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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/lov/lov_pack.c
  40  *
  41  * (Un)packing of OST/MDS requests
  42  *
  43  * Author: Andreas Dilger <adilger@clusterfs.com>
  44  */
  45
  46 #define DEBUG_SUBSYSTEM S_LOV
  47 #ifndef __KERNEL__
  48 #include <liblustre.h>
  49 #endif
  50
  51 #include <lustre_net.h>
  52 #include <obd.h>
  53 #include <obd_lov.h>
  54 #include <obd_class.h>
  55 #include <obd_support.h>
  56 #include <lustre/lustre_user.h>
  57
  58 #include "lov_internal.h"
  59
  60 static void lov_dump_lmm_common(int level, void *lmmp)
  61 {
  62         struct lov_mds_md *lmm = lmmp;
  63
  64         CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
  65                (__u64)le64_to_cpu(lmm->lmm_object_id),
  66                le32_to_cpu(lmm->lmm_magic),
  67                le32_to_cpu(lmm->lmm_pattern));
  68         CDEBUG(level,"stripe_size %u, stripe_count %u\n",
  69                le32_to_cpu(lmm->lmm_stripe_size),
  70                le32_to_cpu(lmm->lmm_stripe_count));
  71 }
  72
  73 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  74                                  int stripe_count)
  75 {
  76         int i;
  77
  78         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  79                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  80                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  81         }
  82
  83         for (i = 0; i < stripe_count; ++i, ++lod) {
  84                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n", i,
  85                        le32_to_cpu(lod->l_ost_idx),
  86                        (__u64)le64_to_cpu(lod->l_object_seq),
  87                        (__u64)le64_to_cpu(lod->l_object_id));
  88         }
  89 }
  90
  91 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  92 {
  93         lov_dump_lmm_common(level, lmm);
  94         lov_dump_lmm_objects(level, lmm->lmm_objects,
  95                              le32_to_cpu(lmm->lmm_stripe_count));
  96 }
  97
  98 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
  99 {
 100         lov_dump_lmm_common(level, lmm);
 101         CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
 102         lov_dump_lmm_objects(level, lmm->lmm_objects,
 103                              le32_to_cpu(lmm->lmm_stripe_count));
 104 }
 105
 106 void lov_dump_lmm(int level, void *lmm)
 107 {
 108         int magic;
 109
 110         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 111         switch (magic) {
 112         case LOV_MAGIC_V1:
 113                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 114         case LOV_MAGIC_V3:
 115                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 116         default:
 117                 CERROR("Cannot recognize lmm_magic %x", magic);
 118         }
 119         return;
 120 }
 121
 122 #define LMM_ASSERT(test)                                                \
 123 do {                                                                    \
 124         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 125         LASSERT(test); /* so we know what assertion failed */           \
 126 } while(0)
 127
 128 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 129  * order and is opaque to the networking layer.
 130  *
 131  * XXX In the future, this will be enhanced to get the EA size from the
 132  *     underlying OSC device(s) to get their EA sizes so we can stack
 133  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 134  *     per stripe.
 135  */
 136 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 137                struct lov_stripe_md *lsm)
 138 {
 139         struct obd_device *obd = class_exp2obd(exp);
 140         struct lov_obd *lov = &obd->u.lov;
 141         struct lov_mds_md_v1 *lmmv1;
 142         struct lov_mds_md_v3 *lmmv3;
 143         int stripe_count;
 144         struct lov_ost_data_v1 *lmm_objects;
 145         int lmm_size, lmm_magic;
 146         int i;
 147         ENTRY;
 148
 149         if (lsm) {
 150                 lmm_magic = lsm->lsm_magic;
 151
 152                 /* If we are just sizing the EA, limit the stripe count
 153                  * to the actual number of OSTs in this filesystem. */
 154                 if (!lmmp) {
 155                         stripe_count = lov_get_stripecnt(lov,
 156                                                          lsm->lsm_stripe_count);
 157                         lsm->lsm_stripe_count = stripe_count;
 158                 } else {
 159                         stripe_count = lsm->lsm_stripe_count;
 160                 }
 161         } else {
 162                 /* No needs to allocated more than LOV_MAX_STRIPE_COUNT.
 163                  * Anyway, this is pretty inaccurate since ld_tgt_count now
 164                  * represents max index and we should rely on the actual number
 165                  * of OSTs instead */
 166                 stripe_count = min((__u32)LOV_MAX_STRIPE_COUNT,
 167                                    lov->desc.ld_tgt_count);
 168
 169                 if (lmmp && *lmmp)
 170                         lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 171                 else
 172                         /* lsm == NULL and lmmp == NULL */
 173                         lmm_magic = LOV_MAGIC;
 174         }
 175
 176         if ((lmm_magic != LOV_MAGIC_V1) &&
 177             (lmm_magic != LOV_MAGIC_V3)) {
 178                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 179                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 180                 RETURN(-EINVAL);
 181
 182         }
 183
 184         /* XXX LOV STACKING call into osc for sizes */
 185         lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 186
 187         if (!lmmp)
 188                 RETURN(lmm_size);
 189
 190         if (*lmmp && !lsm) {
 191                 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
 192                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 193                 OBD_FREE_LARGE(*lmmp, lmm_size);
 194                 *lmmp = NULL;
 195                 RETURN(0);
 196         }
 197
 198         if (!*lmmp) {
 199                 OBD_ALLOC_LARGE(*lmmp, lmm_size);
 200                 if (!*lmmp)
 201                         RETURN(-ENOMEM);
 202         }
 203
 204         CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
 205                lmm_magic, lmm_size);
 206
 207         lmmv1 = *lmmp;
 208         lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
 209         if (lmm_magic == LOV_MAGIC_V3)
 210                 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 211         else
 212                 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
 213
 214         if (!lsm)
 215                 RETURN(lmm_size);
 216
 217         /* lmmv1 and lmmv3 point to the same struct and have the
 218          * same first fields
 219          */
 220         lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
 221         lmmv1->lmm_object_seq = cpu_to_le64(lsm->lsm_object_seq);
 222         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 223         lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
 224         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 225         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 226                 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 227                         LOV_MAXPOOLNAME);
 228                 lmm_objects = lmmv3->lmm_objects;
 229         } else {
 230                 lmm_objects = lmmv1->lmm_objects;
 231         }
 232
 233         for (i = 0; i < stripe_count; i++) {
 234                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 235                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 236                 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
 237                          lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
 238                 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
 239                 lmm_objects[i].l_object_seq = cpu_to_le64(loi->loi_seq);
 240                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 241                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 242         }
 243
 244         RETURN(lmm_size);
 245 }
 246
 247 /* Find the max stripecount we should use */
 248 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
 249 {
 250         if (!stripe_count)
 251                 stripe_count = lov->desc.ld_default_stripe_count;
 252         if (stripe_count > lov->desc.ld_active_tgt_count)
 253                 stripe_count = lov->desc.ld_active_tgt_count;
 254         if (!stripe_count)
 255                 stripe_count = 1;
 256         /* for now, we limit the stripe count directly, when bug 4424 is
 257          * fixed this needs to be somewhat dynamic based on whether ext3
 258          * can handle larger EA sizes. */
 259         if (stripe_count > LOV_MAX_STRIPE_COUNT)
 260                 stripe_count = LOV_MAX_STRIPE_COUNT;
 261
 262         return stripe_count;
 263 }
 264
 265
 266 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
 267 {
 268         int rc;
 269
 270         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 271                 char *buffer;
 272                 int sz;
 273
 274                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 275                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 276                 sz = lmm_bytes * 2 + 1;
 277                 OBD_ALLOC_LARGE(buffer, sz);
 278                 if (buffer != NULL) {
 279                         int i;
 280
 281                         for (i = 0; i < lmm_bytes; i++)
 282                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 283                         buffer[sz] = '\0';
 284                         CERROR("%s\n", buffer);
 285                         OBD_FREE_LARGE(buffer, sz);
 286                 }
 287                 return -EINVAL;
 288         }
 289         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 290                                      lmm_bytes, stripe_count);
 291         return rc;
 292 }
 293
 294 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
 295                     int pattern, int magic)
 296 {
 297         int i, lsm_size;
 298         ENTRY;
 299
 300         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 301
 302         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 303         if (!*lsmp) {
 304                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 305                 RETURN(-ENOMEM);
 306         }
 307
 308         cfs_spin_lock_init(&(*lsmp)->lsm_lock);
 309         (*lsmp)->lsm_magic = magic;
 310         (*lsmp)->lsm_stripe_count = stripe_count;
 311         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 312         (*lsmp)->lsm_pattern = pattern;
 313         (*lsmp)->lsm_pool_name[0] = '\0';
 314         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 315
 316         for (i = 0; i < stripe_count; i++)
 317                 loi_init((*lsmp)->lsm_oinfo[i]);
 318
 319         RETURN(lsm_size);
 320 }
 321
 322 void lov_free_memmd(struct lov_stripe_md **lsmp)
 323 {
 324         struct lov_stripe_md *lsm = *lsmp;
 325
 326         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 327         lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 328
 329         *lsmp = NULL;
 330 }
 331
 332
 333 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 334  * order and is opaque to the networking layer.
 335  */
 336 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 337                  struct lov_mds_md *lmm, int lmm_bytes)
 338 {
 339         struct obd_device *obd = class_exp2obd(exp);
 340         struct lov_obd *lov = &obd->u.lov;
 341         int rc = 0, stripe_count, lsm_size;
 342         __u32 magic;
 343         ENTRY;
 344
 345         /* If passed an MDS struct use values from there, otherwise defaults */
 346         if (lmm) {
 347                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 348                 if (rc)
 349                         RETURN(rc);
 350                 magic = le32_to_cpu(lmm->lmm_magic);
 351         } else {
 352                 stripe_count = lov_get_stripecnt(lov, 0);
 353                 magic = LOV_MAGIC;
 354         }
 355
 356         /* If we aren't passed an lsmp struct, we just want the size */
 357         if (!lsmp) {
 358                 /* XXX LOV STACKING call into osc for sizes */
 359                 LBUG();
 360                 RETURN(lov_stripe_md_size(stripe_count));
 361         }
 362         /* If we are passed an allocated struct but nothing to unpack, free */
 363         if (*lsmp && !lmm) {
 364                 lov_free_memmd(lsmp);
 365                 RETURN(0);
 366         }
 367
 368         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 369                                    magic);
 370         if (lsm_size < 0)
 371                 RETURN(lsm_size);
 372
 373         /* If we are passed a pointer but nothing to unpack, we only alloc */
 374         if (!lmm)
 375                 RETURN(lsm_size);
 376
 377         LASSERT(lsm_op_find(magic) != NULL);
 378         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 379         if (rc) {
 380                 lov_free_memmd(lsmp);
 381                 RETURN(rc);
 382         }
 383
 384         RETURN(lsm_size);
 385 }
 386
 387 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
 388                            struct lov_stripe_md **lsmp,
 389                            struct lov_user_md *lump)
 390 {
 391         struct obd_device *obd = class_exp2obd(exp);
 392         struct lov_obd *lov = &obd->u.lov;
 393         char buffer[sizeof(struct lov_user_md_v3)];
 394         struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0];
 395         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0];
 396         int lmm_magic;
 397         int stripe_count;
 398         int rc;
 399         ENTRY;
 400
 401         if (cfs_copy_from_user(lumv3, lump, sizeof(struct lov_user_md_v1)))
 402                 RETURN(-EFAULT);
 403
 404         lmm_magic = lumv1->lmm_magic;
 405
 406         if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
 407                 lustre_swab_lov_user_md_v1(lumv1);
 408                 lmm_magic = LOV_USER_MAGIC_V1;
 409         } else if (lmm_magic == LOV_USER_MAGIC_V3) {
 410                 if (cfs_copy_from_user(lumv3, lump, sizeof(*lumv3)))
 411                         RETURN(-EFAULT);
 412         } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
 413                 if (cfs_copy_from_user(lumv3, lump, sizeof(*lumv3)))
 414                         RETURN(-EFAULT);
 415                 lustre_swab_lov_user_md_v3(lumv3);
 416                 lmm_magic = LOV_USER_MAGIC_V3;
 417         } else if (lmm_magic != LOV_USER_MAGIC_V1) {
 418                 CDEBUG(D_IOCTL,
 419                        "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 420                        lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
 421                        RETURN(-EINVAL);
 422         }
 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, lumv1->lmm_stripe_count);
 454
 455         if (max_lmm_size) {
 456                 int max_stripes = (max_lmm_size -
 457                                    lov_mds_md_size(0, lmm_magic)) /
 458                                    sizeof(struct lov_ost_data_v1);
 459                 if (unlikely(max_stripes < stripe_count)) {
 460                         CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
 461                                stripe_count, max_stripes);
 462                         stripe_count = max_stripes;
 463                 }
 464         }
 465
 466         if (lmm_magic == LOV_USER_MAGIC_V3) {
 467                 struct pool_desc *pool;
 468
 469                 pool = lov_find_pool(lov, lumv3->lmm_pool_name);
 470                 if (pool != NULL) {
 471                         if (lumv3->lmm_stripe_offset !=
 472                             (typeof(lumv3->lmm_stripe_offset))(-1)) {
 473                                 rc = lov_check_index_in_pool(
 474                                         lumv3->lmm_stripe_offset, pool);
 475                                 if (rc < 0) {
 476                                         lov_pool_putref(pool);
 477                                         RETURN(-EINVAL);
 478                                 }
 479                         }
 480
 481                         if (stripe_count > pool_tgt_count(pool))
 482                                 stripe_count = pool_tgt_count(pool);
 483
 484                         lov_pool_putref(pool);
 485                 }
 486         }
 487
 488         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 489
 490         if (rc >= 0) {
 491                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 492                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 493                 if (lmm_magic == LOV_USER_MAGIC_V3)
 494                         strncpy((*lsmp)->lsm_pool_name, lumv3->lmm_pool_name,
 495                                 LOV_MAXPOOLNAME);
 496                 rc = 0;
 497         }
 498
 499         RETURN(rc);
 500 }
 501
 502 /* Configure object striping information on a new file.
 503  *
 504  * @lmmu is a pointer to a user struct with one or more of the fields set to
 505  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 506  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 507  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 508  */
 509 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
 510                   struct lov_stripe_md **lsmp, struct lov_user_md *lump)
 511 {
 512         int rc;
 513         mm_segment_t seg;
 514
 515         seg = get_fs();
 516         set_fs(KERNEL_DS);
 517
 518         rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
 519         set_fs(seg);
 520         RETURN(rc);
 521 }
 522
 523 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 524               struct lov_user_md *lump)
 525 {
 526         int i;
 527         int rc;
 528         struct obd_export *oexp;
 529         struct lov_obd *lov = &exp->exp_obd->u.lov;
 530         obd_id last_id = 0;
 531         struct lov_user_ost_data_v1 *lmm_objects;
 532
 533         ENTRY;
 534
 535         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 536                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 537         else
 538                 lmm_objects = lump->lmm_objects;
 539
 540         for (i = 0; i < lump->lmm_stripe_count; i++) {
 541                 __u32 len = sizeof(last_id);
 542                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 543                 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 544                                   &len, &last_id, NULL);
 545                 if (rc)
 546                         RETURN(rc);
 547                 if (lmm_objects[i].l_object_id > last_id) {
 548                         CERROR("Setting EA for object > than last id on "
 549                                "ost idx %d "LPD64" > "LPD64" \n",
 550                                lmm_objects[i].l_ost_idx,
 551                                lmm_objects[i].l_object_id, last_id);
 552                         RETURN(-EINVAL);
 553                 }
 554         }
 555
 556         rc = lov_setstripe(exp, 0, lsmp, lump);
 557         if (rc)
 558                 RETURN(rc);
 559
 560         for (i = 0; i < lump->lmm_stripe_count; i++) {
 561                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 562                         lmm_objects[i].l_ost_idx;
 563                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 564                 (*lsmp)->lsm_oinfo[i]->loi_seq = lmm_objects[i].l_object_seq;
 565         }
 566         RETURN(0);
 567 }
 568
 569
 570 /* Retrieve object striping information.
 571  *
 572  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 573  * the maximum number of OST indices which will fit in the user buffer.
 574  * lmm_magic must be LOV_USER_MAGIC.
 575  */
 576 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
 577                   struct lov_user_md *lump)
 578 {
 579         /*
 580          * XXX huge struct allocated on stack.
 581          */
 582         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 583         struct lov_user_md_v3 lum;
 584         struct lov_mds_md *lmmk = NULL;
 585         int rc, lmm_size;
 586         int lum_size;
 587         mm_segment_t seg;
 588         ENTRY;
 589
 590         if (!lsm)
 591                 RETURN(-ENODATA);
 592
 593         /*
 594          * "Switch to kernel segment" to allow copying from kernel space by
 595          * copy_{to,from}_user().
 596          */
 597         seg = get_fs();
 598         set_fs(KERNEL_DS);
 599
 600         /* we only need the header part from user space to get lmm_magic and
 601          * lmm_stripe_count, (the header part is common to v1 and v3) */
 602         lum_size = sizeof(struct lov_user_md_v1);
 603         if (cfs_copy_from_user(&lum, lump, lum_size))
 604                 GOTO(out_set, rc = -EFAULT);
 605         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 606                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 607                 GOTO(out_set, rc = -EINVAL);
 608
 609         if (lum.lmm_stripe_count &&
 610             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 611                 /* Return right size of stripe to user */
 612                 lum.lmm_stripe_count = lsm->lsm_stripe_count;
 613                 rc = cfs_copy_to_user(lump, &lum, lum_size);
 614                 GOTO(out_set, rc = -EOVERFLOW);
 615         }
 616         rc = lov_packmd(exp, &lmmk, lsm);
 617         if (rc < 0)
 618                 GOTO(out_set, rc);
 619         lmm_size = rc;
 620         rc = 0;
 621
 622         /* FIXME: Bug 1185 - copy fields properly when structs change */
 623         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 624         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 625         CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
 626
 627         if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
 628             ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
 629             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
 630                 lustre_swab_lov_mds_md(lmmk);
 631                 lustre_swab_lov_user_md_objects(
 632                                 (struct lov_user_ost_data*)lmmk->lmm_objects,
 633                                 lmmk->lmm_stripe_count);
 634         }
 635         if (lum.lmm_magic == LOV_USER_MAGIC) {
 636                 /* User request for v1, we need skip lmm_pool_name */
 637                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 638                         memmove((char*)(&lmmk->lmm_stripe_count) +
 639                                 sizeof(lmmk->lmm_stripe_count),
 640                                 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
 641                                 lmmk->lmm_stripe_count *
 642                                 sizeof(struct lov_ost_data_v1));
 643                         lmm_size -= LOV_MAXPOOLNAME;
 644                 }
 645         } else {
 646                 /* if v3 we just have to update the lum_size */
 647                 lum_size = sizeof(struct lov_user_md_v3);
 648         }
 649
 650         /* User wasn't expecting this many OST entries */
 651         if (lum.lmm_stripe_count == 0)
 652                 lmm_size = lum_size;
 653         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 654                 GOTO(out_set, rc = -EOVERFLOW);
 655         /*
 656          * Have a difference between lov_mds_md & lov_user_md.
 657          * So we have to re-order the data before copy to user.
 658          */
 659         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 660         ((struct lov_user_md*)lmmk)->lmm_stripe_offset = 0;
 661         ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 662         if (cfs_copy_to_user(lump, lmmk, lmm_size))
 663                 rc = -EFAULT;
 664
 665         obd_free_diskmd(exp, &lmmk);
 666 out_set:
 667         set_fs(seg);
 668         RETURN(rc);
 669 }