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/lov/lov_pack.c
38 * (Un)packing of OST/MDS requests
40 * Author: Andreas Dilger <adilger@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_LOV
45 #include <liblustre.h>
48 #include <lustre_net.h>
51 #include <obd_class.h>
52 #include <obd_support.h>
53 #include <lustre/lustre_user.h>
55 #include "lov_internal.h"
57 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
59 struct lov_ost_data_v1 *lod;
62 CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
63 le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
64 le32_to_cpu(lmm->lmm_pattern));
65 CDEBUG(level,"stripe_size %u, stripe_count %u\n",
66 le32_to_cpu(lmm->lmm_stripe_size),
67 le32_to_cpu(lmm->lmm_stripe_count));
69 if (le32_to_cpu(lmm->lmm_stripe_count) <= LOV_V1_INSANE_STRIPE_COUNT) {
70 for (i = 0, lod = lmm->lmm_objects;
71 i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
73 "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
74 i, le32_to_cpu(lod->l_ost_idx),
75 le64_to_cpu(lod->l_object_gr),
76 le64_to_cpu(lod->l_object_id));
78 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
79 le32_to_cpu(lmm->lmm_stripe_count),
80 LOV_V1_INSANE_STRIPE_COUNT);
84 void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
87 CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
88 le64_to_cpu(lmmj->lmmj_md.lmm_object_id),
89 le32_to_cpu(lmmj->lmmj_md.lmm_magic),
90 le32_to_cpu(lmmj->lmmj_md.lmm_pattern));
91 CDEBUG(level,"stripe_size %u, stripe_count %u extent_count %u \n",
92 le32_to_cpu(lmmj->lmmj_md.lmm_stripe_size),
93 le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count),
94 le32_to_cpu(lmmj->lmmj_extent_count));
97 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
99 struct lov_ost_data_v1 *lod;
102 CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
103 le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
104 le32_to_cpu(lmm->lmm_pattern));
105 CDEBUG(level,"stripe_size %u, stripe_count %u\n",
106 le32_to_cpu(lmm->lmm_stripe_size),
107 le32_to_cpu(lmm->lmm_stripe_count));
108 CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
110 if (le32_to_cpu(lmm->lmm_stripe_count) <= LOV_V1_INSANE_STRIPE_COUNT) {
111 for (i = 0, lod = lmm->lmm_objects;
112 i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
114 "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
115 i, le32_to_cpu(lod->l_ost_idx),
116 le64_to_cpu(lod->l_object_gr),
117 le64_to_cpu(lod->l_object_id));
119 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
120 le32_to_cpu(lmm->lmm_stripe_count),
121 LOV_V1_INSANE_STRIPE_COUNT);
125 void lov_dump_lmm(int level, void *lmm)
129 magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
132 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
134 return lov_dump_lmm_join(level, (struct lov_mds_md_join *)(lmm));
136 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
138 CERROR("Cannot recognize lmm_magic %x", magic);
143 #define LMM_ASSERT(test) \
145 if (!(test)) lov_dump_lmm(D_ERROR, lmm); \
146 LASSERT(test); /* so we know what assertion failed */ \
149 /* Pack LOV object metadata for disk storage. It is packed in LE byte
150 * order and is opaque to the networking layer.
152 * XXX In the future, this will be enhanced to get the EA size from the
153 * underlying OSC device(s) to get their EA sizes so we can stack
154 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
157 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
158 struct lov_stripe_md *lsm)
160 struct obd_device *obd = class_exp2obd(exp);
161 struct lov_obd *lov = &obd->u.lov;
162 struct lov_mds_md_v1 *lmmv1;
163 struct lov_mds_md_v3 *lmmv3;
164 struct lov_oinfo *loi;
166 struct lov_ost_data_v1 *lmm_objects;
167 int lmm_size, lmm_magic;
172 lmm_magic = lsm->lsm_magic;
175 stripe_count = lov_get_stripecnt(lov,
176 lsm->lsm_stripe_count);
177 lsm->lsm_stripe_count = stripe_count;
179 stripe_count = lsm->lsm_stripe_count;
182 /* No needs to allocated more than LOV_MAX_STRIPE_COUNT.
183 * Anyway, this is pretty inaccurate since ld_tgt_count now
184 * represents max index and we should rely on the actual number
186 stripe_count = min((__u32)LOV_MAX_STRIPE_COUNT,
187 lov->desc.ld_tgt_count);
190 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
192 /* lsm == NULL and lmmp == NULL */
193 lmm_magic = LOV_MAGIC;
196 if ((lmm_magic != LOV_MAGIC_V1) &&
197 (lmm_magic != LOV_MAGIC_V3)) {
198 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
199 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
203 /* XXX LOV STACKING call into osc for sizes */
204 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
210 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
211 lmm_size = lov_mds_md_size(stripe_count, le32_to_cpu((*lmmp)->lmm_magic));
212 OBD_FREE(*lmmp, lmm_size);
218 OBD_ALLOC(*lmmp, lmm_size);
223 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
224 lmm_magic, lmm_size);
227 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
228 if (lmm_magic == LOV_MAGIC_V3)
229 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
231 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
236 /* lmmv1 and lmmv3 point to the same struct and have the
239 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
240 lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
241 lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
242 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
243 lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
244 if (lsm->lsm_magic == LOV_MAGIC_V3) {
245 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name, LOV_MAXPOOLNAME);
246 lmm_objects = lmmv3->lmm_objects;
248 lmm_objects = lmmv1->lmm_objects;
251 for (i = 0; i < stripe_count; i++) {
252 loi = lsm->lsm_oinfo[i];
253 /* XXX LOV STACKING call down to osc_packmd() to do packing */
254 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
255 lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
256 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
257 lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
258 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
259 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
265 /* Find the max stripecount we should use */
266 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
269 stripe_count = lov->desc.ld_default_stripe_count;
270 if (stripe_count > lov->desc.ld_active_tgt_count)
271 stripe_count = lov->desc.ld_active_tgt_count;
274 /* for now, we limit the stripe count directly, when bug 4424 is
275 * fixed this needs to be somewhat dynamic based on whether ext3
276 * can handle larger EA sizes. */
277 if (stripe_count > LOV_MAX_STRIPE_COUNT)
278 stripe_count = LOV_MAX_STRIPE_COUNT;
284 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
288 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
292 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
293 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
294 sz = lmm_bytes * 2 + 1;
295 OBD_ALLOC(buffer, sz);
296 if (buffer != NULL) {
299 for (i = 0; i < lmm_bytes; i++)
300 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
302 CERROR("%s\n", buffer);
303 OBD_FREE(buffer, sz);
307 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
308 lmm_bytes, stripe_count);
312 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
313 int pattern, int magic)
318 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
320 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
322 CERROR("can't allocate lsmp, stripe_count %d\n", stripe_count);
326 spin_lock_init(&(*lsmp)->lsm_lock);
327 (*lsmp)->lsm_magic = magic;
328 (*lsmp)->lsm_stripe_count = stripe_count;
329 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
330 (*lsmp)->lsm_pattern = pattern;
331 (*lsmp)->lsm_pool_name[0] = '\0';
332 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
334 for (i = 0; i < stripe_count; i++)
335 loi_init((*lsmp)->lsm_oinfo[i]);
340 void lov_free_memmd(struct lov_stripe_md **lsmp)
342 struct lov_stripe_md *lsm = *lsmp;
344 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
345 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
351 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
352 * order and is opaque to the networking layer.
354 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
355 struct lov_mds_md *lmm, int lmm_bytes)
357 struct obd_device *obd = class_exp2obd(exp);
358 struct lov_obd *lov = &obd->u.lov;
359 int rc = 0, stripe_count, lsm_size;
363 /* If passed an MDS struct use values from there, otherwise defaults */
365 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
368 magic = le32_to_cpu(lmm->lmm_magic);
370 stripe_count = lov_get_stripecnt(lov, 0);
374 /* If we aren't passed an lsmp struct, we just want the size */
376 /* XXX LOV STACKING call into osc for sizes */
378 RETURN(lov_stripe_md_size(stripe_count));
381 /* If we are passed an allocated struct but nothing to unpack, free */
383 lov_free_memmd(lsmp);
387 lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
392 /* If we are passed a pointer but nothing to unpack, we only alloc */
396 LASSERT(lsm_op_find(magic) != NULL);
397 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
399 lov_free_memmd(lsmp);
406 /* Configure object striping information on a new file.
408 * @lmmu is a pointer to a user struct with one or more of the fields set to
409 * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
410 * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC.
411 * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
413 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
414 struct lov_user_md *lump)
416 struct obd_device *obd = class_exp2obd(exp);
417 struct lov_obd *lov = &obd->u.lov;
418 struct lov_user_md_v3 lumv3;
419 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
425 rc = copy_from_user(&lumv3, lump, sizeof(struct lov_user_md_v1));
429 lmm_magic = lumv1->lmm_magic;
431 if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
432 lustre_swab_lov_user_md_v1(lumv1);
433 lmm_magic = LOV_USER_MAGIC_V1;
434 } else if (lmm_magic == LOV_USER_MAGIC_V3) {
435 rc = copy_from_user(&lumv3, lump, sizeof(lumv3));
438 } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
439 rc = copy_from_user(&lumv3, lump, sizeof(lumv3));
442 lustre_swab_lov_user_md_v3(&lumv3);
443 lmm_magic = LOV_USER_MAGIC_V3;
444 } else if (lmm_magic != LOV_USER_MAGIC_V1) {
446 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
447 lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
451 /* in the rest of the tests, as *lumv1 and lumv3 have the same
452 * fields, we use lumv1 to avoid code duplication */
454 if (lumv1->lmm_pattern == 0) {
455 lumv1->lmm_pattern = lov->desc.ld_pattern ?
456 lov->desc.ld_pattern : LOV_PATTERN_RAID0;
459 if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
460 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
465 /* 64kB is the largest common page size we see (ia64), and matches the
467 if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
468 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
469 lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
470 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
473 if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
474 (lumv1->lmm_stripe_offset !=
475 (typeof(lumv1->lmm_stripe_offset))(-1))) {
476 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
477 lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
481 stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count);
483 if (lmm_magic == LOV_USER_MAGIC_V3) {
484 struct pool_desc *pool;
486 pool = lov_find_pool(lov, lumv3.lmm_pool_name);
488 if (lumv1->lmm_stripe_offset !=
489 (typeof(lumv1->lmm_stripe_offset))(-1)) {
490 rc = lov_check_index_in_pool(
491 lumv1->lmm_stripe_offset, pool);
493 lov_pool_putref(pool);
498 if (stripe_count > pool_tgt_count(pool))
499 stripe_count = pool_tgt_count(pool);
501 lov_pool_putref(pool);
505 rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
510 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
511 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
513 if (lmm_magic == LOV_USER_MAGIC_V3)
514 strncpy((*lsmp)->lsm_pool_name, lumv3.lmm_pool_name,
520 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
521 struct lov_user_md *lump)
525 struct obd_export *oexp;
526 struct lov_obd *lov = &exp->exp_obd->u.lov;
528 struct lov_user_ost_data_v1 *lmm_objects;
532 if (lump->lmm_magic == LOV_USER_MAGIC_V3)
533 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
535 lmm_objects = lump->lmm_objects;
537 for (i = 0; i < lump->lmm_stripe_count; i++) {
538 __u32 len = sizeof(last_id);
539 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
540 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
541 &len, &last_id, NULL);
544 if (lmm_objects[i].l_object_id > last_id) {
545 CERROR("Setting EA for object > than last id on "
546 "ost idx %d "LPD64" > "LPD64" \n",
547 lmm_objects[i].l_ost_idx,
548 lmm_objects[i].l_object_id, last_id);
553 rc = lov_setstripe(exp, lsmp, lump);
557 for (i = 0; i < lump->lmm_stripe_count; i++) {
558 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
559 lmm_objects[i].l_ost_idx;
560 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
561 (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr;
567 /* Retrieve object striping information.
569 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
570 * the maximum number of OST indices which will fit in the user buffer.
571 * lmm_magic must be LOV_USER_MAGIC.
573 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
574 struct lov_user_md *lump)
576 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
577 struct lov_user_md_v3 lum;
578 struct lov_mds_md *lmmk = NULL;
585 /* we only need the header part from user space to get lmm_magic and
586 * lmm_stripe_count, (the header part is common to v1 and v3) */
587 lum_size = sizeof(struct lov_user_md_v1);
588 rc = copy_from_user(&lum, lump, lum_size);
591 /* if v3 we just have to update the lum_size */
592 if (lum.lmm_magic == LOV_USER_MAGIC_V3)
593 lum_size = sizeof(struct lov_user_md_v3);
595 if ((lum.lmm_magic != LOV_USER_MAGIC_V1) &&
596 (lum.lmm_magic != LOV_USER_MAGIC_V3))
599 if (lum.lmm_stripe_count &&
600 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
601 /* Return right size of stripe to user */
602 lum.lmm_stripe_count = lsm->lsm_stripe_count;
603 rc = copy_to_user(lump, &lum, lum_size);
606 rc = lov_packmd(exp, &lmmk, lsm);
612 /* FIXME: Bug 1185 - copy fields properly when structs change */
613 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
614 LASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
615 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
617 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
618 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
619 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
620 lustre_swab_lov_mds_md(lmmk);
621 lustre_swab_lov_user_md_objects(
622 (struct lov_user_ost_data*)lmmk->lmm_objects,
623 lmmk->lmm_stripe_count);
625 if (lum.lmm_magic == LOV_USER_MAGIC) {
626 /* User request for v1, we need skip lmm_pool_name */
627 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
628 memmove((char*)(&lmmk->lmm_stripe_count) +
629 sizeof(lmmk->lmm_stripe_count),
630 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
631 lmmk->lmm_stripe_count *
632 sizeof(struct lov_ost_data_v1));
633 lmm_size -= LOV_MAXPOOLNAME;
636 /* if v3 we just have to update the lum_size */
637 lum_size = sizeof(struct lov_user_md_v3);
639 /* User wasn't expecting this many OST entries */
640 if (lum.lmm_stripe_count == 0)
642 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
645 * Have a difference between lov_mds_md & lov_user_md.
646 * So we have to re-order the data before copy to user.
648 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
649 ((struct lov_user_md*)lmmk)->lmm_stripe_offset = 0;
650 ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
651 if (copy_to_user(lump, lmmk, lmm_size))
654 obd_free_diskmd(exp, &lmmk);