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.
38 * Author: Wang Di <wangdi@clusterfs.com>
42 # define EXPORT_SYMTAB
44 #define DEBUG_SUBSYSTEM S_LOV
47 #include <asm/div64.h>
48 #include <libcfs/libcfs.h>
50 #include <liblustre.h>
53 #include <obd_class.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_log.h>
58 #include "lov_internal.h"
60 struct lovea_unpack_args {
61 struct lov_stripe_md *lsm;
65 static int lsm_lmm_verify_common(struct lov_mds_md *lmm, int lmm_bytes,
69 if (stripe_count == 0 || stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
70 CERROR("bad stripe count %d\n", stripe_count);
71 lov_dump_lmm(D_WARNING, lmm);
75 if (lmm->lmm_object_id == 0) {
76 CERROR("zero object id\n");
77 lov_dump_lmm(D_WARNING, lmm);
81 if (lmm->lmm_pattern != cpu_to_le32(LOV_PATTERN_RAID0)) {
82 CERROR("bad striping pattern\n");
83 lov_dump_lmm(D_WARNING, lmm);
87 if (lmm->lmm_stripe_size == 0 ||
88 (stripe_count != -1 &&
89 (__u64)le32_to_cpu(lmm->lmm_stripe_size)*stripe_count >
91 CERROR("bad stripe size %u\n",
92 le32_to_cpu(lmm->lmm_stripe_size));
93 lov_dump_lmm(D_WARNING, lmm);
99 struct lov_stripe_md *lsm_alloc_plain(int stripe_count, int *size)
101 struct lov_stripe_md *lsm;
102 int i, oinfo_ptrs_size;
103 struct lov_oinfo *loi;
105 LASSERT(stripe_count > 0);
107 oinfo_ptrs_size = sizeof(struct lov_oinfo *) * stripe_count;
108 *size = sizeof(struct lov_stripe_md) + oinfo_ptrs_size;
110 OBD_ALLOC(lsm, *size);
114 for (i = 0; i < stripe_count; i++) {
115 OBD_SLAB_ALLOC(loi, lov_oinfo_slab, CFS_ALLOC_IO, sizeof(*loi));
118 lsm->lsm_oinfo[i] = loi;
120 lsm->lsm_stripe_count = stripe_count;
121 lsm->lsm_pool_name[0] = '\0';
126 OBD_SLAB_FREE(lsm->lsm_oinfo[i], lov_oinfo_slab, sizeof(*loi));
127 OBD_FREE(lsm, *size);
131 void lsm_free_plain(struct lov_stripe_md *lsm)
133 int stripe_count = lsm->lsm_stripe_count;
136 for (i = 0; i < stripe_count; i++)
137 OBD_SLAB_FREE(lsm->lsm_oinfo[i], lov_oinfo_slab,
138 sizeof(struct lov_oinfo));
139 OBD_FREE(lsm, sizeof(struct lov_stripe_md) +
140 stripe_count * sizeof(struct lov_oinfo *));
143 static void lsm_unpackmd_common(struct lov_stripe_md *lsm,
144 struct lov_mds_md *lmm)
147 * This supposes lov_mds_md_v1/v3 first fields are
150 lsm->lsm_object_id = le64_to_cpu(lmm->lmm_object_id);
151 lsm->lsm_object_gr = le64_to_cpu(lmm->lmm_object_gr);
152 lsm->lsm_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
153 lsm->lsm_pattern = le32_to_cpu(lmm->lmm_pattern);
154 lsm->lsm_pool_name[0] = '\0';
158 lsm_stripe_by_index_plain(struct lov_stripe_md *lsm, int *stripeno,
159 obd_off *lov_off, unsigned long *swidth)
162 *swidth = (unsigned long)lsm->lsm_stripe_size * lsm->lsm_stripe_count;
166 lsm_stripe_by_offset_plain(struct lov_stripe_md *lsm, int *stripeno,
167 obd_off *lov_off, unsigned long *swidth)
170 *swidth = (unsigned long)lsm->lsm_stripe_size * lsm->lsm_stripe_count;
174 lsm_stripe_offset_by_index_plain(struct lov_stripe_md *lsm,
181 lsm_stripe_offset_by_offset_plain(struct lov_stripe_md *lsm,
188 lsm_stripe_index_by_offset_plain(struct lov_stripe_md *lsm,
194 static int lsm_revalidate_plain(struct lov_stripe_md *lsm,
195 struct obd_device *obd)
200 static int lsm_destroy_plain(struct lov_stripe_md *lsm, struct obdo *oa,
201 struct obd_export *md_exp)
206 static int lsm_lmm_verify_v1(struct lov_mds_md_v1 *lmm, int lmm_bytes,
209 if (lmm_bytes < sizeof(*lmm)) {
210 CERROR("lov_mds_md_v1 too small: %d, need at least %d\n",
211 lmm_bytes, (int)sizeof(*lmm));
215 *stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
217 if (lmm_bytes < lov_mds_md_size(*stripe_count, LOV_MAGIC_V1)) {
218 CERROR("LOV EA V1 too small: %d, need %d\n",
219 lmm_bytes, lov_mds_md_size(*stripe_count, LOV_MAGIC_V1));
220 lov_dump_lmm_v1(D_WARNING, lmm);
224 return lsm_lmm_verify_common(lmm, lmm_bytes, *stripe_count);
227 int lsm_unpackmd_v1(struct lov_obd *lov, struct lov_stripe_md *lsm,
228 struct lov_mds_md_v1 *lmm)
230 struct lov_oinfo *loi;
233 lsm_unpackmd_common(lsm, lmm);
235 for (i = 0; i < lsm->lsm_stripe_count; i++) {
236 /* XXX LOV STACKING call down to osc_unpackmd() */
237 loi = lsm->lsm_oinfo[i];
238 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
239 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
240 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
241 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
242 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count) {
243 CERROR("OST index %d more than OST count %d\n",
244 loi->loi_ost_idx, lov->desc.ld_tgt_count);
245 lov_dump_lmm_v1(D_WARNING, lmm);
248 if (!lov->lov_tgts[loi->loi_ost_idx]) {
249 CERROR("OST index %d missing\n", loi->loi_ost_idx);
250 lov_dump_lmm_v1(D_WARNING, lmm);
258 const struct lsm_operations lsm_v1_ops = {
259 .lsm_free = lsm_free_plain,
260 .lsm_destroy = lsm_destroy_plain,
261 .lsm_stripe_by_index = lsm_stripe_by_index_plain,
262 .lsm_stripe_by_offset = lsm_stripe_by_offset_plain,
263 .lsm_revalidate = lsm_revalidate_plain,
264 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_plain,
265 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_plain,
266 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_plain,
267 .lsm_lmm_verify = lsm_lmm_verify_v1,
268 .lsm_unpackmd = lsm_unpackmd_v1,
271 struct lov_extent *lovea_off2le(struct lov_stripe_md *lsm, obd_off lov_off)
273 struct lov_array_info *lai;
274 struct lov_extent *le;
277 LASSERT(lsm->lsm_array != NULL);
278 lai = lsm->lsm_array;
279 LASSERT(lai->lai_ext_count > 1);
281 for (le = lai->lai_ext_array, i = 0;
282 i < lai->lai_ext_count && le->le_start + le->le_len <= lov_off
288 CDEBUG(D_INFO, "off "LPU64" idx %d, ext "LPU64":"LPU64" idx %d sc %d\n",
289 lov_off, i, le->le_start, le->le_len, le->le_loi_idx,
290 le->le_stripe_count);
295 struct lov_extent *lovea_idx2le(struct lov_stripe_md *lsm, int stripe_no)
297 struct lov_extent *le;
298 struct lov_array_info *lai;
301 LASSERT(lsm->lsm_array != NULL);
302 LASSERT(stripe_no >= 0 && stripe_no <= lsm->lsm_stripe_count);
303 lai = lsm->lsm_array;
304 LASSERT(lai->lai_ext_count > 1);
306 for (le = lai->lai_ext_array, i = 0, stripe_index = le->le_stripe_count;
307 i < lai->lai_ext_count && stripe_index <= stripe_no &&
308 le->le_len != -1; i ++, le ++,
309 stripe_index += le->le_stripe_count) {
313 CDEBUG(D_INFO, "stripe %d idx %d, ext "LPU64":"LPU64" idx %d sc %d\n",
314 stripe_no, i, le->le_start, le->le_len, le->le_loi_idx,
315 le->le_stripe_count);
319 static void lovea_free_array_info(struct lov_stripe_md *lsm)
321 if (!lsm || !lsm->lsm_array)
324 if (lsm->lsm_array->lai_ext_array)
325 OBD_FREE(lsm->lsm_array->lai_ext_array,
326 lsm->lsm_array->lai_ext_count *
327 sizeof(struct lov_extent));
329 OBD_FREE_PTR(lsm->lsm_array);
332 static void lsm_free_join(struct lov_stripe_md *lsm)
334 lovea_free_array_info(lsm);
339 lsm_stripe_by_index_join(struct lov_stripe_md *lsm, int *stripeno,
340 obd_off *lov_off, unsigned long *swidth)
342 struct lov_extent *le;
344 LASSERT(stripeno != NULL);
346 le = lovea_idx2le(lsm, *stripeno);
348 LASSERT(le != NULL && le->le_stripe_count != 0);
350 *stripeno -= le->le_loi_idx;
353 *swidth = (unsigned long)lsm->lsm_stripe_size * le->le_stripe_count;
356 struct lov_extent *lov_le = lovea_off2le(lsm, *lov_off);
358 *lov_off = (*lov_off > le->le_start) ?
359 (*lov_off - le->le_start) : 0;
361 *lov_off = (*lov_off > le->le_start) ?
363 LASSERT(*lov_off != -1);
369 lsm_stripe_by_offset_join(struct lov_stripe_md *lsm, int *stripeno,
370 obd_off *lov_off, unsigned long *swidth)
372 struct lov_extent *le;
374 LASSERT(lov_off != NULL);
376 le = lovea_off2le(lsm, *lov_off);
378 LASSERT(le != NULL && le->le_stripe_count != 0);
380 *lov_off = (*lov_off > le->le_start) ? (*lov_off - le->le_start) : 0;
383 *stripeno -= le->le_loi_idx;
386 *swidth = (unsigned long)lsm->lsm_stripe_size * le->le_stripe_count;
390 lsm_stripe_offset_by_index_join(struct lov_stripe_md *lsm,
393 struct lov_extent *le;
395 le = lovea_idx2le(lsm, stripe_index);
397 return le ? le->le_start : 0;
401 lsm_stripe_offset_by_offset_join(struct lov_stripe_md *lsm,
404 struct lov_extent *le;
406 le = lovea_off2le(lsm, lov_off);
408 return le ? le->le_start : 0;
412 lsm_stripe_index_by_offset_join(struct lov_stripe_md *lsm,
415 struct lov_extent *le = NULL;
417 le = lovea_off2le(lsm, lov_off);
419 return le ? le->le_loi_idx : 0;
422 static int lovea_unpack_array(struct llog_handle *handle,
423 struct llog_rec_hdr *rec, void *data)
425 struct lovea_unpack_args *args = (struct lovea_unpack_args *)data;
426 struct llog_array_rec *la_rec = (struct llog_array_rec*)rec;
427 struct mds_extent_desc *med = &la_rec->lmr_med;
428 struct lov_stripe_md *lsm = args->lsm;
429 int cursor = args->cursor++;
430 struct lov_mds_md *lmm;
431 struct lov_array_info *lai;
432 struct lov_oinfo * loi;
437 LASSERT(lsm->lsm_stripe_count != 0);
439 LASSERT(lsm->lsm_array != NULL);
441 lai = lsm->lsm_array;
444 lai->lai_ext_array[cursor].le_loi_idx = 0;
446 int next_loi_index = lai->lai_ext_array[cursor - 1].le_loi_idx +
447 lai->lai_ext_array[cursor - 1].le_stripe_count;
448 lai->lai_ext_array[cursor].le_loi_idx = next_loi_index;
450 /* insert extent desc into lsm extent array */
451 lai->lai_ext_array[cursor].le_start = le64_to_cpu(med->med_start);
452 lai->lai_ext_array[cursor].le_len = le64_to_cpu(med->med_len);
453 lai->lai_ext_array[cursor].le_stripe_count =
454 le32_to_cpu(lmm->lmm_stripe_count);
456 /* unpack extent's lmm to lov_oinfo array */
457 loi_index = lai->lai_ext_array[cursor].le_loi_idx;
458 CDEBUG(D_INFO, "lovea upackmd cursor %d, loi_index %d extent "
459 LPU64":"LPU64"\n", cursor, loi_index, med->med_start,
462 for (i = 0; i < le32_to_cpu(lmm->lmm_stripe_count); i ++, loi_index++) {
463 /* XXX LOV STACKING call down to osc_unpackmd() */
464 loi = lsm->lsm_oinfo[loi_index];
465 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
466 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
467 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
468 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
474 static int lsm_revalidate_join(struct lov_stripe_md *lsm,
475 struct obd_device *obd)
477 struct llog_handle *llh;
478 struct llog_ctxt *ctxt;
479 struct lovea_unpack_args args;
483 LASSERT(lsm->lsm_array != NULL);
485 /*Revalidate lsm might be called from client or MDS server.
486 *So the ctxt might be in different position
488 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
490 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
494 if (lsm->lsm_array && lsm->lsm_array->lai_ext_array)
495 GOTO(release_ctxt, rc = 0);
497 CDEBUG(D_INFO, "get lsm logid: "LPU64":"LPU64"\n",
498 lsm->lsm_array->lai_array_id.lgl_oid,
499 lsm->lsm_array->lai_array_id.lgl_ogr);
500 OBD_ALLOC(lsm->lsm_array->lai_ext_array,lsm->lsm_array->lai_ext_count *
501 sizeof (struct lov_extent));
502 if (!lsm->lsm_array->lai_ext_array)
503 GOTO(release_ctxt, rc = -ENOMEM);
505 CDEBUG(D_INFO, "get lsm logid: "LPU64":"LPU64"\n",
506 lsm->lsm_array->lai_array_id.lgl_oid,
507 lsm->lsm_array->lai_array_id.lgl_ogr);
509 rc = llog_create(ctxt, &llh, &lsm->lsm_array->lai_array_id, NULL);
515 rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
517 rc = llog_process(llh, lovea_unpack_array, &args, NULL);
518 rc2 = llog_close(llh);
523 lovea_free_array_info(lsm);
529 int lsm_destroy_join(struct lov_stripe_md *lsm, struct obdo *oa,
530 struct obd_export *md_exp)
532 struct llog_ctxt *ctxt;
533 struct llog_handle *llh;
537 LASSERT(md_exp != NULL);
538 /*for those orphan inode, we should keep array id*/
539 if (!(oa->o_valid & OBD_MD_FLCOOKIE))
542 ctxt = llog_get_context(md_exp->exp_obd, LLOG_LOVEA_REPL_CTXT);
546 LASSERT(lsm->lsm_array != NULL);
547 rc = llog_create(ctxt, &llh, &lsm->lsm_array->lai_array_id,
552 rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
554 rc = llog_destroy(llh);
556 llog_free_handle(llh);
562 static int lsm_lmm_verify_join(struct lov_mds_md *lmm, int lmm_bytes,
565 struct lov_mds_md_join *lmmj = (struct lov_mds_md_join *)lmm;
567 if (lmm_bytes < sizeof(*lmmj)) {
568 CERROR("lov_mds_md too small: %d, need at least %d\n",
569 lmm_bytes, (int)sizeof(*lmmj));
573 if (lmmj->lmmj_array_id.lgl_oid == 0) {
574 CERROR("zero array object id\n");
578 *stripe_count = le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count);
580 return lsm_lmm_verify_common(&lmmj->lmmj_md, lmm_bytes, *stripe_count);
583 static int lovea_init_array_info(struct lov_stripe_md *lsm,
584 struct llog_logid *logid,
587 struct lov_array_info *lai;
594 lai->lai_array_id.lgl_oid = le64_to_cpu(logid->lgl_oid);
595 lai->lai_array_id.lgl_ogr = le64_to_cpu(logid->lgl_ogr);
596 lai->lai_array_id.lgl_ogen = le32_to_cpu(logid->lgl_ogen);
597 lai->lai_ext_count = le32_to_cpu(extent_count);
598 lsm->lsm_array = lai;
602 static int lsm_unpackmd_join(struct lov_obd *lov, struct lov_stripe_md *lsm,
603 struct lov_mds_md *lmm)
605 struct lov_mds_md_join *lmmj = (struct lov_mds_md_join*)lmm;
609 lsm_unpackmd_common(lsm, &lmmj->lmmj_md);
611 rc = lovea_init_array_info(lsm, &lmmj->lmmj_array_id,
612 lmmj->lmmj_extent_count);
614 CERROR("Init joined lsm id"LPU64" arrary error %d",
615 lsm->lsm_object_id, rc);
622 const struct lsm_operations lsm_join_ops = {
623 .lsm_free = lsm_free_join,
624 .lsm_destroy = lsm_destroy_join,
625 .lsm_stripe_by_index = lsm_stripe_by_index_join,
626 .lsm_stripe_by_offset = lsm_stripe_by_offset_join,
627 .lsm_revalidate = lsm_revalidate_join,
628 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_join,
629 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_join,
630 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_join,
631 .lsm_lmm_verify = lsm_lmm_verify_join,
632 .lsm_unpackmd = lsm_unpackmd_join,
636 static int lsm_lmm_verify_v3(struct lov_mds_md *lmmv1, int lmm_bytes,
639 struct lov_mds_md_v3 *lmm;
641 lmm = (struct lov_mds_md_v3 *)lmmv1;
643 if (lmm_bytes < sizeof(*lmm)) {
644 CERROR("lov_mds_md_v3 too small: %d, need at least %d\n",
645 lmm_bytes, (int)sizeof(*lmm));
649 *stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
651 if (lmm_bytes < lov_mds_md_size(*stripe_count, LOV_MAGIC_V3)) {
652 CERROR("LOV EA V3 too small: %d, need %d\n",
653 lmm_bytes, lov_mds_md_size(*stripe_count, LOV_MAGIC_V3));
654 lov_dump_lmm_v3(D_WARNING, lmm);
658 return lsm_lmm_verify_common((struct lov_mds_md_v1 *)lmm, lmm_bytes,
662 int lsm_unpackmd_v3(struct lov_obd *lov, struct lov_stripe_md *lsm,
663 struct lov_mds_md *lmmv1)
665 struct lov_mds_md_v3 *lmm;
666 struct lov_oinfo *loi;
669 lmm = (struct lov_mds_md_v3 *)lmmv1;
671 lsm_unpackmd_common(lsm, (struct lov_mds_md_v1 *)lmm);
672 strncpy(lsm->lsm_pool_name, lmm->lmm_pool_name, LOV_MAXPOOLNAME);
674 for (i = 0; i < lsm->lsm_stripe_count; i++) {
675 /* XXX LOV STACKING call down to osc_unpackmd() */
676 loi = lsm->lsm_oinfo[i];
677 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
678 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
679 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
680 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
681 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count) {
682 CERROR("OST index %d more than OST count %d\n",
683 loi->loi_ost_idx, lov->desc.ld_tgt_count);
684 lov_dump_lmm_v3(D_WARNING, lmm);
687 if (!lov->lov_tgts[loi->loi_ost_idx]) {
688 CERROR("OST index %d missing\n", loi->loi_ost_idx);
689 lov_dump_lmm_v3(D_WARNING, lmm);
697 const struct lsm_operations lsm_v3_ops = {
698 .lsm_free = lsm_free_plain,
699 .lsm_destroy = lsm_destroy_plain,
700 .lsm_stripe_by_index = lsm_stripe_by_index_plain,
701 .lsm_stripe_by_offset = lsm_stripe_by_offset_plain,
702 .lsm_revalidate = lsm_revalidate_plain,
703 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_plain,
704 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_plain,
705 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_plain,
706 .lsm_lmm_verify = lsm_lmm_verify_v3,
707 .lsm_unpackmd = lsm_unpackmd_v3,