lustre/lov/lov_pack.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_pack.c
  37  *
  38  * (Un)packing of OST/MDS requests
  39  *
  40  * Author: Andreas Dilger <adilger@clusterfs.com>
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_LOV
  44 #ifndef __KERNEL__
  45 #include <liblustre.h>
  46 #endif
  47
  48 #include <lustre_net.h>
  49 #include <obd.h>
  50 #include <obd_lov.h>
  51 #include <obd_class.h>
  52 #include <obd_support.h>
  53 #include <lustre/lustre_user.h>
  54
  55 #include "lov_internal.h"
  56
  57 static void lov_dump_lmm_common(int level, void *lmmp)
  58 {
  59         struct lov_mds_md *lmm = lmmp;
  60
  61         CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n",
  62                le64_to_cpu(lmm->lmm_object_id),
  63                le32_to_cpu(lmm->lmm_magic),
  64                le32_to_cpu(lmm->lmm_pattern));
  65         CDEBUG(level,"stripe_size %u, stripe_count %u\n",
  66                le32_to_cpu(lmm->lmm_stripe_size),
  67                le32_to_cpu(lmm->lmm_stripe_count));
  68 }
  69
  70 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  71                                  int stripe_count)
  72 {
  73         int i;
  74
  75         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  76                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  77                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  78         }
  79
  80         for (i = 0; i < stripe_count; ++i, ++lod) {
  81                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n", i,
  82                        le32_to_cpu(lod->l_ost_idx),
  83                        le64_to_cpu(lod->l_object_gr),
  84                        le64_to_cpu(lod->l_object_id));
  85         }
  86 }
  87
  88 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  89 {
  90         lov_dump_lmm_common(level, lmm);
  91         lov_dump_lmm_objects(level, lmm->lmm_objects,
  92                              le32_to_cpu(lmm->lmm_stripe_count));
  93 }
  94
  95 void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
  96 {
  97         lov_dump_lmm_common(level, &lmmj->lmmj_md);
  98         CDEBUG(level, "extent_count %u\n",
  99                le32_to_cpu(lmmj->lmmj_extent_count));
 100 }
 101
 102 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
 103 {
 104         lov_dump_lmm_common(level, lmm);
 105         CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
 106         lov_dump_lmm_objects(level, lmm->lmm_objects,
 107                              le32_to_cpu(lmm->lmm_stripe_count));
 108 }
 109
 110 void lov_dump_lmm(int level, void *lmm)
 111 {
 112         int magic;
 113
 114         magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic;
 115         switch (magic) {
 116         case LOV_MAGIC_V1:
 117                 return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm));
 118         case LOV_MAGIC_JOIN:
 119                 return lov_dump_lmm_join(level, (struct lov_mds_md_join *)(lmm));
 120         case LOV_MAGIC_V3:
 121                 return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm));
 122         default:
 123                 CERROR("Cannot recognize lmm_magic %x", magic);
 124         }
 125         return;
 126 }
 127
 128 #define LMM_ASSERT(test)                                                \
 129 do {                                                                    \
 130         if (!(test)) lov_dump_lmm(D_ERROR, lmm);                        \
 131         LASSERT(test); /* so we know what assertion failed */           \
 132 } while(0)
 133
 134 /* Pack LOV object metadata for disk storage.  It is packed in LE byte
 135  * order and is opaque to the networking layer.
 136  *
 137  * XXX In the future, this will be enhanced to get the EA size from the
 138  *     underlying OSC device(s) to get their EA sizes so we can stack
 139  *     LOVs properly.  For now lov_mds_md_size() just assumes one obd_id
 140  *     per stripe.
 141  */
 142 int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 143                struct lov_stripe_md *lsm)
 144 {
 145         struct obd_device *obd = class_exp2obd(exp);
 146         struct lov_obd *lov = &obd->u.lov;
 147         struct lov_mds_md_v1 *lmmv1;
 148         struct lov_mds_md_v3 *lmmv3;
 149         int stripe_count;
 150         struct lov_ost_data_v1 *lmm_objects;
 151         int lmm_size, lmm_magic;
 152         int i;
 153         ENTRY;
 154
 155         if (lsm) {
 156                 lmm_magic = lsm->lsm_magic;
 157
 158                 /* If we are just sizing the EA, limit the stripe count
 159                  * to the actual number of OSTs in this filesystem. */
 160                 if (!lmmp) {
 161                         stripe_count = lov_get_stripecnt(lov,
 162                                                          lsm->lsm_stripe_count);
 163                         lsm->lsm_stripe_count = stripe_count;
 164                 } else {
 165                         stripe_count = lsm->lsm_stripe_count;
 166                 }
 167         } else {
 168                 /* No needs to allocated more than LOV_MAX_STRIPE_COUNT.
 169                  * Anyway, this is pretty inaccurate since ld_tgt_count now
 170                  * represents max index and we should rely on the actual number
 171                  * of OSTs instead */
 172                 stripe_count = min((__u32)LOV_MAX_STRIPE_COUNT,
 173                                    lov->desc.ld_tgt_count);
 174
 175                 if (lmmp && *lmmp)
 176                         lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 177                 else
 178                         /* lsm == NULL and lmmp == NULL */
 179                         lmm_magic = LOV_MAGIC;
 180         }
 181
 182         if ((lmm_magic != LOV_MAGIC_V1) &&
 183             (lmm_magic != LOV_MAGIC_V3)) {
 184                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 185                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 186                 RETURN(-EINVAL);
 187
 188         }
 189
 190         /* XXX LOV STACKING call into osc for sizes */
 191         lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 192
 193         if (!lmmp)
 194                 RETURN(lmm_size);
 195
 196         if (*lmmp && !lsm) {
 197                 stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count);
 198                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 199                 OBD_FREE(*lmmp, lmm_size);
 200                 *lmmp = NULL;
 201                 RETURN(0);
 202         }
 203
 204         if (!*lmmp) {
 205                 OBD_ALLOC(*lmmp, lmm_size);
 206                 if (!*lmmp)
 207                         RETURN(-ENOMEM);
 208         }
 209
 210         CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
 211                lmm_magic, lmm_size);
 212
 213         lmmv1 = *lmmp;
 214         lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
 215         if (lmm_magic == LOV_MAGIC_V3)
 216                 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 217         else
 218                 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
 219
 220         if (!lsm)
 221                 RETURN(lmm_size);
 222
 223         /* lmmv1 and lmmv3 point to the same struct and have the
 224          * same first fields
 225          */
 226         lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
 227         lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
 228         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 229         lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count);
 230         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 231         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 232                 strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 233                         LOV_MAXPOOLNAME);
 234                 lmm_objects = lmmv3->lmm_objects;
 235         } else {
 236                 lmm_objects = lmmv1->lmm_objects;
 237         }
 238
 239         for (i = 0; i < stripe_count; i++) {
 240                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 241
 242                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 243                 LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n",
 244                          lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
 245                 lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
 246                 lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
 247                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 248                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 249         }
 250
 251         RETURN(lmm_size);
 252 }
 253
 254 /* Find the max stripecount we should use */
 255 int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count)
 256 {
 257         if (!stripe_count)
 258                 stripe_count = lov->desc.ld_default_stripe_count;
 259         if (stripe_count > lov->desc.ld_active_tgt_count)
 260                 stripe_count = lov->desc.ld_active_tgt_count;
 261         if (!stripe_count)
 262                 stripe_count = 1;
 263         /* for now, we limit the stripe count directly, when bug 4424 is
 264          * fixed this needs to be somewhat dynamic based on whether ext3
 265          * can handle larger EA sizes. */
 266         if (stripe_count > LOV_MAX_STRIPE_COUNT)
 267                 stripe_count = LOV_MAX_STRIPE_COUNT;
 268
 269         return stripe_count;
 270 }
 271
 272
 273 static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
 274 {
 275         int rc;
 276
 277         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 278                 char *buffer;
 279                 int sz;
 280
 281                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 282                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 283                 sz = lmm_bytes * 2 + 1;
 284                 OBD_ALLOC(buffer, sz);
 285                 if (buffer != NULL) {
 286                         int i;
 287
 288                         for (i = 0; i < lmm_bytes; i++)
 289                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 290                         buffer[sz] = '\0';
 291                         CERROR("%s\n", buffer);
 292                         OBD_FREE(buffer, sz);
 293                 }
 294                 return -EINVAL;
 295         }
 296         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 297                                      lmm_bytes, stripe_count);
 298         return rc;
 299 }
 300
 301 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count,
 302                       int pattern, int magic)
 303 {
 304         int i, lsm_size;
 305         ENTRY;
 306
 307         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 308
 309         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 310         if (!*lsmp) {
 311                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 312                 RETURN(-ENOMEM);
 313         }
 314
 315         spin_lock_init(&(*lsmp)->lsm_lock);
 316         (*lsmp)->lsm_magic = magic;
 317         (*lsmp)->lsm_stripe_count = stripe_count;
 318         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 319         (*lsmp)->lsm_pattern = pattern;
 320         (*lsmp)->lsm_pool_name[0] = '\0';
 321         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 322
 323         for (i = 0; i < stripe_count; i++)
 324                 loi_init((*lsmp)->lsm_oinfo[i]);
 325
 326         RETURN(lsm_size);
 327 }
 328
 329 void lov_free_memmd(struct lov_stripe_md **lsmp)
 330 {
 331         struct lov_stripe_md *lsm = *lsmp;
 332
 333         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 334         lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 335
 336         *lsmp = NULL;
 337 }
 338
 339
 340 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 341  * order and is opaque to the networking layer.
 342  */
 343 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 344                  struct lov_mds_md *lmm, int lmm_bytes)
 345 {
 346         struct obd_device *obd = class_exp2obd(exp);
 347         struct lov_obd *lov = &obd->u.lov;
 348         int rc = 0, stripe_count, lsm_size;
 349         __u32 magic;
 350         ENTRY;
 351
 352         /* If passed an MDS struct use values from there, otherwise defaults */
 353         if (lmm) {
 354                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 355                 if (rc)
 356                         RETURN(rc);
 357                 magic = le32_to_cpu(lmm->lmm_magic);
 358         } else {
 359                 stripe_count = lov_get_stripecnt(lov, 0);
 360                 magic = LOV_MAGIC;
 361         }
 362
 363         /* If we aren't passed an lsmp struct, we just want the size */
 364         if (!lsmp) {
 365                 /* XXX LOV STACKING call into osc for sizes */
 366                 LBUG();
 367                 RETURN(lov_stripe_md_size(stripe_count));
 368         }
 369         /* If we are passed an allocated struct but nothing to unpack, free */
 370         if (*lsmp && !lmm) {
 371                 lov_free_memmd(lsmp);
 372                 RETURN(0);
 373         }
 374
 375         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 376                                    magic);
 377         if (lsm_size < 0)
 378                 RETURN(lsm_size);
 379
 380         /* If we are passed a pointer but nothing to unpack, we only alloc */
 381         if (!lmm)
 382                 RETURN(lsm_size);
 383
 384         LASSERT(lsm_op_find(magic) != NULL);
 385         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 386         if (rc) {
 387                 lov_free_memmd(lsmp);
 388                 RETURN(rc);
 389         }
 390
 391         RETURN(lsm_size);
 392 }
 393
 394 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
 395                            struct lov_stripe_md **lsmp,
 396                            struct lov_user_md *lump)
 397 {
 398         struct obd_device *obd = class_exp2obd(exp);
 399         struct lov_obd *lov = &obd->u.lov;
 400         struct lov_user_md_v3 lumv3;
 401         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 402         int lmm_magic;
 403         int stripe_count;
 404         int rc;
 405         ENTRY;
 406
 407         if (copy_from_user(&lumv3, lump, sizeof(struct lov_user_md_v1)))
 408                 RETURN(-EFAULT);
 409
 410         lmm_magic = lumv1->lmm_magic;
 411
 412         if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
 413                 lustre_swab_lov_user_md_v1(lumv1);
 414                 lmm_magic = LOV_USER_MAGIC_V1;
 415         } else if (lmm_magic == LOV_USER_MAGIC_V3) {
 416                 if (copy_from_user(&lumv3, lump, sizeof(lumv3)))
 417                         RETURN(-EFAULT);
 418         } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
 419                 if (copy_from_user(&lumv3, lump, sizeof(lumv3)))
 420                         RETURN(-EFAULT);
 421                 lustre_swab_lov_user_md_v3(&lumv3);
 422                 lmm_magic = LOV_USER_MAGIC_V3;
 423         } else if (lmm_magic != LOV_USER_MAGIC_V1) {
 424                 CDEBUG(D_IOCTL,
 425                        "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 426                        lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
 427                        RETURN(-EINVAL);
 428         }
 429
 430         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 431          * fields, we use lumv1 to avoid code duplication */
 432
 433         if (lumv1->lmm_pattern == 0) {
 434                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 435                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 436         }
 437
 438         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 439                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 440                        lumv1->lmm_pattern);
 441                 RETURN(-EINVAL);
 442         }
 443
 444         /* 64kB is the largest common page size we see (ia64), and matches the
 445          * check in lfs */
 446         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 447                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 448                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 449                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 450         }
 451
 452         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 453             (lumv1->lmm_stripe_offset !=
 454              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 455                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 456                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 457                 RETURN(-EINVAL);
 458         }
 459         stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count);
 460
 461         if (max_lmm_size) {
 462                 int max_stripes = (max_lmm_size -
 463                                    lov_mds_md_size(0, lmm_magic)) /
 464                                    sizeof(struct lov_ost_data_v1);
 465                 if (unlikely(max_stripes < stripe_count)) {
 466                         CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
 467                                stripe_count, max_stripes);
 468                         stripe_count = max_stripes;
 469                 }
 470         }
 471
 472         if (lmm_magic == LOV_USER_MAGIC_V3) {
 473                 struct pool_desc *pool;
 474
 475                 pool = lov_find_pool(lov, lumv3.lmm_pool_name);
 476                 if (pool != NULL) {
 477                         if (lumv3.lmm_stripe_offset !=
 478                             (typeof(lumv3.lmm_stripe_offset))(-1)) {
 479                                 rc = lov_check_index_in_pool(
 480                                         lumv3.lmm_stripe_offset, pool);
 481                                 if (rc < 0) {
 482                                         lov_pool_putref(pool);
 483                                         RETURN(-EINVAL);
 484                                 }
 485                         }
 486
 487                         if (stripe_count > pool_tgt_count(pool))
 488                                 stripe_count = pool_tgt_count(pool);
 489
 490                         lov_pool_putref(pool);
 491                 }
 492         }
 493
 494         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 495
 496         if (rc >= 0) {
 497                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 498                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 499                 if (lmm_magic == LOV_USER_MAGIC_V3)
 500                         strncpy((*lsmp)->lsm_pool_name, lumv3.lmm_pool_name,
 501                                 LOV_MAXPOOLNAME);
 502                 rc = 0;
 503         }
 504
 505         RETURN(rc);
 506 }
 507
 508 /* Configure object striping information on a new file.
 509  *
 510  * @lmmu is a pointer to a user struct with one or more of the fields set to
 511  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 512  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 513  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 514  */
 515 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
 516                   struct lov_stripe_md **lsmp, struct lov_user_md *lump)
 517 {
 518         int rc;
 519         mm_segment_t seg;
 520
 521         seg = get_fs();
 522         set_fs(KERNEL_DS);
 523
 524         rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
 525         set_fs(seg);
 526         RETURN(rc);
 527 }
 528
 529 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 530               struct lov_user_md *lump)
 531 {
 532         int i;
 533         int rc;
 534         struct obd_export *oexp;
 535         struct lov_obd *lov = &exp->exp_obd->u.lov;
 536         obd_id last_id = 0;
 537         struct lov_user_ost_data_v1 *lmm_objects;
 538
 539         ENTRY;
 540
 541         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 542                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 543         else
 544                 lmm_objects = lump->lmm_objects;
 545
 546         for (i = 0; i < lump->lmm_stripe_count; i++) {
 547                 __u32 len = sizeof(last_id);
 548                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 549                 rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 550                                   &len, &last_id, NULL);
 551                 if (rc)
 552                         RETURN(rc);
 553                 if (lmm_objects[i].l_object_id > last_id) {
 554                         CERROR("Setting EA for object > than last id on "
 555                                "ost idx %d "LPD64" > "LPD64" \n",
 556                                lmm_objects[i].l_ost_idx,
 557                                lmm_objects[i].l_object_id, last_id);
 558                         RETURN(-EINVAL);
 559                 }
 560         }
 561
 562         rc = lov_setstripe(exp, 0, lsmp, lump);
 563         if (rc)
 564                 RETURN(rc);
 565
 566         for (i = 0; i < lump->lmm_stripe_count; i++) {
 567                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 568                         lmm_objects[i].l_ost_idx;
 569                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 570                 (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr;
 571         }
 572         RETURN(0);
 573 }
 574
 575
 576 /* Retrieve object striping information.
 577  *
 578  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 579  * the maximum number of OST indices which will fit in the user buffer.
 580  * lmm_magic must be LOV_USER_MAGIC.
 581  */
 582 int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
 583                   struct lov_user_md *lump)
 584 {
 585         /*
 586          * XXX huge struct allocated on stack.
 587          */
 588         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 589         struct lov_user_md_v3 lum;
 590         struct lov_mds_md *lmmk = NULL;
 591         int rc, lmm_size;
 592         int lum_size;
 593         mm_segment_t seg;
 594         ENTRY;
 595
 596         if (!lsm)
 597                 RETURN(-ENODATA);
 598
 599         /*
 600          * "Switch to kernel segment" to allow copying from kernel space by
 601          * copy_{to,from}_user().
 602          */
 603         seg = get_fs();
 604         set_fs(KERNEL_DS);
 605
 606         /* we only need the header part from user space to get lmm_magic and
 607          * lmm_stripe_count, (the header part is common to v1 and v3) */
 608         lum_size = sizeof(struct lov_user_md_v1);
 609         if (copy_from_user(&lum, lump, lum_size))
 610                 GOTO(out_set, rc = -EFAULT);
 611         else if ((lum.lmm_magic != LOV_USER_MAGIC) &&
 612                  (lum.lmm_magic != LOV_USER_MAGIC_V3))
 613                 GOTO(out_set, rc = -EINVAL);
 614
 615         if (lum.lmm_stripe_count &&
 616             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 617                 /* Return right size of stripe to user */
 618                 lum.lmm_stripe_count = lsm->lsm_stripe_count;
 619                 rc = copy_to_user(lump, &lum, lum_size);
 620                 GOTO(out_set, rc = -EOVERFLOW);
 621         }
 622         rc = lov_packmd(exp, &lmmk, lsm);
 623         if (rc < 0)
 624                 GOTO(out_set, 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] == sizeof lmmk->lmm_objects[0]);
 632
 633         if ((cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) &&
 634             ((lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) ||
 635             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)))) {
 636                 lustre_swab_lov_mds_md(lmmk);
 637                 lustre_swab_lov_user_md_objects(
 638                                 (struct lov_user_ost_data*)lmmk->lmm_objects,
 639                                 lmmk->lmm_stripe_count);
 640         }
 641         if (lum.lmm_magic == LOV_USER_MAGIC) {
 642                 /* User request for v1, we need skip lmm_pool_name */
 643                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 644                         memmove((char*)(&lmmk->lmm_stripe_count) +
 645                                 sizeof(lmmk->lmm_stripe_count),
 646                                 ((struct lov_mds_md_v3*)lmmk)->lmm_objects,
 647                                 lmmk->lmm_stripe_count *
 648                                 sizeof(struct lov_ost_data_v1));
 649                         lmm_size -= LOV_MAXPOOLNAME;
 650                 }
 651         } else {
 652                 /* if v3 we just have to update the lum_size */
 653                 lum_size = sizeof(struct lov_user_md_v3);
 654         }
 655
 656         /* User wasn't expecting this many OST entries */
 657         if (lum.lmm_stripe_count == 0)
 658                 lmm_size = lum_size;
 659         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 660                 GOTO(out_set, rc = -EOVERFLOW);
 661         /*
 662          * Have a difference between lov_mds_md & lov_user_md.
 663          * So we have to re-order the data before copy to user.
 664          */
 665         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 666         ((struct lov_user_md*)lmmk)->lmm_stripe_offset = 0;
 667         ((struct lov_user_md*)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 668         if (copy_to_user(lump, lmmk, lmm_size))
 669                 rc = -EFAULT;
 670
 671         obd_free_diskmd(exp, &lmmk);
 672 out_set:
 673         set_fs(seg);
 674         RETURN(rc);
 675 }