1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2001-2005 Cluster File Systems, Inc.
5 * Author: Wang Di <wangdi@clusterfs.com>
7 * This file is part of the Lustre file system, http://www.lustre.org
8 * Lustre is a trademark of Cluster File Systems, Inc.
10 * You may have signed or agreed to another license before downloading
11 * this software. If so, you are bound by the terms and conditions
12 * of that agreement, and the following does not apply to you. See the
13 * LICENSE file included with this distribution for more information.
15 * If you did not agree to a different license, then this copy of Lustre
16 * is open source software; you can redistribute it and/or modify it
17 * under the terms of version 2 of the GNU General Public License as
18 * published by the Free Software Foundation.
20 * In either case, Lustre is distributed in the hope that it will be
21 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
22 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 * license text for more details.
27 # define EXPORT_SYMTAB
29 #define DEBUG_SUBSYSTEM S_LOV
32 #include <asm/div64.h>
33 #include <libcfs/libcfs.h>
35 #include <liblustre.h>
38 #include <obd_class.h>
40 #include <lustre/lustre_idl.h>
41 #include <lustre_log.h>
43 #include "lov_internal.h"
45 struct lovea_unpack_args {
46 struct lov_stripe_md *lsm;
50 static int lsm_lmm_verify_common(struct lov_mds_md *lmm, int lmm_bytes,
54 if (stripe_count == 0) {
55 CERROR("bad stripe count %d\n", stripe_count);
56 lov_dump_lmm_v1(D_WARNING, lmm);
60 if (lmm->lmm_object_id == 0) {
61 CERROR("zero object id\n");
62 lov_dump_lmm_v1(D_WARNING, lmm);
66 if (lmm->lmm_pattern != cpu_to_le32(LOV_PATTERN_RAID0)) {
67 CERROR("bad striping pattern\n");
68 lov_dump_lmm_v1(D_WARNING, lmm);
72 if (lmm->lmm_stripe_size == 0 ||
73 (__u64)le32_to_cpu(lmm->lmm_stripe_size)*stripe_count > 0xffffffff){
74 CERROR("bad stripe size %u\n",
75 le32_to_cpu(lmm->lmm_stripe_size));
76 lov_dump_lmm_v1(D_WARNING, lmm);
82 struct lov_stripe_md *lsm_alloc_plain(int stripe_count, int *size)
84 struct lov_stripe_md *lsm;
85 int i, oinfo_ptrs_size;
86 struct lov_oinfo *loi;
88 LASSERT(stripe_count > 0);
90 oinfo_ptrs_size = sizeof(struct lov_oinfo *) * stripe_count;
91 *size = sizeof(struct lov_stripe_md) + oinfo_ptrs_size;
93 OBD_ALLOC(lsm, *size);
97 for (i = 0; i < stripe_count; i++) {
98 OBD_SLAB_ALLOC(loi, lov_oinfo_slab, GFP_NOFS, sizeof(*loi));
101 lsm->lsm_oinfo[i] = loi;
103 lsm->lsm_stripe_count = stripe_count;
108 OBD_SLAB_FREE(lsm->lsm_oinfo[i], lov_oinfo_slab, sizeof(*loi));
109 OBD_FREE(lsm, *size);
113 void lsm_free_plain(struct lov_stripe_md *lsm)
115 int stripe_count = lsm->lsm_stripe_count;
118 for (i = 0; i < stripe_count; i++)
119 OBD_SLAB_FREE(lsm->lsm_oinfo[i], lov_oinfo_slab,
120 sizeof(struct lov_oinfo));
121 OBD_FREE(lsm, sizeof(struct lov_stripe_md) +
122 stripe_count * sizeof(struct lov_oinfo *));
125 static void lsm_unpackmd_common(struct lov_stripe_md *lsm,
126 struct lov_mds_md *lmm)
128 lsm->lsm_object_id = le64_to_cpu(lmm->lmm_object_id);
129 lsm->lsm_object_gr = le64_to_cpu(lmm->lmm_object_gr);
130 lsm->lsm_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
131 lsm->lsm_pattern = le32_to_cpu(lmm->lmm_pattern);
135 lsm_stripe_by_index_plain(struct lov_stripe_md *lsm, int *stripeno,
136 obd_off *lov_off, unsigned long *swidth)
139 *swidth = (ulong)lsm->lsm_stripe_size * lsm->lsm_stripe_count;
143 lsm_stripe_by_offset_plain(struct lov_stripe_md *lsm, int *stripeno,
144 obd_off *lov_off, unsigned long *swidth)
147 *swidth = (ulong)lsm->lsm_stripe_size * lsm->lsm_stripe_count;
151 lsm_stripe_offset_by_index_plain(struct lov_stripe_md *lsm,
158 lsm_stripe_offset_by_offset_plain(struct lov_stripe_md *lsm,
165 lsm_stripe_index_by_offset_plain(struct lov_stripe_md *lsm,
171 static int lsm_revalidate_plain(struct lov_stripe_md *lsm,
172 struct obd_device *obd)
177 static int lsm_destroy_plain(struct lov_stripe_md *lsm, struct obdo *oa,
178 struct obd_export *md_exp)
183 static int lsm_lmm_verify_plain(struct lov_mds_md *lmm, int lmm_bytes,
186 if (lmm_bytes < sizeof(*lmm)) {
187 CERROR("lov_mds_md too small: %d, need at least %d\n",
188 lmm_bytes, (int)sizeof(*lmm));
192 *stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
194 if (lmm_bytes < lov_mds_md_v1_size(*stripe_count)) {
195 CERROR("LOV EA too small: %d, need %d\n",
196 lmm_bytes, lov_mds_md_v1_size(*stripe_count));
197 lov_dump_lmm_v1(D_WARNING, lmm);
201 return lsm_lmm_verify_common(lmm, lmm_bytes, *stripe_count);
204 int lsm_unpackmd_plain(struct lov_obd *lov, struct lov_stripe_md *lsm,
205 struct lov_mds_md_v1 *lmm)
207 struct lov_oinfo *loi;
210 lsm_unpackmd_common(lsm, lmm);
212 for (i = 0; i < lsm->lsm_stripe_count; i++) {
213 /* XXX LOV STACKING call down to osc_unpackmd() */
214 loi = lsm->lsm_oinfo[i];
215 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
216 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
217 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
218 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
219 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count) {
220 CERROR("OST index %d more than OST count %d\n",
221 loi->loi_ost_idx, lov->desc.ld_tgt_count);
222 lov_dump_lmm_v1(D_WARNING, lmm);
225 if (!lov->lov_tgts[loi->loi_ost_idx]) {
226 CERROR("OST index %d missing\n", loi->loi_ost_idx);
227 lov_dump_lmm_v1(D_WARNING, lmm);
235 struct lsm_operations lsm_plain_ops = {
236 .lsm_free = lsm_free_plain,
237 .lsm_destroy = lsm_destroy_plain,
238 .lsm_stripe_by_index = lsm_stripe_by_index_plain,
239 .lsm_stripe_by_offset = lsm_stripe_by_offset_plain,
240 .lsm_revalidate = lsm_revalidate_plain,
241 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_plain,
242 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_plain,
243 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_plain,
244 .lsm_lmm_verify = lsm_lmm_verify_plain,
245 .lsm_unpackmd = lsm_unpackmd_plain,
248 struct lov_extent *lovea_off2le(struct lov_stripe_md *lsm, obd_off lov_off)
250 struct lov_array_info *lai;
251 struct lov_extent *le;
254 LASSERT(lsm->lsm_array != NULL);
255 lai = lsm->lsm_array;
256 LASSERT(lai->lai_ext_count > 1);
258 for (le = lai->lai_ext_array, i = 0;
259 i < lai->lai_ext_count && le->le_start + le->le_len <= lov_off
265 CDEBUG(D_INFO, "off "LPU64" idx %d, ext "LPU64":"LPU64" idx %d sc %d\n",
266 lov_off, i, le->le_start, le->le_len, le->le_loi_idx,
267 le->le_stripe_count);
272 struct lov_extent *lovea_idx2le(struct lov_stripe_md *lsm, int stripe_no)
274 struct lov_extent *le;
275 struct lov_array_info *lai;
278 LASSERT(lsm->lsm_array != NULL);
279 LASSERT(stripe_no >= 0 && stripe_no <= lsm->lsm_stripe_count);
280 lai = lsm->lsm_array;
281 LASSERT(lai->lai_ext_count > 1);
283 for (le = lai->lai_ext_array, i = 0, stripe_index = le->le_stripe_count;
284 i < lai->lai_ext_count && stripe_index <= stripe_no &&
285 le->le_len != -1; i ++, le ++,
286 stripe_index += le->le_stripe_count) {
290 CDEBUG(D_INFO, "stripe %d idx %d, ext "LPU64":"LPU64" idx %d sc %d\n",
291 stripe_no, i, le->le_start, le->le_len, le->le_loi_idx,
292 le->le_stripe_count);
296 static void lovea_free_array_info(struct lov_stripe_md *lsm)
298 if (!lsm || !lsm->lsm_array)
301 if (lsm->lsm_array->lai_ext_array)
302 OBD_FREE(lsm->lsm_array->lai_ext_array,
303 lsm->lsm_array->lai_ext_count *
304 sizeof(struct lov_extent));
306 OBD_FREE_PTR(lsm->lsm_array);
309 static void lsm_free_join(struct lov_stripe_md *lsm)
311 lovea_free_array_info(lsm);
316 lsm_stripe_by_index_join(struct lov_stripe_md *lsm, int *stripeno,
317 obd_off *lov_off, unsigned long *swidth)
319 struct lov_extent *le;
321 LASSERT(stripeno != NULL);
323 le = lovea_idx2le(lsm, *stripeno);
325 LASSERT(le != NULL && le->le_stripe_count != 0);
327 *stripeno -= le->le_loi_idx;
330 *swidth = (ulong)lsm->lsm_stripe_size * le->le_stripe_count;
333 struct lov_extent *lov_le = lovea_off2le(lsm, *lov_off);
335 *lov_off = (*lov_off > le->le_start) ?
336 (*lov_off - le->le_start) : 0;
338 *lov_off = (*lov_off > le->le_start) ?
340 LASSERT(*lov_off != -1);
346 lsm_stripe_by_offset_join(struct lov_stripe_md *lsm, int *stripeno,
347 obd_off *lov_off, unsigned long *swidth)
349 struct lov_extent *le;
351 LASSERT(lov_off != NULL);
353 le = lovea_off2le(lsm, *lov_off);
355 LASSERT(le != NULL && le->le_stripe_count != 0);
357 *lov_off = (*lov_off > le->le_start) ? (*lov_off - le->le_start) : 0;
360 *stripeno -= le->le_loi_idx;
363 *swidth = (ulong)lsm->lsm_stripe_size * le->le_stripe_count;
367 lsm_stripe_offset_by_index_join(struct lov_stripe_md *lsm,
370 struct lov_extent *le;
372 le = lovea_idx2le(lsm, stripe_index);
374 return le ? le->le_start : 0;
378 lsm_stripe_offset_by_offset_join(struct lov_stripe_md *lsm,
381 struct lov_extent *le;
383 le = lovea_off2le(lsm, lov_off);
385 return le ? le->le_start : 0;
389 lsm_stripe_index_by_offset_join(struct lov_stripe_md *lsm,
392 struct lov_extent *le = NULL;
394 le = lovea_off2le(lsm, lov_off);
396 return le ? le->le_loi_idx : 0;
399 static int lovea_unpack_array(struct llog_handle *handle,
400 struct llog_rec_hdr *rec, void *data)
402 struct lovea_unpack_args *args = (struct lovea_unpack_args *)data;
403 struct llog_array_rec *la_rec = (struct llog_array_rec*)rec;
404 struct mds_extent_desc *med = &la_rec->lmr_med;
405 struct lov_stripe_md *lsm = args->lsm;
406 int cursor = args->cursor++;
407 struct lov_mds_md *lmm;
408 struct lov_array_info *lai;
409 struct lov_oinfo * loi;
414 LASSERT(lsm->lsm_stripe_count != 0);
416 LASSERT(lsm->lsm_array != NULL);
418 lai = lsm->lsm_array;
421 lai->lai_ext_array[cursor].le_loi_idx = 0;
423 int next_loi_index = lai->lai_ext_array[cursor - 1].le_loi_idx +
424 lai->lai_ext_array[cursor - 1].le_stripe_count;
425 lai->lai_ext_array[cursor].le_loi_idx = next_loi_index;
427 /* insert extent desc into lsm extent array */
428 lai->lai_ext_array[cursor].le_start = le64_to_cpu(med->med_start);
429 lai->lai_ext_array[cursor].le_len = le64_to_cpu(med->med_len);
430 lai->lai_ext_array[cursor].le_stripe_count = lmm->lmm_stripe_count;
432 /* unpack extent's lmm to lov_oinfo array */
433 loi_index = lai->lai_ext_array[cursor].le_loi_idx;
434 CDEBUG(D_INFO, "lovea upackmd cursor %d, loi_index %d extent "
435 LPU64":"LPU64"\n", cursor, loi_index, med->med_start,
438 for (i = 0; i < le32_to_cpu(lmm->lmm_stripe_count); i ++, loi_index++) {
439 /* XXX LOV STACKING call down to osc_unpackmd() */
440 loi = lsm->lsm_oinfo[loi_index];
441 loi->loi_id = le64_to_cpu(lmm->lmm_objects[i].l_object_id);
442 loi->loi_gr = le64_to_cpu(lmm->lmm_objects[i].l_object_gr);
443 loi->loi_ost_idx = le32_to_cpu(lmm->lmm_objects[i].l_ost_idx);
444 loi->loi_ost_gen = le32_to_cpu(lmm->lmm_objects[i].l_ost_gen);
450 static int lsm_revalidate_join(struct lov_stripe_md *lsm,
451 struct obd_device *obd)
453 struct llog_handle *llh;
454 struct llog_ctxt *ctxt;
455 struct lovea_unpack_args args;
459 LASSERT(lsm->lsm_array != NULL);
461 /*Revalidate lsm might be called from client or MDS server.
462 *So the ctxt might be in different position
464 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
466 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
470 if (lsm->lsm_array && lsm->lsm_array->lai_ext_array)
473 CDEBUG(D_INFO, "get lsm logid: "LPU64":"LPU64"\n",
474 lsm->lsm_array->lai_array_id.lgl_oid,
475 lsm->lsm_array->lai_array_id.lgl_ogr);
476 OBD_ALLOC(lsm->lsm_array->lai_ext_array,lsm->lsm_array->lai_ext_count *
477 sizeof (struct lov_extent));
478 if (!lsm->lsm_array->lai_ext_array)
481 CDEBUG(D_INFO, "get lsm logid: "LPU64":"LPU64"\n",
482 lsm->lsm_array->lai_array_id.lgl_oid,
483 lsm->lsm_array->lai_array_id.lgl_ogr);
485 rc = llog_create(ctxt, &llh, &lsm->lsm_array->lai_array_id, NULL);
491 rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
493 rc = llog_process(llh, lovea_unpack_array, &args, NULL);
494 rc2 = llog_close(llh);
499 lovea_free_array_info(lsm);
503 int lsm_destroy_join(struct lov_stripe_md *lsm, struct obdo *oa,
504 struct obd_export *md_exp)
506 struct llog_ctxt *ctxt;
507 struct llog_handle *llh;
511 LASSERT(md_exp != NULL);
512 ctxt = llog_get_context(md_exp->exp_obd, LLOG_LOVEA_REPL_CTXT);
514 GOTO(out, rc = -EINVAL);
516 LASSERT(lsm->lsm_array != NULL);
517 /*for those orphan inode, we should keep array id*/
518 if (!(oa->o_valid & OBD_MD_FLCOOKIE))
521 LASSERT(ctxt != NULL);
522 rc = llog_create(ctxt, &llh, &lsm->lsm_array->lai_array_id,
527 rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
529 rc = llog_destroy(llh);
531 llog_free_handle(llh);
536 static int lsm_lmm_verify_join(struct lov_mds_md *lmm, int lmm_bytes,
539 struct lov_mds_md_join *lmmj = (struct lov_mds_md_join *)lmm;
541 if (lmm_bytes < sizeof(*lmmj)) {
542 CERROR("lov_mds_md too small: %d, need at least %d\n",
543 lmm_bytes, (int)sizeof(*lmmj));
547 if (lmmj->lmmj_array_id.lgl_oid == 0) {
548 CERROR("zero array object id\n");
552 *stripe_count = le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count);
554 return lsm_lmm_verify_common(&lmmj->lmmj_md, lmm_bytes, *stripe_count);
557 static int lovea_init_array_info(struct lov_stripe_md *lsm,
558 struct llog_logid *logid,
561 struct lov_array_info *lai;
568 lai->lai_array_id = *logid;
569 lai->lai_ext_count = extent_count;
570 lsm->lsm_array = lai;
574 static int lsm_unpackmd_join(struct lov_obd *lov, struct lov_stripe_md *lsm,
575 struct lov_mds_md *lmm)
577 struct lov_mds_md_join *lmmj = (struct lov_mds_md_join*)lmm;
581 lsm_unpackmd_common(lsm, &lmmj->lmmj_md);
583 rc = lovea_init_array_info(lsm, &lmmj->lmmj_array_id,
584 lmmj->lmmj_extent_count);
586 CERROR("Init joined lsm id"LPU64" arrary error %d",
587 lsm->lsm_object_id, rc);
594 struct lsm_operations lsm_join_ops = {
595 .lsm_free = lsm_free_join,
596 .lsm_destroy = lsm_destroy_join,
597 .lsm_stripe_by_index = lsm_stripe_by_index_join,
598 .lsm_stripe_by_offset = lsm_stripe_by_offset_join,
599 .lsm_revalidate = lsm_revalidate_join,
600 .lsm_stripe_offset_by_index = lsm_stripe_offset_by_index_join,
601 .lsm_stripe_offset_by_offset = lsm_stripe_offset_by_offset_join,
602 .lsm_stripe_index_by_offset = lsm_stripe_index_by_offset_join,
603 .lsm_lmm_verify = lsm_lmm_verify_join,
604 .lsm_unpackmd = lsm_unpackmd_join,