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/lustre_idl.h>
42 #include <lustre/lustre_user.h>
44 #include <lustre_net.h>
45 #include <lustre_swab.h>
47 #include <obd_class.h>
48 #include <obd_support.h>
50 #include "lov_cl_internal.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 void lov_dump_lmm(int level, void *lmm)
107 magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
110 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
113 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
116 CDEBUG(level, "unrecognized lmm_magic %x, assuming %x\n",
117 magic, LOV_MAGIC_V1);
118 lov_dump_lmm_common(level, lmm);
124 * Pack LOV striping metadata for disk storage format (in little
125 * endian byte order).
127 * This follows the getxattr() conventions. If \a buf_size is zero
128 * then return the size needed. If \a buf_size is too small then
129 * return -ERANGE. Otherwise return the size of the result.
131 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
134 struct lov_mds_md_v1 *lmmv1 = buf;
135 struct lov_mds_md_v3 *lmmv3 = buf;
136 struct lov_ost_data_v1 *lmm_objects;
141 lmm_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
145 if (buf_size < lmm_size)
148 /* lmmv1 and lmmv3 point to the same struct and have the
151 lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
152 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
153 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
154 lmmv1->lmm_stripe_count = cpu_to_le16(lsm->lsm_stripe_count);
155 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
156 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
158 if (lsm->lsm_magic == LOV_MAGIC_V3) {
159 CLASSERT(sizeof(lsm->lsm_pool_name) ==
160 sizeof(lmmv3->lmm_pool_name));
161 strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
162 sizeof(lmmv3->lmm_pool_name));
163 lmm_objects = lmmv3->lmm_objects;
165 lmm_objects = lmmv1->lmm_objects;
168 for (i = 0; i < lsm->lsm_stripe_count; i++) {
169 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
171 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
172 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
173 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
179 /* Find the max stripecount we should use */
180 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
182 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
185 stripe_count = lov->desc.ld_default_stripe_count;
186 if (stripe_count > lov->desc.ld_active_tgt_count)
187 stripe_count = lov->desc.ld_active_tgt_count;
191 /* stripe count is based on whether ldiskfs can handle
193 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
194 lov->lov_ocd.ocd_max_easize)
195 max_stripes = lov_mds_md_max_stripe_count(
196 lov->lov_ocd.ocd_max_easize, magic);
198 if (stripe_count > max_stripes)
199 stripe_count = max_stripes;
204 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
208 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
212 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
213 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
214 sz = lmm_bytes * 2 + 1;
215 OBD_ALLOC_LARGE(buffer, sz);
216 if (buffer != NULL) {
219 for (i = 0; i < lmm_bytes; i++)
220 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
221 buffer[sz - 1] = '\0';
222 CERROR("%s\n", buffer);
223 OBD_FREE_LARGE(buffer, sz);
227 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
228 lmm_bytes, stripe_count);
232 struct lov_stripe_md *lov_lsm_alloc(u16 stripe_count, u32 pattern, u32 magic)
234 struct lov_stripe_md *lsm;
238 CDEBUG(D_INFO, "alloc lsm, stripe_count %u\n",
239 (unsigned int)stripe_count);
241 lsm = lsm_alloc_plain(stripe_count);
243 CERROR("cannot allocate LSM stripe_count %u\n",
244 (unsigned int)stripe_count);
245 RETURN(ERR_PTR(-ENOMEM));
248 atomic_set(&lsm->lsm_refc, 1);
249 spin_lock_init(&lsm->lsm_lock);
250 lsm->lsm_magic = magic;
251 lsm->lsm_stripe_count = stripe_count;
252 lsm->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
253 lsm->lsm_pattern = pattern;
254 lsm->lsm_pool_name[0] = '\0';
255 lsm->lsm_layout_gen = 0;
256 if (stripe_count > 0)
257 lsm->lsm_oinfo[0]->loi_ost_idx = ~0;
259 for (i = 0; i < stripe_count; i++)
260 loi_init(lsm->lsm_oinfo[i]);
265 int lov_free_memmd(struct lov_stripe_md **lsmp)
267 struct lov_stripe_md *lsm = *lsmp;
271 refc = atomic_dec_return(&lsm->lsm_refc);
274 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
275 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
280 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
281 * order and is opaque to the networking layer.
283 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, struct lov_mds_md *lmm,
286 struct lov_stripe_md *lsm;
293 rc = lov_verify_lmm(lmm, lmm_size, &stripe_count);
297 magic = le32_to_cpu(lmm->lmm_magic);
298 pattern = le32_to_cpu(lmm->lmm_pattern);
300 lsm = lov_lsm_alloc(stripe_count, pattern, magic);
304 LASSERT(lsm_op_find(magic) != NULL);
305 rc = lsm_op_find(magic)->lsm_unpackmd(lov, lsm, lmm);
307 lov_free_memmd(&lsm);
314 /* Retrieve object striping information.
316 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
317 * the maximum number of OST indices which will fit in the user buffer.
318 * lmm_magic must be LOV_USER_MAGIC.
320 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
321 struct lov_user_md __user *lump)
323 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
324 struct lov_mds_md *lmmk;
325 struct lov_user_md_v3 lum;
333 if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
334 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
335 lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
336 GOTO(out, rc = -EIO);
339 if (!lsm_is_released(lsm))
340 stripe_count = lsm->lsm_stripe_count;
344 /* we only need the header part from user space to get lmm_magic and
345 * lmm_stripe_count, (the header part is common to v1 and v3) */
346 lum_size = sizeof(struct lov_user_md_v1);
347 if (copy_from_user(&lum, lump, lum_size))
348 GOTO(out, rc = -EFAULT);
350 if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
351 lum.lmm_magic != LOV_USER_MAGIC_V3 &&
352 lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC)
353 GOTO(out, rc = -EINVAL);
355 if (lum.lmm_stripe_count != 0 && lum.lmm_stripe_count < stripe_count) {
356 /* Return right size of stripe to user */
357 lum.lmm_stripe_count = stripe_count;
358 rc = copy_to_user(lump, &lum, lum_size);
359 GOTO(out, rc = -EOVERFLOW);
362 lmmk_size = lov_mds_md_size(stripe_count, lsm->lsm_magic);
364 OBD_ALLOC_LARGE(lmmk, lmmk_size);
366 GOTO(out, rc = -ENOMEM);
368 lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
370 GOTO(out_free, rc = lmm_size);
372 /* FIXME: Bug 1185 - copy fields properly when structs change */
373 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
374 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
375 CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
377 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC &&
378 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
379 lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))) {
380 lustre_swab_lov_mds_md(lmmk);
381 lustre_swab_lov_user_md_objects(
382 (struct lov_user_ost_data *)lmmk->lmm_objects,
383 lmmk->lmm_stripe_count);
386 if (lum.lmm_magic == LOV_USER_MAGIC) {
387 /* User request for v1, we need skip lmm_pool_name */
388 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
389 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
390 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
391 lmmk->lmm_stripe_count *
392 sizeof(struct lov_ost_data_v1));
393 lmm_size -= LOV_MAXPOOLNAME;
396 /* if v3 we just have to update the lum_size */
397 lum_size = sizeof(struct lov_user_md_v3);
400 /* User wasn't expecting this many OST entries */
401 if (lum.lmm_stripe_count == 0)
403 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
404 GOTO(out_free, rc = -EOVERFLOW);
406 * Have a difference between lov_mds_md & lov_user_md.
407 * So we have to re-order the data before copy to user.
409 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
410 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
411 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
412 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
413 if (copy_to_user(lump, lmmk, lmm_size))
414 GOTO(out_free, rc = -EFAULT);
416 GOTO(out_free, rc = 0);
418 OBD_FREE_LARGE(lmmk, lmmk_size);