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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/lov/lov_pack.c
40 * (Un)packing of OST/MDS requests
42 * Author: Andreas Dilger <adilger@clusterfs.com>
45 #define DEBUG_SUBSYSTEM S_LOV
47 #include <liblustre.h>
50 #include <lustre_net.h>
53 #include <obd_class.h>
54 #include <obd_support.h>
55 #include <lustre/lustre_user.h>
57 #include "lov_internal.h"
59 static void lov_dump_lmm_common(int level, void *lmmp)
61 struct lov_mds_md *lmm = lmmp;
63 CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
64 (__u64)le64_to_cpu(lmm->lmm_object_id),
65 le32_to_cpu(lmm->lmm_magic),
66 le32_to_cpu(lmm->lmm_pattern));
67 CDEBUG(level,"stripe_size %u, stripe_count %u, layout_gen %u\n",
68 le32_to_cpu(lmm->lmm_stripe_size),
69 le16_to_cpu(lmm->lmm_stripe_count),
70 le16_to_cpu(lmm->lmm_layout_gen));
73 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
78 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
79 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
80 stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
83 for (i = 0; i < stripe_count; ++i, ++lod) {
84 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n", i,
85 le32_to_cpu(lod->l_ost_idx),
86 (__u64)le64_to_cpu(lod->l_object_seq),
87 (__u64)le64_to_cpu(lod->l_object_id));
91 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
93 lov_dump_lmm_common(level, lmm);
94 lov_dump_lmm_objects(level, lmm->lmm_objects,
95 le16_to_cpu(lmm->lmm_stripe_count));
98 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
100 lov_dump_lmm_common(level, lmm);
101 CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
102 lov_dump_lmm_objects(level, lmm->lmm_objects,
103 le16_to_cpu(lmm->lmm_stripe_count));
106 void lov_dump_lmm(int level, void *lmm)
110 magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
113 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
115 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
117 CERROR("Cannot recognize lmm_magic %x", magic);
122 #define LMM_ASSERT(test) \
124 if (!(test)) lov_dump_lmm(D_ERROR, lmm); \
125 LASSERT(test); /* so we know what assertion failed */ \
128 /* Pack LOV object metadata for disk storage. It is packed in LE byte
129 * order and is opaque to the networking layer.
131 * XXX In the future, this will be enhanced to get the EA size from the
132 * underlying OSC device(s) to get their EA sizes so we can stack
133 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
136 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
137 struct lov_stripe_md *lsm)
139 struct obd_device *obd = class_exp2obd(exp);
140 struct lov_obd *lov = &obd->u.lov;
141 struct lov_mds_md_v1 *lmmv1;
142 struct lov_mds_md_v3 *lmmv3;
144 struct lov_ost_data_v1 *lmm_objects;
145 int lmm_size, lmm_magic;
150 lmm_magic = lsm->lsm_magic;
153 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
155 /* lsm == NULL and lmmp == NULL */
156 lmm_magic = LOV_MAGIC;
159 if ((lmm_magic != LOV_MAGIC_V1) &&
160 (lmm_magic != LOV_MAGIC_V3)) {
161 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
162 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
168 /* If we are just sizing the EA, limit the stripe count
169 * to the actual number of OSTs in this filesystem. */
171 stripe_count = lov_get_stripecnt(lov, lmm_magic,
172 lsm->lsm_stripe_count);
173 lsm->lsm_stripe_count = stripe_count;
175 stripe_count = lsm->lsm_stripe_count;
178 /* No need to allocate more than maximum supported stripes.
179 * Anyway, this is pretty inaccurate since ld_tgt_count now
180 * represents max index and we should rely on the actual number
182 stripe_count = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
184 if (stripe_count > lov->desc.ld_tgt_count)
185 stripe_count = lov->desc.ld_tgt_count;
188 /* XXX LOV STACKING call into osc for sizes */
189 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
195 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
196 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
197 OBD_FREE_LARGE(*lmmp, lmm_size);
203 OBD_ALLOC_LARGE(*lmmp, lmm_size);
208 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
209 lmm_magic, lmm_size);
212 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
213 if (lmm_magic == LOV_MAGIC_V3)
214 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
216 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
221 /* lmmv1 and lmmv3 point to the same struct and have the
224 lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
225 lmmv1->lmm_object_seq = cpu_to_le64(lsm->lsm_object_seq);
226 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
227 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
228 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
229 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
230 if (lsm->lsm_magic == LOV_MAGIC_V3) {
231 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
233 lmm_objects = lmmv3->lmm_objects;
235 lmm_objects = lmmv1->lmm_objects;
238 for (i = 0; i < stripe_count; i++) {
239 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
240 /* XXX LOV STACKING call down to osc_packmd() to do packing */
241 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
242 lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
243 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
244 lmm_objects[i].l_object_seq = cpu_to_le64(loi->loi_seq);
245 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
246 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
252 /* Find the max stripecount we should use */
253 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
255 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
258 stripe_count = lov->desc.ld_default_stripe_count;
259 if (stripe_count > lov->desc.ld_active_tgt_count)
260 stripe_count = lov->desc.ld_active_tgt_count;
264 /* stripe count is based on whether ldiskfs can handle
266 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
267 lov->lov_ocd.ocd_max_easize)
268 max_stripes = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
271 if (stripe_count > max_stripes)
272 stripe_count = max_stripes;
278 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
282 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
286 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
287 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
288 sz = lmm_bytes * 2 + 1;
289 OBD_ALLOC_LARGE(buffer, sz);
290 if (buffer != NULL) {
293 for (i = 0; i < lmm_bytes; i++)
294 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
296 CERROR("%s\n", buffer);
297 OBD_FREE_LARGE(buffer, sz);
301 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
302 lmm_bytes, stripe_count);
306 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
307 int pattern, int magic)
312 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
314 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
316 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
320 cfs_spin_lock_init(&(*lsmp)->lsm_lock);
321 (*lsmp)->lsm_magic = magic;
322 (*lsmp)->lsm_stripe_count = stripe_count;
323 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
324 (*lsmp)->lsm_pattern = pattern;
325 (*lsmp)->lsm_pool_name[0] = '\0';
326 (*lsmp)->lsm_layout_gen = 0;
327 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
329 for (i = 0; i < stripe_count; i++)
330 loi_init((*lsmp)->lsm_oinfo[i]);
335 void lov_free_memmd(struct lov_stripe_md **lsmp)
337 struct lov_stripe_md *lsm = *lsmp;
339 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
340 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
346 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
347 * order and is opaque to the networking layer.
349 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
350 struct lov_mds_md *lmm, int lmm_bytes)
352 struct obd_device *obd = class_exp2obd(exp);
353 struct lov_obd *lov = &obd->u.lov;
354 int rc = 0, lsm_size;
359 /* If passed an MDS struct use values from there, otherwise defaults */
361 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
364 magic = le32_to_cpu(lmm->lmm_magic);
367 stripe_count = lov_get_stripecnt(lov, magic, 0);
370 /* If we aren't passed an lsmp struct, we just want the size */
372 /* XXX LOV STACKING call into osc for sizes */
374 RETURN(lov_stripe_md_size(stripe_count));
376 /* If we are passed an allocated struct but nothing to unpack, free */
378 lov_free_memmd(lsmp);
382 lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
387 /* If we are passed a pointer but nothing to unpack, we only alloc */
391 LASSERT(lsm_op_find(magic) != NULL);
392 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
394 lov_free_memmd(lsmp);
401 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
402 struct lov_stripe_md **lsmp,
403 struct lov_user_md *lump)
405 struct obd_device *obd = class_exp2obd(exp);
406 struct lov_obd *lov = &obd->u.lov;
407 char buffer[sizeof(struct lov_user_md_v3)];
408 struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0];
409 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0];
415 rc = lov_lum_swab_if_needed(lumv3, &lmm_magic, lump);
419 /* in the rest of the tests, as *lumv1 and lumv3 have the same
420 * fields, we use lumv1 to avoid code duplication */
422 if (lumv1->lmm_pattern == 0) {
423 lumv1->lmm_pattern = lov->desc.ld_pattern ?
424 lov->desc.ld_pattern : LOV_PATTERN_RAID0;
427 if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
428 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
433 /* 64kB is the largest common page size we see (ia64), and matches the
435 if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
436 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
437 lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
438 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
441 if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
442 (lumv1->lmm_stripe_offset !=
443 (typeof(lumv1->lmm_stripe_offset))(-1))) {
444 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
445 lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
448 stripe_count = lov_get_stripecnt(lov, lmm_magic,
449 lumv1->lmm_stripe_count);
452 int max_stripes = (max_lmm_size -
453 lov_mds_md_size(0, lmm_magic)) /
454 sizeof(struct lov_ost_data_v1);
455 if (unlikely(max_stripes < stripe_count)) {
456 CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
457 stripe_count, max_stripes);
458 stripe_count = max_stripes;
462 if (lmm_magic == LOV_USER_MAGIC_V3) {
463 struct pool_desc *pool;
465 pool = lov_find_pool(lov, lumv3->lmm_pool_name);
467 if (lumv3->lmm_stripe_offset !=
468 (typeof(lumv3->lmm_stripe_offset))(-1)) {
469 rc = lov_check_index_in_pool(
470 lumv3->lmm_stripe_offset, pool);
472 lov_pool_putref(pool);
477 if (stripe_count > pool_tgt_count(pool))
478 stripe_count = pool_tgt_count(pool);
480 lov_pool_putref(pool);
484 rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
487 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
488 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
489 if (lmm_magic == LOV_USER_MAGIC_V3)
490 strncpy((*lsmp)->lsm_pool_name, lumv3->lmm_pool_name,
498 /* Configure object striping information on a new file.
500 * @lmmu is a pointer to a user struct with one or more of the fields set to
501 * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
502 * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC.
503 * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
505 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
506 struct lov_stripe_md **lsmp, struct lov_user_md *lump)
514 rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
519 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
520 struct lov_user_md *lump)
524 struct obd_export *oexp;
525 struct lov_obd *lov = &exp->exp_obd->u.lov;
527 struct lov_user_ost_data_v1 *lmm_objects;
531 if (lump->lmm_magic == LOV_USER_MAGIC_V3)
532 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
534 lmm_objects = lump->lmm_objects;
536 for (i = 0; i < lump->lmm_stripe_count; i++) {
537 __u32 len = sizeof(last_id);
538 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
539 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
540 &len, &last_id, NULL);
543 if (lmm_objects[i].l_object_id > last_id) {
544 CERROR("Setting EA for object > than last id on "
545 "ost idx %d "LPD64" > "LPD64" \n",
546 lmm_objects[i].l_ost_idx,
547 lmm_objects[i].l_object_id, last_id);
552 rc = lov_setstripe(exp, 0, lsmp, lump);
556 for (i = 0; i < lump->lmm_stripe_count; i++) {
557 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
558 lmm_objects[i].l_ost_idx;
559 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
560 (*lsmp)->lsm_oinfo[i]->loi_seq = lmm_objects[i].l_object_seq;
566 /* Retrieve object striping information.
568 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
569 * the maximum number of OST indices which will fit in the user buffer.
570 * lmm_magic must be LOV_USER_MAGIC.
572 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
573 struct lov_user_md *lump)
576 * XXX huge struct allocated on stack.
578 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
579 struct lov_user_md_v3 lum;
580 struct lov_mds_md *lmmk = NULL;
590 * "Switch to kernel segment" to allow copying from kernel space by
591 * copy_{to,from}_user().
596 /* we only need the header part from user space to get lmm_magic and
597 * lmm_stripe_count, (the header part is common to v1 and v3) */
598 lum_size = sizeof(struct lov_user_md_v1);
599 if (cfs_copy_from_user(&lum, lump, lum_size))
600 GOTO(out_set, rc = -EFAULT);
601 else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
602 (lum.lmm_magic != LOV_USER_MAGIC_V3))
603 GOTO(out_set, rc = -EINVAL);
605 if (lum.lmm_stripe_count &&
606 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
607 /* Return right size of stripe to user */
608 lum.lmm_stripe_count = lsm->lsm_stripe_count;
609 rc = cfs_copy_to_user(lump, &lum, lum_size);
610 GOTO(out_set, rc = -EOVERFLOW);
612 rc = lov_packmd(exp, &lmmk, lsm);
618 /* FIXME: Bug 1185 - copy fields properly when structs change */
619 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
620 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
621 CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
623 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
624 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
625 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
626 lustre_swab_lov_mds_md(lmmk);
627 lustre_swab_lov_user_md_objects(
628 (struct lov_user_ost_data*)lmmk->lmm_objects,
629 lmmk->lmm_stripe_count);
631 if (lum.lmm_magic == LOV_USER_MAGIC) {
632 /* User request for v1, we need skip lmm_pool_name */
633 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
634 memmove((char*)(&lmmk->lmm_stripe_count) +
635 sizeof(lmmk->lmm_stripe_count),
636 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
637 lmmk->lmm_stripe_count *
638 sizeof(struct lov_ost_data_v1));
639 lmm_size -= LOV_MAXPOOLNAME;
642 /* if v3 we just have to update the lum_size */
643 lum_size = sizeof(struct lov_user_md_v3);
646 /* User wasn't expecting this many OST entries */
647 if (lum.lmm_stripe_count == 0)
649 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
650 GOTO(out_set, rc = -EOVERFLOW);
652 * Have a difference between lov_mds_md & lov_user_md.
653 * So we have to re-order the data before copy to user.
655 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
656 lum.u.lum_layout_gen = lmmk->lmm_layout_gen;
657 ((struct lov_user_md*)lmmk)->u.lum_layout_gen = lum.u.lum_layout_gen;
658 ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
659 if (cfs_copy_to_user(lump, lmmk, lmm_size))
662 obd_free_diskmd(exp, &lmmk);