Whamcloud - gitweb
266ae99f78cec6c58e7a3e8c46cde4910e2f07fc
[fs/lustre-release.git] / 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  2008 Sun Microsystems, Inc. All rights reserved
30  * Use is subject to license terms.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/llite/dir.c
37  *
38  * Directory code for lustre client.
39  */
40
41 #include <linux/fs.h>
42 #include <linux/pagemap.h>
43 #include <linux/mm.h>
44 #include <linux/version.h>
45 #include <linux/smp_lock.h>
46 #include <asm/uaccess.h>
47 #include <linux/buffer_head.h>   // for wait_on_buffer
48
49 #define DEBUG_SUBSYSTEM S_LLITE
50
51 #include <obd_support.h>
52 #include <obd_class.h>
53 #include <lustre_lib.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_lite.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include "llite_internal.h"
59
60 #ifndef HAVE_PAGE_CHECKED
61 #ifdef HAVE_PG_FS_MISC
62 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
63 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
64 #else
65 #error PageChecked or PageFsMisc not defined in kernel
66 #endif
67 #endif
68
69 /*
70  * (new) readdir implementation overview.
71  *
72  * Original lustre readdir implementation cached exact copy of raw directory
73  * pages on the client. These pages were indexed in client page cache by
74  * logical offset in the directory file. This design, while very simple and
75  * intuitive had some inherent problems:
76  *
77  *     . it implies that byte offset to the directory entry serves as a
78  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
79  *     ext3/htree directory entries may move due to splits, and more
80  *     importantly,
81  *
82  *     . it is incompatible with the design of split directories for cmd3,
83  *     that assumes that names are distributed across nodes based on their
84  *     hash, and so readdir should be done in hash order.
85  *
86  * New readdir implementation does readdir in hash order, and uses hash of a
87  * file name as a telldir/seekdir cookie. This led to number of complications:
88  *
89  *     . hash is not unique, so it cannot be used to index cached directory
90  *     pages on the client (note, that it requires a whole pageful of hash
91  *     collided entries to cause two pages to have identical hashes);
92  *
93  *     . hash is not unique, so it cannot, strictly speaking, be used as an
94  *     entry cookie. ext3/htree has the same problem and lustre implementation
95  *     mimics their solution: seekdir(hash) positions directory at the first
96  *     entry with the given hash.
97  *
98  * Client side.
99  *
100  * 0. caching
101  *
102  * Client caches directory pages using hash of the first entry as an index. As
103  * noted above hash is not unique, so this solution doesn't work as is:
104  * special processing is needed for "page hash chains" (i.e., sequences of
105  * pages filled with entries all having the same hash value).
106  *
107  * First, such chains have to be detected. To this end, server returns to the
108  * client the hash of the first entry on the page next to one returned. When
109  * client detects that this hash is the same as hash of the first entry on the
110  * returned page, page hash collision has to be handled. Pages in the
111  * hash chain, except first one, are termed "overflow pages".
112  *
113  * Solution to index uniqueness problem is to not cache overflow
114  * pages. Instead, when page hash collision is detected, all overflow pages
115  * from emerging chain are immediately requested from the server and placed in
116  * a special data structure (struct ll_dir_chain). This data structure is used
117  * by ll_readdir() to process entries from overflow pages. When readdir
118  * invocation finishes, overflow pages are discarded. If page hash collision
119  * chain weren't completely processed, next call to readdir will again detect
120  * page hash collision, again read overflow pages in, process next portion of
121  * entries and again discard the pages. This is not as wasteful as it looks,
122  * because, given reasonable hash, page hash collisions are extremely rare.
123  *
124  * 1. directory positioning
125  *
126  * When seekdir(hash) is called, original
127  *
128  *
129  *
130  *
131  *
132  *
133  *
134  *
135  * Server.
136  *
137  * identification of and access to overflow pages
138  *
139  * page format
140  *
141  *
142  *
143  *
144  *
145  */
146
147 /* returns the page unlocked, but with a reference */
148 static int ll_dir_readpage(struct file *file, struct page *page)
149 {
150         struct inode *inode = page->mapping->host;
151         struct ptlrpc_request *request;
152         struct mdt_body *body;
153         struct obd_capa *oc;
154         __u64 hash;
155         int rc;
156         ENTRY;
157
158         hash = (__u64)hash_x_index(page->index);
159         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off %lu\n",
160                inode->i_ino, inode->i_generation, inode, (unsigned long)hash);
161
162         oc = ll_mdscapa_get(inode);
163         rc = md_readpage(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode),
164                          oc, hash, page, &request);
165         capa_put(oc);
166         if (!rc) {
167                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
168                 /* Checked by mdc_readpage() */
169                 LASSERT(body != NULL);
170
171                 if (body->valid & OBD_MD_FLSIZE)
172                         cl_isize_write(inode, body->size);
173                 SetPageUptodate(page);
174         }
175         ptlrpc_req_finished(request);
176
177         unlock_page(page);
178         EXIT;
179         return rc;
180 }
181
182 struct address_space_operations ll_dir_aops = {
183         .readpage  = ll_dir_readpage,
184 };
185
186 static void ll_check_page(struct inode *dir, struct page *page)
187 {
188         /* XXX: check page format later */
189         SetPageChecked(page);
190 }
191
192 static void ll_release_page(struct page *page, __u64 hash,
193                             __u64 start, __u64 end)
194 {
195         kunmap(page);
196         lock_page(page);
197         if (likely(page->mapping != NULL)) {
198                 ll_truncate_complete_page(page);
199                 unlock_page(page);
200         } else {
201                 unlock_page(page);
202                 CWARN("NULL mapping page %p, truncated by others: "
203                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
204                       page, hash, start, end);
205         }
206         page_cache_release(page);
207 }
208
209 /*
210  * Find, kmap and return page that contains given hash.
211  */
212 static struct page *ll_dir_page_locate(struct inode *dir, __u64 hash,
213                                        __u64 *start, __u64 *end)
214 {
215         struct address_space *mapping = dir->i_mapping;
216         /*
217          * Complement of hash is used as an index so that
218          * radix_tree_gang_lookup() can be used to find a page with starting
219          * hash _smaller_ than one we are looking for.
220          */
221         unsigned long offset = hash_x_index((unsigned long)hash);
222         struct page *page;
223         int found;
224
225         TREE_READ_LOCK_IRQ(mapping);
226         found = radix_tree_gang_lookup(&mapping->page_tree,
227                                        (void **)&page, offset, 1);
228         if (found > 0) {
229                 struct lu_dirpage *dp;
230
231                 page_cache_get(page);
232                 TREE_READ_UNLOCK_IRQ(mapping);
233                 /*
234                  * In contrast to find_lock_page() we are sure that directory
235                  * page cannot be truncated (while DLM lock is held) and,
236                  * hence, can avoid restart.
237                  *
238                  * In fact, page cannot be locked here at all, because
239                  * ll_dir_readpage() does synchronous io.
240                  */
241                 wait_on_page(page);
242                 if (PageUptodate(page)) {
243                         dp = kmap(page);
244                         *start = le64_to_cpu(dp->ldp_hash_start);
245                         *end   = le64_to_cpu(dp->ldp_hash_end);
246                         LASSERT(*start <= hash);
247                         if (hash > *end || (*end != *start && hash == *end)) {
248                                 ll_release_page(page, hash, *start, *end);
249                                 page = NULL;
250                         }
251                 } else {
252                         page_cache_release(page);
253                         page = ERR_PTR(-EIO);
254                 }
255
256         } else {
257                 TREE_READ_UNLOCK_IRQ(mapping);
258                 page = NULL;
259         }
260         return page;
261 }
262
263 struct page *ll_get_dir_page(struct inode *dir, __u64 hash, int exact,
264                              struct ll_dir_chain *chain)
265 {
266         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
267         struct address_space *mapping = dir->i_mapping;
268         struct lustre_handle lockh;
269         struct lu_dirpage *dp;
270         struct page *page;
271         ldlm_mode_t mode;
272         int rc;
273         __u64 start = 0;
274         __u64 end = 0;
275
276         mode = LCK_PR;
277         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
278                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
279         if (!rc) {
280                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
281                        ll_md_blocking_ast, ldlm_completion_ast,
282                        NULL, NULL, dir };
283                 struct lookup_intent it = { .it_op = IT_READDIR };
284                 struct ptlrpc_request *request;
285                 struct md_op_data *op_data;
286
287                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
288                                              LUSTRE_OPC_ANY, NULL);
289                 if (IS_ERR(op_data))
290                         return (void *)op_data;
291
292                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
293                                 op_data, &lockh, NULL, 0, NULL, 0);
294
295                 ll_finish_md_op_data(op_data);
296
297                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
298                 if (request)
299                         ptlrpc_req_finished(request);
300                 if (rc < 0) {
301                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
302                                PFID(ll_inode2fid(dir)), hash, rc);
303                         return ERR_PTR(rc);
304                 }
305         } else {
306                 /* for cross-ref object, l_ast_data of the lock may not be set,
307                  * we reset it here */
308                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
309                                  dir, NULL);
310         }
311         ldlm_lock_dump_handle(D_OTHER, &lockh);
312
313         page = ll_dir_page_locate(dir, hash, &start, &end);
314         if (IS_ERR(page)) {
315                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
316                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
317                 GOTO(out_unlock, page);
318         }
319
320         if (page != NULL) {
321                 /*
322                  * XXX nikita: not entirely correct handling of a corner case:
323                  * suppose hash chain of entries with hash value HASH crosses
324                  * border between pages P0 and P1. First both P0 and P1 are
325                  * cached, seekdir() is called for some entry from the P0 part
326                  * of the chain. Later P0 goes out of cache. telldir(HASH)
327                  * happens and finds P1, as it starts with matching hash
328                  * value. Remaining entries from P0 part of the chain are
329                  * skipped. (Is that really a bug?)
330                  *
331                  * Possible solutions: 0. don't cache P1 is such case, handle
332                  * it as an "overflow" page. 1. invalidate all pages at
333                  * once. 2. use HASH|1 as an index for P1.
334                  */
335                 if (exact && hash != start) {
336                         /*
337                          * readdir asked for a page starting _exactly_ from
338                          * given hash, but cache contains stale page, with
339                          * entries with smaller hash values. Stale page should
340                          * be invalidated, and new one fetched.
341                          */
342                         CDEBUG(D_OTHER, "Stale readpage page %p: "LPX64" != "LPX64"\n",
343                                page, hash, start);
344                         ll_release_page(page, hash, start, end);
345                 } else {
346                         GOTO(hash_collision, page);
347                 }
348         }
349
350         page = read_cache_page(mapping, hash_x_index((unsigned long)hash),
351                                (filler_t*)mapping->a_ops->readpage, NULL);
352         if (IS_ERR(page)) {
353                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
354                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
355                 GOTO(out_unlock, page);
356         }
357
358         wait_on_page(page);
359         (void)kmap(page);
360         if (!PageUptodate(page)) {
361                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
362                        PFID(ll_inode2fid(dir)), hash, -5);
363                 goto fail;
364         }
365         if (!PageChecked(page))
366                 ll_check_page(dir, page);
367         if (PageError(page)) {
368                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
369                        PFID(ll_inode2fid(dir)), hash, -5);
370                 goto fail;
371         }
372 hash_collision:
373         dp = page_address(page);
374
375         start = le64_to_cpu(dp->ldp_hash_start);
376         end   = le64_to_cpu(dp->ldp_hash_end);
377         if (end == start) {
378                 LASSERT(start == hash);
379                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
380                 /*
381                  * Fetch whole overflow chain...
382                  *
383                  * XXX not yet.
384                  */
385                 goto fail;
386         }
387 out_unlock:
388         ldlm_lock_decref(&lockh, mode);
389         return page;
390
391 fail:
392         ll_put_page(page);
393         page = ERR_PTR(-EIO);
394         goto out_unlock;
395 }
396
397 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
398 {
399         struct inode         *inode = filp->f_dentry->d_inode;
400         struct ll_inode_info *info  = ll_i2info(inode);
401         __u64                 pos   = filp->f_pos;
402         struct page          *page;
403         struct ll_dir_chain   chain;
404         int rc;
405         int done;
406         int shift;
407         __u16 type;
408         ENTRY;
409
410         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
411                inode->i_ino, inode->i_generation, inode,
412                (unsigned long)pos, i_size_read(inode));
413
414         if (pos == DIR_END_OFF)
415                 /*
416                  * end-of-file.
417                  */
418                 RETURN(0);
419
420         rc    = 0;
421         done  = 0;
422         shift = 0;
423         ll_dir_chain_init(&chain);
424
425         page = ll_get_dir_page(inode, pos, 0, &chain);
426
427         while (rc == 0 && !done) {
428                 struct lu_dirpage *dp;
429                 struct lu_dirent  *ent;
430
431                 if (!IS_ERR(page)) {
432                         /*
433                          * If page is empty (end of directory is reached),
434                          * use this value.
435                          */
436                         __u64 hash = DIR_END_OFF;
437                         __u64 next;
438
439                         dp = page_address(page);
440                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
441                              ent = lu_dirent_next(ent)) {
442                                 char          *name;
443                                 int            namelen;
444                                 struct lu_fid  fid;
445                                 ino_t          ino;
446
447                                 /*
448                                  * XXX: implement correct swabbing here.
449                                  */
450
451                                 hash    = le64_to_cpu(ent->lde_hash);
452                                 namelen = le16_to_cpu(ent->lde_namelen);
453
454                                 if (hash < pos)
455                                         /*
456                                          * Skip until we find target hash
457                                          * value.
458                                          */
459                                         continue;
460
461                                 if (namelen == 0)
462                                         /*
463                                          * Skip dummy record.
464                                          */
465                                         continue;
466
467                                 fid  = ent->lde_fid;
468                                 name = ent->lde_name;
469                                 fid_le_to_cpu(&fid, &fid);
470                                 ino  = cl_fid_build_ino(&fid);
471                                 type = ll_dirent_type_get(ent);
472                                 done = filldir(cookie, name, namelen,
473                                                (loff_t)hash, ino, type);
474                         }
475                         next = le64_to_cpu(dp->ldp_hash_end);
476                         ll_put_page(page);
477                         if (!done) {
478                                 pos = next;
479                                 if (pos == DIR_END_OFF)
480                                         /*
481                                          * End of directory reached.
482                                          */
483                                         done = 1;
484                                 else if (1 /* chain is exhausted*/)
485                                         /*
486                                          * Normal case: continue to the next
487                                          * page.
488                                          */
489                                         page = ll_get_dir_page(inode, pos, 1,
490                                                                &chain);
491                                 else {
492                                         /*
493                                          * go into overflow page.
494                                          */
495                                 }
496                         } else
497                                 pos = hash;
498                 } else {
499                         rc = PTR_ERR(page);
500                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
501                                PFID(&info->lli_fid), (unsigned long)pos, rc);
502                 }
503         }
504
505         filp->f_pos = (loff_t)pos;
506         filp->f_version = inode->i_version;
507         touch_atime(filp->f_vfsmnt, filp->f_dentry);
508
509         ll_dir_chain_fini(&chain);
510
511         RETURN(rc);
512 }
513
514 int ll_send_mgc_param(struct obd_export *mgc, char *string)
515 {
516         struct mgs_send_param *msp;
517         int rc = 0;
518
519         OBD_ALLOC_PTR(msp);
520         if (!msp)
521                 return -ENOMEM;
522
523         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
524         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
525                                 sizeof(struct mgs_send_param), msp, NULL);
526         if (rc)
527                 CERROR("Failed to set parameter: %d\n", rc);
528         OBD_FREE_PTR(msp);
529
530         return rc;
531 }
532
533 char *ll_get_fsname(struct inode *inode)
534 {
535         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
536         char *ptr, *fsname;
537         int len;
538
539         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
540         len = strlen(lsi->lsi_lmd->lmd_profile);
541         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
542         if (ptr && (strcmp(ptr, "-client") == 0))
543                 len -= 7;
544         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
545         fsname[len] = '\0';
546
547         return fsname;
548 }
549
550 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
551                      int set_default)
552 {
553         struct ll_sb_info *sbi = ll_i2sbi(inode);
554         struct md_op_data *op_data;
555         struct ptlrpc_request *req = NULL;
556         int rc = 0;
557         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
558         struct obd_device *mgc = lsi->lsi_mgc;
559         char *fsname = NULL, *param = NULL;
560         int lum_size;
561
562         if (lump != NULL) {
563                 /*
564                  * This is coming from userspace, so should be in
565                  * local endian.  But the MDS would like it in little
566                  * endian, so we swab it before we send it.
567                  */
568                 switch (lump->lmm_magic) {
569                 case LOV_USER_MAGIC_V1: {
570                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
571                                 lustre_swab_lov_user_md_v1(lump);
572                         lum_size = sizeof(struct lov_user_md_v1);
573                         break;
574                         }
575                 case LOV_USER_MAGIC_V3: {
576                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
577                                 lustre_swab_lov_user_md_v3(
578                                         (struct lov_user_md_v3 *)lump);
579                         lum_size = sizeof(struct lov_user_md_v3);
580                         break;
581                         }
582                 default: {
583                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
584                                         " %#08x != %#08x nor %#08x\n",
585                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
586                                         LOV_USER_MAGIC_V3);
587                         RETURN(-EINVAL);
588                         }
589                }
590         } else {
591                 lum_size = sizeof(struct lov_user_md_v1);
592         }
593
594         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
595                                      LUSTRE_OPC_ANY, NULL);
596         if (IS_ERR(op_data))
597                 RETURN(PTR_ERR(op_data));
598
599         /* swabbing is done in lov_setstripe() on server side */
600         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
601                         NULL, 0, &req, NULL);
602         ll_finish_md_op_data(op_data);
603         ptlrpc_req_finished(req);
604         if (rc) {
605                 if (rc != -EPERM && rc != -EACCES)
606                         CERROR("mdc_setattr fails: rc = %d\n", rc);
607         }
608
609         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
610          LOV_USER_MAGIC_V3 have the same initial fields so we do not
611          need the make the distiction between the 2 versions */
612         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
613                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
614
615                 /* Get fsname and assume devname to be -MDT0000. */
616                 fsname = ll_get_fsname(inode);
617                 /* Set root stripesize */
618                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
619                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
620                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
621                 if (rc)
622                         goto end;
623
624                 /* Set root stripecount */
625                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
626                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
627                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
628                 if (rc)
629                         goto end;
630
631                 /* Set root stripeoffset */
632                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
633                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
634                         (typeof(lump->lmm_stripe_offset))(-1));
635                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
636                 if (rc)
637                         goto end;
638 end:
639                 if (fsname)
640                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
641                 if (param)
642                         OBD_FREE(param, MGS_PARAM_MAXLEN);
643         }
644         return rc;
645 }
646
647 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
648                      int *lmm_size, struct ptlrpc_request **request)
649 {
650         struct ll_sb_info *sbi = ll_i2sbi(inode);
651         struct mdt_body   *body;
652         struct lov_mds_md *lmm = NULL;
653         struct ptlrpc_request *req = NULL;
654         int rc, lmmsize;
655         struct md_op_data *op_data;
656
657         rc = ll_get_max_mdsize(sbi, &lmmsize);
658         if (rc)
659                 RETURN(rc);
660
661         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
662                                      0, lmmsize, LUSTRE_OPC_ANY,
663                                      NULL);
664         if (op_data == NULL)
665                 RETURN(-ENOMEM);
666
667         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
668         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
669         ll_finish_md_op_data(op_data);
670         if (rc < 0) {
671                 CDEBUG(D_INFO, "md_getattr failed on inode "
672                        "%lu/%u: rc %d\n", inode->i_ino,
673                        inode->i_generation, rc);
674                 GOTO(out, rc);
675         }
676
677         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
678         LASSERT(body != NULL);
679
680         lmmsize = body->eadatasize;
681
682         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
683             lmmsize == 0) {
684                 GOTO(out, rc = -ENODATA);
685         }
686
687         lmm = req_capsule_server_sized_get(&req->rq_pill,
688                                            &RMF_MDT_MD, lmmsize);
689         LASSERT(lmm != NULL);
690
691         /*
692          * This is coming from the MDS, so is probably in
693          * little endian.  We convert it to host endian before
694          * passing it to userspace.
695          */
696         /* We don't swab objects for directories */
697         switch (le32_to_cpu(lmm->lmm_magic)) {
698         case LOV_MAGIC_V1:
699                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
700                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
701                 break;
702         case LOV_MAGIC_V3:
703                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
704                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
705                 break;
706         default:
707                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
708                 rc = -EPROTO;
709         }
710 out:
711         *lmmp = lmm;
712         *lmm_size = lmmsize;
713         *request = req;
714         return rc;
715 }
716
717 /*
718  *  Get MDT index for the inode.
719  */
720 int ll_get_mdt_idx(struct inode *inode)
721 {
722         struct ll_sb_info *sbi = ll_i2sbi(inode);
723         struct md_op_data *op_data;
724         int rc, mdtidx;
725         ENTRY;
726
727         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
728                                      0, LUSTRE_OPC_ANY, NULL);
729         if (op_data == NULL)
730                 RETURN(-ENOMEM);
731
732         op_data->op_valid |= OBD_MD_MDTIDX;
733         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
734         mdtidx = op_data->op_mds;
735         ll_finish_md_op_data(op_data);
736         if (rc < 0) {
737                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
738                 RETURN(rc);
739         }
740         return mdtidx;
741 }
742
743 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
744 {
745         void *ptr;
746         int rc;
747
748         OBD_ALLOC(ptr, len);
749         if (ptr == NULL)
750                 return -ENOMEM;
751         if (cfs_copy_from_user(ptr, data, len)) {
752                 OBD_FREE(ptr, len);
753                 return -EFAULT;
754         }
755         rc = obd_iocontrol(cmd, exp, len, data, NULL);
756         OBD_FREE(ptr, len);
757         return rc;
758 }
759
760 static int ll_dir_ioctl(struct inode *inode, struct file *file,
761                         unsigned int cmd, unsigned long arg)
762 {
763         struct ll_sb_info *sbi = ll_i2sbi(inode);
764         struct obd_ioctl_data *data;
765         int rc = 0;
766         ENTRY;
767
768         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
769                inode->i_ino, inode->i_generation, inode, cmd);
770
771         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
772         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
773                 return -ENOTTY;
774
775         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
776         switch(cmd) {
777         case FSFILT_IOC_GETFLAGS:
778         case FSFILT_IOC_SETFLAGS:
779                 RETURN(ll_iocontrol(inode, file, cmd, arg));
780         case FSFILT_IOC_GETVERSION_OLD:
781         case FSFILT_IOC_GETVERSION:
782                 RETURN(put_user(inode->i_generation, (int *)arg));
783         /* We need to special case any other ioctls we want to handle,
784          * to send them to the MDS/OST as appropriate and to properly
785          * network encode the arg field.
786         case FSFILT_IOC_SETVERSION_OLD:
787         case FSFILT_IOC_SETVERSION:
788         */
789         case LL_IOC_GET_MDTIDX: {
790                 int mdtidx;
791
792                 mdtidx = ll_get_mdt_idx(inode);
793                 if (mdtidx < 0)
794                         RETURN(mdtidx);
795
796                 if (put_user((int)mdtidx, (int*)arg))
797                         RETURN(-EFAULT);
798
799                 return 0;
800         }
801         case IOC_MDC_LOOKUP: {
802                 struct ptlrpc_request *request = NULL;
803                 int namelen, len = 0;
804                 char *buf = NULL;
805                 char *filename;
806                 struct md_op_data *op_data;
807
808                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
809                 if (rc)
810                         RETURN(rc);
811                 data = (void *)buf;
812
813                 filename = data->ioc_inlbuf1;
814                 namelen = strlen(filename);
815
816                 if (namelen < 1) {
817                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
818                         GOTO(out_free, rc = -EINVAL);
819                 }
820
821                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
822                                              0, LUSTRE_OPC_ANY, NULL);
823                 if (op_data == NULL)
824                         GOTO(out_free, rc = -ENOMEM);
825
826                 op_data->op_valid = OBD_MD_FLID;
827                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
828                 ll_finish_md_op_data(op_data);
829                 if (rc < 0) {
830                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
831                         GOTO(out_free, rc);
832                 }
833                 ptlrpc_req_finished(request);
834                 EXIT;
835 out_free:
836                 obd_ioctl_freedata(buf, len);
837                 return rc;
838         }
839         case LL_IOC_LOV_SETSTRIPE: {
840                 struct lov_user_md_v3 lumv3;
841                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
842                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
843                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
844
845                 int set_default = 0;
846
847                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
848                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
849                         sizeof(lumv3p->lmm_objects[0]));
850                 /* first try with v1 which is smaller than v3 */
851                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
852                         RETURN(-EFAULT);
853
854                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
855                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
856                                 RETURN(-EFAULT);
857                 }
858
859                 if (inode->i_sb->s_root == file->f_dentry)
860                         set_default = 1;
861
862                 /* in v1 and v3 cases lumv1 points to data */
863                 rc = ll_dir_setstripe(inode, lumv1, set_default);
864
865                 RETURN(rc);
866         }
867         case LL_IOC_OBD_STATFS:
868                 RETURN(ll_obd_statfs(inode, (void *)arg));
869         case LL_IOC_LOV_GETSTRIPE:
870         case LL_IOC_MDC_GETINFO:
871         case IOC_MDC_GETFILEINFO:
872         case IOC_MDC_GETFILESTRIPE: {
873                 struct ptlrpc_request *request = NULL;
874                 struct lov_user_md *lump;
875                 struct lov_mds_md *lmm = NULL;
876                 struct mdt_body *body;
877                 char *filename = NULL;
878                 int lmmsize;
879
880                 if (cmd == IOC_MDC_GETFILEINFO ||
881                     cmd == IOC_MDC_GETFILESTRIPE) {
882                         filename = getname((const char *)arg);
883                         if (IS_ERR(filename))
884                                 RETURN(PTR_ERR(filename));
885
886                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
887                                                       &lmmsize, &request);
888                 } else {
889                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
890                 }
891
892                 if (request) {
893                         body = req_capsule_server_get(&request->rq_pill,
894                                                       &RMF_MDT_BODY);
895                         LASSERT(body != NULL);
896                 } else {
897                         GOTO(out_req, rc);
898                 }
899
900                 if (rc < 0) {
901                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
902                                                cmd == LL_IOC_MDC_GETINFO))
903                                 GOTO(skip_lmm, rc = 0);
904                         else
905                                 GOTO(out_req, rc);
906                 }
907
908                 if (cmd == IOC_MDC_GETFILESTRIPE ||
909                     cmd == LL_IOC_LOV_GETSTRIPE) {
910                         lump = (struct lov_user_md *)arg;
911                 } else {
912                         struct lov_user_mds_data *lmdp;
913                         lmdp = (struct lov_user_mds_data *)arg;
914                         lump = &lmdp->lmd_lmm;
915                 }
916                 if (cfs_copy_to_user(lump, lmm, lmmsize))
917                         GOTO(out_req, rc = -EFAULT);
918         skip_lmm:
919                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
920                         struct lov_user_mds_data *lmdp;
921                         lstat_t st = { 0 };
922
923                         st.st_dev     = inode->i_sb->s_dev;
924                         st.st_mode    = body->mode;
925                         st.st_nlink   = body->nlink;
926                         st.st_uid     = body->uid;
927                         st.st_gid     = body->gid;
928                         st.st_rdev    = body->rdev;
929                         st.st_size    = body->size;
930                         st.st_blksize = CFS_PAGE_SIZE;
931                         st.st_blocks  = body->blocks;
932                         st.st_atime   = body->atime;
933                         st.st_mtime   = body->mtime;
934                         st.st_ctime   = body->ctime;
935                         st.st_ino     = inode->i_ino;
936
937                         lmdp = (struct lov_user_mds_data *)arg;
938                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
939                                 GOTO(out_req, rc = -EFAULT);
940                 }
941
942                 EXIT;
943         out_req:
944                 ptlrpc_req_finished(request);
945                 if (filename)
946                         putname(filename);
947                 return rc;
948         }
949         case IOC_LOV_GETINFO: {
950                 struct lov_user_mds_data *lumd;
951                 struct lov_stripe_md *lsm;
952                 struct lov_user_md *lum;
953                 struct lov_mds_md *lmm;
954                 int lmmsize;
955                 lstat_t st;
956
957                 lumd = (struct lov_user_mds_data *)arg;
958                 lum = &lumd->lmd_lmm;
959
960                 rc = ll_get_max_mdsize(sbi, &lmmsize);
961                 if (rc)
962                         RETURN(rc);
963
964                 OBD_ALLOC(lmm, lmmsize);
965                 if (cfs_copy_from_user(lmm, lum, lmmsize))
966                         GOTO(free_lmm, rc = -EFAULT);
967
968                 switch (lmm->lmm_magic) {
969                 case LOV_USER_MAGIC_V1:
970                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
971                                 break;
972                         /* swab objects first so that stripes num will be sane */
973                         lustre_swab_lov_user_md_objects(
974                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
975                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
976                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
977                         break;
978                 case LOV_USER_MAGIC_V3:
979                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
980                                 break;
981                         /* swab objects first so that stripes num will be sane */
982                         lustre_swab_lov_user_md_objects(
983                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
984                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
985                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
986                         break;
987                 default:
988                         GOTO(free_lmm, rc = -EINVAL);
989                 }
990
991                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
992                 if (rc < 0)
993                         GOTO(free_lmm, rc = -ENOMEM);
994
995                 /* Perform glimpse_size operation. */
996                 memset(&st, 0, sizeof(st));
997
998                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
999                 if (rc)
1000                         GOTO(free_lsm, rc);
1001
1002                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1003                         GOTO(free_lsm, rc = -EFAULT);
1004
1005                 EXIT;
1006         free_lsm:
1007                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1008         free_lmm:
1009                 OBD_FREE(lmm, lmmsize);
1010                 return rc;
1011         }
1012         case OBD_IOC_LLOG_CATINFO: {
1013                 struct ptlrpc_request *req = NULL;
1014                 char                  *buf = NULL;
1015                 char                  *str;
1016                 int                    len = 0;
1017
1018                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1019                 if (rc)
1020                         RETURN(rc);
1021                 data = (void *)buf;
1022
1023                 if (!data->ioc_inlbuf1) {
1024                         obd_ioctl_freedata(buf, len);
1025                         RETURN(-EINVAL);
1026                 }
1027
1028                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1029                                            &RQF_LLOG_CATINFO);
1030                 if (req == NULL)
1031                         GOTO(out_catinfo, rc = -ENOMEM);
1032
1033                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1034                                      data->ioc_inllen1);
1035                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1036                                      data->ioc_inllen2);
1037
1038                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1039                 if (rc) {
1040                         ptlrpc_request_free(req);
1041                         GOTO(out_catinfo, rc);
1042                 }
1043
1044                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1045                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1046                 if (data->ioc_inllen2) {
1047                         str = req_capsule_client_get(&req->rq_pill,
1048                                                      &RMF_STRING);
1049                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1050                 }
1051
1052                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1053                                      data->ioc_plen1);
1054                 ptlrpc_request_set_replen(req);
1055
1056                 rc = ptlrpc_queue_wait(req);
1057                 if (!rc) {
1058                         str = req_capsule_server_get(&req->rq_pill,
1059                                                      &RMF_STRING);
1060                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1061                                              data->ioc_plen1))
1062                                 rc = -EFAULT;
1063                 }
1064                 ptlrpc_req_finished(req);
1065         out_catinfo:
1066                 obd_ioctl_freedata(buf, len);
1067                 RETURN(rc);
1068         }
1069         case OBD_IOC_QUOTACHECK: {
1070                 struct obd_quotactl *oqctl;
1071                 int error = 0;
1072
1073                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1074                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1075                         RETURN(-EPERM);
1076
1077                 OBD_ALLOC_PTR(oqctl);
1078                 if (!oqctl)
1079                         RETURN(-ENOMEM);
1080                 oqctl->qc_type = arg;
1081                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1082                 if (rc < 0) {
1083                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1084                         error = rc;
1085                 }
1086
1087                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1088                 if (rc < 0)
1089                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1090
1091                 OBD_FREE_PTR(oqctl);
1092                 return error ?: rc;
1093         }
1094         case OBD_IOC_POLL_QUOTACHECK: {
1095                 struct if_quotacheck *check;
1096
1097                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1098                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1099                         RETURN(-EPERM);
1100
1101                 OBD_ALLOC_PTR(check);
1102                 if (!check)
1103                         RETURN(-ENOMEM);
1104
1105                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1106                                    NULL);
1107                 if (rc) {
1108                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1109                         if (cfs_copy_to_user((void *)arg, check,
1110                                              sizeof(*check)))
1111                                 rc = -EFAULT;
1112                         GOTO(out_poll, rc);
1113                 }
1114
1115                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1116                                    NULL);
1117                 if (rc) {
1118                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1119                         if (cfs_copy_to_user((void *)arg, check,
1120                                              sizeof(*check)))
1121                                 rc = -EFAULT;
1122                         GOTO(out_poll, rc);
1123                 }
1124         out_poll:
1125                 OBD_FREE_PTR(check);
1126                 RETURN(rc);
1127         }
1128         case OBD_IOC_QUOTACTL: {
1129                 struct if_quotactl *qctl;
1130                 int cmd, type, id, valid;
1131
1132                 OBD_ALLOC_PTR(qctl);
1133                 if (!qctl)
1134                         RETURN(-ENOMEM);
1135
1136                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1137                         GOTO(out_quotactl, rc = -EFAULT);
1138
1139                 cmd = qctl->qc_cmd;
1140                 type = qctl->qc_type;
1141                 id = qctl->qc_id;
1142                 valid = qctl->qc_valid;
1143
1144                 switch (cmd) {
1145                 case LUSTRE_Q_INVALIDATE:
1146                 case LUSTRE_Q_FINVALIDATE:
1147                 case Q_QUOTAON:
1148                 case Q_QUOTAOFF:
1149                 case Q_SETQUOTA:
1150                 case Q_SETINFO:
1151                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1152                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1153                                 GOTO(out_quotactl, rc = -EPERM);
1154                         break;
1155                 case Q_GETQUOTA:
1156                         if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1157                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1158                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1159                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1160                                 GOTO(out_quotactl, rc = -EPERM);
1161                         break;
1162                 case Q_GETINFO:
1163                         break;
1164                 default:
1165                         CERROR("unsupported quotactl op: %#x\n", cmd);
1166                         GOTO(out_quotactl, rc = -ENOTTY);
1167                 }
1168
1169                 if (valid != QC_GENERAL) {
1170                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1171                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1172
1173                         if (cmd == Q_GETINFO)
1174                                 qctl->qc_cmd = Q_GETOINFO;
1175                         else if (cmd == Q_GETQUOTA)
1176                                 qctl->qc_cmd = Q_GETOQUOTA;
1177                         else
1178                                 GOTO(out_quotactl, rc = -EINVAL);
1179
1180                         switch (valid) {
1181                         case QC_MDTIDX:
1182                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1183                                                    sbi->ll_md_exp,
1184                                                    sizeof(*qctl), qctl, NULL);
1185                                 break;
1186                         case QC_OSTIDX:
1187                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1188                                                    sbi->ll_dt_exp,
1189                                                    sizeof(*qctl), qctl, NULL);
1190                                 break;
1191                         case QC_UUID:
1192                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1193                                                    sbi->ll_md_exp,
1194                                                    sizeof(*qctl), qctl, NULL);
1195                                 if (rc == -EAGAIN)
1196                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1197                                                            sbi->ll_dt_exp,
1198                                                            sizeof(*qctl), qctl,
1199                                                            NULL);
1200                                 break;
1201                         default:
1202                                 rc = -EINVAL;
1203                                 break;
1204                         }
1205
1206                         if (rc)
1207                                 GOTO(out_quotactl, rc);
1208                         else
1209                                 qctl->qc_cmd = cmd;
1210                 } else {
1211                         struct obd_quotactl *oqctl;
1212
1213                         OBD_ALLOC_PTR(oqctl);
1214                         if (!oqctl)
1215                                 GOTO(out_quotactl, rc = -ENOMEM);
1216
1217                         QCTL_COPY(oqctl, qctl);
1218                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1219                         if (rc) {
1220                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1221                                         oqctl->qc_cmd = Q_QUOTAOFF;
1222                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1223                                 }
1224                                 OBD_FREE_PTR(oqctl);
1225                                 GOTO(out_quotactl, rc);
1226                         } else {
1227                                 QCTL_COPY(qctl, oqctl);
1228                                 OBD_FREE_PTR(oqctl);
1229                         }
1230                 }
1231
1232                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1233                         rc = -EFAULT;
1234
1235         out_quotactl:
1236                 OBD_FREE_PTR(qctl);
1237                 RETURN(rc);
1238         }
1239         case OBD_IOC_GETNAME: {
1240                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1241                 if (!obd)
1242                         RETURN(-EFAULT);
1243                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1244                                      strlen(obd->obd_name) + 1))
1245                         RETURN (-EFAULT);
1246                 RETURN(0);
1247         }
1248         case LL_IOC_FLUSHCTX:
1249                 RETURN(ll_flush_ctx(inode));
1250 #ifdef CONFIG_FS_POSIX_ACL
1251         case LL_IOC_RMTACL: {
1252             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1253                 inode == inode->i_sb->s_root->d_inode) {
1254                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1255
1256                 LASSERT(fd != NULL);
1257                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1258                 if (!rc)
1259                         fd->fd_flags |= LL_FILE_RMTACL;
1260                 RETURN(rc);
1261             } else
1262                 RETURN(0);
1263         }
1264 #endif
1265         case LL_IOC_GETOBDCOUNT: {
1266                 int count;
1267
1268                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1269                         RETURN(-EFAULT);
1270
1271                 if (!count) {
1272                         /* get ost count */
1273                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1274                         count = lov->desc.ld_tgt_count;
1275                 } else {
1276                         /* get mdt count */
1277                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1278                         count = lmv->desc.ld_tgt_count;
1279                 }
1280
1281                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1282                         RETURN(-EFAULT);
1283
1284                 RETURN(0);
1285         }
1286         case LL_IOC_PATH2FID:
1287                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1288                                      sizeof(struct lu_fid)))
1289                         RETURN(-EFAULT);
1290                 RETURN(0);
1291         case OBD_IOC_CHANGELOG_SEND:
1292         case OBD_IOC_CHANGELOG_CLEAR:
1293                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1294                                     sizeof(struct ioc_changelog));
1295                 RETURN(rc);
1296         case OBD_IOC_FID2PATH:
1297                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1298         case LL_IOC_HSM_CT_START:
1299                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1300                                     sizeof(struct lustre_kernelcomm));
1301                 RETURN(rc);
1302
1303         default:
1304                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1305         }
1306 }
1307
1308 int ll_dir_open(struct inode *inode, struct file *file)
1309 {
1310         ENTRY;
1311         RETURN(ll_file_open(inode, file));
1312 }
1313
1314 int ll_dir_release(struct inode *inode, struct file *file)
1315 {
1316         ENTRY;
1317         RETURN(ll_file_release(inode, file));
1318 }
1319
1320 struct file_operations ll_dir_operations = {
1321         .open     = ll_dir_open,
1322         .release  = ll_dir_release,
1323         .read     = generic_read_dir,
1324         .readdir  = ll_readdir,
1325         .ioctl    = ll_dir_ioctl
1326 };