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.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * lustre/lov/lov_pack.c
37 *
38 * (Un)packing of OST/MDS requests
39 *
40 * Author: Andreas Dilger <adilger@clusterfs.com>
41 */
42
43 #define DEBUG_SUBSYSTEM S_LOV
44
45 #include "../include/lustre_net.h"
46 #include "../include/obd.h"
47 #include "../include/obd_class.h"
48 #include "../include/obd_support.h"
49 #include "../include/lustre/lustre_user.h"
50
51 #include "lov_internal.h"
52
lov_dump_lmm_common(int level,void * lmmp)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
lov_dump_lmm_objects(int level,struct lov_ost_data * lod,int stripe_count)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
lov_dump_lmm_v1(int level,struct lov_mds_md_v1 * lmm)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
lov_dump_lmm_v3(int level,struct lov_mds_md_v3 * lmm)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 /* Pack LOV object metadata for disk storage. It is packed in LE byte
104 * order and is opaque to the networking layer.
105 *
106 * XXX In the future, this will be enhanced to get the EA size from the
107 * underlying OSC device(s) to get their EA sizes so we can stack
108 * LOVs properly. For now lov_mds_md_size() just assumes one u64
109 * per stripe.
110 */
lov_packmd(struct obd_export * exp,struct lov_mds_md ** lmmp,struct lov_stripe_md * lsm)111 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
112 struct lov_stripe_md *lsm)
113 {
114 struct obd_device *obd = class_exp2obd(exp);
115 struct lov_obd *lov = &obd->u.lov;
116 struct lov_mds_md_v1 *lmmv1;
117 struct lov_mds_md_v3 *lmmv3;
118 __u16 stripe_count;
119 struct lov_ost_data_v1 *lmm_objects;
120 int lmm_size, lmm_magic;
121 int i;
122 int cplen = 0;
123
124 if (lsm) {
125 lmm_magic = lsm->lsm_magic;
126 } else {
127 if (lmmp && *lmmp)
128 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
129 else
130 /* lsm == NULL and lmmp == NULL */
131 lmm_magic = LOV_MAGIC;
132 }
133
134 if ((lmm_magic != LOV_MAGIC_V1) &&
135 (lmm_magic != LOV_MAGIC_V3)) {
136 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
137 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
138 return -EINVAL;
139
140 }
141
142 if (lsm) {
143 /* If we are just sizing the EA, limit the stripe count
144 * to the actual number of OSTs in this filesystem. */
145 if (!lmmp) {
146 stripe_count = lov_get_stripecnt(lov, lmm_magic,
147 lsm->lsm_stripe_count);
148 lsm->lsm_stripe_count = stripe_count;
149 } else if (!lsm_is_released(lsm)) {
150 stripe_count = lsm->lsm_stripe_count;
151 } else {
152 stripe_count = 0;
153 }
154 } else {
155 /* No need to allocate more than maximum supported stripes.
156 * Anyway, this is pretty inaccurate since ld_tgt_count now
157 * represents max index and we should rely on the actual number
158 * of OSTs instead */
159 stripe_count = lov_mds_md_max_stripe_count(
160 lov->lov_ocd.ocd_max_easize, lmm_magic);
161
162 if (stripe_count > lov->desc.ld_tgt_count)
163 stripe_count = lov->desc.ld_tgt_count;
164 }
165
166 /* XXX LOV STACKING call into osc for sizes */
167 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
168
169 if (!lmmp)
170 return lmm_size;
171
172 if (*lmmp && !lsm) {
173 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
174 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
175 kvfree(*lmmp);
176 *lmmp = NULL;
177 return 0;
178 }
179
180 if (!*lmmp) {
181 *lmmp = libcfs_kvzalloc(lmm_size, GFP_NOFS);
182 if (!*lmmp)
183 return -ENOMEM;
184 }
185
186 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
187 lmm_magic, lmm_size);
188
189 lmmv1 = *lmmp;
190 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
191 if (lmm_magic == LOV_MAGIC_V3)
192 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
193 else
194 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
195
196 if (!lsm)
197 return lmm_size;
198
199 /* lmmv1 and lmmv3 point to the same struct and have the
200 * same first fields
201 */
202 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
203 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
204 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
205 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
206 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
207 if (lsm->lsm_magic == LOV_MAGIC_V3) {
208 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
209 sizeof(lmmv3->lmm_pool_name));
210 if (cplen >= sizeof(lmmv3->lmm_pool_name))
211 return -E2BIG;
212 lmm_objects = lmmv3->lmm_objects;
213 } else {
214 lmm_objects = lmmv1->lmm_objects;
215 }
216
217 for (i = 0; i < stripe_count; i++) {
218 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
219 /* XXX LOV STACKING call down to osc_packmd() to do packing */
220 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
221 " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
222 i, stripe_count, loi->loi_ost_idx);
223 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
224 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
225 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
226 }
227
228 return lmm_size;
229 }
230
231 /* Find the max stripecount we should use */
lov_get_stripecnt(struct lov_obd * lov,__u32 magic,__u16 stripe_count)232 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
233 {
234 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
235
236 if (!stripe_count)
237 stripe_count = lov->desc.ld_default_stripe_count;
238 if (stripe_count > lov->desc.ld_active_tgt_count)
239 stripe_count = lov->desc.ld_active_tgt_count;
240 if (!stripe_count)
241 stripe_count = 1;
242
243 /* stripe count is based on whether ldiskfs can handle
244 * larger EA sizes */
245 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
246 lov->lov_ocd.ocd_max_easize)
247 max_stripes = lov_mds_md_max_stripe_count(
248 lov->lov_ocd.ocd_max_easize, magic);
249
250 if (stripe_count > max_stripes)
251 stripe_count = max_stripes;
252
253 return stripe_count;
254 }
255
lov_verify_lmm(void * lmm,int lmm_bytes,__u16 * stripe_count)256 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
257 {
258 int rc;
259
260 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
261 char *buffer;
262 int sz;
263
264 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
265 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
266 sz = lmm_bytes * 2 + 1;
267 buffer = libcfs_kvzalloc(sz, GFP_NOFS);
268 if (buffer != NULL) {
269 int i;
270
271 for (i = 0; i < lmm_bytes; i++)
272 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
273 buffer[sz - 1] = '\0';
274 CERROR("%s\n", buffer);
275 kvfree(buffer);
276 }
277 return -EINVAL;
278 }
279 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
280 lmm_bytes, stripe_count);
281 return rc;
282 }
283
lov_alloc_memmd(struct lov_stripe_md ** lsmp,__u16 stripe_count,int pattern,int magic)284 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
285 int pattern, int magic)
286 {
287 int i, lsm_size;
288
289 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
290
291 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
292 if (!*lsmp) {
293 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
294 return -ENOMEM;
295 }
296
297 atomic_set(&(*lsmp)->lsm_refc, 1);
298 spin_lock_init(&(*lsmp)->lsm_lock);
299 (*lsmp)->lsm_magic = magic;
300 (*lsmp)->lsm_stripe_count = stripe_count;
301 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
302 (*lsmp)->lsm_pattern = pattern;
303 (*lsmp)->lsm_pool_name[0] = '\0';
304 (*lsmp)->lsm_layout_gen = 0;
305 if (stripe_count > 0)
306 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
307
308 for (i = 0; i < stripe_count; i++)
309 loi_init((*lsmp)->lsm_oinfo[i]);
310
311 return lsm_size;
312 }
313
lov_free_memmd(struct lov_stripe_md ** lsmp)314 int lov_free_memmd(struct lov_stripe_md **lsmp)
315 {
316 struct lov_stripe_md *lsm = *lsmp;
317 int refc;
318
319 *lsmp = NULL;
320 LASSERT(atomic_read(&lsm->lsm_refc) > 0);
321 refc = atomic_dec_return(&lsm->lsm_refc);
322 if (refc == 0) {
323 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
324 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
325 }
326 return refc;
327 }
328
329 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
330 * order and is opaque to the networking layer.
331 */
lov_unpackmd(struct obd_export * exp,struct lov_stripe_md ** lsmp,struct lov_mds_md * lmm,int lmm_bytes)332 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
333 struct lov_mds_md *lmm, int lmm_bytes)
334 {
335 struct obd_device *obd = class_exp2obd(exp);
336 struct lov_obd *lov = &obd->u.lov;
337 int rc = 0, lsm_size;
338 __u16 stripe_count;
339 __u32 magic;
340 __u32 pattern;
341
342 /* If passed an MDS struct use values from there, otherwise defaults */
343 if (lmm) {
344 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
345 if (rc)
346 return rc;
347 magic = le32_to_cpu(lmm->lmm_magic);
348 pattern = le32_to_cpu(lmm->lmm_pattern);
349 } else {
350 magic = LOV_MAGIC;
351 stripe_count = lov_get_stripecnt(lov, magic, 0);
352 pattern = LOV_PATTERN_RAID0;
353 }
354
355 /* If we aren't passed an lsmp struct, we just want the size */
356 if (!lsmp) {
357 /* XXX LOV STACKING call into osc for sizes */
358 LBUG();
359 return lov_stripe_md_size(stripe_count);
360 }
361 /* If we are passed an allocated struct but nothing to unpack, free */
362 if (*lsmp && !lmm) {
363 lov_free_memmd(lsmp);
364 return 0;
365 }
366
367 lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
368 if (lsm_size < 0)
369 return lsm_size;
370
371 /* If we are passed a pointer but nothing to unpack, we only alloc */
372 if (!lmm)
373 return lsm_size;
374
375 LASSERT(lsm_op_find(magic) != NULL);
376 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
377 if (rc) {
378 lov_free_memmd(lsmp);
379 return rc;
380 }
381
382 return lsm_size;
383 }
384
385 /* Retrieve object striping information.
386 *
387 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
388 * the maximum number of OST indices which will fit in the user buffer.
389 * lmm_magic must be LOV_USER_MAGIC.
390 */
lov_getstripe(struct obd_export * exp,struct lov_stripe_md * lsm,struct lov_user_md * lump)391 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
392 struct lov_user_md *lump)
393 {
394 /*
395 * XXX huge struct allocated on stack.
396 */
397 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
398 struct lov_user_md_v3 lum;
399 struct lov_mds_md *lmmk = NULL;
400 int rc, lmm_size;
401 int lum_size;
402
403 if (!lsm)
404 return -ENODATA;
405
406 /* we only need the header part from user space to get lmm_magic and
407 * lmm_stripe_count, (the header part is common to v1 and v3) */
408 lum_size = sizeof(struct lov_user_md_v1);
409 if (copy_from_user(&lum, lump, lum_size)) {
410 rc = -EFAULT;
411 goto out_set;
412 } else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
413 (lum.lmm_magic != LOV_USER_MAGIC_V3)) {
414 rc = -EINVAL;
415 goto out_set;
416 }
417
418 if (lum.lmm_stripe_count &&
419 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
420 /* Return right size of stripe to user */
421 lum.lmm_stripe_count = lsm->lsm_stripe_count;
422 rc = copy_to_user(lump, &lum, lum_size);
423 rc = -EOVERFLOW;
424 goto out_set;
425 }
426 rc = lov_packmd(exp, &lmmk, lsm);
427 if (rc < 0)
428 goto out_set;
429 lmm_size = rc;
430 rc = 0;
431
432 /* FIXME: Bug 1185 - copy fields properly when structs change */
433 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
434 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
435 CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
436
437 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
438 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
439 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
440 lustre_swab_lov_mds_md(lmmk);
441 lustre_swab_lov_user_md_objects(
442 (struct lov_user_ost_data *)lmmk->lmm_objects,
443 lmmk->lmm_stripe_count);
444 }
445 if (lum.lmm_magic == LOV_USER_MAGIC) {
446 /* User request for v1, we need skip lmm_pool_name */
447 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
448 memmove((char *)(&lmmk->lmm_stripe_count) +
449 sizeof(lmmk->lmm_stripe_count),
450 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
451 lmmk->lmm_stripe_count *
452 sizeof(struct lov_ost_data_v1));
453 lmm_size -= LOV_MAXPOOLNAME;
454 }
455 } else {
456 /* if v3 we just have to update the lum_size */
457 lum_size = sizeof(struct lov_user_md_v3);
458 }
459
460 /* User wasn't expecting this many OST entries */
461 if (lum.lmm_stripe_count == 0)
462 lmm_size = lum_size;
463 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
464 rc = -EOVERFLOW;
465 goto out_set;
466 }
467 /*
468 * Have a difference between lov_mds_md & lov_user_md.
469 * So we have to re-order the data before copy to user.
470 */
471 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
472 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
473 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
474 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
475 if (copy_to_user(lump, lmmk, lmm_size))
476 rc = -EFAULT;
477
478 obd_free_diskmd(exp, &lmmk);
479 out_set:
480 return rc;
481 }
482