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, 2014, 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 <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <asm/uaccess.h>
  50 #include <linux/buffer_head.h>   // for wait_on_buffer
  51 #include <linux/pagevec.h>
  52
  53 #define DEBUG_SUBSYSTEM S_LLITE
  54
  55 #include <lustre/lustre_idl.h>
  56 #include <obd_support.h>
  57 #include <obd_class.h>
  58 #include <lustre_ioctl.h>
  59 #include <lustre_lib.h>
  60 #include <lustre_dlm.h>
  61 #include <lustre_fid.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "llite_internal.h"
  64
  65 /*
  66  * (new) readdir implementation overview.
  67  *
  68  * Original lustre readdir implementation cached exact copy of raw directory
  69  * pages on the client. These pages were indexed in client page cache by
  70  * logical offset in the directory file. This design, while very simple and
  71  * intuitive had some inherent problems:
  72  *
  73  *     . it implies that byte offset to the directory entry serves as a
  74  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  75  *     ext3/htree directory entries may move due to splits, and more
  76  *     importantly,
  77  *
  78  *     . it is incompatible with the design of split directories for cmd3,
  79  *     that assumes that names are distributed across nodes based on their
  80  *     hash, and so readdir should be done in hash order.
  81  *
  82  * New readdir implementation does readdir in hash order, and uses hash of a
  83  * file name as a telldir/seekdir cookie. This led to number of complications:
  84  *
  85  *     . hash is not unique, so it cannot be used to index cached directory
  86  *     pages on the client (note, that it requires a whole pageful of hash
  87  *     collided entries to cause two pages to have identical hashes);
  88  *
  89  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  90  *     entry cookie. ext3/htree has the same problem and lustre implementation
  91  *     mimics their solution: seekdir(hash) positions directory at the first
  92  *     entry with the given hash.
  93  *
  94  * Client side.
  95  *
  96  * 0. caching
  97  *
  98  * Client caches directory pages using hash of the first entry as an index. As
  99  * noted above hash is not unique, so this solution doesn't work as is:
 100  * special processing is needed for "page hash chains" (i.e., sequences of
 101  * pages filled with entries all having the same hash value).
 102  *
 103  * First, such chains have to be detected. To this end, server returns to the
 104  * client the hash of the first entry on the page next to one returned. When
 105  * client detects that this hash is the same as hash of the first entry on the
 106  * returned page, page hash collision has to be handled. Pages in the
 107  * hash chain, except first one, are termed "overflow pages".
 108  *
 109  * Solution to index uniqueness problem is to not cache overflow
 110  * pages. Instead, when page hash collision is detected, all overflow pages
 111  * from emerging chain are immediately requested from the server and placed in
 112  * a special data structure (struct ll_dir_chain). This data structure is used
 113  * by ll_readdir() to process entries from overflow pages. When readdir
 114  * invocation finishes, overflow pages are discarded. If page hash collision
 115  * chain weren't completely processed, next call to readdir will again detect
 116  * page hash collision, again read overflow pages in, process next portion of
 117  * entries and again discard the pages. This is not as wasteful as it looks,
 118  * because, given reasonable hash, page hash collisions are extremely rare.
 119  *
 120  * 1. directory positioning
 121  *
 122  * When seekdir(hash) is called, original
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  *
 130  *
 131  * Server.
 132  *
 133  * identification of and access to overflow pages
 134  *
 135  * page format
 136  *
 137  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 138  * a header lu_dirpage which describes the start/end hash, and whether this
 139  * page is empty (contains no dir entry) or hash collide with next page.
 140  * After client receives reply, several pages will be integrated into dir page
 141  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 142  * lu_dirpage for this integrated page will be adjusted. See
 143  * mdc_adjust_dirpages().
 144  *
 145  */
 146 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 147                              __u64 offset, struct ll_dir_chain *chain)
 148 {
 149         struct md_callback      cb_op;
 150         struct page             *page;
 151         int                     rc;
 152
 153         cb_op.md_blocking_ast = ll_md_blocking_ast;
 154         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 155         if (rc != 0)
 156                 return ERR_PTR(rc);
 157
 158         return page;
 159 }
 160
 161 void ll_release_page(struct inode *inode, struct page *page,
 162                      bool remove)
 163 {
 164         kunmap(page);
 165
 166         /* Always remove the page for striped dir, because the page is
 167          * built from temporarily in LMV layer */
 168         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 169             ll_i2info(inode)->lli_lsm_md != NULL) {
 170                 __free_page(page);
 171                 return;
 172         }
 173
 174         if (remove) {
 175                 lock_page(page);
 176                 if (likely(page->mapping != NULL))
 177                         truncate_complete_page(page->mapping, page);
 178                 unlock_page(page);
 179         }
 180         page_cache_release(page);
 181 }
 182
 183 /**
 184  * return IF_* type for given lu_dirent entry.
 185  * IF_* flag shld be converted to particular OS file type in
 186  * platform llite module.
 187  */
 188 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 189 {
 190         u16 type = 0;
 191         struct luda_type *lt;
 192         int len = 0;
 193
 194         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 195                 const unsigned align = sizeof(struct luda_type) - 1;
 196
 197                 len = le16_to_cpu(ent->lde_namelen);
 198                 len = (len + align) & ~align;
 199                 lt = (void *)ent->lde_name + len;
 200                 type = IFTODT(le16_to_cpu(lt->lt_type));
 201         }
 202
 203         return type;
 204 }
 205
 206 #ifdef HAVE_DIR_CONTEXT
 207 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 208                 struct dir_context *ctx)
 209 {
 210 #else
 211 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 212                 void *cookie, filldir_t filldir)
 213 {
 214 #endif
 215         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 216         __u64                 pos        = *ppos;
 217         bool                  is_api32 = ll_need_32bit_api(sbi);
 218         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 219         struct page          *page;
 220         struct ll_dir_chain   chain;
 221         bool                  done = false;
 222         int                   rc = 0;
 223         ENTRY;
 224
 225         ll_dir_chain_init(&chain);
 226
 227         page = ll_get_dir_page(inode, op_data, pos, &chain);
 228
 229         while (rc == 0 && !done) {
 230                 struct lu_dirpage *dp;
 231                 struct lu_dirent  *ent;
 232                 __u64 hash;
 233                 __u64 next;
 234
 235                 if (IS_ERR(page)) {
 236                         rc = PTR_ERR(page);
 237                         break;
 238                 }
 239
 240                 hash = MDS_DIR_END_OFF;
 241                 dp = page_address(page);
 242                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 243                      ent = lu_dirent_next(ent)) {
 244                         __u16          type;
 245                         int            namelen;
 246                         struct lu_fid  fid;
 247                         __u64          lhash;
 248                         __u64          ino;
 249
 250                         hash = le64_to_cpu(ent->lde_hash);
 251                         if (hash < pos)
 252                                 /*
 253                                  * Skip until we find target hash
 254                                  * value.
 255                                  */
 256                                 continue;
 257
 258                         namelen = le16_to_cpu(ent->lde_namelen);
 259                         if (namelen == 0)
 260                                 /*
 261                                  * Skip dummy record.
 262                                  */
 263                                 continue;
 264
 265                         if (is_api32 && is_hash64)
 266                                 lhash = hash >> 32;
 267                         else
 268                                 lhash = hash;
 269                         fid_le_to_cpu(&fid, &ent->lde_fid);
 270                         ino = cl_fid_build_ino(&fid, is_api32);
 271                         type = ll_dirent_type_get(ent);
 272                         /* For 'll_nfs_get_name_filldir()', it will try
 273                          * to access the 'ent' through its 'lde_name',
 274                          * so the parameter 'name' for 'filldir()' must
 275                          * be part of the 'ent'. */
 276 #ifdef HAVE_DIR_CONTEXT
 277                         ctx->pos = lhash;
 278                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 279                                          type);
 280 #else
 281                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 282                                        ino, type);
 283 #endif
 284                 }
 285
 286                 if (done) {
 287                         pos = hash;
 288                         ll_release_page(inode, page, false);
 289                         break;
 290                 }
 291
 292                 next = le64_to_cpu(dp->ldp_hash_end);
 293                 pos = next;
 294                 if (pos == MDS_DIR_END_OFF) {
 295                         /*
 296                          * End of directory reached.
 297                          */
 298                         done = 1;
 299                         ll_release_page(inode, page, false);
 300                 } else {
 301                         /*
 302                          * Normal case: continue to the next
 303                          * page.
 304                          */
 305                         ll_release_page(inode, page,
 306                                         le32_to_cpu(dp->ldp_flags) &
 307                                         LDF_COLLIDE);
 308                         next = pos;
 309                         page = ll_get_dir_page(inode, op_data, pos,
 310                                                &chain);
 311                 }
 312         }
 313 #ifdef HAVE_DIR_CONTEXT
 314         ctx->pos = pos;
 315 #else
 316         *ppos = pos;
 317 #endif
 318         ll_dir_chain_fini(&chain);
 319         RETURN(rc);
 320 }
 321
 322 #ifdef HAVE_DIR_CONTEXT
 323 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 324 #else
 325 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 326 #endif
 327 {
 328         struct inode            *inode  = filp->f_dentry->d_inode;
 329         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 330         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 331         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 332         int                     api32   = ll_need_32bit_api(sbi);
 333         struct md_op_data       *op_data;
 334         __u64                   pos;
 335         int                     rc;
 336         ENTRY;
 337
 338         if (lfd != NULL)
 339                 pos = lfd->lfd_pos;
 340         else
 341                 pos = 0;
 342
 343         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 344                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 345                inode, (unsigned long)pos, i_size_read(inode), api32);
 346
 347         if (pos == MDS_DIR_END_OFF)
 348                 /*
 349                  * end-of-file.
 350                  */
 351                 GOTO(out, rc = 0);
 352
 353         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 354                                      LUSTRE_OPC_ANY, inode);
 355         if (IS_ERR(op_data))
 356                 GOTO(out, rc = PTR_ERR(op_data));
 357
 358         if (unlikely(op_data->op_mea1 != NULL)) {
 359                 /* This is only needed for striped dir to fill ..,
 360                  * see lmv_read_entry */
 361                 if (filp->f_dentry->d_parent != NULL &&
 362                     filp->f_dentry->d_parent->d_inode != NULL) {
 363                         __u64 ibits = MDS_INODELOCK_UPDATE;
 364                         struct inode *parent =
 365                                 filp->f_dentry->d_parent->d_inode;
 366
 367                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 368                                 op_data->op_fid3 = *ll_inode2fid(parent);
 369                 }
 370
 371                 /* If it can not find in cache, do lookup .. on the master
 372                  * object */
 373                 if (fid_is_zero(&op_data->op_fid3)) {
 374                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 375                         if (rc != 0) {
 376                                 ll_finish_md_op_data(op_data);
 377                                 RETURN(rc);
 378                         }
 379                 }
 380         }
 381         op_data->op_max_pages = sbi->ll_md_brw_pages;
 382 #ifdef HAVE_DIR_CONTEXT
 383         ctx->pos = pos;
 384         rc = ll_dir_read(inode, &pos, op_data, ctx);
 385         pos = ctx->pos;
 386 #else
 387         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 388 #endif
 389         if (lfd != NULL)
 390                 lfd->lfd_pos = pos;
 391
 392         if (pos == MDS_DIR_END_OFF) {
 393                 if (api32)
 394                         pos = LL_DIR_END_OFF_32BIT;
 395                 else
 396                         pos = LL_DIR_END_OFF;
 397         } else {
 398                 if (api32 && hash64)
 399                         pos = pos >> 32;
 400         }
 401 #ifdef HAVE_DIR_CONTEXT
 402         ctx->pos = pos;
 403 #else
 404         filp->f_pos = pos;
 405 #endif
 406         ll_finish_md_op_data(op_data);
 407         filp->f_version = inode->i_version;
 408
 409 out:
 410         if (!rc)
 411                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 412
 413         RETURN(rc);
 414 }
 415
 416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 417 {
 418         struct mgs_send_param *msp;
 419         int rc = 0;
 420
 421         OBD_ALLOC_PTR(msp);
 422         if (!msp)
 423                 return -ENOMEM;
 424
 425         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 426         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 427                                 sizeof(struct mgs_send_param), msp, NULL);
 428         if (rc)
 429                 CERROR("Failed to set parameter: %d\n", rc);
 430         OBD_FREE_PTR(msp);
 431
 432         return rc;
 433 }
 434
 435 /**
 436  * Create striped directory with specified stripe(@lump)
 437  *
 438  * param[in]parent      the parent of the directory.
 439  * param[in]lump        the specified stripes.
 440  * param[in]dirname     the name of the directory.
 441  * param[in]mode        the specified mode of the directory.
 442  *
 443  * retval               =0 if striped directory is being created successfully.
 444  *                      <0 if the creation is failed.
 445  */
 446 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
 447                                const char *dirname, umode_t mode)
 448 {
 449         struct ptlrpc_request *request = NULL;
 450         struct md_op_data *op_data;
 451         struct ll_sb_info *sbi = ll_i2sbi(parent);
 452         int err;
 453         ENTRY;
 454
 455         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 456                 RETURN(-EINVAL);
 457
 458         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 459                "stripe_offset %d, stripe_count: %u\n",
 460                PFID(ll_inode2fid(parent)), parent, dirname,
 461                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 462
 463         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 464                 lustre_swab_lmv_user_md(lump);
 465
 466         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 467                 mode &= ~current_umask();
 468         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 469         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 470                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 471                                      lump);
 472         if (IS_ERR(op_data))
 473                 GOTO(err_exit, err = PTR_ERR(op_data));
 474
 475         op_data->op_cli_flags |= CLI_SET_MEA;
 476         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 477                         from_kuid(&init_user_ns, current_fsuid()),
 478                         from_kgid(&init_user_ns, current_fsgid()),
 479                         cfs_curproc_cap_pack(), 0, &request);
 480         ll_finish_md_op_data(op_data);
 481         if (err)
 482                 GOTO(err_exit, err);
 483 err_exit:
 484         ptlrpc_req_finished(request);
 485         return err;
 486 }
 487
 488 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 489                      int set_default)
 490 {
 491         struct ll_sb_info *sbi = ll_i2sbi(inode);
 492         struct md_op_data *op_data;
 493         struct ptlrpc_request *req = NULL;
 494         int rc = 0;
 495         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 496         struct obd_device *mgc = lsi->lsi_mgc;
 497         int lum_size;
 498         ENTRY;
 499
 500         if (lump != NULL) {
 501                 /*
 502                  * This is coming from userspace, so should be in
 503                  * local endian.  But the MDS would like it in little
 504                  * endian, so we swab it before we send it.
 505                  */
 506                 switch (lump->lmm_magic) {
 507                 case LOV_USER_MAGIC_V1: {
 508                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 509                                 lustre_swab_lov_user_md_v1(lump);
 510                         lum_size = sizeof(struct lov_user_md_v1);
 511                         break;
 512                 }
 513                 case LOV_USER_MAGIC_V3: {
 514                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 515                                 lustre_swab_lov_user_md_v3(
 516                                         (struct lov_user_md_v3 *)lump);
 517                         lum_size = sizeof(struct lov_user_md_v3);
 518                         break;
 519                 }
 520                 case LMV_USER_MAGIC: {
 521                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 522                                 lustre_swab_lmv_user_md(
 523                                         (struct lmv_user_md *)lump);
 524                         lum_size = sizeof(struct lmv_user_md);
 525                         break;
 526                 }
 527                 default: {
 528                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 529                                         " %#08x != %#08x nor %#08x\n",
 530                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 531                                         LOV_USER_MAGIC_V3);
 532                         RETURN(-EINVAL);
 533                 }
 534                 }
 535         } else {
 536                 lum_size = sizeof(struct lov_user_md_v1);
 537         }
 538
 539         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 540                                      LUSTRE_OPC_ANY, NULL);
 541         if (IS_ERR(op_data))
 542                 RETURN(PTR_ERR(op_data));
 543
 544         /* swabbing is done in lov_setstripe() on server side */
 545         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 546         ll_finish_md_op_data(op_data);
 547         ptlrpc_req_finished(req);
 548         if (rc) {
 549                 if (rc != -EPERM && rc != -EACCES)
 550                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 551         }
 552
 553         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 554          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 555          need the make the distiction between the 2 versions */
 556         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 557                 char *param = NULL;
 558                 char *buf;
 559
 560                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 561                 if (param == NULL)
 562                         GOTO(end, rc = -ENOMEM);
 563
 564                 buf = param;
 565                 /* Get fsname and assume devname to be -MDT0000. */
 566                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 567                 strcat(buf, "-MDT0000.lov");
 568                 buf += strlen(buf);
 569
 570                 /* Set root stripesize */
 571                 sprintf(buf, ".stripesize=%u",
 572                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 573                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 574                 if (rc)
 575                         GOTO(end, rc);
 576
 577                 /* Set root stripecount */
 578                 sprintf(buf, ".stripecount=%hd",
 579                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 580                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 581                 if (rc)
 582                         GOTO(end, rc);
 583
 584                 /* Set root stripeoffset */
 585                 sprintf(buf, ".stripeoffset=%hd",
 586                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 587                         (typeof(lump->lmm_stripe_offset))(-1));
 588                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 589
 590 end:
 591                 if (param != NULL)
 592                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 593         }
 594         RETURN(rc);
 595 }
 596
 597 /**
 598  * This function will be used to get default LOV/LMV/Default LMV
 599  * @valid will be used to indicate which stripe it will retrieve
 600  *      OBD_MD_MEA              LMV stripe EA
 601  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 602  *      otherwise               Default LOV EA.
 603  * Each time, it can only retrieve 1 stripe EA
 604  **/
 605 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 606                      struct ptlrpc_request **request, u64 valid)
 607 {
 608         struct ll_sb_info *sbi = ll_i2sbi(inode);
 609         struct mdt_body   *body;
 610         struct lov_mds_md *lmm = NULL;
 611         struct ptlrpc_request *req = NULL;
 612         int rc, lmm_size;
 613         struct md_op_data *op_data;
 614         ENTRY;
 615
 616         rc = ll_get_default_mdsize(sbi, &lmm_size);
 617         if (rc)
 618                 RETURN(rc);
 619
 620         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 621                                      0, lmm_size, LUSTRE_OPC_ANY,
 622                                      NULL);
 623         if (IS_ERR(op_data))
 624                 RETURN(PTR_ERR(op_data));
 625
 626         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 627         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 628         ll_finish_md_op_data(op_data);
 629         if (rc < 0) {
 630                 CDEBUG(D_INFO, "md_getattr failed on inode "
 631                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 632                 GOTO(out, rc);
 633         }
 634
 635         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 636         LASSERT(body != NULL);
 637
 638         lmm_size = body->mbo_eadatasize;
 639
 640         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 641             lmm_size == 0) {
 642                 GOTO(out, rc = -ENODATA);
 643         }
 644
 645         lmm = req_capsule_server_sized_get(&req->rq_pill,
 646                                            &RMF_MDT_MD, lmm_size);
 647         LASSERT(lmm != NULL);
 648
 649         /*
 650          * This is coming from the MDS, so is probably in
 651          * little endian.  We convert it to host endian before
 652          * passing it to userspace.
 653          */
 654         /* We don't swab objects for directories */
 655         switch (le32_to_cpu(lmm->lmm_magic)) {
 656         case LOV_MAGIC_V1:
 657                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 658                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 659                 break;
 660         case LOV_MAGIC_V3:
 661                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 662                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 663                 break;
 664         case LMV_MAGIC_V1:
 665                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 666                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 667                 break;
 668         case LMV_USER_MAGIC:
 669                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 670                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 671                 break;
 672         default:
 673                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 674                 rc = -EPROTO;
 675         }
 676 out:
 677         *plmm = lmm;
 678         *plmm_size = lmm_size;
 679         *request = req;
 680         return rc;
 681 }
 682
 683 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 684 {
 685         struct md_op_data       *op_data;
 686         int                     rc;
 687         int                     mdt_index;
 688         ENTRY;
 689
 690         OBD_ALLOC_PTR(op_data);
 691         if (op_data == NULL)
 692                 RETURN(-ENOMEM);
 693
 694         op_data->op_flags |= MF_GET_MDT_IDX;
 695         op_data->op_fid1 = *fid;
 696         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 697         mdt_index = op_data->op_mds;
 698         OBD_FREE_PTR(op_data);
 699         if (rc < 0)
 700                 RETURN(rc);
 701
 702         RETURN(mdt_index);
 703 }
 704
 705 /*
 706  *  Get MDT index for the inode.
 707  */
 708 int ll_get_mdt_idx(struct inode *inode)
 709 {
 710         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 711 }
 712
 713 /**
 714  * Generic handler to do any pre-copy work.
 715  *
 716  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 717  * first information for it that real work has started.
 718  *
 719  * Moreover, for a ARCHIVE request, it will sample the file data version and
 720  * store it in \a copy.
 721  *
 722  * \return 0 on success.
 723  */
 724 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 725 {
 726         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 727         struct hsm_progress_kernel       hpk;
 728         int                              rc;
 729         ENTRY;
 730
 731         /* Forge a hsm_progress based on data from copy. */
 732         hpk.hpk_fid = copy->hc_hai.hai_fid;
 733         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 734         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 735         hpk.hpk_extent.length = 0;
 736         hpk.hpk_flags = 0;
 737         hpk.hpk_errval = 0;
 738         hpk.hpk_data_version = 0;
 739
 740
 741         /* For archive request, we need to read the current file version. */
 742         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 743                 struct inode    *inode;
 744                 __u64            data_version = 0;
 745
 746                 /* Get inode for this fid */
 747                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 748                 if (IS_ERR(inode)) {
 749                         hpk.hpk_flags |= HP_FLAG_RETRY;
 750                         /* hpk_errval is >= 0 */
 751                         hpk.hpk_errval = -PTR_ERR(inode);
 752                         GOTO(progress, rc = PTR_ERR(inode));
 753                 }
 754
 755                 /* Read current file data version */
 756                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 757                 iput(inode);
 758                 if (rc != 0) {
 759                         CDEBUG(D_HSM, "Could not read file data version of "
 760                                       DFID" (rc = %d). Archive request ("
 761                                       LPX64") could not be done.\n",
 762                                       PFID(&copy->hc_hai.hai_fid), rc,
 763                                       copy->hc_hai.hai_cookie);
 764                         hpk.hpk_flags |= HP_FLAG_RETRY;
 765                         /* hpk_errval must be >= 0 */
 766                         hpk.hpk_errval = -rc;
 767                         GOTO(progress, rc);
 768                 }
 769
 770                 /* Store in the hsm_copy for later copytool use.
 771                  * Always modified even if no lsm. */
 772                 copy->hc_data_version = data_version;
 773         }
 774
 775 progress:
 776         /* On error, the request should be considered as completed */
 777         if (hpk.hpk_errval > 0)
 778                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 779         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 780                            &hpk, NULL);
 781
 782         RETURN(rc);
 783 }
 784
 785 /**
 786  * Generic handler to do any post-copy work.
 787  *
 788  * It will send the last hsm_progress update to coordinator to inform it
 789  * that copy is finished and whether it was successful or not.
 790  *
 791  * Moreover,
 792  * - for ARCHIVE request, it will sample the file data version and compare it
 793  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 794  *   will be considered as failed.
 795  * - for RESTORE request, it will sample the file data version and send it to
 796  *   coordinator which is useful if the file was imported as 'released'.
 797  *
 798  * \return 0 on success.
 799  */
 800 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 801 {
 802         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 803         struct hsm_progress_kernel       hpk;
 804         int                              rc;
 805         ENTRY;
 806
 807         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 808         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 809          * initialized if copy_end was called with copy == NULL.
 810          */
 811
 812         /* Forge a hsm_progress based on data from copy. */
 813         hpk.hpk_fid = copy->hc_hai.hai_fid;
 814         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 815         hpk.hpk_extent = copy->hc_hai.hai_extent;
 816         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 817         hpk.hpk_errval = copy->hc_errval;
 818         hpk.hpk_data_version = 0;
 819
 820         /* For archive request, we need to check the file data was not changed.
 821          *
 822          * For restore request, we need to send the file data version, this is
 823          * useful when the file was created using hsm_import.
 824          */
 825         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 826              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 827             (copy->hc_errval == 0)) {
 828                 struct inode    *inode;
 829                 __u64            data_version = 0;
 830
 831                 /* Get lsm for this fid */
 832                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 833                 if (IS_ERR(inode)) {
 834                         hpk.hpk_flags |= HP_FLAG_RETRY;
 835                         /* hpk_errval must be >= 0 */
 836                         hpk.hpk_errval = -PTR_ERR(inode);
 837                         GOTO(progress, rc = PTR_ERR(inode));
 838                 }
 839
 840                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 841                 iput(inode);
 842                 if (rc) {
 843                         CDEBUG(D_HSM, "Could not read file data version. "
 844                                       "Request could not be confirmed.\n");
 845                         if (hpk.hpk_errval == 0)
 846                                 hpk.hpk_errval = -rc;
 847                         GOTO(progress, rc);
 848                 }
 849
 850                 /* Store in the hsm_copy for later copytool use.
 851                  * Always modified even if no lsm. */
 852                 hpk.hpk_data_version = data_version;
 853
 854                 /* File could have been stripped during archiving, so we need
 855                  * to check anyway. */
 856                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 857                     (copy->hc_data_version != data_version)) {
 858                         CDEBUG(D_HSM, "File data version mismatched. "
 859                               "File content was changed during archiving. "
 860                                DFID", start:"LPX64" current:"LPX64"\n",
 861                                PFID(&copy->hc_hai.hai_fid),
 862                                copy->hc_data_version, data_version);
 863                         /* File was changed, send error to cdt. Do not ask for
 864                          * retry because if a file is modified frequently,
 865                          * the cdt will loop on retried archive requests.
 866                          * The policy engine will ask for a new archive later
 867                          * when the file will not be modified for some tunable
 868                          * time */
 869                         /* we do not notify caller */
 870                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 871                         /* hpk_errval must be >= 0 */
 872                         hpk.hpk_errval = EBUSY;
 873                 }
 874
 875         }
 876
 877 progress:
 878         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 879                            &hpk, NULL);
 880
 881         RETURN(rc);
 882 }
 883
 884
 885 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 886                           const void __user *data, size_t size)
 887 {
 888         void *copy;
 889         int rc;
 890
 891         OBD_ALLOC(copy, size);
 892         if (copy == NULL)
 893                 return -ENOMEM;
 894
 895         if (copy_from_user(copy, data, size)) {
 896                 rc = -EFAULT;
 897                 goto out;
 898         }
 899
 900         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 901 out:
 902         OBD_FREE(copy, size);
 903
 904         return rc;
 905 }
 906
 907 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 908 {
 909         int cmd = qctl->qc_cmd;
 910         int type = qctl->qc_type;
 911         int id = qctl->qc_id;
 912         int valid = qctl->qc_valid;
 913         int rc = 0;
 914         ENTRY;
 915
 916         switch (cmd) {
 917         case LUSTRE_Q_INVALIDATE:
 918         case LUSTRE_Q_FINVALIDATE:
 919         case Q_QUOTAON:
 920         case Q_QUOTAOFF:
 921         case Q_SETQUOTA:
 922         case Q_SETINFO:
 923                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 924                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 925                         RETURN(-EPERM);
 926                 break;
 927         case Q_GETQUOTA:
 928                 if (((type == USRQUOTA &&
 929                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 930                      (type == GRPQUOTA &&
 931                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 932                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 933                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 934                         RETURN(-EPERM);
 935                 break;
 936         case Q_GETINFO:
 937                 break;
 938         default:
 939                 CERROR("unsupported quotactl op: %#x\n", cmd);
 940                 RETURN(-ENOTTY);
 941         }
 942
 943         if (valid != QC_GENERAL) {
 944                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 945                         RETURN(-EOPNOTSUPP);
 946
 947                 if (cmd == Q_GETINFO)
 948                         qctl->qc_cmd = Q_GETOINFO;
 949                 else if (cmd == Q_GETQUOTA)
 950                         qctl->qc_cmd = Q_GETOQUOTA;
 951                 else
 952                         RETURN(-EINVAL);
 953
 954                 switch (valid) {
 955                 case QC_MDTIDX:
 956                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 957                                            sizeof(*qctl), qctl, NULL);
 958                         break;
 959                 case QC_OSTIDX:
 960                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 961                                            sizeof(*qctl), qctl, NULL);
 962                         break;
 963                 case QC_UUID:
 964                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 965                                            sizeof(*qctl), qctl, NULL);
 966                         if (rc == -EAGAIN)
 967                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 968                                                    sbi->ll_dt_exp,
 969                                                    sizeof(*qctl), qctl, NULL);
 970                         break;
 971                 default:
 972                         rc = -EINVAL;
 973                         break;
 974                 }
 975
 976                 if (rc)
 977                         RETURN(rc);
 978
 979                 qctl->qc_cmd = cmd;
 980         } else {
 981                 struct obd_quotactl *oqctl;
 982
 983                 OBD_ALLOC_PTR(oqctl);
 984                 if (oqctl == NULL)
 985                         RETURN(-ENOMEM);
 986
 987                 QCTL_COPY(oqctl, qctl);
 988                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 989                 if (rc) {
 990                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 991                                 oqctl->qc_cmd = Q_QUOTAOFF;
 992                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 993                         }
 994                         OBD_FREE_PTR(oqctl);
 995                         RETURN(rc);
 996                 }
 997                 /* If QIF_SPACE is not set, client should collect the
 998                  * space usage from OSSs by itself */
 999                 if (cmd == Q_GETQUOTA &&
1000                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1001                     !oqctl->qc_dqblk.dqb_curspace) {
1002                         struct obd_quotactl *oqctl_tmp;
1003
1004                         OBD_ALLOC_PTR(oqctl_tmp);
1005                         if (oqctl_tmp == NULL)
1006                                 GOTO(out, rc = -ENOMEM);
1007
1008                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1009                         oqctl_tmp->qc_id = oqctl->qc_id;
1010                         oqctl_tmp->qc_type = oqctl->qc_type;
1011
1012                         /* collect space usage from OSTs */
1013                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1014                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1015                         if (!rc || rc == -EREMOTEIO) {
1016                                 oqctl->qc_dqblk.dqb_curspace =
1017                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1018                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1019                         }
1020
1021                         /* collect space & inode usage from MDTs */
1022                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1023                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1024                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1025                         if (!rc || rc == -EREMOTEIO) {
1026                                 oqctl->qc_dqblk.dqb_curspace +=
1027                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1028                                 oqctl->qc_dqblk.dqb_curinodes =
1029                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1030                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1031                         } else {
1032                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1033                         }
1034
1035                         OBD_FREE_PTR(oqctl_tmp);
1036                 }
1037 out:
1038                 QCTL_COPY(qctl, oqctl);
1039                 OBD_FREE_PTR(oqctl);
1040         }
1041
1042         RETURN(rc);
1043 }
1044
1045 static char *
1046 ll_getname(const char __user *filename)
1047 {
1048         int ret = 0, len;
1049         char *tmp = __getname();
1050
1051         if (!tmp)
1052                 return ERR_PTR(-ENOMEM);
1053
1054         len = strncpy_from_user(tmp, filename, PATH_MAX);
1055         if (len == 0)
1056                 ret = -ENOENT;
1057         else if (len > PATH_MAX)
1058                 ret = -ENAMETOOLONG;
1059
1060         if (ret) {
1061                 __putname(tmp);
1062                 tmp =  ERR_PTR(ret);
1063         }
1064         return tmp;
1065 }
1066
1067 #define ll_putname(filename) __putname(filename)
1068
1069 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1070 {
1071         struct inode *inode = file->f_dentry->d_inode;
1072         struct ll_sb_info *sbi = ll_i2sbi(inode);
1073         struct obd_ioctl_data *data;
1074         int rc = 0;
1075         ENTRY;
1076
1077         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1078                PFID(ll_inode2fid(inode)), inode, cmd);
1079
1080         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1081         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1082                 return -ENOTTY;
1083
1084         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1085         switch(cmd) {
1086         case FSFILT_IOC_GETFLAGS:
1087         case FSFILT_IOC_SETFLAGS:
1088                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1089         case FSFILT_IOC_GETVERSION_OLD:
1090         case FSFILT_IOC_GETVERSION:
1091                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1092         /* We need to special case any other ioctls we want to handle,
1093          * to send them to the MDS/OST as appropriate and to properly
1094          * network encode the arg field.
1095         case FSFILT_IOC_SETVERSION_OLD:
1096         case FSFILT_IOC_SETVERSION:
1097         */
1098         case LL_IOC_GET_MDTIDX: {
1099                 int mdtidx;
1100
1101                 mdtidx = ll_get_mdt_idx(inode);
1102                 if (mdtidx < 0)
1103                         RETURN(mdtidx);
1104
1105                 if (put_user((int)mdtidx, (int __user *)arg))
1106                         RETURN(-EFAULT);
1107
1108                 return 0;
1109         }
1110         case IOC_MDC_LOOKUP: {
1111                 int namelen, len = 0;
1112                 char *buf = NULL;
1113                 char *filename;
1114
1115                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1116                 if (rc != 0)
1117                         RETURN(rc);
1118                 data = (void *)buf;
1119
1120                 filename = data->ioc_inlbuf1;
1121                 namelen = strlen(filename);
1122                 if (namelen < 1) {
1123                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1124                         GOTO(out_free, rc = -EINVAL);
1125                 }
1126
1127                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1128                 if (rc < 0) {
1129                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1130                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1131                                filename, rc);
1132                         GOTO(out_free, rc);
1133                 }
1134 out_free:
1135                 obd_ioctl_freedata(buf, len);
1136                 return rc;
1137         }
1138         case LL_IOC_LMV_SETSTRIPE: {
1139                 struct lmv_user_md  *lum;
1140                 char            *buf = NULL;
1141                 char            *filename;
1142                 int              namelen = 0;
1143                 int              lumlen = 0;
1144                 umode_t          mode;
1145                 int              len;
1146                 int              rc;
1147
1148                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1149                 if (rc)
1150                         RETURN(rc);
1151
1152                 data = (void *)buf;
1153                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1154                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1155                         GOTO(lmv_out_free, rc = -EINVAL);
1156
1157                 filename = data->ioc_inlbuf1;
1158                 namelen = data->ioc_inllen1;
1159
1160                 if (namelen < 1) {
1161                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1162                         GOTO(lmv_out_free, rc = -EINVAL);
1163                 }
1164                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1165                 lumlen = data->ioc_inllen2;
1166
1167                 if (lum->lum_magic != LMV_USER_MAGIC ||
1168                     lumlen != sizeof(*lum)) {
1169                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1170                                filename, lum->lum_magic, lumlen, -EFAULT);
1171                         GOTO(lmv_out_free, rc = -EINVAL);
1172                 }
1173
1174 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1175                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1176 #else
1177                 mode = data->ioc_type;
1178 #endif
1179                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1180 lmv_out_free:
1181                 obd_ioctl_freedata(buf, len);
1182                 RETURN(rc);
1183
1184         }
1185         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1186                 struct lmv_user_md        lum;
1187                 struct lmv_user_md __user *ulump =
1188                                         (struct lmv_user_md __user *)arg;
1189                 int                       rc;
1190
1191                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1192                         RETURN(-EFAULT);
1193
1194                 if (lum.lum_magic != LMV_USER_MAGIC)
1195                         RETURN(-EINVAL);
1196
1197                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1198
1199                 RETURN(rc);
1200         }
1201         case LL_IOC_LOV_SETSTRIPE: {
1202                 struct lov_user_md_v3 lumv3;
1203                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1204                 struct lov_user_md_v1 __user *lumv1p =
1205                         (struct lov_user_md_v1 __user *)arg;
1206                 struct lov_user_md_v3 __user *lumv3p =
1207                         (struct lov_user_md_v3 __user *)arg;
1208
1209                 int set_default = 0;
1210
1211                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1212                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1213                         sizeof(lumv3p->lmm_objects[0]));
1214                 /* first try with v1 which is smaller than v3 */
1215                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1216                         RETURN(-EFAULT);
1217
1218                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1219                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1220                                 RETURN(-EFAULT);
1221                 }
1222
1223                 if (inode->i_sb->s_root == file->f_dentry)
1224                         set_default = 1;
1225
1226                 /* in v1 and v3 cases lumv1 points to data */
1227                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1228
1229                 RETURN(rc);
1230         }
1231         case LL_IOC_LMV_GETSTRIPE: {
1232                 struct lmv_user_md __user *ulmv =
1233                                         (struct lmv_user_md __user *)arg;
1234                 struct lmv_user_md      lum;
1235                 struct ptlrpc_request   *request = NULL;
1236                 union lmv_mds_md        *lmm = NULL;
1237                 int                     lmmsize;
1238                 u64                     valid = 0;
1239                 struct lmv_user_md      *tmp = NULL;
1240                 int                     mdt_index;
1241                 int                     lum_size;
1242                 int                     stripe_count;
1243                 int                     i;
1244                 int                     rc;
1245
1246                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1247                         RETURN(-EFAULT);
1248
1249                 /* lum_magic will indicate which stripe the ioctl will like
1250                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1251                  * is for default LMV stripe */
1252                 if (lum.lum_magic == LMV_MAGIC_V1)
1253                         valid |= OBD_MD_MEA;
1254                 else if (lum.lum_magic == LMV_USER_MAGIC)
1255                         valid |= OBD_MD_DEFAULT_MEA;
1256                 else
1257                         RETURN(-EINVAL);
1258
1259                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1260                                       valid);
1261                 if (rc != 0)
1262                         GOTO(finish_req, rc);
1263
1264                 /* Get default LMV EA */
1265                 if (lum.lum_magic == LMV_USER_MAGIC) {
1266                         if (rc != 0)
1267                                 GOTO(finish_req, rc);
1268
1269                         if (lmmsize > sizeof(*ulmv))
1270                                 GOTO(finish_req, rc = -EINVAL);
1271
1272                         if (copy_to_user(ulmv, lmm, lmmsize))
1273                                 GOTO(finish_req, rc = -EFAULT);
1274
1275                         GOTO(finish_req, rc);
1276                 }
1277
1278                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1279                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1280                 OBD_ALLOC(tmp, lum_size);
1281                 if (tmp == NULL)
1282                         GOTO(finish_req, rc = -ENOMEM);
1283
1284                 mdt_index = ll_get_mdt_idx(inode);
1285                 if (mdt_index < 0)
1286                         GOTO(out_tmp, rc = -ENOMEM);
1287
1288                 tmp->lum_magic = LMV_MAGIC_V1;
1289                 tmp->lum_stripe_count = 0;
1290                 tmp->lum_stripe_offset = mdt_index;
1291                 for (i = 0; i < stripe_count; i++) {
1292                         struct lu_fid   fid;
1293
1294                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1295                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1296                         if (mdt_index < 0)
1297                                 GOTO(out_tmp, rc = mdt_index);
1298
1299                         tmp->lum_objects[i].lum_mds = mdt_index;
1300                         tmp->lum_objects[i].lum_fid = fid;
1301                         tmp->lum_stripe_count++;
1302                 }
1303
1304                 if (copy_to_user(ulmv, tmp, lum_size))
1305                         GOTO(out_tmp, rc = -EFAULT);
1306 out_tmp:
1307                 OBD_FREE(tmp, lum_size);
1308 finish_req:
1309                 ptlrpc_req_finished(request);
1310                 return rc;
1311         }
1312
1313         case LL_IOC_REMOVE_ENTRY: {
1314                 char            *filename = NULL;
1315                 int              namelen = 0;
1316                 int              rc;
1317
1318                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1319                  * unsupported server, which might crash the server(LU-2730),
1320                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1321                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1322                  * server will support REINT_RMENTRY XXX*/
1323                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1324                         RETURN(-ENOTSUPP);
1325
1326                 filename = ll_getname((const char __user *)arg);
1327                 if (IS_ERR(filename))
1328                         RETURN(PTR_ERR(filename));
1329
1330                 namelen = strlen(filename);
1331                 if (namelen < 1)
1332                         GOTO(out_rmdir, rc = -EINVAL);
1333
1334                 rc = ll_rmdir_entry(inode, filename, namelen);
1335 out_rmdir:
1336                 if (filename)
1337                         ll_putname(filename);
1338                 RETURN(rc);
1339         }
1340         case LL_IOC_LOV_SWAP_LAYOUTS:
1341                 RETURN(-EPERM);
1342         case IOC_OBD_STATFS:
1343                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1344         case LL_IOC_LOV_GETSTRIPE:
1345         case LL_IOC_MDC_GETINFO:
1346         case IOC_MDC_GETFILEINFO:
1347         case IOC_MDC_GETFILESTRIPE: {
1348                 struct ptlrpc_request *request = NULL;
1349                 struct lov_user_md __user *lump;
1350                 struct lov_mds_md *lmm = NULL;
1351                 struct mdt_body *body;
1352                 char *filename = NULL;
1353                 int lmmsize;
1354
1355                 if (cmd == IOC_MDC_GETFILEINFO ||
1356                     cmd == IOC_MDC_GETFILESTRIPE) {
1357                         filename = ll_getname((const char __user *)arg);
1358                         if (IS_ERR(filename))
1359                                 RETURN(PTR_ERR(filename));
1360
1361                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1362                                                       &lmmsize, &request);
1363                 } else {
1364                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1365                                               &request, 0);
1366                 }
1367
1368                 if (request) {
1369                         body = req_capsule_server_get(&request->rq_pill,
1370                                                       &RMF_MDT_BODY);
1371                         LASSERT(body != NULL);
1372                 } else {
1373                         GOTO(out_req, rc);
1374                 }
1375
1376                 if (rc < 0) {
1377                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1378                                                cmd == LL_IOC_MDC_GETINFO))
1379                                 GOTO(skip_lmm, rc = 0);
1380                         else
1381                                 GOTO(out_req, rc);
1382                 }
1383
1384                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1385                     cmd == LL_IOC_LOV_GETSTRIPE) {
1386                         lump = (struct lov_user_md __user *)arg;
1387                 } else {
1388                         struct lov_user_mds_data __user *lmdp;
1389                         lmdp = (struct lov_user_mds_data __user *)arg;
1390                         lump = &lmdp->lmd_lmm;
1391                 }
1392                 if (copy_to_user(lump, lmm, lmmsize)) {
1393                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1394                                 GOTO(out_req, rc = -EFAULT);
1395                         rc = -EOVERFLOW;
1396                 }
1397         skip_lmm:
1398                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1399                         struct lov_user_mds_data __user *lmdp;
1400                         lstat_t st = { 0 };
1401
1402                         st.st_dev       = inode->i_sb->s_dev;
1403                         st.st_mode      = body->mbo_mode;
1404                         st.st_nlink     = body->mbo_nlink;
1405                         st.st_uid       = body->mbo_uid;
1406                         st.st_gid       = body->mbo_gid;
1407                         st.st_rdev      = body->mbo_rdev;
1408                         st.st_size      = body->mbo_size;
1409                         st.st_blksize   = PAGE_CACHE_SIZE;
1410                         st.st_blocks    = body->mbo_blocks;
1411                         st.st_atime     = body->mbo_atime;
1412                         st.st_mtime     = body->mbo_mtime;
1413                         st.st_ctime     = body->mbo_ctime;
1414                         st.st_ino       = inode->i_ino;
1415
1416                         lmdp = (struct lov_user_mds_data __user *)arg;
1417                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1418                                 GOTO(out_req, rc = -EFAULT);
1419                 }
1420
1421                 EXIT;
1422         out_req:
1423                 ptlrpc_req_finished(request);
1424                 if (filename)
1425                         ll_putname(filename);
1426                 return rc;
1427         }
1428         case OBD_IOC_QUOTACHECK: {
1429                 struct obd_quotactl *oqctl;
1430                 int error = 0;
1431
1432                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1433                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1434                         RETURN(-EPERM);
1435
1436                 OBD_ALLOC_PTR(oqctl);
1437                 if (!oqctl)
1438                         RETURN(-ENOMEM);
1439                 oqctl->qc_type = arg;
1440                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1441                 if (rc < 0) {
1442                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1443                         error = rc;
1444                 }
1445
1446                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1447                 if (rc < 0)
1448                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1449
1450                 OBD_FREE_PTR(oqctl);
1451                 return error ?: rc;
1452         }
1453         case OBD_IOC_POLL_QUOTACHECK: {
1454                 struct if_quotacheck *check;
1455
1456                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1457                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1458                         RETURN(-EPERM);
1459
1460                 OBD_ALLOC_PTR(check);
1461                 if (!check)
1462                         RETURN(-ENOMEM);
1463
1464                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1465                                    NULL);
1466                 if (rc) {
1467                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1468                         if (copy_to_user((void __user *)arg, check,
1469                                              sizeof(*check)))
1470                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1471                         GOTO(out_poll, rc);
1472                 }
1473
1474                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1475                                    NULL);
1476                 if (rc) {
1477                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1478                         if (copy_to_user((void __user *)arg, check,
1479                                              sizeof(*check)))
1480                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1481                         GOTO(out_poll, rc);
1482                 }
1483         out_poll:
1484                 OBD_FREE_PTR(check);
1485                 RETURN(rc);
1486         }
1487 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
1488         case LL_IOC_QUOTACTL_18: {
1489                 /* copy the old 1.x quota struct for internal use, then copy
1490                  * back into old format struct.  For 1.8 compatibility. */
1491                 struct if_quotactl_18 *qctl_18;
1492                 struct if_quotactl *qctl_20;
1493
1494                 OBD_ALLOC_PTR(qctl_18);
1495                 if (!qctl_18)
1496                         RETURN(-ENOMEM);
1497
1498                 OBD_ALLOC_PTR(qctl_20);
1499                 if (!qctl_20)
1500                         GOTO(out_quotactl_18, rc = -ENOMEM);
1501
1502                 if (copy_from_user(qctl_18, (void __user *)arg,
1503                                    sizeof(*qctl_18)))
1504                         GOTO(out_quotactl_20, rc = -ENOMEM);
1505
1506                 QCTL_COPY(qctl_20, qctl_18);
1507                 qctl_20->qc_idx = 0;
1508
1509                 /* XXX: dqb_valid was borrowed as a flag to mark that
1510                  *      only mds quota is wanted */
1511                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1512                     qctl_18->qc_dqblk.dqb_valid) {
1513                         qctl_20->qc_valid = QC_MDTIDX;
1514                         qctl_20->qc_dqblk.dqb_valid = 0;
1515                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1516                         qctl_20->qc_valid = QC_UUID;
1517                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1518                 } else {
1519                         qctl_20->qc_valid = QC_GENERAL;
1520                 }
1521
1522                 rc = quotactl_ioctl(sbi, qctl_20);
1523
1524                 if (rc == 0) {
1525                         QCTL_COPY(qctl_18, qctl_20);
1526                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1527
1528                         if (copy_to_user((void __user *)arg, qctl_18,
1529                                              sizeof(*qctl_18)))
1530                                 rc = -EFAULT;
1531                 }
1532
1533         out_quotactl_20:
1534                 OBD_FREE_PTR(qctl_20);
1535         out_quotactl_18:
1536                 OBD_FREE_PTR(qctl_18);
1537                 RETURN(rc);
1538         }
1539 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */
1540         case OBD_IOC_QUOTACTL: {
1541                 struct if_quotactl *qctl;
1542
1543                 OBD_ALLOC_PTR(qctl);
1544                 if (!qctl)
1545                         RETURN(-ENOMEM);
1546
1547                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1548                         GOTO(out_quotactl, rc = -EFAULT);
1549
1550                 rc = quotactl_ioctl(sbi, qctl);
1551
1552                 if (rc == 0 &&
1553                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1554                         rc = -EFAULT;
1555
1556         out_quotactl:
1557                 OBD_FREE_PTR(qctl);
1558                 RETURN(rc);
1559         }
1560         case OBD_IOC_GETDTNAME:
1561         case OBD_IOC_GETMDNAME:
1562                 RETURN(ll_get_obd_name(inode, cmd, arg));
1563         case LL_IOC_FLUSHCTX:
1564                 RETURN(ll_flush_ctx(inode));
1565 #ifdef CONFIG_FS_POSIX_ACL
1566         case LL_IOC_RMTACL: {
1567             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1568                 inode == inode->i_sb->s_root->d_inode) {
1569                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1570
1571                 LASSERT(fd != NULL);
1572                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1573                 if (!rc)
1574                         fd->fd_flags |= LL_FILE_RMTACL;
1575                 RETURN(rc);
1576             } else
1577                 RETURN(0);
1578         }
1579 #endif
1580         case LL_IOC_GETOBDCOUNT: {
1581                 u32 count, vallen;
1582                 struct obd_export *exp;
1583
1584                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1585                         RETURN(-EFAULT);
1586
1587                 /* get ost count when count is zero, get mdt count otherwise */
1588                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1589                 vallen = sizeof(count);
1590                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1591                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1592                 if (rc) {
1593                         CERROR("get target count failed: %d\n", rc);
1594                         RETURN(rc);
1595                 }
1596
1597                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1598                         RETURN(-EFAULT);
1599
1600                 RETURN(0);
1601         }
1602         case LL_IOC_PATH2FID:
1603                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1604                                      sizeof(struct lu_fid)))
1605                         RETURN(-EFAULT);
1606                 RETURN(0);
1607         case LL_IOC_GET_CONNECT_FLAGS: {
1608                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1609                                      (void __user *)arg));
1610         }
1611         case OBD_IOC_CHANGELOG_SEND:
1612         case OBD_IOC_CHANGELOG_CLEAR:
1613                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1614                                     sizeof(struct ioc_changelog));
1615                 RETURN(rc);
1616         case OBD_IOC_FID2PATH:
1617                 RETURN(ll_fid2path(inode, (void __user *)arg));
1618         case LL_IOC_GETPARENT:
1619                 RETURN(ll_getparent(file, (void __user *)arg));
1620         case LL_IOC_FID2MDTIDX: {
1621                 struct obd_export *exp = ll_i2mdexp(inode);
1622                 struct lu_fid     fid;
1623                 __u32             index;
1624
1625                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1626                                    sizeof(fid)))
1627                         RETURN(-EFAULT);
1628
1629                 /* Call mdc_iocontrol */
1630                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1631                                    (__u32 __user *)&index);
1632                 if (rc != 0)
1633                         RETURN(rc);
1634
1635                 RETURN(index);
1636         }
1637         case LL_IOC_HSM_REQUEST: {
1638                 struct hsm_user_request *hur;
1639                 ssize_t                  totalsize;
1640
1641                 OBD_ALLOC_PTR(hur);
1642                 if (hur == NULL)
1643                         RETURN(-ENOMEM);
1644
1645                 /* We don't know the true size yet; copy the fixed-size part */
1646                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1647                         OBD_FREE_PTR(hur);
1648                         RETURN(-EFAULT);
1649                 }
1650
1651                 /* Compute the whole struct size */
1652                 totalsize = hur_len(hur);
1653                 OBD_FREE_PTR(hur);
1654                 if (totalsize < 0)
1655                         RETURN(-E2BIG);
1656
1657                 /* Final size will be more than double totalsize */
1658                 if (totalsize >= MDS_MAXREQSIZE / 3)
1659                         RETURN(-E2BIG);
1660
1661                 OBD_ALLOC_LARGE(hur, totalsize);
1662                 if (hur == NULL)
1663                         RETURN(-ENOMEM);
1664
1665                 /* Copy the whole struct */
1666                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1667                         OBD_FREE_LARGE(hur, totalsize);
1668                         RETURN(-EFAULT);
1669                 }
1670
1671                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1672                         const struct lu_fid *fid;
1673                         struct inode *f;
1674                         int i;
1675
1676                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1677                                 fid = &hur->hur_user_item[i].hui_fid;
1678                                 f = search_inode_for_lustre(inode->i_sb, fid);
1679                                 if (IS_ERR(f)) {
1680                                         rc = PTR_ERR(f);
1681                                         break;
1682                                 }
1683
1684                                 rc = ll_hsm_release(f);
1685                                 iput(f);
1686                                 if (rc != 0)
1687                                         break;
1688                         }
1689                 } else {
1690                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1691                                            hur, NULL);
1692                 }
1693
1694                 OBD_FREE_LARGE(hur, totalsize);
1695
1696                 RETURN(rc);
1697         }
1698         case LL_IOC_HSM_PROGRESS: {
1699                 struct hsm_progress_kernel      hpk;
1700                 struct hsm_progress             hp;
1701
1702                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1703                         RETURN(-EFAULT);
1704
1705                 hpk.hpk_fid = hp.hp_fid;
1706                 hpk.hpk_cookie = hp.hp_cookie;
1707                 hpk.hpk_extent = hp.hp_extent;
1708                 hpk.hpk_flags = hp.hp_flags;
1709                 hpk.hpk_errval = hp.hp_errval;
1710                 hpk.hpk_data_version = 0;
1711
1712                 /* File may not exist in Lustre; all progress
1713                  * reported to Lustre root */
1714                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1715                                    NULL);
1716                 RETURN(rc);
1717         }
1718         case LL_IOC_HSM_CT_START:
1719                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1720                         RETURN(-EPERM);
1721
1722                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1723                                     sizeof(struct lustre_kernelcomm));
1724                 RETURN(rc);
1725
1726         case LL_IOC_HSM_COPY_START: {
1727                 struct hsm_copy *copy;
1728                 int              rc;
1729
1730                 OBD_ALLOC_PTR(copy);
1731                 if (copy == NULL)
1732                         RETURN(-ENOMEM);
1733                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1734                         OBD_FREE_PTR(copy);
1735                         RETURN(-EFAULT);
1736                 }
1737
1738                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1739                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1740                         rc = -EFAULT;
1741
1742                 OBD_FREE_PTR(copy);
1743                 RETURN(rc);
1744         }
1745         case LL_IOC_HSM_COPY_END: {
1746                 struct hsm_copy *copy;
1747                 int              rc;
1748
1749                 OBD_ALLOC_PTR(copy);
1750                 if (copy == NULL)
1751                         RETURN(-ENOMEM);
1752                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1753                         OBD_FREE_PTR(copy);
1754                         RETURN(-EFAULT);
1755                 }
1756
1757                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1758                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1759                         rc = -EFAULT;
1760
1761                 OBD_FREE_PTR(copy);
1762                 RETURN(rc);
1763         }
1764         case LL_IOC_MIGRATE: {
1765                 char            *buf = NULL;
1766                 const char      *filename;
1767                 int             namelen = 0;
1768                 int             len;
1769                 int             rc;
1770                 int             mdtidx;
1771
1772                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1773                 if (rc < 0)
1774                         RETURN(rc);
1775
1776                 data = (struct obd_ioctl_data *)buf;
1777                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1778                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1779                         GOTO(migrate_free, rc = -EINVAL);
1780
1781                 filename = data->ioc_inlbuf1;
1782                 namelen = data->ioc_inllen1;
1783                 /* \0 is packed at the end of filename */
1784                 if (namelen < 1 || namelen != strlen(filename) + 1)
1785                         GOTO(migrate_free, rc = -EINVAL);
1786
1787                 if (data->ioc_inllen2 != sizeof(mdtidx))
1788                         GOTO(migrate_free, rc = -EINVAL);
1789                 mdtidx = *(int *)data->ioc_inlbuf2;
1790
1791                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1792 migrate_free:
1793                 obd_ioctl_freedata(buf, len);
1794
1795                 RETURN(rc);
1796         }
1797         default:
1798                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1799                                      (void __user *)arg));
1800         }
1801 }
1802
1803 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1804 {
1805         struct inode *inode = file->f_mapping->host;
1806         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1807         struct ll_sb_info *sbi = ll_i2sbi(inode);
1808         int api32 = ll_need_32bit_api(sbi);
1809         loff_t ret = -EINVAL;
1810         ENTRY;
1811
1812         mutex_lock(&inode->i_mutex);
1813         switch (origin) {
1814                 case SEEK_SET:
1815                         break;
1816                 case SEEK_CUR:
1817                         offset += file->f_pos;
1818                         break;
1819                 case SEEK_END:
1820                         if (offset > 0)
1821                                 GOTO(out, ret);
1822                         if (api32)
1823                                 offset += LL_DIR_END_OFF_32BIT;
1824                         else
1825                                 offset += LL_DIR_END_OFF;
1826                         break;
1827                 default:
1828                         GOTO(out, ret);
1829         }
1830
1831         if (offset >= 0 &&
1832             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1833              (!api32 && offset <= LL_DIR_END_OFF))) {
1834                 if (offset != file->f_pos) {
1835                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1836                             (!api32 && offset == LL_DIR_END_OFF))
1837                                 fd->lfd_pos = MDS_DIR_END_OFF;
1838                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1839                                 fd->lfd_pos = offset << 32;
1840                         else
1841                                 fd->lfd_pos = offset;
1842                         file->f_pos = offset;
1843                         file->f_version = 0;
1844                 }
1845                 ret = offset;
1846         }
1847         GOTO(out, ret);
1848
1849 out:
1850         mutex_unlock(&inode->i_mutex);
1851         return ret;
1852 }
1853
1854 static int ll_dir_open(struct inode *inode, struct file *file)
1855 {
1856         ENTRY;
1857         RETURN(ll_file_open(inode, file));
1858 }
1859
1860 static int ll_dir_release(struct inode *inode, struct file *file)
1861 {
1862         ENTRY;
1863         RETURN(ll_file_release(inode, file));
1864 }
1865
1866 const struct file_operations ll_dir_operations = {
1867         .llseek         = ll_dir_seek,
1868         .open           = ll_dir_open,
1869         .release        = ll_dir_release,
1870         .read           = generic_read_dir,
1871 #ifdef HAVE_DIR_CONTEXT
1872         .iterate        = ll_iterate,
1873 #else
1874         .readdir        = ll_readdir,
1875 #endif
1876         .unlocked_ioctl = ll_dir_ioctl,
1877         .fsync          = ll_fsync,
1878 };