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_EX(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_EX(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 < le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
53 CDEBUG_EX(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_EX(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_EX(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_oinfo *loi;
92 struct lov_mds_md *lmm;
93 int stripe_count = lov->desc.ld_tgt_count;
99 if (lsm->lsm_magic != LOV_MAGIC) {
100 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X\n",
101 lsm->lsm_magic, LOV_MAGIC);
104 stripe_count = lsm->lsm_stripe_count;
107 /* XXX LOV STACKING call into osc for sizes */
108 lmm_size = lov_mds_md_size(stripe_count);
114 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
115 OBD_FREE(*lmmp, lov_mds_md_size(stripe_count));
121 OBD_ALLOC(*lmmp, lmm_size);
127 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC); /* only write new format */
132 lmm->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
133 lmm->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
134 lmm->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
135 lmm->lmm_stripe_count = cpu_to_le32(stripe_count);
136 lmm->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
138 for (i = 0, loi = lsm->lsm_oinfo; i < stripe_count; i++, loi++) {
139 /* XXX LOV STACKING call down to osc_packmd() to do packing */
140 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
141 lmm->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
142 lmm->lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
143 lmm->lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
144 lmm->lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
145 lmm->lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
151 /* Find the max stripecount we should use */
152 int lov_get_stripecnt(struct lov_obd *lov, int stripe_count)
155 stripe_count = lov->desc.ld_default_stripe_count;
156 if (stripe_count > lov->desc.ld_active_tgt_count)
157 stripe_count = lov->desc.ld_active_tgt_count;
160 /* for now, we limit the stripe count directly, when bug 4424 is
161 * fixed this needs to be somewhat dynamic based on whether ext3
162 * can handle larger EA sizes. */
163 if (stripe_count > LOV_MAX_STRIPE_COUNT)
164 stripe_count = LOV_MAX_STRIPE_COUNT;
170 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
174 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
175 CERROR("bad disk LOV MAGIC: 0x%08X; dumping V1 LMM:\n",
176 le32_to_cpu(*(__u32 *)lmm));
177 lov_dump_lmm_v1(D_WARNING, lmm);
180 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
181 lmm_bytes, stripe_count);
185 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
186 int pattern, int magic)
188 int lsm_size = lov_stripe_md_size(stripe_count);
189 struct lov_oinfo *loi;
193 CDEBUG(D_INFO, "alloc lsm, stripe_count %d, lsm_size %d\n",
194 stripe_count, lsm_size);
196 OBD_ALLOC(*lsmp, lsm_size);
198 CERROR("can not allocate lsmp lsm_size %d stripe_count %d\n",
199 lsm_size, stripe_count);
203 spin_lock_init(&(*lsmp)->lsm_lock);
204 (*lsmp)->lsm_magic = magic;
205 (*lsmp)->lsm_stripe_count = stripe_count;
206 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
207 (*lsmp)->lsm_xfersize = PTLRPC_MAX_BRW_SIZE * stripe_count;
208 (*lsmp)->lsm_pattern = pattern;
209 (*lsmp)->lsm_oinfo[0].loi_ost_idx = ~0;
211 for (i = 0, loi = (*lsmp)->lsm_oinfo; i < stripe_count; i++, loi++)
217 void lov_free_memmd(struct lov_stripe_md **lsmp)
219 struct lov_stripe_md *lsm = *lsmp;
221 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
222 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
228 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
229 * order and is opaque to the networking layer.
231 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
232 struct lov_mds_md *lmm, int lmm_bytes)
234 struct obd_device *obd = class_exp2obd(exp);
235 struct lov_obd *lov = &obd->u.lov;
236 int rc = 0, stripe_count, lsm_size;
240 /* If passed an MDS struct use values from there, otherwise defaults */
242 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
245 magic = le32_to_cpu(lmm->lmm_magic);
247 stripe_count = lov_get_stripecnt(lov, 0);
251 /* If we aren't passed an lsmp struct, we just want the size */
253 /* XXX LOV STACKING call into osc for sizes */
254 RETURN(lov_stripe_md_size(stripe_count));
256 /* If we are passed an allocated struct but nothing to unpack, free */
258 lov_free_memmd(lsmp);
262 lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
267 /* If we are passed a pointer but nothing to unpack, we only alloc */
271 LASSERT(lsm_op_find(magic) != NULL);
272 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
274 lov_free_memmd(lsmp);
281 /* Configure object striping information on a new file.
283 * @lmmu is a pointer to a user struct with one or more of the fields set to
284 * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
285 * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC.
286 * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
288 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
289 struct lov_user_md *lump)
291 struct obd_device *obd = class_exp2obd(exp);
292 struct lov_obd *lov = &obd->u.lov;
293 struct lov_user_md lum;
298 rc = copy_from_user(&lum, lump, sizeof(lum));
302 if (lum.lmm_magic != LOV_USER_MAGIC) {
303 if (lum.lmm_magic == __swab32(LOV_USER_MAGIC)) {
304 lustre_swab_lov_user_md(&lum);
306 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
308 lum.lmm_magic, LOV_USER_MAGIC);
313 if (lum.lmm_pattern == 0) {
314 lum.lmm_pattern = lov->desc.ld_pattern ?
315 lov->desc.ld_pattern : LOV_PATTERN_RAID0;
318 if (lum.lmm_pattern != LOV_PATTERN_RAID0) {
319 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
324 /* 64kB is the largest common page size we see (ia64), and matches the
326 if (lum.lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
327 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
328 lum.lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
329 lum.lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
332 if ((lum.lmm_stripe_offset >= lov->desc.ld_active_tgt_count) &&
333 (lum.lmm_stripe_offset != (typeof(lum.lmm_stripe_offset))(-1))) {
334 CDEBUG(D_IOCTL, "stripe offset %u > number of active OSTs %u\n",
335 lum.lmm_stripe_offset, lov->desc.ld_active_tgt_count);
338 stripe_count = lov_get_stripecnt(lov, lum.lmm_stripe_count);
340 if ((__u64)lum.lmm_stripe_size * stripe_count > ~0UL) {
341 CDEBUG(D_IOCTL, "stripe width %ux%u > %lu on 32-bit system\n",
342 lum.lmm_stripe_size, (int)lum.lmm_stripe_count, ~0UL);
346 rc = lov_alloc_memmd(lsmp, stripe_count, lum.lmm_pattern, LOV_MAGIC);
351 (*lsmp)->lsm_oinfo[0].loi_ost_idx = lum.lmm_stripe_offset;
352 (*lsmp)->lsm_stripe_size = lum.lmm_stripe_size;
353 (*lsmp)->lsm_xfersize = lum.lmm_stripe_size * stripe_count;
358 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
359 struct lov_user_md *lump)
363 struct obd_export *oexp;
364 struct lov_obd *lov = &exp->exp_obd->u.lov;
368 for (i = 0; i < lump->lmm_stripe_count; i++) {
369 __u32 len = sizeof(last_id);
370 oexp = lov->tgts[lump->lmm_objects[i].l_ost_idx].ltd_exp;
371 rc = obd_get_info(oexp, strlen("last_id"), "last_id",
375 if (lump->lmm_objects[i].l_object_id > last_id) {
376 CERROR("Setting EA for object > than last id on "
377 "ost idx %d "LPD64" > "LPD64" \n",
378 lump->lmm_objects[i].l_ost_idx,
379 lump->lmm_objects[i].l_object_id, last_id);
384 rc = lov_setstripe(exp, lsmp, lump);
388 for (i = 0; i < lump->lmm_stripe_count; i++) {
389 (*lsmp)->lsm_oinfo[i].loi_ost_idx =
390 lump->lmm_objects[i].l_ost_idx;
391 (*lsmp)->lsm_oinfo[i].loi_id = lump->lmm_objects[i].l_object_id;
392 (*lsmp)->lsm_oinfo[i].loi_gr = lump->lmm_objects[i].l_object_gr;
398 /* Retrieve object striping information.
400 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
401 * the maximum number of OST indices which will fit in the user buffer.
402 * lmm_magic must be LOV_USER_MAGIC.
404 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
405 struct lov_user_md *lump)
407 struct lov_user_md lum;
408 struct lov_mds_md *lmmk = NULL;
415 rc = copy_from_user(&lum, lump, sizeof(lum));
419 if (lum.lmm_magic != LOV_USER_MAGIC)
422 rc = lov_packmd(exp, &lmmk, lsm);
428 /* FIXME: Bug 1185 - copy fields properly when structs change */
429 LASSERT(sizeof(lum) == sizeof(*lmmk));
430 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
432 /* User wasn't expecting this many OST entries */
433 if (lum.lmm_stripe_count == 0) {
434 if (copy_to_user(lump, lmmk, sizeof(lum)))
436 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
438 } else if (copy_to_user(lump, lmmk, lmm_size)) {
442 obd_free_diskmd(exp, &lmmk);