lustre/llite/dir.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, 2013, 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/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <asm/uaccess.h>
  46 #include <linux/buffer_head.h>   // for wait_on_buffer
  47 #include <linux/pagevec.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <lustre/lustre_idl.h>
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted. See
 139  * mdc_adjust_dirpages().
 140  *
 141  */
 142 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 143                              __u64 offset, struct ll_dir_chain *chain)
 144 {
 145         struct md_callback      cb_op;
 146         struct page             *page;
 147         int                     rc;
 148
 149         cb_op.md_blocking_ast = ll_md_blocking_ast;
 150         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 151         if (rc != 0)
 152                 return ERR_PTR(rc);
 153
 154         return page;
 155 }
 156
 157 void ll_release_page(struct inode *inode, struct page *page,
 158                      bool remove)
 159 {
 160         kunmap(page);
 161
 162         /* Always remove the page for striped dir, because the page is
 163          * built from temporarily in LMV layer */
 164         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 165             ll_i2info(inode)->lli_lsm_md != NULL) {
 166                 __free_page(page);
 167                 return;
 168         }
 169
 170         if (remove) {
 171                 lock_page(page);
 172                 if (likely(page->mapping != NULL))
 173                         truncate_complete_page(page->mapping, page);
 174                 unlock_page(page);
 175         }
 176         page_cache_release(page);
 177 }
 178
 179 #ifdef HAVE_DIR_CONTEXT
 180 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 181                 struct dir_context *ctx)
 182 {
 183 #else
 184 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 185                 void *cookie, filldir_t filldir)
 186 {
 187 #endif
 188         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 189         __u64                 pos        = *ppos;
 190         bool                  is_api32 = ll_need_32bit_api(sbi);
 191         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 192         struct page          *page;
 193         struct ll_dir_chain   chain;
 194         bool                  done = false;
 195         int                   rc = 0;
 196         ENTRY;
 197
 198         ll_dir_chain_init(&chain);
 199
 200         page = ll_get_dir_page(inode, op_data, pos, &chain);
 201
 202         while (rc == 0 && !done) {
 203                 struct lu_dirpage *dp;
 204                 struct lu_dirent  *ent;
 205                 __u64 hash;
 206                 __u64 next;
 207
 208                 if (IS_ERR(page)) {
 209                         rc = PTR_ERR(page);
 210                         break;
 211                 }
 212
 213                 hash = MDS_DIR_END_OFF;
 214                 dp = page_address(page);
 215                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 216                      ent = lu_dirent_next(ent)) {
 217                         __u16          type;
 218                         int            namelen;
 219                         struct lu_fid  fid;
 220                         __u64          lhash;
 221                         __u64          ino;
 222
 223                         hash = le64_to_cpu(ent->lde_hash);
 224                         if (hash < pos)
 225                                 /*
 226                                  * Skip until we find target hash
 227                                  * value.
 228                                  */
 229                                 continue;
 230
 231                         namelen = le16_to_cpu(ent->lde_namelen);
 232                         if (namelen == 0)
 233                                 /*
 234                                  * Skip dummy record.
 235                                  */
 236                                 continue;
 237
 238                         if (is_api32 && is_hash64)
 239                                 lhash = hash >> 32;
 240                         else
 241                                 lhash = hash;
 242                         fid_le_to_cpu(&fid, &ent->lde_fid);
 243                         ino = cl_fid_build_ino(&fid, is_api32);
 244                         type = ll_dirent_type_get(ent);
 245                         /* For 'll_nfs_get_name_filldir()', it will try
 246                          * to access the 'ent' through its 'lde_name',
 247                          * so the parameter 'name' for 'filldir()' must
 248                          * be part of the 'ent'. */
 249 #ifdef HAVE_DIR_CONTEXT
 250                         ctx->pos = lhash;
 251                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 252                                          type);
 253 #else
 254                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 255                                        ino, type);
 256 #endif
 257                 }
 258
 259                 if (done) {
 260                         pos = hash;
 261                         ll_release_page(inode, page, false);
 262                         break;
 263                 }
 264
 265                 next = le64_to_cpu(dp->ldp_hash_end);
 266                 pos = next;
 267                 if (pos == MDS_DIR_END_OFF) {
 268                         /*
 269                          * End of directory reached.
 270                          */
 271                         done = 1;
 272                         ll_release_page(inode, page, false);
 273                 } else {
 274                         /*
 275                          * Normal case: continue to the next
 276                          * page.
 277                          */
 278                         ll_release_page(inode, page,
 279                                         le32_to_cpu(dp->ldp_flags) &
 280                                         LDF_COLLIDE);
 281                         next = pos;
 282                         page = ll_get_dir_page(inode, op_data, pos,
 283                                                &chain);
 284                 }
 285         }
 286 #ifdef HAVE_DIR_CONTEXT
 287         ctx->pos = pos;
 288 #else
 289         *ppos = pos;
 290 #endif
 291         ll_dir_chain_fini(&chain);
 292         RETURN(rc);
 293 }
 294
 295 #ifdef HAVE_DIR_CONTEXT
 296 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 297 #else
 298 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 299 #endif
 300 {
 301         struct inode            *inode  = filp->f_dentry->d_inode;
 302         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 303         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 304         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 305         int                     api32   = ll_need_32bit_api(sbi);
 306         struct md_op_data       *op_data;
 307         __u64                   pos;
 308         int                     rc;
 309         ENTRY;
 310
 311         if (lfd != NULL)
 312                 pos = lfd->lfd_pos;
 313         else
 314                 pos = 0;
 315
 316         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 317                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 318                inode, (unsigned long)pos, i_size_read(inode), api32);
 319
 320         if (pos == MDS_DIR_END_OFF)
 321                 /*
 322                  * end-of-file.
 323                  */
 324                 GOTO(out, rc = 0);
 325
 326         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 327                                      LUSTRE_OPC_ANY, inode);
 328         if (IS_ERR(op_data))
 329                 GOTO(out, rc = PTR_ERR(op_data));
 330
 331         if (unlikely(op_data->op_mea1 != NULL)) {
 332                 /* This is only needed for striped dir to fill ..,
 333                  * see lmv_read_entry */
 334                 if (filp->f_dentry->d_parent != NULL &&
 335                     filp->f_dentry->d_parent->d_inode != NULL) {
 336                         __u64 ibits = MDS_INODELOCK_UPDATE;
 337                         struct inode *parent =
 338                                 filp->f_dentry->d_parent->d_inode;
 339
 340                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 341                                 op_data->op_fid3 = *ll_inode2fid(parent);
 342                 }
 343
 344                 /* If it can not find in cache, do lookup .. on the master
 345                  * object */
 346                 if (fid_is_zero(&op_data->op_fid3)) {
 347                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 348                         if (rc != 0) {
 349                                 ll_finish_md_op_data(op_data);
 350                                 RETURN(rc);
 351                         }
 352                 }
 353         }
 354         op_data->op_max_pages = sbi->ll_md_brw_pages;
 355 #ifdef HAVE_DIR_CONTEXT
 356         ctx->pos = pos;
 357         rc = ll_dir_read(inode, &pos, op_data, ctx);
 358         pos = ctx->pos;
 359 #else
 360         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 361 #endif
 362         if (lfd != NULL)
 363                 lfd->lfd_pos = pos;
 364
 365         if (pos == MDS_DIR_END_OFF) {
 366                 if (api32)
 367                         pos = LL_DIR_END_OFF_32BIT;
 368                 else
 369                         pos = LL_DIR_END_OFF;
 370         } else {
 371                 if (api32 && hash64)
 372                         pos = pos >> 32;
 373         }
 374 #ifdef HAVE_DIR_CONTEXT
 375         ctx->pos = pos;
 376 #else
 377         filp->f_pos = pos;
 378 #endif
 379         ll_finish_md_op_data(op_data);
 380         filp->f_version = inode->i_version;
 381
 382 out:
 383         if (!rc)
 384                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 385
 386         RETURN(rc);
 387 }
 388
 389 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 390 {
 391         struct mgs_send_param *msp;
 392         int rc = 0;
 393
 394         OBD_ALLOC_PTR(msp);
 395         if (!msp)
 396                 return -ENOMEM;
 397
 398         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 399         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 400                                 sizeof(struct mgs_send_param), msp, NULL);
 401         if (rc)
 402                 CERROR("Failed to set parameter: %d\n", rc);
 403         OBD_FREE_PTR(msp);
 404
 405         return rc;
 406 }
 407
 408 /**
 409  * Create striped directory with specified stripe(@lump)
 410  *
 411  * param[in]parent      the parent of the directory.
 412  * param[in]lump        the specified stripes.
 413  * param[in]dirname     the name of the directory.
 414  * param[in]mode        the specified mode of the directory.
 415  *
 416  * retval               =0 if striped directory is being created successfully.
 417  *                      <0 if the creation is failed.
 418  */
 419 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
 420                                const char *dirname, umode_t mode)
 421 {
 422         struct ptlrpc_request *request = NULL;
 423         struct md_op_data *op_data;
 424         struct ll_sb_info *sbi = ll_i2sbi(parent);
 425         int err;
 426         ENTRY;
 427
 428         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 429                 RETURN(-EINVAL);
 430
 431         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 432                "stripe_offset %d, stripe_count: %u\n",
 433                PFID(ll_inode2fid(parent)), parent, dirname,
 434                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 435
 436         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 437                 lustre_swab_lmv_user_md(lump);
 438
 439         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 440                 mode &= ~current_umask();
 441         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 442         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 443                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 444                                      lump);
 445         if (IS_ERR(op_data))
 446                 GOTO(err_exit, err = PTR_ERR(op_data));
 447
 448         op_data->op_cli_flags |= CLI_SET_MEA;
 449         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 450                         from_kuid(&init_user_ns, current_fsuid()),
 451                         from_kgid(&init_user_ns, current_fsgid()),
 452                         cfs_curproc_cap_pack(), 0, &request);
 453         ll_finish_md_op_data(op_data);
 454         if (err)
 455                 GOTO(err_exit, err);
 456 err_exit:
 457         ptlrpc_req_finished(request);
 458         return err;
 459 }
 460
 461 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 462                      int set_default)
 463 {
 464         struct ll_sb_info *sbi = ll_i2sbi(inode);
 465         struct md_op_data *op_data;
 466         struct ptlrpc_request *req = NULL;
 467         int rc = 0;
 468         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 469         struct obd_device *mgc = lsi->lsi_mgc;
 470         int lum_size;
 471         ENTRY;
 472
 473         if (lump != NULL) {
 474                 /*
 475                  * This is coming from userspace, so should be in
 476                  * local endian.  But the MDS would like it in little
 477                  * endian, so we swab it before we send it.
 478                  */
 479                 switch (lump->lmm_magic) {
 480                 case LOV_USER_MAGIC_V1: {
 481                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 482                                 lustre_swab_lov_user_md_v1(lump);
 483                         lum_size = sizeof(struct lov_user_md_v1);
 484                         break;
 485                 }
 486                 case LOV_USER_MAGIC_V3: {
 487                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 488                                 lustre_swab_lov_user_md_v3(
 489                                         (struct lov_user_md_v3 *)lump);
 490                         lum_size = sizeof(struct lov_user_md_v3);
 491                         break;
 492                 }
 493                 case LMV_USER_MAGIC: {
 494                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 495                                 lustre_swab_lmv_user_md(
 496                                         (struct lmv_user_md *)lump);
 497                         lum_size = sizeof(struct lmv_user_md);
 498                         break;
 499                 }
 500                 default: {
 501                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 502                                         " %#08x != %#08x nor %#08x\n",
 503                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 504                                         LOV_USER_MAGIC_V3);
 505                         RETURN(-EINVAL);
 506                 }
 507                 }
 508         } else {
 509                 lum_size = sizeof(struct lov_user_md_v1);
 510         }
 511
 512         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 513                                      LUSTRE_OPC_ANY, NULL);
 514         if (IS_ERR(op_data))
 515                 RETURN(PTR_ERR(op_data));
 516
 517         /* swabbing is done in lov_setstripe() on server side */
 518         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 519                         NULL, 0, &req, NULL);
 520         ll_finish_md_op_data(op_data);
 521         ptlrpc_req_finished(req);
 522         if (rc) {
 523                 if (rc != -EPERM && rc != -EACCES)
 524                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 525         }
 526
 527         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 528          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 529          need the make the distiction between the 2 versions */
 530         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 531                 char *param = NULL;
 532                 char *buf;
 533
 534                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 535                 if (param == NULL)
 536                         GOTO(end, rc = -ENOMEM);
 537
 538                 buf = param;
 539                 /* Get fsname and assume devname to be -MDT0000. */
 540                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 541                 strcat(buf, "-MDT0000.lov");
 542                 buf += strlen(buf);
 543
 544                 /* Set root stripesize */
 545                 sprintf(buf, ".stripesize=%u",
 546                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 547                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 548                 if (rc)
 549                         GOTO(end, rc);
 550
 551                 /* Set root stripecount */
 552                 sprintf(buf, ".stripecount=%hd",
 553                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 554                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 555                 if (rc)
 556                         GOTO(end, rc);
 557
 558                 /* Set root stripeoffset */
 559                 sprintf(buf, ".stripeoffset=%hd",
 560                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 561                         (typeof(lump->lmm_stripe_offset))(-1));
 562                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 563
 564 end:
 565                 if (param != NULL)
 566                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 567         }
 568         RETURN(rc);
 569 }
 570
 571 /**
 572  * This function will be used to get default LOV/LMV/Default LMV
 573  * @valid will be used to indicate which stripe it will retrieve
 574  *      OBD_MD_MEA              LMV stripe EA
 575  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 576  *      otherwise               Default LOV EA.
 577  * Each time, it can only retrieve 1 stripe EA
 578  **/
 579 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 580                      struct ptlrpc_request **request, obd_valid valid)
 581 {
 582         struct ll_sb_info *sbi = ll_i2sbi(inode);
 583         struct mdt_body   *body;
 584         struct lov_mds_md *lmm = NULL;
 585         struct ptlrpc_request *req = NULL;
 586         int rc, lmm_size;
 587         struct md_op_data *op_data;
 588         ENTRY;
 589
 590         rc = ll_get_default_mdsize(sbi, &lmm_size);
 591         if (rc)
 592                 RETURN(rc);
 593
 594         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 595                                      0, lmm_size, LUSTRE_OPC_ANY,
 596                                      NULL);
 597         if (IS_ERR(op_data))
 598                 RETURN(PTR_ERR(op_data));
 599
 600         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 601         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 602         ll_finish_md_op_data(op_data);
 603         if (rc < 0) {
 604                 CDEBUG(D_INFO, "md_getattr failed on inode "
 605                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 606                 GOTO(out, rc);
 607         }
 608
 609         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 610         LASSERT(body != NULL);
 611
 612         lmm_size = body->mbo_eadatasize;
 613
 614         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 615             lmm_size == 0) {
 616                 GOTO(out, rc = -ENODATA);
 617         }
 618
 619         lmm = req_capsule_server_sized_get(&req->rq_pill,
 620                                            &RMF_MDT_MD, lmm_size);
 621         LASSERT(lmm != NULL);
 622
 623         /*
 624          * This is coming from the MDS, so is probably in
 625          * little endian.  We convert it to host endian before
 626          * passing it to userspace.
 627          */
 628         /* We don't swab objects for directories */
 629         switch (le32_to_cpu(lmm->lmm_magic)) {
 630         case LOV_MAGIC_V1:
 631                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 632                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 633                 break;
 634         case LOV_MAGIC_V3:
 635                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 636                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 637                 break;
 638         case LMV_MAGIC_V1:
 639                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 640                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 641                 break;
 642         case LMV_USER_MAGIC:
 643                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 644                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 645                 break;
 646         default:
 647                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 648                 rc = -EPROTO;
 649         }
 650 out:
 651         *plmm = lmm;
 652         *plmm_size = lmm_size;
 653         *request = req;
 654         return rc;
 655 }
 656
 657 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 658 {
 659         struct md_op_data       *op_data;
 660         int                     rc;
 661         int                     mdt_index;
 662         ENTRY;
 663
 664         OBD_ALLOC_PTR(op_data);
 665         if (op_data == NULL)
 666                 RETURN(-ENOMEM);
 667
 668         op_data->op_flags |= MF_GET_MDT_IDX;
 669         op_data->op_fid1 = *fid;
 670         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 671         mdt_index = op_data->op_mds;
 672         OBD_FREE_PTR(op_data);
 673         if (rc < 0)
 674                 RETURN(rc);
 675
 676         RETURN(mdt_index);
 677 }
 678
 679 /*
 680  *  Get MDT index for the inode.
 681  */
 682 int ll_get_mdt_idx(struct inode *inode)
 683 {
 684         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 685 }
 686
 687 /**
 688  * Generic handler to do any pre-copy work.
 689  *
 690  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 691  * first information for it that real work has started.
 692  *
 693  * Moreover, for a ARCHIVE request, it will sample the file data version and
 694  * store it in \a copy.
 695  *
 696  * \return 0 on success.
 697  */
 698 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 699 {
 700         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 701         struct hsm_progress_kernel       hpk;
 702         int                              rc;
 703         ENTRY;
 704
 705         /* Forge a hsm_progress based on data from copy. */
 706         hpk.hpk_fid = copy->hc_hai.hai_fid;
 707         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 708         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 709         hpk.hpk_extent.length = 0;
 710         hpk.hpk_flags = 0;
 711         hpk.hpk_errval = 0;
 712         hpk.hpk_data_version = 0;
 713
 714
 715         /* For archive request, we need to read the current file version. */
 716         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 717                 struct inode    *inode;
 718                 __u64            data_version = 0;
 719
 720                 /* Get inode for this fid */
 721                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 722                 if (IS_ERR(inode)) {
 723                         hpk.hpk_flags |= HP_FLAG_RETRY;
 724                         /* hpk_errval is >= 0 */
 725                         hpk.hpk_errval = -PTR_ERR(inode);
 726                         GOTO(progress, rc = PTR_ERR(inode));
 727                 }
 728
 729                 /* Read current file data version */
 730                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 731                 iput(inode);
 732                 if (rc != 0) {
 733                         CDEBUG(D_HSM, "Could not read file data version of "
 734                                       DFID" (rc = %d). Archive request ("
 735                                       LPX64") could not be done.\n",
 736                                       PFID(&copy->hc_hai.hai_fid), rc,
 737                                       copy->hc_hai.hai_cookie);
 738                         hpk.hpk_flags |= HP_FLAG_RETRY;
 739                         /* hpk_errval must be >= 0 */
 740                         hpk.hpk_errval = -rc;
 741                         GOTO(progress, rc);
 742                 }
 743
 744                 /* Store it the hsm_copy for later copytool use.
 745                  * Always modified even if no lsm. */
 746                 copy->hc_data_version = data_version;
 747         }
 748
 749 progress:
 750         /* On error, the request should be considered as completed */
 751         if (hpk.hpk_errval > 0)
 752                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 753         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 754                            &hpk, NULL);
 755
 756         RETURN(rc);
 757 }
 758
 759 /**
 760  * Generic handler to do any post-copy work.
 761  *
 762  * It will send the last hsm_progress update to coordinator to inform it
 763  * that copy is finished and whether it was successful or not.
 764  *
 765  * Moreover,
 766  * - for ARCHIVE request, it will sample the file data version and compare it
 767  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 768  *   will be considered as failed.
 769  * - for RESTORE request, it will sample the file data version and send it to
 770  *   coordinator which is useful if the file was imported as 'released'.
 771  *
 772  * \return 0 on success.
 773  */
 774 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 775 {
 776         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 777         struct hsm_progress_kernel       hpk;
 778         int                              rc;
 779         ENTRY;
 780
 781         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 782         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 783          * initialized if copy_end was called with copy == NULL.
 784          */
 785
 786         /* Forge a hsm_progress based on data from copy. */
 787         hpk.hpk_fid = copy->hc_hai.hai_fid;
 788         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 789         hpk.hpk_extent = copy->hc_hai.hai_extent;
 790         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 791         hpk.hpk_errval = copy->hc_errval;
 792         hpk.hpk_data_version = 0;
 793
 794         /* For archive request, we need to check the file data was not changed.
 795          *
 796          * For restore request, we need to send the file data version, this is
 797          * useful when the file was created using hsm_import.
 798          */
 799         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 800              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 801             (copy->hc_errval == 0)) {
 802                 struct inode    *inode;
 803                 __u64            data_version = 0;
 804
 805                 /* Get lsm for this fid */
 806                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 807                 if (IS_ERR(inode)) {
 808                         hpk.hpk_flags |= HP_FLAG_RETRY;
 809                         /* hpk_errval must be >= 0 */
 810                         hpk.hpk_errval = -PTR_ERR(inode);
 811                         GOTO(progress, rc = PTR_ERR(inode));
 812                 }
 813
 814                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 815                 iput(inode);
 816                 if (rc) {
 817                         CDEBUG(D_HSM, "Could not read file data version. "
 818                                       "Request could not be confirmed.\n");
 819                         if (hpk.hpk_errval == 0)
 820                                 hpk.hpk_errval = -rc;
 821                         GOTO(progress, rc);
 822                 }
 823
 824                 /* Store it the hsm_copy for later copytool use.
 825                  * Always modified even if no lsm. */
 826                 hpk.hpk_data_version = data_version;
 827
 828                 /* File could have been stripped during archiving, so we need
 829                  * to check anyway. */
 830                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 831                     (copy->hc_data_version != data_version)) {
 832                         CDEBUG(D_HSM, "File data version mismatched. "
 833                               "File content was changed during archiving. "
 834                                DFID", start:"LPX64" current:"LPX64"\n",
 835                                PFID(&copy->hc_hai.hai_fid),
 836                                copy->hc_data_version, data_version);
 837                         /* File was changed, send error to cdt. Do not ask for
 838                          * retry because if a file is modified frequently,
 839                          * the cdt will loop on retried archive requests.
 840                          * The policy engine will ask for a new archive later
 841                          * when the file will not be modified for some tunable
 842                          * time */
 843                         /* we do not notify caller */
 844                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 845                         /* hpk_errval must be >= 0 */
 846                         hpk.hpk_errval = EBUSY;
 847                 }
 848
 849         }
 850
 851 progress:
 852         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 853                            &hpk, NULL);
 854
 855         RETURN(rc);
 856 }
 857
 858
 859 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 860                           const void __user *data, size_t size)
 861 {
 862         void *copy;
 863         int rc;
 864
 865         OBD_ALLOC(copy, size);
 866         if (copy == NULL)
 867                 return -ENOMEM;
 868
 869         if (copy_from_user(copy, data, size)) {
 870                 rc = -EFAULT;
 871                 goto out;
 872         }
 873
 874         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 875 out:
 876         OBD_FREE(copy, size);
 877
 878         return rc;
 879 }
 880
 881 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 882 {
 883         int cmd = qctl->qc_cmd;
 884         int type = qctl->qc_type;
 885         int id = qctl->qc_id;
 886         int valid = qctl->qc_valid;
 887         int rc = 0;
 888         ENTRY;
 889
 890         switch (cmd) {
 891         case LUSTRE_Q_INVALIDATE:
 892         case LUSTRE_Q_FINVALIDATE:
 893         case Q_QUOTAON:
 894         case Q_QUOTAOFF:
 895         case Q_SETQUOTA:
 896         case Q_SETINFO:
 897                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 898                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 899                         RETURN(-EPERM);
 900                 break;
 901         case Q_GETQUOTA:
 902                 if (((type == USRQUOTA &&
 903                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 904                      (type == GRPQUOTA &&
 905                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 906                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 907                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 908                         RETURN(-EPERM);
 909                 break;
 910         case Q_GETINFO:
 911                 break;
 912         default:
 913                 CERROR("unsupported quotactl op: %#x\n", cmd);
 914                 RETURN(-ENOTTY);
 915         }
 916
 917         if (valid != QC_GENERAL) {
 918                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 919                         RETURN(-EOPNOTSUPP);
 920
 921                 if (cmd == Q_GETINFO)
 922                         qctl->qc_cmd = Q_GETOINFO;
 923                 else if (cmd == Q_GETQUOTA)
 924                         qctl->qc_cmd = Q_GETOQUOTA;
 925                 else
 926                         RETURN(-EINVAL);
 927
 928                 switch (valid) {
 929                 case QC_MDTIDX:
 930                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 931                                            sizeof(*qctl), qctl, NULL);
 932                         break;
 933                 case QC_OSTIDX:
 934                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 935                                            sizeof(*qctl), qctl, NULL);
 936                         break;
 937                 case QC_UUID:
 938                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 939                                            sizeof(*qctl), qctl, NULL);
 940                         if (rc == -EAGAIN)
 941                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 942                                                    sbi->ll_dt_exp,
 943                                                    sizeof(*qctl), qctl, NULL);
 944                         break;
 945                 default:
 946                         rc = -EINVAL;
 947                         break;
 948                 }
 949
 950                 if (rc)
 951                         RETURN(rc);
 952
 953                 qctl->qc_cmd = cmd;
 954         } else {
 955                 struct obd_quotactl *oqctl;
 956
 957                 OBD_ALLOC_PTR(oqctl);
 958                 if (oqctl == NULL)
 959                         RETURN(-ENOMEM);
 960
 961                 QCTL_COPY(oqctl, qctl);
 962                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 963                 if (rc) {
 964                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 965                                 oqctl->qc_cmd = Q_QUOTAOFF;
 966                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 967                         }
 968                         OBD_FREE_PTR(oqctl);
 969                         RETURN(rc);
 970                 }
 971                 /* If QIF_SPACE is not set, client should collect the
 972                  * space usage from OSSs by itself */
 973                 if (cmd == Q_GETQUOTA &&
 974                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 975                     !oqctl->qc_dqblk.dqb_curspace) {
 976                         struct obd_quotactl *oqctl_tmp;
 977
 978                         OBD_ALLOC_PTR(oqctl_tmp);
 979                         if (oqctl_tmp == NULL)
 980                                 GOTO(out, rc = -ENOMEM);
 981
 982                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 983                         oqctl_tmp->qc_id = oqctl->qc_id;
 984                         oqctl_tmp->qc_type = oqctl->qc_type;
 985
 986                         /* collect space usage from OSTs */
 987                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 988                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 989                         if (!rc || rc == -EREMOTEIO) {
 990                                 oqctl->qc_dqblk.dqb_curspace =
 991                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 992                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 993                         }
 994
 995                         /* collect space & inode usage from MDTs */
 996                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 997                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 998                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 999                         if (!rc || rc == -EREMOTEIO) {
1000                                 oqctl->qc_dqblk.dqb_curspace +=
1001                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1002                                 oqctl->qc_dqblk.dqb_curinodes =
1003                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1004                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1005                         } else {
1006                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1007                         }
1008
1009                         OBD_FREE_PTR(oqctl_tmp);
1010                 }
1011 out:
1012                 QCTL_COPY(qctl, oqctl);
1013                 OBD_FREE_PTR(oqctl);
1014         }
1015
1016         RETURN(rc);
1017 }
1018
1019 static char *
1020 ll_getname(const char __user *filename)
1021 {
1022         int ret = 0, len;
1023         char *tmp = __getname();
1024
1025         if (!tmp)
1026                 return ERR_PTR(-ENOMEM);
1027
1028         len = strncpy_from_user(tmp, filename, PATH_MAX);
1029         if (len == 0)
1030                 ret = -ENOENT;
1031         else if (len > PATH_MAX)
1032                 ret = -ENAMETOOLONG;
1033
1034         if (ret) {
1035                 __putname(tmp);
1036                 tmp =  ERR_PTR(ret);
1037         }
1038         return tmp;
1039 }
1040
1041 #define ll_putname(filename) __putname(filename)
1042
1043 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1044 {
1045         struct inode *inode = file->f_dentry->d_inode;
1046         struct ll_sb_info *sbi = ll_i2sbi(inode);
1047         struct obd_ioctl_data *data;
1048         int rc = 0;
1049         ENTRY;
1050
1051         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1052                PFID(ll_inode2fid(inode)), inode, cmd);
1053
1054         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1055         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1056                 return -ENOTTY;
1057
1058         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1059         switch(cmd) {
1060         case FSFILT_IOC_GETFLAGS:
1061         case FSFILT_IOC_SETFLAGS:
1062                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1063         case FSFILT_IOC_GETVERSION_OLD:
1064         case FSFILT_IOC_GETVERSION:
1065                 RETURN(put_user(inode->i_generation, (int *)arg));
1066         /* We need to special case any other ioctls we want to handle,
1067          * to send them to the MDS/OST as appropriate and to properly
1068          * network encode the arg field.
1069         case FSFILT_IOC_SETVERSION_OLD:
1070         case FSFILT_IOC_SETVERSION:
1071         */
1072         case LL_IOC_GET_MDTIDX: {
1073                 int mdtidx;
1074
1075                 mdtidx = ll_get_mdt_idx(inode);
1076                 if (mdtidx < 0)
1077                         RETURN(mdtidx);
1078
1079                 if (put_user((int)mdtidx, (int*)arg))
1080                         RETURN(-EFAULT);
1081
1082                 return 0;
1083         }
1084         case IOC_MDC_LOOKUP: {
1085                 int namelen, len = 0;
1086                 char *buf = NULL;
1087                 char *filename;
1088
1089                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1090                 if (rc != 0)
1091                         RETURN(rc);
1092                 data = (void *)buf;
1093
1094                 filename = data->ioc_inlbuf1;
1095                 namelen = strlen(filename);
1096                 if (namelen < 1) {
1097                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1098                         GOTO(out_free, rc = -EINVAL);
1099                 }
1100
1101                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1102                 if (rc < 0) {
1103                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1104                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1105                                filename, rc);
1106                         GOTO(out_free, rc);
1107                 }
1108 out_free:
1109                 obd_ioctl_freedata(buf, len);
1110                 return rc;
1111         }
1112         case LL_IOC_LMV_SETSTRIPE: {
1113                 struct lmv_user_md  *lum;
1114                 char            *buf = NULL;
1115                 char            *filename;
1116                 int              namelen = 0;
1117                 int              lumlen = 0;
1118                 umode_t          mode;
1119                 int              len;
1120                 int              rc;
1121
1122                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1123                 if (rc)
1124                         RETURN(rc);
1125
1126                 data = (void *)buf;
1127                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1128                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1129                         GOTO(lmv_out_free, rc = -EINVAL);
1130
1131                 filename = data->ioc_inlbuf1;
1132                 namelen = data->ioc_inllen1;
1133
1134                 if (namelen < 1) {
1135                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1136                         GOTO(lmv_out_free, rc = -EINVAL);
1137                 }
1138                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1139                 lumlen = data->ioc_inllen2;
1140
1141                 if (lum->lum_magic != LMV_USER_MAGIC ||
1142                     lumlen != sizeof(*lum)) {
1143                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1144                                filename, lum->lum_magic, lumlen, -EFAULT);
1145                         GOTO(lmv_out_free, rc = -EINVAL);
1146                 }
1147
1148 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1149                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1150 #else
1151                 mode = data->ioc_type;
1152 #endif
1153                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1154 lmv_out_free:
1155                 obd_ioctl_freedata(buf, len);
1156                 RETURN(rc);
1157
1158         }
1159         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1160                 struct lmv_user_md        lum;
1161                 struct lmv_user_md __user *ulump =
1162                                         (struct lmv_user_md __user *)arg;
1163                 int                       rc;
1164
1165                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1166                         RETURN(-EFAULT);
1167
1168                 if (lum.lum_magic != LMV_USER_MAGIC)
1169                         RETURN(-EINVAL);
1170
1171                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1172
1173                 RETURN(rc);
1174         }
1175         case LL_IOC_LOV_SETSTRIPE: {
1176                 struct lov_user_md_v3 lumv3;
1177                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1178                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1179                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1180
1181                 int set_default = 0;
1182
1183                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1184                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1185                         sizeof(lumv3p->lmm_objects[0]));
1186                 /* first try with v1 which is smaller than v3 */
1187                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1188                         RETURN(-EFAULT);
1189
1190                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1191                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1192                                 RETURN(-EFAULT);
1193                 }
1194
1195                 if (inode->i_sb->s_root == file->f_dentry)
1196                         set_default = 1;
1197
1198                 /* in v1 and v3 cases lumv1 points to data */
1199                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1200
1201                 RETURN(rc);
1202         }
1203         case LL_IOC_LMV_GETSTRIPE: {
1204                 struct lmv_user_md __user *ulmv =
1205                                         (struct lmv_user_md __user *)arg;
1206                 struct lmv_user_md      lum;
1207                 struct ptlrpc_request   *request = NULL;
1208                 union lmv_mds_md        *lmm = NULL;
1209                 int                     lmmsize;
1210                 obd_valid               valid = 0;
1211                 struct lmv_user_md      *tmp = NULL;
1212                 int                     mdt_index;
1213                 int                     lum_size;
1214                 int                     stripe_count;
1215                 int                     i;
1216                 int                     rc;
1217
1218                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1219                         RETURN(-EFAULT);
1220
1221                 /* lum_magic will indicate which stripe the ioctl will like
1222                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1223                  * is for default LMV stripe */
1224                 if (lum.lum_magic == LMV_MAGIC_V1)
1225                         valid |= OBD_MD_MEA;
1226                 else if (lum.lum_magic == LMV_USER_MAGIC)
1227                         valid |= OBD_MD_DEFAULT_MEA;
1228                 else
1229                         RETURN(-EINVAL);
1230
1231                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1232                                       valid);
1233                 if (rc != 0)
1234                         GOTO(finish_req, rc);
1235
1236                 /* Get default LMV EA */
1237                 if (lum.lum_magic == LMV_USER_MAGIC) {
1238                         if (rc != 0)
1239                                 GOTO(finish_req, rc);
1240
1241                         if (lmmsize > sizeof(*ulmv))
1242                                 GOTO(finish_req, rc = -EINVAL);
1243
1244                         if (copy_to_user(ulmv, lmm, lmmsize))
1245                                 GOTO(finish_req, rc = -EFAULT);
1246
1247                         GOTO(finish_req, rc);
1248                 }
1249
1250                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1251                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1252                 OBD_ALLOC(tmp, lum_size);
1253                 if (tmp == NULL)
1254                         GOTO(finish_req, rc = -ENOMEM);
1255
1256                 mdt_index = ll_get_mdt_idx(inode);
1257                 if (mdt_index < 0)
1258                         GOTO(out_tmp, rc = -ENOMEM);
1259
1260                 tmp->lum_magic = LMV_MAGIC_V1;
1261                 tmp->lum_stripe_count = 0;
1262                 tmp->lum_stripe_offset = mdt_index;
1263                 for (i = 0; i < stripe_count; i++) {
1264                         struct lu_fid   fid;
1265
1266                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1267                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1268                         if (mdt_index < 0)
1269                                 GOTO(out_tmp, rc = mdt_index);
1270
1271                         tmp->lum_objects[i].lum_mds = mdt_index;
1272                         tmp->lum_objects[i].lum_fid = fid;
1273                         tmp->lum_stripe_count++;
1274                 }
1275
1276                 if (copy_to_user(ulmv, tmp, lum_size))
1277                         GOTO(out_tmp, rc = -EFAULT);
1278 out_tmp:
1279                 OBD_FREE(tmp, lum_size);
1280 finish_req:
1281                 ptlrpc_req_finished(request);
1282                 return rc;
1283         }
1284
1285         case LL_IOC_REMOVE_ENTRY: {
1286                 char            *filename = NULL;
1287                 int              namelen = 0;
1288                 int              rc;
1289
1290                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1291                  * unsupported server, which might crash the server(LU-2730),
1292                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1293                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1294                  * server will support REINT_RMENTRY XXX*/
1295                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1296                         RETURN(-ENOTSUPP);
1297
1298                 filename = ll_getname((const char *)arg);
1299                 if (IS_ERR(filename))
1300                         RETURN(PTR_ERR(filename));
1301
1302                 namelen = strlen(filename);
1303                 if (namelen < 1)
1304                         GOTO(out_rmdir, rc = -EINVAL);
1305
1306                 rc = ll_rmdir_entry(inode, filename, namelen);
1307 out_rmdir:
1308                 if (filename)
1309                         ll_putname(filename);
1310                 RETURN(rc);
1311         }
1312         case LL_IOC_LOV_SWAP_LAYOUTS:
1313                 RETURN(-EPERM);
1314         case IOC_OBD_STATFS:
1315                 RETURN(ll_obd_statfs(inode, (void *)arg));
1316         case LL_IOC_LOV_GETSTRIPE:
1317         case LL_IOC_MDC_GETINFO:
1318         case IOC_MDC_GETFILEINFO:
1319         case IOC_MDC_GETFILESTRIPE: {
1320                 struct ptlrpc_request *request = NULL;
1321                 struct lov_user_md *lump;
1322                 struct lov_mds_md *lmm = NULL;
1323                 struct mdt_body *body;
1324                 char *filename = NULL;
1325                 int lmmsize;
1326
1327                 if (cmd == IOC_MDC_GETFILEINFO ||
1328                     cmd == IOC_MDC_GETFILESTRIPE) {
1329                         filename = ll_getname((const char *)arg);
1330                         if (IS_ERR(filename))
1331                                 RETURN(PTR_ERR(filename));
1332
1333                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1334                                                       &lmmsize, &request);
1335                 } else {
1336                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1337                                               &request, 0);
1338                 }
1339
1340                 if (request) {
1341                         body = req_capsule_server_get(&request->rq_pill,
1342                                                       &RMF_MDT_BODY);
1343                         LASSERT(body != NULL);
1344                 } else {
1345                         GOTO(out_req, rc);
1346                 }
1347
1348                 if (rc < 0) {
1349                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1350                                                cmd == LL_IOC_MDC_GETINFO))
1351                                 GOTO(skip_lmm, rc = 0);
1352                         else
1353                                 GOTO(out_req, rc);
1354                 }
1355
1356                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1357                     cmd == LL_IOC_LOV_GETSTRIPE) {
1358                         lump = (struct lov_user_md *)arg;
1359                 } else {
1360                         struct lov_user_mds_data *lmdp;
1361                         lmdp = (struct lov_user_mds_data *)arg;
1362                         lump = &lmdp->lmd_lmm;
1363                 }
1364                 if (copy_to_user(lump, lmm, lmmsize)) {
1365                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1366                                 GOTO(out_req, rc = -EFAULT);
1367                         rc = -EOVERFLOW;
1368                 }
1369         skip_lmm:
1370                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1371                         struct lov_user_mds_data *lmdp;
1372                         lstat_t st = { 0 };
1373
1374                         st.st_dev       = inode->i_sb->s_dev;
1375                         st.st_mode      = body->mbo_mode;
1376                         st.st_nlink     = body->mbo_nlink;
1377                         st.st_uid       = body->mbo_uid;
1378                         st.st_gid       = body->mbo_gid;
1379                         st.st_rdev      = body->mbo_rdev;
1380                         st.st_size      = body->mbo_size;
1381                         st.st_blksize   = PAGE_CACHE_SIZE;
1382                         st.st_blocks    = body->mbo_blocks;
1383                         st.st_atime     = body->mbo_atime;
1384                         st.st_mtime     = body->mbo_mtime;
1385                         st.st_ctime     = body->mbo_ctime;
1386                         st.st_ino       = inode->i_ino;
1387
1388                         lmdp = (struct lov_user_mds_data *)arg;
1389                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1390                                 GOTO(out_req, rc = -EFAULT);
1391                 }
1392
1393                 EXIT;
1394         out_req:
1395                 ptlrpc_req_finished(request);
1396                 if (filename)
1397                         ll_putname(filename);
1398                 return rc;
1399         }
1400         case IOC_LOV_GETINFO: {
1401                 struct lov_user_mds_data *lumd;
1402                 struct lov_stripe_md *lsm;
1403                 struct lov_user_md *lum;
1404                 struct lov_mds_md *lmm;
1405                 int lmmsize;
1406                 lstat_t st;
1407
1408                 lumd = (struct lov_user_mds_data *)arg;
1409                 lum = &lumd->lmd_lmm;
1410
1411                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1412                 if (rc)
1413                         RETURN(rc);
1414
1415                 OBD_ALLOC_LARGE(lmm, lmmsize);
1416                 if (lmm == NULL)
1417                         RETURN(-ENOMEM);
1418
1419                 if (copy_from_user(lmm, lum, lmmsize))
1420                         GOTO(free_lmm, rc = -EFAULT);
1421
1422                 switch (lmm->lmm_magic) {
1423                 case LOV_USER_MAGIC_V1:
1424                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1425                                 break;
1426                         /* swab objects first so that stripes num will be sane */
1427                         lustre_swab_lov_user_md_objects(
1428                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1429                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1430                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1431                         break;
1432                 case LOV_USER_MAGIC_V3:
1433                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1434                                 break;
1435                         /* swab objects first so that stripes num will be sane */
1436                         lustre_swab_lov_user_md_objects(
1437                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1438                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1439                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1440                         break;
1441                 default:
1442                         GOTO(free_lmm, rc = -EINVAL);
1443                 }
1444
1445                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1446                 if (rc < 0)
1447                         GOTO(free_lmm, rc = -ENOMEM);
1448
1449                 /* Perform glimpse_size operation. */
1450                 memset(&st, 0, sizeof(st));
1451
1452                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1453                 if (rc)
1454                         GOTO(free_lsm, rc);
1455
1456                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1457                         GOTO(free_lsm, rc = -EFAULT);
1458
1459                 EXIT;
1460         free_lsm:
1461                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1462         free_lmm:
1463                 OBD_FREE_LARGE(lmm, lmmsize);
1464                 return rc;
1465         }
1466         case OBD_IOC_QUOTACHECK: {
1467                 struct obd_quotactl *oqctl;
1468                 int error = 0;
1469
1470                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1471                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1472                         RETURN(-EPERM);
1473
1474                 OBD_ALLOC_PTR(oqctl);
1475                 if (!oqctl)
1476                         RETURN(-ENOMEM);
1477                 oqctl->qc_type = arg;
1478                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1479                 if (rc < 0) {
1480                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1481                         error = rc;
1482                 }
1483
1484                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1485                 if (rc < 0)
1486                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1487
1488                 OBD_FREE_PTR(oqctl);
1489                 return error ?: rc;
1490         }
1491         case OBD_IOC_POLL_QUOTACHECK: {
1492                 struct if_quotacheck *check;
1493
1494                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1495                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1496                         RETURN(-EPERM);
1497
1498                 OBD_ALLOC_PTR(check);
1499                 if (!check)
1500                         RETURN(-ENOMEM);
1501
1502                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1503                                    NULL);
1504                 if (rc) {
1505                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1506                         if (copy_to_user((void *)arg, check,
1507                                              sizeof(*check)))
1508                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1509                         GOTO(out_poll, rc);
1510                 }
1511
1512                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1513                                    NULL);
1514                 if (rc) {
1515                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1516                         if (copy_to_user((void *)arg, check,
1517                                              sizeof(*check)))
1518                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1519                         GOTO(out_poll, rc);
1520                 }
1521         out_poll:
1522                 OBD_FREE_PTR(check);
1523                 RETURN(rc);
1524         }
1525 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
1526         case LL_IOC_QUOTACTL_18: {
1527                 /* copy the old 1.x quota struct for internal use, then copy
1528                  * back into old format struct.  For 1.8 compatibility. */
1529                 struct if_quotactl_18 *qctl_18;
1530                 struct if_quotactl *qctl_20;
1531
1532                 OBD_ALLOC_PTR(qctl_18);
1533                 if (!qctl_18)
1534                         RETURN(-ENOMEM);
1535
1536                 OBD_ALLOC_PTR(qctl_20);
1537                 if (!qctl_20)
1538                         GOTO(out_quotactl_18, rc = -ENOMEM);
1539
1540                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1541                         GOTO(out_quotactl_20, rc = -ENOMEM);
1542
1543                 QCTL_COPY(qctl_20, qctl_18);
1544                 qctl_20->qc_idx = 0;
1545
1546                 /* XXX: dqb_valid was borrowed as a flag to mark that
1547                  *      only mds quota is wanted */
1548                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1549                     qctl_18->qc_dqblk.dqb_valid) {
1550                         qctl_20->qc_valid = QC_MDTIDX;
1551                         qctl_20->qc_dqblk.dqb_valid = 0;
1552                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1553                         qctl_20->qc_valid = QC_UUID;
1554                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1555                 } else {
1556                         qctl_20->qc_valid = QC_GENERAL;
1557                 }
1558
1559                 rc = quotactl_ioctl(sbi, qctl_20);
1560
1561                 if (rc == 0) {
1562                         QCTL_COPY(qctl_18, qctl_20);
1563                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1564
1565                         if (copy_to_user((void *)arg, qctl_18,
1566                                              sizeof(*qctl_18)))
1567                                 rc = -EFAULT;
1568                 }
1569
1570         out_quotactl_20:
1571                 OBD_FREE_PTR(qctl_20);
1572         out_quotactl_18:
1573                 OBD_FREE_PTR(qctl_18);
1574                 RETURN(rc);
1575         }
1576 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */
1577         case OBD_IOC_QUOTACTL: {
1578                 struct if_quotactl *qctl;
1579
1580                 OBD_ALLOC_PTR(qctl);
1581                 if (!qctl)
1582                         RETURN(-ENOMEM);
1583
1584                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1585                         GOTO(out_quotactl, rc = -EFAULT);
1586
1587                 rc = quotactl_ioctl(sbi, qctl);
1588
1589                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1590                         rc = -EFAULT;
1591
1592         out_quotactl:
1593                 OBD_FREE_PTR(qctl);
1594                 RETURN(rc);
1595         }
1596         case OBD_IOC_GETDTNAME:
1597         case OBD_IOC_GETMDNAME:
1598                 RETURN(ll_get_obd_name(inode, cmd, arg));
1599         case LL_IOC_FLUSHCTX:
1600                 RETURN(ll_flush_ctx(inode));
1601 #ifdef CONFIG_FS_POSIX_ACL
1602         case LL_IOC_RMTACL: {
1603             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1604                 inode == inode->i_sb->s_root->d_inode) {
1605                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1606
1607                 LASSERT(fd != NULL);
1608                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1609                 if (!rc)
1610                         fd->fd_flags |= LL_FILE_RMTACL;
1611                 RETURN(rc);
1612             } else
1613                 RETURN(0);
1614         }
1615 #endif
1616         case LL_IOC_GETOBDCOUNT: {
1617                 int count, vallen;
1618                 struct obd_export *exp;
1619
1620                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1621                         RETURN(-EFAULT);
1622
1623                 /* get ost count when count is zero, get mdt count otherwise */
1624                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1625                 vallen = sizeof(count);
1626                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1627                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1628                 if (rc) {
1629                         CERROR("get target count failed: %d\n", rc);
1630                         RETURN(rc);
1631                 }
1632
1633                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1634                         RETURN(-EFAULT);
1635
1636                 RETURN(0);
1637         }
1638         case LL_IOC_PATH2FID:
1639                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1640                                      sizeof(struct lu_fid)))
1641                         RETURN(-EFAULT);
1642                 RETURN(0);
1643         case LL_IOC_GET_CONNECT_FLAGS: {
1644                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1645         }
1646         case OBD_IOC_CHANGELOG_SEND:
1647         case OBD_IOC_CHANGELOG_CLEAR:
1648                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1649                                     sizeof(struct ioc_changelog));
1650                 RETURN(rc);
1651         case OBD_IOC_FID2PATH:
1652                 RETURN(ll_fid2path(inode, (void *)arg));
1653         case LL_IOC_FID2MDTIDX: {
1654                 struct obd_export *exp = ll_i2mdexp(inode);
1655                 struct lu_fid     fid;
1656                 __u32             index;
1657
1658                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1659                                    sizeof(fid)))
1660                         RETURN(-EFAULT);
1661
1662                 /* Call mdc_iocontrol */
1663                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1664                                    &index);
1665                 if (rc != 0)
1666                         RETURN(rc);
1667
1668                 RETURN(index);
1669         }
1670         case LL_IOC_HSM_REQUEST: {
1671                 struct hsm_user_request *hur;
1672                 ssize_t                  totalsize;
1673
1674                 OBD_ALLOC_PTR(hur);
1675                 if (hur == NULL)
1676                         RETURN(-ENOMEM);
1677
1678                 /* We don't know the true size yet; copy the fixed-size part */
1679                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1680                         OBD_FREE_PTR(hur);
1681                         RETURN(-EFAULT);
1682                 }
1683
1684                 /* Compute the whole struct size */
1685                 totalsize = hur_len(hur);
1686                 OBD_FREE_PTR(hur);
1687                 if (totalsize < 0)
1688                         RETURN(-E2BIG);
1689
1690                 /* Final size will be more than double totalsize */
1691                 if (totalsize >= MDS_MAXREQSIZE / 3)
1692                         RETURN(-E2BIG);
1693
1694                 OBD_ALLOC_LARGE(hur, totalsize);
1695                 if (hur == NULL)
1696                         RETURN(-ENOMEM);
1697
1698                 /* Copy the whole struct */
1699                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1700                         OBD_FREE_LARGE(hur, totalsize);
1701                         RETURN(-EFAULT);
1702                 }
1703
1704                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1705                         const struct lu_fid *fid;
1706                         struct inode *f;
1707                         int i;
1708
1709                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1710                                 fid = &hur->hur_user_item[i].hui_fid;
1711                                 f = search_inode_for_lustre(inode->i_sb, fid);
1712                                 if (IS_ERR(f)) {
1713                                         rc = PTR_ERR(f);
1714                                         break;
1715                                 }
1716
1717                                 rc = ll_hsm_release(f);
1718                                 iput(f);
1719                                 if (rc != 0)
1720                                         break;
1721                         }
1722                 } else {
1723                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1724                                            hur, NULL);
1725                 }
1726
1727                 OBD_FREE_LARGE(hur, totalsize);
1728
1729                 RETURN(rc);
1730         }
1731         case LL_IOC_HSM_PROGRESS: {
1732                 struct hsm_progress_kernel      hpk;
1733                 struct hsm_progress             hp;
1734
1735                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1736                         RETURN(-EFAULT);
1737
1738                 hpk.hpk_fid = hp.hp_fid;
1739                 hpk.hpk_cookie = hp.hp_cookie;
1740                 hpk.hpk_extent = hp.hp_extent;
1741                 hpk.hpk_flags = hp.hp_flags;
1742                 hpk.hpk_errval = hp.hp_errval;
1743                 hpk.hpk_data_version = 0;
1744
1745                 /* File may not exist in Lustre; all progress
1746                  * reported to Lustre root */
1747                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1748                                    NULL);
1749                 RETURN(rc);
1750         }
1751         case LL_IOC_HSM_CT_START:
1752                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1753                         RETURN(-EPERM);
1754
1755                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1756                                     sizeof(struct lustre_kernelcomm));
1757                 RETURN(rc);
1758
1759         case LL_IOC_HSM_COPY_START: {
1760                 struct hsm_copy *copy;
1761                 int              rc;
1762
1763                 OBD_ALLOC_PTR(copy);
1764                 if (copy == NULL)
1765                         RETURN(-ENOMEM);
1766                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1767                         OBD_FREE_PTR(copy);
1768                         RETURN(-EFAULT);
1769                 }
1770
1771                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1772                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1773                         rc = -EFAULT;
1774
1775                 OBD_FREE_PTR(copy);
1776                 RETURN(rc);
1777         }
1778         case LL_IOC_HSM_COPY_END: {
1779                 struct hsm_copy *copy;
1780                 int              rc;
1781
1782                 OBD_ALLOC_PTR(copy);
1783                 if (copy == NULL)
1784                         RETURN(-ENOMEM);
1785                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1786                         OBD_FREE_PTR(copy);
1787                         RETURN(-EFAULT);
1788                 }
1789
1790                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1791                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1792                         rc = -EFAULT;
1793
1794                 OBD_FREE_PTR(copy);
1795                 RETURN(rc);
1796         }
1797         case LL_IOC_MIGRATE: {
1798                 char            *buf = NULL;
1799                 const char      *filename;
1800                 int             namelen = 0;
1801                 int             len;
1802                 int             rc;
1803                 int             mdtidx;
1804
1805                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1806                 if (rc < 0)
1807                         RETURN(rc);
1808
1809                 data = (struct obd_ioctl_data *)buf;
1810                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1811                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1812                         GOTO(migrate_free, rc = -EINVAL);
1813
1814                 filename = data->ioc_inlbuf1;
1815                 namelen = data->ioc_inllen1;
1816                 /* \0 is packed at the end of filename */
1817                 if (namelen < 1 || namelen != strlen(filename) + 1)
1818                         GOTO(migrate_free, rc = -EINVAL);
1819
1820                 if (data->ioc_inllen2 != sizeof(mdtidx))
1821                         GOTO(migrate_free, rc = -EINVAL);
1822                 mdtidx = *(int *)data->ioc_inlbuf2;
1823
1824                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1825 migrate_free:
1826                 obd_ioctl_freedata(buf, len);
1827
1828                 RETURN(rc);
1829         }
1830         default:
1831                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1832                                      (void *)arg));
1833         }
1834 }
1835
1836 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1837 {
1838         struct inode *inode = file->f_mapping->host;
1839         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1840         struct ll_sb_info *sbi = ll_i2sbi(inode);
1841         int api32 = ll_need_32bit_api(sbi);
1842         loff_t ret = -EINVAL;
1843         ENTRY;
1844
1845         mutex_lock(&inode->i_mutex);
1846         switch (origin) {
1847                 case SEEK_SET:
1848                         break;
1849                 case SEEK_CUR:
1850                         offset += file->f_pos;
1851                         break;
1852                 case SEEK_END:
1853                         if (offset > 0)
1854                                 GOTO(out, ret);
1855                         if (api32)
1856                                 offset += LL_DIR_END_OFF_32BIT;
1857                         else
1858                                 offset += LL_DIR_END_OFF;
1859                         break;
1860                 default:
1861                         GOTO(out, ret);
1862         }
1863
1864         if (offset >= 0 &&
1865             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1866              (!api32 && offset <= LL_DIR_END_OFF))) {
1867                 if (offset != file->f_pos) {
1868                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1869                             (!api32 && offset == LL_DIR_END_OFF))
1870                                 fd->lfd_pos = MDS_DIR_END_OFF;
1871                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1872                                 fd->lfd_pos = offset << 32;
1873                         else
1874                                 fd->lfd_pos = offset;
1875                         file->f_pos = offset;
1876                         file->f_version = 0;
1877                 }
1878                 ret = offset;
1879         }
1880         GOTO(out, ret);
1881
1882 out:
1883         mutex_unlock(&inode->i_mutex);
1884         return ret;
1885 }
1886
1887 static int ll_dir_open(struct inode *inode, struct file *file)
1888 {
1889         ENTRY;
1890         RETURN(ll_file_open(inode, file));
1891 }
1892
1893 static int ll_dir_release(struct inode *inode, struct file *file)
1894 {
1895         ENTRY;
1896         RETURN(ll_file_release(inode, file));
1897 }
1898
1899 const struct file_operations ll_dir_operations = {
1900         .llseek         = ll_dir_seek,
1901         .open           = ll_dir_open,
1902         .release        = ll_dir_release,
1903         .read           = generic_read_dir,
1904 #ifdef HAVE_DIR_CONTEXT
1905         .iterate        = ll_iterate,
1906 #else
1907         .readdir        = ll_readdir,
1908 #endif
1909         .unlocked_ioctl = ll_dir_ioctl,
1910         .fsync          = ll_fsync,
1911 };