lustre/llite/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  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/smp_lock.h>
  46 #include <asm/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <obd_support.h>
  52 #include <obd_class.h>
  53 #include <lustre_lib.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_lite.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include "llite_internal.h"
  59
  60 #ifndef HAVE_PAGE_CHECKED
  61 #ifdef HAVE_PG_FS_MISC
  62 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
  63 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
  64 #else
  65 #error PageChecked or PageFsMisc not defined in kernel
  66 #endif
  67 #endif
  68
  69 /*
  70  * (new) readdir implementation overview.
  71  *
  72  * Original lustre readdir implementation cached exact copy of raw directory
  73  * pages on the client. These pages were indexed in client page cache by
  74  * logical offset in the directory file. This design, while very simple and
  75  * intuitive had some inherent problems:
  76  *
  77  *     . it implies that byte offset to the directory entry serves as a
  78  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  79  *     ext3/htree directory entries may move due to splits, and more
  80  *     importantly,
  81  *
  82  *     . it is incompatible with the design of split directories for cmd3,
  83  *     that assumes that names are distributed across nodes based on their
  84  *     hash, and so readdir should be done in hash order.
  85  *
  86  * New readdir implementation does readdir in hash order, and uses hash of a
  87  * file name as a telldir/seekdir cookie. This led to number of complications:
  88  *
  89  *     . hash is not unique, so it cannot be used to index cached directory
  90  *     pages on the client (note, that it requires a whole pageful of hash
  91  *     collided entries to cause two pages to have identical hashes);
  92  *
  93  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  94  *     entry cookie. ext3/htree has the same problem and lustre implementation
  95  *     mimics their solution: seekdir(hash) positions directory at the first
  96  *     entry with the given hash.
  97  *
  98  * Client side.
  99  *
 100  * 0. caching
 101  *
 102  * Client caches directory pages using hash of the first entry as an index. As
 103  * noted above hash is not unique, so this solution doesn't work as is:
 104  * special processing is needed for "page hash chains" (i.e., sequences of
 105  * pages filled with entries all having the same hash value).
 106  *
 107  * First, such chains have to be detected. To this end, server returns to the
 108  * client the hash of the first entry on the page next to one returned. When
 109  * client detects that this hash is the same as hash of the first entry on the
 110  * returned page, page hash collision has to be handled. Pages in the
 111  * hash chain, except first one, are termed "overflow pages".
 112  *
 113  * Solution to index uniqueness problem is to not cache overflow
 114  * pages. Instead, when page hash collision is detected, all overflow pages
 115  * from emerging chain are immediately requested from the server and placed in
 116  * a special data structure (struct ll_dir_chain). This data structure is used
 117  * by ll_readdir() to process entries from overflow pages. When readdir
 118  * invocation finishes, overflow pages are discarded. If page hash collision
 119  * chain weren't completely processed, next call to readdir will again detect
 120  * page hash collision, again read overflow pages in, process next portion of
 121  * entries and again discard the pages. This is not as wasteful as it looks,
 122  * because, given reasonable hash, page hash collisions are extremely rare.
 123  *
 124  * 1. directory positioning
 125  *
 126  * When seekdir(hash) is called, original
 127  *
 128  *
 129  *
 130  *
 131  *
 132  *
 133  *
 134  *
 135  * Server.
 136  *
 137  * identification of and access to overflow pages
 138  *
 139  * page format
 140  *
 141  *
 142  *
 143  *
 144  *
 145  */
 146
 147 /* returns the page unlocked, but with a reference */
 148 static int ll_dir_readpage(struct file *file, struct page *page)
 149 {
 150         struct inode *inode = page->mapping->host;
 151         struct ptlrpc_request *request;
 152         struct mdt_body *body;
 153         struct obd_capa *oc;
 154         __u64 hash;
 155         int rc;
 156         ENTRY;
 157
 158         if (file) {
 159                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
 160
 161                 hash = fd->fd_dir.lfd_next;
 162         } else {
 163                 struct ll_inode_info *lli = ll_i2info(inode);
 164
 165                 cfs_spin_lock(&lli->lli_sa_lock);
 166                 if (lli->lli_sai)
 167                         LASSERT(lli->lli_sai->sai_pid == cfs_curproc_pid());
 168                 else
 169                         LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
 170                 hash = lli->lli_sa_pos;
 171                 cfs_spin_unlock(&lli->lli_sa_lock);
 172         }
 173         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off %lu\n",
 174                inode->i_ino, inode->i_generation, inode, (unsigned long)hash);
 175
 176         oc = ll_mdscapa_get(inode);
 177         rc = md_readpage(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode),
 178                          oc, hash, page, &request);
 179         capa_put(oc);
 180         if (!rc) {
 181                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 182                 /* Checked by mdc_readpage() */
 183                 LASSERT(body != NULL);
 184
 185                 if (body->valid & OBD_MD_FLSIZE)
 186                         cl_isize_write(inode, body->size);
 187                 SetPageUptodate(page);
 188         }
 189         ptlrpc_req_finished(request);
 190
 191         unlock_page(page);
 192         EXIT;
 193         return rc;
 194 }
 195
 196 struct address_space_operations ll_dir_aops = {
 197         .readpage  = ll_dir_readpage,
 198 };
 199
 200 static void ll_check_page(struct inode *dir, struct page *page)
 201 {
 202         /* XXX: check page format later */
 203         SetPageChecked(page);
 204 }
 205
 206 static void ll_release_page(struct page *page, __u64 hash,
 207                             __u64 start, __u64 end)
 208 {
 209         kunmap(page);
 210         lock_page(page);
 211         if (likely(page->mapping != NULL)) {
 212                 ll_truncate_complete_page(page);
 213                 unlock_page(page);
 214         } else {
 215                 unlock_page(page);
 216                 CWARN("NULL mapping page %p, truncated by others: "
 217                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
 218                       page, hash, start, end);
 219         }
 220         page_cache_release(page);
 221 }
 222
 223 /*
 224  * Find, kmap and return page that contains given hash.
 225  */
 226 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 227                                        __u64 *start, __u64 *end)
 228 {
 229         struct address_space *mapping = dir->i_mapping;
 230         /*
 231          * Complement of hash is used as an index so that
 232          * radix_tree_gang_lookup() can be used to find a page with starting
 233          * hash _smaller_ than one we are looking for.
 234          */
 235         unsigned long offset = hash_x_index(*hash);
 236         struct page *page;
 237         int found;
 238
 239         TREE_READ_LOCK_IRQ(mapping);
 240         found = radix_tree_gang_lookup(&mapping->page_tree,
 241                                        (void **)&page, offset, 1);
 242         if (found > 0) {
 243                 struct lu_dirpage *dp;
 244
 245                 page_cache_get(page);
 246                 TREE_READ_UNLOCK_IRQ(mapping);
 247                 /*
 248                  * In contrast to find_lock_page() we are sure that directory
 249                  * page cannot be truncated (while DLM lock is held) and,
 250                  * hence, can avoid restart.
 251                  *
 252                  * In fact, page cannot be locked here at all, because
 253                  * ll_dir_readpage() does synchronous io.
 254                  */
 255                 wait_on_page(page);
 256                 if (PageUptodate(page)) {
 257                         dp = kmap(page);
 258 #if BITS_PER_LONG == 32
 259                         *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 260                         *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 261                         *hash  = *hash >> 32;
 262 #else
 263                         *start = le64_to_cpu(dp->ldp_hash_start);
 264                         *end   = le64_to_cpu(dp->ldp_hash_end);
 265 #endif
 266                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 267                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 268                         if (*hash > *end || (*end != *start && *hash == *end)) {
 269                                 ll_release_page(page, *hash, *start, *end);
 270                                 page = NULL;
 271                         }
 272                 } else {
 273                         page_cache_release(page);
 274                         page = ERR_PTR(-EIO);
 275                 }
 276
 277         } else {
 278                 TREE_READ_UNLOCK_IRQ(mapping);
 279                 page = NULL;
 280         }
 281         return page;
 282 }
 283
 284 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
 285                              int exact, struct ll_dir_chain *chain)
 286 {
 287         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 288         struct address_space *mapping = dir->i_mapping;
 289         struct lustre_handle lockh;
 290         struct lu_dirpage *dp;
 291         struct page *page;
 292         ldlm_mode_t mode;
 293         int rc;
 294         __u64 start = 0;
 295         __u64 end = 0;
 296         __u64 lhash = hash;
 297         struct ll_inode_info *lli = ll_i2info(dir);
 298
 299         mode = LCK_PR;
 300         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 301                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 302         if (!rc) {
 303                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 304                        ll_md_blocking_ast, ldlm_completion_ast,
 305                        NULL, NULL, dir };
 306                 struct lookup_intent it = { .it_op = IT_READDIR };
 307                 struct ptlrpc_request *request;
 308                 struct md_op_data *op_data;
 309
 310                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 311                                              LUSTRE_OPC_ANY, NULL);
 312                 if (IS_ERR(op_data))
 313                         return (void *)op_data;
 314
 315                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 316                                 op_data, &lockh, NULL, 0, NULL, 0);
 317
 318                 ll_finish_md_op_data(op_data);
 319
 320                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 321                 if (request)
 322                         ptlrpc_req_finished(request);
 323                 if (rc < 0) {
 324                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 325                                PFID(ll_inode2fid(dir)), hash, rc);
 326                         return ERR_PTR(rc);
 327                 }
 328         } else {
 329                 /* for cross-ref object, l_ast_data of the lock may not be set,
 330                  * we reset it here */
 331                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 332                                  dir, NULL);
 333         }
 334         ldlm_lock_dump_handle(D_OTHER, &lockh);
 335
 336         cfs_down(&lli->lli_readdir_sem);
 337         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 338         if (IS_ERR(page)) {
 339                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 340                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 341                 GOTO(out_unlock, page);
 342         }
 343
 344         if (page != NULL) {
 345                 /*
 346                  * XXX nikita: not entirely correct handling of a corner case:
 347                  * suppose hash chain of entries with hash value HASH crosses
 348                  * border between pages P0 and P1. First both P0 and P1 are
 349                  * cached, seekdir() is called for some entry from the P0 part
 350                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 351                  * happens and finds P1, as it starts with matching hash
 352                  * value. Remaining entries from P0 part of the chain are
 353                  * skipped. (Is that really a bug?)
 354                  *
 355                  * Possible solutions: 0. don't cache P1 is such case, handle
 356                  * it as an "overflow" page. 1. invalidate all pages at
 357                  * once. 2. use HASH|1 as an index for P1.
 358                  */
 359                 if (exact && lhash != start) {
 360                         /*
 361                          * readdir asked for a page starting _exactly_ from
 362                          * given hash, but cache contains stale page, with
 363                          * entries with smaller hash values. Stale page should
 364                          * be invalidated, and new one fetched.
 365                          */
 366                         CDEBUG(D_OTHER, "Stale readpage page %p: "
 367                                "start = "LPX64",end = "LPX64"hash ="LPX64"\n",
 368                                page, start, end, lhash);
 369                         ll_release_page(page, lhash, start, end);
 370                 } else {
 371                         GOTO(hash_collision, page);
 372                 }
 373         }
 374
 375         page = read_cache_page(mapping, hash_x_index(hash),
 376                                (filler_t*)mapping->a_ops->readpage, filp);
 377         if (IS_ERR(page)) {
 378                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 379                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 380                 GOTO(out_unlock, page);
 381         }
 382
 383         wait_on_page(page);
 384         (void)kmap(page);
 385         if (!PageUptodate(page)) {
 386                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 387                        PFID(ll_inode2fid(dir)), hash, -5);
 388                 goto fail;
 389         }
 390         if (!PageChecked(page))
 391                 ll_check_page(dir, page);
 392         if (PageError(page)) {
 393                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 394                        PFID(ll_inode2fid(dir)), hash, -5);
 395                 goto fail;
 396         }
 397 hash_collision:
 398         dp = page_address(page);
 399 #if BITS_PER_LONG == 32
 400         start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 401         end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 402         lhash = hash >> 32;
 403 #else
 404         start = le64_to_cpu(dp->ldp_hash_start);
 405         end   = le64_to_cpu(dp->ldp_hash_end);
 406         lhash = hash;
 407 #endif
 408         if (end == start) {
 409                 LASSERT(start == lhash);
 410                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 411 #if BITS_PER_LONG == 32
 412                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
 413                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
 414                       le64_to_cpu(dp->ldp_hash_end), hash);
 415 #endif
 416                 /*
 417                  * Fetch whole overflow chain...
 418                  *
 419                  * XXX not yet.
 420                  */
 421                 goto fail;
 422         }
 423 out_unlock:
 424         cfs_up(&lli->lli_readdir_sem);
 425         ldlm_lock_decref(&lockh, mode);
 426         return page;
 427
 428 fail:
 429         ll_put_page(page);
 430         page = ERR_PTR(-EIO);
 431         goto out_unlock;
 432 }
 433
 434 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 435 {
 436         struct inode         *inode = filp->f_dentry->d_inode;
 437         struct ll_inode_info *info  = ll_i2info(inode);
 438         struct ll_sb_info    *sbi   = ll_i2sbi(inode);
 439         struct ll_file_data  *fd    = LUSTRE_FPRIVATE(filp);
 440         __u64                 pos   = fd->fd_dir.lfd_pos;
 441         struct page          *page;
 442         struct ll_dir_chain   chain;
 443         int rc, need_32bit;
 444         int done;
 445         int shift;
 446         __u16 type;
 447         ENTRY;
 448
 449         need_32bit = ll_need_32bit_api(sbi);
 450         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 451                inode->i_ino, inode->i_generation, inode,
 452                (unsigned long)pos, i_size_read(inode), need_32bit);
 453
 454         if (pos == DIR_END_OFF)
 455                 /*
 456                  * end-of-file.
 457                  */
 458                 RETURN(0);
 459
 460         rc    = 0;
 461         done  = 0;
 462         shift = 0;
 463         ll_dir_chain_init(&chain);
 464
 465         fd->fd_dir.lfd_next = pos;
 466         page = ll_get_dir_page(filp, inode, pos, 0, &chain);
 467
 468         while (rc == 0 && !done) {
 469                 struct lu_dirpage *dp;
 470                 struct lu_dirent  *ent;
 471
 472                 if (!IS_ERR(page)) {
 473                         /*
 474                          * If page is empty (end of directory is reached),
 475                          * use this value.
 476                          */
 477                         __u64 hash = DIR_END_OFF;
 478                         __u64 next;
 479
 480                         dp = page_address(page);
 481                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 482                              ent = lu_dirent_next(ent)) {
 483                                 char          *name;
 484                                 int            namelen;
 485                                 struct lu_fid  fid;
 486                                 __u64          ino;
 487                                 __u64          lhash;
 488
 489                                 /*
 490                                  * XXX: implement correct swabbing here.
 491                                  */
 492
 493                                 hash = le64_to_cpu(ent->lde_hash);
 494                                 if (hash < pos)
 495                                         /*
 496                                          * Skip until we find target hash
 497                                          * value.
 498                                          */
 499                                         continue;
 500
 501                                 namelen = le16_to_cpu(ent->lde_namelen);
 502                                 if (namelen == 0)
 503                                         /*
 504                                          * Skip dummy record.
 505                                          */
 506                                         continue;
 507
 508                                 name = ent->lde_name;
 509                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 510                                 if (need_32bit) {
 511                                         lhash = hash >> 32;
 512                                         ino = cl_fid_build_ino32(&fid);
 513                                 } else {
 514                                         lhash = hash;
 515                                         ino = cl_fid_build_ino(&fid);
 516                                 }
 517                                 type = ll_dirent_type_get(ent);
 518                                 done = filldir(cookie, name, namelen,
 519                                                lhash, ino, type);
 520                         }
 521                         next = le64_to_cpu(dp->ldp_hash_end);
 522                         ll_put_page(page);
 523                         if (!done) {
 524                                 pos = next;
 525                                 if (pos == DIR_END_OFF) {
 526                                         /*
 527                                          * End of directory reached.
 528                                          */
 529                                         done = 1;
 530                                 } else if (1 /* chain is exhausted*/) {
 531                                         /*
 532                                          * Normal case: continue to the next
 533                                          * page.
 534                                          */
 535                                         fd->fd_dir.lfd_next = pos;
 536                                         page = ll_get_dir_page(filp, inode, pos,
 537                                                                1, &chain);
 538                                 } else {
 539                                         /*
 540                                          * go into overflow page.
 541                                          */
 542                                 }
 543                         } else {
 544                                 pos = hash;
 545                         }
 546                 } else {
 547                         rc = PTR_ERR(page);
 548                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 549                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 550                 }
 551         }
 552
 553         fd->fd_dir.lfd_pos = pos;
 554         if (need_32bit)
 555                 filp->f_pos = pos >> 32;
 556         else
 557                 filp->f_pos = pos;
 558         filp->f_version = inode->i_version;
 559         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 560
 561         ll_dir_chain_fini(&chain);
 562
 563         RETURN(rc);
 564 }
 565
 566 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 567 {
 568         struct mgs_send_param *msp;
 569         int rc = 0;
 570
 571         OBD_ALLOC_PTR(msp);
 572         if (!msp)
 573                 return -ENOMEM;
 574
 575         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 576         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 577                                 sizeof(struct mgs_send_param), msp, NULL);
 578         if (rc)
 579                 CERROR("Failed to set parameter: %d\n", rc);
 580         OBD_FREE_PTR(msp);
 581
 582         return rc;
 583 }
 584
 585 char *ll_get_fsname(struct inode *inode)
 586 {
 587         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 588         char *ptr, *fsname;
 589         int len;
 590
 591         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 592         len = strlen(lsi->lsi_lmd->lmd_profile);
 593         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 594         if (ptr && (strcmp(ptr, "-client") == 0))
 595                 len -= 7;
 596         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 597         fsname[len] = '\0';
 598
 599         return fsname;
 600 }
 601
 602 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 603                      int set_default)
 604 {
 605         struct ll_sb_info *sbi = ll_i2sbi(inode);
 606         struct md_op_data *op_data;
 607         struct ptlrpc_request *req = NULL;
 608         int rc = 0;
 609         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 610         struct obd_device *mgc = lsi->lsi_mgc;
 611         char *fsname = NULL, *param = NULL;
 612         int lum_size;
 613
 614         if (lump != NULL) {
 615                 /*
 616                  * This is coming from userspace, so should be in
 617                  * local endian.  But the MDS would like it in little
 618                  * endian, so we swab it before we send it.
 619                  */
 620                 switch (lump->lmm_magic) {
 621                 case LOV_USER_MAGIC_V1: {
 622                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 623                                 lustre_swab_lov_user_md_v1(lump);
 624                         lum_size = sizeof(struct lov_user_md_v1);
 625                         break;
 626                         }
 627                 case LOV_USER_MAGIC_V3: {
 628                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 629                                 lustre_swab_lov_user_md_v3(
 630                                         (struct lov_user_md_v3 *)lump);
 631                         lum_size = sizeof(struct lov_user_md_v3);
 632                         break;
 633                         }
 634                 default: {
 635                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 636                                         " %#08x != %#08x nor %#08x\n",
 637                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 638                                         LOV_USER_MAGIC_V3);
 639                         RETURN(-EINVAL);
 640                         }
 641                }
 642         } else {
 643                 lum_size = sizeof(struct lov_user_md_v1);
 644         }
 645
 646         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 647                                      LUSTRE_OPC_ANY, NULL);
 648         if (IS_ERR(op_data))
 649                 RETURN(PTR_ERR(op_data));
 650
 651         /* swabbing is done in lov_setstripe() on server side */
 652         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 653                         NULL, 0, &req, NULL);
 654         ll_finish_md_op_data(op_data);
 655         ptlrpc_req_finished(req);
 656         if (rc) {
 657                 if (rc != -EPERM && rc != -EACCES)
 658                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 659         }
 660
 661         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 662          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 663          need the make the distiction between the 2 versions */
 664         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 665                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 666
 667                 /* Get fsname and assume devname to be -MDT0000. */
 668                 fsname = ll_get_fsname(inode);
 669                 /* Set root stripesize */
 670                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 671                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 672                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 673                 if (rc)
 674                         goto end;
 675
 676                 /* Set root stripecount */
 677                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 678                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 679                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 680                 if (rc)
 681                         goto end;
 682
 683                 /* Set root stripeoffset */
 684                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 685                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 686                         (typeof(lump->lmm_stripe_offset))(-1));
 687                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 688                 if (rc)
 689                         goto end;
 690 end:
 691                 if (fsname)
 692                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 693                 if (param)
 694                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 695         }
 696         return rc;
 697 }
 698
 699 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 700                      int *lmm_size, struct ptlrpc_request **request)
 701 {
 702         struct ll_sb_info *sbi = ll_i2sbi(inode);
 703         struct mdt_body   *body;
 704         struct lov_mds_md *lmm = NULL;
 705         struct ptlrpc_request *req = NULL;
 706         int rc, lmmsize;
 707         struct md_op_data *op_data;
 708
 709         rc = ll_get_max_mdsize(sbi, &lmmsize);
 710         if (rc)
 711                 RETURN(rc);
 712
 713         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 714                                      0, lmmsize, LUSTRE_OPC_ANY,
 715                                      NULL);
 716         if (op_data == NULL)
 717                 RETURN(-ENOMEM);
 718
 719         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 720         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 721         ll_finish_md_op_data(op_data);
 722         if (rc < 0) {
 723                 CDEBUG(D_INFO, "md_getattr failed on inode "
 724                        "%lu/%u: rc %d\n", inode->i_ino,
 725                        inode->i_generation, rc);
 726                 GOTO(out, rc);
 727         }
 728
 729         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 730         LASSERT(body != NULL);
 731
 732         lmmsize = body->eadatasize;
 733
 734         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 735             lmmsize == 0) {
 736                 GOTO(out, rc = -ENODATA);
 737         }
 738
 739         lmm = req_capsule_server_sized_get(&req->rq_pill,
 740                                            &RMF_MDT_MD, lmmsize);
 741         LASSERT(lmm != NULL);
 742
 743         /*
 744          * This is coming from the MDS, so is probably in
 745          * little endian.  We convert it to host endian before
 746          * passing it to userspace.
 747          */
 748         /* We don't swab objects for directories */
 749         switch (le32_to_cpu(lmm->lmm_magic)) {
 750         case LOV_MAGIC_V1:
 751                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 752                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 753                 break;
 754         case LOV_MAGIC_V3:
 755                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 756                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 757                 break;
 758         default:
 759                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 760                 rc = -EPROTO;
 761         }
 762 out:
 763         *lmmp = lmm;
 764         *lmm_size = lmmsize;
 765         *request = req;
 766         return rc;
 767 }
 768
 769 /*
 770  *  Get MDT index for the inode.
 771  */
 772 int ll_get_mdt_idx(struct inode *inode)
 773 {
 774         struct ll_sb_info *sbi = ll_i2sbi(inode);
 775         struct md_op_data *op_data;
 776         int rc, mdtidx;
 777         ENTRY;
 778
 779         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 780                                      0, LUSTRE_OPC_ANY, NULL);
 781         if (op_data == NULL)
 782                 RETURN(-ENOMEM);
 783
 784         op_data->op_valid |= OBD_MD_MDTIDX;
 785         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 786         mdtidx = op_data->op_mds;
 787         ll_finish_md_op_data(op_data);
 788         if (rc < 0) {
 789                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 790                 RETURN(rc);
 791         }
 792         return mdtidx;
 793 }
 794
 795 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 796 {
 797         void *ptr;
 798         int rc;
 799
 800         OBD_ALLOC(ptr, len);
 801         if (ptr == NULL)
 802                 return -ENOMEM;
 803         if (cfs_copy_from_user(ptr, data, len)) {
 804                 OBD_FREE(ptr, len);
 805                 return -EFAULT;
 806         }
 807         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 808         OBD_FREE(ptr, len);
 809         return rc;
 810 }
 811
 812 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 813                         unsigned int cmd, unsigned long arg)
 814 {
 815         struct ll_sb_info *sbi = ll_i2sbi(inode);
 816         struct obd_ioctl_data *data;
 817         int rc = 0;
 818         ENTRY;
 819
 820         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 821                inode->i_ino, inode->i_generation, inode, cmd);
 822
 823         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 824         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 825                 return -ENOTTY;
 826
 827         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 828         switch(cmd) {
 829         case FSFILT_IOC_GETFLAGS:
 830         case FSFILT_IOC_SETFLAGS:
 831                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 832         case FSFILT_IOC_GETVERSION_OLD:
 833         case FSFILT_IOC_GETVERSION:
 834                 RETURN(put_user(inode->i_generation, (int *)arg));
 835         /* We need to special case any other ioctls we want to handle,
 836          * to send them to the MDS/OST as appropriate and to properly
 837          * network encode the arg field.
 838         case FSFILT_IOC_SETVERSION_OLD:
 839         case FSFILT_IOC_SETVERSION:
 840         */
 841         case LL_IOC_GET_MDTIDX: {
 842                 int mdtidx;
 843
 844                 mdtidx = ll_get_mdt_idx(inode);
 845                 if (mdtidx < 0)
 846                         RETURN(mdtidx);
 847
 848                 if (put_user((int)mdtidx, (int*)arg))
 849                         RETURN(-EFAULT);
 850
 851                 return 0;
 852         }
 853         case IOC_MDC_LOOKUP: {
 854                 struct ptlrpc_request *request = NULL;
 855                 int namelen, len = 0;
 856                 char *buf = NULL;
 857                 char *filename;
 858                 struct md_op_data *op_data;
 859
 860                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 861                 if (rc)
 862                         RETURN(rc);
 863                 data = (void *)buf;
 864
 865                 filename = data->ioc_inlbuf1;
 866                 namelen = strlen(filename);
 867
 868                 if (namelen < 1) {
 869                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 870                         GOTO(out_free, rc = -EINVAL);
 871                 }
 872
 873                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
 874                                              0, LUSTRE_OPC_ANY, NULL);
 875                 if (op_data == NULL)
 876                         GOTO(out_free, rc = -ENOMEM);
 877
 878                 op_data->op_valid = OBD_MD_FLID;
 879                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
 880                 ll_finish_md_op_data(op_data);
 881                 if (rc < 0) {
 882                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 883                         GOTO(out_free, rc);
 884                 }
 885                 ptlrpc_req_finished(request);
 886                 EXIT;
 887 out_free:
 888                 obd_ioctl_freedata(buf, len);
 889                 return rc;
 890         }
 891         case LL_IOC_LOV_SETSTRIPE: {
 892                 struct lov_user_md_v3 lumv3;
 893                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 894                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 895                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 896
 897                 int set_default = 0;
 898
 899                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 900                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 901                         sizeof(lumv3p->lmm_objects[0]));
 902                 /* first try with v1 which is smaller than v3 */
 903                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 904                         RETURN(-EFAULT);
 905
 906                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 907                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 908                                 RETURN(-EFAULT);
 909                 }
 910
 911                 if (inode->i_sb->s_root == file->f_dentry)
 912                         set_default = 1;
 913
 914                 /* in v1 and v3 cases lumv1 points to data */
 915                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 916
 917                 RETURN(rc);
 918         }
 919         case LL_IOC_OBD_STATFS:
 920                 RETURN(ll_obd_statfs(inode, (void *)arg));
 921         case LL_IOC_LOV_GETSTRIPE:
 922         case LL_IOC_MDC_GETINFO:
 923         case IOC_MDC_GETFILEINFO:
 924         case IOC_MDC_GETFILESTRIPE: {
 925                 struct ptlrpc_request *request = NULL;
 926                 struct lov_user_md *lump;
 927                 struct lov_mds_md *lmm = NULL;
 928                 struct mdt_body *body;
 929                 char *filename = NULL;
 930                 int lmmsize;
 931
 932                 if (cmd == IOC_MDC_GETFILEINFO ||
 933                     cmd == IOC_MDC_GETFILESTRIPE) {
 934                         filename = getname((const char *)arg);
 935                         if (IS_ERR(filename))
 936                                 RETURN(PTR_ERR(filename));
 937
 938                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 939                                                       &lmmsize, &request);
 940                 } else {
 941                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 942                 }
 943
 944                 if (request) {
 945                         body = req_capsule_server_get(&request->rq_pill,
 946                                                       &RMF_MDT_BODY);
 947                         LASSERT(body != NULL);
 948                 } else {
 949                         GOTO(out_req, rc);
 950                 }
 951
 952                 if (rc < 0) {
 953                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 954                                                cmd == LL_IOC_MDC_GETINFO))
 955                                 GOTO(skip_lmm, rc = 0);
 956                         else
 957                                 GOTO(out_req, rc);
 958                 }
 959
 960                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 961                     cmd == LL_IOC_LOV_GETSTRIPE) {
 962                         lump = (struct lov_user_md *)arg;
 963                 } else {
 964                         struct lov_user_mds_data *lmdp;
 965                         lmdp = (struct lov_user_mds_data *)arg;
 966                         lump = &lmdp->lmd_lmm;
 967                 }
 968                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
 969                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
 970                                 GOTO(out_req, rc = -EFAULT);
 971                         rc = -EOVERFLOW;
 972                 }
 973         skip_lmm:
 974                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 975                         struct lov_user_mds_data *lmdp;
 976                         lstat_t st = { 0 };
 977
 978                         st.st_dev     = inode->i_sb->s_dev;
 979                         st.st_mode    = body->mode;
 980                         st.st_nlink   = body->nlink;
 981                         st.st_uid     = body->uid;
 982                         st.st_gid     = body->gid;
 983                         st.st_rdev    = body->rdev;
 984                         st.st_size    = body->size;
 985                         st.st_blksize = CFS_PAGE_SIZE;
 986                         st.st_blocks  = body->blocks;
 987                         st.st_atime   = body->atime;
 988                         st.st_mtime   = body->mtime;
 989                         st.st_ctime   = body->ctime;
 990                         st.st_ino     = inode->i_ino;
 991
 992                         lmdp = (struct lov_user_mds_data *)arg;
 993                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 994                                 GOTO(out_req, rc = -EFAULT);
 995                 }
 996
 997                 EXIT;
 998         out_req:
 999                 ptlrpc_req_finished(request);
1000                 if (filename)
1001                         putname(filename);
1002                 return rc;
1003         }
1004         case IOC_LOV_GETINFO: {
1005                 struct lov_user_mds_data *lumd;
1006                 struct lov_stripe_md *lsm;
1007                 struct lov_user_md *lum;
1008                 struct lov_mds_md *lmm;
1009                 int lmmsize;
1010                 lstat_t st;
1011
1012                 lumd = (struct lov_user_mds_data *)arg;
1013                 lum = &lumd->lmd_lmm;
1014
1015                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1016                 if (rc)
1017                         RETURN(rc);
1018
1019                 OBD_ALLOC_LARGE(lmm, lmmsize);
1020                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1021                         GOTO(free_lmm, rc = -EFAULT);
1022
1023                 switch (lmm->lmm_magic) {
1024                 case LOV_USER_MAGIC_V1:
1025                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1026                                 break;
1027                         /* swab objects first so that stripes num will be sane */
1028                         lustre_swab_lov_user_md_objects(
1029                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1030                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1031                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1032                         break;
1033                 case LOV_USER_MAGIC_V3:
1034                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1035                                 break;
1036                         /* swab objects first so that stripes num will be sane */
1037                         lustre_swab_lov_user_md_objects(
1038                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1039                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1040                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1041                         break;
1042                 default:
1043                         GOTO(free_lmm, rc = -EINVAL);
1044                 }
1045
1046                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1047                 if (rc < 0)
1048                         GOTO(free_lmm, rc = -ENOMEM);
1049
1050                 /* Perform glimpse_size operation. */
1051                 memset(&st, 0, sizeof(st));
1052
1053                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1054                 if (rc)
1055                         GOTO(free_lsm, rc);
1056
1057                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1058                         GOTO(free_lsm, rc = -EFAULT);
1059
1060                 EXIT;
1061         free_lsm:
1062                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1063         free_lmm:
1064                 OBD_FREE_LARGE(lmm, lmmsize);
1065                 return rc;
1066         }
1067         case OBD_IOC_LLOG_CATINFO: {
1068                 struct ptlrpc_request *req = NULL;
1069                 char                  *buf = NULL;
1070                 char                  *str;
1071                 int                    len = 0;
1072
1073                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1074                 if (rc)
1075                         RETURN(rc);
1076                 data = (void *)buf;
1077
1078                 if (!data->ioc_inlbuf1) {
1079                         obd_ioctl_freedata(buf, len);
1080                         RETURN(-EINVAL);
1081                 }
1082
1083                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1084                                            &RQF_LLOG_CATINFO);
1085                 if (req == NULL)
1086                         GOTO(out_catinfo, rc = -ENOMEM);
1087
1088                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1089                                      data->ioc_inllen1);
1090                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1091                                      data->ioc_inllen2);
1092
1093                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1094                 if (rc) {
1095                         ptlrpc_request_free(req);
1096                         GOTO(out_catinfo, rc);
1097                 }
1098
1099                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1100                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1101                 if (data->ioc_inllen2) {
1102                         str = req_capsule_client_get(&req->rq_pill,
1103                                                      &RMF_STRING);
1104                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1105                 }
1106
1107                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1108                                      data->ioc_plen1);
1109                 ptlrpc_request_set_replen(req);
1110
1111                 rc = ptlrpc_queue_wait(req);
1112                 if (!rc) {
1113                         str = req_capsule_server_get(&req->rq_pill,
1114                                                      &RMF_STRING);
1115                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1116                                              data->ioc_plen1))
1117                                 rc = -EFAULT;
1118                 }
1119                 ptlrpc_req_finished(req);
1120         out_catinfo:
1121                 obd_ioctl_freedata(buf, len);
1122                 RETURN(rc);
1123         }
1124         case OBD_IOC_QUOTACHECK: {
1125                 struct obd_quotactl *oqctl;
1126                 int error = 0;
1127
1128                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1129                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1130                         RETURN(-EPERM);
1131
1132                 OBD_ALLOC_PTR(oqctl);
1133                 if (!oqctl)
1134                         RETURN(-ENOMEM);
1135                 oqctl->qc_type = arg;
1136                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1137                 if (rc < 0) {
1138                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1139                         error = rc;
1140                 }
1141
1142                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1143                 if (rc < 0)
1144                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1145
1146                 OBD_FREE_PTR(oqctl);
1147                 return error ?: rc;
1148         }
1149         case OBD_IOC_POLL_QUOTACHECK: {
1150                 struct if_quotacheck *check;
1151
1152                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1153                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1154                         RETURN(-EPERM);
1155
1156                 OBD_ALLOC_PTR(check);
1157                 if (!check)
1158                         RETURN(-ENOMEM);
1159
1160                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1161                                    NULL);
1162                 if (rc) {
1163                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1164                         if (cfs_copy_to_user((void *)arg, check,
1165                                              sizeof(*check)))
1166                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1167                         GOTO(out_poll, rc);
1168                 }
1169
1170                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1171                                    NULL);
1172                 if (rc) {
1173                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1174                         if (cfs_copy_to_user((void *)arg, check,
1175                                              sizeof(*check)))
1176                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1177                         GOTO(out_poll, rc);
1178                 }
1179         out_poll:
1180                 OBD_FREE_PTR(check);
1181                 RETURN(rc);
1182         }
1183         case OBD_IOC_QUOTACTL: {
1184                 struct if_quotactl *qctl;
1185                 int cmd, type, id, valid;
1186
1187                 OBD_ALLOC_PTR(qctl);
1188                 if (!qctl)
1189                         RETURN(-ENOMEM);
1190
1191                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1192                         GOTO(out_quotactl, rc = -EFAULT);
1193
1194                 cmd = qctl->qc_cmd;
1195                 type = qctl->qc_type;
1196                 id = qctl->qc_id;
1197                 valid = qctl->qc_valid;
1198
1199                 switch (cmd) {
1200                 case LUSTRE_Q_INVALIDATE:
1201                 case LUSTRE_Q_FINVALIDATE:
1202                 case Q_QUOTAON:
1203                 case Q_QUOTAOFF:
1204                 case Q_SETQUOTA:
1205                 case Q_SETINFO:
1206                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1207                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1208                                 GOTO(out_quotactl, rc = -EPERM);
1209                         break;
1210                 case Q_GETQUOTA:
1211                         if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1212                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1213                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1214                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1215                                 GOTO(out_quotactl, rc = -EPERM);
1216                         break;
1217                 case Q_GETINFO:
1218                         break;
1219                 default:
1220                         CERROR("unsupported quotactl op: %#x\n", cmd);
1221                         GOTO(out_quotactl, rc = -ENOTTY);
1222                 }
1223
1224                 if (valid != QC_GENERAL) {
1225                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1226                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1227
1228                         if (cmd == Q_GETINFO)
1229                                 qctl->qc_cmd = Q_GETOINFO;
1230                         else if (cmd == Q_GETQUOTA)
1231                                 qctl->qc_cmd = Q_GETOQUOTA;
1232                         else
1233                                 GOTO(out_quotactl, rc = -EINVAL);
1234
1235                         switch (valid) {
1236                         case QC_MDTIDX:
1237                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1238                                                    sbi->ll_md_exp,
1239                                                    sizeof(*qctl), qctl, NULL);
1240                                 break;
1241                         case QC_OSTIDX:
1242                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1243                                                    sbi->ll_dt_exp,
1244                                                    sizeof(*qctl), qctl, NULL);
1245                                 break;
1246                         case QC_UUID:
1247                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1248                                                    sbi->ll_md_exp,
1249                                                    sizeof(*qctl), qctl, NULL);
1250                                 if (rc == -EAGAIN)
1251                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1252                                                            sbi->ll_dt_exp,
1253                                                            sizeof(*qctl), qctl,
1254                                                            NULL);
1255                                 break;
1256                         default:
1257                                 rc = -EINVAL;
1258                                 break;
1259                         }
1260
1261                         if (rc)
1262                                 GOTO(out_quotactl, rc);
1263                         else
1264                                 qctl->qc_cmd = cmd;
1265                 } else {
1266                         struct obd_quotactl *oqctl;
1267
1268                         OBD_ALLOC_PTR(oqctl);
1269                         if (!oqctl)
1270                                 GOTO(out_quotactl, rc = -ENOMEM);
1271
1272                         QCTL_COPY(oqctl, qctl);
1273                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1274                         if (rc) {
1275                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1276                                         oqctl->qc_cmd = Q_QUOTAOFF;
1277                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1278                                 }
1279                                 OBD_FREE_PTR(oqctl);
1280                                 GOTO(out_quotactl, rc);
1281                         } else {
1282                                 QCTL_COPY(qctl, oqctl);
1283                                 OBD_FREE_PTR(oqctl);
1284                         }
1285                 }
1286
1287                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1288                         rc = -EFAULT;
1289
1290         out_quotactl:
1291                 OBD_FREE_PTR(qctl);
1292                 RETURN(rc);
1293         }
1294         case OBD_IOC_GETNAME: {
1295                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1296                 if (!obd)
1297                         RETURN(-EFAULT);
1298                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1299                                      strlen(obd->obd_name) + 1))
1300                         RETURN (-EFAULT);
1301                 RETURN(0);
1302         }
1303         case LL_IOC_FLUSHCTX:
1304                 RETURN(ll_flush_ctx(inode));
1305 #ifdef CONFIG_FS_POSIX_ACL
1306         case LL_IOC_RMTACL: {
1307             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1308                 inode == inode->i_sb->s_root->d_inode) {
1309                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1310
1311                 LASSERT(fd != NULL);
1312                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1313                 if (!rc)
1314                         fd->fd_flags |= LL_FILE_RMTACL;
1315                 RETURN(rc);
1316             } else
1317                 RETURN(0);
1318         }
1319 #endif
1320         case LL_IOC_GETOBDCOUNT: {
1321                 int count;
1322
1323                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1324                         RETURN(-EFAULT);
1325
1326                 if (!count) {
1327                         /* get ost count */
1328                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1329                         count = lov->desc.ld_tgt_count;
1330                 } else {
1331                         /* get mdt count */
1332                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1333                         count = lmv->desc.ld_tgt_count;
1334                 }
1335
1336                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1337                         RETURN(-EFAULT);
1338
1339                 RETURN(0);
1340         }
1341         case LL_IOC_PATH2FID:
1342                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1343                                      sizeof(struct lu_fid)))
1344                         RETURN(-EFAULT);
1345                 RETURN(0);
1346         case LL_IOC_GET_CONNECT_FLAGS: {
1347                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1348         }
1349         case OBD_IOC_CHANGELOG_SEND:
1350         case OBD_IOC_CHANGELOG_CLEAR:
1351                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1352                                     sizeof(struct ioc_changelog));
1353                 RETURN(rc);
1354         case OBD_IOC_FID2PATH:
1355                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1356         case LL_IOC_HSM_CT_START:
1357                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1358                                     sizeof(struct lustre_kernelcomm));
1359                 RETURN(rc);
1360
1361         default:
1362                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1363         }
1364 }
1365
1366 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1367 {
1368         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1369         loff_t pos = file->f_pos;
1370         loff_t ret;
1371         ENTRY;
1372
1373         if (origin == 1 && offset >= 0 && file->f_pos == DIR_END_OFF) {
1374                 CWARN("end of dir hash, DIR_END_OFF(-2) is returned\n");
1375                 RETURN(DIR_END_OFF);
1376         }
1377
1378         ret = default_llseek(file, offset, origin);
1379         if (ret >= 0) {
1380                 struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
1381
1382                 if (ll_need_32bit_api(sbi)) {
1383                         if (file->f_pos >> 32) {
1384                                 /* hash overflow, simple revert */
1385                                 file->f_pos = pos;
1386                                 RETURN(-EOVERFLOW);
1387                         } else {
1388                                 fd->fd_dir.lfd_pos = file->f_pos << 32;
1389                         }
1390                 } else {
1391                         fd->fd_dir.lfd_pos = file->f_pos;
1392                 }
1393         }
1394         RETURN(ret);
1395 }
1396
1397 int ll_dir_open(struct inode *inode, struct file *file)
1398 {
1399         ENTRY;
1400         RETURN(ll_file_open(inode, file));
1401 }
1402
1403 int ll_dir_release(struct inode *inode, struct file *file)
1404 {
1405         ENTRY;
1406         RETURN(ll_file_release(inode, file));
1407 }
1408
1409 struct file_operations ll_dir_operations = {
1410         .llseek   = ll_dir_seek,
1411         .open     = ll_dir_open,
1412         .release  = ll_dir_release,
1413         .read     = generic_read_dir,
1414         .readdir  = ll_readdir,
1415         .ioctl    = ll_dir_ioctl
1416 };