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
lov_dump_lmm(int level,void * lmm)103 void lov_dump_lmm(int level, void *lmm)
104 {
105 int magic;
106
107 magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
108 switch (magic) {
109 case LOV_MAGIC_V1:
110 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
111 break;
112 case LOV_MAGIC_V3:
113 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
114 break;
115 default:
116 CDEBUG(level, "unrecognized lmm_magic %x, assuming %x\n",
117 magic, LOV_MAGIC_V1);
118 lov_dump_lmm_common(level, lmm);
119 break;
120 }
121 }
122
123 /* Pack LOV object metadata for disk storage. It is packed in LE byte
124 * order and is opaque to the networking layer.
125 *
126 * XXX In the future, this will be enhanced to get the EA size from the
127 * underlying OSC device(s) to get their EA sizes so we can stack
128 * LOVs properly. For now lov_mds_md_size() just assumes one u64
129 * per stripe.
130 */
lov_packmd(struct obd_export * exp,struct lov_mds_md ** lmmp,struct lov_stripe_md * lsm)131 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
132 struct lov_stripe_md *lsm)
133 {
134 struct obd_device *obd = class_exp2obd(exp);
135 struct lov_obd *lov = &obd->u.lov;
136 struct lov_mds_md_v1 *lmmv1;
137 struct lov_mds_md_v3 *lmmv3;
138 __u16 stripe_count;
139 struct lov_ost_data_v1 *lmm_objects;
140 int lmm_size, lmm_magic;
141 int i;
142 int cplen = 0;
143
144 if (lsm) {
145 lmm_magic = lsm->lsm_magic;
146 } else {
147 if (lmmp && *lmmp)
148 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
149 else
150 /* lsm == NULL and lmmp == NULL */
151 lmm_magic = LOV_MAGIC;
152 }
153
154 if ((lmm_magic != LOV_MAGIC_V1) &&
155 (lmm_magic != LOV_MAGIC_V3)) {
156 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
157 lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
158 return -EINVAL;
159
160 }
161
162 if (lsm) {
163 /* If we are just sizing the EA, limit the stripe count
164 * to the actual number of OSTs in this filesystem. */
165 if (!lmmp) {
166 stripe_count = lov_get_stripecnt(lov, lmm_magic,
167 lsm->lsm_stripe_count);
168 lsm->lsm_stripe_count = stripe_count;
169 } else if (!lsm_is_released(lsm)) {
170 stripe_count = lsm->lsm_stripe_count;
171 } else {
172 stripe_count = 0;
173 }
174 } else {
175 /* No need to allocate more than maximum supported stripes.
176 * Anyway, this is pretty inaccurate since ld_tgt_count now
177 * represents max index and we should rely on the actual number
178 * of OSTs instead */
179 stripe_count = lov_mds_md_max_stripe_count(
180 lov->lov_ocd.ocd_max_easize, lmm_magic);
181
182 if (stripe_count > lov->desc.ld_tgt_count)
183 stripe_count = lov->desc.ld_tgt_count;
184 }
185
186 /* XXX LOV STACKING call into osc for sizes */
187 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
188
189 if (!lmmp)
190 return lmm_size;
191
192 if (*lmmp && !lsm) {
193 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
194 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
195 OBD_FREE_LARGE(*lmmp, lmm_size);
196 *lmmp = NULL;
197 return 0;
198 }
199
200 if (!*lmmp) {
201 OBD_ALLOC_LARGE(*lmmp, lmm_size);
202 if (!*lmmp)
203 return -ENOMEM;
204 }
205
206 CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
207 lmm_magic, lmm_size);
208
209 lmmv1 = *lmmp;
210 lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
211 if (lmm_magic == LOV_MAGIC_V3)
212 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
213 else
214 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
215
216 if (!lsm)
217 return lmm_size;
218
219 /* lmmv1 and lmmv3 point to the same struct and have the
220 * same first fields
221 */
222 lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
223 lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
224 lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
225 lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
226 lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
227 if (lsm->lsm_magic == LOV_MAGIC_V3) {
228 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
229 sizeof(lmmv3->lmm_pool_name));
230 if (cplen >= sizeof(lmmv3->lmm_pool_name))
231 return -E2BIG;
232 lmm_objects = lmmv3->lmm_objects;
233 } else {
234 lmm_objects = lmmv1->lmm_objects;
235 }
236
237 for (i = 0; i < stripe_count; i++) {
238 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
239 /* XXX LOV STACKING call down to osc_packmd() to do packing */
240 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
241 " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
242 i, stripe_count, loi->loi_ost_idx);
243 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
244 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
245 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
246 }
247
248 return lmm_size;
249 }
250
251 /* Find the max stripecount we should use */
lov_get_stripecnt(struct lov_obd * lov,__u32 magic,__u16 stripe_count)252 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
253 {
254 __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
255
256 if (!stripe_count)
257 stripe_count = lov->desc.ld_default_stripe_count;
258 if (stripe_count > lov->desc.ld_active_tgt_count)
259 stripe_count = lov->desc.ld_active_tgt_count;
260 if (!stripe_count)
261 stripe_count = 1;
262
263 /* stripe count is based on whether ldiskfs can handle
264 * larger EA sizes */
265 if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
266 lov->lov_ocd.ocd_max_easize)
267 max_stripes = lov_mds_md_max_stripe_count(
268 lov->lov_ocd.ocd_max_easize, magic);
269
270 if (stripe_count > max_stripes)
271 stripe_count = max_stripes;
272
273 return stripe_count;
274 }
275
276
lov_verify_lmm(void * lmm,int lmm_bytes,__u16 * stripe_count)277 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
278 {
279 int rc;
280
281 if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
282 char *buffer;
283 int sz;
284
285 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
286 le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
287 sz = lmm_bytes * 2 + 1;
288 OBD_ALLOC_LARGE(buffer, sz);
289 if (buffer != NULL) {
290 int i;
291
292 for (i = 0; i < lmm_bytes; i++)
293 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
294 buffer[sz - 1] = '\0';
295 CERROR("%s\n", buffer);
296 OBD_FREE_LARGE(buffer, sz);
297 }
298 return -EINVAL;
299 }
300 rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
301 lmm_bytes, stripe_count);
302 return rc;
303 }
304
lov_alloc_memmd(struct lov_stripe_md ** lsmp,__u16 stripe_count,int pattern,int magic)305 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
306 int pattern, int magic)
307 {
308 int i, lsm_size;
309
310 CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
311
312 *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
313 if (!*lsmp) {
314 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
315 return -ENOMEM;
316 }
317
318 atomic_set(&(*lsmp)->lsm_refc, 1);
319 spin_lock_init(&(*lsmp)->lsm_lock);
320 (*lsmp)->lsm_magic = magic;
321 (*lsmp)->lsm_stripe_count = stripe_count;
322 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
323 (*lsmp)->lsm_pattern = pattern;
324 (*lsmp)->lsm_pool_name[0] = '\0';
325 (*lsmp)->lsm_layout_gen = 0;
326 if (stripe_count > 0)
327 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
328
329 for (i = 0; i < stripe_count; i++)
330 loi_init((*lsmp)->lsm_oinfo[i]);
331
332 return lsm_size;
333 }
334
lov_free_memmd(struct lov_stripe_md ** lsmp)335 int lov_free_memmd(struct lov_stripe_md **lsmp)
336 {
337 struct lov_stripe_md *lsm = *lsmp;
338 int refc;
339
340 *lsmp = NULL;
341 LASSERT(atomic_read(&lsm->lsm_refc) > 0);
342 refc = atomic_dec_return(&lsm->lsm_refc);
343 if (refc == 0) {
344 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
345 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
346 }
347 return refc;
348 }
349
350
351 /* Unpack LOV object metadata from disk storage. It is packed in LE byte
352 * order and is opaque to the networking layer.
353 */
lov_unpackmd(struct obd_export * exp,struct lov_stripe_md ** lsmp,struct lov_mds_md * lmm,int lmm_bytes)354 int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
355 struct lov_mds_md *lmm, int lmm_bytes)
356 {
357 struct obd_device *obd = class_exp2obd(exp);
358 struct lov_obd *lov = &obd->u.lov;
359 int rc = 0, lsm_size;
360 __u16 stripe_count;
361 __u32 magic;
362 __u32 pattern;
363
364 /* If passed an MDS struct use values from there, otherwise defaults */
365 if (lmm) {
366 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
367 if (rc)
368 return rc;
369 magic = le32_to_cpu(lmm->lmm_magic);
370 } else {
371 magic = LOV_MAGIC;
372 stripe_count = lov_get_stripecnt(lov, magic, 0);
373 }
374
375 /* If we aren't passed an lsmp struct, we just want the size */
376 if (!lsmp) {
377 /* XXX LOV STACKING call into osc for sizes */
378 LBUG();
379 return lov_stripe_md_size(stripe_count);
380 }
381 /* If we are passed an allocated struct but nothing to unpack, free */
382 if (*lsmp && !lmm) {
383 lov_free_memmd(lsmp);
384 return 0;
385 }
386
387 pattern = le32_to_cpu(lmm->lmm_pattern);
388 lsm_size = lov_alloc_memmd(lsmp, stripe_count, pattern, magic);
389 if (lsm_size < 0)
390 return lsm_size;
391
392 /* If we are passed a pointer but nothing to unpack, we only alloc */
393 if (!lmm)
394 return lsm_size;
395
396 LASSERT(lsm_op_find(magic) != NULL);
397 rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
398 if (rc) {
399 lov_free_memmd(lsmp);
400 return rc;
401 }
402
403 return lsm_size;
404 }
405
406 /* Retrieve object striping information.
407 *
408 * @lump is a pointer to an in-core struct with lmm_ost_count indicating
409 * the maximum number of OST indices which will fit in the user buffer.
410 * lmm_magic must be LOV_USER_MAGIC.
411 */
lov_getstripe(struct obd_export * exp,struct lov_stripe_md * lsm,struct lov_user_md * lump)412 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
413 struct lov_user_md *lump)
414 {
415 /*
416 * XXX huge struct allocated on stack.
417 */
418 /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
419 struct lov_user_md_v3 lum;
420 struct lov_mds_md *lmmk = NULL;
421 int rc, lmm_size;
422 int lum_size;
423 mm_segment_t seg;
424
425 if (!lsm)
426 return -ENODATA;
427
428 /*
429 * "Switch to kernel segment" to allow copying from kernel space by
430 * copy_{to,from}_user().
431 */
432 seg = get_fs();
433 set_fs(KERNEL_DS);
434
435 /* we only need the header part from user space to get lmm_magic and
436 * lmm_stripe_count, (the header part is common to v1 and v3) */
437 lum_size = sizeof(struct lov_user_md_v1);
438 if (copy_from_user(&lum, lump, lum_size)) {
439 rc = -EFAULT;
440 goto out_set;
441 } else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
442 (lum.lmm_magic != LOV_USER_MAGIC_V3)) {
443 rc = -EINVAL;
444 goto out_set;
445 }
446
447 if (lum.lmm_stripe_count &&
448 (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
449 /* Return right size of stripe to user */
450 lum.lmm_stripe_count = lsm->lsm_stripe_count;
451 rc = copy_to_user(lump, &lum, lum_size);
452 rc = -EOVERFLOW;
453 goto out_set;
454 }
455 rc = lov_packmd(exp, &lmmk, lsm);
456 if (rc < 0)
457 goto out_set;
458 lmm_size = rc;
459 rc = 0;
460
461 /* FIXME: Bug 1185 - copy fields properly when structs change */
462 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
463 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
464 CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
465
466 if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
467 ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
468 (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
469 lustre_swab_lov_mds_md(lmmk);
470 lustre_swab_lov_user_md_objects(
471 (struct lov_user_ost_data *)lmmk->lmm_objects,
472 lmmk->lmm_stripe_count);
473 }
474 if (lum.lmm_magic == LOV_USER_MAGIC) {
475 /* User request for v1, we need skip lmm_pool_name */
476 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
477 memmove((char *)(&lmmk->lmm_stripe_count) +
478 sizeof(lmmk->lmm_stripe_count),
479 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
480 lmmk->lmm_stripe_count *
481 sizeof(struct lov_ost_data_v1));
482 lmm_size -= LOV_MAXPOOLNAME;
483 }
484 } else {
485 /* if v3 we just have to update the lum_size */
486 lum_size = sizeof(struct lov_user_md_v3);
487 }
488
489 /* User wasn't expecting this many OST entries */
490 if (lum.lmm_stripe_count == 0)
491 lmm_size = lum_size;
492 else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
493 rc = -EOVERFLOW;
494 goto out_set;
495 }
496 /*
497 * Have a difference between lov_mds_md & lov_user_md.
498 * So we have to re-order the data before copy to user.
499 */
500 lum.lmm_stripe_count = lmmk->lmm_stripe_count;
501 lum.lmm_layout_gen = lmmk->lmm_layout_gen;
502 ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
503 ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
504 if (copy_to_user(lump, lmmk, lmm_size))
505 rc = -EFAULT;
506
507 obd_free_diskmd(exp, &lmmk);
508 out_set:
509 set_fs(seg);
510 return rc;
511 }
512