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, 2013, 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_net.h>
47 #include <obd_class.h>
48 #include <obd_support.h>
49 #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 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);
123 /* Pack LOV object metadata for disk storage. It is packed in LE byte
124 * order and is opaque to the networking layer.
126 * XXX In the future, this will be enhanced to get the EA size from the
127 * underlying OSC device(s) to get their EA sizes so we can stack
128 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
131 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
132 struct lov_stripe_md *lsm)
134 struct obd_device *obd = class_exp2obd(exp);
135 struct lov_obd *lov = &obd->u.lov;
136 struct lov_mds_md_v1 *lmmv1;
137 struct lov_mds_md_v3 *lmmv3;
139 struct lov_ost_data_v1 *lmm_objects;
140 int lmm_size, lmm_magic;
146 lmm_magic = lsm->lsm_magic;
149 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
151 /* lsm == NULL and lmmp == NULL */
152 lmm_magic = LOV_MAGIC;
155 if ((lmm_magic != LOV_MAGIC_V1) &&
156 (lmm_magic != LOV_MAGIC_V3)) {
157 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
158 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
164 /* If we are just sizing the EA, limit the stripe count
165 * to the actual number of OSTs in this filesystem. */
167 stripe_count = lov_get_stripecnt(lov, lmm_magic,
168 lsm->lsm_stripe_count);
169 lsm->lsm_stripe_count = stripe_count;
170 } else if (!lsm_is_released(lsm)) {
171 stripe_count = lsm->lsm_stripe_count;
176 /* To calculate maximum easize by active targets at present,
177 * which is exactly the maximum easize to be seen by LOV */
178 stripe_count = lov->desc.ld_active_tgt_count;
181 /* XXX LOV STACKING call into osc for sizes */
182 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
188 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
189 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
190 OBD_FREE_LARGE(*lmmp, lmm_size);
196 OBD_ALLOC_LARGE(*lmmp, lmm_size);
201 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
202 lmm_magic, lmm_size);
205 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
206 if (lmm_magic == LOV_MAGIC_V3)
207 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
209 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
214 /* lmmv1 and lmmv3 point to the same struct and have the
217 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
218 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
219 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
220 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
221 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
222 if (lsm->lsm_magic == LOV_MAGIC_V3) {
223 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
224 sizeof(lmmv3->lmm_pool_name));
225 if (cplen >= sizeof(lmmv3->lmm_pool_name))
227 lmm_objects = lmmv3->lmm_objects;
229 lmm_objects = lmmv1->lmm_objects;
232 for (i = 0; i < stripe_count; i++) {
233 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
234 /* XXX LOV STACKING call down to osc_packmd() to do packing */
235 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
236 " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
237 i, stripe_count, loi->loi_ost_idx);
238 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
239 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
240 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
246 /* Find the max stripecount we should use */
247 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
249 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
252 stripe_count = lov->desc.ld_default_stripe_count;
253 if (stripe_count > lov->desc.ld_active_tgt_count)
254 stripe_count = lov->desc.ld_active_tgt_count;
258 /* stripe count is based on whether ldiskfs can handle
260 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
261 lov->lov_ocd.ocd_max_easize)
262 max_stripes = lov_mds_md_max_stripe_count(
263 lov->lov_ocd.ocd_max_easize, magic);
265 if (stripe_count > max_stripes)
266 stripe_count = max_stripes;
272 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
276 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
280 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
281 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
282 sz = lmm_bytes * 2 + 1;
283 OBD_ALLOC_LARGE(buffer, sz);
284 if (buffer != NULL) {
287 for (i = 0; i < lmm_bytes; i++)
288 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
289 buffer[sz - 1] = '\0';
290 CERROR("%s\n", buffer);
291 OBD_FREE_LARGE(buffer, sz);
295 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
296 lmm_bytes, stripe_count);
300 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
301 int pattern, int magic)
306 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
308 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
310 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
314 atomic_set(&(*lsmp)->lsm_refc, 1);
315 spin_lock_init(&(*lsmp)->lsm_lock);
316 (*lsmp)->lsm_magic = magic;
317 (*lsmp)->lsm_stripe_count = stripe_count;
318 (*lsmp)->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
319 (*lsmp)->lsm_pattern = pattern;
320 (*lsmp)->lsm_pool_name[0] = '\0';
321 (*lsmp)->lsm_layout_gen = 0;
322 if (stripe_count > 0)
323 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
325 for (i = 0; i < stripe_count; i++)
326 loi_init((*lsmp)->lsm_oinfo[i]);
331 int lov_free_memmd(struct lov_stripe_md **lsmp)
333 struct lov_stripe_md *lsm = *lsmp;
337 refc = atomic_dec_return(&lsm->lsm_refc);
340 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
341 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
347 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
348 * order and is opaque to the networking layer.
350 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
351 struct lov_mds_md *lmm, int lmm_bytes)
353 struct obd_device *obd = class_exp2obd(exp);
354 struct lov_obd *lov = &obd->u.lov;
355 int rc = 0, lsm_size;
361 /* If passed an MDS struct use values from there, otherwise defaults */
363 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
366 magic = le32_to_cpu(lmm->lmm_magic);
367 pattern = le32_to_cpu(lmm->lmm_pattern);
370 stripe_count = lov_get_stripecnt(lov, magic, 0);
371 pattern = LOV_PATTERN_RAID0;
374 /* If we aren't passed an lsmp struct, we just want the size */
376 /* XXX LOV STACKING call into osc for sizes */
378 RETURN(lov_stripe_md_size(stripe_count));
380 /* If we are passed an allocated struct but nothing to unpack, free */
382 lov_free_memmd(lsmp);
386 lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
390 /* If we are passed a pointer but nothing to unpack, we only alloc */
394 LASSERT(lsm_op_find(magic) != NULL);
395 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
397 lov_free_memmd(lsmp);
404 /* Retrieve object striping information.
406 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
407 * the maximum number of OST indices which will fit in the user buffer.
408 * lmm_magic must be LOV_USER_MAGIC.
410 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
411 struct lov_user_md __user *lump)
414 * XXX huge struct allocated on stack.
416 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
417 struct lov_user_md_v3 lum;
418 struct lov_mds_md *lmmk = NULL;
426 /* we only need the header part from user space to get lmm_magic and
427 * lmm_stripe_count, (the header part is common to v1 and v3) */
428 lum_size = sizeof(struct lov_user_md_v1);
429 if (copy_from_user(&lum, lump, lum_size))
430 GOTO(out_set, rc = -EFAULT);
432 if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
433 lum.lmm_magic != LOV_USER_MAGIC_V3 &&
434 lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC)
435 GOTO(out_set, rc = -EINVAL);
437 if (lum.lmm_stripe_count &&
438 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
439 /* Return right size of stripe to user */
440 lum.lmm_stripe_count = lsm->lsm_stripe_count;
441 rc = copy_to_user(lump, &lum, lum_size);
442 GOTO(out_set, rc = -EOVERFLOW);
444 rc = lov_packmd(exp, &lmmk, lsm);
450 /* FIXME: Bug 1185 - copy fields properly when structs change */
451 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
452 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
453 CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
455 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
456 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
457 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
458 lustre_swab_lov_mds_md(lmmk);
459 lustre_swab_lov_user_md_objects(
460 (struct lov_user_ost_data*)lmmk->lmm_objects,
461 lmmk->lmm_stripe_count);
463 if (lum.lmm_magic == LOV_USER_MAGIC) {
464 /* User request for v1, we need skip lmm_pool_name */
465 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
466 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
467 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
468 lmmk->lmm_stripe_count *
469 sizeof(struct lov_ost_data_v1));
470 lmm_size -= LOV_MAXPOOLNAME;
473 /* if v3 we just have to update the lum_size */
474 lum_size = sizeof(struct lov_user_md_v3);
477 /* User wasn't expecting this many OST entries */
478 if (lum.lmm_stripe_count == 0)
480 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
481 GOTO(out_set, rc = -EOVERFLOW);
483 * Have a difference between lov_mds_md & lov_user_md.
484 * So we have to re-order the data before copy to user.
486 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
487 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
488 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
489 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
490 if (copy_to_user(lump, lmmk, lmm_size))
493 obd_free_diskmd(exp, &lmmk);