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;
44 CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
45 le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
46 le32_to_cpu(lmm->lmm_pattern));
47 CDEBUG(level,"stripe_size %u, stripe_count %u\n",
48 le32_to_cpu(lmm->lmm_stripe_size),
49 le32_to_cpu(lmm->lmm_stripe_count));
50 for (i = 0, lod = lmm->lmm_objects;
51 i < le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
52 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
53 i, le32_to_cpu(lod->l_ost_idx),
54 le64_to_cpu(lod->l_object_gr),
55 le64_to_cpu(lod->l_object_id));
58 void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
61 CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
62 le64_to_cpu(lmmj->lmmj_md.lmm_object_id),
63 le32_to_cpu(lmmj->lmmj_md.lmm_magic),
64 le32_to_cpu(lmmj->lmmj_md.lmm_pattern));
65 CDEBUG(level,"stripe_size %u, stripe_count %u extent_count %u \n",
66 le32_to_cpu(lmmj->lmmj_md.lmm_stripe_size),
67 le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count),
68 le32_to_cpu(lmmj->lmmj_extent_count));
71 #define LMM_ASSERT(test) \
73 if (!(test)) lov_dump_lmm(D_ERROR, lmm); \
74 LASSERT(test); /* so we know what assertion failed */ \
77 /* Pack LOV object metadata for disk storage. It is packed in LE byte
78 * order and is opaque to the networking layer.
80 * XXX In the future, this will be enhanced to get the EA size from the
81 * underlying OSC device(s) to get their EA sizes so we can stack
82 * LOVs properly. For now lov_mds_md_size() just assumes one obd_id
85 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
86 struct lov_stripe_md *lsm)
88 struct obd_device *obd = class_exp2obd(exp);
89 struct lov_obd *lov = &obd->u.lov;
90 struct lov_oinfo *loi;
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, loi = lsm->lsm_oinfo; i < stripe_count; i++, loi++) {
138 /* XXX LOV STACKING call down to osc_packmd() to do packing */
139 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
140 lmm->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
141 lmm->lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
142 lmm->lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
143 lmm->lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
144 lmm->lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
150 /* Find the max stripecount we should use */
151 int lov_get_stripecnt(struct lov_obd *lov, int stripe_count)
154 stripe_count = lov->desc.ld_default_stripe_count;
155 if (stripe_count > lov->desc.ld_active_tgt_count)
156 stripe_count = lov->desc.ld_active_tgt_count;
159 /* for now, we limit the stripe count directly, when bug 4424 is
160 * fixed this needs to be somewhat dynamic based on whether ext3
161 * can handle larger EA sizes. */
162 if (stripe_count > LOV_MAX_STRIPE_COUNT)
163 stripe_count = LOV_MAX_STRIPE_COUNT;
169 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
173 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
174 CERROR("bad disk LOV MAGIC: 0x%08X; dumping V1 LMM:\n",
175 le32_to_cpu(*(__u32 *)lmm));
176 lov_dump_lmm_v1(D_WARNING, lmm);
179 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
180 lmm_bytes, stripe_count);
184 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
185 int pattern, int magic)
187 int lsm_size = lov_stripe_md_size(stripe_count);
188 struct lov_oinfo *loi;
192 CDEBUG(D_INFO, "alloc lsm, stripe_count %d, lsm_size %d\n",
193 stripe_count, lsm_size);
195 OBD_ALLOC(*lsmp, lsm_size);
197 CERROR("can not allocate lsmp lsm_size %d stripe_count %d\n",
198 lsm_size, stripe_count);
202 spin_lock_init(&(*lsmp)->lsm_lock);
203 (*lsmp)->lsm_magic = magic;
204 (*lsmp)->lsm_stripe_count = stripe_count;
205 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
206 (*lsmp)->lsm_pattern = pattern;
207 (*lsmp)->lsm_oinfo[0].loi_ost_idx = ~0;
209 for (i = 0, loi = (*lsmp)->lsm_oinfo; i < stripe_count; i++, loi++)
215 void lov_free_memmd(struct lov_stripe_md **lsmp)
217 struct lov_stripe_md *lsm = *lsmp;
219 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
220 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
226 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
227 * order and is opaque to the networking layer.
229 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
230 struct lov_mds_md *lmm, int lmm_bytes)
232 struct obd_device *obd = class_exp2obd(exp);
233 struct lov_obd *lov = &obd->u.lov;
234 int rc = 0, stripe_count, lsm_size;
238 /* If passed an MDS struct use values from there, otherwise defaults */
240 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
243 magic = le32_to_cpu(lmm->lmm_magic);
245 stripe_count = lov_get_stripecnt(lov, 0);
249 /* If we aren't passed an lsmp struct, we just want the size */
251 /* XXX LOV STACKING call into osc for sizes */
252 RETURN(lov_stripe_md_size(stripe_count));
254 /* If we are passed an allocated struct but nothing to unpack, free */
256 lov_free_memmd(lsmp);
260 lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
265 /* If we are passed a pointer but nothing to unpack, we only alloc */
269 LASSERT(lsm_op_find(magic) != NULL);
270 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
272 lov_free_memmd(lsmp);
279 static int __lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
280 struct lov_user_md *lump)
282 struct obd_device *obd = class_exp2obd(exp);
283 struct lov_obd *lov = &obd->u.lov;
284 struct lov_user_md lum;
289 rc = copy_from_user(&lum, lump, sizeof(lum));
293 if (lum.lmm_magic != LOV_USER_MAGIC) {
294 if (lum.lmm_magic == __swab32(LOV_USER_MAGIC)) {
295 lustre_swab_lov_user_md(&lum);
297 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
299 lum.lmm_magic, LOV_USER_MAGIC);
304 if (lum.lmm_pattern == 0) {
305 lum.lmm_pattern = lov->desc.ld_pattern ?
306 lov->desc.ld_pattern : LOV_PATTERN_RAID0;
309 if (lum.lmm_pattern != LOV_PATTERN_RAID0) {
310 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
315 /* 64kB is the largest common page size we see (ia64), and matches the
317 if (lum.lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
318 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
319 lum.lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
320 lum.lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
323 if ((lum.lmm_stripe_offset >= lov->desc.ld_active_tgt_count) &&
324 (lum.lmm_stripe_offset != (typeof(lum.lmm_stripe_offset))(-1))) {
325 CDEBUG(D_IOCTL, "stripe offset %u > number of active OSTs %u\n",
326 lum.lmm_stripe_offset, lov->desc.ld_active_tgt_count);
329 stripe_count = lov_get_stripecnt(lov, lum.lmm_stripe_count);
331 if ((__u64)lum.lmm_stripe_size * stripe_count > ~0UL) {
332 CDEBUG(D_IOCTL, "stripe width %ux%u > %lu on 32-bit system\n",
333 lum.lmm_stripe_size, (int)lum.lmm_stripe_count, ~0UL);
337 rc = lov_alloc_memmd(lsmp, stripe_count, lum.lmm_pattern, LOV_MAGIC);
340 (*lsmp)->lsm_oinfo[0].loi_ost_idx = lum.lmm_stripe_offset;
341 (*lsmp)->lsm_stripe_size = lum.lmm_stripe_size;
348 /* Configure object striping information on a new file.
350 * @lmmu is a pointer to a user struct with one or more of the fields set to
351 * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
352 * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC.
353 * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
355 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
356 struct lov_user_md *lump)
364 rc = __lov_setstripe(exp, lsmp, lump);
369 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
370 struct lov_user_md *lump)
374 struct obd_export *oexp;
375 struct lov_obd *lov = &exp->exp_obd->u.lov;
379 for (i = 0; i < lump->lmm_stripe_count; i++) {
380 __u32 len = sizeof(last_id);
381 oexp = lov->lov_tgts[lump->lmm_objects[i].l_ost_idx]->ltd_exp;
382 rc = obd_get_info(oexp, strlen("last_id"), "last_id",
386 if (lump->lmm_objects[i].l_object_id > last_id) {
387 CERROR("Setting EA for object > than last id on "
388 "ost idx %d "LPD64" > "LPD64" \n",
389 lump->lmm_objects[i].l_ost_idx,
390 lump->lmm_objects[i].l_object_id, last_id);
395 rc = lov_setstripe(exp, lsmp, lump);
399 for (i = 0; i < lump->lmm_stripe_count; i++) {
400 (*lsmp)->lsm_oinfo[i].loi_ost_idx =
401 lump->lmm_objects[i].l_ost_idx;
402 (*lsmp)->lsm_oinfo[i].loi_id = lump->lmm_objects[i].l_object_id;
403 (*lsmp)->lsm_oinfo[i].loi_gr = lump->lmm_objects[i].l_object_gr;
409 /* Retrieve object striping information.
411 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
412 * the maximum number of OST indices which will fit in the user buffer.
413 * lmm_magic must be LOV_USER_MAGIC.
415 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
416 struct lov_user_md *lump)
419 * XXX huge struct allocated on stack.
421 struct lov_user_md lum;
422 struct lov_mds_md *lmmk = NULL;
431 * "Switch to kernel segment" to allow copying from kernel space by
432 * copy_{to,from}_user().
436 rc = copy_from_user(&lum, lump, sizeof(lum));
439 else if (lum.lmm_magic != LOV_USER_MAGIC)
442 rc = lov_packmd(exp, &lmmk, lsm);
448 /* FIXME: Bug 1185 - copy fields properly when structs change */
449 CLASSERT(sizeof lum == sizeof *lmmk);
450 CLASSERT(sizeof lum.lmm_objects[0] ==
451 sizeof lmmk->lmm_objects[0]);
453 /* User wasn't expecting this many OST entries */
454 if (lum.lmm_stripe_count == 0) {
455 if (copy_to_user(lump, lmmk, sizeof lum))
457 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
459 } else if (copy_to_user(lump, lmmk, sizeof lum))
462 obd_free_diskmd(exp, &lmmk);