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, 2012, 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 static void lov_dump_lmm_common(int level, void *lmmp)
  58 {
  59         struct lov_mds_md *lmm = lmmp;
  60         struct ost_id   oi;
  61
  62         ostid_le_to_cpu(&lmm->lmm_oi, &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         }
  81
  82         for (i = 0; i < stripe_count; ++i, ++lod) {
  83                 struct ost_id   oi;
  84
  85                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  86                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  87                        le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  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                              le16_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                              le16_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         __u16 stripe_count;
 144         struct lov_ost_data_v1 *lmm_objects;
 145         int lmm_size, lmm_magic;
 146         int i;
 147         int cplen = 0;
 148         ENTRY;
 149
 150         if (lsm) {
 151                 lmm_magic = lsm->lsm_magic;
 152         } else {
 153                 if (lmmp && *lmmp)
 154                         lmm_magic = le32_to_cpu((*lmmp)->lmm_magic);
 155                 else
 156                         /* lsm == NULL and lmmp == NULL */
 157                         lmm_magic = LOV_MAGIC;
 158         }
 159
 160         if ((lmm_magic != LOV_MAGIC_V1) &&
 161             (lmm_magic != LOV_MAGIC_V3)) {
 162                 CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 163                         lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 164                 RETURN(-EINVAL);
 165
 166         }
 167
 168         if (lsm) {
 169                 /* If we are just sizing the EA, limit the stripe count
 170                  * to the actual number of OSTs in this filesystem. */
 171                 if (!lmmp) {
 172                         stripe_count = lov_get_stripecnt(lov, lmm_magic,
 173                                                          lsm->lsm_stripe_count);
 174                         lsm->lsm_stripe_count = stripe_count;
 175                 } else {
 176                         stripe_count = lsm->lsm_stripe_count;
 177                 }
 178         } else {
 179                 /* No need to allocate more than maximum supported stripes.
 180                  * Anyway, this is pretty inaccurate since ld_tgt_count now
 181                  * represents max index and we should rely on the actual number
 182                  * of OSTs instead */
 183                 stripe_count = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
 184                                                     lmm_magic);
 185                 if (stripe_count > lov->desc.ld_tgt_count)
 186                         stripe_count = lov->desc.ld_tgt_count;
 187         }
 188
 189         /* XXX LOV STACKING call into osc for sizes */
 190         lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 191
 192         if (!lmmp)
 193                 RETURN(lmm_size);
 194
 195         if (*lmmp && !lsm) {
 196                 stripe_count = le16_to_cpu((*lmmp)->lmm_stripe_count);
 197                 lmm_size = lov_mds_md_size(stripe_count, lmm_magic);
 198                 OBD_FREE_LARGE(*lmmp, lmm_size);
 199                 *lmmp = NULL;
 200                 RETURN(0);
 201         }
 202
 203         if (!*lmmp) {
 204                 OBD_ALLOC_LARGE(*lmmp, lmm_size);
 205                 if (!*lmmp)
 206                         RETURN(-ENOMEM);
 207         }
 208
 209         CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n",
 210                lmm_magic, lmm_size);
 211
 212         lmmv1 = *lmmp;
 213         lmmv3 = (struct lov_mds_md_v3 *)*lmmp;
 214         if (lmm_magic == LOV_MAGIC_V3)
 215                 lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
 216         else
 217                 lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
 218
 219         if (!lsm)
 220                 RETURN(lmm_size);
 221
 222         /* lmmv1 and lmmv3 point to the same struct and have the
 223          * same first fields
 224          */
 225         ostid_cpu_to_le(&lsm->lsm_oi, &lmmv1->lmm_oi);
 226         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 227         lmmv1->lmm_stripe_count = cpu_to_le16(stripe_count);
 228         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 229         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 230         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 231                 cplen = strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 232                                 sizeof(lmmv3->lmm_pool_name));
 233                 if (cplen >= sizeof(lmmv3->lmm_pool_name))
 234                         RETURN(-E2BIG);
 235                 lmm_objects = lmmv3->lmm_objects;
 236         } else {
 237                 lmm_objects = lmmv1->lmm_objects;
 238         }
 239
 240         for (i = 0; i < stripe_count; i++) {
 241                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 242                 /* XXX LOV STACKING call down to osc_packmd() to do packing */
 243                 LASSERTF(loi->loi_id != 0, "lmm_oi "DOSTID" stripe %u/%u"
 244                          " idx %u\n", POSTID(&lmmv1->lmm_oi), i, stripe_count,
 245                          loi->loi_ost_idx);
 246                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 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 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 256 {
 257         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 258
 259         if (!stripe_count)
 260                 stripe_count = lov->desc.ld_default_stripe_count;
 261         if (stripe_count > lov->desc.ld_active_tgt_count)
 262                 stripe_count = lov->desc.ld_active_tgt_count;
 263         if (!stripe_count)
 264                 stripe_count = 1;
 265
 266         /* stripe count is based on whether ldiskfs can handle
 267          * larger EA sizes */
 268         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 269             lov->lov_ocd.ocd_max_easize)
 270                 max_stripes = lov_mds_md_stripecnt(lov->lov_ocd.ocd_max_easize,
 271                                                    magic);
 272
 273         if (stripe_count > max_stripes)
 274                 stripe_count = max_stripes;
 275
 276         return stripe_count;
 277 }
 278
 279
 280 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
 281 {
 282         int rc;
 283
 284         if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
 285                 char *buffer;
 286                 int sz;
 287
 288                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 289                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 290                 sz = lmm_bytes * 2 + 1;
 291                 OBD_ALLOC_LARGE(buffer, sz);
 292                 if (buffer != NULL) {
 293                         int i;
 294
 295                         for (i = 0; i < lmm_bytes; i++)
 296                                 sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]);
 297                         buffer[sz - 1] = '\0';
 298                         CERROR("%s\n", buffer);
 299                         OBD_FREE_LARGE(buffer, sz);
 300                 }
 301                 return -EINVAL;
 302         }
 303         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 304                                      lmm_bytes, stripe_count);
 305         return rc;
 306 }
 307
 308 int lov_alloc_memmd(struct lov_stripe_md **lsmp, __u16 stripe_count,
 309                     int pattern, int magic)
 310 {
 311         int i, lsm_size;
 312         ENTRY;
 313
 314         CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count);
 315
 316         *lsmp = lsm_alloc_plain(stripe_count, &lsm_size);
 317         if (!*lsmp) {
 318                 CERROR("can't allocate lsmp stripe_count %d\n", stripe_count);
 319                 RETURN(-ENOMEM);
 320         }
 321
 322         cfs_atomic_set(&(*lsmp)->lsm_refc, 1);
 323         spin_lock_init(&(*lsmp)->lsm_lock);
 324         (*lsmp)->lsm_magic = magic;
 325         (*lsmp)->lsm_stripe_count = stripe_count;
 326         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
 327         (*lsmp)->lsm_pattern = pattern;
 328         (*lsmp)->lsm_pool_name[0] = '\0';
 329         (*lsmp)->lsm_layout_gen = 0;
 330         (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0;
 331
 332         for (i = 0; i < stripe_count; i++)
 333                 loi_init((*lsmp)->lsm_oinfo[i]);
 334
 335         RETURN(lsm_size);
 336 }
 337
 338 int lov_free_memmd(struct lov_stripe_md **lsmp)
 339 {
 340         struct lov_stripe_md *lsm = *lsmp;
 341         int refc;
 342
 343         *lsmp = NULL;
 344         LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 0);
 345         if ((refc = cfs_atomic_dec_return(&lsm->lsm_refc)) == 0) {
 346                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
 347                 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 348         }
 349         return refc;
 350 }
 351
 352
 353 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 354  * order and is opaque to the networking layer.
 355  */
 356 int lov_unpackmd(struct obd_export *exp,  struct lov_stripe_md **lsmp,
 357                  struct lov_mds_md *lmm, int lmm_bytes)
 358 {
 359         struct obd_device *obd = class_exp2obd(exp);
 360         struct lov_obd *lov = &obd->u.lov;
 361         int rc = 0, lsm_size;
 362         __u16 stripe_count;
 363         __u32 magic;
 364         ENTRY;
 365
 366         /* If passed an MDS struct use values from there, otherwise defaults */
 367         if (lmm) {
 368                 rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count);
 369                 if (rc)
 370                         RETURN(rc);
 371                 magic = le32_to_cpu(lmm->lmm_magic);
 372         } else {
 373                 magic = LOV_MAGIC;
 374                 stripe_count = lov_get_stripecnt(lov, magic, 0);
 375         }
 376
 377         /* If we aren't passed an lsmp struct, we just want the size */
 378         if (!lsmp) {
 379                 /* XXX LOV STACKING call into osc for sizes */
 380                 LBUG();
 381                 RETURN(lov_stripe_md_size(stripe_count));
 382         }
 383         /* If we are passed an allocated struct but nothing to unpack, free */
 384         if (*lsmp && !lmm) {
 385                 lov_free_memmd(lsmp);
 386                 RETURN(0);
 387         }
 388
 389         lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
 390                                    magic);
 391         if (lsm_size < 0)
 392                 RETURN(lsm_size);
 393
 394         /* If we are passed a pointer but nothing to unpack, we only alloc */
 395         if (!lmm)
 396                 RETURN(lsm_size);
 397
 398         LASSERT(lsm_op_find(magic) != NULL);
 399         rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm);
 400         if (rc) {
 401                 lov_free_memmd(lsmp);
 402                 RETURN(rc);
 403         }
 404
 405         RETURN(lsm_size);
 406 }
 407
 408 static int __lov_setstripe(struct obd_export *exp, int max_lmm_size,
 409                            struct lov_stripe_md **lsmp,
 410                            struct lov_user_md *lump)
 411 {
 412         struct obd_device *obd = class_exp2obd(exp);
 413         struct lov_obd *lov = &obd->u.lov;
 414         char buffer[sizeof(struct lov_user_md_v3)];
 415         struct lov_user_md_v3 *lumv3 = (struct lov_user_md_v3 *)&buffer[0];
 416         struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&buffer[0];
 417         int lmm_magic;
 418         __u16 stripe_count;
 419         int rc;
 420         int cplen = 0;
 421         ENTRY;
 422
 423         rc = lov_lum_swab_if_needed(lumv3, &lmm_magic, lump);
 424         if (rc)
 425                 RETURN(rc);
 426
 427         /* in the rest of the tests, as *lumv1 and lumv3 have the same
 428          * fields, we use lumv1 to avoid code duplication */
 429
 430         if (lumv1->lmm_pattern == 0) {
 431                 lumv1->lmm_pattern = lov->desc.ld_pattern ?
 432                         lov->desc.ld_pattern : LOV_PATTERN_RAID0;
 433         }
 434
 435         if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) {
 436                 CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
 437                        lumv1->lmm_pattern);
 438                 RETURN(-EINVAL);
 439         }
 440
 441         /* 64kB is the largest common page size we see (ia64), and matches the
 442          * check in lfs */
 443         if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 444                 CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
 445                        lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
 446                 lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
 447         }
 448
 449         if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) &&
 450             (lumv1->lmm_stripe_offset !=
 451              (typeof(lumv1->lmm_stripe_offset))(-1))) {
 452                 CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n",
 453                        lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count);
 454                 RETURN(-EINVAL);
 455         }
 456         stripe_count = lov_get_stripecnt(lov, lmm_magic,
 457                                          lumv1->lmm_stripe_count);
 458
 459         if (max_lmm_size) {
 460                 int max_stripes = (max_lmm_size -
 461                                    lov_mds_md_size(0, lmm_magic)) /
 462                                    sizeof(struct lov_ost_data_v1);
 463                 if (unlikely(max_stripes < stripe_count)) {
 464                         CDEBUG(D_IOCTL, "stripe count reset from %d to %d\n",
 465                                stripe_count, max_stripes);
 466                         stripe_count = max_stripes;
 467                 }
 468         }
 469
 470         if (lmm_magic == LOV_USER_MAGIC_V3) {
 471                 struct pool_desc *pool;
 472
 473                 /* In the function below, .hs_keycmp resolves to
 474                  * pool_hashkey_keycmp() */
 475                 /* coverity[overrun-buffer-val] */
 476                 pool = lov_find_pool(lov, lumv3->lmm_pool_name);
 477                 if (pool != NULL) {
 478                         if (lumv3->lmm_stripe_offset !=
 479                             (typeof(lumv3->lmm_stripe_offset))(-1)) {
 480                                 rc = lov_check_index_in_pool(
 481                                         lumv3->lmm_stripe_offset, pool);
 482                                 if (rc < 0) {
 483                                         lov_pool_putref(pool);
 484                                         RETURN(-EINVAL);
 485                                 }
 486                         }
 487
 488                         if (stripe_count > pool_tgt_count(pool))
 489                                 stripe_count = pool_tgt_count(pool);
 490
 491                         lov_pool_putref(pool);
 492                 }
 493         }
 494
 495         rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic);
 496
 497         if (rc >= 0) {
 498                 (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset;
 499                 (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size;
 500                 if (lmm_magic == LOV_USER_MAGIC_V3) {
 501                         cplen = strlcpy((*lsmp)->lsm_pool_name,
 502                                         lumv3->lmm_pool_name,
 503                                         sizeof((*lsmp)->lsm_pool_name));
 504                         if (cplen >= sizeof((*lsmp)->lsm_pool_name))
 505                                 rc = -E2BIG;
 506                 }
 507                 rc = 0;
 508         }
 509
 510         RETURN(rc);
 511 }
 512
 513 /* Configure object striping information on a new file.
 514  *
 515  * @lmmu is a pointer to a user struct with one or more of the fields set to
 516  * indicate the application preference: lmm_stripe_count, lmm_stripe_size,
 517  * lmm_stripe_offset, and lmm_stripe_pattern.  lmm_magic must be LOV_MAGIC.
 518  * @lsmp is a pointer to an in-core stripe MD that needs to be filled in.
 519  */
 520 int lov_setstripe(struct obd_export *exp, int max_lmm_size,
 521                   struct lov_stripe_md **lsmp, struct lov_user_md *lump)
 522 {
 523         int rc;
 524         mm_segment_t seg;
 525
 526         seg = get_fs();
 527         set_fs(KERNEL_DS);
 528
 529         rc = __lov_setstripe(exp, max_lmm_size, lsmp, lump);
 530         set_fs(seg);
 531         RETURN(rc);
 532 }
 533
 534 int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp,
 535               struct lov_user_md *lump)
 536 {
 537         int i;
 538         int rc;
 539         struct obd_export *oexp;
 540         struct lov_obd *lov = &exp->exp_obd->u.lov;
 541         obd_id last_id = 0;
 542         struct lov_user_ost_data_v1 *lmm_objects;
 543
 544         ENTRY;
 545
 546         if (lump->lmm_magic == LOV_USER_MAGIC_V3)
 547                 lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects;
 548         else
 549                 lmm_objects = lump->lmm_objects;
 550
 551         for (i = 0; i < lump->lmm_stripe_count; i++) {
 552                 __u32 len = sizeof(last_id);
 553                 oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp;
 554                 rc = obd_get_info(NULL, oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
 555                                   &len, &last_id, NULL);
 556                 if (rc)
 557                         RETURN(rc);
 558                 if (lmm_objects[i].l_object_id > last_id) {
 559                         CERROR("Setting EA for object > than last id on "
 560                                "ost idx %d "LPD64" > "LPD64" \n",
 561                                lmm_objects[i].l_ost_idx,
 562                                lmm_objects[i].l_object_id, last_id);
 563                         RETURN(-EINVAL);
 564                 }
 565         }
 566
 567         rc = lov_setstripe(exp, 0, lsmp, lump);
 568         if (rc)
 569                 RETURN(rc);
 570
 571         for (i = 0; i < lump->lmm_stripe_count; i++) {
 572                 (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
 573                         lmm_objects[i].l_ost_idx;
 574                 (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id;
 575                 (*lsmp)->lsm_oinfo[i]->loi_seq = lmm_objects[i].l_object_seq;
 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 }