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                 hash = ll_i2info(inode)->lli_sa_pos;
 183         }
 184         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 185                inode->i_ino, inode->i_generation, inode, hash);
 186
 187         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
 188
 189         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 190         if (page_pool != NULL) {
 191                 page_pool[0] = page0;
 192         } else {
 193                 page_pool = &page0;
 194                 max_pages = 1;
 195         }
 196         for (npages = 1; npages < max_pages; npages++) {
 197                 page = page_cache_alloc_cold(inode->i_mapping);
 198                 if (!page)
 199                         break;
 200                 page_pool[npages] = page;
 201         }
 202
 203         oc = ll_mdscapa_get(inode);
 204         rc = md_readpage(exp, ll_inode2fid(inode), oc, hash, page_pool, npages,
 205                          &request);
 206         capa_put(oc);
 207         if (rc == 0) {
 208                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 209                 /* Checked by mdc_readpage() */
 210                 LASSERT(body != NULL);
 211
 212                 if (body->valid & OBD_MD_FLSIZE)
 213                         cl_isize_write(inode, body->size);
 214
 215                 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
 216                          >> CFS_PAGE_SHIFT;
 217                 SetPageUptodate(page0);
 218         }
 219         unlock_page(page0);
 220         ptlrpc_req_finished(request);
 221
 222         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 223
 224         ll_pagevec_init(&lru_pvec, 0);
 225         for (i = 1; i < npages; i++) {
 226                 unsigned long offset;
 227                 int ret;
 228
 229                 page = page_pool[i];
 230
 231                 if (rc < 0 || i >= nrdpgs) {
 232                         page_cache_release(page);
 233                         continue;
 234                 }
 235
 236                 SetPageUptodate(page);
 237
 238                 dp = cfs_kmap(page);
 239                 hash = le64_to_cpu(dp->ldp_hash_start);
 240                 cfs_kunmap(page);
 241
 242                 offset = hash_x_index(hash, hash64);
 243
 244                 prefetchw(&page->flags);
 245                 ret = ll_add_to_page_cache_lru(page, inode->i_mapping, offset,
 246                                                GFP_KERNEL);
 247                 if (ret == 0) {
 248                         unlock_page(page);
 249                         page_cache_get(page);
 250                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 251                                 ll_pagevec_lru_add_file(&lru_pvec);
 252                 } else {
 253                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 254                                " %d\n", offset, ret);
 255                 }
 256                 page_cache_release(page);
 257         }
 258         ll_pagevec_lru_add_file(&lru_pvec);
 259
 260         if (page_pool != &page0)
 261                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 262         EXIT;
 263         return rc;
 264 }
 265
 266 #ifndef MS_HAS_NEW_AOPS
 267 struct address_space_operations ll_dir_aops = {
 268         .readpage  = ll_dir_readpage,
 269 };
 270 #else
 271 struct address_space_operations_ext ll_dir_aops = {
 272         .orig_aops.readpage  = ll_dir_readpage,
 273 };
 274 #endif
 275
 276 static void ll_check_page(struct inode *dir, struct page *page)
 277 {
 278         /* XXX: check page format later */
 279         SetPageChecked(page);
 280 }
 281
 282 void ll_release_page(struct page *page, int remove)
 283 {
 284         kunmap(page);
 285         if (remove) {
 286                 lock_page(page);
 287                 if (likely(page->mapping != NULL))
 288                         truncate_complete_page(page->mapping, page);
 289                 unlock_page(page);
 290         }
 291         page_cache_release(page);
 292 }
 293
 294 /*
 295  * Find, kmap and return page that contains given hash.
 296  */
 297 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 298                                        __u64 *start, __u64 *end)
 299 {
 300         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 301         struct address_space *mapping = dir->i_mapping;
 302         /*
 303          * Complement of hash is used as an index so that
 304          * radix_tree_gang_lookup() can be used to find a page with starting
 305          * hash _smaller_ than one we are looking for.
 306          */
 307         unsigned long offset = hash_x_index(*hash, hash64);
 308         struct page *page;
 309         int found;
 310
 311         TREE_READ_LOCK_IRQ(mapping);
 312         found = radix_tree_gang_lookup(&mapping->page_tree,
 313                                        (void **)&page, offset, 1);
 314         if (found > 0) {
 315                 struct lu_dirpage *dp;
 316
 317                 page_cache_get(page);
 318                 TREE_READ_UNLOCK_IRQ(mapping);
 319                 /*
 320                  * In contrast to find_lock_page() we are sure that directory
 321                  * page cannot be truncated (while DLM lock is held) and,
 322                  * hence, can avoid restart.
 323                  *
 324                  * In fact, page cannot be locked here at all, because
 325                  * ll_dir_readpage() does synchronous io.
 326                  */
 327                 wait_on_page(page);
 328                 if (PageUptodate(page)) {
 329                         dp = cfs_kmap(page);
 330                         if (BITS_PER_LONG == 32 && hash64) {
 331                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 332                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 333                                 *hash  = *hash >> 32;
 334                         } else {
 335                                 *start = le64_to_cpu(dp->ldp_hash_start);
 336                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 337                         }
 338                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 339                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 340                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 341                                offset, *start, *end, *hash);
 342                         if (*hash > *end) {
 343                                 ll_release_page(page, 0);
 344                                 page = NULL;
 345                         } else if (*end != *start && *hash == *end) {
 346                                 /*
 347                                  * upon hash collision, remove this page,
 348                                  * otherwise put page reference, and
 349                                  * ll_get_dir_page() will issue RPC to fetch
 350                                  * the page we want.
 351                                  */
 352                                 ll_release_page(page,
 353                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 354                                 page = NULL;
 355                         }
 356                 } else {
 357                         page_cache_release(page);
 358                         page = ERR_PTR(-EIO);
 359                 }
 360
 361         } else {
 362                 TREE_READ_UNLOCK_IRQ(mapping);
 363                 page = NULL;
 364         }
 365         return page;
 366 }
 367
 368 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
 369                              struct ll_dir_chain *chain)
 370 {
 371         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 372         struct address_space *mapping = dir->i_mapping;
 373         struct lustre_handle lockh;
 374         struct lu_dirpage *dp;
 375         struct page *page;
 376         ldlm_mode_t mode;
 377         int rc;
 378         __u64 start = 0;
 379         __u64 end = 0;
 380         __u64 lhash = hash;
 381         struct ll_inode_info *lli = ll_i2info(dir);
 382         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 383
 384         mode = LCK_PR;
 385         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 386                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 387         if (!rc) {
 388                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 389                        ll_md_blocking_ast, ldlm_completion_ast,
 390                        NULL, NULL, dir };
 391                 struct lookup_intent it = { .it_op = IT_READDIR };
 392                 struct ptlrpc_request *request;
 393                 struct md_op_data *op_data;
 394
 395                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 396                                              LUSTRE_OPC_ANY, NULL);
 397                 if (IS_ERR(op_data))
 398                         return (void *)op_data;
 399
 400                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 401                                 op_data, &lockh, NULL, 0, NULL, 0);
 402
 403                 ll_finish_md_op_data(op_data);
 404
 405                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 406                 if (request)
 407                         ptlrpc_req_finished(request);
 408                 if (rc < 0) {
 409                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 410                                PFID(ll_inode2fid(dir)), hash, rc);
 411                         return ERR_PTR(rc);
 412                 }
 413         } else {
 414                 /* for cross-ref object, l_ast_data of the lock may not be set,
 415                  * we reset it here */
 416                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 417                                  dir, NULL);
 418         }
 419         ldlm_lock_dump_handle(D_OTHER, &lockh);
 420
 421         cfs_down(&lli->lli_readdir_sem);
 422         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 423         if (IS_ERR(page)) {
 424                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 425                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 426                 GOTO(out_unlock, page);
 427         } else if (page != NULL) {
 428                 /*
 429                  * XXX nikita: not entirely correct handling of a corner case:
 430                  * suppose hash chain of entries with hash value HASH crosses
 431                  * border between pages P0 and P1. First both P0 and P1 are
 432                  * cached, seekdir() is called for some entry from the P0 part
 433                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 434                  * happens and finds P1, as it starts with matching hash
 435                  * value. Remaining entries from P0 part of the chain are
 436                  * skipped. (Is that really a bug?)
 437                  *
 438                  * Possible solutions: 0. don't cache P1 is such case, handle
 439                  * it as an "overflow" page. 1. invalidate all pages at
 440                  * once. 2. use HASH|1 as an index for P1.
 441                  */
 442                 GOTO(hash_collision, page);
 443         }
 444
 445         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 446                                (filler_t*)mapping->a_ops->readpage, filp);
 447         if (IS_ERR(page)) {
 448                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 449                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 450                 GOTO(out_unlock, page);
 451         }
 452
 453         wait_on_page(page);
 454         (void)kmap(page);
 455         if (!PageUptodate(page)) {
 456                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 457                        PFID(ll_inode2fid(dir)), hash, -5);
 458                 goto fail;
 459         }
 460         if (!PageChecked(page))
 461                 ll_check_page(dir, page);
 462         if (PageError(page)) {
 463                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 464                        PFID(ll_inode2fid(dir)), hash, -5);
 465                 goto fail;
 466         }
 467 hash_collision:
 468         dp = page_address(page);
 469         if (BITS_PER_LONG == 32 && hash64) {
 470                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 471                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 472                 lhash = hash >> 32;
 473         } else {
 474                 start = le64_to_cpu(dp->ldp_hash_start);
 475                 end   = le64_to_cpu(dp->ldp_hash_end);
 476                 lhash = hash;
 477         }
 478         if (end == start) {
 479                 LASSERT(start == lhash);
 480                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 481                 if (BITS_PER_LONG == 32 && hash64)
 482                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 483                               "] with hash "LPU64"\n",
 484                               le64_to_cpu(dp->ldp_hash_start),
 485                               le64_to_cpu(dp->ldp_hash_end), hash);
 486                 /*
 487                  * Fetch whole overflow chain...
 488                  *
 489                  * XXX not yet.
 490                  */
 491                 goto fail;
 492         }
 493 out_unlock:
 494         cfs_up(&lli->lli_readdir_sem);
 495         ldlm_lock_decref(&lockh, mode);
 496         return page;
 497
 498 fail:
 499         ll_release_page(page, 1);
 500         page = ERR_PTR(-EIO);
 501         goto out_unlock;
 502 }
 503
 504 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 505 {
 506         struct inode         *inode      = filp->f_dentry->d_inode;
 507         struct ll_inode_info *info       = ll_i2info(inode);
 508         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 509         struct ll_file_data  *fd         = LUSTRE_FPRIVATE(filp);
 510         __u64                 pos        = fd->fd_dir.lfd_pos;
 511         int                   api32      = ll_need_32bit_api(sbi);
 512         int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 513         struct page          *page;
 514         struct ll_dir_chain   chain;
 515         int                   done;
 516         int                   rc;
 517         ENTRY;
 518
 519         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 520                inode->i_ino, inode->i_generation, inode,
 521                (unsigned long)pos, i_size_read(inode), api32);
 522
 523         if (pos == MDS_DIR_END_OFF)
 524                 /*
 525                  * end-of-file.
 526                  */
 527                 RETURN(0);
 528
 529         rc    = 0;
 530         done  = 0;
 531         ll_dir_chain_init(&chain);
 532
 533         fd->fd_dir.lfd_next = pos;
 534         page = ll_get_dir_page(filp, inode, pos, &chain);
 535
 536         while (rc == 0 && !done) {
 537                 struct lu_dirpage *dp;
 538                 struct lu_dirent  *ent;
 539
 540                 if (!IS_ERR(page)) {
 541                         /*
 542                          * If page is empty (end of directory is reached),
 543                          * use this value.
 544                          */
 545                         __u64 hash = MDS_DIR_END_OFF;
 546                         __u64 next;
 547
 548                         dp = page_address(page);
 549                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 550                              ent = lu_dirent_next(ent)) {
 551                                 __u16          type;
 552                                 int            namelen;
 553                                 struct lu_fid  fid;
 554                                 __u64          lhash;
 555                                 __u64          ino;
 556
 557                                 /*
 558                                  * XXX: implement correct swabbing here.
 559                                  */
 560
 561                                 hash = le64_to_cpu(ent->lde_hash);
 562                                 if (hash < pos)
 563                                         /*
 564                                          * Skip until we find target hash
 565                                          * value.
 566                                          */
 567                                         continue;
 568
 569                                 namelen = le16_to_cpu(ent->lde_namelen);
 570                                 if (namelen == 0)
 571                                         /*
 572                                          * Skip dummy record.
 573                                          */
 574                                         continue;
 575
 576                                 if (api32 && hash64)
 577                                         lhash = hash >> 32;
 578                                 else
 579                                         lhash = hash;
 580                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 581                                 ino = cl_fid_build_ino(&fid, api32);
 582                                 type = ll_dirent_type_get(ent);
 583                                 /* For 'll_nfs_get_name_filldir()', it will try
 584                                  * to access the 'ent' through its 'lde_name',
 585                                  * so the parameter 'name' for 'filldir()' must
 586                                  * be part of the 'ent'. */
 587                                 done = filldir(cookie, ent->lde_name, namelen,
 588                                                lhash, ino, type);
 589                         }
 590                         next = le64_to_cpu(dp->ldp_hash_end);
 591                         if (!done) {
 592                                 pos = next;
 593                                 if (pos == MDS_DIR_END_OFF) {
 594                                         /*
 595                                          * End of directory reached.
 596                                          */
 597                                         done = 1;
 598                                         ll_release_page(page, 0);
 599                                 } else if (1 /* chain is exhausted*/) {
 600                                         /*
 601                                          * Normal case: continue to the next
 602                                          * page.
 603                                          */
 604                                         ll_release_page(page,
 605                                             le32_to_cpu(dp->ldp_flags) &
 606                                                         LDF_COLLIDE);
 607                                         fd->fd_dir.lfd_next = pos;
 608                                         page = ll_get_dir_page(filp, inode, pos,
 609                                                                &chain);
 610                                 } else {
 611                                         /*
 612                                          * go into overflow page.
 613                                          */
 614                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 615                                                 LDF_COLLIDE);
 616                                         ll_release_page(page, 1);
 617                                 }
 618                         } else {
 619                                 pos = hash;
 620                                 ll_release_page(page, 0);
 621                         }
 622                 } else {
 623                         rc = PTR_ERR(page);
 624                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 625                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 626                 }
 627         }
 628
 629         fd->fd_dir.lfd_pos = pos;
 630         if (pos == MDS_DIR_END_OFF) {
 631                 if (api32)
 632                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 633                 else
 634                         filp->f_pos = LL_DIR_END_OFF;
 635         } else {
 636                 if (api32 && hash64)
 637                         filp->f_pos = pos >> 32;
 638                 else
 639                         filp->f_pos = pos;
 640         }
 641         filp->f_version = inode->i_version;
 642         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 643
 644         ll_dir_chain_fini(&chain);
 645
 646         RETURN(rc);
 647 }
 648
 649 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 650 {
 651         struct mgs_send_param *msp;
 652         int rc = 0;
 653
 654         OBD_ALLOC_PTR(msp);
 655         if (!msp)
 656                 return -ENOMEM;
 657
 658         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 659         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 660                                 sizeof(struct mgs_send_param), msp, NULL);
 661         if (rc)
 662                 CERROR("Failed to set parameter: %d\n", rc);
 663         OBD_FREE_PTR(msp);
 664
 665         return rc;
 666 }
 667
 668 char *ll_get_fsname(struct inode *inode)
 669 {
 670         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 671         char *ptr, *fsname;
 672         int len;
 673
 674         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 675         len = strlen(lsi->lsi_lmd->lmd_profile);
 676         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 677         if (ptr && (strcmp(ptr, "-client") == 0))
 678                 len -= 7;
 679         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 680         fsname[len] = '\0';
 681
 682         return fsname;
 683 }
 684
 685 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 686                      int set_default)
 687 {
 688         struct ll_sb_info *sbi = ll_i2sbi(inode);
 689         struct md_op_data *op_data;
 690         struct ptlrpc_request *req = NULL;
 691         int rc = 0;
 692         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 693         struct obd_device *mgc = lsi->lsi_mgc;
 694         char *fsname = NULL, *param = NULL;
 695         int lum_size;
 696
 697         if (lump != NULL) {
 698                 /*
 699                  * This is coming from userspace, so should be in
 700                  * local endian.  But the MDS would like it in little
 701                  * endian, so we swab it before we send it.
 702                  */
 703                 switch (lump->lmm_magic) {
 704                 case LOV_USER_MAGIC_V1: {
 705                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 706                                 lustre_swab_lov_user_md_v1(lump);
 707                         lum_size = sizeof(struct lov_user_md_v1);
 708                         break;
 709                         }
 710                 case LOV_USER_MAGIC_V3: {
 711                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 712                                 lustre_swab_lov_user_md_v3(
 713                                         (struct lov_user_md_v3 *)lump);
 714                         lum_size = sizeof(struct lov_user_md_v3);
 715                         break;
 716                         }
 717                 default: {
 718                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 719                                         " %#08x != %#08x nor %#08x\n",
 720                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 721                                         LOV_USER_MAGIC_V3);
 722                         RETURN(-EINVAL);
 723                         }
 724                }
 725         } else {
 726                 lum_size = sizeof(struct lov_user_md_v1);
 727         }
 728
 729         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 730                                      LUSTRE_OPC_ANY, NULL);
 731         if (IS_ERR(op_data))
 732                 RETURN(PTR_ERR(op_data));
 733
 734         /* swabbing is done in lov_setstripe() on server side */
 735         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 736                         NULL, 0, &req, NULL);
 737         ll_finish_md_op_data(op_data);
 738         ptlrpc_req_finished(req);
 739         if (rc) {
 740                 if (rc != -EPERM && rc != -EACCES)
 741                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 742         }
 743
 744         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 745          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 746          need the make the distiction between the 2 versions */
 747         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 748                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 749
 750                 /* Get fsname and assume devname to be -MDT0000. */
 751                 fsname = ll_get_fsname(inode);
 752                 /* Set root stripesize */
 753                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 754                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 755                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 756                 if (rc)
 757                         goto end;
 758
 759                 /* Set root stripecount */
 760                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 761                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 762                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 763                 if (rc)
 764                         goto end;
 765
 766                 /* Set root stripeoffset */
 767                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 768                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 769                         (typeof(lump->lmm_stripe_offset))(-1));
 770                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 771                 if (rc)
 772                         goto end;
 773 end:
 774                 if (fsname)
 775                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 776                 if (param)
 777                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 778         }
 779         return rc;
 780 }
 781
 782 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 783                      int *lmm_size, struct ptlrpc_request **request)
 784 {
 785         struct ll_sb_info *sbi = ll_i2sbi(inode);
 786         struct mdt_body   *body;
 787         struct lov_mds_md *lmm = NULL;
 788         struct ptlrpc_request *req = NULL;
 789         int rc, lmmsize;
 790         struct md_op_data *op_data;
 791
 792         rc = ll_get_max_mdsize(sbi, &lmmsize);
 793         if (rc)
 794                 RETURN(rc);
 795
 796         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 797                                      0, lmmsize, LUSTRE_OPC_ANY,
 798                                      NULL);
 799         if (IS_ERR(op_data))
 800                 RETURN(PTR_ERR(op_data));
 801
 802         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 803         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 804         ll_finish_md_op_data(op_data);
 805         if (rc < 0) {
 806                 CDEBUG(D_INFO, "md_getattr failed on inode "
 807                        "%lu/%u: rc %d\n", inode->i_ino,
 808                        inode->i_generation, rc);
 809                 GOTO(out, rc);
 810         }
 811
 812         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 813         LASSERT(body != NULL);
 814
 815         lmmsize = body->eadatasize;
 816
 817         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 818             lmmsize == 0) {
 819                 GOTO(out, rc = -ENODATA);
 820         }
 821
 822         lmm = req_capsule_server_sized_get(&req->rq_pill,
 823                                            &RMF_MDT_MD, lmmsize);
 824         LASSERT(lmm != NULL);
 825
 826         /*
 827          * This is coming from the MDS, so is probably in
 828          * little endian.  We convert it to host endian before
 829          * passing it to userspace.
 830          */
 831         /* We don't swab objects for directories */
 832         switch (le32_to_cpu(lmm->lmm_magic)) {
 833         case LOV_MAGIC_V1:
 834                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 835                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 836                 break;
 837         case LOV_MAGIC_V3:
 838                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 839                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 840                 break;
 841         default:
 842                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 843                 rc = -EPROTO;
 844         }
 845 out:
 846         *lmmp = lmm;
 847         *lmm_size = lmmsize;
 848         *request = req;
 849         return rc;
 850 }
 851
 852 /*
 853  *  Get MDT index for the inode.
 854  */
 855 int ll_get_mdt_idx(struct inode *inode)
 856 {
 857         struct ll_sb_info *sbi = ll_i2sbi(inode);
 858         struct md_op_data *op_data;
 859         int rc, mdtidx;
 860         ENTRY;
 861
 862         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 863                                      0, LUSTRE_OPC_ANY, NULL);
 864         if (IS_ERR(op_data))
 865                 RETURN(PTR_ERR(op_data));
 866
 867         op_data->op_valid |= OBD_MD_MDTIDX;
 868         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 869         mdtidx = op_data->op_mds;
 870         ll_finish_md_op_data(op_data);
 871         if (rc < 0) {
 872                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 873                 RETURN(rc);
 874         }
 875         return mdtidx;
 876 }
 877
 878 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 879 {
 880         void *ptr;
 881         int rc;
 882
 883         OBD_ALLOC(ptr, len);
 884         if (ptr == NULL)
 885                 return -ENOMEM;
 886         if (cfs_copy_from_user(ptr, data, len)) {
 887                 OBD_FREE(ptr, len);
 888                 return -EFAULT;
 889         }
 890         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 891         OBD_FREE(ptr, len);
 892         return rc;
 893 }
 894
 895 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 896 {
 897         int cmd = qctl->qc_cmd;
 898         int type = qctl->qc_type;
 899         int id = qctl->qc_id;
 900         int valid = qctl->qc_valid;
 901         int rc = 0;
 902         ENTRY;
 903
 904         switch (cmd) {
 905         case LUSTRE_Q_INVALIDATE:
 906         case LUSTRE_Q_FINVALIDATE:
 907         case Q_QUOTAON:
 908         case Q_QUOTAOFF:
 909         case Q_SETQUOTA:
 910         case Q_SETINFO:
 911                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 912                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 913                         RETURN(-EPERM);
 914                 break;
 915         case Q_GETQUOTA:
 916                 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
 917                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 918                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 919                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 920                         RETURN(-EPERM);
 921                 break;
 922         case Q_GETINFO:
 923                 break;
 924         default:
 925                 CERROR("unsupported quotactl op: %#x\n", cmd);
 926                 RETURN(-ENOTTY);
 927         }
 928
 929         if (valid != QC_GENERAL) {
 930                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 931                         RETURN(-EOPNOTSUPP);
 932
 933                 if (cmd == Q_GETINFO)
 934                         qctl->qc_cmd = Q_GETOINFO;
 935                 else if (cmd == Q_GETQUOTA)
 936                         qctl->qc_cmd = Q_GETOQUOTA;
 937                 else
 938                         RETURN(-EINVAL);
 939
 940                 switch (valid) {
 941                 case QC_MDTIDX:
 942                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 943                                            sizeof(*qctl), qctl, NULL);
 944                         break;
 945                 case QC_OSTIDX:
 946                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 947                                            sizeof(*qctl), qctl, NULL);
 948                         break;
 949                 case QC_UUID:
 950                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 951                                            sizeof(*qctl), qctl, NULL);
 952                         if (rc == -EAGAIN)
 953                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 954                                                    sbi->ll_dt_exp,
 955                                                    sizeof(*qctl), qctl, NULL);
 956                         break;
 957                 default:
 958                         rc = -EINVAL;
 959                         break;
 960                 }
 961
 962                 if (rc)
 963                         RETURN(rc);
 964
 965                 qctl->qc_cmd = cmd;
 966         } else {
 967                 struct obd_quotactl *oqctl;
 968
 969                 OBD_ALLOC_PTR(oqctl);
 970                 if (oqctl == NULL)
 971                         RETURN(-ENOMEM);
 972
 973                 QCTL_COPY(oqctl, qctl);
 974                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 975                 if (rc) {
 976                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 977                                 oqctl->qc_cmd = Q_QUOTAOFF;
 978                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 979                         }
 980                         OBD_FREE_PTR(oqctl);
 981                         RETURN(rc);
 982                 }
 983                 /* If QIF_SPACE is not set, client should collect the
 984                  * space usage from OSSs by itself */
 985                 if (cmd == Q_GETQUOTA &&
 986                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 987                     !oqctl->qc_dqblk.dqb_curspace) {
 988                         struct obd_quotactl *oqctl_tmp;
 989
 990                         OBD_ALLOC_PTR(oqctl_tmp);
 991                         if (oqctl_tmp == NULL)
 992                                 GOTO(out, rc = -ENOMEM);
 993
 994                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 995                         oqctl_tmp->qc_id = oqctl->qc_id;
 996                         oqctl_tmp->qc_type = oqctl->qc_type;
 997
 998                         /* collect space usage from OSTs */
 999                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1000                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1001                         if (!rc || rc == -EREMOTEIO) {
1002                                 oqctl->qc_dqblk.dqb_curspace =
1003                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1004                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1005                         }
1006
1007                         /* collect space & inode usage from MDTs */
1008                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1009                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1010                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1011                         if (!rc || rc == -EREMOTEIO) {
1012                                 oqctl->qc_dqblk.dqb_curspace +=
1013                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1014                                 oqctl->qc_dqblk.dqb_curinodes =
1015                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1016                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1017                         } else {
1018                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1019                         }
1020
1021                         OBD_FREE_PTR(oqctl_tmp);
1022                 }
1023 out:
1024                 QCTL_COPY(qctl, oqctl);
1025                 OBD_FREE_PTR(oqctl);
1026         }
1027
1028         RETURN(rc);
1029 }
1030
1031 #ifdef HAVE_UNLOCKED_IOCTL
1032 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1033 #else
1034 static int  ll_dir_ioctl(struct inode *unuse, struct file *file,
1035                         unsigned int cmd, unsigned long arg)
1036 #endif
1037 {
1038         struct inode *inode = file->f_dentry->d_inode;
1039         struct ll_sb_info *sbi = ll_i2sbi(inode);
1040         struct obd_ioctl_data *data;
1041         int rc = 0;
1042         ENTRY;
1043
1044         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1045                inode->i_ino, inode->i_generation, inode, cmd);
1046
1047         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1048         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1049                 return -ENOTTY;
1050
1051         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1052         switch(cmd) {
1053         case FSFILT_IOC_GETFLAGS:
1054         case FSFILT_IOC_SETFLAGS:
1055                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1056         case FSFILT_IOC_GETVERSION_OLD:
1057         case FSFILT_IOC_GETVERSION:
1058                 RETURN(put_user(inode->i_generation, (int *)arg));
1059         /* We need to special case any other ioctls we want to handle,
1060          * to send them to the MDS/OST as appropriate and to properly
1061          * network encode the arg field.
1062         case FSFILT_IOC_SETVERSION_OLD:
1063         case FSFILT_IOC_SETVERSION:
1064         */
1065         case LL_IOC_GET_MDTIDX: {
1066                 int mdtidx;
1067
1068                 mdtidx = ll_get_mdt_idx(inode);
1069                 if (mdtidx < 0)
1070                         RETURN(mdtidx);
1071
1072                 if (put_user((int)mdtidx, (int*)arg))
1073                         RETURN(-EFAULT);
1074
1075                 return 0;
1076         }
1077         case IOC_MDC_LOOKUP: {
1078                 struct ptlrpc_request *request = NULL;
1079                 int namelen, len = 0;
1080                 char *buf = NULL;
1081                 char *filename;
1082                 struct md_op_data *op_data;
1083
1084                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1085                 if (rc)
1086                         RETURN(rc);
1087                 data = (void *)buf;
1088
1089                 filename = data->ioc_inlbuf1;
1090                 namelen = strlen(filename);
1091
1092                 if (namelen < 1) {
1093                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1094                         GOTO(out_free, rc = -EINVAL);
1095                 }
1096
1097                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1098                                              0, LUSTRE_OPC_ANY, NULL);
1099                 if (IS_ERR(op_data))
1100                         GOTO(out_free, rc = PTR_ERR(op_data));
1101
1102                 op_data->op_valid = OBD_MD_FLID;
1103                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1104                 ll_finish_md_op_data(op_data);
1105                 if (rc < 0) {
1106                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1107                         GOTO(out_free, rc);
1108                 }
1109                 ptlrpc_req_finished(request);
1110                 EXIT;
1111 out_free:
1112                 obd_ioctl_freedata(buf, len);
1113                 return rc;
1114         }
1115         case LL_IOC_LOV_SETSTRIPE: {
1116                 struct lov_user_md_v3 lumv3;
1117                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1118                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1119                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1120
1121                 int set_default = 0;
1122
1123                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1124                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1125                         sizeof(lumv3p->lmm_objects[0]));
1126                 /* first try with v1 which is smaller than v3 */
1127                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1128                         RETURN(-EFAULT);
1129
1130                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1131                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1132                                 RETURN(-EFAULT);
1133                 }
1134
1135                 if (inode->i_sb->s_root == file->f_dentry)
1136                         set_default = 1;
1137
1138                 /* in v1 and v3 cases lumv1 points to data */
1139                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1140
1141                 RETURN(rc);
1142         }
1143         case LL_IOC_OBD_STATFS:
1144                 RETURN(ll_obd_statfs(inode, (void *)arg));
1145         case LL_IOC_LOV_GETSTRIPE:
1146         case LL_IOC_MDC_GETINFO:
1147         case IOC_MDC_GETFILEINFO:
1148         case IOC_MDC_GETFILESTRIPE: {
1149                 struct ptlrpc_request *request = NULL;
1150                 struct lov_user_md *lump;
1151                 struct lov_mds_md *lmm = NULL;
1152                 struct mdt_body *body;
1153                 char *filename = NULL;
1154                 int lmmsize;
1155
1156                 if (cmd == IOC_MDC_GETFILEINFO ||
1157                     cmd == IOC_MDC_GETFILESTRIPE) {
1158                         filename = getname((const char *)arg);
1159                         if (IS_ERR(filename))
1160                                 RETURN(PTR_ERR(filename));
1161
1162                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1163                                                       &lmmsize, &request);
1164                 } else {
1165                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1166                 }
1167
1168                 if (request) {
1169                         body = req_capsule_server_get(&request->rq_pill,
1170                                                       &RMF_MDT_BODY);
1171                         LASSERT(body != NULL);
1172                 } else {
1173                         GOTO(out_req, rc);
1174                 }
1175
1176                 if (rc < 0) {
1177                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1178                                                cmd == LL_IOC_MDC_GETINFO))
1179                                 GOTO(skip_lmm, rc = 0);
1180                         else
1181                                 GOTO(out_req, rc);
1182                 }
1183
1184                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1185                     cmd == LL_IOC_LOV_GETSTRIPE) {
1186                         lump = (struct lov_user_md *)arg;
1187                 } else {
1188                         struct lov_user_mds_data *lmdp;
1189                         lmdp = (struct lov_user_mds_data *)arg;
1190                         lump = &lmdp->lmd_lmm;
1191                 }
1192                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1193                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1194                                 GOTO(out_req, rc = -EFAULT);
1195                         rc = -EOVERFLOW;
1196                 }
1197         skip_lmm:
1198                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1199                         struct lov_user_mds_data *lmdp;
1200                         lstat_t st = { 0 };
1201
1202                         st.st_dev     = inode->i_sb->s_dev;
1203                         st.st_mode    = body->mode;
1204                         st.st_nlink   = body->nlink;
1205                         st.st_uid     = body->uid;
1206                         st.st_gid     = body->gid;
1207                         st.st_rdev    = body->rdev;
1208                         st.st_size    = body->size;
1209                         st.st_blksize = CFS_PAGE_SIZE;
1210                         st.st_blocks  = body->blocks;
1211                         st.st_atime   = body->atime;
1212                         st.st_mtime   = body->mtime;
1213                         st.st_ctime   = body->ctime;
1214                         st.st_ino     = inode->i_ino;
1215
1216                         lmdp = (struct lov_user_mds_data *)arg;
1217                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1218                                 GOTO(out_req, rc = -EFAULT);
1219                 }
1220
1221                 EXIT;
1222         out_req:
1223                 ptlrpc_req_finished(request);
1224                 if (filename)
1225                         putname(filename);
1226                 return rc;
1227         }
1228         case IOC_LOV_GETINFO: {
1229                 struct lov_user_mds_data *lumd;
1230                 struct lov_stripe_md *lsm;
1231                 struct lov_user_md *lum;
1232                 struct lov_mds_md *lmm;
1233                 int lmmsize;
1234                 lstat_t st;
1235
1236                 lumd = (struct lov_user_mds_data *)arg;
1237                 lum = &lumd->lmd_lmm;
1238
1239                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1240                 if (rc)
1241                         RETURN(rc);
1242
1243                 OBD_ALLOC_LARGE(lmm, lmmsize);
1244                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1245                         GOTO(free_lmm, rc = -EFAULT);
1246
1247                 switch (lmm->lmm_magic) {
1248                 case LOV_USER_MAGIC_V1:
1249                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1250                                 break;
1251                         /* swab objects first so that stripes num will be sane */
1252                         lustre_swab_lov_user_md_objects(
1253                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1254                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1255                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1256                         break;
1257                 case LOV_USER_MAGIC_V3:
1258                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1259                                 break;
1260                         /* swab objects first so that stripes num will be sane */
1261                         lustre_swab_lov_user_md_objects(
1262                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1263                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1264                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1265                         break;
1266                 default:
1267                         GOTO(free_lmm, rc = -EINVAL);
1268                 }
1269
1270                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1271                 if (rc < 0)
1272                         GOTO(free_lmm, rc = -ENOMEM);
1273
1274                 /* Perform glimpse_size operation. */
1275                 memset(&st, 0, sizeof(st));
1276
1277                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1278                 if (rc)
1279                         GOTO(free_lsm, rc);
1280
1281                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1282                         GOTO(free_lsm, rc = -EFAULT);
1283
1284                 EXIT;
1285         free_lsm:
1286                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1287         free_lmm:
1288                 OBD_FREE_LARGE(lmm, lmmsize);
1289                 return rc;
1290         }
1291         case OBD_IOC_LLOG_CATINFO: {
1292                 struct ptlrpc_request *req = NULL;
1293                 char                  *buf = NULL;
1294                 char                  *str;
1295                 int                    len = 0;
1296
1297                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1298                 if (rc)
1299                         RETURN(rc);
1300                 data = (void *)buf;
1301
1302                 if (!data->ioc_inlbuf1) {
1303                         obd_ioctl_freedata(buf, len);
1304                         RETURN(-EINVAL);
1305                 }
1306
1307                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1308                                            &RQF_LLOG_CATINFO);
1309                 if (req == NULL)
1310                         GOTO(out_catinfo, rc = -ENOMEM);
1311
1312                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1313                                      data->ioc_inllen1);
1314                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1315                                      data->ioc_inllen2);
1316
1317                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1318                 if (rc) {
1319                         ptlrpc_request_free(req);
1320                         GOTO(out_catinfo, rc);
1321                 }
1322
1323                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1324                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1325                 if (data->ioc_inllen2) {
1326                         str = req_capsule_client_get(&req->rq_pill,
1327                                                      &RMF_STRING);
1328                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1329                 }
1330
1331                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1332                                      data->ioc_plen1);
1333                 ptlrpc_request_set_replen(req);
1334
1335                 rc = ptlrpc_queue_wait(req);
1336                 if (!rc) {
1337                         str = req_capsule_server_get(&req->rq_pill,
1338                                                      &RMF_STRING);
1339                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1340                                              data->ioc_plen1))
1341                                 rc = -EFAULT;
1342                 }
1343                 ptlrpc_req_finished(req);
1344         out_catinfo:
1345                 obd_ioctl_freedata(buf, len);
1346                 RETURN(rc);
1347         }
1348         case OBD_IOC_QUOTACHECK: {
1349                 struct obd_quotactl *oqctl;
1350                 int error = 0;
1351
1352                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1353                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1354                         RETURN(-EPERM);
1355
1356                 OBD_ALLOC_PTR(oqctl);
1357                 if (!oqctl)
1358                         RETURN(-ENOMEM);
1359                 oqctl->qc_type = arg;
1360                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1361                 if (rc < 0) {
1362                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1363                         error = rc;
1364                 }
1365
1366                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1367                 if (rc < 0)
1368                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1369
1370                 OBD_FREE_PTR(oqctl);
1371                 return error ?: rc;
1372         }
1373         case OBD_IOC_POLL_QUOTACHECK: {
1374                 struct if_quotacheck *check;
1375
1376                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1377                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1378                         RETURN(-EPERM);
1379
1380                 OBD_ALLOC_PTR(check);
1381                 if (!check)
1382                         RETURN(-ENOMEM);
1383
1384                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1385                                    NULL);
1386                 if (rc) {
1387                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1388                         if (cfs_copy_to_user((void *)arg, check,
1389                                              sizeof(*check)))
1390                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1391                         GOTO(out_poll, rc);
1392                 }
1393
1394                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1395                                    NULL);
1396                 if (rc) {
1397                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1398                         if (cfs_copy_to_user((void *)arg, check,
1399                                              sizeof(*check)))
1400                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1401                         GOTO(out_poll, rc);
1402                 }
1403         out_poll:
1404                 OBD_FREE_PTR(check);
1405                 RETURN(rc);
1406         }
1407 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
1408         case LL_IOC_QUOTACTL_18: {
1409                 /* copy the old 1.x quota struct for internal use, then copy
1410                  * back into old format struct.  For 1.8 compatibility. */
1411                 struct if_quotactl_18 *qctl_18;
1412                 struct if_quotactl *qctl_20;
1413
1414                 OBD_ALLOC_PTR(qctl_18);
1415                 if (!qctl_18)
1416                         RETURN(-ENOMEM);
1417
1418                 OBD_ALLOC_PTR(qctl_20);
1419                 if (!qctl_20)
1420                         GOTO(out_quotactl_18, rc = -ENOMEM);
1421
1422                 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1423                         GOTO(out_quotactl_20, rc = -ENOMEM);
1424
1425                 QCTL_COPY(qctl_20, qctl_18);
1426                 qctl_20->qc_idx = 0;
1427
1428                 /* XXX: dqb_valid was borrowed as a flag to mark that
1429                  *      only mds quota is wanted */
1430                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1431                     qctl_18->qc_dqblk.dqb_valid) {
1432                         qctl_20->qc_valid = QC_MDTIDX;
1433                         qctl_20->qc_dqblk.dqb_valid = 0;
1434                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1435                         qctl_20->qc_valid = QC_UUID;
1436                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1437                 } else {
1438                         qctl_20->qc_valid = QC_GENERAL;
1439                 }
1440
1441                 rc = quotactl_ioctl(sbi, qctl_20);
1442
1443                 if (rc == 0) {
1444                         QCTL_COPY(qctl_18, qctl_20);
1445                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1446
1447                         if (cfs_copy_to_user((void *)arg, qctl_18,
1448                                              sizeof(*qctl_18)))
1449                                 rc = -EFAULT;
1450                 }
1451
1452         out_quotactl_20:
1453                 OBD_FREE_PTR(qctl_20);
1454         out_quotactl_18:
1455                 OBD_FREE_PTR(qctl_18);
1456                 RETURN(rc);
1457         }
1458 #else
1459 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1460 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
1461         case LL_IOC_QUOTACTL: {
1462                 struct if_quotactl *qctl;
1463
1464                 OBD_ALLOC_PTR(qctl);
1465                 if (!qctl)
1466                         RETURN(-ENOMEM);
1467
1468                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1469                         GOTO(out_quotactl, rc = -EFAULT);
1470
1471                 rc = quotactl_ioctl(sbi, qctl);
1472
1473                 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1474                         rc = -EFAULT;
1475
1476         out_quotactl:
1477                 OBD_FREE_PTR(qctl);
1478                 RETURN(rc);
1479         }
1480         case OBD_IOC_GETDTNAME:
1481         case OBD_IOC_GETMDNAME:
1482                 RETURN(ll_get_obd_name(inode, cmd, arg));
1483         case LL_IOC_FLUSHCTX:
1484                 RETURN(ll_flush_ctx(inode));
1485 #ifdef CONFIG_FS_POSIX_ACL
1486         case LL_IOC_RMTACL: {
1487             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1488                 inode == inode->i_sb->s_root->d_inode) {
1489                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1490
1491                 LASSERT(fd != NULL);
1492                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1493                 if (!rc)
1494                         fd->fd_flags |= LL_FILE_RMTACL;
1495                 RETURN(rc);
1496             } else
1497                 RETURN(0);
1498         }
1499 #endif
1500         case LL_IOC_GETOBDCOUNT: {
1501                 int count, vallen;
1502                 struct obd_export *exp;
1503
1504                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1505                         RETURN(-EFAULT);
1506
1507                 /* get ost count when count is zero, get mdt count otherwise */
1508                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1509                 vallen = sizeof(count);
1510                 rc = obd_get_info(exp, sizeof(KEY_TGT_COUNT), KEY_TGT_COUNT,
1511                                   &vallen, &count, NULL);
1512                 if (rc) {
1513                         CERROR("get target count failed: %d\n", rc);
1514                         RETURN(rc);
1515                 }
1516
1517                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1518                         RETURN(-EFAULT);
1519
1520                 RETURN(0);
1521         }
1522         case LL_IOC_PATH2FID:
1523                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1524                                      sizeof(struct lu_fid)))
1525                         RETURN(-EFAULT);
1526                 RETURN(0);
1527         case LL_IOC_GET_CONNECT_FLAGS: {
1528                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1529         }
1530         case OBD_IOC_CHANGELOG_SEND:
1531         case OBD_IOC_CHANGELOG_CLEAR:
1532                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1533                                     sizeof(struct ioc_changelog));
1534                 RETURN(rc);
1535         case OBD_IOC_FID2PATH:
1536                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1537         case LL_IOC_HSM_CT_START:
1538                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1539                                     sizeof(struct lustre_kernelcomm));
1540                 RETURN(rc);
1541
1542         default:
1543                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1544         }
1545 }
1546
1547 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1548 {
1549         struct inode *inode = file->f_mapping->host;
1550         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1551         struct ll_sb_info *sbi = ll_i2sbi(inode);
1552         int api32 = ll_need_32bit_api(sbi);
1553         loff_t ret = -EINVAL;
1554         ENTRY;
1555
1556         cfs_mutex_lock(&inode->i_mutex);
1557         switch (origin) {
1558                 case SEEK_SET:
1559                         break;
1560                 case SEEK_CUR:
1561                         offset += file->f_pos;
1562                         break;
1563                 case SEEK_END:
1564                         if (offset > 0)
1565                                 GOTO(out, ret);
1566                         if (api32)
1567                                 offset += LL_DIR_END_OFF_32BIT;
1568                         else
1569                                 offset += LL_DIR_END_OFF;
1570                         break;
1571                 default:
1572                         GOTO(out, ret);
1573         }
1574
1575         if (offset >= 0 &&
1576             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1577              (!api32 && offset <= LL_DIR_END_OFF))) {
1578                 if (offset != file->f_pos) {
1579                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1580                             (!api32 && offset == LL_DIR_END_OFF))
1581                                 fd->fd_dir.lfd_pos = MDS_DIR_END_OFF;
1582                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1583                                 fd->fd_dir.lfd_pos = offset << 32;
1584                         else
1585                                 fd->fd_dir.lfd_pos = offset;
1586                         file->f_pos = offset;
1587                         file->f_version = 0;
1588                 }
1589                 ret = offset;
1590         }
1591         GOTO(out, ret);
1592
1593 out:
1594         cfs_mutex_unlock(&inode->i_mutex);
1595         return ret;
1596 }
1597
1598 int ll_dir_open(struct inode *inode, struct file *file)
1599 {
1600         ENTRY;
1601         RETURN(ll_file_open(inode, file));
1602 }
1603
1604 int ll_dir_release(struct inode *inode, struct file *file)
1605 {
1606         ENTRY;
1607         RETURN(ll_file_release(inode, file));
1608 }
1609
1610 struct file_operations ll_dir_operations = {
1611         .llseek   = ll_dir_seek,
1612         .open     = ll_dir_open,
1613         .release  = ll_dir_release,
1614         .read     = generic_read_dir,
1615         .readdir  = ll_readdir,
1616 #ifdef HAVE_UNLOCKED_IOCTL
1617         .unlocked_ioctl   = ll_dir_ioctl,
1618 #else
1619         .ioctl          = ll_dir_ioctl,
1620 #endif
1621         .fsync    = ll_fsync,
1622 };