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 = ll_full_name_hash(dparent, dirname,
 452                                                   strlen(dirname)),
 453                 },
 454         };
 455         int err;
 456         ENTRY;
 457
 458         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 459                 RETURN(-EINVAL);
 460
 461         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
 462                "stripe_offset %d, stripe_count: %u\n",
 463                PFID(ll_inode2fid(parent)), parent, dirname,
 464                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 465
 466         if (lump->lum_stripe_count > 1 &&
 467             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 468                 RETURN(-EINVAL);
 469
 470         if (IS_DEADDIR(parent) &&
 471             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 472                 RETURN(-ENOENT);
 473
 474         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 475                 lustre_swab_lmv_user_md(lump);
 476
 477         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 478                 mode &= ~current_umask();
 479         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 480         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 481                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 482                                      lump);
 483         if (IS_ERR(op_data))
 484                 RETURN(PTR_ERR(op_data));
 485
 486         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 487                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 488                  * to determine the security context for the file. So our fake
 489                  * dentry should be real enough for this purpose. */
 490                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 491                                               &op_data->op_file_secctx_name,
 492                                               &op_data->op_file_secctx,
 493                                               &op_data->op_file_secctx_size);
 494                 if (err < 0)
 495                         GOTO(out_op_data, err);
 496         }
 497
 498         op_data->op_cli_flags |= CLI_SET_MEA;
 499         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 500                         from_kuid(&init_user_ns, current_fsuid()),
 501                         from_kgid(&init_user_ns, current_fsgid()),
 502                         cfs_curproc_cap_pack(), 0, &request);
 503         if (err)
 504                 GOTO(out_request, err);
 505
 506         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 507
 508         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 509         if (err)
 510                 GOTO(out_inode, err);
 511
 512         dentry.d_inode = inode;
 513
 514         if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
 515                 err = ll_inode_init_security(&dentry, inode, parent);
 516                 if (err)
 517                         GOTO(out_inode, err);
 518         }
 519
 520 out_inode:
 521         if (inode != NULL)
 522                 iput(inode);
 523 out_request:
 524         ptlrpc_req_finished(request);
 525 out_op_data:
 526         ll_finish_md_op_data(op_data);
 527
 528         return err;
 529 }
 530
 531 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 532                      int set_default)
 533 {
 534         struct ll_sb_info *sbi = ll_i2sbi(inode);
 535         struct md_op_data *op_data;
 536         struct ptlrpc_request *req = NULL;
 537         int rc = 0;
 538 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 539         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 540         struct obd_device *mgc = lsi->lsi_mgc;
 541 #endif
 542         int lum_size;
 543         ENTRY;
 544
 545         if (lump != NULL) {
 546                 /*
 547                  * This is coming from userspace, so should be in
 548                  * local endian.  But the MDS would like it in little
 549                  * endian, so we swab it before we send it.
 550                  */
 551                 switch (lump->lmm_magic) {
 552                 case LOV_USER_MAGIC_V1: {
 553                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 554                                 lustre_swab_lov_user_md_v1(lump);
 555                         lum_size = sizeof(struct lov_user_md_v1);
 556                         break;
 557                 }
 558                 case LOV_USER_MAGIC_V3: {
 559                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 560                                 lustre_swab_lov_user_md_v3(
 561                                         (struct lov_user_md_v3 *)lump);
 562                         lum_size = sizeof(struct lov_user_md_v3);
 563                         break;
 564                 }
 565                 case LOV_USER_MAGIC_COMP_V1: {
 566                         if (lump->lmm_magic !=
 567                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 568                                 lustre_swab_lov_comp_md_v1(
 569                                         (struct lov_comp_md_v1 *)lump);
 570                         lum_size = le32_to_cpu(
 571                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 572                         break;
 573                 }
 574                 case LMV_USER_MAGIC: {
 575                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 576                                 lustre_swab_lmv_user_md(
 577                                         (struct lmv_user_md *)lump);
 578                         lum_size = sizeof(struct lmv_user_md);
 579                         break;
 580                 }
 581                 default: {
 582                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 583                                         " %#08x != %#08x nor %#08x\n",
 584                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 585                                         LOV_USER_MAGIC_V3);
 586                         RETURN(-EINVAL);
 587                 }
 588                 }
 589         } else {
 590                 lum_size = sizeof(struct lov_user_md_v1);
 591         }
 592
 593         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 594                                      LUSTRE_OPC_ANY, NULL);
 595         if (IS_ERR(op_data))
 596                 RETURN(PTR_ERR(op_data));
 597
 598         /* swabbing is done in lov_setstripe() on server side */
 599         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 600         ll_finish_md_op_data(op_data);
 601         ptlrpc_req_finished(req);
 602         if (rc)
 603                 RETURN(rc);
 604
 605 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 606         /*
 607          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 608          * and it's stored into system config for backward compatibility.
 609          *
 610          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 611          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 612          * need the make the distiction between the 2 versions
 613          */
 614         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 615             (lump == NULL ||
 616              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 617              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 618                 char *param = NULL;
 619                 char *buf;
 620
 621                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 622                 if (param == NULL)
 623                         GOTO(end, rc = -ENOMEM);
 624
 625                 buf = param;
 626                 /* Get fsname and assume devname to be -MDT0000. */
 627                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 628                 strcat(buf, "-MDT0000.lov");
 629                 buf += strlen(buf);
 630
 631                 /* Set root stripesize */
 632                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 633                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 634                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 635                 if (rc)
 636                         GOTO(end, rc);
 637
 638                 /* Set root stripecount */
 639                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 640                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 641                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 642                 if (rc)
 643                         GOTO(end, rc);
 644
 645                 /* Set root stripeoffset */
 646                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 647                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 648                                 (typeof(lump->lmm_stripe_offset))(-1));
 649                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 650
 651 end:
 652                 if (param != NULL)
 653                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 654         }
 655 #endif
 656         RETURN(rc);
 657 }
 658
 659 /**
 660  * This function will be used to get default LOV/LMV/Default LMV
 661  * @valid will be used to indicate which stripe it will retrieve
 662  *      OBD_MD_MEA              LMV stripe EA
 663  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 664  *      otherwise               Default LOV EA.
 665  * Each time, it can only retrieve 1 stripe EA
 666  **/
 667 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 668                      struct ptlrpc_request **request, u64 valid)
 669 {
 670         struct ll_sb_info *sbi = ll_i2sbi(inode);
 671         struct mdt_body   *body;
 672         struct lov_mds_md *lmm = NULL;
 673         struct ptlrpc_request *req = NULL;
 674         int rc, lmm_size;
 675         struct md_op_data *op_data;
 676         ENTRY;
 677
 678         rc = ll_get_default_mdsize(sbi, &lmm_size);
 679         if (rc)
 680                 RETURN(rc);
 681
 682         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 683                                      0, lmm_size, LUSTRE_OPC_ANY,
 684                                      NULL);
 685         if (IS_ERR(op_data))
 686                 RETURN(PTR_ERR(op_data));
 687
 688         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 689         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 690         ll_finish_md_op_data(op_data);
 691         if (rc < 0) {
 692                 CDEBUG(D_INFO, "md_getattr failed on inode "
 693                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 694                 GOTO(out, rc);
 695         }
 696
 697         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 698         LASSERT(body != NULL);
 699
 700         lmm_size = body->mbo_eadatasize;
 701
 702         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 703             lmm_size == 0) {
 704                 GOTO(out, rc = -ENODATA);
 705         }
 706
 707         lmm = req_capsule_server_sized_get(&req->rq_pill,
 708                                            &RMF_MDT_MD, lmm_size);
 709         LASSERT(lmm != NULL);
 710
 711         /*
 712          * This is coming from the MDS, so is probably in
 713          * little endian.  We convert it to host endian before
 714          * passing it to userspace.
 715          */
 716         /* We don't swab objects for directories */
 717         switch (le32_to_cpu(lmm->lmm_magic)) {
 718         case LOV_MAGIC_V1:
 719                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 720                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 721                 break;
 722         case LOV_MAGIC_V3:
 723                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 724                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 725                 break;
 726         case LOV_MAGIC_COMP_V1:
 727                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 728                         lustre_swab_lov_comp_md_v1(
 729                                         (struct lov_comp_md_v1 *)lmm);
 730                 break;
 731         case LMV_MAGIC_V1:
 732                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 733                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 734                 break;
 735         case LMV_USER_MAGIC:
 736                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 737                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 738                 break;
 739         default:
 740                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 741                 rc = -EPROTO;
 742         }
 743 out:
 744         *plmm = lmm;
 745         *plmm_size = lmm_size;
 746         *request = req;
 747         return rc;
 748 }
 749
 750 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 751 {
 752         struct md_op_data       *op_data;
 753         int                     rc;
 754         int                     mdt_index;
 755         ENTRY;
 756
 757         OBD_ALLOC_PTR(op_data);
 758         if (op_data == NULL)
 759                 RETURN(-ENOMEM);
 760
 761         op_data->op_flags |= MF_GET_MDT_IDX;
 762         op_data->op_fid1 = *fid;
 763         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 764         mdt_index = op_data->op_mds;
 765         OBD_FREE_PTR(op_data);
 766         if (rc < 0)
 767                 RETURN(rc);
 768
 769         RETURN(mdt_index);
 770 }
 771
 772 /*
 773  *  Get MDT index for the inode.
 774  */
 775 int ll_get_mdt_idx(struct inode *inode)
 776 {
 777         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 778 }
 779
 780 /**
 781  * Generic handler to do any pre-copy work.
 782  *
 783  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 784  * first information for it that real work has started.
 785  *
 786  * Moreover, for a ARCHIVE request, it will sample the file data version and
 787  * store it in \a copy.
 788  *
 789  * \return 0 on success.
 790  */
 791 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 792 {
 793         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 794         struct hsm_progress_kernel       hpk;
 795         int                              rc = 0;
 796         int                              rc2;
 797         ENTRY;
 798
 799         /* Forge a hsm_progress based on data from copy. */
 800         hpk.hpk_fid = copy->hc_hai.hai_fid;
 801         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 802         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 803         hpk.hpk_extent.length = 0;
 804         hpk.hpk_flags = 0;
 805         hpk.hpk_errval = 0;
 806         hpk.hpk_data_version = 0;
 807
 808
 809         /* For archive request, we need to read the current file version. */
 810         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 811                 struct inode    *inode;
 812                 __u64            data_version = 0;
 813
 814                 /* Get inode for this fid */
 815                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 816                 if (IS_ERR(inode)) {
 817                         hpk.hpk_flags |= HP_FLAG_RETRY;
 818                         /* hpk_errval is >= 0 */
 819                         hpk.hpk_errval = -PTR_ERR(inode);
 820                         GOTO(progress, rc = PTR_ERR(inode));
 821                 }
 822
 823                 /* Read current file data version */
 824                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 825                 iput(inode);
 826                 if (rc != 0) {
 827                         CDEBUG(D_HSM, "Could not read file data version of "
 828                                       DFID" (rc = %d). Archive request ("
 829                                       "%#llx) could not be done.\n",
 830                                       PFID(&copy->hc_hai.hai_fid), rc,
 831                                       copy->hc_hai.hai_cookie);
 832                         hpk.hpk_flags |= HP_FLAG_RETRY;
 833                         /* hpk_errval must be >= 0 */
 834                         hpk.hpk_errval = -rc;
 835                         GOTO(progress, rc);
 836                 }
 837
 838                 /* Store in the hsm_copy for later copytool use.
 839                  * Always modified even if no lsm. */
 840                 copy->hc_data_version = data_version;
 841         }
 842
 843 progress:
 844         /* On error, the request should be considered as completed */
 845         if (hpk.hpk_errval > 0)
 846                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 847
 848         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 849                             &hpk, NULL);
 850
 851         /* Return first error */
 852         RETURN(rc != 0 ? rc : rc2);
 853 }
 854
 855 /**
 856  * Generic handler to do any post-copy work.
 857  *
 858  * It will send the last hsm_progress update to coordinator to inform it
 859  * that copy is finished and whether it was successful or not.
 860  *
 861  * Moreover,
 862  * - for ARCHIVE request, it will sample the file data version and compare it
 863  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 864  *   will be considered as failed.
 865  * - for RESTORE request, it will sample the file data version and send it to
 866  *   coordinator which is useful if the file was imported as 'released'.
 867  *
 868  * \return 0 on success.
 869  */
 870 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 871 {
 872         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 873         struct hsm_progress_kernel       hpk;
 874         int                              rc = 0;
 875         int                              rc2;
 876         ENTRY;
 877
 878         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 879         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 880          * initialized if copy_end was called with copy == NULL.
 881          */
 882
 883         /* Forge a hsm_progress based on data from copy. */
 884         hpk.hpk_fid = copy->hc_hai.hai_fid;
 885         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 886         hpk.hpk_extent = copy->hc_hai.hai_extent;
 887         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 888         hpk.hpk_errval = copy->hc_errval;
 889         hpk.hpk_data_version = 0;
 890
 891         /* For archive request, we need to check the file data was not changed.
 892          *
 893          * For restore request, we need to send the file data version, this is
 894          * useful when the file was created using hsm_import.
 895          */
 896         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 897              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 898             (copy->hc_errval == 0)) {
 899                 struct inode    *inode;
 900                 __u64            data_version = 0;
 901
 902                 /* Get lsm for this fid */
 903                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 904                 if (IS_ERR(inode)) {
 905                         hpk.hpk_flags |= HP_FLAG_RETRY;
 906                         /* hpk_errval must be >= 0 */
 907                         hpk.hpk_errval = -PTR_ERR(inode);
 908                         GOTO(progress, rc = PTR_ERR(inode));
 909                 }
 910
 911                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 912                 iput(inode);
 913                 if (rc) {
 914                         CDEBUG(D_HSM, "Could not read file data version. "
 915                                       "Request could not be confirmed.\n");
 916                         if (hpk.hpk_errval == 0)
 917                                 hpk.hpk_errval = -rc;
 918                         GOTO(progress, rc);
 919                 }
 920
 921                 /* Store in the hsm_copy for later copytool use.
 922                  * Always modified even if no lsm. */
 923                 hpk.hpk_data_version = data_version;
 924
 925                 /* File could have been stripped during archiving, so we need
 926                  * to check anyway. */
 927                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 928                     (copy->hc_data_version != data_version)) {
 929                         CDEBUG(D_HSM, "File data version mismatched. "
 930                               "File content was changed during archiving. "
 931                                DFID", start:%#llx current:%#llx\n",
 932                                PFID(&copy->hc_hai.hai_fid),
 933                                copy->hc_data_version, data_version);
 934                         /* File was changed, send error to cdt. Do not ask for
 935                          * retry because if a file is modified frequently,
 936                          * the cdt will loop on retried archive requests.
 937                          * The policy engine will ask for a new archive later
 938                          * when the file will not be modified for some tunable
 939                          * time */
 940                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 941                         rc = -EBUSY;
 942                         /* hpk_errval must be >= 0 */
 943                         hpk.hpk_errval = -rc;
 944                         GOTO(progress, rc);
 945                 }
 946
 947         }
 948
 949 progress:
 950         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 951                             &hpk, NULL);
 952
 953         /* Return first error */
 954         RETURN(rc != 0 ? rc : rc2);
 955 }
 956
 957
 958 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 959                           const void __user *data, size_t size)
 960 {
 961         void *copy;
 962         int rc;
 963
 964         OBD_ALLOC(copy, size);
 965         if (copy == NULL)
 966                 return -ENOMEM;
 967
 968         if (copy_from_user(copy, data, size)) {
 969                 rc = -EFAULT;
 970                 goto out;
 971         }
 972
 973         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 974 out:
 975         OBD_FREE(copy, size);
 976
 977         return rc;
 978 }
 979
 980 static int check_owner(int type, int id)
 981 {
 982         switch (type) {
 983         case USRQUOTA:
 984                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
 985                         return -EPERM;
 986                 break;
 987         case GRPQUOTA:
 988                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
 989                         return -EPERM;
 990                 break;
 991         case PRJQUOTA:
 992                 break;
 993         }
 994         return 0;
 995 }
 996
 997 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 998 {
 999         int cmd = qctl->qc_cmd;
1000         int type = qctl->qc_type;
1001         int id = qctl->qc_id;
1002         int valid = qctl->qc_valid;
1003         int rc = 0;
1004         ENTRY;
1005
1006         switch (cmd) {
1007         case Q_SETQUOTA:
1008         case Q_SETINFO:
1009                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1010                         RETURN(-EPERM);
1011                 break;
1012         case Q_GETQUOTA:
1013                 if (check_owner(type, id) &&
1014                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1015                         RETURN(-EPERM);
1016                 break;
1017         case Q_GETINFO:
1018                 break;
1019         default:
1020                 CERROR("unsupported quotactl op: %#x\n", cmd);
1021                 RETURN(-ENOTTY);
1022         }
1023
1024         if (valid != QC_GENERAL) {
1025                 if (cmd == Q_GETINFO)
1026                         qctl->qc_cmd = Q_GETOINFO;
1027                 else if (cmd == Q_GETQUOTA)
1028                         qctl->qc_cmd = Q_GETOQUOTA;
1029                 else
1030                         RETURN(-EINVAL);
1031
1032                 switch (valid) {
1033                 case QC_MDTIDX:
1034                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1035                                            sizeof(*qctl), qctl, NULL);
1036                         break;
1037                 case QC_OSTIDX:
1038                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1039                                            sizeof(*qctl), qctl, NULL);
1040                         break;
1041                 case QC_UUID:
1042                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1043                                            sizeof(*qctl), qctl, NULL);
1044                         if (rc == -EAGAIN)
1045                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1046                                                    sbi->ll_dt_exp,
1047                                                    sizeof(*qctl), qctl, NULL);
1048                         break;
1049                 default:
1050                         rc = -EINVAL;
1051                         break;
1052                 }
1053
1054                 if (rc)
1055                         RETURN(rc);
1056
1057                 qctl->qc_cmd = cmd;
1058         } else {
1059                 struct obd_quotactl *oqctl;
1060
1061                 OBD_ALLOC_PTR(oqctl);
1062                 if (oqctl == NULL)
1063                         RETURN(-ENOMEM);
1064
1065                 QCTL_COPY(oqctl, qctl);
1066                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1067                 if (rc) {
1068                         OBD_FREE_PTR(oqctl);
1069                         RETURN(rc);
1070                 }
1071                 /* If QIF_SPACE is not set, client should collect the
1072                  * space usage from OSSs by itself */
1073                 if (cmd == Q_GETQUOTA &&
1074                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1075                     !oqctl->qc_dqblk.dqb_curspace) {
1076                         struct obd_quotactl *oqctl_tmp;
1077
1078                         OBD_ALLOC_PTR(oqctl_tmp);
1079                         if (oqctl_tmp == NULL)
1080                                 GOTO(out, rc = -ENOMEM);
1081
1082                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1083                         oqctl_tmp->qc_id = oqctl->qc_id;
1084                         oqctl_tmp->qc_type = oqctl->qc_type;
1085
1086                         /* collect space usage from OSTs */
1087                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1088                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1089                         if (!rc || rc == -EREMOTEIO) {
1090                                 oqctl->qc_dqblk.dqb_curspace =
1091                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1092                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1093                         }
1094
1095                         /* collect space & inode usage from MDTs */
1096                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1097                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1098                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1099                         if (!rc || rc == -EREMOTEIO) {
1100                                 oqctl->qc_dqblk.dqb_curspace +=
1101                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1102                                 oqctl->qc_dqblk.dqb_curinodes =
1103                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1104                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1105                         } else {
1106                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1107                         }
1108
1109                         OBD_FREE_PTR(oqctl_tmp);
1110                 }
1111 out:
1112                 QCTL_COPY(qctl, oqctl);
1113                 OBD_FREE_PTR(oqctl);
1114         }
1115
1116         RETURN(rc);
1117 }
1118
1119 /* This function tries to get a single name component,
1120  * to send to the server. No actual path traversal involved,
1121  * so we limit to NAME_MAX */
1122 static char *ll_getname(const char __user *filename)
1123 {
1124         int ret = 0, len;
1125         char *tmp;
1126
1127         OBD_ALLOC(tmp, NAME_MAX + 1);
1128
1129         if (!tmp)
1130                 return ERR_PTR(-ENOMEM);
1131
1132         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1133         if (len < 0)
1134                 ret = -ENOENT;
1135         else if (len > NAME_MAX)
1136                 ret = -ENAMETOOLONG;
1137
1138         if (ret) {
1139                 OBD_FREE(tmp, NAME_MAX + 1);
1140                 tmp =  ERR_PTR(ret);
1141         }
1142         return tmp;
1143 }
1144
1145 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1146
1147 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1148 {
1149         struct dentry *dentry = file_dentry(file);
1150         struct inode *inode = file_inode(file);
1151         struct ll_sb_info *sbi = ll_i2sbi(inode);
1152         struct obd_ioctl_data *data;
1153         int rc = 0;
1154         ENTRY;
1155
1156         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1157                PFID(ll_inode2fid(inode)), inode, cmd);
1158
1159         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1160         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1161                 return -ENOTTY;
1162
1163         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1164         switch(cmd) {
1165         case FSFILT_IOC_GETFLAGS:
1166         case FSFILT_IOC_SETFLAGS:
1167                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1168         case FSFILT_IOC_GETVERSION_OLD:
1169         case FSFILT_IOC_GETVERSION:
1170                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1171         /* We need to special case any other ioctls we want to handle,
1172          * to send them to the MDS/OST as appropriate and to properly
1173          * network encode the arg field.
1174         case FSFILT_IOC_SETVERSION_OLD:
1175         case FSFILT_IOC_SETVERSION:
1176         */
1177         case LL_IOC_GET_MDTIDX: {
1178                 int mdtidx;
1179
1180                 mdtidx = ll_get_mdt_idx(inode);
1181                 if (mdtidx < 0)
1182                         RETURN(mdtidx);
1183
1184                 if (put_user((int)mdtidx, (int __user *)arg))
1185                         RETURN(-EFAULT);
1186
1187                 return 0;
1188         }
1189         case IOC_MDC_LOOKUP: {
1190                 int namelen, len = 0;
1191                 char *buf = NULL;
1192                 char *filename;
1193
1194                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1195                 if (rc != 0)
1196                         RETURN(rc);
1197                 data = (void *)buf;
1198
1199                 filename = data->ioc_inlbuf1;
1200                 namelen = strlen(filename);
1201                 if (namelen < 1) {
1202                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1203                         GOTO(out_free, rc = -EINVAL);
1204                 }
1205
1206                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1207                 if (rc < 0) {
1208                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1209                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1210                                filename, rc);
1211                         GOTO(out_free, rc);
1212                 }
1213 out_free:
1214                 OBD_FREE_LARGE(buf, len);
1215                 return rc;
1216         }
1217         case LL_IOC_LMV_SETSTRIPE: {
1218                 struct lmv_user_md  *lum;
1219                 char            *buf = NULL;
1220                 char            *filename;
1221                 int              namelen = 0;
1222                 int              lumlen = 0;
1223                 umode_t          mode;
1224                 int              len;
1225                 int              rc;
1226
1227                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1228                 if (rc)
1229                         RETURN(rc);
1230
1231                 data = (void *)buf;
1232                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1233                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1234                         GOTO(lmv_out_free, rc = -EINVAL);
1235
1236                 filename = data->ioc_inlbuf1;
1237                 namelen = data->ioc_inllen1;
1238
1239                 if (namelen < 1) {
1240                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1241                         GOTO(lmv_out_free, rc = -EINVAL);
1242                 }
1243                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1244                 lumlen = data->ioc_inllen2;
1245
1246                 if (lum->lum_magic != LMV_USER_MAGIC ||
1247                     lumlen != sizeof(*lum)) {
1248                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1249                                filename, lum->lum_magic, lumlen, -EFAULT);
1250                         GOTO(lmv_out_free, rc = -EINVAL);
1251                 }
1252
1253 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1254                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1255 #else
1256                 mode = data->ioc_type;
1257 #endif
1258                 rc = ll_dir_setdirstripe(dentry, lum, filename, mode);
1259 lmv_out_free:
1260                 OBD_FREE_LARGE(buf, len);
1261                 RETURN(rc);
1262
1263         }
1264         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1265                 struct lmv_user_md        lum;
1266                 struct lmv_user_md __user *ulump =
1267                                         (struct lmv_user_md __user *)arg;
1268                 int                       rc;
1269
1270                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1271                         RETURN(-EFAULT);
1272
1273                 if (lum.lum_magic != LMV_USER_MAGIC)
1274                         RETURN(-EINVAL);
1275
1276                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1277
1278                 RETURN(rc);
1279         }
1280         case LL_IOC_LOV_SETSTRIPE: {
1281                 struct lov_user_md_v3 lumv3;
1282                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1283                 struct lov_user_md_v1 __user *lumv1p =
1284                         (struct lov_user_md_v1 __user *)arg;
1285                 struct lov_user_md_v3 __user *lumv3p =
1286                         (struct lov_user_md_v3 __user *)arg;
1287
1288                 int set_default = 0;
1289
1290                 CLASSERT(sizeof(struct lov_user_md_v3) >
1291                          sizeof(struct lov_comp_md_v1));
1292                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1293                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1294                                 sizeof(lumv3p->lmm_objects[0]));
1295                 /* first try with v1 which is smaller than v3 */
1296                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1297                         RETURN(-EFAULT);
1298
1299                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3)
1300                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1301                                 RETURN(-EFAULT);
1302
1303                 if (inode->i_sb->s_root == file_dentry(file))
1304                         set_default = 1;
1305
1306                 /* in v1 and v3 cases lumv1 points to data */
1307                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1308
1309                 RETURN(rc);
1310         }
1311         case LL_IOC_LMV_GETSTRIPE: {
1312                 struct lmv_user_md __user *ulmv =
1313                                         (struct lmv_user_md __user *)arg;
1314                 struct lmv_user_md      lum;
1315                 struct ptlrpc_request   *request = NULL;
1316                 union lmv_mds_md        *lmm = NULL;
1317                 int                     lmmsize;
1318                 u64                     valid = 0;
1319                 struct lmv_user_md      *tmp = NULL;
1320                 int                     mdt_index;
1321                 int                     lum_size;
1322                 int                     stripe_count;
1323                 int                     max_stripe_count;
1324                 int                     i;
1325                 int                     rc;
1326
1327                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1328                         RETURN(-EFAULT);
1329
1330                 max_stripe_count = lum.lum_stripe_count;
1331                 /* lum_magic will indicate which stripe the ioctl will like
1332                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1333                  * is for default LMV stripe */
1334                 if (lum.lum_magic == LMV_MAGIC_V1)
1335                         valid |= OBD_MD_MEA;
1336                 else if (lum.lum_magic == LMV_USER_MAGIC)
1337                         valid |= OBD_MD_DEFAULT_MEA;
1338                 else
1339                         RETURN(-EINVAL);
1340
1341                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1342                                       valid);
1343                 if (rc != 0)
1344                         GOTO(finish_req, rc);
1345
1346                 /* Get default LMV EA */
1347                 if (lum.lum_magic == LMV_USER_MAGIC) {
1348                         if (lmmsize > sizeof(*ulmv))
1349                                 GOTO(finish_req, rc = -EINVAL);
1350
1351                         if (copy_to_user(ulmv, lmm, lmmsize))
1352                                 GOTO(finish_req, rc = -EFAULT);
1353
1354                         GOTO(finish_req, rc);
1355                 }
1356
1357                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1358                 if (max_stripe_count < stripe_count) {
1359                         lum.lum_stripe_count = stripe_count;
1360                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1361                                 GOTO(finish_req, rc = -EFAULT);
1362                         GOTO(finish_req, rc = -E2BIG);
1363                 }
1364
1365                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1366                 OBD_ALLOC(tmp, lum_size);
1367                 if (tmp == NULL)
1368                         GOTO(finish_req, rc = -ENOMEM);
1369
1370                 mdt_index = ll_get_mdt_idx(inode);
1371                 if (mdt_index < 0)
1372                         GOTO(out_tmp, rc = -ENOMEM);
1373
1374                 tmp->lum_magic = LMV_MAGIC_V1;
1375                 tmp->lum_stripe_count = 0;
1376                 tmp->lum_stripe_offset = mdt_index;
1377                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1378                 for (i = 0; i < stripe_count; i++) {
1379                         struct lu_fid   fid;
1380
1381                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1382                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1383                         if (mdt_index < 0)
1384                                 GOTO(out_tmp, rc = mdt_index);
1385
1386                         tmp->lum_objects[i].lum_mds = mdt_index;
1387                         tmp->lum_objects[i].lum_fid = fid;
1388                         tmp->lum_stripe_count++;
1389                 }
1390
1391                 if (copy_to_user(ulmv, tmp, lum_size))
1392                         GOTO(out_tmp, rc = -EFAULT);
1393 out_tmp:
1394                 OBD_FREE(tmp, lum_size);
1395 finish_req:
1396                 ptlrpc_req_finished(request);
1397                 return rc;
1398         }
1399
1400         case LL_IOC_REMOVE_ENTRY: {
1401                 char            *filename = NULL;
1402                 int              namelen = 0;
1403                 int              rc;
1404
1405                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1406                  * unsupported server, which might crash the server(LU-2730),
1407                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1408                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1409                  * server will support REINT_RMENTRY XXX*/
1410                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1411                         RETURN(-ENOTSUPP);
1412
1413                 filename = ll_getname((const char __user *)arg);
1414                 if (IS_ERR(filename))
1415                         RETURN(PTR_ERR(filename));
1416
1417                 namelen = strlen(filename);
1418                 if (namelen < 1)
1419                         GOTO(out_rmdir, rc = -EINVAL);
1420
1421                 rc = ll_rmdir_entry(inode, filename, namelen);
1422 out_rmdir:
1423                 if (filename)
1424                         ll_putname(filename);
1425                 RETURN(rc);
1426         }
1427         case LL_IOC_LOV_SWAP_LAYOUTS:
1428                 RETURN(-EPERM);
1429         case IOC_OBD_STATFS:
1430                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1431         case LL_IOC_LOV_GETSTRIPE:
1432         case LL_IOC_MDC_GETINFO:
1433         case IOC_MDC_GETFILEINFO:
1434         case IOC_MDC_GETFILESTRIPE: {
1435                 struct ptlrpc_request *request = NULL;
1436                 struct lov_user_md __user *lump;
1437                 struct lov_mds_md *lmm = NULL;
1438                 struct mdt_body *body;
1439                 char *filename = NULL;
1440                 int lmmsize;
1441
1442                 if (cmd == IOC_MDC_GETFILEINFO ||
1443                     cmd == IOC_MDC_GETFILESTRIPE) {
1444                         filename = ll_getname((const char __user *)arg);
1445                         if (IS_ERR(filename))
1446                                 RETURN(PTR_ERR(filename));
1447
1448                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1449                                                       &lmmsize, &request);
1450                 } else {
1451                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1452                                               &request, 0);
1453                 }
1454
1455                 if (request) {
1456                         body = req_capsule_server_get(&request->rq_pill,
1457                                                       &RMF_MDT_BODY);
1458                         LASSERT(body != NULL);
1459                 } else {
1460                         GOTO(out_req, rc);
1461                 }
1462
1463                 if (rc < 0) {
1464                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1465                                                cmd == LL_IOC_MDC_GETINFO))
1466                                 GOTO(skip_lmm, rc = 0);
1467                         else
1468                                 GOTO(out_req, rc);
1469                 }
1470
1471                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1472                     cmd == LL_IOC_LOV_GETSTRIPE) {
1473                         lump = (struct lov_user_md __user *)arg;
1474                 } else {
1475                         struct lov_user_mds_data __user *lmdp;
1476                         lmdp = (struct lov_user_mds_data __user *)arg;
1477                         lump = &lmdp->lmd_lmm;
1478                 }
1479                 if (copy_to_user(lump, lmm, lmmsize)) {
1480                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1481                                 GOTO(out_req, rc = -EFAULT);
1482                         rc = -EOVERFLOW;
1483                 }
1484         skip_lmm:
1485                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1486                         struct lov_user_mds_data __user *lmdp;
1487                         lstat_t st = { 0 };
1488
1489                         st.st_dev       = inode->i_sb->s_dev;
1490                         st.st_mode      = body->mbo_mode;
1491                         st.st_nlink     = body->mbo_nlink;
1492                         st.st_uid       = body->mbo_uid;
1493                         st.st_gid       = body->mbo_gid;
1494                         st.st_rdev      = body->mbo_rdev;
1495                         st.st_size      = body->mbo_size;
1496                         st.st_blksize   = PAGE_SIZE;
1497                         st.st_blocks    = body->mbo_blocks;
1498                         st.st_atime     = body->mbo_atime;
1499                         st.st_mtime     = body->mbo_mtime;
1500                         st.st_ctime     = body->mbo_ctime;
1501                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1502                                                 sbi->ll_flags &
1503                                                 LL_SBI_32BIT_API);
1504
1505                         lmdp = (struct lov_user_mds_data __user *)arg;
1506                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1507                                 GOTO(out_req, rc = -EFAULT);
1508                 }
1509
1510                 EXIT;
1511         out_req:
1512                 ptlrpc_req_finished(request);
1513                 if (filename)
1514                         ll_putname(filename);
1515                 return rc;
1516         }
1517         case OBD_IOC_QUOTACTL: {
1518                 struct if_quotactl *qctl;
1519
1520                 OBD_ALLOC_PTR(qctl);
1521                 if (!qctl)
1522                         RETURN(-ENOMEM);
1523
1524                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1525                         GOTO(out_quotactl, rc = -EFAULT);
1526
1527                 rc = quotactl_ioctl(sbi, qctl);
1528
1529                 if (rc == 0 &&
1530                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1531                         rc = -EFAULT;
1532
1533         out_quotactl:
1534                 OBD_FREE_PTR(qctl);
1535                 RETURN(rc);
1536         }
1537         case OBD_IOC_GETDTNAME:
1538         case OBD_IOC_GETMDNAME:
1539                 RETURN(ll_get_obd_name(inode, cmd, arg));
1540         case LL_IOC_FLUSHCTX:
1541                 RETURN(ll_flush_ctx(inode));
1542         case LL_IOC_GETOBDCOUNT: {
1543                 u32 count, vallen;
1544                 struct obd_export *exp;
1545
1546                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1547                         RETURN(-EFAULT);
1548
1549                 /* get ost count when count is zero, get mdt count otherwise */
1550                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1551                 vallen = sizeof(count);
1552                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1553                                   KEY_TGT_COUNT, &vallen, &count);
1554                 if (rc) {
1555                         CERROR("get target count failed: %d\n", rc);
1556                         RETURN(rc);
1557                 }
1558
1559                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1560                         RETURN(-EFAULT);
1561
1562                 RETURN(0);
1563         }
1564         case LL_IOC_PATH2FID:
1565                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1566                                      sizeof(struct lu_fid)))
1567                         RETURN(-EFAULT);
1568                 RETURN(0);
1569         case LL_IOC_GET_CONNECT_FLAGS: {
1570                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1571                                      (void __user *)arg));
1572         }
1573         case OBD_IOC_FID2PATH:
1574                 RETURN(ll_fid2path(inode, (void __user *)arg));
1575         case LL_IOC_GETPARENT:
1576                 RETURN(ll_getparent(file, (void __user *)arg));
1577         case LL_IOC_FID2MDTIDX: {
1578                 struct obd_export *exp = ll_i2mdexp(inode);
1579                 struct lu_fid     fid;
1580                 __u32             index;
1581
1582                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1583                                    sizeof(fid)))
1584                         RETURN(-EFAULT);
1585
1586                 /* Call mdc_iocontrol */
1587                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1588                                    (__u32 __user *)&index);
1589                 if (rc != 0)
1590                         RETURN(rc);
1591
1592                 RETURN(index);
1593         }
1594         case LL_IOC_HSM_REQUEST: {
1595                 struct hsm_user_request *hur;
1596                 ssize_t                  totalsize;
1597
1598                 OBD_ALLOC_PTR(hur);
1599                 if (hur == NULL)
1600                         RETURN(-ENOMEM);
1601
1602                 /* We don't know the true size yet; copy the fixed-size part */
1603                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1604                         OBD_FREE_PTR(hur);
1605                         RETURN(-EFAULT);
1606                 }
1607
1608                 /* Compute the whole struct size */
1609                 totalsize = hur_len(hur);
1610                 OBD_FREE_PTR(hur);
1611                 if (totalsize < 0)
1612                         RETURN(-E2BIG);
1613
1614                 /* Final size will be more than double totalsize */
1615                 if (totalsize >= MDS_MAXREQSIZE / 3)
1616                         RETURN(-E2BIG);
1617
1618                 OBD_ALLOC_LARGE(hur, totalsize);
1619                 if (hur == NULL)
1620                         RETURN(-ENOMEM);
1621
1622                 /* Copy the whole struct */
1623                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1624                         GOTO(out_hur, rc = -EFAULT);
1625
1626                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1627                         const struct lu_fid *fid;
1628                         struct inode *f;
1629                         int i;
1630
1631                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1632                                 fid = &hur->hur_user_item[i].hui_fid;
1633                                 f = search_inode_for_lustre(inode->i_sb, fid);
1634                                 if (IS_ERR(f)) {
1635                                         rc = PTR_ERR(f);
1636                                         break;
1637                                 }
1638
1639                                 rc = ll_hsm_release(f);
1640                                 iput(f);
1641                                 if (rc != 0)
1642                                         break;
1643                         }
1644                 } else {
1645                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1646                                            hur, NULL);
1647                 }
1648
1649 out_hur:
1650                 OBD_FREE_LARGE(hur, totalsize);
1651
1652                 RETURN(rc);
1653         }
1654         case LL_IOC_HSM_PROGRESS: {
1655                 struct hsm_progress_kernel      hpk;
1656                 struct hsm_progress             hp;
1657
1658                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1659                         RETURN(-EFAULT);
1660
1661                 hpk.hpk_fid = hp.hp_fid;
1662                 hpk.hpk_cookie = hp.hp_cookie;
1663                 hpk.hpk_extent = hp.hp_extent;
1664                 hpk.hpk_flags = hp.hp_flags;
1665                 hpk.hpk_errval = hp.hp_errval;
1666                 hpk.hpk_data_version = 0;
1667
1668                 /* File may not exist in Lustre; all progress
1669                  * reported to Lustre root */
1670                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1671                                    NULL);
1672                 RETURN(rc);
1673         }
1674         case LL_IOC_HSM_CT_START:
1675                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1676                         RETURN(-EPERM);
1677
1678                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1679                                     sizeof(struct lustre_kernelcomm));
1680                 RETURN(rc);
1681
1682         case LL_IOC_HSM_COPY_START: {
1683                 struct hsm_copy *copy;
1684                 int              rc;
1685
1686                 OBD_ALLOC_PTR(copy);
1687                 if (copy == NULL)
1688                         RETURN(-ENOMEM);
1689                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1690                         OBD_FREE_PTR(copy);
1691                         RETURN(-EFAULT);
1692                 }
1693
1694                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1695                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1696                         rc = -EFAULT;
1697
1698                 OBD_FREE_PTR(copy);
1699                 RETURN(rc);
1700         }
1701         case LL_IOC_HSM_COPY_END: {
1702                 struct hsm_copy *copy;
1703                 int              rc;
1704
1705                 OBD_ALLOC_PTR(copy);
1706                 if (copy == NULL)
1707                         RETURN(-ENOMEM);
1708                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1709                         OBD_FREE_PTR(copy);
1710                         RETURN(-EFAULT);
1711                 }
1712
1713                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1714                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1715                         rc = -EFAULT;
1716
1717                 OBD_FREE_PTR(copy);
1718                 RETURN(rc);
1719         }
1720         case LL_IOC_MIGRATE: {
1721                 char            *buf = NULL;
1722                 const char      *filename;
1723                 int             namelen = 0;
1724                 int             len;
1725                 int             rc;
1726                 int             mdtidx;
1727
1728                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1729                 if (rc < 0)
1730                         RETURN(rc);
1731
1732                 data = (struct obd_ioctl_data *)buf;
1733                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1734                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1735                         GOTO(migrate_free, rc = -EINVAL);
1736
1737                 filename = data->ioc_inlbuf1;
1738                 namelen = data->ioc_inllen1;
1739                 /* \0 is packed at the end of filename */
1740                 if (namelen < 1 || namelen != strlen(filename) + 1)
1741                         GOTO(migrate_free, rc = -EINVAL);
1742
1743                 if (data->ioc_inllen2 != sizeof(mdtidx))
1744                         GOTO(migrate_free, rc = -EINVAL);
1745                 mdtidx = *(int *)data->ioc_inlbuf2;
1746
1747                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1748 migrate_free:
1749                 OBD_FREE_LARGE(buf, len);
1750
1751                 RETURN(rc);
1752         }
1753         case LL_IOC_FSGETXATTR:
1754                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1755         case LL_IOC_FSSETXATTR:
1756                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1757         default:
1758                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1759                                      (void __user *)arg));
1760         }
1761 }
1762
1763 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1764 {
1765         struct inode *inode = file->f_mapping->host;
1766         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1767         struct ll_sb_info *sbi = ll_i2sbi(inode);
1768         int api32 = ll_need_32bit_api(sbi);
1769         loff_t ret = -EINVAL;
1770         ENTRY;
1771
1772         inode_lock(inode);
1773         switch (origin) {
1774                 case SEEK_SET:
1775                         break;
1776                 case SEEK_CUR:
1777                         offset += file->f_pos;
1778                         break;
1779                 case SEEK_END:
1780                         if (offset > 0)
1781                                 GOTO(out, ret);
1782                         if (api32)
1783                                 offset += LL_DIR_END_OFF_32BIT;
1784                         else
1785                                 offset += LL_DIR_END_OFF;
1786                         break;
1787                 default:
1788                         GOTO(out, ret);
1789         }
1790
1791         if (offset >= 0 &&
1792             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1793              (!api32 && offset <= LL_DIR_END_OFF))) {
1794                 if (offset != file->f_pos) {
1795                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1796                             (!api32 && offset == LL_DIR_END_OFF))
1797                                 fd->lfd_pos = MDS_DIR_END_OFF;
1798                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1799                                 fd->lfd_pos = offset << 32;
1800                         else
1801                                 fd->lfd_pos = offset;
1802                         file->f_pos = offset;
1803                         file->f_version = 0;
1804                 }
1805                 ret = offset;
1806         }
1807         GOTO(out, ret);
1808
1809 out:
1810         inode_unlock(inode);
1811         return ret;
1812 }
1813
1814 static int ll_dir_open(struct inode *inode, struct file *file)
1815 {
1816         ENTRY;
1817         RETURN(ll_file_open(inode, file));
1818 }
1819
1820 static int ll_dir_release(struct inode *inode, struct file *file)
1821 {
1822         ENTRY;
1823         RETURN(ll_file_release(inode, file));
1824 }
1825
1826 const struct file_operations ll_dir_operations = {
1827         .llseek         = ll_dir_seek,
1828         .open           = ll_dir_open,
1829         .release        = ll_dir_release,
1830         .read           = generic_read_dir,
1831 #ifdef HAVE_DIR_CONTEXT
1832         .iterate        = ll_iterate,
1833 #else
1834         .readdir        = ll_readdir,
1835 #endif
1836         .unlocked_ioctl = ll_dir_ioctl,
1837         .fsync          = ll_fsync,
1838 };