Whamcloud - gitweb
ca9cfd17594331e680d2a6d763f2f65d2f8a2689
[fs/lustre-release.git] / lustre / llite / llite_lib.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 [sun.com URL with a
20  * copy of GPLv2].
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/llite_lib.c
37  *
38  * Lustre Light Super operations
39  */
40
41 #define DEBUG_SUBSYSTEM S_LLITE
42
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/random.h>
46 #include <linux/version.h>
47 #include <linux/mm.h>
48
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <obd_cksum.h>
57 #include <lustre_cache.h>
58 #include "llite_internal.h"
59
60 cfs_mem_cache_t *ll_file_data_slab;
61
62 LIST_HEAD(ll_super_blocks);
63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
64
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
67
68 #ifndef log2
69 #define log2(n) ffz(~(n))
70 #endif
71
72
73 static struct ll_sb_info *ll_init_sbi(void)
74 {
75         struct ll_sb_info *sbi = NULL;
76         unsigned long pages;
77         struct sysinfo si;
78         class_uuid_t uuid;
79         int i;
80         ENTRY;
81
82         OBD_ALLOC(sbi, sizeof(*sbi));
83         if (!sbi)
84                 RETURN(NULL);
85
86         spin_lock_init(&sbi->ll_lock);
87         spin_lock_init(&sbi->ll_lco.lco_lock);
88         spin_lock_init(&sbi->ll_pp_extent_lock);
89         spin_lock_init(&sbi->ll_process_lock);
90         sbi->ll_rw_stats_on = 0;
91         INIT_LIST_HEAD(&sbi->ll_pglist);
92
93         si_meminfo(&si);
94         pages = si.totalram - si.totalhigh;
95         if (pages >> (20 - CFS_PAGE_SHIFT) < 512)
96                 sbi->ll_async_page_max = pages / 2;
97         else
98                 sbi->ll_async_page_max = (pages / 4) * 3;
99         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
100                                            SBI_DEFAULT_READAHEAD_MAX);
101         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
102                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
103         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
104         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
105         INIT_LIST_HEAD(&sbi->ll_conn_chain);
106         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
107
108         ll_generate_random_uuid(uuid);
109         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
110         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
111
112         spin_lock(&ll_sb_lock);
113         list_add_tail(&sbi->ll_list, &ll_super_blocks);
114         spin_unlock(&ll_sb_lock);
115
116 #ifdef ENABLE_LLITE_CHECKSUM
117         sbi->ll_flags |= LL_SBI_CHECKSUM;
118 #endif
119
120 #ifdef HAVE_LRU_RESIZE_SUPPORT
121         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
122 #endif
123
124 #ifdef HAVE_EXPORT___IGET
125         INIT_LIST_HEAD(&sbi->ll_deathrow);
126         spin_lock_init(&sbi->ll_deathrow_lock);
127 #endif
128         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
129                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
130                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
131         }
132
133         /* metadata statahead is enabled by default */
134         sbi->ll_sa_max = LL_SA_RPC_DEF;
135
136         RETURN(sbi);
137 }
138
139 void ll_free_sbi(struct super_block *sb)
140 {
141         struct ll_sb_info *sbi = ll_s2sbi(sb);
142         ENTRY;
143
144         if (sbi != NULL) {
145                 spin_lock(&ll_sb_lock);
146                 list_del(&sbi->ll_list);
147                 spin_unlock(&ll_sb_lock);
148                 OBD_FREE(sbi, sizeof(*sbi));
149         }
150         EXIT;
151 }
152
153 static struct dentry_operations ll_d_root_ops = {
154 #ifdef DCACHE_LUSTRE_INVALID
155         .d_compare = ll_dcompare,
156 #endif
157 };
158
159 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
160  * us to make MDS RPCs with large enough reply buffers to hold the
161  * maximum-sized (= maximum striped) EA and cookie without having to
162  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
163 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
164 {
165         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
166         __u32 valsize = sizeof(struct lov_desc);
167         int rc, easize, def_easize, cookiesize;
168         struct lov_desc desc;
169         __u32 stripes;
170         ENTRY;
171
172         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
173                           &valsize, &desc);
174         if (rc)
175                 RETURN(rc);
176
177         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
178         lsm.lsm_stripe_count = stripes;
179         easize = obd_size_diskmd(dt_exp, &lsm);
180
181         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
182         def_easize = obd_size_diskmd(dt_exp, &lsm);
183
184         cookiesize = stripes * sizeof(struct llog_cookie);
185
186         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
187                easize, cookiesize);
188
189         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
190         RETURN(rc);
191 }
192
193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
194 {
195         struct inode *root = 0;
196         struct ll_sb_info *sbi = ll_s2sbi(sb);
197         struct obd_device *obd;
198         struct lu_fid rootfid;
199         struct obd_capa *oc = NULL;
200         struct obd_statfs osfs;
201         struct ptlrpc_request *request = NULL;
202         struct lustre_handle dt_conn = {0, };
203         struct lustre_handle md_conn = {0, };
204         struct obd_connect_data *data = NULL;
205         struct lustre_md lmd;
206         obd_valid valid;
207         int size, err, checksum;
208         ENTRY;
209
210         obd = class_name2obd(md);
211         if (!obd) {
212                 CERROR("MD %s: not setup or attached\n", md);
213                 RETURN(-EINVAL);
214         }
215
216         OBD_ALLOC_PTR(data);
217         if (data == NULL)
218                 RETURN(-ENOMEM);
219
220         if (proc_lustre_fs_root) {
221                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
222                                                   dt, md);
223                 if (err < 0)
224                         CERROR("could not register mount in /proc/fs/lustre\n");
225         }
226
227         /* indicate the features supported by this client */
228         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
229                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
230                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
231                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
232                                   OBD_CONNECT_FID      | OBD_CONNECT_AT;
233
234 #ifdef HAVE_LRU_RESIZE_SUPPORT
235         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
236                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
237 #endif
238 #ifdef CONFIG_FS_POSIX_ACL
239         data->ocd_connect_flags |= OBD_CONNECT_ACL;
240 #endif
241         data->ocd_ibits_known = MDS_INODELOCK_FULL;
242         data->ocd_version = LUSTRE_VERSION_CODE;
243
244         if (sb->s_flags & MS_RDONLY)
245                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
246         if (sbi->ll_flags & LL_SBI_USER_XATTR)
247                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
248
249 #ifdef HAVE_MS_FLOCK_LOCK
250         /* force vfs to use lustre handler for flock() calls - bug 10743 */
251         sb->s_flags |= MS_FLOCK_LOCK;
252 #endif
253         
254         if (sbi->ll_flags & LL_SBI_FLOCK)
255                 sbi->ll_fop = &ll_file_operations_flock;
256         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
257                 sbi->ll_fop = &ll_file_operations;
258         else
259                 sbi->ll_fop = &ll_file_operations_noflock;
260
261         /* real client */
262         data->ocd_connect_flags |= OBD_CONNECT_REAL;
263         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
264                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
265                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
266         } else {
267                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
268                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
269         }
270
271         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
272         if (err == -EBUSY) {
273                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
274                                    "recovery, of which this client is not a "
275                                    "part. Please wait for recovery to complete,"
276                                    " abort, or time out.\n", md);
277                 GOTO(out, err);
278         } else if (err) {
279                 CERROR("cannot connect to %s: rc = %d\n", md, err);
280                 GOTO(out, err);
281         }
282         sbi->ll_md_exp = class_conn2export(&md_conn);
283
284         err = obd_fid_init(sbi->ll_md_exp);
285         if (err) {
286                 CERROR("Can't init metadata layer FID infrastructure, "
287                        "rc %d\n", err);
288                 GOTO(out_md, err);
289         }
290
291         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
292         if (err)
293                 GOTO(out_md_fid, err);
294
295         size = sizeof(*data);
296         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
297                            KEY_CONN_DATA,  &size, data);
298         if (err) {
299                 CERROR("Get connect data failed: %d \n", err);
300                 GOTO(out_md, err);
301         }
302
303         LASSERT(osfs.os_bsize);
304         sb->s_blocksize = osfs.os_bsize;
305         sb->s_blocksize_bits = log2(osfs.os_bsize);
306         sb->s_magic = LL_SUPER_MAGIC;
307
308         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
309          *         retval = in_file->f_op->sendfile(...);
310          *         if (*ppos > max)
311          *                 retval = -EOVERFLOW;
312          *
313          * it will check if *ppos is greater than max. However, max equals to
314          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
315          * has been defined as a signed long long ineger in linux kernel. */
316 #if BITS_PER_LONG == 64
317         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
318 #else
319         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
320 #endif
321         sbi->ll_namelen = osfs.os_namelen;
322         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
323
324         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
325             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
326                 LCONSOLE_INFO("Disabling user_xattr feature because "
327                               "it is not supported on the server\n");
328                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
329         }
330
331         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
332 #ifdef MS_POSIXACL
333                 sb->s_flags |= MS_POSIXACL;
334 #endif
335                 sbi->ll_flags |= LL_SBI_ACL;
336         } else {
337                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
338 #ifdef MS_POSIXACL
339                 sb->s_flags &= ~MS_POSIXACL;
340 #endif
341                 sbi->ll_flags &= ~LL_SBI_ACL;
342         }
343
344         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
345                 sbi->ll_flags |= LL_SBI_JOIN;
346
347         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
348                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
349                         /* sometimes local client claims to be remote, but mdt
350                          * will disagree when client gss not applied. */
351                         LCONSOLE_INFO("client claims to be remote, but server "
352                                       "rejected, forced to be local.\n");
353                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
354                 }
355         } else {
356                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
357                         /* with gss applied, remote client can not claim to be
358                          * local, so mdt maybe force client to be remote. */
359                         LCONSOLE_INFO("client claims to be local, but server "
360                                       "rejected, forced to be remote.\n");
361                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
362                 }
363         }
364
365         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
366                 LCONSOLE_INFO("client enabled MDS capability!\n");
367                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
368         }
369
370         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
371                 LCONSOLE_INFO("client enabled OSS capability!\n");
372                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
373         }
374
375         sbi->ll_sdev_orig = sb->s_dev;
376 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
377         /* We set sb->s_dev equal on all lustre clients in order to support
378          * NFS export clustering.  NFSD requires that the FSID be the same
379          * on all clients. */
380         /* s_dev is also used in lt_compare() to compare two fs, but that is
381          * only a node-local comparison. */
382
383         /* XXX: this will not work with LMV */
384         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
385                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
386 #endif
387
388         obd = class_name2obd(dt);
389         if (!obd) {
390                 CERROR("DT %s: not setup or attached\n", dt);
391                 GOTO(out_md_fid, err = -ENODEV);
392         }
393
394         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
395                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
396                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
397                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
398                                   OBD_CONNECT_AT;
399         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
400                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
401
402         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
403                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
404                  * disabled by default, because it can still be enabled on the
405                  * fly via /proc. As a consequence, we still need to come to an
406                  * agreement on the supported algorithms at connect time */
407                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
408
409                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
410                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
411                 else
412                         /* send the list of supported checksum types */
413                         data->ocd_cksum_types = OBD_CKSUM_ALL;
414         }
415
416 #ifdef HAVE_LRU_RESIZE_SUPPORT
417         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
418 #endif
419         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
420                "ocd_grant: %d\n", data->ocd_connect_flags,
421                data->ocd_version, data->ocd_grant);
422
423         obd->obd_upcall.onu_owner = &sbi->ll_lco;
424         obd->obd_upcall.onu_upcall = ll_ocd_update;
425         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
426
427         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
428         if (err == -EBUSY) {
429                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
430                                    "recovery, of which this client is not a "
431                                    "part.  Please wait for recovery to "
432                                    "complete, abort, or time out.\n", dt);
433                 GOTO(out_md_fid, err);
434         } else if (err) {
435                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
436                 GOTO(out_md_fid, err);
437         }
438
439         sbi->ll_dt_exp = class_conn2export(&dt_conn);
440
441         err = obd_fid_init(sbi->ll_dt_exp);
442         if (err) {
443                 CERROR("Can't init data layer FID infrastructure, "
444                        "rc %d\n", err);
445                 GOTO(out_dt, err);
446         }
447         
448         spin_lock(&sbi->ll_lco.lco_lock);
449         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
450         spin_unlock(&sbi->ll_lco.lco_lock);
451
452         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
453                                            ll_page_removal_cb, 
454                                            ll_pin_extent_cb);
455         if (err) {
456                 CERROR("cannot register page removal callback: rc = %d\n",err);
457                 GOTO(out_dt, err);
458         }
459         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
460                                           ll_extent_lock_cancel_cb);
461         if (err) {
462                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
463                 GOTO(out_page_rm_cb, err);
464         }
465
466         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
467         if (err) {
468                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
469                 GOTO(out_lock_cn_cb, err);
470         }
471
472         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
473                                   0, NULL, NULL, NULL, 0, NULL);
474         if (err < 0) {
475                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
476                                    "filesystem. There must be at least one "
477                                    "active OST for a client to start.\n");
478                 GOTO(out_lock_cn_cb, err);
479         }
480
481         if (!ll_async_page_slab) {
482                 ll_async_page_slab_size =
483                         size_round(sizeof(struct ll_async_page)) + err;
484                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
485                                                           ll_async_page_slab_size,
486                                                           0, 0);
487                 if (!ll_async_page_slab)
488                         GOTO(out_lock_cn_cb, err = -ENOMEM);
489         }
490
491         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
492         if (err) {
493                 CERROR("cannot mds_connect: rc = %d\n", err);
494                 GOTO(out_lock_cn_cb, err);
495         }
496         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
497         sbi->ll_root_fid = rootfid;
498
499         sb->s_op = &lustre_super_operations;
500         sb->s_export_op = &lustre_export_operations;
501
502         /* make root inode
503          * XXX: move this to after cbd setup? */
504         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
505         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
506                 valid |= OBD_MD_FLRMTPERM;
507         else if (sbi->ll_flags & LL_SBI_ACL)
508                 valid |= OBD_MD_FLACL;
509
510         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
511         if (oc)
512                 free_capa(oc);
513         if (err) {
514                 CERROR("md_getattr failed for root: rc = %d\n", err);
515                 GOTO(out_lock_cn_cb, err);
516         }
517         memset(&lmd, 0, sizeof(lmd));
518         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
519                                sbi->ll_md_exp, &lmd);
520         if (err) {
521                 CERROR("failed to understand root inode md: rc = %d\n", err);
522                 ptlrpc_req_finished (request);
523                 GOTO(out_lock_cn_cb, err);
524         }
525
526         LASSERT(fid_is_sane(&sbi->ll_root_fid));
527         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
528         md_free_lustre_md(sbi->ll_md_exp, &lmd);
529         ptlrpc_req_finished(request);
530
531         if (root == NULL || is_bad_inode(root)) {
532                 if (lmd.lsm)
533                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
534 #ifdef CONFIG_FS_POSIX_ACL
535                 if (lmd.posix_acl) {
536                         posix_acl_release(lmd.posix_acl);
537                         lmd.posix_acl = NULL;
538                 }
539 #endif
540                 CERROR("lustre_lite: bad iget4 for root\n");
541                 GOTO(out_root, err = -EBADF);
542         }
543
544         err = ll_close_thread_start(&sbi->ll_lcq);
545         if (err) {
546                 CERROR("cannot start close thread: rc %d\n", err);
547                 GOTO(out_root, err);
548         }
549
550 #ifdef CONFIG_FS_POSIX_ACL
551         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
552                 rct_init(&sbi->ll_rct);
553                 et_init(&sbi->ll_et);
554         }
555 #endif
556
557         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
558         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
559                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
560                                  NULL);
561
562         sb->s_root = d_alloc_root(root);
563         if (data != NULL)
564                 OBD_FREE(data, sizeof(*data));
565         sb->s_root->d_op = &ll_d_root_ops;
566         RETURN(err);
567 out_root:
568         if (root)
569                 iput(root);
570 out_lock_cn_cb:
571         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
572                                       ll_extent_lock_cancel_cb);
573 out_page_rm_cb:
574         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
575                                        ll_page_removal_cb);
576         obd_fid_fini(sbi->ll_dt_exp);
577 out_dt:
578         obd_disconnect(sbi->ll_dt_exp);
579         sbi->ll_dt_exp = NULL;
580 out_md_fid:
581         obd_fid_fini(sbi->ll_md_exp);
582 out_md:
583         obd_disconnect(sbi->ll_md_exp);
584         sbi->ll_md_exp = NULL;
585 out:
586         if (data != NULL)
587                 OBD_FREE_PTR(data);
588         lprocfs_unregister_mountpoint(sbi);
589         return err;
590 }
591
592 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
593 {
594         int size, rc;
595
596         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
597         size = sizeof(int);
598         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
599                           KEY_MAX_EASIZE, &size, lmmsize);
600         if (rc)
601                 CERROR("Get max mdsize error rc %d \n", rc);
602
603         RETURN(rc);
604 }
605
606 void ll_dump_inode(struct inode *inode)
607 {
608         struct list_head *tmp;
609         int dentry_count = 0;
610
611         LASSERT(inode != NULL);
612
613         list_for_each(tmp, &inode->i_dentry)
614                 dentry_count++;
615
616         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
617                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
618                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
619 }
620
621 void lustre_dump_dentry(struct dentry *dentry, int recur)
622 {
623         struct list_head *tmp;
624         int subdirs = 0;
625
626         LASSERT(dentry != NULL);
627
628         list_for_each(tmp, &dentry->d_subdirs)
629                 subdirs++;
630
631         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
632                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
633                dentry->d_name.len, dentry->d_name.name,
634                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
635                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
636                dentry->d_flags, dentry->d_fsdata, subdirs);
637         if (dentry->d_inode != NULL)
638                 ll_dump_inode(dentry->d_inode);
639
640         if (recur == 0)
641                 return;
642
643         list_for_each(tmp, &dentry->d_subdirs) {
644                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
645                 lustre_dump_dentry(d, recur - 1);
646         }
647 }
648
649 #ifdef HAVE_EXPORT___IGET
650 static void prune_dir_dentries(struct inode *inode)
651 {
652         struct dentry *dentry, *prev = NULL;
653
654         /* due to lustre specific logic, a directory
655          * can have few dentries - a bug from VFS POV */
656 restart:
657         spin_lock(&dcache_lock);
658         if (!list_empty(&inode->i_dentry)) {
659                 dentry = list_entry(inode->i_dentry.prev,
660                                     struct dentry, d_alias);
661                 /* in order to prevent infinite loops we
662                  * break if previous dentry is busy */
663                 if (dentry != prev) {
664                         prev = dentry;
665                         dget_locked(dentry);
666                         spin_unlock(&dcache_lock);
667
668                         /* try to kill all child dentries */
669                         lock_dentry(dentry);
670                         shrink_dcache_parent(dentry);
671                         unlock_dentry(dentry);
672                         dput(dentry);
673
674                         /* now try to get rid of current dentry */
675                         d_prune_aliases(inode);
676                         goto restart;
677                 }
678         }
679         spin_unlock(&dcache_lock);
680 }
681
682 static void prune_deathrow_one(struct ll_inode_info *lli)
683 {
684         struct inode *inode = ll_info2i(lli);
685
686         /* first, try to drop any dentries - they hold a ref on the inode */
687         if (S_ISDIR(inode->i_mode))
688                 prune_dir_dentries(inode);
689         else
690                 d_prune_aliases(inode);
691
692
693         /* if somebody still uses it, leave it */
694         LASSERT(atomic_read(&inode->i_count) > 0);
695         if (atomic_read(&inode->i_count) > 1)
696                 goto out;
697
698         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
699                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
700
701         /* seems nobody uses it anymore */
702         inode->i_nlink = 0;
703
704 out:
705         iput(inode);
706         return;
707 }
708
709 static void prune_deathrow(struct ll_sb_info *sbi, int try)
710 {
711         struct ll_inode_info *lli;
712         int empty;
713
714         do {
715                 if (need_resched() && try)
716                         break;
717
718                 if (try) {
719                         if (!spin_trylock(&sbi->ll_deathrow_lock))
720                                 break;
721                 } else {
722                         spin_lock(&sbi->ll_deathrow_lock);
723                 }
724
725                 empty = 1;
726                 lli = NULL;
727                 if (!list_empty(&sbi->ll_deathrow)) {
728                         lli = list_entry(sbi->ll_deathrow.next,
729                                          struct ll_inode_info,
730                                          lli_dead_list);
731                         list_del_init(&lli->lli_dead_list);
732                         if (!list_empty(&sbi->ll_deathrow))
733                                 empty = 0;
734                 }
735                 spin_unlock(&sbi->ll_deathrow_lock);
736
737                 if (lli)
738                         prune_deathrow_one(lli);
739
740         } while (empty == 0);
741 }
742 #else /* !HAVE_EXPORT___IGET */
743 #define prune_deathrow(sbi, try) do {} while (0)
744 #endif /* HAVE_EXPORT___IGET */
745
746 void client_common_put_super(struct super_block *sb)
747 {
748         struct ll_sb_info *sbi = ll_s2sbi(sb);
749         ENTRY;
750
751 #ifdef CONFIG_FS_POSIX_ACL
752         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
753                 et_fini(&sbi->ll_et);
754                 rct_fini(&sbi->ll_rct);
755         }
756 #endif
757
758         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
759
760         ll_close_thread_shutdown(sbi->ll_lcq);
761
762         /* destroy inodes in deathrow */
763         prune_deathrow(sbi, 0);
764
765         list_del(&sbi->ll_conn_chain);
766
767         obd_fid_fini(sbi->ll_dt_exp);
768         obd_disconnect(sbi->ll_dt_exp);
769         sbi->ll_dt_exp = NULL;
770
771         lprocfs_unregister_mountpoint(sbi);
772
773         obd_fid_fini(sbi->ll_md_exp);
774         obd_disconnect(sbi->ll_md_exp);
775         sbi->ll_md_exp = NULL;
776
777         EXIT;
778 }
779
780 void ll_kill_super(struct super_block *sb)
781 {
782         struct ll_sb_info *sbi;
783
784         ENTRY;
785
786         /* not init sb ?*/
787         if (!(sb->s_flags & MS_ACTIVE))
788                 return;
789
790         sbi = ll_s2sbi(sb);
791         /* we need restore s_dev from changed for clustred NFS before put_super
792          * because new kernels have cached s_dev and change sb->s_dev in
793          * put_super not affected real removing devices */
794         if (sbi)
795                 sb->s_dev = sbi->ll_sdev_orig;
796         EXIT;
797 }
798
799 char *ll_read_opt(const char *opt, char *data)
800 {
801         char *value;
802         char *retval;
803         ENTRY;
804
805         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
806         if (strncmp(opt, data, strlen(opt)))
807                 RETURN(NULL);
808         if ((value = strchr(data, '=')) == NULL)
809                 RETURN(NULL);
810
811         value++;
812         OBD_ALLOC(retval, strlen(value) + 1);
813         if (!retval) {
814                 CERROR("out of memory!\n");
815                 RETURN(NULL);
816         }
817
818         memcpy(retval, value, strlen(value)+1);
819         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
820         RETURN(retval);
821 }
822
823 static inline int ll_set_opt(const char *opt, char *data, int fl)
824 {
825         if (strncmp(opt, data, strlen(opt)) != 0)
826                 return(0);
827         else
828                 return(fl);
829 }
830
831 /* non-client-specific mount options are parsed in lmd_parse */
832 static int ll_options(char *options, int *flags)
833 {
834         int tmp;
835         char *s1 = options, *s2;
836         ENTRY;
837
838         if (!options) 
839                 RETURN(0);
840
841         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
842
843         while (*s1) {
844                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
845                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
846                 if (tmp) {
847                         *flags |= tmp;
848                         goto next;
849                 }
850                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
851                 if (tmp) {
852                         *flags |= tmp;
853                         goto next;
854                 }
855                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
856                 if (tmp) {
857                         *flags |= tmp;
858                         goto next;
859                 }
860                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
861                 if (tmp) {
862                         *flags &= ~tmp;
863                         goto next;
864                 }
865                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
866                 if (tmp) {
867                         *flags |= tmp;
868                         goto next;
869                 }
870                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
871                 if (tmp) {
872                         *flags &= ~tmp;
873                         goto next;
874                 }
875                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
876                 if (tmp) {
877                         /* Ignore deprecated mount option.  The client will
878                          * always try to mount with ACL support, whether this
879                          * is used depends on whether server supports it. */
880                         goto next;
881                 }
882                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
883                 if (tmp) {
884                         goto next;
885                 }
886                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
887                 if (tmp) {
888                         *flags |= tmp;
889                         goto next;
890                 }
891
892                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
893                 if (tmp) {
894                         *flags |= tmp;
895                         goto next;
896                 }
897                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
898                 if (tmp) {
899                         *flags &= ~tmp;
900                         goto next;
901                 }
902                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
903                 if (tmp) {
904                         *flags |= tmp;
905                         goto next;
906                 }
907                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
908                 if (tmp) {
909                         *flags &= ~tmp;
910                         goto next;
911                 }
912
913                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
914                                    s1);
915                 RETURN(-EINVAL);
916
917 next:
918                 /* Find next opt */
919                 s2 = strchr(s1, ',');
920                 if (s2 == NULL)
921                         break;
922                 s1 = s2 + 1;
923         }
924         RETURN(0);
925 }
926
927 void ll_lli_init(struct ll_inode_info *lli)
928 {
929         lli->lli_inode_magic = LLI_INODE_MAGIC;
930         sema_init(&lli->lli_size_sem, 1);
931         sema_init(&lli->lli_write_sem, 1);
932         lli->lli_flags = 0;
933         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
934         spin_lock_init(&lli->lli_lock);
935         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
936         INIT_LIST_HEAD(&lli->lli_close_list);
937         lli->lli_inode_magic = LLI_INODE_MAGIC;
938         sema_init(&lli->lli_och_sem, 1);
939         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
940         lli->lli_mds_exec_och = NULL;
941         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
942         lli->lli_open_fd_exec_count = 0;
943         INIT_LIST_HEAD(&lli->lli_dead_list);
944         lli->lli_remote_perms = NULL;
945         lli->lli_rmtperm_utime = 0;
946         sema_init(&lli->lli_rmtperm_sem, 1);
947         INIT_LIST_HEAD(&lli->lli_oss_capas);
948 }
949
950 int ll_fill_super(struct super_block *sb)
951 {
952         struct lustre_profile *lprof;
953         struct lustre_sb_info *lsi = s2lsi(sb);
954         struct ll_sb_info *sbi;
955         char  *dt = NULL, *md = NULL;
956         char  *profilenm = get_profile_name(sb);
957         struct config_llog_instance cfg = {0, };
958         char   ll_instance[sizeof(sb) * 2 + 1];
959         int    err;
960         ENTRY;
961
962         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
963
964         cfs_module_get();
965
966         /* client additional sb info */
967         lsi->lsi_llsbi = sbi = ll_init_sbi();
968         if (!sbi) {
969                 cfs_module_put();
970                 RETURN(-ENOMEM);
971         }
972
973         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
974         if (err) 
975                 GOTO(out_free, err);
976
977         /* Generate a string unique to this super, in case some joker tries
978            to mount the same fs at two mount points.
979            Use the address of the super itself.*/
980         sprintf(ll_instance, "%p", sb);
981         cfg.cfg_instance = ll_instance;
982         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
983
984         /* set up client obds */
985         err = lustre_process_log(sb, profilenm, &cfg);
986         if (err < 0) {
987                 CERROR("Unable to process log: %d\n", err);
988                 GOTO(out_free, err);
989         }
990
991         lprof = class_get_profile(profilenm);
992         if (lprof == NULL) {
993                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
994                                    " read from the MGS.  Does that filesystem "
995                                    "exist?\n", profilenm);
996                 GOTO(out_free, err = -EINVAL);
997         }
998         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
999                lprof->lp_md, lprof->lp_dt);
1000
1001         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1002                   strlen(ll_instance) + 2);
1003         if (!dt)
1004                 GOTO(out_free, err = -ENOMEM);
1005         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1006
1007         OBD_ALLOC(md, strlen(lprof->lp_md) +
1008                   strlen(ll_instance) + 2);
1009         if (!md)
1010                 GOTO(out_free, err = -ENOMEM);
1011         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1012
1013         /* connections, registrations, sb setup */
1014         err = client_common_fill_super(sb, md, dt);
1015
1016 out_free:
1017         if (md)
1018                 OBD_FREE(md, strlen(md) + 1);
1019         if (dt)
1020                 OBD_FREE(dt, strlen(dt) + 1);
1021         if (err) 
1022                 ll_put_super(sb);
1023         else
1024                 LCONSOLE_WARN("Client %s has started\n", profilenm);        
1025
1026         RETURN(err);
1027 } /* ll_fill_super */
1028
1029
1030 void ll_put_super(struct super_block *sb)
1031 {
1032         struct config_llog_instance cfg;
1033         char   ll_instance[sizeof(sb) * 2 + 1];
1034         struct obd_device *obd;
1035         struct lustre_sb_info *lsi = s2lsi(sb);
1036         struct ll_sb_info *sbi = ll_s2sbi(sb);
1037         char *profilenm = get_profile_name(sb);
1038         int force = 1, next;
1039         ENTRY;
1040
1041         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1042
1043         ll_print_capa_stat(sbi);
1044
1045         sprintf(ll_instance, "%p", sb);
1046         cfg.cfg_instance = ll_instance;
1047         lustre_end_log(sb, NULL, &cfg);
1048         
1049         if (sbi->ll_md_exp) {
1050                 obd = class_exp2obd(sbi->ll_md_exp);
1051                 if (obd) 
1052                         force = obd->obd_force;
1053         }
1054         
1055         /* We need to set force before the lov_disconnect in 
1056            lustre_common_put_super, since l_d cleans up osc's as well. */
1057         if (force) {
1058                 next = 0;
1059                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1060                                                      &next)) != NULL) {
1061                         obd->obd_force = force;
1062                 }
1063         }                       
1064
1065         if (sbi->ll_lcq) {
1066                 /* Only if client_common_fill_super succeeded */
1067                 client_common_put_super(sb);
1068         }
1069         next = 0;
1070         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1071                 class_manual_cleanup(obd);
1072         }
1073
1074         if (profilenm)
1075                 class_del_profile(profilenm);
1076
1077         ll_free_sbi(sb);
1078         lsi->lsi_llsbi = NULL;
1079
1080         lustre_common_put_super(sb);
1081
1082         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1083         
1084         cfs_module_put();
1085
1086         EXIT;
1087 } /* client_put_super */
1088
1089 #ifdef HAVE_REGISTER_CACHE
1090 #include <linux/cache_def.h>
1091 #ifdef HAVE_CACHE_RETURN_INT
1092 static int
1093 #else
1094 static void
1095 #endif
1096 ll_shrink_cache(int priority, unsigned int gfp_mask)
1097 {
1098         struct ll_sb_info *sbi;
1099         int count = 0;
1100
1101         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1102                 count += llap_shrink_cache(sbi, priority);
1103
1104 #ifdef HAVE_CACHE_RETURN_INT
1105         return count;
1106 #endif
1107 }
1108
1109 struct cache_definition ll_cache_definition = {
1110         .name = "llap_cache",
1111         .shrink = ll_shrink_cache
1112 };
1113 #endif /* HAVE_REGISTER_CACHE */
1114
1115 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1116 {
1117         struct inode *inode = NULL;
1118         /* NOTE: we depend on atomic igrab() -bzzz */
1119         lock_res_and_lock(lock);
1120         if (lock->l_ast_data) {
1121                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1122                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1123                         inode = igrab(lock->l_ast_data);
1124                 } else {
1125                         inode = lock->l_ast_data;
1126                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1127                                                 D_INFO : D_WARNING,
1128                                         lock, __FILE__, __func__, __LINE__,
1129                                         "l_ast_data %p is bogus: magic %08x",
1130                                         lock->l_ast_data, lli->lli_inode_magic);
1131                         inode = NULL;
1132                 }
1133         }
1134         unlock_res_and_lock(lock);
1135         return inode;
1136 }
1137
1138 static int null_if_equal(struct ldlm_lock *lock, void *data)
1139 {
1140         if (data == lock->l_ast_data) {
1141                 lock->l_ast_data = NULL;
1142
1143                 if (lock->l_req_mode != lock->l_granted_mode)
1144                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1145         }
1146
1147         return LDLM_ITER_CONTINUE;
1148 }
1149
1150 void ll_clear_inode(struct inode *inode)
1151 {
1152         struct ll_inode_info *lli = ll_i2info(inode);
1153         struct ll_sb_info *sbi = ll_i2sbi(inode);
1154         ENTRY;
1155
1156         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1157                inode->i_generation, inode);
1158
1159         if (S_ISDIR(inode->i_mode)) {
1160                 /* these should have been cleared in ll_file_release */
1161                 LASSERT(lli->lli_sai == NULL);
1162                 LASSERT(lli->lli_opendir_key == NULL);
1163                 LASSERT(lli->lli_opendir_pid == 0);
1164         }
1165
1166         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1167         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1168                          null_if_equal, inode);
1169
1170         LASSERT(!lli->lli_open_fd_write_count);
1171         LASSERT(!lli->lli_open_fd_read_count);
1172         LASSERT(!lli->lli_open_fd_exec_count);
1173
1174         if (lli->lli_mds_write_och)
1175                 ll_md_real_close(inode, FMODE_WRITE);
1176         if (lli->lli_mds_exec_och)
1177                 ll_md_real_close(inode, FMODE_EXEC);
1178         if (lli->lli_mds_read_och)
1179                 ll_md_real_close(inode, FMODE_READ);
1180
1181         if (lli->lli_smd) {
1182                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1183                                   null_if_equal, inode);
1184
1185                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1186                 lli->lli_smd = NULL;
1187         }
1188
1189         if (lli->lli_symlink_name) {
1190                 OBD_FREE(lli->lli_symlink_name,
1191                          strlen(lli->lli_symlink_name) + 1);
1192                 lli->lli_symlink_name = NULL;
1193         }
1194
1195         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1196                 LASSERT(lli->lli_posix_acl == NULL);
1197                 if (lli->lli_remote_perms) {
1198                         free_rmtperm_hash(lli->lli_remote_perms);
1199                         lli->lli_remote_perms = NULL;
1200                 }
1201         }
1202 #ifdef CONFIG_FS_POSIX_ACL
1203         else if (lli->lli_posix_acl) {
1204                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1205                 LASSERT(lli->lli_remote_perms == NULL);
1206                 posix_acl_release(lli->lli_posix_acl);
1207                 lli->lli_posix_acl = NULL;
1208         }
1209 #endif
1210         lli->lli_inode_magic = LLI_INODE_DEAD;
1211
1212 #ifdef HAVE_EXPORT___IGET
1213         spin_lock(&sbi->ll_deathrow_lock);
1214         list_del_init(&lli->lli_dead_list);
1215         spin_unlock(&sbi->ll_deathrow_lock);
1216 #endif
1217         ll_clear_inode_capas(inode);
1218
1219         EXIT;
1220 }
1221
1222 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1223                   struct md_open_data **mod)
1224 {
1225         struct lustre_md md;
1226         struct ll_sb_info *sbi = ll_i2sbi(inode);
1227         struct ptlrpc_request *request = NULL;
1228         int rc;
1229         ENTRY;
1230         
1231         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, 
1232                                      LUSTRE_OPC_ANY, NULL);
1233         if (IS_ERR(op_data))
1234                 RETURN(PTR_ERR(op_data));
1235
1236         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0, 
1237                         &request, mod);
1238         if (rc) {
1239                 ptlrpc_req_finished(request);
1240                 if (rc == -ENOENT) {
1241                         inode->i_nlink = 0;
1242                         /* Unlinked special device node? Or just a race?
1243                          * Pretend we done everything. */
1244                         if (!S_ISREG(inode->i_mode) &&
1245                             !S_ISDIR(inode->i_mode))
1246                                 rc = inode_setattr(inode, &op_data->op_attr);
1247                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1248                         CERROR("md_setattr fails: rc = %d\n", rc);
1249                 }
1250                 RETURN(rc);
1251         }
1252
1253         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1254                               sbi->ll_md_exp, &md);
1255         if (rc) {
1256                 ptlrpc_req_finished(request);
1257                 RETURN(rc);
1258         }
1259
1260         /* We call inode_setattr to adjust timestamps.
1261          * If there is at least some data in file, we cleared ATTR_SIZE
1262          * above to avoid invoking vmtruncate, otherwise it is important
1263          * to call vmtruncate in inode_setattr to update inode->i_size
1264          * (bug 6196) */
1265         rc = inode_setattr(inode, &op_data->op_attr);
1266
1267         /* Extract epoch data if obtained. */
1268         op_data->op_handle = md.body->handle;
1269         op_data->op_ioepoch = md.body->ioepoch;
1270
1271         ll_update_inode(inode, &md);
1272         ptlrpc_req_finished(request);
1273
1274         RETURN(rc);
1275 }
1276
1277 /* Close IO epoch and send Size-on-MDS attribute update. */
1278 static int ll_setattr_done_writing(struct inode *inode,
1279                                    struct md_op_data *op_data,
1280                                    struct md_open_data *mod)
1281 {
1282         struct ll_inode_info *lli = ll_i2info(inode);
1283         int rc = 0;
1284         ENTRY;
1285         
1286         LASSERT(op_data != NULL);
1287         if (!S_ISREG(inode->i_mode))
1288                 RETURN(0);
1289
1290         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1291                op_data->op_ioepoch, PFID(&lli->lli_fid));
1292
1293         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1294         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1295         if (rc == -EAGAIN) {
1296                 /* MDS has instructed us to obtain Size-on-MDS attribute
1297                  * from OSTs and send setattr to back to MDS. */
1298                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1299                                          op_data->op_ioepoch);
1300         } else if (rc) {
1301                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1302                        inode->i_ino, rc);
1303         }
1304         RETURN(rc);
1305 }
1306
1307 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1308 {
1309         struct ll_sb_info *sbi = ll_i2sbi(inode);
1310         struct ll_inode_info *lli = ll_i2info(inode);
1311         struct lov_stripe_md *lsm = lli->lli_smd;
1312         int rc;
1313         ldlm_policy_data_t policy = { .l_extent = {new_size,
1314                                                    OBD_OBJECT_EOF } };
1315         struct lustre_handle lockh = { 0 };
1316         int local_lock = 0; /* 0 - no local lock;
1317                              * 1 - lock taken by lock_extent;
1318                              * 2 - by obd_match*/
1319         int ast_flags;
1320         int err;
1321         ENTRY;
1322
1323         UNLOCK_INODE_MUTEX(inode);
1324         UP_WRITE_I_ALLOC_SEM(inode);
1325
1326         if (sbi->ll_lockless_truncate_enable &&
1327             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1328                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1329                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1330                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1331                 if (rc > 0) {
1332                         local_lock = 2;
1333                         rc = 0;
1334                 } else if (rc == 0) {
1335                         rc = ll_file_punch(inode, new_size, 1);
1336                 }
1337         } else {
1338                 /* XXX when we fix the AST intents to pass the discard-range
1339                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1340                  * XXX here. */
1341                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1342                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1343                                     &lockh, ast_flags);
1344                 if (likely(rc == 0))
1345                         local_lock = 1;
1346         }
1347
1348         LOCK_INODE_MUTEX(inode);
1349         DOWN_WRITE_I_ALLOC_SEM(inode);
1350
1351         if (likely(rc == 0)) {
1352                 /* Only ll_inode_size_lock is taken at this level.
1353                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1354                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1355                  * ll_truncate dropped ll_inode_size_lock() */
1356                 ll_inode_size_lock(inode, 0);
1357                 if (!local_lock) {
1358                         spin_lock(&lli->lli_lock);
1359                         lli->lli_flags |= LLIF_SRVLOCK;
1360                         spin_unlock(&lli->lli_lock);
1361                 }
1362                 rc = vmtruncate(inode, new_size);
1363                 if (!local_lock) {
1364                         spin_lock(&lli->lli_lock);
1365                         lli->lli_flags &= ~LLIF_SRVLOCK;
1366                         spin_unlock(&lli->lli_lock);
1367                 }
1368                 if (rc != 0) {
1369                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1370                         ll_inode_size_unlock(inode, 0);
1371                 }
1372         }
1373
1374         if (local_lock) {
1375                 if (local_lock == 2)
1376                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1377                 else
1378                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1379                 if (unlikely(err != 0)){
1380                         CERROR("extent unlock failed: err=%d,"
1381                                " unlock method =%d\n", err, local_lock);
1382                         if (rc == 0)
1383                                 rc = err;
1384                 }
1385         }
1386         RETURN(rc);
1387 }
1388
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1391  * keep these values until such a time that objects are allocated for it.
1392  * We do the MDS operations first, as it is checking permissions for us.
1393  * We don't to the MDS RPC if there is nothing that we want to store there,
1394  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395  * going to do an RPC anyways.
1396  *
1397  * If we are doing a truncate, we will send the mtime and ctime updates
1398  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1400  * at the same time.
1401  */
1402 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1403 {
1404         struct ll_inode_info *lli = ll_i2info(inode);
1405         struct lov_stripe_md *lsm = lli->lli_smd;
1406         struct ll_sb_info *sbi = ll_i2sbi(inode);
1407         struct md_op_data *op_data = NULL;
1408         struct md_open_data *mod = NULL;
1409         int ia_valid = attr->ia_valid;
1410         int rc = 0, rc1 = 0;
1411         ENTRY;
1412
1413         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1414                attr->ia_valid);
1415         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1416
1417         if (ia_valid & ATTR_SIZE) {
1418                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1419                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1420                                attr->ia_size, ll_file_maxbytes(inode));
1421                         RETURN(-EFBIG);
1422                 }
1423
1424                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1425         }
1426
1427         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1428         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1429                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1430                         RETURN(-EPERM);
1431         }
1432
1433         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1434         if (attr->ia_valid & ATTR_CTIME) {
1435                 attr->ia_ctime = CURRENT_TIME;
1436                 attr->ia_valid |= ATTR_CTIME_SET;
1437         }
1438         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1439                 attr->ia_atime = CURRENT_TIME;
1440                 attr->ia_valid |= ATTR_ATIME_SET;
1441         }
1442         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1443                 attr->ia_mtime = CURRENT_TIME;
1444                 attr->ia_valid |= ATTR_MTIME_SET;
1445         }
1446         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1447                 /* To avoid stale mtime on mds, obtain it from ost and send 
1448                    to mds. */
1449                 rc = ll_glimpse_size(inode, 0);
1450                 if (rc) 
1451                         RETURN(rc);
1452                 
1453                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1454                 attr->ia_mtime = inode->i_mtime;
1455         }
1456
1457         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1458                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1459                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1460                        cfs_time_current_sec());
1461
1462         /* NB: ATTR_SIZE will only be set after this point if the size
1463          * resides on the MDS, ie, this file has no objects. */
1464         if (lsm)
1465                 attr->ia_valid &= ~ATTR_SIZE;
1466
1467         /* We always do an MDS RPC, even if we're only changing the size;
1468          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1469
1470         OBD_ALLOC_PTR(op_data);
1471         if (op_data == NULL)
1472                 RETURN(-ENOMEM);
1473
1474         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1475
1476         /* Open epoch for truncate. */
1477         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1478             (ia_valid & ATTR_SIZE))
1479                 op_data->op_flags = MF_EPOCH_OPEN;
1480
1481         rc = ll_md_setattr(inode, op_data, &mod);
1482         if (rc)
1483                 GOTO(out, rc);
1484
1485         if (op_data->op_ioepoch)
1486                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1487                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1488
1489         if (!lsm || !S_ISREG(inode->i_mode)) {
1490                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1491                 GOTO(out, rc = 0);
1492         }
1493
1494         /* We really need to get our PW lock before we change inode->i_size.
1495          * If we don't we can race with other i_size updaters on our node, like
1496          * ll_file_read.  We can also race with i_size propogation to other
1497          * nodes through dirtying and writeback of final cached pages.  This
1498          * last one is especially bad for racing o_append users on other
1499          * nodes. */
1500         if (ia_valid & ATTR_SIZE) {
1501                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1502         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1503                 obd_flag flags;
1504                 struct obd_info oinfo = { { { 0 } } };
1505                 struct obdo *oa;
1506
1507                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1508                        inode->i_ino, LTIME_S(attr->ia_mtime));
1509
1510                 OBDO_ALLOC(oa);
1511                 if (oa) {
1512                         oa->o_id = lsm->lsm_object_id;
1513                         oa->o_gr = lsm->lsm_object_gr;
1514                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1515
1516                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1517                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1518                                 OBD_MD_FLFID | OBD_MD_FLGENER | 
1519                                 OBD_MD_FLGROUP;
1520
1521                         obdo_from_inode(oa, inode, flags);
1522
1523                         oinfo.oi_oa = oa;
1524                         oinfo.oi_md = lsm;
1525                         oinfo.oi_capa = ll_mdscapa_get(inode);
1526
1527                         /* XXX: this looks unnecessary now. */
1528                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1529                         capa_put(oinfo.oi_capa);
1530                         if (rc)
1531                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1532                         OBDO_FREE(oa);
1533                 } else {
1534                         rc = -ENOMEM;
1535                 }
1536         }
1537         EXIT;
1538 out:
1539         if (op_data) {
1540                 if (op_data->op_ioepoch)
1541                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1542                 ll_finish_md_op_data(op_data);
1543         }
1544         return rc ? rc : rc1;
1545 }
1546
1547 int ll_setattr(struct dentry *de, struct iattr *attr)
1548 {
1549         int mode;
1550
1551         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1552             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1553                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1554         if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) == 
1555             (ATTR_SIZE|ATTR_MODE)) {
1556                 mode = de->d_inode->i_mode;
1557                 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1558                     ((mode & S_ISGID) && (mode & S_IXGRP) &&
1559                     (!(attr->ia_mode & S_ISGID))))
1560                         attr->ia_valid |= ATTR_FORCE;
1561         }
1562
1563         return ll_setattr_raw(de->d_inode, attr);
1564 }
1565
1566 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1567                        __u64 max_age, __u32 flags)
1568 {
1569         struct ll_sb_info *sbi = ll_s2sbi(sb);
1570         struct obd_statfs obd_osfs;
1571         int rc;
1572         ENTRY;
1573
1574         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1575         if (rc) {
1576                 CERROR("md_statfs fails: rc = %d\n", rc);
1577                 RETURN(rc);
1578         }
1579
1580         osfs->os_type = sb->s_magic;
1581
1582         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1583                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1584
1585         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1586                               &obd_osfs, max_age, flags);
1587         if (rc) {
1588                 CERROR("obd_statfs fails: rc = %d\n", rc);
1589                 RETURN(rc);
1590         }
1591
1592         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1593                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1594                obd_osfs.os_files);
1595
1596         osfs->os_bsize = obd_osfs.os_bsize;
1597         osfs->os_blocks = obd_osfs.os_blocks;
1598         osfs->os_bfree = obd_osfs.os_bfree;
1599         osfs->os_bavail = obd_osfs.os_bavail;
1600
1601         /* If we don't have as many objects free on the OST as inodes
1602          * on the MDS, we reduce the total number of inodes to
1603          * compensate, so that the "inodes in use" number is correct.
1604          */
1605         if (obd_osfs.os_ffree < osfs->os_ffree) {
1606                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1607                         obd_osfs.os_ffree;
1608                 osfs->os_ffree = obd_osfs.os_ffree;
1609         }
1610
1611         RETURN(rc);
1612 }
1613 #ifndef HAVE_STATFS_DENTRY_PARAM
1614 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1615 {
1616 #else
1617 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1618 {
1619         struct super_block *sb = de->d_sb;
1620 #endif
1621         struct obd_statfs osfs;
1622         int rc;
1623
1624         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1625         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1626
1627         /* For now we will always get up-to-date statfs values, but in the
1628          * future we may allow some amount of caching on the client (e.g.
1629          * from QOS or lprocfs updates). */
1630         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1631         if (rc)
1632                 return rc;
1633
1634         statfs_unpack(sfs, &osfs);
1635
1636         /* We need to downshift for all 32-bit kernels, because we can't
1637          * tell if the kernel is being called via sys_statfs64() or not.
1638          * Stop before overflowing f_bsize - in which case it is better
1639          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1640         if (sizeof(long) < 8) {
1641                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1642                         sfs->f_bsize <<= 1;
1643
1644                         osfs.os_blocks >>= 1;
1645                         osfs.os_bfree >>= 1;
1646                         osfs.os_bavail >>= 1;
1647                 }
1648         }
1649
1650         sfs->f_blocks = osfs.os_blocks;
1651         sfs->f_bfree = osfs.os_bfree;
1652         sfs->f_bavail = osfs.os_bavail;
1653
1654         return 0;
1655 }
1656
1657 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1658 {
1659         struct ll_inode_info *lli;
1660         struct lov_stripe_md *lsm;
1661
1662         lli = ll_i2info(inode);
1663         LASSERT(lli->lli_size_sem_owner != current);
1664         down(&lli->lli_size_sem);
1665         LASSERT(lli->lli_size_sem_owner == NULL);
1666         lli->lli_size_sem_owner = current;
1667         lsm = lli->lli_smd;
1668         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1669                  lsm, lock_lsm);
1670         if (lock_lsm)
1671                 lov_stripe_lock(lsm);
1672 }
1673
1674 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1675 {
1676         struct ll_inode_info *lli;
1677         struct lov_stripe_md *lsm;
1678
1679         lli = ll_i2info(inode);
1680         lsm = lli->lli_smd;
1681         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1682                  lsm, unlock_lsm);
1683         if (unlock_lsm)
1684                 lov_stripe_unlock(lsm);
1685         LASSERT(lli->lli_size_sem_owner == current);
1686         lli->lli_size_sem_owner = NULL;
1687         up(&lli->lli_size_sem);
1688 }
1689
1690 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1691 {
1692         struct ll_inode_info *lli = ll_i2info(inode);
1693
1694         dump_lsm(D_INODE, lsm);
1695         dump_lsm(D_INODE, lli->lli_smd);
1696         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1697                  "lsm must be joined lsm %p\n", lsm);
1698         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1699         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1700                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1701         lli->lli_smd = lsm;
1702         lli->lli_maxbytes = lsm->lsm_maxbytes;
1703         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1704                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1705 }
1706
1707 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1708 {
1709         struct ll_inode_info *lli = ll_i2info(inode);
1710         struct mdt_body *body = md->body;
1711         struct lov_stripe_md *lsm = md->lsm;
1712         struct ll_sb_info *sbi = ll_i2sbi(inode);
1713
1714         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1715         if (lsm != NULL) {
1716                 if (lli->lli_smd == NULL) {
1717                         if (lsm->lsm_magic != LOV_MAGIC &&
1718                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1719                                 dump_lsm(D_ERROR, lsm);
1720                                 LBUG();
1721                         }
1722                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1723                                lsm, inode->i_ino, inode->i_generation, inode);
1724                         /* ll_inode_size_lock() requires it is only called
1725                          * with lli_smd != NULL or lock_lsm == 0 or we can
1726                          * race between lock/unlock.  bug 9547 */
1727                         lli->lli_smd = lsm;
1728                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1729                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1730                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1731                 } else {
1732                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1733                              lli->lli_smd->lsm_stripe_count ==
1734                                         lsm->lsm_stripe_count) {
1735                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1736                                         CERROR("lsm mismatch for inode %ld\n",
1737                                                 inode->i_ino);
1738                                         CERROR("lli_smd:\n");
1739                                         dump_lsm(D_ERROR, lli->lli_smd);
1740                                         CERROR("lsm:\n");
1741                                         dump_lsm(D_ERROR, lsm);
1742                                         LBUG();
1743                                 }
1744                         } else
1745                                 ll_replace_lsm(inode, lsm);
1746                 }
1747                 if (lli->lli_smd != lsm)
1748                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1749         }
1750
1751         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1752                 if (body->valid & OBD_MD_FLRMTPERM)
1753                         ll_update_remote_perm(inode, md->remote_perm);
1754         }
1755 #ifdef CONFIG_FS_POSIX_ACL
1756         else if (body->valid & OBD_MD_FLACL) {
1757                 spin_lock(&lli->lli_lock);
1758                 if (lli->lli_posix_acl)
1759                         posix_acl_release(lli->lli_posix_acl);
1760                 lli->lli_posix_acl = md->posix_acl;
1761                 spin_unlock(&lli->lli_lock);
1762         }
1763 #endif
1764         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1765
1766         if (body->valid & OBD_MD_FLATIME &&
1767             body->atime > LTIME_S(inode->i_atime))
1768                 LTIME_S(inode->i_atime) = body->atime;
1769         
1770         /* mtime is always updated with ctime, but can be set in past.
1771            As write and utime(2) may happen within 1 second, and utime's
1772            mtime has a priority over write's one, so take mtime from mds 
1773            for the same ctimes. */
1774         if (body->valid & OBD_MD_FLCTIME &&
1775             body->ctime >= LTIME_S(inode->i_ctime)) {
1776                 LTIME_S(inode->i_ctime) = body->ctime;
1777                 if (body->valid & OBD_MD_FLMTIME) {
1778                         CDEBUG(D_INODE, "setting ino %lu mtime "
1779                                "from %lu to "LPU64"\n", inode->i_ino, 
1780                                LTIME_S(inode->i_mtime), body->mtime);
1781                         LTIME_S(inode->i_mtime) = body->mtime;
1782                 }
1783         }
1784         if (body->valid & OBD_MD_FLMODE)
1785                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1786         if (body->valid & OBD_MD_FLTYPE)
1787                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1788         if (S_ISREG(inode->i_mode)) {
1789                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1790         } else {
1791                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1792         }
1793 #ifdef HAVE_INODE_BLKSIZE
1794         inode->i_blksize = 1<<inode->i_blkbits;
1795 #endif
1796         if (body->valid & OBD_MD_FLUID)
1797                 inode->i_uid = body->uid;
1798         if (body->valid & OBD_MD_FLGID)
1799                 inode->i_gid = body->gid;
1800         if (body->valid & OBD_MD_FLFLAGS)
1801                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1802         if (body->valid & OBD_MD_FLNLINK)
1803                 inode->i_nlink = body->nlink;
1804         if (body->valid & OBD_MD_FLRDEV)
1805                 inode->i_rdev = old_decode_dev(body->rdev);
1806
1807         if (body->valid & OBD_MD_FLID) {
1808                 /* FID shouldn't be changed! */
1809                 if (fid_is_sane(&lli->lli_fid)) {
1810                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1811                                  "Trying to change FID "DFID
1812                                  " to the "DFID", inode %lu/%u(%p)\n",
1813                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1814                                  inode->i_ino, inode->i_generation, inode);
1815                 } else 
1816                         lli->lli_fid = body->fid1;
1817         }
1818
1819         LASSERT(fid_seq(&lli->lli_fid) != 0);
1820
1821         if (body->valid & OBD_MD_FLSIZE) {
1822                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1823                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1824                         struct lustre_handle lockh;
1825                         ldlm_mode_t mode;
1826                         
1827                         /* As it is possible a blocking ast has been processed
1828                          * by this time, we need to check there is an UPDATE 
1829                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1830                          * it. */
1831                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1832                                                &lockh);
1833                         if (mode) {
1834                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1835                                                       LLIF_EPOCH_PENDING |
1836                                                       LLIF_SOM_DIRTY)) {
1837                                         CERROR("ino %lu flags %lu still has "
1838                                                "size authority! do not trust "
1839                                                "the size got from MDS\n",
1840                                                inode->i_ino, lli->lli_flags);
1841                                 } else {
1842                                         /* Use old size assignment to avoid
1843                                          * deadlock bz14138 & bz14326 */
1844                                         inode->i_size = body->size;
1845                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1846                                 }
1847                                 ldlm_lock_decref(&lockh, mode);
1848                         }
1849                 } else {
1850                         /* Use old size assignment to avoid
1851                          * deadlock bz14138 & bz14326 */
1852                         inode->i_size = body->size;
1853                 }
1854
1855                 if (body->valid & OBD_MD_FLBLOCKS)
1856                         inode->i_blocks = body->blocks;
1857         }
1858
1859         if (body->valid & OBD_MD_FLMDSCAPA) {
1860                 LASSERT(md->mds_capa);
1861                 ll_add_capa(inode, md->mds_capa);
1862         }
1863         if (body->valid & OBD_MD_FLOSSCAPA) {
1864                 LASSERT(md->oss_capa);
1865                 ll_add_capa(inode, md->oss_capa);
1866         }
1867 }
1868
1869 static struct backing_dev_info ll_backing_dev_info = {
1870         .ra_pages       = 0,    /* No readahead */
1871 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1872         .capabilities   = 0,    /* Does contribute to dirty memory */
1873 #else
1874         .memory_backed  = 0,    /* Does contribute to dirty memory */
1875 #endif
1876 };
1877
1878 void ll_read_inode2(struct inode *inode, void *opaque)
1879 {
1880         struct lustre_md *md = opaque;
1881         struct ll_inode_info *lli = ll_i2info(inode);
1882         ENTRY;
1883
1884         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1885                inode->i_ino, inode->i_generation, inode);
1886
1887         ll_lli_init(lli);
1888
1889         LASSERT(!lli->lli_smd);
1890
1891         /* Core attributes from the MDS first.  This is a new inode, and
1892          * the VFS doesn't zero times in the core inode so we have to do
1893          * it ourselves.  They will be overwritten by either MDS or OST
1894          * attributes - we just need to make sure they aren't newer. */
1895         LTIME_S(inode->i_mtime) = 0;
1896         LTIME_S(inode->i_atime) = 0;
1897         LTIME_S(inode->i_ctime) = 0;
1898         inode->i_rdev = 0;
1899         ll_update_inode(inode, md);
1900
1901         /* OIDEBUG(inode); */
1902
1903         if (S_ISREG(inode->i_mode)) {
1904                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1905                 inode->i_op = &ll_file_inode_operations;
1906                 inode->i_fop = sbi->ll_fop;
1907                 inode->i_mapping->a_ops = &ll_aops;
1908                 EXIT;
1909         } else if (S_ISDIR(inode->i_mode)) {
1910                 inode->i_op = &ll_dir_inode_operations;
1911                 inode->i_fop = &ll_dir_operations;
1912                 inode->i_mapping->a_ops = &ll_dir_aops;
1913                 EXIT;
1914         } else if (S_ISLNK(inode->i_mode)) {
1915                 inode->i_op = &ll_fast_symlink_inode_operations;
1916                 EXIT;
1917         } else {
1918                 inode->i_op = &ll_special_inode_operations;
1919
1920                 init_special_inode(inode, inode->i_mode,
1921                                    kdev_t_to_nr(inode->i_rdev));
1922
1923                 /* initializing backing dev info. */
1924                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1925
1926                 EXIT;
1927         }
1928 }
1929
1930 void ll_delete_inode(struct inode *inode)
1931 {
1932         struct ll_sb_info *sbi = ll_i2sbi(inode);
1933         int rc;
1934         ENTRY;
1935
1936         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1937         if (rc) {
1938                 CERROR("fid_delete() failed, rc %d\n", rc);
1939         }
1940         truncate_inode_pages(&inode->i_data, 0);
1941         clear_inode(inode);
1942
1943         EXIT;
1944 }
1945
1946 int ll_iocontrol(struct inode *inode, struct file *file,
1947                  unsigned int cmd, unsigned long arg)
1948 {
1949         struct ll_sb_info *sbi = ll_i2sbi(inode);
1950         struct ptlrpc_request *req = NULL;
1951         int rc, flags = 0;
1952         ENTRY;
1953
1954         switch(cmd) {
1955         case EXT3_IOC_GETFLAGS: {
1956                 struct mdt_body *body;
1957                 struct obd_capa *oc;
1958
1959                 oc = ll_mdscapa_get(inode);
1960                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1961                                 OBD_MD_FLFLAGS, 0, &req);
1962                 capa_put(oc);
1963                 if (rc) {
1964                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1965                         RETURN(-abs(rc));
1966                 }
1967
1968                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1969
1970                 flags = body->flags;
1971
1972                 ptlrpc_req_finished(req);
1973
1974                 RETURN(put_user(flags, (int *)arg));
1975         }
1976         case EXT3_IOC_SETFLAGS: {
1977                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1978                 struct obd_info oinfo = { { { 0 } } };
1979                 struct md_op_data *op_data;
1980
1981                 if (get_user(flags, (int *)arg))
1982                         RETURN(-EFAULT);
1983
1984                 oinfo.oi_md = lsm;
1985                 OBDO_ALLOC(oinfo.oi_oa);
1986                 if (!oinfo.oi_oa)
1987                         RETURN(-ENOMEM);
1988
1989                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1990                                              LUSTRE_OPC_ANY, NULL);
1991                 if (IS_ERR(op_data))
1992                         RETURN(PTR_ERR(op_data));
1993
1994                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1995                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1996                 rc = md_setattr(sbi->ll_md_exp, op_data,
1997                                 NULL, 0, NULL, 0, &req, NULL);
1998                 ll_finish_md_op_data(op_data);
1999                 ptlrpc_req_finished(req);
2000                 if (rc || lsm == NULL) {
2001                         OBDO_FREE(oinfo.oi_oa);
2002                         RETURN(rc);
2003                 }
2004
2005                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2006                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2007                 oinfo.oi_oa->o_flags = flags;
2008                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | 
2009                                        OBD_MD_FLGROUP;
2010                 oinfo.oi_capa = ll_mdscapa_get(inode);
2011
2012                 obdo_from_inode(oinfo.oi_oa, inode,
2013                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2014                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2015                 capa_put(oinfo.oi_capa);
2016                 OBDO_FREE(oinfo.oi_oa);
2017                 if (rc) {
2018                         if (rc != -EPERM && rc != -EACCES)
2019                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2020                         RETURN(rc);
2021                 }
2022
2023                 inode->i_flags = ll_ext_to_inode_flags(flags |
2024                                                        MDS_BFLAG_EXT_FLAGS);
2025                 RETURN(0);
2026         }
2027         default:
2028                 RETURN(-ENOSYS);
2029         }
2030
2031         RETURN(0);
2032 }
2033
2034 int ll_flush_ctx(struct inode *inode)
2035 {
2036         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2037
2038         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2039
2040         obd_set_info_async(sbi->ll_md_exp,
2041                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2042                            0, NULL, NULL);
2043         obd_set_info_async(sbi->ll_dt_exp,
2044                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2045                            0, NULL, NULL);
2046         return 0;
2047 }
2048
2049 /* umount -f client means force down, don't save state */
2050 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2051 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2052 {
2053         struct super_block *sb = vfsmnt->mnt_sb;
2054 #else
2055 void ll_umount_begin(struct super_block *sb)
2056 {
2057 #endif
2058         struct lustre_sb_info *lsi = s2lsi(sb);
2059         struct ll_sb_info *sbi = ll_s2sbi(sb);
2060         struct obd_device *obd;
2061         struct obd_ioctl_data ioc_data = { 0 };
2062         ENTRY;
2063
2064 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2065         if (!(flags & MNT_FORCE)) {
2066                 EXIT;
2067                 return;
2068         }
2069 #endif
2070
2071         /* Tell the MGC we got umount -f */
2072         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2073
2074         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2075                sb->s_count, atomic_read(&sb->s_active));
2076
2077         obd = class_exp2obd(sbi->ll_md_exp);
2078         if (obd == NULL) {
2079                 CERROR("Invalid MDC connection handle "LPX64"\n",
2080                        sbi->ll_md_exp->exp_handle.h_cookie);
2081                 EXIT;
2082                 return;
2083         }
2084         obd->obd_force = 1;
2085         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2086                       &ioc_data, NULL);
2087
2088         obd = class_exp2obd(sbi->ll_dt_exp);
2089         if (obd == NULL) {
2090                 CERROR("Invalid LOV connection handle "LPX64"\n",
2091                        sbi->ll_dt_exp->exp_handle.h_cookie);
2092                 EXIT;
2093                 return;
2094         }
2095
2096         obd->obd_force = 1;
2097         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2098                       &ioc_data, NULL);
2099
2100         /* Really, we'd like to wait until there are no requests outstanding,
2101          * and then continue.  For now, we just invalidate the requests,
2102          * schedule, and hope.
2103          */
2104         schedule();
2105
2106         EXIT;
2107 }
2108
2109 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2110 {
2111         struct ll_sb_info *sbi = ll_s2sbi(sb);
2112         int err;
2113         __u32 read_only;
2114
2115         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2116                 read_only = *flags & MS_RDONLY;
2117                 err = obd_set_info_async(sbi->ll_md_exp,
2118                                          sizeof(KEY_READ_ONLY),
2119                                          KEY_READ_ONLY, sizeof(read_only),
2120                                          &read_only, NULL);
2121                 if (err) {
2122                         CERROR("Failed to change the read-only flag during "
2123                                "remount: %d\n", err);
2124                         return err;
2125                 }
2126
2127                 if (read_only)
2128                         sb->s_flags |= MS_RDONLY;
2129                 else
2130                         sb->s_flags &= ~MS_RDONLY;
2131         }
2132         return 0;
2133 }
2134
2135 int ll_prep_inode(struct inode **inode,
2136                   struct ptlrpc_request *req,
2137                   struct super_block *sb)
2138 {
2139         struct ll_sb_info *sbi = NULL;
2140         struct lustre_md md;
2141         int rc = 0;
2142         ENTRY;
2143
2144         LASSERT(*inode || sb);
2145         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2146         prune_deathrow(sbi, 1);
2147         memset(&md, 0, sizeof(struct lustre_md));
2148
2149         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2150                               sbi->ll_md_exp, &md);
2151         if (rc)
2152                 RETURN(rc);
2153
2154         if (*inode) {
2155                 ll_update_inode(*inode, &md);
2156         } else {
2157                 LASSERT(sb != NULL);
2158
2159                 /*
2160                  * At this point server returns to client's same fid as client
2161                  * generated for creating. So using ->fid1 is okay here.
2162                  */
2163                 LASSERT(fid_is_sane(&md.body->fid1));
2164
2165                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2166                 if (*inode == NULL || is_bad_inode(*inode)) {
2167                         if (md.lsm)
2168                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2169 #ifdef CONFIG_FS_POSIX_ACL
2170                         if (md.posix_acl) {
2171                                 posix_acl_release(md.posix_acl);
2172                                 md.posix_acl = NULL;
2173                         }
2174 #endif
2175                         rc = -ENOMEM;
2176                         CERROR("new_inode -fatal: rc %d\n", rc);
2177                         GOTO(out, rc);
2178                 }
2179         }
2180
2181         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2182                          ll_i2info(*inode)->lli_smd);
2183 out:
2184         md_free_lustre_md(sbi->ll_md_exp, &md);
2185         RETURN(rc);
2186 }
2187
2188 char *llap_origins[] = {
2189         [LLAP_ORIGIN_UNKNOWN] = "--",
2190         [LLAP_ORIGIN_READPAGE] = "rp",
2191         [LLAP_ORIGIN_READAHEAD] = "ra",
2192         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2193         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2194         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2195 };
2196
2197 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2198                                              struct list_head *list)
2199 {
2200         struct ll_async_page *llap;
2201         struct list_head *pos;
2202
2203         list_for_each(pos, list) {
2204                 if (pos == &sbi->ll_pglist)
2205                         return NULL;
2206                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2207                 if (llap->llap_page == NULL)
2208                         continue;
2209                 return llap;
2210         }
2211         LBUG();
2212         return NULL;
2213 }
2214
2215 int ll_obd_statfs(struct inode *inode, void *arg)
2216 {
2217         struct ll_sb_info *sbi = NULL;
2218         struct obd_export *exp;
2219         char *buf = NULL;
2220         struct obd_ioctl_data *data = NULL;
2221         __u32 type;
2222         int len = 0, rc;
2223
2224         if (!inode || !(sbi = ll_i2sbi(inode)))
2225                 GOTO(out_statfs, rc = -EINVAL);
2226
2227         rc = obd_ioctl_getdata(&buf, &len, arg);
2228         if (rc)
2229                 GOTO(out_statfs, rc);
2230
2231         data = (void*)buf;
2232         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2233             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2234                 GOTO(out_statfs, rc = -EINVAL);
2235
2236         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2237         if (type == LL_STATFS_MDC)
2238                 exp = sbi->ll_md_exp;
2239         else if (type == LL_STATFS_LOV)
2240                 exp = sbi->ll_dt_exp;
2241         else 
2242                 GOTO(out_statfs, rc = -ENODEV);
2243
2244         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2245         if (rc)
2246                 GOTO(out_statfs, rc);
2247 out_statfs:
2248         if (buf)
2249                 obd_ioctl_freedata(buf, len);
2250         return rc;
2251 }
2252
2253 int ll_process_config(struct lustre_cfg *lcfg)
2254 {
2255         char *ptr;
2256         void *sb;
2257         struct lprocfs_static_vars lvars;
2258         unsigned long x; 
2259         int rc = 0;
2260
2261         lprocfs_llite_init_vars(&lvars);
2262
2263         /* The instance name contains the sb: lustre-client-aacfe000 */
2264         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2265         if (!ptr || !*(++ptr)) 
2266                 return -EINVAL;
2267         if (sscanf(ptr, "%lx", &x) != 1)
2268                 return -EINVAL;
2269         sb = (void *)x;
2270         /* This better be a real Lustre superblock! */
2271         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2272
2273         /* Note we have not called client_common_fill_super yet, so 
2274            proc fns must be able to handle that! */
2275         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2276                                       lcfg, sb);
2277         return(rc);
2278 }
2279
2280 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2281 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2282                                        struct inode *i1, struct inode *i2,
2283                                        const char *name, int namelen,
2284                                        int mode, __u32 opc, void *data)
2285 {
2286         LASSERT(i1 != NULL);
2287
2288         if (namelen > ll_i2sbi(i1)->ll_namelen)
2289                 return ERR_PTR(-ENAMETOOLONG);
2290         
2291         if (op_data == NULL)
2292                 OBD_ALLOC_PTR(op_data);
2293         
2294         if (op_data == NULL)
2295                 return ERR_PTR(-ENOMEM);
2296
2297         ll_i2gids(op_data->op_suppgids, i1, i2);
2298         op_data->op_fid1 = *ll_inode2fid(i1);
2299         op_data->op_capa1 = ll_mdscapa_get(i1);
2300
2301         if (i2) {
2302                 op_data->op_fid2 = *ll_inode2fid(i2);
2303                 op_data->op_capa2 = ll_mdscapa_get(i2);
2304         } else {
2305                 fid_zero(&op_data->op_fid2);
2306                 op_data->op_capa2 = NULL;
2307         }
2308
2309         op_data->op_name = name;
2310         op_data->op_namelen = namelen;
2311         op_data->op_mode = mode;
2312         op_data->op_mod_time = cfs_time_current_sec();
2313         op_data->op_fsuid = current->fsuid;
2314         op_data->op_fsgid = current->fsgid;
2315         op_data->op_cap = current->cap_effective;
2316         op_data->op_bias = MDS_CHECK_SPLIT;
2317         op_data->op_opc = opc;
2318         op_data->op_mds = 0;
2319         op_data->op_data = data;
2320
2321         return op_data;
2322 }
2323
2324 void ll_finish_md_op_data(struct md_op_data *op_data)
2325 {
2326         capa_put(op_data->op_capa1);
2327         capa_put(op_data->op_capa2);
2328         OBD_FREE_PTR(op_data);
2329 }