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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/dir.c
  33  *
  34  * Directory code for lustre client.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/mm.h>
  40 #include <linux/version.h>
  41 #include <linux/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  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
  53 #include <obd_support.h>
  54 #include <obd_class.h>
  55 #include <lustre_ioctl.h>
  56 #include <lustre_lib.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include <lustre_kernelcomm.h>
  60 #include <lustre_swab.h>
  61
  62 #include "llite_internal.h"
  63
  64 /*
  65  * (new) readdir implementation overview.
  66  *
  67  * Original lustre readdir implementation cached exact copy of raw directory
  68  * pages on the client. These pages were indexed in client page cache by
  69  * logical offset in the directory file. This design, while very simple and
  70  * intuitive had some inherent problems:
  71  *
  72  *     . it implies that byte offset to the directory entry serves as a
  73  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  74  *     ext3/htree directory entries may move due to splits, and more
  75  *     importantly,
  76  *
  77  *     . it is incompatible with the design of split directories for cmd3,
  78  *     that assumes that names are distributed across nodes based on their
  79  *     hash, and so readdir should be done in hash order.
  80  *
  81  * New readdir implementation does readdir in hash order, and uses hash of a
  82  * file name as a telldir/seekdir cookie. This led to number of complications:
  83  *
  84  *     . hash is not unique, so it cannot be used to index cached directory
  85  *     pages on the client (note, that it requires a whole pageful of hash
  86  *     collided entries to cause two pages to have identical hashes);
  87  *
  88  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  89  *     entry cookie. ext3/htree has the same problem and lustre implementation
  90  *     mimics their solution: seekdir(hash) positions directory at the first
  91  *     entry with the given hash.
  92  *
  93  * Client side.
  94  *
  95  * 0. caching
  96  *
  97  * Client caches directory pages using hash of the first entry as an index. As
  98  * noted above hash is not unique, so this solution doesn't work as is:
  99  * special processing is needed for "page hash chains" (i.e., sequences of
 100  * pages filled with entries all having the same hash value).
 101  *
 102  * First, such chains have to be detected. To this end, server returns to the
 103  * client the hash of the first entry on the page next to one returned. When
 104  * client detects that this hash is the same as hash of the first entry on the
 105  * returned page, page hash collision has to be handled. Pages in the
 106  * hash chain, except first one, are termed "overflow pages".
 107  *
 108  * Solution to index uniqueness problem is to not cache overflow
 109  * pages. Instead, when page hash collision is detected, all overflow pages
 110  * from emerging chain are immediately requested from the server and placed in
 111  * a special data structure (struct ll_dir_chain). This data structure is used
 112  * by ll_readdir() to process entries from overflow pages. When readdir
 113  * invocation finishes, overflow pages are discarded. If page hash collision
 114  * chain weren't completely processed, next call to readdir will again detect
 115  * page hash collision, again read overflow pages in, process next portion of
 116  * entries and again discard the pages. This is not as wasteful as it looks,
 117  * because, given reasonable hash, page hash collisions are extremely rare.
 118  *
 119  * 1. directory positioning
 120  *
 121  * When seekdir(hash) is called, original
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  *
 130  * Server.
 131  *
 132  * identification of and access to overflow pages
 133  *
 134  * page format
 135  *
 136  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 137  * a header lu_dirpage which describes the start/end hash, and whether this
 138  * page is empty (contains no dir entry) or hash collide with next page.
 139  * After client receives reply, several pages will be integrated into dir page
 140  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 141  * lu_dirpage for this integrated page will be adjusted. See
 142  * mdc_adjust_dirpages().
 143  *
 144  */
 145 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 146                              __u64 offset, struct ll_dir_chain *chain)
 147 {
 148         struct md_callback      cb_op;
 149         struct page             *page;
 150         int                     rc;
 151
 152         cb_op.md_blocking_ast = ll_md_blocking_ast;
 153         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 154         if (rc != 0)
 155                 return ERR_PTR(rc);
 156
 157         return page;
 158 }
 159
 160 void ll_release_page(struct inode *inode, struct page *page,
 161                      bool remove)
 162 {
 163         kunmap(page);
 164
 165         /* Always remove the page for striped dir, because the page is
 166          * built from temporarily in LMV layer */
 167         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 168             ll_i2info(inode)->lli_lsm_md != NULL) {
 169                 __free_page(page);
 170                 return;
 171         }
 172
 173         if (remove) {
 174                 lock_page(page);
 175                 if (likely(page->mapping != NULL))
 176                         truncate_complete_page(page->mapping, page);
 177                 unlock_page(page);
 178         }
 179         put_page(page);
 180 }
 181
 182 /**
 183  * return IF_* type for given lu_dirent entry.
 184  * IF_* flag shld be converted to particular OS file type in
 185  * platform llite module.
 186  */
 187 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 188 {
 189         u16 type = 0;
 190         struct luda_type *lt;
 191         int len = 0;
 192
 193         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 194                 const unsigned align = sizeof(struct luda_type) - 1;
 195
 196                 len = le16_to_cpu(ent->lde_namelen);
 197                 len = (len + align) & ~align;
 198                 lt = (void *)ent->lde_name + len;
 199                 type = IFTODT(le16_to_cpu(lt->lt_type));
 200         }
 201
 202         return type;
 203 }
 204
 205 #ifdef HAVE_DIR_CONTEXT
 206 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 207                 struct dir_context *ctx)
 208 {
 209 #else
 210 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 211                 void *cookie, filldir_t filldir)
 212 {
 213 #endif
 214         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 215         __u64                 pos        = *ppos;
 216         bool                  is_api32 = ll_need_32bit_api(sbi);
 217         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 218         struct page          *page;
 219         struct ll_dir_chain   chain;
 220         bool                  done = false;
 221         int                   rc = 0;
 222         ENTRY;
 223
 224         ll_dir_chain_init(&chain);
 225
 226         page = ll_get_dir_page(inode, op_data, pos, &chain);
 227
 228         while (rc == 0 && !done) {
 229                 struct lu_dirpage *dp;
 230                 struct lu_dirent  *ent;
 231                 __u64 hash;
 232                 __u64 next;
 233
 234                 if (IS_ERR(page)) {
 235                         rc = PTR_ERR(page);
 236                         break;
 237                 }
 238
 239                 hash = MDS_DIR_END_OFF;
 240                 dp = page_address(page);
 241                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 242                      ent = lu_dirent_next(ent)) {
 243                         __u16          type;
 244                         int            namelen;
 245                         struct lu_fid  fid;
 246                         __u64          lhash;
 247                         __u64          ino;
 248
 249                         hash = le64_to_cpu(ent->lde_hash);
 250                         if (hash < pos)
 251                                 /*
 252                                  * Skip until we find target hash
 253                                  * value.
 254                                  */
 255                                 continue;
 256
 257                         namelen = le16_to_cpu(ent->lde_namelen);
 258                         if (namelen == 0)
 259                                 /*
 260                                  * Skip dummy record.
 261                                  */
 262                                 continue;
 263
 264                         if (is_api32 && is_hash64)
 265                                 lhash = hash >> 32;
 266                         else
 267                                 lhash = hash;
 268                         fid_le_to_cpu(&fid, &ent->lde_fid);
 269                         ino = cl_fid_build_ino(&fid, is_api32);
 270                         type = ll_dirent_type_get(ent);
 271                         /* For 'll_nfs_get_name_filldir()', it will try
 272                          * to access the 'ent' through its 'lde_name',
 273                          * so the parameter 'name' for 'filldir()' must
 274                          * be part of the 'ent'. */
 275 #ifdef HAVE_DIR_CONTEXT
 276                         ctx->pos = lhash;
 277                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 278                                          type);
 279 #else
 280                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 281                                        ino, type);
 282 #endif
 283                 }
 284
 285                 if (done) {
 286                         pos = hash;
 287                         ll_release_page(inode, page, false);
 288                         break;
 289                 }
 290
 291                 next = le64_to_cpu(dp->ldp_hash_end);
 292                 pos = next;
 293                 if (pos == MDS_DIR_END_OFF) {
 294                         /*
 295                          * End of directory reached.
 296                          */
 297                         done = 1;
 298                         ll_release_page(inode, page, false);
 299                 } else {
 300                         /*
 301                          * Normal case: continue to the next
 302                          * page.
 303                          */
 304                         ll_release_page(inode, page,
 305                                         le32_to_cpu(dp->ldp_flags) &
 306                                         LDF_COLLIDE);
 307                         next = pos;
 308                         page = ll_get_dir_page(inode, op_data, pos,
 309                                                &chain);
 310                 }
 311         }
 312 #ifdef HAVE_DIR_CONTEXT
 313         ctx->pos = pos;
 314 #else
 315         *ppos = pos;
 316 #endif
 317         ll_dir_chain_fini(&chain);
 318         RETURN(rc);
 319 }
 320
 321 #ifdef HAVE_DIR_CONTEXT
 322 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 323 #else
 324 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 325 #endif
 326 {
 327         struct inode            *inode  = file_inode(filp);
 328         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 329         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 330         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 331         int                     api32   = ll_need_32bit_api(sbi);
 332         struct md_op_data       *op_data;
 333         __u64                   pos;
 334         int                     rc;
 335         ENTRY;
 336
 337         if (lfd != NULL)
 338                 pos = lfd->lfd_pos;
 339         else
 340                 pos = 0;
 341
 342         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 343                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 344                inode, (unsigned long)pos, i_size_read(inode), api32);
 345
 346         if (pos == MDS_DIR_END_OFF)
 347                 /*
 348                  * end-of-file.
 349                  */
 350                 GOTO(out, rc = 0);
 351
 352         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 353                                      LUSTRE_OPC_ANY, inode);
 354         if (IS_ERR(op_data))
 355                 GOTO(out, rc = PTR_ERR(op_data));
 356
 357         if (unlikely(op_data->op_mea1 != NULL)) {
 358                 /* This is only needed for striped dir to fill ..,
 359                  * see lmv_read_entry */
 360                 if (file_dentry(filp)->d_parent != NULL &&
 361                     file_dentry(filp)->d_parent->d_inode != NULL) {
 362                         __u64 ibits = MDS_INODELOCK_UPDATE;
 363                         struct inode *parent =
 364                                 file_dentry(filp)->d_parent->d_inode;
 365
 366                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 367                                 op_data->op_fid3 = *ll_inode2fid(parent);
 368                 }
 369
 370                 /* If it can not find in cache, do lookup .. on the master
 371                  * object */
 372                 if (fid_is_zero(&op_data->op_fid3)) {
 373                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 374                         if (rc != 0) {
 375                                 ll_finish_md_op_data(op_data);
 376                                 RETURN(rc);
 377                         }
 378                 }
 379         }
 380         op_data->op_max_pages = sbi->ll_md_brw_pages;
 381 #ifdef HAVE_DIR_CONTEXT
 382         ctx->pos = pos;
 383         rc = ll_dir_read(inode, &pos, op_data, ctx);
 384         pos = ctx->pos;
 385 #else
 386         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 387 #endif
 388         if (lfd != NULL)
 389                 lfd->lfd_pos = pos;
 390
 391         if (pos == MDS_DIR_END_OFF) {
 392                 if (api32)
 393                         pos = LL_DIR_END_OFF_32BIT;
 394                 else
 395                         pos = LL_DIR_END_OFF;
 396         } else {
 397                 if (api32 && hash64)
 398                         pos = pos >> 32;
 399         }
 400 #ifdef HAVE_DIR_CONTEXT
 401         ctx->pos = pos;
 402 #else
 403         filp->f_pos = pos;
 404 #endif
 405         ll_finish_md_op_data(op_data);
 406         filp->f_version = inode->i_version;
 407
 408 out:
 409         if (!rc)
 410                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 411
 412         RETURN(rc);
 413 }
 414
 415 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 417 {
 418         struct mgs_send_param *msp;
 419         int rc = 0;
 420
 421         OBD_ALLOC_PTR(msp);
 422         if (!msp)
 423                 return -ENOMEM;
 424
 425         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 426         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 427                                 sizeof(struct mgs_send_param), msp, NULL);
 428         if (rc)
 429                 CERROR("Failed to set parameter: %d\n", rc);
 430         OBD_FREE_PTR(msp);
 431
 432         return rc;
 433 }
 434 #endif
 435
 436 /**
 437  * Create striped directory with specified stripe(@lump)
 438  *
 439  * \param[in] dparent   the parent of the directory.
 440  * \param[in] lump      the specified stripes.
 441  * \param[in] dirname   the name of the directory.
 442  * \param[in] mode      the specified mode of the directory.
 443  *
 444  * \retval              =0 if striped directory is being created successfully.
 445  *                      <0 if the creation is failed.
 446  */
 447 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 448                                const char *dirname, umode_t mode)
 449 {
 450         struct inode *parent = dparent->d_inode;
 451         struct ptlrpc_request *request = NULL;
 452         struct md_op_data *op_data;
 453         struct ll_sb_info *sbi = ll_i2sbi(parent);
 454         struct inode *inode = NULL;
 455         struct dentry dentry = {
 456                 .d_parent = dparent,
 457                 .d_name = {
 458                         .name = dirname,
 459                         .len = strlen(dirname),
 460                         .hash = full_name_hash(dirname, strlen(dirname)),
 461                 },
 462         };
 463         int err;
 464         ENTRY;
 465
 466         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 467                 RETURN(-EINVAL);
 468
 469         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 470                "stripe_offset %d, stripe_count: %u\n",
 471                PFID(ll_inode2fid(parent)), parent, dirname,
 472                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 473
 474         if (lump->lum_stripe_count > 1 &&
 475             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 476                 RETURN(-EINVAL);
 477
 478         if (IS_DEADDIR(parent) &&
 479             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 480                 RETURN(-ENOENT);
 481
 482         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 483                 lustre_swab_lmv_user_md(lump);
 484
 485         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 486                 mode &= ~current_umask();
 487         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 488         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 489                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 490                                      lump);
 491         if (IS_ERR(op_data))
 492                 RETURN(PTR_ERR(op_data));
 493
 494         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 495                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 496                  * to determine the security context for the file. So our fake
 497                  * dentry should be real enough for this purpose. */
 498                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 499                                               &op_data->op_file_secctx_name,
 500                                               &op_data->op_file_secctx,
 501                                               &op_data->op_file_secctx_size);
 502                 if (err < 0)
 503                         GOTO(out_op_data, err);
 504         }
 505
 506         op_data->op_cli_flags |= CLI_SET_MEA;
 507         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 508                         from_kuid(&init_user_ns, current_fsuid()),
 509                         from_kgid(&init_user_ns, current_fsgid()),
 510                         cfs_curproc_cap_pack(), 0, &request);
 511         if (err)
 512                 GOTO(out_request, err);
 513
 514         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 515
 516         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 517         if (err)
 518                 GOTO(out_inode, err);
 519
 520         dentry.d_inode = inode;
 521
 522         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
 523                 err = ll_inode_init_security(&dentry, inode, parent);
 524                 if (err)
 525                         GOTO(out_inode, err);
 526         }
 527
 528 out_inode:
 529         if (inode != NULL)
 530                 iput(inode);
 531 out_request:
 532         ptlrpc_req_finished(request);
 533 out_op_data:
 534         ll_finish_md_op_data(op_data);
 535
 536         return err;
 537 }
 538
 539 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 540                      int set_default)
 541 {
 542         struct ll_sb_info *sbi = ll_i2sbi(inode);
 543         struct md_op_data *op_data;
 544         struct ptlrpc_request *req = NULL;
 545         int rc = 0;
 546 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 547         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 548         struct obd_device *mgc = lsi->lsi_mgc;
 549 #endif
 550         int lum_size;
 551         ENTRY;
 552
 553         if (lump != NULL) {
 554                 /*
 555                  * This is coming from userspace, so should be in
 556                  * local endian.  But the MDS would like it in little
 557                  * endian, so we swab it before we send it.
 558                  */
 559                 switch (lump->lmm_magic) {
 560                 case LOV_USER_MAGIC_V1: {
 561                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 562                                 lustre_swab_lov_user_md_v1(lump);
 563                         lum_size = sizeof(struct lov_user_md_v1);
 564                         break;
 565                 }
 566                 case LOV_USER_MAGIC_V3: {
 567                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 568                                 lustre_swab_lov_user_md_v3(
 569                                         (struct lov_user_md_v3 *)lump);
 570                         lum_size = sizeof(struct lov_user_md_v3);
 571                         break;
 572                 }
 573                 case LOV_USER_MAGIC_COMP_V1: {
 574                         if (lump->lmm_magic !=
 575                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 576                                 lustre_swab_lov_comp_md_v1(
 577                                         (struct lov_comp_md_v1 *)lump);
 578                         lum_size = le32_to_cpu(
 579                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 580                         break;
 581                 }
 582                 case LMV_USER_MAGIC: {
 583                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 584                                 lustre_swab_lmv_user_md(
 585                                         (struct lmv_user_md *)lump);
 586                         lum_size = sizeof(struct lmv_user_md);
 587                         break;
 588                 }
 589                 default: {
 590                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 591                                         " %#08x != %#08x nor %#08x\n",
 592                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 593                                         LOV_USER_MAGIC_V3);
 594                         RETURN(-EINVAL);
 595                 }
 596                 }
 597         } else {
 598                 lum_size = sizeof(struct lov_user_md_v1);
 599         }
 600
 601         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 602                                      LUSTRE_OPC_ANY, NULL);
 603         if (IS_ERR(op_data))
 604                 RETURN(PTR_ERR(op_data));
 605
 606         /* swabbing is done in lov_setstripe() on server side */
 607         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 608         ll_finish_md_op_data(op_data);
 609         ptlrpc_req_finished(req);
 610         if (rc)
 611                 RETURN(rc);
 612
 613 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 614         /*
 615          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 616          * and it's stored into system config for backward compatibility.
 617          *
 618          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 619          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 620          * need the make the distiction between the 2 versions
 621          */
 622         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 623             (lump == NULL ||
 624              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 625              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 626                 char *param = NULL;
 627                 char *buf;
 628
 629                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 630                 if (param == NULL)
 631                         GOTO(end, rc = -ENOMEM);
 632
 633                 buf = param;
 634                 /* Get fsname and assume devname to be -MDT0000. */
 635                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 636                 strcat(buf, "-MDT0000.lov");
 637                 buf += strlen(buf);
 638
 639                 /* Set root stripesize */
 640                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 641                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 642                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 643                 if (rc)
 644                         GOTO(end, rc);
 645
 646                 /* Set root stripecount */
 647                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 648                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 649                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 650                 if (rc)
 651                         GOTO(end, rc);
 652
 653                 /* Set root stripeoffset */
 654                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 655                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 656                                 (typeof(lump->lmm_stripe_offset))(-1));
 657                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 658
 659 end:
 660                 if (param != NULL)
 661                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 662         }
 663 #endif
 664         RETURN(rc);
 665 }
 666
 667 /**
 668  * This function will be used to get default LOV/LMV/Default LMV
 669  * @valid will be used to indicate which stripe it will retrieve
 670  *      OBD_MD_MEA              LMV stripe EA
 671  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 672  *      otherwise               Default LOV EA.
 673  * Each time, it can only retrieve 1 stripe EA
 674  **/
 675 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 676                      struct ptlrpc_request **request, u64 valid)
 677 {
 678         struct ll_sb_info *sbi = ll_i2sbi(inode);
 679         struct mdt_body   *body;
 680         struct lov_mds_md *lmm = NULL;
 681         struct ptlrpc_request *req = NULL;
 682         int rc, lmm_size;
 683         struct md_op_data *op_data;
 684         ENTRY;
 685
 686         rc = ll_get_default_mdsize(sbi, &lmm_size);
 687         if (rc)
 688                 RETURN(rc);
 689
 690         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 691                                      0, lmm_size, LUSTRE_OPC_ANY,
 692                                      NULL);
 693         if (IS_ERR(op_data))
 694                 RETURN(PTR_ERR(op_data));
 695
 696         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 697         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 698         ll_finish_md_op_data(op_data);
 699         if (rc < 0) {
 700                 CDEBUG(D_INFO, "md_getattr failed on inode "
 701                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 702                 GOTO(out, rc);
 703         }
 704
 705         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 706         LASSERT(body != NULL);
 707
 708         lmm_size = body->mbo_eadatasize;
 709
 710         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 711             lmm_size == 0) {
 712                 GOTO(out, rc = -ENODATA);
 713         }
 714
 715         lmm = req_capsule_server_sized_get(&req->rq_pill,
 716                                            &RMF_MDT_MD, lmm_size);
 717         LASSERT(lmm != NULL);
 718
 719         /*
 720          * This is coming from the MDS, so is probably in
 721          * little endian.  We convert it to host endian before
 722          * passing it to userspace.
 723          */
 724         /* We don't swab objects for directories */
 725         switch (le32_to_cpu(lmm->lmm_magic)) {
 726         case LOV_MAGIC_V1:
 727                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 728                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 729                 break;
 730         case LOV_MAGIC_V3:
 731                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 732                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 733                 break;
 734         case LOV_MAGIC_COMP_V1:
 735                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 736                         lustre_swab_lov_comp_md_v1(
 737                                         (struct lov_comp_md_v1 *)lmm);
 738                 break;
 739         case LMV_MAGIC_V1:
 740                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 741                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 742                 break;
 743         case LMV_USER_MAGIC:
 744                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 745                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 746                 break;
 747         default:
 748                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 749                 rc = -EPROTO;
 750         }
 751 out:
 752         *plmm = lmm;
 753         *plmm_size = lmm_size;
 754         *request = req;
 755         return rc;
 756 }
 757
 758 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 759 {
 760         struct md_op_data       *op_data;
 761         int                     rc;
 762         int                     mdt_index;
 763         ENTRY;
 764
 765         OBD_ALLOC_PTR(op_data);
 766         if (op_data == NULL)
 767                 RETURN(-ENOMEM);
 768
 769         op_data->op_flags |= MF_GET_MDT_IDX;
 770         op_data->op_fid1 = *fid;
 771         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 772         mdt_index = op_data->op_mds;
 773         OBD_FREE_PTR(op_data);
 774         if (rc < 0)
 775                 RETURN(rc);
 776
 777         RETURN(mdt_index);
 778 }
 779
 780 /*
 781  *  Get MDT index for the inode.
 782  */
 783 int ll_get_mdt_idx(struct inode *inode)
 784 {
 785         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 786 }
 787
 788 /**
 789  * Generic handler to do any pre-copy work.
 790  *
 791  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 792  * first information for it that real work has started.
 793  *
 794  * Moreover, for a ARCHIVE request, it will sample the file data version and
 795  * store it in \a copy.
 796  *
 797  * \return 0 on success.
 798  */
 799 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 800 {
 801         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 802         struct hsm_progress_kernel       hpk;
 803         int                              rc = 0;
 804         int                              rc2;
 805         ENTRY;
 806
 807         /* Forge a hsm_progress based on data from copy. */
 808         hpk.hpk_fid = copy->hc_hai.hai_fid;
 809         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 810         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 811         hpk.hpk_extent.length = 0;
 812         hpk.hpk_flags = 0;
 813         hpk.hpk_errval = 0;
 814         hpk.hpk_data_version = 0;
 815
 816
 817         /* For archive request, we need to read the current file version. */
 818         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 819                 struct inode    *inode;
 820                 __u64            data_version = 0;
 821
 822                 /* Get inode for this fid */
 823                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 824                 if (IS_ERR(inode)) {
 825                         hpk.hpk_flags |= HP_FLAG_RETRY;
 826                         /* hpk_errval is >= 0 */
 827                         hpk.hpk_errval = -PTR_ERR(inode);
 828                         GOTO(progress, rc = PTR_ERR(inode));
 829                 }
 830
 831                 /* Read current file data version */
 832                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 833                 iput(inode);
 834                 if (rc != 0) {
 835                         CDEBUG(D_HSM, "Could not read file data version of "
 836                                       DFID" (rc = %d). Archive request ("
 837                                       "%#llx) could not be done.\n",
 838                                       PFID(&copy->hc_hai.hai_fid), rc,
 839                                       copy->hc_hai.hai_cookie);
 840                         hpk.hpk_flags |= HP_FLAG_RETRY;
 841                         /* hpk_errval must be >= 0 */
 842                         hpk.hpk_errval = -rc;
 843                         GOTO(progress, rc);
 844                 }
 845
 846                 /* Store in the hsm_copy for later copytool use.
 847                  * Always modified even if no lsm. */
 848                 copy->hc_data_version = data_version;
 849         }
 850
 851 progress:
 852         /* On error, the request should be considered as completed */
 853         if (hpk.hpk_errval > 0)
 854                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 855
 856         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 857                             &hpk, NULL);
 858
 859         /* Return first error */
 860         RETURN(rc != 0 ? rc : rc2);
 861 }
 862
 863 /**
 864  * Generic handler to do any post-copy work.
 865  *
 866  * It will send the last hsm_progress update to coordinator to inform it
 867  * that copy is finished and whether it was successful or not.
 868  *
 869  * Moreover,
 870  * - for ARCHIVE request, it will sample the file data version and compare it
 871  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 872  *   will be considered as failed.
 873  * - for RESTORE request, it will sample the file data version and send it to
 874  *   coordinator which is useful if the file was imported as 'released'.
 875  *
 876  * \return 0 on success.
 877  */
 878 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 879 {
 880         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 881         struct hsm_progress_kernel       hpk;
 882         int                              rc = 0;
 883         int                              rc2;
 884         ENTRY;
 885
 886         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 887         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 888          * initialized if copy_end was called with copy == NULL.
 889          */
 890
 891         /* Forge a hsm_progress based on data from copy. */
 892         hpk.hpk_fid = copy->hc_hai.hai_fid;
 893         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 894         hpk.hpk_extent = copy->hc_hai.hai_extent;
 895         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 896         hpk.hpk_errval = copy->hc_errval;
 897         hpk.hpk_data_version = 0;
 898
 899         /* For archive request, we need to check the file data was not changed.
 900          *
 901          * For restore request, we need to send the file data version, this is
 902          * useful when the file was created using hsm_import.
 903          */
 904         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 905              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 906             (copy->hc_errval == 0)) {
 907                 struct inode    *inode;
 908                 __u64            data_version = 0;
 909
 910                 /* Get lsm for this fid */
 911                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 912                 if (IS_ERR(inode)) {
 913                         hpk.hpk_flags |= HP_FLAG_RETRY;
 914                         /* hpk_errval must be >= 0 */
 915                         hpk.hpk_errval = -PTR_ERR(inode);
 916                         GOTO(progress, rc = PTR_ERR(inode));
 917                 }
 918
 919                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 920                 iput(inode);
 921                 if (rc) {
 922                         CDEBUG(D_HSM, "Could not read file data version. "
 923                                       "Request could not be confirmed.\n");
 924                         if (hpk.hpk_errval == 0)
 925                                 hpk.hpk_errval = -rc;
 926                         GOTO(progress, rc);
 927                 }
 928
 929                 /* Store in the hsm_copy for later copytool use.
 930                  * Always modified even if no lsm. */
 931                 hpk.hpk_data_version = data_version;
 932
 933                 /* File could have been stripped during archiving, so we need
 934                  * to check anyway. */
 935                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 936                     (copy->hc_data_version != data_version)) {
 937                         CDEBUG(D_HSM, "File data version mismatched. "
 938                               "File content was changed during archiving. "
 939                                DFID", start:%#llx current:%#llx\n",
 940                                PFID(&copy->hc_hai.hai_fid),
 941                                copy->hc_data_version, data_version);
 942                         /* File was changed, send error to cdt. Do not ask for
 943                          * retry because if a file is modified frequently,
 944                          * the cdt will loop on retried archive requests.
 945                          * The policy engine will ask for a new archive later
 946                          * when the file will not be modified for some tunable
 947                          * time */
 948                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 949                         rc = -EBUSY;
 950                         /* hpk_errval must be >= 0 */
 951                         hpk.hpk_errval = -rc;
 952                         GOTO(progress, rc);
 953                 }
 954
 955         }
 956
 957 progress:
 958         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 959                             &hpk, NULL);
 960
 961         /* Return first error */
 962         RETURN(rc != 0 ? rc : rc2);
 963 }
 964
 965
 966 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 967                           const void __user *data, size_t size)
 968 {
 969         void *copy;
 970         int rc;
 971
 972         OBD_ALLOC(copy, size);
 973         if (copy == NULL)
 974                 return -ENOMEM;
 975
 976         if (copy_from_user(copy, data, size)) {
 977                 rc = -EFAULT;
 978                 goto out;
 979         }
 980
 981         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 982 out:
 983         OBD_FREE(copy, size);
 984
 985         return rc;
 986 }
 987
 988 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 989 {
 990         int cmd = qctl->qc_cmd;
 991         int type = qctl->qc_type;
 992         int id = qctl->qc_id;
 993         int valid = qctl->qc_valid;
 994         int rc = 0;
 995         ENTRY;
 996
 997         switch (cmd) {
 998         case Q_SETQUOTA:
 999         case Q_SETINFO:
1000                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1001                         RETURN(-EPERM);
1002                 break;
1003         case Q_GETQUOTA:
1004                 if (((type == USRQUOTA &&
1005                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1006                      (type == GRPQUOTA &&
1007                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1008                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1009                         RETURN(-EPERM);
1010                 break;
1011         case Q_GETINFO:
1012                 break;
1013         default:
1014                 CERROR("unsupported quotactl op: %#x\n", cmd);
1015                 RETURN(-ENOTTY);
1016         }
1017
1018         if (valid != QC_GENERAL) {
1019                 if (cmd == Q_GETINFO)
1020                         qctl->qc_cmd = Q_GETOINFO;
1021                 else if (cmd == Q_GETQUOTA)
1022                         qctl->qc_cmd = Q_GETOQUOTA;
1023                 else
1024                         RETURN(-EINVAL);
1025
1026                 switch (valid) {
1027                 case QC_MDTIDX:
1028                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1029                                            sizeof(*qctl), qctl, NULL);
1030                         break;
1031                 case QC_OSTIDX:
1032                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1033                                            sizeof(*qctl), qctl, NULL);
1034                         break;
1035                 case QC_UUID:
1036                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1037                                            sizeof(*qctl), qctl, NULL);
1038                         if (rc == -EAGAIN)
1039                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1040                                                    sbi->ll_dt_exp,
1041                                                    sizeof(*qctl), qctl, NULL);
1042                         break;
1043                 default:
1044                         rc = -EINVAL;
1045                         break;
1046                 }
1047
1048                 if (rc)
1049                         RETURN(rc);
1050
1051                 qctl->qc_cmd = cmd;
1052         } else {
1053                 struct obd_quotactl *oqctl;
1054
1055                 OBD_ALLOC_PTR(oqctl);
1056                 if (oqctl == NULL)
1057                         RETURN(-ENOMEM);
1058
1059                 QCTL_COPY(oqctl, qctl);
1060                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1061                 if (rc) {
1062                         OBD_FREE_PTR(oqctl);
1063                         RETURN(rc);
1064                 }
1065                 /* If QIF_SPACE is not set, client should collect the
1066                  * space usage from OSSs by itself */
1067                 if (cmd == Q_GETQUOTA &&
1068                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1069                     !oqctl->qc_dqblk.dqb_curspace) {
1070                         struct obd_quotactl *oqctl_tmp;
1071
1072                         OBD_ALLOC_PTR(oqctl_tmp);
1073                         if (oqctl_tmp == NULL)
1074                                 GOTO(out, rc = -ENOMEM);
1075
1076                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1077                         oqctl_tmp->qc_id = oqctl->qc_id;
1078                         oqctl_tmp->qc_type = oqctl->qc_type;
1079
1080                         /* collect space usage from OSTs */
1081                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1082                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1083                         if (!rc || rc == -EREMOTEIO) {
1084                                 oqctl->qc_dqblk.dqb_curspace =
1085                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1086                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1087                         }
1088
1089                         /* collect space & inode usage from MDTs */
1090                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1091                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1092                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1093                         if (!rc || rc == -EREMOTEIO) {
1094                                 oqctl->qc_dqblk.dqb_curspace +=
1095                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1096                                 oqctl->qc_dqblk.dqb_curinodes =
1097                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1098                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1099                         } else {
1100                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1101                         }
1102
1103                         OBD_FREE_PTR(oqctl_tmp);
1104                 }
1105 out:
1106                 QCTL_COPY(qctl, oqctl);
1107                 OBD_FREE_PTR(oqctl);
1108         }
1109
1110         RETURN(rc);
1111 }
1112
1113 /* This function tries to get a single name component,
1114  * to send to the server. No actual path traversal involved,
1115  * so we limit to NAME_MAX */
1116 static char *ll_getname(const char __user *filename)
1117 {
1118         int ret = 0, len;
1119         char *tmp;
1120
1121         OBD_ALLOC(tmp, NAME_MAX + 1);
1122
1123         if (!tmp)
1124                 return ERR_PTR(-ENOMEM);
1125
1126         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1127         if (len < 0)
1128                 ret = -ENOENT;
1129         else if (len > NAME_MAX)
1130                 ret = -ENAMETOOLONG;
1131
1132         if (ret) {
1133                 OBD_FREE(tmp, NAME_MAX + 1);
1134                 tmp =  ERR_PTR(ret);
1135         }
1136         return tmp;
1137 }
1138
1139 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1140
1141 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1142 {
1143         struct dentry *dentry = file_dentry(file);
1144         struct inode *inode = file_inode(file);
1145         struct ll_sb_info *sbi = ll_i2sbi(inode);
1146         struct obd_ioctl_data *data;
1147         int rc = 0;
1148         ENTRY;
1149
1150         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1151                PFID(ll_inode2fid(inode)), inode, cmd);
1152
1153         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1154         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1155                 return -ENOTTY;
1156
1157         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1158         switch(cmd) {
1159         case FSFILT_IOC_GETFLAGS:
1160         case FSFILT_IOC_SETFLAGS:
1161                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1162         case FSFILT_IOC_GETVERSION_OLD:
1163         case FSFILT_IOC_GETVERSION:
1164                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1165         /* We need to special case any other ioctls we want to handle,
1166          * to send them to the MDS/OST as appropriate and to properly
1167          * network encode the arg field.
1168         case FSFILT_IOC_SETVERSION_OLD:
1169         case FSFILT_IOC_SETVERSION:
1170         */
1171         case LL_IOC_GET_MDTIDX: {
1172                 int mdtidx;
1173
1174                 mdtidx = ll_get_mdt_idx(inode);
1175                 if (mdtidx < 0)
1176                         RETURN(mdtidx);
1177
1178                 if (put_user((int)mdtidx, (int __user *)arg))
1179                         RETURN(-EFAULT);
1180
1181                 return 0;
1182         }
1183         case IOC_MDC_LOOKUP: {
1184                 int namelen, len = 0;
1185                 char *buf = NULL;
1186                 char *filename;
1187
1188                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1189                 if (rc != 0)
1190                         RETURN(rc);
1191                 data = (void *)buf;
1192
1193                 filename = data->ioc_inlbuf1;
1194                 namelen = strlen(filename);
1195                 if (namelen < 1) {
1196                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1197                         GOTO(out_free, rc = -EINVAL);
1198                 }
1199
1200                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1201                 if (rc < 0) {
1202                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1203                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1204                                filename, rc);
1205                         GOTO(out_free, rc);
1206                 }
1207 out_free:
1208                 obd_ioctl_freedata(buf, len);
1209                 return rc;
1210         }
1211         case LL_IOC_LMV_SETSTRIPE: {
1212                 struct lmv_user_md  *lum;
1213                 char            *buf = NULL;
1214                 char            *filename;
1215                 int              namelen = 0;
1216                 int              lumlen = 0;
1217                 umode_t          mode;
1218                 int              len;
1219                 int              rc;
1220
1221                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1222                 if (rc)
1223                         RETURN(rc);
1224
1225                 data = (void *)buf;
1226                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1227                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1228                         GOTO(lmv_out_free, rc = -EINVAL);
1229
1230                 filename = data->ioc_inlbuf1;
1231                 namelen = data->ioc_inllen1;
1232
1233                 if (namelen < 1) {
1234                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1235                         GOTO(lmv_out_free, rc = -EINVAL);
1236                 }
1237                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1238                 lumlen = data->ioc_inllen2;
1239
1240                 if (lum->lum_magic != LMV_USER_MAGIC ||
1241                     lumlen != sizeof(*lum)) {
1242                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1243                                filename, lum->lum_magic, lumlen, -EFAULT);
1244                         GOTO(lmv_out_free, rc = -EINVAL);
1245                 }
1246
1247 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1248                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1249 #else
1250                 mode = data->ioc_type;
1251 #endif
1252                 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1253 lmv_out_free:
1254                 obd_ioctl_freedata(buf, len);
1255                 RETURN(rc);
1256
1257         }
1258         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1259                 struct lmv_user_md        lum;
1260                 struct lmv_user_md __user *ulump =
1261                                         (struct lmv_user_md __user *)arg;
1262                 int                       rc;
1263
1264                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1265                         RETURN(-EFAULT);
1266
1267                 if (lum.lum_magic != LMV_USER_MAGIC)
1268                         RETURN(-EINVAL);
1269
1270                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1271
1272                 RETURN(rc);
1273         }
1274         case LL_IOC_LOV_SETSTRIPE: {
1275                 struct lov_user_md_v3 lumv3;
1276                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1277                 struct lov_user_md_v1 __user *lumv1p =
1278                         (struct lov_user_md_v1 __user *)arg;
1279                 struct lov_user_md_v3 __user *lumv3p =
1280                         (struct lov_user_md_v3 __user *)arg;
1281
1282                 int set_default = 0;
1283
1284                 CLASSERT(sizeof(struct lov_user_md_v3) >
1285                          sizeof(struct lov_comp_md_v1));
1286                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1287                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1288                                 sizeof(lumv3p->lmm_objects[0]));
1289                 /* first try with v1 which is smaller than v3 */
1290                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1291                         RETURN(-EFAULT);
1292
1293                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3)
1294                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1295                                 RETURN(-EFAULT);
1296
1297                 if (inode->i_sb->s_root == file_dentry(file))
1298                         set_default = 1;
1299
1300                 /* in v1 and v3 cases lumv1 points to data */
1301                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1302
1303                 RETURN(rc);
1304         }
1305         case LL_IOC_LMV_GETSTRIPE: {
1306                 struct lmv_user_md __user *ulmv =
1307                                         (struct lmv_user_md __user *)arg;
1308                 struct lmv_user_md      lum;
1309                 struct ptlrpc_request   *request = NULL;
1310                 union lmv_mds_md        *lmm = NULL;
1311                 int                     lmmsize;
1312                 u64                     valid = 0;
1313                 struct lmv_user_md      *tmp = NULL;
1314                 int                     mdt_index;
1315                 int                     lum_size;
1316                 int                     stripe_count;
1317                 int                     max_stripe_count;
1318                 int                     i;
1319                 int                     rc;
1320
1321                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1322                         RETURN(-EFAULT);
1323
1324                 max_stripe_count = lum.lum_stripe_count;
1325                 /* lum_magic will indicate which stripe the ioctl will like
1326                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1327                  * is for default LMV stripe */
1328                 if (lum.lum_magic == LMV_MAGIC_V1)
1329                         valid |= OBD_MD_MEA;
1330                 else if (lum.lum_magic == LMV_USER_MAGIC)
1331                         valid |= OBD_MD_DEFAULT_MEA;
1332                 else
1333                         RETURN(-EINVAL);
1334
1335                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1336                                       valid);
1337                 if (rc != 0)
1338                         GOTO(finish_req, rc);
1339
1340                 /* Get default LMV EA */
1341                 if (lum.lum_magic == LMV_USER_MAGIC) {
1342                         if (lmmsize > sizeof(*ulmv))
1343                                 GOTO(finish_req, rc = -EINVAL);
1344
1345                         if (copy_to_user(ulmv, lmm, lmmsize))
1346                                 GOTO(finish_req, rc = -EFAULT);
1347
1348                         GOTO(finish_req, rc);
1349                 }
1350
1351                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1352                 if (max_stripe_count < stripe_count) {
1353                         lum.lum_stripe_count = stripe_count;
1354                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1355                                 GOTO(finish_req, rc = -EFAULT);
1356                         GOTO(finish_req, rc = -E2BIG);
1357                 }
1358
1359                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1360                 OBD_ALLOC(tmp, lum_size);
1361                 if (tmp == NULL)
1362                         GOTO(finish_req, rc = -ENOMEM);
1363
1364                 mdt_index = ll_get_mdt_idx(inode);
1365                 if (mdt_index < 0)
1366                         GOTO(out_tmp, rc = -ENOMEM);
1367
1368                 tmp->lum_magic = LMV_MAGIC_V1;
1369                 tmp->lum_stripe_count = 0;
1370                 tmp->lum_stripe_offset = mdt_index;
1371                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1372                 for (i = 0; i < stripe_count; i++) {
1373                         struct lu_fid   fid;
1374
1375                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1376                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1377                         if (mdt_index < 0)
1378                                 GOTO(out_tmp, rc = mdt_index);
1379
1380                         tmp->lum_objects[i].lum_mds = mdt_index;
1381                         tmp->lum_objects[i].lum_fid = fid;
1382                         tmp->lum_stripe_count++;
1383                 }
1384
1385                 if (copy_to_user(ulmv, tmp, lum_size))
1386                         GOTO(out_tmp, rc = -EFAULT);
1387 out_tmp:
1388                 OBD_FREE(tmp, lum_size);
1389 finish_req:
1390                 ptlrpc_req_finished(request);
1391                 return rc;
1392         }
1393
1394         case LL_IOC_REMOVE_ENTRY: {
1395                 char            *filename = NULL;
1396                 int              namelen = 0;
1397                 int              rc;
1398
1399                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1400                  * unsupported server, which might crash the server(LU-2730),
1401                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1402                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1403                  * server will support REINT_RMENTRY XXX*/
1404                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1405                         RETURN(-ENOTSUPP);
1406
1407                 filename = ll_getname((const char __user *)arg);
1408                 if (IS_ERR(filename))
1409                         RETURN(PTR_ERR(filename));
1410
1411                 namelen = strlen(filename);
1412                 if (namelen < 1)
1413                         GOTO(out_rmdir, rc = -EINVAL);
1414
1415                 rc = ll_rmdir_entry(inode, filename, namelen);
1416 out_rmdir:
1417                 if (filename)
1418                         ll_putname(filename);
1419                 RETURN(rc);
1420         }
1421         case LL_IOC_LOV_SWAP_LAYOUTS:
1422                 RETURN(-EPERM);
1423         case IOC_OBD_STATFS:
1424                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1425         case LL_IOC_LOV_GETSTRIPE:
1426         case LL_IOC_MDC_GETINFO:
1427         case IOC_MDC_GETFILEINFO:
1428         case IOC_MDC_GETFILESTRIPE: {
1429                 struct ptlrpc_request *request = NULL;
1430                 struct lov_user_md __user *lump;
1431                 struct lov_mds_md *lmm = NULL;
1432                 struct mdt_body *body;
1433                 char *filename = NULL;
1434                 int lmmsize;
1435
1436                 if (cmd == IOC_MDC_GETFILEINFO ||
1437                     cmd == IOC_MDC_GETFILESTRIPE) {
1438                         filename = ll_getname((const char __user *)arg);
1439                         if (IS_ERR(filename))
1440                                 RETURN(PTR_ERR(filename));
1441
1442                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1443                                                       &lmmsize, &request);
1444                 } else {
1445                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1446                                               &request, 0);
1447                 }
1448
1449                 if (request) {
1450                         body = req_capsule_server_get(&request->rq_pill,
1451                                                       &RMF_MDT_BODY);
1452                         LASSERT(body != NULL);
1453                 } else {
1454                         GOTO(out_req, rc);
1455                 }
1456
1457                 if (rc < 0) {
1458                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1459                                                cmd == LL_IOC_MDC_GETINFO))
1460                                 GOTO(skip_lmm, rc = 0);
1461                         else
1462                                 GOTO(out_req, rc);
1463                 }
1464
1465                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1466                     cmd == LL_IOC_LOV_GETSTRIPE) {
1467                         lump = (struct lov_user_md __user *)arg;
1468                 } else {
1469                         struct lov_user_mds_data __user *lmdp;
1470                         lmdp = (struct lov_user_mds_data __user *)arg;
1471                         lump = &lmdp->lmd_lmm;
1472                 }
1473                 if (copy_to_user(lump, lmm, lmmsize)) {
1474                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1475                                 GOTO(out_req, rc = -EFAULT);
1476                         rc = -EOVERFLOW;
1477                 }
1478         skip_lmm:
1479                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1480                         struct lov_user_mds_data __user *lmdp;
1481                         lstat_t st = { 0 };
1482
1483                         st.st_dev       = inode->i_sb->s_dev;
1484                         st.st_mode      = body->mbo_mode;
1485                         st.st_nlink     = body->mbo_nlink;
1486                         st.st_uid       = body->mbo_uid;
1487                         st.st_gid       = body->mbo_gid;
1488                         st.st_rdev      = body->mbo_rdev;
1489                         st.st_size      = body->mbo_size;
1490                         st.st_blksize   = PAGE_SIZE;
1491                         st.st_blocks    = body->mbo_blocks;
1492                         st.st_atime     = body->mbo_atime;
1493                         st.st_mtime     = body->mbo_mtime;
1494                         st.st_ctime     = body->mbo_ctime;
1495                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1496                                                 sbi->ll_flags &
1497                                                 LL_SBI_32BIT_API);
1498
1499                         lmdp = (struct lov_user_mds_data __user *)arg;
1500                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1501                                 GOTO(out_req, rc = -EFAULT);
1502                 }
1503
1504                 EXIT;
1505         out_req:
1506                 ptlrpc_req_finished(request);
1507                 if (filename)
1508                         ll_putname(filename);
1509                 return rc;
1510         }
1511         case OBD_IOC_QUOTACTL: {
1512                 struct if_quotactl *qctl;
1513
1514                 OBD_ALLOC_PTR(qctl);
1515                 if (!qctl)
1516                         RETURN(-ENOMEM);
1517
1518                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1519                         GOTO(out_quotactl, rc = -EFAULT);
1520
1521                 rc = quotactl_ioctl(sbi, qctl);
1522
1523                 if (rc == 0 &&
1524                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1525                         rc = -EFAULT;
1526
1527         out_quotactl:
1528                 OBD_FREE_PTR(qctl);
1529                 RETURN(rc);
1530         }
1531         case OBD_IOC_GETDTNAME:
1532         case OBD_IOC_GETMDNAME:
1533                 RETURN(ll_get_obd_name(inode, cmd, arg));
1534         case LL_IOC_FLUSHCTX:
1535                 RETURN(ll_flush_ctx(inode));
1536         case LL_IOC_GETOBDCOUNT: {
1537                 u32 count, vallen;
1538                 struct obd_export *exp;
1539
1540                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1541                         RETURN(-EFAULT);
1542
1543                 /* get ost count when count is zero, get mdt count otherwise */
1544                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1545                 vallen = sizeof(count);
1546                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1547                                   KEY_TGT_COUNT, &vallen, &count);
1548                 if (rc) {
1549                         CERROR("get target count failed: %d\n", rc);
1550                         RETURN(rc);
1551                 }
1552
1553                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1554                         RETURN(-EFAULT);
1555
1556                 RETURN(0);
1557         }
1558         case LL_IOC_PATH2FID:
1559                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1560                                      sizeof(struct lu_fid)))
1561                         RETURN(-EFAULT);
1562                 RETURN(0);
1563         case LL_IOC_GET_CONNECT_FLAGS: {
1564                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1565                                      (void __user *)arg));
1566         }
1567         case OBD_IOC_FID2PATH:
1568                 RETURN(ll_fid2path(inode, (void __user *)arg));
1569         case LL_IOC_GETPARENT:
1570                 RETURN(ll_getparent(file, (void __user *)arg));
1571         case LL_IOC_FID2MDTIDX: {
1572                 struct obd_export *exp = ll_i2mdexp(inode);
1573                 struct lu_fid     fid;
1574                 __u32             index;
1575
1576                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1577                                    sizeof(fid)))
1578                         RETURN(-EFAULT);
1579
1580                 /* Call mdc_iocontrol */
1581                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1582                                    (__u32 __user *)&index);
1583                 if (rc != 0)
1584                         RETURN(rc);
1585
1586                 RETURN(index);
1587         }
1588         case LL_IOC_HSM_REQUEST: {
1589                 struct hsm_user_request *hur;
1590                 ssize_t                  totalsize;
1591
1592                 OBD_ALLOC_PTR(hur);
1593                 if (hur == NULL)
1594                         RETURN(-ENOMEM);
1595
1596                 /* We don't know the true size yet; copy the fixed-size part */
1597                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1598                         OBD_FREE_PTR(hur);
1599                         RETURN(-EFAULT);
1600                 }
1601
1602                 /* Compute the whole struct size */
1603                 totalsize = hur_len(hur);
1604                 OBD_FREE_PTR(hur);
1605                 if (totalsize < 0)
1606                         RETURN(-E2BIG);
1607
1608                 /* Final size will be more than double totalsize */
1609                 if (totalsize >= MDS_MAXREQSIZE / 3)
1610                         RETURN(-E2BIG);
1611
1612                 OBD_ALLOC_LARGE(hur, totalsize);
1613                 if (hur == NULL)
1614                         RETURN(-ENOMEM);
1615
1616                 /* Copy the whole struct */
1617                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1618                         GOTO(out_hur, rc = -EFAULT);
1619
1620                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1621                         const struct lu_fid *fid;
1622                         struct inode *f;
1623                         int i;
1624
1625                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1626                                 fid = &hur->hur_user_item[i].hui_fid;
1627                                 f = search_inode_for_lustre(inode->i_sb, fid);
1628                                 if (IS_ERR(f)) {
1629                                         rc = PTR_ERR(f);
1630                                         break;
1631                                 }
1632
1633                                 rc = ll_hsm_release(f);
1634                                 iput(f);
1635                                 if (rc != 0)
1636                                         break;
1637                         }
1638                 } else {
1639                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1640                                            hur, NULL);
1641                 }
1642
1643 out_hur:
1644                 OBD_FREE_LARGE(hur, totalsize);
1645
1646                 RETURN(rc);
1647         }
1648         case LL_IOC_HSM_PROGRESS: {
1649                 struct hsm_progress_kernel      hpk;
1650                 struct hsm_progress             hp;
1651
1652                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1653                         RETURN(-EFAULT);
1654
1655                 hpk.hpk_fid = hp.hp_fid;
1656                 hpk.hpk_cookie = hp.hp_cookie;
1657                 hpk.hpk_extent = hp.hp_extent;
1658                 hpk.hpk_flags = hp.hp_flags;
1659                 hpk.hpk_errval = hp.hp_errval;
1660                 hpk.hpk_data_version = 0;
1661
1662                 /* File may not exist in Lustre; all progress
1663                  * reported to Lustre root */
1664                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1665                                    NULL);
1666                 RETURN(rc);
1667         }
1668         case LL_IOC_HSM_CT_START:
1669                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1670                         RETURN(-EPERM);
1671
1672                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1673                                     sizeof(struct lustre_kernelcomm));
1674                 RETURN(rc);
1675
1676         case LL_IOC_HSM_COPY_START: {
1677                 struct hsm_copy *copy;
1678                 int              rc;
1679
1680                 OBD_ALLOC_PTR(copy);
1681                 if (copy == NULL)
1682                         RETURN(-ENOMEM);
1683                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1684                         OBD_FREE_PTR(copy);
1685                         RETURN(-EFAULT);
1686                 }
1687
1688                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1689                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1690                         rc = -EFAULT;
1691
1692                 OBD_FREE_PTR(copy);
1693                 RETURN(rc);
1694         }
1695         case LL_IOC_HSM_COPY_END: {
1696                 struct hsm_copy *copy;
1697                 int              rc;
1698
1699                 OBD_ALLOC_PTR(copy);
1700                 if (copy == NULL)
1701                         RETURN(-ENOMEM);
1702                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1703                         OBD_FREE_PTR(copy);
1704                         RETURN(-EFAULT);
1705                 }
1706
1707                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1708                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1709                         rc = -EFAULT;
1710
1711                 OBD_FREE_PTR(copy);
1712                 RETURN(rc);
1713         }
1714         case LL_IOC_MIGRATE: {
1715                 char            *buf = NULL;
1716                 const char      *filename;
1717                 int             namelen = 0;
1718                 int             len;
1719                 int             rc;
1720                 int             mdtidx;
1721
1722                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1723                 if (rc < 0)
1724                         RETURN(rc);
1725
1726                 data = (struct obd_ioctl_data *)buf;
1727                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1728                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1729                         GOTO(migrate_free, rc = -EINVAL);
1730
1731                 filename = data->ioc_inlbuf1;
1732                 namelen = data->ioc_inllen1;
1733                 /* \0 is packed at the end of filename */
1734                 if (namelen < 1 || namelen != strlen(filename) + 1)
1735                         GOTO(migrate_free, rc = -EINVAL);
1736
1737                 if (data->ioc_inllen2 != sizeof(mdtidx))
1738                         GOTO(migrate_free, rc = -EINVAL);
1739                 mdtidx = *(int *)data->ioc_inlbuf2;
1740
1741                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1742 migrate_free:
1743                 obd_ioctl_freedata(buf, len);
1744
1745                 RETURN(rc);
1746         }
1747         default:
1748                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1749                                      (void __user *)arg));
1750         }
1751 }
1752
1753 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1754 {
1755         struct inode *inode = file->f_mapping->host;
1756         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1757         struct ll_sb_info *sbi = ll_i2sbi(inode);
1758         int api32 = ll_need_32bit_api(sbi);
1759         loff_t ret = -EINVAL;
1760         ENTRY;
1761
1762         inode_lock(inode);
1763         switch (origin) {
1764                 case SEEK_SET:
1765                         break;
1766                 case SEEK_CUR:
1767                         offset += file->f_pos;
1768                         break;
1769                 case SEEK_END:
1770                         if (offset > 0)
1771                                 GOTO(out, ret);
1772                         if (api32)
1773                                 offset += LL_DIR_END_OFF_32BIT;
1774                         else
1775                                 offset += LL_DIR_END_OFF;
1776                         break;
1777                 default:
1778                         GOTO(out, ret);
1779         }
1780
1781         if (offset >= 0 &&
1782             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1783              (!api32 && offset <= LL_DIR_END_OFF))) {
1784                 if (offset != file->f_pos) {
1785                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1786                             (!api32 && offset == LL_DIR_END_OFF))
1787                                 fd->lfd_pos = MDS_DIR_END_OFF;
1788                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1789                                 fd->lfd_pos = offset << 32;
1790                         else
1791                                 fd->lfd_pos = offset;
1792                         file->f_pos = offset;
1793                         file->f_version = 0;
1794                 }
1795                 ret = offset;
1796         }
1797         GOTO(out, ret);
1798
1799 out:
1800         inode_unlock(inode);
1801         return ret;
1802 }
1803
1804 static int ll_dir_open(struct inode *inode, struct file *file)
1805 {
1806         ENTRY;
1807         RETURN(ll_file_open(inode, file));
1808 }
1809
1810 static int ll_dir_release(struct inode *inode, struct file *file)
1811 {
1812         ENTRY;
1813         RETURN(ll_file_release(inode, file));
1814 }
1815
1816 const struct file_operations ll_dir_operations = {
1817         .llseek         = ll_dir_seek,
1818         .open           = ll_dir_open,
1819         .release        = ll_dir_release,
1820         .read           = generic_read_dir,
1821 #ifdef HAVE_DIR_CONTEXT
1822         .iterate        = ll_iterate,
1823 #else
1824         .readdir        = ll_readdir,
1825 #endif
1826         .unlocked_ioctl = ll_dir_ioctl,
1827         .fsync          = ll_fsync,
1828 };