1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
5 * Author: Andreas Dilger <adilger@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 * (Un)packing of OST/MDS requests
26 #define DEBUG_SUBSYSTEM S_LOV
28 #include <liblustre.h>
31 #include <lustre_net.h>
34 #include <obd_class.h>
35 #include <obd_support.h>
36 #include <lustre/lustre_user.h>
38 #include "lov_internal.h"
40 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
42 struct lov_ost_data_v1 *lod;
45 CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
46 le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
47 le32_to_cpu(lmm->lmm_pattern));
48 CDEBUG(level,"stripe_size %u, stripe_count %u\n",
49 le32_to_cpu(lmm->lmm_stripe_size),
50 le32_to_cpu(lmm->lmm_stripe_count));
51 for (i = 0, lod = lmm->lmm_objects;
52 i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
53 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
54 i, le32_to_cpu(lod->l_ost_idx),
55 le64_to_cpu(lod->l_object_gr),
56 le64_to_cpu(lod->l_object_id));
59 void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
62 CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
63 le64_to_cpu(lmmj->lmmj_md.lmm_object_id),
64 le32_to_cpu(lmmj->lmmj_md.lmm_magic),
65 le32_to_cpu(lmmj->lmmj_md.lmm_pattern));
66 CDEBUG(level,"stripe_size %u, stripe_count %u extent_count %u \n",
67 le32_to_cpu(lmmj->lmmj_md.lmm_stripe_size),
68 le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count),
69 le32_to_cpu(lmmj->lmmj_extent_count));
72 #define LMM_ASSERT(test) \
74 if (!(test)) lov_dump_lmm(D_ERROR, lmm); \
75 LASSERT(test); /* so we know what assertion failed */ \
78 /* Pack LOV object metadata for disk storage. It is packed in LE byte
79 * order and is opaque to the networking layer.
81 * XXX In the future, this will be enhanced to get the EA size from the
82 * underlying OSC device(s) to get their EA sizes so we can stack
83 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
86 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
87 struct lov_stripe_md *lsm)
89 struct obd_device *obd = class_exp2obd(exp);
90 struct lov_obd *lov = &obd->u.lov;
91 struct lov_mds_md *lmm;
92 int stripe_count = lov->desc.ld_tgt_count;
98 if (lsm->lsm_magic != LOV_MAGIC) {
99 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X\n",
100 lsm->lsm_magic, LOV_MAGIC);
103 stripe_count = lsm->lsm_stripe_count;
106 /* XXX LOV STACKING call into osc for sizes */
107 lmm_size = lov_mds_md_size(stripe_count);
113 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
114 OBD_FREE(*lmmp, lov_mds_md_size(stripe_count));
120 OBD_ALLOC(*lmmp, lmm_size);
126 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC); /* only write new format */
131 lmm->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
132 lmm->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
133 lmm->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
134 lmm->lmm_stripe_count = cpu_to_le32(stripe_count);
135 lmm->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
137 for (i = 0; i < stripe_count; i++) {
138 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
140 /* XXX LOV STACKING call down to osc_packmd() to do packing */
141 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
142 lmm->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
143 lmm->lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
144 lmm->lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
145 lmm->lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
146 lmm->lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
152 /* Find the max stripecount we should use */
153 int lov_get_stripecnt(struct lov_obd *lov, int stripe_count)
156 stripe_count = lov->desc.ld_default_stripe_count;
157 if (stripe_count > lov->desc.ld_active_tgt_count)
158 stripe_count = lov->desc.ld_active_tgt_count;
161 /* for now, we limit the stripe count directly, when bug 4424 is
162 * fixed this needs to be somewhat dynamic based on whether ext3
163 * can handle larger EA sizes. */
164 if (stripe_count > LOV_MAX_STRIPE_COUNT)
165 stripe_count = LOV_MAX_STRIPE_COUNT;
171 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
175 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
176 CERROR("bad disk LOV MAGIC: 0x%08X; dumping V1 LMM:\n",
177 le32_to_cpu(*(__u32 *)lmm));
178 lov_dump_lmm_v1(D_WARNING, lmm);
181 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
182 lmm_bytes, stripe_count);
186 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
187 int pattern, int magic)
192 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
194 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
196 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
200 spin_lock_init(&(*lsmp)->lsm_lock);
201 (*lsmp)->lsm_magic = magic;
202 (*lsmp)->lsm_stripe_count = stripe_count;
203 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
204 (*lsmp)->lsm_pattern = pattern;
205 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
207 for (i = 0; i < stripe_count; i++)
208 loi_init((*lsmp)->lsm_oinfo[i]);
213 void lov_free_memmd(struct lov_stripe_md **lsmp)
215 struct lov_stripe_md *lsm = *lsmp;
217 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
218 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
224 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
225 * order and is opaque to the networking layer.
227 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
228 struct lov_mds_md *lmm, int lmm_bytes)
230 struct obd_device *obd = class_exp2obd(exp);
231 struct lov_obd *lov = &obd->u.lov;
232 int rc = 0, stripe_count, lsm_size;
236 /* If passed an MDS struct use values from there, otherwise defaults */
238 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
241 magic = le32_to_cpu(lmm->lmm_magic);
243 stripe_count = lov_get_stripecnt(lov, 0);
247 /* If we aren't passed an lsmp struct, we just want the size */
249 /* XXX LOV STACKING call into osc for sizes */
251 RETURN(lov_stripe_md_size(stripe_count));
253 /* If we are passed an allocated struct but nothing to unpack, free */
255 lov_free_memmd(lsmp);
259 lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
264 /* If we are passed a pointer but nothing to unpack, we only alloc */
268 LASSERT(lsm_op_find(magic) != NULL);
269 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
271 lov_free_memmd(lsmp);
278 static int __lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
279 struct lov_user_md *lump)
281 struct obd_device *obd = class_exp2obd(exp);
282 struct lov_obd *lov = &obd->u.lov;
283 struct lov_user_md lum;
288 rc = copy_from_user(&lum, lump, sizeof(lum));
292 if (lum.lmm_magic != LOV_USER_MAGIC) {
293 if (lum.lmm_magic == __swab32(LOV_USER_MAGIC)) {
294 lustre_swab_lov_user_md(&lum);
296 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
298 lum.lmm_magic, LOV_USER_MAGIC);
303 if (lum.lmm_pattern == 0) {
304 lum.lmm_pattern = lov->desc.ld_pattern ?
305 lov->desc.ld_pattern : LOV_PATTERN_RAID0;
308 if (lum.lmm_pattern != LOV_PATTERN_RAID0) {
309 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
314 /* 64kB is the largest common page size we see (ia64), and matches the
316 if (lum.lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
317 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
318 lum.lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
319 lum.lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
322 if ((lum.lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
323 (lum.lmm_stripe_offset != (typeof(lum.lmm_stripe_offset))(-1))) {
324 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
325 lum.lmm_stripe_offset, lov->desc.ld_tgt_count);
328 stripe_count = lov_get_stripecnt(lov, lum.lmm_stripe_count);
330 if ((__u64)lum.lmm_stripe_size * stripe_count > ~0UL) {
331 CDEBUG(D_IOCTL, "stripe width %ux%i exeeds %lu bytes\n",
332 lum.lmm_stripe_size, (int)lum.lmm_stripe_count, ~0UL);
336 rc = lov_alloc_memmd(lsmp, stripe_count, lum.lmm_pattern, LOV_MAGIC);
339 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lum.lmm_stripe_offset;
340 (*lsmp)->lsm_stripe_size = lum.lmm_stripe_size;
347 /* Configure object striping information on a new file.
349 * @lmmu is a pointer to a user struct with one or more of the fields set to
350 * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
351 * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC.
352 * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
354 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
355 struct lov_user_md *lump)
363 rc = __lov_setstripe(exp, lsmp, lump);
368 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
369 struct lov_user_md *lump)
373 struct obd_export *oexp;
374 struct lov_obd *lov = &exp->exp_obd->u.lov;
378 for (i = 0; i < lump->lmm_stripe_count; i++) {
379 __u32 len = sizeof(last_id);
380 oexp = lov->lov_tgts[lump->lmm_objects[i].l_ost_idx]->ltd_exp;
381 rc = obd_get_info(oexp, strlen("last_id"), "last_id",
385 if (lump->lmm_objects[i].l_object_id > last_id) {
386 CERROR("Setting EA for object > than last id on "
387 "ost idx %d "LPD64" > "LPD64" \n",
388 lump->lmm_objects[i].l_ost_idx,
389 lump->lmm_objects[i].l_object_id, last_id);
394 rc = lov_setstripe(exp, lsmp, lump);
398 for (i = 0; i < lump->lmm_stripe_count; i++) {
399 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
400 lump->lmm_objects[i].l_ost_idx;
401 (*lsmp)->lsm_oinfo[i]->loi_id = lump->lmm_objects[i].l_object_id;
402 (*lsmp)->lsm_oinfo[i]->loi_gr = lump->lmm_objects[i].l_object_gr;
408 /* Retrieve object striping information.
410 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
411 * the maximum number of OST indices which will fit in the user buffer.
412 * lmm_magic must be LOV_USER_MAGIC.
414 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
415 struct lov_user_md *lump)
418 * XXX huge struct allocated on stack.
420 struct lov_user_md lum;
421 struct lov_mds_md *lmmk = NULL;
430 * "Switch to kernel segment" to allow copying from kernel space by
431 * copy_{to,from}_user().
435 rc = copy_from_user(&lum, lump, sizeof(lum));
438 else if (lum.lmm_magic != LOV_USER_MAGIC)
441 rc = lov_packmd(exp, &lmmk, lsm);
447 /* FIXME: Bug 1185 - copy fields properly when structs change */
448 CLASSERT(sizeof lum == sizeof *lmmk);
449 CLASSERT(sizeof lum.lmm_objects[0] ==
450 sizeof lmmk->lmm_objects[0]);
452 /* User wasn't expecting this many OST entries */
453 if (lum.lmm_stripe_count == 0) {
454 if (copy_to_user(lump, lmmk, sizeof lum))
456 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
458 } else if (copy_to_user(lump, lmmk, sizeof lum))
461 obd_free_diskmd(exp, &lmmk);