Whamcloud - gitweb
LU-761: Check return of ll_prep_md_op_data() using IS_ERR().
[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
20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
21  *
22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23  * CA 95054 USA or visit www.sun.com if you need additional information or
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30  * Use is subject to license terms.
31  */
32 /*
33  * Copyright (c) 2011 Whamcloud, Inc.
34  */
35 /*
36  * This file is part of Lustre, http://www.lustre.org/
37  * Lustre is a trademark of Sun Microsystems, Inc.
38  *
39  * lustre/llite/llite_lib.c
40  *
41  * Lustre Light Super operations
42  */
43
44 #define DEBUG_SUBSYSTEM S_LLITE
45
46 #include <linux/module.h>
47 #include <linux/types.h>
48 #include <linux/version.h>
49 #include <linux/mm.h>
50
51 #include <lustre_lite.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
61
62 cfs_mem_cache_t *ll_file_data_slab;
63
64 CFS_LIST_HEAD(ll_super_blocks);
65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
66
67 #ifndef MS_HAS_NEW_AOPS
68 extern struct address_space_operations ll_aops;
69 extern struct address_space_operations ll_dir_aops;
70 #else
71 extern struct address_space_operations_ext ll_aops;
72 extern struct address_space_operations_ext ll_dir_aops;
73 #endif
74
75 #ifndef log2
76 #define log2(n) cfs_ffz(~(n))
77 #endif
78
79 static struct ll_sb_info *ll_init_sbi(void)
80 {
81         struct ll_sb_info *sbi = NULL;
82         unsigned long pages;
83         struct sysinfo si;
84         class_uuid_t uuid;
85         int i;
86         ENTRY;
87
88         OBD_ALLOC(sbi, sizeof(*sbi));
89         if (!sbi)
90                 RETURN(NULL);
91
92         cfs_spin_lock_init(&sbi->ll_lock);
93         cfs_init_mutex(&sbi->ll_lco.lco_lock);
94         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
95         cfs_spin_lock_init(&sbi->ll_process_lock);
96         sbi->ll_rw_stats_on = 0;
97
98         si_meminfo(&si);
99         pages = si.totalram - si.totalhigh;
100         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
101 #ifdef HAVE_BGL_SUPPORT
102                 sbi->ll_async_page_max = pages / 4;
103 #else
104                 sbi->ll_async_page_max = pages / 2;
105 #endif
106         } else {
107                 sbi->ll_async_page_max = (pages / 4) * 3;
108         }
109
110         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
111                                            SBI_DEFAULT_READAHEAD_MAX);
112         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
113         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
114                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
115         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
116         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
117
118         ll_generate_random_uuid(uuid);
119         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
120         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
121
122         cfs_spin_lock(&ll_sb_lock);
123         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
124         cfs_spin_unlock(&ll_sb_lock);
125
126 #ifdef ENABLE_CHECKSUM
127         sbi->ll_flags |= LL_SBI_CHECKSUM;
128 #endif
129
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
132 #endif
133
134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
136                                    pp_r_hist.oh_lock);
137                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
138                                    pp_w_hist.oh_lock);
139         }
140
141         /* metadata statahead is enabled by default */
142         sbi->ll_sa_max = LL_SA_RPC_DEF;
143         cfs_atomic_set(&sbi->ll_sa_total, 0);
144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
145
146         RETURN(sbi);
147 }
148
149 void ll_free_sbi(struct super_block *sb)
150 {
151         struct ll_sb_info *sbi = ll_s2sbi(sb);
152         ENTRY;
153
154         if (sbi != NULL) {
155                 cfs_spin_lock(&ll_sb_lock);
156                 cfs_list_del(&sbi->ll_list);
157                 cfs_spin_unlock(&ll_sb_lock);
158                 OBD_FREE(sbi, sizeof(*sbi));
159         }
160         EXIT;
161 }
162
163 static struct dentry_operations ll_d_root_ops = {
164         .d_compare = ll_dcompare,
165         .d_revalidate = ll_revalidate_nd,
166 };
167
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169                                     struct vfsmount *mnt)
170 {
171         struct inode *root = 0;
172         struct ll_sb_info *sbi = ll_s2sbi(sb);
173         struct obd_device *obd;
174         struct obd_capa *oc = NULL;
175         struct obd_statfs *osfs = NULL;
176         struct ptlrpc_request *request = NULL;
177         struct obd_connect_data *data = NULL;
178         struct obd_uuid *uuid;
179         struct md_op_data *op_data;
180         struct lustre_md lmd;
181         obd_valid valid;
182         int size, err, checksum;
183         ENTRY;
184
185         obd = class_name2obd(md);
186         if (!obd) {
187                 CERROR("MD %s: not setup or attached\n", md);
188                 RETURN(-EINVAL);
189         }
190
191         OBD_ALLOC_PTR(data);
192         if (data == NULL)
193                 RETURN(-ENOMEM);
194
195         OBD_ALLOC_PTR(osfs);
196         if (osfs == NULL) {
197                 OBD_FREE_PTR(data);
198                 RETURN(-ENOMEM);
199         }
200
201         if (proc_lustre_fs_root) {
202                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
203                                                   dt, md);
204                 if (err < 0)
205                         CERROR("could not register mount in /proc/fs/lustre\n");
206         }
207
208         /* indicate the features supported by this client */
209         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
210                                   OBD_CONNECT_ATTRFID  |
211                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
212                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
213                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
214                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
215                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
216                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH;
217
218         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
219                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
220
221 #ifdef HAVE_LRU_RESIZE_SUPPORT
222         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
223                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
224 #endif
225 #ifdef CONFIG_FS_POSIX_ACL
226         data->ocd_connect_flags |= OBD_CONNECT_ACL;
227 #endif
228         data->ocd_ibits_known = MDS_INODELOCK_FULL;
229         data->ocd_version = LUSTRE_VERSION_CODE;
230
231         if (sb->s_flags & MS_RDONLY)
232                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
233         if (sbi->ll_flags & LL_SBI_USER_XATTR)
234                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
235
236 #ifdef HAVE_MS_FLOCK_LOCK
237         /* force vfs to use lustre handler for flock() calls - bug 10743 */
238         sb->s_flags |= MS_FLOCK_LOCK;
239 #endif
240 #ifdef MS_HAS_NEW_AOPS
241         sb->s_flags |= MS_HAS_NEW_AOPS;
242 #endif
243
244         if (sbi->ll_flags & LL_SBI_FLOCK)
245                 sbi->ll_fop = &ll_file_operations_flock;
246         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
247                 sbi->ll_fop = &ll_file_operations;
248         else
249                 sbi->ll_fop = &ll_file_operations_noflock;
250
251         /* real client */
252         data->ocd_connect_flags |= OBD_CONNECT_REAL;
253         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
254                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
255
256         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
257
258         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
259         if (err == -EBUSY) {
260                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
261                                    "recovery, of which this client is not a "
262                                    "part. Please wait for recovery to complete,"
263                                    " abort, or time out.\n", md);
264                 GOTO(out, err);
265         } else if (err) {
266                 CERROR("cannot connect to %s: rc = %d\n", md, err);
267                 GOTO(out, err);
268         }
269
270         err = obd_fid_init(sbi->ll_md_exp);
271         if (err) {
272                 CERROR("Can't init metadata layer FID infrastructure, "
273                        "rc %d\n", err);
274                 GOTO(out_md, err);
275         }
276
277         err = obd_statfs(obd, osfs,
278                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
279         if (err)
280                 GOTO(out_md_fid, err);
281
282         /* This needs to be after statfs to ensure connect has finished.
283          * Note that "data" does NOT contain the valid connect reply.
284          * If connecting to a 1.8 server there will be no LMV device, so
285          * we can access the MDC export directly and exp_connect_flags will
286          * be non-zero, but if accessing an upgraded 2.1 server it will
287          * have the correct flags filled in.
288          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
289         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
290         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
291             valid != CLIENT_CONNECT_MDT_REQD) {
292                 char *buf;
293
294                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
295                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
296                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
297                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
298                                    "feature(s) needed for correct operation "
299                                    "of this client (%s). Please upgrade "
300                                    "server or downgrade client.\n",
301                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
302                 OBD_FREE(buf, CFS_PAGE_SIZE);
303                 GOTO(out_md, err = -EPROTO);
304         }
305
306         size = sizeof(*data);
307         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
308                            KEY_CONN_DATA,  &size, data, NULL);
309         if (err) {
310                 CERROR("Get connect data failed: %d \n", err);
311                 GOTO(out_md, err);
312         }
313
314         LASSERT(osfs->os_bsize);
315         sb->s_blocksize = osfs->os_bsize;
316         sb->s_blocksize_bits = log2(osfs->os_bsize);
317         sb->s_magic = LL_SUPER_MAGIC;
318
319         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
320          *         retval = in_file->f_op->sendfile(...);
321          *         if (*ppos > max)
322          *                 retval = -EOVERFLOW;
323          *
324          * it will check if *ppos is greater than max. However, max equals to
325          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
326          * has been defined as a signed long long integer in linux kernel. */
327 #if BITS_PER_LONG == 64
328         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
329 #else
330         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
331 #endif
332         sbi->ll_namelen = osfs->os_namelen;
333         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
334
335         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
336             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
337                 LCONSOLE_INFO("Disabling user_xattr feature because "
338                               "it is not supported on the server\n");
339                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
340         }
341
342         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
343 #ifdef MS_POSIXACL
344                 sb->s_flags |= MS_POSIXACL;
345 #endif
346                 sbi->ll_flags |= LL_SBI_ACL;
347         } else {
348                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
349 #ifdef MS_POSIXACL
350                 sb->s_flags &= ~MS_POSIXACL;
351 #endif
352                 sbi->ll_flags &= ~LL_SBI_ACL;
353         }
354
355         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
356                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
357                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
358                         LCONSOLE_INFO("client is set as remote by default.\n");
359                 }
360         } else {
361                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
362                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
363                         LCONSOLE_INFO("client claims to be remote, but server "
364                                       "rejected, forced to be local.\n");
365                 }
366         }
367
368         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
369                 LCONSOLE_INFO("client enabled MDS capability!\n");
370                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
371         }
372
373         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
374                 LCONSOLE_INFO("client enabled OSS capability!\n");
375                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
376         }
377
378         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
379                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
380
381         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
382                 sbi->ll_md_brw_size = data->ocd_brw_size;
383         else
384                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
385
386         obd = class_name2obd(dt);
387         if (!obd) {
388                 CERROR("DT %s: not setup or attached\n", dt);
389                 GOTO(out_md_fid, err = -ENODEV);
390         }
391
392         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
393                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
394                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
395                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
396                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
397                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
398                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
399                                   OBD_CONNECT_MAXBYTES;
400
401         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
402                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
403
404         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
405                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
406                  * disabled by default, because it can still be enabled on the
407                  * fly via /proc. As a consequence, we still need to come to an
408                  * agreement on the supported algorithms at connect time */
409                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
410
411                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
412                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
413                 else
414                         data->ocd_cksum_types = cksum_types_supported();
415         }
416
417 #ifdef HAVE_LRU_RESIZE_SUPPORT
418         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
419 #endif
420         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
421                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
422
423         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
424                "ocd_grant: %d\n", data->ocd_connect_flags,
425                data->ocd_version, data->ocd_grant);
426
427         obd->obd_upcall.onu_owner = &sbi->ll_lco;
428         obd->obd_upcall.onu_upcall = cl_ocd_update;
429
430         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
431
432         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
433         if (err == -EBUSY) {
434                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
435                                    "recovery, of which this client is not a "
436                                    "part.  Please wait for recovery to "
437                                    "complete, abort, or time out.\n", dt);
438                 GOTO(out_md_fid, err);
439         } else if (err) {
440                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
441                 GOTO(out_md_fid, err);
442         }
443
444         err = obd_fid_init(sbi->ll_dt_exp);
445         if (err) {
446                 CERROR("Can't init data layer FID infrastructure, "
447                        "rc %d\n", err);
448                 GOTO(out_dt, err);
449         }
450
451         cfs_mutex_down(&sbi->ll_lco.lco_lock);
452         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
453         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
454         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
455         cfs_mutex_up(&sbi->ll_lco.lco_lock);
456
457         fid_zero(&sbi->ll_root_fid);
458         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
459         if (err) {
460                 CERROR("cannot mds_connect: rc = %d\n", err);
461                 GOTO(out_lock_cn_cb, err);
462         }
463         if (!fid_is_sane(&sbi->ll_root_fid)) {
464                 CERROR("Invalid root fid during mount\n");
465                 GOTO(out_lock_cn_cb, err = -EINVAL);
466         }
467         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
468
469         sb->s_op = &lustre_super_operations;
470 #if THREAD_SIZE >= 8192
471         sb->s_export_op = &lustre_export_operations;
472 #endif
473
474         /* make root inode
475          * XXX: move this to after cbd setup? */
476         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
477         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
478                 valid |= OBD_MD_FLRMTPERM;
479         else if (sbi->ll_flags & LL_SBI_ACL)
480                 valid |= OBD_MD_FLACL;
481
482         OBD_ALLOC_PTR(op_data);
483         if (op_data == NULL)
484                 GOTO(out_lock_cn_cb, err = -ENOMEM);
485
486         op_data->op_fid1 = sbi->ll_root_fid;
487         op_data->op_mode = 0;
488         op_data->op_capa1 = oc;
489         op_data->op_valid = valid;
490
491         err = md_getattr(sbi->ll_md_exp, op_data, &request);
492         if (oc)
493                 capa_put(oc);
494         OBD_FREE_PTR(op_data);
495         if (err) {
496                 CERROR("md_getattr failed for root: rc = %d\n", err);
497                 GOTO(out_lock_cn_cb, err);
498         }
499         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
500                                sbi->ll_md_exp, &lmd);
501         if (err) {
502                 CERROR("failed to understand root inode md: rc = %d\n", err);
503                 ptlrpc_req_finished (request);
504                 GOTO(out_lock_cn_cb, err);
505         }
506
507         LASSERT(fid_is_sane(&sbi->ll_root_fid));
508         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
509         md_free_lustre_md(sbi->ll_md_exp, &lmd);
510         ptlrpc_req_finished(request);
511
512         if (root == NULL || IS_ERR(root)) {
513                 if (lmd.lsm)
514                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
515 #ifdef CONFIG_FS_POSIX_ACL
516                 if (lmd.posix_acl) {
517                         posix_acl_release(lmd.posix_acl);
518                         lmd.posix_acl = NULL;
519                 }
520 #endif
521                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
522                 root = NULL;
523                 CERROR("lustre_lite: bad iget4 for root\n");
524                 GOTO(out_root, err);
525         }
526
527         err = ll_close_thread_start(&sbi->ll_lcq);
528         if (err) {
529                 CERROR("cannot start close thread: rc %d\n", err);
530                 GOTO(out_root, err);
531         }
532
533 #ifdef CONFIG_FS_POSIX_ACL
534         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
535                 rct_init(&sbi->ll_rct);
536                 et_init(&sbi->ll_et);
537         }
538 #endif
539
540         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
541         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
542                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
543                                  NULL);
544         cl_sb_init(sb);
545
546         sb->s_root = d_alloc_root(root);
547         sb->s_root->d_op = &ll_d_root_ops;
548
549         sbi->ll_sdev_orig = sb->s_dev;
550
551         /* We set sb->s_dev equal on all lustre clients in order to support
552          * NFS export clustering.  NFSD requires that the FSID be the same
553          * on all clients. */
554         /* s_dev is also used in lt_compare() to compare two fs, but that is
555          * only a node-local comparison. */
556         uuid = obd_get_uuid(sbi->ll_md_exp);
557         if (uuid != NULL)
558                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
559         sbi->ll_mnt = mnt;
560
561         if (data != NULL)
562                 OBD_FREE_PTR(data);
563         if (osfs != NULL)
564                 OBD_FREE_PTR(osfs);
565
566         RETURN(err);
567 out_root:
568         if (root)
569                 iput(root);
570 out_lock_cn_cb:
571         obd_fid_fini(sbi->ll_dt_exp);
572 out_dt:
573         obd_disconnect(sbi->ll_dt_exp);
574         sbi->ll_dt_exp = NULL;
575 out_md_fid:
576         obd_fid_fini(sbi->ll_md_exp);
577 out_md:
578         obd_disconnect(sbi->ll_md_exp);
579         sbi->ll_md_exp = NULL;
580 out:
581         if (data != NULL)
582                 OBD_FREE_PTR(data);
583         if (osfs != NULL)
584                 OBD_FREE_PTR(osfs);
585         lprocfs_unregister_mountpoint(sbi);
586         return err;
587 }
588
589 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
590 {
591         int size, rc;
592
593         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
594         size = sizeof(int);
595         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
596                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
597         if (rc)
598                 CERROR("Get max mdsize error rc %d \n", rc);
599
600         RETURN(rc);
601 }
602
603 void ll_dump_inode(struct inode *inode)
604 {
605         struct list_head *tmp;
606         int dentry_count = 0;
607
608         LASSERT(inode != NULL);
609
610         list_for_each(tmp, &inode->i_dentry)
611                 dentry_count++;
612
613         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
614                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
615                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
616 }
617
618 void lustre_dump_dentry(struct dentry *dentry, int recur)
619 {
620         struct list_head *tmp;
621         int subdirs = 0;
622
623         LASSERT(dentry != NULL);
624
625         list_for_each(tmp, &dentry->d_subdirs)
626                 subdirs++;
627
628         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
629                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
630                dentry->d_name.len, dentry->d_name.name,
631                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
632                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
633                dentry->d_flags, dentry->d_fsdata, subdirs);
634         if (dentry->d_inode != NULL)
635                 ll_dump_inode(dentry->d_inode);
636
637         if (recur == 0)
638                 return;
639
640        list_for_each(tmp, &dentry->d_subdirs) {
641                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
642                 lustre_dump_dentry(d, recur - 1);
643         }
644 }
645
646 void client_common_put_super(struct super_block *sb)
647 {
648         struct ll_sb_info *sbi = ll_s2sbi(sb);
649         ENTRY;
650
651 #ifdef CONFIG_FS_POSIX_ACL
652         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
653                 et_fini(&sbi->ll_et);
654                 rct_fini(&sbi->ll_rct);
655         }
656 #endif
657
658         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
659
660         ll_close_thread_shutdown(sbi->ll_lcq);
661
662         cl_sb_fini(sb);
663
664         cfs_list_del(&sbi->ll_conn_chain);
665
666         obd_fid_fini(sbi->ll_dt_exp);
667         obd_disconnect(sbi->ll_dt_exp);
668         sbi->ll_dt_exp = NULL;
669
670         lprocfs_unregister_mountpoint(sbi);
671
672         obd_fid_fini(sbi->ll_md_exp);
673         obd_disconnect(sbi->ll_md_exp);
674         sbi->ll_md_exp = NULL;
675
676         EXIT;
677 }
678
679 void ll_kill_super(struct super_block *sb)
680 {
681         struct ll_sb_info *sbi;
682
683         ENTRY;
684
685         /* not init sb ?*/
686         if (!(sb->s_flags & MS_ACTIVE))
687                 return;
688
689         sbi = ll_s2sbi(sb);
690         /* we need restore s_dev from changed for clustred NFS before put_super
691          * because new kernels have cached s_dev and change sb->s_dev in
692          * put_super not affected real removing devices */
693         if (sbi)
694                 sb->s_dev = sbi->ll_sdev_orig;
695         EXIT;
696 }
697
698 char *ll_read_opt(const char *opt, char *data)
699 {
700         char *value;
701         char *retval;
702         ENTRY;
703
704         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
705         if (strncmp(opt, data, strlen(opt)))
706                 RETURN(NULL);
707         if ((value = strchr(data, '=')) == NULL)
708                 RETURN(NULL);
709
710         value++;
711         OBD_ALLOC(retval, strlen(value) + 1);
712         if (!retval) {
713                 CERROR("out of memory!\n");
714                 RETURN(NULL);
715         }
716
717         memcpy(retval, value, strlen(value)+1);
718         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
719         RETURN(retval);
720 }
721
722 static inline int ll_set_opt(const char *opt, char *data, int fl)
723 {
724         if (strncmp(opt, data, strlen(opt)) != 0)
725                 return(0);
726         else
727                 return(fl);
728 }
729
730 /* non-client-specific mount options are parsed in lmd_parse */
731 static int ll_options(char *options, int *flags)
732 {
733         int tmp;
734         char *s1 = options, *s2;
735         ENTRY;
736
737         if (!options)
738                 RETURN(0);
739
740         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
741
742         while (*s1) {
743                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
744                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
745                 if (tmp) {
746                         *flags |= tmp;
747                         goto next;
748                 }
749                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
750                 if (tmp) {
751                         *flags |= tmp;
752                         goto next;
753                 }
754                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
755                 if (tmp) {
756                         *flags |= tmp;
757                         goto next;
758                 }
759                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
760                 if (tmp) {
761                         *flags &= ~tmp;
762                         goto next;
763                 }
764                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
765                 if (tmp) {
766                         *flags |= tmp;
767                         goto next;
768                 }
769                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
770                 if (tmp) {
771                         *flags &= ~tmp;
772                         goto next;
773                 }
774 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
775                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
776                 if (tmp) {
777                         /* Ignore deprecated mount option.  The client will
778                          * always try to mount with ACL support, whether this
779                          * is used depends on whether server supports it. */
780                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
781                                                   "mount option 'acl'.\n");
782                         goto next;
783                 }
784                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
785                 if (tmp) {
786                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
787                                                   "mount option 'noacl'.\n");
788                         goto next;
789                 }
790 #else
791 #warning "{no}acl options have been deprecated since 1.8, please remove them"
792 #endif
793                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
794                 if (tmp) {
795                         *flags |= tmp;
796                         goto next;
797                 }
798
799                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
800                 if (tmp) {
801                         *flags |= tmp;
802                         goto next;
803                 }
804                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
805                 if (tmp) {
806                         *flags &= ~tmp;
807                         goto next;
808                 }
809                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
810                 if (tmp) {
811                         *flags |= tmp;
812                         goto next;
813                 }
814                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
815                 if (tmp) {
816                         *flags &= ~tmp;
817                         goto next;
818                 }
819                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
820                 if (tmp) {
821                         *flags |= tmp;
822                         goto next;
823                 }
824                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
825                 if (tmp) {
826                         *flags &= ~tmp;
827                         goto next;
828                 }
829                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
830                 if (tmp) {
831                         *flags |= tmp;
832                         goto next;
833                 }
834                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
835                 if (tmp) {
836                         *flags |= tmp;
837                         goto next;
838                 }
839
840                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
841                                    s1);
842                 RETURN(-EINVAL);
843
844 next:
845                 /* Find next opt */
846                 s2 = strchr(s1, ',');
847                 if (s2 == NULL)
848                         break;
849                 s1 = s2 + 1;
850         }
851         RETURN(0);
852 }
853
854 void ll_lli_init(struct ll_inode_info *lli)
855 {
856         lli->lli_inode_magic = LLI_INODE_MAGIC;
857         cfs_sema_init(&lli->lli_size_sem, 1);
858         cfs_sema_init(&lli->lli_write_sem, 1);
859         cfs_init_rwsem(&lli->lli_trunc_sem);
860         lli->lli_flags = 0;
861         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
862         cfs_spin_lock_init(&lli->lli_lock);
863         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
864         lli->lli_inode_magic = LLI_INODE_MAGIC;
865         cfs_sema_init(&lli->lli_och_sem, 1);
866         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
867         lli->lli_mds_exec_och = NULL;
868         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
869         lli->lli_open_fd_exec_count = 0;
870         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
871         lli->lli_remote_perms = NULL;
872         lli->lli_rmtperm_utime = 0;
873         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
874         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
875         cfs_spin_lock_init(&lli->lli_sa_lock);
876         lli->lli_clob = NULL;
877         cfs_sema_init(&lli->lli_readdir_sem, 1);
878         fid_zero(&lli->lli_pfid);
879 }
880
881 static inline int ll_bdi_register(struct backing_dev_info *bdi)
882 {
883 #ifdef HAVE_BDI_REGISTER
884         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
885
886 #ifdef HAVE_BDI_NAME
887         bdi->name = "lustre";
888 #endif
889         return bdi_register(bdi, NULL, "lustre-%d",
890                             atomic_inc_return(&ll_bdi_num));
891 #else
892         return 0;
893 #endif
894 }
895
896 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
897 {
898         struct lustre_profile *lprof = NULL;
899         struct lustre_sb_info *lsi = s2lsi(sb);
900         struct ll_sb_info *sbi;
901         char  *dt = NULL, *md = NULL;
902         char  *profilenm = get_profile_name(sb);
903         struct config_llog_instance *cfg;
904         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
905         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
906         int    err;
907         ENTRY;
908
909         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
910
911         OBD_ALLOC_PTR(cfg);
912         if (cfg == NULL)
913                 RETURN(-ENOMEM);
914
915         cfs_module_get();
916
917         /* client additional sb info */
918         lsi->lsi_llsbi = sbi = ll_init_sbi();
919         if (!sbi) {
920                 cfs_module_put(THIS_MODULE);
921                 OBD_FREE_PTR(cfg);
922                 RETURN(-ENOMEM);
923         }
924
925         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
926         if (err)
927                 GOTO(out_free, err);
928
929         err = ll_bdi_init(&lsi->lsi_bdi);
930         if (err)
931                 GOTO(out_free, err);
932         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
933         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
934         err = ll_bdi_register(&lsi->lsi_bdi);
935         if (err)
936                 GOTO(out_free, err);
937
938 #ifdef HAVE_SB_BDI
939         sb->s_bdi = &lsi->lsi_bdi;
940 #endif
941
942         /* Generate a string unique to this super, in case some joker tries
943            to mount the same fs at two mount points.
944            Use the address of the super itself.*/
945         cfg->cfg_instance = sb;
946         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
947
948         /* set up client obds */
949         err = lustre_process_log(sb, profilenm, cfg);
950         if (err < 0) {
951                 CERROR("Unable to process log: %d\n", err);
952                 GOTO(out_free, err);
953         }
954
955         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
956         lprof = class_get_profile(profilenm);
957         if (lprof == NULL) {
958                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
959                                    " read from the MGS.  Does that filesystem "
960                                    "exist?\n", profilenm);
961                 GOTO(out_free, err = -EINVAL);
962         }
963         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
964                lprof->lp_md, lprof->lp_dt);
965
966         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
967         if (!dt)
968                 GOTO(out_free, err = -ENOMEM);
969         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
970
971         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
972         if (!md)
973                 GOTO(out_free, err = -ENOMEM);
974         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
975
976         /* connections, registrations, sb setup */
977         err = client_common_fill_super(sb, md, dt, mnt);
978
979 out_free:
980         if (md)
981                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
982         if (dt)
983                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
984         if (err)
985                 ll_put_super(sb);
986         else
987                 LCONSOLE_WARN("Client %s has started\n", profilenm);
988
989         OBD_FREE_PTR(cfg);
990         RETURN(err);
991 } /* ll_fill_super */
992
993
994 void lu_context_keys_dump(void);
995
996 void ll_put_super(struct super_block *sb)
997 {
998         struct config_llog_instance cfg;
999         struct obd_device *obd;
1000         struct lustre_sb_info *lsi = s2lsi(sb);
1001         struct ll_sb_info *sbi = ll_s2sbi(sb);
1002         char *profilenm = get_profile_name(sb);
1003         int force = 1, next;
1004         ENTRY;
1005
1006         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1007
1008         ll_print_capa_stat(sbi);
1009
1010         cfg.cfg_instance = sb;
1011         lustre_end_log(sb, profilenm, &cfg);
1012
1013         if (sbi->ll_md_exp) {
1014                 obd = class_exp2obd(sbi->ll_md_exp);
1015                 if (obd)
1016                         force = obd->obd_force;
1017         }
1018
1019         /* We need to set force before the lov_disconnect in
1020            lustre_common_put_super, since l_d cleans up osc's as well. */
1021         if (force) {
1022                 next = 0;
1023                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1024                                                      &next)) != NULL) {
1025                         obd->obd_force = force;
1026                 }
1027         }
1028
1029         if (sbi->ll_lcq) {
1030                 /* Only if client_common_fill_super succeeded */
1031                 client_common_put_super(sb);
1032         }
1033
1034         next = 0;
1035         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1036                 class_manual_cleanup(obd);
1037         }
1038
1039         if (profilenm)
1040                 class_del_profile(profilenm);
1041
1042         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1043                 ll_bdi_destroy(&lsi->lsi_bdi);
1044                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1045         }
1046
1047         ll_free_sbi(sb);
1048         lsi->lsi_llsbi = NULL;
1049
1050         lustre_common_put_super(sb);
1051
1052         cl_env_cache_purge(~0);
1053
1054         LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1055
1056         cfs_module_put(THIS_MODULE);
1057
1058         EXIT;
1059 } /* client_put_super */
1060
1061 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1062 {
1063         struct inode *inode = NULL;
1064         /* NOTE: we depend on atomic igrab() -bzzz */
1065         lock_res_and_lock(lock);
1066         if (lock->l_ast_data) {
1067                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1068                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1069                         inode = igrab(lock->l_ast_data);
1070                 } else {
1071                         inode = lock->l_ast_data;
1072                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1073                                          D_WARNING, lock, "l_ast_data %p is "
1074                                          "bogus: magic %08x", lock->l_ast_data,
1075                                          lli->lli_inode_magic);
1076                         inode = NULL;
1077                 }
1078         }
1079         unlock_res_and_lock(lock);
1080         return inode;
1081 }
1082
1083 static int null_if_equal(struct ldlm_lock *lock, void *data)
1084 {
1085         if (data == lock->l_ast_data) {
1086                 lock->l_ast_data = NULL;
1087
1088                 if (lock->l_req_mode != lock->l_granted_mode)
1089                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1090         }
1091
1092         return LDLM_ITER_CONTINUE;
1093 }
1094
1095 void ll_clear_inode(struct inode *inode)
1096 {
1097         struct ll_inode_info *lli = ll_i2info(inode);
1098         struct ll_sb_info *sbi = ll_i2sbi(inode);
1099         ENTRY;
1100
1101         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1102                inode->i_generation, inode);
1103
1104         if (S_ISDIR(inode->i_mode)) {
1105                 /* these should have been cleared in ll_file_release */
1106                 LASSERT(lli->lli_sai == NULL);
1107                 LASSERT(lli->lli_opendir_key == NULL);
1108                 LASSERT(lli->lli_opendir_pid == 0);
1109         }
1110
1111         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1112         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1113                          null_if_equal, inode);
1114
1115         LASSERT(!lli->lli_open_fd_write_count);
1116         LASSERT(!lli->lli_open_fd_read_count);
1117         LASSERT(!lli->lli_open_fd_exec_count);
1118
1119         if (lli->lli_mds_write_och)
1120                 ll_md_real_close(inode, FMODE_WRITE);
1121         if (lli->lli_mds_exec_och)
1122                 ll_md_real_close(inode, FMODE_EXEC);
1123         if (lli->lli_mds_read_och)
1124                 ll_md_real_close(inode, FMODE_READ);
1125
1126         if (lli->lli_symlink_name) {
1127                 OBD_FREE(lli->lli_symlink_name,
1128                          strlen(lli->lli_symlink_name) + 1);
1129                 lli->lli_symlink_name = NULL;
1130         }
1131
1132         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1133                 LASSERT(lli->lli_posix_acl == NULL);
1134                 if (lli->lli_remote_perms) {
1135                         free_rmtperm_hash(lli->lli_remote_perms);
1136                         lli->lli_remote_perms = NULL;
1137                 }
1138         }
1139 #ifdef CONFIG_FS_POSIX_ACL
1140         else if (lli->lli_posix_acl) {
1141                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1142                 LASSERT(lli->lli_remote_perms == NULL);
1143                 posix_acl_release(lli->lli_posix_acl);
1144                 lli->lli_posix_acl = NULL;
1145         }
1146 #endif
1147         lli->lli_inode_magic = LLI_INODE_DEAD;
1148
1149         ll_clear_inode_capas(inode);
1150         /*
1151          * XXX This has to be done before lsm is freed below, because
1152          * cl_object still uses inode lsm.
1153          */
1154         cl_inode_fini(inode);
1155
1156         if (lli->lli_smd) {
1157                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1158                 lli->lli_smd = NULL;
1159         }
1160
1161
1162         EXIT;
1163 }
1164
1165 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1166                   struct md_open_data **mod)
1167 {
1168         struct lustre_md md;
1169         struct ll_sb_info *sbi = ll_i2sbi(inode);
1170         struct ptlrpc_request *request = NULL;
1171         int rc;
1172         ENTRY;
1173
1174         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1175                                      LUSTRE_OPC_ANY, NULL);
1176         if (IS_ERR(op_data))
1177                 RETURN(PTR_ERR(op_data));
1178
1179         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1180                         &request, mod);
1181         if (rc) {
1182                 ptlrpc_req_finished(request);
1183                 if (rc == -ENOENT) {
1184                         inode->i_nlink = 0;
1185                         /* Unlinked special device node? Or just a race?
1186                          * Pretend we done everything. */
1187                         if (!S_ISREG(inode->i_mode) &&
1188                             !S_ISDIR(inode->i_mode))
1189                                 rc = inode_setattr(inode, &op_data->op_attr);
1190                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1191                         CERROR("md_setattr fails: rc = %d\n", rc);
1192                 }
1193                 RETURN(rc);
1194         }
1195
1196         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1197                               sbi->ll_md_exp, &md);
1198         if (rc) {
1199                 ptlrpc_req_finished(request);
1200                 RETURN(rc);
1201         }
1202
1203         /* We call inode_setattr to adjust timestamps.
1204          * If there is at least some data in file, we cleared ATTR_SIZE
1205          * above to avoid invoking vmtruncate, otherwise it is important
1206          * to call vmtruncate in inode_setattr to update inode->i_size
1207          * (bug 6196) */
1208         rc = inode_setattr(inode, &op_data->op_attr);
1209
1210         /* Extract epoch data if obtained. */
1211         op_data->op_handle = md.body->handle;
1212         op_data->op_ioepoch = md.body->ioepoch;
1213
1214         ll_update_inode(inode, &md);
1215         ptlrpc_req_finished(request);
1216
1217         RETURN(rc);
1218 }
1219
1220 /* Close IO epoch and send Size-on-MDS attribute update. */
1221 static int ll_setattr_done_writing(struct inode *inode,
1222                                    struct md_op_data *op_data,
1223                                    struct md_open_data *mod)
1224 {
1225         struct ll_inode_info *lli = ll_i2info(inode);
1226         int rc = 0;
1227         ENTRY;
1228
1229         LASSERT(op_data != NULL);
1230         if (!S_ISREG(inode->i_mode))
1231                 RETURN(0);
1232
1233         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1234                op_data->op_ioepoch, PFID(&lli->lli_fid));
1235
1236         op_data->op_flags = MF_EPOCH_CLOSE;
1237         ll_done_writing_attr(inode, op_data);
1238         ll_pack_inode2opdata(inode, op_data, NULL);
1239
1240         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1241         if (rc == -EAGAIN) {
1242                 /* MDS has instructed us to obtain Size-on-MDS attribute
1243                  * from OSTs and send setattr to back to MDS. */
1244                 rc = ll_som_update(inode, op_data);
1245         } else if (rc) {
1246                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1247                        inode->i_ino, rc);
1248         }
1249         RETURN(rc);
1250 }
1251
1252 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1253 {
1254         struct obd_capa *capa;
1255         int rc;
1256
1257         if (attr->ia_valid & ATTR_SIZE)
1258                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1259         else
1260                 capa = ll_mdscapa_get(inode);
1261
1262         rc = cl_setattr_ost(inode, attr, capa);
1263
1264         if (attr->ia_valid & ATTR_SIZE)
1265                 ll_truncate_free_capa(capa);
1266         else
1267                 capa_put(capa);
1268
1269         return rc;
1270 }
1271
1272 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1273  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1274  * keep these values until such a time that objects are allocated for it.
1275  * We do the MDS operations first, as it is checking permissions for us.
1276  * We don't to the MDS RPC if there is nothing that we want to store there,
1277  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1278  * going to do an RPC anyways.
1279  *
1280  * If we are doing a truncate, we will send the mtime and ctime updates
1281  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1282  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1283  * at the same time.
1284  */
1285 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1286 {
1287         struct ll_inode_info *lli = ll_i2info(inode);
1288         struct lov_stripe_md *lsm = lli->lli_smd;
1289         struct md_op_data *op_data = NULL;
1290         struct md_open_data *mod = NULL;
1291         int ia_valid = attr->ia_valid;
1292         int rc = 0, rc1 = 0;
1293         ENTRY;
1294
1295         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1296                attr->ia_valid);
1297         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1298
1299         if (ia_valid & ATTR_SIZE) {
1300                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1301                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1302                                attr->ia_size, ll_file_maxbytes(inode));
1303                         RETURN(-EFBIG);
1304                 }
1305
1306                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1307         }
1308
1309         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1310         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1311                 if (cfs_curproc_fsuid() != inode->i_uid &&
1312                     !cfs_capable(CFS_CAP_FOWNER))
1313                         RETURN(-EPERM);
1314         }
1315
1316         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1317         if (attr->ia_valid & ATTR_CTIME) {
1318                 attr->ia_ctime = CFS_CURRENT_TIME;
1319                 attr->ia_valid |= ATTR_CTIME_SET;
1320         }
1321         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1322                 attr->ia_atime = CFS_CURRENT_TIME;
1323                 attr->ia_valid |= ATTR_ATIME_SET;
1324         }
1325         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1326                 attr->ia_mtime = CFS_CURRENT_TIME;
1327                 attr->ia_valid |= ATTR_MTIME_SET;
1328         }
1329
1330         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1331                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1332                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1333                        cfs_time_current_sec());
1334
1335         /* NB: ATTR_SIZE will only be set after this point if the size
1336          * resides on the MDS, ie, this file has no objects. */
1337         if (lsm)
1338                 attr->ia_valid &= ~ATTR_SIZE;
1339
1340         /* We always do an MDS RPC, even if we're only changing the size;
1341          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1342
1343         OBD_ALLOC_PTR(op_data);
1344         if (op_data == NULL)
1345                 RETURN(-ENOMEM);
1346
1347         UNLOCK_INODE_MUTEX(inode);
1348         if (ia_valid & ATTR_SIZE)
1349                 UP_WRITE_I_ALLOC_SEM(inode);
1350         cfs_down_write(&lli->lli_trunc_sem);
1351         LOCK_INODE_MUTEX(inode);
1352         if (ia_valid & ATTR_SIZE)
1353                 DOWN_WRITE_I_ALLOC_SEM(inode);
1354
1355         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1356
1357         /* Open epoch for truncate. */
1358         if (exp_connect_som(ll_i2mdexp(inode)) &&
1359             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1360                 op_data->op_flags = MF_EPOCH_OPEN;
1361
1362         rc = ll_md_setattr(inode, op_data, &mod);
1363         if (rc)
1364                 GOTO(out, rc);
1365
1366         ll_ioepoch_open(lli, op_data->op_ioepoch);
1367         if (!lsm || !S_ISREG(inode->i_mode)) {
1368                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1369                 GOTO(out, rc = 0);
1370         }
1371
1372         if (ia_valid & ATTR_SIZE)
1373                 attr->ia_valid |= ATTR_SIZE;
1374         if (ia_valid & (ATTR_SIZE |
1375                         ATTR_ATIME | ATTR_ATIME_SET |
1376                         ATTR_MTIME | ATTR_MTIME_SET))
1377                 /* on truncate and utimes send attributes to osts, setting
1378                  * mtime/atime to past will be performed under PW 0:EOF extent
1379                  * lock (new_size:EOF for truncate)
1380                  * it may seem excessive to send mtime/atime updates to osts
1381                  * when not setting times to past, but it is necessary due to
1382                  * possible time de-synchronization */
1383                 rc = ll_setattr_ost(inode, attr);
1384         EXIT;
1385 out:
1386         if (op_data) {
1387                 if (op_data->op_ioepoch)
1388                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1389                 ll_finish_md_op_data(op_data);
1390         }
1391         cfs_up_write(&lli->lli_trunc_sem);
1392         return rc ? rc : rc1;
1393 }
1394
1395 int ll_setattr(struct dentry *de, struct iattr *attr)
1396 {
1397         int mode = de->d_inode->i_mode;
1398
1399         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1400                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1401                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1402
1403         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1404                                (ATTR_SIZE|ATTR_MODE)) &&
1405             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1406              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1407               !(attr->ia_mode & S_ISGID))))
1408                 attr->ia_valid |= ATTR_FORCE;
1409
1410         if ((mode & S_ISUID) &&
1411             !(attr->ia_mode & S_ISUID) &&
1412             !(attr->ia_valid & ATTR_KILL_SUID))
1413                 attr->ia_valid |= ATTR_KILL_SUID;
1414
1415         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1416             !(attr->ia_mode & S_ISGID) &&
1417             !(attr->ia_valid & ATTR_KILL_SGID))
1418                 attr->ia_valid |= ATTR_KILL_SGID;
1419
1420         return ll_setattr_raw(de->d_inode, attr);
1421 }
1422
1423 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1424                        __u64 max_age, __u32 flags)
1425 {
1426         struct ll_sb_info *sbi = ll_s2sbi(sb);
1427         struct obd_statfs obd_osfs;
1428         int rc;
1429         ENTRY;
1430
1431         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1432         if (rc) {
1433                 CERROR("md_statfs fails: rc = %d\n", rc);
1434                 RETURN(rc);
1435         }
1436
1437         osfs->os_type = sb->s_magic;
1438
1439         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1440                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1441
1442         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1443                 flags |= OBD_STATFS_NODELAY;
1444
1445         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1446                               &obd_osfs, max_age, flags);
1447         if (rc) {
1448                 CERROR("obd_statfs fails: rc = %d\n", rc);
1449                 RETURN(rc);
1450         }
1451
1452         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1453                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1454                obd_osfs.os_files);
1455
1456         osfs->os_bsize = obd_osfs.os_bsize;
1457         osfs->os_blocks = obd_osfs.os_blocks;
1458         osfs->os_bfree = obd_osfs.os_bfree;
1459         osfs->os_bavail = obd_osfs.os_bavail;
1460
1461         /* If we don't have as many objects free on the OST as inodes
1462          * on the MDS, we reduce the total number of inodes to
1463          * compensate, so that the "inodes in use" number is correct.
1464          */
1465         if (obd_osfs.os_ffree < osfs->os_ffree) {
1466                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1467                         obd_osfs.os_ffree;
1468                 osfs->os_ffree = obd_osfs.os_ffree;
1469         }
1470
1471         RETURN(rc);
1472 }
1473 #ifndef HAVE_STATFS_DENTRY_PARAM
1474 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1475 {
1476 #else
1477 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1478 {
1479         struct super_block *sb = de->d_sb;
1480 #endif
1481         struct obd_statfs osfs;
1482         int rc;
1483
1484         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1485         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1486
1487         /* Some amount of caching on the client is allowed */
1488         rc = ll_statfs_internal(sb, &osfs,
1489                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1490                                 0);
1491         if (rc)
1492                 return rc;
1493
1494         statfs_unpack(sfs, &osfs);
1495
1496         /* We need to downshift for all 32-bit kernels, because we can't
1497          * tell if the kernel is being called via sys_statfs64() or not.
1498          * Stop before overflowing f_bsize - in which case it is better
1499          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1500         if (sizeof(long) < 8) {
1501                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1502                         sfs->f_bsize <<= 1;
1503
1504                         osfs.os_blocks >>= 1;
1505                         osfs.os_bfree >>= 1;
1506                         osfs.os_bavail >>= 1;
1507                 }
1508         }
1509
1510         sfs->f_blocks = osfs.os_blocks;
1511         sfs->f_bfree = osfs.os_bfree;
1512         sfs->f_bavail = osfs.os_bavail;
1513
1514         return 0;
1515 }
1516
1517 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1518 {
1519         struct ll_inode_info *lli;
1520         struct lov_stripe_md *lsm;
1521
1522         lli = ll_i2info(inode);
1523         LASSERT(lli->lli_size_sem_owner != current);
1524         cfs_down(&lli->lli_size_sem);
1525         LASSERT(lli->lli_size_sem_owner == NULL);
1526         lli->lli_size_sem_owner = current;
1527         lsm = lli->lli_smd;
1528         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1529                  lsm, lock_lsm);
1530         if (lock_lsm)
1531                 lov_stripe_lock(lsm);
1532 }
1533
1534 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1535 {
1536         struct ll_inode_info *lli;
1537         struct lov_stripe_md *lsm;
1538
1539         lli = ll_i2info(inode);
1540         lsm = lli->lli_smd;
1541         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1542                  lsm, unlock_lsm);
1543         if (unlock_lsm)
1544                 lov_stripe_unlock(lsm);
1545         LASSERT(lli->lli_size_sem_owner == current);
1546         lli->lli_size_sem_owner = NULL;
1547         cfs_up(&lli->lli_size_sem);
1548 }
1549
1550 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1551 {
1552         struct ll_inode_info *lli = ll_i2info(inode);
1553         struct mdt_body *body = md->body;
1554         struct lov_stripe_md *lsm = md->lsm;
1555         struct ll_sb_info *sbi = ll_i2sbi(inode);
1556
1557         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1558         if (lsm != NULL) {
1559                 cfs_down(&lli->lli_och_sem);
1560                 if (lli->lli_smd == NULL) {
1561                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1562                             lsm->lsm_magic != LOV_MAGIC_V3) {
1563                                 dump_lsm(D_ERROR, lsm);
1564                                 LBUG();
1565                         }
1566                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1567                                lsm, inode->i_ino, inode->i_generation, inode);
1568                         /* cl_inode_init must go before lli_smd or a race is
1569                          * possible where client thinks the file has stripes,
1570                          * but lov raid0 is not setup yet and parallel e.g.
1571                          * glimpse would try to use uninitialized lov */
1572                         cl_inode_init(inode, md);
1573                         cfs_spin_lock(&lli->lli_lock);
1574                         lli->lli_smd = lsm;
1575                         cfs_spin_unlock(&lli->lli_lock);
1576                         cfs_up(&lli->lli_och_sem);
1577                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1578                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1579                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1580                 } else {
1581                         cfs_up(&lli->lli_och_sem);
1582                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1583                                 lli->lli_smd->lsm_stripe_count ==
1584                                 lsm->lsm_stripe_count);
1585                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1586                                 CERROR("lsm mismatch for inode %ld\n",
1587                                        inode->i_ino);
1588                                 CERROR("lli_smd:\n");
1589                                 dump_lsm(D_ERROR, lli->lli_smd);
1590                                 CERROR("lsm:\n");
1591                                 dump_lsm(D_ERROR, lsm);
1592                                 LBUG();
1593                         }
1594                 }
1595                 if (lli->lli_smd != lsm)
1596                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1597         }
1598
1599         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1600                 if (body->valid & OBD_MD_FLRMTPERM)
1601                         ll_update_remote_perm(inode, md->remote_perm);
1602         }
1603 #ifdef CONFIG_FS_POSIX_ACL
1604         else if (body->valid & OBD_MD_FLACL) {
1605                 cfs_spin_lock(&lli->lli_lock);
1606                 if (lli->lli_posix_acl)
1607                         posix_acl_release(lli->lli_posix_acl);
1608                 lli->lli_posix_acl = md->posix_acl;
1609                 cfs_spin_unlock(&lli->lli_lock);
1610         }
1611 #endif
1612         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1613         inode->i_generation = cl_fid_build_gen(&body->fid1);
1614
1615         if (body->valid & OBD_MD_FLATIME) {
1616                 if (body->atime > LTIME_S(inode->i_atime))
1617                         LTIME_S(inode->i_atime) = body->atime;
1618                 lli->lli_lvb.lvb_atime = body->atime;
1619         }
1620         if (body->valid & OBD_MD_FLMTIME) {
1621                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1622                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1623                                "to "LPU64"\n", inode->i_ino,
1624                                LTIME_S(inode->i_mtime), body->mtime);
1625                         LTIME_S(inode->i_mtime) = body->mtime;
1626                 }
1627                 lli->lli_lvb.lvb_mtime = body->mtime;
1628         }
1629         if (body->valid & OBD_MD_FLCTIME) {
1630                 if (body->ctime > LTIME_S(inode->i_ctime))
1631                         LTIME_S(inode->i_ctime) = body->ctime;
1632                 lli->lli_lvb.lvb_ctime = body->ctime;
1633         }
1634         if (body->valid & OBD_MD_FLMODE)
1635                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1636         if (body->valid & OBD_MD_FLTYPE)
1637                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1638         LASSERT(inode->i_mode != 0);
1639         if (S_ISREG(inode->i_mode)) {
1640                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1641         } else {
1642                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1643         }
1644 #ifdef HAVE_INODE_BLKSIZE
1645         inode->i_blksize = 1<<inode->i_blkbits;
1646 #endif
1647         if (body->valid & OBD_MD_FLUID)
1648                 inode->i_uid = body->uid;
1649         if (body->valid & OBD_MD_FLGID)
1650                 inode->i_gid = body->gid;
1651         if (body->valid & OBD_MD_FLFLAGS)
1652                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1653         if (body->valid & OBD_MD_FLNLINK)
1654                 inode->i_nlink = body->nlink;
1655         if (body->valid & OBD_MD_FLRDEV)
1656                 inode->i_rdev = old_decode_dev(body->rdev);
1657
1658         if (body->valid & OBD_MD_FLID) {
1659                 /* FID shouldn't be changed! */
1660                 if (fid_is_sane(&lli->lli_fid)) {
1661                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1662                                  "Trying to change FID "DFID
1663                                  " to the "DFID", inode %lu/%u(%p)\n",
1664                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1665                                  inode->i_ino, inode->i_generation, inode);
1666                 } else
1667                         lli->lli_fid = body->fid1;
1668         }
1669
1670         LASSERT(fid_seq(&lli->lli_fid) != 0);
1671
1672         if (body->valid & OBD_MD_FLSIZE) {
1673                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1674                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1675                         struct lustre_handle lockh;
1676                         ldlm_mode_t mode;
1677
1678                         /* As it is possible a blocking ast has been processed
1679                          * by this time, we need to check there is an UPDATE
1680                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1681                          * it. */
1682                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1683                                                &lockh);
1684                         if (mode) {
1685                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1686                                                       LLIF_EPOCH_PENDING |
1687                                                       LLIF_SOM_DIRTY)) {
1688                                         CERROR("ino %lu flags %lu still has "
1689                                                "size authority! do not trust "
1690                                                "the size got from MDS\n",
1691                                                inode->i_ino, lli->lli_flags);
1692                                 } else {
1693                                         /* Use old size assignment to avoid
1694                                          * deadlock bz14138 & bz14326 */
1695                                         inode->i_size = body->size;
1696                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1697                                 }
1698                                 ldlm_lock_decref(&lockh, mode);
1699                         }
1700                 } else {
1701                         /* Use old size assignment to avoid
1702                          * deadlock bz14138 & bz14326 */
1703                         inode->i_size = body->size;
1704
1705                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1706                                inode->i_ino, (unsigned long long)body->size);
1707                 }
1708
1709                 if (body->valid & OBD_MD_FLBLOCKS)
1710                         inode->i_blocks = body->blocks;
1711         }
1712
1713         if (body->valid & OBD_MD_FLMDSCAPA) {
1714                 LASSERT(md->mds_capa);
1715                 ll_add_capa(inode, md->mds_capa);
1716         }
1717         if (body->valid & OBD_MD_FLOSSCAPA) {
1718                 LASSERT(md->oss_capa);
1719                 ll_add_capa(inode, md->oss_capa);
1720         }
1721 }
1722
1723 void ll_read_inode2(struct inode *inode, void *opaque)
1724 {
1725         struct lustre_md *md = opaque;
1726         struct ll_inode_info *lli = ll_i2info(inode);
1727         ENTRY;
1728
1729         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1730                PFID(&lli->lli_fid), inode);
1731
1732         ll_lli_init(lli);
1733
1734         LASSERT(!lli->lli_smd);
1735
1736         /* Core attributes from the MDS first.  This is a new inode, and
1737          * the VFS doesn't zero times in the core inode so we have to do
1738          * it ourselves.  They will be overwritten by either MDS or OST
1739          * attributes - we just need to make sure they aren't newer. */
1740         LTIME_S(inode->i_mtime) = 0;
1741         LTIME_S(inode->i_atime) = 0;
1742         LTIME_S(inode->i_ctime) = 0;
1743         inode->i_rdev = 0;
1744         ll_update_inode(inode, md);
1745
1746         /* OIDEBUG(inode); */
1747
1748         /* initializing backing dev info. */
1749         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1750
1751
1752         if (S_ISREG(inode->i_mode)) {
1753                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1754                 inode->i_op = &ll_file_inode_operations;
1755                 inode->i_fop = sbi->ll_fop;
1756                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1757                 EXIT;
1758         } else if (S_ISDIR(inode->i_mode)) {
1759                 inode->i_op = &ll_dir_inode_operations;
1760                 inode->i_fop = &ll_dir_operations;
1761                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1762                 EXIT;
1763         } else if (S_ISLNK(inode->i_mode)) {
1764                 inode->i_op = &ll_fast_symlink_inode_operations;
1765                 EXIT;
1766         } else {
1767                 inode->i_op = &ll_special_inode_operations;
1768
1769                 init_special_inode(inode, inode->i_mode,
1770                                    kdev_t_to_nr(inode->i_rdev));
1771
1772                 EXIT;
1773         }
1774 }
1775
1776 void ll_delete_inode(struct inode *inode)
1777 {
1778         struct ll_sb_info *sbi = ll_i2sbi(inode);
1779         int rc;
1780         ENTRY;
1781
1782         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1783         if (rc)
1784                 CERROR("fid_delete() failed, rc %d\n", rc);
1785
1786         truncate_inode_pages(&inode->i_data, 0);
1787
1788         /* Workaround for LU-118 */
1789         if (inode->i_data.nrpages) {
1790                 TREE_READ_LOCK_IRQ(&inode->i_data);
1791                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1792                 LASSERTF(inode->i_data.nrpages == 0,
1793                          "inode=%lu/%u(%p) nrpages=%lu, see "
1794                          "http://jira.whamcloud.com/browse/LU-118\n",
1795                          inode->i_ino, inode->i_generation, inode,
1796                          inode->i_data.nrpages);
1797         }
1798         /* Workaround end */
1799
1800         clear_inode(inode);
1801
1802         EXIT;
1803 }
1804
1805 int ll_iocontrol(struct inode *inode, struct file *file,
1806                  unsigned int cmd, unsigned long arg)
1807 {
1808         struct ll_sb_info *sbi = ll_i2sbi(inode);
1809         struct ptlrpc_request *req = NULL;
1810         int rc, flags = 0;
1811         ENTRY;
1812
1813         switch(cmd) {
1814         case FSFILT_IOC_GETFLAGS: {
1815                 struct mdt_body *body;
1816                 struct md_op_data *op_data;
1817
1818                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1819                                              0, 0, LUSTRE_OPC_ANY,
1820                                              NULL);
1821                 if (IS_ERR(op_data))
1822                         RETURN(PTR_ERR(op_data));
1823
1824                 op_data->op_valid = OBD_MD_FLFLAGS;
1825                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1826                 ll_finish_md_op_data(op_data);
1827                 if (rc) {
1828                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1829                         RETURN(-abs(rc));
1830                 }
1831
1832                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1833
1834                 flags = body->flags;
1835
1836                 ptlrpc_req_finished(req);
1837
1838                 RETURN(put_user(flags, (int *)arg));
1839         }
1840         case FSFILT_IOC_SETFLAGS: {
1841                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1842                 struct obd_info oinfo = { { { 0 } } };
1843                 struct md_op_data *op_data;
1844
1845                 if (get_user(flags, (int *)arg))
1846                         RETURN(-EFAULT);
1847
1848                 oinfo.oi_md = lsm;
1849                 OBDO_ALLOC(oinfo.oi_oa);
1850                 if (!oinfo.oi_oa)
1851                         RETURN(-ENOMEM);
1852
1853                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1854                                              LUSTRE_OPC_ANY, NULL);
1855                 if (IS_ERR(op_data))
1856                         RETURN(PTR_ERR(op_data));
1857
1858                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1859                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1860                 rc = md_setattr(sbi->ll_md_exp, op_data,
1861                                 NULL, 0, NULL, 0, &req, NULL);
1862                 ll_finish_md_op_data(op_data);
1863                 ptlrpc_req_finished(req);
1864                 if (rc) {
1865                         OBDO_FREE(oinfo.oi_oa);
1866                         RETURN(rc);
1867                 }
1868
1869                 if (lsm == NULL) {
1870                         OBDO_FREE(oinfo.oi_oa);
1871                         GOTO(update_cache, rc);
1872                 }
1873
1874                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1875                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1876                 oinfo.oi_oa->o_flags = flags;
1877                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1878                                        OBD_MD_FLGROUP;
1879                 oinfo.oi_capa = ll_mdscapa_get(inode);
1880                 obdo_from_inode(oinfo.oi_oa, inode,
1881                                 &ll_i2info(inode)->lli_fid, 0);
1882                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1883                 capa_put(oinfo.oi_capa);
1884                 OBDO_FREE(oinfo.oi_oa);
1885                 if (rc) {
1886                         if (rc != -EPERM && rc != -EACCES)
1887                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1888                         RETURN(rc);
1889                 }
1890
1891                 EXIT;
1892 update_cache:
1893                 inode->i_flags = ll_ext_to_inode_flags(flags);
1894                 return 0;
1895         }
1896         default:
1897                 RETURN(-ENOSYS);
1898         }
1899
1900         RETURN(0);
1901 }
1902
1903 int ll_flush_ctx(struct inode *inode)
1904 {
1905         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1906
1907         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1908
1909         obd_set_info_async(sbi->ll_md_exp,
1910                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1911                            0, NULL, NULL);
1912         obd_set_info_async(sbi->ll_dt_exp,
1913                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1914                            0, NULL, NULL);
1915         return 0;
1916 }
1917
1918 /* umount -f client means force down, don't save state */
1919 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1920 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1921 {
1922         struct super_block *sb = vfsmnt->mnt_sb;
1923 #else
1924 void ll_umount_begin(struct super_block *sb)
1925 {
1926 #endif
1927         struct lustre_sb_info *lsi = s2lsi(sb);
1928         struct ll_sb_info *sbi = ll_s2sbi(sb);
1929         struct obd_device *obd;
1930         struct obd_ioctl_data *ioc_data;
1931         ENTRY;
1932
1933 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1934         if (!(flags & MNT_FORCE)) {
1935                 EXIT;
1936                 return;
1937         }
1938 #endif
1939
1940         /* Tell the MGC we got umount -f */
1941         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1942
1943         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1944                sb->s_count, atomic_read(&sb->s_active));
1945
1946         obd = class_exp2obd(sbi->ll_md_exp);
1947         if (obd == NULL) {
1948                 CERROR("Invalid MDC connection handle "LPX64"\n",
1949                        sbi->ll_md_exp->exp_handle.h_cookie);
1950                 EXIT;
1951                 return;
1952         }
1953         obd->obd_force = 1;
1954
1955         obd = class_exp2obd(sbi->ll_dt_exp);
1956         if (obd == NULL) {
1957                 CERROR("Invalid LOV connection handle "LPX64"\n",
1958                        sbi->ll_dt_exp->exp_handle.h_cookie);
1959                 EXIT;
1960                 return;
1961         }
1962         obd->obd_force = 1;
1963
1964         OBD_ALLOC_PTR(ioc_data);
1965         if (ioc_data) {
1966                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1967                               sizeof ioc_data, ioc_data, NULL);
1968
1969                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1970                               sizeof ioc_data, ioc_data, NULL);
1971
1972                 OBD_FREE_PTR(ioc_data);
1973         }
1974
1975
1976         /* Really, we'd like to wait until there are no requests outstanding,
1977          * and then continue.  For now, we just invalidate the requests,
1978          * schedule() and sleep one second if needed, and hope.
1979          */
1980         cfs_schedule();
1981 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1982         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1983                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1984                                                    cfs_time_seconds(1));
1985                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1986                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1987                                       "may try to umount it a bit later\n",
1988                                       atomic_read(&vfsmnt->mnt_count));
1989         }
1990 #endif
1991
1992         EXIT;
1993 }
1994
1995 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1996 {
1997         struct ll_sb_info *sbi = ll_s2sbi(sb);
1998         int err;
1999         __u32 read_only;
2000
2001         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2002                 read_only = *flags & MS_RDONLY;
2003                 err = obd_set_info_async(sbi->ll_md_exp,
2004                                          sizeof(KEY_READ_ONLY),
2005                                          KEY_READ_ONLY, sizeof(read_only),
2006                                          &read_only, NULL);
2007                 if (err) {
2008                         CERROR("Failed to change the read-only flag during "
2009                                "remount: %d\n", err);
2010                         return err;
2011                 }
2012
2013                 if (read_only)
2014                         sb->s_flags |= MS_RDONLY;
2015                 else
2016                         sb->s_flags &= ~MS_RDONLY;
2017         }
2018         return 0;
2019 }
2020
2021 int ll_prep_inode(struct inode **inode,
2022                   struct ptlrpc_request *req,
2023                   struct super_block *sb)
2024 {
2025         struct ll_sb_info *sbi = NULL;
2026         struct lustre_md md;
2027         int rc;
2028         ENTRY;
2029
2030         LASSERT(*inode || sb);
2031         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2032         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2033                               sbi->ll_md_exp, &md);
2034         if (rc)
2035                 RETURN(rc);
2036
2037         if (*inode) {
2038                 ll_update_inode(*inode, &md);
2039         } else {
2040                 LASSERT(sb != NULL);
2041
2042                 /*
2043                  * At this point server returns to client's same fid as client
2044                  * generated for creating. So using ->fid1 is okay here.
2045                  */
2046                 LASSERT(fid_is_sane(&md.body->fid1));
2047
2048                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2049                 if (*inode == NULL || IS_ERR(*inode)) {
2050                         if (md.lsm)
2051                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2052 #ifdef CONFIG_FS_POSIX_ACL
2053                         if (md.posix_acl) {
2054                                 posix_acl_release(md.posix_acl);
2055                                 md.posix_acl = NULL;
2056                         }
2057 #endif
2058                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2059                         *inode = NULL;
2060                         CERROR("new_inode -fatal: rc %d\n", rc);
2061                         GOTO(out, rc);
2062                 }
2063         }
2064
2065 out:
2066         md_free_lustre_md(sbi->ll_md_exp, &md);
2067         RETURN(rc);
2068 }
2069
2070 int ll_obd_statfs(struct inode *inode, void *arg)
2071 {
2072         struct ll_sb_info *sbi = NULL;
2073         struct obd_export *exp;
2074         char *buf = NULL;
2075         struct obd_ioctl_data *data = NULL;
2076         __u32 type;
2077         int len = 0, rc;
2078
2079         if (!inode || !(sbi = ll_i2sbi(inode)))
2080                 GOTO(out_statfs, rc = -EINVAL);
2081
2082         rc = obd_ioctl_getdata(&buf, &len, arg);
2083         if (rc)
2084                 GOTO(out_statfs, rc);
2085
2086         data = (void*)buf;
2087         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2088             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2089                 GOTO(out_statfs, rc = -EINVAL);
2090
2091         if (data->ioc_inllen1 != sizeof(__u32) ||
2092             data->ioc_inllen2 != sizeof(__u32) ||
2093             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2094             data->ioc_plen2 != sizeof(struct obd_uuid))
2095                 GOTO(out_statfs, rc = -EINVAL);
2096
2097         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2098         if (type == LL_STATFS_MDC)
2099                 exp = sbi->ll_md_exp;
2100         else if (type == LL_STATFS_LOV)
2101                 exp = sbi->ll_dt_exp;
2102         else
2103                 GOTO(out_statfs, rc = -ENODEV);
2104
2105         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2106         if (rc)
2107                 GOTO(out_statfs, rc);
2108 out_statfs:
2109         if (buf)
2110                 obd_ioctl_freedata(buf, len);
2111         return rc;
2112 }
2113
2114 int ll_process_config(struct lustre_cfg *lcfg)
2115 {
2116         char *ptr;
2117         void *sb;
2118         struct lprocfs_static_vars lvars;
2119         unsigned long x;
2120         int rc = 0;
2121
2122         lprocfs_llite_init_vars(&lvars);
2123
2124         /* The instance name contains the sb: lustre-client-aacfe000 */
2125         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2126         if (!ptr || !*(++ptr))
2127                 return -EINVAL;
2128         if (sscanf(ptr, "%lx", &x) != 1)
2129                 return -EINVAL;
2130         sb = (void *)x;
2131         /* This better be a real Lustre superblock! */
2132         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2133
2134         /* Note we have not called client_common_fill_super yet, so
2135            proc fns must be able to handle that! */
2136         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2137                                       lcfg, sb);
2138         if (rc > 0)
2139                 rc = 0;
2140         return(rc);
2141 }
2142
2143 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2144 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2145                                        struct inode *i1, struct inode *i2,
2146                                        const char *name, int namelen,
2147                                        int mode, __u32 opc, void *data)
2148 {
2149         LASSERT(i1 != NULL);
2150
2151         if (namelen > ll_i2sbi(i1)->ll_namelen)
2152                 return ERR_PTR(-ENAMETOOLONG);
2153
2154         if (op_data == NULL)
2155                 OBD_ALLOC_PTR(op_data);
2156
2157         if (op_data == NULL)
2158                 return ERR_PTR(-ENOMEM);
2159
2160         ll_i2gids(op_data->op_suppgids, i1, i2);
2161         op_data->op_fid1 = *ll_inode2fid(i1);
2162         op_data->op_capa1 = ll_mdscapa_get(i1);
2163
2164         if (i2) {
2165                 op_data->op_fid2 = *ll_inode2fid(i2);
2166                 op_data->op_capa2 = ll_mdscapa_get(i2);
2167         } else {
2168                 fid_zero(&op_data->op_fid2);
2169                 op_data->op_capa2 = NULL;
2170         }
2171
2172         op_data->op_name = name;
2173         op_data->op_namelen = namelen;
2174         op_data->op_mode = mode;
2175         op_data->op_mod_time = cfs_time_current_sec();
2176         op_data->op_fsuid = cfs_curproc_fsuid();
2177         op_data->op_fsgid = cfs_curproc_fsgid();
2178         op_data->op_cap = cfs_curproc_cap_pack();
2179         op_data->op_bias = MDS_CHECK_SPLIT;
2180         op_data->op_opc = opc;
2181         op_data->op_mds = 0;
2182         op_data->op_data = data;
2183
2184         /* If the file is being opened after mknod() (normally due to NFS)
2185          * try to use the default stripe data from parent directory for
2186          * allocating OST objects.  Try to pass the parent FID to MDS. */
2187         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2188             ll_i2info(i2)->lli_smd == NULL) {
2189                 struct ll_inode_info *lli = ll_i2info(i2);
2190
2191                 cfs_spin_lock(&lli->lli_lock);
2192                 if (likely(lli->lli_smd == NULL &&
2193                            !fid_is_zero(&lli->lli_pfid)))
2194                         op_data->op_fid1 = lli->lli_pfid;
2195                 cfs_spin_unlock(&lli->lli_lock);
2196                 /** We ignore parent's capability temporary. */
2197         }
2198
2199         return op_data;
2200 }
2201
2202 void ll_finish_md_op_data(struct md_op_data *op_data)
2203 {
2204         capa_put(op_data->op_capa1);
2205         capa_put(op_data->op_capa2);
2206         OBD_FREE_PTR(op_data);
2207 }
2208
2209 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2210 {
2211         struct ll_sb_info *sbi;
2212
2213         LASSERT((seq != NULL) && (vfs != NULL));
2214         sbi = ll_s2sbi(vfs->mnt_sb);
2215
2216         if (sbi->ll_flags & LL_SBI_NOLCK)
2217                 seq_puts(seq, ",nolock");
2218
2219         if (sbi->ll_flags & LL_SBI_FLOCK)
2220                 seq_puts(seq, ",flock");
2221
2222         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2223                 seq_puts(seq, ",localflock");
2224
2225         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2226                 seq_puts(seq, ",user_xattr");
2227
2228         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2229                 seq_puts(seq, ",lazystatfs");
2230
2231         RETURN(0);
2232 }