4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
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 "../include/lustre_net.h"
46 #include "../include/obd.h"
47 #include "../include/obd_class.h"
48 #include "../include/obd_support.h"
49 #include "../include/lustre/lustre_user.h"
51 #include "lov_internal.h"
53 void lov_dump_lmm_common(int level, void *lmmp)
55 struct lov_mds_md *lmm = lmmp;
58 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
59 CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
60 POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
61 le32_to_cpu(lmm->lmm_pattern));
62 CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
63 le32_to_cpu(lmm->lmm_stripe_size),
64 le16_to_cpu(lmm->lmm_stripe_count),
65 le16_to_cpu(lmm->lmm_layout_gen));
68 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
73 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
74 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
75 stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
79 for (i = 0; i < stripe_count; ++i, ++lod) {
82 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
83 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
84 le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
88 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
90 lov_dump_lmm_common(level, lmm);
91 lov_dump_lmm_objects(level, lmm->lmm_objects,
92 le16_to_cpu(lmm->lmm_stripe_count));
95 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
97 lov_dump_lmm_common(level, lmm);
98 CDEBUG(level, "pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
99 lov_dump_lmm_objects(level, lmm->lmm_objects,
100 le16_to_cpu(lmm->lmm_stripe_count));
103 /* Pack LOV object metadata for disk storage. It is packed in LE byte
104 * order and is opaque to the networking layer.
106 * XXX In the future, this will be enhanced to get the EA size from the
107 * underlying OSC device(s) to get their EA sizes so we can stack
108 * LOVs properly. For now lov_mds_md_size() just assumes one u64
111 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
112 struct lov_stripe_md *lsm)
114 struct obd_device *obd = class_exp2obd(exp);
115 struct lov_obd *lov = &obd->u.lov;
116 struct lov_mds_md_v1 *lmmv1;
117 struct lov_mds_md_v3 *lmmv3;
119 struct lov_ost_data_v1 *lmm_objects;
120 int lmm_size, lmm_magic;
125 lmm_magic = lsm->lsm_magic;
128 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
130 /* lsm == NULL and lmmp == NULL */
131 lmm_magic = LOV_MAGIC;
134 if ((lmm_magic != LOV_MAGIC_V1) &&
135 (lmm_magic != LOV_MAGIC_V3)) {
136 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
137 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
142 /* If we are just sizing the EA, limit the stripe count
143 * to the actual number of OSTs in this filesystem.
146 stripe_count = lov_get_stripecnt(lov, lmm_magic,
147 lsm->lsm_stripe_count);
148 lsm->lsm_stripe_count = stripe_count;
149 } else if (!lsm_is_released(lsm)) {
150 stripe_count = lsm->lsm_stripe_count;
155 /* No need to allocate more than maximum supported stripes.
156 * Anyway, this is pretty inaccurate since ld_tgt_count now
157 * represents max index and we should rely on the actual number
160 stripe_count = lov_mds_md_max_stripe_count(
161 lov->lov_ocd.ocd_max_easize, lmm_magic);
163 if (stripe_count > lov->desc.ld_tgt_count)
164 stripe_count = lov->desc.ld_tgt_count;
167 /* XXX LOV STACKING call into osc for sizes */
168 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
174 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
175 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
182 *lmmp = libcfs_kvzalloc(lmm_size, GFP_NOFS);
187 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d\n",
188 lmm_magic, lmm_size);
191 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
192 if (lmm_magic == LOV_MAGIC_V3)
193 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
195 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
200 /* lmmv1 and lmmv3 point to the same struct and have the
203 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
204 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
205 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
206 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
207 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
208 if (lsm->lsm_magic == LOV_MAGIC_V3) {
209 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
210 sizeof(lmmv3->lmm_pool_name));
211 if (cplen >= sizeof(lmmv3->lmm_pool_name))
213 lmm_objects = lmmv3->lmm_objects;
215 lmm_objects = lmmv1->lmm_objects;
218 for (i = 0; i < stripe_count; i++) {
219 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
220 /* XXX LOV STACKING call down to osc_packmd() to do packing */
221 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
222 " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
223 i, stripe_count, loi->loi_ost_idx);
224 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
225 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
226 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
232 /* Find the max stripecount we should use */
233 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
235 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
238 stripe_count = lov->desc.ld_default_stripe_count;
239 if (stripe_count > lov->desc.ld_active_tgt_count)
240 stripe_count = lov->desc.ld_active_tgt_count;
244 /* stripe count is based on whether ldiskfs can handle
247 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
248 lov->lov_ocd.ocd_max_easize)
249 max_stripes = lov_mds_md_max_stripe_count(
250 lov->lov_ocd.ocd_max_easize, magic);
252 if (stripe_count > max_stripes)
253 stripe_count = max_stripes;
258 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
262 if (!lsm_op_find(le32_to_cpu(*(__u32 *)lmm))) {
263 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
264 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
265 CERROR("%*phN\n", lmm_bytes, lmm);
268 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
274 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
275 int pattern, int magic)
279 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
281 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
283 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
287 atomic_set(&(*lsmp)->lsm_refc, 1);
288 spin_lock_init(&(*lsmp)->lsm_lock);
289 (*lsmp)->lsm_magic = magic;
290 (*lsmp)->lsm_stripe_count = stripe_count;
291 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
292 (*lsmp)->lsm_pattern = pattern;
293 (*lsmp)->lsm_pool_name[0] = '\0';
294 (*lsmp)->lsm_layout_gen = 0;
295 if (stripe_count > 0)
296 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
298 for (i = 0; i < stripe_count; i++)
299 loi_init((*lsmp)->lsm_oinfo[i]);
304 int lov_free_memmd(struct lov_stripe_md **lsmp)
306 struct lov_stripe_md *lsm = *lsmp;
310 LASSERT(atomic_read(&lsm->lsm_refc) > 0);
311 refc = atomic_dec_return(&lsm->lsm_refc);
313 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
318 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
319 * order and is opaque to the networking layer.
321 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
322 struct lov_mds_md *lmm, int lmm_bytes)
324 struct obd_device *obd = class_exp2obd(exp);
325 struct lov_obd *lov = &obd->u.lov;
326 int rc = 0, lsm_size;
331 /* If passed an MDS struct use values from there, otherwise defaults */
333 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
336 magic = le32_to_cpu(lmm->lmm_magic);
337 pattern = le32_to_cpu(lmm->lmm_pattern);
340 stripe_count = lov_get_stripecnt(lov, magic, 0);
341 pattern = LOV_PATTERN_RAID0;
344 /* If we aren't passed an lsmp struct, we just want the size */
346 /* XXX LOV STACKING call into osc for sizes */
348 return lov_stripe_md_size(stripe_count);
350 /* If we are passed an allocated struct but nothing to unpack, free */
352 lov_free_memmd(lsmp);
356 lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
360 /* If we are passed a pointer but nothing to unpack, we only alloc */
364 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
366 lov_free_memmd(lsmp);
373 /* Retrieve object striping information.
375 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
376 * the maximum number of OST indices which will fit in the user buffer.
377 * lmm_magic must be LOV_USER_MAGIC.
379 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
380 struct lov_user_md __user *lump)
383 * XXX huge struct allocated on stack.
385 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
386 struct lov_user_md_v3 lum;
387 struct lov_mds_md *lmmk = NULL;
396 * "Switch to kernel segment" to allow copying from kernel space by
397 * copy_{to,from}_user().
402 /* we only need the header part from user space to get lmm_magic and
403 * lmm_stripe_count, (the header part is common to v1 and v3)
405 lum_size = sizeof(struct lov_user_md_v1);
406 if (copy_from_user(&lum, lump, lum_size)) {
410 if ((lum.lmm_magic != LOV_USER_MAGIC) &&
411 (lum.lmm_magic != LOV_USER_MAGIC_V3)) {
416 if (lum.lmm_stripe_count &&
417 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
418 /* Return right size of stripe to user */
419 lum.lmm_stripe_count = lsm->lsm_stripe_count;
420 rc = copy_to_user(lump, &lum, lum_size);
424 rc = lov_packmd(exp, &lmmk, lsm);
430 /* FIXME: Bug 1185 - copy fields properly when structs change */
431 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
432 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
433 CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
435 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
436 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
437 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
438 lustre_swab_lov_mds_md(lmmk);
439 lustre_swab_lov_user_md_objects(
440 (struct lov_user_ost_data *)lmmk->lmm_objects,
441 lmmk->lmm_stripe_count);
443 if (lum.lmm_magic == LOV_USER_MAGIC) {
444 /* User request for v1, we need skip lmm_pool_name */
445 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
446 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
447 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
448 lmmk->lmm_stripe_count *
449 sizeof(struct lov_ost_data_v1));
450 lmm_size -= LOV_MAXPOOLNAME;
453 /* if v3 we just have to update the lum_size */
454 lum_size = sizeof(struct lov_user_md_v3);
457 /* User wasn't expecting this many OST entries */
458 if (lum.lmm_stripe_count == 0) {
460 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
465 * Have a difference between lov_mds_md & lov_user_md.
466 * So we have to re-order the data before copy to user.
468 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
469 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
470 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
471 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
472 if (copy_to_user(lump, lmmk, lmm_size))
475 obd_free_diskmd(exp, &lmmk);