed in lod_putref() when the
448 * last reference is released.
450 * \param[in] env execution environment for this thread
451 * \param[in] lod LOD device the target table belongs to
452 * \param[in] ltd target table
453 * \param[in] idx index of the target
454 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
456 static void __lod_del_device(const struct lu_env *env, struct lod_device *lod,
457 struct lod_tgt_descs *ltd, unsigned idx,
460 LASSERT(LTD_TGT(ltd, idx));
462 lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt,
465 if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) {
466 struct ptlrpc_thread *thread;
468 thread = LTD_TGT(ltd, idx)->ltd_recovery_thread;
469 OBD_FREE_PTR(thread);
472 if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
473 LTD_TGT(ltd, idx)->ltd_reap = 1;
474 ltd->ltd_death_row++;
479 * Schedule removal of all the targets from the given target table.
481 * See more details in the description for __lod_del_device()
483 * \param[in] env execution environment for this thread
484 * \param[in] lod LOD device the target table belongs to
485 * \param[in] ltd target table
486 * \param[in] for_ost type of the target: MDT or OST
490 int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod,
491 struct lod_tgt_descs *ltd, bool for_ost)
495 if (ltd->ltd_tgts_size <= 0)
498 mutex_lock(<d->ltd_mutex);
499 cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
500 __lod_del_device(env, lod, ltd, idx, for_ost);
501 mutex_unlock(<d->ltd_mutex);
502 lod_putref(lod, ltd);
503 CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
504 for (idx = 0; idx < TGT_PTRS; idx++) {
505 if (ltd->ltd_tgt_idx[idx])
506 OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
508 ltd->ltd_tgts_size = 0;
513 * Remove device by name.
515 * Remove a device identified by \a osp from the target table. Given
516 * the device can be in use, the real deletion happens in lod_putref().
518 * \param[in] env execution environment for this thread
519 * \param[in] lod LOD device to be connected to the new OSP
520 * \param[in] ltd target table
521 * \param[in] osp name of OSP device to be removed
522 * \param[in] idx index of the target
523 * \param[in] gen generation number, not used currently
524 * \param[in] for_ost type of the target: 0 - MDT, 1 - OST
526 * \retval 0 if the device was scheduled for removal
527 * \retval -EINVAL if no device was found
529 int lod_del_device(const struct lu_env *env, struct lod_device *lod,
530 struct lod_tgt_descs *ltd, char *osp, unsigned idx,
531 unsigned gen, bool for_ost)
533 struct obd_device *obd;
535 struct obd_uuid uuid;
538 CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
540 obd_str2uuid(&uuid, osp);
542 obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
543 &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
545 CERROR("can't find %s device\n", osp);
550 CERROR("%s: request to remove OBD %s with invalid generation %d"
551 "\n", obd->obd_name, osp, gen);
555 obd_str2uuid(&uuid, osp);
558 mutex_lock(<d->ltd_mutex);
559 /* check that the index is allocated in the bitmap */
560 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
561 !LTD_TGT(ltd, idx)) {
562 CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
563 GOTO(out, rc = -EINVAL);
566 /* check that the UUID matches */
567 if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
568 CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
569 obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
571 GOTO(out, rc = -EINVAL);
574 __lod_del_device(env, lod, ltd, idx, for_ost);
577 mutex_unlock(<d->ltd_mutex);
578 lod_putref(lod, ltd);
583 * Resize per-thread storage to hold specified size.
585 * A helper function to resize per-thread temporary storage. This storage
586 * is used to process LOV/LVM EAs and may be quite large. We do not want to
587 * allocate/release it every time, so instead we put it into the env and
588 * reallocate on demand. The memory is released when the correspondent thread
591 * \param[in] info LOD-specific storage in the environment
592 * \param[in] size new size to grow the buffer to
594 * \retval 0 on success, -ENOMEM if reallocation failed
596 int lod_ea_store_resize(struct lod_thread_info *info, size_t size)
598 __u32 round = size_roundup_power2(size);
601 lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
602 if (info->lti_ea_store) {
603 LASSERT(info->lti_ea_store_size);
604 LASSERT(info->lti_ea_store_size < round);
605 CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
606 info->lti_ea_store_size, round);
607 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
608 info->lti_ea_store = NULL;
609 info->lti_ea_store_size = 0;
612 OBD_ALLOC_LARGE(info->lti_ea_store, round);
613 if (info->lti_ea_store == NULL)
615 info->lti_ea_store_size = round;
621 * Make LOV EA for striped object.
623 * Generate striping information and store it in the LOV EA of the given
624 * object. The caller must ensure nobody else is calling the function
625 * against the object concurrently. The transaction must be started.
626 * FLDB service must be running as well; it's used to map FID to the target,
627 * which is stored in LOV EA.
629 * \param[in] env execution environment for this thread
630 * \param[in] lo LOD object
631 * \param[in] th transaction handle
633 * \retval 0 if LOV EA is stored successfully
634 * \retval negative error number on failure
636 int lod_generate_and_set_lovea(const struct lu_env *env,
637 struct lod_object *lo, struct thandle *th)
639 struct lod_thread_info *info = lod_env_info(env);
640 struct dt_object *next = dt_object_child(&lo->ldo_obj);
641 const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
642 struct lov_mds_md_v1 *lmm;
643 struct lov_ost_data_v1 *objs;
651 magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
652 lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
653 if (info->lti_ea_store_size < lmm_size) {
654 rc = lod_ea_store_resize(info, lmm_size);
659 if (lo->ldo_pattern == 0) /* default striping */
660 lo->ldo_pattern = LOV_PATTERN_RAID0;
662 lmm = info->lti_ea_store;
664 lmm->lmm_magic = cpu_to_le32(magic);
665 lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
666 fid_to_lmm_oi(fid, &lmm->lmm_oi);
667 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_LMMOI))
668 lmm->lmm_oi.oi.oi_id++;
669 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
670 lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
671 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
672 if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
673 lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
674 lmm->lmm_layout_gen = 0;
675 if (magic == LOV_MAGIC_V1) {
676 objs = &lmm->lmm_objects[0];
678 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
679 size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
680 sizeof(v3->lmm_pool_name));
681 if (cplen >= sizeof(v3->lmm_pool_name))
683 objs = &v3->lmm_objects[0];
686 for (i = 0; i < lo->ldo_stripenr; i++) {
687 struct lu_fid *fid = &info->lti_fid;
688 struct lod_device *lod;
690 int type = LU_SEQ_RANGE_OST;
692 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
693 LASSERT(lo->ldo_stripe[i]);
695 *fid = *lu_object_fid(&lo->ldo_stripe[i]->do_lu);
696 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MULTIPLE_REF)) {
697 if (cfs_fail_val == 0)
698 cfs_fail_val = fid->f_oid;
700 fid->f_oid = cfs_fail_val;
703 rc = fid_to_ostid(fid, &info->lti_ostid);
706 ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
707 objs[i].l_ost_gen = cpu_to_le32(0);
708 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FLD_LOOKUP))
711 rc = lod_fld_lookup(env, lod, fid,
714 CERROR("%s: Can not locate "DFID": rc = %d\n",
715 lod2obd(lod)->obd_name, PFID(fid), rc);
716 lod_object_free_striping(env, lo);
719 objs[i].l_ost_idx = cpu_to_le32(index);
722 info->lti_buf.lb_buf = lmm;
723 info->lti_buf.lb_len = lmm_size;
724 rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
727 lod_object_free_striping(env, lo);
737 * Fill lti_ea_store buffer in the environment with a value for the given
738 * EA. The buffer is reallocated if the value doesn't fit.
740 * \param[in,out] env execution environment for this thread
741 * .lti_ea_store buffer is filled with EA's value
742 * \param[in] lo LOD object
743 * \param[in] name name of the EA
745 * \retval 0 if EA is fetched successfully
746 * \retval negative error number on failure
748 int lod_get_ea(const struct lu_env *env, struct lod_object *lo,
751 struct lod_thread_info *info = lod_env_info(env);
752 struct dt_object *next = dt_object_child(&lo->ldo_obj);
758 if (unlikely(info->lti_ea_store == NULL)) {
759 /* just to enter in allocation block below */
763 info->lti_buf.lb_buf = info->lti_ea_store;
764 info->lti_buf.lb_len = info->lti_ea_store_size;
765 rc = dt_xattr_get(env, next, &info->lti_buf, name);
768 /* if object is not striped or inaccessible */
769 if (rc == -ENODATA || rc == -ENOENT)
773 /* EA doesn't fit, reallocate new buffer */
774 rc = dt_xattr_get(env, next, &LU_BUF_NULL, name);
775 if (rc == -ENODATA || rc == -ENOENT)
781 rc = lod_ea_store_resize(info, rc);
791 * Verify the target index is present in the current configuration.
793 * \param[in] md LOD device where the target table is stored
794 * \param[in] idx target's index
796 * \retval 0 if the index is present
797 * \retval -EINVAL if not
799 static int validate_lod_and_idx(struct lod_device *md, __u32 idx)
801 if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
802 !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
803 CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
804 md->lod_ost_descs.ltd_tgts_size);
808 if (unlikely(OST_TGT(md, idx) == NULL)) {
809 CERROR("%s: bad lod_tgt_desc for idx: %d\n",
810 lod2obd(md)->obd_name, idx);
814 if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
815 CERROR("%s: invalid lod device, for idx: %d\n",
816 lod2obd(md)->obd_name , idx);
824 * Instantiate objects for stripes.
826 * Allocate and initialize LU-objects representing the stripes. The number
827 * of the stripes (ldo_stripenr) must be initialized already. The caller
828 * must ensure nobody else is calling the function on the object at the same
829 * time. FLDB service must be running to be able to map a FID to the targets
830 * and find appropriate device representing that target.
832 * \param[in] env execution environment for this thread
833 * \param[in,out] lo LOD object
834 * \param[in] objs an array of IDs to creates the objects from
836 * \retval 0 if the objects are instantiated successfully
837 * \retval negative error number on failure
839 int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
840 struct lov_ost_data_v1 *objs)
842 struct lod_thread_info *info = lod_env_info(env);
843 struct lod_device *md;
844 struct lu_object *o, *n;
845 struct lu_device *nd;
846 struct dt_object **stripe;
853 md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
854 LASSERT(lo->ldo_stripe == NULL);
855 LASSERT(lo->ldo_stripenr > 0);
856 LASSERT(lo->ldo_stripe_size > 0);
858 stripe_len = lo->ldo_stripenr;
859 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
863 for (i = 0; i < lo->ldo_stripenr; i++) {
864 if (unlikely(lovea_slot_is_dummy(&objs[i])))
867 ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
868 idx = le32_to_cpu(objs[i].l_ost_idx);
869 rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
872 LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
873 PFID(&info->lti_fid));
874 lod_getref(&md->lod_ost_descs);
876 rc = validate_lod_and_idx(md, idx);
877 if (unlikely(rc != 0)) {
878 lod_putref(md, &md->lod_ost_descs);
882 nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
883 lod_putref(md, &md->lod_ost_descs);
885 /* In the function below, .hs_keycmp resolves to
886 * u_obj_hop_keycmp() */
887 /* coverity[overrun-buffer-val] */
888 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
890 GOTO(out, rc = PTR_ERR(o));
892 n = lu_object_locate(o->lo_header, nd->ld_type);
895 stripe[i] = container_of(n, struct dt_object, do_lu);
900 for (i = 0; i < stripe_len; i++)
901 if (stripe[i] != NULL)
902 lu_object_put(env, &stripe[i]->do_lu);
904 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
905 lo->ldo_stripenr = 0;
907 lo->ldo_stripe = stripe;
908 lo->ldo_stripes_allocated = stripe_len;
915 * Instantiate objects for striping.
917 * Parse striping information in \a buf and instantiate the objects
918 * representing the stripes.
920 * \param[in] env execution environment for this thread
921 * \param[in] lo LOD object
922 * \param[in] buf buffer storing LOV EA to parse
924 * \retval 0 if parsing and objects creation succeed
925 * \retval negative error number on failure
927 int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
928 const struct lu_buf *buf)
930 struct lov_mds_md_v1 *lmm;
931 struct lov_ost_data_v1 *objs;
938 LASSERT(buf->lb_buf);
939 LASSERT(buf->lb_len);
941 lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
942 magic = le32_to_cpu(lmm->lmm_magic);
943 pattern = le32_to_cpu(lmm->lmm_pattern);
945 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
946 GOTO(out, rc = -EINVAL);
947 if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
948 GOTO(out, rc = -EINVAL);
950 lo->ldo_pattern = pattern;
951 lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
952 lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
953 lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
954 /* released file stripenr fixup. */
955 if (pattern & LOV_PATTERN_F_RELEASED)
956 lo->ldo_stripenr = 0;
958 LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
960 if (magic == LOV_MAGIC_V3) {
961 struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
962 objs = &v3->lmm_objects[0];
963 /* no need to set pool, which is used in create only */
965 objs = &lmm->lmm_objects[0];
968 if (lo->ldo_stripenr > 0)
969 rc = lod_initialize_objects(env, lo, objs);
976 * Initialize the object representing the stripes.
978 * Unless the stripes are initialized already, fetch LOV (for regular
979 * objects) or LMV (for directory objects) EA and call lod_parse_striping()
980 * to instantiate the objects representing the stripes.
982 * \param[in] env execution environment for this thread
983 * \param[in,out] lo LOD object
985 * \retval 0 if parsing and object creation succeed
986 * \retval negative error number on failure
988 int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo)
990 struct lod_thread_info *info = lod_env_info(env);
991 struct lu_buf *buf = &info->lti_buf;
992 struct dt_object *next = dt_object_child(&lo->ldo_obj);
996 /* already initialized? */
997 if (lo->ldo_stripe != NULL)
1000 if (!dt_object_exists(next))
1003 /* Do not load stripe for slaves of striped dir */
1004 if (lo->ldo_dir_slave_stripe)
1007 if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) {
1008 rc = lod_get_lov_ea(env, lo);
1012 * there is LOV EA (striping information) in this object
1013 * let's parse it and create in-core objects for the stripes
1015 buf->lb_buf = info->lti_ea_store;
1016 buf->lb_len = info->lti_ea_store_size;
1017 rc = lod_parse_striping(env, lo, buf);
1018 } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) {
1019 rc = lod_get_lmv_ea(env, lo);
1020 if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1))
1021 GOTO(out, rc = rc > 0 ? -EINVAL : rc);
1023 buf->lb_buf = info->lti_ea_store;
1024 buf->lb_len = info->lti_ea_store_size;
1025 if (rc == sizeof(struct lmv_mds_md_v1)) {
1026 rc = lod_load_lmv_shards(env, lo, buf, true);
1027 if (buf->lb_buf != info->lti_ea_store) {
1028 OBD_FREE_LARGE(info->lti_ea_store,
1029 info->lti_ea_store_size);
1030 info->lti_ea_store = buf->lb_buf;
1031 info->lti_ea_store_size = buf->lb_len;
1039 * there is LOV EA (striping information) in this object
1040 * let's parse it and create in-core objects for the stripes
1042 rc = lod_parse_dir_striping(env, lo, buf);
1049 * A generic function to initialize the stripe objects.
1051 * A protected version of lod_load_striping_locked() - load the striping
1052 * information from storage, parse that and instantiate LU objects to
1053 * represent the stripes. The LOD object \a lo supplies a pointer to the
1054 * next sub-object in the LU stack so we can lock it. Also use \a lo to
1055 * return an array of references to the newly instantiated objects.
1057 * \param[in] env execution environment for this thread
1058 * \param[in,out] lo LOD object, where striping is stored and
1059 * which gets an array of references
1061 * \retval 0 if parsing and object creation succeed
1062 * \retval negative error number on failure
1064 int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
1066 struct dt_object *next = dt_object_child(&lo->ldo_obj);
1069 /* currently this code is supposed to be called from declaration
1070 * phase only, thus the object is not expected to be locked by caller */
1071 dt_write_lock(env, next, 0);
1072 rc = lod_load_striping_locked(env, lo);
1073 dt_write_unlock(env, next);
1080 * Check the validity of all fields including the magic, stripe size,
1081 * stripe count, stripe offset and that the pool is present. Also check
1082 * that each target index points to an existing target. The additional
1083 * \a is_from_disk turns additional checks. In some cases zero fields
1084 * are allowed (like pattern=0).
1086 * \param[in] d LOD device
1087 * \param[in] buf buffer with LOV EA to verify
1088 * \param[in] is_from_disk 0 - from user, allow some fields to be 0
1089 * 1 - from disk, do not allow
1091 * \retval 0 if the striping is valid
1092 * \retval -EINVAL if striping is invalid
1094 int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
1097 struct lov_user_md_v1 *lum;
1098 struct lov_user_md_v3 *lum3;
1099 struct pool_desc *pool = NULL;
1103 __u16 stripe_offset;
1110 LASSERT(sizeof(*lum) < sizeof(*lum3));
1112 if (buf->lb_len < sizeof(*lum)) {
1113 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n",
1115 GOTO(out, rc = -EINVAL);
1118 magic = le32_to_cpu(lum->lmm_magic);
1119 if (magic != LOV_USER_MAGIC_V1 &&
1120 magic != LOV_USER_MAGIC_V3 &&
1121 magic != LOV_MAGIC_V1_DEF &&
1122 magic != LOV_MAGIC_V3_DEF) {
1123 CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
1124 GOTO(out, rc = -EINVAL);
1127 /* the user uses "0" for default stripe pattern normally. */
1128 if (!is_from_disk && lum->lmm_pattern == 0)
1129 lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1131 if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) {
1132 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
1133 le32_to_cpu(lum->lmm_pattern));
1134 GOTO(out, rc = -EINVAL);
1137 /* 64kB is the largest common page size we see (ia64), and matches the
1139 stripe_size = le32_to_cpu(lum->lmm_stripe_size);
1140 if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
1141 CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
1142 stripe_size, LOV_MIN_STRIPE_SIZE);
1143 GOTO(out, rc = -EINVAL);
1146 stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
1147 if (stripe_offset != LOV_OFFSET_DEFAULT) {
1148 /* if offset is not within valid range [0, osts_size) */
1149 if (stripe_offset >= d->lod_osts_size) {
1150 CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
1151 stripe_offset, d->lod_osts_size);
1152 GOTO(out, rc = -EINVAL);
1155 /* if lmm_stripe_offset is *not* in bitmap */
1156 if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
1157 CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
1159 GOTO(out, rc = -EINVAL);
1163 if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
1164 lum_size = offsetof(struct lov_user_md_v1,
1166 else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
1167 lum_size = offsetof(struct lov_user_md_v3,
1170 GOTO(out, rc = -EINVAL);
1172 stripe_count = le16_to_cpu(lum->lmm_stripe_count);
1173 if (buf->lb_len != lum_size) {
1174 CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with "
1175 "magic %#x and stripe_count %u\n",
1176 buf->lb_len, magic, stripe_count);
1177 GOTO(out, rc = -EINVAL);
1180 if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
1184 if (buf->lb_len < sizeof(*lum3)) {
1185 CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n",
1187 GOTO(out, rc = -EINVAL);
1190 /* In the function below, .hs_keycmp resolves to
1191 * pool_hashkey_keycmp() */
1192 /* coverity[overrun-buffer-val] */
1193 pool = lod_find_pool(d, lum3->lmm_pool_name);
1197 if (stripe_offset != LOV_OFFSET_DEFAULT) {
1198 rc = lod_check_index_in_pool(stripe_offset, pool);
1200 GOTO(out, rc = -EINVAL);
1203 if (is_from_disk && stripe_count > pool_tgt_count(pool)) {
1205 "stripe count %u > # OSTs %u in the pool\n",
1206 stripe_count, pool_tgt_count(pool));
1207 GOTO(out, rc = -EINVAL);
1212 lod_pool_putref(pool);
1217 void lod_fix_desc_stripe_size(__u64 *val)
1219 if (*val < LOV_MIN_STRIPE_SIZE) {
1221 LCONSOLE_INFO("Increasing default stripe size to "
1222 "minimum value %u\n",
1223 LOV_DESC_STRIPE_SIZE_DEFAULT);
1224 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
1225 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
1226 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
1227 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
1228 "multiple of %u)\n",
1229 *val, LOV_MIN_STRIPE_SIZE);
1233 void lod_fix_desc_stripe_count(__u32 *val)
1239 void lod_fix_desc_pattern(__u32 *val)
1241 /* from lov_setstripe */
1242 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
1243 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
1248 void lod_fix_desc_qos_maxage(__u32 *val)
1250 /* fix qos_maxage */
1252 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
1256 * Used to fix insane default striping.
1258 * \param[in] desc striping description
1260 void lod_fix_desc(struct lov_desc *desc)
1262 lod_fix_desc_stripe_size(&desc->ld_default_stripe_size);
1263 lod_fix_desc_stripe_count(&desc->ld_default_stripe_count);
1264 lod_fix_desc_pattern(&desc->ld_pattern);
1265 lod_fix_desc_qos_maxage(&desc->ld_qos_maxage);
1269 * Initialize the structures used to store pools and default striping.
1271 * \param[in] lod LOD device
1272 * \param[in] lcfg configuration structure storing default striping.
1274 * \retval 0 if initialization succeeds
1275 * \retval negative error number on failure
1277 int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
1279 struct obd_device *obd;
1280 struct lov_desc *desc;
1284 obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1285 LASSERT(obd != NULL);
1286 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1288 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1289 CERROR("LOD setup requires a descriptor\n");
1293 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
1295 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1296 CERROR("descriptor size wrong: %d > %d\n",
1297 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1301 if (desc->ld_magic != LOV_DESC_MAGIC) {
1302 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
1303 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
1304 obd->obd_name, desc);
1305 lustre_swab_lov_desc(desc);
1307 CERROR("%s: Bad lov desc magic: %#x\n",
1308 obd->obd_name, desc->ld_magic);
1315 desc->ld_active_tgt_count = 0;
1316 lod->lod_desc = *desc;
1318 lod->lod_sp_me = LUSTRE_SP_CLI;
1320 /* Set up allocation policy (QoS and RR) */
1321 INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
1322 init_rwsem(&lod->lod_qos.lq_rw_sem);
1323 lod->lod_qos.lq_dirty = 1;
1324 lod->lod_qos.lq_rr.lqr_dirty = 1;
1325 lod->lod_qos.lq_reset = 1;
1326 /* Default priority is toward free space balance */
1327 lod->lod_qos.lq_prio_free = 232;
1328 /* Default threshold for rr (roughly 17%) */
1329 lod->lod_qos.lq_threshold_rr = 43;
1331 /* Set up OST pool environment */
1332 lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
1333 HASH_POOLS_MAX_BITS,
1334 HASH_POOLS_BKT_BITS, 0,
1337 &pool_hash_operations,
1339 if (lod->lod_pools_hash_body == NULL)
1342 INIT_LIST_HEAD(&lod->lod_pool_list);
1343 lod->lod_pool_count = 0;
1344 rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
1347 lod_qos_rr_init(&lod->lod_qos.lq_rr);
1348 rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0);
1350 GOTO(out_pool_info, rc);
1355 lod_ost_pool_free(&lod->lod_pool_info);
1357 cfs_hash_putref(lod->lod_pools_hash_body);
1363 * Release the structures describing the pools.
1365 * \param[in] lod LOD device from which we release the structures
1369 int lod_pools_fini(struct lod_device *lod)
1371 struct obd_device *obd = lod2obd(lod);
1372 struct pool_desc *pool, *tmp;
1375 list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) {
1376 /* free pool structs */
1377 CDEBUG(D_INFO, "delete pool %p\n", pool);
1378 /* In the function below, .hs_keycmp resolves to
1379 * pool_hashkey_keycmp() */
1380 /* coverity[overrun-buffer-val] */
1381 lod_pool_del(obd, pool->pool_name);
1384 cfs_hash_putref(lod->lod_pools_hash_body);
1385 lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
1386 lod_ost_pool_free(&lod->lod_pool_info);