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.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, Whamcloud, Inc.
  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 #ifndef __KERNEL__
  45 #include <liblustre.h>
  46 #endif
  47
  48 #include <lustre_net.h>
  49 #include <obd.h>
  50 #include <obd_lov.h>
  51 #include <obd_class.h>
  52 #include <obd_support.h>
  53 #include <lustre/lustre_user.h>
  54
  55 #include "lov_internal.h"
  56
  57 static void lov_dump_lmm_common(int level, void *lmmp)
  58 {
  59         struct lov_mds_md *lmm = lmmp;
  60
  61         CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
  62                (__u64)le64_to_cpu(lmm->lmm_object_id),
  63                le32_to_cpu(lmm->lmm_magic),
  64                le32_to_cpu(lmm->lmm_pattern));
  65         CDEBUG(level,"stripe_size %u, stripe_count %u, layout_gen %u\n",
  66                le32_to_cpu(lmm->lmm_stripe_size),
  67                le16_to_cpu(lmm->lmm_stripe_count),
  68                le16_to_cpu(lmm->lmm_layout_gen));
  69 }
  70
  71 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  72                                  int stripe_count)
  73 {
  74         int i;
  75
  76         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  77                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  78                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  79         }
  80
  81         for (i = 0; i < stripe_count; ++i, ++lod) {
  82                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n", i,
  83                        le32_to_cpu(lod->l_ost_idx),
  84                        (__u64)le64_to_cpu(lod->l_object_seq),
  85                        (__u64)le64_to_cpu(lod->l_object_id));
  86         }
  87 }
  88
  89 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  90 {
  91         lov_dump_lmm_common(level, lmm);
  92         lov_dump_lmm_objects(level, lmm->lmm_objects,
  93                              le16_to_cpu(lmm->lmm_stripe_count));
  94 }
  95
  96 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
  97 {
  98         lov_dump_lmm_common(level, lmm);
  99         CDEBUG(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));
 102 }
 103
 104 void lov_dump_lmm(int level, void *lmm)
 105 {
 106         int magic;
 107
 108         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 109         switch (magic) {
 110         case LOV_MAGIC_V1:
 111                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 112         case LOV_MAGIC_V3:
 113                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 114         default:
 115                 CERROR("Cannot recognize lmm_magic %x", magic);
 116         }
 117         return;
 118 }
 119
 120 #define LMM_ASSERT(test)                                                \
 121 do {                                                                    \
 122         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 123         LASSERT(test); /* so we know what assertion failed */           \
 124 } while(0)
 125
 126 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 127  * order and is opaque to the networking layer.
 128  *
 129  * XXX In the future, this will be enhanced to get the EA size from the
 130  *     underlying OSC device(s) to get their EA sizes so we can stack
 131  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 132  *     per stripe.
 133  */
 134 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 135                struct lov_stripe_md *lsm)
 136 {
 137         struct obd_device *obd = class_exp2obd(exp);
 138         struct lov_obd *lov = &obd->u.lov;
 139         struct lov_mds_md_v1 *lmmv1;
 140         struct lov_mds_md_v3 *lmmv3;
 141         __u16 stripe_count;
 142         struct lov_ost_data_v1 *lmm_objects;
 143         int lmm_size, lmm_magic;
 144         int i;
 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         lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
 223         lmmv1->lmm_object_seq = cpu_to_le64(lsm->lsm_object_seq);
 224         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 225         lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
 226         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 227         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 228         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 229                 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 230                         LOV_MAXPOOLNAME);
 231                 lmm_objects = lmmv3->lmm_objects;
 232         } else {
 233                 lmm_objects = lmmv1->lmm_objects;
 234         }
 235
 236         for (i = 0; i < stripe_count; i++) {
 237                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 238                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 239                 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
 240                          lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
 241                 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
 242                 lmm_objects[i].l_object_seq = cpu_to_le64(loi->loi_seq);
 243                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 244                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 245         }
 246
 247         RETURN(lmm_size);
 248 }
 249
 250 /* Find the max stripecount we should use */
 251 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 252 {
 253         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 254
 255         if (!stripe_count)
 256                 stripe_count = lov->desc.ld_default_stripe_count;
 257         if (stripe_count > lov->desc.ld_active_tgt_count)
 258                 stripe_count = lov->desc.ld_active_tgt_count;
 259         if (!stripe_count)
 260                 stripe_count = 1;
 261
 262         /* stripe count is based on whether ldiskfs can handle
 263          * larger EA sizes */
 264         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 265             lov->lov_ocd.ocd_max_easize)
 266                 max_stripes = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
 267                                                    magic);
 268
 269         if (stripe_count > max_stripes)
 270                 stripe_count = max_stripes;
 271
 272         return stripe_count;
 273 }
 274
 275
 276 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
 277 {
 278         int rc;
 279
 280         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 281                 char *buffer;
 282                 int sz;
 283
 284                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 285                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 286                 sz = lmm_bytes * 2 + 1;
 287                 OBD_ALLOC_LARGE(buffer, sz);
 288                 if (buffer != NULL) {
 289                         int i;
 290
 291                         for (i = 0; i < lmm_bytes; i++)
 292                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 293                         buffer[sz] = '\0';
 294                         CERROR("%s\n", buffer);
 295                         OBD_FREE_LARGE(buffer, sz);
 296                 }
 297                 return -EINVAL;
 298         }
 299         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 300                                      lmm_bytes, stripe_count);
 301         return rc;
 302 }
 303
 304 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
 305                     int pattern, int magic)
 306 {
 307         int i, lsm_size;
 308         ENTRY;
 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         cfs_spin_lock_init(&(*lsmp)->lsm_lock);
 319         (*lsmp)->lsm_magic = magic;
 320         (*lsmp)->lsm_stripe_count = stripe_count;
 321         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 322         (*lsmp)->lsm_pattern = pattern;
 323         (*lsmp)->lsm_pool_name[0] = '\0';
 324         (*lsmp)->lsm_layout_gen = 0;
 325         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 326
 327         for (i = 0; i < stripe_count; i++)
 328                 loi_init((*lsmp)->lsm_oinfo[i]);
 329
 330         RETURN(lsm_size);
 331 }
 332
 333 void lov_free_memmd(struct lov_stripe_md **lsmp)
 334 {
 335         struct lov_stripe_md *lsm = *lsmp;
 336
 337         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 338         lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 339
 340         *lsmp = NULL;
 341 }
 342
 343
 344 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 345  * order and is opaque to the networking layer.
 346  */
 347 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 348                  struct lov_mds_md *lmm, int lmm_bytes)
 349 {
 350         struct obd_device *obd = class_exp2obd(exp);
 351         struct lov_obd *lov = &obd->u.lov;
 352         int rc = 0, lsm_size;
 353         __u16 stripe_count;
 354         __u32 magic;
 355         ENTRY;
 356
 357         /* If passed an MDS struct use values from there, otherwise defaults */
 358         if (lmm) {
 359                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 360                 if (rc)
 361                         RETURN(rc);
 362                 magic = le32_to_cpu(lmm->lmm_magic);
 363         } else {
 364                 magic = LOV_MAGIC;
 365                 stripe_count = lov_get_stripecnt(lov, magic, 0);
 366         }
 367
 368         /* If we aren't passed an lsmp struct, we just want the size */
 369         if (!lsmp) {
 370                 /* XXX LOV STACKING call into osc for sizes */
 371                 LBUG();
 372                 RETURN(lov_stripe_md_size(stripe_count));
 373         }
 374         /* If we are passed an allocated struct but nothing to unpack, free */
 375         if (*lsmp && !lmm) {
 376                 lov_free_memmd(lsmp);
 377                 RETURN(0);
 378         }
 379
 380         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 381                                    magic);
 382         if (lsm_size < 0)
 383                 RETURN(lsm_size);
 384
 385         /* If we are passed a pointer but nothing to unpack, we only alloc */
 386         if (!lmm)
 387                 RETURN(lsm_size);
 388
 389         LASSERT(lsm_op_find(magic) != NULL);
 390         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 391         if (rc) {
 392                 lov_free_memmd(lsmp);
 393                 RETURN(rc);
 394         }
 395
 396         RETURN(lsm_size);
 397 }
 398
 399 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
 400                            struct lov_stripe_md **lsmp,
 401                            struct lov_user_md *lump)
 402 {
 403         struct obd_device *obd = class_exp2obd(exp);
 404         struct lov_obd *lov = &obd->u.lov;
 405         char buffer[sizeof(struct lov_user_md_v3)];
 406         struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0];
 407         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0];
 408         int lmm_magic;
 409         __u16 stripe_count;
 410         int rc;
 411         ENTRY;
 412
 413         rc = lov_lum_swab_if_needed(lumv3, &lmm_magic, lump);
 414         if (rc)
 415                 RETURN(rc);
 416
 417         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 418          * fields, we use lumv1 to avoid code duplication */
 419
 420         if (lumv1->lmm_pattern == 0) {
 421                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 422                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 423         }
 424
 425         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 426                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 427                        lumv1->lmm_pattern);
 428                 RETURN(-EINVAL);
 429         }
 430
 431         /* 64kB is the largest common page size we see (ia64), and matches the
 432          * check in lfs */
 433         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 434                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 435                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 436                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 437         }
 438
 439         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 440             (lumv1->lmm_stripe_offset !=
 441              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 442                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 443                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 444                 RETURN(-EINVAL);
 445         }
 446         stripe_count = lov_get_stripecnt(lov, lmm_magic,
 447                                          lumv1->lmm_stripe_count);
 448
 449         if (max_lmm_size) {
 450                 int max_stripes = (max_lmm_size -
 451                                    lov_mds_md_size(0, lmm_magic)) /
 452                                    sizeof(struct lov_ost_data_v1);
 453                 if (unlikely(max_stripes < stripe_count)) {
 454                         CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
 455                                stripe_count, max_stripes);
 456                         stripe_count = max_stripes;
 457                 }
 458         }
 459
 460         if (lmm_magic == LOV_USER_MAGIC_V3) {
 461                 struct pool_desc *pool;
 462
 463                 pool = lov_find_pool(lov, lumv3->lmm_pool_name);
 464                 if (pool != NULL) {
 465                         if (lumv3->lmm_stripe_offset !=
 466                             (typeof(lumv3->lmm_stripe_offset))(-1)) {
 467                                 rc = lov_check_index_in_pool(
 468                                         lumv3->lmm_stripe_offset, pool);
 469                                 if (rc < 0) {
 470                                         lov_pool_putref(pool);
 471                                         RETURN(-EINVAL);
 472                                 }
 473                         }
 474
 475                         if (stripe_count > pool_tgt_count(pool))
 476                                 stripe_count = pool_tgt_count(pool);
 477
 478                         lov_pool_putref(pool);
 479                 }
 480         }
 481
 482         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 483
 484         if (rc >= 0) {
 485                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 486                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 487                 if (lmm_magic == LOV_USER_MAGIC_V3)
 488                         strncpy((*lsmp)->lsm_pool_name, lumv3->lmm_pool_name,
 489                                 LOV_MAXPOOLNAME);
 490                 rc = 0;
 491         }
 492
 493         RETURN(rc);
 494 }
 495
 496 /* Configure object striping information on a new file.
 497  *
 498  * @lmmu is a pointer to a user struct with one or more of the fields set to
 499  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 500  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 501  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 502  */
 503 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
 504                   struct lov_stripe_md **lsmp, struct lov_user_md *lump)
 505 {
 506         int rc;
 507         mm_segment_t seg;
 508
 509         seg = get_fs();
 510         set_fs(KERNEL_DS);
 511
 512         rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
 513         set_fs(seg);
 514         RETURN(rc);
 515 }
 516
 517 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 518               struct lov_user_md *lump)
 519 {
 520         int i;
 521         int rc;
 522         struct obd_export *oexp;
 523         struct lov_obd *lov = &exp->exp_obd->u.lov;
 524         obd_id last_id = 0;
 525         struct lov_user_ost_data_v1 *lmm_objects;
 526
 527         ENTRY;
 528
 529         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 530                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 531         else
 532                 lmm_objects = lump->lmm_objects;
 533
 534         for (i = 0; i < lump->lmm_stripe_count; i++) {
 535                 __u32 len = sizeof(last_id);
 536                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 537                 rc = obd_get_info(NULL, oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 538                                   &len, &last_id, NULL);
 539                 if (rc)
 540                         RETURN(rc);
 541                 if (lmm_objects[i].l_object_id > last_id) {
 542                         CERROR("Setting EA for object > than last id on "
 543                                "ost idx %d "LPD64" > "LPD64" \n",
 544                                lmm_objects[i].l_ost_idx,
 545                                lmm_objects[i].l_object_id, last_id);
 546                         RETURN(-EINVAL);
 547                 }
 548         }
 549
 550         rc = lov_setstripe(exp, 0, lsmp, lump);
 551         if (rc)
 552                 RETURN(rc);
 553
 554         for (i = 0; i < lump->lmm_stripe_count; i++) {
 555                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 556                         lmm_objects[i].l_ost_idx;
 557                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 558                 (*lsmp)->lsm_oinfo[i]->loi_seq = lmm_objects[i].l_object_seq;
 559         }
 560         RETURN(0);
 561 }
 562
 563
 564 /* Retrieve object striping information.
 565  *
 566  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 567  * the maximum number of OST indices which will fit in the user buffer.
 568  * lmm_magic must be LOV_USER_MAGIC.
 569  */
 570 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
 571                   struct lov_user_md *lump)
 572 {
 573         /*
 574          * XXX huge struct allocated on stack.
 575          */
 576         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 577         struct lov_user_md_v3 lum;
 578         struct lov_mds_md *lmmk = NULL;
 579         int rc, lmm_size;
 580         int lum_size;
 581         mm_segment_t seg;
 582         ENTRY;
 583
 584         if (!lsm)
 585                 RETURN(-ENODATA);
 586
 587         /*
 588          * "Switch to kernel segment" to allow copying from kernel space by
 589          * copy_{to,from}_user().
 590          */
 591         seg = get_fs();
 592         set_fs(KERNEL_DS);
 593
 594         /* we only need the header part from user space to get lmm_magic and
 595          * lmm_stripe_count, (the header part is common to v1 and v3) */
 596         lum_size = sizeof(struct lov_user_md_v1);
 597         if (cfs_copy_from_user(&lum, lump, lum_size))
 598                 GOTO(out_set, rc = -EFAULT);
 599         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 600                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 601                 GOTO(out_set, rc = -EINVAL);
 602
 603         if (lum.lmm_stripe_count &&
 604             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 605                 /* Return right size of stripe to user */
 606                 lum.lmm_stripe_count = lsm->lsm_stripe_count;
 607                 rc = cfs_copy_to_user(lump, &lum, lum_size);
 608                 GOTO(out_set, rc = -EOVERFLOW);
 609         }
 610         rc = lov_packmd(exp, &lmmk, lsm);
 611         if (rc < 0)
 612                 GOTO(out_set, rc);
 613         lmm_size = rc;
 614         rc = 0;
 615
 616         /* FIXME: Bug 1185 - copy fields properly when structs change */
 617         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 618         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 619         CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
 620
 621         if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
 622             ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
 623             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
 624                 lustre_swab_lov_mds_md(lmmk);
 625                 lustre_swab_lov_user_md_objects(
 626                                 (struct lov_user_ost_data*)lmmk->lmm_objects,
 627                                 lmmk->lmm_stripe_count);
 628         }
 629         if (lum.lmm_magic == LOV_USER_MAGIC) {
 630                 /* User request for v1, we need skip lmm_pool_name */
 631                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 632                         memmove((char*)(&lmmk->lmm_stripe_count) +
 633                                 sizeof(lmmk->lmm_stripe_count),
 634                                 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
 635                                 lmmk->lmm_stripe_count *
 636                                 sizeof(struct lov_ost_data_v1));
 637                         lmm_size -= LOV_MAXPOOLNAME;
 638                 }
 639         } else {
 640                 /* if v3 we just have to update the lum_size */
 641                 lum_size = sizeof(struct lov_user_md_v3);
 642         }
 643
 644         /* User wasn't expecting this many OST entries */
 645         if (lum.lmm_stripe_count == 0)
 646                 lmm_size = lum_size;
 647         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 648                 GOTO(out_set, rc = -EOVERFLOW);
 649         /*
 650          * Have a difference between lov_mds_md & lov_user_md.
 651          * So we have to re-order the data before copy to user.
 652          */
 653         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 654         lum.u.lum_layout_gen = lmmk->lmm_layout_gen;
 655         ((struct lov_user_md*)lmmk)->u.lum_layout_gen = lum.u.lum_layout_gen;
 656         ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 657         if (cfs_copy_to_user(lump, lmmk, lmm_size))
 658                 rc = -EFAULT;
 659
 660         obd_free_diskmd(exp, &lmmk);
 661 out_set:
 662         set_fs(seg);
 663         RETURN(rc);
 664 }