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 <lustre/lustre_idl.h>
46 #include <lustre/lustre_user.h>
48 #include <lustre_net.h>
49 #include <lustre_swab.h>
51 #include <obd_class.h>
52 #include <obd_support.h>
54 #include "lov_cl_internal.h"
55 #include "lov_internal.h"
57 void lov_dump_lmm_common(int level, void *lmmp)
59 struct lov_mds_md *lmm = lmmp;
62 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
63 CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
64 POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
65 le32_to_cpu(lmm->lmm_pattern));
66 CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
67 le32_to_cpu(lmm->lmm_stripe_size),
68 le16_to_cpu(lmm->lmm_stripe_count),
69 le16_to_cpu(lmm->lmm_layout_gen));
72 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
77 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
78 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
79 stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
83 for (i = 0; i < stripe_count; ++i, ++lod) {
86 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
87 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
88 le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
92 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
94 lov_dump_lmm_common(level, lmm);
95 lov_dump_lmm_objects(level, lmm->lmm_objects,
96 le16_to_cpu(lmm->lmm_stripe_count));
99 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
101 lov_dump_lmm_common(level, lmm);
102 CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
103 lov_dump_lmm_objects(level, lmm->lmm_objects,
104 le16_to_cpu(lmm->lmm_stripe_count));
107 void lov_dump_lmm(int level, void *lmm)
111 magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
114 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
117 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
120 CDEBUG(level, "unrecognized lmm_magic %x, assuming %x\n",
121 magic, LOV_MAGIC_V1);
122 lov_dump_lmm_common(level, lmm);
128 * Pack LOV striping metadata for disk storage format (in little
129 * endian byte order).
131 * This follows the getxattr() conventions. If \a buf_size is zero
132 * then return the size needed. If \a buf_size is too small then
133 * return -ERANGE. Otherwise return the size of the result.
135 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
138 struct lov_mds_md_v1 *lmmv1 = buf;
139 struct lov_mds_md_v3 *lmmv3 = buf;
140 struct lov_ost_data_v1 *lmm_objects;
145 lmm_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
149 if (buf_size < lmm_size)
152 /* lmmv1 and lmmv3 point to the same struct and have the
155 lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
156 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
157 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
158 lmmv1->lmm_stripe_count = cpu_to_le16(lsm->lsm_stripe_count);
159 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
160 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
162 if (lsm->lsm_magic == LOV_MAGIC_V3) {
163 CLASSERT(sizeof(lsm->lsm_pool_name) ==
164 sizeof(lmmv3->lmm_pool_name));
165 strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
166 sizeof(lmmv3->lmm_pool_name));
167 lmm_objects = lmmv3->lmm_objects;
169 lmm_objects = lmmv1->lmm_objects;
172 for (i = 0; i < lsm->lsm_stripe_count; i++) {
173 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
175 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
176 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
177 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
183 /* Find the max stripecount we should use */
184 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
186 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
189 stripe_count = lov->desc.ld_default_stripe_count;
190 if (stripe_count > lov->desc.ld_active_tgt_count)
191 stripe_count = lov->desc.ld_active_tgt_count;
195 /* stripe count is based on whether ldiskfs can handle
197 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
198 lov->lov_ocd.ocd_max_easize)
199 max_stripes = lov_mds_md_max_stripe_count(
200 lov->lov_ocd.ocd_max_easize, magic);
202 if (stripe_count > max_stripes)
203 stripe_count = max_stripes;
208 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
212 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
216 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
217 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
218 sz = lmm_bytes * 2 + 1;
219 OBD_ALLOC_LARGE(buffer, sz);
220 if (buffer != NULL) {
223 for (i = 0; i < lmm_bytes; i++)
224 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
225 buffer[sz - 1] = '\0';
226 CERROR("%s\n", buffer);
227 OBD_FREE_LARGE(buffer, sz);
231 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
232 lmm_bytes, stripe_count);
236 struct lov_stripe_md *lov_lsm_alloc(u16 stripe_count, u32 pattern, u32 magic)
238 struct lov_stripe_md *lsm;
242 CDEBUG(D_INFO, "alloc lsm, stripe_count %u\n",
243 (unsigned int)stripe_count);
245 lsm = lsm_alloc_plain(stripe_count);
247 CERROR("cannot allocate LSM stripe_count %u\n",
248 (unsigned int)stripe_count);
249 RETURN(ERR_PTR(-ENOMEM));
252 atomic_set(&lsm->lsm_refc, 1);
253 spin_lock_init(&lsm->lsm_lock);
254 lsm->lsm_magic = magic;
255 lsm->lsm_stripe_count = stripe_count;
256 lsm->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
257 lsm->lsm_pattern = pattern;
258 lsm->lsm_pool_name[0] = '\0';
259 lsm->lsm_layout_gen = 0;
260 if (stripe_count > 0)
261 lsm->lsm_oinfo[0]->loi_ost_idx = ~0;
263 for (i = 0; i < stripe_count; i++)
264 loi_init(lsm->lsm_oinfo[i]);
269 int lov_free_memmd(struct lov_stripe_md **lsmp)
271 struct lov_stripe_md *lsm = *lsmp;
275 refc = atomic_dec_return(&lsm->lsm_refc);
278 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
279 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
284 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
285 * order and is opaque to the networking layer.
287 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, struct lov_mds_md *lmm,
290 struct lov_stripe_md *lsm;
297 rc = lov_verify_lmm(lmm, lmm_size, &stripe_count);
301 magic = le32_to_cpu(lmm->lmm_magic);
302 pattern = le32_to_cpu(lmm->lmm_pattern);
304 lsm = lov_lsm_alloc(stripe_count, pattern, magic);
308 LASSERT(lsm_op_find(magic) != NULL);
309 rc = lsm_op_find(magic)->lsm_unpackmd(lov, lsm, lmm);
311 lov_free_memmd(&lsm);
318 /* Retrieve object striping information.
320 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
321 * the maximum number of OST indices which will fit in the user buffer.
322 * lmm_magic must be LOV_USER_MAGIC.
324 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
325 struct lov_user_md __user *lump)
327 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
328 struct lov_mds_md *lmmk;
329 struct lov_user_md_v3 lum;
337 if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
338 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
339 lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
340 GOTO(out, rc = -EIO);
343 if (!lsm_is_released(lsm))
344 stripe_count = lsm->lsm_stripe_count;
348 /* we only need the header part from user space to get lmm_magic and
349 * lmm_stripe_count, (the header part is common to v1 and v3) */
350 lum_size = sizeof(struct lov_user_md_v1);
351 if (copy_from_user(&lum, lump, lum_size))
352 GOTO(out, rc = -EFAULT);
354 if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
355 lum.lmm_magic != LOV_USER_MAGIC_V3 &&
356 lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC)
357 GOTO(out, rc = -EINVAL);
359 if (lum.lmm_stripe_count != 0 && lum.lmm_stripe_count < stripe_count) {
360 /* Return right size of stripe to user */
361 lum.lmm_stripe_count = stripe_count;
362 rc = copy_to_user(lump, &lum, lum_size);
363 GOTO(out, rc = -EOVERFLOW);
366 lmmk_size = lov_mds_md_size(stripe_count, lsm->lsm_magic);
368 OBD_ALLOC_LARGE(lmmk, lmmk_size);
370 GOTO(out, rc = -ENOMEM);
372 lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
374 GOTO(out_free, rc = lmm_size);
376 /* FIXME: Bug 1185 - copy fields properly when structs change */
377 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
378 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
379 CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
381 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC &&
382 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
383 lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))) {
384 lustre_swab_lov_mds_md(lmmk);
385 lustre_swab_lov_user_md_objects(
386 (struct lov_user_ost_data *)lmmk->lmm_objects,
387 lmmk->lmm_stripe_count);
390 if (lum.lmm_magic == LOV_USER_MAGIC) {
391 /* User request for v1, we need skip lmm_pool_name */
392 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
393 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
394 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
395 lmmk->lmm_stripe_count *
396 sizeof(struct lov_ost_data_v1));
397 lmm_size -= LOV_MAXPOOLNAME;
400 /* if v3 we just have to update the lum_size */
401 lum_size = sizeof(struct lov_user_md_v3);
404 /* User wasn't expecting this many OST entries */
405 if (lum.lmm_stripe_count == 0)
407 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
408 GOTO(out_free, rc = -EOVERFLOW);
410 * Have a difference between lov_mds_md & lov_user_md.
411 * So we have to re-order the data before copy to user.
413 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
414 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
415 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
416 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
417 if (copy_to_user(lump, lmmk, lmm_size))
418 GOTO(out_free, rc = -EFAULT);
420 GOTO(out_free, rc = 0);
422 OBD_FREE_LARGE(lmmk, lmmk_size);