Commit | Line | Data |
---|---|---|
d7e09d03 PT |
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 <lustre_net.h> | |
46 | #include <obd.h> | |
47 | #include <obd_lov.h> | |
48 | #include <obd_class.h> | |
49 | #include <obd_support.h> | |
50 | #include <lustre/lustre_user.h> | |
51 | ||
52 | #include "lov_internal.h" | |
53 | ||
54 | static void lov_dump_lmm_common(int level, void *lmmp) | |
55 | { | |
56 | struct lov_mds_md *lmm = lmmp; | |
57 | struct ost_id oi; | |
58 | ||
59 | lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi); | |
60 | CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n", | |
61 | POSTID(&oi), le32_to_cpu(lmm->lmm_magic), | |
62 | le32_to_cpu(lmm->lmm_pattern)); | |
63 | CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n", | |
64 | le32_to_cpu(lmm->lmm_stripe_size), | |
65 | le16_to_cpu(lmm->lmm_stripe_count), | |
66 | le16_to_cpu(lmm->lmm_layout_gen)); | |
67 | } | |
68 | ||
69 | static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod, | |
70 | int stripe_count) | |
71 | { | |
72 | int i; | |
73 | ||
74 | if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) { | |
75 | CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n", | |
76 | stripe_count, LOV_V1_INSANE_STRIPE_COUNT); | |
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 | void lov_dump_lmm(int level, void *lmm) | |
104 | { | |
105 | int magic; | |
106 | ||
107 | magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic; | |
108 | switch (magic) { | |
109 | case LOV_MAGIC_V1: | |
110 | return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm)); | |
111 | case LOV_MAGIC_V3: | |
112 | return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm)); | |
113 | default: | |
114 | CERROR("Cannot recognize lmm_magic %x", magic); | |
115 | } | |
116 | return; | |
117 | } | |
118 | ||
119 | #define LMM_ASSERT(test) \ | |
120 | do { \ | |
121 | if (!(test)) lov_dump_lmm(D_ERROR, lmm); \ | |
122 | LASSERT(test); /* so we know what assertion failed */ \ | |
123 | } while(0) | |
124 | ||
125 | /* Pack LOV object metadata for disk storage. It is packed in LE byte | |
126 | * order and is opaque to the networking layer. | |
127 | * | |
128 | * XXX In the future, this will be enhanced to get the EA size from the | |
129 | * underlying OSC device(s) to get their EA sizes so we can stack | |
130 | * LOVs properly. For now lov_mds_md_size() just assumes one obd_id | |
131 | * per stripe. | |
132 | */ | |
133 | int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, | |
134 | struct lov_stripe_md *lsm) | |
135 | { | |
136 | struct obd_device *obd = class_exp2obd(exp); | |
137 | struct lov_obd *lov = &obd->u.lov; | |
138 | struct lov_mds_md_v1 *lmmv1; | |
139 | struct lov_mds_md_v3 *lmmv3; | |
140 | __u16 stripe_count; | |
141 | struct lov_ost_data_v1 *lmm_objects; | |
142 | int lmm_size, lmm_magic; | |
143 | int i; | |
144 | int cplen = 0; | |
145 | ENTRY; | |
146 | ||
147 | if (lsm) { | |
148 | lmm_magic = lsm->lsm_magic; | |
149 | } else { | |
150 | if (lmmp && *lmmp) | |
151 | lmm_magic = le32_to_cpu((*lmmp)->lmm_magic); | |
152 | else | |
153 | /* lsm == NULL and lmmp == NULL */ | |
154 | lmm_magic = LOV_MAGIC; | |
155 | } | |
156 | ||
157 | if ((lmm_magic != LOV_MAGIC_V1) && | |
158 | (lmm_magic != LOV_MAGIC_V3)) { | |
159 | CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n", | |
160 | lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3); | |
161 | RETURN(-EINVAL); | |
162 | ||
163 | } | |
164 | ||
165 | if (lsm) { | |
166 | /* If we are just sizing the EA, limit the stripe count | |
167 | * to the actual number of OSTs in this filesystem. */ | |
168 | if (!lmmp) { | |
169 | stripe_count = lov_get_stripecnt(lov, lmm_magic, | |
170 | lsm->lsm_stripe_count); | |
171 | lsm->lsm_stripe_count = stripe_count; | |
172 | } else { | |
173 | stripe_count = lsm->lsm_stripe_count; | |
174 | } | |
175 | } else { | |
176 | /* No need to allocate more than maximum supported stripes. | |
177 | * Anyway, this is pretty inaccurate since ld_tgt_count now | |
178 | * represents max index and we should rely on the actual number | |
179 | * of OSTs instead */ | |
180 | stripe_count = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize, | |
181 | lmm_magic); | |
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 */ | |
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_stripecnt(lov->lov_ocd.ocd_max_easize, | |
268 | magic); | |
269 | ||
270 | if (stripe_count > max_stripes) | |
271 | stripe_count = max_stripes; | |
272 | ||
273 | return stripe_count; | |
274 | } | |
275 | ||
276 | ||
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 | ||
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 | ENTRY; | |
310 | ||
311 | CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count); | |
312 | ||
313 | *lsmp = lsm_alloc_plain(stripe_count, &lsm_size); | |
314 | if (!*lsmp) { | |
315 | CERROR("can't allocate lsmp stripe_count %d\n", stripe_count); | |
316 | RETURN(-ENOMEM); | |
317 | } | |
318 | ||
319 | atomic_set(&(*lsmp)->lsm_refc, 1); | |
320 | spin_lock_init(&(*lsmp)->lsm_lock); | |
321 | (*lsmp)->lsm_magic = magic; | |
322 | (*lsmp)->lsm_stripe_count = stripe_count; | |
323 | (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count; | |
324 | (*lsmp)->lsm_pattern = pattern; | |
325 | (*lsmp)->lsm_pool_name[0] = '\0'; | |
326 | (*lsmp)->lsm_layout_gen = 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 | ||
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 | if ((refc = atomic_dec_return(&lsm->lsm_refc)) == 0) { | |
343 | LASSERT(lsm_op_find(lsm->lsm_magic) != NULL); | |
344 | lsm_op_find(lsm->lsm_magic)->lsm_free(lsm); | |
345 | } | |
346 | return refc; | |
347 | } | |
348 | ||
349 | ||
350 | /* Unpack LOV object metadata from disk storage. It is packed in LE byte | |
351 | * order and is opaque to the networking layer. | |
352 | */ | |
353 | int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, | |
354 | struct lov_mds_md *lmm, int lmm_bytes) | |
355 | { | |
356 | struct obd_device *obd = class_exp2obd(exp); | |
357 | struct lov_obd *lov = &obd->u.lov; | |
358 | int rc = 0, lsm_size; | |
359 | __u16 stripe_count; | |
360 | __u32 magic; | |
361 | ENTRY; | |
362 | ||
363 | /* If passed an MDS struct use values from there, otherwise defaults */ | |
364 | if (lmm) { | |
365 | rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count); | |
366 | if (rc) | |
367 | RETURN(rc); | |
368 | magic = le32_to_cpu(lmm->lmm_magic); | |
369 | } else { | |
370 | magic = LOV_MAGIC; | |
371 | stripe_count = lov_get_stripecnt(lov, magic, 0); | |
372 | } | |
373 | ||
374 | /* If we aren't passed an lsmp struct, we just want the size */ | |
375 | if (!lsmp) { | |
376 | /* XXX LOV STACKING call into osc for sizes */ | |
377 | LBUG(); | |
378 | RETURN(lov_stripe_md_size(stripe_count)); | |
379 | } | |
380 | /* If we are passed an allocated struct but nothing to unpack, free */ | |
381 | if (*lsmp && !lmm) { | |
382 | lov_free_memmd(lsmp); | |
383 | RETURN(0); | |
384 | } | |
385 | ||
386 | lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0, | |
387 | magic); | |
388 | if (lsm_size < 0) | |
389 | RETURN(lsm_size); | |
390 | ||
391 | /* If we are passed a pointer but nothing to unpack, we only alloc */ | |
392 | if (!lmm) | |
393 | RETURN(lsm_size); | |
394 | ||
395 | LASSERT(lsm_op_find(magic) != NULL); | |
396 | rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm); | |
397 | if (rc) { | |
398 | lov_free_memmd(lsmp); | |
399 | RETURN(rc); | |
400 | } | |
401 | ||
402 | RETURN(lsm_size); | |
403 | } | |
404 | ||
405 | static int __lov_setstripe(struct obd_export *exp, int max_lmm_size, | |
406 | struct lov_stripe_md **lsmp, | |
407 | struct lov_user_md *lump) | |
408 | { | |
409 | struct obd_device *obd = class_exp2obd(exp); | |
410 | struct lov_obd *lov = &obd->u.lov; | |
411 | char buffer[sizeof(struct lov_user_md_v3)]; | |
412 | struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0]; | |
413 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0]; | |
414 | int lmm_magic; | |
415 | __u16 stripe_count; | |
416 | int rc; | |
417 | int cplen = 0; | |
418 | ENTRY; | |
419 | ||
420 | rc = lov_lum_swab_if_needed(lumv3, &lmm_magic, lump); | |
421 | if (rc) | |
422 | RETURN(rc); | |
423 | ||
424 | /* in the rest of the tests, as *lumv1 and lumv3 have the same | |
425 | * fields, we use lumv1 to avoid code duplication */ | |
426 | ||
427 | if (lumv1->lmm_pattern == 0) { | |
428 | lumv1->lmm_pattern = lov->desc.ld_pattern ? | |
429 | lov->desc.ld_pattern : LOV_PATTERN_RAID0; | |
430 | } | |
431 | ||
432 | if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) { | |
433 | CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n", | |
434 | lumv1->lmm_pattern); | |
435 | RETURN(-EINVAL); | |
436 | } | |
437 | ||
438 | /* 64kB is the largest common page size we see (ia64), and matches the | |
439 | * check in lfs */ | |
440 | if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) { | |
441 | CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n", | |
442 | lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE); | |
443 | lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE; | |
444 | } | |
445 | ||
446 | if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) && | |
447 | (lumv1->lmm_stripe_offset != | |
448 | (typeof(lumv1->lmm_stripe_offset))(-1))) { | |
449 | CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n", | |
450 | lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count); | |
451 | RETURN(-EINVAL); | |
452 | } | |
453 | stripe_count = lov_get_stripecnt(lov, lmm_magic, | |
454 | lumv1->lmm_stripe_count); | |
455 | ||
456 | if (max_lmm_size) { | |
457 | int max_stripes = (max_lmm_size - | |
458 | lov_mds_md_size(0, lmm_magic)) / | |
459 | sizeof(struct lov_ost_data_v1); | |
460 | if (unlikely(max_stripes < stripe_count)) { | |
461 | CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n", | |
462 | stripe_count, max_stripes); | |
463 | stripe_count = max_stripes; | |
464 | } | |
465 | } | |
466 | ||
467 | if (lmm_magic == LOV_USER_MAGIC_V3) { | |
468 | struct pool_desc *pool; | |
469 | ||
470 | /* In the function below, .hs_keycmp resolves to | |
471 | * pool_hashkey_keycmp() */ | |
472 | /* coverity[overrun-buffer-val] */ | |
473 | pool = lov_find_pool(lov, lumv3->lmm_pool_name); | |
474 | if (pool != NULL) { | |
475 | if (lumv3->lmm_stripe_offset != | |
476 | (typeof(lumv3->lmm_stripe_offset))(-1)) { | |
477 | rc = lov_check_index_in_pool( | |
478 | lumv3->lmm_stripe_offset, pool); | |
479 | if (rc < 0) { | |
480 | lov_pool_putref(pool); | |
481 | RETURN(-EINVAL); | |
482 | } | |
483 | } | |
484 | ||
485 | if (stripe_count > pool_tgt_count(pool)) | |
486 | stripe_count = pool_tgt_count(pool); | |
487 | ||
488 | lov_pool_putref(pool); | |
489 | } | |
490 | } | |
491 | ||
492 | rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic); | |
493 | ||
494 | if (rc >= 0) { | |
495 | (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset; | |
496 | (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size; | |
497 | if (lmm_magic == LOV_USER_MAGIC_V3) { | |
498 | cplen = strlcpy((*lsmp)->lsm_pool_name, | |
499 | lumv3->lmm_pool_name, | |
500 | sizeof((*lsmp)->lsm_pool_name)); | |
501 | if (cplen >= sizeof((*lsmp)->lsm_pool_name)) | |
502 | rc = -E2BIG; | |
503 | } | |
504 | rc = 0; | |
505 | } | |
506 | ||
507 | RETURN(rc); | |
508 | } | |
509 | ||
510 | /* Configure object striping information on a new file. | |
511 | * | |
512 | * @lmmu is a pointer to a user struct with one or more of the fields set to | |
513 | * indicate the application preference: lmm_stripe_count, lmm_stripe_size, | |
514 | * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC. | |
515 | * @lsmp is a pointer to an in-core stripe MD that needs to be filled in. | |
516 | */ | |
517 | int lov_setstripe(struct obd_export *exp, int max_lmm_size, | |
518 | struct lov_stripe_md **lsmp, struct lov_user_md *lump) | |
519 | { | |
520 | int rc; | |
521 | mm_segment_t seg; | |
522 | ||
523 | seg = get_fs(); | |
524 | set_fs(KERNEL_DS); | |
525 | ||
526 | rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump); | |
527 | set_fs(seg); | |
528 | RETURN(rc); | |
529 | } | |
530 | ||
531 | int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp, | |
532 | struct lov_user_md *lump) | |
533 | { | |
534 | int i; | |
535 | int rc; | |
536 | struct obd_export *oexp; | |
537 | struct lov_obd *lov = &exp->exp_obd->u.lov; | |
538 | obd_id last_id = 0; | |
539 | struct lov_user_ost_data_v1 *lmm_objects; | |
540 | ||
541 | ENTRY; | |
542 | ||
543 | if (lump->lmm_magic == LOV_USER_MAGIC_V3) | |
544 | lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects; | |
545 | else | |
546 | lmm_objects = lump->lmm_objects; | |
547 | ||
548 | for (i = 0; i < lump->lmm_stripe_count; i++) { | |
549 | __u32 len = sizeof(last_id); | |
550 | oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp; | |
551 | rc = obd_get_info(NULL, oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID, | |
552 | &len, &last_id, NULL); | |
553 | if (rc) | |
554 | RETURN(rc); | |
555 | if (ostid_id(&lmm_objects[i].l_ost_oi) > last_id) { | |
556 | CERROR("Setting EA for object > than last id on" | |
557 | " ost idx %d "DOSTID" > "LPD64" \n", | |
558 | lmm_objects[i].l_ost_idx, | |
559 | POSTID(&lmm_objects[i].l_ost_oi), last_id); | |
560 | RETURN(-EINVAL); | |
561 | } | |
562 | } | |
563 | ||
564 | rc = lov_setstripe(exp, 0, lsmp, lump); | |
565 | if (rc) | |
566 | RETURN(rc); | |
567 | ||
568 | for (i = 0; i < lump->lmm_stripe_count; i++) { | |
569 | (*lsmp)->lsm_oinfo[i]->loi_ost_idx = | |
570 | lmm_objects[i].l_ost_idx; | |
571 | (*lsmp)->lsm_oinfo[i]->loi_oi = lmm_objects[i].l_ost_oi; | |
572 | } | |
573 | RETURN(0); | |
574 | } | |
575 | ||
576 | ||
577 | /* Retrieve object striping information. | |
578 | * | |
579 | * @lump is a pointer to an in-core struct with lmm_ost_count indicating | |
580 | * the maximum number of OST indices which will fit in the user buffer. | |
581 | * lmm_magic must be LOV_USER_MAGIC. | |
582 | */ | |
583 | int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm, | |
584 | struct lov_user_md *lump) | |
585 | { | |
586 | /* | |
587 | * XXX huge struct allocated on stack. | |
588 | */ | |
589 | /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */ | |
590 | struct lov_user_md_v3 lum; | |
591 | struct lov_mds_md *lmmk = NULL; | |
592 | int rc, lmm_size; | |
593 | int lum_size; | |
594 | mm_segment_t seg; | |
595 | ENTRY; | |
596 | ||
597 | if (!lsm) | |
598 | RETURN(-ENODATA); | |
599 | ||
600 | /* | |
601 | * "Switch to kernel segment" to allow copying from kernel space by | |
602 | * copy_{to,from}_user(). | |
603 | */ | |
604 | seg = get_fs(); | |
605 | set_fs(KERNEL_DS); | |
606 | ||
607 | /* we only need the header part from user space to get lmm_magic and | |
608 | * lmm_stripe_count, (the header part is common to v1 and v3) */ | |
609 | lum_size = sizeof(struct lov_user_md_v1); | |
610 | if (copy_from_user(&lum, lump, lum_size)) | |
611 | GOTO(out_set, rc = -EFAULT); | |
612 | else if ((lum.lmm_magic != LOV_USER_MAGIC) && | |
613 | (lum.lmm_magic != LOV_USER_MAGIC_V3)) | |
614 | GOTO(out_set, rc = -EINVAL); | |
615 | ||
616 | if (lum.lmm_stripe_count && | |
617 | (lum.lmm_stripe_count < lsm->lsm_stripe_count)) { | |
618 | /* Return right size of stripe to user */ | |
619 | lum.lmm_stripe_count = lsm->lsm_stripe_count; | |
620 | rc = copy_to_user(lump, &lum, lum_size); | |
621 | GOTO(out_set, rc = -EOVERFLOW); | |
622 | } | |
623 | rc = lov_packmd(exp, &lmmk, lsm); | |
624 | if (rc < 0) | |
625 | GOTO(out_set, rc); | |
626 | lmm_size = rc; | |
627 | rc = 0; | |
628 | ||
629 | /* FIXME: Bug 1185 - copy fields properly when structs change */ | |
630 | /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */ | |
631 | CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3)); | |
632 | CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]); | |
633 | ||
634 | if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) && | |
635 | ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) || | |
636 | (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) { | |
637 | lustre_swab_lov_mds_md(lmmk); | |
638 | lustre_swab_lov_user_md_objects( | |
639 | (struct lov_user_ost_data*)lmmk->lmm_objects, | |
640 | lmmk->lmm_stripe_count); | |
641 | } | |
642 | if (lum.lmm_magic == LOV_USER_MAGIC) { | |
643 | /* User request for v1, we need skip lmm_pool_name */ | |
644 | if (lmmk->lmm_magic == LOV_MAGIC_V3) { | |
645 | memmove((char*)(&lmmk->lmm_stripe_count) + | |
646 | sizeof(lmmk->lmm_stripe_count), | |
647 | ((struct lov_mds_md_v3*)lmmk)->lmm_objects, | |
648 | lmmk->lmm_stripe_count * | |
649 | sizeof(struct lov_ost_data_v1)); | |
650 | lmm_size -= LOV_MAXPOOLNAME; | |
651 | } | |
652 | } else { | |
653 | /* if v3 we just have to update the lum_size */ | |
654 | lum_size = sizeof(struct lov_user_md_v3); | |
655 | } | |
656 | ||
657 | /* User wasn't expecting this many OST entries */ | |
658 | if (lum.lmm_stripe_count == 0) | |
659 | lmm_size = lum_size; | |
660 | else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) | |
661 | GOTO(out_set, rc = -EOVERFLOW); | |
662 | /* | |
663 | * Have a difference between lov_mds_md & lov_user_md. | |
664 | * So we have to re-order the data before copy to user. | |
665 | */ | |
666 | lum.lmm_stripe_count = lmmk->lmm_stripe_count; | |
667 | lum.lmm_layout_gen = lmmk->lmm_layout_gen; | |
668 | ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen; | |
669 | ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count; | |
670 | if (copy_to_user(lump, lmmk, lmm_size)) | |
671 | rc = -EFAULT; | |
672 | ||
673 | obd_free_diskmd(exp, &lmmk); | |
674 | out_set: | |
675 | set_fs(seg); | |
676 | RETURN(rc); | |
677 | } |