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 <uapi/linux/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) /* Skip until we find target hash */
 251                                 continue;
 252
 253                         namelen = le16_to_cpu(ent->lde_namelen);
 254                         if (namelen == 0) /* Skip dummy record */
 255                                 continue;
 256
 257                         if (is_api32 && is_hash64)
 258                                 lhash = hash >> 32;
 259                         else
 260                                 lhash = hash;
 261                         fid_le_to_cpu(&fid, &ent->lde_fid);
 262                         ino = cl_fid_build_ino(&fid, is_api32);
 263                         type = ll_dirent_type_get(ent);
 264                         /* For ll_nfs_get_name_filldir(), it will try to access
 265                          * 'ent' through 'lde_name', so the parameter 'name'
 266                          * for 'filldir()' must be part of the 'ent'. */
 267 #ifdef HAVE_DIR_CONTEXT
 268                         ctx->pos = lhash;
 269                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 270                                          type);
 271 #else
 272                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 273                                        ino, type);
 274 #endif
 275                 }
 276
 277                 if (done) {
 278                         pos = hash;
 279                         ll_release_page(inode, page, false);
 280                         break;
 281                 }
 282
 283                 next = le64_to_cpu(dp->ldp_hash_end);
 284                 pos = next;
 285                 if (pos == MDS_DIR_END_OFF) {
 286                         /*
 287                          * End of directory reached.
 288                          */
 289                         done = 1;
 290                         ll_release_page(inode, page, false);
 291                 } else {
 292                         /*
 293                          * Normal case: continue to the next
 294                          * page.
 295                          */
 296                         ll_release_page(inode, page,
 297                                         le32_to_cpu(dp->ldp_flags) &
 298                                         LDF_COLLIDE);
 299                         next = pos;
 300                         page = ll_get_dir_page(inode, op_data, pos,
 301                                                &chain);
 302                 }
 303         }
 304 #ifdef HAVE_DIR_CONTEXT
 305         ctx->pos = pos;
 306 #else
 307         *ppos = pos;
 308 #endif
 309         ll_dir_chain_fini(&chain);
 310         RETURN(rc);
 311 }
 312
 313 #ifdef HAVE_DIR_CONTEXT
 314 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 315 #else
 316 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 317 #endif
 318 {
 319         struct inode            *inode  = file_inode(filp);
 320         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 321         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 322         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 323         int                     api32   = ll_need_32bit_api(sbi);
 324         struct md_op_data       *op_data;
 325         __u64                   pos;
 326         int                     rc;
 327         ENTRY;
 328
 329         if (lfd != NULL)
 330                 pos = lfd->lfd_pos;
 331         else
 332                 pos = 0;
 333
 334         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 335                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 336                inode, (unsigned long)pos, i_size_read(inode), api32);
 337
 338         if (pos == MDS_DIR_END_OFF)
 339                 /*
 340                  * end-of-file.
 341                  */
 342                 GOTO(out, rc = 0);
 343
 344         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 345                                      LUSTRE_OPC_ANY, inode);
 346         if (IS_ERR(op_data))
 347                 GOTO(out, rc = PTR_ERR(op_data));
 348
 349         if (unlikely(op_data->op_mea1 != NULL)) {
 350                 /* This is only needed for striped dir to fill ..,
 351                  * see lmv_read_entry */
 352                 if (file_dentry(filp)->d_parent != NULL &&
 353                     file_dentry(filp)->d_parent->d_inode != NULL) {
 354                         __u64 ibits = MDS_INODELOCK_UPDATE;
 355                         struct inode *parent =
 356                                 file_dentry(filp)->d_parent->d_inode;
 357
 358                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 359                                 op_data->op_fid3 = *ll_inode2fid(parent);
 360                 }
 361
 362                 /* If it can not find in cache, do lookup .. on the master
 363                  * object */
 364                 if (fid_is_zero(&op_data->op_fid3)) {
 365                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 366                         if (rc != 0) {
 367                                 ll_finish_md_op_data(op_data);
 368                                 RETURN(rc);
 369                         }
 370                 }
 371         }
 372 #ifdef HAVE_DIR_CONTEXT
 373         ctx->pos = pos;
 374         rc = ll_dir_read(inode, &pos, op_data, ctx);
 375         pos = ctx->pos;
 376 #else
 377         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 378 #endif
 379         if (lfd != NULL)
 380                 lfd->lfd_pos = pos;
 381
 382         if (pos == MDS_DIR_END_OFF) {
 383                 if (api32)
 384                         pos = LL_DIR_END_OFF_32BIT;
 385                 else
 386                         pos = LL_DIR_END_OFF;
 387         } else {
 388                 if (api32 && hash64)
 389                         pos = pos >> 32;
 390         }
 391 #ifdef HAVE_DIR_CONTEXT
 392         ctx->pos = pos;
 393 #else
 394         filp->f_pos = pos;
 395 #endif
 396         ll_finish_md_op_data(op_data);
 397         filp->f_version = inode->i_version;
 398
 399 out:
 400         if (!rc)
 401                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 402
 403         RETURN(rc);
 404 }
 405
 406 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 407 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 408 {
 409         struct mgs_send_param *msp;
 410         int rc = 0;
 411
 412         OBD_ALLOC_PTR(msp);
 413         if (!msp)
 414                 return -ENOMEM;
 415
 416         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 417         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 418                                 sizeof(struct mgs_send_param), msp, NULL);
 419         if (rc)
 420                 CERROR("Failed to set parameter: %d\n", rc);
 421         OBD_FREE_PTR(msp);
 422
 423         return rc;
 424 }
 425 #endif
 426
 427 /**
 428  * Create striped directory with specified stripe(@lump)
 429  *
 430  * \param[in] dparent   the parent of the directory.
 431  * \param[in] lump      the specified stripes.
 432  * \param[in] dirname   the name of the directory.
 433  * \param[in] mode      the specified mode of the directory.
 434  *
 435  * \retval              =0 if striped directory is being created successfully.
 436  *                      <0 if the creation is failed.
 437  */
 438 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 439                                const char *dirname, umode_t mode)
 440 {
 441         struct inode *parent = dparent->d_inode;
 442         struct ptlrpc_request *request = NULL;
 443         struct md_op_data *op_data;
 444         struct ll_sb_info *sbi = ll_i2sbi(parent);
 445         struct inode *inode = NULL;
 446         struct dentry dentry = {
 447                 .d_parent = dparent,
 448                 .d_name = {
 449                         .name = dirname,
 450                         .len = strlen(dirname),
 451                         .hash = full_name_hash(dirname, strlen(dirname)),
 452                 },
 453         };
 454         int err;
 455         ENTRY;
 456
 457         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 458                 RETURN(-EINVAL);
 459
 460         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 461                "stripe_offset %d, stripe_count: %u\n",
 462                PFID(ll_inode2fid(parent)), parent, dirname,
 463                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 464
 465         if (lump->lum_stripe_count > 1 &&
 466             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 467                 RETURN(-EINVAL);
 468
 469         if (IS_DEADDIR(parent) &&
 470             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 471                 RETURN(-ENOENT);
 472
 473         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 474                 lustre_swab_lmv_user_md(lump);
 475
 476         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 477                 mode &= ~current_umask();
 478         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 479         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 480                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 481                                      lump);
 482         if (IS_ERR(op_data))
 483                 RETURN(PTR_ERR(op_data));
 484
 485         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 486                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 487                  * to determine the security context for the file. So our fake
 488                  * dentry should be real enough for this purpose. */
 489                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 490                                               &op_data->op_file_secctx_name,
 491                                               &op_data->op_file_secctx,
 492                                               &op_data->op_file_secctx_size);
 493                 if (err < 0)
 494                         GOTO(out_op_data, err);
 495         }
 496
 497         op_data->op_cli_flags |= CLI_SET_MEA;
 498         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 499                         from_kuid(&init_user_ns, current_fsuid()),
 500                         from_kgid(&init_user_ns, current_fsgid()),
 501                         cfs_curproc_cap_pack(), 0, &request);
 502         if (err)
 503                 GOTO(out_request, err);
 504
 505         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 506
 507         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 508         if (err)
 509                 GOTO(out_inode, err);
 510
 511         dentry.d_inode = inode;
 512
 513         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
 514                 err = ll_inode_init_security(&dentry, inode, parent);
 515                 if (err)
 516                         GOTO(out_inode, err);
 517         }
 518
 519 out_inode:
 520         if (inode != NULL)
 521                 iput(inode);
 522 out_request:
 523         ptlrpc_req_finished(request);
 524 out_op_data:
 525         ll_finish_md_op_data(op_data);
 526
 527         return err;
 528 }
 529
 530 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 531                      int set_default)
 532 {
 533         struct ll_sb_info *sbi = ll_i2sbi(inode);
 534         struct md_op_data *op_data;
 535         struct ptlrpc_request *req = NULL;
 536         int rc = 0;
 537 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 538         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 539         struct obd_device *mgc = lsi->lsi_mgc;
 540 #endif
 541         int lum_size;
 542         ENTRY;
 543
 544         if (lump != NULL) {
 545                 /*
 546                  * This is coming from userspace, so should be in
 547                  * local endian.  But the MDS would like it in little
 548                  * endian, so we swab it before we send it.
 549                  */
 550                 switch (lump->lmm_magic) {
 551                 case LOV_USER_MAGIC_V1: {
 552                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 553                                 lustre_swab_lov_user_md_v1(lump);
 554                         lum_size = sizeof(struct lov_user_md_v1);
 555                         break;
 556                 }
 557                 case LOV_USER_MAGIC_V3: {
 558                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 559                                 lustre_swab_lov_user_md_v3(
 560                                         (struct lov_user_md_v3 *)lump);
 561                         lum_size = sizeof(struct lov_user_md_v3);
 562                         break;
 563                 }
 564                 case LOV_USER_MAGIC_COMP_V1: {
 565                         if (lump->lmm_magic !=
 566                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 567                                 lustre_swab_lov_comp_md_v1(
 568                                         (struct lov_comp_md_v1 *)lump);
 569                         lum_size = le32_to_cpu(
 570                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 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             (lump == NULL ||
 615              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 616              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 617                 char *param = NULL;
 618                 char *buf;
 619
 620                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 621                 if (param == NULL)
 622                         GOTO(end, rc = -ENOMEM);
 623
 624                 buf = param;
 625                 /* Get fsname and assume devname to be -MDT0000. */
 626                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 627                 strcat(buf, "-MDT0000.lov");
 628                 buf += strlen(buf);
 629
 630                 /* Set root stripesize */
 631                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 632                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 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 stripecount */
 638                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 639                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 640                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 641                 if (rc)
 642                         GOTO(end, rc);
 643
 644                 /* Set root stripeoffset */
 645                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 646                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 647                                 (typeof(lump->lmm_stripe_offset))(-1));
 648                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 649
 650 end:
 651                 if (param != NULL)
 652                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 653         }
 654 #endif
 655         RETURN(rc);
 656 }
 657
 658 /**
 659  * This function will be used to get default LOV/LMV/Default LMV
 660  * @valid will be used to indicate which stripe it will retrieve
 661  *      OBD_MD_MEA              LMV stripe EA
 662  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 663  *      otherwise               Default LOV EA.
 664  * Each time, it can only retrieve 1 stripe EA
 665  **/
 666 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 667                      struct ptlrpc_request **request, u64 valid)
 668 {
 669         struct ll_sb_info *sbi = ll_i2sbi(inode);
 670         struct mdt_body   *body;
 671         struct lov_mds_md *lmm = NULL;
 672         struct ptlrpc_request *req = NULL;
 673         int rc, lmm_size;
 674         struct md_op_data *op_data;
 675         ENTRY;
 676
 677         rc = ll_get_default_mdsize(sbi, &lmm_size);
 678         if (rc)
 679                 RETURN(rc);
 680
 681         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 682                                      0, lmm_size, LUSTRE_OPC_ANY,
 683                                      NULL);
 684         if (IS_ERR(op_data))
 685                 RETURN(PTR_ERR(op_data));
 686
 687         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 688         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 689         ll_finish_md_op_data(op_data);
 690         if (rc < 0) {
 691                 CDEBUG(D_INFO, "md_getattr failed on inode "
 692                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 693                 GOTO(out, rc);
 694         }
 695
 696         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 697         LASSERT(body != NULL);
 698
 699         lmm_size = body->mbo_eadatasize;
 700
 701         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 702             lmm_size == 0) {
 703                 GOTO(out, rc = -ENODATA);
 704         }
 705
 706         lmm = req_capsule_server_sized_get(&req->rq_pill,
 707                                            &RMF_MDT_MD, lmm_size);
 708         LASSERT(lmm != NULL);
 709
 710         /*
 711          * This is coming from the MDS, so is probably in
 712          * little endian.  We convert it to host endian before
 713          * passing it to userspace.
 714          */
 715         /* We don't swab objects for directories */
 716         switch (le32_to_cpu(lmm->lmm_magic)) {
 717         case LOV_MAGIC_V1:
 718                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 719                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 720                 break;
 721         case LOV_MAGIC_V3:
 722                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 723                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 724                 break;
 725         case LOV_MAGIC_COMP_V1:
 726                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 727                         lustre_swab_lov_comp_md_v1(
 728                                         (struct lov_comp_md_v1 *)lmm);
 729                 break;
 730         case LMV_MAGIC_V1:
 731                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 732                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 733                 break;
 734         case LMV_USER_MAGIC:
 735                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 736                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 737                 break;
 738         default:
 739                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 740                 rc = -EPROTO;
 741         }
 742 out:
 743         *plmm = lmm;
 744         *plmm_size = lmm_size;
 745         *request = req;
 746         return rc;
 747 }
 748
 749 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 750 {
 751         struct md_op_data       *op_data;
 752         int                     rc;
 753         int                     mdt_index;
 754         ENTRY;
 755
 756         OBD_ALLOC_PTR(op_data);
 757         if (op_data == NULL)
 758                 RETURN(-ENOMEM);
 759
 760         op_data->op_flags |= MF_GET_MDT_IDX;
 761         op_data->op_fid1 = *fid;
 762         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 763         mdt_index = op_data->op_mds;
 764         OBD_FREE_PTR(op_data);
 765         if (rc < 0)
 766                 RETURN(rc);
 767
 768         RETURN(mdt_index);
 769 }
 770
 771 /*
 772  *  Get MDT index for the inode.
 773  */
 774 int ll_get_mdt_idx(struct inode *inode)
 775 {
 776         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 777 }
 778
 779 /**
 780  * Generic handler to do any pre-copy work.
 781  *
 782  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 783  * first information for it that real work has started.
 784  *
 785  * Moreover, for a ARCHIVE request, it will sample the file data version and
 786  * store it in \a copy.
 787  *
 788  * \return 0 on success.
 789  */
 790 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 791 {
 792         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 793         struct hsm_progress_kernel       hpk;
 794         int                              rc = 0;
 795         int                              rc2;
 796         ENTRY;
 797
 798         /* Forge a hsm_progress based on data from copy. */
 799         hpk.hpk_fid = copy->hc_hai.hai_fid;
 800         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 801         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 802         hpk.hpk_extent.length = 0;
 803         hpk.hpk_flags = 0;
 804         hpk.hpk_errval = 0;
 805         hpk.hpk_data_version = 0;
 806
 807
 808         /* For archive request, we need to read the current file version. */
 809         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 810                 struct inode    *inode;
 811                 __u64            data_version = 0;
 812
 813                 /* Get inode for this fid */
 814                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 815                 if (IS_ERR(inode)) {
 816                         hpk.hpk_flags |= HP_FLAG_RETRY;
 817                         /* hpk_errval is >= 0 */
 818                         hpk.hpk_errval = -PTR_ERR(inode);
 819                         GOTO(progress, rc = PTR_ERR(inode));
 820                 }
 821
 822                 /* Read current file data version */
 823                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 824                 iput(inode);
 825                 if (rc != 0) {
 826                         CDEBUG(D_HSM, "Could not read file data version of "
 827                                       DFID" (rc = %d). Archive request ("
 828                                       "%#llx) could not be done.\n",
 829                                       PFID(&copy->hc_hai.hai_fid), rc,
 830                                       copy->hc_hai.hai_cookie);
 831                         hpk.hpk_flags |= HP_FLAG_RETRY;
 832                         /* hpk_errval must be >= 0 */
 833                         hpk.hpk_errval = -rc;
 834                         GOTO(progress, rc);
 835                 }
 836
 837                 /* Store in the hsm_copy for later copytool use.
 838                  * Always modified even if no lsm. */
 839                 copy->hc_data_version = data_version;
 840         }
 841
 842 progress:
 843         /* On error, the request should be considered as completed */
 844         if (hpk.hpk_errval > 0)
 845                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 846
 847         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 848                             &hpk, NULL);
 849
 850         /* Return first error */
 851         RETURN(rc != 0 ? rc : rc2);
 852 }
 853
 854 /**
 855  * Generic handler to do any post-copy work.
 856  *
 857  * It will send the last hsm_progress update to coordinator to inform it
 858  * that copy is finished and whether it was successful or not.
 859  *
 860  * Moreover,
 861  * - for ARCHIVE request, it will sample the file data version and compare it
 862  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 863  *   will be considered as failed.
 864  * - for RESTORE request, it will sample the file data version and send it to
 865  *   coordinator which is useful if the file was imported as 'released'.
 866  *
 867  * \return 0 on success.
 868  */
 869 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 870 {
 871         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 872         struct hsm_progress_kernel       hpk;
 873         int                              rc = 0;
 874         int                              rc2;
 875         ENTRY;
 876
 877         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 878         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 879          * initialized if copy_end was called with copy == NULL.
 880          */
 881
 882         /* Forge a hsm_progress based on data from copy. */
 883         hpk.hpk_fid = copy->hc_hai.hai_fid;
 884         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 885         hpk.hpk_extent = copy->hc_hai.hai_extent;
 886         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 887         hpk.hpk_errval = copy->hc_errval;
 888         hpk.hpk_data_version = 0;
 889
 890         /* For archive request, we need to check the file data was not changed.
 891          *
 892          * For restore request, we need to send the file data version, this is
 893          * useful when the file was created using hsm_import.
 894          */
 895         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 896              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 897             (copy->hc_errval == 0)) {
 898                 struct inode    *inode;
 899                 __u64            data_version = 0;
 900
 901                 /* Get lsm for this fid */
 902                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 903                 if (IS_ERR(inode)) {
 904                         hpk.hpk_flags |= HP_FLAG_RETRY;
 905                         /* hpk_errval must be >= 0 */
 906                         hpk.hpk_errval = -PTR_ERR(inode);
 907                         GOTO(progress, rc = PTR_ERR(inode));
 908                 }
 909
 910                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 911                 iput(inode);
 912                 if (rc) {
 913                         CDEBUG(D_HSM, "Could not read file data version. "
 914                                       "Request could not be confirmed.\n");
 915                         if (hpk.hpk_errval == 0)
 916                                 hpk.hpk_errval = -rc;
 917                         GOTO(progress, rc);
 918                 }
 919
 920                 /* Store in the hsm_copy for later copytool use.
 921                  * Always modified even if no lsm. */
 922                 hpk.hpk_data_version = data_version;
 923
 924                 /* File could have been stripped during archiving, so we need
 925                  * to check anyway. */
 926                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 927                     (copy->hc_data_version != data_version)) {
 928                         CDEBUG(D_HSM, "File data version mismatched. "
 929                               "File content was changed during archiving. "
 930                                DFID", start:%#llx current:%#llx\n",
 931                                PFID(&copy->hc_hai.hai_fid),
 932                                copy->hc_data_version, data_version);
 933                         /* File was changed, send error to cdt. Do not ask for
 934                          * retry because if a file is modified frequently,
 935                          * the cdt will loop on retried archive requests.
 936                          * The policy engine will ask for a new archive later
 937                          * when the file will not be modified for some tunable
 938                          * time */
 939                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 940                         rc = -EBUSY;
 941                         /* hpk_errval must be >= 0 */
 942                         hpk.hpk_errval = -rc;
 943                         GOTO(progress, rc);
 944                 }
 945
 946         }
 947
 948 progress:
 949         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 950                             &hpk, NULL);
 951
 952         /* Return first error */
 953         RETURN(rc != 0 ? rc : rc2);
 954 }
 955
 956
 957 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 958                           const void __user *data, size_t size)
 959 {
 960         void *copy;
 961         int rc;
 962
 963         OBD_ALLOC(copy, size);
 964         if (copy == NULL)
 965                 return -ENOMEM;
 966
 967         if (copy_from_user(copy, data, size)) {
 968                 rc = -EFAULT;
 969                 goto out;
 970         }
 971
 972         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 973 out:
 974         OBD_FREE(copy, size);
 975
 976         return rc;
 977 }
 978
 979 static int check_owner(int type, int id)
 980 {
 981         switch (type) {
 982         case USRQUOTA:
 983                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
 984                         return -EPERM;
 985                 break;
 986         case GRPQUOTA:
 987                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
 988                         return -EPERM;
 989                 break;
 990         case PRJQUOTA:
 991                 break;
 992         }
 993         return 0;
 994 }
 995
 996 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 997 {
 998         int cmd = qctl->qc_cmd;
 999         int type = qctl->qc_type;
1000         int id = qctl->qc_id;
1001         int valid = qctl->qc_valid;
1002         int rc = 0;
1003         ENTRY;
1004
1005         switch (cmd) {
1006         case Q_SETQUOTA:
1007         case Q_SETINFO:
1008                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1009                         RETURN(-EPERM);
1010                 break;
1011         case Q_GETQUOTA:
1012                 if (check_owner(type, id) &&
1013                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1014                         RETURN(-EPERM);
1015                 break;
1016         case Q_GETINFO:
1017                 break;
1018         default:
1019                 CERROR("unsupported quotactl op: %#x\n", cmd);
1020                 RETURN(-ENOTTY);
1021         }
1022
1023         if (valid != QC_GENERAL) {
1024                 if (cmd == Q_GETINFO)
1025                         qctl->qc_cmd = Q_GETOINFO;
1026                 else if (cmd == Q_GETQUOTA)
1027                         qctl->qc_cmd = Q_GETOQUOTA;
1028                 else
1029                         RETURN(-EINVAL);
1030
1031                 switch (valid) {
1032                 case QC_MDTIDX:
1033                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1034                                            sizeof(*qctl), qctl, NULL);
1035                         break;
1036                 case QC_OSTIDX:
1037                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1038                                            sizeof(*qctl), qctl, NULL);
1039                         break;
1040                 case QC_UUID:
1041                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1042                                            sizeof(*qctl), qctl, NULL);
1043                         if (rc == -EAGAIN)
1044                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1045                                                    sbi->ll_dt_exp,
1046                                                    sizeof(*qctl), qctl, NULL);
1047                         break;
1048                 default:
1049                         rc = -EINVAL;
1050                         break;
1051                 }
1052
1053                 if (rc)
1054                         RETURN(rc);
1055
1056                 qctl->qc_cmd = cmd;
1057         } else {
1058                 struct obd_quotactl *oqctl;
1059
1060                 OBD_ALLOC_PTR(oqctl);
1061                 if (oqctl == NULL)
1062                         RETURN(-ENOMEM);
1063
1064                 QCTL_COPY(oqctl, qctl);
1065                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1066                 if (rc) {
1067                         OBD_FREE_PTR(oqctl);
1068                         RETURN(rc);
1069                 }
1070                 /* If QIF_SPACE is not set, client should collect the
1071                  * space usage from OSSs by itself */
1072                 if (cmd == Q_GETQUOTA &&
1073                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1074                     !oqctl->qc_dqblk.dqb_curspace) {
1075                         struct obd_quotactl *oqctl_tmp;
1076
1077                         OBD_ALLOC_PTR(oqctl_tmp);
1078                         if (oqctl_tmp == NULL)
1079                                 GOTO(out, rc = -ENOMEM);
1080
1081                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1082                         oqctl_tmp->qc_id = oqctl->qc_id;
1083                         oqctl_tmp->qc_type = oqctl->qc_type;
1084
1085                         /* collect space usage from OSTs */
1086                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1087                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1088                         if (!rc || rc == -EREMOTEIO) {
1089                                 oqctl->qc_dqblk.dqb_curspace =
1090                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1091                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1092                         }
1093
1094                         /* collect space & inode usage from MDTs */
1095                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1096                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1097                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1098                         if (!rc || rc == -EREMOTEIO) {
1099                                 oqctl->qc_dqblk.dqb_curspace +=
1100                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1101                                 oqctl->qc_dqblk.dqb_curinodes =
1102                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1103                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1104                         } else {
1105                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1106                         }
1107
1108                         OBD_FREE_PTR(oqctl_tmp);
1109                 }
1110 out:
1111                 QCTL_COPY(qctl, oqctl);
1112                 OBD_FREE_PTR(oqctl);
1113         }
1114
1115         RETURN(rc);
1116 }
1117
1118 /* This function tries to get a single name component,
1119  * to send to the server. No actual path traversal involved,
1120  * so we limit to NAME_MAX */
1121 static char *ll_getname(const char __user *filename)
1122 {
1123         int ret = 0, len;
1124         char *tmp;
1125
1126         OBD_ALLOC(tmp, NAME_MAX + 1);
1127
1128         if (!tmp)
1129                 return ERR_PTR(-ENOMEM);
1130
1131         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1132         if (len < 0)
1133                 ret = -ENOENT;
1134         else if (len > NAME_MAX)
1135                 ret = -ENAMETOOLONG;
1136
1137         if (ret) {
1138                 OBD_FREE(tmp, NAME_MAX + 1);
1139                 tmp =  ERR_PTR(ret);
1140         }
1141         return tmp;
1142 }
1143
1144 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1145
1146 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1147 {
1148         struct dentry *dentry = file_dentry(file);
1149         struct inode *inode = file_inode(file);
1150         struct ll_sb_info *sbi = ll_i2sbi(inode);
1151         struct obd_ioctl_data *data;
1152         int rc = 0;
1153         ENTRY;
1154
1155         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1156                PFID(ll_inode2fid(inode)), inode, cmd);
1157
1158         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1159         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1160                 return -ENOTTY;
1161
1162         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1163         switch(cmd) {
1164         case FSFILT_IOC_GETFLAGS:
1165         case FSFILT_IOC_SETFLAGS:
1166                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1167         case FSFILT_IOC_GETVERSION_OLD:
1168         case FSFILT_IOC_GETVERSION:
1169                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1170         /* We need to special case any other ioctls we want to handle,
1171          * to send them to the MDS/OST as appropriate and to properly
1172          * network encode the arg field.
1173         case FSFILT_IOC_SETVERSION_OLD:
1174         case FSFILT_IOC_SETVERSION:
1175         */
1176         case LL_IOC_GET_MDTIDX: {
1177                 int mdtidx;
1178
1179                 mdtidx = ll_get_mdt_idx(inode);
1180                 if (mdtidx < 0)
1181                         RETURN(mdtidx);
1182
1183                 if (put_user((int)mdtidx, (int __user *)arg))
1184                         RETURN(-EFAULT);
1185
1186                 return 0;
1187         }
1188         case IOC_MDC_LOOKUP: {
1189                 int namelen, len = 0;
1190                 char *buf = NULL;
1191                 char *filename;
1192
1193                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1194                 if (rc != 0)
1195                         RETURN(rc);
1196                 data = (void *)buf;
1197
1198                 filename = data->ioc_inlbuf1;
1199                 namelen = strlen(filename);
1200                 if (namelen < 1) {
1201                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1202                         GOTO(out_free, rc = -EINVAL);
1203                 }
1204
1205                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1206                 if (rc < 0) {
1207                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1208                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1209                                filename, rc);
1210                         GOTO(out_free, rc);
1211                 }
1212 out_free:
1213                 OBD_FREE_LARGE(buf, len);
1214                 return rc;
1215         }
1216         case LL_IOC_LMV_SETSTRIPE: {
1217                 struct lmv_user_md  *lum;
1218                 char            *buf = NULL;
1219                 char            *filename;
1220                 int              namelen = 0;
1221                 int              lumlen = 0;
1222                 umode_t          mode;
1223                 int              len;
1224                 int              rc;
1225
1226                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1227                 if (rc)
1228                         RETURN(rc);
1229
1230                 data = (void *)buf;
1231                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1232                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1233                         GOTO(lmv_out_free, rc = -EINVAL);
1234
1235                 filename = data->ioc_inlbuf1;
1236                 namelen = data->ioc_inllen1;
1237
1238                 if (namelen < 1) {
1239                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1240                         GOTO(lmv_out_free, rc = -EINVAL);
1241                 }
1242                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1243                 lumlen = data->ioc_inllen2;
1244
1245                 if (lum->lum_magic != LMV_USER_MAGIC ||
1246                     lumlen != sizeof(*lum)) {
1247                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1248                                filename, lum->lum_magic, lumlen, -EFAULT);
1249                         GOTO(lmv_out_free, rc = -EINVAL);
1250                 }
1251
1252 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1253                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1254 #else
1255                 mode = data->ioc_type;
1256 #endif
1257                 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1258 lmv_out_free:
1259                 OBD_FREE_LARGE(buf, len);
1260                 RETURN(rc);
1261
1262         }
1263         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1264                 struct lmv_user_md        lum;
1265                 struct lmv_user_md __user *ulump =
1266                                         (struct lmv_user_md __user *)arg;
1267                 int                       rc;
1268
1269                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1270                         RETURN(-EFAULT);
1271
1272                 if (lum.lum_magic != LMV_USER_MAGIC)
1273                         RETURN(-EINVAL);
1274
1275                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1276
1277                 RETURN(rc);
1278         }
1279         case LL_IOC_LOV_SETSTRIPE: {
1280                 struct lov_user_md_v3 lumv3;
1281                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1282                 struct lov_user_md_v1 __user *lumv1p =
1283                         (struct lov_user_md_v1 __user *)arg;
1284                 struct lov_user_md_v3 __user *lumv3p =
1285                         (struct lov_user_md_v3 __user *)arg;
1286
1287                 int set_default = 0;
1288
1289                 CLASSERT(sizeof(struct lov_user_md_v3) >
1290                          sizeof(struct lov_comp_md_v1));
1291                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1292                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1293                                 sizeof(lumv3p->lmm_objects[0]));
1294                 /* first try with v1 which is smaller than v3 */
1295                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1296                         RETURN(-EFAULT);
1297
1298                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3)
1299                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1300                                 RETURN(-EFAULT);
1301
1302                 if (inode->i_sb->s_root == file_dentry(file))
1303                         set_default = 1;
1304
1305                 /* in v1 and v3 cases lumv1 points to data */
1306                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1307
1308                 RETURN(rc);
1309         }
1310         case LL_IOC_LMV_GETSTRIPE: {
1311                 struct lmv_user_md __user *ulmv =
1312                                         (struct lmv_user_md __user *)arg;
1313                 struct lmv_user_md      lum;
1314                 struct ptlrpc_request   *request = NULL;
1315                 union lmv_mds_md        *lmm = NULL;
1316                 int                     lmmsize;
1317                 u64                     valid = 0;
1318                 struct lmv_user_md      *tmp = NULL;
1319                 int                     mdt_index;
1320                 int                     lum_size;
1321                 int                     stripe_count;
1322                 int                     max_stripe_count;
1323                 int                     i;
1324                 int                     rc;
1325
1326                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1327                         RETURN(-EFAULT);
1328
1329                 max_stripe_count = lum.lum_stripe_count;
1330                 /* lum_magic will indicate which stripe the ioctl will like
1331                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1332                  * is for default LMV stripe */
1333                 if (lum.lum_magic == LMV_MAGIC_V1)
1334                         valid |= OBD_MD_MEA;
1335                 else if (lum.lum_magic == LMV_USER_MAGIC)
1336                         valid |= OBD_MD_DEFAULT_MEA;
1337                 else
1338                         RETURN(-EINVAL);
1339
1340                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1341                                       valid);
1342                 if (rc != 0)
1343                         GOTO(finish_req, rc);
1344
1345                 /* Get default LMV EA */
1346                 if (lum.lum_magic == LMV_USER_MAGIC) {
1347                         if (lmmsize > sizeof(*ulmv))
1348                                 GOTO(finish_req, rc = -EINVAL);
1349
1350                         if (copy_to_user(ulmv, lmm, lmmsize))
1351                                 GOTO(finish_req, rc = -EFAULT);
1352
1353                         GOTO(finish_req, rc);
1354                 }
1355
1356                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1357                 if (max_stripe_count < stripe_count) {
1358                         lum.lum_stripe_count = stripe_count;
1359                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1360                                 GOTO(finish_req, rc = -EFAULT);
1361                         GOTO(finish_req, rc = -E2BIG);
1362                 }
1363
1364                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1365                 OBD_ALLOC(tmp, lum_size);
1366                 if (tmp == NULL)
1367                         GOTO(finish_req, rc = -ENOMEM);
1368
1369                 mdt_index = ll_get_mdt_idx(inode);
1370                 if (mdt_index < 0)
1371                         GOTO(out_tmp, rc = -ENOMEM);
1372
1373                 tmp->lum_magic = LMV_MAGIC_V1;
1374                 tmp->lum_stripe_count = 0;
1375                 tmp->lum_stripe_offset = mdt_index;
1376                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1377                 for (i = 0; i < stripe_count; i++) {
1378                         struct lu_fid   fid;
1379
1380                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1381                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1382                         if (mdt_index < 0)
1383                                 GOTO(out_tmp, rc = mdt_index);
1384
1385                         tmp->lum_objects[i].lum_mds = mdt_index;
1386                         tmp->lum_objects[i].lum_fid = fid;
1387                         tmp->lum_stripe_count++;
1388                 }
1389
1390                 if (copy_to_user(ulmv, tmp, lum_size))
1391                         GOTO(out_tmp, rc = -EFAULT);
1392 out_tmp:
1393                 OBD_FREE(tmp, lum_size);
1394 finish_req:
1395                 ptlrpc_req_finished(request);
1396                 return rc;
1397         }
1398
1399         case LL_IOC_REMOVE_ENTRY: {
1400                 char            *filename = NULL;
1401                 int              namelen = 0;
1402                 int              rc;
1403
1404                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1405                  * unsupported server, which might crash the server(LU-2730),
1406                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1407                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1408                  * server will support REINT_RMENTRY XXX*/
1409                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1410                         RETURN(-ENOTSUPP);
1411
1412                 filename = ll_getname((const char __user *)arg);
1413                 if (IS_ERR(filename))
1414                         RETURN(PTR_ERR(filename));
1415
1416                 namelen = strlen(filename);
1417                 if (namelen < 1)
1418                         GOTO(out_rmdir, rc = -EINVAL);
1419
1420                 rc = ll_rmdir_entry(inode, filename, namelen);
1421 out_rmdir:
1422                 if (filename)
1423                         ll_putname(filename);
1424                 RETURN(rc);
1425         }
1426         case LL_IOC_LOV_SWAP_LAYOUTS:
1427                 RETURN(-EPERM);
1428         case IOC_OBD_STATFS:
1429                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1430         case LL_IOC_LOV_GETSTRIPE:
1431         case LL_IOC_MDC_GETINFO:
1432         case IOC_MDC_GETFILEINFO:
1433         case IOC_MDC_GETFILESTRIPE: {
1434                 struct ptlrpc_request *request = NULL;
1435                 struct lov_user_md __user *lump;
1436                 struct lov_mds_md *lmm = NULL;
1437                 struct mdt_body *body;
1438                 char *filename = NULL;
1439                 int lmmsize;
1440
1441                 if (cmd == IOC_MDC_GETFILEINFO ||
1442                     cmd == IOC_MDC_GETFILESTRIPE) {
1443                         filename = ll_getname((const char __user *)arg);
1444                         if (IS_ERR(filename))
1445                                 RETURN(PTR_ERR(filename));
1446
1447                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1448                                                       &lmmsize, &request);
1449                 } else {
1450                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1451                                               &request, 0);
1452                 }
1453
1454                 if (request) {
1455                         body = req_capsule_server_get(&request->rq_pill,
1456                                                       &RMF_MDT_BODY);
1457                         LASSERT(body != NULL);
1458                 } else {
1459                         GOTO(out_req, rc);
1460                 }
1461
1462                 if (rc < 0) {
1463                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1464                                                cmd == LL_IOC_MDC_GETINFO))
1465                                 GOTO(skip_lmm, rc = 0);
1466                         else
1467                                 GOTO(out_req, rc);
1468                 }
1469
1470                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1471                     cmd == LL_IOC_LOV_GETSTRIPE) {
1472                         lump = (struct lov_user_md __user *)arg;
1473                 } else {
1474                         struct lov_user_mds_data __user *lmdp;
1475                         lmdp = (struct lov_user_mds_data __user *)arg;
1476                         lump = &lmdp->lmd_lmm;
1477                 }
1478                 if (copy_to_user(lump, lmm, lmmsize)) {
1479                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1480                                 GOTO(out_req, rc = -EFAULT);
1481                         rc = -EOVERFLOW;
1482                 }
1483         skip_lmm:
1484                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1485                         struct lov_user_mds_data __user *lmdp;
1486                         lstat_t st = { 0 };
1487
1488                         st.st_dev       = inode->i_sb->s_dev;
1489                         st.st_mode      = body->mbo_mode;
1490                         st.st_nlink     = body->mbo_nlink;
1491                         st.st_uid       = body->mbo_uid;
1492                         st.st_gid       = body->mbo_gid;
1493                         st.st_rdev      = body->mbo_rdev;
1494                         st.st_size      = body->mbo_size;
1495                         st.st_blksize   = PAGE_SIZE;
1496                         st.st_blocks    = body->mbo_blocks;
1497                         st.st_atime     = body->mbo_atime;
1498                         st.st_mtime     = body->mbo_mtime;
1499                         st.st_ctime     = body->mbo_ctime;
1500                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1501                                                 sbi->ll_flags &
1502                                                 LL_SBI_32BIT_API);
1503
1504                         lmdp = (struct lov_user_mds_data __user *)arg;
1505                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1506                                 GOTO(out_req, rc = -EFAULT);
1507                 }
1508
1509                 EXIT;
1510         out_req:
1511                 ptlrpc_req_finished(request);
1512                 if (filename)
1513                         ll_putname(filename);
1514                 return rc;
1515         }
1516         case OBD_IOC_QUOTACTL: {
1517                 struct if_quotactl *qctl;
1518
1519                 OBD_ALLOC_PTR(qctl);
1520                 if (!qctl)
1521                         RETURN(-ENOMEM);
1522
1523                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1524                         GOTO(out_quotactl, rc = -EFAULT);
1525
1526                 rc = quotactl_ioctl(sbi, qctl);
1527
1528                 if (rc == 0 &&
1529                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1530                         rc = -EFAULT;
1531
1532         out_quotactl:
1533                 OBD_FREE_PTR(qctl);
1534                 RETURN(rc);
1535         }
1536         case OBD_IOC_GETDTNAME:
1537         case OBD_IOC_GETMDNAME:
1538                 RETURN(ll_get_obd_name(inode, cmd, arg));
1539         case LL_IOC_FLUSHCTX:
1540                 RETURN(ll_flush_ctx(inode));
1541         case LL_IOC_GETOBDCOUNT: {
1542                 u32 count, vallen;
1543                 struct obd_export *exp;
1544
1545                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1546                         RETURN(-EFAULT);
1547
1548                 /* get ost count when count is zero, get mdt count otherwise */
1549                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1550                 vallen = sizeof(count);
1551                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1552                                   KEY_TGT_COUNT, &vallen, &count);
1553                 if (rc) {
1554                         CERROR("get target count failed: %d\n", rc);
1555                         RETURN(rc);
1556                 }
1557
1558                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1559                         RETURN(-EFAULT);
1560
1561                 RETURN(0);
1562         }
1563         case LL_IOC_PATH2FID:
1564                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1565                                      sizeof(struct lu_fid)))
1566                         RETURN(-EFAULT);
1567                 RETURN(0);
1568         case LL_IOC_GET_CONNECT_FLAGS: {
1569                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1570                                      (void __user *)arg));
1571         }
1572         case OBD_IOC_FID2PATH:
1573                 RETURN(ll_fid2path(inode, (void __user *)arg));
1574         case LL_IOC_GETPARENT:
1575                 RETURN(ll_getparent(file, (void __user *)arg));
1576         case LL_IOC_FID2MDTIDX: {
1577                 struct obd_export *exp = ll_i2mdexp(inode);
1578                 struct lu_fid     fid;
1579                 __u32             index;
1580
1581                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1582                                    sizeof(fid)))
1583                         RETURN(-EFAULT);
1584
1585                 /* Call mdc_iocontrol */
1586                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1587                                    (__u32 __user *)&index);
1588                 if (rc != 0)
1589                         RETURN(rc);
1590
1591                 RETURN(index);
1592         }
1593         case LL_IOC_HSM_REQUEST: {
1594                 struct hsm_user_request *hur;
1595                 ssize_t                  totalsize;
1596
1597                 OBD_ALLOC_PTR(hur);
1598                 if (hur == NULL)
1599                         RETURN(-ENOMEM);
1600
1601                 /* We don't know the true size yet; copy the fixed-size part */
1602                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1603                         OBD_FREE_PTR(hur);
1604                         RETURN(-EFAULT);
1605                 }
1606
1607                 /* Compute the whole struct size */
1608                 totalsize = hur_len(hur);
1609                 OBD_FREE_PTR(hur);
1610                 if (totalsize < 0)
1611                         RETURN(-E2BIG);
1612
1613                 /* Final size will be more than double totalsize */
1614                 if (totalsize >= MDS_MAXREQSIZE / 3)
1615                         RETURN(-E2BIG);
1616
1617                 OBD_ALLOC_LARGE(hur, totalsize);
1618                 if (hur == NULL)
1619                         RETURN(-ENOMEM);
1620
1621                 /* Copy the whole struct */
1622                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1623                         GOTO(out_hur, rc = -EFAULT);
1624
1625                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1626                         const struct lu_fid *fid;
1627                         struct inode *f;
1628                         int i;
1629
1630                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1631                                 fid = &hur->hur_user_item[i].hui_fid;
1632                                 f = search_inode_for_lustre(inode->i_sb, fid);
1633                                 if (IS_ERR(f)) {
1634                                         rc = PTR_ERR(f);
1635                                         break;
1636                                 }
1637
1638                                 rc = ll_hsm_release(f);
1639                                 iput(f);
1640                                 if (rc != 0)
1641                                         break;
1642                         }
1643                 } else {
1644                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1645                                            hur, NULL);
1646                 }
1647
1648 out_hur:
1649                 OBD_FREE_LARGE(hur, totalsize);
1650
1651                 RETURN(rc);
1652         }
1653         case LL_IOC_HSM_PROGRESS: {
1654                 struct hsm_progress_kernel      hpk;
1655                 struct hsm_progress             hp;
1656
1657                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1658                         RETURN(-EFAULT);
1659
1660                 hpk.hpk_fid = hp.hp_fid;
1661                 hpk.hpk_cookie = hp.hp_cookie;
1662                 hpk.hpk_extent = hp.hp_extent;
1663                 hpk.hpk_flags = hp.hp_flags;
1664                 hpk.hpk_errval = hp.hp_errval;
1665                 hpk.hpk_data_version = 0;
1666
1667                 /* File may not exist in Lustre; all progress
1668                  * reported to Lustre root */
1669                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1670                                    NULL);
1671                 RETURN(rc);
1672         }
1673         case LL_IOC_HSM_CT_START:
1674                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1675                         RETURN(-EPERM);
1676
1677                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1678                                     sizeof(struct lustre_kernelcomm));
1679                 RETURN(rc);
1680
1681         case LL_IOC_HSM_COPY_START: {
1682                 struct hsm_copy *copy;
1683                 int              rc;
1684
1685                 OBD_ALLOC_PTR(copy);
1686                 if (copy == NULL)
1687                         RETURN(-ENOMEM);
1688                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1689                         OBD_FREE_PTR(copy);
1690                         RETURN(-EFAULT);
1691                 }
1692
1693                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1694                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1695                         rc = -EFAULT;
1696
1697                 OBD_FREE_PTR(copy);
1698                 RETURN(rc);
1699         }
1700         case LL_IOC_HSM_COPY_END: {
1701                 struct hsm_copy *copy;
1702                 int              rc;
1703
1704                 OBD_ALLOC_PTR(copy);
1705                 if (copy == NULL)
1706                         RETURN(-ENOMEM);
1707                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1708                         OBD_FREE_PTR(copy);
1709                         RETURN(-EFAULT);
1710                 }
1711
1712                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1713                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1714                         rc = -EFAULT;
1715
1716                 OBD_FREE_PTR(copy);
1717                 RETURN(rc);
1718         }
1719         case LL_IOC_MIGRATE: {
1720                 char            *buf = NULL;
1721                 const char      *filename;
1722                 int             namelen = 0;
1723                 int             len;
1724                 int             rc;
1725                 int             mdtidx;
1726
1727                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1728                 if (rc < 0)
1729                         RETURN(rc);
1730
1731                 data = (struct obd_ioctl_data *)buf;
1732                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1733                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1734                         GOTO(migrate_free, rc = -EINVAL);
1735
1736                 filename = data->ioc_inlbuf1;
1737                 namelen = data->ioc_inllen1;
1738                 /* \0 is packed at the end of filename */
1739                 if (namelen < 1 || namelen != strlen(filename) + 1)
1740                         GOTO(migrate_free, rc = -EINVAL);
1741
1742                 if (data->ioc_inllen2 != sizeof(mdtidx))
1743                         GOTO(migrate_free, rc = -EINVAL);
1744                 mdtidx = *(int *)data->ioc_inlbuf2;
1745
1746                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1747 migrate_free:
1748                 OBD_FREE_LARGE(buf, len);
1749
1750                 RETURN(rc);
1751         }
1752         case LL_IOC_FSGETXATTR:
1753         case LL_IOC_FSSETXATTR:
1754                 RETURN(ll_ioctl_projid(inode, cmd, arg));
1755         default:
1756                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1757                                      (void __user *)arg));
1758         }
1759 }
1760
1761 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1762 {
1763         struct inode *inode = file->f_mapping->host;
1764         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1765         struct ll_sb_info *sbi = ll_i2sbi(inode);
1766         int api32 = ll_need_32bit_api(sbi);
1767         loff_t ret = -EINVAL;
1768         ENTRY;
1769
1770         inode_lock(inode);
1771         switch (origin) {
1772                 case SEEK_SET:
1773                         break;
1774                 case SEEK_CUR:
1775                         offset += file->f_pos;
1776                         break;
1777                 case SEEK_END:
1778                         if (offset > 0)
1779                                 GOTO(out, ret);
1780                         if (api32)
1781                                 offset += LL_DIR_END_OFF_32BIT;
1782                         else
1783                                 offset += LL_DIR_END_OFF;
1784                         break;
1785                 default:
1786                         GOTO(out, ret);
1787         }
1788
1789         if (offset >= 0 &&
1790             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1791              (!api32 && offset <= LL_DIR_END_OFF))) {
1792                 if (offset != file->f_pos) {
1793                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1794                             (!api32 && offset == LL_DIR_END_OFF))
1795                                 fd->lfd_pos = MDS_DIR_END_OFF;
1796                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1797                                 fd->lfd_pos = offset << 32;
1798                         else
1799                                 fd->lfd_pos = offset;
1800                         file->f_pos = offset;
1801                         file->f_version = 0;
1802                 }
1803                 ret = offset;
1804         }
1805         GOTO(out, ret);
1806
1807 out:
1808         inode_unlock(inode);
1809         return ret;
1810 }
1811
1812 static int ll_dir_open(struct inode *inode, struct file *file)
1813 {
1814         ENTRY;
1815         RETURN(ll_file_open(inode, file));
1816 }
1817
1818 static int ll_dir_release(struct inode *inode, struct file *file)
1819 {
1820         ENTRY;
1821         RETURN(ll_file_release(inode, file));
1822 }
1823
1824 const struct file_operations ll_dir_operations = {
1825         .llseek         = ll_dir_seek,
1826         .open           = ll_dir_open,
1827         .release        = ll_dir_release,
1828         .read           = generic_read_dir,
1829 #ifdef HAVE_DIR_CONTEXT
1830         .iterate        = ll_iterate,
1831 #else
1832         .readdir        = ll_readdir,
1833 #endif
1834         .unlocked_ioctl = ll_dir_ioctl,
1835         .fsync          = ll_fsync,
1836 };