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 <liblustre.h>
48 #include <lustre_net.h>
51 #include <obd_class.h>
52 #include <obd_support.h>
53 #include <lustre/lustre_user.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);
127 /* Pack LOV object metadata for disk storage. It is packed in LE byte
128 * order and is opaque to the networking layer.
130 * XXX In the future, this will be enhanced to get the EA size from the
131 * underlying OSC device(s) to get their EA sizes so we can stack
132 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
135 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
136 struct lov_stripe_md *lsm)
138 struct obd_device *obd = class_exp2obd(exp);
139 struct lov_obd *lov = &obd->u.lov;
140 struct lov_mds_md_v1 *lmmv1;
141 struct lov_mds_md_v3 *lmmv3;
143 struct lov_ost_data_v1 *lmm_objects;
144 int lmm_size, lmm_magic;
150 lmm_magic = lsm->lsm_magic;
153 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
155 /* lsm == NULL and lmmp == NULL */
156 lmm_magic = LOV_MAGIC;
159 if ((lmm_magic != LOV_MAGIC_V1) &&
160 (lmm_magic != LOV_MAGIC_V3)) {
161 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
162 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
168 /* If we are just sizing the EA, limit the stripe count
169 * to the actual number of OSTs in this filesystem. */
171 stripe_count = lov_get_stripecnt(lov, lmm_magic,
172 lsm->lsm_stripe_count);
173 lsm->lsm_stripe_count = stripe_count;
174 } else if (!lsm_is_released(lsm)) {
175 stripe_count = lsm->lsm_stripe_count;
180 /* No need to allocate more than maximum supported stripes.
181 * Anyway, this is pretty inaccurate since ld_tgt_count now
182 * represents max index and we should rely on the actual number
184 stripe_count = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
186 if (stripe_count > lov->desc.ld_tgt_count)
187 stripe_count = lov->desc.ld_tgt_count;
190 /* XXX LOV STACKING call into osc for sizes */
191 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
197 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
198 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
199 OBD_FREE_LARGE(*lmmp, lmm_size);
205 OBD_ALLOC_LARGE(*lmmp, lmm_size);
210 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
211 lmm_magic, lmm_size);
214 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
215 if (lmm_magic == LOV_MAGIC_V3)
216 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
218 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
223 /* lmmv1 and lmmv3 point to the same struct and have the
226 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
227 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
228 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
229 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
230 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
231 if (lsm->lsm_magic == LOV_MAGIC_V3) {
232 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
233 sizeof(lmmv3->lmm_pool_name));
234 if (cplen >= sizeof(lmmv3->lmm_pool_name))
236 lmm_objects = lmmv3->lmm_objects;
238 lmm_objects = lmmv1->lmm_objects;
241 for (i = 0; i < stripe_count; i++) {
242 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
243 /* XXX LOV STACKING call down to osc_packmd() to do packing */
244 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
245 " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
246 i, stripe_count, loi->loi_ost_idx);
247 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
248 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
249 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
255 /* Find the max stripecount we should use */
256 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
258 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
261 stripe_count = lov->desc.ld_default_stripe_count;
262 if (stripe_count > lov->desc.ld_active_tgt_count)
263 stripe_count = lov->desc.ld_active_tgt_count;
267 /* stripe count is based on whether ldiskfs can handle
269 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
270 lov->lov_ocd.ocd_max_easize)
271 max_stripes = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
274 if (stripe_count > max_stripes)
275 stripe_count = max_stripes;
281 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
285 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
289 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
290 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
291 sz = lmm_bytes * 2 + 1;
292 OBD_ALLOC_LARGE(buffer, sz);
293 if (buffer != NULL) {
296 for (i = 0; i < lmm_bytes; i++)
297 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
298 buffer[sz - 1] = '\0';
299 CERROR("%s\n", buffer);
300 OBD_FREE_LARGE(buffer, sz);
304 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
305 lmm_bytes, stripe_count);
309 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
310 int pattern, int magic)
315 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
317 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
319 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
323 cfs_atomic_set(&(*lsmp)->lsm_refc, 1);
324 spin_lock_init(&(*lsmp)->lsm_lock);
325 (*lsmp)->lsm_magic = magic;
326 (*lsmp)->lsm_stripe_count = stripe_count;
327 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
328 (*lsmp)->lsm_pattern = pattern;
329 (*lsmp)->lsm_pool_name[0] = '\0';
330 (*lsmp)->lsm_layout_gen = 0;
331 if (stripe_count > 0)
332 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
334 for (i = 0; i < stripe_count; i++)
335 loi_init((*lsmp)->lsm_oinfo[i]);
340 int lov_free_memmd(struct lov_stripe_md **lsmp)
342 struct lov_stripe_md *lsm = *lsmp;
346 LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 0);
347 if ((refc = cfs_atomic_dec_return(&lsm->lsm_refc)) == 0) {
348 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
349 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
355 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
356 * order and is opaque to the networking layer.
358 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
359 struct lov_mds_md *lmm, int lmm_bytes)
361 struct obd_device *obd = class_exp2obd(exp);
362 struct lov_obd *lov = &obd->u.lov;
363 int rc = 0, lsm_size;
369 /* If passed an MDS struct use values from there, otherwise defaults */
371 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
374 magic = le32_to_cpu(lmm->lmm_magic);
375 pattern = le32_to_cpu(lmm->lmm_pattern);
378 stripe_count = lov_get_stripecnt(lov, magic, 0);
379 pattern = LOV_PATTERN_RAID0;
382 /* If we aren't passed an lsmp struct, we just want the size */
384 /* XXX LOV STACKING call into osc for sizes */
386 RETURN(lov_stripe_md_size(stripe_count));
388 /* If we are passed an allocated struct but nothing to unpack, free */
390 lov_free_memmd(lsmp);
394 lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
398 /* If we are passed a pointer but nothing to unpack, we only alloc */
402 LASSERT(lsm_op_find(magic) != NULL);
403 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
405 lov_free_memmd(lsmp);
412 /* Retrieve object striping information.
414 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
415 * the maximum number of OST indices which will fit in the user buffer.
416 * lmm_magic must be LOV_USER_MAGIC.
418 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
419 struct lov_user_md *lump)
422 * XXX huge struct allocated on stack.
424 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
425 struct lov_user_md_v3 lum;
426 struct lov_mds_md *lmmk = NULL;
434 /* we only need the header part from user space to get lmm_magic and
435 * lmm_stripe_count, (the header part is common to v1 and v3) */
436 lum_size = sizeof(struct lov_user_md_v1);
437 if (copy_from_user(&lum, lump, lum_size))
438 GOTO(out_set, rc = -EFAULT);
439 else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
440 (lum.lmm_magic != LOV_USER_MAGIC_V3))
441 GOTO(out_set, rc = -EINVAL);
443 if (lum.lmm_stripe_count &&
444 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
445 /* Return right size of stripe to user */
446 lum.lmm_stripe_count = lsm->lsm_stripe_count;
447 rc = copy_to_user(lump, &lum, lum_size);
448 GOTO(out_set, rc = -EOVERFLOW);
450 rc = lov_packmd(exp, &lmmk, lsm);
456 /* FIXME: Bug 1185 - copy fields properly when structs change */
457 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
458 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
459 CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
461 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
462 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
463 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
464 lustre_swab_lov_mds_md(lmmk);
465 lustre_swab_lov_user_md_objects(
466 (struct lov_user_ost_data*)lmmk->lmm_objects,
467 lmmk->lmm_stripe_count);
469 if (lum.lmm_magic == LOV_USER_MAGIC) {
470 /* User request for v1, we need skip lmm_pool_name */
471 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
472 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
473 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
474 lmmk->lmm_stripe_count *
475 sizeof(struct lov_ost_data_v1));
476 lmm_size -= LOV_MAXPOOLNAME;
479 /* if v3 we just have to update the lum_size */
480 lum_size = sizeof(struct lov_user_md_v3);
483 /* User wasn't expecting this many OST entries */
484 if (lum.lmm_stripe_count == 0)
486 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
487 GOTO(out_set, rc = -EOVERFLOW);
489 * Have a difference between lov_mds_md & lov_user_md.
490 * So we have to re-order the data before copy to user.
492 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
493 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
494 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
495 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
496 if (copy_to_user(lump, lmmk, lmm_size))
499 obd_free_diskmd(exp, &lmmk);