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 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  58 {
  59         struct lov_ost_data_v1 *lod;
  60         int i;
  61
  62         CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
  63                le64_to_cpu(lmm->lmm_object_id), 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         if (le32_to_cpu(lmm->lmm_stripe_count) <= LOV_V1_INSANE_STRIPE_COUNT) {
  70                 for (i = 0, lod = lmm->lmm_objects;
  71                      i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
  72                          CDEBUG(level,
  73                                 "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
  74                                 i, le32_to_cpu(lod->l_ost_idx),
  75                                 le64_to_cpu(lod->l_object_gr),
  76                                 le64_to_cpu(lod->l_object_id));
  77         } else {
  78                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  79                        le32_to_cpu(lmm->lmm_stripe_count),
  80                        LOV_V1_INSANE_STRIPE_COUNT);
  81         }
  82 }
  83
  84 void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
  85 {
  86
  87         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
  88                le64_to_cpu(lmmj->lmmj_md.lmm_object_id),
  89                le32_to_cpu(lmmj->lmmj_md.lmm_magic),
  90                le32_to_cpu(lmmj->lmmj_md.lmm_pattern));
  91         CDEBUG(level,"stripe_size %u, stripe_count %u extent_count %u \n",
  92                le32_to_cpu(lmmj->lmmj_md.lmm_stripe_size),
  93                le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count),
  94                le32_to_cpu(lmmj->lmmj_extent_count));
  95 }
  96
  97 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
  98 {
  99         struct lov_ost_data_v1 *lod;
 100         int i;
 101
 102         CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
 103                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
 104                le32_to_cpu(lmm->lmm_pattern));
 105         CDEBUG(level,"stripe_size %u, stripe_count %u\n",
 106                le32_to_cpu(lmm->lmm_stripe_size),
 107                le32_to_cpu(lmm->lmm_stripe_count));
 108         CDEBUG(level,"pool_name "POOLNAMEF"\n", lmm->lmm_pool_name);
 109
 110         if (le32_to_cpu(lmm->lmm_stripe_count) <= LOV_V1_INSANE_STRIPE_COUNT) {
 111                 for (i = 0, lod = lmm->lmm_objects;
 112                      i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++)
 113                          CDEBUG(level,
 114                                 "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
 115                                 i, le32_to_cpu(lod->l_ost_idx),
 116                                 le64_to_cpu(lod->l_object_gr),
 117                                 le64_to_cpu(lod->l_object_id));
 118         } else {
 119                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
 120                        le32_to_cpu(lmm->lmm_stripe_count),
 121                        LOV_V1_INSANE_STRIPE_COUNT);
 122         }
 123 }
 124
 125 void lov_dump_lmm(int level, void *lmm)
 126 {
 127         int magic;
 128
 129         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 130         switch (magic) {
 131         case LOV_MAGIC_V1:
 132                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 133         case LOV_MAGIC_JOIN:
 134                 return lov_dump_lmm_join(level, (struct lov_mds_md_join *)(lmm));
 135         case LOV_MAGIC_V3:
 136                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 137         default:
 138                 CERROR("Cannot recognize lmm_magic %x", magic);
 139         }
 140         return;
 141 }
 142
 143 #define LMM_ASSERT(test)                                                \
 144 do {                                                                    \
 145         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 146         LASSERT(test); /* so we know what assertion failed */           \
 147 } while(0)
 148
 149 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 150  * order and is opaque to the networking layer.
 151  *
 152  * XXX In the future, this will be enhanced to get the EA size from the
 153  *     underlying OSC device(s) to get their EA sizes so we can stack
 154  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 155  *     per stripe.
 156  */
 157 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 158                struct lov_stripe_md *lsm)
 159 {
 160         struct obd_device *obd = class_exp2obd(exp);
 161         struct lov_obd *lov = &obd->u.lov;
 162         struct lov_mds_md_v1 *lmmv1;
 163         struct lov_mds_md_v3 *lmmv3;
 164         int stripe_count = lov->desc.ld_tgt_count;
 165         struct lov_ost_data_v1 *lmm_objects;
 166         int lmm_size, lmm_magic;
 167         int i;
 168         ENTRY;
 169
 170         if (lsm) {
 171                 lmm_magic = lsm->lsm_magic;
 172
 173                 /* If we are just sizing the EA, limit the stripe count
 174                  * to the actual number of OSTs in this filesystem. */
 175                 if (!lmmp) {
 176                         stripe_count = lov_get_stripecnt(lov,
 177                                                          lsm->lsm_stripe_count);
 178                         lsm->lsm_stripe_count = stripe_count;
 179                 } else {
 180                         stripe_count = lsm->lsm_stripe_count;
 181                 }
 182         } else if (lmmp && *lmmp) {
 183                 lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 184         } else {
 185                 /* lsm == NULL and lmmp == NULL */
 186                 lmm_magic = LOV_MAGIC;
 187         }
 188
 189         if ((lmm_magic != LOV_MAGIC_V1) &&
 190             (lmm_magic != LOV_MAGIC_V3)) {
 191                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 192                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 193                 RETURN(-EINVAL);
 194
 195         }
 196
 197         /* XXX LOV STACKING call into osc for sizes */
 198         lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 199
 200         if (!lmmp)
 201                 RETURN(lmm_size);
 202
 203         if (*lmmp && !lsm) {
 204                 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
 205                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 206                 OBD_FREE(*lmmp, lmm_size);
 207                 *lmmp = NULL;
 208                 RETURN(0);
 209         }
 210
 211         if (!*lmmp) {
 212                 OBD_ALLOC(*lmmp, lmm_size);
 213                 if (!*lmmp)
 214                         RETURN(-ENOMEM);
 215         }
 216
 217         CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
 218                lmm_magic, lmm_size);
 219
 220         lmmv1 = *lmmp;
 221         lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
 222         if (lmm_magic == LOV_MAGIC_V3)
 223                 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 224         else
 225                 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
 226
 227         if (!lsm)
 228                 RETURN(lmm_size);
 229
 230         /* lmmv1 and lmmv3 point to the same struct and have the
 231          * same first fields
 232          */
 233         lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
 234         lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
 235         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 236         lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
 237         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 238         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 239                 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name, MAXPOOLNAME);
 240                 lmm_objects = lmmv3->lmm_objects;
 241         } else {
 242                 lmm_objects = lmmv1->lmm_objects;
 243         }
 244
 245         for (i = 0; i < stripe_count; i++) {
 246                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 247
 248                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 249                 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
 250                          lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
 251                 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
 252                 lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
 253                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 254                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 255         }
 256
 257         RETURN(lmm_size);
 258 }
 259
 260 /* Find the max stripecount we should use */
 261 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
 262 {
 263         if (!stripe_count)
 264                 stripe_count = lov->desc.ld_default_stripe_count;
 265         if (stripe_count > lov->desc.ld_active_tgt_count)
 266                 stripe_count = lov->desc.ld_active_tgt_count;
 267         if (!stripe_count)
 268                 stripe_count = 1;
 269         /* for now, we limit the stripe count directly, when bug 4424 is
 270          * fixed this needs to be somewhat dynamic based on whether ext3
 271          * can handle larger EA sizes. */
 272         if (stripe_count > LOV_MAX_STRIPE_COUNT)
 273                 stripe_count = LOV_MAX_STRIPE_COUNT;
 274
 275         return stripe_count;
 276 }
 277
 278
 279 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
 280 {
 281         int rc;
 282
 283         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 284                 char *buffer;
 285                 int sz;
 286
 287                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 288                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 289                 sz = lmm_bytes * 2 + 1;
 290                 OBD_ALLOC(buffer, sz);
 291                 if (buffer != NULL) {
 292                         int i;
 293
 294                         for (i = 0; i < lmm_bytes; i++)
 295                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 296                         buffer[sz] = '\0';
 297                         CERROR("%s\n", buffer);
 298                         OBD_FREE(buffer, sz);
 299                 }
 300                 return -EINVAL;
 301         }
 302         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 303                                      lmm_bytes, stripe_count);
 304         return rc;
 305 }
 306
 307 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
 308                       int pattern, int magic)
 309 {
 310         int i, lsm_size;
 311         ENTRY;
 312
 313         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 314
 315         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 316         if (!*lsmp) {
 317                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 318                 RETURN(-ENOMEM);
 319         }
 320
 321         spin_lock_init(&(*lsmp)->lsm_lock);
 322         (*lsmp)->lsm_magic = magic;
 323         (*lsmp)->lsm_stripe_count = stripe_count;
 324         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 325         (*lsmp)->lsm_pattern = pattern;
 326         (*lsmp)->lsm_pool_name[0] = '\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 void lov_free_memmd(struct lov_stripe_md **lsmp)
 336 {
 337         struct lov_stripe_md *lsm = *lsmp;
 338
 339         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 340         lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 341
 342         *lsmp = NULL;
 343 }
 344
 345
 346 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 347  * order and is opaque to the networking layer.
 348  */
 349 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 350                  struct lov_mds_md *lmm, int lmm_bytes)
 351 {
 352         struct obd_device *obd = class_exp2obd(exp);
 353         struct lov_obd *lov = &obd->u.lov;
 354         int rc = 0, stripe_count, lsm_size;
 355         __u32 magic;
 356         ENTRY;
 357
 358         /* If passed an MDS struct use values from there, otherwise defaults */
 359         if (lmm) {
 360                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 361                 if (rc)
 362                         RETURN(rc);
 363                 magic = le32_to_cpu(lmm->lmm_magic);
 364         } else {
 365                 stripe_count = lov_get_stripecnt(lov, 0);
 366                 magic = LOV_MAGIC;
 367         }
 368
 369         /* If we aren't passed an lsmp struct, we just want the size */
 370         if (!lsmp) {
 371                 /* XXX LOV STACKING call into osc for sizes */
 372                 LBUG();
 373                 RETURN(lov_stripe_md_size(stripe_count));
 374         }
 375         /* If we are passed an allocated struct but nothing to unpack, free */
 376         if (*lsmp && !lmm) {
 377                 lov_free_memmd(lsmp);
 378                 RETURN(0);
 379         }
 380
 381         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 382                                    magic);
 383         if (lsm_size < 0)
 384                 RETURN(lsm_size);
 385
 386         /* If we are passed a pointer but nothing to unpack, we only alloc */
 387         if (!lmm)
 388                 RETURN(lsm_size);
 389
 390         LASSERT(lsm_op_find(magic) != NULL);
 391         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 392         if (rc) {
 393                 lov_free_memmd(lsmp);
 394                 RETURN(rc);
 395         }
 396
 397         RETURN(lsm_size);
 398 }
 399
 400 static int __lov_setstripe(struct obd_export *exp, 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         struct lov_user_md_v3 lumv3;
 406         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 407         int lmm_magic;
 408         int stripe_count;
 409         int rc;
 410         ENTRY;
 411
 412         rc = copy_from_user(&lumv3, lump, sizeof(struct lov_user_md_v1));
 413         if (rc)
 414                 RETURN(-EFAULT);
 415
 416         lmm_magic = lumv1->lmm_magic;
 417
 418         if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
 419                 lustre_swab_lov_user_md_v1(lumv1);
 420                 lmm_magic = LOV_USER_MAGIC_V1;
 421         } else if (lmm_magic == LOV_USER_MAGIC_V3) {
 422                 rc = copy_from_user(&lumv3, lump, sizeof(lumv3));
 423                 if (rc)
 424                         RETURN(-EFAULT);
 425         } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
 426                 rc = copy_from_user(&lumv3, lump, sizeof(lumv3));
 427                 if (rc)
 428                         RETURN(-EFAULT);
 429                 lustre_swab_lov_user_md_v3(&lumv3);
 430                 lmm_magic = LOV_USER_MAGIC_V3;
 431         } else if (lmm_magic != LOV_USER_MAGIC_V1) {
 432                 CDEBUG(D_IOCTL,
 433                        "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 434                        lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
 435                        RETURN(-EINVAL);
 436         }
 437
 438         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 439          * fields, we use lumv1 to avoid code duplication */
 440
 441         if (lumv1->lmm_pattern == 0) {
 442                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 443                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 444         }
 445
 446         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 447                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 448                        lumv1->lmm_pattern);
 449                 RETURN(-EINVAL);
 450         }
 451
 452         /* 64kB is the largest common page size we see (ia64), and matches the
 453          * check in lfs */
 454         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 455                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 456                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 457                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 458         }
 459
 460         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 461             (lumv1->lmm_stripe_offset !=
 462              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 463                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 464                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 465                 RETURN(-EINVAL);
 466         }
 467         stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count);
 468
 469         if (lmm_magic == LOV_USER_MAGIC_V3) {
 470                 struct pool_desc *pool;
 471
 472                 pool = lov_find_pool(lov, lumv3.lmm_pool_name);
 473                 if (pool == NULL)
 474                         RETURN(-EINVAL);
 475
 476                 if (lumv3.lmm_stripe_offset !=
 477                     (typeof(lumv3.lmm_stripe_offset))(-1)) {
 478                         rc = lov_check_index_in_pool(lumv3.lmm_stripe_offset,
 479                                                      pool);
 480                         if (rc < 0)
 481                                 RETURN(-EINVAL);
 482                 }
 483
 484                 if (stripe_count > pool_tgt_count(pool))
 485                         stripe_count = pool_tgt_count(pool);
 486         }
 487
 488         if ((__u64)lumv1->lmm_stripe_size * stripe_count > ~0UL) {
 489                 CDEBUG(D_IOCTL, "stripe width %ux%i exeeds %lu bytes\n",
 490                        lumv1->lmm_stripe_size, (int)lumv1->lmm_stripe_count,
 491                        ~0UL);
 492                 RETURN(-EINVAL);
 493         }
 494
 495         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 496
 497         if (rc >= 0) {
 498                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 499                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 500                 if (lmm_magic == LOV_USER_MAGIC_V3)
 501                         strncpy((*lsmp)->lsm_pool_name, lumv3.lmm_pool_name,
 502                                 MAXPOOLNAME);
 503                 rc = 0;
 504         }
 505
 506         RETURN(rc);
 507 }
 508
 509 /* Configure object striping information on a new file.
 510  *
 511  * @lmmu is a pointer to a user struct with one or more of the fields set to
 512  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 513  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 514  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 515  */
 516 int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp,
 517                   struct lov_user_md *lump)
 518 {
 519         int rc;
 520         mm_segment_t seg;
 521
 522         seg = get_fs();
 523         set_fs(KERNEL_DS);
 524
 525         rc = __lov_setstripe(exp, lsmp, lump);
 526         set_fs(seg);
 527         RETURN(rc);
 528 }
 529
 530 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 531               struct lov_user_md *lump)
 532 {
 533         int i;
 534         int rc;
 535         struct obd_export *oexp;
 536         struct lov_obd *lov = &exp->exp_obd->u.lov;
 537         obd_id last_id = 0;
 538         struct lov_user_ost_data_v1 *lmm_objects;
 539
 540         ENTRY;
 541
 542         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 543                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 544         else
 545                 lmm_objects = lump->lmm_objects;
 546
 547         for (i = 0; i < lump->lmm_stripe_count; i++) {
 548                 __u32 len = sizeof(last_id);
 549                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 550                 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 551                                   &len, &last_id, NULL);
 552                 if (rc)
 553                         RETURN(rc);
 554                 if (lmm_objects[i].l_object_id > last_id) {
 555                         CERROR("Setting EA for object > than last id on "
 556                                "ost idx %d "LPD64" > "LPD64" \n",
 557                                lmm_objects[i].l_ost_idx,
 558                                lmm_objects[i].l_object_id, last_id);
 559                         RETURN(-EINVAL);
 560                 }
 561         }
 562
 563         rc = lov_setstripe(exp, lsmp, lump);
 564         if (rc)
 565                 RETURN(rc);
 566
 567         for (i = 0; i < lump->lmm_stripe_count; i++) {
 568                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 569                         lmm_objects[i].l_ost_idx;
 570                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 571                 (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr;
 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         rc = copy_from_user(&lum, lump, lum_size);
 611
 612         if (rc)
 613                 rc = -EFAULT;
 614         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 615                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 616                 rc = -EINVAL;
 617         else {
 618                 /* if v3 we just have to update the lum_size */
 619                 if (lum.lmm_magic == LOV_USER_MAGIC_V3)
 620                         lum_size = sizeof(struct lov_user_md_v3);
 621
 622                 rc = lov_packmd(exp, &lmmk, lsm);
 623                 if (rc < 0)
 624                         RETURN(rc);
 625                 lmm_size = rc;
 626                 rc = 0;
 627
 628                 /* FIXME: Bug 1185 - copy fields properly when structs change */
 629                 /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 630                 CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 631                 CLASSERT(sizeof lum.lmm_objects[0] ==
 632                          sizeof lmmk->lmm_objects[0]);
 633
 634                 /* User wasn't expecting this many OST entries */
 635                 if (lum.lmm_stripe_count == 0) {
 636                         if (copy_to_user(lump, lmmk, lum_size))
 637                                 rc = -EFAULT;
 638                 } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
 639                         rc = -EOVERFLOW;
 640                 } else if (copy_to_user(lump, lmmk, lmm_size))
 641                         rc = -EFAULT;
 642
 643                 obd_free_diskmd(exp, &lmmk);
 644         }
 645         set_fs(seg);
 646         RETURN(rc);
 647 }