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, 2017, 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_net.h>
  42 #include <lustre_swab.h>
  43 #include <obd.h>
  44 #include <obd_class.h>
  45 #include <obd_support.h>
  46
  47 #include "lov_cl_internal.h"
  48 #include "lov_internal.h"
  49
  50 void lov_dump_lmm_common(int level, void *lmmp)
  51 {
  52         struct lov_mds_md *lmm = lmmp;
  53         struct ost_id oi;
  54
  55         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
  56         CDEBUG_LIMIT(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
  57                      POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
  58                      le32_to_cpu(lmm->lmm_pattern));
  59         CDEBUG_LIMIT(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
  60                      le32_to_cpu(lmm->lmm_stripe_size),
  61                      le16_to_cpu(lmm->lmm_stripe_count),
  62                      le16_to_cpu(lmm->lmm_layout_gen));
  63 }
  64
  65 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  66                                  int stripe_count)
  67 {
  68         int i;
  69
  70         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  71                 CDEBUG_LIMIT(level,
  72                              "bad stripe_count %u > max_stripe_count %u\n",
  73                              stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
  74                 return;
  75         }
  76
  77         for (i = 0; i < stripe_count; ++i, ++lod) {
  78                 struct ost_id oi;
  79
  80                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  81                 CDEBUG_LIMIT(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  82                              le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  83         }
  84 }
  85
  86 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  87 {
  88         lov_dump_lmm_common(level, lmm);
  89         lov_dump_lmm_objects(level, lmm->lmm_objects,
  90                              le16_to_cpu(lmm->lmm_stripe_count));
  91 }
  92
  93 /**
  94  * Pack LOV striping metadata for disk storage format (in little
  95  * endian byte order).
  96  *
  97  * This follows the getxattr() conventions. If \a buf_size is zero
  98  * then return the size needed. If \a buf_size is too small then
  99  * return -ERANGE. Otherwise return the size of the result.
 100  */
 101 static ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
 102                                  size_t buf_size)
 103 {
 104         struct lov_mds_md_v1 *lmmv1 = buf;
 105         struct lov_mds_md_v3 *lmmv3 = buf;
 106         struct lov_ost_data_v1 *lmm_objects;
 107         size_t lmm_size;
 108         unsigned int i;
 109
 110         ENTRY;
 111
 112         lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
 113                                    lsm->lsm_magic);
 114         if (buf_size == 0)
 115                 RETURN(lmm_size);
 116
 117         if (buf_size < lmm_size)
 118                 RETURN(-ERANGE);
 119
 120         /*
 121          * lmmv1 and lmmv3 point to the same struct and have the
 122          * same first fields
 123          */
 124         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
 125         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 126         lmmv1->lmm_stripe_size = cpu_to_le32(
 127                                 lsm->lsm_entries[0]->lsme_stripe_size);
 128         lmmv1->lmm_stripe_count = cpu_to_le16(
 129                                 lsm->lsm_entries[0]->lsme_stripe_count);
 130         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
 131         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 132
 133         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 134                 BUILD_BUG_ON(sizeof(lsm->lsm_entries[0]->lsme_pool_name) !=
 135                                     sizeof(lmmv3->lmm_pool_name));
 136                 strlcpy(lmmv3->lmm_pool_name,
 137                         lsm->lsm_entries[0]->lsme_pool_name,
 138                         sizeof(lmmv3->lmm_pool_name));
 139                 lmm_objects = lmmv3->lmm_objects;
 140         } else {
 141                 lmm_objects = lmmv1->lmm_objects;
 142         }
 143
 144         if (lsm->lsm_is_released)
 145                 RETURN(lmm_size);
 146
 147         for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
 148                 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
 149
 150                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 151                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 152                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 153         }
 154
 155         RETURN(lmm_size);
 156 }
 157
 158 static ssize_t lov_lsm_pack_foreign(const struct lov_stripe_md *lsm, void *buf,
 159                                     size_t buf_size)
 160 {
 161         struct lov_foreign_md *lfm = buf;
 162         size_t lfm_size;
 163
 164         lfm_size = lsm->lsm_foreign_size;
 165
 166         if (buf_size == 0)
 167                 RETURN(lfm_size);
 168
 169         if (buf_size < lfm_size)
 170                 RETURN(-ERANGE);
 171
 172         /* full foreign LOV is already avail in its cache
 173          * no need to translate format fields to little-endian
 174          */
 175         memcpy(lfm, lsm_foreign(lsm), lsm->lsm_foreign_size);
 176
 177         RETURN(lfm_size);
 178 }
 179
 180 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
 181                      size_t buf_size)
 182 {
 183         struct lov_comp_md_v1 *lcmv1 = buf;
 184         struct lov_comp_md_entry_v1 *lcme;
 185         struct lov_ost_data_v1 *lmm_objects;
 186         size_t lmm_size;
 187         unsigned int entry;
 188         unsigned int offset;
 189         unsigned int size;
 190         unsigned int i;
 191
 192         ENTRY;
 193
 194         if (lsm->lsm_magic == LOV_MAGIC_V1 || lsm->lsm_magic == LOV_MAGIC_V3)
 195                 return lov_lsm_pack_v1v3(lsm, buf, buf_size);
 196
 197         if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
 198                 return lov_lsm_pack_foreign(lsm, buf, buf_size);
 199
 200         lmm_size = lov_comp_md_size(lsm);
 201         if (buf_size == 0)
 202                 RETURN(lmm_size);
 203
 204         if (buf_size < lmm_size)
 205                 RETURN(-ERANGE);
 206
 207         lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
 208         lcmv1->lcm_size = cpu_to_le32(lmm_size);
 209         lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
 210         lcmv1->lcm_flags = cpu_to_le16(lsm->lsm_flags);
 211         lcmv1->lcm_mirror_count = cpu_to_le16(lsm->lsm_mirror_count);
 212         lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
 213
 214         offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
 215
 216         for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
 217                 struct lov_stripe_md_entry *lsme;
 218                 struct lov_mds_md *lmm;
 219                 __u16 stripe_count;
 220
 221                 lsme = lsm->lsm_entries[entry];
 222                 lcme = &lcmv1->lcm_entries[entry];
 223
 224                 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
 225                 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
 226                 if (lsme->lsme_flags & LCME_FL_NOSYNC)
 227                         lcme->lcme_timestamp =
 228                                 cpu_to_le64(lsme->lsme_timestamp);
 229                 lcme->lcme_extent.e_start =
 230                         cpu_to_le64(lsme->lsme_extent.e_start);
 231                 lcme->lcme_extent.e_end =
 232                         cpu_to_le64(lsme->lsme_extent.e_end);
 233                 lcme->lcme_offset = cpu_to_le32(offset);
 234
 235                 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
 236                 lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
 237                 /* lmm->lmm_oi not set */
 238                 lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
 239                 lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
 240                 lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
 241                 lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
 242
 243                 if (lsme->lsme_magic == LOV_MAGIC_V3) {
 244                         struct lov_mds_md_v3 *lmmv3 =
 245                                                 (struct lov_mds_md_v3 *)lmm;
 246
 247                         strlcpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
 248                                 sizeof(lmmv3->lmm_pool_name));
 249                         lmm_objects = lmmv3->lmm_objects;
 250                 } else {
 251                         lmm_objects =
 252                                 ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
 253                 }
 254
 255                 if (lsme_inited(lsme) &&
 256                     !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
 257                         stripe_count = lsme->lsme_stripe_count;
 258                 else
 259                         stripe_count = 0;
 260
 261                 for (i = 0; i < stripe_count; i++) {
 262                         struct lov_oinfo *loi = lsme->lsme_oinfo[i];
 263
 264                         ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 265                         lmm_objects[i].l_ost_gen =
 266                                         cpu_to_le32(loi->loi_ost_gen);
 267                         lmm_objects[i].l_ost_idx =
 268                                         cpu_to_le32(loi->loi_ost_idx);
 269                 }
 270
 271                 size = lov_mds_md_size(stripe_count, lsme->lsme_magic);
 272                 lcme->lcme_size = cpu_to_le32(size);
 273                 offset += size;
 274         } /* for each layout component */
 275
 276         RETURN(lmm_size);
 277 }
 278
 279 /* Find the max stripecount we should use */
 280 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 281 {
 282         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 283
 284         if (!stripe_count)
 285                 stripe_count = lov->desc.ld_default_stripe_count;
 286         if (stripe_count > lov->desc.ld_active_tgt_count)
 287                 stripe_count = lov->desc.ld_active_tgt_count;
 288         if (!stripe_count)
 289                 stripe_count = 1;
 290
 291         /*
 292          * stripe count is based on whether ldiskfs can handle
 293          * larger EA sizes
 294          */
 295         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 296             lov->lov_ocd.ocd_max_easize)
 297                 max_stripes = lov_mds_md_max_stripe_count(
 298                         lov->lov_ocd.ocd_max_easize, magic);
 299
 300         if (stripe_count > max_stripes)
 301                 stripe_count = max_stripes;
 302
 303         return stripe_count;
 304 }
 305
 306 int lov_free_memmd(struct lov_stripe_md **lsmp)
 307 {
 308         struct lov_stripe_md *lsm = *lsmp;
 309         int refc;
 310
 311         *lsmp = NULL;
 312         refc = atomic_dec_return(&lsm->lsm_refc);
 313         LASSERT(refc >= 0);
 314         if (refc == 0)
 315                 lsm_free(lsm);
 316
 317         return refc;
 318 }
 319
 320 /*
 321  * Unpack LOV object metadata from disk storage.  It is packed in LE byte
 322  * order and is opaque to the networking layer.
 323  */
 324 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
 325                                    size_t buf_size)
 326 {
 327         const struct lsm_operations *op;
 328         struct lov_stripe_md *lsm;
 329         u32 magic;
 330
 331         ENTRY;
 332
 333         if (buf_size < sizeof(magic))
 334                 RETURN(ERR_PTR(-EINVAL));
 335
 336         magic = le32_to_cpu(*(u32 *)buf);
 337         op = lsm_op_find(magic);
 338         if (!op)
 339                 RETURN(ERR_PTR(-EINVAL));
 340
 341         lsm = op->lsm_unpackmd(lov, buf, buf_size);
 342
 343         RETURN(lsm);
 344 }
 345
 346 /*
 347  * Retrieve object striping information.
 348  *
 349  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 350  * the maximum number of OST indices which will fit in the user buffer.
 351  * lmm_magic must be LOV_USER_MAGIC.
 352  *
 353  * If @size > 0, User specified limited buffer size, usually the buffer is from
 354  * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
 355  */
 356 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
 357                   struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
 358                   size_t size)
 359 {
 360         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 361         struct lov_mds_md *lmmk, *lmm;
 362         struct lov_foreign_md *lfm;
 363         struct lov_user_md_v1 lum;
 364         size_t lmmk_size, lum_size = 0;
 365         ssize_t lmm_size;
 366         static bool printed;
 367         int rc = 0;
 368
 369         ENTRY;
 370
 371         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
 372             lsm->lsm_magic != LOV_MAGIC_COMP_V1 &&
 373             lsm->lsm_magic != LOV_MAGIC_FOREIGN) {
 374                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 375                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 376                 GOTO(out, rc = -EIO);
 377         }
 378
 379         if (!printed) {
 380                 LCONSOLE_WARN("%s: using old ioctl(LL_IOC_LOV_GETSTRIPE) on "
 381                               DFID", use llapi_layout_get_by_path()\n",
 382                               current->comm,
 383                               PFID(&obj->lo_cl.co_lu.lo_header->loh_fid));
 384                 printed = true;
 385         }
 386
 387         lmmk_size = lov_comp_md_size(lsm);
 388
 389         OBD_ALLOC_LARGE(lmmk, lmmk_size);
 390         if (!lmmk)
 391                 GOTO(out, rc = -ENOMEM);
 392
 393         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
 394         if (lmm_size < 0)
 395                 GOTO(out_free, rc = lmm_size);
 396
 397         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
 398                 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
 399                     lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
 400                         lustre_swab_lov_mds_md(lmmk);
 401                         lustre_swab_lov_user_md_objects(
 402                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
 403                                 lmmk->lmm_stripe_count);
 404                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
 405                         lustre_swab_lov_comp_md_v1(
 406                                         (struct lov_comp_md_v1 *)lmmk);
 407                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_FOREIGN)) {
 408                         lfm = (struct lov_foreign_md *)lmmk;
 409                         __swab32s(&lfm->lfm_magic);
 410                         __swab32s(&lfm->lfm_length);
 411                         __swab32s(&lfm->lfm_type);
 412                         __swab32s(&lfm->lfm_flags);
 413                 }
 414         }
 415
 416         /*
 417          * Legacy appication passes limited buffer, we need to figure out
 418          * the user buffer size by the passed in lmm_stripe_count.
 419          */
 420         if (lsm->lsm_magic != LOV_MAGIC_FOREIGN)
 421                 if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
 422                         GOTO(out_free, rc = -EFAULT);
 423
 424         if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
 425             lum.lmm_magic == LOV_USER_MAGIC_V3)
 426                 lum_size = lov_user_md_size(lum.lmm_stripe_count,
 427                                             lum.lmm_magic);
 428
 429         if (lum_size != 0) {
 430                 struct lov_mds_md *comp_md = lmmk;
 431
 432                 /*
 433                  * Legacy app (ADIO for instance) treats the layout as V1/V3
 434                  * blindly, we'd return a reasonable V1/V3 for them.
 435                  */
 436                 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
 437                         struct lov_comp_md_v1 *comp_v1;
 438                         struct cl_object *cl_obj;
 439                         struct cl_attr attr;
 440                         int i;
 441
 442                         attr.cat_size = 0;
 443                         cl_obj = cl_object_top(&obj->lo_cl);
 444                         cl_object_attr_lock(cl_obj);
 445                         cl_object_attr_get(env, cl_obj, &attr);
 446                         cl_object_attr_unlock(cl_obj);
 447
 448                         /*
 449                          * return the last instantiated component if file size
 450                          * is non-zero, otherwise, return the last component.
 451                          */
 452                         comp_v1 = (struct lov_comp_md_v1 *)lmmk;
 453                         i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
 454                         for (; i < comp_v1->lcm_entry_count; i++) {
 455                                 if (!(comp_v1->lcm_entries[i].lcme_flags &
 456                                                 LCME_FL_INIT))
 457                                         break;
 458                         }
 459                         if (i > 0)
 460                                 i--;
 461                         comp_md = (struct lov_mds_md *)((char *)comp_v1 +
 462                                         comp_v1->lcm_entries[i].lcme_offset);
 463                         lum_size = comp_v1->lcm_entries[i].lcme_size;
 464                 }
 465
 466                 lmm = comp_md;
 467                 lmm_size = min(lum_size, lmmk_size);
 468         } else {
 469                 lmm = lmmk;
 470                 lmm_size = lmmk_size;
 471         }
 472
 473         /**
 474          * Return stripe_count=1 instead of 0 for DoM files to avoid
 475          * divide-by-zero for older userspace that calls this ioctl,
 476          * e.g. lustre ADIO driver.
 477          */
 478         if ((lum.lmm_stripe_count == 0) && (lum.lmm_pattern & LOV_PATTERN_MDT))
 479                 lum.lmm_stripe_count = 1;
 480         /**
 481          * User specified limited buffer size, usually the buffer is
 482          * from ll_lov_setstripe(), and the buffer can only hold basic
 483          * layout template info.
 484          */
 485         if (size == 0 || size > lmm_size)
 486                 size = lmm_size;
 487         if (copy_to_user(lump, lmm, size))
 488                 GOTO(out_free, rc = -EFAULT);
 489
 490 out_free:
 491         OBD_FREE_LARGE(lmmk, lmmk_size);
 492 out:
 493         RETURN(rc);
 494 }