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, 2015, 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 <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <asm/uaccess.h>
  50 #include <linux/buffer_head.h>   // for wait_on_buffer
  51 #include <linux/pagevec.h>
  52
  53 #define DEBUG_SUBSYSTEM S_LLITE
  54
  55 #include <lustre/lustre_idl.h>
  56
  57 #include <obd_support.h>
  58 #include <obd_class.h>
  59 #include <lustre_ioctl.h>
  60 #include <lustre_lib.h>
  61 #include <lustre_dlm.h>
  62 #include <lustre_fid.h>
  63 #include <lustre_kernelcomm.h>
  64 #include <lustre_swab.h>
  65
  66 #include "llite_internal.h"
  67
  68 /*
  69  * (new) readdir implementation overview.
  70  *
  71  * Original lustre readdir implementation cached exact copy of raw directory
  72  * pages on the client. These pages were indexed in client page cache by
  73  * logical offset in the directory file. This design, while very simple and
  74  * intuitive had some inherent problems:
  75  *
  76  *     . it implies that byte offset to the directory entry serves as a
  77  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  78  *     ext3/htree directory entries may move due to splits, and more
  79  *     importantly,
  80  *
  81  *     . it is incompatible with the design of split directories for cmd3,
  82  *     that assumes that names are distributed across nodes based on their
  83  *     hash, and so readdir should be done in hash order.
  84  *
  85  * New readdir implementation does readdir in hash order, and uses hash of a
  86  * file name as a telldir/seekdir cookie. This led to number of complications:
  87  *
  88  *     . hash is not unique, so it cannot be used to index cached directory
  89  *     pages on the client (note, that it requires a whole pageful of hash
  90  *     collided entries to cause two pages to have identical hashes);
  91  *
  92  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  93  *     entry cookie. ext3/htree has the same problem and lustre implementation
  94  *     mimics their solution: seekdir(hash) positions directory at the first
  95  *     entry with the given hash.
  96  *
  97  * Client side.
  98  *
  99  * 0. caching
 100  *
 101  * Client caches directory pages using hash of the first entry as an index. As
 102  * noted above hash is not unique, so this solution doesn't work as is:
 103  * special processing is needed for "page hash chains" (i.e., sequences of
 104  * pages filled with entries all having the same hash value).
 105  *
 106  * First, such chains have to be detected. To this end, server returns to the
 107  * client the hash of the first entry on the page next to one returned. When
 108  * client detects that this hash is the same as hash of the first entry on the
 109  * returned page, page hash collision has to be handled. Pages in the
 110  * hash chain, except first one, are termed "overflow pages".
 111  *
 112  * Solution to index uniqueness problem is to not cache overflow
 113  * pages. Instead, when page hash collision is detected, all overflow pages
 114  * from emerging chain are immediately requested from the server and placed in
 115  * a special data structure (struct ll_dir_chain). This data structure is used
 116  * by ll_readdir() to process entries from overflow pages. When readdir
 117  * invocation finishes, overflow pages are discarded. If page hash collision
 118  * chain weren't completely processed, next call to readdir will again detect
 119  * page hash collision, again read overflow pages in, process next portion of
 120  * entries and again discard the pages. This is not as wasteful as it looks,
 121  * because, given reasonable hash, page hash collisions are extremely rare.
 122  *
 123  * 1. directory positioning
 124  *
 125  * When seekdir(hash) is called, original
 126  *
 127  *
 128  *
 129  *
 130  *
 131  *
 132  *
 133  *
 134  * Server.
 135  *
 136  * identification of and access to overflow pages
 137  *
 138  * page format
 139  *
 140  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 141  * a header lu_dirpage which describes the start/end hash, and whether this
 142  * page is empty (contains no dir entry) or hash collide with next page.
 143  * After client receives reply, several pages will be integrated into dir page
 144  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 145  * lu_dirpage for this integrated page will be adjusted. See
 146  * mdc_adjust_dirpages().
 147  *
 148  */
 149 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 150                              __u64 offset, struct ll_dir_chain *chain)
 151 {
 152         struct md_callback      cb_op;
 153         struct page             *page;
 154         int                     rc;
 155
 156         cb_op.md_blocking_ast = ll_md_blocking_ast;
 157         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 158         if (rc != 0)
 159                 return ERR_PTR(rc);
 160
 161         return page;
 162 }
 163
 164 void ll_release_page(struct inode *inode, struct page *page,
 165                      bool remove)
 166 {
 167         kunmap(page);
 168
 169         /* Always remove the page for striped dir, because the page is
 170          * built from temporarily in LMV layer */
 171         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 172             ll_i2info(inode)->lli_lsm_md != NULL) {
 173                 __free_page(page);
 174                 return;
 175         }
 176
 177         if (remove) {
 178                 lock_page(page);
 179                 if (likely(page->mapping != NULL))
 180                         truncate_complete_page(page->mapping, page);
 181                 unlock_page(page);
 182         }
 183         page_cache_release(page);
 184 }
 185
 186 /**
 187  * return IF_* type for given lu_dirent entry.
 188  * IF_* flag shld be converted to particular OS file type in
 189  * platform llite module.
 190  */
 191 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 192 {
 193         u16 type = 0;
 194         struct luda_type *lt;
 195         int len = 0;
 196
 197         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 198                 const unsigned align = sizeof(struct luda_type) - 1;
 199
 200                 len = le16_to_cpu(ent->lde_namelen);
 201                 len = (len + align) & ~align;
 202                 lt = (void *)ent->lde_name + len;
 203                 type = IFTODT(le16_to_cpu(lt->lt_type));
 204         }
 205
 206         return type;
 207 }
 208
 209 #ifdef HAVE_DIR_CONTEXT
 210 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 211                 struct dir_context *ctx)
 212 {
 213 #else
 214 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 215                 void *cookie, filldir_t filldir)
 216 {
 217 #endif
 218         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 219         __u64                 pos        = *ppos;
 220         bool                  is_api32 = ll_need_32bit_api(sbi);
 221         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 222         struct page          *page;
 223         struct ll_dir_chain   chain;
 224         bool                  done = false;
 225         int                   rc = 0;
 226         ENTRY;
 227
 228         ll_dir_chain_init(&chain);
 229
 230         page = ll_get_dir_page(inode, op_data, pos, &chain);
 231
 232         while (rc == 0 && !done) {
 233                 struct lu_dirpage *dp;
 234                 struct lu_dirent  *ent;
 235                 __u64 hash;
 236                 __u64 next;
 237
 238                 if (IS_ERR(page)) {
 239                         rc = PTR_ERR(page);
 240                         break;
 241                 }
 242
 243                 hash = MDS_DIR_END_OFF;
 244                 dp = page_address(page);
 245                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 246                      ent = lu_dirent_next(ent)) {
 247                         __u16          type;
 248                         int            namelen;
 249                         struct lu_fid  fid;
 250                         __u64          lhash;
 251                         __u64          ino;
 252
 253                         hash = le64_to_cpu(ent->lde_hash);
 254                         if (hash < pos)
 255                                 /*
 256                                  * Skip until we find target hash
 257                                  * value.
 258                                  */
 259                                 continue;
 260
 261                         namelen = le16_to_cpu(ent->lde_namelen);
 262                         if (namelen == 0)
 263                                 /*
 264                                  * Skip dummy record.
 265                                  */
 266                                 continue;
 267
 268                         if (is_api32 && is_hash64)
 269                                 lhash = hash >> 32;
 270                         else
 271                                 lhash = hash;
 272                         fid_le_to_cpu(&fid, &ent->lde_fid);
 273                         ino = cl_fid_build_ino(&fid, is_api32);
 274                         type = ll_dirent_type_get(ent);
 275                         /* For 'll_nfs_get_name_filldir()', it will try
 276                          * to access the 'ent' through its 'lde_name',
 277                          * so the parameter 'name' for 'filldir()' must
 278                          * be part of the 'ent'. */
 279 #ifdef HAVE_DIR_CONTEXT
 280                         ctx->pos = lhash;
 281                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 282                                          type);
 283 #else
 284                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 285                                        ino, type);
 286 #endif
 287                 }
 288
 289                 if (done) {
 290                         pos = hash;
 291                         ll_release_page(inode, page, false);
 292                         break;
 293                 }
 294
 295                 next = le64_to_cpu(dp->ldp_hash_end);
 296                 pos = next;
 297                 if (pos == MDS_DIR_END_OFF) {
 298                         /*
 299                          * End of directory reached.
 300                          */
 301                         done = 1;
 302                         ll_release_page(inode, page, false);
 303                 } else {
 304                         /*
 305                          * Normal case: continue to the next
 306                          * page.
 307                          */
 308                         ll_release_page(inode, page,
 309                                         le32_to_cpu(dp->ldp_flags) &
 310                                         LDF_COLLIDE);
 311                         next = pos;
 312                         page = ll_get_dir_page(inode, op_data, pos,
 313                                                &chain);
 314                 }
 315         }
 316 #ifdef HAVE_DIR_CONTEXT
 317         ctx->pos = pos;
 318 #else
 319         *ppos = pos;
 320 #endif
 321         ll_dir_chain_fini(&chain);
 322         RETURN(rc);
 323 }
 324
 325 #ifdef HAVE_DIR_CONTEXT
 326 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 327 #else
 328 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 329 #endif
 330 {
 331         struct inode            *inode  = file_inode(filp);
 332         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 333         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 334         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 335         int                     api32   = ll_need_32bit_api(sbi);
 336         struct md_op_data       *op_data;
 337         __u64                   pos;
 338         int                     rc;
 339         ENTRY;
 340
 341         if (lfd != NULL)
 342                 pos = lfd->lfd_pos;
 343         else
 344                 pos = 0;
 345
 346         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 347                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 348                inode, (unsigned long)pos, i_size_read(inode), api32);
 349
 350         if (pos == MDS_DIR_END_OFF)
 351                 /*
 352                  * end-of-file.
 353                  */
 354                 GOTO(out, rc = 0);
 355
 356         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 357                                      LUSTRE_OPC_ANY, inode);
 358         if (IS_ERR(op_data))
 359                 GOTO(out, rc = PTR_ERR(op_data));
 360
 361         if (unlikely(op_data->op_mea1 != NULL)) {
 362                 /* This is only needed for striped dir to fill ..,
 363                  * see lmv_read_entry */
 364                 if (file_dentry(filp)->d_parent != NULL &&
 365                     file_dentry(filp)->d_parent->d_inode != NULL) {
 366                         __u64 ibits = MDS_INODELOCK_UPDATE;
 367                         struct inode *parent =
 368                                 file_dentry(filp)->d_parent->d_inode;
 369
 370                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 371                                 op_data->op_fid3 = *ll_inode2fid(parent);
 372                 }
 373
 374                 /* If it can not find in cache, do lookup .. on the master
 375                  * object */
 376                 if (fid_is_zero(&op_data->op_fid3)) {
 377                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 378                         if (rc != 0) {
 379                                 ll_finish_md_op_data(op_data);
 380                                 RETURN(rc);
 381                         }
 382                 }
 383         }
 384         op_data->op_max_pages = sbi->ll_md_brw_pages;
 385 #ifdef HAVE_DIR_CONTEXT
 386         ctx->pos = pos;
 387         rc = ll_dir_read(inode, &pos, op_data, ctx);
 388         pos = ctx->pos;
 389 #else
 390         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 391 #endif
 392         if (lfd != NULL)
 393                 lfd->lfd_pos = pos;
 394
 395         if (pos == MDS_DIR_END_OFF) {
 396                 if (api32)
 397                         pos = LL_DIR_END_OFF_32BIT;
 398                 else
 399                         pos = LL_DIR_END_OFF;
 400         } else {
 401                 if (api32 && hash64)
 402                         pos = pos >> 32;
 403         }
 404 #ifdef HAVE_DIR_CONTEXT
 405         ctx->pos = pos;
 406 #else
 407         filp->f_pos = pos;
 408 #endif
 409         ll_finish_md_op_data(op_data);
 410         filp->f_version = inode->i_version;
 411
 412 out:
 413         if (!rc)
 414                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 415
 416         RETURN(rc);
 417 }
 418
 419 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 420 {
 421         struct mgs_send_param *msp;
 422         int rc = 0;
 423
 424         OBD_ALLOC_PTR(msp);
 425         if (!msp)
 426                 return -ENOMEM;
 427
 428         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 429         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 430                                 sizeof(struct mgs_send_param), msp, NULL);
 431         if (rc)
 432                 CERROR("Failed to set parameter: %d\n", rc);
 433         OBD_FREE_PTR(msp);
 434
 435         return rc;
 436 }
 437
 438 /**
 439  * Create striped directory with specified stripe(@lump)
 440  *
 441  * \param[in] dparent   the parent of the directory.
 442  * \param[in] lump      the specified stripes.
 443  * \param[in] dirname   the name of the directory.
 444  * \param[in] mode      the specified mode of the directory.
 445  *
 446  * \retval              =0 if striped directory is being created successfully.
 447  *                      <0 if the creation is failed.
 448  */
 449 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 450                                const char *dirname, umode_t mode)
 451 {
 452         struct inode *parent = dparent->d_inode;
 453         struct ptlrpc_request *request = NULL;
 454         struct md_op_data *op_data;
 455         struct ll_sb_info *sbi = ll_i2sbi(parent);
 456         struct inode *inode = NULL;
 457         struct dentry dentry = {
 458                 .d_parent = dparent,
 459                 .d_name = {
 460                         .name = dirname,
 461                         .len = strlen(dirname),
 462                         .hash = full_name_hash(dirname, strlen(dirname)),
 463                 },
 464         };
 465         int err;
 466         ENTRY;
 467
 468         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 469                 RETURN(-EINVAL);
 470
 471         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 472                "stripe_offset %d, stripe_count: %u\n",
 473                PFID(ll_inode2fid(parent)), parent, dirname,
 474                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 475
 476         if (lump->lum_stripe_count > 1 &&
 477             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 478                 RETURN(-EINVAL);
 479
 480         if (IS_DEADDIR(parent) &&
 481             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 482                 RETURN(-ENOENT);
 483
 484         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 485                 lustre_swab_lmv_user_md(lump);
 486
 487         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 488                 mode &= ~current_umask();
 489         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 490         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 491                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 492                                      lump);
 493         if (IS_ERR(op_data))
 494                 RETURN(PTR_ERR(op_data));
 495
 496         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 497                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 498                  * to determine the security context for the file. So our fake
 499                  * dentry should be real enough for this purpose. */
 500                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 501                                               &op_data->op_file_secctx_name,
 502                                               &op_data->op_file_secctx,
 503                                               &op_data->op_file_secctx_size);
 504                 if (err < 0)
 505                         GOTO(out_op_data, err);
 506         }
 507
 508         op_data->op_cli_flags |= CLI_SET_MEA;
 509         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 510                         from_kuid(&init_user_ns, current_fsuid()),
 511                         from_kgid(&init_user_ns, current_fsgid()),
 512                         cfs_curproc_cap_pack(), 0, &request);
 513         if (err)
 514                 GOTO(out_request, err);
 515
 516         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 517
 518         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 519         if (err)
 520                 GOTO(out_inode, err);
 521
 522         dentry.d_inode = inode;
 523
 524         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
 525                 err = ll_inode_init_security(&dentry, inode, parent);
 526                 if (err)
 527                         GOTO(out_inode, err);
 528         }
 529
 530 out_inode:
 531         if (inode != NULL)
 532                 iput(inode);
 533 out_request:
 534         ptlrpc_req_finished(request);
 535 out_op_data:
 536         ll_finish_md_op_data(op_data);
 537
 538         return err;
 539 }
 540
 541 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 542                      int set_default)
 543 {
 544         struct ll_sb_info *sbi = ll_i2sbi(inode);
 545         struct md_op_data *op_data;
 546         struct ptlrpc_request *req = NULL;
 547         int rc = 0;
 548         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 549         struct obd_device *mgc = lsi->lsi_mgc;
 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 LMV_USER_MAGIC: {
 574                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 575                                 lustre_swab_lmv_user_md(
 576                                         (struct lmv_user_md *)lump);
 577                         lum_size = sizeof(struct lmv_user_md);
 578                         break;
 579                 }
 580                 default: {
 581                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 582                                         " %#08x != %#08x nor %#08x\n",
 583                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 584                                         LOV_USER_MAGIC_V3);
 585                         RETURN(-EINVAL);
 586                 }
 587                 }
 588         } else {
 589                 lum_size = sizeof(struct lov_user_md_v1);
 590         }
 591
 592         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 593                                      LUSTRE_OPC_ANY, NULL);
 594         if (IS_ERR(op_data))
 595                 RETURN(PTR_ERR(op_data));
 596
 597         /* swabbing is done in lov_setstripe() on server side */
 598         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 599         ll_finish_md_op_data(op_data);
 600         ptlrpc_req_finished(req);
 601         if (rc) {
 602                 if (rc != -EPERM && rc != -EACCES)
 603                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 604         }
 605
 606         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 607          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 608          need the make the distiction between the 2 versions */
 609         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 610                 char *param = NULL;
 611                 char *buf;
 612
 613                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 614                 if (param == NULL)
 615                         GOTO(end, rc = -ENOMEM);
 616
 617                 buf = param;
 618                 /* Get fsname and assume devname to be -MDT0000. */
 619                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 620                 strcat(buf, "-MDT0000.lov");
 621                 buf += strlen(buf);
 622
 623                 /* Set root stripesize */
 624                 sprintf(buf, ".stripesize=%u",
 625                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 626                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 627                 if (rc)
 628                         GOTO(end, rc);
 629
 630                 /* Set root stripecount */
 631                 sprintf(buf, ".stripecount=%hd",
 632                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 633                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 634                 if (rc)
 635                         GOTO(end, rc);
 636
 637                 /* Set root stripeoffset */
 638                 sprintf(buf, ".stripeoffset=%hd",
 639                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 640                         (typeof(lump->lmm_stripe_offset))(-1));
 641                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 642
 643 end:
 644                 if (param != NULL)
 645                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 646         }
 647         RETURN(rc);
 648 }
 649
 650 /**
 651  * This function will be used to get default LOV/LMV/Default LMV
 652  * @valid will be used to indicate which stripe it will retrieve
 653  *      OBD_MD_MEA              LMV stripe EA
 654  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 655  *      otherwise               Default LOV EA.
 656  * Each time, it can only retrieve 1 stripe EA
 657  **/
 658 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 659                      struct ptlrpc_request **request, u64 valid)
 660 {
 661         struct ll_sb_info *sbi = ll_i2sbi(inode);
 662         struct mdt_body   *body;
 663         struct lov_mds_md *lmm = NULL;
 664         struct ptlrpc_request *req = NULL;
 665         int rc, lmm_size;
 666         struct md_op_data *op_data;
 667         ENTRY;
 668
 669         rc = ll_get_default_mdsize(sbi, &lmm_size);
 670         if (rc)
 671                 RETURN(rc);
 672
 673         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 674                                      0, lmm_size, LUSTRE_OPC_ANY,
 675                                      NULL);
 676         if (IS_ERR(op_data))
 677                 RETURN(PTR_ERR(op_data));
 678
 679         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 680         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 681         ll_finish_md_op_data(op_data);
 682         if (rc < 0) {
 683                 CDEBUG(D_INFO, "md_getattr failed on inode "
 684                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 685                 GOTO(out, rc);
 686         }
 687
 688         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 689         LASSERT(body != NULL);
 690
 691         lmm_size = body->mbo_eadatasize;
 692
 693         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 694             lmm_size == 0) {
 695                 GOTO(out, rc = -ENODATA);
 696         }
 697
 698         lmm = req_capsule_server_sized_get(&req->rq_pill,
 699                                            &RMF_MDT_MD, lmm_size);
 700         LASSERT(lmm != NULL);
 701
 702         /*
 703          * This is coming from the MDS, so is probably in
 704          * little endian.  We convert it to host endian before
 705          * passing it to userspace.
 706          */
 707         /* We don't swab objects for directories */
 708         switch (le32_to_cpu(lmm->lmm_magic)) {
 709         case LOV_MAGIC_V1:
 710                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 711                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 712                 break;
 713         case LOV_MAGIC_V3:
 714                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 715                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 716                 break;
 717         case LMV_MAGIC_V1:
 718                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 719                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 720                 break;
 721         case LMV_USER_MAGIC:
 722                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 723                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 724                 break;
 725         default:
 726                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 727                 rc = -EPROTO;
 728         }
 729 out:
 730         *plmm = lmm;
 731         *plmm_size = lmm_size;
 732         *request = req;
 733         return rc;
 734 }
 735
 736 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 737 {
 738         struct md_op_data       *op_data;
 739         int                     rc;
 740         int                     mdt_index;
 741         ENTRY;
 742
 743         OBD_ALLOC_PTR(op_data);
 744         if (op_data == NULL)
 745                 RETURN(-ENOMEM);
 746
 747         op_data->op_flags |= MF_GET_MDT_IDX;
 748         op_data->op_fid1 = *fid;
 749         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 750         mdt_index = op_data->op_mds;
 751         OBD_FREE_PTR(op_data);
 752         if (rc < 0)
 753                 RETURN(rc);
 754
 755         RETURN(mdt_index);
 756 }
 757
 758 /*
 759  *  Get MDT index for the inode.
 760  */
 761 int ll_get_mdt_idx(struct inode *inode)
 762 {
 763         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 764 }
 765
 766 /**
 767  * Generic handler to do any pre-copy work.
 768  *
 769  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 770  * first information for it that real work has started.
 771  *
 772  * Moreover, for a ARCHIVE request, it will sample the file data version and
 773  * store it in \a copy.
 774  *
 775  * \return 0 on success.
 776  */
 777 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 778 {
 779         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 780         struct hsm_progress_kernel       hpk;
 781         int                              rc = 0;
 782         int                              rc2;
 783         ENTRY;
 784
 785         /* Forge a hsm_progress based on data from copy. */
 786         hpk.hpk_fid = copy->hc_hai.hai_fid;
 787         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 788         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 789         hpk.hpk_extent.length = 0;
 790         hpk.hpk_flags = 0;
 791         hpk.hpk_errval = 0;
 792         hpk.hpk_data_version = 0;
 793
 794
 795         /* For archive request, we need to read the current file version. */
 796         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 797                 struct inode    *inode;
 798                 __u64            data_version = 0;
 799
 800                 /* Get inode for this fid */
 801                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 802                 if (IS_ERR(inode)) {
 803                         hpk.hpk_flags |= HP_FLAG_RETRY;
 804                         /* hpk_errval is >= 0 */
 805                         hpk.hpk_errval = -PTR_ERR(inode);
 806                         GOTO(progress, rc = PTR_ERR(inode));
 807                 }
 808
 809                 /* Read current file data version */
 810                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 811                 iput(inode);
 812                 if (rc != 0) {
 813                         CDEBUG(D_HSM, "Could not read file data version of "
 814                                       DFID" (rc = %d). Archive request ("
 815                                       LPX64") could not be done.\n",
 816                                       PFID(&copy->hc_hai.hai_fid), rc,
 817                                       copy->hc_hai.hai_cookie);
 818                         hpk.hpk_flags |= HP_FLAG_RETRY;
 819                         /* hpk_errval must be >= 0 */
 820                         hpk.hpk_errval = -rc;
 821                         GOTO(progress, rc);
 822                 }
 823
 824                 /* Store in the hsm_copy for later copytool use.
 825                  * Always modified even if no lsm. */
 826                 copy->hc_data_version = data_version;
 827         }
 828
 829 progress:
 830         /* On error, the request should be considered as completed */
 831         if (hpk.hpk_errval > 0)
 832                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 833
 834         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 835                             &hpk, NULL);
 836
 837         /* Return first error */
 838         RETURN(rc != 0 ? rc : rc2);
 839 }
 840
 841 /**
 842  * Generic handler to do any post-copy work.
 843  *
 844  * It will send the last hsm_progress update to coordinator to inform it
 845  * that copy is finished and whether it was successful or not.
 846  *
 847  * Moreover,
 848  * - for ARCHIVE request, it will sample the file data version and compare it
 849  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 850  *   will be considered as failed.
 851  * - for RESTORE request, it will sample the file data version and send it to
 852  *   coordinator which is useful if the file was imported as 'released'.
 853  *
 854  * \return 0 on success.
 855  */
 856 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 857 {
 858         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 859         struct hsm_progress_kernel       hpk;
 860         int                              rc = 0;
 861         int                              rc2;
 862         ENTRY;
 863
 864         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 865         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 866          * initialized if copy_end was called with copy == NULL.
 867          */
 868
 869         /* Forge a hsm_progress based on data from copy. */
 870         hpk.hpk_fid = copy->hc_hai.hai_fid;
 871         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 872         hpk.hpk_extent = copy->hc_hai.hai_extent;
 873         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 874         hpk.hpk_errval = copy->hc_errval;
 875         hpk.hpk_data_version = 0;
 876
 877         /* For archive request, we need to check the file data was not changed.
 878          *
 879          * For restore request, we need to send the file data version, this is
 880          * useful when the file was created using hsm_import.
 881          */
 882         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 883              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 884             (copy->hc_errval == 0)) {
 885                 struct inode    *inode;
 886                 __u64            data_version = 0;
 887
 888                 /* Get lsm for this fid */
 889                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 890                 if (IS_ERR(inode)) {
 891                         hpk.hpk_flags |= HP_FLAG_RETRY;
 892                         /* hpk_errval must be >= 0 */
 893                         hpk.hpk_errval = -PTR_ERR(inode);
 894                         GOTO(progress, rc = PTR_ERR(inode));
 895                 }
 896
 897                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 898                 iput(inode);
 899                 if (rc) {
 900                         CDEBUG(D_HSM, "Could not read file data version. "
 901                                       "Request could not be confirmed.\n");
 902                         if (hpk.hpk_errval == 0)
 903                                 hpk.hpk_errval = -rc;
 904                         GOTO(progress, rc);
 905                 }
 906
 907                 /* Store in the hsm_copy for later copytool use.
 908                  * Always modified even if no lsm. */
 909                 hpk.hpk_data_version = data_version;
 910
 911                 /* File could have been stripped during archiving, so we need
 912                  * to check anyway. */
 913                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 914                     (copy->hc_data_version != data_version)) {
 915                         CDEBUG(D_HSM, "File data version mismatched. "
 916                               "File content was changed during archiving. "
 917                                DFID", start:"LPX64" current:"LPX64"\n",
 918                                PFID(&copy->hc_hai.hai_fid),
 919                                copy->hc_data_version, data_version);
 920                         /* File was changed, send error to cdt. Do not ask for
 921                          * retry because if a file is modified frequently,
 922                          * the cdt will loop on retried archive requests.
 923                          * The policy engine will ask for a new archive later
 924                          * when the file will not be modified for some tunable
 925                          * time */
 926                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 927                         rc = -EBUSY;
 928                         /* hpk_errval must be >= 0 */
 929                         hpk.hpk_errval = -rc;
 930                         GOTO(progress, rc);
 931                 }
 932
 933         }
 934
 935 progress:
 936         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 937                             &hpk, NULL);
 938
 939         /* Return first error */
 940         RETURN(rc != 0 ? rc : rc2);
 941 }
 942
 943
 944 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 945                           const void __user *data, size_t size)
 946 {
 947         void *copy;
 948         int rc;
 949
 950         OBD_ALLOC(copy, size);
 951         if (copy == NULL)
 952                 return -ENOMEM;
 953
 954         if (copy_from_user(copy, data, size)) {
 955                 rc = -EFAULT;
 956                 goto out;
 957         }
 958
 959         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 960 out:
 961         OBD_FREE(copy, size);
 962
 963         return rc;
 964 }
 965
 966 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 967 {
 968         int cmd = qctl->qc_cmd;
 969         int type = qctl->qc_type;
 970         int id = qctl->qc_id;
 971         int valid = qctl->qc_valid;
 972         int rc = 0;
 973         ENTRY;
 974
 975         switch (cmd) {
 976         case Q_SETQUOTA:
 977         case Q_SETINFO:
 978                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
 979                         RETURN(-EPERM);
 980                 break;
 981         case Q_GETQUOTA:
 982                 if (((type == USRQUOTA &&
 983                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 984                      (type == GRPQUOTA &&
 985                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 986                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
 987                         RETURN(-EPERM);
 988                 break;
 989         case Q_GETINFO:
 990                 break;
 991         default:
 992                 CERROR("unsupported quotactl op: %#x\n", cmd);
 993                 RETURN(-ENOTTY);
 994         }
 995
 996         if (valid != QC_GENERAL) {
 997                 if (cmd == Q_GETINFO)
 998                         qctl->qc_cmd = Q_GETOINFO;
 999                 else if (cmd == Q_GETQUOTA)
1000                         qctl->qc_cmd = Q_GETOQUOTA;
1001                 else
1002                         RETURN(-EINVAL);
1003
1004                 switch (valid) {
1005                 case QC_MDTIDX:
1006                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1007                                            sizeof(*qctl), qctl, NULL);
1008                         break;
1009                 case QC_OSTIDX:
1010                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1011                                            sizeof(*qctl), qctl, NULL);
1012                         break;
1013                 case QC_UUID:
1014                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1015                                            sizeof(*qctl), qctl, NULL);
1016                         if (rc == -EAGAIN)
1017                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1018                                                    sbi->ll_dt_exp,
1019                                                    sizeof(*qctl), qctl, NULL);
1020                         break;
1021                 default:
1022                         rc = -EINVAL;
1023                         break;
1024                 }
1025
1026                 if (rc)
1027                         RETURN(rc);
1028
1029                 qctl->qc_cmd = cmd;
1030         } else {
1031                 struct obd_quotactl *oqctl;
1032
1033                 OBD_ALLOC_PTR(oqctl);
1034                 if (oqctl == NULL)
1035                         RETURN(-ENOMEM);
1036
1037                 QCTL_COPY(oqctl, qctl);
1038                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1039                 if (rc) {
1040                         OBD_FREE_PTR(oqctl);
1041                         RETURN(rc);
1042                 }
1043                 /* If QIF_SPACE is not set, client should collect the
1044                  * space usage from OSSs by itself */
1045                 if (cmd == Q_GETQUOTA &&
1046                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1047                     !oqctl->qc_dqblk.dqb_curspace) {
1048                         struct obd_quotactl *oqctl_tmp;
1049
1050                         OBD_ALLOC_PTR(oqctl_tmp);
1051                         if (oqctl_tmp == NULL)
1052                                 GOTO(out, rc = -ENOMEM);
1053
1054                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1055                         oqctl_tmp->qc_id = oqctl->qc_id;
1056                         oqctl_tmp->qc_type = oqctl->qc_type;
1057
1058                         /* collect space usage from OSTs */
1059                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1060                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1061                         if (!rc || rc == -EREMOTEIO) {
1062                                 oqctl->qc_dqblk.dqb_curspace =
1063                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1064                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1065                         }
1066
1067                         /* collect space & inode usage from MDTs */
1068                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1069                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1070                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1071                         if (!rc || rc == -EREMOTEIO) {
1072                                 oqctl->qc_dqblk.dqb_curspace +=
1073                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1074                                 oqctl->qc_dqblk.dqb_curinodes =
1075                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1076                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1077                         } else {
1078                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1079                         }
1080
1081                         OBD_FREE_PTR(oqctl_tmp);
1082                 }
1083 out:
1084                 QCTL_COPY(qctl, oqctl);
1085                 OBD_FREE_PTR(oqctl);
1086         }
1087
1088         RETURN(rc);
1089 }
1090
1091 /* This function tries to get a single name component,
1092  * to send to the server. No actual path traversal involved,
1093  * so we limit to NAME_MAX */
1094 static char *ll_getname(const char __user *filename)
1095 {
1096         int ret = 0, len;
1097         char *tmp;
1098
1099         OBD_ALLOC(tmp, NAME_MAX + 1);
1100
1101         if (!tmp)
1102                 return ERR_PTR(-ENOMEM);
1103
1104         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1105         if (len < 0)
1106                 ret = -ENOENT;
1107         else if (len > NAME_MAX)
1108                 ret = -ENAMETOOLONG;
1109
1110         if (ret) {
1111                 OBD_FREE(tmp, NAME_MAX + 1);
1112                 tmp =  ERR_PTR(ret);
1113         }
1114         return tmp;
1115 }
1116
1117 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1118
1119 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1120 {
1121         struct dentry *dentry = file_dentry(file);
1122         struct inode *inode = file_inode(file);
1123         struct ll_sb_info *sbi = ll_i2sbi(inode);
1124         struct obd_ioctl_data *data;
1125         int rc = 0;
1126         ENTRY;
1127
1128         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1129                PFID(ll_inode2fid(inode)), inode, cmd);
1130
1131         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1132         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1133                 return -ENOTTY;
1134
1135         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1136         switch(cmd) {
1137         case FSFILT_IOC_GETFLAGS:
1138         case FSFILT_IOC_SETFLAGS:
1139                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1140         case FSFILT_IOC_GETVERSION_OLD:
1141         case FSFILT_IOC_GETVERSION:
1142                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1143         /* We need to special case any other ioctls we want to handle,
1144          * to send them to the MDS/OST as appropriate and to properly
1145          * network encode the arg field.
1146         case FSFILT_IOC_SETVERSION_OLD:
1147         case FSFILT_IOC_SETVERSION:
1148         */
1149         case LL_IOC_GET_MDTIDX: {
1150                 int mdtidx;
1151
1152                 mdtidx = ll_get_mdt_idx(inode);
1153                 if (mdtidx < 0)
1154                         RETURN(mdtidx);
1155
1156                 if (put_user((int)mdtidx, (int __user *)arg))
1157                         RETURN(-EFAULT);
1158
1159                 return 0;
1160         }
1161         case IOC_MDC_LOOKUP: {
1162                 int namelen, len = 0;
1163                 char *buf = NULL;
1164                 char *filename;
1165
1166                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1167                 if (rc != 0)
1168                         RETURN(rc);
1169                 data = (void *)buf;
1170
1171                 filename = data->ioc_inlbuf1;
1172                 namelen = strlen(filename);
1173                 if (namelen < 1) {
1174                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1175                         GOTO(out_free, rc = -EINVAL);
1176                 }
1177
1178                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1179                 if (rc < 0) {
1180                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1181                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1182                                filename, rc);
1183                         GOTO(out_free, rc);
1184                 }
1185 out_free:
1186                 obd_ioctl_freedata(buf, len);
1187                 return rc;
1188         }
1189         case LL_IOC_LMV_SETSTRIPE: {
1190                 struct lmv_user_md  *lum;
1191                 char            *buf = NULL;
1192                 char            *filename;
1193                 int              namelen = 0;
1194                 int              lumlen = 0;
1195                 umode_t          mode;
1196                 int              len;
1197                 int              rc;
1198
1199                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1200                 if (rc)
1201                         RETURN(rc);
1202
1203                 data = (void *)buf;
1204                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1205                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1206                         GOTO(lmv_out_free, rc = -EINVAL);
1207
1208                 filename = data->ioc_inlbuf1;
1209                 namelen = data->ioc_inllen1;
1210
1211                 if (namelen < 1) {
1212                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1213                         GOTO(lmv_out_free, rc = -EINVAL);
1214                 }
1215                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1216                 lumlen = data->ioc_inllen2;
1217
1218                 if (lum->lum_magic != LMV_USER_MAGIC ||
1219                     lumlen != sizeof(*lum)) {
1220                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1221                                filename, lum->lum_magic, lumlen, -EFAULT);
1222                         GOTO(lmv_out_free, rc = -EINVAL);
1223                 }
1224
1225 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1226                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1227 #else
1228                 mode = data->ioc_type;
1229 #endif
1230                 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1231 lmv_out_free:
1232                 obd_ioctl_freedata(buf, len);
1233                 RETURN(rc);
1234
1235         }
1236         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1237                 struct lmv_user_md        lum;
1238                 struct lmv_user_md __user *ulump =
1239                                         (struct lmv_user_md __user *)arg;
1240                 int                       rc;
1241
1242                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1243                         RETURN(-EFAULT);
1244
1245                 if (lum.lum_magic != LMV_USER_MAGIC)
1246                         RETURN(-EINVAL);
1247
1248                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1249
1250                 RETURN(rc);
1251         }
1252         case LL_IOC_LOV_SETSTRIPE: {
1253                 struct lov_user_md_v3 lumv3;
1254                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1255                 struct lov_user_md_v1 __user *lumv1p =
1256                         (struct lov_user_md_v1 __user *)arg;
1257                 struct lov_user_md_v3 __user *lumv3p =
1258                         (struct lov_user_md_v3 __user *)arg;
1259
1260                 int set_default = 0;
1261
1262                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1263                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1264                         sizeof(lumv3p->lmm_objects[0]));
1265                 /* first try with v1 which is smaller than v3 */
1266                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1267                         RETURN(-EFAULT);
1268
1269                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1270                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1271                                 RETURN(-EFAULT);
1272                 }
1273
1274                 if (inode->i_sb->s_root == file_dentry(file))
1275                         set_default = 1;
1276
1277                 /* in v1 and v3 cases lumv1 points to data */
1278                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1279
1280                 RETURN(rc);
1281         }
1282         case LL_IOC_LMV_GETSTRIPE: {
1283                 struct lmv_user_md __user *ulmv =
1284                                         (struct lmv_user_md __user *)arg;
1285                 struct lmv_user_md      lum;
1286                 struct ptlrpc_request   *request = NULL;
1287                 union lmv_mds_md        *lmm = NULL;
1288                 int                     lmmsize;
1289                 u64                     valid = 0;
1290                 struct lmv_user_md      *tmp = NULL;
1291                 int                     mdt_index;
1292                 int                     lum_size;
1293                 int                     stripe_count;
1294                 int                     max_stripe_count;
1295                 int                     i;
1296                 int                     rc;
1297
1298                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1299                         RETURN(-EFAULT);
1300
1301                 max_stripe_count = lum.lum_stripe_count;
1302                 /* lum_magic will indicate which stripe the ioctl will like
1303                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1304                  * is for default LMV stripe */
1305                 if (lum.lum_magic == LMV_MAGIC_V1)
1306                         valid |= OBD_MD_MEA;
1307                 else if (lum.lum_magic == LMV_USER_MAGIC)
1308                         valid |= OBD_MD_DEFAULT_MEA;
1309                 else
1310                         RETURN(-EINVAL);
1311
1312                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1313                                       valid);
1314                 if (rc != 0)
1315                         GOTO(finish_req, rc);
1316
1317                 /* Get default LMV EA */
1318                 if (lum.lum_magic == LMV_USER_MAGIC) {
1319                         if (lmmsize > sizeof(*ulmv))
1320                                 GOTO(finish_req, rc = -EINVAL);
1321
1322                         if (copy_to_user(ulmv, lmm, lmmsize))
1323                                 GOTO(finish_req, rc = -EFAULT);
1324
1325                         GOTO(finish_req, rc);
1326                 }
1327
1328                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1329                 if (max_stripe_count < stripe_count) {
1330                         lum.lum_stripe_count = stripe_count;
1331                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1332                                 GOTO(finish_req, rc = -EFAULT);
1333                         GOTO(finish_req, rc = -E2BIG);
1334                 }
1335
1336                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1337                 OBD_ALLOC(tmp, lum_size);
1338                 if (tmp == NULL)
1339                         GOTO(finish_req, rc = -ENOMEM);
1340
1341                 mdt_index = ll_get_mdt_idx(inode);
1342                 if (mdt_index < 0)
1343                         GOTO(out_tmp, rc = -ENOMEM);
1344
1345                 tmp->lum_magic = LMV_MAGIC_V1;
1346                 tmp->lum_stripe_count = 0;
1347                 tmp->lum_stripe_offset = mdt_index;
1348                 for (i = 0; i < stripe_count; i++) {
1349                         struct lu_fid   fid;
1350
1351                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1352                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1353                         if (mdt_index < 0)
1354                                 GOTO(out_tmp, rc = mdt_index);
1355
1356                         tmp->lum_objects[i].lum_mds = mdt_index;
1357                         tmp->lum_objects[i].lum_fid = fid;
1358                         tmp->lum_stripe_count++;
1359                 }
1360
1361                 if (copy_to_user(ulmv, tmp, lum_size))
1362                         GOTO(out_tmp, rc = -EFAULT);
1363 out_tmp:
1364                 OBD_FREE(tmp, lum_size);
1365 finish_req:
1366                 ptlrpc_req_finished(request);
1367                 return rc;
1368         }
1369
1370         case LL_IOC_REMOVE_ENTRY: {
1371                 char            *filename = NULL;
1372                 int              namelen = 0;
1373                 int              rc;
1374
1375                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1376                  * unsupported server, which might crash the server(LU-2730),
1377                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1378                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1379                  * server will support REINT_RMENTRY XXX*/
1380                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1381                         RETURN(-ENOTSUPP);
1382
1383                 filename = ll_getname((const char __user *)arg);
1384                 if (IS_ERR(filename))
1385                         RETURN(PTR_ERR(filename));
1386
1387                 namelen = strlen(filename);
1388                 if (namelen < 1)
1389                         GOTO(out_rmdir, rc = -EINVAL);
1390
1391                 rc = ll_rmdir_entry(inode, filename, namelen);
1392 out_rmdir:
1393                 if (filename)
1394                         ll_putname(filename);
1395                 RETURN(rc);
1396         }
1397         case LL_IOC_LOV_SWAP_LAYOUTS:
1398                 RETURN(-EPERM);
1399         case IOC_OBD_STATFS:
1400                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1401         case LL_IOC_LOV_GETSTRIPE:
1402         case LL_IOC_MDC_GETINFO:
1403         case IOC_MDC_GETFILEINFO:
1404         case IOC_MDC_GETFILESTRIPE: {
1405                 struct ptlrpc_request *request = NULL;
1406                 struct lov_user_md __user *lump;
1407                 struct lov_mds_md *lmm = NULL;
1408                 struct mdt_body *body;
1409                 char *filename = NULL;
1410                 int lmmsize;
1411
1412                 if (cmd == IOC_MDC_GETFILEINFO ||
1413                     cmd == IOC_MDC_GETFILESTRIPE) {
1414                         filename = ll_getname((const char __user *)arg);
1415                         if (IS_ERR(filename))
1416                                 RETURN(PTR_ERR(filename));
1417
1418                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1419                                                       &lmmsize, &request);
1420                 } else {
1421                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1422                                               &request, 0);
1423                 }
1424
1425                 if (request) {
1426                         body = req_capsule_server_get(&request->rq_pill,
1427                                                       &RMF_MDT_BODY);
1428                         LASSERT(body != NULL);
1429                 } else {
1430                         GOTO(out_req, rc);
1431                 }
1432
1433                 if (rc < 0) {
1434                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1435                                                cmd == LL_IOC_MDC_GETINFO))
1436                                 GOTO(skip_lmm, rc = 0);
1437                         else
1438                                 GOTO(out_req, rc);
1439                 }
1440
1441                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1442                     cmd == LL_IOC_LOV_GETSTRIPE) {
1443                         lump = (struct lov_user_md __user *)arg;
1444                 } else {
1445                         struct lov_user_mds_data __user *lmdp;
1446                         lmdp = (struct lov_user_mds_data __user *)arg;
1447                         lump = &lmdp->lmd_lmm;
1448                 }
1449                 if (copy_to_user(lump, lmm, lmmsize)) {
1450                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1451                                 GOTO(out_req, rc = -EFAULT);
1452                         rc = -EOVERFLOW;
1453                 }
1454         skip_lmm:
1455                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1456                         struct lov_user_mds_data __user *lmdp;
1457                         lstat_t st = { 0 };
1458
1459                         st.st_dev       = inode->i_sb->s_dev;
1460                         st.st_mode      = body->mbo_mode;
1461                         st.st_nlink     = body->mbo_nlink;
1462                         st.st_uid       = body->mbo_uid;
1463                         st.st_gid       = body->mbo_gid;
1464                         st.st_rdev      = body->mbo_rdev;
1465                         st.st_size      = body->mbo_size;
1466                         st.st_blksize   = PAGE_CACHE_SIZE;
1467                         st.st_blocks    = body->mbo_blocks;
1468                         st.st_atime     = body->mbo_atime;
1469                         st.st_mtime     = body->mbo_mtime;
1470                         st.st_ctime     = body->mbo_ctime;
1471                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1472                                                 sbi->ll_flags &
1473                                                 LL_SBI_32BIT_API);
1474
1475                         lmdp = (struct lov_user_mds_data __user *)arg;
1476                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1477                                 GOTO(out_req, rc = -EFAULT);
1478                 }
1479
1480                 EXIT;
1481         out_req:
1482                 ptlrpc_req_finished(request);
1483                 if (filename)
1484                         ll_putname(filename);
1485                 return rc;
1486         }
1487         case OBD_IOC_QUOTACTL: {
1488                 struct if_quotactl *qctl;
1489
1490                 OBD_ALLOC_PTR(qctl);
1491                 if (!qctl)
1492                         RETURN(-ENOMEM);
1493
1494                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1495                         GOTO(out_quotactl, rc = -EFAULT);
1496
1497                 rc = quotactl_ioctl(sbi, qctl);
1498
1499                 if (rc == 0 &&
1500                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1501                         rc = -EFAULT;
1502
1503         out_quotactl:
1504                 OBD_FREE_PTR(qctl);
1505                 RETURN(rc);
1506         }
1507         case OBD_IOC_GETDTNAME:
1508         case OBD_IOC_GETMDNAME:
1509                 RETURN(ll_get_obd_name(inode, cmd, arg));
1510         case LL_IOC_FLUSHCTX:
1511                 RETURN(ll_flush_ctx(inode));
1512         case LL_IOC_GETOBDCOUNT: {
1513                 u32 count, vallen;
1514                 struct obd_export *exp;
1515
1516                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1517                         RETURN(-EFAULT);
1518
1519                 /* get ost count when count is zero, get mdt count otherwise */
1520                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1521                 vallen = sizeof(count);
1522                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1523                                   KEY_TGT_COUNT, &vallen, &count);
1524                 if (rc) {
1525                         CERROR("get target count failed: %d\n", rc);
1526                         RETURN(rc);
1527                 }
1528
1529                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1530                         RETURN(-EFAULT);
1531
1532                 RETURN(0);
1533         }
1534         case LL_IOC_PATH2FID:
1535                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1536                                      sizeof(struct lu_fid)))
1537                         RETURN(-EFAULT);
1538                 RETURN(0);
1539         case LL_IOC_GET_CONNECT_FLAGS: {
1540                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1541                                      (void __user *)arg));
1542         }
1543         case OBD_IOC_CHANGELOG_SEND:
1544         case OBD_IOC_CHANGELOG_CLEAR:
1545                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1546                         RETURN(-EPERM);
1547
1548                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1549                                     sizeof(struct ioc_changelog));
1550                 RETURN(rc);
1551         case OBD_IOC_FID2PATH:
1552                 RETURN(ll_fid2path(inode, (void __user *)arg));
1553         case LL_IOC_GETPARENT:
1554                 RETURN(ll_getparent(file, (void __user *)arg));
1555         case LL_IOC_FID2MDTIDX: {
1556                 struct obd_export *exp = ll_i2mdexp(inode);
1557                 struct lu_fid     fid;
1558                 __u32             index;
1559
1560                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1561                                    sizeof(fid)))
1562                         RETURN(-EFAULT);
1563
1564                 /* Call mdc_iocontrol */
1565                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1566                                    (__u32 __user *)&index);
1567                 if (rc != 0)
1568                         RETURN(rc);
1569
1570                 RETURN(index);
1571         }
1572         case LL_IOC_HSM_REQUEST: {
1573                 struct hsm_user_request *hur;
1574                 ssize_t                  totalsize;
1575
1576                 OBD_ALLOC_PTR(hur);
1577                 if (hur == NULL)
1578                         RETURN(-ENOMEM);
1579
1580                 /* We don't know the true size yet; copy the fixed-size part */
1581                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1582                         OBD_FREE_PTR(hur);
1583                         RETURN(-EFAULT);
1584                 }
1585
1586                 /* Compute the whole struct size */
1587                 totalsize = hur_len(hur);
1588                 OBD_FREE_PTR(hur);
1589                 if (totalsize < 0)
1590                         RETURN(-E2BIG);
1591
1592                 /* Final size will be more than double totalsize */
1593                 if (totalsize >= MDS_MAXREQSIZE / 3)
1594                         RETURN(-E2BIG);
1595
1596                 OBD_ALLOC_LARGE(hur, totalsize);
1597                 if (hur == NULL)
1598                         RETURN(-ENOMEM);
1599
1600                 /* Copy the whole struct */
1601                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1602                         GOTO(out_hur, rc = -EFAULT);
1603
1604                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1605                         const struct lu_fid *fid;
1606                         struct inode *f;
1607                         int i;
1608
1609                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1610                                 fid = &hur->hur_user_item[i].hui_fid;
1611                                 f = search_inode_for_lustre(inode->i_sb, fid);
1612                                 if (IS_ERR(f)) {
1613                                         rc = PTR_ERR(f);
1614                                         break;
1615                                 }
1616
1617                                 rc = ll_hsm_release(f);
1618                                 iput(f);
1619                                 if (rc != 0)
1620                                         break;
1621                         }
1622                 } else {
1623                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1624                                            hur, NULL);
1625                 }
1626
1627 out_hur:
1628                 OBD_FREE_LARGE(hur, totalsize);
1629
1630                 RETURN(rc);
1631         }
1632         case LL_IOC_HSM_PROGRESS: {
1633                 struct hsm_progress_kernel      hpk;
1634                 struct hsm_progress             hp;
1635
1636                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1637                         RETURN(-EFAULT);
1638
1639                 hpk.hpk_fid = hp.hp_fid;
1640                 hpk.hpk_cookie = hp.hp_cookie;
1641                 hpk.hpk_extent = hp.hp_extent;
1642                 hpk.hpk_flags = hp.hp_flags;
1643                 hpk.hpk_errval = hp.hp_errval;
1644                 hpk.hpk_data_version = 0;
1645
1646                 /* File may not exist in Lustre; all progress
1647                  * reported to Lustre root */
1648                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1649                                    NULL);
1650                 RETURN(rc);
1651         }
1652         case LL_IOC_HSM_CT_START:
1653                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1654                         RETURN(-EPERM);
1655
1656                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1657                                     sizeof(struct lustre_kernelcomm));
1658                 RETURN(rc);
1659
1660         case LL_IOC_HSM_COPY_START: {
1661                 struct hsm_copy *copy;
1662                 int              rc;
1663
1664                 OBD_ALLOC_PTR(copy);
1665                 if (copy == NULL)
1666                         RETURN(-ENOMEM);
1667                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1668                         OBD_FREE_PTR(copy);
1669                         RETURN(-EFAULT);
1670                 }
1671
1672                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1673                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1674                         rc = -EFAULT;
1675
1676                 OBD_FREE_PTR(copy);
1677                 RETURN(rc);
1678         }
1679         case LL_IOC_HSM_COPY_END: {
1680                 struct hsm_copy *copy;
1681                 int              rc;
1682
1683                 OBD_ALLOC_PTR(copy);
1684                 if (copy == NULL)
1685                         RETURN(-ENOMEM);
1686                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1687                         OBD_FREE_PTR(copy);
1688                         RETURN(-EFAULT);
1689                 }
1690
1691                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1692                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1693                         rc = -EFAULT;
1694
1695                 OBD_FREE_PTR(copy);
1696                 RETURN(rc);
1697         }
1698         case LL_IOC_MIGRATE: {
1699                 char            *buf = NULL;
1700                 const char      *filename;
1701                 int             namelen = 0;
1702                 int             len;
1703                 int             rc;
1704                 int             mdtidx;
1705
1706                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1707                 if (rc < 0)
1708                         RETURN(rc);
1709
1710                 data = (struct obd_ioctl_data *)buf;
1711                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1712                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1713                         GOTO(migrate_free, rc = -EINVAL);
1714
1715                 filename = data->ioc_inlbuf1;
1716                 namelen = data->ioc_inllen1;
1717                 /* \0 is packed at the end of filename */
1718                 if (namelen < 1 || namelen != strlen(filename) + 1)
1719                         GOTO(migrate_free, rc = -EINVAL);
1720
1721                 if (data->ioc_inllen2 != sizeof(mdtidx))
1722                         GOTO(migrate_free, rc = -EINVAL);
1723                 mdtidx = *(int *)data->ioc_inlbuf2;
1724
1725                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1726 migrate_free:
1727                 obd_ioctl_freedata(buf, len);
1728
1729                 RETURN(rc);
1730         }
1731         default:
1732                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1733                                      (void __user *)arg));
1734         }
1735 }
1736
1737 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1738 {
1739         struct inode *inode = file->f_mapping->host;
1740         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1741         struct ll_sb_info *sbi = ll_i2sbi(inode);
1742         int api32 = ll_need_32bit_api(sbi);
1743         loff_t ret = -EINVAL;
1744         ENTRY;
1745
1746         inode_lock(inode);
1747         switch (origin) {
1748                 case SEEK_SET:
1749                         break;
1750                 case SEEK_CUR:
1751                         offset += file->f_pos;
1752                         break;
1753                 case SEEK_END:
1754                         if (offset > 0)
1755                                 GOTO(out, ret);
1756                         if (api32)
1757                                 offset += LL_DIR_END_OFF_32BIT;
1758                         else
1759                                 offset += LL_DIR_END_OFF;
1760                         break;
1761                 default:
1762                         GOTO(out, ret);
1763         }
1764
1765         if (offset >= 0 &&
1766             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1767              (!api32 && offset <= LL_DIR_END_OFF))) {
1768                 if (offset != file->f_pos) {
1769                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1770                             (!api32 && offset == LL_DIR_END_OFF))
1771                                 fd->lfd_pos = MDS_DIR_END_OFF;
1772                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1773                                 fd->lfd_pos = offset << 32;
1774                         else
1775                                 fd->lfd_pos = offset;
1776                         file->f_pos = offset;
1777                         file->f_version = 0;
1778                 }
1779                 ret = offset;
1780         }
1781         GOTO(out, ret);
1782
1783 out:
1784         inode_unlock(inode);
1785         return ret;
1786 }
1787
1788 static int ll_dir_open(struct inode *inode, struct file *file)
1789 {
1790         ENTRY;
1791         RETURN(ll_file_open(inode, file));
1792 }
1793
1794 static int ll_dir_release(struct inode *inode, struct file *file)
1795 {
1796         ENTRY;
1797         RETURN(ll_file_release(inode, file));
1798 }
1799
1800 const struct file_operations ll_dir_operations = {
1801         .llseek         = ll_dir_seek,
1802         .open           = ll_dir_open,
1803         .release        = ll_dir_release,
1804         .read           = generic_read_dir,
1805 #ifdef HAVE_DIR_CONTEXT
1806         .iterate        = ll_iterate,
1807 #else
1808         .readdir        = ll_readdir,
1809 #endif
1810         .unlocked_ioctl = ll_dir_ioctl,
1811         .fsync          = ll_fsync,
1812 };