Whamcloud - gitweb
050d0e6b47d2812d52acb75ace24cdeb3cf6d07e
[fs/lustre-release.git] / lustre / llite / llite_lib.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2015, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/llite/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/statfs.h>
45 #include <linux/types.h>
46 #include <linux/version.h>
47 #include <linux/mm.h>
48 #include <linux/user_namespace.h>
49 #ifdef HAVE_UIDGID_HEADER
50 # include <linux/uidgid.h>
51 #endif
52
53 #include <lustre_ioctl.h>
54 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
63
64 struct kmem_cache *ll_file_data_slab;
65
66 #ifndef log2
67 #define log2(n) ffz(~(n))
68 #endif
69
70 static struct ll_sb_info *ll_init_sbi(void)
71 {
72         struct ll_sb_info *sbi = NULL;
73         unsigned long pages;
74         unsigned long lru_page_max;
75         struct sysinfo si;
76         class_uuid_t uuid;
77         int i;
78         ENTRY;
79
80         OBD_ALLOC_PTR(sbi);
81         if (sbi == NULL)
82                 RETURN(NULL);
83
84         spin_lock_init(&sbi->ll_lock);
85         mutex_init(&sbi->ll_lco.lco_lock);
86         spin_lock_init(&sbi->ll_pp_extent_lock);
87         spin_lock_init(&sbi->ll_process_lock);
88         sbi->ll_rw_stats_on = 0;
89
90         si_meminfo(&si);
91         pages = si.totalram - si.totalhigh;
92         lru_page_max = pages / 2;
93
94         /* initialize ll_cache data */
95         sbi->ll_cache = cl_cache_init(lru_page_max);
96         if (sbi->ll_cache == NULL) {
97                 OBD_FREE(sbi, sizeof(*sbi));
98                 RETURN(NULL);
99         }
100
101         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
102                                            SBI_DEFAULT_READAHEAD_MAX);
103         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
104         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
105                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
106         INIT_LIST_HEAD(&sbi->ll_conn_chain);
107         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
108
109         ll_generate_random_uuid(uuid);
110         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
111         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
112
113         sbi->ll_flags |= LL_SBI_VERBOSE;
114 #ifdef ENABLE_CHECKSUM
115         sbi->ll_flags |= LL_SBI_CHECKSUM;
116 #endif
117
118 #ifdef HAVE_LRU_RESIZE_SUPPORT
119         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
120 #endif
121
122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
124                                pp_r_hist.oh_lock);
125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126                                pp_w_hist.oh_lock);
127         }
128
129         /* metadata statahead is enabled by default */
130         sbi->ll_sa_max = LL_SA_RPC_DEF;
131         atomic_set(&sbi->ll_sa_total, 0);
132         atomic_set(&sbi->ll_sa_wrong, 0);
133         atomic_set(&sbi->ll_sa_running, 0);
134         atomic_set(&sbi->ll_agl_total, 0);
135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
136
137         /* root squash */
138         sbi->ll_squash.rsi_uid = 0;
139         sbi->ll_squash.rsi_gid = 0;
140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
141         init_rwsem(&sbi->ll_squash.rsi_sem);
142
143         RETURN(sbi);
144 }
145
146 static void ll_free_sbi(struct super_block *sb)
147 {
148         struct ll_sb_info *sbi = ll_s2sbi(sb);
149         ENTRY;
150
151         if (sbi != NULL) {
152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
154                 if (sbi->ll_cache != NULL) {
155                         cl_cache_decref(sbi->ll_cache);
156                         sbi->ll_cache = NULL;
157                 }
158                 OBD_FREE(sbi, sizeof(*sbi));
159         }
160         EXIT;
161 }
162
163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
164                                     struct vfsmount *mnt)
165 {
166         struct inode *root = NULL;
167         struct ll_sb_info *sbi = ll_s2sbi(sb);
168         struct obd_device *obd;
169         struct obd_statfs *osfs = NULL;
170         struct ptlrpc_request *request = NULL;
171         struct obd_connect_data *data = NULL;
172         struct obd_uuid *uuid;
173         struct md_op_data *op_data;
174         struct lustre_md lmd;
175         u64 valid;
176         int size, err, checksum;
177         ENTRY;
178
179         obd = class_name2obd(md);
180         if (!obd) {
181                 CERROR("MD %s: not setup or attached\n", md);
182                 RETURN(-EINVAL);
183         }
184
185         OBD_ALLOC_PTR(data);
186         if (data == NULL)
187                 RETURN(-ENOMEM);
188
189         OBD_ALLOC_PTR(osfs);
190         if (osfs == NULL) {
191                 OBD_FREE_PTR(data);
192                 RETURN(-ENOMEM);
193         }
194
195         /* indicate the features supported by this client */
196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
197                                   OBD_CONNECT_ATTRFID  |
198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
204                                   OBD_CONNECT_EINPROGRESS |
205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
207                                   OBD_CONNECT_MAX_EASIZE |
208                                   OBD_CONNECT_FLOCK_DEAD |
209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
210                                   OBD_CONNECT_OPEN_BY_FID |
211                                   OBD_CONNECT_DIR_STRIPE |
212                                   OBD_CONNECT_BULK_MBITS;
213
214 #ifdef HAVE_LRU_RESIZE_SUPPORT
215         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
216                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
217 #endif
218 #ifdef CONFIG_FS_POSIX_ACL
219         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
220 #endif
221
222         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
223                 /* flag mdc connection as lightweight, only used for test
224                  * purpose, use with care */
225                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
226
227         data->ocd_ibits_known = MDS_INODELOCK_FULL;
228         data->ocd_version = LUSTRE_VERSION_CODE;
229
230         if (sb->s_flags & MS_RDONLY)
231                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
232         if (sbi->ll_flags & LL_SBI_USER_XATTR)
233                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
234
235 #ifdef HAVE_MS_FLOCK_LOCK
236         /* force vfs to use lustre handler for flock() calls - bug 10743 */
237         sb->s_flags |= MS_FLOCK_LOCK;
238 #endif
239 #ifdef MS_HAS_NEW_AOPS
240         sb->s_flags |= MS_HAS_NEW_AOPS;
241 #endif
242
243         if (sbi->ll_flags & LL_SBI_FLOCK)
244                 sbi->ll_fop = &ll_file_operations_flock;
245         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
246                 sbi->ll_fop = &ll_file_operations;
247         else
248                 sbi->ll_fop = &ll_file_operations_noflock;
249
250         /* real client */
251         data->ocd_connect_flags |= OBD_CONNECT_REAL;
252         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
253                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
254
255         data->ocd_brw_size = MD_MAX_BRW_SIZE;
256
257         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
258         if (err == -EBUSY) {
259                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
260                                    "recovery, of which this client is not a "
261                                    "part. Please wait for recovery to complete,"
262                                    " abort, or time out.\n", md);
263                 GOTO(out, err);
264         } else if (err) {
265                 CERROR("cannot connect to %s: rc = %d\n", md, err);
266                 GOTO(out, err);
267         }
268
269         sbi->ll_md_exp->exp_connect_data = *data;
270
271         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
272                            LUSTRE_SEQ_METADATA);
273         if (err) {
274                 CERROR("%s: Can't init metadata layer FID infrastructure, "
275                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
276                 GOTO(out_md, err);
277         }
278
279         /* For mount, we only need fs info from MDT0, and also in DNE, it
280          * can make sure the client can be mounted as long as MDT0 is
281          * avaible */
282         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
283                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
284                         OBD_STATFS_FOR_MDT0);
285         if (err)
286                 GOTO(out_md_fid, err);
287
288         /* This needs to be after statfs to ensure connect has finished.
289          * Note that "data" does NOT contain the valid connect reply.
290          * If connecting to a 1.8 server there will be no LMV device, so
291          * we can access the MDC export directly and exp_connect_flags will
292          * be non-zero, but if accessing an upgraded 2.1 server it will
293          * have the correct flags filled in.
294          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
295         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
296         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
297             valid != CLIENT_CONNECT_MDT_REQD) {
298                 char *buf;
299
300                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
301                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
302                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
303                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
304                                    "feature(s) needed for correct operation "
305                                    "of this client (%s). Please upgrade "
306                                    "server or downgrade client.\n",
307                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
308                 OBD_FREE(buf, PAGE_CACHE_SIZE);
309                 GOTO(out_md_fid, err = -EPROTO);
310         }
311
312         size = sizeof(*data);
313         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
314                            KEY_CONN_DATA,  &size, data);
315         if (err) {
316                 CERROR("%s: Get connect data failed: rc = %d\n",
317                        sbi->ll_md_exp->exp_obd->obd_name, err);
318                 GOTO(out_md_fid, err);
319         }
320
321         LASSERT(osfs->os_bsize);
322         sb->s_blocksize = osfs->os_bsize;
323         sb->s_blocksize_bits = log2(osfs->os_bsize);
324         sb->s_magic = LL_SUPER_MAGIC;
325         sb->s_maxbytes = MAX_LFS_FILESIZE;
326         sbi->ll_namelen = osfs->os_namelen;
327
328         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
329             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
330                 LCONSOLE_INFO("Disabling user_xattr feature because "
331                               "it is not supported on the server\n");
332                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
333         }
334
335         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
336 #ifdef MS_POSIXACL
337                 sb->s_flags |= MS_POSIXACL;
338 #endif
339                 sbi->ll_flags |= LL_SBI_ACL;
340         } else {
341                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
342 #ifdef MS_POSIXACL
343                 sb->s_flags &= ~MS_POSIXACL;
344 #endif
345                 sbi->ll_flags &= ~LL_SBI_ACL;
346         }
347
348         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
349                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
350                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
351                         LCONSOLE_INFO("client is set as remote by default.\n");
352                 }
353         } else {
354                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
355                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
356                         LCONSOLE_INFO("client claims to be remote, but server "
357                                       "rejected, forced to be local.\n");
358                 }
359         }
360
361         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
362                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
363
364         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
365                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
366         else
367                 sbi->ll_md_brw_pages = 1;
368
369         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
370                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
371
372         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
373                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
374                         LCONSOLE_INFO("%s: disabling xattr cache due to "
375                                       "unknown maximum xattr size.\n", dt);
376                 } else {
377                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
378                         sbi->ll_xattr_cache_enabled = 1;
379                 }
380         }
381
382         obd = class_name2obd(dt);
383         if (!obd) {
384                 CERROR("DT %s: not setup or attached\n", dt);
385                 GOTO(out_md_fid, err = -ENODEV);
386         }
387
388         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
389                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
390                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
391                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
392                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
393                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
394                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
395                                   OBD_CONNECT_MAXBYTES |
396                                   OBD_CONNECT_EINPROGRESS |
397                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
398                                   OBD_CONNECT_LAYOUTLOCK |
399                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
400                                   OBD_CONNECT_BULK_MBITS;
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                         data->ocd_cksum_types = cksum_types_supported_client();
413         }
414
415 #ifdef HAVE_LRU_RESIZE_SUPPORT
416         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
417 #endif
418         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
419                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
420
421         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
422                "ocd_grant: %d\n", data->ocd_connect_flags,
423                data->ocd_version, data->ocd_grant);
424
425         obd->obd_upcall.onu_owner = &sbi->ll_lco;
426         obd->obd_upcall.onu_upcall = cl_ocd_update;
427
428         data->ocd_brw_size = DT_MAX_BRW_SIZE;
429
430         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
431                           NULL);
432         if (err == -EBUSY) {
433                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
434                                    "recovery, of which this client is not a "
435                                    "part.  Please wait for recovery to "
436                                    "complete, abort, or time out.\n", dt);
437                 GOTO(out_md, err);
438         } else if (err) {
439                 CERROR("%s: Cannot connect to %s: rc = %d\n",
440                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
441                 GOTO(out_md, err);
442         }
443
444         sbi->ll_dt_exp->exp_connect_data = *data;
445
446         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
447                            LUSTRE_SEQ_METADATA);
448         if (err) {
449                 CERROR("%s: Can't init data layer FID infrastructure, "
450                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
451                 GOTO(out_dt, err);
452         }
453
454         mutex_lock(&sbi->ll_lco.lco_lock);
455         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
456         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
457         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
458         mutex_unlock(&sbi->ll_lco.lco_lock);
459
460         fid_zero(&sbi->ll_root_fid);
461         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
462         if (err) {
463                 CERROR("cannot mds_connect: rc = %d\n", err);
464                 GOTO(out_lock_cn_cb, err);
465         }
466         if (!fid_is_sane(&sbi->ll_root_fid)) {
467                 CERROR("%s: Invalid root fid "DFID" during mount\n",
468                        sbi->ll_md_exp->exp_obd->obd_name,
469                        PFID(&sbi->ll_root_fid));
470                 GOTO(out_lock_cn_cb, err = -EINVAL);
471         }
472         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
473
474         sb->s_op = &lustre_super_operations;
475 #if THREAD_SIZE >= 8192 /*b=17630*/
476         sb->s_export_op = &lustre_export_operations;
477 #endif
478
479         /* make root inode
480          * XXX: move this to after cbd setup? */
481         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
482         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
483                 valid |= OBD_MD_FLRMTPERM;
484         else if (sbi->ll_flags & LL_SBI_ACL)
485                 valid |= OBD_MD_FLACL;
486
487         OBD_ALLOC_PTR(op_data);
488         if (op_data == NULL)
489                 GOTO(out_lock_cn_cb, err = -ENOMEM);
490
491         op_data->op_fid1 = sbi->ll_root_fid;
492         op_data->op_mode = 0;
493         op_data->op_valid = valid;
494
495         err = md_getattr(sbi->ll_md_exp, op_data, &request);
496
497         OBD_FREE_PTR(op_data);
498         if (err) {
499                 CERROR("%s: md_getattr failed for root: rc = %d\n",
500                        sbi->ll_md_exp->exp_obd->obd_name, err);
501                 GOTO(out_lock_cn_cb, err);
502         }
503
504         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
505                                sbi->ll_md_exp, &lmd);
506         if (err) {
507                 CERROR("failed to understand root inode md: rc = %d\n", err);
508                 ptlrpc_req_finished(request);
509                 GOTO(out_lock_cn_cb, err);
510         }
511
512         LASSERT(fid_is_sane(&sbi->ll_root_fid));
513         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
514                                             sbi->ll_flags & LL_SBI_32BIT_API),
515                        &lmd);
516         md_free_lustre_md(sbi->ll_md_exp, &lmd);
517         ptlrpc_req_finished(request);
518
519         if (IS_ERR(root)) {
520 #ifdef CONFIG_FS_POSIX_ACL
521                 if (lmd.posix_acl) {
522                         posix_acl_release(lmd.posix_acl);
523                         lmd.posix_acl = NULL;
524                 }
525 #endif
526                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
527                 root = NULL;
528                 CERROR("lustre_lite: bad iget4 for root\n");
529                 GOTO(out_root, err);
530         }
531
532 #ifdef CONFIG_FS_POSIX_ACL
533         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
534                 rct_init(&sbi->ll_rct);
535                 et_init(&sbi->ll_et);
536         }
537 #endif
538
539         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
540         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
541                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
542                                  NULL);
543         cl_sb_init(sb);
544
545         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
546                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
547                                  sbi->ll_cache, NULL);
548
549         sb->s_root = d_make_root(root);
550         if (sb->s_root == NULL) {
551                 CERROR("%s: can't make root dentry\n",
552                         ll_get_fsname(sb, NULL, 0));
553                 GOTO(out_root, err = -ENOMEM);
554         }
555 #ifdef HAVE_DCACHE_LOCK
556         sb->s_root->d_op = &ll_d_ops;
557 #endif
558
559         sbi->ll_sdev_orig = sb->s_dev;
560
561         /* We set sb->s_dev equal on all lustre clients in order to support
562          * NFS export clustering.  NFSD requires that the FSID be the same
563          * on all clients. */
564         /* s_dev is also used in lt_compare() to compare two fs, but that is
565          * only a node-local comparison. */
566         uuid = obd_get_uuid(sbi->ll_md_exp);
567         if (uuid != NULL)
568                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
569
570         if (data != NULL)
571                 OBD_FREE_PTR(data);
572         if (osfs != NULL)
573                 OBD_FREE_PTR(osfs);
574         if (proc_lustre_fs_root != NULL) {
575                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
576                                                   dt, md);
577                 if (err < 0) {
578                         CERROR("%s: could not register mount in lprocfs: "
579                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
580                         err = 0;
581                 }
582         }
583
584         RETURN(err);
585 out_root:
586         if (root)
587                 iput(root);
588 out_lock_cn_cb:
589         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
590 out_dt:
591         obd_disconnect(sbi->ll_dt_exp);
592         sbi->ll_dt_exp = NULL;
593 out_md_fid:
594         obd_fid_fini(sbi->ll_md_exp->exp_obd);
595 out_md:
596         obd_disconnect(sbi->ll_md_exp);
597         sbi->ll_md_exp = NULL;
598 out:
599         if (data != NULL)
600                 OBD_FREE_PTR(data);
601         if (osfs != NULL)
602                 OBD_FREE_PTR(osfs);
603         return err;
604 }
605
606 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
607 {
608         int size, rc;
609
610         size = sizeof(*lmmsize);
611         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
612                           KEY_MAX_EASIZE, &size, lmmsize);
613         if (rc != 0) {
614                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
615                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
616                 RETURN(rc);
617         }
618
619         size = sizeof(int);
620         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
621                           KEY_MAX_EASIZE, &size, lmmsize);
622         if (rc)
623                 CERROR("Get max mdsize error rc %d\n", rc);
624
625         RETURN(rc);
626 }
627
628 /**
629  * Get the value of the default_easize parameter.
630  *
631  * \see client_obd::cl_default_mds_easize
632  *
633  * \param[in] sbi       superblock info for this filesystem
634  * \param[out] lmmsize  pointer to storage location for value
635  *
636  * \retval 0            on success
637  * \retval negative     negated errno on failure
638  */
639 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
640 {
641         int size, rc;
642
643         size = sizeof(int);
644         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
645                          KEY_DEFAULT_EASIZE, &size, lmmsize);
646         if (rc)
647                 CERROR("Get default mdsize error rc %d\n", rc);
648
649         RETURN(rc);
650 }
651
652 /**
653  * Set the default_easize parameter to the given value.
654  *
655  * \see client_obd::cl_default_mds_easize
656  *
657  * \param[in] sbi       superblock info for this filesystem
658  * \param[in] lmmsize   the size to set
659  *
660  * \retval 0            on success
661  * \retval negative     negated errno on failure
662  */
663 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
664 {
665         int rc;
666
667         if (lmmsize < sizeof(struct lov_mds_md) ||
668             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
669                 return -EINVAL;
670
671         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
672                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
673                                 sizeof(int), &lmmsize, NULL);
674
675         RETURN(rc);
676 }
677
678 static void ll_dump_inode(struct inode *inode)
679 {
680         struct ll_d_hlist_node *tmp;
681         int dentry_count = 0;
682
683         LASSERT(inode != NULL);
684
685         ll_d_hlist_for_each(tmp, &inode->i_dentry)
686                 dentry_count++;
687
688         CERROR("%s: inode %p dump: dev=%s fid="DFID
689                " mode=%o count=%u, %d dentries\n",
690                ll_get_fsname(inode->i_sb, NULL, 0), inode,
691                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
692                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
693 }
694
695 void lustre_dump_dentry(struct dentry *dentry, int recur)
696 {
697         struct list_head *tmp;
698         int subdirs = 0;
699
700         LASSERT(dentry != NULL);
701
702         list_for_each(tmp, &dentry->d_subdirs)
703                 subdirs++;
704
705         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
706                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
707                dentry->d_name.len, dentry->d_name.name,
708                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
709                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
710                dentry->d_flags, dentry->d_fsdata, subdirs);
711         if (dentry->d_inode != NULL)
712                 ll_dump_inode(dentry->d_inode);
713
714         if (recur == 0)
715                 return;
716
717         list_for_each(tmp, &dentry->d_subdirs) {
718                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
719                 lustre_dump_dentry(d, recur - 1);
720         }
721 }
722
723 static void client_common_put_super(struct super_block *sb)
724 {
725         struct ll_sb_info *sbi = ll_s2sbi(sb);
726         ENTRY;
727
728 #ifdef CONFIG_FS_POSIX_ACL
729         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
730                 et_fini(&sbi->ll_et);
731                 rct_fini(&sbi->ll_rct);
732         }
733 #endif
734
735         cl_sb_fini(sb);
736
737         list_del(&sbi->ll_conn_chain);
738
739         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
740         obd_disconnect(sbi->ll_dt_exp);
741         sbi->ll_dt_exp = NULL;
742
743         lprocfs_unregister_mountpoint(sbi);
744
745         obd_fid_fini(sbi->ll_md_exp->exp_obd);
746         obd_disconnect(sbi->ll_md_exp);
747         sbi->ll_md_exp = NULL;
748
749         EXIT;
750 }
751
752 void ll_kill_super(struct super_block *sb)
753 {
754         struct ll_sb_info *sbi;
755         ENTRY;
756
757         /* not init sb ?*/
758         if (!(sb->s_flags & MS_ACTIVE))
759                 return;
760
761         sbi = ll_s2sbi(sb);
762         /* we need restore s_dev from changed for clustred NFS before put_super
763          * because new kernels have cached s_dev and change sb->s_dev in
764          * put_super not affected real removing devices */
765         if (sbi) {
766                 sb->s_dev = sbi->ll_sdev_orig;
767                 sbi->ll_umounting = 1;
768
769                 /* wait running statahead threads to quit */
770                 while (atomic_read(&sbi->ll_sa_running) > 0) {
771                         set_current_state(TASK_UNINTERRUPTIBLE);
772                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
773                 }
774         }
775
776         EXIT;
777 }
778
779 static inline int ll_set_opt(const char *opt, char *data, int fl)
780 {
781         if (strncmp(opt, data, strlen(opt)) != 0)
782                 return(0);
783         else
784                 return(fl);
785 }
786
787 /* non-client-specific mount options are parsed in lmd_parse */
788 static int ll_options(char *options, int *flags)
789 {
790         int tmp;
791         char *s1 = options, *s2;
792         ENTRY;
793
794         if (!options)
795                 RETURN(0);
796
797         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
798
799         while (*s1) {
800                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
801                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
802                 if (tmp) {
803                         *flags |= tmp;
804                         goto next;
805                 }
806                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
807                 if (tmp) {
808                         *flags |= tmp;
809                         goto next;
810                 }
811                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
812                 if (tmp) {
813                         *flags |= tmp;
814                         goto next;
815                 }
816                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
817                 if (tmp) {
818                         *flags &= ~tmp;
819                         goto next;
820                 }
821                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
822                 if (tmp) {
823                         *flags |= tmp;
824                         goto next;
825                 }
826                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
827                 if (tmp) {
828                         *flags &= ~tmp;
829                         goto next;
830                 }
831                 tmp = ll_set_opt("context", s1, 1);
832                 if (tmp)
833                         goto next;
834                 tmp = ll_set_opt("fscontext", s1, 1);
835                 if (tmp)
836                         goto next;
837                 tmp = ll_set_opt("defcontext", s1, 1);
838                 if (tmp)
839                         goto next;
840                 tmp = ll_set_opt("rootcontext", s1, 1);
841                 if (tmp)
842                         goto next;
843                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
844                 if (tmp) {
845                         *flags |= tmp;
846                         goto next;
847                 }
848                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
849                 if (tmp) {
850                         *flags |= tmp;
851                         goto next;
852                 }
853                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
854                 if (tmp) {
855                         *flags &= ~tmp;
856                         goto next;
857                 }
858
859                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
860                 if (tmp) {
861                         *flags |= tmp;
862                         goto next;
863                 }
864                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
865                 if (tmp) {
866                         *flags &= ~tmp;
867                         goto next;
868                 }
869                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
870                 if (tmp) {
871                         *flags |= tmp;
872                         goto next;
873                 }
874                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
875                 if (tmp) {
876                         *flags &= ~tmp;
877                         goto next;
878                 }
879                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
880                 if (tmp) {
881                         *flags |= tmp;
882                         goto next;
883                 }
884                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
885                 if (tmp) {
886                         *flags &= ~tmp;
887                         goto next;
888                 }
889                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
890                 if (tmp) {
891                         *flags |= tmp;
892                         goto next;
893                 }
894                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
895                 if (tmp) {
896                         *flags |= tmp;
897                         goto next;
898                 }
899                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
900                 if (tmp) {
901                         *flags &= ~tmp;
902                         goto next;
903                 }
904                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
905                                    s1);
906                 RETURN(-EINVAL);
907
908 next:
909                 /* Find next opt */
910                 s2 = strchr(s1, ',');
911                 if (s2 == NULL)
912                         break;
913                 s1 = s2 + 1;
914         }
915         RETURN(0);
916 }
917
918 void ll_lli_init(struct ll_inode_info *lli)
919 {
920         lli->lli_inode_magic = LLI_INODE_MAGIC;
921         lli->lli_flags = 0;
922         spin_lock_init(&lli->lli_lock);
923         lli->lli_posix_acl = NULL;
924         lli->lli_remote_perms = NULL;
925         mutex_init(&lli->lli_rmtperm_mutex);
926         /* Do not set lli_fid, it has been initialized already. */
927         fid_zero(&lli->lli_pfid);
928         atomic_set(&lli->lli_open_count, 0);
929         lli->lli_rmtperm_time = 0;
930         lli->lli_mds_read_och = NULL;
931         lli->lli_mds_write_och = NULL;
932         lli->lli_mds_exec_och = NULL;
933         lli->lli_open_fd_read_count = 0;
934         lli->lli_open_fd_write_count = 0;
935         lli->lli_open_fd_exec_count = 0;
936         mutex_init(&lli->lli_och_mutex);
937         spin_lock_init(&lli->lli_agl_lock);
938         spin_lock_init(&lli->lli_layout_lock);
939         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
940         lli->lli_clob = NULL;
941
942         init_rwsem(&lli->lli_xattrs_list_rwsem);
943         mutex_init(&lli->lli_xattrs_enq_lock);
944
945         LASSERT(lli->lli_vfs_inode.i_mode != 0);
946         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
947                 mutex_init(&lli->lli_readdir_mutex);
948                 lli->lli_opendir_key = NULL;
949                 lli->lli_sai = NULL;
950                 spin_lock_init(&lli->lli_sa_lock);
951                 lli->lli_opendir_pid = 0;
952                 lli->lli_sa_enabled = 0;
953                 lli->lli_def_stripe_offset = -1;
954         } else {
955                 mutex_init(&lli->lli_size_mutex);
956                 lli->lli_symlink_name = NULL;
957                 init_rwsem(&lli->lli_trunc_sem);
958                 range_lock_tree_init(&lli->lli_write_tree);
959                 init_rwsem(&lli->lli_glimpse_sem);
960                 lli->lli_glimpse_time = 0;
961                 INIT_LIST_HEAD(&lli->lli_agl_list);
962                 lli->lli_agl_index = 0;
963                 lli->lli_async_rc = 0;
964         }
965         mutex_init(&lli->lli_layout_mutex);
966 }
967
968 static inline int ll_bdi_register(struct backing_dev_info *bdi)
969 {
970         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
971
972         bdi->name = "lustre";
973         return bdi_register(bdi, NULL, "lustre-%d",
974                             atomic_inc_return(&ll_bdi_num));
975 }
976
977 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
978 {
979         struct lustre_profile *lprof = NULL;
980         struct lustre_sb_info *lsi = s2lsi(sb);
981         struct ll_sb_info *sbi;
982         char  *dt = NULL, *md = NULL;
983         char  *profilenm = get_profile_name(sb);
984         struct config_llog_instance *cfg;
985         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
986         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
987         int    err;
988         ENTRY;
989
990         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
991
992         OBD_ALLOC_PTR(cfg);
993         if (cfg == NULL)
994                 RETURN(-ENOMEM);
995
996         try_module_get(THIS_MODULE);
997
998         /* client additional sb info */
999         lsi->lsi_llsbi = sbi = ll_init_sbi();
1000         if (!sbi) {
1001                 module_put(THIS_MODULE);
1002                 OBD_FREE_PTR(cfg);
1003                 RETURN(-ENOMEM);
1004         }
1005
1006         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1007         if (err)
1008                 GOTO(out_free, err);
1009
1010         err = bdi_init(&lsi->lsi_bdi);
1011         if (err)
1012                 GOTO(out_free, err);
1013         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1014 #ifdef HAVE_BDI_CAP_MAP_COPY
1015         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1016 #else
1017         lsi->lsi_bdi.capabilities = 0;
1018 #endif
1019         err = ll_bdi_register(&lsi->lsi_bdi);
1020         if (err)
1021                 GOTO(out_free, err);
1022
1023         sb->s_bdi = &lsi->lsi_bdi;
1024 #ifndef HAVE_DCACHE_LOCK
1025         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1026         sb->s_d_op = &ll_d_ops;
1027 #endif
1028
1029         /* Generate a string unique to this super, in case some joker tries
1030            to mount the same fs at two mount points.
1031            Use the address of the super itself.*/
1032         cfg->cfg_instance = sb;
1033         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1034         cfg->cfg_callback = class_config_llog_handler;
1035         /* set up client obds */
1036         err = lustre_process_log(sb, profilenm, cfg);
1037         if (err < 0)
1038                 GOTO(out_free, err);
1039
1040         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1041         lprof = class_get_profile(profilenm);
1042         if (lprof == NULL) {
1043                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1044                                    " read from the MGS.  Does that filesystem "
1045                                    "exist?\n", profilenm);
1046                 GOTO(out_free, err = -EINVAL);
1047         }
1048         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1049                lprof->lp_md, lprof->lp_dt);
1050
1051         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1052         if (!dt)
1053                 GOTO(out_free, err = -ENOMEM);
1054         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1055
1056         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1057         if (!md)
1058                 GOTO(out_free, err = -ENOMEM);
1059         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1060
1061         /* connections, registrations, sb setup */
1062         err = client_common_fill_super(sb, md, dt, mnt);
1063         if (err < 0)
1064                 GOTO(out_free, err);
1065
1066         sbi->ll_client_common_fill_super_succeeded = 1;
1067
1068 out_free:
1069         if (md)
1070                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1071         if (dt)
1072                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1073         if (lprof != NULL)
1074                 class_put_profile(lprof);
1075         if (err)
1076                 ll_put_super(sb);
1077         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1078                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1079
1080         OBD_FREE_PTR(cfg);
1081         RETURN(err);
1082 } /* ll_fill_super */
1083
1084 void ll_put_super(struct super_block *sb)
1085 {
1086         struct config_llog_instance cfg, params_cfg;
1087         struct obd_device *obd;
1088         struct lustre_sb_info *lsi = s2lsi(sb);
1089         struct ll_sb_info *sbi = ll_s2sbi(sb);
1090         char *profilenm = get_profile_name(sb);
1091         long ccc_count;
1092         int next, force = 1, rc = 0;
1093         ENTRY;
1094
1095         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1096
1097         cfg.cfg_instance = sb;
1098         lustre_end_log(sb, profilenm, &cfg);
1099
1100         params_cfg.cfg_instance = sb;
1101         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1102
1103         if (sbi->ll_md_exp) {
1104                 obd = class_exp2obd(sbi->ll_md_exp);
1105                 if (obd)
1106                         force = obd->obd_force;
1107         }
1108
1109         /* Wait for unstable pages to be committed to stable storage */
1110         if (force == 0) {
1111                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1112                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1113                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1114                         &lwi);
1115         }
1116
1117         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1118         if (force == 0 && rc != -EINTR)
1119                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1120
1121
1122         /* We need to set force before the lov_disconnect in
1123            lustre_common_put_super, since l_d cleans up osc's as well. */
1124         if (force) {
1125                 next = 0;
1126                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1127                                                      &next)) != NULL) {
1128                         obd->obd_force = force;
1129                 }
1130         }
1131
1132         if (sbi->ll_client_common_fill_super_succeeded) {
1133                 /* Only if client_common_fill_super succeeded */
1134                 client_common_put_super(sb);
1135         }
1136
1137         next = 0;
1138         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1139                 class_manual_cleanup(obd);
1140         }
1141
1142         if (sbi->ll_flags & LL_SBI_VERBOSE)
1143                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1144
1145         if (profilenm)
1146                 class_del_profile(profilenm);
1147
1148         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1149                 bdi_destroy(&lsi->lsi_bdi);
1150                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1151         }
1152
1153         ll_free_sbi(sb);
1154         lsi->lsi_llsbi = NULL;
1155
1156         lustre_common_put_super(sb);
1157
1158         cl_env_cache_purge(~0);
1159
1160         module_put(THIS_MODULE);
1161
1162         EXIT;
1163 } /* client_put_super */
1164
1165 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1166 {
1167         struct inode *inode = NULL;
1168
1169         /* NOTE: we depend on atomic igrab() -bzzz */
1170         lock_res_and_lock(lock);
1171         if (lock->l_resource->lr_lvb_inode) {
1172                 struct ll_inode_info * lli;
1173                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1174                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1175                         inode = igrab(lock->l_resource->lr_lvb_inode);
1176                 } else {
1177                         inode = lock->l_resource->lr_lvb_inode;
1178                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1179                                          D_WARNING, lock, "lr_lvb_inode %p is "
1180                                          "bogus: magic %08x",
1181                                          lock->l_resource->lr_lvb_inode,
1182                                          lli->lli_inode_magic);
1183                         inode = NULL;
1184                 }
1185         }
1186         unlock_res_and_lock(lock);
1187         return inode;
1188 }
1189
1190 void ll_dir_clear_lsm_md(struct inode *inode)
1191 {
1192         struct ll_inode_info *lli = ll_i2info(inode);
1193
1194         LASSERT(S_ISDIR(inode->i_mode));
1195
1196         if (lli->lli_lsm_md != NULL) {
1197                 lmv_free_memmd(lli->lli_lsm_md);
1198                 lli->lli_lsm_md = NULL;
1199         }
1200 }
1201
1202 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1203                                       const struct lu_fid *fid,
1204                                       struct lustre_md *md)
1205 {
1206         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1207         struct mdt_body         *body = md->body;
1208         struct inode            *inode;
1209         ino_t                   ino;
1210         ENTRY;
1211
1212         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1213         inode = iget_locked(sb, ino);
1214         if (inode == NULL) {
1215                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1216                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1217                 RETURN(ERR_PTR(-ENOENT));
1218         }
1219
1220         if (inode->i_state & I_NEW) {
1221                 struct ll_inode_info *lli = ll_i2info(inode);
1222                 struct lmv_stripe_md *lsm = md->lmv;
1223
1224                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1225                                 (body->mbo_mode & S_IFMT);
1226                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1227                          PFID(fid));
1228
1229                 LTIME_S(inode->i_mtime) = 0;
1230                 LTIME_S(inode->i_atime) = 0;
1231                 LTIME_S(inode->i_ctime) = 0;
1232                 inode->i_rdev = 0;
1233
1234 #ifdef HAVE_BACKING_DEV_INFO
1235                 /* initializing backing dev info. */
1236                 inode->i_mapping->backing_dev_info =
1237                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1238 #endif
1239                 inode->i_op = &ll_dir_inode_operations;
1240                 inode->i_fop = &ll_dir_operations;
1241                 lli->lli_fid = *fid;
1242                 ll_lli_init(lli);
1243
1244                 LASSERT(lsm != NULL);
1245                 /* master object FID */
1246                 lli->lli_pfid = body->mbo_fid1;
1247                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1248                        lli, PFID(fid), PFID(&lli->lli_pfid));
1249                 unlock_new_inode(inode);
1250         }
1251
1252         RETURN(inode);
1253 }
1254
1255 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1256 {
1257         struct lu_fid *fid;
1258         struct lmv_stripe_md *lsm = md->lmv;
1259         int i;
1260
1261         LASSERT(lsm != NULL);
1262         /* XXX sigh, this lsm_root initialization should be in
1263          * LMV layer, but it needs ll_iget right now, so we
1264          * put this here right now. */
1265         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1266                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1267                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1268                 /* Unfortunately ll_iget will call ll_update_inode,
1269                  * where the initialization of slave inode is slightly
1270                  * different, so it reset lsm_md to NULL to avoid
1271                  * initializing lsm for slave inode. */
1272                 /* For migrating inode, master stripe and master object will
1273                  * be same, so we only need assign this inode */
1274                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1275                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1276                 else
1277                         lsm->lsm_md_oinfo[i].lmo_root =
1278                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1279
1280                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1281                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1282
1283                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1284                         return rc;
1285                 }
1286         }
1287
1288         return 0;
1289 }
1290
1291 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1292                                 const struct lmv_stripe_md *lsm_md2)
1293 {
1294         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1295                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1296                lsm_md1->lsm_md_master_mdt_index ==
1297                                         lsm_md2->lsm_md_master_mdt_index &&
1298                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1299                lsm_md1->lsm_md_layout_version ==
1300                                         lsm_md2->lsm_md_layout_version &&
1301                strcmp(lsm_md1->lsm_md_pool_name,
1302                       lsm_md2->lsm_md_pool_name) == 0;
1303 }
1304
1305 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1306 {
1307         struct ll_inode_info *lli = ll_i2info(inode);
1308         struct lmv_stripe_md *lsm = md->lmv;
1309         int     rc;
1310         ENTRY;
1311
1312         LASSERT(S_ISDIR(inode->i_mode));
1313         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1314                PFID(ll_inode2fid(inode)));
1315
1316         /* no striped information from request. */
1317         if (lsm == NULL) {
1318                 if (lli->lli_lsm_md == NULL) {
1319                         RETURN(0);
1320                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1321                                                 LMV_HASH_FLAG_MIGRATION) {
1322                         /* migration is done, the temporay MIGRATE layout has
1323                          * been removed */
1324                         CDEBUG(D_INODE, DFID" finish migration.\n",
1325                                PFID(ll_inode2fid(inode)));
1326                         lmv_free_memmd(lli->lli_lsm_md);
1327                         lli->lli_lsm_md = NULL;
1328                         RETURN(0);
1329                 } else {
1330                         /* The lustre_md from req does not include stripeEA,
1331                          * see ll_md_setattr */
1332                         RETURN(0);
1333                 }
1334         }
1335
1336         /* set the directory layout */
1337         if (lli->lli_lsm_md == NULL) {
1338                 struct cl_attr  *attr;
1339
1340                 rc = ll_init_lsm_md(inode, md);
1341                 if (rc != 0)
1342                         RETURN(rc);
1343
1344                 /* set md->lmv to NULL, so the following free lustre_md
1345                  * will not free this lsm */
1346                 md->lmv = NULL;
1347                 lli->lli_lsm_md = lsm;
1348
1349                 OBD_ALLOC_PTR(attr);
1350                 if (attr == NULL)
1351                         RETURN(-ENOMEM);
1352
1353                 /* validate the lsm */
1354                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1355                                    ll_md_blocking_ast);
1356                 if (rc != 0) {
1357                         OBD_FREE_PTR(attr);
1358                         RETURN(rc);
1359                 }
1360
1361                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1362                         md->body->mbo_nlink = attr->cat_nlink;
1363                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1364                         md->body->mbo_size = attr->cat_size;
1365                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1366                         md->body->mbo_atime = attr->cat_atime;
1367                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1368                         md->body->mbo_ctime = attr->cat_ctime;
1369                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1370                         md->body->mbo_mtime = attr->cat_mtime;
1371
1372                 OBD_FREE_PTR(attr);
1373
1374                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1375                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1376                 RETURN(0);
1377         }
1378
1379         /* Compare the old and new stripe information */
1380         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1381                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1382                 int                     idx;
1383
1384                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1385                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1386                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1387                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1388                        inode, lsm, old_lsm,
1389                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1390                        lsm->lsm_md_stripe_count,
1391                        old_lsm->lsm_md_stripe_count,
1392                        lsm->lsm_md_master_mdt_index,
1393                        old_lsm->lsm_md_master_mdt_index,
1394                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1395                        lsm->lsm_md_layout_version,
1396                        old_lsm->lsm_md_layout_version,
1397                        lsm->lsm_md_pool_name,
1398                        old_lsm->lsm_md_pool_name);
1399
1400                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1401                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1402                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1403                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1404                 }
1405
1406                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1407                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1408                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1409                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1410                 }
1411
1412                 RETURN(-EIO);
1413         }
1414
1415         RETURN(0);
1416 }
1417
1418 void ll_clear_inode(struct inode *inode)
1419 {
1420         struct ll_inode_info *lli = ll_i2info(inode);
1421         struct ll_sb_info *sbi = ll_i2sbi(inode);
1422         ENTRY;
1423
1424         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1425                PFID(ll_inode2fid(inode)), inode);
1426
1427         if (S_ISDIR(inode->i_mode)) {
1428                 /* these should have been cleared in ll_file_release */
1429                 LASSERT(lli->lli_opendir_key == NULL);
1430                 LASSERT(lli->lli_sai == NULL);
1431                 LASSERT(lli->lli_opendir_pid == 0);
1432         }
1433
1434         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1435
1436         LASSERT(!lli->lli_open_fd_write_count);
1437         LASSERT(!lli->lli_open_fd_read_count);
1438         LASSERT(!lli->lli_open_fd_exec_count);
1439
1440         if (lli->lli_mds_write_och)
1441                 ll_md_real_close(inode, FMODE_WRITE);
1442         if (lli->lli_mds_exec_och)
1443                 ll_md_real_close(inode, FMODE_EXEC);
1444         if (lli->lli_mds_read_och)
1445                 ll_md_real_close(inode, FMODE_READ);
1446
1447         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1448                 OBD_FREE(lli->lli_symlink_name,
1449                          strlen(lli->lli_symlink_name) + 1);
1450                 lli->lli_symlink_name = NULL;
1451         }
1452
1453         ll_xattr_cache_destroy(inode);
1454
1455         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1456                 LASSERT(lli->lli_posix_acl == NULL);
1457                 if (lli->lli_remote_perms) {
1458                         free_rmtperm_hash(lli->lli_remote_perms);
1459                         lli->lli_remote_perms = NULL;
1460                 }
1461         }
1462 #ifdef CONFIG_FS_POSIX_ACL
1463         else if (lli->lli_posix_acl) {
1464                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1465                 LASSERT(lli->lli_remote_perms == NULL);
1466                 posix_acl_release(lli->lli_posix_acl);
1467                 lli->lli_posix_acl = NULL;
1468         }
1469 #endif
1470         lli->lli_inode_magic = LLI_INODE_DEAD;
1471
1472         if (S_ISDIR(inode->i_mode))
1473                 ll_dir_clear_lsm_md(inode);
1474         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1475                 LASSERT(list_empty(&lli->lli_agl_list));
1476
1477         /*
1478          * XXX This has to be done before lsm is freed below, because
1479          * cl_object still uses inode lsm.
1480          */
1481         cl_inode_fini(inode);
1482
1483         EXIT;
1484 }
1485
1486 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1487 {
1488         struct lustre_md md;
1489         struct inode *inode = dentry->d_inode;
1490         struct ll_sb_info *sbi = ll_i2sbi(inode);
1491         struct ptlrpc_request *request = NULL;
1492         int rc, ia_valid;
1493         ENTRY;
1494
1495         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1496                                      LUSTRE_OPC_ANY, NULL);
1497         if (IS_ERR(op_data))
1498                 RETURN(PTR_ERR(op_data));
1499
1500         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1501         if (rc) {
1502                 ptlrpc_req_finished(request);
1503                 if (rc == -ENOENT) {
1504                         clear_nlink(inode);
1505                         /* Unlinked special device node? Or just a race?
1506                          * Pretend we done everything. */
1507                         if (!S_ISREG(inode->i_mode) &&
1508                             !S_ISDIR(inode->i_mode)) {
1509                                 ia_valid = op_data->op_attr.ia_valid;
1510                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1511                                 rc = simple_setattr(dentry, &op_data->op_attr);
1512                                 op_data->op_attr.ia_valid = ia_valid;
1513                         }
1514                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1515                         CERROR("md_setattr fails: rc = %d\n", rc);
1516                 }
1517                 RETURN(rc);
1518         }
1519
1520         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1521                               sbi->ll_md_exp, &md);
1522         if (rc) {
1523                 ptlrpc_req_finished(request);
1524                 RETURN(rc);
1525         }
1526
1527         ia_valid = op_data->op_attr.ia_valid;
1528         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1529          * cache is not cleared yet. */
1530         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1531         if (S_ISREG(inode->i_mode))
1532                 mutex_lock(&inode->i_mutex);
1533         rc = simple_setattr(dentry, &op_data->op_attr);
1534         if (S_ISREG(inode->i_mode))
1535                 mutex_unlock(&inode->i_mutex);
1536         op_data->op_attr.ia_valid = ia_valid;
1537
1538         rc = ll_update_inode(inode, &md);
1539         ptlrpc_req_finished(request);
1540
1541         RETURN(rc);
1542 }
1543
1544 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1545  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1546  * keep these values until such a time that objects are allocated for it.
1547  * We do the MDS operations first, as it is checking permissions for us.
1548  * We don't to the MDS RPC if there is nothing that we want to store there,
1549  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1550  * going to do an RPC anyways.
1551  *
1552  * If we are doing a truncate, we will send the mtime and ctime updates
1553  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1554  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1555  * at the same time.
1556  *
1557  * In case of HSMimport, we only set attr on MDS.
1558  */
1559 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1560 {
1561         struct inode *inode = dentry->d_inode;
1562         struct ll_inode_info *lli = ll_i2info(inode);
1563         struct md_op_data *op_data = NULL;
1564         int rc = 0;
1565         ENTRY;
1566
1567         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1568                "valid %x, hsm_import %d\n",
1569                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1570                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1571                hsm_import);
1572
1573         if (attr->ia_valid & ATTR_SIZE) {
1574                 /* Check new size against VFS/VM file size limit and rlimit */
1575                 rc = inode_newsize_ok(inode, attr->ia_size);
1576                 if (rc)
1577                         RETURN(rc);
1578
1579                 /* The maximum Lustre file size is variable, based on the
1580                  * OST maximum object size and number of stripes.  This
1581                  * needs another check in addition to the VFS check above. */
1582                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1583                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1584                                PFID(&lli->lli_fid), attr->ia_size,
1585                                ll_file_maxbytes(inode));
1586                         RETURN(-EFBIG);
1587                 }
1588
1589                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1590         }
1591
1592         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1593         if (attr->ia_valid & TIMES_SET_FLAGS) {
1594                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1595                     !cfs_capable(CFS_CAP_FOWNER))
1596                         RETURN(-EPERM);
1597         }
1598
1599         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1600         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1601             (attr->ia_valid & ATTR_CTIME)) {
1602                 attr->ia_ctime = CFS_CURRENT_TIME;
1603                 attr->ia_valid |= ATTR_CTIME_SET;
1604         }
1605         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1606             (attr->ia_valid & ATTR_ATIME)) {
1607                 attr->ia_atime = CFS_CURRENT_TIME;
1608                 attr->ia_valid |= ATTR_ATIME_SET;
1609         }
1610         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1611             (attr->ia_valid & ATTR_MTIME)) {
1612                 attr->ia_mtime = CFS_CURRENT_TIME;
1613                 attr->ia_valid |= ATTR_MTIME_SET;
1614         }
1615
1616         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1617                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1618                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1619                        cfs_time_current_sec());
1620
1621         if (S_ISREG(inode->i_mode)) {
1622                 if (attr->ia_valid & ATTR_SIZE)
1623                         inode_dio_write_done(inode);
1624                 mutex_unlock(&inode->i_mutex);
1625         }
1626
1627         /* We always do an MDS RPC, even if we're only changing the size;
1628          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1629
1630         OBD_ALLOC_PTR(op_data);
1631         if (op_data == NULL)
1632                 GOTO(out, rc = -ENOMEM);
1633
1634         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1635                 /* If we are changing file size, file content is
1636                  * modified, flag it. */
1637                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1638                 op_data->op_bias |= MDS_DATA_MODIFIED;
1639                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1640         }
1641
1642         op_data->op_attr = *attr;
1643
1644         rc = ll_md_setattr(dentry, op_data);
1645         if (rc)
1646                 GOTO(out, rc);
1647
1648         if (!S_ISREG(inode->i_mode) || hsm_import)
1649                 GOTO(out, rc = 0);
1650
1651         if (attr->ia_valid & (ATTR_SIZE |
1652                               ATTR_ATIME | ATTR_ATIME_SET |
1653                               ATTR_MTIME | ATTR_MTIME_SET |
1654                               ATTR_CTIME | ATTR_CTIME_SET)) {
1655                 /* For truncate and utimes sending attributes to OSTs, setting
1656                  * mtime/atime to the past will be performed under PW [0:EOF]
1657                  * extent lock (new_size:EOF for truncate).  It may seem
1658                  * excessive to send mtime/atime updates to OSTs when not
1659                  * setting times to past, but it is necessary due to possible
1660                  * time de-synchronization between MDT inode and OST objects */
1661                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1662         }
1663
1664         /* If the file was restored, it needs to set dirty flag.
1665          *
1666          * We've already sent MDS_DATA_MODIFIED flag in
1667          * ll_md_setattr() for truncate. However, the MDT refuses to
1668          * set the HS_DIRTY flag on released files, so we have to set
1669          * it again if the file has been restored. Please check how
1670          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1671          *
1672          * Please notice that if the file is not released, the previous
1673          * MDS_DATA_MODIFIED has taken effect and usually
1674          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1675          * This way we can save an RPC for common open + trunc
1676          * operation. */
1677         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1678                 struct hsm_state_set hss = {
1679                         .hss_valid = HSS_SETMASK,
1680                         .hss_setmask = HS_DIRTY,
1681                 };
1682                 int rc2;
1683
1684                 rc2 = ll_hsm_state_set(inode, &hss);
1685                 /* truncate and write can happen at the same time, so that
1686                  * the file can be set modified even though the file is not
1687                  * restored from released state, and ll_hsm_state_set() is
1688                  * not applicable for the file, and rc2 < 0 is normal in this
1689                  * case. */
1690                 if (rc2 < 0)
1691                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1692                                PFID(ll_inode2fid(inode)), rc2);
1693         }
1694
1695         EXIT;
1696 out:
1697         if (op_data != NULL)
1698                 ll_finish_md_op_data(op_data);
1699
1700         if (S_ISREG(inode->i_mode)) {
1701                 mutex_lock(&inode->i_mutex);
1702                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1703                         inode_dio_wait(inode);
1704         }
1705
1706         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1707                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1708
1709         return rc;
1710 }
1711
1712 int ll_setattr(struct dentry *de, struct iattr *attr)
1713 {
1714         int mode = de->d_inode->i_mode;
1715
1716         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1717                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1718                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1719
1720         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1721                                (ATTR_SIZE|ATTR_MODE)) &&
1722             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1723              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1724               !(attr->ia_mode & S_ISGID))))
1725                 attr->ia_valid |= ATTR_FORCE;
1726
1727         if ((attr->ia_valid & ATTR_MODE) &&
1728             (mode & S_ISUID) &&
1729             !(attr->ia_mode & S_ISUID) &&
1730             !(attr->ia_valid & ATTR_KILL_SUID))
1731                 attr->ia_valid |= ATTR_KILL_SUID;
1732
1733         if ((attr->ia_valid & ATTR_MODE) &&
1734             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1735             !(attr->ia_mode & S_ISGID) &&
1736             !(attr->ia_valid & ATTR_KILL_SGID))
1737                 attr->ia_valid |= ATTR_KILL_SGID;
1738
1739         return ll_setattr_raw(de, attr, false);
1740 }
1741
1742 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1743                        __u64 max_age, __u32 flags)
1744 {
1745         struct ll_sb_info *sbi = ll_s2sbi(sb);
1746         struct obd_statfs obd_osfs;
1747         int rc;
1748         ENTRY;
1749
1750         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1751         if (rc) {
1752                 CERROR("md_statfs fails: rc = %d\n", rc);
1753                 RETURN(rc);
1754         }
1755
1756         osfs->os_type = sb->s_magic;
1757
1758         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1759                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1760
1761         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1762                 flags |= OBD_STATFS_NODELAY;
1763
1764         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1765         if (rc) {
1766                 CERROR("obd_statfs fails: rc = %d\n", rc);
1767                 RETURN(rc);
1768         }
1769
1770         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1771                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1772                obd_osfs.os_files);
1773
1774         osfs->os_bsize = obd_osfs.os_bsize;
1775         osfs->os_blocks = obd_osfs.os_blocks;
1776         osfs->os_bfree = obd_osfs.os_bfree;
1777         osfs->os_bavail = obd_osfs.os_bavail;
1778
1779         /* If we don't have as many objects free on the OST as inodes
1780          * on the MDS, we reduce the total number of inodes to
1781          * compensate, so that the "inodes in use" number is correct.
1782          */
1783         if (obd_osfs.os_ffree < osfs->os_ffree) {
1784                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1785                         obd_osfs.os_ffree;
1786                 osfs->os_ffree = obd_osfs.os_ffree;
1787         }
1788
1789         RETURN(rc);
1790 }
1791 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1792 {
1793         struct super_block *sb = de->d_sb;
1794         struct obd_statfs osfs;
1795         __u64 fsid = huge_encode_dev(sb->s_dev);
1796         int rc;
1797
1798         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1799         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1800
1801         /* Some amount of caching on the client is allowed */
1802         rc = ll_statfs_internal(sb, &osfs,
1803                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1804                                 0);
1805         if (rc)
1806                 return rc;
1807
1808         statfs_unpack(sfs, &osfs);
1809
1810         /* We need to downshift for all 32-bit kernels, because we can't
1811          * tell if the kernel is being called via sys_statfs64() or not.
1812          * Stop before overflowing f_bsize - in which case it is better
1813          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1814         if (sizeof(long) < 8) {
1815                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1816                         sfs->f_bsize <<= 1;
1817
1818                         osfs.os_blocks >>= 1;
1819                         osfs.os_bfree >>= 1;
1820                         osfs.os_bavail >>= 1;
1821                 }
1822         }
1823
1824         sfs->f_blocks = osfs.os_blocks;
1825         sfs->f_bfree = osfs.os_bfree;
1826         sfs->f_bavail = osfs.os_bavail;
1827         sfs->f_fsid.val[0] = (__u32)fsid;
1828         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1829         return 0;
1830 }
1831
1832 void ll_inode_size_lock(struct inode *inode)
1833 {
1834         struct ll_inode_info *lli;
1835
1836         LASSERT(!S_ISDIR(inode->i_mode));
1837
1838         lli = ll_i2info(inode);
1839         mutex_lock(&lli->lli_size_mutex);
1840 }
1841
1842 void ll_inode_size_unlock(struct inode *inode)
1843 {
1844         struct ll_inode_info *lli;
1845
1846         lli = ll_i2info(inode);
1847         mutex_unlock(&lli->lli_size_mutex);
1848 }
1849
1850 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1851 {
1852         struct ll_inode_info *lli = ll_i2info(inode);
1853         struct mdt_body *body = md->body;
1854         struct ll_sb_info *sbi = ll_i2sbi(inode);
1855
1856         if (body->mbo_valid & OBD_MD_FLEASIZE)
1857                 cl_file_inode_init(inode, md);
1858
1859         if (S_ISDIR(inode->i_mode)) {
1860                 int     rc;
1861
1862                 rc = ll_update_lsm_md(inode, md);
1863                 if (rc != 0)
1864                         return rc;
1865         }
1866
1867         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1868                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1869                         ll_update_remote_perm(inode, md->remote_perm);
1870         }
1871 #ifdef CONFIG_FS_POSIX_ACL
1872         else if (body->mbo_valid & OBD_MD_FLACL) {
1873                 spin_lock(&lli->lli_lock);
1874                 if (lli->lli_posix_acl)
1875                         posix_acl_release(lli->lli_posix_acl);
1876                 lli->lli_posix_acl = md->posix_acl;
1877                 spin_unlock(&lli->lli_lock);
1878         }
1879 #endif
1880         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1881                                         sbi->ll_flags & LL_SBI_32BIT_API);
1882         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1883
1884         if (body->mbo_valid & OBD_MD_FLATIME) {
1885                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1886                         LTIME_S(inode->i_atime) = body->mbo_atime;
1887                 lli->lli_atime = body->mbo_atime;
1888         }
1889
1890         if (body->mbo_valid & OBD_MD_FLMTIME) {
1891                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1892                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1893                                "to "LPU64"\n", inode->i_ino,
1894                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1895                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1896                 }
1897                 lli->lli_mtime = body->mbo_mtime;
1898         }
1899
1900         if (body->mbo_valid & OBD_MD_FLCTIME) {
1901                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1902                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1903                 lli->lli_ctime = body->mbo_ctime;
1904         }
1905
1906         if (body->mbo_valid & OBD_MD_FLMODE)
1907                 inode->i_mode = (inode->i_mode & S_IFMT) |
1908                                 (body->mbo_mode & ~S_IFMT);
1909
1910         if (body->mbo_valid & OBD_MD_FLTYPE)
1911                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1912                                 (body->mbo_mode & S_IFMT);
1913
1914         LASSERT(inode->i_mode != 0);
1915         if (S_ISREG(inode->i_mode))
1916                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1917                                        LL_MAX_BLKSIZE_BITS);
1918         else
1919                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1920
1921         if (body->mbo_valid & OBD_MD_FLUID)
1922                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1923         if (body->mbo_valid & OBD_MD_FLGID)
1924                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1925         if (body->mbo_valid & OBD_MD_FLFLAGS)
1926                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1927         if (body->mbo_valid & OBD_MD_FLNLINK)
1928                 set_nlink(inode, body->mbo_nlink);
1929         if (body->mbo_valid & OBD_MD_FLRDEV)
1930                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1931
1932         if (body->mbo_valid & OBD_MD_FLID) {
1933                 /* FID shouldn't be changed! */
1934                 if (fid_is_sane(&lli->lli_fid)) {
1935                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1936                                  "Trying to change FID "DFID
1937                                  " to the "DFID", inode "DFID"(%p)\n",
1938                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1939                                  PFID(ll_inode2fid(inode)), inode);
1940                 } else {
1941                         lli->lli_fid = body->mbo_fid1;
1942                 }
1943         }
1944
1945         LASSERT(fid_seq(&lli->lli_fid) != 0);
1946
1947         if (body->mbo_valid & OBD_MD_FLSIZE) {
1948                 i_size_write(inode, body->mbo_size);
1949
1950                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1951                        PFID(ll_inode2fid(inode)),
1952                        (unsigned long long)body->mbo_size);
1953
1954                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1955                         inode->i_blocks = body->mbo_blocks;
1956         }
1957
1958         if (body->mbo_valid & OBD_MD_TSTATE) {
1959                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1960                  * clear it when done to ensure to start again
1961                  * glimpsing updated attrs
1962                  */
1963                 if (body->mbo_t_state & MS_RESTORE)
1964                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1965                 else
1966                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1967         }
1968
1969         return 0;
1970 }
1971
1972 int ll_read_inode2(struct inode *inode, void *opaque)
1973 {
1974         struct lustre_md *md = opaque;
1975         struct ll_inode_info *lli = ll_i2info(inode);
1976         int     rc;
1977         ENTRY;
1978
1979         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1980                PFID(&lli->lli_fid), inode);
1981
1982         /* Core attributes from the MDS first.  This is a new inode, and
1983          * the VFS doesn't zero times in the core inode so we have to do
1984          * it ourselves.  They will be overwritten by either MDS or OST
1985          * attributes - we just need to make sure they aren't newer. */
1986         LTIME_S(inode->i_mtime) = 0;
1987         LTIME_S(inode->i_atime) = 0;
1988         LTIME_S(inode->i_ctime) = 0;
1989         inode->i_rdev = 0;
1990         rc = ll_update_inode(inode, md);
1991         if (rc != 0)
1992                 RETURN(rc);
1993
1994         /* OIDEBUG(inode); */
1995
1996 #ifdef HAVE_BACKING_DEV_INFO
1997         /* initializing backing dev info. */
1998         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1999 #endif
2000         if (S_ISREG(inode->i_mode)) {
2001                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2002                 inode->i_op = &ll_file_inode_operations;
2003                 inode->i_fop = sbi->ll_fop;
2004                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2005                 EXIT;
2006         } else if (S_ISDIR(inode->i_mode)) {
2007                 inode->i_op = &ll_dir_inode_operations;
2008                 inode->i_fop = &ll_dir_operations;
2009                 EXIT;
2010         } else if (S_ISLNK(inode->i_mode)) {
2011                 inode->i_op = &ll_fast_symlink_inode_operations;
2012                 EXIT;
2013         } else {
2014                 inode->i_op = &ll_special_inode_operations;
2015
2016                 init_special_inode(inode, inode->i_mode,
2017                                    inode->i_rdev);
2018
2019                 EXIT;
2020         }
2021
2022         return 0;
2023 }
2024
2025 void ll_delete_inode(struct inode *inode)
2026 {
2027         struct ll_inode_info *lli = ll_i2info(inode);
2028         ENTRY;
2029
2030         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2031                 /* It is last chance to write out dirty pages,
2032                  * otherwise we may lose data while umount */
2033                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2034
2035         truncate_inode_pages_final(&inode->i_data);
2036
2037         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2038                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2039                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2040
2041 #ifdef HAVE_SBOPS_EVICT_INODE
2042         ll_clear_inode(inode);
2043 #endif
2044         clear_inode(inode);
2045
2046         EXIT;
2047 }
2048
2049 int ll_iocontrol(struct inode *inode, struct file *file,
2050                  unsigned int cmd, unsigned long arg)
2051 {
2052         struct ll_sb_info *sbi = ll_i2sbi(inode);
2053         struct ptlrpc_request *req = NULL;
2054         int rc, flags = 0;
2055         ENTRY;
2056
2057         switch(cmd) {
2058         case FSFILT_IOC_GETFLAGS: {
2059                 struct mdt_body *body;
2060                 struct md_op_data *op_data;
2061
2062                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2063                                              0, 0, LUSTRE_OPC_ANY,
2064                                              NULL);
2065                 if (IS_ERR(op_data))
2066                         RETURN(PTR_ERR(op_data));
2067
2068                 op_data->op_valid = OBD_MD_FLFLAGS;
2069                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2070                 ll_finish_md_op_data(op_data);
2071                 if (rc) {
2072                         CERROR("%s: failure inode "DFID": rc = %d\n",
2073                                sbi->ll_md_exp->exp_obd->obd_name,
2074                                PFID(ll_inode2fid(inode)), rc);
2075                         RETURN(-abs(rc));
2076                 }
2077
2078                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2079
2080                 flags = body->mbo_flags;
2081
2082                 ptlrpc_req_finished(req);
2083
2084                 RETURN(put_user(flags, (int __user *)arg));
2085         }
2086         case FSFILT_IOC_SETFLAGS: {
2087                 struct iattr *attr;
2088                 struct md_op_data *op_data;
2089                 struct cl_object *obj;
2090
2091                 if (get_user(flags, (int __user *)arg))
2092                         RETURN(-EFAULT);
2093
2094                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2095                                              LUSTRE_OPC_ANY, NULL);
2096                 if (IS_ERR(op_data))
2097                         RETURN(PTR_ERR(op_data));
2098
2099                 op_data->op_attr_flags = flags;
2100                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2101                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2102                 ll_finish_md_op_data(op_data);
2103                 ptlrpc_req_finished(req);
2104                 if (rc)
2105                         RETURN(rc);
2106
2107                 inode->i_flags = ll_ext_to_inode_flags(flags);
2108
2109                 obj = ll_i2info(inode)->lli_clob;
2110                 if (obj == NULL)
2111                         RETURN(0);
2112
2113                 OBD_ALLOC_PTR(attr);
2114                 if (attr == NULL)
2115                         RETURN(-ENOMEM);
2116
2117                 attr->ia_valid = ATTR_ATTR_FLAG;
2118                 rc = cl_setattr_ost(obj, attr, flags);
2119
2120                 OBD_FREE_PTR(attr);
2121                 RETURN(rc);
2122         }
2123         default:
2124                 RETURN(-ENOSYS);
2125         }
2126
2127         RETURN(0);
2128 }
2129
2130 int ll_flush_ctx(struct inode *inode)
2131 {
2132         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2133
2134         CDEBUG(D_SEC, "flush context for user %d\n",
2135                from_kuid(&init_user_ns, current_uid()));
2136
2137         obd_set_info_async(NULL, sbi->ll_md_exp,
2138                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2139                            0, NULL, NULL);
2140         obd_set_info_async(NULL, sbi->ll_dt_exp,
2141                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2142                            0, NULL, NULL);
2143         return 0;
2144 }
2145
2146 /* umount -f client means force down, don't save state */
2147 void ll_umount_begin(struct super_block *sb)
2148 {
2149         struct ll_sb_info *sbi = ll_s2sbi(sb);
2150         struct obd_device *obd;
2151         struct obd_ioctl_data *ioc_data;
2152         ENTRY;
2153
2154         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2155                sb->s_count, atomic_read(&sb->s_active));
2156
2157         obd = class_exp2obd(sbi->ll_md_exp);
2158         if (obd == NULL) {
2159                 CERROR("Invalid MDC connection handle "LPX64"\n",
2160                        sbi->ll_md_exp->exp_handle.h_cookie);
2161                 EXIT;
2162                 return;
2163         }
2164         obd->obd_force = 1;
2165
2166         obd = class_exp2obd(sbi->ll_dt_exp);
2167         if (obd == NULL) {
2168                 CERROR("Invalid LOV connection handle "LPX64"\n",
2169                        sbi->ll_dt_exp->exp_handle.h_cookie);
2170                 EXIT;
2171                 return;
2172         }
2173         obd->obd_force = 1;
2174
2175         OBD_ALLOC_PTR(ioc_data);
2176         if (ioc_data) {
2177                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2178                               sizeof *ioc_data, ioc_data, NULL);
2179
2180                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2181                               sizeof *ioc_data, ioc_data, NULL);
2182
2183                 OBD_FREE_PTR(ioc_data);
2184         }
2185
2186         /* Really, we'd like to wait until there are no requests outstanding,
2187          * and then continue.  For now, we just invalidate the requests,
2188          * schedule() and sleep one second if needed, and hope.
2189          */
2190         schedule();
2191         EXIT;
2192 }
2193
2194 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2195 {
2196         struct ll_sb_info *sbi = ll_s2sbi(sb);
2197         char *profilenm = get_profile_name(sb);
2198         int err;
2199         __u32 read_only;
2200
2201         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2202                 read_only = *flags & MS_RDONLY;
2203                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2204                                          sizeof(KEY_READ_ONLY),
2205                                          KEY_READ_ONLY, sizeof(read_only),
2206                                          &read_only, NULL);
2207                 if (err) {
2208                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2209                                       profilenm, read_only ?
2210                                       "read-only" : "read-write", err);
2211                         return err;
2212                 }
2213
2214                 if (read_only)
2215                         sb->s_flags |= MS_RDONLY;
2216                 else
2217                         sb->s_flags &= ~MS_RDONLY;
2218
2219                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2220                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2221                                       read_only ?  "read-only" : "read-write");
2222         }
2223         return 0;
2224 }
2225
2226 /**
2227  * Cleanup the open handle that is cached on MDT-side.
2228  *
2229  * For open case, the client side open handling thread may hit error
2230  * after the MDT grant the open. Under such case, the client should
2231  * send close RPC to the MDT as cleanup; otherwise, the open handle
2232  * on the MDT will be leaked there until the client umount or evicted.
2233  *
2234  * In further, if someone unlinked the file, because the open handle
2235  * holds the reference on such file/object, then it will block the
2236  * subsequent threads that want to locate such object via FID.
2237  *
2238  * \param[in] sb        super block for this file-system
2239  * \param[in] open_req  pointer to the original open request
2240  */
2241 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2242 {
2243         struct mdt_body                 *body;
2244         struct md_op_data               *op_data;
2245         struct ptlrpc_request           *close_req = NULL;
2246         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2247         ENTRY;
2248
2249         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2250         OBD_ALLOC_PTR(op_data);
2251         if (op_data == NULL) {
2252                 CWARN("%s: cannot allocate op_data to release open handle for "
2253                       DFID"\n",
2254                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2255
2256                 RETURN_EXIT;
2257         }
2258
2259         op_data->op_fid1 = body->mbo_fid1;
2260         op_data->op_handle = body->mbo_handle;
2261         op_data->op_mod_time = cfs_time_current_sec();
2262         md_close(exp, op_data, NULL, &close_req);
2263         ptlrpc_req_finished(close_req);
2264         ll_finish_md_op_data(op_data);
2265
2266         EXIT;
2267 }
2268
2269 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2270                   struct super_block *sb, struct lookup_intent *it)
2271 {
2272         struct ll_sb_info *sbi = NULL;
2273         struct lustre_md md = { NULL };
2274         int rc;
2275         ENTRY;
2276
2277         LASSERT(*inode || sb);
2278         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2279         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2280                               sbi->ll_md_exp, &md);
2281         if (rc != 0)
2282                 GOTO(cleanup, rc);
2283
2284         if (*inode) {
2285                 rc = ll_update_inode(*inode, &md);
2286                 if (rc != 0)
2287                         GOTO(out, rc);
2288         } else {
2289                 LASSERT(sb != NULL);
2290
2291                 /*
2292                  * At this point server returns to client's same fid as client
2293                  * generated for creating. So using ->fid1 is okay here.
2294                  */
2295                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2296
2297                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2298                                              sbi->ll_flags & LL_SBI_32BIT_API),
2299                                  &md);
2300                 if (IS_ERR(*inode)) {
2301 #ifdef CONFIG_FS_POSIX_ACL
2302                         if (md.posix_acl) {
2303                                 posix_acl_release(md.posix_acl);
2304                                 md.posix_acl = NULL;
2305                         }
2306 #endif
2307                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2308                         *inode = NULL;
2309                         CERROR("new_inode -fatal: rc %d\n", rc);
2310                         GOTO(out, rc);
2311                 }
2312         }
2313
2314         /* Handling piggyback layout lock.
2315          * Layout lock can be piggybacked by getattr and open request.
2316          * The lsm can be applied to inode only if it comes with a layout lock
2317          * otherwise correct layout may be overwritten, for example:
2318          * 1. proc1: mdt returns a lsm but not granting layout
2319          * 2. layout was changed by another client
2320          * 3. proc2: refresh layout and layout lock granted
2321          * 4. proc1: to apply a stale layout */
2322         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2323                 struct lustre_handle lockh;
2324                 struct ldlm_lock *lock;
2325
2326                 lockh.cookie = it->d.lustre.it_lock_handle;
2327                 lock = ldlm_handle2lock(&lockh);
2328                 LASSERT(lock != NULL);
2329                 if (ldlm_has_layout(lock)) {
2330                         struct cl_object_conf conf;
2331
2332                         memset(&conf, 0, sizeof(conf));
2333                         conf.coc_opc = OBJECT_CONF_SET;
2334                         conf.coc_inode = *inode;
2335                         conf.coc_lock = lock;
2336                         conf.u.coc_layout = md.layout;
2337                         (void)ll_layout_conf(*inode, &conf);
2338                 }
2339                 LDLM_LOCK_PUT(lock);
2340         }
2341
2342         GOTO(out, rc = 0);
2343
2344 out:
2345         md_free_lustre_md(sbi->ll_md_exp, &md);
2346
2347 cleanup:
2348         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2349                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2350
2351         return rc;
2352 }
2353
2354 int ll_obd_statfs(struct inode *inode, void __user *arg)
2355 {
2356         struct ll_sb_info *sbi = NULL;
2357         struct obd_export *exp;
2358         char *buf = NULL;
2359         struct obd_ioctl_data *data = NULL;
2360         __u32 type;
2361         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2362         int len = 0, rc;
2363
2364         if (!inode || !(sbi = ll_i2sbi(inode)))
2365                 GOTO(out_statfs, rc = -EINVAL);
2366
2367         rc = obd_ioctl_getdata(&buf, &len, arg);
2368         if (rc)
2369                 GOTO(out_statfs, rc);
2370
2371         data = (void*)buf;
2372         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2373             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2374                 GOTO(out_statfs, rc = -EINVAL);
2375
2376         if (data->ioc_inllen1 != sizeof(__u32) ||
2377             data->ioc_inllen2 != sizeof(__u32) ||
2378             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2379             data->ioc_plen2 != sizeof(struct obd_uuid))
2380                 GOTO(out_statfs, rc = -EINVAL);
2381
2382         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2383         if (type & LL_STATFS_LMV)
2384                 exp = sbi->ll_md_exp;
2385         else if (type & LL_STATFS_LOV)
2386                 exp = sbi->ll_dt_exp;
2387         else
2388                 GOTO(out_statfs, rc = -ENODEV);
2389
2390         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2391         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2392         if (rc)
2393                 GOTO(out_statfs, rc);
2394 out_statfs:
2395         if (buf)
2396                 obd_ioctl_freedata(buf, len);
2397         return rc;
2398 }
2399
2400 int ll_process_config(struct lustre_cfg *lcfg)
2401 {
2402         struct super_block *sb;
2403         unsigned long x;
2404         int rc = 0;
2405         char *ptr;
2406
2407         /* The instance name contains the sb: lustre-client-aacfe000 */
2408         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2409         if (!ptr || !*(++ptr))
2410                 return -EINVAL;
2411         if (sscanf(ptr, "%lx", &x) != 1)
2412                 return -EINVAL;
2413         sb = (struct super_block *)x;
2414         /* This better be a real Lustre superblock! */
2415         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2416
2417         /* Note we have not called client_common_fill_super yet, so
2418            proc fns must be able to handle that! */
2419         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2420                                       lcfg, sb);
2421         if (rc > 0)
2422                 rc = 0;
2423         return rc;
2424 }
2425
2426 /* this function prepares md_op_data hint for passing it down to MD stack. */
2427 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2428                                       struct inode *i1, struct inode *i2,
2429                                       const char *name, size_t namelen,
2430                                       __u32 mode, __u32 opc, void *data)
2431 {
2432         LASSERT(i1 != NULL);
2433
2434         if (name == NULL) {
2435                 /* Do not reuse namelen for something else. */
2436                 if (namelen != 0)
2437                         return ERR_PTR(-EINVAL);
2438         } else {
2439                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2440                         return ERR_PTR(-ENAMETOOLONG);
2441
2442                 if (!lu_name_is_valid_2(name, namelen))
2443                         return ERR_PTR(-EINVAL);
2444         }
2445
2446         if (op_data == NULL)
2447                 OBD_ALLOC_PTR(op_data);
2448
2449         if (op_data == NULL)
2450                 return ERR_PTR(-ENOMEM);
2451
2452         ll_i2gids(op_data->op_suppgids, i1, i2);
2453         op_data->op_fid1 = *ll_inode2fid(i1);
2454         op_data->op_default_stripe_offset = -1;
2455         if (S_ISDIR(i1->i_mode)) {
2456                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2457                 if (opc == LUSTRE_OPC_MKDIR)
2458                         op_data->op_default_stripe_offset =
2459                                    ll_i2info(i1)->lli_def_stripe_offset;
2460         }
2461
2462         if (i2) {
2463                 op_data->op_fid2 = *ll_inode2fid(i2);
2464                 if (S_ISDIR(i2->i_mode))
2465                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2466         } else {
2467                 fid_zero(&op_data->op_fid2);
2468         }
2469
2470         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2471                 op_data->op_cli_flags |= CLI_HASH64;
2472
2473         if (ll_need_32bit_api(ll_i2sbi(i1)))
2474                 op_data->op_cli_flags |= CLI_API32;
2475
2476         op_data->op_name = name;
2477         op_data->op_namelen = namelen;
2478         op_data->op_mode = mode;
2479         op_data->op_mod_time = cfs_time_current_sec();
2480         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2481         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2482         op_data->op_cap = cfs_curproc_cap_pack();
2483         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2484              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2485                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2486         } else {
2487                 op_data->op_mds = 0;
2488         }
2489         op_data->op_data = data;
2490
2491         return op_data;
2492 }
2493
2494 void ll_finish_md_op_data(struct md_op_data *op_data)
2495 {
2496         OBD_FREE_PTR(op_data);
2497 }
2498
2499 #ifdef HAVE_SUPEROPS_USE_DENTRY
2500 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2501 #else
2502 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2503 #endif
2504 {
2505         struct ll_sb_info *sbi;
2506
2507 #ifdef HAVE_SUPEROPS_USE_DENTRY
2508         LASSERT((seq != NULL) && (dentry != NULL));
2509         sbi = ll_s2sbi(dentry->d_sb);
2510 #else
2511         LASSERT((seq != NULL) && (vfs != NULL));
2512         sbi = ll_s2sbi(vfs->mnt_sb);
2513 #endif
2514
2515         if (sbi->ll_flags & LL_SBI_NOLCK)
2516                 seq_puts(seq, ",nolock");
2517
2518         if (sbi->ll_flags & LL_SBI_FLOCK)
2519                 seq_puts(seq, ",flock");
2520
2521         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2522                 seq_puts(seq, ",localflock");
2523
2524         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2525                 seq_puts(seq, ",user_xattr");
2526
2527         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2528                 seq_puts(seq, ",lazystatfs");
2529
2530         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2531                 seq_puts(seq, ",user_fid2path");
2532
2533         RETURN(0);
2534 }
2535
2536 /**
2537  * Get obd name by cmd, and copy out to user space
2538  */
2539 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2540 {
2541         struct ll_sb_info *sbi = ll_i2sbi(inode);
2542         struct obd_device *obd;
2543         ENTRY;
2544
2545         if (cmd == OBD_IOC_GETDTNAME)
2546                 obd = class_exp2obd(sbi->ll_dt_exp);
2547         else if (cmd == OBD_IOC_GETMDNAME)
2548                 obd = class_exp2obd(sbi->ll_md_exp);
2549         else
2550                 RETURN(-EINVAL);
2551
2552         if (!obd)
2553                 RETURN(-ENOENT);
2554
2555         if (copy_to_user((void __user *)arg, obd->obd_name,
2556                          strlen(obd->obd_name) + 1))
2557                 RETURN(-EFAULT);
2558
2559         RETURN(0);
2560 }
2561
2562 /**
2563  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2564  * fsname will be returned in this buffer; otherwise, a static buffer will be
2565  * used to store the fsname and returned to caller.
2566  */
2567 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2568 {
2569         static char fsname_static[MTI_NAME_MAXLEN];
2570         struct lustre_sb_info *lsi = s2lsi(sb);
2571         char *ptr;
2572         int len;
2573
2574         if (buf == NULL) {
2575                 /* this means the caller wants to use static buffer
2576                  * and it doesn't care about race. Usually this is
2577                  * in error reporting path */
2578                 buf = fsname_static;
2579                 buflen = sizeof(fsname_static);
2580         }
2581
2582         len = strlen(lsi->lsi_lmd->lmd_profile);
2583         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2584         if (ptr && (strcmp(ptr, "-client") == 0))
2585                 len -= 7;
2586
2587         if (unlikely(len >= buflen))
2588                 len = buflen - 1;
2589         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2590         buf[len] = '\0';
2591
2592         return buf;
2593 }
2594
2595 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2596 {
2597         char *path = NULL;
2598
2599         struct path p;
2600
2601         p.dentry = dentry;
2602         p.mnt = current->fs->root.mnt;
2603         path_get(&p);
2604         path = d_path(&p, buf, bufsize);
2605         path_put(&p);
2606         return path;
2607 }
2608
2609 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2610 {
2611         char *buf, *path = NULL;
2612         struct dentry *dentry = NULL;
2613         struct inode *inode = page->mapping->host;
2614
2615         /* this can be called inside spin lock so use GFP_ATOMIC. */
2616         buf = (char *)__get_free_page(GFP_ATOMIC);
2617         if (buf != NULL) {
2618                 dentry = d_find_alias(page->mapping->host);
2619                 if (dentry != NULL)
2620                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2621         }
2622
2623         CDEBUG(D_WARNING,
2624                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2625                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2626                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2627                PFID(ll_inode2fid(inode)),
2628                (path && !IS_ERR(path)) ? path : "", ioret);
2629
2630         if (dentry != NULL)
2631                 dput(dentry);
2632
2633         if (buf != NULL)
2634                 free_page((unsigned long)buf);
2635 }
2636
2637 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2638                         struct lov_user_md **kbuf)
2639 {
2640         struct lov_user_md      lum;
2641         ssize_t                 lum_size;
2642         ENTRY;
2643
2644         if (copy_from_user(&lum, md, sizeof(lum)))
2645                 RETURN(-EFAULT);
2646
2647         lum_size = ll_lov_user_md_size(&lum);
2648         if (lum_size < 0)
2649                 RETURN(lum_size);
2650
2651         OBD_ALLOC(*kbuf, lum_size);
2652         if (*kbuf == NULL)
2653                 RETURN(-ENOMEM);
2654
2655         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2656                 OBD_FREE(*kbuf, lum_size);
2657                 RETURN(-EFAULT);
2658         }
2659
2660         RETURN(lum_size);
2661 }
2662
2663 /*
2664  * Compute llite root squash state after a change of root squash
2665  * configuration setting or add/remove of a lnet nid
2666  */
2667 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2668 {
2669         struct root_squash_info *squash = &sbi->ll_squash;
2670         int i;
2671         bool matched;
2672         lnet_process_id_t id;
2673
2674         /* Update norootsquash flag */
2675         down_write(&squash->rsi_sem);
2676         if (list_empty(&squash->rsi_nosquash_nids))
2677                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2678         else {
2679                 /* Do not apply root squash as soon as one of our NIDs is
2680                  * in the nosquash_nids list */
2681                 matched = false;
2682                 i = 0;
2683                 while (LNetGetId(i++, &id) != -ENOENT) {
2684                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2685                                 continue;
2686                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2687                                 matched = true;
2688                                 break;
2689                         }
2690                 }
2691                 if (matched)
2692                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2693                 else
2694                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2695         }
2696         up_write(&squash->rsi_sem);
2697 }
2698
2699 /**
2700  * Parse linkea content to extract information about a given hardlink
2701  *
2702  * \param[in]   ldata      - Initialized linkea data
2703  * \param[in]   linkno     - Link identifier
2704  * \param[out]  parent_fid - The entry's parent FID
2705  * \param[out]  ln         - Entry name destination buffer
2706  *
2707  * \retval 0 on success
2708  * \retval Appropriate negative error code on failure
2709  */
2710 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2711                             struct lu_fid *parent_fid, struct lu_name *ln)
2712 {
2713         unsigned int    idx;
2714         int             rc;
2715         ENTRY;
2716
2717         rc = linkea_init(ldata);
2718         if (rc < 0)
2719                 RETURN(rc);
2720
2721         if (linkno >= ldata->ld_leh->leh_reccount)
2722                 /* beyond last link */
2723                 RETURN(-ENODATA);
2724
2725         linkea_first_entry(ldata);
2726         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2727                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2728                                     parent_fid);
2729                 if (idx == linkno)
2730                         break;
2731
2732                 linkea_next_entry(ldata);
2733         }
2734
2735         if (idx < linkno)
2736                 RETURN(-ENODATA);
2737
2738         RETURN(0);
2739 }
2740
2741 /**
2742  * Get parent FID and name of an identified link. Operation is performed for
2743  * a given link number, letting the caller iterate over linkno to list one or
2744  * all links of an entry.
2745  *
2746  * \param[in]     file - File descriptor against which to perform the operation
2747  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2748  *                       on and the available size. It is eventually filled with
2749  *                       the requested information or left untouched on error
2750  *
2751  * \retval - 0 on success
2752  * \retval - Appropriate negative error code on failure
2753  */
2754 int ll_getparent(struct file *file, struct getparent __user *arg)
2755 {
2756         struct dentry           *dentry = file->f_path.dentry;
2757         struct inode            *inode = dentry->d_inode;
2758         struct linkea_data      *ldata;
2759         struct lu_buf            buf = LU_BUF_NULL;
2760         struct lu_name           ln;
2761         struct lu_fid            parent_fid;
2762         __u32                    linkno;
2763         __u32                    name_size;
2764         int                      rc;
2765
2766         ENTRY;
2767
2768         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2769             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2770                 RETURN(-EPERM);
2771
2772         if (get_user(name_size, &arg->gp_name_size))
2773                 RETURN(-EFAULT);
2774
2775         if (get_user(linkno, &arg->gp_linkno))
2776                 RETURN(-EFAULT);
2777
2778         if (name_size > PATH_MAX)
2779                 RETURN(-EINVAL);
2780
2781         OBD_ALLOC(ldata, sizeof(*ldata));
2782         if (ldata == NULL)
2783                 RETURN(-ENOMEM);
2784
2785         rc = linkea_data_new(ldata, &buf);
2786         if (rc < 0)
2787                 GOTO(ldata_free, rc);
2788
2789         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2790         if (rc < 0)
2791                 GOTO(lb_free, rc);
2792
2793         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2794         if (rc < 0)
2795                 GOTO(lb_free, rc);
2796
2797         if (ln.ln_namelen >= name_size)
2798                 GOTO(lb_free, rc = -EOVERFLOW);
2799
2800         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2801                 GOTO(lb_free, rc = -EFAULT);
2802
2803         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2804                 GOTO(lb_free, rc = -EFAULT);
2805
2806         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2807                 GOTO(lb_free, rc = -EFAULT);
2808
2809 lb_free:
2810         lu_buf_free(&buf);
2811 ldata_free:
2812         OBD_FREE(ldata, sizeof(*ldata));
2813
2814         RETURN(rc);
2815 }