1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mds/mds_lov.c
38 * Lustre Metadata Server (mds) handling of striped file data
40 * Author: Peter Braam <braam@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #include <lustre_mds.h>
47 #include <lustre/lustre_idl.h>
48 #include <obd_class.h>
50 #include <lustre_lib.h>
51 #include <lustre_fsfilt.h>
52 #include <obd_cksum.h>
54 #include "mds_internal.h"
56 static void mds_lov_dump_objids(const char *label, struct obd_device *obd)
58 struct mds_obd *mds = &obd->u.mds;
61 CDEBUG(D_INFO, "dump from %s\n", label);
62 if (mds->mds_lov_page_dirty == NULL) {
63 CERROR("NULL bitmap!\n");
67 for(i = 0; i < mds->mds_lov_page_dirty->size / BITS_PER_LONG + 1; i++)
68 CDEBUG(D_INFO, "%u - %lx\n", i,
69 mds->mds_lov_page_dirty->data[i]);
71 if (mds->mds_lov_page_array == NULL) {
72 CERROR("not init page array!\n");
76 for(i = 0;i < MDS_LOV_OBJID_PAGES_COUNT; i++) {
77 obd_id *data = mds->mds_lov_page_array[i];
82 for(j=0; j < OBJID_PER_PAGE(); j++) {
85 CDEBUG(D_INFO,"objid page %u idx %u - "LPU64" \n",
93 int mds_lov_init_objids(struct obd_device *obd)
95 struct mds_obd *mds = &obd->u.mds;
96 int size = MDS_LOV_OBJID_PAGES_COUNT*sizeof(void *);
101 CLASSERT(((MDS_LOV_ALLOC_SIZE % sizeof(obd_id)) == 0));
103 mds->mds_lov_page_dirty = ALLOCATE_BITMAP(MDS_LOV_OBJID_PAGES_COUNT);
104 if (mds->mds_lov_page_dirty == NULL)
108 OBD_ALLOC(mds->mds_lov_page_array, size);
109 if (mds->mds_lov_page_array == NULL)
110 GOTO(err_free_bitmap, rc = -ENOMEM);
112 /* open and test the lov objd file */
113 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
116 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
117 GOTO(err_free, rc = PTR_ERR(file));
119 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
120 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
121 file->f_dentry->d_inode->i_mode);
122 GOTO(err_open, rc = -ENOENT);
124 mds->mds_lov_objid_filp = file;
128 if (filp_close((struct file *)file, 0))
129 CERROR("can't close %s after error\n", LOV_OBJID);
131 OBD_FREE(mds->mds_lov_page_array, size);
133 FREE_BITMAP(mds->mds_lov_page_dirty);
138 void mds_lov_destroy_objids(struct obd_device *obd)
140 struct mds_obd *mds = &obd->u.mds;
144 if (mds->mds_lov_page_array != NULL) {
145 for(i=0;i<MDS_LOV_OBJID_PAGES_COUNT;i++) {
146 obd_id *data = mds->mds_lov_page_array[i];
148 OBD_FREE(data, MDS_LOV_ALLOC_SIZE);
150 OBD_FREE(mds->mds_lov_page_array,
151 MDS_LOV_OBJID_PAGES_COUNT*sizeof(void *));
154 if (mds->mds_lov_objid_filp) {
155 rc = filp_close((struct file *)mds->mds_lov_objid_filp, NULL);
156 mds->mds_lov_objid_filp = NULL;
158 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
161 FREE_BITMAP(mds->mds_lov_page_dirty);
166 * currently exist two ways for know about ost count and max ost index.
167 * first - after ost is connected to mds and sync process finished
168 * second - get from lmm in recovery process, in case when mds not have configs,
169 * and ost isn't registered in mgs.
171 * \param mds pointer to mds structure
172 * \param index maxium ost index
174 * \retval -ENOMEM is not hame memory for new page
175 * \retval 0 is update passed
177 static int mds_lov_update_max_ost(struct mds_obd *mds, obd_id index)
179 __u32 page = index / OBJID_PER_PAGE();
180 __u32 off = index % OBJID_PER_PAGE();
181 obd_id *data = mds->mds_lov_page_array[page];
184 OBD_ALLOC(data, MDS_LOV_ALLOC_SIZE);
188 mds->mds_lov_page_array[page] = data;
191 if (index > mds->mds_lov_objid_max_index) {
192 mds->mds_lov_objid_lastpage = page;
193 mds->mds_lov_objid_lastidx = off;
194 mds->mds_lov_objid_max_index = index;
197 /* workaround - New target not in objids file; increase mdsize */
198 /* ld_tgt_count is used as the max index everywhere, despite its name. */
199 if (data[off] == 0) {
203 mds->mds_lov_objid_count++;
204 stripes = min_t(__u32, LOV_MAX_STRIPE_COUNT,
205 mds->mds_lov_objid_count);
207 mds->mds_max_mdsize = lov_mds_md_size(stripes, LOV_MAGIC_V3);
208 mds->mds_max_cookiesize = stripes * sizeof(struct llog_cookie);
210 CDEBUG(D_CONFIG, "updated max_mdsize/max_cookiesize for %d"
211 " stripes: %d/%d\n", stripes, mds->mds_max_mdsize,
212 mds->mds_max_cookiesize);
219 int mds_lov_prepare_objids(struct obd_device *obd, struct lov_mds_md *lmm)
221 struct lov_ost_data_v1 *data;
226 /* if we create file without objects - lmm is NULL */
230 switch (le32_to_cpu(lmm->lmm_magic)) {
232 count = le32_to_cpu(((struct lov_mds_md_v1*)lmm)->lmm_stripe_count);
233 data = &(((struct lov_mds_md_v1*)lmm)->lmm_objects[0]);
236 count = le32_to_cpu(((struct lov_mds_md_v3*)lmm)->lmm_stripe_count);
237 data = &(((struct lov_mds_md_v3*)lmm)->lmm_objects[0]);
240 CERROR("Unknow lmm type %X!\n", le32_to_cpu(lmm->lmm_magic));
245 mutex_down(&obd->obd_dev_sem);
246 for (j = 0; j < count; j++) {
247 __u32 i = le32_to_cpu(data[j].l_ost_idx);
248 if (mds_lov_update_max_ost(&obd->u.mds, i)) {
253 mutex_up(&obd->obd_dev_sem);
257 EXPORT_SYMBOL(mds_lov_prepare_objids);
260 * write llog orphan record about lost ost object,
261 * Special lsm is allocated with single stripe, caller should deallocated it
264 static int mds_log_lost_precreated(struct obd_device *obd,
265 struct lov_stripe_md **lsmp, int *stripes,
266 obd_id id, obd_count count, int idx)
268 struct lov_stripe_md *lsm = *lsmp;
273 rc = obd_alloc_memmd(obd->u.mds.mds_osc_exp, &lsm);
276 /* need only one stripe, save old value */
277 *stripes = lsm->lsm_stripe_count;
278 lsm->lsm_stripe_count = 1;
282 lsm->lsm_oinfo[0]->loi_id = id;
283 lsm->lsm_oinfo[0]->loi_gr = mdt_to_obd_objgrp(obd->u.mds.mds_id);
284 lsm->lsm_oinfo[0]->loi_ost_idx = idx;
286 rc = mds_log_op_orphan(obd, lsm, count);
290 void mds_lov_update_objids(struct obd_device *obd, struct lov_mds_md *lmm)
292 struct mds_obd *mds = &obd->u.mds;
294 struct lov_ost_data_v1 *obj;
295 struct lov_stripe_md *lsm = NULL;
300 /* if we create file without objects - lmm is NULL */
304 switch (le32_to_cpu(lmm->lmm_magic)) {
306 count = le32_to_cpu(((struct lov_mds_md_v1*)lmm)->lmm_stripe_count);
307 obj = ((struct lov_mds_md_v1*)lmm)->lmm_objects;
310 count = le32_to_cpu(((struct lov_mds_md_v3*)lmm)->lmm_stripe_count);
311 obj = ((struct lov_mds_md_v3*)lmm)->lmm_objects;
314 CERROR("Unknow lmm type %X !\n",
315 le32_to_cpu(lmm->lmm_magic));
319 for (j = 0; j < count; j++) {
320 __u32 i = le32_to_cpu(obj[j].l_ost_idx);
321 obd_id id = le64_to_cpu(obj[j].l_object_id);
322 __u32 page = i / OBJID_PER_PAGE();
323 __u32 idx = i % OBJID_PER_PAGE();
326 data = mds->mds_lov_page_array[page];
328 CDEBUG(D_INODE,"update last object for ost %u"
329 " - new "LPU64" old "LPU64"\n", i, id, data[idx]);
330 if (id > data[idx]) {
331 int lost = id - data[idx] - 1;
332 /* we might have lost precreated objects due to VBR */
333 if (lost > 0 && obd->obd_recovering) {
334 CDEBUG(D_HA, "Gap in objids is %u\n", lost);
335 if (!obd->obd_version_recov)
336 CERROR("Unexpected gap in objids\n");
337 /* lsm is allocated if NULL */
338 mds_log_lost_precreated(obd, &lsm, &stripes,
339 data[idx]+1, lost, i);
342 cfs_bitmap_set(mds->mds_lov_page_dirty, page);
346 /* restore stripes number */
347 lsm->lsm_stripe_count = stripes;
348 obd_free_memmd(mds->mds_osc_exp, &lsm);
353 EXPORT_SYMBOL(mds_lov_update_objids);
355 static int mds_lov_update_from_read(struct mds_obd *mds, obd_id *data,
361 for (i = 0; i < count; i++) {
365 mds->mds_lov_objid_count++;
368 stripes = min_t(__u32, LOV_MAX_STRIPE_COUNT,
369 mds->mds_lov_objid_count);
371 mds->mds_max_mdsize = lov_mds_md_size(stripes, LOV_MAGIC_V3);
372 mds->mds_max_cookiesize = stripes * sizeof(struct llog_cookie);
374 CDEBUG(D_CONFIG, "updated max_mdsize/max_cookiesize for %d stripes: "
375 "%d/%d\n", stripes, mds->mds_max_mdsize,
376 mds->mds_max_cookiesize);
382 static int mds_lov_read_objids(struct obd_device *obd)
384 struct mds_obd *mds = &obd->u.mds;
386 int i, rc = 0, count = 0, page = 0;
390 /* Read everything in the file, even if our current lov desc
391 has fewer targets. Old targets not in the lov descriptor
392 during mds setup may still have valid objids. */
393 size = i_size_read(mds->mds_lov_objid_filp->f_dentry->d_inode);
397 page = (size / (OBJID_PER_PAGE() * sizeof(obd_id))) + 1;
398 CDEBUG(D_INFO, "file size %lu pages %d\n", size, page);
399 for (i = 0; i < page; i++) {
401 loff_t off_old = off;
403 LASSERT(mds->mds_lov_page_array[i] == NULL);
404 OBD_ALLOC(mds->mds_lov_page_array[i], MDS_LOV_ALLOC_SIZE);
405 if (mds->mds_lov_page_array[i] == NULL)
406 GOTO(out, rc = -ENOMEM);
408 data = mds->mds_lov_page_array[i];
410 rc = fsfilt_read_record(obd, mds->mds_lov_objid_filp, data,
411 OBJID_PER_PAGE()*sizeof(obd_id), &off);
413 CERROR("Error reading objids %d\n", rc);
417 count += (off - off_old) / sizeof(obd_id);
418 if (mds_lov_update_from_read(mds, data, count)) {
419 CERROR("Can't update mds data\n");
420 GOTO(out, rc = -EIO);
426 mds->mds_lov_objid_lastpage = i;
427 mds->mds_lov_objid_lastidx = count % OBJID_PER_PAGE();
429 CDEBUG(D_INFO, "Read %u - %u %u objid\n", mds->mds_lov_objid_count,
430 mds->mds_lov_objid_lastpage, mds->mds_lov_objid_lastidx);
432 mds_lov_dump_objids("read",obd);
437 int mds_lov_write_objids(struct obd_device *obd)
439 struct mds_obd *mds = &obd->u.mds;
443 if (cfs_bitmap_check_empty(mds->mds_lov_page_dirty))
446 mds_lov_dump_objids("write", obd);
448 cfs_foreach_bit(mds->mds_lov_page_dirty, i) {
449 obd_id *data = mds->mds_lov_page_array[i];
450 unsigned int size = OBJID_PER_PAGE()*sizeof(obd_id);
451 loff_t off = i * size;
453 LASSERT(data != NULL);
455 if (!cfs_bitmap_test_and_clear(mds->mds_lov_page_dirty, i))
458 /* check for particaly filled last page */
459 if (i == mds->mds_lov_objid_lastpage)
460 size = (mds->mds_lov_objid_lastidx+1) * sizeof(obd_id);
462 CDEBUG(D_INFO, "write %lld - %u\n", off, size);
463 rc = fsfilt_write_record(obd, mds->mds_lov_objid_filp, data,
466 cfs_bitmap_set(mds->mds_lov_page_dirty, i);
475 EXPORT_SYMBOL(mds_lov_write_objids);
477 static int mds_lov_get_objid(struct obd_device * obd,
480 struct mds_obd *mds = &obd->u.mds;
487 page = idx / OBJID_PER_PAGE();
488 off = idx % OBJID_PER_PAGE();
489 data = mds->mds_lov_page_array[page];
491 /* We never read this lastid; ask the osc */
492 struct obd_id_info lastid;
493 __u32 size = sizeof(lastid);
496 lastid.data = &data[off];
497 rc = obd_get_info(mds->mds_osc_exp, sizeof(KEY_LAST_ID),
498 KEY_LAST_ID, &size, &lastid, NULL);
502 /* workaround for clean filter */
506 cfs_bitmap_set(mds->mds_lov_page_dirty, page);
508 CDEBUG(D_INFO, "idx "LPU64" - %p - %d/%d - "LPU64"\n",
509 idx, data, page, off, data[off]);
514 int mds_lov_clear_orphans(struct mds_obd *mds, struct obd_uuid *ost_uuid)
517 struct obdo oa = { 0 };
518 struct obd_trans_info oti = {0};
519 struct lov_stripe_md *empty_ea = NULL;
522 LASSERT(mds->mds_lov_page_array != NULL);
524 /* This create will in fact either create or destroy: If the OST is
525 * missing objects below this ID, they will be created. If it finds
526 * objects above this ID, they will be removed. */
527 memset(&oa, 0, sizeof(oa));
528 oa.o_flags = OBD_FL_DELORPHAN;
529 oa.o_gr = mdt_to_obd_objgrp(mds->mds_id);
530 oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
531 if (ost_uuid != NULL)
532 oti.oti_ost_uuid = ost_uuid;
534 rc = obd_create(mds->mds_osc_exp, &oa, &empty_ea, &oti);
540 static int mds_lov_set_one_nextid(struct obd_device *obd, __u32 idx, obd_id *id)
542 struct mds_obd *mds = &obd->u.mds;
544 struct obd_id_info info;
547 LASSERT(!obd->obd_recovering);
551 rc = obd_set_info_async(mds->mds_osc_exp, sizeof(KEY_NEXT_ID),
552 KEY_NEXT_ID, sizeof(info), &info, NULL);
554 CERROR ("%s: mds_lov_set_nextid failed (%d)\n",
560 /* Update the lov desc for a new size lov. */
561 static int mds_lov_update_desc(struct obd_device *obd, int idx,
562 struct obd_uuid *uuid, enum obd_notify_event ev)
564 struct mds_obd *mds = &obd->u.mds;
566 __u32 valsize = sizeof(mds->mds_lov_desc);
570 OBD_ALLOC(ld, sizeof(*ld));
574 rc = obd_get_info(mds->mds_osc_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
579 /* Don't change the mds_lov_desc until the objids size matches the
581 mds->mds_lov_desc = *ld;
582 CDEBUG(D_CONFIG, "updated lov_desc, tgt_count: %d - idx %d / uuid %s\n",
583 mds->mds_lov_desc.ld_tgt_count, idx, uuid->uuid);
585 mutex_down(&obd->obd_dev_sem);
586 rc = mds_lov_update_max_ost(mds, idx);
587 mutex_up(&obd->obd_dev_sem);
591 /* If we added a target we have to reconnect the llogs */
592 /* We only _need_ to do this at first add (idx), or the first time
593 after recovery. However, it should now be safe to call anytime. */
594 rc = obd_llog_init(obd, &obd->obd_olg, obd, &idx);
598 /*XXX this notifies the MDD until lov handling use old mds code */
599 if (obd->obd_upcall.onu_owner) {
600 LASSERT(obd->obd_upcall.onu_upcall != NULL);
601 rc = obd->obd_upcall.onu_upcall(obd, NULL, ev,
602 obd->obd_upcall.onu_owner,
603 &mds->mds_mount_count);
606 OBD_FREE(ld, sizeof(*ld));
610 /* Inform MDS about new/updated target */
611 static int mds_lov_update_mds(struct obd_device *obd,
612 struct obd_device *watched,
613 __u32 idx, enum obd_notify_event ev)
615 struct mds_obd *mds = &obd->u.mds;
623 /* Don't let anyone else mess with mds_lov_objids now */
624 rc = mds_lov_update_desc(obd, idx, &watched->u.cli.cl_target_uuid, ev);
628 CDEBUG(D_CONFIG, "idx=%d, recov=%d/%d, cnt=%d\n",
629 idx, obd->obd_recovering, obd->obd_async_recov,
630 mds->mds_lov_desc.ld_tgt_count);
632 /* idx is set as data from lov_notify. */
633 if (obd->obd_recovering)
636 if (idx >= mds->mds_lov_desc.ld_tgt_count) {
637 CERROR("index %d > count %d!\n", idx,
638 mds->mds_lov_desc.ld_tgt_count);
639 GOTO(out, rc = -EINVAL);
642 rc = mds_lov_get_objid(obd, idx);
646 page = idx / OBJID_PER_PAGE();
647 off = idx % OBJID_PER_PAGE();
648 data = mds->mds_lov_page_array[page];
650 /* We have read this lastid from disk; tell the osc.
651 Don't call this during recovery. */
652 rc = mds_lov_set_one_nextid(obd, idx, &data[off]);
654 CERROR("Failed to set next id, idx=%d rc=%d\n", idx,rc);
655 /* Don't abort the rest of the sync */
658 CDEBUG(D_CONFIG, "last object "LPU64" from OST %d rc=%d\n",
665 /* update the LOV-OSC knowledge of the last used object id's */
666 int mds_lov_connect(struct obd_device *obd, char * lov_name)
668 struct mds_obd *mds = &obd->u.mds;
669 struct obd_connect_data *data;
673 if (IS_ERR(mds->mds_osc_obd))
674 RETURN(PTR_ERR(mds->mds_osc_obd));
676 if (mds->mds_osc_obd)
679 mds->mds_osc_obd = class_name2obd(lov_name);
680 if (!mds->mds_osc_obd) {
681 CERROR("MDS cannot locate LOV %s\n", lov_name);
682 mds->mds_osc_obd = ERR_PTR(-ENOTCONN);
686 mutex_down(&obd->obd_dev_sem);
687 rc = mds_lov_read_objids(obd);
688 mutex_up(&obd->obd_dev_sem);
690 CERROR("cannot read %s: rc = %d\n", "lov_objids", rc);
694 rc = obd_register_observer(mds->mds_osc_obd, obd);
696 CERROR("MDS cannot register as observer of LOV %s (%d)\n",
701 /* try init too early */
702 rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
706 mds->mds_osc_obd->u.lov.lov_sp_me = LUSTRE_SP_MDT;
708 OBD_ALLOC(data, sizeof(*data));
710 GOTO(err_exit, rc = -ENOMEM);
712 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX |
713 OBD_CONNECT_REQPORTAL | OBD_CONNECT_QUOTA64 |
714 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_FID |
715 OBD_CONNECT_BRW_SIZE | OBD_CONNECT_CKSUM |
716 OBD_CONNECT_CHANGE_QS | OBD_CONNECT_AT |
717 OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN |
719 #ifdef HAVE_LRU_RESIZE_SUPPORT
720 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
722 data->ocd_version = LUSTRE_VERSION_CODE;
723 data->ocd_group = mdt_to_obd_objgrp(mds->mds_id);
724 /* send max bytes per rpc */
725 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
726 /* send the list of supported checksum types */
727 data->ocd_cksum_types = OBD_CKSUM_ALL;
728 /* NB: lov_connect() needs to fill in .ocd_index for each OST */
729 rc = obd_connect(NULL, &mds->mds_osc_exp, mds->mds_osc_obd, &obd->obd_uuid, data, NULL);
730 OBD_FREE(data, sizeof(*data));
732 CERROR("MDS cannot connect to LOV %s (%d)\n", lov_name, rc);
733 mds->mds_osc_obd = ERR_PTR(rc);
737 /* I want to see a callback happen when the OBD moves to a
738 * "For General Use" state, and that's when we'll call
739 * set_nextid(). The class driver can help us here, because
740 * it can use the obd_recovering flag to determine when the
741 * the OBD is full available. */
742 /* MDD device will care about that
743 if (!obd->obd_recovering)
744 rc = mds_postrecov(obd);
749 mds->mds_osc_exp = NULL;
750 mds->mds_osc_obd = ERR_PTR(rc);
754 int mds_lov_disconnect(struct obd_device *obd)
756 struct mds_obd *mds = &obd->u.mds;
760 if (!IS_ERR(mds->mds_osc_obd) && mds->mds_osc_exp != NULL) {
761 obd_register_observer(mds->mds_osc_obd, NULL);
763 /* The actual disconnect of the mds_lov will be called from
764 * class_disconnect_exports from mds_lov_clean. So we have to
765 * ensure that class_cleanup doesn't fail due to the extra ref
766 * we're holding now. The mechanism to do that already exists -
767 * the obd_force flag. We'll drop the final ref to the
768 * mds_osc_exp in mds_cleanup. */
769 mds->mds_osc_obd->obd_force = 1;
775 struct mds_lov_sync_info {
776 struct obd_device *mlsi_obd; /* the lov device to sync */
777 struct obd_device *mlsi_watched; /* target osc */
778 __u32 mlsi_index; /* index of target */
779 enum obd_notify_event mlsi_ev; /* event type */
782 static int mds_propagate_capa_keys(struct mds_obd *mds, struct obd_uuid *uuid)
784 struct mds_capa_info info = { .uuid = uuid };
785 struct lustre_capa_key *key;
790 if (!mds->mds_capa_keys)
793 for (i = 0; i < 2; i++) {
794 key = &mds->mds_capa_keys[i];
795 DEBUG_CAPA_KEY(D_SEC, key, "propagate");
798 rc = obd_set_info_async(mds->mds_osc_exp, sizeof(KEY_CAPA_KEY),
799 KEY_CAPA_KEY, sizeof(info), &info, NULL);
801 DEBUG_CAPA_KEY(D_ERROR, key,
802 "propagate failed (rc = %d) for", rc);
810 /* We only sync one osc at a time, so that we don't have to hold
811 any kind of lock on the whole mds_lov_desc, which may change
812 (grow) as a result of mds_lov_add_ost. This also avoids any
813 kind of mismatch between the lov_desc and the mds_lov_desc,
814 which are not in lock-step during lov_add_obd */
815 static int __mds_lov_synchronize(void *data)
817 struct mds_lov_sync_info *mlsi = data;
818 struct obd_device *obd = mlsi->mlsi_obd;
819 struct obd_device *watched = mlsi->mlsi_watched;
820 struct mds_obd *mds = &obd->u.mds;
821 struct obd_uuid *uuid;
822 __u32 idx = mlsi->mlsi_index;
823 enum obd_notify_event ev = mlsi->mlsi_ev;
824 struct mds_group_info mgi;
825 struct llog_ctxt *ctxt;
833 uuid = &watched->u.cli.cl_target_uuid;
836 down_read(&mds->mds_notify_lock);
837 if (obd->obd_stopping || obd->obd_fail)
838 GOTO(out, rc = -ENODEV);
840 OBD_RACE(OBD_FAIL_MDS_LOV_SYNC_RACE);
841 rc = mds_lov_update_mds(obd, watched, idx, ev);
843 CERROR("%s failed at update_mds: %d\n", obd_uuid2str(uuid), rc);
846 mgi.group = mdt_to_obd_objgrp(mds->mds_id);
849 rc = obd_set_info_async(mds->mds_osc_exp, sizeof(KEY_MDS_CONN),
850 KEY_MDS_CONN, sizeof(mgi), &mgi, NULL);
853 /* propagate capability keys */
854 rc = mds_propagate_capa_keys(mds, uuid);
858 ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
860 GOTO(out, rc = -ENODEV);
862 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT, 60);
863 rc = llog_connect(ctxt, NULL, NULL, uuid);
866 CERROR("%s failed at llog_origin_connect: %d\n",
867 obd_uuid2str(uuid), rc);
871 LCONSOLE_INFO("MDS %s: %s now active, resetting orphans\n",
872 obd->obd_name, obd_uuid2str(uuid));
873 rc = mds_lov_clear_orphans(mds, uuid);
875 CERROR("%s failed at mds_lov_clear_orphans: %d\n",
876 obd_uuid2str(uuid), rc);
880 #ifdef HAVE_QUOTA_SUPPORT
881 if (obd->obd_upcall.onu_owner) {
883 * This is a hack for mds_notify->mdd_notify. When the mds obd
884 * in mdd is removed, This hack should be removed.
886 LASSERT(obd->obd_upcall.onu_upcall != NULL);
887 rc = obd->obd_upcall.onu_upcall(obd, NULL, OBD_NOTIFY_QUOTA,
888 obd->obd_upcall.onu_owner,NULL);
893 up_read(&mds->mds_notify_lock);
895 /* Deactivate it for safety */
896 CERROR("%s sync failed %d, deactivating\n", obd_uuid2str(uuid),
898 if (!obd->obd_stopping && mds->mds_osc_obd &&
899 !mds->mds_osc_obd->obd_stopping && !watched->obd_stopping)
900 obd_notify(mds->mds_osc_obd, watched,
901 OBD_NOTIFY_INACTIVE, NULL);
904 class_decref(obd, "mds_lov_synchronize", obd);
908 int mds_lov_synchronize(void *data)
910 struct mds_lov_sync_info *mlsi = data;
913 snprintf(name, sizeof(name), "ll_sync_%02u", mlsi->mlsi_index);
914 cfs_daemonize_ctxt(name);
916 RETURN(__mds_lov_synchronize(data));
919 int mds_lov_start_synchronize(struct obd_device *obd,
920 struct obd_device *watched,
921 void *data, enum obd_notify_event ev)
923 struct mds_lov_sync_info *mlsi;
925 struct obd_uuid *uuid;
929 uuid = &watched->u.cli.cl_target_uuid;
931 OBD_ALLOC(mlsi, sizeof(*mlsi));
936 mlsi->mlsi_obd = obd;
937 mlsi->mlsi_watched = watched;
938 mlsi->mlsi_index = *(__u32 *)data;
941 /* Although class_export_get(obd->obd_self_export) would lock
942 the MDS in place, since it's only a self-export
943 it doesn't lock the LOV in place. The LOV can be disconnected
944 during MDS precleanup, leaving nothing for __mds_lov_synchronize.
945 Simply taking an export ref on the LOV doesn't help, because it's
946 still disconnected. Taking an obd reference insures that we don't
947 disconnect the LOV. This of course means a cleanup won't
948 finish for as long as the sync is blocking. */
949 class_incref(obd, "mds_lov_synchronize", obd);
951 if (ev != OBD_NOTIFY_SYNC) {
952 /* Synchronize in the background */
953 rc = cfs_kernel_thread(mds_lov_synchronize, mlsi,
954 CLONE_VM | CLONE_FILES);
956 CERROR("%s: error starting mds_lov_synchronize: %d\n",
958 class_decref(obd, "mds_lov_synchronize", obd);
960 CDEBUG(D_HA, "%s: mds_lov_synchronize idx=%d "
961 "thread=%d\n", obd->obd_name,
962 mlsi->mlsi_index, rc);
966 rc = __mds_lov_synchronize((void *)mlsi);
972 int mds_notify(struct obd_device *obd, struct obd_device *watched,
973 enum obd_notify_event ev, void *data)
978 CDEBUG(D_CONFIG, "notify %s ev=%d\n", watched->obd_name, ev);
981 /* We only handle these: */
982 case OBD_NOTIFY_ACTIVE:
983 /* lov want one or more _active_ targets for work */
984 /* activate event should be pass lov idx as argument */
985 case OBD_NOTIFY_SYNC:
986 case OBD_NOTIFY_SYNC_NONBLOCK:
987 /* sync event should be pass lov idx as argument */
993 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME) != 0) {
994 CERROR("unexpected notification of %s %s!\n",
995 watched->obd_type->typ_name, watched->obd_name);
999 if (obd->obd_recovering) {
1000 CWARN("MDS %s: in recovery, not resetting orphans on %s\n",
1002 obd_uuid2str(&watched->u.cli.cl_target_uuid));
1003 /* We still have to fix the lov descriptor for ost's added
1004 after the mdt in the config log. They didn't make it into
1006 rc = mds_lov_update_desc(obd, *(__u32 *)data,
1007 &watched->u.cli.cl_target_uuid, ev);
1009 rc = mds_lov_start_synchronize(obd, watched, data, ev);