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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/lov/lov_pack.c
34 * (Un)packing of OST/MDS requests
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LOV
41 #include <lustre_net.h>
42 #include <lustre_swab.h>
44 #include <obd_class.h>
45 #include <obd_support.h>
47 #include "lov_cl_internal.h"
48 #include "lov_internal.h"
50 void lov_dump_lmm_common(int level, void *lmmp)
52 struct lov_mds_md *lmm = lmmp;
55 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
56 CDEBUG(level, "objid " DOSTID ", magic 0x%08x, pattern %#x\n",
57 POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
58 le32_to_cpu(lmm->lmm_pattern));
59 CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
60 le32_to_cpu(lmm->lmm_stripe_size),
61 le16_to_cpu(lmm->lmm_stripe_count),
62 le16_to_cpu(lmm->lmm_layout_gen));
65 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
70 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
71 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
72 stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
76 for (i = 0; i < stripe_count; ++i, ++lod) {
79 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
80 CDEBUG(level, "stripe %u idx %u subobj " DOSTID "\n", i,
81 le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
85 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
87 lov_dump_lmm_common(level, lmm);
88 lov_dump_lmm_objects(level, lmm->lmm_objects,
89 le16_to_cpu(lmm->lmm_stripe_count));
92 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
94 lov_dump_lmm_common(level, lmm);
95 CDEBUG(level, "pool_name " LOV_POOLNAMEF "\n", lmm->lmm_pool_name);
96 lov_dump_lmm_objects(level, lmm->lmm_objects,
97 le16_to_cpu(lmm->lmm_stripe_count));
101 * Pack LOV striping metadata for disk storage format (in little
102 * endian byte order).
104 * This follows the getxattr() conventions. If \a buf_size is zero
105 * then return the size needed. If \a buf_size is too small then
106 * return -ERANGE. Otherwise return the size of the result.
108 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
111 struct lov_ost_data_v1 *lmm_objects;
112 struct lov_mds_md_v1 *lmmv1 = buf;
113 struct lov_mds_md_v3 *lmmv3 = buf;
117 lmm_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
121 if (buf_size < lmm_size)
125 * lmmv1 and lmmv3 point to the same struct and have the
128 lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
129 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
130 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
131 lmmv1->lmm_stripe_count = cpu_to_le16(lsm->lsm_stripe_count);
132 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
133 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
135 if (lsm->lsm_magic == LOV_MAGIC_V3) {
136 BUILD_BUG_ON(sizeof(lsm->lsm_pool_name) !=
137 sizeof(lmmv3->lmm_pool_name));
138 strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
139 sizeof(lmmv3->lmm_pool_name));
140 lmm_objects = lmmv3->lmm_objects;
142 lmm_objects = lmmv1->lmm_objects;
145 for (i = 0; i < lsm->lsm_stripe_count; i++) {
146 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
148 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
149 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
150 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
156 /* Find the max stripecount we should use */
157 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
159 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
162 stripe_count = lov->desc.ld_default_stripe_count;
163 if (stripe_count > lov->desc.ld_active_tgt_count)
164 stripe_count = lov->desc.ld_active_tgt_count;
168 /* stripe count is based on whether ldiskfs can handle
171 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
172 lov->lov_ocd.ocd_max_easize)
173 max_stripes = lov_mds_md_max_stripe_count(
174 lov->lov_ocd.ocd_max_easize, magic);
176 if (stripe_count > max_stripes)
177 stripe_count = max_stripes;
182 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
186 if (!lsm_op_find(le32_to_cpu(*(__u32 *)lmm))) {
187 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
188 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
189 CERROR("%*phN\n", lmm_bytes, lmm);
192 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
198 static struct lov_stripe_md *lov_lsm_alloc(u16 stripe_count, u32 pattern,
201 struct lov_stripe_md *lsm;
204 CDEBUG(D_INFO, "alloc lsm, stripe_count %u\n", stripe_count);
206 lsm = lsm_alloc_plain(stripe_count);
208 CERROR("cannot allocate LSM stripe_count %u\n", stripe_count);
209 return ERR_PTR(-ENOMEM);
212 atomic_set(&lsm->lsm_refc, 1);
213 spin_lock_init(&lsm->lsm_lock);
214 lsm->lsm_magic = magic;
215 lsm->lsm_stripe_count = stripe_count;
216 lsm->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
217 lsm->lsm_pattern = pattern;
218 lsm->lsm_pool_name[0] = '\0';
219 lsm->lsm_layout_gen = 0;
220 if (stripe_count > 0)
221 lsm->lsm_oinfo[0]->loi_ost_idx = ~0;
223 for (i = 0; i < stripe_count; i++)
224 loi_init(lsm->lsm_oinfo[i]);
229 int lov_free_memmd(struct lov_stripe_md **lsmp)
231 struct lov_stripe_md *lsm = *lsmp;
235 LASSERT(atomic_read(&lsm->lsm_refc) > 0);
236 refc = atomic_dec_return(&lsm->lsm_refc);
238 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
243 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
244 * order and is opaque to the networking layer.
246 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, struct lov_mds_md *lmm,
249 struct lov_stripe_md *lsm;
255 rc = lov_verify_lmm(lmm, lmm_size, &stripe_count);
259 magic = le32_to_cpu(lmm->lmm_magic);
260 pattern = le32_to_cpu(lmm->lmm_pattern);
262 lsm = lov_lsm_alloc(stripe_count, pattern, magic);
266 LASSERT(lsm_op_find(magic));
267 rc = lsm_op_find(magic)->lsm_unpackmd(lov, lsm, lmm);
269 lov_free_memmd(&lsm);
276 /* Retrieve object striping information.
278 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
279 * the maximum number of OST indices which will fit in the user buffer.
280 * lmm_magic must be LOV_USER_MAGIC.
282 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
283 struct lov_user_md __user *lump)
285 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
286 struct lov_user_md_v3 lum;
287 struct lov_mds_md *lmmk;
297 if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
298 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
299 lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
304 if (!lsm_is_released(lsm))
305 stripe_count = lsm->lsm_stripe_count;
309 /* we only need the header part from user space to get lmm_magic and
310 * lmm_stripe_count, (the header part is common to v1 and v3)
312 lum_size = sizeof(struct lov_user_md_v1);
313 if (copy_from_user(&lum, lump, lum_size)) {
317 if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
318 lum.lmm_magic != LOV_USER_MAGIC_V3 &&
319 lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC) {
324 if (lum.lmm_stripe_count &&
325 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
326 /* Return right size of stripe to user */
327 lum.lmm_stripe_count = stripe_count;
328 rc = copy_to_user(lump, &lum, lum_size);
332 lmmk_size = lov_mds_md_size(stripe_count, lsm->lsm_magic);
335 lmmk = libcfs_kvzalloc(lmmk_size, GFP_NOFS);
341 lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
347 /* FIXME: Bug 1185 - copy fields properly when structs change */
348 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
349 BUILD_BUG_ON(sizeof(lum) != sizeof(struct lov_mds_md_v3));
350 BUILD_BUG_ON(sizeof(lum.lmm_objects[0]) != sizeof(lmmk->lmm_objects[0]));
352 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC &&
353 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
354 lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))) {
355 lustre_swab_lov_mds_md(lmmk);
356 lustre_swab_lov_user_md_objects(
357 (struct lov_user_ost_data *)lmmk->lmm_objects,
358 lmmk->lmm_stripe_count);
361 if (lum.lmm_magic == LOV_USER_MAGIC) {
362 /* User request for v1, we need skip lmm_pool_name */
363 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
364 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
365 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
366 lmmk->lmm_stripe_count *
367 sizeof(struct lov_ost_data_v1));
368 lmm_size -= LOV_MAXPOOLNAME;
371 /* if v3 we just have to update the lum_size */
372 lum_size = sizeof(struct lov_user_md_v3);
375 /* User wasn't expecting this many OST entries */
376 if (lum.lmm_stripe_count == 0) {
378 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
383 * Have a difference between lov_mds_md & lov_user_md.
384 * So we have to re-order the data before copy to user.
386 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
387 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
388 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
389 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
390 if (copy_to_user(lump, lmmk, lmm_size))