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 <lustre/lustre_idl.h>
  42 #include <lustre/lustre_user.h>
  43
  44 #include <lustre_net.h>
  45 #include <lustre_swab.h>
  46 #include <obd.h>
  47 #include <obd_class.h>
  48 #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 void lov_dump_lmm(int level, void *lmm)
 104 {
 105         int magic;
 106
 107         magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
 108         switch (magic) {
 109         case LOV_MAGIC_V1:
 110                 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
 111                 break;
 112         case LOV_MAGIC_V3:
 113                 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
 114                 break;
 115         default:
 116                 CDEBUG(level, "unrecognized lmm_magic %x, assuming %x\n",
 117                        magic, LOV_MAGIC_V1);
 118                 lov_dump_lmm_common(level, lmm);
 119                 break;
 120         }
 121 }
 122
 123 /**
 124  * Pack LOV striping metadata for disk storage format (in little
 125  * endian byte order).
 126  *
 127  * This follows the getxattr() conventions. If \a buf_size is zero
 128  * then return the size needed. If \a buf_size is too small then
 129  * return -ERANGE. Otherwise return the size of the result.
 130  */
 131 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
 132                      size_t buf_size)
 133 {
 134         struct lov_mds_md_v1 *lmmv1 = buf;
 135         struct lov_mds_md_v3 *lmmv3 = buf;
 136         struct lov_ost_data_v1 *lmm_objects;
 137         size_t lmm_size;
 138         unsigned int i;
 139         ENTRY;
 140
 141         lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
 142                                    lsm->lsm_magic);
 143         if (buf_size == 0)
 144                 RETURN(lmm_size);
 145
 146         if (buf_size < lmm_size)
 147                 RETURN(-ERANGE);
 148
 149         /* lmmv1 and lmmv3 point to the same struct and have the
 150          * same first fields
 151          */
 152         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
 153         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 154         lmmv1->lmm_stripe_size = cpu_to_le32(
 155                                 lsm->lsm_entries[0]->lsme_stripe_size);
 156         lmmv1->lmm_stripe_count = cpu_to_le16(
 157                                 lsm->lsm_entries[0]->lsme_stripe_count);
 158         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
 159         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 160
 161         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 162                 CLASSERT(sizeof(lsm->lsm_entries[0]->lsme_pool_name) ==
 163                          sizeof(lmmv3->lmm_pool_name));
 164                 strlcpy(lmmv3->lmm_pool_name,
 165                         lsm->lsm_entries[0]->lsme_pool_name,
 166                         sizeof(lmmv3->lmm_pool_name));
 167                 lmm_objects = lmmv3->lmm_objects;
 168         } else {
 169                 lmm_objects = lmmv1->lmm_objects;
 170         }
 171
 172         for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
 173                 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
 174
 175                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 176                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 177                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 178         }
 179
 180         RETURN(lmm_size);
 181 }
 182
 183 /* Find the max stripecount we should use */
 184 __u16 lov_get_stripecnt(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 185 {
 186         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 187
 188         if (!stripe_count)
 189                 stripe_count = lov->desc.ld_default_stripe_count;
 190         if (stripe_count > lov->desc.ld_active_tgt_count)
 191                 stripe_count = lov->desc.ld_active_tgt_count;
 192         if (!stripe_count)
 193                 stripe_count = 1;
 194
 195         /* stripe count is based on whether ldiskfs can handle
 196          * larger EA sizes */
 197         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 198             lov->lov_ocd.ocd_max_easize)
 199                 max_stripes = lov_mds_md_max_stripe_count(
 200                         lov->lov_ocd.ocd_max_easize, magic);
 201
 202         if (stripe_count > max_stripes)
 203                 stripe_count = max_stripes;
 204
 205         return stripe_count;
 206 }
 207
 208 int lov_free_memmd(struct lov_stripe_md **lsmp)
 209 {
 210         struct lov_stripe_md *lsm = *lsmp;
 211         int refc;
 212
 213         *lsmp = NULL;
 214         refc = atomic_dec_return(&lsm->lsm_refc);
 215         LASSERT(refc >= 0);
 216         if (refc == 0)
 217                 lsm_free(lsm);
 218
 219         return refc;
 220 }
 221
 222 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
 223  * order and is opaque to the networking layer.
 224  */
 225 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
 226                                    size_t buf_size)
 227 {
 228         const struct lsm_operations *op;
 229         struct lov_stripe_md *lsm;
 230         u32 magic;
 231         ENTRY;
 232
 233         if (buf_size < sizeof(magic))
 234                 RETURN(ERR_PTR(-EINVAL));
 235
 236         magic = le32_to_cpu(*(u32 *)buf);
 237         op = lsm_op_find(magic);
 238         if (op == NULL)
 239                 RETURN(ERR_PTR(-EINVAL));
 240
 241         lsm = lsm_op_find(magic)->lsm_unpackmd(lov, buf, buf_size);
 242
 243         RETURN(lsm);
 244 }
 245
 246 /* Retrieve object striping information.
 247  *
 248  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 249  * the maximum number of OST indices which will fit in the user buffer.
 250  * lmm_magic must be LOV_USER_MAGIC.
 251  */
 252 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
 253                   struct lov_user_md __user *lump)
 254 {
 255         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 256         struct lov_mds_md       *lmmk;
 257         struct lov_user_md_v3   lum;
 258         u32                     stripe_count;
 259         size_t                  lum_size;
 260         size_t                  lmmk_size;
 261         ssize_t                 lmm_size;
 262         int                     rc;
 263         ENTRY;
 264
 265         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
 266                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 267                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 268                 GOTO(out, rc = -EIO);
 269         }
 270
 271         if (!lsm->lsm_is_released)
 272                 stripe_count = lsm->lsm_entries[0]->lsme_stripe_count;
 273         else
 274                 stripe_count = 0;
 275
 276         /* we only need the header part from user space to get lmm_magic and
 277          * lmm_stripe_count, (the header part is common to v1 and v3) */
 278         lum_size = sizeof(struct lov_user_md_v1);
 279         if (copy_from_user(&lum, lump, lum_size))
 280                 GOTO(out, rc = -EFAULT);
 281
 282         if (lum.lmm_magic != LOV_USER_MAGIC_V1 &&
 283             lum.lmm_magic != LOV_USER_MAGIC_V3 &&
 284             lum.lmm_magic != LOV_USER_MAGIC_SPECIFIC)
 285                 GOTO(out, rc = -EINVAL);
 286
 287         if (lum.lmm_stripe_count != 0 && lum.lmm_stripe_count < stripe_count) {
 288                 /* Return right size of stripe to user */
 289                 lum.lmm_stripe_count = stripe_count;
 290                 rc = copy_to_user(lump, &lum, lum_size);
 291                 GOTO(out, rc = -EOVERFLOW);
 292         }
 293
 294         lmmk_size = lov_mds_md_size(stripe_count, lsm->lsm_magic);
 295
 296         OBD_ALLOC_LARGE(lmmk, lmmk_size);
 297         if (lmmk == NULL)
 298                 GOTO(out, rc = -ENOMEM);
 299
 300         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
 301         if (lmm_size < 0)
 302                 GOTO(out_free, rc = lmm_size);
 303
 304         /* FIXME: Bug 1185 - copy fields properly when structs change */
 305         /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */
 306         CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3));
 307         CLASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lmmk->lmm_objects[0]));
 308
 309         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC &&
 310             (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
 311              lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3))) {
 312                 lustre_swab_lov_mds_md(lmmk);
 313                 lustre_swab_lov_user_md_objects(
 314                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
 315                                 lmmk->lmm_stripe_count);
 316         }
 317
 318         if (lum.lmm_magic == LOV_USER_MAGIC) {
 319                 /* User request for v1, we need skip lmm_pool_name */
 320                 if (lmmk->lmm_magic == LOV_MAGIC_V3) {
 321                         memmove(((struct lov_mds_md_v1 *)lmmk)->lmm_objects,
 322                                 ((struct lov_mds_md_v3 *)lmmk)->lmm_objects,
 323                                 lmmk->lmm_stripe_count *
 324                                 sizeof(struct lov_ost_data_v1));
 325                         lmm_size -= LOV_MAXPOOLNAME;
 326                 }
 327         } else {
 328                 /* if v3 we just have to update the lum_size */
 329                 lum_size = sizeof(struct lov_user_md_v3);
 330         }
 331
 332         /* User wasn't expecting this many OST entries */
 333         if (lum.lmm_stripe_count == 0)
 334                 lmm_size = lum_size;
 335         else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count)
 336                 GOTO(out_free, rc = -EOVERFLOW);
 337         /*
 338          * Have a difference between lov_mds_md & lov_user_md.
 339          * So we have to re-order the data before copy to user.
 340          */
 341         lum.lmm_stripe_count = lmmk->lmm_stripe_count;
 342         lum.lmm_layout_gen = lmmk->lmm_layout_gen;
 343         ((struct lov_user_md *)lmmk)->lmm_layout_gen = lum.lmm_layout_gen;
 344         ((struct lov_user_md *)lmmk)->lmm_stripe_count = lum.lmm_stripe_count;
 345         if (copy_to_user(lump, lmmk, lmm_size))
 346                 GOTO(out_free, rc = -EFAULT);
 347
 348         GOTO(out_free, rc = 0);
 349 out_free:
 350         OBD_FREE_LARGE(lmmk, lmmk_size);
 351 out:
 352         return rc;
 353 }