drivers/staging/lustre/lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <asm/uaccess.h>
  45 #include <linux/buffer_head.h>   // for wait_on_buffer
  46 #include <linux/pagevec.h>
  47 #include <linux/prefetch.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include "../include/obd_support.h"
  52 #include "../include/obd_class.h"
  53 #include "../include/lustre_lib.h"
  54 #include "../include/lustre/lustre_idl.h"
  55 #include "../include/lustre_lite.h"
  56 #include "../include/lustre_dlm.h"
  57 #include "../include/lustre_fid.h"
  58 #include "llite_internal.h"
  59
  60 /*
  61  * (new) readdir implementation overview.
  62  *
  63  * Original lustre readdir implementation cached exact copy of raw directory
  64  * pages on the client. These pages were indexed in client page cache by
  65  * logical offset in the directory file. This design, while very simple and
  66  * intuitive had some inherent problems:
  67  *
  68  *     . it implies that byte offset to the directory entry serves as a
  69  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  70  *     ext3/htree directory entries may move due to splits, and more
  71  *     importantly,
  72  *
  73  *     . it is incompatible with the design of split directories for cmd3,
  74  *     that assumes that names are distributed across nodes based on their
  75  *     hash, and so readdir should be done in hash order.
  76  *
  77  * New readdir implementation does readdir in hash order, and uses hash of a
  78  * file name as a telldir/seekdir cookie. This led to number of complications:
  79  *
  80  *     . hash is not unique, so it cannot be used to index cached directory
  81  *     pages on the client (note, that it requires a whole pageful of hash
  82  *     collided entries to cause two pages to have identical hashes);
  83  *
  84  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  85  *     entry cookie. ext3/htree has the same problem and lustre implementation
  86  *     mimics their solution: seekdir(hash) positions directory at the first
  87  *     entry with the given hash.
  88  *
  89  * Client side.
  90  *
  91  * 0. caching
  92  *
  93  * Client caches directory pages using hash of the first entry as an index. As
  94  * noted above hash is not unique, so this solution doesn't work as is:
  95  * special processing is needed for "page hash chains" (i.e., sequences of
  96  * pages filled with entries all having the same hash value).
  97  *
  98  * First, such chains have to be detected. To this end, server returns to the
  99  * client the hash of the first entry on the page next to one returned. When
 100  * client detects that this hash is the same as hash of the first entry on the
 101  * returned page, page hash collision has to be handled. Pages in the
 102  * hash chain, except first one, are termed "overflow pages".
 103  *
 104  * Solution to index uniqueness problem is to not cache overflow
 105  * pages. Instead, when page hash collision is detected, all overflow pages
 106  * from emerging chain are immediately requested from the server and placed in
 107  * a special data structure (struct ll_dir_chain). This data structure is used
 108  * by ll_readdir() to process entries from overflow pages. When readdir
 109  * invocation finishes, overflow pages are discarded. If page hash collision
 110  * chain weren't completely processed, next call to readdir will again detect
 111  * page hash collision, again read overflow pages in, process next portion of
 112  * entries and again discard the pages. This is not as wasteful as it looks,
 113  * because, given reasonable hash, page hash collisions are extremely rare.
 114  *
 115  * 1. directory positioning
 116  *
 117  * When seekdir(hash) is called, original
 118  *
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  * Server.
 127  *
 128  * identification of and access to overflow pages
 129  *
 130  * page format
 131  *
 132  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 133  * a header lu_dirpage which describes the start/end hash, and whether this
 134  * page is empty (contains no dir entry) or hash collide with next page.
 135  * After client receives reply, several pages will be integrated into dir page
 136  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 137  * lu_dirpage for this integrated page will be adjusted. See
 138  * lmv_adjust_dirpages().
 139  *
 140  */
 141
 142 /* returns the page unlocked, but with a reference */
 143 static int ll_dir_filler(void *_hash, struct page *page0)
 144 {
 145         struct inode *inode = page0->mapping->host;
 146         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 147         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 148         struct ptlrpc_request *request;
 149         struct mdt_body *body;
 150         struct md_op_data *op_data;
 151         __u64 hash = *((__u64 *)_hash);
 152         struct page **page_pool;
 153         struct page *page;
 154         struct lu_dirpage *dp;
 155         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 156         int nrdpgs = 0; /* number of pages read actually */
 157         int npages;
 158         int i;
 159         int rc;
 160
 161         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
 162                inode->i_ino, inode->i_generation, inode, hash);
 163
 164         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 165
 166         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 167         if (page_pool != NULL) {
 168                 page_pool[0] = page0;
 169         } else {
 170                 page_pool = &page0;
 171                 max_pages = 1;
 172         }
 173         for (npages = 1; npages < max_pages; npages++) {
 174                 page = page_cache_alloc_cold(inode->i_mapping);
 175                 if (!page)
 176                         break;
 177                 page_pool[npages] = page;
 178         }
 179
 180         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 181                                      LUSTRE_OPC_ANY, NULL);
 182         op_data->op_npages = npages;
 183         op_data->op_offset = hash;
 184         rc = md_readpage(exp, op_data, page_pool, &request);
 185         ll_finish_md_op_data(op_data);
 186         if (rc == 0) {
 187                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 188                 /* Checked by mdc_readpage() */
 189                 LASSERT(body != NULL);
 190
 191                 if (body->valid & OBD_MD_FLSIZE)
 192                         cl_isize_write(inode, body->size);
 193
 194                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 195                          >> PAGE_CACHE_SHIFT;
 196                 SetPageUptodate(page0);
 197         }
 198         unlock_page(page0);
 199         ptlrpc_req_finished(request);
 200
 201         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 202
 203         ll_pagevec_init(&lru_pvec, 0);
 204         for (i = 1; i < npages; i++) {
 205                 unsigned long offset;
 206                 int ret;
 207
 208                 page = page_pool[i];
 209
 210                 if (rc < 0 || i >= nrdpgs) {
 211                         page_cache_release(page);
 212                         continue;
 213                 }
 214
 215                 SetPageUptodate(page);
 216
 217                 dp = kmap(page);
 218                 hash = le64_to_cpu(dp->ldp_hash_start);
 219                 kunmap(page);
 220
 221                 offset = hash_x_index(hash, hash64);
 222
 223                 prefetchw(&page->flags);
 224                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 225                                             GFP_KERNEL);
 226                 if (ret == 0) {
 227                         unlock_page(page);
 228                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 229                                 ll_pagevec_lru_add_file(&lru_pvec);
 230                 } else {
 231                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 232                                " %d\n", offset, ret);
 233                 }
 234                 page_cache_release(page);
 235         }
 236         ll_pagevec_lru_add_file(&lru_pvec);
 237
 238         if (page_pool != &page0)
 239                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 240         return rc;
 241 }
 242
 243 static void ll_check_page(struct inode *dir, struct page *page)
 244 {
 245         /* XXX: check page format later */
 246         SetPageChecked(page);
 247 }
 248
 249 void ll_release_page(struct page *page, int remove)
 250 {
 251         kunmap(page);
 252         if (remove) {
 253                 lock_page(page);
 254                 if (likely(page->mapping != NULL))
 255                         truncate_complete_page(page->mapping, page);
 256                 unlock_page(page);
 257         }
 258         page_cache_release(page);
 259 }
 260
 261 /*
 262  * Find, kmap and return page that contains given hash.
 263  */
 264 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 265                                        __u64 *start, __u64 *end)
 266 {
 267         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 268         struct address_space *mapping = dir->i_mapping;
 269         /*
 270          * Complement of hash is used as an index so that
 271          * radix_tree_gang_lookup() can be used to find a page with starting
 272          * hash _smaller_ than one we are looking for.
 273          */
 274         unsigned long offset = hash_x_index(*hash, hash64);
 275         struct page *page;
 276         int found;
 277
 278         TREE_READ_LOCK_IRQ(mapping);
 279         found = radix_tree_gang_lookup(&mapping->page_tree,
 280                                        (void **)&page, offset, 1);
 281         if (found > 0) {
 282                 struct lu_dirpage *dp;
 283
 284                 page_cache_get(page);
 285                 TREE_READ_UNLOCK_IRQ(mapping);
 286                 /*
 287                  * In contrast to find_lock_page() we are sure that directory
 288                  * page cannot be truncated (while DLM lock is held) and,
 289                  * hence, can avoid restart.
 290                  *
 291                  * In fact, page cannot be locked here at all, because
 292                  * ll_dir_filler() does synchronous io.
 293                  */
 294                 wait_on_page_locked(page);
 295                 if (PageUptodate(page)) {
 296                         dp = kmap(page);
 297                         if (BITS_PER_LONG == 32 && hash64) {
 298                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 299                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 300                                 *hash  = *hash >> 32;
 301                         } else {
 302                                 *start = le64_to_cpu(dp->ldp_hash_start);
 303                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 304                         }
 305                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 306                                  *start, *end, *hash);
 307                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 308                                offset, *start, *end, *hash);
 309                         if (*hash > *end) {
 310                                 ll_release_page(page, 0);
 311                                 page = NULL;
 312                         } else if (*end != *start && *hash == *end) {
 313                                 /*
 314                                  * upon hash collision, remove this page,
 315                                  * otherwise put page reference, and
 316                                  * ll_get_dir_page() will issue RPC to fetch
 317                                  * the page we want.
 318                                  */
 319                                 ll_release_page(page,
 320                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 321                                 page = NULL;
 322                         }
 323                 } else {
 324                         page_cache_release(page);
 325                         page = ERR_PTR(-EIO);
 326                 }
 327
 328         } else {
 329                 TREE_READ_UNLOCK_IRQ(mapping);
 330                 page = NULL;
 331         }
 332         return page;
 333 }
 334
 335 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 336                              struct ll_dir_chain *chain)
 337 {
 338         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 339         struct address_space *mapping = dir->i_mapping;
 340         struct lustre_handle lockh;
 341         struct lu_dirpage *dp;
 342         struct page *page;
 343         ldlm_mode_t mode;
 344         int rc;
 345         __u64 start = 0;
 346         __u64 end = 0;
 347         __u64 lhash = hash;
 348         struct ll_inode_info *lli = ll_i2info(dir);
 349         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 350
 351         mode = LCK_PR;
 352         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 353                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 354         if (!rc) {
 355                 struct ldlm_enqueue_info einfo = {
 356                         .ei_type = LDLM_IBITS,
 357                         .ei_mode = mode,
 358                         .ei_cb_bl = ll_md_blocking_ast,
 359                         .ei_cb_cp = ldlm_completion_ast,
 360                 };
 361                 struct lookup_intent it = { .it_op = IT_READDIR };
 362                 struct ptlrpc_request *request;
 363                 struct md_op_data *op_data;
 364
 365                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 366                 LUSTRE_OPC_ANY, NULL);
 367                 if (IS_ERR(op_data))
 368                         return (void *)op_data;
 369
 370                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 371                                 op_data, &lockh, NULL, 0, NULL, 0);
 372
 373                 ll_finish_md_op_data(op_data);
 374
 375                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 376                 if (request)
 377                         ptlrpc_req_finished(request);
 378                 if (rc < 0) {
 379                         CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
 380                                 PFID(ll_inode2fid(dir)), hash, rc);
 381                         return ERR_PTR(rc);
 382                 }
 383
 384                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 385                        dir, dir->i_ino, dir->i_generation);
 386                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 387                                  &it.d.lustre.it_lock_handle, dir, NULL);
 388         } else {
 389                 /* for cross-ref object, l_ast_data of the lock may not be set,
 390                  * we reset it here */
 391                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 392                                  dir, NULL);
 393         }
 394         ldlm_lock_dump_handle(D_OTHER, &lockh);
 395
 396         mutex_lock(&lli->lli_readdir_mutex);
 397         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 398         if (IS_ERR(page)) {
 399                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 400                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 401                 GOTO(out_unlock, page);
 402         } else if (page != NULL) {
 403                 /*
 404                  * XXX nikita: not entirely correct handling of a corner case:
 405                  * suppose hash chain of entries with hash value HASH crosses
 406                  * border between pages P0 and P1. First both P0 and P1 are
 407                  * cached, seekdir() is called for some entry from the P0 part
 408                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 409                  * happens and finds P1, as it starts with matching hash
 410                  * value. Remaining entries from P0 part of the chain are
 411                  * skipped. (Is that really a bug?)
 412                  *
 413                  * Possible solutions: 0. don't cache P1 is such case, handle
 414                  * it as an "overflow" page. 1. invalidate all pages at
 415                  * once. 2. use HASH|1 as an index for P1.
 416                  */
 417                 GOTO(hash_collision, page);
 418         }
 419
 420         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 421                                ll_dir_filler, &lhash);
 422         if (IS_ERR(page)) {
 423                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 424                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 425                 GOTO(out_unlock, page);
 426         }
 427
 428         wait_on_page_locked(page);
 429         (void)kmap(page);
 430         if (!PageUptodate(page)) {
 431                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 432                        PFID(ll_inode2fid(dir)), hash, -5);
 433                 goto fail;
 434         }
 435         if (!PageChecked(page))
 436                 ll_check_page(dir, page);
 437         if (PageError(page)) {
 438                 CERROR("page error: "DFID" at %llu: rc %d\n",
 439                        PFID(ll_inode2fid(dir)), hash, -5);
 440                 goto fail;
 441         }
 442 hash_collision:
 443         dp = page_address(page);
 444         if (BITS_PER_LONG == 32 && hash64) {
 445                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 446                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 447                 lhash = hash >> 32;
 448         } else {
 449                 start = le64_to_cpu(dp->ldp_hash_start);
 450                 end   = le64_to_cpu(dp->ldp_hash_end);
 451                 lhash = hash;
 452         }
 453         if (end == start) {
 454                 LASSERT(start == lhash);
 455                 CWARN("Page-wide hash collision: %llu\n", end);
 456                 if (BITS_PER_LONG == 32 && hash64)
 457                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 458                               le64_to_cpu(dp->ldp_hash_start),
 459                               le64_to_cpu(dp->ldp_hash_end), hash);
 460                 /*
 461                  * Fetch whole overflow chain...
 462                  *
 463                  * XXX not yet.
 464                  */
 465                 goto fail;
 466         }
 467 out_unlock:
 468         mutex_unlock(&lli->lli_readdir_mutex);
 469         ldlm_lock_decref(&lockh, mode);
 470         return page;
 471
 472 fail:
 473         ll_release_page(page, 1);
 474         page = ERR_PTR(-EIO);
 475         goto out_unlock;
 476 }
 477
 478 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 479 {
 480         struct ll_inode_info *info       = ll_i2info(inode);
 481         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 482         __u64              pos          = ctx->pos;
 483         int                api32      = ll_need_32bit_api(sbi);
 484         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 485         struct page       *page;
 486         struct ll_dir_chain   chain;
 487         int                done = 0;
 488         int                rc = 0;
 489
 490         ll_dir_chain_init(&chain);
 491
 492         page = ll_get_dir_page(inode, pos, &chain);
 493
 494         while (rc == 0 && !done) {
 495                 struct lu_dirpage *dp;
 496                 struct lu_dirent  *ent;
 497
 498                 if (!IS_ERR(page)) {
 499                         /*
 500                          * If page is empty (end of directory is reached),
 501                          * use this value.
 502                          */
 503                         __u64 hash = MDS_DIR_END_OFF;
 504                         __u64 next;
 505
 506                         dp = page_address(page);
 507                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 508                              ent = lu_dirent_next(ent)) {
 509                                 __u16     type;
 510                                 int         namelen;
 511                                 struct lu_fid  fid;
 512                                 __u64     lhash;
 513                                 __u64     ino;
 514
 515                                 /*
 516                                  * XXX: implement correct swabbing here.
 517                                  */
 518
 519                                 hash = le64_to_cpu(ent->lde_hash);
 520                                 if (hash < pos)
 521                                         /*
 522                                          * Skip until we find target hash
 523                                          * value.
 524                                          */
 525                                         continue;
 526
 527                                 namelen = le16_to_cpu(ent->lde_namelen);
 528                                 if (namelen == 0)
 529                                         /*
 530                                          * Skip dummy record.
 531                                          */
 532                                         continue;
 533
 534                                 if (api32 && hash64)
 535                                         lhash = hash >> 32;
 536                                 else
 537                                         lhash = hash;
 538                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 539                                 ino = cl_fid_build_ino(&fid, api32);
 540                                 type = ll_dirent_type_get(ent);
 541                                 ctx->pos = lhash;
 542                                 /* For 'll_nfs_get_name_filldir()', it will try
 543                                  * to access the 'ent' through its 'lde_name',
 544                                  * so the parameter 'name' for 'ctx->actor()'
 545                                  * must be part of the 'ent'.
 546                                  */
 547                                 done = !dir_emit(ctx, ent->lde_name,
 548                                                  namelen, ino, type);
 549                         }
 550                         next = le64_to_cpu(dp->ldp_hash_end);
 551                         if (!done) {
 552                                 pos = next;
 553                                 if (pos == MDS_DIR_END_OFF) {
 554                                         /*
 555                                          * End of directory reached.
 556                                          */
 557                                         done = 1;
 558                                         ll_release_page(page, 0);
 559                                 } else if (1 /* chain is exhausted*/) {
 560                                         /*
 561                                          * Normal case: continue to the next
 562                                          * page.
 563                                          */
 564                                         ll_release_page(page,
 565                                             le32_to_cpu(dp->ldp_flags) &
 566                                                         LDF_COLLIDE);
 567                                         next = pos;
 568                                         page = ll_get_dir_page(inode, pos,
 569                                                                &chain);
 570                                 } else {
 571                                         /*
 572                                          * go into overflow page.
 573                                          */
 574                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 575                                                 LDF_COLLIDE);
 576                                         ll_release_page(page, 1);
 577                                 }
 578                         } else {
 579                                 pos = hash;
 580                                 ll_release_page(page, 0);
 581                         }
 582                 } else {
 583                         rc = PTR_ERR(page);
 584                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 585                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 586                 }
 587         }
 588
 589         ctx->pos = pos;
 590         ll_dir_chain_fini(&chain);
 591         return rc;
 592 }
 593
 594 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 595 {
 596         struct inode            *inode  = filp->f_dentry->d_inode;
 597         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 598         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 599         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 600         int                     api32   = ll_need_32bit_api(sbi);
 601         int                     rc;
 602
 603         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
 604                " 32bit_api %d\n", inode->i_ino, inode->i_generation,
 605                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 606
 607         if (lfd->lfd_pos == MDS_DIR_END_OFF)
 608                 /*
 609                  * end-of-file.
 610                  */
 611                 GOTO(out, rc = 0);
 612
 613         ctx->pos = lfd->lfd_pos;
 614         rc = ll_dir_read(inode, ctx);
 615         lfd->lfd_pos = ctx->pos;
 616         if (ctx->pos == MDS_DIR_END_OFF) {
 617                 if (api32)
 618                         ctx->pos = LL_DIR_END_OFF_32BIT;
 619                 else
 620                         ctx->pos = LL_DIR_END_OFF;
 621         } else {
 622                 if (api32 && hash64)
 623                         ctx->pos >>= 32;
 624         }
 625         filp->f_version = inode->i_version;
 626
 627 out:
 628         if (!rc)
 629                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 630
 631         return rc;
 632 }
 633
 634 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 635 {
 636         struct mgs_send_param *msp;
 637         int rc = 0;
 638
 639         OBD_ALLOC_PTR(msp);
 640         if (!msp)
 641                 return -ENOMEM;
 642
 643         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 644         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 645                                 sizeof(struct mgs_send_param), msp, NULL);
 646         if (rc)
 647                 CERROR("Failed to set parameter: %d\n", rc);
 648         OBD_FREE_PTR(msp);
 649
 650         return rc;
 651 }
 652
 653 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 654                         char *filename)
 655 {
 656         struct ptlrpc_request *request = NULL;
 657         struct md_op_data *op_data;
 658         struct ll_sb_info *sbi = ll_i2sbi(dir);
 659         int mode;
 660         int err;
 661
 662         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 663         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 664                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 665                                      lump);
 666         if (IS_ERR(op_data))
 667                 GOTO(err_exit, err = PTR_ERR(op_data));
 668
 669         op_data->op_cli_flags |= CLI_SET_MEA;
 670         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 671                         from_kuid(&init_user_ns, current_fsuid()),
 672                         from_kgid(&init_user_ns, current_fsgid()),
 673                         cfs_curproc_cap_pack(), 0, &request);
 674         ll_finish_md_op_data(op_data);
 675         if (err)
 676                 GOTO(err_exit, err);
 677 err_exit:
 678         ptlrpc_req_finished(request);
 679         return err;
 680 }
 681
 682 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 683                      int set_default)
 684 {
 685         struct ll_sb_info *sbi = ll_i2sbi(inode);
 686         struct md_op_data *op_data;
 687         struct ptlrpc_request *req = NULL;
 688         int rc = 0;
 689         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 690         struct obd_device *mgc = lsi->lsi_mgc;
 691         int lum_size;
 692
 693         if (lump != NULL) {
 694                 /*
 695                  * This is coming from userspace, so should be in
 696                  * local endian.  But the MDS would like it in little
 697                  * endian, so we swab it before we send it.
 698                  */
 699                 switch (lump->lmm_magic) {
 700                 case LOV_USER_MAGIC_V1: {
 701                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 702                                 lustre_swab_lov_user_md_v1(lump);
 703                         lum_size = sizeof(struct lov_user_md_v1);
 704                         break;
 705                 }
 706                 case LOV_USER_MAGIC_V3: {
 707                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 708                                 lustre_swab_lov_user_md_v3(
 709                                         (struct lov_user_md_v3 *)lump);
 710                         lum_size = sizeof(struct lov_user_md_v3);
 711                         break;
 712                 }
 713                 default: {
 714                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 715                                         " %#08x != %#08x nor %#08x\n",
 716                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 717                                         LOV_USER_MAGIC_V3);
 718                         return -EINVAL;
 719                 }
 720                 }
 721         } else {
 722                 lum_size = sizeof(struct lov_user_md_v1);
 723         }
 724
 725         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 726                                      LUSTRE_OPC_ANY, NULL);
 727         if (IS_ERR(op_data))
 728                 return PTR_ERR(op_data);
 729
 730         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 731                 op_data->op_cli_flags |= CLI_SET_MEA;
 732
 733         /* swabbing is done in lov_setstripe() on server side */
 734         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 735                         NULL, 0, &req, NULL);
 736         ll_finish_md_op_data(op_data);
 737         ptlrpc_req_finished(req);
 738         if (rc) {
 739                 if (rc != -EPERM && rc != -EACCES)
 740                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 741         }
 742
 743         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 744          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 745          need to make the distinction between the 2 versions */
 746         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 747                 char *param = NULL;
 748                 char *buf;
 749
 750                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 751                 if (param == NULL)
 752                         GOTO(end, rc = -ENOMEM);
 753
 754                 buf = param;
 755                 /* Get fsname and assume devname to be -MDT0000. */
 756                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 757                 strcat(buf, "-MDT0000.lov");
 758                 buf += strlen(buf);
 759
 760                 /* Set root stripesize */
 761                 sprintf(buf, ".stripesize=%u",
 762                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 763                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 764                 if (rc)
 765                         GOTO(end, rc);
 766
 767                 /* Set root stripecount */
 768                 sprintf(buf, ".stripecount=%hd",
 769                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 770                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 771                 if (rc)
 772                         GOTO(end, rc);
 773
 774                 /* Set root stripeoffset */
 775                 sprintf(buf, ".stripeoffset=%hd",
 776                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 777                         (typeof(lump->lmm_stripe_offset))(-1));
 778                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 779
 780 end:
 781                 if (param != NULL)
 782                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 783         }
 784         return rc;
 785 }
 786
 787 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 788                      int *lmm_size, struct ptlrpc_request **request)
 789 {
 790         struct ll_sb_info *sbi = ll_i2sbi(inode);
 791         struct mdt_body   *body;
 792         struct lov_mds_md *lmm = NULL;
 793         struct ptlrpc_request *req = NULL;
 794         int rc, lmmsize;
 795         struct md_op_data *op_data;
 796
 797         rc = ll_get_default_mdsize(sbi, &lmmsize);
 798         if (rc)
 799                 return rc;
 800
 801         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 802                                      0, lmmsize, LUSTRE_OPC_ANY,
 803                                      NULL);
 804         if (IS_ERR(op_data))
 805                 return PTR_ERR(op_data);
 806
 807         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 808         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 809         ll_finish_md_op_data(op_data);
 810         if (rc < 0) {
 811                 CDEBUG(D_INFO, "md_getattr failed on inode "
 812                        "%lu/%u: rc %d\n", inode->i_ino,
 813                        inode->i_generation, rc);
 814                 GOTO(out, rc);
 815         }
 816
 817         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 818         LASSERT(body != NULL);
 819
 820         lmmsize = body->eadatasize;
 821
 822         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 823             lmmsize == 0) {
 824                 GOTO(out, rc = -ENODATA);
 825         }
 826
 827         lmm = req_capsule_server_sized_get(&req->rq_pill,
 828                                            &RMF_MDT_MD, lmmsize);
 829         LASSERT(lmm != NULL);
 830
 831         /*
 832          * This is coming from the MDS, so is probably in
 833          * little endian.  We convert it to host endian before
 834          * passing it to userspace.
 835          */
 836         /* We don't swab objects for directories */
 837         switch (le32_to_cpu(lmm->lmm_magic)) {
 838         case LOV_MAGIC_V1:
 839                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 840                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 841                 break;
 842         case LOV_MAGIC_V3:
 843                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 844                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 845                 break;
 846         default:
 847                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 848                 rc = -EPROTO;
 849         }
 850 out:
 851         *lmmp = lmm;
 852         *lmm_size = lmmsize;
 853         *request = req;
 854         return rc;
 855 }
 856
 857 /*
 858  *  Get MDT index for the inode.
 859  */
 860 int ll_get_mdt_idx(struct inode *inode)
 861 {
 862         struct ll_sb_info *sbi = ll_i2sbi(inode);
 863         struct md_op_data *op_data;
 864         int rc, mdtidx;
 865
 866         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 867                                      0, LUSTRE_OPC_ANY, NULL);
 868         if (IS_ERR(op_data))
 869                 return PTR_ERR(op_data);
 870
 871         op_data->op_flags |= MF_GET_MDT_IDX;
 872         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 873         mdtidx = op_data->op_mds;
 874         ll_finish_md_op_data(op_data);
 875         if (rc < 0) {
 876                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 877                 return rc;
 878         }
 879         return mdtidx;
 880 }
 881
 882 /**
 883  * Generic handler to do any pre-copy work.
 884  *
 885  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 886  * first information for it that real work has started.
 887  *
 888  * Moreover, for a ARCHIVE request, it will sample the file data version and
 889  * store it in \a copy.
 890  *
 891  * \return 0 on success.
 892  */
 893 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 894 {
 895         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 896         struct hsm_progress_kernel       hpk;
 897         int                              rc;
 898
 899         /* Forge a hsm_progress based on data from copy. */
 900         hpk.hpk_fid = copy->hc_hai.hai_fid;
 901         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 902         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 903         hpk.hpk_extent.length = 0;
 904         hpk.hpk_flags = 0;
 905         hpk.hpk_errval = 0;
 906         hpk.hpk_data_version = 0;
 907
 908
 909         /* For archive request, we need to read the current file version. */
 910         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 911                 struct inode    *inode;
 912                 __u64            data_version = 0;
 913
 914                 /* Get inode for this fid */
 915                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 916                 if (IS_ERR(inode)) {
 917                         hpk.hpk_flags |= HP_FLAG_RETRY;
 918                         /* hpk_errval is >= 0 */
 919                         hpk.hpk_errval = -PTR_ERR(inode);
 920                         GOTO(progress, rc = PTR_ERR(inode));
 921                 }
 922
 923                 /* Read current file data version */
 924                 rc = ll_data_version(inode, &data_version, 1);
 925                 iput(inode);
 926                 if (rc != 0) {
 927                         CDEBUG(D_HSM, "Could not read file data version of "
 928                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 929                                       PFID(&copy->hc_hai.hai_fid), rc,
 930                                       copy->hc_hai.hai_cookie);
 931                         hpk.hpk_flags |= HP_FLAG_RETRY;
 932                         /* hpk_errval must be >= 0 */
 933                         hpk.hpk_errval = -rc;
 934                         GOTO(progress, rc);
 935                 }
 936
 937                 /* Store it the hsm_copy for later copytool use.
 938                  * Always modified even if no lsm. */
 939                 copy->hc_data_version = data_version;
 940         }
 941
 942 progress:
 943         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 944                            &hpk, NULL);
 945
 946         return rc;
 947 }
 948
 949 /**
 950  * Generic handler to do any post-copy work.
 951  *
 952  * It will send the last hsm_progress update to coordinator to inform it
 953  * that copy is finished and whether it was successful or not.
 954  *
 955  * Moreover,
 956  * - for ARCHIVE request, it will sample the file data version and compare it
 957  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 958  *   will be considered as failed.
 959  * - for RESTORE request, it will sample the file data version and send it to
 960  *   coordinator which is useful if the file was imported as 'released'.
 961  *
 962  * \return 0 on success.
 963  */
 964 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 965 {
 966         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 967         struct hsm_progress_kernel       hpk;
 968         int                              rc;
 969
 970         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 971         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 972          * initialized if copy_end was called with copy == NULL.
 973          */
 974
 975         /* Forge a hsm_progress based on data from copy. */
 976         hpk.hpk_fid = copy->hc_hai.hai_fid;
 977         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 978         hpk.hpk_extent = copy->hc_hai.hai_extent;
 979         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 980         hpk.hpk_errval = copy->hc_errval;
 981         hpk.hpk_data_version = 0;
 982
 983         /* For archive request, we need to check the file data was not changed.
 984          *
 985          * For restore request, we need to send the file data version, this is
 986          * useful when the file was created using hsm_import.
 987          */
 988         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 989              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 990             (copy->hc_errval == 0)) {
 991                 struct inode    *inode;
 992                 __u64            data_version = 0;
 993
 994                 /* Get lsm for this fid */
 995                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 996                 if (IS_ERR(inode)) {
 997                         hpk.hpk_flags |= HP_FLAG_RETRY;
 998                         /* hpk_errval must be >= 0 */
 999                         hpk.hpk_errval = -PTR_ERR(inode);
1000                         GOTO(progress, rc = PTR_ERR(inode));
1001                 }
1002
1003                 rc = ll_data_version(inode, &data_version,
1004                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1005                 iput(inode);
1006                 if (rc) {
1007                         CDEBUG(D_HSM, "Could not read file data version. "
1008                                       "Request could not be confirmed.\n");
1009                         if (hpk.hpk_errval == 0)
1010                                 hpk.hpk_errval = -rc;
1011                         GOTO(progress, rc);
1012                 }
1013
1014                 /* Store it the hsm_copy for later copytool use.
1015                  * Always modified even if no lsm. */
1016                 hpk.hpk_data_version = data_version;
1017
1018                 /* File could have been stripped during archiving, so we need
1019                  * to check anyway. */
1020                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1021                     (copy->hc_data_version != data_version)) {
1022                         CDEBUG(D_HSM, "File data version mismatched. "
1023                               "File content was changed during archiving. "
1024                                DFID", start:%#llx current:%#llx\n",
1025                                PFID(&copy->hc_hai.hai_fid),
1026                                copy->hc_data_version, data_version);
1027                         /* File was changed, send error to cdt. Do not ask for
1028                          * retry because if a file is modified frequently,
1029                          * the cdt will loop on retried archive requests.
1030                          * The policy engine will ask for a new archive later
1031                          * when the file will not be modified for some tunable
1032                          * time */
1033                         /* we do not notify caller */
1034                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1035                         /* hpk_errval must be >= 0 */
1036                         hpk.hpk_errval = EBUSY;
1037                 }
1038
1039         }
1040
1041 progress:
1042         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1043                            &hpk, NULL);
1044
1045         return rc;
1046 }
1047
1048
1049 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1050                           const void __user *data, size_t size)
1051 {
1052         void *copy;
1053         int rc;
1054
1055         OBD_ALLOC(copy, size);
1056         if (copy == NULL)
1057                 return -ENOMEM;
1058
1059         if (copy_from_user(copy, data, size)) {
1060                 rc = -EFAULT;
1061                 goto out;
1062         }
1063
1064         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1065 out:
1066         OBD_FREE(copy, size);
1067
1068         return rc;
1069 }
1070
1071 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1072 {
1073         int cmd = qctl->qc_cmd;
1074         int type = qctl->qc_type;
1075         int id = qctl->qc_id;
1076         int valid = qctl->qc_valid;
1077         int rc = 0;
1078
1079         switch (cmd) {
1080         case LUSTRE_Q_INVALIDATE:
1081         case LUSTRE_Q_FINVALIDATE:
1082         case Q_QUOTAON:
1083         case Q_QUOTAOFF:
1084         case Q_SETQUOTA:
1085         case Q_SETINFO:
1086                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1087                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1088                         return -EPERM;
1089                 break;
1090         case Q_GETQUOTA:
1091                 if (((type == USRQUOTA &&
1092                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1093                      (type == GRPQUOTA &&
1094                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1095                     (!capable(CFS_CAP_SYS_ADMIN) ||
1096                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1097                         return -EPERM;
1098                 break;
1099         case Q_GETINFO:
1100                 break;
1101         default:
1102                 CERROR("unsupported quotactl op: %#x\n", cmd);
1103                 return -ENOTTY;
1104         }
1105
1106         if (valid != QC_GENERAL) {
1107                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1108                         return -EOPNOTSUPP;
1109
1110                 if (cmd == Q_GETINFO)
1111                         qctl->qc_cmd = Q_GETOINFO;
1112                 else if (cmd == Q_GETQUOTA)
1113                         qctl->qc_cmd = Q_GETOQUOTA;
1114                 else
1115                         return -EINVAL;
1116
1117                 switch (valid) {
1118                 case QC_MDTIDX:
1119                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1120                                            sizeof(*qctl), qctl, NULL);
1121                         break;
1122                 case QC_OSTIDX:
1123                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1124                                            sizeof(*qctl), qctl, NULL);
1125                         break;
1126                 case QC_UUID:
1127                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1128                                            sizeof(*qctl), qctl, NULL);
1129                         if (rc == -EAGAIN)
1130                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1131                                                    sbi->ll_dt_exp,
1132                                                    sizeof(*qctl), qctl, NULL);
1133                         break;
1134                 default:
1135                         rc = -EINVAL;
1136                         break;
1137                 }
1138
1139                 if (rc)
1140                         return rc;
1141
1142                 qctl->qc_cmd = cmd;
1143         } else {
1144                 struct obd_quotactl *oqctl;
1145
1146                 OBD_ALLOC_PTR(oqctl);
1147                 if (oqctl == NULL)
1148                         return -ENOMEM;
1149
1150                 QCTL_COPY(oqctl, qctl);
1151                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1152                 if (rc) {
1153                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1154                                 oqctl->qc_cmd = Q_QUOTAOFF;
1155                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1156                         }
1157                         OBD_FREE_PTR(oqctl);
1158                         return rc;
1159                 }
1160                 /* If QIF_SPACE is not set, client should collect the
1161                  * space usage from OSSs by itself */
1162                 if (cmd == Q_GETQUOTA &&
1163                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1164                     !oqctl->qc_dqblk.dqb_curspace) {
1165                         struct obd_quotactl *oqctl_tmp;
1166
1167                         OBD_ALLOC_PTR(oqctl_tmp);
1168                         if (oqctl_tmp == NULL)
1169                                 GOTO(out, rc = -ENOMEM);
1170
1171                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1172                         oqctl_tmp->qc_id = oqctl->qc_id;
1173                         oqctl_tmp->qc_type = oqctl->qc_type;
1174
1175                         /* collect space usage from OSTs */
1176                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1177                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1178                         if (!rc || rc == -EREMOTEIO) {
1179                                 oqctl->qc_dqblk.dqb_curspace =
1180                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1181                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1182                         }
1183
1184                         /* collect space & inode usage from MDTs */
1185                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1186                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1187                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1188                         if (!rc || rc == -EREMOTEIO) {
1189                                 oqctl->qc_dqblk.dqb_curspace +=
1190                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1191                                 oqctl->qc_dqblk.dqb_curinodes =
1192                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1193                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1194                         } else {
1195                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1196                         }
1197
1198                         OBD_FREE_PTR(oqctl_tmp);
1199                 }
1200 out:
1201                 QCTL_COPY(qctl, oqctl);
1202                 OBD_FREE_PTR(oqctl);
1203         }
1204
1205         return rc;
1206 }
1207
1208 static char *
1209 ll_getname(const char __user *filename)
1210 {
1211         int ret = 0, len;
1212         char *tmp = __getname();
1213
1214         if (!tmp)
1215                 return ERR_PTR(-ENOMEM);
1216
1217         len = strncpy_from_user(tmp, filename, PATH_MAX);
1218         if (len == 0)
1219                 ret = -ENOENT;
1220         else if (len > PATH_MAX)
1221                 ret = -ENAMETOOLONG;
1222
1223         if (ret) {
1224                 __putname(tmp);
1225                 tmp =  ERR_PTR(ret);
1226         }
1227         return tmp;
1228 }
1229
1230 #define ll_putname(filename) __putname(filename)
1231
1232 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1233 {
1234         struct inode *inode = file->f_dentry->d_inode;
1235         struct ll_sb_info *sbi = ll_i2sbi(inode);
1236         struct obd_ioctl_data *data;
1237         int rc = 0;
1238
1239         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1240                inode->i_ino, inode->i_generation, inode, cmd);
1241
1242         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1243         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1244                 return -ENOTTY;
1245
1246         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1247         switch(cmd) {
1248         case FSFILT_IOC_GETFLAGS:
1249         case FSFILT_IOC_SETFLAGS:
1250                 return ll_iocontrol(inode, file, cmd, arg);
1251         case FSFILT_IOC_GETVERSION_OLD:
1252         case FSFILT_IOC_GETVERSION:
1253                 return put_user(inode->i_generation, (int *)arg);
1254         /* We need to special case any other ioctls we want to handle,
1255          * to send them to the MDS/OST as appropriate and to properly
1256          * network encode the arg field.
1257         case FSFILT_IOC_SETVERSION_OLD:
1258         case FSFILT_IOC_SETVERSION:
1259         */
1260         case LL_IOC_GET_MDTIDX: {
1261                 int mdtidx;
1262
1263                 mdtidx = ll_get_mdt_idx(inode);
1264                 if (mdtidx < 0)
1265                         return mdtidx;
1266
1267                 if (put_user((int)mdtidx, (int *)arg))
1268                         return -EFAULT;
1269
1270                 return 0;
1271         }
1272         case IOC_MDC_LOOKUP: {
1273                 struct ptlrpc_request *request = NULL;
1274                 int namelen, len = 0;
1275                 char *buf = NULL;
1276                 char *filename;
1277                 struct md_op_data *op_data;
1278
1279                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1280                 if (rc)
1281                         return rc;
1282                 data = (void *)buf;
1283
1284                 filename = data->ioc_inlbuf1;
1285                 namelen = strlen(filename);
1286
1287                 if (namelen < 1) {
1288                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1289                         GOTO(out_free, rc = -EINVAL);
1290                 }
1291
1292                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1293                                              0, LUSTRE_OPC_ANY, NULL);
1294                 if (IS_ERR(op_data))
1295                         GOTO(out_free, rc = PTR_ERR(op_data));
1296
1297                 op_data->op_valid = OBD_MD_FLID;
1298                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1299                 ll_finish_md_op_data(op_data);
1300                 if (rc < 0) {
1301                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1302                         GOTO(out_free, rc);
1303                 }
1304                 ptlrpc_req_finished(request);
1305 out_free:
1306                 obd_ioctl_freedata(buf, len);
1307                 return rc;
1308         }
1309         case LL_IOC_LMV_SETSTRIPE: {
1310                 struct lmv_user_md  *lum;
1311                 char            *buf = NULL;
1312                 char            *filename;
1313                 int              namelen = 0;
1314                 int              lumlen = 0;
1315                 int              len;
1316                 int              rc;
1317
1318                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1319                 if (rc)
1320                         return rc;
1321
1322                 data = (void *)buf;
1323                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1324                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1325                         GOTO(lmv_out_free, rc = -EINVAL);
1326
1327                 filename = data->ioc_inlbuf1;
1328                 namelen = data->ioc_inllen1;
1329
1330                 if (namelen < 1) {
1331                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1332                         GOTO(lmv_out_free, rc = -EINVAL);
1333                 }
1334                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1335                 lumlen = data->ioc_inllen2;
1336
1337                 if (lum->lum_magic != LMV_USER_MAGIC ||
1338                     lumlen != sizeof(*lum)) {
1339                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1340                                filename, lum->lum_magic, lumlen, -EFAULT);
1341                         GOTO(lmv_out_free, rc = -EINVAL);
1342                 }
1343
1344                 /**
1345                  * ll_dir_setdirstripe will be used to set dir stripe
1346                  *  mdc_create--->mdt_reint_create (with dirstripe)
1347                  */
1348                 rc = ll_dir_setdirstripe(inode, lum, filename);
1349 lmv_out_free:
1350                 obd_ioctl_freedata(buf, len);
1351                 return rc;
1352
1353         }
1354         case LL_IOC_LOV_SETSTRIPE: {
1355                 struct lov_user_md_v3 lumv3;
1356                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1357                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1358                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1359
1360                 int set_default = 0;
1361
1362                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1363                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1364                         sizeof(lumv3p->lmm_objects[0]));
1365                 /* first try with v1 which is smaller than v3 */
1366                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1367                         return -EFAULT;
1368
1369                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1370                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1371                                 return -EFAULT;
1372                 }
1373
1374                 if (inode->i_sb->s_root == file->f_dentry)
1375                         set_default = 1;
1376
1377                 /* in v1 and v3 cases lumv1 points to data */
1378                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1379
1380                 return rc;
1381         }
1382         case LL_IOC_LMV_GETSTRIPE: {
1383                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1384                 struct lmv_user_md lum;
1385                 struct lmv_user_md *tmp;
1386                 int lum_size;
1387                 int rc = 0;
1388                 int mdtindex;
1389
1390                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1391                         return -EFAULT;
1392
1393                 if (lum.lum_magic != LMV_MAGIC_V1)
1394                         return -EINVAL;
1395
1396                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1397                 OBD_ALLOC(tmp, lum_size);
1398                 if (tmp == NULL)
1399                         GOTO(free_lmv, rc = -ENOMEM);
1400
1401                 *tmp = lum;
1402                 tmp->lum_type = LMV_STRIPE_TYPE;
1403                 tmp->lum_stripe_count = 1;
1404                 mdtindex = ll_get_mdt_idx(inode);
1405                 if (mdtindex < 0)
1406                         GOTO(free_lmv, rc = -ENOMEM);
1407
1408                 tmp->lum_stripe_offset = mdtindex;
1409                 tmp->lum_objects[0].lum_mds = mdtindex;
1410                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1411                        sizeof(struct lu_fid));
1412                 if (copy_to_user((void *)arg, tmp, lum_size))
1413                         GOTO(free_lmv, rc = -EFAULT);
1414 free_lmv:
1415                 if (tmp)
1416                         OBD_FREE(tmp, lum_size);
1417                 return rc;
1418         }
1419         case LL_IOC_REMOVE_ENTRY: {
1420                 char            *filename = NULL;
1421                 int              namelen = 0;
1422                 int              rc;
1423
1424                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1425                  * unsupported server, which might crash the server(LU-2730),
1426                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1427                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1428                  * server will support REINT_RMENTRY XXX*/
1429                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1430                         return -ENOTSUPP;
1431
1432                 filename = ll_getname((const char *)arg);
1433                 if (IS_ERR(filename))
1434                         return PTR_ERR(filename);
1435
1436                 namelen = strlen(filename);
1437                 if (namelen < 1)
1438                         GOTO(out_rmdir, rc = -EINVAL);
1439
1440                 rc = ll_rmdir_entry(inode, filename, namelen);
1441 out_rmdir:
1442                 if (filename)
1443                         ll_putname(filename);
1444                 return rc;
1445         }
1446         case LL_IOC_LOV_SWAP_LAYOUTS:
1447                 return -EPERM;
1448         case LL_IOC_OBD_STATFS:
1449                 return ll_obd_statfs(inode, (void *)arg);
1450         case LL_IOC_LOV_GETSTRIPE:
1451         case LL_IOC_MDC_GETINFO:
1452         case IOC_MDC_GETFILEINFO:
1453         case IOC_MDC_GETFILESTRIPE: {
1454                 struct ptlrpc_request *request = NULL;
1455                 struct lov_user_md *lump;
1456                 struct lov_mds_md *lmm = NULL;
1457                 struct mdt_body *body;
1458                 char *filename = NULL;
1459                 int lmmsize;
1460
1461                 if (cmd == IOC_MDC_GETFILEINFO ||
1462                     cmd == IOC_MDC_GETFILESTRIPE) {
1463                         filename = ll_getname((const char *)arg);
1464                         if (IS_ERR(filename))
1465                                 return PTR_ERR(filename);
1466
1467                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1468                                                       &lmmsize, &request);
1469                 } else {
1470                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1471                 }
1472
1473                 if (request) {
1474                         body = req_capsule_server_get(&request->rq_pill,
1475                                                       &RMF_MDT_BODY);
1476                         LASSERT(body != NULL);
1477                 } else {
1478                         GOTO(out_req, rc);
1479                 }
1480
1481                 if (rc < 0) {
1482                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1483                                                cmd == LL_IOC_MDC_GETINFO))
1484                                 GOTO(skip_lmm, rc = 0);
1485                         else
1486                                 GOTO(out_req, rc);
1487                 }
1488
1489                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1490                     cmd == LL_IOC_LOV_GETSTRIPE) {
1491                         lump = (struct lov_user_md *)arg;
1492                 } else {
1493                         struct lov_user_mds_data *lmdp;
1494                         lmdp = (struct lov_user_mds_data *)arg;
1495                         lump = &lmdp->lmd_lmm;
1496                 }
1497                 if (copy_to_user(lump, lmm, lmmsize)) {
1498                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1499                                 GOTO(out_req, rc = -EFAULT);
1500                         rc = -EOVERFLOW;
1501                 }
1502 skip_lmm:
1503                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1504                         struct lov_user_mds_data *lmdp;
1505                         lstat_t st = { 0 };
1506
1507                         st.st_dev     = inode->i_sb->s_dev;
1508                         st.st_mode    = body->mode;
1509                         st.st_nlink   = body->nlink;
1510                         st.st_uid     = body->uid;
1511                         st.st_gid     = body->gid;
1512                         st.st_rdev    = body->rdev;
1513                         st.st_size    = body->size;
1514                         st.st_blksize = PAGE_CACHE_SIZE;
1515                         st.st_blocks  = body->blocks;
1516                         st.st_atime   = body->atime;
1517                         st.st_mtime   = body->mtime;
1518                         st.st_ctime   = body->ctime;
1519                         st.st_ino     = inode->i_ino;
1520
1521                         lmdp = (struct lov_user_mds_data *)arg;
1522                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1523                                 GOTO(out_req, rc = -EFAULT);
1524                 }
1525
1526 out_req:
1527                 ptlrpc_req_finished(request);
1528                 if (filename)
1529                         ll_putname(filename);
1530                 return rc;
1531         }
1532         case IOC_LOV_GETINFO: {
1533                 struct lov_user_mds_data *lumd;
1534                 struct lov_stripe_md *lsm;
1535                 struct lov_user_md *lum;
1536                 struct lov_mds_md *lmm;
1537                 int lmmsize;
1538                 lstat_t st;
1539
1540                 lumd = (struct lov_user_mds_data *)arg;
1541                 lum = &lumd->lmd_lmm;
1542
1543                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1544                 if (rc)
1545                         return rc;
1546
1547                 OBD_ALLOC_LARGE(lmm, lmmsize);
1548                 if (lmm == NULL)
1549                         return -ENOMEM;
1550                 if (copy_from_user(lmm, lum, lmmsize))
1551                         GOTO(free_lmm, rc = -EFAULT);
1552
1553                 switch (lmm->lmm_magic) {
1554                 case LOV_USER_MAGIC_V1:
1555                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1556                                 break;
1557                         /* swab objects first so that stripes num will be sane */
1558                         lustre_swab_lov_user_md_objects(
1559                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1560                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1561                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1562                         break;
1563                 case LOV_USER_MAGIC_V3:
1564                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1565                                 break;
1566                         /* swab objects first so that stripes num will be sane */
1567                         lustre_swab_lov_user_md_objects(
1568                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1569                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1570                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1571                         break;
1572                 default:
1573                         GOTO(free_lmm, rc = -EINVAL);
1574                 }
1575
1576                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1577                 if (rc < 0)
1578                         GOTO(free_lmm, rc = -ENOMEM);
1579
1580                 /* Perform glimpse_size operation. */
1581                 memset(&st, 0, sizeof(st));
1582
1583                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1584                 if (rc)
1585                         GOTO(free_lsm, rc);
1586
1587                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1588                         GOTO(free_lsm, rc = -EFAULT);
1589
1590 free_lsm:
1591                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1592 free_lmm:
1593                 OBD_FREE_LARGE(lmm, lmmsize);
1594                 return rc;
1595         }
1596         case OBD_IOC_LLOG_CATINFO: {
1597                 return -EOPNOTSUPP;
1598         }
1599         case OBD_IOC_QUOTACHECK: {
1600                 struct obd_quotactl *oqctl;
1601                 int error = 0;
1602
1603                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1604                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1605                         return -EPERM;
1606
1607                 OBD_ALLOC_PTR(oqctl);
1608                 if (!oqctl)
1609                         return -ENOMEM;
1610                 oqctl->qc_type = arg;
1611                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1612                 if (rc < 0) {
1613                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1614                         error = rc;
1615                 }
1616
1617                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1618                 if (rc < 0)
1619                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1620
1621                 OBD_FREE_PTR(oqctl);
1622                 return error ?: rc;
1623         }
1624         case OBD_IOC_POLL_QUOTACHECK: {
1625                 struct if_quotacheck *check;
1626
1627                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1628                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1629                         return -EPERM;
1630
1631                 OBD_ALLOC_PTR(check);
1632                 if (!check)
1633                         return -ENOMEM;
1634
1635                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1636                                    NULL);
1637                 if (rc) {
1638                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1639                         if (copy_to_user((void *)arg, check,
1640                                              sizeof(*check)))
1641                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1642                         GOTO(out_poll, rc);
1643                 }
1644
1645                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1646                                    NULL);
1647                 if (rc) {
1648                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1649                         if (copy_to_user((void *)arg, check,
1650                                              sizeof(*check)))
1651                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1652                         GOTO(out_poll, rc);
1653                 }
1654 out_poll:
1655                 OBD_FREE_PTR(check);
1656                 return rc;
1657         }
1658 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1659         case LL_IOC_QUOTACTL_18: {
1660                 /* copy the old 1.x quota struct for internal use, then copy
1661                  * back into old format struct.  For 1.8 compatibility. */
1662                 struct if_quotactl_18 *qctl_18;
1663                 struct if_quotactl *qctl_20;
1664
1665                 OBD_ALLOC_PTR(qctl_18);
1666                 if (!qctl_18)
1667                         return -ENOMEM;
1668
1669                 OBD_ALLOC_PTR(qctl_20);
1670                 if (!qctl_20)
1671                         GOTO(out_quotactl_18, rc = -ENOMEM);
1672
1673                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1674                         GOTO(out_quotactl_20, rc = -ENOMEM);
1675
1676                 QCTL_COPY(qctl_20, qctl_18);
1677                 qctl_20->qc_idx = 0;
1678
1679                 /* XXX: dqb_valid was borrowed as a flag to mark that
1680                  *      only mds quota is wanted */
1681                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1682                     qctl_18->qc_dqblk.dqb_valid) {
1683                         qctl_20->qc_valid = QC_MDTIDX;
1684                         qctl_20->qc_dqblk.dqb_valid = 0;
1685                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1686                         qctl_20->qc_valid = QC_UUID;
1687                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1688                 } else {
1689                         qctl_20->qc_valid = QC_GENERAL;
1690                 }
1691
1692                 rc = quotactl_ioctl(sbi, qctl_20);
1693
1694                 if (rc == 0) {
1695                         QCTL_COPY(qctl_18, qctl_20);
1696                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1697
1698                         if (copy_to_user((void *)arg, qctl_18,
1699                                              sizeof(*qctl_18)))
1700                                 rc = -EFAULT;
1701                 }
1702
1703 out_quotactl_20:
1704                 OBD_FREE_PTR(qctl_20);
1705 out_quotactl_18:
1706                 OBD_FREE_PTR(qctl_18);
1707                 return rc;
1708         }
1709 #else
1710 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1711 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1712         case LL_IOC_QUOTACTL: {
1713                 struct if_quotactl *qctl;
1714
1715                 OBD_ALLOC_PTR(qctl);
1716                 if (!qctl)
1717                         return -ENOMEM;
1718
1719                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1720                         GOTO(out_quotactl, rc = -EFAULT);
1721
1722                 rc = quotactl_ioctl(sbi, qctl);
1723
1724                 if (rc == 0 && copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1725                         rc = -EFAULT;
1726
1727 out_quotactl:
1728                 OBD_FREE_PTR(qctl);
1729                 return rc;
1730         }
1731         case OBD_IOC_GETDTNAME:
1732         case OBD_IOC_GETMDNAME:
1733                 return ll_get_obd_name(inode, cmd, arg);
1734         case LL_IOC_FLUSHCTX:
1735                 return ll_flush_ctx(inode);
1736 #ifdef CONFIG_FS_POSIX_ACL
1737         case LL_IOC_RMTACL: {
1738             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1739                 inode == inode->i_sb->s_root->d_inode) {
1740                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1741
1742                 LASSERT(fd != NULL);
1743                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1744                 if (!rc)
1745                         fd->fd_flags |= LL_FILE_RMTACL;
1746                 return rc;
1747             } else
1748                 return 0;
1749         }
1750 #endif
1751         case LL_IOC_GETOBDCOUNT: {
1752                 int count, vallen;
1753                 struct obd_export *exp;
1754
1755                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1756                         return -EFAULT;
1757
1758                 /* get ost count when count is zero, get mdt count otherwise */
1759                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1760                 vallen = sizeof(count);
1761                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1762                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1763                 if (rc) {
1764                         CERROR("get target count failed: %d\n", rc);
1765                         return rc;
1766                 }
1767
1768                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1769                         return -EFAULT;
1770
1771                 return 0;
1772         }
1773         case LL_IOC_PATH2FID:
1774                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1775                                      sizeof(struct lu_fid)))
1776                         return -EFAULT;
1777                 return 0;
1778         case LL_IOC_GET_CONNECT_FLAGS: {
1779                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1780         }
1781         case OBD_IOC_CHANGELOG_SEND:
1782         case OBD_IOC_CHANGELOG_CLEAR:
1783                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1784                                     sizeof(struct ioc_changelog));
1785                 return rc;
1786         case OBD_IOC_FID2PATH:
1787                 return ll_fid2path(inode, (void *)arg);
1788         case LL_IOC_HSM_REQUEST: {
1789                 struct hsm_user_request *hur;
1790                 int                      totalsize;
1791
1792                 OBD_ALLOC_PTR(hur);
1793                 if (hur == NULL)
1794                         return -ENOMEM;
1795
1796                 /* We don't know the true size yet; copy the fixed-size part */
1797                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1798                         OBD_FREE_PTR(hur);
1799                         return -EFAULT;
1800                 }
1801
1802                 /* Compute the whole struct size */
1803                 totalsize = hur_len(hur);
1804                 OBD_FREE_PTR(hur);
1805
1806                 /* Final size will be more than double totalsize */
1807                 if (totalsize >= MDS_MAXREQSIZE / 3)
1808                         return -E2BIG;
1809
1810                 OBD_ALLOC_LARGE(hur, totalsize);
1811                 if (hur == NULL)
1812                         return -ENOMEM;
1813
1814                 /* Copy the whole struct */
1815                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1816                         OBD_FREE_LARGE(hur, totalsize);
1817                         return -EFAULT;
1818                 }
1819
1820                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1821                         const struct lu_fid *fid;
1822                         struct inode *f;
1823                         int i;
1824
1825                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1826                                 fid = &hur->hur_user_item[i].hui_fid;
1827                                 f = search_inode_for_lustre(inode->i_sb, fid);
1828                                 if (IS_ERR(f)) {
1829                                         rc = PTR_ERR(f);
1830                                         break;
1831                                 }
1832
1833                                 rc = ll_hsm_release(f);
1834                                 iput(f);
1835                                 if (rc != 0)
1836                                         break;
1837                         }
1838                 } else {
1839                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1840                                            hur, NULL);
1841                 }
1842
1843                 OBD_FREE_LARGE(hur, totalsize);
1844
1845                 return rc;
1846         }
1847         case LL_IOC_HSM_PROGRESS: {
1848                 struct hsm_progress_kernel      hpk;
1849                 struct hsm_progress             hp;
1850
1851                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1852                         return -EFAULT;
1853
1854                 hpk.hpk_fid = hp.hp_fid;
1855                 hpk.hpk_cookie = hp.hp_cookie;
1856                 hpk.hpk_extent = hp.hp_extent;
1857                 hpk.hpk_flags = hp.hp_flags;
1858                 hpk.hpk_errval = hp.hp_errval;
1859                 hpk.hpk_data_version = 0;
1860
1861                 /* File may not exist in Lustre; all progress
1862                  * reported to Lustre root */
1863                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1864                                    NULL);
1865                 return rc;
1866         }
1867         case LL_IOC_HSM_CT_START:
1868                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1869                                     sizeof(struct lustre_kernelcomm));
1870                 return rc;
1871
1872         case LL_IOC_HSM_COPY_START: {
1873                 struct hsm_copy *copy;
1874                 int              rc;
1875
1876                 OBD_ALLOC_PTR(copy);
1877                 if (copy == NULL)
1878                         return -ENOMEM;
1879                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1880                         OBD_FREE_PTR(copy);
1881                         return -EFAULT;
1882                 }
1883
1884                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1885                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1886                         rc = -EFAULT;
1887
1888                 OBD_FREE_PTR(copy);
1889                 return rc;
1890         }
1891         case LL_IOC_HSM_COPY_END: {
1892                 struct hsm_copy *copy;
1893                 int              rc;
1894
1895                 OBD_ALLOC_PTR(copy);
1896                 if (copy == NULL)
1897                         return -ENOMEM;
1898                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1899                         OBD_FREE_PTR(copy);
1900                         return -EFAULT;
1901                 }
1902
1903                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1904                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1905                         rc = -EFAULT;
1906
1907                 OBD_FREE_PTR(copy);
1908                 return rc;
1909         }
1910         default:
1911                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1912         }
1913 }
1914
1915 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1916 {
1917         struct inode *inode = file->f_mapping->host;
1918         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1919         struct ll_sb_info *sbi = ll_i2sbi(inode);
1920         int api32 = ll_need_32bit_api(sbi);
1921         loff_t ret = -EINVAL;
1922
1923         mutex_lock(&inode->i_mutex);
1924         switch (origin) {
1925                 case SEEK_SET:
1926                         break;
1927                 case SEEK_CUR:
1928                         offset += file->f_pos;
1929                         break;
1930                 case SEEK_END:
1931                         if (offset > 0)
1932                                 GOTO(out, ret);
1933                         if (api32)
1934                                 offset += LL_DIR_END_OFF_32BIT;
1935                         else
1936                                 offset += LL_DIR_END_OFF;
1937                         break;
1938                 default:
1939                         GOTO(out, ret);
1940         }
1941
1942         if (offset >= 0 &&
1943             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1944              (!api32 && offset <= LL_DIR_END_OFF))) {
1945                 if (offset != file->f_pos) {
1946                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1947                             (!api32 && offset == LL_DIR_END_OFF))
1948                                 fd->lfd_pos = MDS_DIR_END_OFF;
1949                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1950                                 fd->lfd_pos = offset << 32;
1951                         else
1952                                 fd->lfd_pos = offset;
1953                         file->f_pos = offset;
1954                         file->f_version = 0;
1955                 }
1956                 ret = offset;
1957         }
1958         GOTO(out, ret);
1959
1960 out:
1961         mutex_unlock(&inode->i_mutex);
1962         return ret;
1963 }
1964
1965 static int ll_dir_open(struct inode *inode, struct file *file)
1966 {
1967         return ll_file_open(inode, file);
1968 }
1969
1970 static int ll_dir_release(struct inode *inode, struct file *file)
1971 {
1972         return ll_file_release(inode, file);
1973 }
1974
1975 const struct file_operations ll_dir_operations = {
1976         .llseek   = ll_dir_seek,
1977         .open     = ll_dir_open,
1978         .release  = ll_dir_release,
1979         .read     = generic_read_dir,
1980         .iterate  = ll_readdir,
1981         .unlocked_ioctl   = ll_dir_ioctl,
1982         .fsync    = ll_fsync,
1983 };