lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <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 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_lib.h>
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted. See
 139  * lmv_adjust_dirpages().
 140  *
 141  */
 142 /**
 143  * The following three APIs will be used by llite to iterate directory
 144  * entries from MDC dir page caches.
 145  *
 146  * ll_dir_entry_start(next) will lookup(return) entry by op_hash_offset.
 147  * To avoid extra memory allocation, the @entry will be pointed to
 148  * the dir entries in MDC page directly, so these pages can not be released
 149  * until the entry has been accessed in ll_readdir(or statahead).
 150  *
 151  * The iterate process will be
 152  *
 153  * ll_dir_entry_start: locate the page in MDC, and return the first entry.
 154  *                     hold the page.
 155  *
 156  * ll_dir_entry_next: return the next entry in the current page, if it reaches
 157  *                    to the end, release current page.
 158  *
 159  * ll_dir_entry_end: release the last page.
 160  **/
 161 struct lu_dirent *ll_dir_entry_start(struct inode *dir,
 162                                      struct md_op_data *op_data,
 163                                      struct page **ppage)
 164 {
 165         struct lu_dirent *entry;
 166         struct md_callback cb_op;
 167         int rc;
 168
 169         LASSERT(*ppage == NULL);
 170         cb_op.md_blocking_ast = ll_md_blocking_ast;
 171         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 172         if (rc != 0)
 173                 entry = ERR_PTR(rc);
 174         return entry;
 175 }
 176
 177 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
 178                                     struct md_op_data *op_data,
 179                                     struct lu_dirent *ent,
 180                                     struct page **ppage)
 181 {
 182         struct lu_dirent *entry;
 183         struct md_callback cb_op;
 184         int rc;
 185
 186         LASSERT(*ppage != NULL);
 187         cb_op.md_blocking_ast = ll_md_blocking_ast;
 188         op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
 189         kunmap(*ppage);
 190         page_cache_release(*ppage);
 191         *ppage = NULL;
 192         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 193         if (rc != 0)
 194                 entry = ERR_PTR(rc);
 195         return entry;
 196 }
 197
 198 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 199                 void *cookie, filldir_t filldir)
 200 {
 201         struct ll_sb_info       *sbi = ll_i2sbi(inode);
 202         struct ll_dir_chain     chain;
 203         struct lu_dirent        *ent;
 204         int                     api32 = ll_need_32bit_api(sbi);
 205         int                     hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 206         int                     done = 0;
 207         int                     rc = 0;
 208         __u64                   hash = MDS_DIR_END_OFF;
 209         __u64                   last_hash = MDS_DIR_END_OFF;
 210         struct page             *page = NULL;
 211         ENTRY;
 212
 213         ll_dir_chain_init(&chain);
 214         for (ent = ll_dir_entry_start(inode, op_data, &page);
 215              ent != NULL && !IS_ERR(ent) && !done;
 216              ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
 217                 __u16          type;
 218                 int            namelen;
 219                 struct lu_fid  fid;
 220                 __u64          lhash;
 221                 __u64          ino;
 222
 223                 hash = le64_to_cpu(ent->lde_hash);
 224                 if (hash < op_data->op_hash_offset)
 225                         /*
 226                          * Skip until we find target hash
 227                          * value.
 228                          */
 229                         continue;
 230                 namelen = le16_to_cpu(ent->lde_namelen);
 231                 if (namelen == 0)
 232                         /*
 233                          * Skip dummy record.
 234                          */
 235                         continue;
 236
 237                 if (api32 && hash64)
 238                         lhash = hash >> 32;
 239                 else
 240                         lhash = hash;
 241                 fid_le_to_cpu(&fid, &ent->lde_fid);
 242                 ino = cl_fid_build_ino(&fid, api32);
 243                 type = ll_dirent_type_get(ent);
 244                 /* For 'll_nfs_get_name_filldir()', it will try
 245                  * to access the 'ent' through its 'lde_name',
 246                  * so the parameter 'name' for 'filldir()' must
 247                  * be part of the 'ent'. */
 248                 done = filldir(cookie, ent->lde_name, namelen, lhash,
 249                                ino, type);
 250                 if (done) {
 251                         if (op_data->op_hash_offset != MDS_DIR_END_OFF)
 252                                 op_data->op_hash_offset = last_hash;
 253                         break;
 254                 } else {
 255                         last_hash = hash;
 256                 }
 257         }
 258
 259         if (IS_ERR(ent))
 260                 rc = PTR_ERR(ent);
 261
 262         if (page != NULL) {
 263                 kunmap(page);
 264                 page_cache_release(page);
 265         }
 266
 267         ll_dir_chain_fini(&chain);
 268         RETURN(rc);
 269 }
 270
 271 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 272 {
 273         struct inode            *inode  = filp->f_dentry->d_inode;
 274         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 275         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 276         __u64                   pos     = lfd->lfd_pos;
 277         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 278         int                     api32   = ll_need_32bit_api(sbi);
 279         struct md_op_data       *op_data;
 280         int                     rc;
 281 #ifdef HAVE_TOUCH_ATIME_1ARG
 282         struct path             path;
 283 #endif
 284         ENTRY;
 285
 286         if (lfd != NULL)
 287                 pos = lfd->lfd_pos;
 288         else
 289                 pos = 0;
 290
 291         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 292                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 293                inode, (unsigned long)pos, i_size_read(inode), api32);
 294
 295         if (pos == MDS_DIR_END_OFF)
 296                 /*
 297                  * end-of-file.
 298                  */
 299                 GOTO(out, rc = 0);
 300
 301         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 302                                      LUSTRE_OPC_ANY, inode);
 303         if (IS_ERR(op_data))
 304                 GOTO(out, rc = PTR_ERR(op_data));
 305
 306         op_data->op_hash_offset = pos;
 307         op_data->op_max_pages = sbi->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 308         rc = ll_dir_read(inode, op_data, cookie, filldir);
 309         if (lfd != NULL)
 310                 lfd->lfd_pos = op_data->op_hash_offset;
 311
 312         if (pos == MDS_DIR_END_OFF) {
 313                 if (api32)
 314                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 315                 else
 316                         filp->f_pos = LL_DIR_END_OFF;
 317         } else {
 318                 if (api32 && hash64)
 319                         filp->f_pos = op_data->op_hash_offset >> 32;
 320                 else
 321                         filp->f_pos = op_data->op_hash_offset;
 322         }
 323
 324         ll_finish_md_op_data(op_data);
 325         filp->f_version = inode->i_version;
 326 #ifdef HAVE_TOUCH_ATIME_1ARG
 327 #ifdef HAVE_F_PATH_MNT
 328         path.mnt = filp->f_path.mnt;
 329 #else
 330         path.mnt = filp->f_vfsmnt;
 331 #endif
 332         path.dentry = filp->f_dentry;
 333         touch_atime(&path);
 334 #else
 335         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 336 #endif
 337
 338 out:
 339         if (!rc)
 340                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 341
 342         RETURN(rc);
 343 }
 344
 345 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 346 {
 347         struct mgs_send_param *msp;
 348         int rc = 0;
 349
 350         OBD_ALLOC_PTR(msp);
 351         if (!msp)
 352                 return -ENOMEM;
 353
 354         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 355         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 356                                 sizeof(struct mgs_send_param), msp, NULL);
 357         if (rc)
 358                 CERROR("Failed to set parameter: %d\n", rc);
 359         OBD_FREE_PTR(msp);
 360
 361         return rc;
 362 }
 363
 364 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 365                                const char *filename)
 366 {
 367         struct ptlrpc_request *request = NULL;
 368         struct md_op_data *op_data;
 369         struct ll_sb_info *sbi = ll_i2sbi(dir);
 370         int mode;
 371         int err;
 372         ENTRY;
 373
 374         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 375                 RETURN(-EINVAL);
 376
 377         if (lump->lum_stripe_offset == (__u32)-1) {
 378                 int mdtidx;
 379
 380                 mdtidx = ll_get_mdt_idx(dir);
 381                 if (mdtidx < 0)
 382                         RETURN(mdtidx);
 383
 384                 lump->lum_stripe_offset = mdtidx;
 385         }
 386
 387         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 388                "stripe_offset %d, stripe_count: %u\n",
 389                PFID(ll_inode2fid(dir)), dir, filename,
 390                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 391
 392         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 393                 lustre_swab_lmv_user_md(lump);
 394
 395         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 396         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 397                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 398                                      lump);
 399         if (IS_ERR(op_data))
 400                 GOTO(err_exit, err = PTR_ERR(op_data));
 401
 402         op_data->op_cli_flags |= CLI_SET_MEA;
 403         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 404                         current_fsuid(), current_fsgid(),
 405                         cfs_curproc_cap_pack(), 0, &request);
 406         ll_finish_md_op_data(op_data);
 407         if (err)
 408                 GOTO(err_exit, err);
 409 err_exit:
 410         ptlrpc_req_finished(request);
 411         return err;
 412 }
 413
 414 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 415                      int set_default)
 416 {
 417         struct ll_sb_info *sbi = ll_i2sbi(inode);
 418         struct md_op_data *op_data;
 419         struct ptlrpc_request *req = NULL;
 420         int rc = 0;
 421         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 422         struct obd_device *mgc = lsi->lsi_mgc;
 423         int lum_size;
 424         ENTRY;
 425
 426         if (lump != NULL) {
 427                 /*
 428                  * This is coming from userspace, so should be in
 429                  * local endian.  But the MDS would like it in little
 430                  * endian, so we swab it before we send it.
 431                  */
 432                 switch (lump->lmm_magic) {
 433                 case LOV_USER_MAGIC_V1: {
 434                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 435                                 lustre_swab_lov_user_md_v1(lump);
 436                         lum_size = sizeof(struct lov_user_md_v1);
 437                         break;
 438                 }
 439                 case LOV_USER_MAGIC_V3: {
 440                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 441                                 lustre_swab_lov_user_md_v3(
 442                                         (struct lov_user_md_v3 *)lump);
 443                         lum_size = sizeof(struct lov_user_md_v3);
 444                         break;
 445                 }
 446                 case LMV_USER_MAGIC: {
 447                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 448                                 lustre_swab_lmv_user_md(
 449                                         (struct lmv_user_md *)lump);
 450                         lum_size = sizeof(struct lmv_user_md);
 451                         break;
 452                 }
 453                 default: {
 454                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 455                                         " %#08x != %#08x nor %#08x\n",
 456                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 457                                         LOV_USER_MAGIC_V3);
 458                         RETURN(-EINVAL);
 459                 }
 460                 }
 461         } else {
 462                 lum_size = sizeof(struct lov_user_md_v1);
 463         }
 464
 465         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 466                                      LUSTRE_OPC_ANY, NULL);
 467         if (IS_ERR(op_data))
 468                 RETURN(PTR_ERR(op_data));
 469
 470         /* swabbing is done in lov_setstripe() on server side */
 471         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 472                         NULL, 0, &req, NULL);
 473         ll_finish_md_op_data(op_data);
 474         ptlrpc_req_finished(req);
 475         if (rc) {
 476                 if (rc != -EPERM && rc != -EACCES)
 477                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 478         }
 479
 480         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 481          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 482          need the make the distiction between the 2 versions */
 483         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 484                 char *param = NULL;
 485                 char *buf;
 486
 487                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 488                 if (param == NULL)
 489                         GOTO(end, rc = -ENOMEM);
 490
 491                 buf = param;
 492                 /* Get fsname and assume devname to be -MDT0000. */
 493                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 494                 strcat(buf, "-MDT0000.lov");
 495                 buf += strlen(buf);
 496
 497                 /* Set root stripesize */
 498                 sprintf(buf, ".stripesize=%u",
 499                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 500                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 501                 if (rc)
 502                         GOTO(end, rc);
 503
 504                 /* Set root stripecount */
 505                 sprintf(buf, ".stripecount=%hd",
 506                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 507                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 508                 if (rc)
 509                         GOTO(end, rc);
 510
 511                 /* Set root stripeoffset */
 512                 sprintf(buf, ".stripeoffset=%hd",
 513                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 514                         (typeof(lump->lmm_stripe_offset))(-1));
 515                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 516
 517 end:
 518                 if (param != NULL)
 519                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 520         }
 521         RETURN(rc);
 522 }
 523
 524 /**
 525  * This function will be used to get default LOV/LMV/Default LMV
 526  * @valid will be used to indicate which stripe it will retrieve
 527  *      OBD_MD_MEA              LMV stripe EA
 528  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 529  *      otherwise               Default LOV EA.
 530  * Each time, it can only retrieve 1 stripe EA
 531  **/
 532 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 533                      struct ptlrpc_request **request, obd_valid valid)
 534 {
 535         struct ll_sb_info *sbi = ll_i2sbi(inode);
 536         struct mdt_body   *body;
 537         struct lov_mds_md *lmm = NULL;
 538         struct ptlrpc_request *req = NULL;
 539         int rc, lmm_size;
 540         struct md_op_data *op_data;
 541         ENTRY;
 542
 543         rc = ll_get_max_mdsize(sbi, &lmm_size);
 544         if (rc)
 545                 RETURN(rc);
 546
 547         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 548                                      0, lmm_size, LUSTRE_OPC_ANY,
 549                                      NULL);
 550         if (IS_ERR(op_data))
 551                 RETURN(PTR_ERR(op_data));
 552
 553         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 554         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 555         ll_finish_md_op_data(op_data);
 556         if (rc < 0) {
 557                 CDEBUG(D_INFO, "md_getattr failed on inode "
 558                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 559                 GOTO(out, rc);
 560         }
 561
 562         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 563         LASSERT(body != NULL);
 564
 565         lmm_size = body->eadatasize;
 566
 567         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 568             lmm_size == 0) {
 569                 GOTO(out, rc = -ENODATA);
 570         }
 571
 572         lmm = req_capsule_server_sized_get(&req->rq_pill,
 573                                            &RMF_MDT_MD, lmm_size);
 574         LASSERT(lmm != NULL);
 575
 576         /*
 577          * This is coming from the MDS, so is probably in
 578          * little endian.  We convert it to host endian before
 579          * passing it to userspace.
 580          */
 581         /* We don't swab objects for directories */
 582         switch (le32_to_cpu(lmm->lmm_magic)) {
 583         case LOV_MAGIC_V1:
 584                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 585                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 586                 break;
 587         case LOV_MAGIC_V3:
 588                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 589                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 590                 break;
 591         case LMV_MAGIC:
 592                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 593                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 594                 break;
 595         case LMV_USER_MAGIC:
 596                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 597                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 598                 break;
 599         default:
 600                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 601                 rc = -EPROTO;
 602         }
 603 out:
 604         *plmm = lmm;
 605         *plmm_size = lmm_size;
 606         *request = req;
 607         return rc;
 608 }
 609
 610 static int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi,
 611                                  const struct lu_fid *fid)
 612 {
 613         struct md_op_data       *op_data;
 614         int                     rc;
 615         int                     mdt_index;
 616         ENTRY;
 617
 618         OBD_ALLOC_PTR(op_data);
 619         if (op_data == NULL)
 620                 RETURN(-ENOMEM);
 621
 622         op_data->op_flags |= MF_GET_MDT_IDX;
 623         op_data->op_fid1 = *fid;
 624         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 625         mdt_index = op_data->op_mds;
 626         OBD_FREE_PTR(op_data);
 627         if (rc < 0)
 628                 RETURN(rc);
 629
 630         RETURN(mdt_index);
 631 }
 632
 633 /*
 634  *  Get MDT index for the inode.
 635  */
 636 int ll_get_mdt_idx(struct inode *inode)
 637 {
 638         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 639 }
 640
 641 /**
 642  * Generic handler to do any pre-copy work.
 643  *
 644  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 645  * first information for it that real work has started.
 646  *
 647  * Moreover, for a ARCHIVE request, it will sample the file data version and
 648  * store it in \a copy.
 649  *
 650  * \return 0 on success.
 651  */
 652 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 653 {
 654         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 655         struct hsm_progress_kernel       hpk;
 656         int                              rc;
 657         ENTRY;
 658
 659         /* Forge a hsm_progress based on data from copy. */
 660         hpk.hpk_fid = copy->hc_hai.hai_fid;
 661         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 662         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 663         hpk.hpk_extent.length = 0;
 664         hpk.hpk_flags = 0;
 665         hpk.hpk_errval = 0;
 666         hpk.hpk_data_version = 0;
 667
 668
 669         /* For archive request, we need to read the current file version. */
 670         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 671                 struct inode    *inode;
 672                 __u64            data_version = 0;
 673
 674                 /* Get inode for this fid */
 675                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 676                 if (IS_ERR(inode)) {
 677                         hpk.hpk_flags |= HP_FLAG_RETRY;
 678                         /* hpk_errval is >= 0 */
 679                         hpk.hpk_errval = -PTR_ERR(inode);
 680                         GOTO(progress, rc = PTR_ERR(inode));
 681                 }
 682
 683                 /* Read current file data version */
 684                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 685                 iput(inode);
 686                 if (rc != 0) {
 687                         CDEBUG(D_HSM, "Could not read file data version of "
 688                                       DFID" (rc = %d). Archive request ("
 689                                       LPX64") could not be done.\n",
 690                                       PFID(&copy->hc_hai.hai_fid), rc,
 691                                       copy->hc_hai.hai_cookie);
 692                         hpk.hpk_flags |= HP_FLAG_RETRY;
 693                         /* hpk_errval must be >= 0 */
 694                         hpk.hpk_errval = -rc;
 695                         GOTO(progress, rc);
 696                 }
 697
 698                 /* Store it the hsm_copy for later copytool use.
 699                  * Always modified even if no lsm. */
 700                 copy->hc_data_version = data_version;
 701         }
 702
 703 progress:
 704         /* On error, the request should be considered as completed */
 705         if (hpk.hpk_errval > 0)
 706                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 707         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 708                            &hpk, NULL);
 709
 710         RETURN(rc);
 711 }
 712
 713 /**
 714  * Generic handler to do any post-copy work.
 715  *
 716  * It will send the last hsm_progress update to coordinator to inform it
 717  * that copy is finished and whether it was successful or not.
 718  *
 719  * Moreover,
 720  * - for ARCHIVE request, it will sample the file data version and compare it
 721  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 722  *   will be considered as failed.
 723  * - for RESTORE request, it will sample the file data version and send it to
 724  *   coordinator which is useful if the file was imported as 'released'.
 725  *
 726  * \return 0 on success.
 727  */
 728 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 729 {
 730         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 731         struct hsm_progress_kernel       hpk;
 732         int                              rc;
 733         ENTRY;
 734
 735         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 736         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 737          * initialized if copy_end was called with copy == NULL.
 738          */
 739
 740         /* Forge a hsm_progress based on data from copy. */
 741         hpk.hpk_fid = copy->hc_hai.hai_fid;
 742         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 743         hpk.hpk_extent = copy->hc_hai.hai_extent;
 744         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 745         hpk.hpk_errval = copy->hc_errval;
 746         hpk.hpk_data_version = 0;
 747
 748         /* For archive request, we need to check the file data was not changed.
 749          *
 750          * For restore request, we need to send the file data version, this is
 751          * useful when the file was created using hsm_import.
 752          */
 753         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 754              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 755             (copy->hc_errval == 0)) {
 756                 struct inode    *inode;
 757                 __u64            data_version = 0;
 758
 759                 /* Get lsm for this fid */
 760                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 761                 if (IS_ERR(inode)) {
 762                         hpk.hpk_flags |= HP_FLAG_RETRY;
 763                         /* hpk_errval must be >= 0 */
 764                         hpk.hpk_errval = -PTR_ERR(inode);
 765                         GOTO(progress, rc = PTR_ERR(inode));
 766                 }
 767
 768                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 769                 iput(inode);
 770                 if (rc) {
 771                         CDEBUG(D_HSM, "Could not read file data version. "
 772                                       "Request could not be confirmed.\n");
 773                         if (hpk.hpk_errval == 0)
 774                                 hpk.hpk_errval = -rc;
 775                         GOTO(progress, rc);
 776                 }
 777
 778                 /* Store it the hsm_copy for later copytool use.
 779                  * Always modified even if no lsm. */
 780                 hpk.hpk_data_version = data_version;
 781
 782                 /* File could have been stripped during archiving, so we need
 783                  * to check anyway. */
 784                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 785                     (copy->hc_data_version != data_version)) {
 786                         CDEBUG(D_HSM, "File data version mismatched. "
 787                               "File content was changed during archiving. "
 788                                DFID", start:"LPX64" current:"LPX64"\n",
 789                                PFID(&copy->hc_hai.hai_fid),
 790                                copy->hc_data_version, data_version);
 791                         /* File was changed, send error to cdt. Do not ask for
 792                          * retry because if a file is modified frequently,
 793                          * the cdt will loop on retried archive requests.
 794                          * The policy engine will ask for a new archive later
 795                          * when the file will not be modified for some tunable
 796                          * time */
 797                         /* we do not notify caller */
 798                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 799                         /* hpk_errval must be >= 0 */
 800                         hpk.hpk_errval = EBUSY;
 801                 }
 802
 803         }
 804
 805 progress:
 806         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 807                            &hpk, NULL);
 808
 809         RETURN(rc);
 810 }
 811
 812
 813 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 814                           const void __user *data, size_t size)
 815 {
 816         void *copy;
 817         int rc;
 818
 819         OBD_ALLOC(copy, size);
 820         if (copy == NULL)
 821                 return -ENOMEM;
 822
 823         if (copy_from_user(copy, data, size)) {
 824                 rc = -EFAULT;
 825                 goto out;
 826         }
 827
 828         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 829 out:
 830         OBD_FREE(copy, size);
 831
 832         return rc;
 833 }
 834
 835 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 836 {
 837         int cmd = qctl->qc_cmd;
 838         int type = qctl->qc_type;
 839         int id = qctl->qc_id;
 840         int valid = qctl->qc_valid;
 841         int rc = 0;
 842         ENTRY;
 843
 844         switch (cmd) {
 845         case LUSTRE_Q_INVALIDATE:
 846         case LUSTRE_Q_FINVALIDATE:
 847         case Q_QUOTAON:
 848         case Q_QUOTAOFF:
 849         case Q_SETQUOTA:
 850         case Q_SETINFO:
 851                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 852                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 853                         RETURN(-EPERM);
 854                 break;
 855         case Q_GETQUOTA:
 856                 if (((type == USRQUOTA && current_euid() != id) ||
 857                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 858                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 859                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 860                         RETURN(-EPERM);
 861                 break;
 862         case Q_GETINFO:
 863                 break;
 864         default:
 865                 CERROR("unsupported quotactl op: %#x\n", cmd);
 866                 RETURN(-ENOTTY);
 867         }
 868
 869         if (valid != QC_GENERAL) {
 870                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 871                         RETURN(-EOPNOTSUPP);
 872
 873                 if (cmd == Q_GETINFO)
 874                         qctl->qc_cmd = Q_GETOINFO;
 875                 else if (cmd == Q_GETQUOTA)
 876                         qctl->qc_cmd = Q_GETOQUOTA;
 877                 else
 878                         RETURN(-EINVAL);
 879
 880                 switch (valid) {
 881                 case QC_MDTIDX:
 882                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 883                                            sizeof(*qctl), qctl, NULL);
 884                         break;
 885                 case QC_OSTIDX:
 886                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 887                                            sizeof(*qctl), qctl, NULL);
 888                         break;
 889                 case QC_UUID:
 890                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 891                                            sizeof(*qctl), qctl, NULL);
 892                         if (rc == -EAGAIN)
 893                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 894                                                    sbi->ll_dt_exp,
 895                                                    sizeof(*qctl), qctl, NULL);
 896                         break;
 897                 default:
 898                         rc = -EINVAL;
 899                         break;
 900                 }
 901
 902                 if (rc)
 903                         RETURN(rc);
 904
 905                 qctl->qc_cmd = cmd;
 906         } else {
 907                 struct obd_quotactl *oqctl;
 908
 909                 OBD_ALLOC_PTR(oqctl);
 910                 if (oqctl == NULL)
 911                         RETURN(-ENOMEM);
 912
 913                 QCTL_COPY(oqctl, qctl);
 914                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 915                 if (rc) {
 916                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 917                                 oqctl->qc_cmd = Q_QUOTAOFF;
 918                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 919                         }
 920                         OBD_FREE_PTR(oqctl);
 921                         RETURN(rc);
 922                 }
 923                 /* If QIF_SPACE is not set, client should collect the
 924                  * space usage from OSSs by itself */
 925                 if (cmd == Q_GETQUOTA &&
 926                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 927                     !oqctl->qc_dqblk.dqb_curspace) {
 928                         struct obd_quotactl *oqctl_tmp;
 929
 930                         OBD_ALLOC_PTR(oqctl_tmp);
 931                         if (oqctl_tmp == NULL)
 932                                 GOTO(out, rc = -ENOMEM);
 933
 934                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 935                         oqctl_tmp->qc_id = oqctl->qc_id;
 936                         oqctl_tmp->qc_type = oqctl->qc_type;
 937
 938                         /* collect space usage from OSTs */
 939                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 940                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 941                         if (!rc || rc == -EREMOTEIO) {
 942                                 oqctl->qc_dqblk.dqb_curspace =
 943                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 944                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 945                         }
 946
 947                         /* collect space & inode usage from MDTs */
 948                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 949                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 950                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 951                         if (!rc || rc == -EREMOTEIO) {
 952                                 oqctl->qc_dqblk.dqb_curspace +=
 953                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 954                                 oqctl->qc_dqblk.dqb_curinodes =
 955                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
 956                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
 957                         } else {
 958                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
 959                         }
 960
 961                         OBD_FREE_PTR(oqctl_tmp);
 962                 }
 963 out:
 964                 QCTL_COPY(qctl, oqctl);
 965                 OBD_FREE_PTR(oqctl);
 966         }
 967
 968         RETURN(rc);
 969 }
 970
 971 static char *
 972 ll_getname(const char __user *filename)
 973 {
 974         int ret = 0, len;
 975         char *tmp = __getname();
 976
 977         if (!tmp)
 978                 return ERR_PTR(-ENOMEM);
 979
 980         len = strncpy_from_user(tmp, filename, PATH_MAX);
 981         if (len == 0)
 982                 ret = -ENOENT;
 983         else if (len > PATH_MAX)
 984                 ret = -ENAMETOOLONG;
 985
 986         if (ret) {
 987                 __putname(tmp);
 988                 tmp =  ERR_PTR(ret);
 989         }
 990         return tmp;
 991 }
 992
 993 #define ll_putname(filename) __putname(filename)
 994
 995 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 996 {
 997         struct inode *inode = file->f_dentry->d_inode;
 998         struct ll_sb_info *sbi = ll_i2sbi(inode);
 999         struct obd_ioctl_data *data;
1000         int rc = 0;
1001         ENTRY;
1002
1003         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1004                PFID(ll_inode2fid(inode)), inode, cmd);
1005
1006         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1007         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1008                 return -ENOTTY;
1009
1010         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1011         switch(cmd) {
1012         case FSFILT_IOC_GETFLAGS:
1013         case FSFILT_IOC_SETFLAGS:
1014                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1015         case FSFILT_IOC_GETVERSION_OLD:
1016         case FSFILT_IOC_GETVERSION:
1017                 RETURN(put_user(inode->i_generation, (int *)arg));
1018         /* We need to special case any other ioctls we want to handle,
1019          * to send them to the MDS/OST as appropriate and to properly
1020          * network encode the arg field.
1021         case FSFILT_IOC_SETVERSION_OLD:
1022         case FSFILT_IOC_SETVERSION:
1023         */
1024         case LL_IOC_GET_MDTIDX: {
1025                 int mdtidx;
1026
1027                 mdtidx = ll_get_mdt_idx(inode);
1028                 if (mdtidx < 0)
1029                         RETURN(mdtidx);
1030
1031                 if (put_user((int)mdtidx, (int*)arg))
1032                         RETURN(-EFAULT);
1033
1034                 return 0;
1035         }
1036         case IOC_MDC_LOOKUP: {
1037                 struct ptlrpc_request *request = NULL;
1038                 int namelen, len = 0;
1039                 char *buf = NULL;
1040                 char *filename;
1041                 struct md_op_data *op_data;
1042
1043                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1044                 if (rc)
1045                         RETURN(rc);
1046                 data = (void *)buf;
1047
1048                 filename = data->ioc_inlbuf1;
1049                 namelen = strlen(filename);
1050
1051                 if (namelen < 1) {
1052                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1053                         GOTO(out_free, rc = -EINVAL);
1054                 }
1055
1056                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1057                                              0, LUSTRE_OPC_ANY, NULL);
1058                 if (IS_ERR(op_data))
1059                         GOTO(out_free, rc = PTR_ERR(op_data));
1060
1061                 op_data->op_valid = OBD_MD_FLID;
1062                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1063                 ll_finish_md_op_data(op_data);
1064                 if (rc < 0) {
1065                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1066                         GOTO(out_free, rc);
1067                 }
1068                 ptlrpc_req_finished(request);
1069                 EXIT;
1070 out_free:
1071                 obd_ioctl_freedata(buf, len);
1072                 return rc;
1073         }
1074         case LL_IOC_LMV_SETSTRIPE: {
1075                 struct lmv_user_md  *lum;
1076                 char            *buf = NULL;
1077                 char            *filename;
1078                 int              namelen = 0;
1079                 int              lumlen = 0;
1080                 int              len;
1081                 int              rc;
1082
1083                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1084                 if (rc)
1085                         RETURN(rc);
1086
1087                 data = (void *)buf;
1088                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1089                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1090                         GOTO(lmv_out_free, rc = -EINVAL);
1091
1092                 filename = data->ioc_inlbuf1;
1093                 namelen = data->ioc_inllen1;
1094
1095                 if (namelen < 1) {
1096                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1097                         GOTO(lmv_out_free, rc = -EINVAL);
1098                 }
1099                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1100                 lumlen = data->ioc_inllen2;
1101
1102                 if (lum->lum_magic != LMV_USER_MAGIC ||
1103                     lumlen != sizeof(*lum)) {
1104                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1105                                filename, lum->lum_magic, lumlen, -EFAULT);
1106                         GOTO(lmv_out_free, rc = -EINVAL);
1107                 }
1108
1109                 /**
1110                  * ll_dir_setdirstripe will be used to set dir stripe
1111                  *  mdc_create--->mdt_reint_create (with dirstripe)
1112                  */
1113                 rc = ll_dir_setdirstripe(inode, lum, filename);
1114 lmv_out_free:
1115                 obd_ioctl_freedata(buf, len);
1116                 RETURN(rc);
1117
1118         }
1119         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1120                 struct lmv_user_md        lum;
1121                 struct lmv_user_md __user *ulump =
1122                                         (struct lmv_user_md __user *)arg;
1123                 int                       rc;
1124
1125                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1126                         RETURN(-EFAULT);
1127
1128                 if (lum.lum_magic != LMV_USER_MAGIC)
1129                         RETURN(-EINVAL);
1130
1131                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1132
1133                 RETURN(rc);
1134         }
1135         case LL_IOC_LOV_SETSTRIPE: {
1136                 struct lov_user_md_v3 lumv3;
1137                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1138                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1139                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1140
1141                 int set_default = 0;
1142
1143                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1144                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1145                         sizeof(lumv3p->lmm_objects[0]));
1146                 /* first try with v1 which is smaller than v3 */
1147                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1148                         RETURN(-EFAULT);
1149
1150                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1151                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1152                                 RETURN(-EFAULT);
1153                 }
1154
1155                 if (inode->i_sb->s_root == file->f_dentry)
1156                         set_default = 1;
1157
1158                 /* in v1 and v3 cases lumv1 points to data */
1159                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1160
1161                 RETURN(rc);
1162         }
1163         case LL_IOC_LMV_GETSTRIPE: {
1164                 struct lmv_user_md __user *ulmv =
1165                                         (struct lmv_user_md __user *)arg;
1166                 struct lmv_user_md      lum;
1167                 struct ptlrpc_request   *request = NULL;
1168                 union lmv_mds_md        *lmm = NULL;
1169                 int                     lmmsize;
1170                 obd_valid               valid = 0;
1171                 struct lmv_user_md      *tmp = NULL;
1172                 int                     mdt_index;
1173                 int                     lum_size;
1174                 int                     stripe_count;
1175                 int                     i;
1176                 int                     rc;
1177
1178                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1179                         RETURN(-EFAULT);
1180
1181                 /* lum_magic will indicate which stripe the ioctl will like
1182                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1183                  * is for default LMV stripe */
1184                 if (lum.lum_magic == LMV_MAGIC_V1)
1185                         valid |= OBD_MD_MEA;
1186                 else if (lum.lum_magic == LMV_USER_MAGIC)
1187                         valid |= OBD_MD_DEFAULT_MEA;
1188                 else
1189                         RETURN(-EINVAL);
1190
1191                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1192                                       valid);
1193                 if (rc != 0 && rc != -ENODATA)
1194                         GOTO(finish_req, rc);
1195
1196                 /* Get default LMV EA */
1197                 if (lum.lum_magic == LMV_USER_MAGIC) {
1198                         if (rc != 0)
1199                                 GOTO(finish_req, rc);
1200
1201                         if (lmmsize > sizeof(*ulmv))
1202                                 GOTO(finish_req, rc = -EINVAL);
1203
1204                         if (copy_to_user(ulmv, lmm, lmmsize))
1205                                 GOTO(finish_req, rc = -EFAULT);
1206
1207                         GOTO(finish_req, rc);
1208                 }
1209
1210                 /* Get normal LMV EA */
1211                 if (rc == -ENODATA) {
1212                         stripe_count = 1;
1213                 } else {
1214                         LASSERT(lmm != NULL);
1215                         stripe_count = lmv_mds_md_stripe_count_get(lmm);
1216                 }
1217
1218                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1219                 OBD_ALLOC(tmp, lum_size);
1220                 if (tmp == NULL)
1221                         GOTO(finish_req, rc = -ENOMEM);
1222
1223                 tmp->lum_magic = LMV_MAGIC_V1;
1224                 tmp->lum_stripe_count = 1;
1225                 mdt_index = ll_get_mdt_idx(inode);
1226                 if (mdt_index < 0)
1227                         GOTO(out_tmp, rc = -ENOMEM);
1228                 tmp->lum_stripe_offset = mdt_index;
1229                 tmp->lum_objects[0].lum_mds = mdt_index;
1230                 tmp->lum_objects[0].lum_fid = *ll_inode2fid(inode);
1231                 for (i = 1; i < stripe_count; i++) {
1232                         struct lmv_mds_md_v1 *lmm1;
1233
1234                         lmm1 = &lmm->lmv_md_v1;
1235                         mdt_index = ll_get_mdt_idx_by_fid(sbi,
1236                                                      &lmm1->lmv_stripe_fids[i]);
1237                         if (mdt_index < 0)
1238                                 GOTO(out_tmp, rc = mdt_index);
1239
1240                         tmp->lum_objects[i].lum_mds = mdt_index;
1241                         tmp->lum_objects[i].lum_fid = lmm1->lmv_stripe_fids[i];
1242                         tmp->lum_stripe_count++;
1243                 }
1244
1245                 if (copy_to_user(ulmv, tmp, lum_size))
1246                         GOTO(out_tmp, rc = -EFAULT);
1247 out_tmp:
1248                 OBD_FREE(tmp, lum_size);
1249 finish_req:
1250                 ptlrpc_req_finished(request);
1251                 return rc;
1252         }
1253
1254         case LL_IOC_REMOVE_ENTRY: {
1255                 char            *filename = NULL;
1256                 int              namelen = 0;
1257                 int              rc;
1258
1259                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1260                  * unsupported server, which might crash the server(LU-2730),
1261                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1262                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1263                  * server will support REINT_RMENTRY XXX*/
1264                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1265                         RETURN(-ENOTSUPP);
1266
1267                 filename = ll_getname((const char *)arg);
1268                 if (IS_ERR(filename))
1269                         RETURN(PTR_ERR(filename));
1270
1271                 namelen = strlen(filename);
1272                 if (namelen < 1)
1273                         GOTO(out_rmdir, rc = -EINVAL);
1274
1275                 rc = ll_rmdir_entry(inode, filename, namelen);
1276 out_rmdir:
1277                 if (filename)
1278                         ll_putname(filename);
1279                 RETURN(rc);
1280         }
1281         case LL_IOC_LOV_SWAP_LAYOUTS:
1282                 RETURN(-EPERM);
1283         case LL_IOC_OBD_STATFS:
1284                 RETURN(ll_obd_statfs(inode, (void *)arg));
1285         case LL_IOC_LOV_GETSTRIPE:
1286         case LL_IOC_MDC_GETINFO:
1287         case IOC_MDC_GETFILEINFO:
1288         case IOC_MDC_GETFILESTRIPE: {
1289                 struct ptlrpc_request *request = NULL;
1290                 struct lov_user_md *lump;
1291                 struct lov_mds_md *lmm = NULL;
1292                 struct mdt_body *body;
1293                 char *filename = NULL;
1294                 int lmmsize;
1295
1296                 if (cmd == IOC_MDC_GETFILEINFO ||
1297                     cmd == IOC_MDC_GETFILESTRIPE) {
1298                         filename = ll_getname((const char *)arg);
1299                         if (IS_ERR(filename))
1300                                 RETURN(PTR_ERR(filename));
1301
1302                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1303                                                       &lmmsize, &request);
1304                 } else {
1305                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1306                                               &request, 0);
1307                 }
1308
1309                 if (request) {
1310                         body = req_capsule_server_get(&request->rq_pill,
1311                                                       &RMF_MDT_BODY);
1312                         LASSERT(body != NULL);
1313                 } else {
1314                         GOTO(out_req, rc);
1315                 }
1316
1317                 if (rc < 0) {
1318                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1319                                                cmd == LL_IOC_MDC_GETINFO))
1320                                 GOTO(skip_lmm, rc = 0);
1321                         else
1322                                 GOTO(out_req, rc);
1323                 }
1324
1325                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1326                     cmd == LL_IOC_LOV_GETSTRIPE) {
1327                         lump = (struct lov_user_md *)arg;
1328                 } else {
1329                         struct lov_user_mds_data *lmdp;
1330                         lmdp = (struct lov_user_mds_data *)arg;
1331                         lump = &lmdp->lmd_lmm;
1332                 }
1333                 if (copy_to_user(lump, lmm, lmmsize)) {
1334                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1335                                 GOTO(out_req, rc = -EFAULT);
1336                         rc = -EOVERFLOW;
1337                 }
1338         skip_lmm:
1339                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1340                         struct lov_user_mds_data *lmdp;
1341                         lstat_t st = { 0 };
1342
1343                         st.st_dev     = inode->i_sb->s_dev;
1344                         st.st_mode    = body->mode;
1345                         st.st_nlink   = body->nlink;
1346                         st.st_uid     = body->uid;
1347                         st.st_gid     = body->gid;
1348                         st.st_rdev    = body->rdev;
1349                         st.st_size    = body->size;
1350                         st.st_blksize = PAGE_CACHE_SIZE;
1351                         st.st_blocks  = body->blocks;
1352                         st.st_atime   = body->atime;
1353                         st.st_mtime   = body->mtime;
1354                         st.st_ctime   = body->ctime;
1355                         st.st_ino     = inode->i_ino;
1356
1357                         lmdp = (struct lov_user_mds_data *)arg;
1358                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1359                                 GOTO(out_req, rc = -EFAULT);
1360                 }
1361
1362                 EXIT;
1363         out_req:
1364                 ptlrpc_req_finished(request);
1365                 if (filename)
1366                         ll_putname(filename);
1367                 return rc;
1368         }
1369         case IOC_LOV_GETINFO: {
1370                 struct lov_user_mds_data *lumd;
1371                 struct lov_stripe_md *lsm;
1372                 struct lov_user_md *lum;
1373                 struct lov_mds_md *lmm;
1374                 int lmmsize;
1375                 lstat_t st;
1376
1377                 lumd = (struct lov_user_mds_data *)arg;
1378                 lum = &lumd->lmd_lmm;
1379
1380                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1381                 if (rc)
1382                         RETURN(rc);
1383
1384                 OBD_ALLOC_LARGE(lmm, lmmsize);
1385                 if (lmm == NULL)
1386                         RETURN(-ENOMEM);
1387
1388                 if (copy_from_user(lmm, lum, lmmsize))
1389                         GOTO(free_lmm, rc = -EFAULT);
1390
1391                 switch (lmm->lmm_magic) {
1392                 case LOV_USER_MAGIC_V1:
1393                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1394                                 break;
1395                         /* swab objects first so that stripes num will be sane */
1396                         lustre_swab_lov_user_md_objects(
1397                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1398                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1399                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1400                         break;
1401                 case LOV_USER_MAGIC_V3:
1402                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1403                                 break;
1404                         /* swab objects first so that stripes num will be sane */
1405                         lustre_swab_lov_user_md_objects(
1406                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1407                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1408                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1409                         break;
1410                 default:
1411                         GOTO(free_lmm, rc = -EINVAL);
1412                 }
1413
1414                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1415                 if (rc < 0)
1416                         GOTO(free_lmm, rc = -ENOMEM);
1417
1418                 /* Perform glimpse_size operation. */
1419                 memset(&st, 0, sizeof(st));
1420
1421                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1422                 if (rc)
1423                         GOTO(free_lsm, rc);
1424
1425                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1426                         GOTO(free_lsm, rc = -EFAULT);
1427
1428                 EXIT;
1429         free_lsm:
1430                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1431         free_lmm:
1432                 OBD_FREE_LARGE(lmm, lmmsize);
1433                 return rc;
1434         }
1435         case OBD_IOC_LLOG_CATINFO: {
1436                 RETURN(-EOPNOTSUPP);
1437         }
1438         case OBD_IOC_QUOTACHECK: {
1439                 struct obd_quotactl *oqctl;
1440                 int error = 0;
1441
1442                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1443                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1444                         RETURN(-EPERM);
1445
1446                 OBD_ALLOC_PTR(oqctl);
1447                 if (!oqctl)
1448                         RETURN(-ENOMEM);
1449                 oqctl->qc_type = arg;
1450                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1451                 if (rc < 0) {
1452                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1453                         error = rc;
1454                 }
1455
1456                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1457                 if (rc < 0)
1458                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1459
1460                 OBD_FREE_PTR(oqctl);
1461                 return error ?: rc;
1462         }
1463         case OBD_IOC_POLL_QUOTACHECK: {
1464                 struct if_quotacheck *check;
1465
1466                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1467                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1468                         RETURN(-EPERM);
1469
1470                 OBD_ALLOC_PTR(check);
1471                 if (!check)
1472                         RETURN(-ENOMEM);
1473
1474                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1475                                    NULL);
1476                 if (rc) {
1477                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1478                         if (copy_to_user((void *)arg, check,
1479                                              sizeof(*check)))
1480                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1481                         GOTO(out_poll, rc);
1482                 }
1483
1484                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1485                                    NULL);
1486                 if (rc) {
1487                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1488                         if (copy_to_user((void *)arg, check,
1489                                              sizeof(*check)))
1490                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1491                         GOTO(out_poll, rc);
1492                 }
1493         out_poll:
1494                 OBD_FREE_PTR(check);
1495                 RETURN(rc);
1496         }
1497 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1498         case LL_IOC_QUOTACTL_18: {
1499                 /* copy the old 1.x quota struct for internal use, then copy
1500                  * back into old format struct.  For 1.8 compatibility. */
1501                 struct if_quotactl_18 *qctl_18;
1502                 struct if_quotactl *qctl_20;
1503
1504                 OBD_ALLOC_PTR(qctl_18);
1505                 if (!qctl_18)
1506                         RETURN(-ENOMEM);
1507
1508                 OBD_ALLOC_PTR(qctl_20);
1509                 if (!qctl_20)
1510                         GOTO(out_quotactl_18, rc = -ENOMEM);
1511
1512                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1513                         GOTO(out_quotactl_20, rc = -ENOMEM);
1514
1515                 QCTL_COPY(qctl_20, qctl_18);
1516                 qctl_20->qc_idx = 0;
1517
1518                 /* XXX: dqb_valid was borrowed as a flag to mark that
1519                  *      only mds quota is wanted */
1520                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1521                     qctl_18->qc_dqblk.dqb_valid) {
1522                         qctl_20->qc_valid = QC_MDTIDX;
1523                         qctl_20->qc_dqblk.dqb_valid = 0;
1524                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1525                         qctl_20->qc_valid = QC_UUID;
1526                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1527                 } else {
1528                         qctl_20->qc_valid = QC_GENERAL;
1529                 }
1530
1531                 rc = quotactl_ioctl(sbi, qctl_20);
1532
1533                 if (rc == 0) {
1534                         QCTL_COPY(qctl_18, qctl_20);
1535                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1536
1537                         if (copy_to_user((void *)arg, qctl_18,
1538                                              sizeof(*qctl_18)))
1539                                 rc = -EFAULT;
1540                 }
1541
1542         out_quotactl_20:
1543                 OBD_FREE_PTR(qctl_20);
1544         out_quotactl_18:
1545                 OBD_FREE_PTR(qctl_18);
1546                 RETURN(rc);
1547         }
1548 #else
1549 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1550 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1551         case LL_IOC_QUOTACTL: {
1552                 struct if_quotactl *qctl;
1553
1554                 OBD_ALLOC_PTR(qctl);
1555                 if (!qctl)
1556                         RETURN(-ENOMEM);
1557
1558                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1559                         GOTO(out_quotactl, rc = -EFAULT);
1560
1561                 rc = quotactl_ioctl(sbi, qctl);
1562
1563                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1564                         rc = -EFAULT;
1565
1566         out_quotactl:
1567                 OBD_FREE_PTR(qctl);
1568                 RETURN(rc);
1569         }
1570         case OBD_IOC_GETDTNAME:
1571         case OBD_IOC_GETMDNAME:
1572                 RETURN(ll_get_obd_name(inode, cmd, arg));
1573         case LL_IOC_FLUSHCTX:
1574                 RETURN(ll_flush_ctx(inode));
1575 #ifdef CONFIG_FS_POSIX_ACL
1576         case LL_IOC_RMTACL: {
1577             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1578                 inode == inode->i_sb->s_root->d_inode) {
1579                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1580
1581                 LASSERT(fd != NULL);
1582                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1583                 if (!rc)
1584                         fd->fd_flags |= LL_FILE_RMTACL;
1585                 RETURN(rc);
1586             } else
1587                 RETURN(0);
1588         }
1589 #endif
1590         case LL_IOC_GETOBDCOUNT: {
1591                 int count, vallen;
1592                 struct obd_export *exp;
1593
1594                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1595                         RETURN(-EFAULT);
1596
1597                 /* get ost count when count is zero, get mdt count otherwise */
1598                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1599                 vallen = sizeof(count);
1600                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1601                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1602                 if (rc) {
1603                         CERROR("get target count failed: %d\n", rc);
1604                         RETURN(rc);
1605                 }
1606
1607                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1608                         RETURN(-EFAULT);
1609
1610                 RETURN(0);
1611         }
1612         case LL_IOC_PATH2FID:
1613                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1614                                      sizeof(struct lu_fid)))
1615                         RETURN(-EFAULT);
1616                 RETURN(0);
1617         case LL_IOC_GET_CONNECT_FLAGS: {
1618                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1619         }
1620         case OBD_IOC_CHANGELOG_SEND:
1621         case OBD_IOC_CHANGELOG_CLEAR:
1622                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1623                                     sizeof(struct ioc_changelog));
1624                 RETURN(rc);
1625         case OBD_IOC_FID2PATH:
1626                 RETURN(ll_fid2path(inode, (void *)arg));
1627         case LL_IOC_HSM_REQUEST: {
1628                 struct hsm_user_request *hur;
1629                 int                      totalsize;
1630
1631                 OBD_ALLOC_PTR(hur);
1632                 if (hur == NULL)
1633                         RETURN(-ENOMEM);
1634
1635                 /* We don't know the true size yet; copy the fixed-size part */
1636                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1637                         OBD_FREE_PTR(hur);
1638                         RETURN(-EFAULT);
1639                 }
1640
1641                 /* Compute the whole struct size */
1642                 totalsize = hur_len(hur);
1643                 OBD_FREE_PTR(hur);
1644
1645                 /* Make sure the size is reasonable */
1646                 if (totalsize >= MDS_MAXREQSIZE)
1647                         RETURN(-E2BIG);
1648
1649                 OBD_ALLOC_LARGE(hur, totalsize);
1650                 if (hur == NULL)
1651                         RETURN(-ENOMEM);
1652
1653                 /* Copy the whole struct */
1654                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1655                         OBD_FREE_LARGE(hur, totalsize);
1656                         RETURN(-EFAULT);
1657                 }
1658
1659                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1660                         const struct lu_fid *fid;
1661                         struct inode *f;
1662                         int i;
1663
1664                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1665                                 fid = &hur->hur_user_item[i].hui_fid;
1666                                 f = search_inode_for_lustre(inode->i_sb, fid);
1667                                 if (IS_ERR(f)) {
1668                                         rc = PTR_ERR(f);
1669                                         break;
1670                                 }
1671
1672                                 rc = ll_hsm_release(f);
1673                                 iput(f);
1674                                 if (rc != 0)
1675                                         break;
1676                         }
1677                 } else {
1678                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1679                                            hur, NULL);
1680                 }
1681
1682                 OBD_FREE_LARGE(hur, totalsize);
1683
1684                 RETURN(rc);
1685         }
1686         case LL_IOC_HSM_PROGRESS: {
1687                 struct hsm_progress_kernel      hpk;
1688                 struct hsm_progress             hp;
1689
1690                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1691                         RETURN(-EFAULT);
1692
1693                 hpk.hpk_fid = hp.hp_fid;
1694                 hpk.hpk_cookie = hp.hp_cookie;
1695                 hpk.hpk_extent = hp.hp_extent;
1696                 hpk.hpk_flags = hp.hp_flags;
1697                 hpk.hpk_errval = hp.hp_errval;
1698                 hpk.hpk_data_version = 0;
1699
1700                 /* File may not exist in Lustre; all progress
1701                  * reported to Lustre root */
1702                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1703                                    NULL);
1704                 RETURN(rc);
1705         }
1706         case LL_IOC_HSM_CT_START:
1707                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1708                         RETURN(-EPERM);
1709
1710                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1711                                     sizeof(struct lustre_kernelcomm));
1712                 RETURN(rc);
1713
1714         case LL_IOC_HSM_COPY_START: {
1715                 struct hsm_copy *copy;
1716                 int              rc;
1717
1718                 OBD_ALLOC_PTR(copy);
1719                 if (copy == NULL)
1720                         RETURN(-ENOMEM);
1721                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1722                         OBD_FREE_PTR(copy);
1723                         RETURN(-EFAULT);
1724                 }
1725
1726                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1727                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1728                         rc = -EFAULT;
1729
1730                 OBD_FREE_PTR(copy);
1731                 RETURN(rc);
1732         }
1733         case LL_IOC_HSM_COPY_END: {
1734                 struct hsm_copy *copy;
1735                 int              rc;
1736
1737                 OBD_ALLOC_PTR(copy);
1738                 if (copy == NULL)
1739                         RETURN(-ENOMEM);
1740                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1741                         OBD_FREE_PTR(copy);
1742                         RETURN(-EFAULT);
1743                 }
1744
1745                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1746                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1747                         rc = -EFAULT;
1748
1749                 OBD_FREE_PTR(copy);
1750                 RETURN(rc);
1751         }
1752         default:
1753                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1754                                      (void *)arg));
1755         }
1756 }
1757
1758 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1759 {
1760         struct inode *inode = file->f_mapping->host;
1761         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1762         struct ll_sb_info *sbi = ll_i2sbi(inode);
1763         int api32 = ll_need_32bit_api(sbi);
1764         loff_t ret = -EINVAL;
1765         ENTRY;
1766
1767         mutex_lock(&inode->i_mutex);
1768         switch (origin) {
1769                 case SEEK_SET:
1770                         break;
1771                 case SEEK_CUR:
1772                         offset += file->f_pos;
1773                         break;
1774                 case SEEK_END:
1775                         if (offset > 0)
1776                                 GOTO(out, ret);
1777                         if (api32)
1778                                 offset += LL_DIR_END_OFF_32BIT;
1779                         else
1780                                 offset += LL_DIR_END_OFF;
1781                         break;
1782                 default:
1783                         GOTO(out, ret);
1784         }
1785
1786         if (offset >= 0 &&
1787             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1788              (!api32 && offset <= LL_DIR_END_OFF))) {
1789                 if (offset != file->f_pos) {
1790                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1791                             (!api32 && offset == LL_DIR_END_OFF))
1792                                 fd->lfd_pos = MDS_DIR_END_OFF;
1793                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1794                                 fd->lfd_pos = offset << 32;
1795                         else
1796                                 fd->lfd_pos = offset;
1797                         file->f_pos = offset;
1798                         file->f_version = 0;
1799                 }
1800                 ret = offset;
1801         }
1802         GOTO(out, ret);
1803
1804 out:
1805         mutex_unlock(&inode->i_mutex);
1806         return ret;
1807 }
1808
1809 int ll_dir_open(struct inode *inode, struct file *file)
1810 {
1811         ENTRY;
1812         RETURN(ll_file_open(inode, file));
1813 }
1814
1815 int ll_dir_release(struct inode *inode, struct file *file)
1816 {
1817         ENTRY;
1818         RETURN(ll_file_release(inode, file));
1819 }
1820
1821 struct file_operations ll_dir_operations = {
1822         .llseek   = ll_dir_seek,
1823         .open     = ll_dir_open,
1824         .release  = ll_dir_release,
1825         .read     = generic_read_dir,
1826         .readdir  = ll_readdir,
1827         .unlocked_ioctl   = ll_dir_ioctl,
1828         .fsync    = ll_fsync,
1829 };