lustre/lov/lov_pack.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 /*
   4  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
   5  * Use is subject to license terms.
   6  *
   7  * Copyright (c) 2011, 2017, Intel Corporation.
   8  */
   9
  10 /*
  11  * This file is part of Lustre, http://www.lustre.org/
  12  *
  13  * (Un)packing of OST/MDS requests
  14  *
  15  * Author: Andreas Dilger <adilger@clusterfs.com>
  16  */
  17
  18 #define DEBUG_SUBSYSTEM S_LOV
  19
  20 #include <lustre_net.h>
  21 #include <lustre_swab.h>
  22 #include <obd.h>
  23 #include <obd_class.h>
  24 #include <obd_support.h>
  25
  26 #include "lov_cl_internal.h"
  27 #include "lov_internal.h"
  28
  29 void lov_dump_lmm_common(int level, void *lmmp)
  30 {
  31         struct lov_mds_md *lmm = lmmp;
  32         struct ost_id oi;
  33
  34         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
  35         CDEBUG_LIMIT(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
  36                      POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
  37                      le32_to_cpu(lmm->lmm_pattern));
  38         CDEBUG_LIMIT(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
  39                      le32_to_cpu(lmm->lmm_stripe_size),
  40                      le16_to_cpu(lmm->lmm_stripe_count),
  41                      le16_to_cpu(lmm->lmm_layout_gen));
  42 }
  43
  44 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
  45                                  int stripe_count)
  46 {
  47         int i;
  48
  49         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
  50                 CDEBUG_LIMIT(level,
  51                              "bad stripe_count %u > max_stripe_count %u\n",
  52                              stripe_count, LOV_V1_INSANE_STRIPE_INDEX);
  53                 return;
  54         }
  55
  56         for (i = 0; i < stripe_count; ++i, ++lod) {
  57                 struct ost_id oi;
  58
  59                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
  60                 CDEBUG_LIMIT(level, "stripe %u idx %u subobj "DOSTID"\n", i,
  61                              le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
  62         }
  63 }
  64
  65 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
  66 {
  67         lov_dump_lmm_common(level, lmm);
  68         lov_dump_lmm_objects(level, lmm->lmm_objects,
  69                              le16_to_cpu(lmm->lmm_stripe_count));
  70 }
  71
  72 /**
  73  * Pack LOV striping metadata for disk storage format (in little
  74  * endian byte order).
  75  *
  76  * This follows the getxattr() conventions. If \a buf_size is zero
  77  * then return the size needed. If \a buf_size is too small then
  78  * return -ERANGE. Otherwise return the size of the result.
  79  */
  80 static ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
  81                                  size_t buf_size)
  82 {
  83         struct lov_mds_md_v1 *lmmv1 = buf;
  84         struct lov_mds_md_v3 *lmmv3 = buf;
  85         struct lov_ost_data_v1 *lmm_objects;
  86         size_t lmm_size;
  87         unsigned int i;
  88
  89         ENTRY;
  90
  91         lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
  92                                    lsm->lsm_magic);
  93         if (buf_size == 0)
  94                 RETURN(lmm_size);
  95
  96         if (buf_size < lmm_size)
  97                 RETURN(-ERANGE);
  98
  99         /*
 100          * lmmv1 and lmmv3 point to the same struct and have the
 101          * same first fields
 102          */
 103         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
 104         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
 105         lmmv1->lmm_stripe_size = cpu_to_le32(
 106                                 lsm->lsm_entries[0]->lsme_stripe_size);
 107         lmmv1->lmm_stripe_count = cpu_to_le16(
 108                                 lsm->lsm_entries[0]->lsme_stripe_count);
 109         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
 110         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
 111
 112         if (lsm->lsm_magic == LOV_MAGIC_V3) {
 113                 BUILD_BUG_ON(sizeof(lsm->lsm_entries[0]->lsme_pool_name) !=
 114                                     sizeof(lmmv3->lmm_pool_name));
 115                 strscpy(lmmv3->lmm_pool_name,
 116                         lsm->lsm_entries[0]->lsme_pool_name,
 117                         sizeof(lmmv3->lmm_pool_name));
 118                 lmm_objects = lmmv3->lmm_objects;
 119         } else {
 120                 lmm_objects = lmmv1->lmm_objects;
 121         }
 122
 123         if (lsm->lsm_is_released)
 124                 RETURN(lmm_size);
 125
 126         for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
 127                 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
 128
 129                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 130                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 131                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 132         }
 133
 134         RETURN(lmm_size);
 135 }
 136
 137 static ssize_t lov_lsm_pack_foreign(const struct lov_stripe_md *lsm, void *buf,
 138                                     size_t buf_size)
 139 {
 140         struct lov_foreign_md *lfm = buf;
 141         size_t lfm_size;
 142
 143         lfm_size = lsm->lsm_foreign_size;
 144
 145         if (buf_size == 0)
 146                 RETURN(lfm_size);
 147
 148         /* if buffer too small return ERANGE but copy the size the
 149          * caller has requested anyway. This may be useful to get
 150          * only the header without the need to alloc the full size
 151          */
 152         if (buf_size < lfm_size) {
 153                 memcpy(lfm, lsm_foreign(lsm), buf_size);
 154                 RETURN(-ERANGE);
 155         }
 156
 157         /* full foreign LOV is already avail in its cache
 158          * no need to translate format fields to little-endian
 159          */
 160         memcpy(lfm, lsm_foreign(lsm), lsm->lsm_foreign_size);
 161
 162         RETURN(lfm_size);
 163 }
 164
 165 static unsigned int lov_lsme_pack_foreign(struct lov_stripe_md_entry *lsme,
 166                                           void *lmm)
 167 {
 168         struct lov_foreign_md *lfm = (struct lov_foreign_md *)lmm;
 169
 170         lfm->lfm_magic = cpu_to_le32(lsme->lsme_magic);
 171         lfm->lfm_length = cpu_to_le32(lsme->lsme_length);
 172         lfm->lfm_type = cpu_to_le32(lsme->lsme_type);
 173         lfm->lfm_flags = cpu_to_le32(lsme->lsme_foreign_flags);
 174
 175         /* TODO: support for foreign layout other than HSM, i.e. DAOS. */
 176         if (lov_hsm_type_supported(lsme->lsme_type))
 177                 lov_foreign_hsm_to_le(lfm, &lsme->lsme_hsm);
 178
 179         return lov_foreign_md_size(lsme->lsme_length);
 180 }
 181
 182 static unsigned int lov_lsme_pack_v1v3(struct lov_stripe_md_entry *lsme,
 183                                        struct lov_mds_md *lmm)
 184 {
 185         struct lov_ost_data_v1 *lmm_objects;
 186         __u16 stripe_count;
 187         unsigned int i;
 188
 189         lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
 190         /* lmm->lmm_oi not set */
 191         lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
 192         lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
 193         lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
 194         lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
 195
 196         if (lsme->lsme_magic == LOV_MAGIC_V3) {
 197                 struct lov_mds_md_v3 *lmmv3 = (struct lov_mds_md_v3 *)lmm;
 198
 199                 strscpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
 200                         sizeof(lmmv3->lmm_pool_name));
 201                 lmm_objects = lmmv3->lmm_objects;
 202         } else {
 203                 lmm_objects = ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
 204         }
 205
 206         if (lsme_inited(lsme) && !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
 207                 stripe_count = lsme->lsme_stripe_count;
 208         else
 209                 stripe_count = 0;
 210
 211         for (i = 0; i < stripe_count; i++) {
 212                 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
 213
 214                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
 215                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
 216                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
 217         }
 218
 219         return lov_mds_md_size(stripe_count, lsme->lsme_magic);
 220 }
 221
 222 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
 223                      size_t buf_size)
 224 {
 225         struct lov_comp_md_v1 *lcmv1 = buf;
 226         struct lov_comp_md_entry_v1 *lcme;
 227         size_t lmm_size;
 228         unsigned int entry;
 229         unsigned int offset;
 230         unsigned int size;
 231
 232         ENTRY;
 233
 234         if (lsm->lsm_magic == LOV_MAGIC_V1 || lsm->lsm_magic == LOV_MAGIC_V3)
 235                 return lov_lsm_pack_v1v3(lsm, buf, buf_size);
 236
 237         if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
 238                 return lov_lsm_pack_foreign(lsm, buf, buf_size);
 239
 240         lmm_size = lov_comp_md_size(lsm);
 241         if (buf_size == 0)
 242                 RETURN(lmm_size);
 243
 244         if (buf_size < lmm_size)
 245                 RETURN(-ERANGE);
 246
 247         lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
 248         lcmv1->lcm_size = cpu_to_le32(lmm_size);
 249         lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
 250         lcmv1->lcm_flags = cpu_to_le16(lsm->lsm_flags);
 251         lcmv1->lcm_mirror_count = cpu_to_le16(lsm->lsm_mirror_count);
 252         lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
 253
 254         offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
 255
 256         for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
 257                 struct lov_stripe_md_entry *lsme;
 258                 struct lov_mds_md *lmm;
 259
 260                 lsme = lsm->lsm_entries[entry];
 261                 lcme = &lcmv1->lcm_entries[entry];
 262
 263                 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
 264                 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
 265                 if (lsme->lsme_flags & LCME_FL_NOSYNC)
 266                         lcme->lcme_timestamp =
 267                                 cpu_to_le64(lsme->lsme_timestamp);
 268                 lcme->lcme_extent.e_start =
 269                         cpu_to_le64(lsme->lsme_extent.e_start);
 270                 lcme->lcme_extent.e_end =
 271                         cpu_to_le64(lsme->lsme_extent.e_end);
 272                 lcme->lcme_offset = cpu_to_le32(offset);
 273
 274                 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
 275                 if (lsme->lsme_magic == LOV_MAGIC_FOREIGN)
 276                         size = lov_lsme_pack_foreign(lsme, lmm);
 277                 else
 278                         size = lov_lsme_pack_v1v3(lsme, lmm);
 279                 lcme->lcme_size = cpu_to_le32(size);
 280                 offset += size;
 281         } /* for each layout component */
 282
 283         RETURN(lmm_size);
 284 }
 285
 286 /* Find the max stripecount we should use */
 287 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
 288 {
 289         struct lu_tgt_descs *ltd = &lov->lov_ost_descs;
 290         u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
 291
 292         if (!stripe_count)
 293                 stripe_count = ltd->ltd_lov_desc.ld_default_stripe_count;
 294         if (stripe_count > ltd->ltd_lov_desc.ld_active_tgt_count)
 295                 stripe_count = ltd->ltd_lov_desc.ld_active_tgt_count;
 296         if (!stripe_count)
 297                 stripe_count = 1;
 298
 299         /*
 300          * stripe count is based on whether ldiskfs can handle
 301          * larger EA sizes
 302          */
 303         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
 304             lov->lov_ocd.ocd_max_easize)
 305                 max_stripes = lov_mds_md_max_stripe_count(
 306                         lov->lov_ocd.ocd_max_easize, magic);
 307
 308         if (stripe_count > max_stripes)
 309                 stripe_count = max_stripes;
 310
 311         return stripe_count;
 312 }
 313
 314 void lov_free_memmd(struct lov_stripe_md **lsmp)
 315 {
 316         struct lov_stripe_md *lsm = *lsmp;
 317
 318         *lsmp = NULL;
 319         kref_put(&lsm->lsm_refc, lsm_free);
 320 }
 321
 322 /*
 323  * Unpack LOV object metadata from disk storage.  It is packed in LE byte
 324  * order and is opaque to the networking layer.
 325  */
 326 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
 327                                    size_t buf_size)
 328 {
 329         const struct lsm_operations *op;
 330         struct lov_stripe_md *lsm;
 331         u32 magic;
 332
 333         ENTRY;
 334
 335         if (buf_size < sizeof(magic))
 336                 RETURN(ERR_PTR(-EINVAL));
 337
 338         magic = le32_to_cpu(*(u32 *)buf);
 339         op = lsm_op_find(magic);
 340         if (!op)
 341                 RETURN(ERR_PTR(-EINVAL));
 342
 343         lsm = op->lsm_unpackmd(lov, buf, buf_size);
 344
 345         RETURN(lsm);
 346 }
 347
 348 /*
 349  * Retrieve object striping information.
 350  *
 351  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
 352  * the maximum number of OST indices which will fit in the user buffer.
 353  * lmm_magic must be LOV_USER_MAGIC.
 354  *
 355  * If @size > 0, User specified limited buffer size, usually the buffer is from
 356  * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
 357  */
 358 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
 359                   struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
 360                   size_t size)
 361 {
 362         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
 363         struct lov_mds_md *lmmk, *lmm;
 364         struct lov_foreign_md *lfm;
 365         struct lov_user_md_v1 lum;
 366         size_t lmmk_size, lum_size = 0;
 367         ssize_t lmm_size;
 368         int rc = 0;
 369
 370         ENTRY;
 371
 372         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
 373             lsm->lsm_magic != LOV_MAGIC_COMP_V1 &&
 374             lsm->lsm_magic != LOV_MAGIC_FOREIGN) {
 375                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
 376                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
 377                 GOTO(out, rc = -EIO);
 378         }
 379
 380         lmmk_size = lov_comp_md_size(lsm);
 381
 382         OBD_ALLOC_LARGE(lmmk, lmmk_size);
 383         if (!lmmk)
 384                 GOTO(out, rc = -ENOMEM);
 385
 386         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
 387         if (lmm_size < 0)
 388                 GOTO(out_free, rc = lmm_size);
 389
 390         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
 391                 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
 392                     lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
 393                         lustre_swab_lov_mds_md(lmmk);
 394                         lustre_swab_lov_user_md_objects(
 395                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
 396                                 lmmk->lmm_stripe_count);
 397                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
 398                         lustre_swab_lov_comp_md_v1(
 399                                         (struct lov_comp_md_v1 *)lmmk);
 400                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_FOREIGN)) {
 401                         lfm = (struct lov_foreign_md *)lmmk;
 402                         __swab32s(&lfm->lfm_magic);
 403                         __swab32s(&lfm->lfm_length);
 404                         __swab32s(&lfm->lfm_type);
 405                         __swab32s(&lfm->lfm_flags);
 406                 }
 407         }
 408
 409         /*
 410          * Legacy appication passes limited buffer, we need to figure out
 411          * the user buffer size by the passed in lmm_stripe_count.
 412          */
 413         if (lsm->lsm_magic != LOV_MAGIC_FOREIGN)
 414                 if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
 415                         GOTO(out_free, rc = -EFAULT);
 416
 417         if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
 418             lum.lmm_magic == LOV_USER_MAGIC_V3)
 419                 lum_size = lov_user_md_size(lum.lmm_stripe_count,
 420                                             lum.lmm_magic);
 421
 422         if (lum_size != 0) {
 423                 struct lov_mds_md *comp_md = lmmk;
 424
 425                 /*
 426                  * Legacy app (ADIO for instance) treats the layout as V1/V3
 427                  * blindly, we'd return a reasonable V1/V3 for them.
 428                  */
 429                 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
 430                         struct lov_comp_md_v1 *comp_v1;
 431                         struct cl_object *cl_obj;
 432                         struct cl_attr attr;
 433                         int i;
 434
 435                         attr.cat_size = 0;
 436                         cl_obj = cl_object_top(&obj->lo_cl);
 437                         cl_object_attr_lock(cl_obj);
 438                         cl_object_attr_get(env, cl_obj, &attr);
 439                         cl_object_attr_unlock(cl_obj);
 440
 441                         /*
 442                          * return the last instantiated component if file size
 443                          * is non-zero, otherwise, return the last component.
 444                          */
 445                         comp_v1 = (struct lov_comp_md_v1 *)lmmk;
 446                         i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
 447                         for (; i < comp_v1->lcm_entry_count; i++) {
 448                                 if (!(comp_v1->lcm_entries[i].lcme_flags &
 449                                                 LCME_FL_INIT))
 450                                         break;
 451                         }
 452                         if (i > 0)
 453                                 i--;
 454                         comp_md = (struct lov_mds_md *)((char *)comp_v1 +
 455                                         comp_v1->lcm_entries[i].lcme_offset);
 456                         lum_size = comp_v1->lcm_entries[i].lcme_size;
 457                 }
 458
 459                 lmm = comp_md;
 460                 lmm_size = min(lum_size, lmmk_size);
 461         } else {
 462                 lmm = lmmk;
 463                 lmm_size = lmmk_size;
 464         }
 465
 466         /**
 467          * User specified limited buffer size, usually the buffer is
 468          * from ll_lov_setstripe(), and the buffer can only hold basic
 469          * layout template info.
 470          */
 471         if (size == 0 || size > lmm_size)
 472                 size = lmm_size;
 473         if (copy_to_user(lump, lmm, size))
 474                 GOTO(out_free, rc = -EFAULT);
 475
 476 out_free:
 477         OBD_FREE_LARGE(lmmk, lmmk_size);
 478 out:
 479         RETURN(rc);
 480 }