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  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/dir.c
  40  *
  41  * Directory code for lustre client.
  42  */
  43
  44 #include <linux/fs.h>
  45 #include <linux/pagemap.h>
  46 #include <linux/mm.h>
  47 #include <linux/version.h>
  48 #include <linux/smp_lock.h>
  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 <obd_support.h>
  56 #include <obd_class.h>
  57 #include <lustre_lib.h>
  58 #include <lustre/lustre_idl.h>
  59 #include <lustre_lite.h>
  60 #include <lustre_dlm.h>
  61 #include <lustre_fid.h>
  62 #include "llite_internal.h"
  63
  64 #ifndef HAVE_PAGE_CHECKED
  65 #ifdef HAVE_PG_FS_MISC
  66 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
  67 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
  68 #else
  69 #error PageChecked or PageFsMisc not defined in kernel
  70 #endif
  71 #endif
  72
  73 /*
  74  * (new) readdir implementation overview.
  75  *
  76  * Original lustre readdir implementation cached exact copy of raw directory
  77  * pages on the client. These pages were indexed in client page cache by
  78  * logical offset in the directory file. This design, while very simple and
  79  * intuitive had some inherent problems:
  80  *
  81  *     . it implies that byte offset to the directory entry serves as a
  82  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  83  *     ext3/htree directory entries may move due to splits, and more
  84  *     importantly,
  85  *
  86  *     . it is incompatible with the design of split directories for cmd3,
  87  *     that assumes that names are distributed across nodes based on their
  88  *     hash, and so readdir should be done in hash order.
  89  *
  90  * New readdir implementation does readdir in hash order, and uses hash of a
  91  * file name as a telldir/seekdir cookie. This led to number of complications:
  92  *
  93  *     . hash is not unique, so it cannot be used to index cached directory
  94  *     pages on the client (note, that it requires a whole pageful of hash
  95  *     collided entries to cause two pages to have identical hashes);
  96  *
  97  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  98  *     entry cookie. ext3/htree has the same problem and lustre implementation
  99  *     mimics their solution: seekdir(hash) positions directory at the first
 100  *     entry with the given hash.
 101  *
 102  * Client side.
 103  *
 104  * 0. caching
 105  *
 106  * Client caches directory pages using hash of the first entry as an index. As
 107  * noted above hash is not unique, so this solution doesn't work as is:
 108  * special processing is needed for "page hash chains" (i.e., sequences of
 109  * pages filled with entries all having the same hash value).
 110  *
 111  * First, such chains have to be detected. To this end, server returns to the
 112  * client the hash of the first entry on the page next to one returned. When
 113  * client detects that this hash is the same as hash of the first entry on the
 114  * returned page, page hash collision has to be handled. Pages in the
 115  * hash chain, except first one, are termed "overflow pages".
 116  *
 117  * Solution to index uniqueness problem is to not cache overflow
 118  * pages. Instead, when page hash collision is detected, all overflow pages
 119  * from emerging chain are immediately requested from the server and placed in
 120  * a special data structure (struct ll_dir_chain). This data structure is used
 121  * by ll_readdir() to process entries from overflow pages. When readdir
 122  * invocation finishes, overflow pages are discarded. If page hash collision
 123  * chain weren't completely processed, next call to readdir will again detect
 124  * page hash collision, again read overflow pages in, process next portion of
 125  * entries and again discard the pages. This is not as wasteful as it looks,
 126  * because, given reasonable hash, page hash collisions are extremely rare.
 127  *
 128  * 1. directory positioning
 129  *
 130  * When seekdir(hash) is called, original
 131  *
 132  *
 133  *
 134  *
 135  *
 136  *
 137  *
 138  *
 139  * Server.
 140  *
 141  * identification of and access to overflow pages
 142  *
 143  * page format
 144  *
 145  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 146  * a header lu_dirpage which describes the start/end hash, and whether this
 147  * page is empty (contains no dir entry) or hash collide with next page.
 148  * After client receives reply, several pages will be integrated into dir page
 149  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
 150  * lu_dirpage for this integrated page will be adjusted.
 151  *
 152  */
 153
 154 /* returns the page unlocked, but with a reference */
 155 static int ll_dir_readpage(struct file *file, struct page *page0)
 156 {
 157         struct inode *inode = page0->mapping->host;
 158         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 159         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 160         struct ptlrpc_request *request;
 161         struct mdt_body *body;
 162         struct obd_capa *oc;
 163         __u64 hash;
 164         struct page **page_pool;
 165         struct page *page;
 166 #ifndef HAVE_ADD_TO_PAGE_CACHE_LRU
 167         struct pagevec lru_pvec;
 168 #endif
 169         struct lu_dirpage *dp;
 170         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> CFS_PAGE_SHIFT;
 171         int nrdpgs = 0; /* number of pages read actually */
 172         int npages;
 173         int i;
 174         int rc;
 175         ENTRY;
 176
 177         if (file) {
 178                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
 179
 180                 hash = fd->fd_dir.lfd_next;
 181         } else {
 182                 struct ll_inode_info *lli = ll_i2info(inode);
 183
 184                 cfs_spin_lock(&lli->lli_sa_lock);
 185                 if (lli->lli_sai)
 186                         LASSERT(lli->lli_sai->sai_pid == cfs_curproc_pid());
 187                 else
 188                         LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
 189                 hash = lli->lli_sa_pos;
 190                 cfs_spin_unlock(&lli->lli_sa_lock);
 191         }
 192         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 193                inode->i_ino, inode->i_generation, inode, hash);
 194
 195         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
 196
 197         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 198         if (page_pool != NULL) {
 199                 page_pool[0] = page0;
 200         } else {
 201                 page_pool = &page0;
 202                 max_pages = 1;
 203         }
 204         for (npages = 1; npages < max_pages; npages++) {
 205                 page = page_cache_alloc_cold(inode->i_mapping);
 206                 if (!page)
 207                         break;
 208                 page_pool[npages] = page;
 209         }
 210
 211         oc = ll_mdscapa_get(inode);
 212         rc = md_readpage(exp, ll_inode2fid(inode), oc, hash, page_pool, npages,
 213                          &request);
 214         capa_put(oc);
 215         if (rc == 0) {
 216                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 217                 /* Checked by mdc_readpage() */
 218                 LASSERT(body != NULL);
 219
 220                 if (body->valid & OBD_MD_FLSIZE)
 221                         cl_isize_write(inode, body->size);
 222
 223                 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
 224                          >> CFS_PAGE_SHIFT;
 225                 SetPageUptodate(page0);
 226         }
 227         unlock_page(page0);
 228         ptlrpc_req_finished(request);
 229
 230         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 231
 232         ll_pagevec_init(&lru_pvec, 0);
 233         for (i = 1; i < npages; i++) {
 234                 unsigned long offset;
 235                 int ret;
 236
 237                 page = page_pool[i];
 238
 239                 if (rc < 0 || i >= nrdpgs) {
 240                         page_cache_release(page);
 241                         continue;
 242                 }
 243
 244                 SetPageUptodate(page);
 245
 246                 dp = cfs_kmap(page);
 247                 hash = le64_to_cpu(dp->ldp_hash_start);
 248                 cfs_kunmap(page);
 249
 250                 offset = hash_x_index(hash, hash64);
 251
 252                 prefetchw(&page->flags);
 253                 ret = ll_add_to_page_cache_lru(page, inode->i_mapping, offset,
 254                                                GFP_KERNEL);
 255                 if (ret == 0) {
 256                         unlock_page(page);
 257                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 258                                 ll_pagevec_lru_add_file(&lru_pvec);
 259                 } else {
 260                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 261                                " %d\n", offset, ret);
 262                 }
 263                 page_cache_release(page);
 264         }
 265         ll_pagevec_lru_add_file(&lru_pvec);
 266
 267         if (page_pool != &page0)
 268                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 269         EXIT;
 270         return rc;
 271 }
 272
 273 #ifndef MS_HAS_NEW_AOPS
 274 struct address_space_operations ll_dir_aops = {
 275         .readpage  = ll_dir_readpage,
 276 };
 277 #else
 278 struct address_space_operations_ext ll_dir_aops = {
 279         .orig_aops.readpage  = ll_dir_readpage,
 280 };
 281 #endif
 282
 283 static void ll_check_page(struct inode *dir, struct page *page)
 284 {
 285         /* XXX: check page format later */
 286         SetPageChecked(page);
 287 }
 288
 289 static void ll_release_page(struct page *page, __u64 hash,
 290                             __u64 start, __u64 end)
 291 {
 292         kunmap(page);
 293         lock_page(page);
 294         if (likely(page->mapping != NULL)) {
 295                 truncate_complete_page(page->mapping, page);
 296                 unlock_page(page);
 297         } else {
 298                 unlock_page(page);
 299                 CWARN("NULL mapping page %p, truncated by others: "
 300                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
 301                       page, hash, start, end);
 302         }
 303         page_cache_release(page);
 304 }
 305
 306 /*
 307  * Find, kmap and return page that contains given hash.
 308  */
 309 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 310                                        __u64 *start, __u64 *end)
 311 {
 312         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 313         struct address_space *mapping = dir->i_mapping;
 314         /*
 315          * Complement of hash is used as an index so that
 316          * radix_tree_gang_lookup() can be used to find a page with starting
 317          * hash _smaller_ than one we are looking for.
 318          */
 319         unsigned long offset = hash_x_index(*hash, hash64);
 320         struct page *page;
 321         int found;
 322
 323         TREE_READ_LOCK_IRQ(mapping);
 324         found = radix_tree_gang_lookup(&mapping->page_tree,
 325                                        (void **)&page, offset, 1);
 326         if (found > 0) {
 327                 struct lu_dirpage *dp;
 328
 329                 page_cache_get(page);
 330                 TREE_READ_UNLOCK_IRQ(mapping);
 331                 /*
 332                  * In contrast to find_lock_page() we are sure that directory
 333                  * page cannot be truncated (while DLM lock is held) and,
 334                  * hence, can avoid restart.
 335                  *
 336                  * In fact, page cannot be locked here at all, because
 337                  * ll_dir_readpage() does synchronous io.
 338                  */
 339                 wait_on_page(page);
 340                 if (PageUptodate(page)) {
 341                         dp = cfs_kmap(page);
 342                         if (BITS_PER_LONG == 32 && hash64) {
 343                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 344                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 345                                 *hash  = *hash >> 32;
 346                         } else {
 347                                 *start = le64_to_cpu(dp->ldp_hash_start);
 348                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 349                         }
 350                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 351                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 352                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 353                                offset, *start, *end, *hash);
 354                         if (*hash > *end || (*end != *start && *hash == *end)) {
 355                                 /*
 356                                  * upon hash collision, remove this page,
 357                                  * otherwise put page reference, and
 358                                  * ll_get_dir_page() will issue RPC to fetch
 359                                  * the page we want.
 360                                  */
 361                                 if (dp->ldp_flags & cpu_to_le32(LDF_COLLIDE)) {
 362                                         ll_release_page(page, *hash, *start, *end);
 363                                 } else {
 364                                         cfs_kunmap(page);
 365                                         page_cache_release(page);
 366                                 }
 367                                 page = NULL;
 368                         }
 369                 } else {
 370                         page_cache_release(page);
 371                         page = ERR_PTR(-EIO);
 372                 }
 373
 374         } else {
 375                 TREE_READ_UNLOCK_IRQ(mapping);
 376                 page = NULL;
 377         }
 378         return page;
 379 }
 380
 381 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
 382                              int exact, struct ll_dir_chain *chain)
 383 {
 384         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 385         struct address_space *mapping = dir->i_mapping;
 386         struct lustre_handle lockh;
 387         struct lu_dirpage *dp;
 388         struct page *page;
 389         ldlm_mode_t mode;
 390         int rc;
 391         __u64 start = 0;
 392         __u64 end = 0;
 393         __u64 lhash = hash;
 394         struct ll_inode_info *lli = ll_i2info(dir);
 395         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 396
 397         mode = LCK_PR;
 398         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 399                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 400         if (!rc) {
 401                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 402                        ll_md_blocking_ast, ldlm_completion_ast,
 403                        NULL, NULL, dir };
 404                 struct lookup_intent it = { .it_op = IT_READDIR };
 405                 struct ptlrpc_request *request;
 406                 struct md_op_data *op_data;
 407
 408                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 409                                              LUSTRE_OPC_ANY, NULL);
 410                 if (IS_ERR(op_data))
 411                         return (void *)op_data;
 412
 413                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 414                                 op_data, &lockh, NULL, 0, NULL, 0);
 415
 416                 ll_finish_md_op_data(op_data);
 417
 418                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 419                 if (request)
 420                         ptlrpc_req_finished(request);
 421                 if (rc < 0) {
 422                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 423                                PFID(ll_inode2fid(dir)), hash, rc);
 424                         return ERR_PTR(rc);
 425                 }
 426         } else {
 427                 /* for cross-ref object, l_ast_data of the lock may not be set,
 428                  * we reset it here */
 429                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 430                                  dir, NULL);
 431         }
 432         ldlm_lock_dump_handle(D_OTHER, &lockh);
 433
 434         cfs_down(&lli->lli_readdir_sem);
 435         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 436         if (IS_ERR(page)) {
 437                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 438                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 439                 GOTO(out_unlock, page);
 440         }
 441
 442         if (page != NULL) {
 443                 /*
 444                  * XXX nikita: not entirely correct handling of a corner case:
 445                  * suppose hash chain of entries with hash value HASH crosses
 446                  * border between pages P0 and P1. First both P0 and P1 are
 447                  * cached, seekdir() is called for some entry from the P0 part
 448                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 449                  * happens and finds P1, as it starts with matching hash
 450                  * value. Remaining entries from P0 part of the chain are
 451                  * skipped. (Is that really a bug?)
 452                  *
 453                  * Possible solutions: 0. don't cache P1 is such case, handle
 454                  * it as an "overflow" page. 1. invalidate all pages at
 455                  * once. 2. use HASH|1 as an index for P1.
 456                  */
 457                 if (exact && lhash != start) {
 458                         /*
 459                          * readdir asked for a page starting _exactly_ from
 460                          * given hash, but cache contains stale page, with
 461                          * entries with smaller hash values. Stale page should
 462                          * be invalidated, and new one fetched.
 463                          */
 464                         CDEBUG(D_OTHER, "Stale readpage page %p: "
 465                                "start = "LPX64",end = "LPX64"hash ="LPX64"\n",
 466                                page, start, end, lhash);
 467                         ll_release_page(page, lhash, start, end);
 468                 } else {
 469                         GOTO(hash_collision, page);
 470                 }
 471         }
 472
 473         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 474                                (filler_t*)mapping->a_ops->readpage, filp);
 475         if (IS_ERR(page)) {
 476                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 477                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 478                 GOTO(out_unlock, page);
 479         }
 480
 481         wait_on_page(page);
 482         (void)kmap(page);
 483         if (!PageUptodate(page)) {
 484                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 485                        PFID(ll_inode2fid(dir)), hash, -5);
 486                 goto fail;
 487         }
 488         if (!PageChecked(page))
 489                 ll_check_page(dir, page);
 490         if (PageError(page)) {
 491                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 492                        PFID(ll_inode2fid(dir)), hash, -5);
 493                 goto fail;
 494         }
 495 hash_collision:
 496         dp = page_address(page);
 497         if (BITS_PER_LONG == 32 && hash64) {
 498                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 499                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 500                 lhash = hash >> 32;
 501         } else {
 502                 start = le64_to_cpu(dp->ldp_hash_start);
 503                 end   = le64_to_cpu(dp->ldp_hash_end);
 504                 lhash = hash;
 505         }
 506         if (end == start) {
 507                 LASSERT(start == lhash);
 508                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 509                 if (BITS_PER_LONG == 32 && hash64)
 510                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 511                               "] with hash "LPU64"\n",
 512                               le64_to_cpu(dp->ldp_hash_start),
 513                               le64_to_cpu(dp->ldp_hash_end), hash);
 514                 /*
 515                  * Fetch whole overflow chain...
 516                  *
 517                  * XXX not yet.
 518                  */
 519                 goto fail;
 520         }
 521 out_unlock:
 522         cfs_up(&lli->lli_readdir_sem);
 523         ldlm_lock_decref(&lockh, mode);
 524         return page;
 525
 526 fail:
 527         ll_put_page(page);
 528         page = ERR_PTR(-EIO);
 529         goto out_unlock;
 530 }
 531
 532 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 533 {
 534         struct inode         *inode      = filp->f_dentry->d_inode;
 535         struct ll_inode_info *info       = ll_i2info(inode);
 536         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 537         struct ll_file_data  *fd         = LUSTRE_FPRIVATE(filp);
 538         __u64                 pos        = fd->fd_dir.lfd_pos;
 539         int                   api32      = ll_need_32bit_api(sbi);
 540         int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 541         struct page          *page;
 542         struct ll_dir_chain   chain;
 543         int                   done;
 544         int                   shift;
 545         int                   rc;
 546         ENTRY;
 547
 548         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 549                inode->i_ino, inode->i_generation, inode,
 550                (unsigned long)pos, i_size_read(inode), api32);
 551
 552         if (pos == MDS_DIR_END_OFF)
 553                 /*
 554                  * end-of-file.
 555                  */
 556                 RETURN(0);
 557
 558         rc    = 0;
 559         done  = 0;
 560         shift = 0;
 561         ll_dir_chain_init(&chain);
 562
 563         fd->fd_dir.lfd_next = pos;
 564         page = ll_get_dir_page(filp, inode, pos, 0, &chain);
 565
 566         while (rc == 0 && !done) {
 567                 struct lu_dirpage *dp;
 568                 struct lu_dirent  *ent;
 569
 570                 if (!IS_ERR(page)) {
 571                         /*
 572                          * If page is empty (end of directory is reached),
 573                          * use this value.
 574                          */
 575                         __u64 hash = MDS_DIR_END_OFF;
 576                         __u64 next;
 577
 578                         dp = page_address(page);
 579                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 580                              ent = lu_dirent_next(ent)) {
 581                                 __u16          type;
 582                                 int            namelen;
 583                                 struct lu_fid  fid;
 584                                 __u64          lhash;
 585                                 __u64          ino;
 586
 587                                 /*
 588                                  * XXX: implement correct swabbing here.
 589                                  */
 590
 591                                 hash = le64_to_cpu(ent->lde_hash);
 592                                 if (hash < pos)
 593                                         /*
 594                                          * Skip until we find target hash
 595                                          * value.
 596                                          */
 597                                         continue;
 598
 599                                 namelen = le16_to_cpu(ent->lde_namelen);
 600                                 if (namelen == 0)
 601                                         /*
 602                                          * Skip dummy record.
 603                                          */
 604                                         continue;
 605
 606                                 if (api32 && hash64)
 607                                         lhash = hash >> 32;
 608                                 else
 609                                         lhash = hash;
 610                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 611                                 ino = cl_fid_build_ino(&fid, api32);
 612                                 type = ll_dirent_type_get(ent);
 613                                 /* For 'll_nfs_get_name_filldir()', it will try
 614                                  * to access the 'ent' through its 'lde_name',
 615                                  * so the parameter 'name' for 'filldir()' must
 616                                  * be part of the 'ent'. */
 617                                 done = filldir(cookie, ent->lde_name, namelen,
 618                                                lhash, ino, type);
 619                         }
 620                         next = le64_to_cpu(dp->ldp_hash_end);
 621                         ll_put_page(page);
 622                         if (!done) {
 623                                 pos = next;
 624                                 if (pos == MDS_DIR_END_OFF) {
 625                                         /*
 626                                          * End of directory reached.
 627                                          */
 628                                         done = 1;
 629                                 } else if (1 /* chain is exhausted*/) {
 630                                         /*
 631                                          * Normal case: continue to the next
 632                                          * page.
 633                                          */
 634                                         fd->fd_dir.lfd_next = pos;
 635                                         page = ll_get_dir_page(filp, inode, pos,
 636                                                                1, &chain);
 637                                 } else {
 638                                         /*
 639                                          * go into overflow page.
 640                                          */
 641                                 }
 642                         } else {
 643                                 pos = hash;
 644                         }
 645                 } else {
 646                         rc = PTR_ERR(page);
 647                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 648                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 649                 }
 650         }
 651
 652         fd->fd_dir.lfd_pos = pos;
 653         if (pos == MDS_DIR_END_OFF) {
 654                 if (api32)
 655                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 656                 else
 657                         filp->f_pos = LL_DIR_END_OFF;
 658         } else {
 659                 if (api32 && hash64)
 660                         filp->f_pos = pos >> 32;
 661                 else
 662                         filp->f_pos = pos;
 663         }
 664         filp->f_version = inode->i_version;
 665         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 666
 667         ll_dir_chain_fini(&chain);
 668
 669         RETURN(rc);
 670 }
 671
 672 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 673 {
 674         struct mgs_send_param *msp;
 675         int rc = 0;
 676
 677         OBD_ALLOC_PTR(msp);
 678         if (!msp)
 679                 return -ENOMEM;
 680
 681         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 682         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 683                                 sizeof(struct mgs_send_param), msp, NULL);
 684         if (rc)
 685                 CERROR("Failed to set parameter: %d\n", rc);
 686         OBD_FREE_PTR(msp);
 687
 688         return rc;
 689 }
 690
 691 char *ll_get_fsname(struct inode *inode)
 692 {
 693         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 694         char *ptr, *fsname;
 695         int len;
 696
 697         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 698         len = strlen(lsi->lsi_lmd->lmd_profile);
 699         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 700         if (ptr && (strcmp(ptr, "-client") == 0))
 701                 len -= 7;
 702         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 703         fsname[len] = '\0';
 704
 705         return fsname;
 706 }
 707
 708 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 709                      int set_default)
 710 {
 711         struct ll_sb_info *sbi = ll_i2sbi(inode);
 712         struct md_op_data *op_data;
 713         struct ptlrpc_request *req = NULL;
 714         int rc = 0;
 715         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 716         struct obd_device *mgc = lsi->lsi_mgc;
 717         char *fsname = NULL, *param = NULL;
 718         int lum_size;
 719
 720         if (lump != NULL) {
 721                 /*
 722                  * This is coming from userspace, so should be in
 723                  * local endian.  But the MDS would like it in little
 724                  * endian, so we swab it before we send it.
 725                  */
 726                 switch (lump->lmm_magic) {
 727                 case LOV_USER_MAGIC_V1: {
 728                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 729                                 lustre_swab_lov_user_md_v1(lump);
 730                         lum_size = sizeof(struct lov_user_md_v1);
 731                         break;
 732                         }
 733                 case LOV_USER_MAGIC_V3: {
 734                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 735                                 lustre_swab_lov_user_md_v3(
 736                                         (struct lov_user_md_v3 *)lump);
 737                         lum_size = sizeof(struct lov_user_md_v3);
 738                         break;
 739                         }
 740                 default: {
 741                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 742                                         " %#08x != %#08x nor %#08x\n",
 743                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 744                                         LOV_USER_MAGIC_V3);
 745                         RETURN(-EINVAL);
 746                         }
 747                }
 748         } else {
 749                 lum_size = sizeof(struct lov_user_md_v1);
 750         }
 751
 752         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 753                                      LUSTRE_OPC_ANY, NULL);
 754         if (IS_ERR(op_data))
 755                 RETURN(PTR_ERR(op_data));
 756
 757         /* swabbing is done in lov_setstripe() on server side */
 758         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 759                         NULL, 0, &req, NULL);
 760         ll_finish_md_op_data(op_data);
 761         ptlrpc_req_finished(req);
 762         if (rc) {
 763                 if (rc != -EPERM && rc != -EACCES)
 764                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 765         }
 766
 767         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 768          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 769          need the make the distiction between the 2 versions */
 770         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 771                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 772
 773                 /* Get fsname and assume devname to be -MDT0000. */
 774                 fsname = ll_get_fsname(inode);
 775                 /* Set root stripesize */
 776                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 777                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 778                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 779                 if (rc)
 780                         goto end;
 781
 782                 /* Set root stripecount */
 783                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 784                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 785                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 786                 if (rc)
 787                         goto end;
 788
 789                 /* Set root stripeoffset */
 790                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 791                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 792                         (typeof(lump->lmm_stripe_offset))(-1));
 793                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 794                 if (rc)
 795                         goto end;
 796 end:
 797                 if (fsname)
 798                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 799                 if (param)
 800                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 801         }
 802         return rc;
 803 }
 804
 805 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 806                      int *lmm_size, struct ptlrpc_request **request)
 807 {
 808         struct ll_sb_info *sbi = ll_i2sbi(inode);
 809         struct mdt_body   *body;
 810         struct lov_mds_md *lmm = NULL;
 811         struct ptlrpc_request *req = NULL;
 812         int rc, lmmsize;
 813         struct md_op_data *op_data;
 814
 815         rc = ll_get_max_mdsize(sbi, &lmmsize);
 816         if (rc)
 817                 RETURN(rc);
 818
 819         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 820                                      0, lmmsize, LUSTRE_OPC_ANY,
 821                                      NULL);
 822         if (op_data == NULL)
 823                 RETURN(-ENOMEM);
 824
 825         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 826         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 827         ll_finish_md_op_data(op_data);
 828         if (rc < 0) {
 829                 CDEBUG(D_INFO, "md_getattr failed on inode "
 830                        "%lu/%u: rc %d\n", inode->i_ino,
 831                        inode->i_generation, rc);
 832                 GOTO(out, rc);
 833         }
 834
 835         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 836         LASSERT(body != NULL);
 837
 838         lmmsize = body->eadatasize;
 839
 840         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 841             lmmsize == 0) {
 842                 GOTO(out, rc = -ENODATA);
 843         }
 844
 845         lmm = req_capsule_server_sized_get(&req->rq_pill,
 846                                            &RMF_MDT_MD, lmmsize);
 847         LASSERT(lmm != NULL);
 848
 849         /*
 850          * This is coming from the MDS, so is probably in
 851          * little endian.  We convert it to host endian before
 852          * passing it to userspace.
 853          */
 854         /* We don't swab objects for directories */
 855         switch (le32_to_cpu(lmm->lmm_magic)) {
 856         case LOV_MAGIC_V1:
 857                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 858                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 859                 break;
 860         case LOV_MAGIC_V3:
 861                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 862                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 863                 break;
 864         default:
 865                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 866                 rc = -EPROTO;
 867         }
 868 out:
 869         *lmmp = lmm;
 870         *lmm_size = lmmsize;
 871         *request = req;
 872         return rc;
 873 }
 874
 875 /*
 876  *  Get MDT index for the inode.
 877  */
 878 int ll_get_mdt_idx(struct inode *inode)
 879 {
 880         struct ll_sb_info *sbi = ll_i2sbi(inode);
 881         struct md_op_data *op_data;
 882         int rc, mdtidx;
 883         ENTRY;
 884
 885         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 886                                      0, LUSTRE_OPC_ANY, NULL);
 887         if (op_data == NULL)
 888                 RETURN(-ENOMEM);
 889
 890         op_data->op_valid |= OBD_MD_MDTIDX;
 891         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 892         mdtidx = op_data->op_mds;
 893         ll_finish_md_op_data(op_data);
 894         if (rc < 0) {
 895                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 896                 RETURN(rc);
 897         }
 898         return mdtidx;
 899 }
 900
 901 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 902 {
 903         void *ptr;
 904         int rc;
 905
 906         OBD_ALLOC(ptr, len);
 907         if (ptr == NULL)
 908                 return -ENOMEM;
 909         if (cfs_copy_from_user(ptr, data, len)) {
 910                 OBD_FREE(ptr, len);
 911                 return -EFAULT;
 912         }
 913         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 914         OBD_FREE(ptr, len);
 915         return rc;
 916 }
 917
 918 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 919 {
 920         int cmd = qctl->qc_cmd;
 921         int type = qctl->qc_type;
 922         int id = qctl->qc_id;
 923         int valid = qctl->qc_valid;
 924         int rc = 0;
 925         ENTRY;
 926
 927         switch (cmd) {
 928         case LUSTRE_Q_INVALIDATE:
 929         case LUSTRE_Q_FINVALIDATE:
 930         case Q_QUOTAON:
 931         case Q_QUOTAOFF:
 932         case Q_SETQUOTA:
 933         case Q_SETINFO:
 934                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 935                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 936                         RETURN(-EPERM);
 937                 break;
 938         case Q_GETQUOTA:
 939                 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
 940                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 941                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 942                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 943                         RETURN(-EPERM);
 944                 break;
 945         case Q_GETINFO:
 946                 break;
 947         default:
 948                 CERROR("unsupported quotactl op: %#x\n", cmd);
 949                 RETURN(-ENOTTY);
 950         }
 951
 952         if (valid != QC_GENERAL) {
 953                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 954                         RETURN(-EOPNOTSUPP);
 955
 956                 if (cmd == Q_GETINFO)
 957                         qctl->qc_cmd = Q_GETOINFO;
 958                 else if (cmd == Q_GETQUOTA)
 959                         qctl->qc_cmd = Q_GETOQUOTA;
 960                 else
 961                         RETURN(-EINVAL);
 962
 963                 switch (valid) {
 964                 case QC_MDTIDX:
 965                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 966                                            sizeof(*qctl), qctl, NULL);
 967                         break;
 968                 case QC_OSTIDX:
 969                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 970                                            sizeof(*qctl), qctl, NULL);
 971                         break;
 972                 case QC_UUID:
 973                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 974                                            sizeof(*qctl), qctl, NULL);
 975                         if (rc == -EAGAIN)
 976                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 977                                                    sbi->ll_dt_exp,
 978                                                    sizeof(*qctl), qctl, NULL);
 979                         break;
 980                 default:
 981                         rc = -EINVAL;
 982                         break;
 983                 }
 984
 985                 if (rc)
 986                         RETURN(rc);
 987
 988                 qctl->qc_cmd = cmd;
 989         } else {
 990                 struct obd_quotactl *oqctl;
 991
 992                 OBD_ALLOC_PTR(oqctl);
 993                 if (oqctl == NULL)
 994                         RETURN(-ENOMEM);
 995
 996                 QCTL_COPY(oqctl, qctl);
 997                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 998                 if (rc) {
 999                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1000                                 oqctl->qc_cmd = Q_QUOTAOFF;
1001                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1002                         }
1003                         OBD_FREE_PTR(oqctl);
1004                         RETURN(rc);
1005                 }
1006
1007                 QCTL_COPY(qctl, oqctl);
1008                 OBD_FREE_PTR(oqctl);
1009         }
1010
1011         RETURN(rc);
1012 }
1013
1014 static int ll_dir_ioctl(struct inode *inode, struct file *file,
1015                         unsigned int cmd, unsigned long arg)
1016 {
1017         struct ll_sb_info *sbi = ll_i2sbi(inode);
1018         struct obd_ioctl_data *data;
1019         int rc = 0;
1020         ENTRY;
1021
1022         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1023                inode->i_ino, inode->i_generation, inode, cmd);
1024
1025         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1026         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1027                 return -ENOTTY;
1028
1029         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1030         switch(cmd) {
1031         case FSFILT_IOC_GETFLAGS:
1032         case FSFILT_IOC_SETFLAGS:
1033                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1034         case FSFILT_IOC_GETVERSION_OLD:
1035         case FSFILT_IOC_GETVERSION:
1036                 RETURN(put_user(inode->i_generation, (int *)arg));
1037         /* We need to special case any other ioctls we want to handle,
1038          * to send them to the MDS/OST as appropriate and to properly
1039          * network encode the arg field.
1040         case FSFILT_IOC_SETVERSION_OLD:
1041         case FSFILT_IOC_SETVERSION:
1042         */
1043         case LL_IOC_GET_MDTIDX: {
1044                 int mdtidx;
1045
1046                 mdtidx = ll_get_mdt_idx(inode);
1047                 if (mdtidx < 0)
1048                         RETURN(mdtidx);
1049
1050                 if (put_user((int)mdtidx, (int*)arg))
1051                         RETURN(-EFAULT);
1052
1053                 return 0;
1054         }
1055         case IOC_MDC_LOOKUP: {
1056                 struct ptlrpc_request *request = NULL;
1057                 int namelen, len = 0;
1058                 char *buf = NULL;
1059                 char *filename;
1060                 struct md_op_data *op_data;
1061
1062                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1063                 if (rc)
1064                         RETURN(rc);
1065                 data = (void *)buf;
1066
1067                 filename = data->ioc_inlbuf1;
1068                 namelen = strlen(filename);
1069
1070                 if (namelen < 1) {
1071                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1072                         GOTO(out_free, rc = -EINVAL);
1073                 }
1074
1075                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1076                                              0, LUSTRE_OPC_ANY, NULL);
1077                 if (op_data == NULL)
1078                         GOTO(out_free, rc = -ENOMEM);
1079
1080                 op_data->op_valid = OBD_MD_FLID;
1081                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1082                 ll_finish_md_op_data(op_data);
1083                 if (rc < 0) {
1084                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1085                         GOTO(out_free, rc);
1086                 }
1087                 ptlrpc_req_finished(request);
1088                 EXIT;
1089 out_free:
1090                 obd_ioctl_freedata(buf, len);
1091                 return rc;
1092         }
1093         case LL_IOC_LOV_SETSTRIPE: {
1094                 struct lov_user_md_v3 lumv3;
1095                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1096                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1097                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1098
1099                 int set_default = 0;
1100
1101                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1102                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1103                         sizeof(lumv3p->lmm_objects[0]));
1104                 /* first try with v1 which is smaller than v3 */
1105                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1106                         RETURN(-EFAULT);
1107
1108                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1109                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1110                                 RETURN(-EFAULT);
1111                 }
1112
1113                 if (inode->i_sb->s_root == file->f_dentry)
1114                         set_default = 1;
1115
1116                 /* in v1 and v3 cases lumv1 points to data */
1117                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1118
1119                 RETURN(rc);
1120         }
1121         case LL_IOC_OBD_STATFS:
1122                 RETURN(ll_obd_statfs(inode, (void *)arg));
1123         case LL_IOC_LOV_GETSTRIPE:
1124         case LL_IOC_MDC_GETINFO:
1125         case IOC_MDC_GETFILEINFO:
1126         case IOC_MDC_GETFILESTRIPE: {
1127                 struct ptlrpc_request *request = NULL;
1128                 struct lov_user_md *lump;
1129                 struct lov_mds_md *lmm = NULL;
1130                 struct mdt_body *body;
1131                 char *filename = NULL;
1132                 int lmmsize;
1133
1134                 if (cmd == IOC_MDC_GETFILEINFO ||
1135                     cmd == IOC_MDC_GETFILESTRIPE) {
1136                         filename = getname((const char *)arg);
1137                         if (IS_ERR(filename))
1138                                 RETURN(PTR_ERR(filename));
1139
1140                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1141                                                       &lmmsize, &request);
1142                 } else {
1143                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1144                 }
1145
1146                 if (request) {
1147                         body = req_capsule_server_get(&request->rq_pill,
1148                                                       &RMF_MDT_BODY);
1149                         LASSERT(body != NULL);
1150                 } else {
1151                         GOTO(out_req, rc);
1152                 }
1153
1154                 if (rc < 0) {
1155                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1156                                                cmd == LL_IOC_MDC_GETINFO))
1157                                 GOTO(skip_lmm, rc = 0);
1158                         else
1159                                 GOTO(out_req, rc);
1160                 }
1161
1162                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1163                     cmd == LL_IOC_LOV_GETSTRIPE) {
1164                         lump = (struct lov_user_md *)arg;
1165                 } else {
1166                         struct lov_user_mds_data *lmdp;
1167                         lmdp = (struct lov_user_mds_data *)arg;
1168                         lump = &lmdp->lmd_lmm;
1169                 }
1170                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1171                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1172                                 GOTO(out_req, rc = -EFAULT);
1173                         rc = -EOVERFLOW;
1174                 }
1175         skip_lmm:
1176                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1177                         struct lov_user_mds_data *lmdp;
1178                         lstat_t st = { 0 };
1179
1180                         st.st_dev     = inode->i_sb->s_dev;
1181                         st.st_mode    = body->mode;
1182                         st.st_nlink   = body->nlink;
1183                         st.st_uid     = body->uid;
1184                         st.st_gid     = body->gid;
1185                         st.st_rdev    = body->rdev;
1186                         st.st_size    = body->size;
1187                         st.st_blksize = CFS_PAGE_SIZE;
1188                         st.st_blocks  = body->blocks;
1189                         st.st_atime   = body->atime;
1190                         st.st_mtime   = body->mtime;
1191                         st.st_ctime   = body->ctime;
1192                         st.st_ino     = inode->i_ino;
1193
1194                         lmdp = (struct lov_user_mds_data *)arg;
1195                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1196                                 GOTO(out_req, rc = -EFAULT);
1197                 }
1198
1199                 EXIT;
1200         out_req:
1201                 ptlrpc_req_finished(request);
1202                 if (filename)
1203                         putname(filename);
1204                 return rc;
1205         }
1206         case IOC_LOV_GETINFO: {
1207                 struct lov_user_mds_data *lumd;
1208                 struct lov_stripe_md *lsm;
1209                 struct lov_user_md *lum;
1210                 struct lov_mds_md *lmm;
1211                 int lmmsize;
1212                 lstat_t st;
1213
1214                 lumd = (struct lov_user_mds_data *)arg;
1215                 lum = &lumd->lmd_lmm;
1216
1217                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1218                 if (rc)
1219                         RETURN(rc);
1220
1221                 OBD_ALLOC_LARGE(lmm, lmmsize);
1222                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1223                         GOTO(free_lmm, rc = -EFAULT);
1224
1225                 switch (lmm->lmm_magic) {
1226                 case LOV_USER_MAGIC_V1:
1227                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1228                                 break;
1229                         /* swab objects first so that stripes num will be sane */
1230                         lustre_swab_lov_user_md_objects(
1231                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1232                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1233                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1234                         break;
1235                 case LOV_USER_MAGIC_V3:
1236                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1237                                 break;
1238                         /* swab objects first so that stripes num will be sane */
1239                         lustre_swab_lov_user_md_objects(
1240                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1241                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1242                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1243                         break;
1244                 default:
1245                         GOTO(free_lmm, rc = -EINVAL);
1246                 }
1247
1248                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1249                 if (rc < 0)
1250                         GOTO(free_lmm, rc = -ENOMEM);
1251
1252                 /* Perform glimpse_size operation. */
1253                 memset(&st, 0, sizeof(st));
1254
1255                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1256                 if (rc)
1257                         GOTO(free_lsm, rc);
1258
1259                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1260                         GOTO(free_lsm, rc = -EFAULT);
1261
1262                 EXIT;
1263         free_lsm:
1264                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1265         free_lmm:
1266                 OBD_FREE_LARGE(lmm, lmmsize);
1267                 return rc;
1268         }
1269         case OBD_IOC_LLOG_CATINFO: {
1270                 struct ptlrpc_request *req = NULL;
1271                 char                  *buf = NULL;
1272                 char                  *str;
1273                 int                    len = 0;
1274
1275                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1276                 if (rc)
1277                         RETURN(rc);
1278                 data = (void *)buf;
1279
1280                 if (!data->ioc_inlbuf1) {
1281                         obd_ioctl_freedata(buf, len);
1282                         RETURN(-EINVAL);
1283                 }
1284
1285                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1286                                            &RQF_LLOG_CATINFO);
1287                 if (req == NULL)
1288                         GOTO(out_catinfo, rc = -ENOMEM);
1289
1290                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1291                                      data->ioc_inllen1);
1292                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1293                                      data->ioc_inllen2);
1294
1295                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1296                 if (rc) {
1297                         ptlrpc_request_free(req);
1298                         GOTO(out_catinfo, rc);
1299                 }
1300
1301                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1302                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1303                 if (data->ioc_inllen2) {
1304                         str = req_capsule_client_get(&req->rq_pill,
1305                                                      &RMF_STRING);
1306                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1307                 }
1308
1309                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1310                                      data->ioc_plen1);
1311                 ptlrpc_request_set_replen(req);
1312
1313                 rc = ptlrpc_queue_wait(req);
1314                 if (!rc) {
1315                         str = req_capsule_server_get(&req->rq_pill,
1316                                                      &RMF_STRING);
1317                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1318                                              data->ioc_plen1))
1319                                 rc = -EFAULT;
1320                 }
1321                 ptlrpc_req_finished(req);
1322         out_catinfo:
1323                 obd_ioctl_freedata(buf, len);
1324                 RETURN(rc);
1325         }
1326         case OBD_IOC_QUOTACHECK: {
1327                 struct obd_quotactl *oqctl;
1328                 int error = 0;
1329
1330                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1331                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1332                         RETURN(-EPERM);
1333
1334                 OBD_ALLOC_PTR(oqctl);
1335                 if (!oqctl)
1336                         RETURN(-ENOMEM);
1337                 oqctl->qc_type = arg;
1338                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1339                 if (rc < 0) {
1340                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1341                         error = rc;
1342                 }
1343
1344                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1345                 if (rc < 0)
1346                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1347
1348                 OBD_FREE_PTR(oqctl);
1349                 return error ?: rc;
1350         }
1351         case OBD_IOC_POLL_QUOTACHECK: {
1352                 struct if_quotacheck *check;
1353
1354                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1355                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1356                         RETURN(-EPERM);
1357
1358                 OBD_ALLOC_PTR(check);
1359                 if (!check)
1360                         RETURN(-ENOMEM);
1361
1362                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1363                                    NULL);
1364                 if (rc) {
1365                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1366                         if (cfs_copy_to_user((void *)arg, check,
1367                                              sizeof(*check)))
1368                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1369                         GOTO(out_poll, rc);
1370                 }
1371
1372                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1373                                    NULL);
1374                 if (rc) {
1375                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1376                         if (cfs_copy_to_user((void *)arg, check,
1377                                              sizeof(*check)))
1378                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1379                         GOTO(out_poll, rc);
1380                 }
1381         out_poll:
1382                 OBD_FREE_PTR(check);
1383                 RETURN(rc);
1384         }
1385 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
1386         case LL_IOC_QUOTACTL_18: {
1387                 /* copy the old 1.x quota struct for internal use, then copy
1388                  * back into old format struct.  For 1.8 compatibility. */
1389                 struct if_quotactl_18 *qctl_18;
1390                 struct if_quotactl *qctl_20;
1391
1392                 OBD_ALLOC_PTR(qctl_18);
1393                 if (!qctl_18)
1394                         RETURN(-ENOMEM);
1395
1396                 OBD_ALLOC_PTR(qctl_20);
1397                 if (!qctl_20)
1398                         GOTO(out_quotactl_18, rc = -ENOMEM);
1399
1400                 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1401                         GOTO(out_quotactl_20, rc = -ENOMEM);
1402
1403                 QCTL_COPY(qctl_20, qctl_18);
1404                 qctl_20->qc_idx = 0;
1405
1406                 /* XXX: dqb_valid was borrowed as a flag to mark that
1407                  *      only mds quota is wanted */
1408                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1409                     qctl_18->qc_dqblk.dqb_valid) {
1410                         qctl_20->qc_valid = QC_MDTIDX;
1411                         qctl_20->qc_dqblk.dqb_valid = 0;
1412                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1413                         qctl_20->qc_valid = QC_UUID;
1414                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1415                 } else {
1416                         qctl_20->qc_valid = QC_GENERAL;
1417                 }
1418
1419                 rc = quotactl_ioctl(sbi, qctl_20);
1420
1421                 if (rc == 0) {
1422                         QCTL_COPY(qctl_18, qctl_20);
1423                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1424
1425                         if (cfs_copy_to_user((void *)arg, qctl_18,
1426                                              sizeof(*qctl_18)))
1427                                 rc = -EFAULT;
1428                 }
1429
1430         out_quotactl_20:
1431                 OBD_FREE_PTR(qctl_20);
1432         out_quotactl_18:
1433                 OBD_FREE_PTR(qctl_18);
1434                 RETURN(rc);
1435         }
1436 #else
1437 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1438 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
1439         case LL_IOC_QUOTACTL: {
1440                 struct if_quotactl *qctl;
1441
1442                 OBD_ALLOC_PTR(qctl);
1443                 if (!qctl)
1444                         RETURN(-ENOMEM);
1445
1446                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1447                         GOTO(out_quotactl, rc = -EFAULT);
1448
1449                 rc = quotactl_ioctl(sbi, qctl);
1450
1451                 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1452                         rc = -EFAULT;
1453
1454         out_quotactl:
1455                 OBD_FREE_PTR(qctl);
1456                 RETURN(rc);
1457         }
1458         case OBD_IOC_GETNAME: {
1459                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1460                 if (!obd)
1461                         RETURN(-EFAULT);
1462                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1463                                      strlen(obd->obd_name) + 1))
1464                         RETURN (-EFAULT);
1465                 RETURN(0);
1466         }
1467         case LL_IOC_FLUSHCTX:
1468                 RETURN(ll_flush_ctx(inode));
1469 #ifdef CONFIG_FS_POSIX_ACL
1470         case LL_IOC_RMTACL: {
1471             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1472                 inode == inode->i_sb->s_root->d_inode) {
1473                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1474
1475                 LASSERT(fd != NULL);
1476                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1477                 if (!rc)
1478                         fd->fd_flags |= LL_FILE_RMTACL;
1479                 RETURN(rc);
1480             } else
1481                 RETURN(0);
1482         }
1483 #endif
1484         case LL_IOC_GETOBDCOUNT: {
1485                 int count, vallen;
1486                 struct obd_export *exp;
1487
1488                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1489                         RETURN(-EFAULT);
1490
1491                 /* get ost count when count is zero, get mdt count otherwise */
1492                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1493                 vallen = sizeof(count);
1494                 rc = obd_get_info(exp, sizeof(KEY_TGT_COUNT), KEY_TGT_COUNT,
1495                                   &vallen, &count, NULL);
1496                 if (rc) {
1497                         CERROR("get target count failed: %d\n", rc);
1498                         RETURN(rc);
1499                 }
1500
1501                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1502                         RETURN(-EFAULT);
1503
1504                 RETURN(0);
1505         }
1506         case LL_IOC_PATH2FID:
1507                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1508                                      sizeof(struct lu_fid)))
1509                         RETURN(-EFAULT);
1510                 RETURN(0);
1511         case LL_IOC_GET_CONNECT_FLAGS: {
1512                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1513         }
1514         case OBD_IOC_CHANGELOG_SEND:
1515         case OBD_IOC_CHANGELOG_CLEAR:
1516                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1517                                     sizeof(struct ioc_changelog));
1518                 RETURN(rc);
1519         case OBD_IOC_FID2PATH:
1520                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1521         case LL_IOC_HSM_CT_START:
1522                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1523                                     sizeof(struct lustre_kernelcomm));
1524                 RETURN(rc);
1525
1526         default:
1527                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1528         }
1529 }
1530
1531 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1532 {
1533         struct inode *inode = file->f_mapping->host;
1534         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1535         struct ll_sb_info *sbi = ll_i2sbi(inode);
1536         int api32 = ll_need_32bit_api(sbi);
1537         loff_t ret = -EINVAL;
1538         ENTRY;
1539
1540         cfs_mutex_lock(&inode->i_mutex);
1541         switch (origin) {
1542                 case SEEK_SET:
1543                         break;
1544                 case SEEK_CUR:
1545                         offset += file->f_pos;
1546                         break;
1547                 case SEEK_END:
1548                         if (offset > 0)
1549                                 GOTO(out, ret);
1550                         if (api32)
1551                                 offset += LL_DIR_END_OFF_32BIT;
1552                         else
1553                                 offset += LL_DIR_END_OFF;
1554                         break;
1555                 default:
1556                         GOTO(out, ret);
1557         }
1558
1559         if (offset >= 0 &&
1560             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1561              (!api32 && offset <= LL_DIR_END_OFF))) {
1562                 if (offset != file->f_pos) {
1563                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1564                             (!api32 && offset == LL_DIR_END_OFF))
1565                                 fd->fd_dir.lfd_pos = MDS_DIR_END_OFF;
1566                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1567                                 fd->fd_dir.lfd_pos = offset << 32;
1568                         else
1569                                 fd->fd_dir.lfd_pos = offset;
1570                         file->f_pos = offset;
1571                         file->f_version = 0;
1572                 }
1573                 ret = offset;
1574         }
1575         GOTO(out, ret);
1576
1577 out:
1578         cfs_mutex_unlock(&inode->i_mutex);
1579         return ret;
1580 }
1581
1582 int ll_dir_open(struct inode *inode, struct file *file)
1583 {
1584         ENTRY;
1585         RETURN(ll_file_open(inode, file));
1586 }
1587
1588 int ll_dir_release(struct inode *inode, struct file *file)
1589 {
1590         ENTRY;
1591         RETURN(ll_file_release(inode, file));
1592 }
1593
1594 struct file_operations ll_dir_operations = {
1595         .llseek   = ll_dir_seek,
1596         .open     = ll_dir_open,
1597         .release  = ll_dir_release,
1598         .read     = generic_read_dir,
1599         .readdir  = ll_readdir,
1600         .ioctl    = ll_dir_ioctl,
1601         .fsync    = ll_fsync
1602 };