lustre/lov/lov_pack.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  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_common(int level, void *lmmp)
  58 {
  59         struct lov_mds_md *lmm = lmmp;
  60         struct ost_id   oi;
  61
  62         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
  63         CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
  64                POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
  65                le32_to_cpu(lmm->lmm_pattern));
  66         CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
  67                le32_to_cpu(lmm->lmm_stripe_size),
  68                le16_to_cpu(lmm->lmm_stripe_count),
  69                le16_to_cpu(lmm->lmm_layout_gen));
  70 }
  71
  72 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  73                                  int stripe_count)
  74 {
  75         int i;
  76
  77         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  78                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  79                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  80                 return;
  81         }
  82
  83         for (i = 0; i < stripe_count; ++i, ++lod) {
  84                 struct ost_id   oi;
  85
  86                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  87                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  88                        le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  89         }
  90 }
  91
  92 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  93 {
  94         lov_dump_lmm_common(level, lmm);
  95         lov_dump_lmm_objects(level, lmm->lmm_objects,
  96                              le16_to_cpu(lmm->lmm_stripe_count));
  97 }
  98
  99 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
 100 {
 101         lov_dump_lmm_common(level, lmm);
 102         CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
 103         lov_dump_lmm_objects(level, lmm->lmm_objects,
 104                              le16_to_cpu(lmm->lmm_stripe_count));
 105 }
 106
 107 void lov_dump_lmm(int level, void *lmm)
 108 {
 109         int magic;
 110
 111         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 112         switch (magic) {
 113         case LOV_MAGIC_V1:
 114                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 115         case LOV_MAGIC_V3:
 116                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 117         default:
 118                 CERROR("Cannot recognize lmm_magic %x", magic);
 119         }
 120         return;
 121 }
 122
 123 #define LMM_ASSERT(test)                                                \
 124 do {                                                                    \
 125         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 126         LASSERT(test); /* so we know what assertion failed */           \
 127 } while(0)
 128
 129 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 130  * order and is opaque to the networking layer.
 131  *
 132  * XXX In the future, this will be enhanced to get the EA size from the
 133  *     underlying OSC device(s) to get their EA sizes so we can stack
 134  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 135  *     per stripe.
 136  */
 137 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 138                struct lov_stripe_md *lsm)
 139 {
 140         struct obd_device *obd = class_exp2obd(exp);
 141         struct lov_obd *lov = &obd->u.lov;
 142         struct lov_mds_md_v1 *lmmv1;
 143         struct lov_mds_md_v3 *lmmv3;
 144         __u16 stripe_count;
 145         struct lov_ost_data_v1 *lmm_objects;
 146         int lmm_size, lmm_magic;
 147         int i;
 148         int cplen = 0;
 149         ENTRY;
 150
 151         if (lsm) {
 152                 lmm_magic = lsm->lsm_magic;
 153         } else {
 154                 if (lmmp && *lmmp)
 155                         lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 156                 else
 157                         /* lsm == NULL and lmmp == NULL */
 158                         lmm_magic = LOV_MAGIC;
 159         }
 160
 161         if ((lmm_magic != LOV_MAGIC_V1) &&
 162             (lmm_magic != LOV_MAGIC_V3)) {
 163                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 164                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 165                 RETURN(-EINVAL);
 166
 167         }
 168
 169         if (lsm) {
 170                 /* If we are just sizing the EA, limit the stripe count
 171                  * to the actual number of OSTs in this filesystem. */
 172                 if (!lmmp) {
 173                         stripe_count = lov_get_stripecnt(lov, lmm_magic,
 174                                                          lsm->lsm_stripe_count);
 175                         lsm->lsm_stripe_count = stripe_count;
 176                 } else {
 177                         stripe_count = lsm->lsm_stripe_count;
 178                 }
 179         } else {
 180                 /* No need to allocate more than maximum supported stripes.
 181                  * Anyway, this is pretty inaccurate since ld_tgt_count now
 182                  * represents max index and we should rely on the actual number
 183                  * of OSTs instead */
 184                 stripe_count = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
 185                                                     lmm_magic);
 186                 if (stripe_count > lov->desc.ld_tgt_count)
 187                         stripe_count = lov->desc.ld_tgt_count;
 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 = le16_to_cpu((*lmmp)->lmm_stripe_count);
 198                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 199                 OBD_FREE_LARGE(*lmmp, lmm_size);
 200                 *lmmp = NULL;
 201                 RETURN(0);
 202         }
 203
 204         if (!*lmmp) {
 205                 OBD_ALLOC_LARGE(*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         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 227         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 228         lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
 229         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 230         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 231         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 232                 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 233                                 sizeof(lmmv3->lmm_pool_name));
 234                 if (cplen >= sizeof(lmmv3->lmm_pool_name))
 235                         RETURN(-E2BIG);
 236                 lmm_objects = lmmv3->lmm_objects;
 237         } else {
 238                 lmm_objects = lmmv1->lmm_objects;
 239         }
 240
 241         for (i = 0; i < stripe_count; i++) {
 242                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 243                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 244                 LASSERTF(ostid_id(&loi->loi_oi) != 0, "lmm_oi "DOSTID
 245                          " stripe %u/%u idx %u\n", POSTID(&lmmv1->lmm_oi),
 246                          i, stripe_count, loi->loi_ost_idx);
 247                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 248                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 249                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 250         }
 251
 252         RETURN(lmm_size);
 253 }
 254
 255 /* Find the max stripecount we should use */
 256 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 257 {
 258         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 259
 260         if (!stripe_count)
 261                 stripe_count = lov->desc.ld_default_stripe_count;
 262         if (stripe_count > lov->desc.ld_active_tgt_count)
 263                 stripe_count = lov->desc.ld_active_tgt_count;
 264         if (!stripe_count)
 265                 stripe_count = 1;
 266
 267         /* stripe count is based on whether ldiskfs can handle
 268          * larger EA sizes */
 269         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 270             lov->lov_ocd.ocd_max_easize)
 271                 max_stripes = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
 272                                                    magic);
 273
 274         if (stripe_count > max_stripes)
 275                 stripe_count = max_stripes;
 276
 277         return stripe_count;
 278 }
 279
 280
 281 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
 282 {
 283         int rc;
 284
 285         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 286                 char *buffer;
 287                 int sz;
 288
 289                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 290                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 291                 sz = lmm_bytes * 2 + 1;
 292                 OBD_ALLOC_LARGE(buffer, sz);
 293                 if (buffer != NULL) {
 294                         int i;
 295
 296                         for (i = 0; i < lmm_bytes; i++)
 297                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 298                         buffer[sz - 1] = '\0';
 299                         CERROR("%s\n", buffer);
 300                         OBD_FREE_LARGE(buffer, sz);
 301                 }
 302                 return -EINVAL;
 303         }
 304         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 305                                      lmm_bytes, stripe_count);
 306         return rc;
 307 }
 308
 309 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
 310                     int pattern, int magic)
 311 {
 312         int i, lsm_size;
 313         ENTRY;
 314
 315         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 316
 317         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 318         if (!*lsmp) {
 319                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 320                 RETURN(-ENOMEM);
 321         }
 322
 323         cfs_atomic_set(&(*lsmp)->lsm_refc, 1);
 324         spin_lock_init(&(*lsmp)->lsm_lock);
 325         (*lsmp)->lsm_magic = magic;
 326         (*lsmp)->lsm_stripe_count = stripe_count;
 327         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 328         (*lsmp)->lsm_pattern = pattern;
 329         (*lsmp)->lsm_pool_name[0] = '\0';
 330         (*lsmp)->lsm_layout_gen = 0;
 331         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 332
 333         for (i = 0; i < stripe_count; i++)
 334                 loi_init((*lsmp)->lsm_oinfo[i]);
 335
 336         RETURN(lsm_size);
 337 }
 338
 339 int lov_free_memmd(struct lov_stripe_md **lsmp)
 340 {
 341         struct lov_stripe_md *lsm = *lsmp;
 342         int refc;
 343
 344         *lsmp = NULL;
 345         LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 0);
 346         if ((refc = cfs_atomic_dec_return(&lsm->lsm_refc)) == 0) {
 347                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 348                 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 349         }
 350         return refc;
 351 }
 352
 353
 354 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 355  * order and is opaque to the networking layer.
 356  */
 357 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 358                  struct lov_mds_md *lmm, int lmm_bytes)
 359 {
 360         struct obd_device *obd = class_exp2obd(exp);
 361         struct lov_obd *lov = &obd->u.lov;
 362         int rc = 0, lsm_size;
 363         __u16 stripe_count;
 364         __u32 magic;
 365         ENTRY;
 366
 367         /* If passed an MDS struct use values from there, otherwise defaults */
 368         if (lmm) {
 369                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 370                 if (rc)
 371                         RETURN(rc);
 372                 magic = le32_to_cpu(lmm->lmm_magic);
 373         } else {
 374                 magic = LOV_MAGIC;
 375                 stripe_count = lov_get_stripecnt(lov, magic, 0);
 376         }
 377
 378         /* If we aren't passed an lsmp struct, we just want the size */
 379         if (!lsmp) {
 380                 /* XXX LOV STACKING call into osc for sizes */
 381                 LBUG();
 382                 RETURN(lov_stripe_md_size(stripe_count));
 383         }
 384         /* If we are passed an allocated struct but nothing to unpack, free */
 385         if (*lsmp && !lmm) {
 386                 lov_free_memmd(lsmp);
 387                 RETURN(0);
 388         }
 389
 390         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 391                                    magic);
 392         if (lsm_size < 0)
 393                 RETURN(lsm_size);
 394
 395         /* If we are passed a pointer but nothing to unpack, we only alloc */
 396         if (!lmm)
 397                 RETURN(lsm_size);
 398
 399         LASSERT(lsm_op_find(magic) != NULL);
 400         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 401         if (rc) {
 402                 lov_free_memmd(lsmp);
 403                 RETURN(rc);
 404         }
 405
 406         RETURN(lsm_size);
 407 }
 408
 409 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
 410                            struct lov_stripe_md **lsmp,
 411                            struct lov_user_md *lump)
 412 {
 413         struct obd_device *obd = class_exp2obd(exp);
 414         struct lov_obd *lov = &obd->u.lov;
 415         char buffer[sizeof(struct lov_user_md_v3)];
 416         struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0];
 417         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0];
 418         int lmm_magic;
 419         __u16 stripe_count;
 420         int rc;
 421         int cplen = 0;
 422         ENTRY;
 423
 424         rc = lov_lum_swab_if_needed(lumv3, &lmm_magic, lump);
 425         if (rc)
 426                 RETURN(rc);
 427
 428         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 429          * fields, we use lumv1 to avoid code duplication */
 430
 431         if (lumv1->lmm_pattern == 0) {
 432                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 433                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 434         }
 435
 436         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 437                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 438                        lumv1->lmm_pattern);
 439                 RETURN(-EINVAL);
 440         }
 441
 442         /* 64kB is the largest common page size we see (ia64), and matches the
 443          * check in lfs */
 444         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 445                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 446                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 447                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 448         }
 449
 450         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 451             (lumv1->lmm_stripe_offset !=
 452              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 453                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 454                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 455                 RETURN(-EINVAL);
 456         }
 457         stripe_count = lov_get_stripecnt(lov, lmm_magic,
 458                                          lumv1->lmm_stripe_count);
 459
 460         if (max_lmm_size) {
 461                 int max_stripes = (max_lmm_size -
 462                                    lov_mds_md_size(0, lmm_magic)) /
 463                                    sizeof(struct lov_ost_data_v1);
 464                 if (unlikely(max_stripes < stripe_count)) {
 465                         CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
 466                                stripe_count, max_stripes);
 467                         stripe_count = max_stripes;
 468                 }
 469         }
 470
 471         if (lmm_magic == LOV_USER_MAGIC_V3) {
 472                 struct pool_desc *pool;
 473
 474                 /* In the function below, .hs_keycmp resolves to
 475                  * pool_hashkey_keycmp() */
 476                 /* coverity[overrun-buffer-val] */
 477                 pool = lov_find_pool(lov, lumv3->lmm_pool_name);
 478                 if (pool != NULL) {
 479                         if (lumv3->lmm_stripe_offset !=
 480                             (typeof(lumv3->lmm_stripe_offset))(-1)) {
 481                                 rc = lov_check_index_in_pool(
 482                                         lumv3->lmm_stripe_offset, pool);
 483                                 if (rc < 0) {
 484                                         lov_pool_putref(pool);
 485                                         RETURN(-EINVAL);
 486                                 }
 487                         }
 488
 489                         if (stripe_count > pool_tgt_count(pool))
 490                                 stripe_count = pool_tgt_count(pool);
 491
 492                         lov_pool_putref(pool);
 493                 }
 494         }
 495
 496         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 497
 498         if (rc >= 0) {
 499                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 500                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 501                 if (lmm_magic == LOV_USER_MAGIC_V3) {
 502                         cplen = strlcpy((*lsmp)->lsm_pool_name,
 503                                         lumv3->lmm_pool_name,
 504                                         sizeof((*lsmp)->lsm_pool_name));
 505                         if (cplen >= sizeof((*lsmp)->lsm_pool_name))
 506                                 rc = -E2BIG;
 507                 }
 508                 rc = 0;
 509         }
 510
 511         RETURN(rc);
 512 }
 513
 514 /* Configure object striping information on a new file.
 515  *
 516  * @lmmu is a pointer to a user struct with one or more of the fields set to
 517  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 518  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 519  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 520  */
 521 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
 522                   struct lov_stripe_md **lsmp, struct lov_user_md *lump)
 523 {
 524         int rc;
 525         mm_segment_t seg;
 526
 527         seg = get_fs();
 528         set_fs(KERNEL_DS);
 529
 530         rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
 531         set_fs(seg);
 532         RETURN(rc);
 533 }
 534
 535 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 536               struct lov_user_md *lump)
 537 {
 538         int i;
 539         int rc;
 540         struct obd_export *oexp;
 541         struct lov_obd *lov = &exp->exp_obd->u.lov;
 542         obd_id last_id = 0;
 543         struct lov_user_ost_data_v1 *lmm_objects;
 544
 545         ENTRY;
 546
 547         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 548                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 549         else
 550                 lmm_objects = lump->lmm_objects;
 551
 552         for (i = 0; i < lump->lmm_stripe_count; i++) {
 553                 __u32 len = sizeof(last_id);
 554                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 555                 rc = obd_get_info(NULL, oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 556                                   &len, &last_id, NULL);
 557                 if (rc)
 558                         RETURN(rc);
 559                 if (ostid_id(&lmm_objects[i].l_ost_oi) > last_id) {
 560                         CERROR("Setting EA for object > than last id on"
 561                                " ost idx %d "DOSTID" > "LPD64" \n",
 562                                lmm_objects[i].l_ost_idx,
 563                                POSTID(&lmm_objects[i].l_ost_oi), last_id);
 564                         RETURN(-EINVAL);
 565                 }
 566         }
 567
 568         rc = lov_setstripe(exp, 0, lsmp, lump);
 569         if (rc)
 570                 RETURN(rc);
 571
 572         for (i = 0; i < lump->lmm_stripe_count; i++) {
 573                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 574                         lmm_objects[i].l_ost_idx;
 575                 (*lsmp)->lsm_oinfo[i]->loi_oi = lmm_objects[i].l_ost_oi;
 576         }
 577         RETURN(0);
 578 }
 579
 580
 581 /* Retrieve object striping information.
 582  *
 583  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 584  * the maximum number of OST indices which will fit in the user buffer.
 585  * lmm_magic must be LOV_USER_MAGIC.
 586  */
 587 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
 588                   struct lov_user_md *lump)
 589 {
 590         /*
 591          * XXX huge struct allocated on stack.
 592          */
 593         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 594         struct lov_user_md_v3 lum;
 595         struct lov_mds_md *lmmk = NULL;
 596         int rc, lmm_size;
 597         int lum_size;
 598         mm_segment_t seg;
 599         ENTRY;
 600
 601         if (!lsm)
 602                 RETURN(-ENODATA);
 603
 604         /*
 605          * "Switch to kernel segment" to allow copying from kernel space by
 606          * copy_{to,from}_user().
 607          */
 608         seg = get_fs();
 609         set_fs(KERNEL_DS);
 610
 611         /* we only need the header part from user space to get lmm_magic and
 612          * lmm_stripe_count, (the header part is common to v1 and v3) */
 613         lum_size = sizeof(struct lov_user_md_v1);
 614         if (cfs_copy_from_user(&lum, lump, lum_size))
 615                 GOTO(out_set, rc = -EFAULT);
 616         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 617                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 618                 GOTO(out_set, rc = -EINVAL);
 619
 620         if (lum.lmm_stripe_count &&
 621             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 622                 /* Return right size of stripe to user */
 623                 lum.lmm_stripe_count = lsm->lsm_stripe_count;
 624                 rc = cfs_copy_to_user(lump, &lum, lum_size);
 625                 GOTO(out_set, rc = -EOVERFLOW);
 626         }
 627         rc = lov_packmd(exp, &lmmk, lsm);
 628         if (rc < 0)
 629                 GOTO(out_set, rc);
 630         lmm_size = rc;
 631         rc = 0;
 632
 633         /* FIXME: Bug 1185 - copy fields properly when structs change */
 634         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 635         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 636         CLASSERT(sizeof lum.lmm_objects[0] == sizeof lmmk->lmm_objects[0]);
 637
 638         if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
 639             ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
 640             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
 641                 lustre_swab_lov_mds_md(lmmk);
 642                 lustre_swab_lov_user_md_objects(
 643                                 (struct lov_user_ost_data*)lmmk->lmm_objects,
 644                                 lmmk->lmm_stripe_count);
 645         }
 646         if (lum.lmm_magic == LOV_USER_MAGIC) {
 647                 /* User request for v1, we need skip lmm_pool_name */
 648                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 649                         memmove((char*)(&lmmk->lmm_stripe_count) +
 650                                 sizeof(lmmk->lmm_stripe_count),
 651                                 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
 652                                 lmmk->lmm_stripe_count *
 653                                 sizeof(struct lov_ost_data_v1));
 654                         lmm_size -= LOV_MAXPOOLNAME;
 655                 }
 656         } else {
 657                 /* if v3 we just have to update the lum_size */
 658                 lum_size = sizeof(struct lov_user_md_v3);
 659         }
 660
 661         /* User wasn't expecting this many OST entries */
 662         if (lum.lmm_stripe_count == 0)
 663                 lmm_size = lum_size;
 664         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 665                 GOTO(out_set, rc = -EOVERFLOW);
 666         /*
 667          * Have a difference between lov_mds_md & lov_user_md.
 668          * So we have to re-order the data before copy to user.
 669          */
 670         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 671         lum.lmm_layout_gen = lmmk->lmm_layout_gen;
 672         ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
 673         ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 674         if (cfs_copy_to_user(lump, lmmk, lmm_size))
 675                 rc = -EFAULT;
 676
 677         obd_free_diskmd(exp, &lmmk);
 678 out_set:
 679         set_fs(seg);
 680         RETURN(rc);
 681 }