1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mds) handling of striped file data
7 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
8 * Author: Peter Braam <braam@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org.
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 # define EXPORT_SYMTAB
29 #define DEBUG_SUBSYSTEM S_MDS
31 #include <linux/module.h>
32 #include <linux/lustre_mds.h>
33 #include <linux/obd_ost.h>
34 #include <linux/lustre_idl.h>
35 #include <linux/obd_class.h>
36 #include <linux/obd_lov.h>
37 #include <linux/lustre_lib.h>
38 #include <linux/lustre_fsfilt.h>
40 #include "mds_internal.h"
42 void le_lov_desc_to_cpu (struct lov_desc *ld)
44 ld->ld_tgt_count = le32_to_cpu (ld->ld_tgt_count);
45 ld->ld_default_stripe_count = le32_to_cpu (ld->ld_default_stripe_count);
46 ld->ld_default_stripe_size = le32_to_cpu (ld->ld_default_stripe_size);
47 ld->ld_pattern = le32_to_cpu (ld->ld_pattern);
50 void cpu_to_le_lov_desc (struct lov_desc *ld)
52 ld->ld_tgt_count = cpu_to_le32 (ld->ld_tgt_count);
53 ld->ld_default_stripe_count = cpu_to_le32 (ld->ld_default_stripe_count);
54 ld->ld_default_stripe_size = cpu_to_le32 (ld->ld_default_stripe_size);
55 ld->ld_pattern = cpu_to_le32 (ld->ld_pattern);
58 void mds_lov_update_objids(struct obd_device *obd, obd_id *ids)
60 struct mds_obd *mds = &obd->u.mds;
64 spin_lock(&mds->mds_lov_lock);
65 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
66 if (ids[i] > (mds->mds_lov_objids)[i])
67 (mds->mds_lov_objids)[i] = ids[i];
68 spin_unlock(&mds->mds_lov_lock);
72 static int mds_lov_read_objids(struct obd_device *obd)
74 struct mds_obd *mds = &obd->u.mds;
77 int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids);
80 if (mds->mds_lov_objids != NULL)
86 mds->mds_lov_objids = ids;
88 if (mds->mds_lov_objid_filp->f_dentry->d_inode->i_size == 0)
90 rc = fsfilt_read_record(obd, mds->mds_lov_objid_filp, ids, size, &off);
92 CERROR("Error reading objids %d\n", rc);
94 mds->mds_lov_objids_valid = 1;
98 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
99 CDEBUG(D_INFO, "read last object "LPU64" for idx %d\n",
100 mds->mds_lov_objids[i], i);
105 int mds_lov_write_objids(struct obd_device *obd)
107 struct mds_obd *mds = &obd->u.mds;
109 int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(obd_id);
112 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
113 CDEBUG(D_INFO, "writing last object "LPU64" for idx %d\n",
114 mds->mds_lov_objids[i], i);
116 rc = fsfilt_write_record(obd, mds->mds_lov_objid_filp,
117 mds->mds_lov_objids, size, &off, 0);
121 int mds_lov_clearorphans(struct mds_obd *mds, struct obd_uuid *ost_uuid)
124 struct obdo *oa = NULL;
125 struct obd_trans_info oti = {0};
126 struct lov_stripe_md *empty_ea = NULL;
129 LASSERT(mds->mds_lov_objids != NULL);
132 * this create will in fact either create or destroy: If the OST is
133 * missing objects below this ID, they will be created. If it finds
134 * objects above this ID, they will be removed.
140 memset(oa, 0, sizeof(*oa));
141 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
142 oa->o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
143 oa->o_flags = OBD_FL_DELORPHAN;
145 if (ost_uuid != NULL) {
146 memcpy(&oa->o_inline, ost_uuid, sizeof(*ost_uuid));
147 oa->o_valid |= OBD_MD_FLINLINE;
149 rc = obd_create(mds->mds_lov_exp, oa, &empty_ea, &oti);
154 /* update the LOV-OSC knowledge of the last used object id's */
155 int mds_lov_set_nextid(struct obd_device *obd)
157 struct mds_obd *mds = &obd->u.mds;
161 LASSERT(!obd->obd_recovering);
163 LASSERT(mds->mds_lov_objids != NULL);
165 rc = obd_set_info(mds->mds_lov_exp, strlen("next_id"), "next_id",
166 mds->mds_lov_desc.ld_tgt_count, mds->mds_lov_objids);
170 /* tell the LOV-OSC by how much to pre-create */
171 int mds_lov_set_growth(struct mds_obd *mds, int count)
176 rc = obd_set_info(mds->mds_lov_exp, strlen("growth_count"),
177 "growth_count", sizeof(count), &count);
182 static int mds_lov_update_desc(struct obd_device *obd, struct obd_export *lov)
184 struct mds_obd *mds = &obd->u.mds;
185 int valsize = sizeof(mds->mds_lov_desc), rc, i;
186 int old_count = mds->mds_lov_desc.ld_tgt_count;
189 rc = obd_get_info(lov, strlen("lovdesc") + 1, "lovdesc", &valsize,
194 /* The size of the LOV target table may have increased. */
195 if (old_count >= mds->mds_lov_desc.ld_tgt_count) {
199 size = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids);
200 OBD_ALLOC(ids, size);
204 memset(ids, 0, size);
206 if (mds->mds_lov_objids != NULL) {
207 int oldsize = old_count * sizeof(*ids);
209 memcpy(ids, mds->mds_lov_objids, oldsize);
210 OBD_FREE(mds->mds_lov_objids, oldsize);
212 mds->mds_lov_objids = ids;
215 i = lov_mds_md_size(mds->mds_lov_desc.ld_tgt_count);
216 if (i > mds->mds_max_mdsize)
217 mds->mds_max_mdsize = i;
218 mds->mds_max_cookiesize = mds->mds_lov_desc.ld_tgt_count *
219 sizeof(struct llog_cookie);
220 mds->mds_has_lov_desc = 1;
224 int mds_lov_connect(struct obd_device *obd, char * lov_name)
226 struct mds_obd *mds = &obd->u.mds;
227 struct lustre_handle conn = {0,};
231 if (IS_ERR(mds->mds_lov_obd))
232 RETURN(PTR_ERR(mds->mds_lov_obd));
234 if (mds->mds_lov_obd)
237 spin_lock_init(&mds->mds_lov_lock);
238 mds->mds_lov_obd = class_name2obd(lov_name);
239 if (!mds->mds_lov_obd) {
240 CERROR("MDS cannot locate LOV %s\n", lov_name);
241 mds->mds_lov_obd = ERR_PTR(-ENOTCONN);
245 CDEBUG(D_HA, "obd: %s osc: %s lov_name: %s\n",
246 obd->obd_name, mds->mds_lov_obd->obd_name, lov_name);
248 rc = obd_connect(&conn, mds->mds_lov_obd, &obd->obd_uuid,
249 mds->mds_num + FILTER_GROUP_FIRST_MDS);
251 CERROR("MDS cannot connect to LOV %s (%d)\n", lov_name, rc);
252 mds->mds_lov_obd = ERR_PTR(rc);
255 mds->mds_lov_exp = class_conn2export(&conn);
257 rc = obd_register_observer(mds->mds_lov_obd, obd);
259 CERROR("MDS cannot register as observer of LOV %s (%d)\n",
261 GOTO(err_discon, rc);
264 rc = mds_lov_update_desc(obd, mds->mds_lov_exp);
268 rc = mds_lov_read_objids(obd);
270 CERROR("cannot read %s: rc = %d\n", "lov_objids", rc);
274 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs,
275 mds->mds_lov_desc.ld_tgt_count, CATLIST);
277 CERROR("failed to initialize catalog %d\n", rc);
281 /* If we're mounting this code for the first time on an existing FS,
282 * we need to populate the objids array from the real OST values */
283 if (!mds->mds_lov_objids_valid) {
284 int size = sizeof(obd_id) * mds->mds_lov_desc.ld_tgt_count;
285 rc = obd_get_info(mds->mds_lov_exp, strlen("last_id"),
286 "last_id", &size, mds->mds_lov_objids);
288 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
289 CWARN("got last object "LPU64" from OST %d\n",
290 mds->mds_lov_objids[i], i);
291 mds->mds_lov_objids_valid = 1;
292 rc = mds_lov_write_objids(obd);
294 CERROR("got last objids from OSTs, but error "
295 "writing objids file: %d\n", rc);
299 /* I want to see a callback happen when the OBD moves to a
300 * "For General Use" state, and that's when we'll call
301 * set_nextid(). The class driver can help us here, because
302 * it can use the obd_recovering flag to determine when the
303 * the OBD is full available. */
304 if (!obd->obd_recovering)
305 rc = mds_postrecov(obd);
309 obd_register_observer(mds->mds_lov_obd, NULL);
311 obd_disconnect(mds->mds_lov_exp, 0);
312 mds->mds_lov_exp = NULL;
313 mds->mds_lov_obd = ERR_PTR(rc);
317 int mds_lov_disconnect(struct obd_device *obd, int flags)
319 struct mds_obd *mds = &obd->u.mds;
323 if (!IS_ERR(mds->mds_lov_obd) && mds->mds_lov_exp != NULL) {
324 /* cleanup all llogging subsystems */
325 rc = obd_llog_finish(obd, &obd->obd_llogs,
326 mds->mds_lov_desc.ld_tgt_count);
328 CERROR("failed to cleanup llogging subsystems\n");
330 obd_register_observer(mds->mds_lov_obd, NULL);
332 rc = obd_disconnect(mds->mds_lov_exp, flags);
333 /* if obd_disconnect fails (probably because the
334 * export was disconnected by class_disconnect_exports)
335 * then we just need to drop our ref. */
337 class_export_put(mds->mds_lov_exp);
338 mds->mds_lov_exp = NULL;
339 mds->mds_lov_obd = NULL;
345 int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
346 void *karg, void *uarg)
348 static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" };
349 struct obd_device *obd = exp->exp_obd;
350 struct mds_obd *mds = &obd->u.mds;
351 struct obd_ioctl_data *data = karg;
352 struct lvfs_run_ctxt saved;
357 case OBD_IOC_RECORD: {
358 char *name = data->ioc_inlbuf1;
359 if (mds->mds_cfg_llh)
362 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
363 rc = llog_open(llog_get_context(&obd->obd_llogs,
364 LLOG_CONFIG_ORIG_CTXT),
365 &mds->mds_cfg_llh, NULL, name,
368 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
371 mds->mds_cfg_llh = NULL;
372 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
377 case OBD_IOC_ENDRECORD: {
378 if (!mds->mds_cfg_llh)
381 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
382 rc = llog_close(mds->mds_cfg_llh);
383 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
385 mds->mds_cfg_llh = NULL;
389 case OBD_IOC_CLEAR_LOG: {
390 char *name = data->ioc_inlbuf1;
391 if (mds->mds_cfg_llh)
394 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
395 rc = llog_open(llog_get_context(&obd->obd_llogs,
396 LLOG_CONFIG_ORIG_CTXT),
397 &mds->mds_cfg_llh, NULL, name,
400 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
403 rc = llog_destroy(mds->mds_cfg_llh);
404 llog_free_handle(mds->mds_cfg_llh);
406 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
408 mds->mds_cfg_llh = NULL;
412 case OBD_IOC_DORECORD: {
414 struct llog_rec_hdr rec;
415 if (!mds->mds_cfg_llh)
418 /* XXX - this probably should be a parameter to this ioctl.
419 * For now, just use llh_max_transno for expediency. */
420 rec.lrh_len = llog_data_len(data->ioc_plen1);
422 if (data->ioc_type == LUSTRE_CFG_TYPE) {
423 rec.lrh_type = OBD_CFG_REC;
424 } else if (data->ioc_type == PORTALS_CFG_TYPE) {
425 rec.lrh_type = PTL_CFG_REC;
427 CERROR("unknown cfg record type:%d \n", data->ioc_type);
431 OBD_ALLOC(cfg_buf, data->ioc_plen1);
434 rc = copy_from_user(cfg_buf, data->ioc_pbuf1, data->ioc_plen1);
436 OBD_FREE(cfg_buf, data->ioc_plen1);
440 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
441 rc = llog_write_rec(mds->mds_cfg_llh, &rec, NULL, 0,
443 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
445 OBD_FREE(cfg_buf, data->ioc_plen1);
448 case OBD_IOC_PARSE: {
449 struct llog_ctxt *ctxt =
450 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
451 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
452 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
453 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
459 case OBD_IOC_DUMP_LOG: {
460 struct llog_ctxt *ctxt =
461 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
462 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
463 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
464 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
468 case OBD_IOC_SET_READONLY: {
470 struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode;
471 BDEVNAME_DECLARE_STORAGE(tmp);
472 CERROR("setting device %s read-only\n",
473 ll_bdevname(obd->u.mds.mds_sb, tmp));
475 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
477 rc = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 1);
479 dev_set_rdonly(ll_sbdev(obd->u.mds.mds_sb), 2);
483 case OBD_IOC_CATLOGLIST: {
484 int count = mds->mds_lov_desc.ld_tgt_count;
485 rc = llog_catalog_list(obd, count, data);
489 case OBD_IOC_LLOG_CHECK:
490 case OBD_IOC_LLOG_CANCEL:
491 case OBD_IOC_LLOG_REMOVE: {
492 struct llog_ctxt *ctxt =
493 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
497 obd_llog_finish(obd, &obd->obd_llogs,
498 mds->mds_lov_desc.ld_tgt_count);
499 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
500 rc = llog_ioctl(ctxt, cmd, data);
501 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
502 obd_llog_cat_initialize(obd, &obd->obd_llogs,
503 mds->mds_lov_desc.ld_tgt_count,
505 group = FILTER_GROUP_FIRST_MDS + mds->mds_num;
506 valsize = sizeof(group);
507 rc2 = obd_set_info(mds->mds_lov_exp, strlen("mds_conn"),
508 "mds_conn", valsize, &group);
513 case OBD_IOC_LLOG_INFO:
514 case OBD_IOC_LLOG_PRINT: {
515 struct llog_ctxt *ctxt =
516 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
518 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
519 rc = llog_ioctl(ctxt, cmd, data);
520 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
525 case OBD_IOC_ABORT_RECOVERY:
526 target_stop_recovery_thread(obd);
528 case OBD_IOC_ROOT_SQUASH: {
529 __u32 *p = (__u32 *) data->ioc_inlbuf1;
531 if (data->ioc_inllen1 !=
532 (sizeof(__u32) * 4 + sizeof(ptl_nid_t)))
535 if (*p == 0) { /* get */
537 *p++ = mds->mds_squash_uid;
538 *p++ = mds->mds_squash_gid;
539 *((ptl_nid_t*) p) = mds->mds_nosquash_nid;
542 mds->mds_squash_uid = *p++;
543 mds->mds_squash_gid = *p++;
544 mds->mds_nosquash_nid = *((ptl_nid_t*) p);
545 CWARN("MDS: squash root to %d:%d, except nid 0x%llx\n",
546 mds->mds_squash_uid, mds->mds_squash_gid,
547 mds->mds_nosquash_nid);
558 struct mds_lov_sync_info {
559 struct obd_device *mlsi_obd; /* the mds to sync */
560 struct obd_device *mlsi_watched; /* new lov target */
561 int mlsi_index; /* index into mds_lov_objids */
564 int mds_lov_synchronize(void *data)
566 struct mds_lov_sync_info *mlsi = data;
567 struct llog_ctxt *ctxt;
568 struct obd_device *obd;
569 struct obd_device *watched;
571 struct obd_uuid *uuid;
580 char name[32] = "CATLIST";
586 SIGNAL_MASK_LOCK(current, flags);
587 sigfillset(¤t->blocked);
589 SIGNAL_MASK_UNLOCK(current, flags);
593 obd = mlsi->mlsi_obd;
594 watched = mlsi->mlsi_watched;
595 index = mlsi->mlsi_index;
597 LASSERT(obd != NULL);
598 LASSERT(watched != NULL);
600 OBD_FREE(mlsi, sizeof(*mlsi));
603 uuid = &watched->u.cli.cl_import->imp_target_uuid;
605 group = FILTER_GROUP_FIRST_MDS + mds->mds_num;
606 rc = obd_set_info(watched->obd_self_export, strlen("mds_conn"),
607 "mds_conn", sizeof(group), &group);
611 old_count = mds->mds_lov_desc.ld_tgt_count;
613 rc = mds_lov_update_desc(obd, mds->mds_lov_exp);
617 count = mds->mds_lov_desc.ld_tgt_count;
618 LASSERT(count >= old_count);
620 vallen = sizeof(vals[1]);
621 rc = obd_get_info(watched->obd_self_export, strlen("last_id"),
622 "last_id", &vallen, &vals[1]);
627 rc = mds_lov_set_info(obd->obd_self_export, strlen("next_id"),
632 obd_llog_finish(obd, &obd->obd_llogs, old_count);
633 obd_llog_cat_initialize(obd, &obd->obd_llogs, count, name);
635 ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT);
636 LASSERT(ctxt != NULL);
638 rc = llog_connect(ctxt, count, NULL, NULL, uuid);
640 CERROR("%s: failed at llog_origin_connect: %d\n",
645 CWARN("MDS %s: %s now active, resetting orphans\n",
646 obd->obd_name, uuid->uuid);
648 rc = mds_lov_clearorphans(&obd->u.mds, uuid);
650 CERROR("%s: failed at mds_lov_clearorphans: %d\n",
658 int mds_lov_start_synchronize(struct obd_device *obd,
659 struct obd_device *watched, void *data)
661 struct mds_lov_sync_info *mlsi;
666 OBD_ALLOC(mlsi, sizeof(*mlsi));
670 mlsi->mlsi_obd = obd;
671 mlsi->mlsi_watched = watched;
672 mlsi->mlsi_index = (int)data;
674 rc = kernel_thread(mds_lov_synchronize, mlsi, CLONE_VM | CLONE_FILES);
676 CERROR("%s: error starting mds_lov_synchronize: %d\n",
679 CDEBUG(D_HA, "%s: mds_lov_synchronize thread: %d\n",
687 int mds_notify(struct obd_device *obd, struct obd_device *watched,
688 int active, void *data)
696 if (!strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME))
699 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
700 CERROR("unexpected notification of %s %s!\n",
701 watched->obd_type->typ_name, watched->obd_name);
705 if (obd->obd_recovering) {
706 struct obd_uuid *uuid;
707 uuid = &watched->u.cli.cl_import->imp_target_uuid;
709 CWARN("MDS %s: in recovery, not resetting orphans on %s\n",
710 obd->obd_name, uuid->uuid);
712 rc = mds_lov_start_synchronize(obd, watched, data);
717 int mds_lov_set_info(struct obd_export *exp, obd_count keylen,
718 void *key, obd_count vallen, void *val)
720 struct obd_device *obd = class_exp2obd(exp);
721 struct mds_obd *mds = &obd->u.mds;
724 #define KEY_IS(str) \
725 (keylen == strlen(str) && memcmp(key, str, keylen) == 0)
727 if (KEY_IS("next_id")) {
728 obd_id *id = (obd_id *)val;
731 /* XXX - this really should be vallen != (2 * sizeof(*id)) *
732 * Just following the precedent set by lov_set_info. */
737 if ((idx != *id) || (idx >= mds->mds_lov_desc.ld_tgt_count))
740 CDEBUG(D_CONFIG, "idx: %d id: %llu\n", idx, *(id + 1));
742 mds->mds_lov_objids[idx] = *++id;
743 CDEBUG(D_CONFIG, "objid: %d: %lld\n", idx, *id);
744 /* XXX - should we be writing this out here ? */
745 RETURN(mds_lov_write_objids(obd));
751 int mds_lov_update_config(struct obd_device *obd, int clean)
753 struct mds_obd *mds = &obd->u.mds;
754 struct lvfs_run_ctxt saved;
755 struct config_llog_instance cfg;
756 struct llog_ctxt *ctxt;
757 char *profile = mds->mds_profile, *name;
758 int rc, version, namelen;
764 cfg.cfg_instance = NULL;
765 cfg.cfg_uuid = mds->mds_lov_uuid;
767 namelen = strlen(profile) + 20; /* -clean-######### */
768 OBD_ALLOC(name, namelen);
773 version = mds->mds_config_version - 1;
774 sprintf(name, "%s-clean-%d", profile, version);
776 version = mds->mds_config_version + 1;
777 sprintf(name, "%s-%d", profile, version);
780 CWARN("Applying configuration log %s\n", name);
782 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
783 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
784 rc = class_config_process_llog(ctxt, name, &cfg);
785 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
787 mds->mds_config_version = version;
788 CWARN("Finished applying configuration log %s: %d\n", name, rc);
790 OBD_FREE(name, namelen);
794 /* Convert the on-disk LOV EA structre.
795 * We always try to convert from an old LOV EA format to the common in-memory
796 * (lsm) format (obd_unpackmd() understands the old on-disk (lmm) format) and
797 * then convert back to the new on-disk format and save it back to disk
798 * (obd_packmd() only ever saves to the new on-disk format) so we don't have
799 * to convert it each time this inode is accessed.
801 * This function is a bit interesting in the error handling. We can safely
802 * ship the old lmm to the client in case of failure, since it uses the same
803 * obd_unpackmd() code and can do the conversion if the MDS fails for some
804 * reason. We will not delete the old lmm data until we have written the
805 * new format lmm data in fsfilt_set_md(). */
806 int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode,
807 struct lov_mds_md *lmm, int lmm_size)
809 struct lov_stripe_md *lsm = NULL;
814 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC)
817 CDEBUG(D_INODE, "converting LOV EA on %lu/%u from V0 to V1\n",
818 inode->i_ino, inode->i_generation);
819 rc = obd_unpackmd(obd->u.mds.mds_lov_exp, &lsm, lmm, lmm_size);
823 rc = obd_packmd(obd->u.mds.mds_lov_exp, &lmm, lsm);
828 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
829 if (IS_ERR(handle)) {
830 rc = PTR_ERR(handle);
834 rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size);
836 err = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 0);
838 rc = err ? err : lmm_size;
841 obd_free_memmd(obd->u.mds.mds_lov_exp, &lsm);
846 /* Must be called with i_sem held */
847 int mds_revalidate_lov_ea(struct obd_device *obd, struct inode *inode,
848 struct lustre_msg *msg, int offset)
850 struct mds_obd *mds = &obd->u.mds;
851 struct obd_export *lov_exp = mds->mds_lov_exp;
852 struct lov_mds_md *lmm= NULL;
853 struct lov_stripe_md *lsm = NULL;
854 struct obdo *oa = NULL;
855 struct obd_trans_info oti = {0};
857 int lmm_size = 0, lsm_size = 0, err, rc;
861 LASSERT(down_trylock(&inode->i_sem) != 0);
863 /* XXX - add way to know if EA is already up to date & return
864 * without doing anything. Easy to do since we get notified of
867 lmm = lustre_msg_buf(msg, offset, 0);
869 CDEBUG(D_INFO, "no space reserved for inode %lu MD\n",
873 lmm_size = msg->buflens[offset];
875 rc = obd_unpackmd(lov_exp, &lsm, lmm, lmm_size);
881 LASSERT(lsm->lsm_magic == LOV_MAGIC);
885 GOTO(out_lsm, rc = -ENOMEM);
886 oa->o_mode = S_IFREG | 0600;
887 oa->o_id = inode->i_ino;
888 oa->o_generation = inode->i_generation;
889 oa->o_uid = 0; /* must have 0 uid / gid on OST */
891 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
893 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLTYPE |
894 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID |
896 valid = OBD_MD_FLTYPE | OBD_MD_FLATIME | OBD_MD_FLMTIME |
898 obdo_from_inode(oa, inode, valid);
900 rc = obd_revalidate_md(lov_exp, oa, lsm, &oti);
904 CERROR("Error validating LOV EA on %lu/%u: %d\n",
905 inode->i_ino, inode->i_generation, rc);
909 rc = obd_packmd(lov_exp, &lmm, lsm);
914 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
915 if (IS_ERR(handle)) {
916 rc = PTR_ERR(handle);
920 rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size);
921 err = fsfilt_commit(obd, inode->i_sb, inode, handle, 0);
929 obd_free_memmd(lov_exp, &lsm);