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, 2017, 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/security.h>
  42 #include <linux/user_namespace.h>
  43 #ifdef HAVE_UIDGID_HEADER
  44 # include <linux/uidgid.h>
  45 #endif
  46 #include <linux/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48 #include <linux/pagevec.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <uapi/linux/lustre/lustre_ioctl.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <lustre_swab.h>
  60
  61 #include "llite_internal.h"
  62
  63 /*
  64  * (new) readdir implementation overview.
  65  *
  66  * Original lustre readdir implementation cached exact copy of raw directory
  67  * pages on the client. These pages were indexed in client page cache by
  68  * logical offset in the directory file. This design, while very simple and
  69  * intuitive had some inherent problems:
  70  *
  71  *     . it implies that byte offset to the directory entry serves as a
  72  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  73  *     ext3/htree directory entries may move due to splits, and more
  74  *     importantly,
  75  *
  76  *     . it is incompatible with the design of split directories for cmd3,
  77  *     that assumes that names are distributed across nodes based on their
  78  *     hash, and so readdir should be done in hash order.
  79  *
  80  * New readdir implementation does readdir in hash order, and uses hash of a
  81  * file name as a telldir/seekdir cookie. This led to number of complications:
  82  *
  83  *     . hash is not unique, so it cannot be used to index cached directory
  84  *     pages on the client (note, that it requires a whole pageful of hash
  85  *     collided entries to cause two pages to have identical hashes);
  86  *
  87  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  88  *     entry cookie. ext3/htree has the same problem and lustre implementation
  89  *     mimics their solution: seekdir(hash) positions directory at the first
  90  *     entry with the given hash.
  91  *
  92  * Client side.
  93  *
  94  * 0. caching
  95  *
  96  * Client caches directory pages using hash of the first entry as an index. As
  97  * noted above hash is not unique, so this solution doesn't work as is:
  98  * special processing is needed for "page hash chains" (i.e., sequences of
  99  * pages filled with entries all having the same hash value).
 100  *
 101  * First, such chains have to be detected. To this end, server returns to the
 102  * client the hash of the first entry on the page next to one returned. When
 103  * client detects that this hash is the same as hash of the first entry on the
 104  * returned page, page hash collision has to be handled. Pages in the
 105  * hash chain, except first one, are termed "overflow pages".
 106  *
 107  * Solution to index uniqueness problem is to not cache overflow
 108  * pages. Instead, when page hash collision is detected, all overflow pages
 109  * from emerging chain are immediately requested from the server and placed in
 110  * a special data structure (struct ll_dir_chain). This data structure is used
 111  * by ll_readdir() to process entries from overflow pages. When readdir
 112  * invocation finishes, overflow pages are discarded. If page hash collision
 113  * chain weren't completely processed, next call to readdir will again detect
 114  * page hash collision, again read overflow pages in, process next portion of
 115  * entries and again discard the pages. This is not as wasteful as it looks,
 116  * because, given reasonable hash, page hash collisions are extremely rare.
 117  *
 118  * 1. directory positioning
 119  *
 120  * When seekdir(hash) is called, original
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  * Server.
 130  *
 131  * identification of and access to overflow pages
 132  *
 133  * page format
 134  *
 135  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 136  * a header lu_dirpage which describes the start/end hash, and whether this
 137  * page is empty (contains no dir entry) or hash collide with next page.
 138  * After client receives reply, several pages will be integrated into dir page
 139  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 140  * lu_dirpage for this integrated page will be adjusted. See
 141  * mdc_adjust_dirpages().
 142  *
 143  */
 144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 145                              __u64 offset, struct ll_dir_chain *chain)
 146 {
 147         struct md_callback      cb_op;
 148         struct page             *page;
 149         int                     rc;
 150
 151         cb_op.md_blocking_ast = ll_md_blocking_ast;
 152         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 153         if (rc != 0)
 154                 return ERR_PTR(rc);
 155
 156         return page;
 157 }
 158
 159 void ll_release_page(struct inode *inode, struct page *page,
 160                      bool remove)
 161 {
 162         kunmap(page);
 163
 164         /* Always remove the page for striped dir, because the page is
 165          * built from temporarily in LMV layer */
 166         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 167             ll_i2info(inode)->lli_lsm_md != NULL) {
 168                 __free_page(page);
 169                 return;
 170         }
 171
 172         if (remove) {
 173                 lock_page(page);
 174                 if (likely(page->mapping != NULL))
 175                         truncate_complete_page(page->mapping, page);
 176                 unlock_page(page);
 177         }
 178         put_page(page);
 179 }
 180
 181 /**
 182  * return IF_* type for given lu_dirent entry.
 183  * IF_* flag shld be converted to particular OS file type in
 184  * platform llite module.
 185  */
 186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 187 {
 188         u16 type = 0;
 189         struct luda_type *lt;
 190         int len = 0;
 191
 192         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 193                 const unsigned align = sizeof(struct luda_type) - 1;
 194
 195                 len = le16_to_cpu(ent->lde_namelen);
 196                 len = (len + align) & ~align;
 197                 lt = (void *)ent->lde_name + len;
 198                 type = IFTODT(le16_to_cpu(lt->lt_type));
 199         }
 200
 201         return type;
 202 }
 203
 204 #ifdef HAVE_DIR_CONTEXT
 205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 206                 struct dir_context *ctx)
 207 {
 208 #else
 209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 210                 void *cookie, filldir_t filldir)
 211 {
 212 #endif
 213         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 214         __u64                 pos        = *ppos;
 215         bool                  is_api32 = ll_need_32bit_api(sbi);
 216         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 217         struct page          *page;
 218         struct ll_dir_chain   chain;
 219         bool                  done = false;
 220         int                   rc = 0;
 221         ENTRY;
 222
 223         ll_dir_chain_init(&chain);
 224
 225         page = ll_get_dir_page(inode, op_data, pos, &chain);
 226
 227         while (rc == 0 && !done) {
 228                 struct lu_dirpage *dp;
 229                 struct lu_dirent  *ent;
 230                 __u64 hash;
 231                 __u64 next;
 232
 233                 if (IS_ERR(page)) {
 234                         rc = PTR_ERR(page);
 235                         break;
 236                 }
 237
 238                 hash = MDS_DIR_END_OFF;
 239                 dp = page_address(page);
 240                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 241                      ent = lu_dirent_next(ent)) {
 242                         __u16          type;
 243                         int            namelen;
 244                         struct lu_fid  fid;
 245                         __u64          lhash;
 246                         __u64          ino;
 247
 248                         hash = le64_to_cpu(ent->lde_hash);
 249                         if (hash < pos) /* Skip until we find target hash */
 250                                 continue;
 251
 252                         namelen = le16_to_cpu(ent->lde_namelen);
 253                         if (namelen == 0) /* Skip dummy record */
 254                                 continue;
 255
 256                         if (is_api32 && is_hash64)
 257                                 lhash = hash >> 32;
 258                         else
 259                                 lhash = hash;
 260                         fid_le_to_cpu(&fid, &ent->lde_fid);
 261                         ino = cl_fid_build_ino(&fid, is_api32);
 262                         type = ll_dirent_type_get(ent);
 263                         /* For ll_nfs_get_name_filldir(), it will try to access
 264                          * 'ent' through 'lde_name', so the parameter 'name'
 265                          * for 'filldir()' must be part of the 'ent'. */
 266 #ifdef HAVE_DIR_CONTEXT
 267                         ctx->pos = lhash;
 268                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 269                                          type);
 270 #else
 271                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 272                                        ino, type);
 273 #endif
 274                 }
 275
 276                 if (done) {
 277                         pos = hash;
 278                         ll_release_page(inode, page, false);
 279                         break;
 280                 }
 281
 282                 next = le64_to_cpu(dp->ldp_hash_end);
 283                 pos = next;
 284                 if (pos == MDS_DIR_END_OFF) {
 285                         /*
 286                          * End of directory reached.
 287                          */
 288                         done = 1;
 289                         ll_release_page(inode, page, false);
 290                 } else {
 291                         /*
 292                          * Normal case: continue to the next
 293                          * page.
 294                          */
 295                         ll_release_page(inode, page,
 296                                         le32_to_cpu(dp->ldp_flags) &
 297                                         LDF_COLLIDE);
 298                         next = pos;
 299                         page = ll_get_dir_page(inode, op_data, pos,
 300                                                &chain);
 301                 }
 302         }
 303 #ifdef HAVE_DIR_CONTEXT
 304         ctx->pos = pos;
 305 #else
 306         *ppos = pos;
 307 #endif
 308         ll_dir_chain_fini(&chain);
 309         RETURN(rc);
 310 }
 311
 312 #ifdef HAVE_DIR_CONTEXT
 313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 314 #else
 315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 316 #endif
 317 {
 318         struct inode            *inode  = file_inode(filp);
 319         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 320         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 321         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 322         int                     api32   = ll_need_32bit_api(sbi);
 323         struct md_op_data       *op_data;
 324         struct lu_fid           pfid = { 0 };
 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         if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
 345                 /* This is only needed for striped dir to fill ..,
 346                  * see lmv_read_entry */
 347                 if (file_dentry(filp)->d_parent != NULL &&
 348                     file_dentry(filp)->d_parent->d_inode != NULL) {
 349                         __u64 ibits = MDS_INODELOCK_UPDATE;
 350                         struct inode *parent =
 351                                 file_dentry(filp)->d_parent->d_inode;
 352
 353                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 354                                 pfid = *ll_inode2fid(parent);
 355                 }
 356
 357                 /* If it can not find in cache, do lookup .. on the master
 358                  * object */
 359                 if (fid_is_zero(&pfid)) {
 360                         rc = ll_dir_get_parent_fid(inode, &pfid);
 361                         if (rc != 0)
 362                                 RETURN(rc);
 363                 }
 364         }
 365
 366         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 367                                      LUSTRE_OPC_ANY, inode);
 368         if (IS_ERR(op_data))
 369                 GOTO(out, rc = PTR_ERR(op_data));
 370         op_data->op_fid3 = pfid;
 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
 398 out:
 399         if (!rc)
 400                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 401
 402         RETURN(rc);
 403 }
 404
 405 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 406 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 407 {
 408         struct mgs_send_param *msp;
 409         int rc = 0;
 410
 411         OBD_ALLOC_PTR(msp);
 412         if (!msp)
 413                 return -ENOMEM;
 414
 415         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 416         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 417                                 sizeof(struct mgs_send_param), msp, NULL);
 418         if (rc)
 419                 CERROR("Failed to set parameter: %d\n", rc);
 420         OBD_FREE_PTR(msp);
 421
 422         return rc;
 423 }
 424 #endif
 425
 426 /**
 427  * Create striped directory with specified stripe(@lump)
 428  *
 429  * \param[in] dparent   the parent of the directory.
 430  * \param[in] lump      the specified stripes.
 431  * \param[in] dirname   the name of the directory.
 432  * \param[in] mode      the specified mode of the directory.
 433  *
 434  * \retval              =0 if striped directory is being created successfully.
 435  *                      <0 if the creation is failed.
 436  */
 437 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 438                                size_t len, const char *dirname, umode_t mode)
 439 {
 440         struct inode *parent = dparent->d_inode;
 441         struct ptlrpc_request *request = NULL;
 442         struct md_op_data *op_data;
 443         struct ll_sb_info *sbi = ll_i2sbi(parent);
 444         struct inode *inode = NULL;
 445         struct dentry dentry = {
 446                 .d_parent = dparent,
 447                 .d_name = {
 448                         .name = dirname,
 449                         .len = strlen(dirname),
 450                         .hash = ll_full_name_hash(dparent, dirname,
 451                                                   strlen(dirname)),
 452                 },
 453         };
 454         int err;
 455         ENTRY;
 456
 457         if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
 458                      lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
 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             lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
 476                 lustre_swab_lmv_user_md(lump);
 477
 478         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 479                 mode &= ~current_umask();
 480         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 481         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 482                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 483                                      lump);
 484         if (IS_ERR(op_data))
 485                 RETURN(PTR_ERR(op_data));
 486
 487         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 488                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 489                  * to determine the security context for the file. So our fake
 490                  * dentry should be real enough for this purpose. */
 491                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 492                                               &op_data->op_file_secctx_name,
 493                                               &op_data->op_file_secctx,
 494                                               &op_data->op_file_secctx_size);
 495                 if (err < 0)
 496                         GOTO(out_op_data, err);
 497         }
 498
 499         op_data->op_cli_flags |= CLI_SET_MEA;
 500         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 501                         from_kuid(&init_user_ns, current_fsuid()),
 502                         from_kgid(&init_user_ns, current_fsgid()),
 503                         cfs_curproc_cap_pack(), 0, &request);
 504         if (err)
 505                 GOTO(out_request, err);
 506
 507         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 508
 509         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 510         if (err)
 511                 GOTO(out_inode, err);
 512
 513         dentry.d_inode = inode;
 514
 515         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 516                 inode_lock(inode);
 517                 err = security_inode_notifysecctx(inode,
 518                                                   op_data->op_file_secctx,
 519                                                   op_data->op_file_secctx_size);
 520                 inode_unlock(inode);
 521         } else {
 522                 err = ll_inode_init_security(&dentry, inode, parent);
 523         }
 524         if (err)
 525                 GOTO(out_inode, err);
 526
 527 out_inode:
 528         if (inode != NULL)
 529                 iput(inode);
 530 out_request:
 531         ptlrpc_req_finished(request);
 532 out_op_data:
 533         ll_finish_md_op_data(op_data);
 534
 535         return err;
 536 }
 537
 538 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 539                      int set_default)
 540 {
 541         struct ll_sb_info *sbi = ll_i2sbi(inode);
 542         struct md_op_data *op_data;
 543         struct ptlrpc_request *req = NULL;
 544         int rc = 0;
 545 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 546         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 547         struct obd_device *mgc = lsi->lsi_mgc;
 548 #endif
 549         int lum_size;
 550         ENTRY;
 551
 552         if (lump != NULL) {
 553                 /*
 554                  * This is coming from userspace, so should be in
 555                  * local endian.  But the MDS would like it in little
 556                  * endian, so we swab it before we send it.
 557                  */
 558                 switch (lump->lmm_magic) {
 559                 case LOV_USER_MAGIC_V1: {
 560                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 561                                 lustre_swab_lov_user_md_v1(lump);
 562                         lum_size = sizeof(struct lov_user_md_v1);
 563                         break;
 564                 }
 565                 case LOV_USER_MAGIC_V3: {
 566                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 567                                 lustre_swab_lov_user_md_v3(
 568                                         (struct lov_user_md_v3 *)lump);
 569                         lum_size = sizeof(struct lov_user_md_v3);
 570                         break;
 571                 }
 572                 case LOV_USER_MAGIC_COMP_V1: {
 573                         if (lump->lmm_magic !=
 574                             cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
 575                                 lustre_swab_lov_comp_md_v1(
 576                                         (struct lov_comp_md_v1 *)lump);
 577                         lum_size = le32_to_cpu(
 578                                 ((struct lov_comp_md_v1 *)lump)->lcm_size);
 579                         break;
 580                 }
 581                 case LMV_USER_MAGIC: {
 582                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 583                                 lustre_swab_lmv_user_md(
 584                                         (struct lmv_user_md *)lump);
 585                         lum_size = sizeof(struct lmv_user_md);
 586                         break;
 587                 }
 588                 case LOV_USER_MAGIC_SPECIFIC: {
 589                         struct lov_user_md_v3 *v3 =
 590                                         (struct lov_user_md_v3 *)lump;
 591                         if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
 592                                 RETURN(-EINVAL);
 593                         if (lump->lmm_magic !=
 594                             cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
 595                                 lustre_swab_lov_user_md_v3(v3);
 596                                 lustre_swab_lov_user_md_objects(v3->lmm_objects,
 597                                                 v3->lmm_stripe_count);
 598                         }
 599                         lum_size = lov_user_md_size(v3->lmm_stripe_count,
 600                                                     LOV_USER_MAGIC_SPECIFIC);
 601                         break;
 602                 }
 603                 default: {
 604                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 605                                         " %#08x != %#08x nor %#08x\n",
 606                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 607                                         LOV_USER_MAGIC_V3);
 608                         RETURN(-EINVAL);
 609                 }
 610                 }
 611         } else {
 612                 lum_size = sizeof(struct lov_user_md_v1);
 613         }
 614
 615         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 616                                      LUSTRE_OPC_ANY, NULL);
 617         if (IS_ERR(op_data))
 618                 RETURN(PTR_ERR(op_data));
 619
 620         /* swabbing is done in lov_setstripe() on server side */
 621         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 622         ll_finish_md_op_data(op_data);
 623         ptlrpc_req_finished(req);
 624         if (rc)
 625                 RETURN(rc);
 626
 627 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
 628         /*
 629          * 2.9 server has stored filesystem default stripe in ROOT xattr,
 630          * and it's stored into system config for backward compatibility.
 631          *
 632          * In the following we use the fact that LOV_USER_MAGIC_V1 and
 633          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 634          * need the make the distiction between the 2 versions
 635          */
 636         if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
 637             (lump == NULL ||
 638              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
 639              le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
 640                 char *param = NULL;
 641                 char *buf;
 642
 643                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 644                 if (param == NULL)
 645                         GOTO(end, rc = -ENOMEM);
 646
 647                 buf = param;
 648                 /* Get fsname and assume devname to be -MDT0000. */
 649                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 650                 strcat(buf, "-MDT0000.lov");
 651                 buf += strlen(buf);
 652
 653                 /* Set root stripesize */
 654                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
 655                          lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 656                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 657                 if (rc)
 658                         GOTO(end, rc);
 659
 660                 /* Set root stripecount */
 661                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
 662                          lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 663                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 664                 if (rc)
 665                         GOTO(end, rc);
 666
 667                 /* Set root stripeoffset */
 668                 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
 669                          lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 670                                 (typeof(lump->lmm_stripe_offset))(-1));
 671                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 672
 673 end:
 674                 if (param != NULL)
 675                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 676         }
 677 #endif
 678         RETURN(rc);
 679 }
 680
 681 /**
 682  * This function will be used to get default LOV/LMV/Default LMV
 683  * @valid will be used to indicate which stripe it will retrieve
 684  *      OBD_MD_MEA              LMV stripe EA
 685  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 686  *      otherwise               Default LOV EA.
 687  * Each time, it can only retrieve 1 stripe EA
 688  **/
 689 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 690                      struct ptlrpc_request **request, u64 valid)
 691 {
 692         struct ll_sb_info *sbi = ll_i2sbi(inode);
 693         struct mdt_body   *body;
 694         struct lov_mds_md *lmm = NULL;
 695         struct ptlrpc_request *req = NULL;
 696         int rc, lmm_size;
 697         struct md_op_data *op_data;
 698         ENTRY;
 699
 700         rc = ll_get_default_mdsize(sbi, &lmm_size);
 701         if (rc)
 702                 RETURN(rc);
 703
 704         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 705                                      0, lmm_size, LUSTRE_OPC_ANY,
 706                                      NULL);
 707         if (IS_ERR(op_data))
 708                 RETURN(PTR_ERR(op_data));
 709
 710         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 711         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 712         ll_finish_md_op_data(op_data);
 713         if (rc < 0) {
 714                 CDEBUG(D_INFO, "md_getattr failed on inode "
 715                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 716                 GOTO(out, rc);
 717         }
 718
 719         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 720         LASSERT(body != NULL);
 721
 722         lmm_size = body->mbo_eadatasize;
 723
 724         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 725             lmm_size == 0) {
 726                 GOTO(out, rc = -ENODATA);
 727         }
 728
 729         lmm = req_capsule_server_sized_get(&req->rq_pill,
 730                                            &RMF_MDT_MD, lmm_size);
 731         LASSERT(lmm != NULL);
 732
 733         /*
 734          * This is coming from the MDS, so is probably in
 735          * little endian.  We convert it to host endian before
 736          * passing it to userspace.
 737          */
 738         /* We don't swab objects for directories */
 739         switch (le32_to_cpu(lmm->lmm_magic)) {
 740         case LOV_MAGIC_V1:
 741                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 742                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 743                 break;
 744         case LOV_MAGIC_V3:
 745                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 746                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 747                 break;
 748         case LOV_MAGIC_COMP_V1:
 749                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 750                         lustre_swab_lov_comp_md_v1(
 751                                         (struct lov_comp_md_v1 *)lmm);
 752                 break;
 753         case LMV_MAGIC_V1:
 754                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 755                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 756                 break;
 757         case LMV_USER_MAGIC:
 758                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 759                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 760                 break;
 761         case LOV_USER_MAGIC_SPECIFIC: {
 762                 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
 763
 764                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
 765                         lustre_swab_lov_user_md_v3(v3);
 766                         lustre_swab_lov_user_md_objects(v3->lmm_objects,
 767                                                         v3->lmm_stripe_count);
 768                         }
 769                 }
 770                 break;
 771         default:
 772                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 773                 rc = -EPROTO;
 774         }
 775 out:
 776         *plmm = lmm;
 777         *plmm_size = lmm_size;
 778         *request = req;
 779         return rc;
 780 }
 781
 782 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 783 {
 784         struct md_op_data       *op_data;
 785         int                     rc;
 786         int                     mdt_index;
 787         ENTRY;
 788
 789         OBD_ALLOC_PTR(op_data);
 790         if (op_data == NULL)
 791                 RETURN(-ENOMEM);
 792
 793         op_data->op_flags |= MF_GET_MDT_IDX;
 794         op_data->op_fid1 = *fid;
 795         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 796         mdt_index = op_data->op_mds;
 797         OBD_FREE_PTR(op_data);
 798         if (rc < 0)
 799                 RETURN(rc);
 800
 801         RETURN(mdt_index);
 802 }
 803
 804 /*
 805  *  Get MDT index for the inode.
 806  */
 807 int ll_get_mdt_idx(struct inode *inode)
 808 {
 809         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 810 }
 811
 812 /**
 813  * Generic handler to do any pre-copy work.
 814  *
 815  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 816  * first information for it that real work has started.
 817  *
 818  * Moreover, for a ARCHIVE request, it will sample the file data version and
 819  * store it in \a copy.
 820  *
 821  * \return 0 on success.
 822  */
 823 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 824 {
 825         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 826         struct hsm_progress_kernel       hpk;
 827         int                              rc = 0;
 828         int                              rc2;
 829         ENTRY;
 830
 831         /* Forge a hsm_progress based on data from copy. */
 832         hpk.hpk_fid = copy->hc_hai.hai_fid;
 833         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 834         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 835         hpk.hpk_extent.length = 0;
 836         hpk.hpk_flags = 0;
 837         hpk.hpk_errval = 0;
 838         hpk.hpk_data_version = 0;
 839
 840
 841         /* For archive request, we need to read the current file version. */
 842         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 843                 struct inode    *inode;
 844                 __u64            data_version = 0;
 845
 846                 /* Get inode for this fid */
 847                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 848                 if (IS_ERR(inode)) {
 849                         hpk.hpk_flags |= HP_FLAG_RETRY;
 850                         /* hpk_errval is >= 0 */
 851                         hpk.hpk_errval = -PTR_ERR(inode);
 852                         GOTO(progress, rc = PTR_ERR(inode));
 853                 }
 854
 855                 /* Read current file data version */
 856                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 857                 iput(inode);
 858                 if (rc != 0) {
 859                         CDEBUG(D_HSM, "Could not read file data version of "
 860                                       DFID" (rc = %d). Archive request ("
 861                                       "%#llx) could not be done.\n",
 862                                       PFID(&copy->hc_hai.hai_fid), rc,
 863                                       copy->hc_hai.hai_cookie);
 864                         hpk.hpk_flags |= HP_FLAG_RETRY;
 865                         /* hpk_errval must be >= 0 */
 866                         hpk.hpk_errval = -rc;
 867                         GOTO(progress, rc);
 868                 }
 869
 870                 /* Store in the hsm_copy for later copytool use.
 871                  * Always modified even if no lsm. */
 872                 copy->hc_data_version = data_version;
 873         }
 874
 875 progress:
 876         /* On error, the request should be considered as completed */
 877         if (hpk.hpk_errval > 0)
 878                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 879
 880         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 881                             &hpk, NULL);
 882
 883         /* Return first error */
 884         RETURN(rc != 0 ? rc : rc2);
 885 }
 886
 887 /**
 888  * Generic handler to do any post-copy work.
 889  *
 890  * It will send the last hsm_progress update to coordinator to inform it
 891  * that copy is finished and whether it was successful or not.
 892  *
 893  * Moreover,
 894  * - for ARCHIVE request, it will sample the file data version and compare it
 895  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 896  *   will be considered as failed.
 897  * - for RESTORE request, it will sample the file data version and send it to
 898  *   coordinator which is useful if the file was imported as 'released'.
 899  *
 900  * \return 0 on success.
 901  */
 902 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 903 {
 904         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 905         struct hsm_progress_kernel       hpk;
 906         int                              rc = 0;
 907         int                              rc2;
 908         ENTRY;
 909
 910         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 911         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 912          * initialized if copy_end was called with copy == NULL.
 913          */
 914
 915         /* Forge a hsm_progress based on data from copy. */
 916         hpk.hpk_fid = copy->hc_hai.hai_fid;
 917         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 918         hpk.hpk_extent = copy->hc_hai.hai_extent;
 919         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 920         hpk.hpk_errval = copy->hc_errval;
 921         hpk.hpk_data_version = 0;
 922
 923         /* For archive request, we need to check the file data was not changed.
 924          *
 925          * For restore request, we need to send the file data version, this is
 926          * useful when the file was created using hsm_import.
 927          */
 928         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 929              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 930             (copy->hc_errval == 0)) {
 931                 struct inode    *inode;
 932                 __u64            data_version = 0;
 933
 934                 /* Get lsm for this fid */
 935                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 936                 if (IS_ERR(inode)) {
 937                         hpk.hpk_flags |= HP_FLAG_RETRY;
 938                         /* hpk_errval must be >= 0 */
 939                         hpk.hpk_errval = -PTR_ERR(inode);
 940                         GOTO(progress, rc = PTR_ERR(inode));
 941                 }
 942
 943                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 944                 iput(inode);
 945                 if (rc) {
 946                         CDEBUG(D_HSM, "Could not read file data version. "
 947                                       "Request could not be confirmed.\n");
 948                         if (hpk.hpk_errval == 0)
 949                                 hpk.hpk_errval = -rc;
 950                         GOTO(progress, rc);
 951                 }
 952
 953                 /* Store in the hsm_copy for later copytool use.
 954                  * Always modified even if no lsm. */
 955                 hpk.hpk_data_version = data_version;
 956
 957                 /* File could have been stripped during archiving, so we need
 958                  * to check anyway. */
 959                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 960                     (copy->hc_data_version != data_version)) {
 961                         CDEBUG(D_HSM, "File data version mismatched. "
 962                               "File content was changed during archiving. "
 963                                DFID", start:%#llx current:%#llx\n",
 964                                PFID(&copy->hc_hai.hai_fid),
 965                                copy->hc_data_version, data_version);
 966                         /* File was changed, send error to cdt. Do not ask for
 967                          * retry because if a file is modified frequently,
 968                          * the cdt will loop on retried archive requests.
 969                          * The policy engine will ask for a new archive later
 970                          * when the file will not be modified for some tunable
 971                          * time */
 972                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 973                         rc = -EBUSY;
 974                         /* hpk_errval must be >= 0 */
 975                         hpk.hpk_errval = -rc;
 976                         GOTO(progress, rc);
 977                 }
 978
 979         }
 980
 981 progress:
 982         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 983                             &hpk, NULL);
 984
 985         /* Return first error */
 986         RETURN(rc != 0 ? rc : rc2);
 987 }
 988
 989
 990 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 991                           const void __user *data, size_t size)
 992 {
 993         void *copy;
 994         int rc;
 995
 996         OBD_ALLOC(copy, size);
 997         if (copy == NULL)
 998                 return -ENOMEM;
 999
1000         if (copy_from_user(copy, data, size)) {
1001                 rc = -EFAULT;
1002                 goto out;
1003         }
1004
1005         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1006 out:
1007         OBD_FREE(copy, size);
1008
1009         return rc;
1010 }
1011
1012 static int check_owner(int type, int id)
1013 {
1014         switch (type) {
1015         case USRQUOTA:
1016                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1017                         return -EPERM;
1018                 break;
1019         case GRPQUOTA:
1020                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1021                         return -EPERM;
1022                 break;
1023         case PRJQUOTA:
1024                 break;
1025         }
1026         return 0;
1027 }
1028
1029 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1030 {
1031         int cmd = qctl->qc_cmd;
1032         int type = qctl->qc_type;
1033         int id = qctl->qc_id;
1034         int valid = qctl->qc_valid;
1035         int rc = 0;
1036         ENTRY;
1037
1038         switch (cmd) {
1039         case Q_SETQUOTA:
1040         case Q_SETINFO:
1041         case LUSTRE_Q_SETDEFAULT:
1042                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1043                         RETURN(-EPERM);
1044                 break;
1045         case Q_GETQUOTA:
1046         case LUSTRE_Q_GETDEFAULT:
1047                 if (check_owner(type, id) &&
1048                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1049                         RETURN(-EPERM);
1050                 break;
1051         case Q_GETINFO:
1052                 break;
1053         default:
1054                 CERROR("unsupported quotactl op: %#x\n", cmd);
1055                 RETURN(-ENOTSUPP);
1056         }
1057
1058         if (valid != QC_GENERAL) {
1059                 if (cmd == Q_GETINFO)
1060                         qctl->qc_cmd = Q_GETOINFO;
1061                 else if (cmd == Q_GETQUOTA)
1062                         qctl->qc_cmd = Q_GETOQUOTA;
1063                 else
1064                         RETURN(-EINVAL);
1065
1066                 switch (valid) {
1067                 case QC_MDTIDX:
1068                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1069                                            sizeof(*qctl), qctl, NULL);
1070                         break;
1071                 case QC_OSTIDX:
1072                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1073                                            sizeof(*qctl), qctl, NULL);
1074                         break;
1075                 case QC_UUID:
1076                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1077                                            sizeof(*qctl), qctl, NULL);
1078                         if (rc == -EAGAIN)
1079                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1080                                                    sbi->ll_dt_exp,
1081                                                    sizeof(*qctl), qctl, NULL);
1082                         break;
1083                 default:
1084                         rc = -EINVAL;
1085                         break;
1086                 }
1087
1088                 if (rc)
1089                         RETURN(rc);
1090
1091                 qctl->qc_cmd = cmd;
1092         } else {
1093                 struct obd_quotactl *oqctl;
1094
1095                 OBD_ALLOC_PTR(oqctl);
1096                 if (oqctl == NULL)
1097                         RETURN(-ENOMEM);
1098
1099                 QCTL_COPY(oqctl, qctl);
1100                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1101                 if (rc) {
1102                         OBD_FREE_PTR(oqctl);
1103                         RETURN(rc);
1104                 }
1105                 /* If QIF_SPACE is not set, client should collect the
1106                  * space usage from OSSs by itself */
1107                 if (cmd == Q_GETQUOTA &&
1108                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1109                     !oqctl->qc_dqblk.dqb_curspace) {
1110                         struct obd_quotactl *oqctl_tmp;
1111
1112                         OBD_ALLOC_PTR(oqctl_tmp);
1113                         if (oqctl_tmp == NULL)
1114                                 GOTO(out, rc = -ENOMEM);
1115
1116                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1117                         oqctl_tmp->qc_id = oqctl->qc_id;
1118                         oqctl_tmp->qc_type = oqctl->qc_type;
1119
1120                         /* collect space usage from OSTs */
1121                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1122                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1123                         if (!rc || rc == -EREMOTEIO) {
1124                                 oqctl->qc_dqblk.dqb_curspace =
1125                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1126                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1127                         }
1128
1129                         /* collect space & inode usage from MDTs */
1130                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1131                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1132                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1133                         if (!rc || rc == -EREMOTEIO) {
1134                                 oqctl->qc_dqblk.dqb_curspace +=
1135                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1136                                 oqctl->qc_dqblk.dqb_curinodes =
1137                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1138                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1139                         } else {
1140                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1141                         }
1142
1143                         OBD_FREE_PTR(oqctl_tmp);
1144                 }
1145 out:
1146                 QCTL_COPY(qctl, oqctl);
1147                 OBD_FREE_PTR(oqctl);
1148         }
1149
1150         RETURN(rc);
1151 }
1152
1153 /* This function tries to get a single name component,
1154  * to send to the server. No actual path traversal involved,
1155  * so we limit to NAME_MAX */
1156 static char *ll_getname(const char __user *filename)
1157 {
1158         int ret = 0, len;
1159         char *tmp;
1160
1161         OBD_ALLOC(tmp, NAME_MAX + 1);
1162
1163         if (!tmp)
1164                 return ERR_PTR(-ENOMEM);
1165
1166         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1167         if (len < 0)
1168                 ret = -ENOENT;
1169         else if (len > NAME_MAX)
1170                 ret = -ENAMETOOLONG;
1171
1172         if (ret) {
1173                 OBD_FREE(tmp, NAME_MAX + 1);
1174                 tmp =  ERR_PTR(ret);
1175         }
1176         return tmp;
1177 }
1178
1179 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1180
1181 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1182 {
1183         struct dentry *dentry = file_dentry(file);
1184         struct inode *inode = file_inode(file);
1185         struct ll_sb_info *sbi = ll_i2sbi(inode);
1186         struct obd_ioctl_data *data;
1187         int rc = 0;
1188         ENTRY;
1189
1190         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1191                PFID(ll_inode2fid(inode)), inode, cmd);
1192
1193         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1194         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1195                 return -ENOTTY;
1196
1197         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1198         switch (cmd) {
1199         case FS_IOC_GETFLAGS:
1200         case FS_IOC_SETFLAGS:
1201                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1202         case FSFILT_IOC_GETVERSION:
1203         case FS_IOC_GETVERSION:
1204                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1205         /* We need to special case any other ioctls we want to handle,
1206          * to send them to the MDS/OST as appropriate and to properly
1207          * network encode the arg field. */
1208         case FS_IOC_SETVERSION:
1209                 RETURN(-ENOTSUPP);
1210
1211         case LL_IOC_GET_MDTIDX: {
1212                 int mdtidx;
1213
1214                 mdtidx = ll_get_mdt_idx(inode);
1215                 if (mdtidx < 0)
1216                         RETURN(mdtidx);
1217
1218                 if (put_user((int)mdtidx, (int __user *)arg))
1219                         RETURN(-EFAULT);
1220
1221                 return 0;
1222         }
1223         case IOC_MDC_LOOKUP: {
1224                                      int namelen, len = 0;
1225                 char *buf = NULL;
1226                 char *filename;
1227
1228                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1229                 if (rc != 0)
1230                         RETURN(rc);
1231                 data = (void *)buf;
1232
1233                 filename = data->ioc_inlbuf1;
1234                 namelen = strlen(filename);
1235                 if (namelen < 1) {
1236                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1237                         GOTO(out_free, rc = -EINVAL);
1238                 }
1239
1240                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1241                 if (rc < 0) {
1242                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1243                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1244                                filename, rc);
1245                         GOTO(out_free, rc);
1246                 }
1247 out_free:
1248                 OBD_FREE_LARGE(buf, len);
1249                 return rc;
1250         }
1251         case LL_IOC_LMV_SETSTRIPE: {
1252                 struct lmv_user_md  *lum;
1253                 char            *buf = NULL;
1254                 char            *filename;
1255                 int              namelen = 0;
1256                 int              lumlen = 0;
1257                 umode_t          mode;
1258                 int              len;
1259                 int              rc;
1260
1261                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1262                 if (rc)
1263                         RETURN(rc);
1264
1265                 data = (void *)buf;
1266                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1267                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1268                         GOTO(lmv_out_free, rc = -EINVAL);
1269
1270                 filename = data->ioc_inlbuf1;
1271                 namelen = data->ioc_inllen1;
1272
1273                 if (namelen < 1) {
1274                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1275                         GOTO(lmv_out_free, rc = -EINVAL);
1276                 }
1277                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1278                 lumlen = data->ioc_inllen2;
1279
1280                 if ((lum->lum_magic != LMV_USER_MAGIC &&
1281                      lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
1282                     lumlen < sizeof(*lum)) {
1283                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1284                                filename, lum->lum_magic, lumlen, -EFAULT);
1285                         GOTO(lmv_out_free, rc = -EINVAL);
1286                 }
1287
1288 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1289                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1290 #else
1291                 mode = data->ioc_type;
1292 #endif
1293                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1294 lmv_out_free:
1295                 OBD_FREE_LARGE(buf, len);
1296                 RETURN(rc);
1297
1298         }
1299         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1300                 struct lmv_user_md        lum;
1301                 struct lmv_user_md __user *ulump =
1302                                         (struct lmv_user_md __user *)arg;
1303                 int                       rc;
1304
1305                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1306                         RETURN(-EFAULT);
1307
1308                 if (lum.lum_magic != LMV_USER_MAGIC)
1309                         RETURN(-EINVAL);
1310
1311                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1312
1313                 RETURN(rc);
1314         }
1315         case LL_IOC_LOV_SETSTRIPE_NEW:
1316         case LL_IOC_LOV_SETSTRIPE: {
1317                 struct lov_user_md_v3 *lumv3 = NULL;
1318                 struct lov_user_md_v1 lumv1;
1319                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1320                 struct lov_user_md_v1 __user *lumv1p =
1321                         (struct lov_user_md_v1 __user *)arg;
1322                 struct lov_user_md_v3 __user *lumv3p =
1323                         (struct lov_user_md_v3 __user *)arg;
1324                 int lum_size;
1325
1326                 int set_default = 0;
1327
1328                 CLASSERT(sizeof(struct lov_user_md_v3) >
1329                          sizeof(struct lov_comp_md_v1));
1330                 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1331                 /* first try with v1 which is smaller than v3 */
1332                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1333                         RETURN(-EFAULT);
1334
1335                 if (inode->i_sb->s_root == file_dentry(file))
1336                         set_default = 1;
1337
1338                 switch (lumv1.lmm_magic) {
1339                 case LOV_USER_MAGIC_V3:
1340                 case LOV_USER_MAGIC_SPECIFIC:
1341                         lum_size = ll_lov_user_md_size(&lumv1);
1342                         if (lum_size < 0)
1343                                 RETURN(lum_size);
1344                         OBD_ALLOC(lumv3, lum_size);
1345                         if (!lumv3)
1346                                 RETURN(-ENOMEM);
1347                         if (copy_from_user(lumv3, lumv3p, lum_size))
1348                                 GOTO(out, rc = -EFAULT);
1349                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1350                         break;
1351                 case LOV_USER_MAGIC_V1:
1352                         break;
1353                 default:
1354                         GOTO(out, rc = -ENOTSUPP);
1355                 }
1356
1357                 /* in v1 and v3 cases lumv1 points to data */
1358                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1359 out:
1360                 if (lumv3)
1361                         OBD_FREE(lumv3, lum_size);
1362                 RETURN(rc);
1363         }
1364         case LL_IOC_LMV_GETSTRIPE: {
1365                 struct lmv_user_md __user *ulmv =
1366                                         (struct lmv_user_md __user *)arg;
1367                 struct lmv_user_md      lum;
1368                 struct ptlrpc_request   *request = NULL;
1369                 union lmv_mds_md        *lmm = NULL;
1370                 int                     lmmsize;
1371                 u64                     valid = 0;
1372                 struct lmv_user_md      *tmp = NULL;
1373                 int                     mdt_index;
1374                 int                     lum_size;
1375                 int                     stripe_count;
1376                 int                     max_stripe_count;
1377                 int                     i;
1378                 int                     rc;
1379
1380                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1381                         RETURN(-EFAULT);
1382
1383                 max_stripe_count = lum.lum_stripe_count;
1384                 /* lum_magic will indicate which stripe the ioctl will like
1385                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1386                  * is for default LMV stripe */
1387                 if (lum.lum_magic == LMV_MAGIC_V1)
1388                         valid |= OBD_MD_MEA;
1389                 else if (lum.lum_magic == LMV_USER_MAGIC)
1390                         valid |= OBD_MD_DEFAULT_MEA;
1391                 else
1392                         RETURN(-EINVAL);
1393
1394                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1395                                       valid);
1396                 if (rc != 0)
1397                         GOTO(finish_req, rc);
1398
1399                 /* Get default LMV EA */
1400                 if (lum.lum_magic == LMV_USER_MAGIC) {
1401                         if (lmmsize > sizeof(*ulmv))
1402                                 GOTO(finish_req, rc = -EINVAL);
1403
1404                         if (copy_to_user(ulmv, lmm, lmmsize))
1405                                 GOTO(finish_req, rc = -EFAULT);
1406
1407                         GOTO(finish_req, rc);
1408                 }
1409
1410                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1411                 if (max_stripe_count < stripe_count) {
1412                         lum.lum_stripe_count = stripe_count;
1413                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1414                                 GOTO(finish_req, rc = -EFAULT);
1415                         GOTO(finish_req, rc = -E2BIG);
1416                 }
1417
1418                 lum_size = lmv_user_md_size(stripe_count,
1419                                             LMV_USER_MAGIC_SPECIFIC);
1420                 OBD_ALLOC(tmp, lum_size);
1421                 if (tmp == NULL)
1422                         GOTO(finish_req, rc = -ENOMEM);
1423
1424                 mdt_index = ll_get_mdt_idx(inode);
1425                 if (mdt_index < 0)
1426                         GOTO(out_tmp, rc = -ENOMEM);
1427
1428                 tmp->lum_magic = LMV_MAGIC_V1;
1429                 tmp->lum_stripe_count = 0;
1430                 tmp->lum_stripe_offset = mdt_index;
1431                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1432                 for (i = 0; i < stripe_count; i++) {
1433                         struct lu_fid   fid;
1434
1435                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1436                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1437                         if (mdt_index < 0)
1438                                 GOTO(out_tmp, rc = mdt_index);
1439
1440                         tmp->lum_objects[i].lum_mds = mdt_index;
1441                         tmp->lum_objects[i].lum_fid = fid;
1442                         tmp->lum_stripe_count++;
1443                 }
1444
1445                 if (copy_to_user(ulmv, tmp, lum_size))
1446                         GOTO(out_tmp, rc = -EFAULT);
1447 out_tmp:
1448                 OBD_FREE(tmp, lum_size);
1449 finish_req:
1450                 ptlrpc_req_finished(request);
1451                 return rc;
1452         }
1453
1454         case LL_IOC_REMOVE_ENTRY: {
1455                 char            *filename = NULL;
1456                 int              namelen = 0;
1457                 int              rc;
1458
1459                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1460                  * unsupported server, which might crash the server(LU-2730),
1461                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1462                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1463                  * server will support REINT_RMENTRY XXX*/
1464                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1465                         RETURN(-ENOTSUPP);
1466
1467                 filename = ll_getname((const char __user *)arg);
1468                 if (IS_ERR(filename))
1469                         RETURN(PTR_ERR(filename));
1470
1471                 namelen = strlen(filename);
1472                 if (namelen < 1)
1473                         GOTO(out_rmdir, rc = -EINVAL);
1474
1475                 rc = ll_rmdir_entry(inode, filename, namelen);
1476 out_rmdir:
1477                 if (filename)
1478                         ll_putname(filename);
1479                 RETURN(rc);
1480         }
1481         case LL_IOC_LOV_SWAP_LAYOUTS:
1482                 RETURN(-EPERM);
1483         case IOC_OBD_STATFS:
1484                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1485         case LL_IOC_LOV_GETSTRIPE:
1486         case LL_IOC_LOV_GETSTRIPE_NEW:
1487         case LL_IOC_MDC_GETINFO:
1488         case IOC_MDC_GETFILEINFO:
1489         case IOC_MDC_GETFILESTRIPE: {
1490                 struct ptlrpc_request *request = NULL;
1491                 struct lov_user_md __user *lump;
1492                 struct lov_mds_md *lmm = NULL;
1493                 struct mdt_body *body;
1494                 char *filename = NULL;
1495                 int lmmsize;
1496
1497                 if (cmd == IOC_MDC_GETFILEINFO ||
1498                     cmd == IOC_MDC_GETFILESTRIPE) {
1499                         filename = ll_getname((const char __user *)arg);
1500                         if (IS_ERR(filename))
1501                                 RETURN(PTR_ERR(filename));
1502
1503                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1504                                                       &lmmsize, &request);
1505                 } else {
1506                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1507                                               &request, 0);
1508                 }
1509
1510                 if (request) {
1511                         body = req_capsule_server_get(&request->rq_pill,
1512                                                       &RMF_MDT_BODY);
1513                         LASSERT(body != NULL);
1514                 } else {
1515                         GOTO(out_req, rc);
1516                 }
1517
1518                 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1519                                        cmd == LL_IOC_MDC_GETINFO)) {
1520                         lmmsize = 0;
1521                         rc = 0;
1522                 }
1523
1524                 if (rc < 0)
1525                         GOTO(out_req, rc);
1526
1527                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1528                     cmd == LL_IOC_LOV_GETSTRIPE ||
1529                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1530                         lump = (struct lov_user_md __user *)arg;
1531                 } else {
1532                         struct lov_user_mds_data __user *lmdp;
1533                         lmdp = (struct lov_user_mds_data __user *)arg;
1534                         lump = &lmdp->lmd_lmm;
1535                 }
1536
1537                 if (lmmsize == 0) {
1538                         /* If the file has no striping then zero out *lump so
1539                          * that the caller isn't confused by garbage. */
1540                         if (clear_user(lump, sizeof(*lump)))
1541                                 GOTO(out_req, rc = -EFAULT);
1542                 } else if (copy_to_user(lump, lmm, lmmsize)) {
1543                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1544                                 GOTO(out_req, rc = -EFAULT);
1545                         rc = -EOVERFLOW;
1546                 }
1547
1548                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1549                         struct lov_user_mds_data __user *lmdp;
1550                         lstat_t st = { 0 };
1551
1552                         st.st_dev       = inode->i_sb->s_dev;
1553                         st.st_mode      = body->mbo_mode;
1554                         st.st_nlink     = body->mbo_nlink;
1555                         st.st_uid       = body->mbo_uid;
1556                         st.st_gid       = body->mbo_gid;
1557                         st.st_rdev      = body->mbo_rdev;
1558                         st.st_size      = body->mbo_size;
1559                         st.st_blksize   = PAGE_SIZE;
1560                         st.st_blocks    = body->mbo_blocks;
1561                         st.st_atime     = body->mbo_atime;
1562                         st.st_mtime     = body->mbo_mtime;
1563                         st.st_ctime     = body->mbo_ctime;
1564                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1565                                                 sbi->ll_flags &
1566                                                 LL_SBI_32BIT_API);
1567
1568                         lmdp = (struct lov_user_mds_data __user *)arg;
1569                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1570                                 GOTO(out_req, rc = -EFAULT);
1571                 }
1572
1573                 EXIT;
1574         out_req:
1575                 ptlrpc_req_finished(request);
1576                 if (filename)
1577                         ll_putname(filename);
1578                 return rc;
1579         }
1580         case OBD_IOC_QUOTACTL: {
1581                 struct if_quotactl *qctl;
1582
1583                 OBD_ALLOC_PTR(qctl);
1584                 if (!qctl)
1585                         RETURN(-ENOMEM);
1586
1587                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1588                         GOTO(out_quotactl, rc = -EFAULT);
1589
1590                 rc = quotactl_ioctl(sbi, qctl);
1591
1592                 if (rc == 0 &&
1593                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1594                         rc = -EFAULT;
1595
1596         out_quotactl:
1597                 OBD_FREE_PTR(qctl);
1598                 RETURN(rc);
1599         }
1600         case OBD_IOC_GETDTNAME:
1601         case OBD_IOC_GETMDNAME:
1602                 RETURN(ll_get_obd_name(inode, cmd, arg));
1603         case LL_IOC_FLUSHCTX:
1604                 RETURN(ll_flush_ctx(inode));
1605         case LL_IOC_GETOBDCOUNT: {
1606                 u32 count, vallen;
1607                 struct obd_export *exp;
1608
1609                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1610                         RETURN(-EFAULT);
1611
1612                 /* get ost count when count is zero, get mdt count otherwise */
1613                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1614                 vallen = sizeof(count);
1615                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1616                                   KEY_TGT_COUNT, &vallen, &count);
1617                 if (rc) {
1618                         CERROR("get target count failed: %d\n", rc);
1619                         RETURN(rc);
1620                 }
1621
1622                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1623                         RETURN(-EFAULT);
1624
1625                 RETURN(0);
1626         }
1627         case LL_IOC_PATH2FID:
1628                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1629                                      sizeof(struct lu_fid)))
1630                         RETURN(-EFAULT);
1631                 RETURN(0);
1632         case LL_IOC_GET_CONNECT_FLAGS: {
1633                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1634                                      (void __user *)arg));
1635         }
1636         case OBD_IOC_FID2PATH:
1637                 RETURN(ll_fid2path(inode, (void __user *)arg));
1638         case LL_IOC_GETPARENT:
1639                 RETURN(ll_getparent(file, (void __user *)arg));
1640         case LL_IOC_FID2MDTIDX: {
1641                 struct obd_export *exp = ll_i2mdexp(inode);
1642                 struct lu_fid     fid;
1643                 __u32             index;
1644
1645                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1646                                    sizeof(fid)))
1647                         RETURN(-EFAULT);
1648
1649                 /* Call mdc_iocontrol */
1650                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1651                                    (__u32 __user *)&index);
1652                 if (rc != 0)
1653                         RETURN(rc);
1654
1655                 RETURN(index);
1656         }
1657         case LL_IOC_HSM_REQUEST: {
1658                 struct hsm_user_request *hur;
1659                 ssize_t                  totalsize;
1660
1661                 OBD_ALLOC_PTR(hur);
1662                 if (hur == NULL)
1663                         RETURN(-ENOMEM);
1664
1665                 /* We don't know the true size yet; copy the fixed-size part */
1666                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1667                         OBD_FREE_PTR(hur);
1668                         RETURN(-EFAULT);
1669                 }
1670
1671                 /* Compute the whole struct size */
1672                 totalsize = hur_len(hur);
1673                 OBD_FREE_PTR(hur);
1674                 if (totalsize < 0)
1675                         RETURN(-E2BIG);
1676
1677                 /* Final size will be more than double totalsize */
1678                 if (totalsize >= MDS_MAXREQSIZE / 3)
1679                         RETURN(-E2BIG);
1680
1681                 OBD_ALLOC_LARGE(hur, totalsize);
1682                 if (hur == NULL)
1683                         RETURN(-ENOMEM);
1684
1685                 /* Copy the whole struct */
1686                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1687                         GOTO(out_hur, rc = -EFAULT);
1688
1689                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1690                         const struct lu_fid *fid;
1691                         struct inode *f;
1692                         int i;
1693
1694                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1695                                 fid = &hur->hur_user_item[i].hui_fid;
1696                                 f = search_inode_for_lustre(inode->i_sb, fid);
1697                                 if (IS_ERR(f)) {
1698                                         rc = PTR_ERR(f);
1699                                         break;
1700                                 }
1701
1702                                 rc = ll_hsm_release(f);
1703                                 iput(f);
1704                                 if (rc != 0)
1705                                         break;
1706                         }
1707                 } else {
1708                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1709                                            hur, NULL);
1710                 }
1711
1712 out_hur:
1713                 OBD_FREE_LARGE(hur, totalsize);
1714
1715                 RETURN(rc);
1716         }
1717         case LL_IOC_HSM_PROGRESS: {
1718                 struct hsm_progress_kernel      hpk;
1719                 struct hsm_progress             hp;
1720
1721                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1722                         RETURN(-EFAULT);
1723
1724                 hpk.hpk_fid = hp.hp_fid;
1725                 hpk.hpk_cookie = hp.hp_cookie;
1726                 hpk.hpk_extent = hp.hp_extent;
1727                 hpk.hpk_flags = hp.hp_flags;
1728                 hpk.hpk_errval = hp.hp_errval;
1729                 hpk.hpk_data_version = 0;
1730
1731                 /* File may not exist in Lustre; all progress
1732                  * reported to Lustre root */
1733                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1734                                    NULL);
1735                 RETURN(rc);
1736         }
1737         case LL_IOC_HSM_CT_START:
1738                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1739                         RETURN(-EPERM);
1740
1741                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1742                                     sizeof(struct lustre_kernelcomm));
1743                 RETURN(rc);
1744
1745         case LL_IOC_HSM_COPY_START: {
1746                 struct hsm_copy *copy;
1747                 int              rc;
1748
1749                 OBD_ALLOC_PTR(copy);
1750                 if (copy == NULL)
1751                         RETURN(-ENOMEM);
1752                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1753                         OBD_FREE_PTR(copy);
1754                         RETURN(-EFAULT);
1755                 }
1756
1757                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1758                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1759                         rc = -EFAULT;
1760
1761                 OBD_FREE_PTR(copy);
1762                 RETURN(rc);
1763         }
1764         case LL_IOC_HSM_COPY_END: {
1765                 struct hsm_copy *copy;
1766                 int              rc;
1767
1768                 OBD_ALLOC_PTR(copy);
1769                 if (copy == NULL)
1770                         RETURN(-ENOMEM);
1771                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1772                         OBD_FREE_PTR(copy);
1773                         RETURN(-EFAULT);
1774                 }
1775
1776                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1777                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1778                         rc = -EFAULT;
1779
1780                 OBD_FREE_PTR(copy);
1781                 RETURN(rc);
1782         }
1783         case LL_IOC_MIGRATE: {
1784                 struct lmv_user_md *lum;
1785                 char *buf = NULL;
1786                 int len;
1787                 char *filename;
1788                 int namelen = 0;
1789                 int rc;
1790
1791                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1792                 if (rc)
1793                         RETURN(rc);
1794
1795                 data = (struct obd_ioctl_data *)buf;
1796                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1797                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1798                         GOTO(migrate_free, rc = -EINVAL);
1799
1800                 filename = data->ioc_inlbuf1;
1801                 namelen = data->ioc_inllen1;
1802
1803                 if (namelen < 1 || namelen != strlen(filename) + 1) {
1804                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1805                         GOTO(migrate_free, rc = -EINVAL);
1806                 }
1807
1808                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1809                 if (lum->lum_magic != LMV_USER_MAGIC &&
1810                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1811                         rc = -EINVAL;
1812                         CERROR("%s: wrong lum magic %x: rc = %d\n",
1813                                filename, lum->lum_magic, rc);
1814                         GOTO(migrate_free, rc);
1815                 }
1816
1817                 rc = ll_migrate(inode, file, lum, filename);
1818 migrate_free:
1819                 OBD_FREE_LARGE(buf, len);
1820
1821                 RETURN(rc);
1822         }
1823         case LL_IOC_FSGETXATTR:
1824                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1825         case LL_IOC_FSSETXATTR:
1826                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1827         default:
1828                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1829                                      (void __user *)arg));
1830         }
1831 }
1832
1833 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1834 {
1835         struct inode *inode = file->f_mapping->host;
1836         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1837         struct ll_sb_info *sbi = ll_i2sbi(inode);
1838         int api32 = ll_need_32bit_api(sbi);
1839         loff_t ret = -EINVAL;
1840         ENTRY;
1841
1842         inode_lock(inode);
1843         switch (origin) {
1844                 case SEEK_SET:
1845                         break;
1846                 case SEEK_CUR:
1847                         offset += file->f_pos;
1848                         break;
1849                 case SEEK_END:
1850                         if (offset > 0)
1851                                 GOTO(out, ret);
1852                         if (api32)
1853                                 offset += LL_DIR_END_OFF_32BIT;
1854                         else
1855                                 offset += LL_DIR_END_OFF;
1856                         break;
1857                 default:
1858                         GOTO(out, ret);
1859         }
1860
1861         if (offset >= 0 &&
1862             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1863              (!api32 && offset <= LL_DIR_END_OFF))) {
1864                 if (offset != file->f_pos) {
1865                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1866                             (!api32 && offset == LL_DIR_END_OFF))
1867                                 fd->lfd_pos = MDS_DIR_END_OFF;
1868                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1869                                 fd->lfd_pos = offset << 32;
1870                         else
1871                                 fd->lfd_pos = offset;
1872                         file->f_pos = offset;
1873                         file->f_version = 0;
1874                 }
1875                 ret = offset;
1876         }
1877         GOTO(out, ret);
1878
1879 out:
1880         inode_unlock(inode);
1881         return ret;
1882 }
1883
1884 static int ll_dir_open(struct inode *inode, struct file *file)
1885 {
1886         ENTRY;
1887         RETURN(ll_file_open(inode, file));
1888 }
1889
1890 static int ll_dir_release(struct inode *inode, struct file *file)
1891 {
1892         ENTRY;
1893         RETURN(ll_file_release(inode, file));
1894 }
1895
1896 const struct file_operations ll_dir_operations = {
1897         .llseek         = ll_dir_seek,
1898         .open           = ll_dir_open,
1899         .release        = ll_dir_release,
1900         .read           = generic_read_dir,
1901 #ifdef HAVE_DIR_CONTEXT
1902         .iterate        = ll_iterate,
1903 #else
1904         .readdir        = ll_readdir,
1905 #endif
1906         .unlocked_ioctl = ll_dir_ioctl,
1907         .fsync          = ll_fsync,
1908 };