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_ioctl.h>
  55 #include <lustre_lib.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 = NULL;
 166         struct md_callback cb_op;
 167         int rc;
 168         ENTRY;
 169
 170         LASSERT(*ppage == NULL);
 171         cb_op.md_blocking_ast = ll_md_blocking_ast;
 172         op_data->op_cli_flags &= ~CLI_NEXT_ENTRY;
 173         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 174         if (rc != 0)
 175                 entry = ERR_PTR(rc);
 176         RETURN(entry);
 177 }
 178
 179 struct lu_dirent *ll_dir_entry_next(struct inode *dir,
 180                                     struct md_op_data *op_data,
 181                                     struct lu_dirent *ent,
 182                                     struct page **ppage)
 183 {
 184         struct lu_dirent *entry = NULL;
 185         struct md_callback cb_op;
 186         int rc;
 187         ENTRY;
 188
 189         op_data->op_hash_offset = le64_to_cpu(ent->lde_hash);
 190
 191         /* release last page */
 192         LASSERT(*ppage != NULL);
 193         kunmap(*ppage);
 194         page_cache_release(*ppage);
 195
 196         cb_op.md_blocking_ast = ll_md_blocking_ast;
 197         op_data->op_cli_flags |= CLI_NEXT_ENTRY;
 198         rc = md_read_entry(ll_i2mdexp(dir), op_data, &cb_op, &entry, ppage);
 199         if (rc != 0)
 200                 entry = ERR_PTR(rc);
 201
 202         RETURN(entry);
 203 }
 204
 205 #ifdef HAVE_DIR_CONTEXT
 206 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 207                 struct dir_context *ctx)
 208 {
 209 #else
 210 int ll_dir_read(struct inode *inode, struct md_op_data *op_data,
 211                 void *cookie, filldir_t filldir)
 212 {
 213 #endif
 214         struct ll_sb_info       *sbi = ll_i2sbi(inode);
 215         struct ll_dir_chain     chain;
 216         struct lu_dirent        *ent;
 217         int                     api32 = ll_need_32bit_api(sbi);
 218         int                     hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 219         int                     done = 0;
 220         int                     rc = 0;
 221         __u64                   hash = MDS_DIR_END_OFF;
 222         struct page             *page = NULL;
 223         ENTRY;
 224
 225         ll_dir_chain_init(&chain);
 226         for (ent = ll_dir_entry_start(inode, op_data, &page);
 227              ent != NULL && !IS_ERR(ent) && !done;
 228              ent = ll_dir_entry_next(inode, op_data, ent, &page)) {
 229                 __u16          type;
 230                 int            namelen;
 231                 struct lu_fid  fid;
 232                 __u64          lhash;
 233                 __u64          ino;
 234
 235                 hash = le64_to_cpu(ent->lde_hash);
 236                 if (hash < op_data->op_hash_offset)
 237                         /*
 238                          * Skip until we find target hash
 239                          * value.
 240                          */
 241                         continue;
 242                 namelen = le16_to_cpu(ent->lde_namelen);
 243                 if (namelen == 0)
 244                         /*
 245                          * Skip dummy record.
 246                          */
 247                         continue;
 248
 249                 if (api32 && hash64)
 250                         lhash = hash >> 32;
 251                 else
 252                         lhash = hash;
 253                 fid_le_to_cpu(&fid, &ent->lde_fid);
 254                 ino = cl_fid_build_ino(&fid, api32);
 255                 type = ll_dirent_type_get(ent);
 256
 257 #ifdef HAVE_DIR_CONTEXT
 258                 /* For 'll_nfs_get_name_filldir()', it will try
 259                  * to access the 'ent' through its 'lde_name',
 260                  * so the parameter 'name' for 'filldir()' must
 261                  * be part of the 'ent'. */
 262                 done = !dir_emit(ctx, ent->lde_name, namelen, ino, type);
 263 #else
 264                 done = filldir(cookie, ent->lde_name, namelen, lhash,
 265                                ino, type);
 266 #endif
 267                 if (done) {
 268                         if (op_data->op_hash_offset != MDS_DIR_END_OFF)
 269                                 op_data->op_hash_offset = hash;
 270                         break;
 271                 }
 272         }
 273
 274         if (IS_ERR(ent))
 275                 rc = PTR_ERR(ent);
 276         else if (ent == NULL)
 277                 op_data->op_hash_offset = MDS_DIR_END_OFF;
 278
 279         if (page != NULL) {
 280                 kunmap(page);
 281                 page_cache_release(page);
 282         }
 283
 284         ll_dir_chain_fini(&chain);
 285         RETURN(rc);
 286 }
 287
 288 #ifdef HAVE_DIR_CONTEXT
 289 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 290 #else
 291 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 292 #endif
 293 {
 294         struct inode            *inode  = filp->f_dentry->d_inode;
 295         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 296         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 297         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 298         int                     api32   = ll_need_32bit_api(sbi);
 299         struct md_op_data       *op_data;
 300         __u64                   pos;
 301         int                     rc;
 302         ENTRY;
 303
 304         if (lfd != NULL)
 305                 pos = lfd->lfd_pos;
 306         else
 307                 pos = 0;
 308
 309         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 310                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 311                inode, (unsigned long)pos, i_size_read(inode), api32);
 312
 313         if (pos == MDS_DIR_END_OFF)
 314                 /*
 315                  * end-of-file.
 316                  */
 317                 GOTO(out, rc = 0);
 318
 319         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 320                                      LUSTRE_OPC_ANY, inode);
 321         if (IS_ERR(op_data))
 322                 GOTO(out, rc = PTR_ERR(op_data));
 323
 324         if (unlikely(op_data->op_mea1 != NULL)) {
 325                 /* This is only needed for striped dir to fill ..,
 326                  * see lmv_read_entry */
 327                 if (filp->f_dentry->d_parent != NULL &&
 328                     filp->f_dentry->d_parent->d_inode != NULL) {
 329                         __u64 ibits = MDS_INODELOCK_UPDATE;
 330                         struct inode *parent =
 331                                 filp->f_dentry->d_parent->d_inode;
 332
 333                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 334                                 op_data->op_fid3 = *ll_inode2fid(parent);
 335                 }
 336
 337                 /* If it can not find in cache, do lookup .. on the master
 338                  * object */
 339                 if (fid_is_zero(&op_data->op_fid3)) {
 340                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 341                         if (rc != 0)
 342                                 RETURN(rc);
 343                 }
 344         }
 345         op_data->op_hash_offset = pos;
 346         op_data->op_max_pages = sbi->ll_md_brw_pages;
 347 #ifdef HAVE_DIR_CONTEXT
 348         ctx->pos = pos;
 349         rc = ll_dir_read(inode, op_data, ctx);
 350         pos = ctx->pos;
 351 #else
 352         rc = ll_dir_read(inode, op_data, cookie, filldir);
 353 #endif
 354         if (lfd != NULL)
 355                 lfd->lfd_pos = op_data->op_hash_offset;
 356
 357         if (pos == MDS_DIR_END_OFF) {
 358                 if (api32)
 359                         pos = LL_DIR_END_OFF_32BIT;
 360                 else
 361                         pos = LL_DIR_END_OFF;
 362         } else {
 363                 if (api32 && hash64)
 364                         pos = op_data->op_hash_offset >> 32;
 365                 else
 366                         pos = op_data->op_hash_offset;
 367         }
 368 #ifdef HAVE_DIR_CONTEXT
 369         ctx->pos = pos;
 370 #else
 371         filp->f_pos = pos;
 372 #endif
 373         ll_finish_md_op_data(op_data);
 374         filp->f_version = inode->i_version;
 375
 376 out:
 377         if (!rc)
 378                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 379
 380         RETURN(rc);
 381 }
 382
 383 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 384 {
 385         struct mgs_send_param *msp;
 386         int rc = 0;
 387
 388         OBD_ALLOC_PTR(msp);
 389         if (!msp)
 390                 return -ENOMEM;
 391
 392         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 393         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 394                                 sizeof(struct mgs_send_param), msp, NULL);
 395         if (rc)
 396                 CERROR("Failed to set parameter: %d\n", rc);
 397         OBD_FREE_PTR(msp);
 398
 399         return rc;
 400 }
 401
 402 /**
 403  * Create striped directory with specified stripe(@lump)
 404  *
 405  * param[in]parent      the parent of the directory.
 406  * param[in]lump        the specified stripes.
 407  * param[in]dirname     the name of the directory.
 408  * param[in]mode        the specified mode of the directory.
 409  *
 410  * retval               =0 if striped directory is being created successfully.
 411  *                      <0 if the creation is failed.
 412  */
 413 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
 414                                const char *dirname, umode_t mode)
 415 {
 416         struct ptlrpc_request *request = NULL;
 417         struct md_op_data *op_data;
 418         struct ll_sb_info *sbi = ll_i2sbi(parent);
 419         int err;
 420         ENTRY;
 421
 422         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 423                 RETURN(-EINVAL);
 424
 425         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 426                "stripe_offset %d, stripe_count: %u\n",
 427                PFID(ll_inode2fid(parent)), parent, dirname,
 428                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 429
 430         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 431                 lustre_swab_lmv_user_md(lump);
 432
 433         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 434                 mode &= ~current_umask();
 435         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 436         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 437                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 438                                      lump);
 439         if (IS_ERR(op_data))
 440                 GOTO(err_exit, err = PTR_ERR(op_data));
 441
 442         op_data->op_cli_flags |= CLI_SET_MEA;
 443         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 444                         from_kuid(&init_user_ns, current_fsuid()),
 445                         from_kgid(&init_user_ns, current_fsgid()),
 446                         cfs_curproc_cap_pack(), 0, &request);
 447         ll_finish_md_op_data(op_data);
 448         if (err)
 449                 GOTO(err_exit, err);
 450 err_exit:
 451         ptlrpc_req_finished(request);
 452         return err;
 453 }
 454
 455 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 456                      int set_default)
 457 {
 458         struct ll_sb_info *sbi = ll_i2sbi(inode);
 459         struct md_op_data *op_data;
 460         struct ptlrpc_request *req = NULL;
 461         int rc = 0;
 462         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 463         struct obd_device *mgc = lsi->lsi_mgc;
 464         int lum_size;
 465         ENTRY;
 466
 467         if (lump != NULL) {
 468                 /*
 469                  * This is coming from userspace, so should be in
 470                  * local endian.  But the MDS would like it in little
 471                  * endian, so we swab it before we send it.
 472                  */
 473                 switch (lump->lmm_magic) {
 474                 case LOV_USER_MAGIC_V1: {
 475                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 476                                 lustre_swab_lov_user_md_v1(lump);
 477                         lum_size = sizeof(struct lov_user_md_v1);
 478                         break;
 479                 }
 480                 case LOV_USER_MAGIC_V3: {
 481                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 482                                 lustre_swab_lov_user_md_v3(
 483                                         (struct lov_user_md_v3 *)lump);
 484                         lum_size = sizeof(struct lov_user_md_v3);
 485                         break;
 486                 }
 487                 case LMV_USER_MAGIC: {
 488                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 489                                 lustre_swab_lmv_user_md(
 490                                         (struct lmv_user_md *)lump);
 491                         lum_size = sizeof(struct lmv_user_md);
 492                         break;
 493                 }
 494                 default: {
 495                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 496                                         " %#08x != %#08x nor %#08x\n",
 497                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 498                                         LOV_USER_MAGIC_V3);
 499                         RETURN(-EINVAL);
 500                 }
 501                 }
 502         } else {
 503                 lum_size = sizeof(struct lov_user_md_v1);
 504         }
 505
 506         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 507                                      LUSTRE_OPC_ANY, NULL);
 508         if (IS_ERR(op_data))
 509                 RETURN(PTR_ERR(op_data));
 510
 511         /* swabbing is done in lov_setstripe() on server side */
 512         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 513                         NULL, 0, &req, NULL);
 514         ll_finish_md_op_data(op_data);
 515         ptlrpc_req_finished(req);
 516         if (rc) {
 517                 if (rc != -EPERM && rc != -EACCES)
 518                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 519         }
 520
 521         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 522          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 523          need the make the distiction between the 2 versions */
 524         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 525                 char *param = NULL;
 526                 char *buf;
 527
 528                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 529                 if (param == NULL)
 530                         GOTO(end, rc = -ENOMEM);
 531
 532                 buf = param;
 533                 /* Get fsname and assume devname to be -MDT0000. */
 534                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 535                 strcat(buf, "-MDT0000.lov");
 536                 buf += strlen(buf);
 537
 538                 /* Set root stripesize */
 539                 sprintf(buf, ".stripesize=%u",
 540                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 541                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 542                 if (rc)
 543                         GOTO(end, rc);
 544
 545                 /* Set root stripecount */
 546                 sprintf(buf, ".stripecount=%hd",
 547                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 548                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 549                 if (rc)
 550                         GOTO(end, rc);
 551
 552                 /* Set root stripeoffset */
 553                 sprintf(buf, ".stripeoffset=%hd",
 554                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 555                         (typeof(lump->lmm_stripe_offset))(-1));
 556                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 557
 558 end:
 559                 if (param != NULL)
 560                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 561         }
 562         RETURN(rc);
 563 }
 564
 565 /**
 566  * This function will be used to get default LOV/LMV/Default LMV
 567  * @valid will be used to indicate which stripe it will retrieve
 568  *      OBD_MD_MEA              LMV stripe EA
 569  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 570  *      otherwise               Default LOV EA.
 571  * Each time, it can only retrieve 1 stripe EA
 572  **/
 573 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 574                      struct ptlrpc_request **request, obd_valid valid)
 575 {
 576         struct ll_sb_info *sbi = ll_i2sbi(inode);
 577         struct mdt_body   *body;
 578         struct lov_mds_md *lmm = NULL;
 579         struct ptlrpc_request *req = NULL;
 580         int rc, lmm_size;
 581         struct md_op_data *op_data;
 582         ENTRY;
 583
 584         rc = ll_get_default_mdsize(sbi, &lmm_size);
 585         if (rc)
 586                 RETURN(rc);
 587
 588         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 589                                      0, lmm_size, LUSTRE_OPC_ANY,
 590                                      NULL);
 591         if (IS_ERR(op_data))
 592                 RETURN(PTR_ERR(op_data));
 593
 594         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 595         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 596         ll_finish_md_op_data(op_data);
 597         if (rc < 0) {
 598                 CDEBUG(D_INFO, "md_getattr failed on inode "
 599                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 600                 GOTO(out, rc);
 601         }
 602
 603         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 604         LASSERT(body != NULL);
 605
 606         lmm_size = body->mbo_eadatasize;
 607
 608         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 609             lmm_size == 0) {
 610                 GOTO(out, rc = -ENODATA);
 611         }
 612
 613         lmm = req_capsule_server_sized_get(&req->rq_pill,
 614                                            &RMF_MDT_MD, lmm_size);
 615         LASSERT(lmm != NULL);
 616
 617         /*
 618          * This is coming from the MDS, so is probably in
 619          * little endian.  We convert it to host endian before
 620          * passing it to userspace.
 621          */
 622         /* We don't swab objects for directories */
 623         switch (le32_to_cpu(lmm->lmm_magic)) {
 624         case LOV_MAGIC_V1:
 625                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 626                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 627                 break;
 628         case LOV_MAGIC_V3:
 629                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 630                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 631                 break;
 632         case LMV_MAGIC_V1:
 633                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 634                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 635                 break;
 636         case LMV_USER_MAGIC:
 637                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 638                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 639                 break;
 640         default:
 641                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 642                 rc = -EPROTO;
 643         }
 644 out:
 645         *plmm = lmm;
 646         *plmm_size = lmm_size;
 647         *request = req;
 648         return rc;
 649 }
 650
 651 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 652 {
 653         struct md_op_data       *op_data;
 654         int                     rc;
 655         int                     mdt_index;
 656         ENTRY;
 657
 658         OBD_ALLOC_PTR(op_data);
 659         if (op_data == NULL)
 660                 RETURN(-ENOMEM);
 661
 662         op_data->op_flags |= MF_GET_MDT_IDX;
 663         op_data->op_fid1 = *fid;
 664         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 665         mdt_index = op_data->op_mds;
 666         OBD_FREE_PTR(op_data);
 667         if (rc < 0)
 668                 RETURN(rc);
 669
 670         RETURN(mdt_index);
 671 }
 672
 673 /*
 674  *  Get MDT index for the inode.
 675  */
 676 int ll_get_mdt_idx(struct inode *inode)
 677 {
 678         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 679 }
 680
 681 /**
 682  * Generic handler to do any pre-copy work.
 683  *
 684  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 685  * first information for it that real work has started.
 686  *
 687  * Moreover, for a ARCHIVE request, it will sample the file data version and
 688  * store it in \a copy.
 689  *
 690  * \return 0 on success.
 691  */
 692 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 693 {
 694         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 695         struct hsm_progress_kernel       hpk;
 696         int                              rc;
 697         ENTRY;
 698
 699         /* Forge a hsm_progress based on data from copy. */
 700         hpk.hpk_fid = copy->hc_hai.hai_fid;
 701         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 702         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 703         hpk.hpk_extent.length = 0;
 704         hpk.hpk_flags = 0;
 705         hpk.hpk_errval = 0;
 706         hpk.hpk_data_version = 0;
 707
 708
 709         /* For archive request, we need to read the current file version. */
 710         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 711                 struct inode    *inode;
 712                 __u64            data_version = 0;
 713
 714                 /* Get inode for this fid */
 715                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 716                 if (IS_ERR(inode)) {
 717                         hpk.hpk_flags |= HP_FLAG_RETRY;
 718                         /* hpk_errval is >= 0 */
 719                         hpk.hpk_errval = -PTR_ERR(inode);
 720                         GOTO(progress, rc = PTR_ERR(inode));
 721                 }
 722
 723                 /* Read current file data version */
 724                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 725                 iput(inode);
 726                 if (rc != 0) {
 727                         CDEBUG(D_HSM, "Could not read file data version of "
 728                                       DFID" (rc = %d). Archive request ("
 729                                       LPX64") could not be done.\n",
 730                                       PFID(&copy->hc_hai.hai_fid), rc,
 731                                       copy->hc_hai.hai_cookie);
 732                         hpk.hpk_flags |= HP_FLAG_RETRY;
 733                         /* hpk_errval must be >= 0 */
 734                         hpk.hpk_errval = -rc;
 735                         GOTO(progress, rc);
 736                 }
 737
 738                 /* Store it the hsm_copy for later copytool use.
 739                  * Always modified even if no lsm. */
 740                 copy->hc_data_version = data_version;
 741         }
 742
 743 progress:
 744         /* On error, the request should be considered as completed */
 745         if (hpk.hpk_errval > 0)
 746                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 747         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 748                            &hpk, NULL);
 749
 750         RETURN(rc);
 751 }
 752
 753 /**
 754  * Generic handler to do any post-copy work.
 755  *
 756  * It will send the last hsm_progress update to coordinator to inform it
 757  * that copy is finished and whether it was successful or not.
 758  *
 759  * Moreover,
 760  * - for ARCHIVE request, it will sample the file data version and compare it
 761  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 762  *   will be considered as failed.
 763  * - for RESTORE request, it will sample the file data version and send it to
 764  *   coordinator which is useful if the file was imported as 'released'.
 765  *
 766  * \return 0 on success.
 767  */
 768 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 769 {
 770         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 771         struct hsm_progress_kernel       hpk;
 772         int                              rc;
 773         ENTRY;
 774
 775         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 776         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 777          * initialized if copy_end was called with copy == NULL.
 778          */
 779
 780         /* Forge a hsm_progress based on data from copy. */
 781         hpk.hpk_fid = copy->hc_hai.hai_fid;
 782         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 783         hpk.hpk_extent = copy->hc_hai.hai_extent;
 784         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 785         hpk.hpk_errval = copy->hc_errval;
 786         hpk.hpk_data_version = 0;
 787
 788         /* For archive request, we need to check the file data was not changed.
 789          *
 790          * For restore request, we need to send the file data version, this is
 791          * useful when the file was created using hsm_import.
 792          */
 793         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 794              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 795             (copy->hc_errval == 0)) {
 796                 struct inode    *inode;
 797                 __u64            data_version = 0;
 798
 799                 /* Get lsm for this fid */
 800                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 801                 if (IS_ERR(inode)) {
 802                         hpk.hpk_flags |= HP_FLAG_RETRY;
 803                         /* hpk_errval must be >= 0 */
 804                         hpk.hpk_errval = -PTR_ERR(inode);
 805                         GOTO(progress, rc = PTR_ERR(inode));
 806                 }
 807
 808                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 809                 iput(inode);
 810                 if (rc) {
 811                         CDEBUG(D_HSM, "Could not read file data version. "
 812                                       "Request could not be confirmed.\n");
 813                         if (hpk.hpk_errval == 0)
 814                                 hpk.hpk_errval = -rc;
 815                         GOTO(progress, rc);
 816                 }
 817
 818                 /* Store it the hsm_copy for later copytool use.
 819                  * Always modified even if no lsm. */
 820                 hpk.hpk_data_version = data_version;
 821
 822                 /* File could have been stripped during archiving, so we need
 823                  * to check anyway. */
 824                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 825                     (copy->hc_data_version != data_version)) {
 826                         CDEBUG(D_HSM, "File data version mismatched. "
 827                               "File content was changed during archiving. "
 828                                DFID", start:"LPX64" current:"LPX64"\n",
 829                                PFID(&copy->hc_hai.hai_fid),
 830                                copy->hc_data_version, data_version);
 831                         /* File was changed, send error to cdt. Do not ask for
 832                          * retry because if a file is modified frequently,
 833                          * the cdt will loop on retried archive requests.
 834                          * The policy engine will ask for a new archive later
 835                          * when the file will not be modified for some tunable
 836                          * time */
 837                         /* we do not notify caller */
 838                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 839                         /* hpk_errval must be >= 0 */
 840                         hpk.hpk_errval = EBUSY;
 841                 }
 842
 843         }
 844
 845 progress:
 846         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 847                            &hpk, NULL);
 848
 849         RETURN(rc);
 850 }
 851
 852
 853 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 854                           const void __user *data, size_t size)
 855 {
 856         void *copy;
 857         int rc;
 858
 859         OBD_ALLOC(copy, size);
 860         if (copy == NULL)
 861                 return -ENOMEM;
 862
 863         if (copy_from_user(copy, data, size)) {
 864                 rc = -EFAULT;
 865                 goto out;
 866         }
 867
 868         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 869 out:
 870         OBD_FREE(copy, size);
 871
 872         return rc;
 873 }
 874
 875 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 876 {
 877         int cmd = qctl->qc_cmd;
 878         int type = qctl->qc_type;
 879         int id = qctl->qc_id;
 880         int valid = qctl->qc_valid;
 881         int rc = 0;
 882         ENTRY;
 883
 884         switch (cmd) {
 885         case LUSTRE_Q_INVALIDATE:
 886         case LUSTRE_Q_FINVALIDATE:
 887         case Q_QUOTAON:
 888         case Q_QUOTAOFF:
 889         case Q_SETQUOTA:
 890         case Q_SETINFO:
 891                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 892                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 893                         RETURN(-EPERM);
 894                 break;
 895         case Q_GETQUOTA:
 896                 if (((type == USRQUOTA &&
 897                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 898                      (type == GRPQUOTA &&
 899                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 900                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 901                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 902                         RETURN(-EPERM);
 903                 break;
 904         case Q_GETINFO:
 905                 break;
 906         default:
 907                 CERROR("unsupported quotactl op: %#x\n", cmd);
 908                 RETURN(-ENOTTY);
 909         }
 910
 911         if (valid != QC_GENERAL) {
 912                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 913                         RETURN(-EOPNOTSUPP);
 914
 915                 if (cmd == Q_GETINFO)
 916                         qctl->qc_cmd = Q_GETOINFO;
 917                 else if (cmd == Q_GETQUOTA)
 918                         qctl->qc_cmd = Q_GETOQUOTA;
 919                 else
 920                         RETURN(-EINVAL);
 921
 922                 switch (valid) {
 923                 case QC_MDTIDX:
 924                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 925                                            sizeof(*qctl), qctl, NULL);
 926                         break;
 927                 case QC_OSTIDX:
 928                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 929                                            sizeof(*qctl), qctl, NULL);
 930                         break;
 931                 case QC_UUID:
 932                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 933                                            sizeof(*qctl), qctl, NULL);
 934                         if (rc == -EAGAIN)
 935                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 936                                                    sbi->ll_dt_exp,
 937                                                    sizeof(*qctl), qctl, NULL);
 938                         break;
 939                 default:
 940                         rc = -EINVAL;
 941                         break;
 942                 }
 943
 944                 if (rc)
 945                         RETURN(rc);
 946
 947                 qctl->qc_cmd = cmd;
 948         } else {
 949                 struct obd_quotactl *oqctl;
 950
 951                 OBD_ALLOC_PTR(oqctl);
 952                 if (oqctl == NULL)
 953                         RETURN(-ENOMEM);
 954
 955                 QCTL_COPY(oqctl, qctl);
 956                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 957                 if (rc) {
 958                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 959                                 oqctl->qc_cmd = Q_QUOTAOFF;
 960                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 961                         }
 962                         OBD_FREE_PTR(oqctl);
 963                         RETURN(rc);
 964                 }
 965                 /* If QIF_SPACE is not set, client should collect the
 966                  * space usage from OSSs by itself */
 967                 if (cmd == Q_GETQUOTA &&
 968                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 969                     !oqctl->qc_dqblk.dqb_curspace) {
 970                         struct obd_quotactl *oqctl_tmp;
 971
 972                         OBD_ALLOC_PTR(oqctl_tmp);
 973                         if (oqctl_tmp == NULL)
 974                                 GOTO(out, rc = -ENOMEM);
 975
 976                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 977                         oqctl_tmp->qc_id = oqctl->qc_id;
 978                         oqctl_tmp->qc_type = oqctl->qc_type;
 979
 980                         /* collect space usage from OSTs */
 981                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 982                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
 983                         if (!rc || rc == -EREMOTEIO) {
 984                                 oqctl->qc_dqblk.dqb_curspace =
 985                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 986                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
 987                         }
 988
 989                         /* collect space & inode usage from MDTs */
 990                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
 991                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
 992                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
 993                         if (!rc || rc == -EREMOTEIO) {
 994                                 oqctl->qc_dqblk.dqb_curspace +=
 995                                         oqctl_tmp->qc_dqblk.dqb_curspace;
 996                                 oqctl->qc_dqblk.dqb_curinodes =
 997                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
 998                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
 999                         } else {
1000                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1001                         }
1002
1003                         OBD_FREE_PTR(oqctl_tmp);
1004                 }
1005 out:
1006                 QCTL_COPY(qctl, oqctl);
1007                 OBD_FREE_PTR(oqctl);
1008         }
1009
1010         RETURN(rc);
1011 }
1012
1013 static char *
1014 ll_getname(const char __user *filename)
1015 {
1016         int ret = 0, len;
1017         char *tmp = __getname();
1018
1019         if (!tmp)
1020                 return ERR_PTR(-ENOMEM);
1021
1022         len = strncpy_from_user(tmp, filename, PATH_MAX);
1023         if (len == 0)
1024                 ret = -ENOENT;
1025         else if (len > PATH_MAX)
1026                 ret = -ENAMETOOLONG;
1027
1028         if (ret) {
1029                 __putname(tmp);
1030                 tmp =  ERR_PTR(ret);
1031         }
1032         return tmp;
1033 }
1034
1035 #define ll_putname(filename) __putname(filename)
1036
1037 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1038 {
1039         struct inode *inode = file->f_dentry->d_inode;
1040         struct ll_sb_info *sbi = ll_i2sbi(inode);
1041         struct obd_ioctl_data *data;
1042         int rc = 0;
1043         ENTRY;
1044
1045         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1046                PFID(ll_inode2fid(inode)), inode, cmd);
1047
1048         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1049         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1050                 return -ENOTTY;
1051
1052         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1053         switch(cmd) {
1054         case FSFILT_IOC_GETFLAGS:
1055         case FSFILT_IOC_SETFLAGS:
1056                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1057         case FSFILT_IOC_GETVERSION_OLD:
1058         case FSFILT_IOC_GETVERSION:
1059                 RETURN(put_user(inode->i_generation, (int *)arg));
1060         /* We need to special case any other ioctls we want to handle,
1061          * to send them to the MDS/OST as appropriate and to properly
1062          * network encode the arg field.
1063         case FSFILT_IOC_SETVERSION_OLD:
1064         case FSFILT_IOC_SETVERSION:
1065         */
1066         case LL_IOC_GET_MDTIDX: {
1067                 int mdtidx;
1068
1069                 mdtidx = ll_get_mdt_idx(inode);
1070                 if (mdtidx < 0)
1071                         RETURN(mdtidx);
1072
1073                 if (put_user((int)mdtidx, (int*)arg))
1074                         RETURN(-EFAULT);
1075
1076                 return 0;
1077         }
1078         case IOC_MDC_LOOKUP: {
1079                 int namelen, len = 0;
1080                 char *buf = NULL;
1081                 char *filename;
1082
1083                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1084                 if (rc != 0)
1085                         RETURN(rc);
1086                 data = (void *)buf;
1087
1088                 filename = data->ioc_inlbuf1;
1089                 namelen = strlen(filename);
1090                 if (namelen < 1) {
1091                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1092                         GOTO(out_free, rc = -EINVAL);
1093                 }
1094
1095                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1096                 if (rc < 0) {
1097                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1098                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1099                                filename, rc);
1100                         GOTO(out_free, rc);
1101                 }
1102 out_free:
1103                 obd_ioctl_freedata(buf, len);
1104                 return rc;
1105         }
1106         case LL_IOC_LMV_SETSTRIPE: {
1107                 struct lmv_user_md  *lum;
1108                 char            *buf = NULL;
1109                 char            *filename;
1110                 int              namelen = 0;
1111                 int              lumlen = 0;
1112                 umode_t          mode;
1113                 int              len;
1114                 int              rc;
1115
1116                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1117                 if (rc)
1118                         RETURN(rc);
1119
1120                 data = (void *)buf;
1121                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1122                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1123                         GOTO(lmv_out_free, rc = -EINVAL);
1124
1125                 filename = data->ioc_inlbuf1;
1126                 namelen = data->ioc_inllen1;
1127
1128                 if (namelen < 1) {
1129                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1130                         GOTO(lmv_out_free, rc = -EINVAL);
1131                 }
1132                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1133                 lumlen = data->ioc_inllen2;
1134
1135                 if (lum->lum_magic != LMV_USER_MAGIC ||
1136                     lumlen != sizeof(*lum)) {
1137                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1138                                filename, lum->lum_magic, lumlen, -EFAULT);
1139                         GOTO(lmv_out_free, rc = -EINVAL);
1140                 }
1141
1142 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1143                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1144 #else
1145                 mode = data->ioc_type;
1146 #endif
1147                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1148 lmv_out_free:
1149                 obd_ioctl_freedata(buf, len);
1150                 RETURN(rc);
1151
1152         }
1153         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1154                 struct lmv_user_md        lum;
1155                 struct lmv_user_md __user *ulump =
1156                                         (struct lmv_user_md __user *)arg;
1157                 int                       rc;
1158
1159                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1160                         RETURN(-EFAULT);
1161
1162                 if (lum.lum_magic != LMV_USER_MAGIC)
1163                         RETURN(-EINVAL);
1164
1165                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1166
1167                 RETURN(rc);
1168         }
1169         case LL_IOC_LOV_SETSTRIPE: {
1170                 struct lov_user_md_v3 lumv3;
1171                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1172                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1173                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1174
1175                 int set_default = 0;
1176
1177                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1178                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1179                         sizeof(lumv3p->lmm_objects[0]));
1180                 /* first try with v1 which is smaller than v3 */
1181                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1182                         RETURN(-EFAULT);
1183
1184                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1185                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1186                                 RETURN(-EFAULT);
1187                 }
1188
1189                 if (inode->i_sb->s_root == file->f_dentry)
1190                         set_default = 1;
1191
1192                 /* in v1 and v3 cases lumv1 points to data */
1193                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1194
1195                 RETURN(rc);
1196         }
1197         case LL_IOC_LMV_GETSTRIPE: {
1198                 struct lmv_user_md __user *ulmv =
1199                                         (struct lmv_user_md __user *)arg;
1200                 struct lmv_user_md      lum;
1201                 struct ptlrpc_request   *request = NULL;
1202                 union lmv_mds_md        *lmm = NULL;
1203                 int                     lmmsize;
1204                 obd_valid               valid = 0;
1205                 struct lmv_user_md      *tmp = NULL;
1206                 int                     mdt_index;
1207                 int                     lum_size;
1208                 int                     stripe_count;
1209                 int                     i;
1210                 int                     rc;
1211
1212                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1213                         RETURN(-EFAULT);
1214
1215                 /* lum_magic will indicate which stripe the ioctl will like
1216                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1217                  * is for default LMV stripe */
1218                 if (lum.lum_magic == LMV_MAGIC_V1)
1219                         valid |= OBD_MD_MEA;
1220                 else if (lum.lum_magic == LMV_USER_MAGIC)
1221                         valid |= OBD_MD_DEFAULT_MEA;
1222                 else
1223                         RETURN(-EINVAL);
1224
1225                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1226                                       valid);
1227                 if (rc != 0)
1228                         GOTO(finish_req, rc);
1229
1230                 /* Get default LMV EA */
1231                 if (lum.lum_magic == LMV_USER_MAGIC) {
1232                         if (rc != 0)
1233                                 GOTO(finish_req, rc);
1234
1235                         if (lmmsize > sizeof(*ulmv))
1236                                 GOTO(finish_req, rc = -EINVAL);
1237
1238                         if (copy_to_user(ulmv, lmm, lmmsize))
1239                                 GOTO(finish_req, rc = -EFAULT);
1240
1241                         GOTO(finish_req, rc);
1242                 }
1243
1244                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1245                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1246                 OBD_ALLOC(tmp, lum_size);
1247                 if (tmp == NULL)
1248                         GOTO(finish_req, rc = -ENOMEM);
1249
1250                 mdt_index = ll_get_mdt_idx(inode);
1251                 if (mdt_index < 0)
1252                         GOTO(out_tmp, rc = -ENOMEM);
1253
1254                 tmp->lum_magic = LMV_MAGIC_V1;
1255                 tmp->lum_stripe_count = 0;
1256                 tmp->lum_stripe_offset = mdt_index;
1257                 for (i = 0; i < stripe_count; i++) {
1258                         struct lu_fid   *fid;
1259
1260                         fid = &lmm->lmv_md_v1.lmv_stripe_fids[i];
1261                         mdt_index = ll_get_mdt_idx_by_fid(sbi, fid);
1262                         if (mdt_index < 0)
1263                                 GOTO(out_tmp, rc = mdt_index);
1264
1265                         tmp->lum_objects[i].lum_mds = mdt_index;
1266                         tmp->lum_objects[i].lum_fid = *fid;
1267                         tmp->lum_stripe_count++;
1268                 }
1269
1270                 if (copy_to_user(ulmv, tmp, lum_size))
1271                         GOTO(out_tmp, rc = -EFAULT);
1272 out_tmp:
1273                 OBD_FREE(tmp, lum_size);
1274 finish_req:
1275                 ptlrpc_req_finished(request);
1276                 return rc;
1277         }
1278
1279         case LL_IOC_REMOVE_ENTRY: {
1280                 char            *filename = NULL;
1281                 int              namelen = 0;
1282                 int              rc;
1283
1284                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1285                  * unsupported server, which might crash the server(LU-2730),
1286                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1287                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1288                  * server will support REINT_RMENTRY XXX*/
1289                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1290                         RETURN(-ENOTSUPP);
1291
1292                 filename = ll_getname((const char *)arg);
1293                 if (IS_ERR(filename))
1294                         RETURN(PTR_ERR(filename));
1295
1296                 namelen = strlen(filename);
1297                 if (namelen < 1)
1298                         GOTO(out_rmdir, rc = -EINVAL);
1299
1300                 rc = ll_rmdir_entry(inode, filename, namelen);
1301 out_rmdir:
1302                 if (filename)
1303                         ll_putname(filename);
1304                 RETURN(rc);
1305         }
1306         case LL_IOC_LOV_SWAP_LAYOUTS:
1307                 RETURN(-EPERM);
1308         case IOC_OBD_STATFS:
1309                 RETURN(ll_obd_statfs(inode, (void *)arg));
1310         case LL_IOC_LOV_GETSTRIPE:
1311         case LL_IOC_MDC_GETINFO:
1312         case IOC_MDC_GETFILEINFO:
1313         case IOC_MDC_GETFILESTRIPE: {
1314                 struct ptlrpc_request *request = NULL;
1315                 struct lov_user_md *lump;
1316                 struct lov_mds_md *lmm = NULL;
1317                 struct mdt_body *body;
1318                 char *filename = NULL;
1319                 int lmmsize;
1320
1321                 if (cmd == IOC_MDC_GETFILEINFO ||
1322                     cmd == IOC_MDC_GETFILESTRIPE) {
1323                         filename = ll_getname((const char *)arg);
1324                         if (IS_ERR(filename))
1325                                 RETURN(PTR_ERR(filename));
1326
1327                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1328                                                       &lmmsize, &request);
1329                 } else {
1330                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1331                                               &request, 0);
1332                 }
1333
1334                 if (request) {
1335                         body = req_capsule_server_get(&request->rq_pill,
1336                                                       &RMF_MDT_BODY);
1337                         LASSERT(body != NULL);
1338                 } else {
1339                         GOTO(out_req, rc);
1340                 }
1341
1342                 if (rc < 0) {
1343                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1344                                                cmd == LL_IOC_MDC_GETINFO))
1345                                 GOTO(skip_lmm, rc = 0);
1346                         else
1347                                 GOTO(out_req, rc);
1348                 }
1349
1350                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1351                     cmd == LL_IOC_LOV_GETSTRIPE) {
1352                         lump = (struct lov_user_md *)arg;
1353                 } else {
1354                         struct lov_user_mds_data *lmdp;
1355                         lmdp = (struct lov_user_mds_data *)arg;
1356                         lump = &lmdp->lmd_lmm;
1357                 }
1358                 if (copy_to_user(lump, lmm, lmmsize)) {
1359                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1360                                 GOTO(out_req, rc = -EFAULT);
1361                         rc = -EOVERFLOW;
1362                 }
1363         skip_lmm:
1364                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1365                         struct lov_user_mds_data *lmdp;
1366                         lstat_t st = { 0 };
1367
1368                         st.st_dev       = inode->i_sb->s_dev;
1369                         st.st_mode      = body->mbo_mode;
1370                         st.st_nlink     = body->mbo_nlink;
1371                         st.st_uid       = body->mbo_uid;
1372                         st.st_gid       = body->mbo_gid;
1373                         st.st_rdev      = body->mbo_rdev;
1374                         st.st_size      = body->mbo_size;
1375                         st.st_blksize   = PAGE_CACHE_SIZE;
1376                         st.st_blocks    = body->mbo_blocks;
1377                         st.st_atime     = body->mbo_atime;
1378                         st.st_mtime     = body->mbo_mtime;
1379                         st.st_ctime     = body->mbo_ctime;
1380                         st.st_ino       = inode->i_ino;
1381
1382                         lmdp = (struct lov_user_mds_data *)arg;
1383                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1384                                 GOTO(out_req, rc = -EFAULT);
1385                 }
1386
1387                 EXIT;
1388         out_req:
1389                 ptlrpc_req_finished(request);
1390                 if (filename)
1391                         ll_putname(filename);
1392                 return rc;
1393         }
1394         case IOC_LOV_GETINFO: {
1395                 struct lov_user_mds_data *lumd;
1396                 struct lov_stripe_md *lsm;
1397                 struct lov_user_md *lum;
1398                 struct lov_mds_md *lmm;
1399                 int lmmsize;
1400                 lstat_t st;
1401
1402                 lumd = (struct lov_user_mds_data *)arg;
1403                 lum = &lumd->lmd_lmm;
1404
1405                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1406                 if (rc)
1407                         RETURN(rc);
1408
1409                 OBD_ALLOC_LARGE(lmm, lmmsize);
1410                 if (lmm == NULL)
1411                         RETURN(-ENOMEM);
1412
1413                 if (copy_from_user(lmm, lum, lmmsize))
1414                         GOTO(free_lmm, rc = -EFAULT);
1415
1416                 switch (lmm->lmm_magic) {
1417                 case LOV_USER_MAGIC_V1:
1418                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1419                                 break;
1420                         /* swab objects first so that stripes num will be sane */
1421                         lustre_swab_lov_user_md_objects(
1422                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1423                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1424                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1425                         break;
1426                 case LOV_USER_MAGIC_V3:
1427                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1428                                 break;
1429                         /* swab objects first so that stripes num will be sane */
1430                         lustre_swab_lov_user_md_objects(
1431                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1432                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1433                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1434                         break;
1435                 default:
1436                         GOTO(free_lmm, rc = -EINVAL);
1437                 }
1438
1439                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1440                 if (rc < 0)
1441                         GOTO(free_lmm, rc = -ENOMEM);
1442
1443                 /* Perform glimpse_size operation. */
1444                 memset(&st, 0, sizeof(st));
1445
1446                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1447                 if (rc)
1448                         GOTO(free_lsm, rc);
1449
1450                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1451                         GOTO(free_lsm, rc = -EFAULT);
1452
1453                 EXIT;
1454         free_lsm:
1455                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1456         free_lmm:
1457                 OBD_FREE_LARGE(lmm, lmmsize);
1458                 return rc;
1459         }
1460         case OBD_IOC_QUOTACHECK: {
1461                 struct obd_quotactl *oqctl;
1462                 int error = 0;
1463
1464                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1465                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1466                         RETURN(-EPERM);
1467
1468                 OBD_ALLOC_PTR(oqctl);
1469                 if (!oqctl)
1470                         RETURN(-ENOMEM);
1471                 oqctl->qc_type = arg;
1472                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1473                 if (rc < 0) {
1474                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1475                         error = rc;
1476                 }
1477
1478                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1479                 if (rc < 0)
1480                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1481
1482                 OBD_FREE_PTR(oqctl);
1483                 return error ?: rc;
1484         }
1485         case OBD_IOC_POLL_QUOTACHECK: {
1486                 struct if_quotacheck *check;
1487
1488                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1489                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1490                         RETURN(-EPERM);
1491
1492                 OBD_ALLOC_PTR(check);
1493                 if (!check)
1494                         RETURN(-ENOMEM);
1495
1496                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1497                                    NULL);
1498                 if (rc) {
1499                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1500                         if (copy_to_user((void *)arg, check,
1501                                              sizeof(*check)))
1502                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1503                         GOTO(out_poll, rc);
1504                 }
1505
1506                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1507                                    NULL);
1508                 if (rc) {
1509                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1510                         if (copy_to_user((void *)arg, check,
1511                                              sizeof(*check)))
1512                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1513                         GOTO(out_poll, rc);
1514                 }
1515         out_poll:
1516                 OBD_FREE_PTR(check);
1517                 RETURN(rc);
1518         }
1519 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1520         case LL_IOC_QUOTACTL_18: {
1521                 /* copy the old 1.x quota struct for internal use, then copy
1522                  * back into old format struct.  For 1.8 compatibility. */
1523                 struct if_quotactl_18 *qctl_18;
1524                 struct if_quotactl *qctl_20;
1525
1526                 OBD_ALLOC_PTR(qctl_18);
1527                 if (!qctl_18)
1528                         RETURN(-ENOMEM);
1529
1530                 OBD_ALLOC_PTR(qctl_20);
1531                 if (!qctl_20)
1532                         GOTO(out_quotactl_18, rc = -ENOMEM);
1533
1534                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1535                         GOTO(out_quotactl_20, rc = -ENOMEM);
1536
1537                 QCTL_COPY(qctl_20, qctl_18);
1538                 qctl_20->qc_idx = 0;
1539
1540                 /* XXX: dqb_valid was borrowed as a flag to mark that
1541                  *      only mds quota is wanted */
1542                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1543                     qctl_18->qc_dqblk.dqb_valid) {
1544                         qctl_20->qc_valid = QC_MDTIDX;
1545                         qctl_20->qc_dqblk.dqb_valid = 0;
1546                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1547                         qctl_20->qc_valid = QC_UUID;
1548                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1549                 } else {
1550                         qctl_20->qc_valid = QC_GENERAL;
1551                 }
1552
1553                 rc = quotactl_ioctl(sbi, qctl_20);
1554
1555                 if (rc == 0) {
1556                         QCTL_COPY(qctl_18, qctl_20);
1557                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1558
1559                         if (copy_to_user((void *)arg, qctl_18,
1560                                              sizeof(*qctl_18)))
1561                                 rc = -EFAULT;
1562                 }
1563
1564         out_quotactl_20:
1565                 OBD_FREE_PTR(qctl_20);
1566         out_quotactl_18:
1567                 OBD_FREE_PTR(qctl_18);
1568                 RETURN(rc);
1569         }
1570 #else
1571 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1572 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1573         case OBD_IOC_QUOTACTL: {
1574                 struct if_quotactl *qctl;
1575
1576                 OBD_ALLOC_PTR(qctl);
1577                 if (!qctl)
1578                         RETURN(-ENOMEM);
1579
1580                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1581                         GOTO(out_quotactl, rc = -EFAULT);
1582
1583                 rc = quotactl_ioctl(sbi, qctl);
1584
1585                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1586                         rc = -EFAULT;
1587
1588         out_quotactl:
1589                 OBD_FREE_PTR(qctl);
1590                 RETURN(rc);
1591         }
1592         case OBD_IOC_GETDTNAME:
1593         case OBD_IOC_GETMDNAME:
1594                 RETURN(ll_get_obd_name(inode, cmd, arg));
1595         case LL_IOC_FLUSHCTX:
1596                 RETURN(ll_flush_ctx(inode));
1597 #ifdef CONFIG_FS_POSIX_ACL
1598         case LL_IOC_RMTACL: {
1599             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1600                 inode == inode->i_sb->s_root->d_inode) {
1601                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1602
1603                 LASSERT(fd != NULL);
1604                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1605                 if (!rc)
1606                         fd->fd_flags |= LL_FILE_RMTACL;
1607                 RETURN(rc);
1608             } else
1609                 RETURN(0);
1610         }
1611 #endif
1612         case LL_IOC_GETOBDCOUNT: {
1613                 int count, vallen;
1614                 struct obd_export *exp;
1615
1616                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1617                         RETURN(-EFAULT);
1618
1619                 /* get ost count when count is zero, get mdt count otherwise */
1620                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1621                 vallen = sizeof(count);
1622                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1623                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1624                 if (rc) {
1625                         CERROR("get target count failed: %d\n", rc);
1626                         RETURN(rc);
1627                 }
1628
1629                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1630                         RETURN(-EFAULT);
1631
1632                 RETURN(0);
1633         }
1634         case LL_IOC_PATH2FID:
1635                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1636                                      sizeof(struct lu_fid)))
1637                         RETURN(-EFAULT);
1638                 RETURN(0);
1639         case LL_IOC_GET_CONNECT_FLAGS: {
1640                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1641         }
1642         case OBD_IOC_CHANGELOG_SEND:
1643         case OBD_IOC_CHANGELOG_CLEAR:
1644                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1645                                     sizeof(struct ioc_changelog));
1646                 RETURN(rc);
1647         case OBD_IOC_FID2PATH:
1648                 RETURN(ll_fid2path(inode, (void *)arg));
1649         case LL_IOC_HSM_REQUEST: {
1650                 struct hsm_user_request *hur;
1651                 int                      totalsize;
1652
1653                 OBD_ALLOC_PTR(hur);
1654                 if (hur == NULL)
1655                         RETURN(-ENOMEM);
1656
1657                 /* We don't know the true size yet; copy the fixed-size part */
1658                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1659                         OBD_FREE_PTR(hur);
1660                         RETURN(-EFAULT);
1661                 }
1662
1663                 /* Compute the whole struct size */
1664                 totalsize = hur_len(hur);
1665                 OBD_FREE_PTR(hur);
1666
1667                 /* Final size will be more than double totalsize */
1668                 if (totalsize >= MDS_MAXREQSIZE / 3)
1669                         RETURN(-E2BIG);
1670
1671                 OBD_ALLOC_LARGE(hur, totalsize);
1672                 if (hur == NULL)
1673                         RETURN(-ENOMEM);
1674
1675                 /* Copy the whole struct */
1676                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1677                         OBD_FREE_LARGE(hur, totalsize);
1678                         RETURN(-EFAULT);
1679                 }
1680
1681                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1682                         const struct lu_fid *fid;
1683                         struct inode *f;
1684                         int i;
1685
1686                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1687                                 fid = &hur->hur_user_item[i].hui_fid;
1688                                 f = search_inode_for_lustre(inode->i_sb, fid);
1689                                 if (IS_ERR(f)) {
1690                                         rc = PTR_ERR(f);
1691                                         break;
1692                                 }
1693
1694                                 rc = ll_hsm_release(f);
1695                                 iput(f);
1696                                 if (rc != 0)
1697                                         break;
1698                         }
1699                 } else {
1700                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1701                                            hur, NULL);
1702                 }
1703
1704                 OBD_FREE_LARGE(hur, totalsize);
1705
1706                 RETURN(rc);
1707         }
1708         case LL_IOC_HSM_PROGRESS: {
1709                 struct hsm_progress_kernel      hpk;
1710                 struct hsm_progress             hp;
1711
1712                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1713                         RETURN(-EFAULT);
1714
1715                 hpk.hpk_fid = hp.hp_fid;
1716                 hpk.hpk_cookie = hp.hp_cookie;
1717                 hpk.hpk_extent = hp.hp_extent;
1718                 hpk.hpk_flags = hp.hp_flags;
1719                 hpk.hpk_errval = hp.hp_errval;
1720                 hpk.hpk_data_version = 0;
1721
1722                 /* File may not exist in Lustre; all progress
1723                  * reported to Lustre root */
1724                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1725                                    NULL);
1726                 RETURN(rc);
1727         }
1728         case LL_IOC_HSM_CT_START:
1729                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1730                         RETURN(-EPERM);
1731
1732                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1733                                     sizeof(struct lustre_kernelcomm));
1734                 RETURN(rc);
1735
1736         case LL_IOC_HSM_COPY_START: {
1737                 struct hsm_copy *copy;
1738                 int              rc;
1739
1740                 OBD_ALLOC_PTR(copy);
1741                 if (copy == NULL)
1742                         RETURN(-ENOMEM);
1743                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1744                         OBD_FREE_PTR(copy);
1745                         RETURN(-EFAULT);
1746                 }
1747
1748                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1749                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1750                         rc = -EFAULT;
1751
1752                 OBD_FREE_PTR(copy);
1753                 RETURN(rc);
1754         }
1755         case LL_IOC_HSM_COPY_END: {
1756                 struct hsm_copy *copy;
1757                 int              rc;
1758
1759                 OBD_ALLOC_PTR(copy);
1760                 if (copy == NULL)
1761                         RETURN(-ENOMEM);
1762                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1763                         OBD_FREE_PTR(copy);
1764                         RETURN(-EFAULT);
1765                 }
1766
1767                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1768                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1769                         rc = -EFAULT;
1770
1771                 OBD_FREE_PTR(copy);
1772                 RETURN(rc);
1773         }
1774         case LL_IOC_MIGRATE: {
1775                 char            *buf = NULL;
1776                 const char      *filename;
1777                 int             namelen = 0;
1778                 int             len;
1779                 int             rc;
1780                 int             mdtidx;
1781
1782                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1783                 if (rc < 0)
1784                         RETURN(rc);
1785
1786                 data = (struct obd_ioctl_data *)buf;
1787                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1788                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1789                         GOTO(migrate_free, rc = -EINVAL);
1790
1791                 filename = data->ioc_inlbuf1;
1792                 namelen = data->ioc_inllen1;
1793                 /* \0 is packed at the end of filename */
1794                 if (namelen < 1 || namelen != strlen(filename) + 1)
1795                         GOTO(migrate_free, rc = -EINVAL);
1796
1797                 if (data->ioc_inllen2 != sizeof(mdtidx))
1798                         GOTO(migrate_free, rc = -EINVAL);
1799                 mdtidx = *(int *)data->ioc_inlbuf2;
1800
1801                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1802 migrate_free:
1803                 obd_ioctl_freedata(buf, len);
1804
1805                 RETURN(rc);
1806         }
1807         default:
1808                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1809                                      (void *)arg));
1810         }
1811 }
1812
1813 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1814 {
1815         struct inode *inode = file->f_mapping->host;
1816         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1817         struct ll_sb_info *sbi = ll_i2sbi(inode);
1818         int api32 = ll_need_32bit_api(sbi);
1819         loff_t ret = -EINVAL;
1820         ENTRY;
1821
1822         mutex_lock(&inode->i_mutex);
1823         switch (origin) {
1824                 case SEEK_SET:
1825                         break;
1826                 case SEEK_CUR:
1827                         offset += file->f_pos;
1828                         break;
1829                 case SEEK_END:
1830                         if (offset > 0)
1831                                 GOTO(out, ret);
1832                         if (api32)
1833                                 offset += LL_DIR_END_OFF_32BIT;
1834                         else
1835                                 offset += LL_DIR_END_OFF;
1836                         break;
1837                 default:
1838                         GOTO(out, ret);
1839         }
1840
1841         if (offset >= 0 &&
1842             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1843              (!api32 && offset <= LL_DIR_END_OFF))) {
1844                 if (offset != file->f_pos) {
1845                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1846                             (!api32 && offset == LL_DIR_END_OFF))
1847                                 fd->lfd_pos = MDS_DIR_END_OFF;
1848                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1849                                 fd->lfd_pos = offset << 32;
1850                         else
1851                                 fd->lfd_pos = offset;
1852                         file->f_pos = offset;
1853                         file->f_version = 0;
1854                 }
1855                 ret = offset;
1856         }
1857         GOTO(out, ret);
1858
1859 out:
1860         mutex_unlock(&inode->i_mutex);
1861         return ret;
1862 }
1863
1864 static int ll_dir_open(struct inode *inode, struct file *file)
1865 {
1866         ENTRY;
1867         RETURN(ll_file_open(inode, file));
1868 }
1869
1870 static int ll_dir_release(struct inode *inode, struct file *file)
1871 {
1872         ENTRY;
1873         RETURN(ll_file_release(inode, file));
1874 }
1875
1876 const struct file_operations ll_dir_operations = {
1877         .llseek         = ll_dir_seek,
1878         .open           = ll_dir_open,
1879         .release        = ll_dir_release,
1880         .read           = generic_read_dir,
1881 #ifdef HAVE_DIR_CONTEXT
1882         .iterate        = ll_iterate,
1883 #else
1884         .readdir        = ll_readdir,
1885 #endif
1886         .unlocked_ioctl = ll_dir_ioctl,
1887         .fsync          = ll_fsync,
1888 };