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, 2014, 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"
52 #include "lov_cl_internal.h"
54 void lov_dump_lmm_common(int level, void *lmmp)
56 struct lov_mds_md *lmm = lmmp;
59 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
60 CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
61 POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
62 le32_to_cpu(lmm->lmm_pattern));
63 CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
64 le32_to_cpu(lmm->lmm_stripe_size),
65 le16_to_cpu(lmm->lmm_stripe_count),
66 le16_to_cpu(lmm->lmm_layout_gen));
69 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
74 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
75 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
76 stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
80 for (i = 0; i < stripe_count; ++i, ++lod) {
83 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
84 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
85 le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
89 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
91 lov_dump_lmm_common(level, lmm);
92 lov_dump_lmm_objects(level, lmm->lmm_objects,
93 le16_to_cpu(lmm->lmm_stripe_count));
96 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
98 lov_dump_lmm_common(level, lmm);
99 CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
100 lov_dump_lmm_objects(level, lmm->lmm_objects,
101 le16_to_cpu(lmm->lmm_stripe_count));
104 void lov_dump_lmm(int level, void *lmm)
108 magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
111 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
114 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
117 CDEBUG(level, "unrecognized lmm_magic %x, assuming %x\n",
118 magic, LOV_MAGIC_V1);
119 lov_dump_lmm_common(level, lmm);
125 * Pack LOV striping metadata for disk storage format (in little
126 * endian byte order).
128 * This follows the getxattr() conventions. If \a buf_size is zero
129 * then return the size needed. If \a buf_size is too small then
130 * return -ERANGE. Otherwise return the size of the result.
132 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
135 struct lov_mds_md_v1 *lmmv1 = buf;
136 struct lov_mds_md_v3 *lmmv3 = buf;
137 struct lov_ost_data_v1 *lmm_objects;
142 lmm_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
146 if (buf_size < lmm_size)
149 /* lmmv1 and lmmv3 point to the same struct and have the
152 lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
153 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
154 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
155 lmmv1->lmm_stripe_count = cpu_to_le16(lsm->lsm_stripe_count);
156 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
157 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
159 if (lsm->lsm_magic == LOV_MAGIC_V3) {
160 CLASSERT(sizeof(lsm->lsm_pool_name) ==
161 sizeof(lmmv3->lmm_pool_name));
162 strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
163 sizeof(lmmv3->lmm_pool_name));
164 lmm_objects = lmmv3->lmm_objects;
166 lmm_objects = lmmv1->lmm_objects;
169 for (i = 0; i < lsm->lsm_stripe_count; i++) {
170 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
172 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
173 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
174 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
180 /* Pack LOV object metadata for disk storage. It is packed in LE byte
181 * order and is opaque to the networking layer.
183 * XXX In the future, this will be enhanced to get the EA size from the
184 * underlying OSC device(s) to get their EA sizes so we can stack
185 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
188 static int lov_obd_packmd(struct lov_obd *lov, struct lov_mds_md **lmmp,
189 struct lov_stripe_md *lsm)
192 int lmm_size, lmm_magic;
196 lmm_magic = lsm->lsm_magic;
199 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
201 /* lsm == NULL and lmmp == NULL */
202 lmm_magic = LOV_MAGIC;
205 if ((lmm_magic != LOV_MAGIC_V1) &&
206 (lmm_magic != LOV_MAGIC_V3)) {
207 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
208 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
214 /* If we are just sizing the EA, limit the stripe count
215 * to the actual number of OSTs in this filesystem. */
217 stripe_count = lov_get_stripecnt(lov, lmm_magic,
218 lsm->lsm_stripe_count);
219 lsm->lsm_stripe_count = stripe_count;
220 } else if (!lsm_is_released(lsm)) {
221 stripe_count = lsm->lsm_stripe_count;
226 /* To calculate maximum easize by active targets at present,
227 * which is exactly the maximum easize to be seen by LOV */
228 stripe_count = lov->desc.ld_active_tgt_count;
231 /* XXX LOV STACKING call into osc for sizes */
232 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
238 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
239 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
240 OBD_FREE_LARGE(*lmmp, lmm_size);
246 OBD_ALLOC_LARGE(*lmmp, lmm_size);
251 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d\n",
252 lmm_magic, lmm_size);
257 RETURN(lov_lsm_pack(lsm, *lmmp, lmm_size));
260 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
261 struct lov_stripe_md *lsm)
263 struct obd_device *obd = class_exp2obd(exp);
264 struct lov_obd *lov = &obd->u.lov;
266 return lov_obd_packmd(lov, lmmp, lsm);
269 /* Find the max stripecount we should use */
270 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
272 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
275 stripe_count = lov->desc.ld_default_stripe_count;
276 if (stripe_count > lov->desc.ld_active_tgt_count)
277 stripe_count = lov->desc.ld_active_tgt_count;
281 /* stripe count is based on whether ldiskfs can handle
283 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
284 lov->lov_ocd.ocd_max_easize)
285 max_stripes = lov_mds_md_max_stripe_count(
286 lov->lov_ocd.ocd_max_easize, magic);
288 if (stripe_count > max_stripes)
289 stripe_count = max_stripes;
295 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
299 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
303 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
304 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
305 sz = lmm_bytes * 2 + 1;
306 OBD_ALLOC_LARGE(buffer, sz);
307 if (buffer != NULL) {
310 for (i = 0; i < lmm_bytes; i++)
311 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
312 buffer[sz - 1] = '\0';
313 CERROR("%s\n", buffer);
314 OBD_FREE_LARGE(buffer, sz);
318 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
319 lmm_bytes, stripe_count);
323 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
324 int pattern, int magic)
329 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
331 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
333 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
337 atomic_set(&(*lsmp)->lsm_refc, 1);
338 spin_lock_init(&(*lsmp)->lsm_lock);
339 (*lsmp)->lsm_magic = magic;
340 (*lsmp)->lsm_stripe_count = stripe_count;
341 (*lsmp)->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
342 (*lsmp)->lsm_pattern = pattern;
343 (*lsmp)->lsm_pool_name[0] = '\0';
344 (*lsmp)->lsm_layout_gen = 0;
345 if (stripe_count > 0)
346 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
348 for (i = 0; i < stripe_count; i++)
349 loi_init((*lsmp)->lsm_oinfo[i]);
354 int lov_free_memmd(struct lov_stripe_md **lsmp)
356 struct lov_stripe_md *lsm = *lsmp;
360 refc = atomic_dec_return(&lsm->lsm_refc);
363 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
364 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
370 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
371 * order and is opaque to the networking layer.
373 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
374 struct lov_mds_md *lmm, int lmm_bytes)
376 struct obd_device *obd = class_exp2obd(exp);
377 struct lov_obd *lov = &obd->u.lov;
378 int rc = 0, lsm_size;
384 /* If passed an MDS struct use values from there, otherwise defaults */
386 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
389 magic = le32_to_cpu(lmm->lmm_magic);
390 pattern = le32_to_cpu(lmm->lmm_pattern);
393 stripe_count = lov_get_stripecnt(lov, magic, 0);
394 pattern = LOV_PATTERN_RAID0;
397 /* If we aren't passed an lsmp struct, we just want the size */
399 /* XXX LOV STACKING call into osc for sizes */
401 RETURN(lov_stripe_md_size(stripe_count));
403 /* If we are passed an allocated struct but nothing to unpack, free */
405 lov_free_memmd(lsmp);
409 lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
413 /* If we are passed a pointer but nothing to unpack, we only alloc */
417 LASSERT(lsm_op_find(magic) != NULL);
418 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
420 lov_free_memmd(lsmp);
427 /* Retrieve object striping information.
429 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
430 * the maximum number of OST indices which will fit in the user buffer.
431 * lmm_magic must be LOV_USER_MAGIC.
433 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
434 struct lov_user_md __user *lump)
437 * XXX huge struct allocated on stack.
439 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
441 struct lov_mds_md *lmmk = NULL;
442 struct lov_user_md_v3 lum;
449 /* we only need the header part from user space to get lmm_magic and
450 * lmm_stripe_count, (the header part is common to v1 and v3) */
451 lum_size = sizeof(struct lov_user_md_v1);
452 if (copy_from_user(&lum, lump, lum_size))
453 GOTO(out, rc = -EFAULT);
455 if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
456 lum.lmm_magic != LOV_USER_MAGIC_V3 &&
457 lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC)
458 GOTO(out, rc = -EINVAL);
460 if (lum.lmm_stripe_count &&
461 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
462 /* Return right size of stripe to user */
463 lum.lmm_stripe_count = lsm->lsm_stripe_count;
464 rc = copy_to_user(lump, &lum, lum_size);
465 GOTO(out, rc = -EOVERFLOW);
467 lov = lu2lov_dev(obj->lo_cl.co_lu.lo_dev)->ld_lov;
468 rc = lov_obd_packmd(lov, &lmmk, lsm);
471 lmmk_size = lmm_size = rc;
474 /* FIXME: Bug 1185 - copy fields properly when structs change */
475 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
476 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
477 CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
479 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
480 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
481 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
482 lustre_swab_lov_mds_md(lmmk);
483 lustre_swab_lov_user_md_objects(
484 (struct lov_user_ost_data *)lmmk->lmm_objects,
485 lmmk->lmm_stripe_count);
487 if (lum.lmm_magic == LOV_USER_MAGIC) {
488 /* User request for v1, we need skip lmm_pool_name */
489 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
490 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
491 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
492 lmmk->lmm_stripe_count *
493 sizeof(struct lov_ost_data_v1));
494 lmm_size -= LOV_MAXPOOLNAME;
497 /* if v3 we just have to update the lum_size */
498 lum_size = sizeof(struct lov_user_md_v3);
501 /* User wasn't expecting this many OST entries */
502 if (lum.lmm_stripe_count == 0)
504 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
505 GOTO(out_free, rc = -EOVERFLOW);
507 * Have a difference between lov_mds_md & lov_user_md.
508 * So we have to re-order the data before copy to user.
510 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
511 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
512 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
513 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
514 if (copy_to_user(lump, lmmk, lmm_size))
518 OBD_FREE_LARGE(lmmk, lmmk_size);