lustre/lov/lov_pack.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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                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\n",
  66                le32_to_cpu(lmm->lmm_stripe_size),
  67                le32_to_cpu(lmm->lmm_stripe_count));
  68 }
  69
  70 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  71                                  int stripe_count)
  72 {
  73         int i;
  74
  75         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  76                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  77                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  78         }
  79
  80         for (i = 0; i < stripe_count; ++i, ++lod) {
  81                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n", i,
  82                        le32_to_cpu(lod->l_ost_idx),
  83                        le64_to_cpu(lod->l_object_gr),
  84                        le64_to_cpu(lod->l_object_id));
  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                              le32_to_cpu(lmm->lmm_stripe_count));
  93 }
  94
  95 void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
  96 {
  97         lov_dump_lmm_common(level, &lmmj->lmmj_md);
  98         CDEBUG(level, "extent_count %u\n",
  99                le32_to_cpu(lmmj->lmmj_extent_count));
 100 }
 101
 102 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
 103 {
 104         lov_dump_lmm_common(level, lmm);
 105         CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
 106         lov_dump_lmm_objects(level, lmm->lmm_objects,
 107                              le32_to_cpu(lmm->lmm_stripe_count));
 108 }
 109
 110 void lov_dump_lmm(int level, void *lmm)
 111 {
 112         int magic;
 113
 114         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 115         switch (magic) {
 116         case LOV_MAGIC_V1:
 117                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 118         case LOV_MAGIC_JOIN:
 119                 return lov_dump_lmm_join(level, (struct lov_mds_md_join *)(lmm));
 120         case LOV_MAGIC_V3:
 121                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 122         default:
 123                 CERROR("Cannot recognize lmm_magic %x", magic);
 124         }
 125         return;
 126 }
 127
 128 #define LMM_ASSERT(test)                                                \
 129 do {                                                                    \
 130         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 131         LASSERT(test); /* so we know what assertion failed */           \
 132 } while(0)
 133
 134 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 135  * order and is opaque to the networking layer.
 136  *
 137  * XXX In the future, this will be enhanced to get the EA size from the
 138  *     underlying OSC device(s) to get their EA sizes so we can stack
 139  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 140  *     per stripe.
 141  */
 142 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 143                struct lov_stripe_md *lsm)
 144 {
 145         struct obd_device *obd = class_exp2obd(exp);
 146         struct lov_obd *lov = &obd->u.lov;
 147         struct lov_mds_md_v1 *lmmv1;
 148         struct lov_mds_md_v3 *lmmv3;
 149         int stripe_count;
 150         struct lov_ost_data_v1 *lmm_objects;
 151         int lmm_size, lmm_magic;
 152         int i;
 153         ENTRY;
 154
 155         if (lsm) {
 156                 lmm_magic = lsm->lsm_magic;
 157
 158                 /* If we are just sizing the EA, limit the stripe count
 159                  * to the actual number of OSTs in this filesystem. */
 160                 if (!lmmp) {
 161                         stripe_count = lov_get_stripecnt(lov,
 162                                                          lsm->lsm_stripe_count);
 163                         lsm->lsm_stripe_count = stripe_count;
 164                 } else {
 165                         stripe_count = lsm->lsm_stripe_count;
 166                 }
 167         } else {
 168                 /* No needs to allocated more than LOV_MAX_STRIPE_COUNT.
 169                  * Anyway, this is pretty inaccurate since ld_tgt_count now
 170                  * represents max index and we should rely on the actual number
 171                  * of OSTs instead */
 172                 stripe_count = min((__u32)LOV_MAX_STRIPE_COUNT,
 173                                    lov->desc.ld_tgt_count);
 174
 175                 if (lmmp && *lmmp)
 176                         lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 177                 else
 178                         /* lsm == NULL and lmmp == NULL */
 179                         lmm_magic = LOV_MAGIC;
 180         }
 181
 182         if ((lmm_magic != LOV_MAGIC_V1) &&
 183             (lmm_magic != LOV_MAGIC_V3)) {
 184                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 185                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 186                 RETURN(-EINVAL);
 187
 188         }
 189
 190         /* XXX LOV STACKING call into osc for sizes */
 191         lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 192
 193         if (!lmmp)
 194                 RETURN(lmm_size);
 195
 196         if (*lmmp && !lsm) {
 197                 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
 198                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 199                 OBD_FREE(*lmmp, lmm_size);
 200                 *lmmp = NULL;
 201                 RETURN(0);
 202         }
 203
 204         if (!*lmmp) {
 205                 OBD_ALLOC(*lmmp, lmm_size);
 206                 if (!*lmmp)
 207                         RETURN(-ENOMEM);
 208         }
 209
 210         CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
 211                lmm_magic, lmm_size);
 212
 213         lmmv1 = *lmmp;
 214         lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
 215         if (lmm_magic == LOV_MAGIC_V3)
 216                 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 217         else
 218                 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
 219
 220         if (!lsm)
 221                 RETURN(lmm_size);
 222
 223         /* lmmv1 and lmmv3 point to the same struct and have the
 224          * same first fields
 225          */
 226         lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
 227         lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
 228         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 229         lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
 230         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 231         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 232                 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 233                         LOV_MAXPOOLNAME);
 234                 lmm_objects = lmmv3->lmm_objects;
 235         } else {
 236                 lmm_objects = lmmv1->lmm_objects;
 237         }
 238
 239         for (i = 0; i < stripe_count; i++) {
 240                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 241
 242                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 243                 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
 244                          lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
 245                 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
 246                 lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
 247                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 248                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 249         }
 250
 251         RETURN(lmm_size);
 252 }
 253
 254 /* Find the max stripecount we should use */
 255 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
 256 {
 257         if (!stripe_count)
 258                 stripe_count = lov->desc.ld_default_stripe_count;
 259         if (stripe_count > lov->desc.ld_active_tgt_count)
 260                 stripe_count = lov->desc.ld_active_tgt_count;
 261         if (!stripe_count)
 262                 stripe_count = 1;
 263         /* for now, we limit the stripe count directly, when bug 4424 is
 264          * fixed this needs to be somewhat dynamic based on whether ext3
 265          * can handle larger EA sizes. */
 266         if (stripe_count > LOV_MAX_STRIPE_COUNT)
 267                 stripe_count = LOV_MAX_STRIPE_COUNT;
 268
 269         return stripe_count;
 270 }
 271
 272
 273 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
 274 {
 275         int rc;
 276
 277         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 278                 char *buffer;
 279                 int sz;
 280
 281                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 282                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 283                 sz = lmm_bytes * 2 + 1;
 284                 OBD_ALLOC(buffer, sz);
 285                 if (buffer != NULL) {
 286                         int i;
 287
 288                         for (i = 0; i < lmm_bytes; i++)
 289                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 290                         buffer[sz] = '\0';
 291                         CERROR("%s\n", buffer);
 292                         OBD_FREE(buffer, sz);
 293                 }
 294                 return -EINVAL;
 295         }
 296         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 297                                      lmm_bytes, stripe_count);
 298         return rc;
 299 }
 300
 301 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
 302                       int pattern, int magic)
 303 {
 304         int i, lsm_size;
 305         ENTRY;
 306
 307         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 308
 309         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 310         if (!*lsmp) {
 311                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 312                 RETURN(-ENOMEM);
 313         }
 314
 315         spin_lock_init(&(*lsmp)->lsm_lock);
 316         (*lsmp)->lsm_magic = magic;
 317         (*lsmp)->lsm_stripe_count = stripe_count;
 318         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 319         (*lsmp)->lsm_pattern = pattern;
 320         (*lsmp)->lsm_pool_name[0] = '\0';
 321         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 322
 323         for (i = 0; i < stripe_count; i++)
 324                 loi_init((*lsmp)->lsm_oinfo[i]);
 325
 326         RETURN(lsm_size);
 327 }
 328
 329 void lov_free_memmd(struct lov_stripe_md **lsmp)
 330 {
 331         struct lov_stripe_md *lsm = *lsmp;
 332
 333         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 334         lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 335
 336         *lsmp = NULL;
 337 }
 338
 339
 340 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 341  * order and is opaque to the networking layer.
 342  */
 343 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 344                  struct lov_mds_md *lmm, int lmm_bytes)
 345 {
 346         struct obd_device *obd = class_exp2obd(exp);
 347         struct lov_obd *lov = &obd->u.lov;
 348         int rc = 0, stripe_count, lsm_size;
 349         __u32 magic;
 350         ENTRY;
 351
 352         /* If passed an MDS struct use values from there, otherwise defaults */
 353         if (lmm) {
 354                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 355                 if (rc)
 356                         RETURN(rc);
 357                 magic = le32_to_cpu(lmm->lmm_magic);
 358         } else {
 359                 stripe_count = lov_get_stripecnt(lov, 0);
 360                 magic = LOV_MAGIC;
 361         }
 362
 363         /* If we aren't passed an lsmp struct, we just want the size */
 364         if (!lsmp) {
 365                 /* XXX LOV STACKING call into osc for sizes */
 366                 LBUG();
 367                 RETURN(lov_stripe_md_size(stripe_count));
 368         }
 369         /* If we are passed an allocated struct but nothing to unpack, free */
 370         if (*lsmp && !lmm) {
 371                 lov_free_memmd(lsmp);
 372                 RETURN(0);
 373         }
 374
 375         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 376                                    magic);
 377         if (lsm_size < 0)
 378                 RETURN(lsm_size);
 379
 380         /* If we are passed a pointer but nothing to unpack, we only alloc */
 381         if (!lmm)
 382                 RETURN(lsm_size);
 383
 384         LASSERT(lsm_op_find(magic) != NULL);
 385         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 386         if (rc) {
 387                 lov_free_memmd(lsmp);
 388                 RETURN(rc);
 389         }
 390
 391         RETURN(lsm_size);
 392 }
 393
 394 static int __lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
 395                            struct lov_user_md *lump)
 396 {
 397         struct obd_device *obd = class_exp2obd(exp);
 398         struct lov_obd *lov = &obd->u.lov;
 399         struct lov_user_md_v3 lumv3;
 400         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 401         int lmm_magic;
 402         int stripe_count;
 403         int rc;
 404         ENTRY;
 405
 406         if (copy_from_user(&lumv3, lump, sizeof(struct lov_user_md_v1)))
 407                 RETURN(-EFAULT);
 408
 409         lmm_magic = lumv1->lmm_magic;
 410
 411         if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
 412                 lustre_swab_lov_user_md_v1(lumv1);
 413                 lmm_magic = LOV_USER_MAGIC_V1;
 414         } else if (lmm_magic == LOV_USER_MAGIC_V3) {
 415                 if (copy_from_user(&lumv3, lump, sizeof(lumv3)))
 416                         RETURN(-EFAULT);
 417         } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
 418                 if (copy_from_user(&lumv3, lump, sizeof(lumv3)))
 419                         RETURN(-EFAULT);
 420                 lustre_swab_lov_user_md_v3(&lumv3);
 421                 lmm_magic = LOV_USER_MAGIC_V3;
 422         } else if (lmm_magic != LOV_USER_MAGIC_V1) {
 423                 CDEBUG(D_IOCTL,
 424                        "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 425                        lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
 426                        RETURN(-EINVAL);
 427         }
 428
 429         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 430          * fields, we use lumv1 to avoid code duplication */
 431
 432         if (lumv1->lmm_pattern == 0) {
 433                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 434                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 435         }
 436
 437         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 438                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 439                        lumv1->lmm_pattern);
 440                 RETURN(-EINVAL);
 441         }
 442
 443         /* 64kB is the largest common page size we see (ia64), and matches the
 444          * check in lfs */
 445         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 446                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 447                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 448                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 449         }
 450
 451         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 452             (lumv1->lmm_stripe_offset !=
 453              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 454                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 455                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 456                 RETURN(-EINVAL);
 457         }
 458         stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count);
 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         if ((__u64)lumv1->lmm_stripe_size * stripe_count > ~0UL) {
 483                 CDEBUG(D_IOCTL, "stripe width %ux%i exeeds %lu bytes\n",
 484                        lumv1->lmm_stripe_size, (int)lumv1->lmm_stripe_count,
 485                        ~0UL);
 486                 RETURN(-EINVAL);
 487         }
 488
 489         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 490
 491         if (rc >= 0) {
 492                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 493                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 494                 if (lmm_magic == LOV_USER_MAGIC_V3)
 495                         strncpy((*lsmp)->lsm_pool_name, lumv3.lmm_pool_name,
 496                                 LOV_MAXPOOLNAME);
 497                 rc = 0;
 498         }
 499
 500         RETURN(rc);
 501 }
 502
 503 /* Configure object striping information on a new file.
 504  *
 505  * @lmmu is a pointer to a user struct with one or more of the fields set to
 506  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 507  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 508  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 509  */
 510 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
 511                   struct lov_user_md *lump)
 512 {
 513         int rc;
 514         mm_segment_t seg;
 515
 516         seg = get_fs();
 517         set_fs(KERNEL_DS);
 518
 519         rc = __lov_setstripe(exp, lsmp, lump);
 520         set_fs(seg);
 521         RETURN(rc);
 522 }
 523
 524 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 525               struct lov_user_md *lump)
 526 {
 527         int i;
 528         int rc;
 529         struct obd_export *oexp;
 530         struct lov_obd *lov = &exp->exp_obd->u.lov;
 531         obd_id last_id = 0;
 532         struct lov_user_ost_data_v1 *lmm_objects;
 533
 534         ENTRY;
 535
 536         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 537                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 538         else
 539                 lmm_objects = lump->lmm_objects;
 540
 541         for (i = 0; i < lump->lmm_stripe_count; i++) {
 542                 __u32 len = sizeof(last_id);
 543                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 544                 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 545                                   &len, &last_id, NULL);
 546                 if (rc)
 547                         RETURN(rc);
 548                 if (lmm_objects[i].l_object_id > last_id) {
 549                         CERROR("Setting EA for object > than last id on "
 550                                "ost idx %d "LPD64" > "LPD64" \n",
 551                                lmm_objects[i].l_ost_idx,
 552                                lmm_objects[i].l_object_id, last_id);
 553                         RETURN(-EINVAL);
 554                 }
 555         }
 556
 557         rc = lov_setstripe(exp, lsmp, lump);
 558         if (rc)
 559                 RETURN(rc);
 560
 561         for (i = 0; i < lump->lmm_stripe_count; i++) {
 562                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 563                         lmm_objects[i].l_ost_idx;
 564                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 565                 (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr;
 566         }
 567         RETURN(0);
 568 }
 569
 570
 571 /* Retrieve object striping information.
 572  *
 573  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 574  * the maximum number of OST indices which will fit in the user buffer.
 575  * lmm_magic must be LOV_USER_MAGIC.
 576  */
 577 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
 578                   struct lov_user_md *lump)
 579 {
 580         /*
 581          * XXX huge struct allocated on stack.
 582          */
 583         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 584         struct lov_user_md_v3 lum;
 585         struct lov_mds_md *lmmk = NULL;
 586         int rc, lmm_size;
 587         int lum_size;
 588         mm_segment_t seg;
 589         ENTRY;
 590
 591         if (!lsm)
 592                 RETURN(-ENODATA);
 593
 594         /*
 595          * "Switch to kernel segment" to allow copying from kernel space by
 596          * copy_{to,from}_user().
 597          */
 598         seg = get_fs();
 599         set_fs(KERNEL_DS);
 600
 601         /* we only need the header part from user space to get lmm_magic and
 602          * lmm_stripe_count, (the header part is common to v1 and v3) */
 603         lum_size = sizeof(struct lov_user_md_v1);
 604         if (copy_from_user(&lum, lump, lum_size))
 605                 GOTO(out_set, rc = -EFAULT);
 606         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 607                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 608                 GOTO(out_set, rc = -EINVAL);
 609
 610         if (lum.lmm_stripe_count &&
 611             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 612                 /* Return right size of stripe to user */
 613                 lum.lmm_stripe_count = lsm->lsm_stripe_count;
 614                 rc = copy_to_user(lump, &lum, lum_size);
 615                 GOTO(out_set, rc = -EOVERFLOW);
 616         }
 617         rc = lov_packmd(exp, &lmmk, lsm);
 618         if (rc < 0)
 619                 GOTO(out_set, rc);
 620         lmm_size = rc;
 621         rc = 0;
 622
 623         /* FIXME: Bug 1185 - copy fields properly when structs change */
 624         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 625         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 626         CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
 627
 628         if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
 629             ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
 630             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
 631                 lustre_swab_lov_mds_md(lmmk);
 632                 lustre_swab_lov_user_md_objects(
 633                                 (struct lov_user_ost_data*)lmmk->lmm_objects,
 634                                 lmmk->lmm_stripe_count);
 635         }
 636         if (lum.lmm_magic == LOV_USER_MAGIC) {
 637                 /* User request for v1, we need skip lmm_pool_name */
 638                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 639                         memmove((char*)(&lmmk->lmm_stripe_count) +
 640                                 sizeof(lmmk->lmm_stripe_count),
 641                                 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
 642                                 lmmk->lmm_stripe_count *
 643                                 sizeof(struct lov_ost_data_v1));
 644                         lmm_size -= LOV_MAXPOOLNAME;
 645                 }
 646         } else {
 647                 /* if v3 we just have to update the lum_size */
 648                 lum_size = sizeof(struct lov_user_md_v3);
 649         }
 650
 651         /* User wasn't expecting this many OST entries */
 652         if (lum.lmm_stripe_count == 0)
 653                 lmm_size = lum_size;
 654         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 655                 GOTO(out_set, rc = -EOVERFLOW);
 656         /*
 657          * Have a difference between lov_mds_md & lov_user_md.
 658          * So we have to re-order the data before copy to user.
 659          */
 660         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 661         ((struct lov_user_md*)lmmk)->lmm_stripe_offset = 0;
 662         ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 663         if (copy_to_user(lump, lmmk, lmm_size))
 664                 rc = -EFAULT;
 665
 666         obd_free_diskmd(exp, &lmmk);
 667 out_set:
 668         set_fs(seg);
 669         RETURN(rc);
 670 }