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 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                 RETURN(rc);
 603
 604 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 605         /*
 606          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 607          * and it's stored into system config for backward compatibility.
 608          *
 609          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 610          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 611          * need the make the distiction between the 2 versions
 612          */
 613         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 614                 char *param = NULL;
 615                 char *buf;
 616
 617                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 618                 if (param == NULL)
 619                         GOTO(end, rc = -ENOMEM);
 620
 621                 buf = param;
 622                 /* Get fsname and assume devname to be -MDT0000. */
 623                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 624                 strcat(buf, "-MDT0000.lov");
 625                 buf += strlen(buf);
 626
 627                 /* Set root stripesize */
 628                 sprintf(buf, ".stripesize=%u",
 629                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 630                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 631                 if (rc)
 632                         GOTO(end, rc);
 633
 634                 /* Set root stripecount */
 635                 sprintf(buf, ".stripecount=%hd",
 636                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 637                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 638                 if (rc)
 639                         GOTO(end, rc);
 640
 641                 /* Set root stripeoffset */
 642                 sprintf(buf, ".stripeoffset=%hd",
 643                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 644                         (typeof(lump->lmm_stripe_offset))(-1));
 645                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 646
 647 end:
 648                 if (param != NULL)
 649                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 650         }
 651 #endif
 652         RETURN(rc);
 653 }
 654
 655 /**
 656  * This function will be used to get default LOV/LMV/Default LMV
 657  * @valid will be used to indicate which stripe it will retrieve
 658  *      OBD_MD_MEA              LMV stripe EA
 659  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 660  *      otherwise               Default LOV EA.
 661  * Each time, it can only retrieve 1 stripe EA
 662  **/
 663 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 664                      struct ptlrpc_request **request, u64 valid)
 665 {
 666         struct ll_sb_info *sbi = ll_i2sbi(inode);
 667         struct mdt_body   *body;
 668         struct lov_mds_md *lmm = NULL;
 669         struct ptlrpc_request *req = NULL;
 670         int rc, lmm_size;
 671         struct md_op_data *op_data;
 672         ENTRY;
 673
 674         rc = ll_get_default_mdsize(sbi, &lmm_size);
 675         if (rc)
 676                 RETURN(rc);
 677
 678         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 679                                      0, lmm_size, LUSTRE_OPC_ANY,
 680                                      NULL);
 681         if (IS_ERR(op_data))
 682                 RETURN(PTR_ERR(op_data));
 683
 684         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 685         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 686         ll_finish_md_op_data(op_data);
 687         if (rc < 0) {
 688                 CDEBUG(D_INFO, "md_getattr failed on inode "
 689                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 690                 GOTO(out, rc);
 691         }
 692
 693         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 694         LASSERT(body != NULL);
 695
 696         lmm_size = body->mbo_eadatasize;
 697
 698         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 699             lmm_size == 0) {
 700                 GOTO(out, rc = -ENODATA);
 701         }
 702
 703         lmm = req_capsule_server_sized_get(&req->rq_pill,
 704                                            &RMF_MDT_MD, lmm_size);
 705         LASSERT(lmm != NULL);
 706
 707         /*
 708          * This is coming from the MDS, so is probably in
 709          * little endian.  We convert it to host endian before
 710          * passing it to userspace.
 711          */
 712         /* We don't swab objects for directories */
 713         switch (le32_to_cpu(lmm->lmm_magic)) {
 714         case LOV_MAGIC_V1:
 715                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 716                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 717                 break;
 718         case LOV_MAGIC_V3:
 719                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 720                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 721                 break;
 722         case LMV_MAGIC_V1:
 723                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 724                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 725                 break;
 726         case LMV_USER_MAGIC:
 727                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 728                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 729                 break;
 730         default:
 731                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 732                 rc = -EPROTO;
 733         }
 734 out:
 735         *plmm = lmm;
 736         *plmm_size = lmm_size;
 737         *request = req;
 738         return rc;
 739 }
 740
 741 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 742 {
 743         struct md_op_data       *op_data;
 744         int                     rc;
 745         int                     mdt_index;
 746         ENTRY;
 747
 748         OBD_ALLOC_PTR(op_data);
 749         if (op_data == NULL)
 750                 RETURN(-ENOMEM);
 751
 752         op_data->op_flags |= MF_GET_MDT_IDX;
 753         op_data->op_fid1 = *fid;
 754         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 755         mdt_index = op_data->op_mds;
 756         OBD_FREE_PTR(op_data);
 757         if (rc < 0)
 758                 RETURN(rc);
 759
 760         RETURN(mdt_index);
 761 }
 762
 763 /*
 764  *  Get MDT index for the inode.
 765  */
 766 int ll_get_mdt_idx(struct inode *inode)
 767 {
 768         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 769 }
 770
 771 /**
 772  * Generic handler to do any pre-copy work.
 773  *
 774  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 775  * first information for it that real work has started.
 776  *
 777  * Moreover, for a ARCHIVE request, it will sample the file data version and
 778  * store it in \a copy.
 779  *
 780  * \return 0 on success.
 781  */
 782 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 783 {
 784         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 785         struct hsm_progress_kernel       hpk;
 786         int                              rc = 0;
 787         int                              rc2;
 788         ENTRY;
 789
 790         /* Forge a hsm_progress based on data from copy. */
 791         hpk.hpk_fid = copy->hc_hai.hai_fid;
 792         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 793         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 794         hpk.hpk_extent.length = 0;
 795         hpk.hpk_flags = 0;
 796         hpk.hpk_errval = 0;
 797         hpk.hpk_data_version = 0;
 798
 799
 800         /* For archive request, we need to read the current file version. */
 801         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 802                 struct inode    *inode;
 803                 __u64            data_version = 0;
 804
 805                 /* Get inode for this fid */
 806                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 807                 if (IS_ERR(inode)) {
 808                         hpk.hpk_flags |= HP_FLAG_RETRY;
 809                         /* hpk_errval is >= 0 */
 810                         hpk.hpk_errval = -PTR_ERR(inode);
 811                         GOTO(progress, rc = PTR_ERR(inode));
 812                 }
 813
 814                 /* Read current file data version */
 815                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 816                 iput(inode);
 817                 if (rc != 0) {
 818                         CDEBUG(D_HSM, "Could not read file data version of "
 819                                       DFID" (rc = %d). Archive request ("
 820                                       "%#llx) could not be done.\n",
 821                                       PFID(&copy->hc_hai.hai_fid), rc,
 822                                       copy->hc_hai.hai_cookie);
 823                         hpk.hpk_flags |= HP_FLAG_RETRY;
 824                         /* hpk_errval must be >= 0 */
 825                         hpk.hpk_errval = -rc;
 826                         GOTO(progress, rc);
 827                 }
 828
 829                 /* Store in the hsm_copy for later copytool use.
 830                  * Always modified even if no lsm. */
 831                 copy->hc_data_version = data_version;
 832         }
 833
 834 progress:
 835         /* On error, the request should be considered as completed */
 836         if (hpk.hpk_errval > 0)
 837                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 838
 839         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 840                             &hpk, NULL);
 841
 842         /* Return first error */
 843         RETURN(rc != 0 ? rc : rc2);
 844 }
 845
 846 /**
 847  * Generic handler to do any post-copy work.
 848  *
 849  * It will send the last hsm_progress update to coordinator to inform it
 850  * that copy is finished and whether it was successful or not.
 851  *
 852  * Moreover,
 853  * - for ARCHIVE request, it will sample the file data version and compare it
 854  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 855  *   will be considered as failed.
 856  * - for RESTORE request, it will sample the file data version and send it to
 857  *   coordinator which is useful if the file was imported as 'released'.
 858  *
 859  * \return 0 on success.
 860  */
 861 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 862 {
 863         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 864         struct hsm_progress_kernel       hpk;
 865         int                              rc = 0;
 866         int                              rc2;
 867         ENTRY;
 868
 869         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 870         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 871          * initialized if copy_end was called with copy == NULL.
 872          */
 873
 874         /* Forge a hsm_progress based on data from copy. */
 875         hpk.hpk_fid = copy->hc_hai.hai_fid;
 876         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 877         hpk.hpk_extent = copy->hc_hai.hai_extent;
 878         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 879         hpk.hpk_errval = copy->hc_errval;
 880         hpk.hpk_data_version = 0;
 881
 882         /* For archive request, we need to check the file data was not changed.
 883          *
 884          * For restore request, we need to send the file data version, this is
 885          * useful when the file was created using hsm_import.
 886          */
 887         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 888              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 889             (copy->hc_errval == 0)) {
 890                 struct inode    *inode;
 891                 __u64            data_version = 0;
 892
 893                 /* Get lsm for this fid */
 894                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 895                 if (IS_ERR(inode)) {
 896                         hpk.hpk_flags |= HP_FLAG_RETRY;
 897                         /* hpk_errval must be >= 0 */
 898                         hpk.hpk_errval = -PTR_ERR(inode);
 899                         GOTO(progress, rc = PTR_ERR(inode));
 900                 }
 901
 902                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 903                 iput(inode);
 904                 if (rc) {
 905                         CDEBUG(D_HSM, "Could not read file data version. "
 906                                       "Request could not be confirmed.\n");
 907                         if (hpk.hpk_errval == 0)
 908                                 hpk.hpk_errval = -rc;
 909                         GOTO(progress, rc);
 910                 }
 911
 912                 /* Store in the hsm_copy for later copytool use.
 913                  * Always modified even if no lsm. */
 914                 hpk.hpk_data_version = data_version;
 915
 916                 /* File could have been stripped during archiving, so we need
 917                  * to check anyway. */
 918                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 919                     (copy->hc_data_version != data_version)) {
 920                         CDEBUG(D_HSM, "File data version mismatched. "
 921                               "File content was changed during archiving. "
 922                                DFID", start:%#llx current:%#llx\n",
 923                                PFID(&copy->hc_hai.hai_fid),
 924                                copy->hc_data_version, data_version);
 925                         /* File was changed, send error to cdt. Do not ask for
 926                          * retry because if a file is modified frequently,
 927                          * the cdt will loop on retried archive requests.
 928                          * The policy engine will ask for a new archive later
 929                          * when the file will not be modified for some tunable
 930                          * time */
 931                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 932                         rc = -EBUSY;
 933                         /* hpk_errval must be >= 0 */
 934                         hpk.hpk_errval = -rc;
 935                         GOTO(progress, rc);
 936                 }
 937
 938         }
 939
 940 progress:
 941         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 942                             &hpk, NULL);
 943
 944         /* Return first error */
 945         RETURN(rc != 0 ? rc : rc2);
 946 }
 947
 948
 949 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 950                           const void __user *data, size_t size)
 951 {
 952         void *copy;
 953         int rc;
 954
 955         OBD_ALLOC(copy, size);
 956         if (copy == NULL)
 957                 return -ENOMEM;
 958
 959         if (copy_from_user(copy, data, size)) {
 960                 rc = -EFAULT;
 961                 goto out;
 962         }
 963
 964         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 965 out:
 966         OBD_FREE(copy, size);
 967
 968         return rc;
 969 }
 970
 971 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 972 {
 973         int cmd = qctl->qc_cmd;
 974         int type = qctl->qc_type;
 975         int id = qctl->qc_id;
 976         int valid = qctl->qc_valid;
 977         int rc = 0;
 978         ENTRY;
 979
 980         switch (cmd) {
 981         case Q_SETQUOTA:
 982         case Q_SETINFO:
 983                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
 984                         RETURN(-EPERM);
 985                 break;
 986         case Q_GETQUOTA:
 987                 if (((type == USRQUOTA &&
 988                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 989                      (type == GRPQUOTA &&
 990                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 991                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
 992                         RETURN(-EPERM);
 993                 break;
 994         case Q_GETINFO:
 995                 break;
 996         default:
 997                 CERROR("unsupported quotactl op: %#x\n", cmd);
 998                 RETURN(-ENOTTY);
 999         }
1000
1001         if (valid != QC_GENERAL) {
1002                 if (cmd == Q_GETINFO)
1003                         qctl->qc_cmd = Q_GETOINFO;
1004                 else if (cmd == Q_GETQUOTA)
1005                         qctl->qc_cmd = Q_GETOQUOTA;
1006                 else
1007                         RETURN(-EINVAL);
1008
1009                 switch (valid) {
1010                 case QC_MDTIDX:
1011                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1012                                            sizeof(*qctl), qctl, NULL);
1013                         break;
1014                 case QC_OSTIDX:
1015                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1016                                            sizeof(*qctl), qctl, NULL);
1017                         break;
1018                 case QC_UUID:
1019                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1020                                            sizeof(*qctl), qctl, NULL);
1021                         if (rc == -EAGAIN)
1022                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1023                                                    sbi->ll_dt_exp,
1024                                                    sizeof(*qctl), qctl, NULL);
1025                         break;
1026                 default:
1027                         rc = -EINVAL;
1028                         break;
1029                 }
1030
1031                 if (rc)
1032                         RETURN(rc);
1033
1034                 qctl->qc_cmd = cmd;
1035         } else {
1036                 struct obd_quotactl *oqctl;
1037
1038                 OBD_ALLOC_PTR(oqctl);
1039                 if (oqctl == NULL)
1040                         RETURN(-ENOMEM);
1041
1042                 QCTL_COPY(oqctl, qctl);
1043                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1044                 if (rc) {
1045                         OBD_FREE_PTR(oqctl);
1046                         RETURN(rc);
1047                 }
1048                 /* If QIF_SPACE is not set, client should collect the
1049                  * space usage from OSSs by itself */
1050                 if (cmd == Q_GETQUOTA &&
1051                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1052                     !oqctl->qc_dqblk.dqb_curspace) {
1053                         struct obd_quotactl *oqctl_tmp;
1054
1055                         OBD_ALLOC_PTR(oqctl_tmp);
1056                         if (oqctl_tmp == NULL)
1057                                 GOTO(out, rc = -ENOMEM);
1058
1059                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1060                         oqctl_tmp->qc_id = oqctl->qc_id;
1061                         oqctl_tmp->qc_type = oqctl->qc_type;
1062
1063                         /* collect space usage from OSTs */
1064                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1065                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1066                         if (!rc || rc == -EREMOTEIO) {
1067                                 oqctl->qc_dqblk.dqb_curspace =
1068                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1069                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1070                         }
1071
1072                         /* collect space & inode usage from MDTs */
1073                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1074                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1075                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1076                         if (!rc || rc == -EREMOTEIO) {
1077                                 oqctl->qc_dqblk.dqb_curspace +=
1078                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1079                                 oqctl->qc_dqblk.dqb_curinodes =
1080                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1081                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1082                         } else {
1083                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1084                         }
1085
1086                         OBD_FREE_PTR(oqctl_tmp);
1087                 }
1088 out:
1089                 QCTL_COPY(qctl, oqctl);
1090                 OBD_FREE_PTR(oqctl);
1091         }
1092
1093         RETURN(rc);
1094 }
1095
1096 /* This function tries to get a single name component,
1097  * to send to the server. No actual path traversal involved,
1098  * so we limit to NAME_MAX */
1099 static char *ll_getname(const char __user *filename)
1100 {
1101         int ret = 0, len;
1102         char *tmp;
1103
1104         OBD_ALLOC(tmp, NAME_MAX + 1);
1105
1106         if (!tmp)
1107                 return ERR_PTR(-ENOMEM);
1108
1109         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1110         if (len < 0)
1111                 ret = -ENOENT;
1112         else if (len > NAME_MAX)
1113                 ret = -ENAMETOOLONG;
1114
1115         if (ret) {
1116                 OBD_FREE(tmp, NAME_MAX + 1);
1117                 tmp =  ERR_PTR(ret);
1118         }
1119         return tmp;
1120 }
1121
1122 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1123
1124 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1125 {
1126         struct dentry *dentry = file_dentry(file);
1127         struct inode *inode = file_inode(file);
1128         struct ll_sb_info *sbi = ll_i2sbi(inode);
1129         struct obd_ioctl_data *data;
1130         int rc = 0;
1131         ENTRY;
1132
1133         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1134                PFID(ll_inode2fid(inode)), inode, cmd);
1135
1136         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1137         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1138                 return -ENOTTY;
1139
1140         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1141         switch(cmd) {
1142         case FSFILT_IOC_GETFLAGS:
1143         case FSFILT_IOC_SETFLAGS:
1144                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1145         case FSFILT_IOC_GETVERSION_OLD:
1146         case FSFILT_IOC_GETVERSION:
1147                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1148         /* We need to special case any other ioctls we want to handle,
1149          * to send them to the MDS/OST as appropriate and to properly
1150          * network encode the arg field.
1151         case FSFILT_IOC_SETVERSION_OLD:
1152         case FSFILT_IOC_SETVERSION:
1153         */
1154         case LL_IOC_GET_MDTIDX: {
1155                 int mdtidx;
1156
1157                 mdtidx = ll_get_mdt_idx(inode);
1158                 if (mdtidx < 0)
1159                         RETURN(mdtidx);
1160
1161                 if (put_user((int)mdtidx, (int __user *)arg))
1162                         RETURN(-EFAULT);
1163
1164                 return 0;
1165         }
1166         case IOC_MDC_LOOKUP: {
1167                 int namelen, len = 0;
1168                 char *buf = NULL;
1169                 char *filename;
1170
1171                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1172                 if (rc != 0)
1173                         RETURN(rc);
1174                 data = (void *)buf;
1175
1176                 filename = data->ioc_inlbuf1;
1177                 namelen = strlen(filename);
1178                 if (namelen < 1) {
1179                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1180                         GOTO(out_free, rc = -EINVAL);
1181                 }
1182
1183                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1184                 if (rc < 0) {
1185                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1186                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1187                                filename, rc);
1188                         GOTO(out_free, rc);
1189                 }
1190 out_free:
1191                 obd_ioctl_freedata(buf, len);
1192                 return rc;
1193         }
1194         case LL_IOC_LMV_SETSTRIPE: {
1195                 struct lmv_user_md  *lum;
1196                 char            *buf = NULL;
1197                 char            *filename;
1198                 int              namelen = 0;
1199                 int              lumlen = 0;
1200                 umode_t          mode;
1201                 int              len;
1202                 int              rc;
1203
1204                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1205                 if (rc)
1206                         RETURN(rc);
1207
1208                 data = (void *)buf;
1209                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1210                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1211                         GOTO(lmv_out_free, rc = -EINVAL);
1212
1213                 filename = data->ioc_inlbuf1;
1214                 namelen = data->ioc_inllen1;
1215
1216                 if (namelen < 1) {
1217                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1218                         GOTO(lmv_out_free, rc = -EINVAL);
1219                 }
1220                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1221                 lumlen = data->ioc_inllen2;
1222
1223                 if (lum->lum_magic != LMV_USER_MAGIC ||
1224                     lumlen != sizeof(*lum)) {
1225                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1226                                filename, lum->lum_magic, lumlen, -EFAULT);
1227                         GOTO(lmv_out_free, rc = -EINVAL);
1228                 }
1229
1230 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1231                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1232 #else
1233                 mode = data->ioc_type;
1234 #endif
1235                 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1236 lmv_out_free:
1237                 obd_ioctl_freedata(buf, len);
1238                 RETURN(rc);
1239
1240         }
1241         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1242                 struct lmv_user_md        lum;
1243                 struct lmv_user_md __user *ulump =
1244                                         (struct lmv_user_md __user *)arg;
1245                 int                       rc;
1246
1247                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1248                         RETURN(-EFAULT);
1249
1250                 if (lum.lum_magic != LMV_USER_MAGIC)
1251                         RETURN(-EINVAL);
1252
1253                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1254
1255                 RETURN(rc);
1256         }
1257         case LL_IOC_LOV_SETSTRIPE: {
1258                 struct lov_user_md_v3 lumv3;
1259                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1260                 struct lov_user_md_v1 __user *lumv1p =
1261                         (struct lov_user_md_v1 __user *)arg;
1262                 struct lov_user_md_v3 __user *lumv3p =
1263                         (struct lov_user_md_v3 __user *)arg;
1264
1265                 int set_default = 0;
1266
1267                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1268                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1269                         sizeof(lumv3p->lmm_objects[0]));
1270                 /* first try with v1 which is smaller than v3 */
1271                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1272                         RETURN(-EFAULT);
1273
1274                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1275                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1276                                 RETURN(-EFAULT);
1277                 }
1278
1279                 if (inode->i_sb->s_root == file_dentry(file))
1280                         set_default = 1;
1281
1282                 /* in v1 and v3 cases lumv1 points to data */
1283                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1284
1285                 RETURN(rc);
1286         }
1287         case LL_IOC_LMV_GETSTRIPE: {
1288                 struct lmv_user_md __user *ulmv =
1289                                         (struct lmv_user_md __user *)arg;
1290                 struct lmv_user_md      lum;
1291                 struct ptlrpc_request   *request = NULL;
1292                 union lmv_mds_md        *lmm = NULL;
1293                 int                     lmmsize;
1294                 u64                     valid = 0;
1295                 struct lmv_user_md      *tmp = NULL;
1296                 int                     mdt_index;
1297                 int                     lum_size;
1298                 int                     stripe_count;
1299                 int                     max_stripe_count;
1300                 int                     i;
1301                 int                     rc;
1302
1303                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1304                         RETURN(-EFAULT);
1305
1306                 max_stripe_count = lum.lum_stripe_count;
1307                 /* lum_magic will indicate which stripe the ioctl will like
1308                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1309                  * is for default LMV stripe */
1310                 if (lum.lum_magic == LMV_MAGIC_V1)
1311                         valid |= OBD_MD_MEA;
1312                 else if (lum.lum_magic == LMV_USER_MAGIC)
1313                         valid |= OBD_MD_DEFAULT_MEA;
1314                 else
1315                         RETURN(-EINVAL);
1316
1317                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1318                                       valid);
1319                 if (rc != 0)
1320                         GOTO(finish_req, rc);
1321
1322                 /* Get default LMV EA */
1323                 if (lum.lum_magic == LMV_USER_MAGIC) {
1324                         if (lmmsize > sizeof(*ulmv))
1325                                 GOTO(finish_req, rc = -EINVAL);
1326
1327                         if (copy_to_user(ulmv, lmm, lmmsize))
1328                                 GOTO(finish_req, rc = -EFAULT);
1329
1330                         GOTO(finish_req, rc);
1331                 }
1332
1333                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1334                 if (max_stripe_count < stripe_count) {
1335                         lum.lum_stripe_count = stripe_count;
1336                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1337                                 GOTO(finish_req, rc = -EFAULT);
1338                         GOTO(finish_req, rc = -E2BIG);
1339                 }
1340
1341                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1342                 OBD_ALLOC(tmp, lum_size);
1343                 if (tmp == NULL)
1344                         GOTO(finish_req, rc = -ENOMEM);
1345
1346                 mdt_index = ll_get_mdt_idx(inode);
1347                 if (mdt_index < 0)
1348                         GOTO(out_tmp, rc = -ENOMEM);
1349
1350                 tmp->lum_magic = LMV_MAGIC_V1;
1351                 tmp->lum_stripe_count = 0;
1352                 tmp->lum_stripe_offset = mdt_index;
1353                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1354                 for (i = 0; i < stripe_count; i++) {
1355                         struct lu_fid   fid;
1356
1357                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1358                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1359                         if (mdt_index < 0)
1360                                 GOTO(out_tmp, rc = mdt_index);
1361
1362                         tmp->lum_objects[i].lum_mds = mdt_index;
1363                         tmp->lum_objects[i].lum_fid = fid;
1364                         tmp->lum_stripe_count++;
1365                 }
1366
1367                 if (copy_to_user(ulmv, tmp, lum_size))
1368                         GOTO(out_tmp, rc = -EFAULT);
1369 out_tmp:
1370                 OBD_FREE(tmp, lum_size);
1371 finish_req:
1372                 ptlrpc_req_finished(request);
1373                 return rc;
1374         }
1375
1376         case LL_IOC_REMOVE_ENTRY: {
1377                 char            *filename = NULL;
1378                 int              namelen = 0;
1379                 int              rc;
1380
1381                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1382                  * unsupported server, which might crash the server(LU-2730),
1383                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1384                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1385                  * server will support REINT_RMENTRY XXX*/
1386                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1387                         RETURN(-ENOTSUPP);
1388
1389                 filename = ll_getname((const char __user *)arg);
1390                 if (IS_ERR(filename))
1391                         RETURN(PTR_ERR(filename));
1392
1393                 namelen = strlen(filename);
1394                 if (namelen < 1)
1395                         GOTO(out_rmdir, rc = -EINVAL);
1396
1397                 rc = ll_rmdir_entry(inode, filename, namelen);
1398 out_rmdir:
1399                 if (filename)
1400                         ll_putname(filename);
1401                 RETURN(rc);
1402         }
1403         case LL_IOC_LOV_SWAP_LAYOUTS:
1404                 RETURN(-EPERM);
1405         case IOC_OBD_STATFS:
1406                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1407         case LL_IOC_LOV_GETSTRIPE:
1408         case LL_IOC_MDC_GETINFO:
1409         case IOC_MDC_GETFILEINFO:
1410         case IOC_MDC_GETFILESTRIPE: {
1411                 struct ptlrpc_request *request = NULL;
1412                 struct lov_user_md __user *lump;
1413                 struct lov_mds_md *lmm = NULL;
1414                 struct mdt_body *body;
1415                 char *filename = NULL;
1416                 int lmmsize;
1417
1418                 if (cmd == IOC_MDC_GETFILEINFO ||
1419                     cmd == IOC_MDC_GETFILESTRIPE) {
1420                         filename = ll_getname((const char __user *)arg);
1421                         if (IS_ERR(filename))
1422                                 RETURN(PTR_ERR(filename));
1423
1424                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1425                                                       &lmmsize, &request);
1426                 } else {
1427                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1428                                               &request, 0);
1429                 }
1430
1431                 if (request) {
1432                         body = req_capsule_server_get(&request->rq_pill,
1433                                                       &RMF_MDT_BODY);
1434                         LASSERT(body != NULL);
1435                 } else {
1436                         GOTO(out_req, rc);
1437                 }
1438
1439                 if (rc < 0) {
1440                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1441                                                cmd == LL_IOC_MDC_GETINFO))
1442                                 GOTO(skip_lmm, rc = 0);
1443                         else
1444                                 GOTO(out_req, rc);
1445                 }
1446
1447                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1448                     cmd == LL_IOC_LOV_GETSTRIPE) {
1449                         lump = (struct lov_user_md __user *)arg;
1450                 } else {
1451                         struct lov_user_mds_data __user *lmdp;
1452                         lmdp = (struct lov_user_mds_data __user *)arg;
1453                         lump = &lmdp->lmd_lmm;
1454                 }
1455                 if (copy_to_user(lump, lmm, lmmsize)) {
1456                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1457                                 GOTO(out_req, rc = -EFAULT);
1458                         rc = -EOVERFLOW;
1459                 }
1460         skip_lmm:
1461                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1462                         struct lov_user_mds_data __user *lmdp;
1463                         lstat_t st = { 0 };
1464
1465                         st.st_dev       = inode->i_sb->s_dev;
1466                         st.st_mode      = body->mbo_mode;
1467                         st.st_nlink     = body->mbo_nlink;
1468                         st.st_uid       = body->mbo_uid;
1469                         st.st_gid       = body->mbo_gid;
1470                         st.st_rdev      = body->mbo_rdev;
1471                         st.st_size      = body->mbo_size;
1472                         st.st_blksize   = PAGE_SIZE;
1473                         st.st_blocks    = body->mbo_blocks;
1474                         st.st_atime     = body->mbo_atime;
1475                         st.st_mtime     = body->mbo_mtime;
1476                         st.st_ctime     = body->mbo_ctime;
1477                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1478                                                 sbi->ll_flags &
1479                                                 LL_SBI_32BIT_API);
1480
1481                         lmdp = (struct lov_user_mds_data __user *)arg;
1482                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1483                                 GOTO(out_req, rc = -EFAULT);
1484                 }
1485
1486                 EXIT;
1487         out_req:
1488                 ptlrpc_req_finished(request);
1489                 if (filename)
1490                         ll_putname(filename);
1491                 return rc;
1492         }
1493         case OBD_IOC_QUOTACTL: {
1494                 struct if_quotactl *qctl;
1495
1496                 OBD_ALLOC_PTR(qctl);
1497                 if (!qctl)
1498                         RETURN(-ENOMEM);
1499
1500                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1501                         GOTO(out_quotactl, rc = -EFAULT);
1502
1503                 rc = quotactl_ioctl(sbi, qctl);
1504
1505                 if (rc == 0 &&
1506                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1507                         rc = -EFAULT;
1508
1509         out_quotactl:
1510                 OBD_FREE_PTR(qctl);
1511                 RETURN(rc);
1512         }
1513         case OBD_IOC_GETDTNAME:
1514         case OBD_IOC_GETMDNAME:
1515                 RETURN(ll_get_obd_name(inode, cmd, arg));
1516         case LL_IOC_FLUSHCTX:
1517                 RETURN(ll_flush_ctx(inode));
1518         case LL_IOC_GETOBDCOUNT: {
1519                 u32 count, vallen;
1520                 struct obd_export *exp;
1521
1522                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1523                         RETURN(-EFAULT);
1524
1525                 /* get ost count when count is zero, get mdt count otherwise */
1526                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1527                 vallen = sizeof(count);
1528                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1529                                   KEY_TGT_COUNT, &vallen, &count);
1530                 if (rc) {
1531                         CERROR("get target count failed: %d\n", rc);
1532                         RETURN(rc);
1533                 }
1534
1535                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1536                         RETURN(-EFAULT);
1537
1538                 RETURN(0);
1539         }
1540         case LL_IOC_PATH2FID:
1541                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1542                                      sizeof(struct lu_fid)))
1543                         RETURN(-EFAULT);
1544                 RETURN(0);
1545         case LL_IOC_GET_CONNECT_FLAGS: {
1546                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1547                                      (void __user *)arg));
1548         }
1549         case OBD_IOC_CHANGELOG_SEND:
1550         case OBD_IOC_CHANGELOG_CLEAR:
1551                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1552                         RETURN(-EPERM);
1553
1554                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1555                                     sizeof(struct ioc_changelog));
1556                 RETURN(rc);
1557         case OBD_IOC_FID2PATH:
1558                 RETURN(ll_fid2path(inode, (void __user *)arg));
1559         case LL_IOC_GETPARENT:
1560                 RETURN(ll_getparent(file, (void __user *)arg));
1561         case LL_IOC_FID2MDTIDX: {
1562                 struct obd_export *exp = ll_i2mdexp(inode);
1563                 struct lu_fid     fid;
1564                 __u32             index;
1565
1566                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1567                                    sizeof(fid)))
1568                         RETURN(-EFAULT);
1569
1570                 /* Call mdc_iocontrol */
1571                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1572                                    (__u32 __user *)&index);
1573                 if (rc != 0)
1574                         RETURN(rc);
1575
1576                 RETURN(index);
1577         }
1578         case LL_IOC_HSM_REQUEST: {
1579                 struct hsm_user_request *hur;
1580                 ssize_t                  totalsize;
1581
1582                 OBD_ALLOC_PTR(hur);
1583                 if (hur == NULL)
1584                         RETURN(-ENOMEM);
1585
1586                 /* We don't know the true size yet; copy the fixed-size part */
1587                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1588                         OBD_FREE_PTR(hur);
1589                         RETURN(-EFAULT);
1590                 }
1591
1592                 /* Compute the whole struct size */
1593                 totalsize = hur_len(hur);
1594                 OBD_FREE_PTR(hur);
1595                 if (totalsize < 0)
1596                         RETURN(-E2BIG);
1597
1598                 /* Final size will be more than double totalsize */
1599                 if (totalsize >= MDS_MAXREQSIZE / 3)
1600                         RETURN(-E2BIG);
1601
1602                 OBD_ALLOC_LARGE(hur, totalsize);
1603                 if (hur == NULL)
1604                         RETURN(-ENOMEM);
1605
1606                 /* Copy the whole struct */
1607                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1608                         GOTO(out_hur, rc = -EFAULT);
1609
1610                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1611                         const struct lu_fid *fid;
1612                         struct inode *f;
1613                         int i;
1614
1615                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1616                                 fid = &hur->hur_user_item[i].hui_fid;
1617                                 f = search_inode_for_lustre(inode->i_sb, fid);
1618                                 if (IS_ERR(f)) {
1619                                         rc = PTR_ERR(f);
1620                                         break;
1621                                 }
1622
1623                                 rc = ll_hsm_release(f);
1624                                 iput(f);
1625                                 if (rc != 0)
1626                                         break;
1627                         }
1628                 } else {
1629                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1630                                            hur, NULL);
1631                 }
1632
1633 out_hur:
1634                 OBD_FREE_LARGE(hur, totalsize);
1635
1636                 RETURN(rc);
1637         }
1638         case LL_IOC_HSM_PROGRESS: {
1639                 struct hsm_progress_kernel      hpk;
1640                 struct hsm_progress             hp;
1641
1642                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1643                         RETURN(-EFAULT);
1644
1645                 hpk.hpk_fid = hp.hp_fid;
1646                 hpk.hpk_cookie = hp.hp_cookie;
1647                 hpk.hpk_extent = hp.hp_extent;
1648                 hpk.hpk_flags = hp.hp_flags;
1649                 hpk.hpk_errval = hp.hp_errval;
1650                 hpk.hpk_data_version = 0;
1651
1652                 /* File may not exist in Lustre; all progress
1653                  * reported to Lustre root */
1654                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1655                                    NULL);
1656                 RETURN(rc);
1657         }
1658         case LL_IOC_HSM_CT_START:
1659                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1660                         RETURN(-EPERM);
1661
1662                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1663                                     sizeof(struct lustre_kernelcomm));
1664                 RETURN(rc);
1665
1666         case LL_IOC_HSM_COPY_START: {
1667                 struct hsm_copy *copy;
1668                 int              rc;
1669
1670                 OBD_ALLOC_PTR(copy);
1671                 if (copy == NULL)
1672                         RETURN(-ENOMEM);
1673                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1674                         OBD_FREE_PTR(copy);
1675                         RETURN(-EFAULT);
1676                 }
1677
1678                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1679                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1680                         rc = -EFAULT;
1681
1682                 OBD_FREE_PTR(copy);
1683                 RETURN(rc);
1684         }
1685         case LL_IOC_HSM_COPY_END: {
1686                 struct hsm_copy *copy;
1687                 int              rc;
1688
1689                 OBD_ALLOC_PTR(copy);
1690                 if (copy == NULL)
1691                         RETURN(-ENOMEM);
1692                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1693                         OBD_FREE_PTR(copy);
1694                         RETURN(-EFAULT);
1695                 }
1696
1697                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1698                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1699                         rc = -EFAULT;
1700
1701                 OBD_FREE_PTR(copy);
1702                 RETURN(rc);
1703         }
1704         case LL_IOC_MIGRATE: {
1705                 char            *buf = NULL;
1706                 const char      *filename;
1707                 int             namelen = 0;
1708                 int             len;
1709                 int             rc;
1710                 int             mdtidx;
1711
1712                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1713                 if (rc < 0)
1714                         RETURN(rc);
1715
1716                 data = (struct obd_ioctl_data *)buf;
1717                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1718                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1719                         GOTO(migrate_free, rc = -EINVAL);
1720
1721                 filename = data->ioc_inlbuf1;
1722                 namelen = data->ioc_inllen1;
1723                 /* \0 is packed at the end of filename */
1724                 if (namelen < 1 || namelen != strlen(filename) + 1)
1725                         GOTO(migrate_free, rc = -EINVAL);
1726
1727                 if (data->ioc_inllen2 != sizeof(mdtidx))
1728                         GOTO(migrate_free, rc = -EINVAL);
1729                 mdtidx = *(int *)data->ioc_inlbuf2;
1730
1731                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1732 migrate_free:
1733                 obd_ioctl_freedata(buf, len);
1734
1735                 RETURN(rc);
1736         }
1737         default:
1738                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1739                                      (void __user *)arg));
1740         }
1741 }
1742
1743 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1744 {
1745         struct inode *inode = file->f_mapping->host;
1746         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1747         struct ll_sb_info *sbi = ll_i2sbi(inode);
1748         int api32 = ll_need_32bit_api(sbi);
1749         loff_t ret = -EINVAL;
1750         ENTRY;
1751
1752         inode_lock(inode);
1753         switch (origin) {
1754                 case SEEK_SET:
1755                         break;
1756                 case SEEK_CUR:
1757                         offset += file->f_pos;
1758                         break;
1759                 case SEEK_END:
1760                         if (offset > 0)
1761                                 GOTO(out, ret);
1762                         if (api32)
1763                                 offset += LL_DIR_END_OFF_32BIT;
1764                         else
1765                                 offset += LL_DIR_END_OFF;
1766                         break;
1767                 default:
1768                         GOTO(out, ret);
1769         }
1770
1771         if (offset >= 0 &&
1772             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1773              (!api32 && offset <= LL_DIR_END_OFF))) {
1774                 if (offset != file->f_pos) {
1775                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1776                             (!api32 && offset == LL_DIR_END_OFF))
1777                                 fd->lfd_pos = MDS_DIR_END_OFF;
1778                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1779                                 fd->lfd_pos = offset << 32;
1780                         else
1781                                 fd->lfd_pos = offset;
1782                         file->f_pos = offset;
1783                         file->f_version = 0;
1784                 }
1785                 ret = offset;
1786         }
1787         GOTO(out, ret);
1788
1789 out:
1790         inode_unlock(inode);
1791         return ret;
1792 }
1793
1794 static int ll_dir_open(struct inode *inode, struct file *file)
1795 {
1796         ENTRY;
1797         RETURN(ll_file_open(inode, file));
1798 }
1799
1800 static int ll_dir_release(struct inode *inode, struct file *file)
1801 {
1802         ENTRY;
1803         RETURN(ll_file_release(inode, file));
1804 }
1805
1806 const struct file_operations ll_dir_operations = {
1807         .llseek         = ll_dir_seek,
1808         .open           = ll_dir_open,
1809         .release        = ll_dir_release,
1810         .read           = generic_read_dir,
1811 #ifdef HAVE_DIR_CONTEXT
1812         .iterate        = ll_iterate,
1813 #else
1814         .readdir        = ll_readdir,
1815 #endif
1816         .unlocked_ioctl = ll_dir_ioctl,
1817         .fsync          = ll_fsync,
1818 };