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 "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;
165 struct lov_ost_data_v1 *lmm_objects;
166 int lmm_size, lmm_magic;
171 lmm_magic = lsm->lsm_magic;
173 /* If we are just sizing the EA, limit the stripe count
174 * to the actual number of OSTs in this filesystem. */
176 stripe_count = lov_get_stripecnt(lov,
177 lsm->lsm_stripe_count);
178 lsm->lsm_stripe_count = stripe_count;
180 stripe_count = lsm->lsm_stripe_count;
183 /* No needs to allocated more than LOV_MAX_STRIPE_COUNT.
184 * Anyway, this is pretty inaccurate since ld_tgt_count now
185 * represents max index and we should rely on the actual number
187 stripe_count = min(LOV_MAX_STRIPE_COUNT,
188 lov->desc.ld_tgt_count);
191 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
193 /* lsm == NULL and lmmp == NULL */
194 lmm_magic = LOV_MAGIC;
197 if ((lmm_magic != LOV_MAGIC_V1) &&
198 (lmm_magic != LOV_MAGIC_V3)) {
199 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
200 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
205 /* XXX LOV STACKING call into osc for sizes */
206 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
212 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
213 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
214 OBD_FREE(*lmmp, lmm_size);
220 OBD_ALLOC(*lmmp, lmm_size);
225 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
226 lmm_magic, lmm_size);
229 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
230 if (lmm_magic == LOV_MAGIC_V3)
231 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
233 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
238 /* lmmv1 and lmmv3 point to the same struct and have the
241 lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
242 lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
243 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
244 lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
245 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
246 if (lsm->lsm_magic == LOV_MAGIC_V3) {
247 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name, MAXPOOLNAME);
248 lmm_objects = lmmv3->lmm_objects;
250 lmm_objects = lmmv1->lmm_objects;
253 for (i = 0; i < stripe_count; i++) {
254 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
256 /* XXX LOV STACKING call down to osc_packmd() to do packing */
257 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
258 lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
259 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
260 lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
261 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
262 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
268 /* Find the max stripecount we should use */
269 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
272 stripe_count = lov->desc.ld_default_stripe_count;
273 if (stripe_count > lov->desc.ld_active_tgt_count)
274 stripe_count = lov->desc.ld_active_tgt_count;
277 /* for now, we limit the stripe count directly, when bug 4424 is
278 * fixed this needs to be somewhat dynamic based on whether ext3
279 * can handle larger EA sizes. */
280 if (stripe_count > LOV_MAX_STRIPE_COUNT)
281 stripe_count = LOV_MAX_STRIPE_COUNT;
287 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
291 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
295 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
296 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
297 sz = lmm_bytes * 2 + 1;
298 OBD_ALLOC(buffer, sz);
299 if (buffer != NULL) {
302 for (i = 0; i < lmm_bytes; i++)
303 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
305 CERROR("%s\n", buffer);
306 OBD_FREE(buffer, sz);
310 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
311 lmm_bytes, stripe_count);
315 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
316 int pattern, int magic)
321 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
323 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
325 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
329 spin_lock_init(&(*lsmp)->lsm_lock);
330 (*lsmp)->lsm_magic = magic;
331 (*lsmp)->lsm_stripe_count = stripe_count;
332 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
333 (*lsmp)->lsm_pattern = pattern;
334 (*lsmp)->lsm_pool_name[0] = '\0';
335 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
337 for (i = 0; i < stripe_count; i++)
338 loi_init((*lsmp)->lsm_oinfo[i]);
343 void lov_free_memmd(struct lov_stripe_md **lsmp)
345 struct lov_stripe_md *lsm = *lsmp;
347 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
348 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
354 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
355 * order and is opaque to the networking layer.
357 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
358 struct lov_mds_md *lmm, int lmm_bytes)
360 struct obd_device *obd = class_exp2obd(exp);
361 struct lov_obd *lov = &obd->u.lov;
362 int rc = 0, stripe_count, lsm_size;
366 /* If passed an MDS struct use values from there, otherwise defaults */
368 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
371 magic = le32_to_cpu(lmm->lmm_magic);
373 stripe_count = lov_get_stripecnt(lov, 0);
377 /* If we aren't passed an lsmp struct, we just want the size */
379 /* XXX LOV STACKING call into osc for sizes */
381 RETURN(lov_stripe_md_size(stripe_count));
383 /* If we are passed an allocated struct but nothing to unpack, free */
385 lov_free_memmd(lsmp);
389 lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
394 /* If we are passed a pointer but nothing to unpack, we only alloc */
398 LASSERT(lsm_op_find(magic) != NULL);
399 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
401 lov_free_memmd(lsmp);
408 static int __lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
409 struct lov_user_md *lump)
411 struct obd_device *obd = class_exp2obd(exp);
412 struct lov_obd *lov = &obd->u.lov;
413 struct lov_user_md_v3 lumv3;
414 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
420 rc = copy_from_user(&lumv3, lump, sizeof(struct lov_user_md_v1));
424 lmm_magic = lumv1->lmm_magic;
426 if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
427 lustre_swab_lov_user_md_v1(lumv1);
428 lmm_magic = LOV_USER_MAGIC_V1;
429 } else if (lmm_magic == LOV_USER_MAGIC_V3) {
430 rc = copy_from_user(&lumv3, lump, sizeof(lumv3));
433 } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
434 rc = copy_from_user(&lumv3, lump, sizeof(lumv3));
437 lustre_swab_lov_user_md_v3(&lumv3);
438 lmm_magic = LOV_USER_MAGIC_V3;
439 } else if (lmm_magic != LOV_USER_MAGIC_V1) {
441 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
442 lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
446 /* in the rest of the tests, as *lumv1 and lumv3 have the same
447 * fields, we use lumv1 to avoid code duplication */
449 if (lumv1->lmm_pattern == 0) {
450 lumv1->lmm_pattern = lov->desc.ld_pattern ?
451 lov->desc.ld_pattern : LOV_PATTERN_RAID0;
454 if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
455 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
460 /* 64kB is the largest common page size we see (ia64), and matches the
462 if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
463 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
464 lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
465 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
468 if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
469 (lumv1->lmm_stripe_offset !=
470 (typeof(lumv1->lmm_stripe_offset))(-1))) {
471 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
472 lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
475 stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count);
477 if (lmm_magic == LOV_USER_MAGIC_V3) {
478 struct pool_desc *pool;
480 pool = lov_find_pool(lov, lumv3.lmm_pool_name);
484 if (lumv3.lmm_stripe_offset !=
485 (typeof(lumv3.lmm_stripe_offset))(-1)) {
486 rc = lov_check_index_in_pool(lumv3.lmm_stripe_offset,
492 if (stripe_count > pool_tgt_count(pool))
493 stripe_count = pool_tgt_count(pool);
496 if ((__u64)lumv1->lmm_stripe_size * stripe_count > ~0UL) {
497 CDEBUG(D_IOCTL, "stripe width %ux%i exeeds %lu bytes\n",
498 lumv1->lmm_stripe_size, (int)lumv1->lmm_stripe_count,
503 rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
506 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
507 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
508 if (lmm_magic == LOV_USER_MAGIC_V3)
509 strncpy((*lsmp)->lsm_pool_name, lumv3.lmm_pool_name,
517 /* Configure object striping information on a new file.
519 * @lmmu is a pointer to a user struct with one or more of the fields set to
520 * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
521 * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC.
522 * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
524 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
525 struct lov_user_md *lump)
533 rc = __lov_setstripe(exp, lsmp, lump);
538 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
539 struct lov_user_md *lump)
543 struct obd_export *oexp;
544 struct lov_obd *lov = &exp->exp_obd->u.lov;
546 struct lov_user_ost_data_v1 *lmm_objects;
550 if (lump->lmm_magic == LOV_USER_MAGIC_V3)
551 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
553 lmm_objects = lump->lmm_objects;
555 for (i = 0; i < lump->lmm_stripe_count; i++) {
556 __u32 len = sizeof(last_id);
557 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
558 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
559 &len, &last_id, NULL);
562 if (lmm_objects[i].l_object_id > last_id) {
563 CERROR("Setting EA for object > than last id on "
564 "ost idx %d "LPD64" > "LPD64" \n",
565 lmm_objects[i].l_ost_idx,
566 lmm_objects[i].l_object_id, last_id);
571 rc = lov_setstripe(exp, lsmp, lump);
575 for (i = 0; i < lump->lmm_stripe_count; i++) {
576 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
577 lmm_objects[i].l_ost_idx;
578 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
579 (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr;
585 /* Retrieve object striping information.
587 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
588 * the maximum number of OST indices which will fit in the user buffer.
589 * lmm_magic must be LOV_USER_MAGIC.
591 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
592 struct lov_user_md *lump)
595 * XXX huge struct allocated on stack.
597 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
598 struct lov_user_md_v3 lum;
599 struct lov_mds_md *lmmk = NULL;
609 * "Switch to kernel segment" to allow copying from kernel space by
610 * copy_{to,from}_user().
615 /* we only need the header part from user space to get lmm_magic and
616 * lmm_stripe_count, (the header part is common to v1 and v3) */
617 lum_size = sizeof(struct lov_user_md_v1);
618 rc = copy_from_user(&lum, lump, lum_size);
622 else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
623 (lum.lmm_magic != LOV_USER_MAGIC_V3))
626 /* if v3 we just have to update the lum_size */
627 if (lum.lmm_magic == LOV_USER_MAGIC_V3)
628 lum_size = sizeof(struct lov_user_md_v3);
630 rc = lov_packmd(exp, &lmmk, lsm);
636 /* FIXME: Bug 1185 - copy fields properly when structs change */
637 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
638 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
639 CLASSERT(sizeof lum.lmm_objects[0] ==
640 sizeof lmmk->lmm_objects[0]);
642 /* User wasn't expecting this many OST entries */
643 if (lum.lmm_stripe_count == 0) {
644 if (copy_to_user(lump, lmmk, lum_size))
646 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
648 } else if (copy_to_user(lump, lmmk, lmm_size))
651 obd_free_diskmd(exp, &lmmk);