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, 2015, 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/lustre_idl.h>
42 #include <lustre/lustre_user.h>
44 #include <lustre_net.h>
45 #include <lustre_swab.h>
47 #include <obd_class.h>
48 #include <obd_support.h>
50 #include "lov_cl_internal.h"
51 #include "lov_internal.h"
53 void lov_dump_lmm_common(int level, void *lmmp)
55 struct lov_mds_md *lmm = lmmp;
58 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
59 CDEBUG_LIMIT(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
60 POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
61 le32_to_cpu(lmm->lmm_pattern));
62 CDEBUG_LIMIT(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
63 le32_to_cpu(lmm->lmm_stripe_size),
64 le16_to_cpu(lmm->lmm_stripe_count),
65 le16_to_cpu(lmm->lmm_layout_gen));
68 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
73 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
75 "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_LIMIT(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_LIMIT(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_LIMIT(level, "unrecognized lmm_magic %x, assuming %x\n",
118 magic, LOV_MAGIC_V1);
119 lov_dump_lmm_common(level, lmm);
125 * Pack LOV striping metadata for disk storage format (in little
126 * endian byte order).
128 * This follows the getxattr() conventions. If \a buf_size is zero
129 * then return the size needed. If \a buf_size is too small then
130 * return -ERANGE. Otherwise return the size of the result.
132 ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
135 struct lov_mds_md_v1 *lmmv1 = buf;
136 struct lov_mds_md_v3 *lmmv3 = buf;
137 struct lov_ost_data_v1 *lmm_objects;
142 lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
147 if (buf_size < lmm_size)
150 /* lmmv1 and lmmv3 point to the same struct and have the
153 lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
154 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
155 lmmv1->lmm_stripe_size = cpu_to_le32(
156 lsm->lsm_entries[0]->lsme_stripe_size);
157 lmmv1->lmm_stripe_count = cpu_to_le16(
158 lsm->lsm_entries[0]->lsme_stripe_count);
159 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
160 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
162 if (lsm->lsm_magic == LOV_MAGIC_V3) {
163 CLASSERT(sizeof(lsm->lsm_entries[0]->lsme_pool_name) ==
164 sizeof(lmmv3->lmm_pool_name));
165 strlcpy(lmmv3->lmm_pool_name,
166 lsm->lsm_entries[0]->lsme_pool_name,
167 sizeof(lmmv3->lmm_pool_name));
168 lmm_objects = lmmv3->lmm_objects;
170 lmm_objects = lmmv1->lmm_objects;
173 if (lsm->lsm_is_released)
176 for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
177 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
179 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
180 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
181 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
187 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
190 struct lov_comp_md_v1 *lcmv1 = buf;
191 struct lov_comp_md_entry_v1 *lcme;
192 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 lmm_size = lov_comp_md_size(lsm);
207 if (buf_size < lmm_size)
210 lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
211 lcmv1->lcm_size = cpu_to_le32(lmm_size);
212 lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
213 lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
215 offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
217 for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
218 struct lov_stripe_md_entry *lsme;
219 struct lov_mds_md *lmm;
222 lsme = lsm->lsm_entries[entry];
223 lcme = &lcmv1->lcm_entries[entry];
225 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
226 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
227 lcme->lcme_extent.e_start =
228 cpu_to_le64(lsme->lsme_extent.e_start);
229 lcme->lcme_extent.e_end =
230 cpu_to_le64(lsme->lsme_extent.e_end);
231 lcme->lcme_offset = cpu_to_le32(offset);
233 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
234 lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
235 /* lmm->lmm_oi not set */
236 lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
237 lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
238 lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
239 lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
241 if (lsme->lsme_magic == LOV_MAGIC_V3) {
242 struct lov_mds_md_v3 *lmmv3 =
243 (struct lov_mds_md_v3 *)lmm;
245 strlcpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
246 sizeof(lmmv3->lmm_pool_name));
247 lmm_objects = lmmv3->lmm_objects;
250 ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
253 if (lsme_inited(lsme) &&
254 !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
255 stripe_count = lsme->lsme_stripe_count;
259 for (i = 0; i < stripe_count; i++) {
260 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
262 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
263 lmm_objects[i].l_ost_gen =
264 cpu_to_le32(loi->loi_ost_gen);
265 lmm_objects[i].l_ost_idx =
266 cpu_to_le32(loi->loi_ost_idx);
269 size = lov_mds_md_size(stripe_count, lsme->lsme_magic);
270 lcme->lcme_size = cpu_to_le32(size);
272 } /* for each layout component */
277 /* Find the max stripecount we should use */
278 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
280 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
283 stripe_count = lov->desc.ld_default_stripe_count;
284 if (stripe_count > lov->desc.ld_active_tgt_count)
285 stripe_count = lov->desc.ld_active_tgt_count;
289 /* stripe count is based on whether ldiskfs can handle
291 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
292 lov->lov_ocd.ocd_max_easize)
293 max_stripes = lov_mds_md_max_stripe_count(
294 lov->lov_ocd.ocd_max_easize, magic);
296 if (stripe_count > max_stripes)
297 stripe_count = max_stripes;
302 int lov_free_memmd(struct lov_stripe_md **lsmp)
304 struct lov_stripe_md *lsm = *lsmp;
308 refc = atomic_dec_return(&lsm->lsm_refc);
316 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
317 * order and is opaque to the networking layer.
319 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
322 const struct lsm_operations *op;
323 struct lov_stripe_md *lsm;
327 if (buf_size < sizeof(magic))
328 RETURN(ERR_PTR(-EINVAL));
330 magic = le32_to_cpu(*(u32 *)buf);
331 op = lsm_op_find(magic);
333 RETURN(ERR_PTR(-EINVAL));
335 lsm = op->lsm_unpackmd(lov, buf, buf_size);
340 /* Retrieve object striping information.
342 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
343 * the maximum number of OST indices which will fit in the user buffer.
344 * lmm_magic must be LOV_USER_MAGIC.
346 * If @size > 0, User specified limited buffer size, usually the buffer is from
347 * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
349 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
350 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
353 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
354 struct lov_mds_md *lmmk, *lmm;
355 struct lov_user_md_v1 lum;
357 ssize_t lmm_size, lum_size = 0;
362 if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
363 lsm->lsm_magic != LOV_MAGIC_COMP_V1) {
364 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
365 lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
366 GOTO(out, rc = -EIO);
370 LCONSOLE_WARN("%s: using old ioctl(LL_IOC_LOV_GETSTRIPE) on "
371 DFID", use llapi_layout_get_by_path()\n",
373 PFID(&obj->lo_cl.co_lu.lo_header->loh_fid));
377 lmmk_size = lov_comp_md_size(lsm);
379 OBD_ALLOC_LARGE(lmmk, lmmk_size);
381 GOTO(out, rc = -ENOMEM);
383 lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
385 GOTO(out_free, rc = lmm_size);
387 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
388 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
389 lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
390 lustre_swab_lov_mds_md(lmmk);
391 lustre_swab_lov_user_md_objects(
392 (struct lov_user_ost_data *)lmmk->lmm_objects,
393 lmmk->lmm_stripe_count);
394 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
395 lustre_swab_lov_comp_md_v1(
396 (struct lov_comp_md_v1 *)lmmk);
400 /* Legacy appication passes limited buffer, we need to figure out
401 * the user buffer size by the passed in lmm_stripe_count. */
402 if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
403 GOTO(out_free, rc = -EFAULT);
405 if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
406 lum.lmm_magic == LOV_USER_MAGIC_V3)
407 lum_size = lov_user_md_size(lum.lmm_stripe_count,
411 struct lov_mds_md *comp_md = lmmk;
413 /* Legacy app (ADIO for instance) treats the layout as V1/V3
414 * blindly, we'd return a reasonable V1/V3 for them. */
415 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
416 struct lov_comp_md_v1 *comp_v1;
417 struct cl_object *cl_obj;
422 cl_obj = cl_object_top(&obj->lo_cl);
423 cl_object_attr_lock(cl_obj);
424 cl_object_attr_get(env, cl_obj, &attr);
425 cl_object_attr_unlock(cl_obj);
427 /* return the last instantiated component if file size
428 * is non-zero, otherwise, return the last component.*/
429 comp_v1 = (struct lov_comp_md_v1 *)lmmk;
430 i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
431 for (; i < comp_v1->lcm_entry_count; i++) {
432 if (!(comp_v1->lcm_entries[i].lcme_flags &
438 comp_md = (struct lov_mds_md *)((char *)comp_v1 +
439 comp_v1->lcm_entries[i].lcme_offset);
446 lmm_size = lmmk_size;
449 * User specified limited buffer size, usually the buffer is
450 * from ll_lov_setstripe(), and the buffer can only hold basic
451 * layout template info.
453 if (size == 0 || size > lmm_size)
455 if (copy_to_user(lump, lmm, size))
456 GOTO(out_free, rc = -EFAULT);
459 OBD_FREE_LARGE(lmmk, lmmk_size);