6c93d180aef7f03056d631c5415ace823e4f1092
[linux-2.6-microblaze.git] / drivers / staging / lustre / lustre / lov / lov_pack.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2015, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/lov/lov_pack.c
33  *
34  * (Un)packing of OST/MDS requests
35  *
36  * Author: Andreas Dilger <adilger@clusterfs.com>
37  */
38
39 #define DEBUG_SUBSYSTEM S_LOV
40
41 #include "../include/lustre/lustre_idl.h"
42 #include "../include/lustre/lustre_user.h"
43
44 #include "../include/lustre_net.h"
45 #include "../include/lustre_swab.h"
46 #include "../include/obd.h"
47 #include "../include/obd_class.h"
48 #include "../include/obd_support.h"
49
50 #include "lov_cl_internal.h"
51 #include "lov_internal.h"
52
53 void lov_dump_lmm_common(int level, void *lmmp)
54 {
55         struct lov_mds_md *lmm = lmmp;
56         struct ost_id   oi;
57
58         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
59         CDEBUG(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(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));
66 }
67
68 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
69                                  int stripe_count)
70 {
71         int i;
72
73         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
74                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
75                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
76                 return;
77         }
78
79         for (i = 0; i < stripe_count; ++i, ++lod) {
80                 struct ost_id   oi;
81
82                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
83                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
84                        le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
85         }
86 }
87
88 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
89 {
90         lov_dump_lmm_common(level, lmm);
91         lov_dump_lmm_objects(level, lmm->lmm_objects,
92                              le16_to_cpu(lmm->lmm_stripe_count));
93 }
94
95 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
96 {
97         lov_dump_lmm_common(level, lmm);
98         CDEBUG(level, "pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
99         lov_dump_lmm_objects(level, lmm->lmm_objects,
100                              le16_to_cpu(lmm->lmm_stripe_count));
101 }
102
103 /**
104  * Pack LOV striping metadata for disk storage format (in little
105  * endian byte order).
106  *
107  * This follows the getxattr() conventions. If \a buf_size is zero
108  * then return the size needed. If \a buf_size is too small then
109  * return -ERANGE. Otherwise return the size of the result.
110  */
111 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
112                      size_t buf_size)
113 {
114         struct lov_ost_data_v1 *lmm_objects;
115         struct lov_mds_md_v1 *lmmv1 = buf;
116         struct lov_mds_md_v3 *lmmv3 = buf;
117         size_t lmm_size;
118         unsigned int i;
119
120         lmm_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
121         if (!buf_size)
122                 return lmm_size;
123
124         if (buf_size < lmm_size)
125                 return -ERANGE;
126
127         /*
128          * lmmv1 and lmmv3 point to the same struct and have the
129          * same first fields
130          */
131         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
132         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
133         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
134         lmmv1->lmm_stripe_count = cpu_to_le16(lsm->lsm_stripe_count);
135         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
136         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
137
138         if (lsm->lsm_magic == LOV_MAGIC_V3) {
139                 CLASSERT(sizeof(lsm->lsm_pool_name) ==
140                          sizeof(lmmv3->lmm_pool_name));
141                 strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
142                         sizeof(lmmv3->lmm_pool_name));
143                 lmm_objects = lmmv3->lmm_objects;
144         } else {
145                 lmm_objects = lmmv1->lmm_objects;
146         }
147
148         for (i = 0; i < lsm->lsm_stripe_count; i++) {
149                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
150
151                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
152                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
153                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
154         }
155
156         return lmm_size;
157 }
158
159 /* Find the max stripecount we should use */
160 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
161 {
162         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
163
164         if (!stripe_count)
165                 stripe_count = lov->desc.ld_default_stripe_count;
166         if (stripe_count > lov->desc.ld_active_tgt_count)
167                 stripe_count = lov->desc.ld_active_tgt_count;
168         if (!stripe_count)
169                 stripe_count = 1;
170
171         /* stripe count is based on whether ldiskfs can handle
172          * larger EA sizes
173          */
174         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
175             lov->lov_ocd.ocd_max_easize)
176                 max_stripes = lov_mds_md_max_stripe_count(
177                         lov->lov_ocd.ocd_max_easize, magic);
178
179         if (stripe_count > max_stripes)
180                 stripe_count = max_stripes;
181
182         return stripe_count;
183 }
184
185 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
186 {
187         int rc;
188
189         if (!lsm_op_find(le32_to_cpu(*(__u32 *)lmm))) {
190                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
191                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
192                 CERROR("%*phN\n", lmm_bytes, lmm);
193                 return -EINVAL;
194         }
195         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
196                                                                      lmm_bytes,
197                                                                   stripe_count);
198         return rc;
199 }
200
201 struct lov_stripe_md *lov_lsm_alloc(u16 stripe_count, u32 pattern, u32 magic)
202 {
203         struct lov_stripe_md *lsm;
204         unsigned int i;
205
206         CDEBUG(D_INFO, "alloc lsm, stripe_count %u\n", stripe_count);
207
208         lsm = lsm_alloc_plain(stripe_count);
209         if (!lsm) {
210                 CERROR("cannot allocate LSM stripe_count %u\n", stripe_count);
211                 return ERR_PTR(-ENOMEM);
212         }
213
214         atomic_set(&lsm->lsm_refc, 1);
215         spin_lock_init(&lsm->lsm_lock);
216         lsm->lsm_magic = magic;
217         lsm->lsm_stripe_count = stripe_count;
218         lsm->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
219         lsm->lsm_pattern = pattern;
220         lsm->lsm_pool_name[0] = '\0';
221         lsm->lsm_layout_gen = 0;
222         if (stripe_count > 0)
223                 lsm->lsm_oinfo[0]->loi_ost_idx = ~0;
224
225         for (i = 0; i < stripe_count; i++)
226                 loi_init(lsm->lsm_oinfo[i]);
227
228         return lsm;
229 }
230
231 int lov_free_memmd(struct lov_stripe_md **lsmp)
232 {
233         struct lov_stripe_md *lsm = *lsmp;
234         int refc;
235
236         *lsmp = NULL;
237         LASSERT(atomic_read(&lsm->lsm_refc) > 0);
238         refc = atomic_dec_return(&lsm->lsm_refc);
239         if (refc == 0)
240                 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
241
242         return refc;
243 }
244
245 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
246  * order and is opaque to the networking layer.
247  */
248 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, struct lov_mds_md *lmm,
249                                    size_t lmm_size)
250 {
251         struct lov_stripe_md *lsm;
252         u16 stripe_count;
253         u32 pattern;
254         u32 magic;
255         int rc;
256
257         rc = lov_verify_lmm(lmm, lmm_size, &stripe_count);
258         if (rc)
259                 return ERR_PTR(rc);
260
261         magic = le32_to_cpu(lmm->lmm_magic);
262         pattern = le32_to_cpu(lmm->lmm_pattern);
263
264         lsm = lov_lsm_alloc(stripe_count, pattern, magic);
265         if (IS_ERR(lsm))
266                 return lsm;
267
268         LASSERT(lsm_op_find(magic));
269         rc = lsm_op_find(magic)->lsm_unpackmd(lov, lsm, lmm);
270         if (rc) {
271                 lov_free_memmd(&lsm);
272                 return ERR_PTR(rc);
273         }
274
275         return lsm;
276 }
277
278 /* Retrieve object striping information.
279  *
280  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
281  * the maximum number of OST indices which will fit in the user buffer.
282  * lmm_magic must be LOV_USER_MAGIC.
283  */
284 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
285                   struct lov_user_md __user *lump)
286 {
287         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
288         struct lov_user_md_v3 lum;
289         struct lov_mds_md *lmmk;
290         u32 stripe_count;
291         ssize_t lmm_size;
292         size_t lmmk_size;
293         size_t lum_size;
294         int rc;
295         mm_segment_t seg;
296
297         if (!lsm)
298                 return -ENODATA;
299
300         /*
301          * "Switch to kernel segment" to allow copying from kernel space by
302          * copy_{to,from}_user().
303          */
304         seg = get_fs();
305         set_fs(KERNEL_DS);
306
307         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
308                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
309                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
310                 rc = -EIO;
311                 goto out;
312         }
313
314         if (!lsm_is_released(lsm))
315                 stripe_count = lsm->lsm_stripe_count;
316         else
317                 stripe_count = 0;
318
319         /* we only need the header part from user space to get lmm_magic and
320          * lmm_stripe_count, (the header part is common to v1 and v3)
321          */
322         lum_size = sizeof(struct lov_user_md_v1);
323         if (copy_from_user(&lum, lump, lum_size)) {
324                 rc = -EFAULT;
325                 goto out;
326         }
327         if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
328             lum.lmm_magic != LOV_USER_MAGIC_V3 &&
329             lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC) {
330                 rc = -EINVAL;
331                 goto out;
332         }
333
334         if (lum.lmm_stripe_count &&
335             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
336                 /* Return right size of stripe to user */
337                 lum.lmm_stripe_count = stripe_count;
338                 rc = copy_to_user(lump, &lum, lum_size);
339                 rc = -EOVERFLOW;
340                 goto out;
341         }
342         lmmk_size = lov_mds_md_size(stripe_count, lsm->lsm_magic);
343
344
345         lmmk = libcfs_kvzalloc(lmmk_size, GFP_NOFS);
346         if (!lmmk) {
347                 rc = -ENOMEM;
348                 goto out;
349         }
350
351         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
352         if (lmm_size < 0) {
353                 rc = lmm_size;
354                 goto out_free;
355         }
356
357         /* FIXME: Bug 1185 - copy fields properly when structs change */
358         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
359         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
360         CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
361
362         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC &&
363             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
364              lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))) {
365                 lustre_swab_lov_mds_md(lmmk);
366                 lustre_swab_lov_user_md_objects(
367                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
368                                 lmmk->lmm_stripe_count);
369         }
370
371         if (lum.lmm_magic == LOV_USER_MAGIC) {
372                 /* User request for v1, we need skip lmm_pool_name */
373                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
374                         memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
375                                 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
376                                 lmmk->lmm_stripe_count *
377                                 sizeof(struct lov_ost_data_v1));
378                         lmm_size -= LOV_MAXPOOLNAME;
379                 }
380         } else {
381                 /* if v3 we just have to update the lum_size */
382                 lum_size = sizeof(struct lov_user_md_v3);
383         }
384
385         /* User wasn't expecting this many OST entries */
386         if (lum.lmm_stripe_count == 0) {
387                 lmm_size = lum_size;
388         } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
389                 rc = -EOVERFLOW;
390                 goto out_free;
391         }
392         /*
393          * Have a difference between lov_mds_md & lov_user_md.
394          * So we have to re-order the data before copy to user.
395          */
396         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
397         lum.lmm_layout_gen = lmmk->lmm_layout_gen;
398         ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
399         ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
400         if (copy_to_user(lump, lmmk, lmm_size))
401                 rc = -EFAULT;
402         else
403                 rc = 0;
404
405 out_free:
406         kvfree(lmmk);
407 out:
408         set_fs(seg);
409         return rc;
410 }