Whamcloud - gitweb
10fef55c7dcf65921a46aa954295dc7b7d357d5c
[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.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2016, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/llite/llite_lib.c
33  *
34  * Lustre Light Super operations
35  */
36
37 #define DEBUG_SUBSYSTEM S_LLITE
38
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
44 #include <linux/mm.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
48 #endif
49 #include <linux/security.h>
50
51 #include <uapi/linux/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
61
62 struct kmem_cache *ll_file_data_slab;
63
64 #ifndef log2
65 #define log2(n) ffz(~(n))
66 #endif
67
68 static struct ll_sb_info *ll_init_sbi(void)
69 {
70         struct ll_sb_info *sbi = NULL;
71         unsigned long pages;
72         unsigned long lru_page_max;
73         struct sysinfo si;
74         class_uuid_t uuid;
75         int i;
76         ENTRY;
77
78         OBD_ALLOC_PTR(sbi);
79         if (sbi == NULL)
80                 RETURN(NULL);
81
82         spin_lock_init(&sbi->ll_lock);
83         mutex_init(&sbi->ll_lco.lco_lock);
84         spin_lock_init(&sbi->ll_pp_extent_lock);
85         spin_lock_init(&sbi->ll_process_lock);
86         sbi->ll_rw_stats_on = 0;
87
88         si_meminfo(&si);
89         pages = si.totalram - si.totalhigh;
90         lru_page_max = pages / 2;
91
92         /* initialize ll_cache data */
93         sbi->ll_cache = cl_cache_init(lru_page_max);
94         if (sbi->ll_cache == NULL) {
95                 OBD_FREE(sbi, sizeof(*sbi));
96                 RETURN(NULL);
97         }
98
99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100                                            SBI_DEFAULT_READAHEAD_MAX);
101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
103                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
104
105         ll_generate_random_uuid(uuid);
106         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
108
109         sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111         sbi->ll_flags |= LL_SBI_CHECKSUM;
112 #endif
113
114 #ifdef HAVE_LRU_RESIZE_SUPPORT
115         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 #endif
117         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118
119         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121                                pp_r_hist.oh_lock);
122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
123                                pp_w_hist.oh_lock);
124         }
125
126         /* metadata statahead is enabled by default */
127         sbi->ll_sa_max = LL_SA_RPC_DEF;
128         atomic_set(&sbi->ll_sa_total, 0);
129         atomic_set(&sbi->ll_sa_wrong, 0);
130         atomic_set(&sbi->ll_sa_running, 0);
131         atomic_set(&sbi->ll_agl_total, 0);
132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133         sbi->ll_flags |= LL_SBI_FAST_READ;
134
135         /* root squash */
136         sbi->ll_squash.rsi_uid = 0;
137         sbi->ll_squash.rsi_gid = 0;
138         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139         init_rwsem(&sbi->ll_squash.rsi_sem);
140
141         RETURN(sbi);
142 }
143
144 static void ll_free_sbi(struct super_block *sb)
145 {
146         struct ll_sb_info *sbi = ll_s2sbi(sb);
147         ENTRY;
148
149         if (sbi != NULL) {
150                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152                 if (sbi->ll_cache != NULL) {
153                         cl_cache_decref(sbi->ll_cache);
154                         sbi->ll_cache = NULL;
155                 }
156                 OBD_FREE(sbi, sizeof(*sbi));
157         }
158         EXIT;
159 }
160
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
162 {
163         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
165 }
166
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168                                     struct vfsmount *mnt)
169 {
170         struct inode *root = NULL;
171         struct ll_sb_info *sbi = ll_s2sbi(sb);
172         struct obd_device *obd;
173         struct obd_statfs *osfs = NULL;
174         struct ptlrpc_request *request = NULL;
175         struct obd_connect_data *data = NULL;
176         struct obd_uuid *uuid;
177         struct md_op_data *op_data;
178         struct lustre_md lmd;
179         u64 valid;
180         int size, err, checksum;
181         ENTRY;
182
183         obd = class_name2obd(md);
184         if (!obd) {
185                 CERROR("MD %s: not setup or attached\n", md);
186                 RETURN(-EINVAL);
187         }
188
189         OBD_ALLOC_PTR(data);
190         if (data == NULL)
191                 RETURN(-ENOMEM);
192
193         OBD_ALLOC_PTR(osfs);
194         if (osfs == NULL) {
195                 OBD_FREE_PTR(data);
196                 RETURN(-ENOMEM);
197         }
198
199         /* indicate the features supported by this client */
200         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
201                                   OBD_CONNECT_ATTRFID  |
202                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
203                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
205                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
206                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207                                   OBD_CONNECT_64BITHASH |
208                                   OBD_CONNECT_EINPROGRESS |
209                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211                                   OBD_CONNECT_MAX_EASIZE |
212                                   OBD_CONNECT_FLOCK_DEAD |
213                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214                                   OBD_CONNECT_OPEN_BY_FID |
215                                   OBD_CONNECT_DIR_STRIPE |
216                                   OBD_CONNECT_BULK_MBITS |
217                                   OBD_CONNECT_SUBTREE |
218                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
219
220         data->ocd_connect_flags2 = 0;
221
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
225 #endif
226 #ifdef CONFIG_FS_POSIX_ACL
227         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
228                                    OBD_CONNECT_LARGE_ACL;
229 #endif
230
231         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232                 /* flag mdc connection as lightweight, only used for test
233                  * purpose, use with care */
234                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
235
236         data->ocd_ibits_known = MDS_INODELOCK_FULL;
237         data->ocd_version = LUSTRE_VERSION_CODE;
238
239         if (sb->s_flags & MS_RDONLY)
240                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241         if (sbi->ll_flags & LL_SBI_USER_XATTR)
242                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
243
244 #ifdef MS_NOSEC
245         /* Setting this indicates we correctly support S_NOSEC (See kernel
246          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
247          */
248         sb->s_flags |= MS_NOSEC;
249 #endif
250
251         if (sbi->ll_flags & LL_SBI_FLOCK)
252                 sbi->ll_fop = &ll_file_operations_flock;
253         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
254                 sbi->ll_fop = &ll_file_operations;
255         else
256                 sbi->ll_fop = &ll_file_operations_noflock;
257
258         /* always ping even if server suppress_pings */
259         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
260                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
261
262 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
263         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
264 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
265
266         data->ocd_brw_size = MD_MAX_BRW_SIZE;
267
268         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
269         if (err == -EBUSY) {
270                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
271                                    "recovery, of which this client is not a "
272                                    "part. Please wait for recovery to complete,"
273                                    " abort, or time out.\n", md);
274                 GOTO(out, err);
275         } else if (err) {
276                 CERROR("cannot connect to %s: rc = %d\n", md, err);
277                 GOTO(out, err);
278         }
279
280         sbi->ll_md_exp->exp_connect_data = *data;
281
282         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
283                            LUSTRE_SEQ_METADATA);
284         if (err) {
285                 CERROR("%s: Can't init metadata layer FID infrastructure, "
286                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
287                 GOTO(out_md, err);
288         }
289
290         /* For mount, we only need fs info from MDT0, and also in DNE, it
291          * can make sure the client can be mounted as long as MDT0 is
292          * avaible */
293         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
294                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
295                         OBD_STATFS_FOR_MDT0);
296         if (err)
297                 GOTO(out_md_fid, err);
298
299         /* This needs to be after statfs to ensure connect has finished.
300          * Note that "data" does NOT contain the valid connect reply.
301          * If connecting to a 1.8 server there will be no LMV device, so
302          * we can access the MDC export directly and exp_connect_flags will
303          * be non-zero, but if accessing an upgraded 2.1 server it will
304          * have the correct flags filled in.
305          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
306         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
307         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
308             valid != CLIENT_CONNECT_MDT_REQD) {
309                 char *buf;
310
311                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
312                 obd_connect_flags2str(buf, PAGE_SIZE,
313                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
314                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
315                                    "feature(s) needed for correct operation "
316                                    "of this client (%s). Please upgrade "
317                                    "server or downgrade client.\n",
318                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
319                 OBD_FREE(buf, PAGE_SIZE);
320                 GOTO(out_md_fid, err = -EPROTO);
321         }
322
323         size = sizeof(*data);
324         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
325                            KEY_CONN_DATA,  &size, data);
326         if (err) {
327                 CERROR("%s: Get connect data failed: rc = %d\n",
328                        sbi->ll_md_exp->exp_obd->obd_name, err);
329                 GOTO(out_md_fid, err);
330         }
331
332         LASSERT(osfs->os_bsize);
333         sb->s_blocksize = osfs->os_bsize;
334         sb->s_blocksize_bits = log2(osfs->os_bsize);
335         sb->s_magic = LL_SUPER_MAGIC;
336         sb->s_maxbytes = MAX_LFS_FILESIZE;
337         sbi->ll_namelen = osfs->os_namelen;
338         sbi->ll_mnt.mnt = current->fs->root.mnt;
339
340         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
341             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
342                 LCONSOLE_INFO("Disabling user_xattr feature because "
343                               "it is not supported on the server\n");
344                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
345         }
346
347         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
348 #ifdef MS_POSIXACL
349                 sb->s_flags |= MS_POSIXACL;
350 #endif
351                 sbi->ll_flags |= LL_SBI_ACL;
352         } else {
353                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
354 #ifdef MS_POSIXACL
355                 sb->s_flags &= ~MS_POSIXACL;
356 #endif
357                 sbi->ll_flags &= ~LL_SBI_ACL;
358         }
359
360         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
361                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
362
363         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
364                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
365
366         if (obd_connect_has_secctx(data))
367                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
368
369         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
370                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
371                         LCONSOLE_INFO("%s: disabling xattr cache due to "
372                                       "unknown maximum xattr size.\n", dt);
373                 } else if (!sbi->ll_xattr_cache_set) {
374                         /* If xattr_cache is already set (no matter 0 or 1)
375                          * during processing llog, it won't be enabled here. */
376                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
377                         sbi->ll_xattr_cache_enabled = 1;
378                 }
379         }
380
381         obd = class_name2obd(dt);
382         if (!obd) {
383                 CERROR("DT %s: not setup or attached\n", dt);
384                 GOTO(out_md_fid, err = -ENODEV);
385         }
386
387         /* pass client page size via ocd_grant_blkbits, the server should report
388          * back its backend blocksize for grant calculation purpose */
389         data->ocd_grant_blkbits = PAGE_SHIFT;
390
391         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
392                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
393                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
394                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
395                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
396                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
397                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
398                                   OBD_CONNECT_EINPROGRESS |
399                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
400                                   OBD_CONNECT_LAYOUTLOCK |
401                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
402                                   OBD_CONNECT_BULK_MBITS;
403
404         data->ocd_connect_flags2 = 0;
405
406         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
407                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
408
409         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
410          * disabled by default, because it can still be enabled on the
411          * fly via /proc. As a consequence, we still need to come to an
412          * agreement on the supported algorithms at connect time */
413         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
414
415         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
416                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
417         else
418                 data->ocd_cksum_types = cksum_types_supported_client();
419
420 #ifdef HAVE_LRU_RESIZE_SUPPORT
421         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
422 #endif
423         /* always ping even if server suppress_pings */
424         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
425                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
426
427         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
428                "ocd_grant: %d\n", data->ocd_connect_flags,
429                data->ocd_version, data->ocd_grant);
430
431         obd->obd_upcall.onu_owner = &sbi->ll_lco;
432         obd->obd_upcall.onu_upcall = cl_ocd_update;
433
434         data->ocd_brw_size = DT_MAX_BRW_SIZE;
435
436         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
437                           NULL);
438         if (err == -EBUSY) {
439                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
440                                    "recovery, of which this client is not a "
441                                    "part.  Please wait for recovery to "
442                                    "complete, abort, or time out.\n", dt);
443                 GOTO(out_md, err);
444         } else if (err) {
445                 CERROR("%s: Cannot connect to %s: rc = %d\n",
446                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
447                 GOTO(out_md, err);
448         }
449
450         sbi->ll_dt_exp->exp_connect_data = *data;
451
452         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
453                            LUSTRE_SEQ_METADATA);
454         if (err) {
455                 CERROR("%s: Can't init data layer FID infrastructure, "
456                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
457                 GOTO(out_dt, err);
458         }
459
460         mutex_lock(&sbi->ll_lco.lco_lock);
461         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
462         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
463         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
464         mutex_unlock(&sbi->ll_lco.lco_lock);
465
466         fid_zero(&sbi->ll_root_fid);
467         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
468                            &sbi->ll_root_fid);
469         if (err) {
470                 CERROR("cannot mds_connect: rc = %d\n", err);
471                 GOTO(out_lock_cn_cb, err);
472         }
473         if (!fid_is_sane(&sbi->ll_root_fid)) {
474                 CERROR("%s: Invalid root fid "DFID" during mount\n",
475                        sbi->ll_md_exp->exp_obd->obd_name,
476                        PFID(&sbi->ll_root_fid));
477                 GOTO(out_lock_cn_cb, err = -EINVAL);
478         }
479         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
480
481         sb->s_op = &lustre_super_operations;
482 #if THREAD_SIZE >= 8192 /*b=17630*/
483         sb->s_export_op = &lustre_export_operations;
484 #endif
485
486         /* make root inode
487          * XXX: move this to after cbd setup? */
488         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
489         if (sbi->ll_flags & LL_SBI_ACL)
490                 valid |= OBD_MD_FLACL;
491
492         OBD_ALLOC_PTR(op_data);
493         if (op_data == NULL)
494                 GOTO(out_lock_cn_cb, err = -ENOMEM);
495
496         op_data->op_fid1 = sbi->ll_root_fid;
497         op_data->op_mode = 0;
498         op_data->op_valid = valid;
499
500         err = md_getattr(sbi->ll_md_exp, op_data, &request);
501
502         OBD_FREE_PTR(op_data);
503         if (err) {
504                 CERROR("%s: md_getattr failed for root: rc = %d\n",
505                        sbi->ll_md_exp->exp_obd->obd_name, err);
506                 GOTO(out_lock_cn_cb, err);
507         }
508
509         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
510                                sbi->ll_md_exp, &lmd);
511         if (err) {
512                 CERROR("failed to understand root inode md: rc = %d\n", err);
513                 ptlrpc_req_finished(request);
514                 GOTO(out_lock_cn_cb, err);
515         }
516
517         LASSERT(fid_is_sane(&sbi->ll_root_fid));
518         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
519                                             sbi->ll_flags & LL_SBI_32BIT_API),
520                        &lmd);
521         md_free_lustre_md(sbi->ll_md_exp, &lmd);
522         ptlrpc_req_finished(request);
523
524         if (IS_ERR(root)) {
525 #ifdef CONFIG_FS_POSIX_ACL
526                 if (lmd.posix_acl) {
527                         posix_acl_release(lmd.posix_acl);
528                         lmd.posix_acl = NULL;
529                 }
530 #endif
531                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
532                 root = NULL;
533                 CERROR("lustre_lite: bad iget4 for root\n");
534                 GOTO(out_root, err);
535         }
536
537         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
538         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
539                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
540                                  NULL);
541         if (err) {
542                 CERROR("%s: Set checksum failed: rc = %d\n",
543                        sbi->ll_dt_exp->exp_obd->obd_name, err);
544                 GOTO(out_root, err);
545         }
546         cl_sb_init(sb);
547
548         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
549                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
550                                  sbi->ll_cache, NULL);
551         if (err) {
552                 CERROR("%s: Set cache_set failed: rc = %d\n",
553                        sbi->ll_dt_exp->exp_obd->obd_name, err);
554                 GOTO(out_root, err);
555         }
556
557         sb->s_root = d_make_root(root);
558         if (sb->s_root == NULL) {
559                 CERROR("%s: can't make root dentry\n",
560                         ll_get_fsname(sb, NULL, 0));
561                 GOTO(out_root, err = -ENOMEM);
562         }
563 #ifdef HAVE_DCACHE_LOCK
564         sb->s_root->d_op = &ll_d_ops;
565 #endif
566
567         sbi->ll_sdev_orig = sb->s_dev;
568
569         /* We set sb->s_dev equal on all lustre clients in order to support
570          * NFS export clustering.  NFSD requires that the FSID be the same
571          * on all clients. */
572         /* s_dev is also used in lt_compare() to compare two fs, but that is
573          * only a node-local comparison. */
574         uuid = obd_get_uuid(sbi->ll_md_exp);
575         if (uuid != NULL)
576                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
577
578         if (data != NULL)
579                 OBD_FREE_PTR(data);
580         if (osfs != NULL)
581                 OBD_FREE_PTR(osfs);
582
583         if (sbi->ll_proc_root != NULL) {
584                 err = lprocfs_ll_register_obd(sb, dt);
585                 if (err < 0) {
586                         CERROR("%s: could not register %s in llite: rc = %d\n",
587                                dt, ll_get_fsname(sb, NULL, 0), err);
588                         err = 0;
589                 }
590                 err = lprocfs_ll_register_obd(sb, md);
591                 if (err < 0) {
592                         CERROR("%s: could not register %s in llite: rc = %d\n",
593                                md, ll_get_fsname(sb, NULL, 0), err);
594                         err = 0;
595                 }
596         }
597
598         RETURN(err);
599 out_root:
600         if (root)
601                 iput(root);
602 out_lock_cn_cb:
603         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
604 out_dt:
605         obd_disconnect(sbi->ll_dt_exp);
606         sbi->ll_dt_exp = NULL;
607 out_md_fid:
608         obd_fid_fini(sbi->ll_md_exp->exp_obd);
609 out_md:
610         obd_disconnect(sbi->ll_md_exp);
611         sbi->ll_md_exp = NULL;
612 out:
613         if (data != NULL)
614                 OBD_FREE_PTR(data);
615         if (osfs != NULL)
616                 OBD_FREE_PTR(osfs);
617         return err;
618 }
619
620 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
621 {
622         int size, rc;
623
624         size = sizeof(*lmmsize);
625         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
626                           KEY_MAX_EASIZE, &size, lmmsize);
627         if (rc != 0) {
628                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
629                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
630                 RETURN(rc);
631         }
632
633         size = sizeof(int);
634         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
635                           KEY_MAX_EASIZE, &size, lmmsize);
636         if (rc)
637                 CERROR("Get max mdsize error rc %d\n", rc);
638
639         RETURN(rc);
640 }
641
642 /**
643  * Get the value of the default_easize parameter.
644  *
645  * \see client_obd::cl_default_mds_easize
646  *
647  * \param[in] sbi       superblock info for this filesystem
648  * \param[out] lmmsize  pointer to storage location for value
649  *
650  * \retval 0            on success
651  * \retval negative     negated errno on failure
652  */
653 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
654 {
655         int size, rc;
656
657         size = sizeof(int);
658         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
659                          KEY_DEFAULT_EASIZE, &size, lmmsize);
660         if (rc)
661                 CERROR("Get default mdsize error rc %d\n", rc);
662
663         RETURN(rc);
664 }
665
666 /**
667  * Set the default_easize parameter to the given value.
668  *
669  * \see client_obd::cl_default_mds_easize
670  *
671  * \param[in] sbi       superblock info for this filesystem
672  * \param[in] lmmsize   the size to set
673  *
674  * \retval 0            on success
675  * \retval negative     negated errno on failure
676  */
677 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
678 {
679         int rc;
680
681         if (lmmsize < sizeof(struct lov_mds_md) ||
682             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
683                 return -EINVAL;
684
685         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
686                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
687                                 sizeof(int), &lmmsize, NULL);
688
689         RETURN(rc);
690 }
691
692 static void client_common_put_super(struct super_block *sb)
693 {
694         struct ll_sb_info *sbi = ll_s2sbi(sb);
695         ENTRY;
696
697         cl_sb_fini(sb);
698
699         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
700         obd_disconnect(sbi->ll_dt_exp);
701         sbi->ll_dt_exp = NULL;
702
703         lprocfs_ll_unregister_mountpoint(sbi);
704
705         obd_fid_fini(sbi->ll_md_exp->exp_obd);
706         obd_disconnect(sbi->ll_md_exp);
707         sbi->ll_md_exp = NULL;
708
709         EXIT;
710 }
711
712 void ll_kill_super(struct super_block *sb)
713 {
714         struct ll_sb_info *sbi;
715         ENTRY;
716
717         /* not init sb ?*/
718         if (!(sb->s_flags & MS_ACTIVE))
719                 return;
720
721         sbi = ll_s2sbi(sb);
722         /* we need restore s_dev from changed for clustred NFS before put_super
723          * because new kernels have cached s_dev and change sb->s_dev in
724          * put_super not affected real removing devices */
725         if (sbi) {
726                 sb->s_dev = sbi->ll_sdev_orig;
727                 sbi->ll_umounting = 1;
728
729                 /* wait running statahead threads to quit */
730                 while (atomic_read(&sbi->ll_sa_running) > 0) {
731                         set_current_state(TASK_UNINTERRUPTIBLE);
732                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
733                 }
734         }
735
736         EXIT;
737 }
738
739 static inline int ll_set_opt(const char *opt, char *data, int fl)
740 {
741         if (strncmp(opt, data, strlen(opt)) != 0)
742                 return(0);
743         else
744                 return(fl);
745 }
746
747 /* non-client-specific mount options are parsed in lmd_parse */
748 static int ll_options(char *options, int *flags)
749 {
750         int tmp;
751         char *s1 = options, *s2;
752         ENTRY;
753
754         if (!options)
755                 RETURN(0);
756
757         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
758
759         while (*s1) {
760                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
761                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
762                 if (tmp) {
763                         *flags |= tmp;
764                         goto next;
765                 }
766                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
767                 if (tmp) {
768                         *flags |= tmp;
769                         goto next;
770                 }
771                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
772                 if (tmp) {
773                         *flags |= tmp;
774                         goto next;
775                 }
776                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
777                 if (tmp) {
778                         *flags &= ~tmp;
779                         goto next;
780                 }
781                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
782                 if (tmp) {
783                         *flags |= tmp;
784                         goto next;
785                 }
786                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
787                 if (tmp) {
788                         *flags &= ~tmp;
789                         goto next;
790                 }
791                 tmp = ll_set_opt("context", s1, 1);
792                 if (tmp)
793                         goto next;
794                 tmp = ll_set_opt("fscontext", s1, 1);
795                 if (tmp)
796                         goto next;
797                 tmp = ll_set_opt("defcontext", s1, 1);
798                 if (tmp)
799                         goto next;
800                 tmp = ll_set_opt("rootcontext", s1, 1);
801                 if (tmp)
802                         goto next;
803                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
804                 if (tmp) {
805                         *flags |= tmp;
806                         goto next;
807                 }
808                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
809                 if (tmp) {
810                         *flags &= ~tmp;
811                         goto next;
812                 }
813
814                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
815                 if (tmp) {
816                         *flags |= tmp;
817                         goto next;
818                 }
819                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
820                 if (tmp) {
821                         *flags &= ~tmp;
822                         goto next;
823                 }
824                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
825                 if (tmp) {
826                         *flags |= tmp;
827                         goto next;
828                 }
829                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
830                 if (tmp) {
831                         *flags &= ~tmp;
832                         goto next;
833                 }
834                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
835                 if (tmp) {
836                         *flags |= tmp;
837                         goto next;
838                 }
839                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
840                 if (tmp) {
841                         *flags &= ~tmp;
842                         goto next;
843                 }
844                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
845                 if (tmp) {
846                         *flags |= tmp;
847                         goto next;
848                 }
849                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
850                 if (tmp) {
851                         *flags |= tmp;
852                         goto next;
853                 }
854                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
855                 if (tmp) {
856                         *flags &= ~tmp;
857                         goto next;
858                 }
859                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
860                 if (tmp) {
861                         *flags |= tmp;
862                         goto next;
863                 }
864                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
865                                    s1);
866                 RETURN(-EINVAL);
867
868 next:
869                 /* Find next opt */
870                 s2 = strchr(s1, ',');
871                 if (s2 == NULL)
872                         break;
873                 s1 = s2 + 1;
874         }
875         RETURN(0);
876 }
877
878 void ll_lli_init(struct ll_inode_info *lli)
879 {
880         lli->lli_inode_magic = LLI_INODE_MAGIC;
881         lli->lli_flags = 0;
882         spin_lock_init(&lli->lli_lock);
883         lli->lli_posix_acl = NULL;
884         /* Do not set lli_fid, it has been initialized already. */
885         fid_zero(&lli->lli_pfid);
886         lli->lli_mds_read_och = NULL;
887         lli->lli_mds_write_och = NULL;
888         lli->lli_mds_exec_och = NULL;
889         lli->lli_open_fd_read_count = 0;
890         lli->lli_open_fd_write_count = 0;
891         lli->lli_open_fd_exec_count = 0;
892         mutex_init(&lli->lli_och_mutex);
893         spin_lock_init(&lli->lli_agl_lock);
894         spin_lock_init(&lli->lli_layout_lock);
895         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
896         lli->lli_clob = NULL;
897
898         init_rwsem(&lli->lli_xattrs_list_rwsem);
899         mutex_init(&lli->lli_xattrs_enq_lock);
900
901         LASSERT(lli->lli_vfs_inode.i_mode != 0);
902         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
903                 mutex_init(&lli->lli_readdir_mutex);
904                 lli->lli_opendir_key = NULL;
905                 lli->lli_sai = NULL;
906                 spin_lock_init(&lli->lli_sa_lock);
907                 lli->lli_opendir_pid = 0;
908                 lli->lli_sa_enabled = 0;
909                 lli->lli_def_stripe_offset = -1;
910         } else {
911                 mutex_init(&lli->lli_size_mutex);
912                 lli->lli_symlink_name = NULL;
913                 init_rwsem(&lli->lli_trunc_sem);
914                 range_lock_tree_init(&lli->lli_write_tree);
915                 init_rwsem(&lli->lli_glimpse_sem);
916                 lli->lli_glimpse_time = 0;
917                 INIT_LIST_HEAD(&lli->lli_agl_list);
918                 lli->lli_agl_index = 0;
919                 lli->lli_async_rc = 0;
920         }
921         mutex_init(&lli->lli_layout_mutex);
922         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
923 }
924
925 static inline int ll_bdi_register(struct backing_dev_info *bdi)
926 {
927         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
928
929         bdi->name = "lustre";
930         return bdi_register(bdi, NULL, "lustre-%d",
931                             atomic_inc_return(&ll_bdi_num));
932 }
933
934 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
935 {
936         struct  lustre_profile *lprof = NULL;
937         struct  lustre_sb_info *lsi = s2lsi(sb);
938         struct  ll_sb_info *sbi;
939         char    *dt = NULL, *md = NULL;
940         char    *profilenm = get_profile_name(sb);
941         struct config_llog_instance *cfg;
942         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
943         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
944         int     md_len = 0;
945         int     dt_len = 0;
946         int     err;
947         ENTRY;
948
949         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
950
951         OBD_ALLOC_PTR(cfg);
952         if (cfg == NULL)
953                 RETURN(-ENOMEM);
954
955         try_module_get(THIS_MODULE);
956
957         /* client additional sb info */
958         lsi->lsi_llsbi = sbi = ll_init_sbi();
959         if (!sbi) {
960                 module_put(THIS_MODULE);
961                 OBD_FREE_PTR(cfg);
962                 RETURN(-ENOMEM);
963         }
964
965         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
966         if (err)
967                 GOTO(out_free, err);
968
969         err = bdi_init(&lsi->lsi_bdi);
970         if (err)
971                 GOTO(out_free, err);
972         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
973 #ifdef HAVE_BDI_CAP_MAP_COPY
974         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
975 #else
976         lsi->lsi_bdi.capabilities = 0;
977 #endif
978         err = ll_bdi_register(&lsi->lsi_bdi);
979         if (err)
980                 GOTO(out_free, err);
981
982         sb->s_bdi = &lsi->lsi_bdi;
983 #ifndef HAVE_DCACHE_LOCK
984         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
985         sb->s_d_op = &ll_d_ops;
986 #endif
987
988         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
989          * so that "llite.*.*" params can be processed correctly. */
990         if (proc_lustre_fs_root != NULL) {
991                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
992                 if (err < 0) {
993                         CERROR("%s: could not register mountpoint in llite: "
994                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
995                         err = 0;
996                 }
997         }
998
999         /* Generate a string unique to this super, in case some joker tries
1000            to mount the same fs at two mount points.
1001            Use the address of the super itself.*/
1002         cfg->cfg_instance = sb;
1003         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1004         cfg->cfg_callback = class_config_llog_handler;
1005         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1006         /* set up client obds */
1007         err = lustre_process_log(sb, profilenm, cfg);
1008         if (err < 0)
1009                 GOTO(out_proc, err);
1010
1011         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1012         lprof = class_get_profile(profilenm);
1013         if (lprof == NULL) {
1014                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1015                                    " read from the MGS.  Does that filesystem "
1016                                    "exist?\n", profilenm);
1017                 GOTO(out_proc, err = -EINVAL);
1018         }
1019         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1020                lprof->lp_md, lprof->lp_dt);
1021
1022         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1023         OBD_ALLOC(dt, dt_len);
1024         if (!dt)
1025                 GOTO(out_proc, err = -ENOMEM);
1026         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1027
1028         md_len = strlen(lprof->lp_md) + instlen + 2;
1029         OBD_ALLOC(md, md_len);
1030         if (!md)
1031                 GOTO(out_proc, err = -ENOMEM);
1032         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1033
1034         /* connections, registrations, sb setup */
1035         err = client_common_fill_super(sb, md, dt, mnt);
1036         if (err < 0)
1037                 GOTO(out_proc, err);
1038
1039         sbi->ll_client_common_fill_super_succeeded = 1;
1040
1041 out_proc:
1042         if (err < 0)
1043                 lprocfs_ll_unregister_mountpoint(sbi);
1044 out_free:
1045         if (md)
1046                 OBD_FREE(md, md_len);
1047         if (dt)
1048                 OBD_FREE(dt, dt_len);
1049         if (lprof != NULL)
1050                 class_put_profile(lprof);
1051         if (err)
1052                 ll_put_super(sb);
1053         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1054                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1055
1056         OBD_FREE_PTR(cfg);
1057         RETURN(err);
1058 } /* ll_fill_super */
1059
1060 void ll_put_super(struct super_block *sb)
1061 {
1062         struct config_llog_instance cfg, params_cfg;
1063         struct obd_device *obd;
1064         struct lustre_sb_info *lsi = s2lsi(sb);
1065         struct ll_sb_info *sbi = ll_s2sbi(sb);
1066         char *profilenm = get_profile_name(sb);
1067         long ccc_count;
1068         int next, force = 1, rc = 0;
1069         ENTRY;
1070
1071         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1072
1073         cfg.cfg_instance = sb;
1074         lustre_end_log(sb, profilenm, &cfg);
1075
1076         params_cfg.cfg_instance = sb;
1077         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1078
1079         if (sbi->ll_md_exp) {
1080                 obd = class_exp2obd(sbi->ll_md_exp);
1081                 if (obd)
1082                         force = obd->obd_force;
1083         }
1084
1085         /* Wait for unstable pages to be committed to stable storage */
1086         if (force == 0) {
1087                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1088                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1089                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1090                         &lwi);
1091         }
1092
1093         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1094         if (force == 0 && rc != -EINTR)
1095                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1096
1097
1098         /* We need to set force before the lov_disconnect in
1099            lustre_common_put_super, since l_d cleans up osc's as well. */
1100         if (force) {
1101                 next = 0;
1102                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1103                                                      &next)) != NULL) {
1104                         obd->obd_force = force;
1105                 }
1106         }
1107
1108         if (sbi->ll_client_common_fill_super_succeeded) {
1109                 /* Only if client_common_fill_super succeeded */
1110                 client_common_put_super(sb);
1111         }
1112
1113         next = 0;
1114         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1115                 class_manual_cleanup(obd);
1116         }
1117
1118         if (sbi->ll_flags & LL_SBI_VERBOSE)
1119                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1120
1121         if (profilenm)
1122                 class_del_profile(profilenm);
1123
1124         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1125                 bdi_destroy(&lsi->lsi_bdi);
1126                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1127         }
1128
1129         ll_free_sbi(sb);
1130         lsi->lsi_llsbi = NULL;
1131
1132         lustre_common_put_super(sb);
1133
1134         cl_env_cache_purge(~0);
1135
1136         module_put(THIS_MODULE);
1137
1138         EXIT;
1139 } /* client_put_super */
1140
1141 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1142 {
1143         struct inode *inode = NULL;
1144
1145         /* NOTE: we depend on atomic igrab() -bzzz */
1146         lock_res_and_lock(lock);
1147         if (lock->l_resource->lr_lvb_inode) {
1148                 struct ll_inode_info * lli;
1149                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1150                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1151                         inode = igrab(lock->l_resource->lr_lvb_inode);
1152                 } else {
1153                         inode = lock->l_resource->lr_lvb_inode;
1154                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1155                                          D_WARNING, lock, "lr_lvb_inode %p is "
1156                                          "bogus: magic %08x",
1157                                          lock->l_resource->lr_lvb_inode,
1158                                          lli->lli_inode_magic);
1159                         inode = NULL;
1160                 }
1161         }
1162         unlock_res_and_lock(lock);
1163         return inode;
1164 }
1165
1166 void ll_dir_clear_lsm_md(struct inode *inode)
1167 {
1168         struct ll_inode_info *lli = ll_i2info(inode);
1169
1170         LASSERT(S_ISDIR(inode->i_mode));
1171
1172         if (lli->lli_lsm_md != NULL) {
1173                 lmv_free_memmd(lli->lli_lsm_md);
1174                 lli->lli_lsm_md = NULL;
1175         }
1176 }
1177
1178 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1179                                       const struct lu_fid *fid,
1180                                       struct lustre_md *md)
1181 {
1182         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1183         struct mdt_body         *body = md->body;
1184         struct inode            *inode;
1185         ino_t                   ino;
1186         ENTRY;
1187
1188         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1189         inode = iget_locked(sb, ino);
1190         if (inode == NULL) {
1191                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1192                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1193                 RETURN(ERR_PTR(-ENOENT));
1194         }
1195
1196         if (inode->i_state & I_NEW) {
1197                 struct ll_inode_info *lli = ll_i2info(inode);
1198                 struct lmv_stripe_md *lsm = md->lmv;
1199
1200                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1201                                 (body->mbo_mode & S_IFMT);
1202                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1203                          PFID(fid));
1204
1205                 LTIME_S(inode->i_mtime) = 0;
1206                 LTIME_S(inode->i_atime) = 0;
1207                 LTIME_S(inode->i_ctime) = 0;
1208                 inode->i_rdev = 0;
1209
1210 #ifdef HAVE_BACKING_DEV_INFO
1211                 /* initializing backing dev info. */
1212                 inode->i_mapping->backing_dev_info =
1213                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1214 #endif
1215                 inode->i_op = &ll_dir_inode_operations;
1216                 inode->i_fop = &ll_dir_operations;
1217                 lli->lli_fid = *fid;
1218                 ll_lli_init(lli);
1219
1220                 LASSERT(lsm != NULL);
1221                 /* master object FID */
1222                 lli->lli_pfid = body->mbo_fid1;
1223                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1224                        lli, PFID(fid), PFID(&lli->lli_pfid));
1225                 unlock_new_inode(inode);
1226         }
1227
1228         RETURN(inode);
1229 }
1230
1231 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1232 {
1233         struct lu_fid *fid;
1234         struct lmv_stripe_md *lsm = md->lmv;
1235         int i;
1236
1237         LASSERT(lsm != NULL);
1238         /* XXX sigh, this lsm_root initialization should be in
1239          * LMV layer, but it needs ll_iget right now, so we
1240          * put this here right now. */
1241         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1242                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1243                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1244                 /* Unfortunately ll_iget will call ll_update_inode,
1245                  * where the initialization of slave inode is slightly
1246                  * different, so it reset lsm_md to NULL to avoid
1247                  * initializing lsm for slave inode. */
1248                 /* For migrating inode, master stripe and master object will
1249                  * be same, so we only need assign this inode */
1250                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1251                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1252                 else
1253                         lsm->lsm_md_oinfo[i].lmo_root =
1254                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1255
1256                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1257                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1258
1259                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1260                         return rc;
1261                 }
1262         }
1263
1264         return 0;
1265 }
1266
1267 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1268                                 const struct lmv_stripe_md *lsm_md2)
1269 {
1270         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1271                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1272                lsm_md1->lsm_md_master_mdt_index ==
1273                                         lsm_md2->lsm_md_master_mdt_index &&
1274                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1275                lsm_md1->lsm_md_layout_version ==
1276                                         lsm_md2->lsm_md_layout_version &&
1277                strcmp(lsm_md1->lsm_md_pool_name,
1278                       lsm_md2->lsm_md_pool_name) == 0;
1279 }
1280
1281 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1282 {
1283         struct ll_inode_info *lli = ll_i2info(inode);
1284         struct lmv_stripe_md *lsm = md->lmv;
1285         int     rc;
1286         ENTRY;
1287
1288         LASSERT(S_ISDIR(inode->i_mode));
1289         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1290                PFID(ll_inode2fid(inode)));
1291
1292         /* no striped information from request. */
1293         if (lsm == NULL) {
1294                 if (lli->lli_lsm_md == NULL) {
1295                         RETURN(0);
1296                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1297                                                 LMV_HASH_FLAG_MIGRATION) {
1298                         /* migration is done, the temporay MIGRATE layout has
1299                          * been removed */
1300                         CDEBUG(D_INODE, DFID" finish migration.\n",
1301                                PFID(ll_inode2fid(inode)));
1302                         lmv_free_memmd(lli->lli_lsm_md);
1303                         lli->lli_lsm_md = NULL;
1304                         RETURN(0);
1305                 } else {
1306                         /* The lustre_md from req does not include stripeEA,
1307                          * see ll_md_setattr */
1308                         RETURN(0);
1309                 }
1310         }
1311
1312         /* set the directory layout */
1313         if (lli->lli_lsm_md == NULL) {
1314                 struct cl_attr  *attr;
1315
1316                 rc = ll_init_lsm_md(inode, md);
1317                 if (rc != 0)
1318                         RETURN(rc);
1319
1320                 /* set md->lmv to NULL, so the following free lustre_md
1321                  * will not free this lsm */
1322                 md->lmv = NULL;
1323                 lli->lli_lsm_md = lsm;
1324
1325                 OBD_ALLOC_PTR(attr);
1326                 if (attr == NULL)
1327                         RETURN(-ENOMEM);
1328
1329                 /* validate the lsm */
1330                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1331                                    ll_md_blocking_ast);
1332                 if (rc != 0) {
1333                         OBD_FREE_PTR(attr);
1334                         RETURN(rc);
1335                 }
1336
1337                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1338                         md->body->mbo_nlink = attr->cat_nlink;
1339                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1340                         md->body->mbo_size = attr->cat_size;
1341                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1342                         md->body->mbo_atime = attr->cat_atime;
1343                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1344                         md->body->mbo_ctime = attr->cat_ctime;
1345                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1346                         md->body->mbo_mtime = attr->cat_mtime;
1347
1348                 OBD_FREE_PTR(attr);
1349
1350                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1351                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1352                 RETURN(0);
1353         }
1354
1355         /* Compare the old and new stripe information */
1356         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1357                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1358                 int                     idx;
1359
1360                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1361                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1362                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1363                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1364                        inode, lsm, old_lsm,
1365                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1366                        lsm->lsm_md_stripe_count,
1367                        old_lsm->lsm_md_stripe_count,
1368                        lsm->lsm_md_master_mdt_index,
1369                        old_lsm->lsm_md_master_mdt_index,
1370                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1371                        lsm->lsm_md_layout_version,
1372                        old_lsm->lsm_md_layout_version,
1373                        lsm->lsm_md_pool_name,
1374                        old_lsm->lsm_md_pool_name);
1375
1376                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1377                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1378                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1379                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1380                 }
1381
1382                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1383                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1384                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1385                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1386                 }
1387
1388                 RETURN(-EIO);
1389         }
1390
1391         RETURN(0);
1392 }
1393
1394 void ll_clear_inode(struct inode *inode)
1395 {
1396         struct ll_inode_info *lli = ll_i2info(inode);
1397         struct ll_sb_info *sbi = ll_i2sbi(inode);
1398         ENTRY;
1399
1400         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1401                PFID(ll_inode2fid(inode)), inode);
1402
1403         if (S_ISDIR(inode->i_mode)) {
1404                 /* these should have been cleared in ll_file_release */
1405                 LASSERT(lli->lli_opendir_key == NULL);
1406                 LASSERT(lli->lli_sai == NULL);
1407                 LASSERT(lli->lli_opendir_pid == 0);
1408         }
1409
1410         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1411
1412         LASSERT(!lli->lli_open_fd_write_count);
1413         LASSERT(!lli->lli_open_fd_read_count);
1414         LASSERT(!lli->lli_open_fd_exec_count);
1415
1416         if (lli->lli_mds_write_och)
1417                 ll_md_real_close(inode, FMODE_WRITE);
1418         if (lli->lli_mds_exec_och)
1419                 ll_md_real_close(inode, FMODE_EXEC);
1420         if (lli->lli_mds_read_och)
1421                 ll_md_real_close(inode, FMODE_READ);
1422
1423         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1424                 OBD_FREE(lli->lli_symlink_name,
1425                          strlen(lli->lli_symlink_name) + 1);
1426                 lli->lli_symlink_name = NULL;
1427         }
1428
1429         ll_xattr_cache_destroy(inode);
1430
1431 #ifdef CONFIG_FS_POSIX_ACL
1432         if (lli->lli_posix_acl) {
1433                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1434                 posix_acl_release(lli->lli_posix_acl);
1435                 lli->lli_posix_acl = NULL;
1436         }
1437 #endif
1438         lli->lli_inode_magic = LLI_INODE_DEAD;
1439
1440         if (S_ISDIR(inode->i_mode))
1441                 ll_dir_clear_lsm_md(inode);
1442         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1443                 LASSERT(list_empty(&lli->lli_agl_list));
1444
1445         /*
1446          * XXX This has to be done before lsm is freed below, because
1447          * cl_object still uses inode lsm.
1448          */
1449         cl_inode_fini(inode);
1450
1451         EXIT;
1452 }
1453
1454 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1455 {
1456         struct lustre_md md;
1457         struct inode *inode = dentry->d_inode;
1458         struct ll_sb_info *sbi = ll_i2sbi(inode);
1459         struct ptlrpc_request *request = NULL;
1460         int rc, ia_valid;
1461         ENTRY;
1462
1463         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1464                                      LUSTRE_OPC_ANY, NULL);
1465         if (IS_ERR(op_data))
1466                 RETURN(PTR_ERR(op_data));
1467
1468         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1469         if (rc) {
1470                 ptlrpc_req_finished(request);
1471                 if (rc == -ENOENT) {
1472                         clear_nlink(inode);
1473                         /* Unlinked special device node? Or just a race?
1474                          * Pretend we done everything. */
1475                         if (!S_ISREG(inode->i_mode) &&
1476                             !S_ISDIR(inode->i_mode)) {
1477                                 ia_valid = op_data->op_attr.ia_valid;
1478                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1479                                 rc = simple_setattr(dentry, &op_data->op_attr);
1480                                 op_data->op_attr.ia_valid = ia_valid;
1481                         }
1482                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1483                         CERROR("md_setattr fails: rc = %d\n", rc);
1484                 }
1485                 RETURN(rc);
1486         }
1487
1488         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1489                               sbi->ll_md_exp, &md);
1490         if (rc) {
1491                 ptlrpc_req_finished(request);
1492                 RETURN(rc);
1493         }
1494
1495         ia_valid = op_data->op_attr.ia_valid;
1496         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1497          * cache is not cleared yet. */
1498         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1499         if (S_ISREG(inode->i_mode))
1500                 inode_lock(inode);
1501         rc = simple_setattr(dentry, &op_data->op_attr);
1502         if (S_ISREG(inode->i_mode))
1503                 inode_unlock(inode);
1504         op_data->op_attr.ia_valid = ia_valid;
1505
1506         rc = ll_update_inode(inode, &md);
1507         ptlrpc_req_finished(request);
1508
1509         RETURN(rc);
1510 }
1511
1512 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1513  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1514  * keep these values until such a time that objects are allocated for it.
1515  * We do the MDS operations first, as it is checking permissions for us.
1516  * We don't to the MDS RPC if there is nothing that we want to store there,
1517  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1518  * going to do an RPC anyways.
1519  *
1520  * If we are doing a truncate, we will send the mtime and ctime updates
1521  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1522  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1523  * at the same time.
1524  *
1525  * In case of HSMimport, we only set attr on MDS.
1526  */
1527 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1528 {
1529         struct inode *inode = dentry->d_inode;
1530         struct ll_inode_info *lli = ll_i2info(inode);
1531         struct md_op_data *op_data = NULL;
1532         int rc = 0;
1533         ENTRY;
1534
1535         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1536                "valid %x, hsm_import %d\n",
1537                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1538                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1539                hsm_import);
1540
1541         if (attr->ia_valid & ATTR_SIZE) {
1542                 /* Check new size against VFS/VM file size limit and rlimit */
1543                 rc = inode_newsize_ok(inode, attr->ia_size);
1544                 if (rc)
1545                         RETURN(rc);
1546
1547                 /* The maximum Lustre file size is variable, based on the
1548                  * OST maximum object size and number of stripes.  This
1549                  * needs another check in addition to the VFS check above. */
1550                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1551                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1552                                PFID(&lli->lli_fid), attr->ia_size,
1553                                ll_file_maxbytes(inode));
1554                         RETURN(-EFBIG);
1555                 }
1556
1557                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1558         }
1559
1560         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1561         if (attr->ia_valid & TIMES_SET_FLAGS) {
1562                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1563                     !cfs_capable(CFS_CAP_FOWNER))
1564                         RETURN(-EPERM);
1565         }
1566
1567         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1568         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1569             (attr->ia_valid & ATTR_CTIME)) {
1570                 attr->ia_ctime = CURRENT_TIME;
1571                 attr->ia_valid |= ATTR_CTIME_SET;
1572         }
1573         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1574             (attr->ia_valid & ATTR_ATIME)) {
1575                 attr->ia_atime = CURRENT_TIME;
1576                 attr->ia_valid |= ATTR_ATIME_SET;
1577         }
1578         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1579             (attr->ia_valid & ATTR_MTIME)) {
1580                 attr->ia_mtime = CURRENT_TIME;
1581                 attr->ia_valid |= ATTR_MTIME_SET;
1582         }
1583
1584         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1585                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1586                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1587                        (s64)ktime_get_real_seconds());
1588
1589         if (S_ISREG(inode->i_mode)) {
1590                 if (attr->ia_valid & ATTR_SIZE)
1591                         inode_dio_write_done(inode);
1592                 inode_unlock(inode);
1593         }
1594
1595         /* We always do an MDS RPC, even if we're only changing the size;
1596          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1597
1598         OBD_ALLOC_PTR(op_data);
1599         if (op_data == NULL)
1600                 GOTO(out, rc = -ENOMEM);
1601
1602         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1603                 /* If we are changing file size, file content is
1604                  * modified, flag it. */
1605                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1606                 op_data->op_bias |= MDS_DATA_MODIFIED;
1607                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1608         }
1609
1610         op_data->op_attr = *attr;
1611
1612         rc = ll_md_setattr(dentry, op_data);
1613         if (rc)
1614                 GOTO(out, rc);
1615
1616         if (!S_ISREG(inode->i_mode) || hsm_import)
1617                 GOTO(out, rc = 0);
1618
1619         if (attr->ia_valid & (ATTR_SIZE |
1620                               ATTR_ATIME | ATTR_ATIME_SET |
1621                               ATTR_MTIME | ATTR_MTIME_SET |
1622                               ATTR_CTIME | ATTR_CTIME_SET)) {
1623                 /* For truncate and utimes sending attributes to OSTs, setting
1624                  * mtime/atime to the past will be performed under PW [0:EOF]
1625                  * extent lock (new_size:EOF for truncate).  It may seem
1626                  * excessive to send mtime/atime updates to OSTs when not
1627                  * setting times to past, but it is necessary due to possible
1628                  * time de-synchronization between MDT inode and OST objects */
1629                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1630         }
1631
1632         /* If the file was restored, it needs to set dirty flag.
1633          *
1634          * We've already sent MDS_DATA_MODIFIED flag in
1635          * ll_md_setattr() for truncate. However, the MDT refuses to
1636          * set the HS_DIRTY flag on released files, so we have to set
1637          * it again if the file has been restored. Please check how
1638          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1639          *
1640          * Please notice that if the file is not released, the previous
1641          * MDS_DATA_MODIFIED has taken effect and usually
1642          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1643          * This way we can save an RPC for common open + trunc
1644          * operation. */
1645         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1646                 struct hsm_state_set hss = {
1647                         .hss_valid = HSS_SETMASK,
1648                         .hss_setmask = HS_DIRTY,
1649                 };
1650                 int rc2;
1651
1652                 rc2 = ll_hsm_state_set(inode, &hss);
1653                 /* truncate and write can happen at the same time, so that
1654                  * the file can be set modified even though the file is not
1655                  * restored from released state, and ll_hsm_state_set() is
1656                  * not applicable for the file, and rc2 < 0 is normal in this
1657                  * case. */
1658                 if (rc2 < 0)
1659                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1660                                PFID(ll_inode2fid(inode)), rc2);
1661         }
1662
1663         EXIT;
1664 out:
1665         if (op_data != NULL)
1666                 ll_finish_md_op_data(op_data);
1667
1668         if (S_ISREG(inode->i_mode)) {
1669                 inode_lock(inode);
1670                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1671                         inode_dio_wait(inode);
1672                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1673                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1674                  * inode flags, so there is a gap where S_NOSEC is not set.
1675                  * This can cause a writer to take the i_mutex unnecessarily,
1676                  * but this is safe to do and should be rare. */
1677                 inode_has_no_xattr(inode);
1678         }
1679
1680         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1681                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1682
1683         return rc;
1684 }
1685
1686 int ll_setattr(struct dentry *de, struct iattr *attr)
1687 {
1688         int mode = de->d_inode->i_mode;
1689
1690         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1691                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1692                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1693
1694         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1695                                (ATTR_SIZE|ATTR_MODE)) &&
1696             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1697              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1698               !(attr->ia_mode & S_ISGID))))
1699                 attr->ia_valid |= ATTR_FORCE;
1700
1701         if ((attr->ia_valid & ATTR_MODE) &&
1702             (mode & S_ISUID) &&
1703             !(attr->ia_mode & S_ISUID) &&
1704             !(attr->ia_valid & ATTR_KILL_SUID))
1705                 attr->ia_valid |= ATTR_KILL_SUID;
1706
1707         if ((attr->ia_valid & ATTR_MODE) &&
1708             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1709             !(attr->ia_mode & S_ISGID) &&
1710             !(attr->ia_valid & ATTR_KILL_SGID))
1711                 attr->ia_valid |= ATTR_KILL_SGID;
1712
1713         /* avoid polluted from ATTR_TIMES_SET,
1714          * projid is not expected to be set here */
1715         attr->ia_valid &= ~MDS_ATTR_PROJID;
1716
1717         return ll_setattr_raw(de, attr, false);
1718 }
1719
1720 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1721                        __u64 max_age, __u32 flags)
1722 {
1723         struct ll_sb_info *sbi = ll_s2sbi(sb);
1724         struct obd_statfs obd_osfs;
1725         int rc;
1726         ENTRY;
1727
1728         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1729         if (rc) {
1730                 CERROR("md_statfs fails: rc = %d\n", rc);
1731                 RETURN(rc);
1732         }
1733
1734         osfs->os_type = sb->s_magic;
1735
1736         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1737                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1738
1739         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1740                 flags |= OBD_STATFS_NODELAY;
1741
1742         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1743         if (rc) {
1744                 CERROR("obd_statfs fails: rc = %d\n", rc);
1745                 RETURN(rc);
1746         }
1747
1748         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1749                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1750                obd_osfs.os_files);
1751
1752         osfs->os_bsize = obd_osfs.os_bsize;
1753         osfs->os_blocks = obd_osfs.os_blocks;
1754         osfs->os_bfree = obd_osfs.os_bfree;
1755         osfs->os_bavail = obd_osfs.os_bavail;
1756
1757         /* If we don't have as many objects free on the OST as inodes
1758          * on the MDS, we reduce the total number of inodes to
1759          * compensate, so that the "inodes in use" number is correct.
1760          */
1761         if (obd_osfs.os_ffree < osfs->os_ffree) {
1762                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1763                         obd_osfs.os_ffree;
1764                 osfs->os_ffree = obd_osfs.os_ffree;
1765         }
1766
1767         RETURN(rc);
1768 }
1769 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1770 {
1771         struct super_block *sb = de->d_sb;
1772         struct obd_statfs osfs;
1773         __u64 fsid = huge_encode_dev(sb->s_dev);
1774         int rc;
1775
1776         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1777         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1778
1779         /* Some amount of caching on the client is allowed */
1780         rc = ll_statfs_internal(sb, &osfs,
1781                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1782                                 0);
1783         if (rc)
1784                 return rc;
1785
1786         statfs_unpack(sfs, &osfs);
1787
1788         /* We need to downshift for all 32-bit kernels, because we can't
1789          * tell if the kernel is being called via sys_statfs64() or not.
1790          * Stop before overflowing f_bsize - in which case it is better
1791          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1792         if (sizeof(long) < 8) {
1793                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1794                         sfs->f_bsize <<= 1;
1795
1796                         osfs.os_blocks >>= 1;
1797                         osfs.os_bfree >>= 1;
1798                         osfs.os_bavail >>= 1;
1799                 }
1800         }
1801
1802         sfs->f_blocks = osfs.os_blocks;
1803         sfs->f_bfree = osfs.os_bfree;
1804         sfs->f_bavail = osfs.os_bavail;
1805         sfs->f_fsid.val[0] = (__u32)fsid;
1806         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1807         return 0;
1808 }
1809
1810 void ll_inode_size_lock(struct inode *inode)
1811 {
1812         struct ll_inode_info *lli;
1813
1814         LASSERT(!S_ISDIR(inode->i_mode));
1815
1816         lli = ll_i2info(inode);
1817         mutex_lock(&lli->lli_size_mutex);
1818 }
1819
1820 void ll_inode_size_unlock(struct inode *inode)
1821 {
1822         struct ll_inode_info *lli;
1823
1824         lli = ll_i2info(inode);
1825         mutex_unlock(&lli->lli_size_mutex);
1826 }
1827
1828 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1829 {
1830         struct ll_inode_info *lli = ll_i2info(inode);
1831         struct mdt_body *body = md->body;
1832         struct ll_sb_info *sbi = ll_i2sbi(inode);
1833         int rc = 0;
1834
1835         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1836                 rc = cl_file_inode_init(inode, md);
1837                 if (rc)
1838                         return rc;
1839         }
1840
1841         if (S_ISDIR(inode->i_mode)) {
1842                 rc = ll_update_lsm_md(inode, md);
1843                 if (rc != 0)
1844                         return rc;
1845         }
1846
1847 #ifdef CONFIG_FS_POSIX_ACL
1848         if (body->mbo_valid & OBD_MD_FLACL) {
1849                 spin_lock(&lli->lli_lock);
1850                 if (lli->lli_posix_acl)
1851                         posix_acl_release(lli->lli_posix_acl);
1852                 lli->lli_posix_acl = md->posix_acl;
1853                 spin_unlock(&lli->lli_lock);
1854         }
1855 #endif
1856         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1857                                         sbi->ll_flags & LL_SBI_32BIT_API);
1858         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1859
1860         if (body->mbo_valid & OBD_MD_FLATIME) {
1861                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1862                         LTIME_S(inode->i_atime) = body->mbo_atime;
1863                 lli->lli_atime = body->mbo_atime;
1864         }
1865
1866         if (body->mbo_valid & OBD_MD_FLMTIME) {
1867                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1868                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1869                                "to %llu\n", inode->i_ino,
1870                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1871                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1872                 }
1873                 lli->lli_mtime = body->mbo_mtime;
1874         }
1875
1876         if (body->mbo_valid & OBD_MD_FLCTIME) {
1877                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1878                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1879                 lli->lli_ctime = body->mbo_ctime;
1880         }
1881
1882         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1883         if (body->mbo_valid & OBD_MD_FLFLAGS)
1884                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1885         if (body->mbo_valid & OBD_MD_FLMODE)
1886                 inode->i_mode = (inode->i_mode & S_IFMT) |
1887                                 (body->mbo_mode & ~S_IFMT);
1888
1889         if (body->mbo_valid & OBD_MD_FLTYPE)
1890                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1891                                 (body->mbo_mode & S_IFMT);
1892
1893         LASSERT(inode->i_mode != 0);
1894         if (S_ISREG(inode->i_mode))
1895                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1896                                        LL_MAX_BLKSIZE_BITS);
1897         else
1898                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1899
1900         if (body->mbo_valid & OBD_MD_FLUID)
1901                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1902         if (body->mbo_valid & OBD_MD_FLGID)
1903                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1904         if (body->mbo_valid & OBD_MD_FLPROJID)
1905                 lli->lli_projid = body->mbo_projid;
1906         if (body->mbo_valid & OBD_MD_FLNLINK)
1907                 set_nlink(inode, body->mbo_nlink);
1908         if (body->mbo_valid & OBD_MD_FLRDEV)
1909                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1910
1911         if (body->mbo_valid & OBD_MD_FLID) {
1912                 /* FID shouldn't be changed! */
1913                 if (fid_is_sane(&lli->lli_fid)) {
1914                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1915                                  "Trying to change FID "DFID
1916                                  " to the "DFID", inode "DFID"(%p)\n",
1917                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1918                                  PFID(ll_inode2fid(inode)), inode);
1919                 } else {
1920                         lli->lli_fid = body->mbo_fid1;
1921                 }
1922         }
1923
1924         LASSERT(fid_seq(&lli->lli_fid) != 0);
1925
1926         if (body->mbo_valid & OBD_MD_FLSIZE) {
1927                 i_size_write(inode, body->mbo_size);
1928
1929                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1930                        PFID(ll_inode2fid(inode)),
1931                        (unsigned long long)body->mbo_size);
1932
1933                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1934                         inode->i_blocks = body->mbo_blocks;
1935         }
1936
1937         if (body->mbo_valid & OBD_MD_TSTATE) {
1938                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1939                  * clear it when done to ensure to start again
1940                  * glimpsing updated attrs
1941                  */
1942                 if (body->mbo_t_state & MS_RESTORE)
1943                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1944                 else
1945                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1946         }
1947
1948         return 0;
1949 }
1950
1951 int ll_read_inode2(struct inode *inode, void *opaque)
1952 {
1953         struct lustre_md *md = opaque;
1954         struct ll_inode_info *lli = ll_i2info(inode);
1955         int     rc;
1956         ENTRY;
1957
1958         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1959                PFID(&lli->lli_fid), inode);
1960
1961         /* Core attributes from the MDS first.  This is a new inode, and
1962          * the VFS doesn't zero times in the core inode so we have to do
1963          * it ourselves.  They will be overwritten by either MDS or OST
1964          * attributes - we just need to make sure they aren't newer. */
1965         LTIME_S(inode->i_mtime) = 0;
1966         LTIME_S(inode->i_atime) = 0;
1967         LTIME_S(inode->i_ctime) = 0;
1968         inode->i_rdev = 0;
1969         rc = ll_update_inode(inode, md);
1970         if (rc != 0)
1971                 RETURN(rc);
1972
1973         /* OIDEBUG(inode); */
1974
1975 #ifdef HAVE_BACKING_DEV_INFO
1976         /* initializing backing dev info. */
1977         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1978 #endif
1979         if (S_ISREG(inode->i_mode)) {
1980                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1981                 inode->i_op = &ll_file_inode_operations;
1982                 inode->i_fop = sbi->ll_fop;
1983                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1984                 EXIT;
1985         } else if (S_ISDIR(inode->i_mode)) {
1986                 inode->i_op = &ll_dir_inode_operations;
1987                 inode->i_fop = &ll_dir_operations;
1988                 EXIT;
1989         } else if (S_ISLNK(inode->i_mode)) {
1990                 inode->i_op = &ll_fast_symlink_inode_operations;
1991                 EXIT;
1992         } else {
1993                 inode->i_op = &ll_special_inode_operations;
1994
1995                 init_special_inode(inode, inode->i_mode,
1996                                    inode->i_rdev);
1997
1998                 EXIT;
1999         }
2000
2001         return 0;
2002 }
2003
2004 void ll_delete_inode(struct inode *inode)
2005 {
2006         struct ll_inode_info *lli = ll_i2info(inode);
2007         ENTRY;
2008
2009         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2010                 /* It is last chance to write out dirty pages,
2011                  * otherwise we may lose data while umount */
2012                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2013
2014         truncate_inode_pages_final(&inode->i_data);
2015
2016         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2017                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2018                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2019
2020 #ifdef HAVE_SBOPS_EVICT_INODE
2021         ll_clear_inode(inode);
2022 #endif
2023         clear_inode(inode);
2024
2025         EXIT;
2026 }
2027
2028 int ll_iocontrol(struct inode *inode, struct file *file,
2029                  unsigned int cmd, unsigned long arg)
2030 {
2031         struct ll_sb_info *sbi = ll_i2sbi(inode);
2032         struct ptlrpc_request *req = NULL;
2033         int rc, flags = 0;
2034         ENTRY;
2035
2036         switch(cmd) {
2037         case FSFILT_IOC_GETFLAGS: {
2038                 struct mdt_body *body;
2039                 struct md_op_data *op_data;
2040
2041                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2042                                              0, 0, LUSTRE_OPC_ANY,
2043                                              NULL);
2044                 if (IS_ERR(op_data))
2045                         RETURN(PTR_ERR(op_data));
2046
2047                 op_data->op_valid = OBD_MD_FLFLAGS;
2048                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2049                 ll_finish_md_op_data(op_data);
2050                 if (rc) {
2051                         CERROR("%s: failure inode "DFID": rc = %d\n",
2052                                sbi->ll_md_exp->exp_obd->obd_name,
2053                                PFID(ll_inode2fid(inode)), rc);
2054                         RETURN(-abs(rc));
2055                 }
2056
2057                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2058
2059                 flags = body->mbo_flags;
2060
2061                 ptlrpc_req_finished(req);
2062
2063                 RETURN(put_user(flags, (int __user *)arg));
2064         }
2065         case FSFILT_IOC_SETFLAGS: {
2066                 struct iattr *attr;
2067                 struct md_op_data *op_data;
2068                 struct cl_object *obj;
2069
2070                 if (get_user(flags, (int __user *)arg))
2071                         RETURN(-EFAULT);
2072
2073                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2074                                              LUSTRE_OPC_ANY, NULL);
2075                 if (IS_ERR(op_data))
2076                         RETURN(PTR_ERR(op_data));
2077
2078                 op_data->op_attr_flags = flags;
2079                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2080                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2081                 ll_finish_md_op_data(op_data);
2082                 ptlrpc_req_finished(req);
2083                 if (rc)
2084                         RETURN(rc);
2085
2086                 inode->i_flags = ll_ext_to_inode_flags(flags);
2087
2088                 obj = ll_i2info(inode)->lli_clob;
2089                 if (obj == NULL)
2090                         RETURN(0);
2091
2092                 OBD_ALLOC_PTR(attr);
2093                 if (attr == NULL)
2094                         RETURN(-ENOMEM);
2095
2096                 attr->ia_valid = ATTR_ATTR_FLAG;
2097                 rc = cl_setattr_ost(obj, attr, flags);
2098
2099                 OBD_FREE_PTR(attr);
2100                 RETURN(rc);
2101         }
2102         default:
2103                 RETURN(-ENOSYS);
2104         }
2105
2106         RETURN(0);
2107 }
2108
2109 int ll_flush_ctx(struct inode *inode)
2110 {
2111         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2112
2113         CDEBUG(D_SEC, "flush context for user %d\n",
2114                from_kuid(&init_user_ns, current_uid()));
2115
2116         obd_set_info_async(NULL, sbi->ll_md_exp,
2117                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2118                            0, NULL, NULL);
2119         obd_set_info_async(NULL, sbi->ll_dt_exp,
2120                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2121                            0, NULL, NULL);
2122         return 0;
2123 }
2124
2125 /* umount -f client means force down, don't save state */
2126 void ll_umount_begin(struct super_block *sb)
2127 {
2128         struct ll_sb_info *sbi = ll_s2sbi(sb);
2129         struct obd_device *obd;
2130         struct obd_ioctl_data *ioc_data;
2131         struct l_wait_info lwi;
2132         wait_queue_head_t waitq;
2133         ENTRY;
2134
2135         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2136                sb->s_count, atomic_read(&sb->s_active));
2137
2138         obd = class_exp2obd(sbi->ll_md_exp);
2139         if (obd == NULL) {
2140                 CERROR("Invalid MDC connection handle %#llx\n",
2141                        sbi->ll_md_exp->exp_handle.h_cookie);
2142                 EXIT;
2143                 return;
2144         }
2145         obd->obd_force = 1;
2146
2147         obd = class_exp2obd(sbi->ll_dt_exp);
2148         if (obd == NULL) {
2149                 CERROR("Invalid LOV connection handle %#llx\n",
2150                        sbi->ll_dt_exp->exp_handle.h_cookie);
2151                 EXIT;
2152                 return;
2153         }
2154         obd->obd_force = 1;
2155
2156         OBD_ALLOC_PTR(ioc_data);
2157         if (ioc_data) {
2158                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2159                               sizeof *ioc_data, ioc_data, NULL);
2160
2161                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2162                               sizeof *ioc_data, ioc_data, NULL);
2163
2164                 OBD_FREE_PTR(ioc_data);
2165         }
2166
2167         /* Really, we'd like to wait until there are no requests outstanding,
2168          * and then continue.  For now, we just periodically checking for vfs
2169          * to decrement mnt_cnt and hope to finish it within 10sec.
2170          */
2171         init_waitqueue_head(&waitq);
2172         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2173                                    cfs_time_seconds(1), NULL, NULL);
2174         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2175
2176         EXIT;
2177 }
2178
2179 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2180 {
2181         struct ll_sb_info *sbi = ll_s2sbi(sb);
2182         char *profilenm = get_profile_name(sb);
2183         int err;
2184         __u32 read_only;
2185
2186         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2187                 read_only = *flags & MS_RDONLY;
2188                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2189                                          sizeof(KEY_READ_ONLY),
2190                                          KEY_READ_ONLY, sizeof(read_only),
2191                                          &read_only, NULL);
2192                 if (err) {
2193                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2194                                       profilenm, read_only ?
2195                                       "read-only" : "read-write", err);
2196                         return err;
2197                 }
2198
2199                 if (read_only)
2200                         sb->s_flags |= MS_RDONLY;
2201                 else
2202                         sb->s_flags &= ~MS_RDONLY;
2203
2204                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2205                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2206                                       read_only ?  "read-only" : "read-write");
2207         }
2208         return 0;
2209 }
2210
2211 /**
2212  * Cleanup the open handle that is cached on MDT-side.
2213  *
2214  * For open case, the client side open handling thread may hit error
2215  * after the MDT grant the open. Under such case, the client should
2216  * send close RPC to the MDT as cleanup; otherwise, the open handle
2217  * on the MDT will be leaked there until the client umount or evicted.
2218  *
2219  * In further, if someone unlinked the file, because the open handle
2220  * holds the reference on such file/object, then it will block the
2221  * subsequent threads that want to locate such object via FID.
2222  *
2223  * \param[in] sb        super block for this file-system
2224  * \param[in] open_req  pointer to the original open request
2225  */
2226 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2227 {
2228         struct mdt_body                 *body;
2229         struct md_op_data               *op_data;
2230         struct ptlrpc_request           *close_req = NULL;
2231         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2232         ENTRY;
2233
2234         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2235         OBD_ALLOC_PTR(op_data);
2236         if (op_data == NULL) {
2237                 CWARN("%s: cannot allocate op_data to release open handle for "
2238                       DFID"\n",
2239                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2240
2241                 RETURN_EXIT;
2242         }
2243
2244         op_data->op_fid1 = body->mbo_fid1;
2245         op_data->op_handle = body->mbo_handle;
2246         op_data->op_mod_time = ktime_get_real_seconds();
2247         md_close(exp, op_data, NULL, &close_req);
2248         ptlrpc_req_finished(close_req);
2249         ll_finish_md_op_data(op_data);
2250
2251         EXIT;
2252 }
2253
2254 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2255                   struct super_block *sb, struct lookup_intent *it)
2256 {
2257         struct ll_sb_info *sbi = NULL;
2258         struct lustre_md md = { NULL };
2259         int rc;
2260         ENTRY;
2261
2262         LASSERT(*inode || sb);
2263         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2264         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2265                               sbi->ll_md_exp, &md);
2266         if (rc != 0)
2267                 GOTO(cleanup, rc);
2268
2269         if (*inode) {
2270                 rc = ll_update_inode(*inode, &md);
2271                 if (rc != 0)
2272                         GOTO(out, rc);
2273         } else {
2274                 LASSERT(sb != NULL);
2275
2276                 /*
2277                  * At this point server returns to client's same fid as client
2278                  * generated for creating. So using ->fid1 is okay here.
2279                  */
2280                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2281                         CERROR("%s: Fid is insane "DFID"\n",
2282                                 ll_get_fsname(sb, NULL, 0),
2283                                 PFID(&md.body->mbo_fid1));
2284                         GOTO(out, rc = -EINVAL);
2285                 }
2286
2287                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2288                                              sbi->ll_flags & LL_SBI_32BIT_API),
2289                                  &md);
2290                 if (IS_ERR(*inode)) {
2291 #ifdef CONFIG_FS_POSIX_ACL
2292                         if (md.posix_acl) {
2293                                 posix_acl_release(md.posix_acl);
2294                                 md.posix_acl = NULL;
2295                         }
2296 #endif
2297                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2298                         *inode = NULL;
2299                         CERROR("new_inode -fatal: rc %d\n", rc);
2300                         GOTO(out, rc);
2301                 }
2302         }
2303
2304         /* Handling piggyback layout lock.
2305          * Layout lock can be piggybacked by getattr and open request.
2306          * The lsm can be applied to inode only if it comes with a layout lock
2307          * otherwise correct layout may be overwritten, for example:
2308          * 1. proc1: mdt returns a lsm but not granting layout
2309          * 2. layout was changed by another client
2310          * 3. proc2: refresh layout and layout lock granted
2311          * 4. proc1: to apply a stale layout */
2312         if (it != NULL && it->it_lock_mode != 0) {
2313                 struct lustre_handle lockh;
2314                 struct ldlm_lock *lock;
2315
2316                 lockh.cookie = it->it_lock_handle;
2317                 lock = ldlm_handle2lock(&lockh);
2318                 LASSERT(lock != NULL);
2319                 if (ldlm_has_layout(lock)) {
2320                         struct cl_object_conf conf;
2321
2322                         memset(&conf, 0, sizeof(conf));
2323                         conf.coc_opc = OBJECT_CONF_SET;
2324                         conf.coc_inode = *inode;
2325                         conf.coc_lock = lock;
2326                         conf.u.coc_layout = md.layout;
2327                         (void)ll_layout_conf(*inode, &conf);
2328                 }
2329                 LDLM_LOCK_PUT(lock);
2330         }
2331
2332         GOTO(out, rc = 0);
2333
2334 out:
2335         md_free_lustre_md(sbi->ll_md_exp, &md);
2336
2337 cleanup:
2338         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2339                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2340
2341         return rc;
2342 }
2343
2344 int ll_obd_statfs(struct inode *inode, void __user *arg)
2345 {
2346         struct ll_sb_info *sbi = NULL;
2347         struct obd_export *exp;
2348         char *buf = NULL;
2349         struct obd_ioctl_data *data = NULL;
2350         __u32 type;
2351         int len = 0, rc;
2352
2353         if (!inode || !(sbi = ll_i2sbi(inode)))
2354                 GOTO(out_statfs, rc = -EINVAL);
2355
2356         rc = obd_ioctl_getdata(&buf, &len, arg);
2357         if (rc)
2358                 GOTO(out_statfs, rc);
2359
2360         data = (void*)buf;
2361         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2362             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2363                 GOTO(out_statfs, rc = -EINVAL);
2364
2365         if (data->ioc_inllen1 != sizeof(__u32) ||
2366             data->ioc_inllen2 != sizeof(__u32) ||
2367             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2368             data->ioc_plen2 != sizeof(struct obd_uuid))
2369                 GOTO(out_statfs, rc = -EINVAL);
2370
2371         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2372         if (type & LL_STATFS_LMV)
2373                 exp = sbi->ll_md_exp;
2374         else if (type & LL_STATFS_LOV)
2375                 exp = sbi->ll_dt_exp;
2376         else
2377                 GOTO(out_statfs, rc = -ENODEV);
2378
2379         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2380         if (rc)
2381                 GOTO(out_statfs, rc);
2382 out_statfs:
2383         OBD_FREE_LARGE(buf, len);
2384         return rc;
2385 }
2386
2387 int ll_process_config(struct lustre_cfg *lcfg)
2388 {
2389         struct super_block *sb;
2390         unsigned long x;
2391         int rc = 0;
2392         char *ptr;
2393
2394         /* The instance name contains the sb: lustre-client-aacfe000 */
2395         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2396         if (!ptr || !*(++ptr))
2397                 return -EINVAL;
2398         if (sscanf(ptr, "%lx", &x) != 1)
2399                 return -EINVAL;
2400         sb = (struct super_block *)x;
2401         /* This better be a real Lustre superblock! */
2402         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2403
2404         /* Note we have not called client_common_fill_super yet, so
2405            proc fns must be able to handle that! */
2406         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2407                                       lcfg, sb);
2408         if (rc > 0)
2409                 rc = 0;
2410         return rc;
2411 }
2412
2413 /* this function prepares md_op_data hint for passing it down to MD stack. */
2414 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2415                                       struct inode *i1, struct inode *i2,
2416                                       const char *name, size_t namelen,
2417                                       __u32 mode, __u32 opc, void *data)
2418 {
2419         LASSERT(i1 != NULL);
2420
2421         if (name == NULL) {
2422                 /* Do not reuse namelen for something else. */
2423                 if (namelen != 0)
2424                         return ERR_PTR(-EINVAL);
2425         } else {
2426                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2427                         return ERR_PTR(-ENAMETOOLONG);
2428
2429                 if (!lu_name_is_valid_2(name, namelen))
2430                         return ERR_PTR(-EINVAL);
2431         }
2432
2433         if (op_data == NULL)
2434                 OBD_ALLOC_PTR(op_data);
2435
2436         if (op_data == NULL)
2437                 return ERR_PTR(-ENOMEM);
2438
2439         ll_i2gids(op_data->op_suppgids, i1, i2);
2440         op_data->op_fid1 = *ll_inode2fid(i1);
2441         op_data->op_default_stripe_offset = -1;
2442         if (S_ISDIR(i1->i_mode)) {
2443                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2444                 if (opc == LUSTRE_OPC_MKDIR)
2445                         op_data->op_default_stripe_offset =
2446                                    ll_i2info(i1)->lli_def_stripe_offset;
2447         }
2448
2449         if (i2) {
2450                 op_data->op_fid2 = *ll_inode2fid(i2);
2451                 if (S_ISDIR(i2->i_mode))
2452                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2453         } else {
2454                 fid_zero(&op_data->op_fid2);
2455         }
2456
2457         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2458                 op_data->op_cli_flags |= CLI_HASH64;
2459
2460         if (ll_need_32bit_api(ll_i2sbi(i1)))
2461                 op_data->op_cli_flags |= CLI_API32;
2462
2463         op_data->op_name = name;
2464         op_data->op_namelen = namelen;
2465         op_data->op_mode = mode;
2466         op_data->op_mod_time = cfs_time_current_sec();
2467         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2468         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2469         op_data->op_cap = cfs_curproc_cap_pack();
2470         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2471              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2472                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2473         } else {
2474                 op_data->op_mds = 0;
2475         }
2476         op_data->op_data = data;
2477
2478         return op_data;
2479 }
2480
2481 void ll_finish_md_op_data(struct md_op_data *op_data)
2482 {
2483         security_release_secctx(op_data->op_file_secctx,
2484                                 op_data->op_file_secctx_size);
2485         OBD_FREE_PTR(op_data);
2486 }
2487
2488 #ifdef HAVE_SUPEROPS_USE_DENTRY
2489 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2490 #else
2491 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2492 #endif
2493 {
2494         struct ll_sb_info *sbi;
2495
2496 #ifdef HAVE_SUPEROPS_USE_DENTRY
2497         LASSERT((seq != NULL) && (dentry != NULL));
2498         sbi = ll_s2sbi(dentry->d_sb);
2499 #else
2500         LASSERT((seq != NULL) && (vfs != NULL));
2501         sbi = ll_s2sbi(vfs->mnt_sb);
2502 #endif
2503
2504         if (sbi->ll_flags & LL_SBI_NOLCK)
2505                 seq_puts(seq, ",nolock");
2506
2507         if (sbi->ll_flags & LL_SBI_FLOCK)
2508                 seq_puts(seq, ",flock");
2509
2510         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2511                 seq_puts(seq, ",localflock");
2512
2513         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2514                 seq_puts(seq, ",user_xattr");
2515
2516         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2517                 seq_puts(seq, ",lazystatfs");
2518
2519         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2520                 seq_puts(seq, ",user_fid2path");
2521
2522         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2523                 seq_puts(seq, ",always_ping");
2524
2525         RETURN(0);
2526 }
2527
2528 /**
2529  * Get obd name by cmd, and copy out to user space
2530  */
2531 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2532 {
2533         struct ll_sb_info *sbi = ll_i2sbi(inode);
2534         struct obd_device *obd;
2535         ENTRY;
2536
2537         if (cmd == OBD_IOC_GETDTNAME)
2538                 obd = class_exp2obd(sbi->ll_dt_exp);
2539         else if (cmd == OBD_IOC_GETMDNAME)
2540                 obd = class_exp2obd(sbi->ll_md_exp);
2541         else
2542                 RETURN(-EINVAL);
2543
2544         if (!obd)
2545                 RETURN(-ENOENT);
2546
2547         if (copy_to_user((void __user *)arg, obd->obd_name,
2548                          strlen(obd->obd_name) + 1))
2549                 RETURN(-EFAULT);
2550
2551         RETURN(0);
2552 }
2553
2554 /**
2555  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2556  * fsname will be returned in this buffer; otherwise, a static buffer will be
2557  * used to store the fsname and returned to caller.
2558  */
2559 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2560 {
2561         static char fsname_static[MTI_NAME_MAXLEN];
2562         struct lustre_sb_info *lsi = s2lsi(sb);
2563         char *ptr;
2564         int len;
2565
2566         if (buf == NULL) {
2567                 /* this means the caller wants to use static buffer
2568                  * and it doesn't care about race. Usually this is
2569                  * in error reporting path */
2570                 buf = fsname_static;
2571                 buflen = sizeof(fsname_static);
2572         }
2573
2574         len = strlen(lsi->lsi_lmd->lmd_profile);
2575         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2576         if (ptr && (strcmp(ptr, "-client") == 0))
2577                 len -= 7;
2578
2579         if (unlikely(len >= buflen))
2580                 len = buflen - 1;
2581         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2582         buf[len] = '\0';
2583
2584         return buf;
2585 }
2586
2587 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2588 {
2589         char *path = NULL;
2590
2591         struct path p;
2592
2593         p.dentry = dentry;
2594         p.mnt = current->fs->root.mnt;
2595         path_get(&p);
2596         path = d_path(&p, buf, bufsize);
2597         path_put(&p);
2598         return path;
2599 }
2600
2601 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2602 {
2603         char *buf, *path = NULL;
2604         struct dentry *dentry = NULL;
2605         struct inode *inode = page->mapping->host;
2606
2607         /* this can be called inside spin lock so use GFP_ATOMIC. */
2608         buf = (char *)__get_free_page(GFP_ATOMIC);
2609         if (buf != NULL) {
2610                 dentry = d_find_alias(page->mapping->host);
2611                 if (dentry != NULL)
2612                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2613         }
2614
2615         CDEBUG(D_WARNING,
2616                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2617                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2618                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2619                PFID(ll_inode2fid(inode)),
2620                (path && !IS_ERR(path)) ? path : "", ioret);
2621
2622         if (dentry != NULL)
2623                 dput(dentry);
2624
2625         if (buf != NULL)
2626                 free_page((unsigned long)buf);
2627 }
2628
2629 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2630                         struct lov_user_md **kbuf)
2631 {
2632         struct lov_user_md      lum;
2633         ssize_t                 lum_size;
2634         ENTRY;
2635
2636         if (copy_from_user(&lum, md, sizeof(lum)))
2637                 RETURN(-EFAULT);
2638
2639         lum_size = ll_lov_user_md_size(&lum);
2640         if (lum_size < 0)
2641                 RETURN(lum_size);
2642
2643         OBD_ALLOC(*kbuf, lum_size);
2644         if (*kbuf == NULL)
2645                 RETURN(-ENOMEM);
2646
2647         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2648                 OBD_FREE(*kbuf, lum_size);
2649                 RETURN(-EFAULT);
2650         }
2651
2652         RETURN(lum_size);
2653 }
2654
2655 /*
2656  * Compute llite root squash state after a change of root squash
2657  * configuration setting or add/remove of a lnet nid
2658  */
2659 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2660 {
2661         struct root_squash_info *squash = &sbi->ll_squash;
2662         int i;
2663         bool matched;
2664         struct lnet_process_id id;
2665
2666         /* Update norootsquash flag */
2667         down_write(&squash->rsi_sem);
2668         if (list_empty(&squash->rsi_nosquash_nids))
2669                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2670         else {
2671                 /* Do not apply root squash as soon as one of our NIDs is
2672                  * in the nosquash_nids list */
2673                 matched = false;
2674                 i = 0;
2675                 while (LNetGetId(i++, &id) != -ENOENT) {
2676                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2677                                 continue;
2678                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2679                                 matched = true;
2680                                 break;
2681                         }
2682                 }
2683                 if (matched)
2684                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2685                 else
2686                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2687         }
2688         up_write(&squash->rsi_sem);
2689 }
2690
2691 /**
2692  * Parse linkea content to extract information about a given hardlink
2693  *
2694  * \param[in]   ldata      - Initialized linkea data
2695  * \param[in]   linkno     - Link identifier
2696  * \param[out]  parent_fid - The entry's parent FID
2697  * \param[out]  ln         - Entry name destination buffer
2698  *
2699  * \retval 0 on success
2700  * \retval Appropriate negative error code on failure
2701  */
2702 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2703                             struct lu_fid *parent_fid, struct lu_name *ln)
2704 {
2705         unsigned int    idx;
2706         int             rc;
2707         ENTRY;
2708
2709         rc = linkea_init_with_rec(ldata);
2710         if (rc < 0)
2711                 RETURN(rc);
2712
2713         if (linkno >= ldata->ld_leh->leh_reccount)
2714                 /* beyond last link */
2715                 RETURN(-ENODATA);
2716
2717         linkea_first_entry(ldata);
2718         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2719                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2720                                     parent_fid);
2721                 if (idx == linkno)
2722                         break;
2723
2724                 linkea_next_entry(ldata);
2725         }
2726
2727         if (idx < linkno)
2728                 RETURN(-ENODATA);
2729
2730         RETURN(0);
2731 }
2732
2733 /**
2734  * Get parent FID and name of an identified link. Operation is performed for
2735  * a given link number, letting the caller iterate over linkno to list one or
2736  * all links of an entry.
2737  *
2738  * \param[in]     file - File descriptor against which to perform the operation
2739  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2740  *                       on and the available size. It is eventually filled with
2741  *                       the requested information or left untouched on error
2742  *
2743  * \retval - 0 on success
2744  * \retval - Appropriate negative error code on failure
2745  */
2746 int ll_getparent(struct file *file, struct getparent __user *arg)
2747 {
2748         struct dentry           *dentry = file_dentry(file);
2749         struct inode            *inode = file_inode(file);
2750         struct linkea_data      *ldata;
2751         struct lu_buf            buf = LU_BUF_NULL;
2752         struct lu_name           ln;
2753         struct lu_fid            parent_fid;
2754         __u32                    linkno;
2755         __u32                    name_size;
2756         int                      rc;
2757
2758         ENTRY;
2759
2760         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2761             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2762                 RETURN(-EPERM);
2763
2764         if (get_user(name_size, &arg->gp_name_size))
2765                 RETURN(-EFAULT);
2766
2767         if (get_user(linkno, &arg->gp_linkno))
2768                 RETURN(-EFAULT);
2769
2770         if (name_size > PATH_MAX)
2771                 RETURN(-EINVAL);
2772
2773         OBD_ALLOC(ldata, sizeof(*ldata));
2774         if (ldata == NULL)
2775                 RETURN(-ENOMEM);
2776
2777         rc = linkea_data_new(ldata, &buf);
2778         if (rc < 0)
2779                 GOTO(ldata_free, rc);
2780
2781         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2782         if (rc < 0)
2783                 GOTO(lb_free, rc);
2784
2785         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2786         if (rc < 0)
2787                 GOTO(lb_free, rc);
2788
2789         if (ln.ln_namelen >= name_size)
2790                 GOTO(lb_free, rc = -EOVERFLOW);
2791
2792         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2793                 GOTO(lb_free, rc = -EFAULT);
2794
2795         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2796                 GOTO(lb_free, rc = -EFAULT);
2797
2798         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2799                 GOTO(lb_free, rc = -EFAULT);
2800
2801 lb_free:
2802         lu_buf_free(&buf);
2803 ldata_free:
2804         OBD_FREE(ldata, sizeof(*ldata));
2805
2806         RETURN(rc);
2807 }