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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/lov/lov_pack.c
34 * (Un)packing of OST/MDS requests
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LOV
41 #include <lustre_net.h>
42 #include <lustre_swab.h>
44 #include <obd_class.h>
45 #include <obd_support.h>
47 #include "lov_cl_internal.h"
48 #include "lov_internal.h"
50 void lov_dump_lmm_common(int level, void *lmmp)
52 struct lov_mds_md *lmm = lmmp;
55 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
56 CDEBUG_LIMIT(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
57 POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
58 le32_to_cpu(lmm->lmm_pattern));
59 CDEBUG_LIMIT(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
60 le32_to_cpu(lmm->lmm_stripe_size),
61 le16_to_cpu(lmm->lmm_stripe_count),
62 le16_to_cpu(lmm->lmm_layout_gen));
65 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
70 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
72 "bad stripe_count %u > max_stripe_count %u\n",
73 stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
77 for (i = 0; i < stripe_count; ++i, ++lod) {
80 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
81 CDEBUG_LIMIT(level, "stripe %u idx %u subobj "DOSTID"\n", i,
82 le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
86 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
88 lov_dump_lmm_common(level, lmm);
89 lov_dump_lmm_objects(level, lmm->lmm_objects,
90 le16_to_cpu(lmm->lmm_stripe_count));
94 * Pack LOV striping metadata for disk storage format (in little
97 * This follows the getxattr() conventions. If \a buf_size is zero
98 * then return the size needed. If \a buf_size is too small then
99 * return -ERANGE. Otherwise return the size of the result.
101 static ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
104 struct lov_mds_md_v1 *lmmv1 = buf;
105 struct lov_mds_md_v3 *lmmv3 = buf;
106 struct lov_ost_data_v1 *lmm_objects;
112 lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
117 if (buf_size < lmm_size)
121 * lmmv1 and lmmv3 point to the same struct and have the
124 lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
125 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
126 lmmv1->lmm_stripe_size = cpu_to_le32(
127 lsm->lsm_entries[0]->lsme_stripe_size);
128 lmmv1->lmm_stripe_count = cpu_to_le16(
129 lsm->lsm_entries[0]->lsme_stripe_count);
130 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
131 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
133 if (lsm->lsm_magic == LOV_MAGIC_V3) {
134 BUILD_BUG_ON(sizeof(lsm->lsm_entries[0]->lsme_pool_name) !=
135 sizeof(lmmv3->lmm_pool_name));
136 strlcpy(lmmv3->lmm_pool_name,
137 lsm->lsm_entries[0]->lsme_pool_name,
138 sizeof(lmmv3->lmm_pool_name));
139 lmm_objects = lmmv3->lmm_objects;
141 lmm_objects = lmmv1->lmm_objects;
144 if (lsm->lsm_is_released)
147 for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
148 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
150 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
151 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
152 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
158 static ssize_t lov_lsm_pack_foreign(const struct lov_stripe_md *lsm, void *buf,
161 struct lov_foreign_md *lfm = buf;
164 lfm_size = lsm->lsm_foreign_size;
169 /* if buffer too small return ERANGE but copy the size the
170 * caller has requested anyway. This may be useful to get
171 * only the header without the need to alloc the full size
173 if (buf_size < lfm_size) {
174 memcpy(lfm, lsm_foreign(lsm), buf_size);
178 /* full foreign LOV is already avail in its cache
179 * no need to translate format fields to little-endian
181 memcpy(lfm, lsm_foreign(lsm), lsm->lsm_foreign_size);
186 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
189 struct lov_comp_md_v1 *lcmv1 = buf;
190 struct lov_comp_md_entry_v1 *lcme;
191 struct lov_ost_data_v1 *lmm_objects;
200 if (lsm->lsm_magic == LOV_MAGIC_V1 || lsm->lsm_magic == LOV_MAGIC_V3)
201 return lov_lsm_pack_v1v3(lsm, buf, buf_size);
203 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
204 return lov_lsm_pack_foreign(lsm, buf, buf_size);
206 lmm_size = lov_comp_md_size(lsm);
210 if (buf_size < lmm_size)
213 lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
214 lcmv1->lcm_size = cpu_to_le32(lmm_size);
215 lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
216 lcmv1->lcm_flags = cpu_to_le16(lsm->lsm_flags);
217 lcmv1->lcm_mirror_count = cpu_to_le16(lsm->lsm_mirror_count);
218 lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
220 offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
222 for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
223 struct lov_stripe_md_entry *lsme;
224 struct lov_mds_md *lmm;
227 lsme = lsm->lsm_entries[entry];
228 lcme = &lcmv1->lcm_entries[entry];
230 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
231 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
232 if (lsme->lsme_flags & LCME_FL_NOSYNC)
233 lcme->lcme_timestamp =
234 cpu_to_le64(lsme->lsme_timestamp);
235 lcme->lcme_extent.e_start =
236 cpu_to_le64(lsme->lsme_extent.e_start);
237 lcme->lcme_extent.e_end =
238 cpu_to_le64(lsme->lsme_extent.e_end);
239 lcme->lcme_offset = cpu_to_le32(offset);
241 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
242 lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
243 /* lmm->lmm_oi not set */
244 lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
245 lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
246 lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
247 lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
249 if (lsme->lsme_magic == LOV_MAGIC_V3) {
250 struct lov_mds_md_v3 *lmmv3 =
251 (struct lov_mds_md_v3 *)lmm;
253 strlcpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
254 sizeof(lmmv3->lmm_pool_name));
255 lmm_objects = lmmv3->lmm_objects;
258 ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
261 if (lsme_inited(lsme) &&
262 !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
263 stripe_count = lsme->lsme_stripe_count;
267 for (i = 0; i < stripe_count; i++) {
268 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
270 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
271 lmm_objects[i].l_ost_gen =
272 cpu_to_le32(loi->loi_ost_gen);
273 lmm_objects[i].l_ost_idx =
274 cpu_to_le32(loi->loi_ost_idx);
277 size = lov_mds_md_size(stripe_count, lsme->lsme_magic);
278 lcme->lcme_size = cpu_to_le32(size);
280 } /* for each layout component */
285 /* Find the max stripecount we should use */
286 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
288 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
291 stripe_count = lov->desc.ld_default_stripe_count;
292 if (stripe_count > lov->desc.ld_active_tgt_count)
293 stripe_count = lov->desc.ld_active_tgt_count;
298 * stripe count is based on whether ldiskfs can handle
301 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
302 lov->lov_ocd.ocd_max_easize)
303 max_stripes = lov_mds_md_max_stripe_count(
304 lov->lov_ocd.ocd_max_easize, magic);
306 if (stripe_count > max_stripes)
307 stripe_count = max_stripes;
312 int lov_free_memmd(struct lov_stripe_md **lsmp)
314 struct lov_stripe_md *lsm = *lsmp;
318 refc = atomic_dec_return(&lsm->lsm_refc);
327 * Unpack LOV object metadata from disk storage. It is packed in LE byte
328 * order and is opaque to the networking layer.
330 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
333 const struct lsm_operations *op;
334 struct lov_stripe_md *lsm;
339 if (buf_size < sizeof(magic))
340 RETURN(ERR_PTR(-EINVAL));
342 magic = le32_to_cpu(*(u32 *)buf);
343 op = lsm_op_find(magic);
345 RETURN(ERR_PTR(-EINVAL));
347 lsm = op->lsm_unpackmd(lov, buf, buf_size);
353 * Retrieve object striping information.
355 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
356 * the maximum number of OST indices which will fit in the user buffer.
357 * lmm_magic must be LOV_USER_MAGIC.
359 * If @size > 0, User specified limited buffer size, usually the buffer is from
360 * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
362 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
363 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
366 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
367 struct lov_mds_md *lmmk, *lmm;
368 struct lov_foreign_md *lfm;
369 struct lov_user_md_v1 lum;
370 size_t lmmk_size, lum_size = 0;
376 if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
377 lsm->lsm_magic != LOV_MAGIC_COMP_V1 &&
378 lsm->lsm_magic != LOV_MAGIC_FOREIGN) {
379 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
380 lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
381 GOTO(out, rc = -EIO);
384 lmmk_size = lov_comp_md_size(lsm);
386 OBD_ALLOC_LARGE(lmmk, lmmk_size);
388 GOTO(out, rc = -ENOMEM);
390 lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
392 GOTO(out_free, rc = lmm_size);
394 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
395 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
396 lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
397 lustre_swab_lov_mds_md(lmmk);
398 lustre_swab_lov_user_md_objects(
399 (struct lov_user_ost_data *)lmmk->lmm_objects,
400 lmmk->lmm_stripe_count);
401 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
402 lustre_swab_lov_comp_md_v1(
403 (struct lov_comp_md_v1 *)lmmk);
404 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_FOREIGN)) {
405 lfm = (struct lov_foreign_md *)lmmk;
406 __swab32s(&lfm->lfm_magic);
407 __swab32s(&lfm->lfm_length);
408 __swab32s(&lfm->lfm_type);
409 __swab32s(&lfm->lfm_flags);
414 * Legacy appication passes limited buffer, we need to figure out
415 * the user buffer size by the passed in lmm_stripe_count.
417 if (lsm->lsm_magic != LOV_MAGIC_FOREIGN)
418 if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
419 GOTO(out_free, rc = -EFAULT);
421 if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
422 lum.lmm_magic == LOV_USER_MAGIC_V3)
423 lum_size = lov_user_md_size(lum.lmm_stripe_count,
427 struct lov_mds_md *comp_md = lmmk;
430 * Legacy app (ADIO for instance) treats the layout as V1/V3
431 * blindly, we'd return a reasonable V1/V3 for them.
433 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
434 struct lov_comp_md_v1 *comp_v1;
435 struct cl_object *cl_obj;
440 cl_obj = cl_object_top(&obj->lo_cl);
441 cl_object_attr_lock(cl_obj);
442 cl_object_attr_get(env, cl_obj, &attr);
443 cl_object_attr_unlock(cl_obj);
446 * return the last instantiated component if file size
447 * is non-zero, otherwise, return the last component.
449 comp_v1 = (struct lov_comp_md_v1 *)lmmk;
450 i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
451 for (; i < comp_v1->lcm_entry_count; i++) {
452 if (!(comp_v1->lcm_entries[i].lcme_flags &
458 comp_md = (struct lov_mds_md *)((char *)comp_v1 +
459 comp_v1->lcm_entries[i].lcme_offset);
460 lum_size = comp_v1->lcm_entries[i].lcme_size;
464 lmm_size = min(lum_size, lmmk_size);
467 lmm_size = lmmk_size;
471 * User specified limited buffer size, usually the buffer is
472 * from ll_lov_setstripe(), and the buffer can only hold basic
473 * layout template info.
475 if (size == 0 || size > lmm_size)
477 if (copy_to_user(lump, lmm, size))
478 GOTO(out_free, rc = -EFAULT);
481 OBD_FREE_LARGE(lmmk, lmmk_size);