drivers/staging/lustre/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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/lov/lov_pack.c
  33  *
  34  * (Un)packing of OST/MDS requests
  35  *
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_LOV
  40
  41 #include "../include/lustre/lustre_idl.h"
  42 #include "../include/lustre/lustre_user.h"
  43
  44 #include "../include/lustre_net.h"
  45 #include "../include/lustre_swab.h"
  46 #include "../include/obd.h"
  47 #include "../include/obd_class.h"
  48 #include "../include/obd_support.h"
  49
  50 #include "lov_cl_internal.h"
  51 #include "lov_internal.h"
  52
  53 void lov_dump_lmm_common(int level, void *lmmp)
  54 {
  55         struct lov_mds_md *lmm = lmmp;
  56         struct ost_id   oi;
  57
  58         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
  59         CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
  60                POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
  61                le32_to_cpu(lmm->lmm_pattern));
  62         CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
  63                le32_to_cpu(lmm->lmm_stripe_size),
  64                le16_to_cpu(lmm->lmm_stripe_count),
  65                le16_to_cpu(lmm->lmm_layout_gen));
  66 }
  67
  68 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  69                                  int stripe_count)
  70 {
  71         int i;
  72
  73         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  74                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
  75                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  76                 return;
  77         }
  78
  79         for (i = 0; i < stripe_count; ++i, ++lod) {
  80                 struct ost_id   oi;
  81
  82                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  83                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  84                        le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  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                              le16_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                              le16_to_cpu(lmm->lmm_stripe_count));
 101 }
 102
 103 /**
 104  * Pack LOV striping metadata for disk storage format (in little
 105  * endian byte order).
 106  *
 107  * This follows the getxattr() conventions. If \a buf_size is zero
 108  * then return the size needed. If \a buf_size is too small then
 109  * return -ERANGE. Otherwise return the size of the result.
 110  */
 111 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
 112                      size_t buf_size)
 113 {
 114         struct lov_ost_data_v1 *lmm_objects;
 115         struct lov_mds_md_v1 *lmmv1 = buf;
 116         struct lov_mds_md_v3 *lmmv3 = buf;
 117         size_t lmm_size;
 118         unsigned int i;
 119
 120         lmm_size = lov_mds_md_size(lsm->lsm_stripe_count, lsm->lsm_magic);
 121         if (!buf_size)
 122                 return lmm_size;
 123
 124         if (buf_size < lmm_size)
 125                 return -ERANGE;
 126
 127         /*
 128          * lmmv1 and lmmv3 point to the same struct and have the
 129          * same first fields
 130          */
 131         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
 132         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 133         lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
 134         lmmv1->lmm_stripe_count = cpu_to_le16(lsm->lsm_stripe_count);
 135         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
 136         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 137
 138         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 139                 BUILD_BUG_ON(sizeof(lsm->lsm_pool_name) !=
 140                          sizeof(lmmv3->lmm_pool_name));
 141                 strlcpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name,
 142                         sizeof(lmmv3->lmm_pool_name));
 143                 lmm_objects = lmmv3->lmm_objects;
 144         } else {
 145                 lmm_objects = lmmv1->lmm_objects;
 146         }
 147
 148         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 149                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 150
 151                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 152                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 153                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 154         }
 155
 156         return lmm_size;
 157 }
 158
 159 /* Find the max stripecount we should use */
 160 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 161 {
 162         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 163
 164         if (!stripe_count)
 165                 stripe_count = lov->desc.ld_default_stripe_count;
 166         if (stripe_count > lov->desc.ld_active_tgt_count)
 167                 stripe_count = lov->desc.ld_active_tgt_count;
 168         if (!stripe_count)
 169                 stripe_count = 1;
 170
 171         /* stripe count is based on whether ldiskfs can handle
 172          * larger EA sizes
 173          */
 174         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 175             lov->lov_ocd.ocd_max_easize)
 176                 max_stripes = lov_mds_md_max_stripe_count(
 177                         lov->lov_ocd.ocd_max_easize, magic);
 178
 179         if (stripe_count > max_stripes)
 180                 stripe_count = max_stripes;
 181
 182         return stripe_count;
 183 }
 184
 185 static int lov_verify_lmm(void *lmm, int lmm_bytes, __u16 *stripe_count)
 186 {
 187         int rc;
 188
 189         if (!lsm_op_find(le32_to_cpu(*(__u32 *)lmm))) {
 190                 CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n",
 191                        le32_to_cpu(*(__u32 *)lmm), lmm_bytes);
 192                 CERROR("%*phN\n", lmm_bytes, lmm);
 193                 return -EINVAL;
 194         }
 195         rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm,
 196                                                                      lmm_bytes,
 197                                                                   stripe_count);
 198         return rc;
 199 }
 200
 201 static struct lov_stripe_md *lov_lsm_alloc(u16 stripe_count, u32 pattern,
 202                                            u32 magic)
 203 {
 204         struct lov_stripe_md *lsm;
 205         unsigned int i;
 206
 207         CDEBUG(D_INFO, "alloc lsm, stripe_count %u\n", stripe_count);
 208
 209         lsm = lsm_alloc_plain(stripe_count);
 210         if (!lsm) {
 211                 CERROR("cannot allocate LSM stripe_count %u\n", stripe_count);
 212                 return ERR_PTR(-ENOMEM);
 213         }
 214
 215         atomic_set(&lsm->lsm_refc, 1);
 216         spin_lock_init(&lsm->lsm_lock);
 217         lsm->lsm_magic = magic;
 218         lsm->lsm_stripe_count = stripe_count;
 219         lsm->lsm_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES * stripe_count;
 220         lsm->lsm_pattern = pattern;
 221         lsm->lsm_pool_name[0] = '\0';
 222         lsm->lsm_layout_gen = 0;
 223         if (stripe_count > 0)
 224                 lsm->lsm_oinfo[0]->loi_ost_idx = ~0;
 225
 226         for (i = 0; i < stripe_count; i++)
 227                 loi_init(lsm->lsm_oinfo[i]);
 228
 229         return lsm;
 230 }
 231
 232 int lov_free_memmd(struct lov_stripe_md **lsmp)
 233 {
 234         struct lov_stripe_md *lsm = *lsmp;
 235         int refc;
 236
 237         *lsmp = NULL;
 238         LASSERT(atomic_read(&lsm->lsm_refc) > 0);
 239         refc = atomic_dec_return(&lsm->lsm_refc);
 240         if (refc == 0)
 241                 lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
 242
 243         return refc;
 244 }
 245
 246 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 247  * order and is opaque to the networking layer.
 248  */
 249 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, struct lov_mds_md *lmm,
 250                                    size_t lmm_size)
 251 {
 252         struct lov_stripe_md *lsm;
 253         u16 stripe_count;
 254         u32 pattern;
 255         u32 magic;
 256         int rc;
 257
 258         rc = lov_verify_lmm(lmm, lmm_size, &stripe_count);
 259         if (rc)
 260                 return ERR_PTR(rc);
 261
 262         magic = le32_to_cpu(lmm->lmm_magic);
 263         pattern = le32_to_cpu(lmm->lmm_pattern);
 264
 265         lsm = lov_lsm_alloc(stripe_count, pattern, magic);
 266         if (IS_ERR(lsm))
 267                 return lsm;
 268
 269         LASSERT(lsm_op_find(magic));
 270         rc = lsm_op_find(magic)->lsm_unpackmd(lov, lsm, lmm);
 271         if (rc) {
 272                 lov_free_memmd(&lsm);
 273                 return ERR_PTR(rc);
 274         }
 275
 276         return lsm;
 277 }
 278
 279 /* Retrieve object striping information.
 280  *
 281  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 282  * the maximum number of OST indices which will fit in the user buffer.
 283  * lmm_magic must be LOV_USER_MAGIC.
 284  */
 285 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
 286                   struct lov_user_md __user *lump)
 287 {
 288         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 289         struct lov_user_md_v3 lum;
 290         struct lov_mds_md *lmmk;
 291         u32 stripe_count;
 292         ssize_t lmm_size;
 293         size_t lmmk_size;
 294         size_t lum_size;
 295         int rc;
 296
 297         if (!lsm)
 298                 return -ENODATA;
 299
 300         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
 301                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 302                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 303                 rc = -EIO;
 304                 goto out;
 305         }
 306
 307         if (!lsm_is_released(lsm))
 308                 stripe_count = lsm->lsm_stripe_count;
 309         else
 310                 stripe_count = 0;
 311
 312         /* we only need the header part from user space to get lmm_magic and
 313          * lmm_stripe_count, (the header part is common to v1 and v3)
 314          */
 315         lum_size = sizeof(struct lov_user_md_v1);
 316         if (copy_from_user(&lum, lump, lum_size)) {
 317                 rc = -EFAULT;
 318                 goto out;
 319         }
 320         if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
 321             lum.lmm_magic != LOV_USER_MAGIC_V3 &&
 322             lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC) {
 323                 rc = -EINVAL;
 324                 goto out;
 325         }
 326
 327         if (lum.lmm_stripe_count &&
 328             (lum.lmm_stripe_count < lsm->lsm_stripe_count)) {
 329                 /* Return right size of stripe to user */
 330                 lum.lmm_stripe_count = stripe_count;
 331                 rc = copy_to_user(lump, &lum, lum_size);
 332                 rc = -EOVERFLOW;
 333                 goto out;
 334         }
 335         lmmk_size = lov_mds_md_size(stripe_count, lsm->lsm_magic);
 336
 337
 338         lmmk = libcfs_kvzalloc(lmmk_size, GFP_NOFS);
 339         if (!lmmk) {
 340                 rc = -ENOMEM;
 341                 goto out;
 342         }
 343
 344         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
 345         if (lmm_size < 0) {
 346                 rc = lmm_size;
 347                 goto out_free;
 348         }
 349
 350         /* FIXME: Bug 1185 - copy fields properly when structs change */
 351         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 352         BUILD_BUG_ON(sizeof(lum) != sizeof(struct lov_mds_md_v3));
 353         BUILD_BUG_ON(sizeof(lum.lmm_objects[0]) != sizeof(lmmk->lmm_objects[0]));
 354
 355         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC &&
 356             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
 357              lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))) {
 358                 lustre_swab_lov_mds_md(lmmk);
 359                 lustre_swab_lov_user_md_objects(
 360                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
 361                                 lmmk->lmm_stripe_count);
 362         }
 363
 364         if (lum.lmm_magic == LOV_USER_MAGIC) {
 365                 /* User request for v1, we need skip lmm_pool_name */
 366                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 367                         memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
 368                                 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
 369                                 lmmk->lmm_stripe_count *
 370                                 sizeof(struct lov_ost_data_v1));
 371                         lmm_size -= LOV_MAXPOOLNAME;
 372                 }
 373         } else {
 374                 /* if v3 we just have to update the lum_size */
 375                 lum_size = sizeof(struct lov_user_md_v3);
 376         }
 377
 378         /* User wasn't expecting this many OST entries */
 379         if (lum.lmm_stripe_count == 0) {
 380                 lmm_size = lum_size;
 381         } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
 382                 rc = -EOVERFLOW;
 383                 goto out_free;
 384         }
 385         /*
 386          * Have a difference between lov_mds_md & lov_user_md.
 387          * So we have to re-order the data before copy to user.
 388          */
 389         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 390         lum.lmm_layout_gen = lmmk->lmm_layout_gen;
 391         ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
 392         ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 393         if (copy_to_user(lump, lmmk, lmm_size))
 394                 rc = -EFAULT;
 395         else
 396                 rc = 0;
 397
 398 out_free:
 399         kvfree(lmmk);
 400 out:
 401         return rc;
 402 }