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 (le32_to_cpu(lmm->lmm_stripe_size)&(LOV_MIN_STRIPE_SIZE-1)) != 0) {
89 CERROR("bad stripe size %u\n",
90 le32_to_cpu(lmm->lmm_stripe_size));
91 lov_dump_lmm(D_WARNING, lmm);
97 struct lov_stripe_md *lsm_alloc_plain(int stripe_count, int *size)
99 struct lov_stripe_md *lsm;
100 int i, oinfo_ptrs_size;
101 struct lov_oinfo *loi;
103 LASSERT(stripe_count > 0);
105 oinfo_ptrs_size = sizeof(struct lov_oinfo *) * stripe_count;
106 *size = sizeof(struct lov_stripe_md) + oinfo_ptrs_size;
108 OBD_ALLOC(lsm, *size);
112 for (i = 0; i < stripe_count; i++) {
113 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, CFS_ALLOC_IO);
116 lsm->lsm_oinfo[i] = loi;
118 lsm->lsm_stripe_count = stripe_count;
119 lsm->lsm_pool_name[0] = '\0';
124 OBD_SLAB_FREE(lsm->lsm_oinfo[i], lov_oinfo_slab, sizeof(*loi));
125 OBD_FREE(lsm, *size);
129 void lsm_free_plain(struct lov_stripe_md *lsm)
131 int stripe_count = lsm->lsm_stripe_count;
134 for (i = 0; i < stripe_count; i++)
135 OBD_SLAB_FREE(lsm->lsm_oinfo[i], lov_oinfo_slab,
136 sizeof(struct lov_oinfo));
137 OBD_FREE(lsm, sizeof(struct lov_stripe_md) +
138 stripe_count * sizeof(struct lov_oinfo *));
141 static void lsm_unpackmd_common(struct lov_stripe_md *lsm,
142 struct lov_mds_md *lmm)
145 * This supposes lov_mds_md_v1/v3 first fields are
148 lsm->lsm_object_id = le64_to_cpu(lmm->lmm_object_id);
149 lsm->lsm_object_gr = le64_to_cpu(lmm->lmm_object_gr);
150 lsm->lsm_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
151 lsm->lsm_pattern = le32_to_cpu(lmm->lmm_pattern);
152 lsm->lsm_pool_name[0] = '\0';
156 lsm_stripe_by_index_plain(struct lov_stripe_md *lsm, int *stripeno,
157 obd_off *lov_off, obd_off *swidth)
160 *swidth = (obd_off)lsm->lsm_stripe_size * lsm->lsm_stripe_count;
164 lsm_stripe_by_offset_plain(struct lov_stripe_md *lsm, int *stripeno,
165 obd_off *lov_off, obd_off *swidth)
168 *swidth = (obd_off)lsm->lsm_stripe_size * lsm->lsm_stripe_count;
172 lsm_stripe_offset_by_index_plain(struct lov_stripe_md *lsm,
179 lsm_stripe_offset_by_offset_plain(struct lov_stripe_md *lsm,
186 lsm_stripe_index_by_offset_plain(struct lov_stripe_md *lsm,
192 static int lsm_revalidate_plain(struct lov_stripe_md *lsm,
193 struct obd_device *obd)
198 static int lsm_destroy_plain(struct lov_stripe_md *lsm, struct obdo *oa,
199 struct obd_export *md_exp)
204 static int lsm_lmm_verify_v1(struct lov_mds_md_v1 *lmm, int lmm_bytes,
207 if (lmm_bytes < sizeof(*lmm)) {
208 CERROR("lov_mds_md_v1 too small: %d, need at least %d\n",
209 lmm_bytes, (int)sizeof(*lmm));
213 *stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
215 if (lmm_bytes < lov_mds_md_size(*stripe_count, LOV_MAGIC_V1)) {
216 CERROR("LOV EA V1 too small: %d, need %d\n",
217 lmm_bytes, lov_mds_md_size(*stripe_count, LOV_MAGIC_V1));
218 lov_dump_lmm_v1(D_WARNING, lmm);
222 return lsm_lmm_verify_common(lmm, lmm_bytes, *stripe_count);
225 int lsm_unpackmd_v1(struct lov_obd *lov, struct lov_stripe_md *lsm,
226 struct lov_mds_md_v1 *lmm)
228 struct lov_oinfo *loi;
231 lsm_unpackmd_common(lsm, lmm);
233 for (i = 0; i < lsm->lsm_stripe_count; i++) {
234 /* XXX LOV STACKING call down to osc_unpackmd() */
235 loi = lsm->lsm_oinfo[i];
236 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
237 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
238 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
239 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
240 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count) {
241 CERROR("OST index %d more than OST count %d\n",
242 loi->loi_ost_idx, lov->desc.ld_tgt_count);
243 lov_dump_lmm_v1(D_WARNING, lmm);
246 if (!lov->lov_tgts[loi->loi_ost_idx]) {
247 CERROR("OST index %d missing\n", loi->loi_ost_idx);
248 lov_dump_lmm_v1(D_WARNING, lmm);
256 const struct lsm_operations lsm_v1_ops = {
257 .lsm_free = lsm_free_plain,
258 .lsm_destroy = lsm_destroy_plain,
259 .lsm_stripe_by_index = lsm_stripe_by_index_plain,
260 .lsm_stripe_by_offset = lsm_stripe_by_offset_plain,
261 .lsm_revalidate = lsm_revalidate_plain,
262 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_plain,
263 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_plain,
264 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_plain,
265 .lsm_lmm_verify = lsm_lmm_verify_v1,
266 .lsm_unpackmd = lsm_unpackmd_v1,
269 struct lov_extent *lovea_off2le(struct lov_stripe_md *lsm, obd_off lov_off)
271 struct lov_array_info *lai;
272 struct lov_extent *le;
275 LASSERT(lsm->lsm_array != NULL);
276 lai = lsm->lsm_array;
277 LASSERT(lai->lai_ext_count > 1);
279 for (le = lai->lai_ext_array, i = 0;
280 i < lai->lai_ext_count && le->le_start + le->le_len <= lov_off
286 CDEBUG(D_INFO, "off "LPU64" idx %d, ext "LPU64":"LPU64" idx %d sc %d\n",
287 lov_off, i, le->le_start, le->le_len, le->le_loi_idx,
288 le->le_stripe_count);
293 struct lov_extent *lovea_idx2le(struct lov_stripe_md *lsm, int stripe_no)
295 struct lov_extent *le;
296 struct lov_array_info *lai;
299 LASSERT(lsm->lsm_array != NULL);
300 LASSERT(stripe_no >= 0 && stripe_no <= lsm->lsm_stripe_count);
301 lai = lsm->lsm_array;
302 LASSERT(lai->lai_ext_count > 1);
304 for (le = lai->lai_ext_array, i = 0, stripe_index = le->le_stripe_count;
305 i < lai->lai_ext_count && stripe_index <= stripe_no &&
306 le->le_len != -1; i ++, le ++,
307 stripe_index += le->le_stripe_count) {
311 CDEBUG(D_INFO, "stripe %d idx %d, ext "LPU64":"LPU64" idx %d sc %d\n",
312 stripe_no, i, le->le_start, le->le_len, le->le_loi_idx,
313 le->le_stripe_count);
317 static void lovea_free_array_info(struct lov_stripe_md *lsm)
319 if (!lsm || !lsm->lsm_array)
322 if (lsm->lsm_array->lai_ext_array)
323 OBD_FREE(lsm->lsm_array->lai_ext_array,
324 lsm->lsm_array->lai_ext_count *
325 sizeof(struct lov_extent));
327 OBD_FREE_PTR(lsm->lsm_array);
330 static void lsm_free_join(struct lov_stripe_md *lsm)
332 lovea_free_array_info(lsm);
337 lsm_stripe_by_index_join(struct lov_stripe_md *lsm, int *stripeno,
338 obd_off *lov_off, obd_off *swidth)
340 struct lov_extent *le;
342 LASSERT(stripeno != NULL);
344 le = lovea_idx2le(lsm, *stripeno);
346 LASSERT(le != NULL && le->le_stripe_count != 0);
348 *stripeno -= le->le_loi_idx;
351 *swidth = (obd_off)lsm->lsm_stripe_size * le->le_stripe_count;
354 struct lov_extent *lov_le = lovea_off2le(lsm, *lov_off);
356 *lov_off = (*lov_off > le->le_start) ?
357 (*lov_off - le->le_start) : 0;
359 *lov_off = (*lov_off > le->le_start) ?
361 LASSERT(*lov_off != -1);
367 lsm_stripe_by_offset_join(struct lov_stripe_md *lsm, int *stripeno,
368 obd_off *lov_off, obd_off *swidth)
370 struct lov_extent *le;
372 LASSERT(lov_off != NULL);
374 le = lovea_off2le(lsm, *lov_off);
376 LASSERT(le != NULL && le->le_stripe_count != 0);
378 *lov_off = (*lov_off > le->le_start) ? (*lov_off - le->le_start) : 0;
381 *stripeno -= le->le_loi_idx;
384 *swidth = (obd_off)lsm->lsm_stripe_size * le->le_stripe_count;
388 lsm_stripe_offset_by_index_join(struct lov_stripe_md *lsm,
391 struct lov_extent *le;
393 le = lovea_idx2le(lsm, stripe_index);
395 return le ? le->le_start : 0;
399 lsm_stripe_offset_by_offset_join(struct lov_stripe_md *lsm,
402 struct lov_extent *le;
404 le = lovea_off2le(lsm, lov_off);
406 return le ? le->le_start : 0;
410 lsm_stripe_index_by_offset_join(struct lov_stripe_md *lsm,
413 struct lov_extent *le = NULL;
415 le = lovea_off2le(lsm, lov_off);
417 return le ? le->le_loi_idx : 0;
420 static int lovea_unpack_array(struct llog_handle *handle,
421 struct llog_rec_hdr *rec, void *data)
423 struct lovea_unpack_args *args = (struct lovea_unpack_args *)data;
424 struct llog_array_rec *la_rec = (struct llog_array_rec*)rec;
425 struct mds_extent_desc *med = &la_rec->lmr_med;
426 struct lov_stripe_md *lsm = args->lsm;
427 int cursor = args->cursor++;
428 struct lov_mds_md *lmm;
429 struct lov_array_info *lai;
430 struct lov_oinfo * loi;
435 LASSERT(lsm->lsm_stripe_count != 0);
437 LASSERT(lsm->lsm_array != NULL);
439 lai = lsm->lsm_array;
442 lai->lai_ext_array[cursor].le_loi_idx = 0;
444 int next_loi_index = lai->lai_ext_array[cursor - 1].le_loi_idx +
445 lai->lai_ext_array[cursor - 1].le_stripe_count;
446 lai->lai_ext_array[cursor].le_loi_idx = next_loi_index;
448 /* insert extent desc into lsm extent array */
449 lai->lai_ext_array[cursor].le_start = le64_to_cpu(med->med_start);
450 lai->lai_ext_array[cursor].le_len = le64_to_cpu(med->med_len);
451 lai->lai_ext_array[cursor].le_stripe_count =
452 le32_to_cpu(lmm->lmm_stripe_count);
454 /* unpack extent's lmm to lov_oinfo array */
455 loi_index = lai->lai_ext_array[cursor].le_loi_idx;
456 CDEBUG(D_INFO, "lovea upackmd cursor %d, loi_index %d extent "
457 LPU64":"LPU64"\n", cursor, loi_index, med->med_start,
460 for (i = 0; i < le32_to_cpu(lmm->lmm_stripe_count); i ++, loi_index++) {
461 /* XXX LOV STACKING call down to osc_unpackmd() */
462 loi = lsm->lsm_oinfo[loi_index];
463 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
464 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
465 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
466 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
472 static int lsm_revalidate_join(struct lov_stripe_md *lsm,
473 struct obd_device *obd)
475 struct llog_handle *llh;
476 struct llog_ctxt *ctxt;
477 struct lovea_unpack_args args;
481 LASSERT(lsm->lsm_array != NULL);
483 /*Revalidate lsm might be called from client or MDS server.
484 *So the ctxt might be in different position
486 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
488 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
492 if (lsm->lsm_array && lsm->lsm_array->lai_ext_array)
493 GOTO(release_ctxt, rc = 0);
495 CDEBUG(D_INFO, "get lsm logid: "LPU64":"LPU64"\n",
496 lsm->lsm_array->lai_array_id.lgl_oid,
497 lsm->lsm_array->lai_array_id.lgl_ogr);
498 OBD_ALLOC(lsm->lsm_array->lai_ext_array,lsm->lsm_array->lai_ext_count *
499 sizeof (struct lov_extent));
500 if (!lsm->lsm_array->lai_ext_array)
501 GOTO(release_ctxt, rc = -ENOMEM);
503 CDEBUG(D_INFO, "get lsm logid: "LPU64":"LPU64"\n",
504 lsm->lsm_array->lai_array_id.lgl_oid,
505 lsm->lsm_array->lai_array_id.lgl_ogr);
507 rc = llog_create(ctxt, &llh, &lsm->lsm_array->lai_array_id, NULL);
513 rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
515 rc = llog_process(llh, lovea_unpack_array, &args, NULL);
516 rc2 = llog_close(llh);
521 lovea_free_array_info(lsm);
527 int lsm_destroy_join(struct lov_stripe_md *lsm, struct obdo *oa,
528 struct obd_export *md_exp)
530 struct llog_ctxt *ctxt;
531 struct llog_handle *llh;
535 LASSERT(md_exp != NULL);
536 /*for those orphan inode, we should keep array id*/
537 if (!(oa->o_valid & OBD_MD_FLCOOKIE))
540 ctxt = llog_get_context(md_exp->exp_obd, LLOG_LOVEA_REPL_CTXT);
544 LASSERT(lsm->lsm_array != NULL);
545 rc = llog_create(ctxt, &llh, &lsm->lsm_array->lai_array_id,
550 rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
552 rc = llog_destroy(llh);
554 llog_free_handle(llh);
560 static int lsm_lmm_verify_join(struct lov_mds_md *lmm, int lmm_bytes,
563 struct lov_mds_md_join *lmmj = (struct lov_mds_md_join *)lmm;
565 if (lmm_bytes < sizeof(*lmmj)) {
566 CERROR("lov_mds_md too small: %d, need at least %d\n",
567 lmm_bytes, (int)sizeof(*lmmj));
571 if (lmmj->lmmj_array_id.lgl_oid == 0) {
572 CERROR("zero array object id\n");
576 *stripe_count = le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count);
578 return lsm_lmm_verify_common(&lmmj->lmmj_md, lmm_bytes, *stripe_count);
581 static int lovea_init_array_info(struct lov_stripe_md *lsm,
582 struct llog_logid *logid,
585 struct lov_array_info *lai;
592 lai->lai_array_id.lgl_oid = le64_to_cpu(logid->lgl_oid);
593 lai->lai_array_id.lgl_ogr = le64_to_cpu(logid->lgl_ogr);
594 lai->lai_array_id.lgl_ogen = le32_to_cpu(logid->lgl_ogen);
595 lai->lai_ext_count = le32_to_cpu(extent_count);
596 lsm->lsm_array = lai;
600 static int lsm_unpackmd_join(struct lov_obd *lov, struct lov_stripe_md *lsm,
601 struct lov_mds_md *lmm)
603 struct lov_mds_md_join *lmmj = (struct lov_mds_md_join*)lmm;
607 lsm_unpackmd_common(lsm, &lmmj->lmmj_md);
609 rc = lovea_init_array_info(lsm, &lmmj->lmmj_array_id,
610 lmmj->lmmj_extent_count);
612 CERROR("Init joined lsm id"LPU64" arrary error %d",
613 lsm->lsm_object_id, rc);
620 const struct lsm_operations lsm_join_ops = {
621 .lsm_free = lsm_free_join,
622 .lsm_destroy = lsm_destroy_join,
623 .lsm_stripe_by_index = lsm_stripe_by_index_join,
624 .lsm_stripe_by_offset = lsm_stripe_by_offset_join,
625 .lsm_revalidate = lsm_revalidate_join,
626 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_join,
627 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_join,
628 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_join,
629 .lsm_lmm_verify = lsm_lmm_verify_join,
630 .lsm_unpackmd = lsm_unpackmd_join,
634 static int lsm_lmm_verify_v3(struct lov_mds_md *lmmv1, int lmm_bytes,
637 struct lov_mds_md_v3 *lmm;
639 lmm = (struct lov_mds_md_v3 *)lmmv1;
641 if (lmm_bytes < sizeof(*lmm)) {
642 CERROR("lov_mds_md_v3 too small: %d, need at least %d\n",
643 lmm_bytes, (int)sizeof(*lmm));
647 *stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
649 if (lmm_bytes < lov_mds_md_size(*stripe_count, LOV_MAGIC_V3)) {
650 CERROR("LOV EA V3 too small: %d, need %d\n",
651 lmm_bytes, lov_mds_md_size(*stripe_count, LOV_MAGIC_V3));
652 lov_dump_lmm_v3(D_WARNING, lmm);
656 return lsm_lmm_verify_common((struct lov_mds_md_v1 *)lmm, lmm_bytes,
660 int lsm_unpackmd_v3(struct lov_obd *lov, struct lov_stripe_md *lsm,
661 struct lov_mds_md *lmmv1)
663 struct lov_mds_md_v3 *lmm;
664 struct lov_oinfo *loi;
667 lmm = (struct lov_mds_md_v3 *)lmmv1;
669 lsm_unpackmd_common(lsm, (struct lov_mds_md_v1 *)lmm);
670 strncpy(lsm->lsm_pool_name, lmm->lmm_pool_name, LOV_MAXPOOLNAME);
672 for (i = 0; i < lsm->lsm_stripe_count; i++) {
673 /* XXX LOV STACKING call down to osc_unpackmd() */
674 loi = lsm->lsm_oinfo[i];
675 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
676 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
677 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
678 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
679 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count) {
680 CERROR("OST index %d more than OST count %d\n",
681 loi->loi_ost_idx, lov->desc.ld_tgt_count);
682 lov_dump_lmm_v3(D_WARNING, lmm);
685 if (!lov->lov_tgts[loi->loi_ost_idx]) {
686 CERROR("OST index %d missing\n", loi->loi_ost_idx);
687 lov_dump_lmm_v3(D_WARNING, lmm);
695 const struct lsm_operations lsm_v3_ops = {
696 .lsm_free = lsm_free_plain,
697 .lsm_destroy = lsm_destroy_plain,
698 .lsm_stripe_by_index = lsm_stripe_by_index_plain,
699 .lsm_stripe_by_offset = lsm_stripe_by_offset_plain,
700 .lsm_revalidate = lsm_revalidate_plain,
701 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_plain,
702 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_plain,
703 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_plain,
704 .lsm_lmm_verify = lsm_lmm_verify_v3,
705 .lsm_unpackmd = lsm_unpackmd_v3,