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                (__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\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                        (__u64)le64_to_cpu(lod->l_object_gr),
  84                        (__u64)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_v3(int level, struct lov_mds_md_v3 *lmm)
  96 {
  97         lov_dump_lmm_common(level, lmm);
  98         CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
  99         lov_dump_lmm_objects(level, lmm->lmm_objects,
 100                              le32_to_cpu(lmm->lmm_stripe_count));
 101 }
 102
 103 void lov_dump_lmm(int level, void *lmm)
 104 {
 105         int magic;
 106
 107         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 108         switch (magic) {
 109         case LOV_MAGIC_V1:
 110                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 111         case LOV_MAGIC_V3:
 112                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 113         default:
 114                 CERROR("Cannot recognize lmm_magic %x", magic);
 115         }
 116         return;
 117 }
 118
 119 #define LMM_ASSERT(test)                                                \
 120 do {                                                                    \
 121         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 122         LASSERT(test); /* so we know what assertion failed */           \
 123 } while(0)
 124
 125 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 126  * order and is opaque to the networking layer.
 127  *
 128  * XXX In the future, this will be enhanced to get the EA size from the
 129  *     underlying OSC device(s) to get their EA sizes so we can stack
 130  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 131  *     per stripe.
 132  */
 133 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 134                struct lov_stripe_md *lsm)
 135 {
 136         struct obd_device *obd = class_exp2obd(exp);
 137         struct lov_obd *lov = &obd->u.lov;
 138         struct lov_mds_md_v1 *lmmv1;
 139         struct lov_mds_md_v3 *lmmv3;
 140         int stripe_count;
 141         struct lov_ost_data_v1 *lmm_objects;
 142         int lmm_size, lmm_magic;
 143         int i;
 144         ENTRY;
 145
 146         if (lsm) {
 147                 lmm_magic = lsm->lsm_magic;
 148
 149                 /* If we are just sizing the EA, limit the stripe count
 150                  * to the actual number of OSTs in this filesystem. */
 151                 if (!lmmp) {
 152                         stripe_count = lov_get_stripecnt(lov,
 153                                                          lsm->lsm_stripe_count);
 154                         lsm->lsm_stripe_count = stripe_count;
 155                 } else {
 156                         stripe_count = lsm->lsm_stripe_count;
 157                 }
 158         } else {
 159                 /* No needs to allocated more than LOV_MAX_STRIPE_COUNT.
 160                  * Anyway, this is pretty inaccurate since ld_tgt_count now
 161                  * represents max index and we should rely on the actual number
 162                  * of OSTs instead */
 163                 stripe_count = min((__u32)LOV_MAX_STRIPE_COUNT,
 164                                    lov->desc.ld_tgt_count);
 165
 166                 if (lmmp && *lmmp)
 167                         lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 168                 else
 169                         /* lsm == NULL and lmmp == NULL */
 170                         lmm_magic = LOV_MAGIC;
 171         }
 172
 173         if ((lmm_magic != LOV_MAGIC_V1) &&
 174             (lmm_magic != LOV_MAGIC_V3)) {
 175                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 176                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 177                 RETURN(-EINVAL);
 178
 179         }
 180
 181         /* XXX LOV STACKING call into osc for sizes */
 182         lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 183
 184         if (!lmmp)
 185                 RETURN(lmm_size);
 186
 187         if (*lmmp && !lsm) {
 188                 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
 189                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 190                 OBD_FREE(*lmmp, lmm_size);
 191                 *lmmp = NULL;
 192                 RETURN(0);
 193         }
 194
 195         if (!*lmmp) {
 196                 OBD_ALLOC(*lmmp, lmm_size);
 197                 if (!*lmmp)
 198                         RETURN(-ENOMEM);
 199         }
 200
 201         CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
 202                lmm_magic, lmm_size);
 203
 204         lmmv1 = *lmmp;
 205         lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
 206         if (lmm_magic == LOV_MAGIC_V3)
 207                 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 208         else
 209                 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
 210
 211         if (!lsm)
 212                 RETURN(lmm_size);
 213
 214         /* lmmv1 and lmmv3 point to the same struct and have the
 215          * same first fields
 216          */
 217         lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
 218         lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
 219         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 220         lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
 221         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 222         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 223                 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 224                         LOV_MAXPOOLNAME);
 225                 lmm_objects = lmmv3->lmm_objects;
 226         } else {
 227                 lmm_objects = lmmv1->lmm_objects;
 228         }
 229
 230         for (i = 0; i < stripe_count; i++) {
 231                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 232
 233                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 234                 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
 235                          lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
 236                 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
 237                 lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
 238                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 239                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 240         }
 241
 242         RETURN(lmm_size);
 243 }
 244
 245 /* Find the max stripecount we should use */
 246 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
 247 {
 248         if (!stripe_count)
 249                 stripe_count = lov->desc.ld_default_stripe_count;
 250         if (stripe_count > lov->desc.ld_active_tgt_count)
 251                 stripe_count = lov->desc.ld_active_tgt_count;
 252         if (!stripe_count)
 253                 stripe_count = 1;
 254         /* for now, we limit the stripe count directly, when bug 4424 is
 255          * fixed this needs to be somewhat dynamic based on whether ext3
 256          * can handle larger EA sizes. */
 257         if (stripe_count > LOV_MAX_STRIPE_COUNT)
 258                 stripe_count = LOV_MAX_STRIPE_COUNT;
 259
 260         return stripe_count;
 261 }
 262
 263
 264 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
 265 {
 266         int rc;
 267
 268         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 269                 char *buffer;
 270                 int sz;
 271
 272                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 273                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 274                 sz = lmm_bytes * 2 + 1;
 275                 OBD_ALLOC(buffer, sz);
 276                 if (buffer != NULL) {
 277                         int i;
 278
 279                         for (i = 0; i < lmm_bytes; i++)
 280                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 281                         buffer[sz] = '\0';
 282                         CERROR("%s\n", buffer);
 283                         OBD_FREE(buffer, sz);
 284                 }
 285                 return -EINVAL;
 286         }
 287         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 288                                      lmm_bytes, stripe_count);
 289         return rc;
 290 }
 291
 292 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
 293                       int pattern, int magic)
 294 {
 295         int i, lsm_size;
 296         ENTRY;
 297
 298         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 299
 300         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 301         if (!*lsmp) {
 302                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 303                 RETURN(-ENOMEM);
 304         }
 305
 306         cfs_spin_lock_init(&(*lsmp)->lsm_lock);
 307         (*lsmp)->lsm_magic = magic;
 308         (*lsmp)->lsm_stripe_count = stripe_count;
 309         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 310         (*lsmp)->lsm_pattern = pattern;
 311         (*lsmp)->lsm_pool_name[0] = '\0';
 312         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 313
 314         for (i = 0; i < stripe_count; i++)
 315                 loi_init((*lsmp)->lsm_oinfo[i]);
 316
 317         RETURN(lsm_size);
 318 }
 319
 320 void lov_free_memmd(struct lov_stripe_md **lsmp)
 321 {
 322         struct lov_stripe_md *lsm = *lsmp;
 323
 324         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 325         lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 326
 327         *lsmp = NULL;
 328 }
 329
 330
 331 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 332  * order and is opaque to the networking layer.
 333  */
 334 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 335                  struct lov_mds_md *lmm, int lmm_bytes)
 336 {
 337         struct obd_device *obd = class_exp2obd(exp);
 338         struct lov_obd *lov = &obd->u.lov;
 339         int rc = 0, stripe_count, lsm_size;
 340         __u32 magic;
 341         ENTRY;
 342
 343         /* If passed an MDS struct use values from there, otherwise defaults */
 344         if (lmm) {
 345                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 346                 if (rc)
 347                         RETURN(rc);
 348                 magic = le32_to_cpu(lmm->lmm_magic);
 349         } else {
 350                 stripe_count = lov_get_stripecnt(lov, 0);
 351                 magic = LOV_MAGIC;
 352         }
 353
 354         /* If we aren't passed an lsmp struct, we just want the size */
 355         if (!lsmp) {
 356                 /* XXX LOV STACKING call into osc for sizes */
 357                 LBUG();
 358                 RETURN(lov_stripe_md_size(stripe_count));
 359         }
 360         /* If we are passed an allocated struct but nothing to unpack, free */
 361         if (*lsmp && !lmm) {
 362                 lov_free_memmd(lsmp);
 363                 RETURN(0);
 364         }
 365
 366         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 367                                    magic);
 368         if (lsm_size < 0)
 369                 RETURN(lsm_size);
 370
 371         /* If we are passed a pointer but nothing to unpack, we only alloc */
 372         if (!lmm)
 373                 RETURN(lsm_size);
 374
 375         LASSERT(lsm_op_find(magic) != NULL);
 376         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 377         if (rc) {
 378                 lov_free_memmd(lsmp);
 379                 RETURN(rc);
 380         }
 381
 382         RETURN(lsm_size);
 383 }
 384
 385 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
 386                            struct lov_stripe_md **lsmp,
 387                            struct lov_user_md *lump)
 388 {
 389         struct obd_device *obd = class_exp2obd(exp);
 390         struct lov_obd *lov = &obd->u.lov;
 391         char buffer[sizeof(struct lov_user_md_v3)];
 392         struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0];
 393         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0];
 394         int lmm_magic;
 395         int stripe_count;
 396         int rc;
 397         ENTRY;
 398
 399         if (cfs_copy_from_user(lumv3, lump, sizeof(struct lov_user_md_v1)))
 400                 RETURN(-EFAULT);
 401
 402         lmm_magic = lumv1->lmm_magic;
 403
 404         if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
 405                 lustre_swab_lov_user_md_v1(lumv1);
 406                 lmm_magic = LOV_USER_MAGIC_V1;
 407         } else if (lmm_magic == LOV_USER_MAGIC_V3) {
 408                 if (cfs_copy_from_user(lumv3, lump, sizeof(*lumv3)))
 409                         RETURN(-EFAULT);
 410         } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
 411                 if (cfs_copy_from_user(lumv3, lump, sizeof(*lumv3)))
 412                         RETURN(-EFAULT);
 413                 lustre_swab_lov_user_md_v3(lumv3);
 414                 lmm_magic = LOV_USER_MAGIC_V3;
 415         } else if (lmm_magic != LOV_USER_MAGIC_V1) {
 416                 CDEBUG(D_IOCTL,
 417                        "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 418                        lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
 419                        RETURN(-EINVAL);
 420         }
 421
 422         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 423          * fields, we use lumv1 to avoid code duplication */
 424
 425         if (lumv1->lmm_pattern == 0) {
 426                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 427                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 428         }
 429
 430         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 431                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 432                        lumv1->lmm_pattern);
 433                 RETURN(-EINVAL);
 434         }
 435
 436         /* 64kB is the largest common page size we see (ia64), and matches the
 437          * check in lfs */
 438         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 439                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 440                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 441                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 442         }
 443
 444         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 445             (lumv1->lmm_stripe_offset !=
 446              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 447                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 448                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 449                 RETURN(-EINVAL);
 450         }
 451         stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count);
 452
 453         if (max_lmm_size) {
 454                 int max_stripes = (max_lmm_size -
 455                                    lov_mds_md_size(0, lmm_magic)) /
 456                                    sizeof(struct lov_ost_data_v1);
 457                 if (unlikely(max_stripes < stripe_count)) {
 458                         CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
 459                                stripe_count, max_stripes);
 460                         stripe_count = max_stripes;
 461                 }
 462         }
 463
 464         if (lmm_magic == LOV_USER_MAGIC_V3) {
 465                 struct pool_desc *pool;
 466
 467                 pool = lov_find_pool(lov, lumv3->lmm_pool_name);
 468                 if (pool != NULL) {
 469                         if (lumv3->lmm_stripe_offset !=
 470                             (typeof(lumv3->lmm_stripe_offset))(-1)) {
 471                                 rc = lov_check_index_in_pool(
 472                                         lumv3->lmm_stripe_offset, pool);
 473                                 if (rc < 0) {
 474                                         lov_pool_putref(pool);
 475                                         RETURN(-EINVAL);
 476                                 }
 477                         }
 478
 479                         if (stripe_count > pool_tgt_count(pool))
 480                                 stripe_count = pool_tgt_count(pool);
 481
 482                         lov_pool_putref(pool);
 483                 }
 484         }
 485
 486         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 487
 488         if (rc >= 0) {
 489                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 490                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 491                 if (lmm_magic == LOV_USER_MAGIC_V3)
 492                         strncpy((*lsmp)->lsm_pool_name, lumv3->lmm_pool_name,
 493                                 LOV_MAXPOOLNAME);
 494                 rc = 0;
 495         }
 496
 497         RETURN(rc);
 498 }
 499
 500 /* Configure object striping information on a new file.
 501  *
 502  * @lmmu is a pointer to a user struct with one or more of the fields set to
 503  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 504  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 505  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 506  */
 507 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
 508                   struct lov_stripe_md **lsmp, struct lov_user_md *lump)
 509 {
 510         int rc;
 511         mm_segment_t seg;
 512
 513         seg = get_fs();
 514         set_fs(KERNEL_DS);
 515
 516         rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
 517         set_fs(seg);
 518         RETURN(rc);
 519 }
 520
 521 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 522               struct lov_user_md *lump)
 523 {
 524         int i;
 525         int rc;
 526         struct obd_export *oexp;
 527         struct lov_obd *lov = &exp->exp_obd->u.lov;
 528         obd_id last_id = 0;
 529         struct lov_user_ost_data_v1 *lmm_objects;
 530
 531         ENTRY;
 532
 533         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 534                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 535         else
 536                 lmm_objects = lump->lmm_objects;
 537
 538         for (i = 0; i < lump->lmm_stripe_count; i++) {
 539                 __u32 len = sizeof(last_id);
 540                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 541                 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 542                                   &len, &last_id, NULL);
 543                 if (rc)
 544                         RETURN(rc);
 545                 if (lmm_objects[i].l_object_id > last_id) {
 546                         CERROR("Setting EA for object > than last id on "
 547                                "ost idx %d "LPD64" > "LPD64" \n",
 548                                lmm_objects[i].l_ost_idx,
 549                                lmm_objects[i].l_object_id, last_id);
 550                         RETURN(-EINVAL);
 551                 }
 552         }
 553
 554         rc = lov_setstripe(exp, 0, lsmp, lump);
 555         if (rc)
 556                 RETURN(rc);
 557
 558         for (i = 0; i < lump->lmm_stripe_count; i++) {
 559                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 560                         lmm_objects[i].l_ost_idx;
 561                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 562                 (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr;
 563         }
 564         RETURN(0);
 565 }
 566
 567
 568 /* Retrieve object striping information.
 569  *
 570  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 571  * the maximum number of OST indices which will fit in the user buffer.
 572  * lmm_magic must be LOV_USER_MAGIC.
 573  */
 574 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
 575                   struct lov_user_md *lump)
 576 {
 577         /*
 578          * XXX huge struct allocated on stack.
 579          */
 580         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 581         struct lov_user_md_v3 lum;
 582         struct lov_mds_md *lmmk = NULL;
 583         int rc, lmm_size;
 584         int lum_size;
 585         mm_segment_t seg;
 586         ENTRY;
 587
 588         if (!lsm)
 589                 RETURN(-ENODATA);
 590
 591         /*
 592          * "Switch to kernel segment" to allow copying from kernel space by
 593          * copy_{to,from}_user().
 594          */
 595         seg = get_fs();
 596         set_fs(KERNEL_DS);
 597
 598         /* we only need the header part from user space to get lmm_magic and
 599          * lmm_stripe_count, (the header part is common to v1 and v3) */
 600         lum_size = sizeof(struct lov_user_md_v1);
 601         if (cfs_copy_from_user(&lum, lump, lum_size))
 602                 GOTO(out_set, rc = -EFAULT);
 603         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 604                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 605                 GOTO(out_set, rc = -EINVAL);
 606
 607         if (lum.lmm_stripe_count &&
 608             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 609                 /* Return right size of stripe to user */
 610                 lum.lmm_stripe_count = lsm->lsm_stripe_count;
 611                 rc = cfs_copy_to_user(lump, &lum, lum_size);
 612                 GOTO(out_set, rc = -EOVERFLOW);
 613         }
 614         rc = lov_packmd(exp, &lmmk, lsm);
 615         if (rc < 0)
 616                 GOTO(out_set, rc);
 617         lmm_size = rc;
 618         rc = 0;
 619
 620         /* FIXME: Bug 1185 - copy fields properly when structs change */
 621         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 622         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 623         CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
 624
 625         if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
 626             ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
 627             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
 628                 lustre_swab_lov_mds_md(lmmk);
 629                 lustre_swab_lov_user_md_objects(
 630                                 (struct lov_user_ost_data*)lmmk->lmm_objects,
 631                                 lmmk->lmm_stripe_count);
 632         }
 633         if (lum.lmm_magic == LOV_USER_MAGIC) {
 634                 /* User request for v1, we need skip lmm_pool_name */
 635                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 636                         memmove((char*)(&lmmk->lmm_stripe_count) +
 637                                 sizeof(lmmk->lmm_stripe_count),
 638                                 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
 639                                 lmmk->lmm_stripe_count *
 640                                 sizeof(struct lov_ost_data_v1));
 641                         lmm_size -= LOV_MAXPOOLNAME;
 642                 }
 643         } else {
 644                 /* if v3 we just have to update the lum_size */
 645                 lum_size = sizeof(struct lov_user_md_v3);
 646         }
 647
 648         /* User wasn't expecting this many OST entries */
 649         if (lum.lmm_stripe_count == 0)
 650                 lmm_size = lum_size;
 651         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 652                 GOTO(out_set, rc = -EOVERFLOW);
 653         /*
 654          * Have a difference between lov_mds_md & lov_user_md.
 655          * So we have to re-order the data before copy to user.
 656          */
 657         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 658         ((struct lov_user_md*)lmmk)->lmm_stripe_offset = 0;
 659         ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 660         if (cfs_copy_to_user(lump, lmmk, lmm_size))
 661                 rc = -EFAULT;
 662
 663         obd_free_diskmd(exp, &lmmk);
 664 out_set:
 665         set_fs(seg);
 666         RETURN(rc);
 667 }