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);
124 /* Pack LOV object metadata for disk storage. It is packed in LE byte
125 * order and is opaque to the networking layer.
127 * XXX In the future, this will be enhanced to get the EA size from the
128 * underlying OSC device(s) to get their EA sizes so we can stack
129 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
132 static int lov_obd_packmd(struct lov_obd *lov, struct lov_mds_md **lmmp,
133 struct lov_stripe_md *lsm)
135 struct lov_mds_md_v1 *lmmv1;
136 struct lov_mds_md_v3 *lmmv3;
138 struct lov_ost_data_v1 *lmm_objects;
139 int lmm_size, lmm_magic;
145 lmm_magic = lsm->lsm_magic;
148 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
150 /* lsm == NULL and lmmp == NULL */
151 lmm_magic = LOV_MAGIC;
154 if ((lmm_magic != LOV_MAGIC_V1) &&
155 (lmm_magic != LOV_MAGIC_V3)) {
156 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
157 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
163 /* If we are just sizing the EA, limit the stripe count
164 * to the actual number of OSTs in this filesystem. */
166 stripe_count = lov_get_stripecnt(lov, lmm_magic,
167 lsm->lsm_stripe_count);
168 lsm->lsm_stripe_count = stripe_count;
169 } else if (!lsm_is_released(lsm)) {
170 stripe_count = lsm->lsm_stripe_count;
175 /* To calculate maximum easize by active targets at present,
176 * which is exactly the maximum easize to be seen by LOV */
177 stripe_count = lov->desc.ld_active_tgt_count;
180 /* XXX LOV STACKING call into osc for sizes */
181 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
187 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
188 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
189 OBD_FREE_LARGE(*lmmp, lmm_size);
195 OBD_ALLOC_LARGE(*lmmp, lmm_size);
200 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
201 lmm_magic, lmm_size);
204 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
205 if (lmm_magic == LOV_MAGIC_V3)
206 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
208 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
213 /* lmmv1 and lmmv3 point to the same struct and have the
216 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
217 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
218 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
219 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
220 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
221 if (lsm->lsm_magic == LOV_MAGIC_V3) {
222 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
223 sizeof(lmmv3->lmm_pool_name));
224 if (cplen >= sizeof(lmmv3->lmm_pool_name))
226 lmm_objects = lmmv3->lmm_objects;
228 lmm_objects = lmmv1->lmm_objects;
231 for (i = 0; i < stripe_count; i++) {
232 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
233 /* XXX LOV STACKING call down to osc_packmd() to do packing */
234 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
235 " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
236 i, stripe_count, loi->loi_ost_idx);
237 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
238 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
239 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
245 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
246 struct lov_stripe_md *lsm)
248 struct obd_device *obd = class_exp2obd(exp);
249 struct lov_obd *lov = &obd->u.lov;
251 return lov_obd_packmd(lov, lmmp, lsm);
254 /* Find the max stripecount we should use */
255 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
257 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
260 stripe_count = lov->desc.ld_default_stripe_count;
261 if (stripe_count > lov->desc.ld_active_tgt_count)
262 stripe_count = lov->desc.ld_active_tgt_count;
266 /* stripe count is based on whether ldiskfs can handle
268 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
269 lov->lov_ocd.ocd_max_easize)
270 max_stripes = lov_mds_md_max_stripe_count(
271 lov->lov_ocd.ocd_max_easize, magic);
273 if (stripe_count > max_stripes)
274 stripe_count = max_stripes;
280 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
284 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
288 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
289 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
290 sz = lmm_bytes * 2 + 1;
291 OBD_ALLOC_LARGE(buffer, sz);
292 if (buffer != NULL) {
295 for (i = 0; i < lmm_bytes; i++)
296 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
297 buffer[sz - 1] = '\0';
298 CERROR("%s\n", buffer);
299 OBD_FREE_LARGE(buffer, sz);
303 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
304 lmm_bytes, stripe_count);
308 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
309 int pattern, int magic)
314 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
316 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
318 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
322 atomic_set(&(*lsmp)->lsm_refc, 1);
323 spin_lock_init(&(*lsmp)->lsm_lock);
324 (*lsmp)->lsm_magic = magic;
325 (*lsmp)->lsm_stripe_count = stripe_count;
326 (*lsmp)->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
327 (*lsmp)->lsm_pattern = pattern;
328 (*lsmp)->lsm_pool_name[0] = '\0';
329 (*lsmp)->lsm_layout_gen = 0;
330 if (stripe_count > 0)
331 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
333 for (i = 0; i < stripe_count; i++)
334 loi_init((*lsmp)->lsm_oinfo[i]);
339 int lov_free_memmd(struct lov_stripe_md **lsmp)
341 struct lov_stripe_md *lsm = *lsmp;
345 refc = atomic_dec_return(&lsm->lsm_refc);
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 lov_object *obj, struct lov_stripe_md *lsm,
419 struct lov_user_md __user *lump)
422 * XXX huge struct allocated on stack.
424 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
426 struct lov_mds_md *lmmk = NULL;
427 struct lov_user_md_v3 lum;
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, rc = -EFAULT);
440 if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
441 lum.lmm_magic != LOV_USER_MAGIC_V3 &&
442 lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC)
443 GOTO(out, rc = -EINVAL);
445 if (lum.lmm_stripe_count &&
446 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
447 /* Return right size of stripe to user */
448 lum.lmm_stripe_count = lsm->lsm_stripe_count;
449 rc = copy_to_user(lump, &lum, lum_size);
450 GOTO(out, rc = -EOVERFLOW);
452 lov = lu2lov_dev(obj->lo_cl.co_lu.lo_dev)->ld_lov;
453 rc = lov_obd_packmd(lov, &lmmk, lsm);
456 lmmk_size = lmm_size = rc;
459 /* FIXME: Bug 1185 - copy fields properly when structs change */
460 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
461 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
462 CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
464 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
465 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
466 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
467 lustre_swab_lov_mds_md(lmmk);
468 lustre_swab_lov_user_md_objects(
469 (struct lov_user_ost_data *)lmmk->lmm_objects,
470 lmmk->lmm_stripe_count);
472 if (lum.lmm_magic == LOV_USER_MAGIC) {
473 /* User request for v1, we need skip lmm_pool_name */
474 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
475 memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
476 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
477 lmmk->lmm_stripe_count *
478 sizeof(struct lov_ost_data_v1));
479 lmm_size -= LOV_MAXPOOLNAME;
482 /* if v3 we just have to update the lum_size */
483 lum_size = sizeof(struct lov_user_md_v3);
486 /* User wasn't expecting this many OST entries */
487 if (lum.lmm_stripe_count == 0)
489 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
490 GOTO(out_free, rc = -EOVERFLOW);
492 * Have a difference between lov_mds_md & lov_user_md.
493 * So we have to re-order the data before copy to user.
495 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
496 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
497 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
498 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
499 if (copy_to_user(lump, lmmk, lmm_size))
503 OBD_FREE_LARGE(lmmk, lmmk_size);