Whamcloud - gitweb
f1379d46a4ada5827ac8183b9b43a0093a008a34
[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 <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 <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;
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 #endif
229
230         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
231                 /* flag mdc connection as lightweight, only used for test
232                  * purpose, use with care */
233                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
234
235         data->ocd_ibits_known = MDS_INODELOCK_FULL;
236         data->ocd_version = LUSTRE_VERSION_CODE;
237
238         if (sb->s_flags & MS_RDONLY)
239                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
240         if (sbi->ll_flags & LL_SBI_USER_XATTR)
241                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
242
243         if (sbi->ll_flags & LL_SBI_FLOCK)
244                 sbi->ll_fop = &ll_file_operations_flock;
245         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
246                 sbi->ll_fop = &ll_file_operations;
247         else
248                 sbi->ll_fop = &ll_file_operations_noflock;
249
250         /* always ping even if server suppress_pings */
251         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
252                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
253
254 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
255         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
256 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
257
258         data->ocd_brw_size = MD_MAX_BRW_SIZE;
259
260         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
261         if (err == -EBUSY) {
262                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
263                                    "recovery, of which this client is not a "
264                                    "part. Please wait for recovery to complete,"
265                                    " abort, or time out.\n", md);
266                 GOTO(out, err);
267         } else if (err) {
268                 CERROR("cannot connect to %s: rc = %d\n", md, err);
269                 GOTO(out, err);
270         }
271
272         sbi->ll_md_exp->exp_connect_data = *data;
273
274         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
275                            LUSTRE_SEQ_METADATA);
276         if (err) {
277                 CERROR("%s: Can't init metadata layer FID infrastructure, "
278                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
279                 GOTO(out_md, err);
280         }
281
282         /* For mount, we only need fs info from MDT0, and also in DNE, it
283          * can make sure the client can be mounted as long as MDT0 is
284          * avaible */
285         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
286                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
287                         OBD_STATFS_FOR_MDT0);
288         if (err)
289                 GOTO(out_md_fid, err);
290
291         /* This needs to be after statfs to ensure connect has finished.
292          * Note that "data" does NOT contain the valid connect reply.
293          * If connecting to a 1.8 server there will be no LMV device, so
294          * we can access the MDC export directly and exp_connect_flags will
295          * be non-zero, but if accessing an upgraded 2.1 server it will
296          * have the correct flags filled in.
297          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
298         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
299         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
300             valid != CLIENT_CONNECT_MDT_REQD) {
301                 char *buf;
302
303                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
304                 obd_connect_flags2str(buf, PAGE_SIZE,
305                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
306                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
307                                    "feature(s) needed for correct operation "
308                                    "of this client (%s). Please upgrade "
309                                    "server or downgrade client.\n",
310                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
311                 OBD_FREE(buf, PAGE_SIZE);
312                 GOTO(out_md_fid, err = -EPROTO);
313         }
314
315         size = sizeof(*data);
316         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
317                            KEY_CONN_DATA,  &size, data);
318         if (err) {
319                 CERROR("%s: Get connect data failed: rc = %d\n",
320                        sbi->ll_md_exp->exp_obd->obd_name, err);
321                 GOTO(out_md_fid, err);
322         }
323
324         LASSERT(osfs->os_bsize);
325         sb->s_blocksize = osfs->os_bsize;
326         sb->s_blocksize_bits = log2(osfs->os_bsize);
327         sb->s_magic = LL_SUPER_MAGIC;
328         sb->s_maxbytes = MAX_LFS_FILESIZE;
329         sbi->ll_namelen = osfs->os_namelen;
330         sbi->ll_mnt.mnt = current->fs->root.mnt;
331
332         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
333             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
334                 LCONSOLE_INFO("Disabling user_xattr feature because "
335                               "it is not supported on the server\n");
336                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
337         }
338
339         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
340 #ifdef MS_POSIXACL
341                 sb->s_flags |= MS_POSIXACL;
342 #endif
343                 sbi->ll_flags |= LL_SBI_ACL;
344         } else {
345                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
346 #ifdef MS_POSIXACL
347                 sb->s_flags &= ~MS_POSIXACL;
348 #endif
349                 sbi->ll_flags &= ~LL_SBI_ACL;
350         }
351
352         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
353                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
354
355         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
356                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_SHIFT;
357         else
358                 sbi->ll_md_brw_pages = 1;
359
360         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
361                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
362
363         if (obd_connect_has_secctx(data))
364                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
365
366         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
367                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
368                         LCONSOLE_INFO("%s: disabling xattr cache due to "
369                                       "unknown maximum xattr size.\n", dt);
370                 } else {
371                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
372                         sbi->ll_xattr_cache_enabled = 1;
373                 }
374         }
375
376         obd = class_name2obd(dt);
377         if (!obd) {
378                 CERROR("DT %s: not setup or attached\n", dt);
379                 GOTO(out_md_fid, err = -ENODEV);
380         }
381
382         /* pass client page size via ocd_grant_blkbits, the server should report
383          * back its backend blocksize for grant calculation purpose */
384         data->ocd_grant_blkbits = PAGE_SHIFT;
385
386         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
387                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
388                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
389                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
390                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
391                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
392                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
393                                   OBD_CONNECT_EINPROGRESS |
394                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
395                                   OBD_CONNECT_LAYOUTLOCK |
396                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
397                                   OBD_CONNECT_BULK_MBITS;
398
399         data->ocd_connect_flags2 = 0;
400
401         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
402                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
403
404         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
405                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
406                  * disabled by default, because it can still be enabled on the
407                  * fly via /proc. As a consequence, we still need to come to an
408                  * agreement on the supported algorithms at connect time */
409                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
410
411                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
412                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
413                 else
414                         data->ocd_cksum_types = cksum_types_supported_client();
415         }
416
417 #ifdef HAVE_LRU_RESIZE_SUPPORT
418         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
419 #endif
420         /* always ping even if server suppress_pings */
421         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
422                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
423
424         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
425                "ocd_grant: %d\n", data->ocd_connect_flags,
426                data->ocd_version, data->ocd_grant);
427
428         obd->obd_upcall.onu_owner = &sbi->ll_lco;
429         obd->obd_upcall.onu_upcall = cl_ocd_update;
430
431         data->ocd_brw_size = DT_MAX_BRW_SIZE;
432
433         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
434                           NULL);
435         if (err == -EBUSY) {
436                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
437                                    "recovery, of which this client is not a "
438                                    "part.  Please wait for recovery to "
439                                    "complete, abort, or time out.\n", dt);
440                 GOTO(out_md, err);
441         } else if (err) {
442                 CERROR("%s: Cannot connect to %s: rc = %d\n",
443                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
444                 GOTO(out_md, err);
445         }
446
447         sbi->ll_dt_exp->exp_connect_data = *data;
448
449         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
450                            LUSTRE_SEQ_METADATA);
451         if (err) {
452                 CERROR("%s: Can't init data layer FID infrastructure, "
453                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
454                 GOTO(out_dt, err);
455         }
456
457         mutex_lock(&sbi->ll_lco.lco_lock);
458         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
459         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
460         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
461         mutex_unlock(&sbi->ll_lco.lco_lock);
462
463         fid_zero(&sbi->ll_root_fid);
464         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
465                            &sbi->ll_root_fid);
466         if (err) {
467                 CERROR("cannot mds_connect: rc = %d\n", err);
468                 GOTO(out_lock_cn_cb, err);
469         }
470         if (!fid_is_sane(&sbi->ll_root_fid)) {
471                 CERROR("%s: Invalid root fid "DFID" during mount\n",
472                        sbi->ll_md_exp->exp_obd->obd_name,
473                        PFID(&sbi->ll_root_fid));
474                 GOTO(out_lock_cn_cb, err = -EINVAL);
475         }
476         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
477
478         sb->s_op = &lustre_super_operations;
479 #if THREAD_SIZE >= 8192 /*b=17630*/
480         sb->s_export_op = &lustre_export_operations;
481 #endif
482
483         /* make root inode
484          * XXX: move this to after cbd setup? */
485         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
486         if (sbi->ll_flags & LL_SBI_ACL)
487                 valid |= OBD_MD_FLACL;
488
489         OBD_ALLOC_PTR(op_data);
490         if (op_data == NULL)
491                 GOTO(out_lock_cn_cb, err = -ENOMEM);
492
493         op_data->op_fid1 = sbi->ll_root_fid;
494         op_data->op_mode = 0;
495         op_data->op_valid = valid;
496
497         err = md_getattr(sbi->ll_md_exp, op_data, &request);
498
499         OBD_FREE_PTR(op_data);
500         if (err) {
501                 CERROR("%s: md_getattr failed for root: rc = %d\n",
502                        sbi->ll_md_exp->exp_obd->obd_name, err);
503                 GOTO(out_lock_cn_cb, err);
504         }
505
506         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
507                                sbi->ll_md_exp, &lmd);
508         if (err) {
509                 CERROR("failed to understand root inode md: rc = %d\n", err);
510                 ptlrpc_req_finished(request);
511                 GOTO(out_lock_cn_cb, err);
512         }
513
514         LASSERT(fid_is_sane(&sbi->ll_root_fid));
515         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
516                                             sbi->ll_flags & LL_SBI_32BIT_API),
517                        &lmd);
518         md_free_lustre_md(sbi->ll_md_exp, &lmd);
519         ptlrpc_req_finished(request);
520
521         if (IS_ERR(root)) {
522 #ifdef CONFIG_FS_POSIX_ACL
523                 if (lmd.posix_acl) {
524                         posix_acl_release(lmd.posix_acl);
525                         lmd.posix_acl = NULL;
526                 }
527 #endif
528                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
529                 root = NULL;
530                 CERROR("lustre_lite: bad iget4 for root\n");
531                 GOTO(out_root, err);
532         }
533
534         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
535         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
536                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
537                                  NULL);
538         if (err) {
539                 CERROR("%s: Set checksum failed: rc = %d\n",
540                        sbi->ll_dt_exp->exp_obd->obd_name, err);
541                 GOTO(out_root, err);
542         }
543         cl_sb_init(sb);
544
545         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
546                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
547                                  sbi->ll_cache, NULL);
548         if (err) {
549                 CERROR("%s: Set cache_set failed: rc = %d\n",
550                        sbi->ll_dt_exp->exp_obd->obd_name, err);
551                 GOTO(out_root, err);
552         }
553
554         sb->s_root = d_make_root(root);
555         if (sb->s_root == NULL) {
556                 CERROR("%s: can't make root dentry\n",
557                         ll_get_fsname(sb, NULL, 0));
558                 GOTO(out_root, err = -ENOMEM);
559         }
560 #ifdef HAVE_DCACHE_LOCK
561         sb->s_root->d_op = &ll_d_ops;
562 #endif
563
564         sbi->ll_sdev_orig = sb->s_dev;
565
566         /* We set sb->s_dev equal on all lustre clients in order to support
567          * NFS export clustering.  NFSD requires that the FSID be the same
568          * on all clients. */
569         /* s_dev is also used in lt_compare() to compare two fs, but that is
570          * only a node-local comparison. */
571         uuid = obd_get_uuid(sbi->ll_md_exp);
572         if (uuid != NULL)
573                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
574
575         if (data != NULL)
576                 OBD_FREE_PTR(data);
577         if (osfs != NULL)
578                 OBD_FREE_PTR(osfs);
579         if (proc_lustre_fs_root != NULL) {
580                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
581                                                   dt, md);
582                 if (err < 0) {
583                         CERROR("%s: could not register mount in lprocfs: "
584                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
585                         err = 0;
586                 }
587         }
588
589         RETURN(err);
590 out_root:
591         if (root)
592                 iput(root);
593 out_lock_cn_cb:
594         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
595 out_dt:
596         obd_disconnect(sbi->ll_dt_exp);
597         sbi->ll_dt_exp = NULL;
598 out_md_fid:
599         obd_fid_fini(sbi->ll_md_exp->exp_obd);
600 out_md:
601         obd_disconnect(sbi->ll_md_exp);
602         sbi->ll_md_exp = NULL;
603 out:
604         if (data != NULL)
605                 OBD_FREE_PTR(data);
606         if (osfs != NULL)
607                 OBD_FREE_PTR(osfs);
608         return err;
609 }
610
611 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
612 {
613         int size, rc;
614
615         size = sizeof(*lmmsize);
616         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
617                           KEY_MAX_EASIZE, &size, lmmsize);
618         if (rc != 0) {
619                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
620                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
621                 RETURN(rc);
622         }
623
624         size = sizeof(int);
625         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
626                           KEY_MAX_EASIZE, &size, lmmsize);
627         if (rc)
628                 CERROR("Get max mdsize error rc %d\n", rc);
629
630         RETURN(rc);
631 }
632
633 /**
634  * Get the value of the default_easize parameter.
635  *
636  * \see client_obd::cl_default_mds_easize
637  *
638  * \param[in] sbi       superblock info for this filesystem
639  * \param[out] lmmsize  pointer to storage location for value
640  *
641  * \retval 0            on success
642  * \retval negative     negated errno on failure
643  */
644 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
645 {
646         int size, rc;
647
648         size = sizeof(int);
649         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
650                          KEY_DEFAULT_EASIZE, &size, lmmsize);
651         if (rc)
652                 CERROR("Get default mdsize error rc %d\n", rc);
653
654         RETURN(rc);
655 }
656
657 /**
658  * Set the default_easize parameter to the given value.
659  *
660  * \see client_obd::cl_default_mds_easize
661  *
662  * \param[in] sbi       superblock info for this filesystem
663  * \param[in] lmmsize   the size to set
664  *
665  * \retval 0            on success
666  * \retval negative     negated errno on failure
667  */
668 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
669 {
670         int rc;
671
672         if (lmmsize < sizeof(struct lov_mds_md) ||
673             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
674                 return -EINVAL;
675
676         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
677                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
678                                 sizeof(int), &lmmsize, NULL);
679
680         RETURN(rc);
681 }
682
683 static void ll_dump_inode(struct inode *inode)
684 {
685         struct ll_d_hlist_node *tmp;
686         int dentry_count = 0;
687
688         LASSERT(inode != NULL);
689
690         ll_d_hlist_for_each(tmp, &inode->i_dentry)
691                 dentry_count++;
692
693         CERROR("%s: inode %p dump: dev=%s fid="DFID
694                " mode=%o count=%u, %d dentries\n",
695                ll_get_fsname(inode->i_sb, NULL, 0), inode,
696                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
697                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
698 }
699
700 void lustre_dump_dentry(struct dentry *dentry, int recur)
701 {
702         struct list_head *tmp;
703         int subdirs = 0;
704
705         LASSERT(dentry != NULL);
706
707         list_for_each(tmp, &dentry->d_subdirs)
708                 subdirs++;
709
710         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
711                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
712                dentry->d_name.len, dentry->d_name.name,
713                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
714                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
715                dentry->d_flags, dentry->d_fsdata, subdirs);
716         if (dentry->d_inode != NULL)
717                 ll_dump_inode(dentry->d_inode);
718
719         if (recur == 0)
720                 return;
721
722         list_for_each(tmp, &dentry->d_subdirs) {
723                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
724                 lustre_dump_dentry(d, recur - 1);
725         }
726 }
727
728 static void client_common_put_super(struct super_block *sb)
729 {
730         struct ll_sb_info *sbi = ll_s2sbi(sb);
731         ENTRY;
732
733         cl_sb_fini(sb);
734
735         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
736         obd_disconnect(sbi->ll_dt_exp);
737         sbi->ll_dt_exp = NULL;
738
739         lprocfs_unregister_mountpoint(sbi);
740
741         obd_fid_fini(sbi->ll_md_exp->exp_obd);
742         obd_disconnect(sbi->ll_md_exp);
743         sbi->ll_md_exp = NULL;
744
745         EXIT;
746 }
747
748 void ll_kill_super(struct super_block *sb)
749 {
750         struct ll_sb_info *sbi;
751         ENTRY;
752
753         /* not init sb ?*/
754         if (!(sb->s_flags & MS_ACTIVE))
755                 return;
756
757         sbi = ll_s2sbi(sb);
758         /* we need restore s_dev from changed for clustred NFS before put_super
759          * because new kernels have cached s_dev and change sb->s_dev in
760          * put_super not affected real removing devices */
761         if (sbi) {
762                 sb->s_dev = sbi->ll_sdev_orig;
763                 sbi->ll_umounting = 1;
764
765                 /* wait running statahead threads to quit */
766                 while (atomic_read(&sbi->ll_sa_running) > 0) {
767                         set_current_state(TASK_UNINTERRUPTIBLE);
768                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
769                 }
770         }
771
772         EXIT;
773 }
774
775 static inline int ll_set_opt(const char *opt, char *data, int fl)
776 {
777         if (strncmp(opt, data, strlen(opt)) != 0)
778                 return(0);
779         else
780                 return(fl);
781 }
782
783 /* non-client-specific mount options are parsed in lmd_parse */
784 static int ll_options(char *options, int *flags)
785 {
786         int tmp;
787         char *s1 = options, *s2;
788         ENTRY;
789
790         if (!options)
791                 RETURN(0);
792
793         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
794
795         while (*s1) {
796                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
797                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
798                 if (tmp) {
799                         *flags |= tmp;
800                         goto next;
801                 }
802                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
803                 if (tmp) {
804                         *flags |= tmp;
805                         goto next;
806                 }
807                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
808                 if (tmp) {
809                         *flags |= tmp;
810                         goto next;
811                 }
812                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
813                 if (tmp) {
814                         *flags &= ~tmp;
815                         goto next;
816                 }
817                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
818                 if (tmp) {
819                         *flags |= tmp;
820                         goto next;
821                 }
822                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
823                 if (tmp) {
824                         *flags &= ~tmp;
825                         goto next;
826                 }
827                 tmp = ll_set_opt("context", s1, 1);
828                 if (tmp)
829                         goto next;
830                 tmp = ll_set_opt("fscontext", s1, 1);
831                 if (tmp)
832                         goto next;
833                 tmp = ll_set_opt("defcontext", s1, 1);
834                 if (tmp)
835                         goto next;
836                 tmp = ll_set_opt("rootcontext", s1, 1);
837                 if (tmp)
838                         goto next;
839                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
840                 if (tmp) {
841                         *flags |= tmp;
842                         goto next;
843                 }
844                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
845                 if (tmp) {
846                         *flags &= ~tmp;
847                         goto next;
848                 }
849
850                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
851                 if (tmp) {
852                         *flags |= tmp;
853                         goto next;
854                 }
855                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
856                 if (tmp) {
857                         *flags &= ~tmp;
858                         goto next;
859                 }
860                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
861                 if (tmp) {
862                         *flags |= tmp;
863                         goto next;
864                 }
865                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
866                 if (tmp) {
867                         *flags &= ~tmp;
868                         goto next;
869                 }
870                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
871                 if (tmp) {
872                         *flags |= tmp;
873                         goto next;
874                 }
875                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
876                 if (tmp) {
877                         *flags &= ~tmp;
878                         goto next;
879                 }
880                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
881                 if (tmp) {
882                         *flags |= tmp;
883                         goto next;
884                 }
885                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
886                 if (tmp) {
887                         *flags |= tmp;
888                         goto next;
889                 }
890                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
891                 if (tmp) {
892                         *flags &= ~tmp;
893                         goto next;
894                 }
895                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
896                 if (tmp) {
897                         *flags |= tmp;
898                         goto next;
899                 }
900                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
901                                    s1);
902                 RETURN(-EINVAL);
903
904 next:
905                 /* Find next opt */
906                 s2 = strchr(s1, ',');
907                 if (s2 == NULL)
908                         break;
909                 s1 = s2 + 1;
910         }
911         RETURN(0);
912 }
913
914 void ll_lli_init(struct ll_inode_info *lli)
915 {
916         lli->lli_inode_magic = LLI_INODE_MAGIC;
917         lli->lli_flags = 0;
918         spin_lock_init(&lli->lli_lock);
919         lli->lli_posix_acl = NULL;
920         /* Do not set lli_fid, it has been initialized already. */
921         fid_zero(&lli->lli_pfid);
922         lli->lli_mds_read_och = NULL;
923         lli->lli_mds_write_och = NULL;
924         lli->lli_mds_exec_och = NULL;
925         lli->lli_open_fd_read_count = 0;
926         lli->lli_open_fd_write_count = 0;
927         lli->lli_open_fd_exec_count = 0;
928         mutex_init(&lli->lli_och_mutex);
929         spin_lock_init(&lli->lli_agl_lock);
930         spin_lock_init(&lli->lli_layout_lock);
931         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
932         lli->lli_clob = NULL;
933
934         init_rwsem(&lli->lli_xattrs_list_rwsem);
935         mutex_init(&lli->lli_xattrs_enq_lock);
936
937         LASSERT(lli->lli_vfs_inode.i_mode != 0);
938         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
939                 mutex_init(&lli->lli_readdir_mutex);
940                 lli->lli_opendir_key = NULL;
941                 lli->lli_sai = NULL;
942                 spin_lock_init(&lli->lli_sa_lock);
943                 lli->lli_opendir_pid = 0;
944                 lli->lli_sa_enabled = 0;
945                 lli->lli_def_stripe_offset = -1;
946         } else {
947                 mutex_init(&lli->lli_size_mutex);
948                 lli->lli_symlink_name = NULL;
949                 init_rwsem(&lli->lli_trunc_sem);
950                 range_lock_tree_init(&lli->lli_write_tree);
951                 init_rwsem(&lli->lli_glimpse_sem);
952                 lli->lli_glimpse_time = 0;
953                 INIT_LIST_HEAD(&lli->lli_agl_list);
954                 lli->lli_agl_index = 0;
955                 lli->lli_async_rc = 0;
956         }
957         mutex_init(&lli->lli_layout_mutex);
958         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
959 }
960
961 static inline int ll_bdi_register(struct backing_dev_info *bdi)
962 {
963         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
964
965         bdi->name = "lustre";
966         return bdi_register(bdi, NULL, "lustre-%d",
967                             atomic_inc_return(&ll_bdi_num));
968 }
969
970 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
971 {
972         struct lustre_profile *lprof = NULL;
973         struct lustre_sb_info *lsi = s2lsi(sb);
974         struct ll_sb_info *sbi;
975         char  *dt = NULL, *md = NULL;
976         char  *profilenm = get_profile_name(sb);
977         struct config_llog_instance *cfg;
978         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
979         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
980         int    err;
981         ENTRY;
982
983         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
984
985         OBD_ALLOC_PTR(cfg);
986         if (cfg == NULL)
987                 RETURN(-ENOMEM);
988
989         try_module_get(THIS_MODULE);
990
991         /* client additional sb info */
992         lsi->lsi_llsbi = sbi = ll_init_sbi();
993         if (!sbi) {
994                 module_put(THIS_MODULE);
995                 OBD_FREE_PTR(cfg);
996                 RETURN(-ENOMEM);
997         }
998
999         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1000         if (err)
1001                 GOTO(out_free, err);
1002
1003         err = bdi_init(&lsi->lsi_bdi);
1004         if (err)
1005                 GOTO(out_free, err);
1006         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1007 #ifdef HAVE_BDI_CAP_MAP_COPY
1008         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1009 #else
1010         lsi->lsi_bdi.capabilities = 0;
1011 #endif
1012         err = ll_bdi_register(&lsi->lsi_bdi);
1013         if (err)
1014                 GOTO(out_free, err);
1015
1016         sb->s_bdi = &lsi->lsi_bdi;
1017 #ifndef HAVE_DCACHE_LOCK
1018         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1019         sb->s_d_op = &ll_d_ops;
1020 #endif
1021
1022         /* Generate a string unique to this super, in case some joker tries
1023            to mount the same fs at two mount points.
1024            Use the address of the super itself.*/
1025         cfg->cfg_instance = sb;
1026         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1027         cfg->cfg_callback = class_config_llog_handler;
1028         /* set up client obds */
1029         err = lustre_process_log(sb, profilenm, cfg);
1030         if (err < 0)
1031                 GOTO(out_free, err);
1032
1033         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1034         lprof = class_get_profile(profilenm);
1035         if (lprof == NULL) {
1036                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1037                                    " read from the MGS.  Does that filesystem "
1038                                    "exist?\n", profilenm);
1039                 GOTO(out_free, err = -EINVAL);
1040         }
1041         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1042                lprof->lp_md, lprof->lp_dt);
1043
1044         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1045         if (!dt)
1046                 GOTO(out_free, err = -ENOMEM);
1047         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1048
1049         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1050         if (!md)
1051                 GOTO(out_free, err = -ENOMEM);
1052         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1053
1054         /* connections, registrations, sb setup */
1055         err = client_common_fill_super(sb, md, dt, mnt);
1056         if (err < 0)
1057                 GOTO(out_free, err);
1058
1059         sbi->ll_client_common_fill_super_succeeded = 1;
1060
1061 out_free:
1062         if (md)
1063                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1064         if (dt)
1065                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1066         if (lprof != NULL)
1067                 class_put_profile(lprof);
1068         if (err)
1069                 ll_put_super(sb);
1070         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1071                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1072
1073         OBD_FREE_PTR(cfg);
1074         RETURN(err);
1075 } /* ll_fill_super */
1076
1077 void ll_put_super(struct super_block *sb)
1078 {
1079         struct config_llog_instance cfg, params_cfg;
1080         struct obd_device *obd;
1081         struct lustre_sb_info *lsi = s2lsi(sb);
1082         struct ll_sb_info *sbi = ll_s2sbi(sb);
1083         char *profilenm = get_profile_name(sb);
1084         long ccc_count;
1085         int next, force = 1, rc = 0;
1086         ENTRY;
1087
1088         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1089
1090         cfg.cfg_instance = sb;
1091         lustre_end_log(sb, profilenm, &cfg);
1092
1093         params_cfg.cfg_instance = sb;
1094         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1095
1096         if (sbi->ll_md_exp) {
1097                 obd = class_exp2obd(sbi->ll_md_exp);
1098                 if (obd)
1099                         force = obd->obd_force;
1100         }
1101
1102         /* Wait for unstable pages to be committed to stable storage */
1103         if (force == 0) {
1104                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1105                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1106                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1107                         &lwi);
1108         }
1109
1110         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1111         if (force == 0 && rc != -EINTR)
1112                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1113
1114
1115         /* We need to set force before the lov_disconnect in
1116            lustre_common_put_super, since l_d cleans up osc's as well. */
1117         if (force) {
1118                 next = 0;
1119                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1120                                                      &next)) != NULL) {
1121                         obd->obd_force = force;
1122                 }
1123         }
1124
1125         if (sbi->ll_client_common_fill_super_succeeded) {
1126                 /* Only if client_common_fill_super succeeded */
1127                 client_common_put_super(sb);
1128         }
1129
1130         next = 0;
1131         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1132                 class_manual_cleanup(obd);
1133         }
1134
1135         if (sbi->ll_flags & LL_SBI_VERBOSE)
1136                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1137
1138         if (profilenm)
1139                 class_del_profile(profilenm);
1140
1141         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1142                 bdi_destroy(&lsi->lsi_bdi);
1143                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1144         }
1145
1146         ll_free_sbi(sb);
1147         lsi->lsi_llsbi = NULL;
1148
1149         lustre_common_put_super(sb);
1150
1151         cl_env_cache_purge(~0);
1152
1153         module_put(THIS_MODULE);
1154
1155         EXIT;
1156 } /* client_put_super */
1157
1158 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1159 {
1160         struct inode *inode = NULL;
1161
1162         /* NOTE: we depend on atomic igrab() -bzzz */
1163         lock_res_and_lock(lock);
1164         if (lock->l_resource->lr_lvb_inode) {
1165                 struct ll_inode_info * lli;
1166                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1167                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1168                         inode = igrab(lock->l_resource->lr_lvb_inode);
1169                 } else {
1170                         inode = lock->l_resource->lr_lvb_inode;
1171                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1172                                          D_WARNING, lock, "lr_lvb_inode %p is "
1173                                          "bogus: magic %08x",
1174                                          lock->l_resource->lr_lvb_inode,
1175                                          lli->lli_inode_magic);
1176                         inode = NULL;
1177                 }
1178         }
1179         unlock_res_and_lock(lock);
1180         return inode;
1181 }
1182
1183 void ll_dir_clear_lsm_md(struct inode *inode)
1184 {
1185         struct ll_inode_info *lli = ll_i2info(inode);
1186
1187         LASSERT(S_ISDIR(inode->i_mode));
1188
1189         if (lli->lli_lsm_md != NULL) {
1190                 lmv_free_memmd(lli->lli_lsm_md);
1191                 lli->lli_lsm_md = NULL;
1192         }
1193 }
1194
1195 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1196                                       const struct lu_fid *fid,
1197                                       struct lustre_md *md)
1198 {
1199         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1200         struct mdt_body         *body = md->body;
1201         struct inode            *inode;
1202         ino_t                   ino;
1203         ENTRY;
1204
1205         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1206         inode = iget_locked(sb, ino);
1207         if (inode == NULL) {
1208                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1209                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1210                 RETURN(ERR_PTR(-ENOENT));
1211         }
1212
1213         if (inode->i_state & I_NEW) {
1214                 struct ll_inode_info *lli = ll_i2info(inode);
1215                 struct lmv_stripe_md *lsm = md->lmv;
1216
1217                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1218                                 (body->mbo_mode & S_IFMT);
1219                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1220                          PFID(fid));
1221
1222                 LTIME_S(inode->i_mtime) = 0;
1223                 LTIME_S(inode->i_atime) = 0;
1224                 LTIME_S(inode->i_ctime) = 0;
1225                 inode->i_rdev = 0;
1226
1227 #ifdef HAVE_BACKING_DEV_INFO
1228                 /* initializing backing dev info. */
1229                 inode->i_mapping->backing_dev_info =
1230                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1231 #endif
1232                 inode->i_op = &ll_dir_inode_operations;
1233                 inode->i_fop = &ll_dir_operations;
1234                 lli->lli_fid = *fid;
1235                 ll_lli_init(lli);
1236
1237                 LASSERT(lsm != NULL);
1238                 /* master object FID */
1239                 lli->lli_pfid = body->mbo_fid1;
1240                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1241                        lli, PFID(fid), PFID(&lli->lli_pfid));
1242                 unlock_new_inode(inode);
1243         }
1244
1245         RETURN(inode);
1246 }
1247
1248 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1249 {
1250         struct lu_fid *fid;
1251         struct lmv_stripe_md *lsm = md->lmv;
1252         int i;
1253
1254         LASSERT(lsm != NULL);
1255         /* XXX sigh, this lsm_root initialization should be in
1256          * LMV layer, but it needs ll_iget right now, so we
1257          * put this here right now. */
1258         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1259                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1260                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1261                 /* Unfortunately ll_iget will call ll_update_inode,
1262                  * where the initialization of slave inode is slightly
1263                  * different, so it reset lsm_md to NULL to avoid
1264                  * initializing lsm for slave inode. */
1265                 /* For migrating inode, master stripe and master object will
1266                  * be same, so we only need assign this inode */
1267                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1268                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1269                 else
1270                         lsm->lsm_md_oinfo[i].lmo_root =
1271                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1272
1273                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1274                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1275
1276                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1277                         return rc;
1278                 }
1279         }
1280
1281         return 0;
1282 }
1283
1284 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1285                                 const struct lmv_stripe_md *lsm_md2)
1286 {
1287         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1288                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1289                lsm_md1->lsm_md_master_mdt_index ==
1290                                         lsm_md2->lsm_md_master_mdt_index &&
1291                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1292                lsm_md1->lsm_md_layout_version ==
1293                                         lsm_md2->lsm_md_layout_version &&
1294                strcmp(lsm_md1->lsm_md_pool_name,
1295                       lsm_md2->lsm_md_pool_name) == 0;
1296 }
1297
1298 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1299 {
1300         struct ll_inode_info *lli = ll_i2info(inode);
1301         struct lmv_stripe_md *lsm = md->lmv;
1302         int     rc;
1303         ENTRY;
1304
1305         LASSERT(S_ISDIR(inode->i_mode));
1306         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1307                PFID(ll_inode2fid(inode)));
1308
1309         /* no striped information from request. */
1310         if (lsm == NULL) {
1311                 if (lli->lli_lsm_md == NULL) {
1312                         RETURN(0);
1313                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1314                                                 LMV_HASH_FLAG_MIGRATION) {
1315                         /* migration is done, the temporay MIGRATE layout has
1316                          * been removed */
1317                         CDEBUG(D_INODE, DFID" finish migration.\n",
1318                                PFID(ll_inode2fid(inode)));
1319                         lmv_free_memmd(lli->lli_lsm_md);
1320                         lli->lli_lsm_md = NULL;
1321                         RETURN(0);
1322                 } else {
1323                         /* The lustre_md from req does not include stripeEA,
1324                          * see ll_md_setattr */
1325                         RETURN(0);
1326                 }
1327         }
1328
1329         /* set the directory layout */
1330         if (lli->lli_lsm_md == NULL) {
1331                 struct cl_attr  *attr;
1332
1333                 rc = ll_init_lsm_md(inode, md);
1334                 if (rc != 0)
1335                         RETURN(rc);
1336
1337                 /* set md->lmv to NULL, so the following free lustre_md
1338                  * will not free this lsm */
1339                 md->lmv = NULL;
1340                 lli->lli_lsm_md = lsm;
1341
1342                 OBD_ALLOC_PTR(attr);
1343                 if (attr == NULL)
1344                         RETURN(-ENOMEM);
1345
1346                 /* validate the lsm */
1347                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1348                                    ll_md_blocking_ast);
1349                 if (rc != 0) {
1350                         OBD_FREE_PTR(attr);
1351                         RETURN(rc);
1352                 }
1353
1354                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1355                         md->body->mbo_nlink = attr->cat_nlink;
1356                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1357                         md->body->mbo_size = attr->cat_size;
1358                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1359                         md->body->mbo_atime = attr->cat_atime;
1360                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1361                         md->body->mbo_ctime = attr->cat_ctime;
1362                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1363                         md->body->mbo_mtime = attr->cat_mtime;
1364
1365                 OBD_FREE_PTR(attr);
1366
1367                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1368                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1369                 RETURN(0);
1370         }
1371
1372         /* Compare the old and new stripe information */
1373         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1374                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1375                 int                     idx;
1376
1377                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1378                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1379                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1380                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1381                        inode, lsm, old_lsm,
1382                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1383                        lsm->lsm_md_stripe_count,
1384                        old_lsm->lsm_md_stripe_count,
1385                        lsm->lsm_md_master_mdt_index,
1386                        old_lsm->lsm_md_master_mdt_index,
1387                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1388                        lsm->lsm_md_layout_version,
1389                        old_lsm->lsm_md_layout_version,
1390                        lsm->lsm_md_pool_name,
1391                        old_lsm->lsm_md_pool_name);
1392
1393                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1394                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1395                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1396                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1397                 }
1398
1399                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1400                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1401                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1402                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1403                 }
1404
1405                 RETURN(-EIO);
1406         }
1407
1408         RETURN(0);
1409 }
1410
1411 void ll_clear_inode(struct inode *inode)
1412 {
1413         struct ll_inode_info *lli = ll_i2info(inode);
1414         struct ll_sb_info *sbi = ll_i2sbi(inode);
1415         ENTRY;
1416
1417         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1418                PFID(ll_inode2fid(inode)), inode);
1419
1420         if (S_ISDIR(inode->i_mode)) {
1421                 /* these should have been cleared in ll_file_release */
1422                 LASSERT(lli->lli_opendir_key == NULL);
1423                 LASSERT(lli->lli_sai == NULL);
1424                 LASSERT(lli->lli_opendir_pid == 0);
1425         }
1426
1427         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1428
1429         LASSERT(!lli->lli_open_fd_write_count);
1430         LASSERT(!lli->lli_open_fd_read_count);
1431         LASSERT(!lli->lli_open_fd_exec_count);
1432
1433         if (lli->lli_mds_write_och)
1434                 ll_md_real_close(inode, FMODE_WRITE);
1435         if (lli->lli_mds_exec_och)
1436                 ll_md_real_close(inode, FMODE_EXEC);
1437         if (lli->lli_mds_read_och)
1438                 ll_md_real_close(inode, FMODE_READ);
1439
1440         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1441                 OBD_FREE(lli->lli_symlink_name,
1442                          strlen(lli->lli_symlink_name) + 1);
1443                 lli->lli_symlink_name = NULL;
1444         }
1445
1446         ll_xattr_cache_destroy(inode);
1447
1448 #ifdef CONFIG_FS_POSIX_ACL
1449         if (lli->lli_posix_acl) {
1450                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1451                 posix_acl_release(lli->lli_posix_acl);
1452                 lli->lli_posix_acl = NULL;
1453         }
1454 #endif
1455         lli->lli_inode_magic = LLI_INODE_DEAD;
1456
1457         if (S_ISDIR(inode->i_mode))
1458                 ll_dir_clear_lsm_md(inode);
1459         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1460                 LASSERT(list_empty(&lli->lli_agl_list));
1461
1462         /*
1463          * XXX This has to be done before lsm is freed below, because
1464          * cl_object still uses inode lsm.
1465          */
1466         cl_inode_fini(inode);
1467
1468         EXIT;
1469 }
1470
1471 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1472 {
1473         struct lustre_md md;
1474         struct inode *inode = dentry->d_inode;
1475         struct ll_sb_info *sbi = ll_i2sbi(inode);
1476         struct ptlrpc_request *request = NULL;
1477         int rc, ia_valid;
1478         ENTRY;
1479
1480         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1481                                      LUSTRE_OPC_ANY, NULL);
1482         if (IS_ERR(op_data))
1483                 RETURN(PTR_ERR(op_data));
1484
1485         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1486         if (rc) {
1487                 ptlrpc_req_finished(request);
1488                 if (rc == -ENOENT) {
1489                         clear_nlink(inode);
1490                         /* Unlinked special device node? Or just a race?
1491                          * Pretend we done everything. */
1492                         if (!S_ISREG(inode->i_mode) &&
1493                             !S_ISDIR(inode->i_mode)) {
1494                                 ia_valid = op_data->op_attr.ia_valid;
1495                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1496                                 rc = simple_setattr(dentry, &op_data->op_attr);
1497                                 op_data->op_attr.ia_valid = ia_valid;
1498                         }
1499                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1500                         CERROR("md_setattr fails: rc = %d\n", rc);
1501                 }
1502                 RETURN(rc);
1503         }
1504
1505         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1506                               sbi->ll_md_exp, &md);
1507         if (rc) {
1508                 ptlrpc_req_finished(request);
1509                 RETURN(rc);
1510         }
1511
1512         ia_valid = op_data->op_attr.ia_valid;
1513         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1514          * cache is not cleared yet. */
1515         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1516         if (S_ISREG(inode->i_mode))
1517                 inode_lock(inode);
1518         rc = simple_setattr(dentry, &op_data->op_attr);
1519         if (S_ISREG(inode->i_mode))
1520                 inode_unlock(inode);
1521         op_data->op_attr.ia_valid = ia_valid;
1522
1523         rc = ll_update_inode(inode, &md);
1524         ptlrpc_req_finished(request);
1525
1526         RETURN(rc);
1527 }
1528
1529 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1530  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1531  * keep these values until such a time that objects are allocated for it.
1532  * We do the MDS operations first, as it is checking permissions for us.
1533  * We don't to the MDS RPC if there is nothing that we want to store there,
1534  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1535  * going to do an RPC anyways.
1536  *
1537  * If we are doing a truncate, we will send the mtime and ctime updates
1538  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1539  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1540  * at the same time.
1541  *
1542  * In case of HSMimport, we only set attr on MDS.
1543  */
1544 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1545 {
1546         struct inode *inode = dentry->d_inode;
1547         struct ll_inode_info *lli = ll_i2info(inode);
1548         struct md_op_data *op_data = NULL;
1549         int rc = 0;
1550         ENTRY;
1551
1552         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1553                "valid %x, hsm_import %d\n",
1554                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1555                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1556                hsm_import);
1557
1558         if (attr->ia_valid & ATTR_SIZE) {
1559                 /* Check new size against VFS/VM file size limit and rlimit */
1560                 rc = inode_newsize_ok(inode, attr->ia_size);
1561                 if (rc)
1562                         RETURN(rc);
1563
1564                 /* The maximum Lustre file size is variable, based on the
1565                  * OST maximum object size and number of stripes.  This
1566                  * needs another check in addition to the VFS check above. */
1567                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1568                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1569                                PFID(&lli->lli_fid), attr->ia_size,
1570                                ll_file_maxbytes(inode));
1571                         RETURN(-EFBIG);
1572                 }
1573
1574                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1575         }
1576
1577         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1578         if (attr->ia_valid & TIMES_SET_FLAGS) {
1579                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1580                     !cfs_capable(CFS_CAP_FOWNER))
1581                         RETURN(-EPERM);
1582         }
1583
1584         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1585         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1586             (attr->ia_valid & ATTR_CTIME)) {
1587                 attr->ia_ctime = CURRENT_TIME;
1588                 attr->ia_valid |= ATTR_CTIME_SET;
1589         }
1590         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1591             (attr->ia_valid & ATTR_ATIME)) {
1592                 attr->ia_atime = CURRENT_TIME;
1593                 attr->ia_valid |= ATTR_ATIME_SET;
1594         }
1595         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1596             (attr->ia_valid & ATTR_MTIME)) {
1597                 attr->ia_mtime = CURRENT_TIME;
1598                 attr->ia_valid |= ATTR_MTIME_SET;
1599         }
1600
1601         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1602                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1603                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1604                        cfs_time_current_sec());
1605
1606         if (S_ISREG(inode->i_mode)) {
1607                 if (attr->ia_valid & ATTR_SIZE)
1608                         inode_dio_write_done(inode);
1609                 inode_unlock(inode);
1610         }
1611
1612         /* We always do an MDS RPC, even if we're only changing the size;
1613          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1614
1615         OBD_ALLOC_PTR(op_data);
1616         if (op_data == NULL)
1617                 GOTO(out, rc = -ENOMEM);
1618
1619         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1620                 /* If we are changing file size, file content is
1621                  * modified, flag it. */
1622                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1623                 op_data->op_bias |= MDS_DATA_MODIFIED;
1624                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1625         }
1626
1627         op_data->op_attr = *attr;
1628
1629         rc = ll_md_setattr(dentry, op_data);
1630         if (rc)
1631                 GOTO(out, rc);
1632
1633         if (!S_ISREG(inode->i_mode) || hsm_import)
1634                 GOTO(out, rc = 0);
1635
1636         if (attr->ia_valid & (ATTR_SIZE |
1637                               ATTR_ATIME | ATTR_ATIME_SET |
1638                               ATTR_MTIME | ATTR_MTIME_SET |
1639                               ATTR_CTIME | ATTR_CTIME_SET)) {
1640                 /* For truncate and utimes sending attributes to OSTs, setting
1641                  * mtime/atime to the past will be performed under PW [0:EOF]
1642                  * extent lock (new_size:EOF for truncate).  It may seem
1643                  * excessive to send mtime/atime updates to OSTs when not
1644                  * setting times to past, but it is necessary due to possible
1645                  * time de-synchronization between MDT inode and OST objects */
1646                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1647         }
1648
1649         /* If the file was restored, it needs to set dirty flag.
1650          *
1651          * We've already sent MDS_DATA_MODIFIED flag in
1652          * ll_md_setattr() for truncate. However, the MDT refuses to
1653          * set the HS_DIRTY flag on released files, so we have to set
1654          * it again if the file has been restored. Please check how
1655          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1656          *
1657          * Please notice that if the file is not released, the previous
1658          * MDS_DATA_MODIFIED has taken effect and usually
1659          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1660          * This way we can save an RPC for common open + trunc
1661          * operation. */
1662         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1663                 struct hsm_state_set hss = {
1664                         .hss_valid = HSS_SETMASK,
1665                         .hss_setmask = HS_DIRTY,
1666                 };
1667                 int rc2;
1668
1669                 rc2 = ll_hsm_state_set(inode, &hss);
1670                 /* truncate and write can happen at the same time, so that
1671                  * the file can be set modified even though the file is not
1672                  * restored from released state, and ll_hsm_state_set() is
1673                  * not applicable for the file, and rc2 < 0 is normal in this
1674                  * case. */
1675                 if (rc2 < 0)
1676                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1677                                PFID(ll_inode2fid(inode)), rc2);
1678         }
1679
1680         EXIT;
1681 out:
1682         if (op_data != NULL)
1683                 ll_finish_md_op_data(op_data);
1684
1685         if (S_ISREG(inode->i_mode)) {
1686                 inode_lock(inode);
1687                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1688                         inode_dio_wait(inode);
1689         }
1690
1691         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1692                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1693
1694         return rc;
1695 }
1696
1697 int ll_setattr(struct dentry *de, struct iattr *attr)
1698 {
1699         int mode = de->d_inode->i_mode;
1700
1701         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1702                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1703                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1704
1705         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1706                                (ATTR_SIZE|ATTR_MODE)) &&
1707             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1708              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1709               !(attr->ia_mode & S_ISGID))))
1710                 attr->ia_valid |= ATTR_FORCE;
1711
1712         if ((attr->ia_valid & ATTR_MODE) &&
1713             (mode & S_ISUID) &&
1714             !(attr->ia_mode & S_ISUID) &&
1715             !(attr->ia_valid & ATTR_KILL_SUID))
1716                 attr->ia_valid |= ATTR_KILL_SUID;
1717
1718         if ((attr->ia_valid & ATTR_MODE) &&
1719             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1720             !(attr->ia_mode & S_ISGID) &&
1721             !(attr->ia_valid & ATTR_KILL_SGID))
1722                 attr->ia_valid |= ATTR_KILL_SGID;
1723
1724         return ll_setattr_raw(de, attr, false);
1725 }
1726
1727 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1728                        __u64 max_age, __u32 flags)
1729 {
1730         struct ll_sb_info *sbi = ll_s2sbi(sb);
1731         struct obd_statfs obd_osfs;
1732         int rc;
1733         ENTRY;
1734
1735         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1736         if (rc) {
1737                 CERROR("md_statfs fails: rc = %d\n", rc);
1738                 RETURN(rc);
1739         }
1740
1741         osfs->os_type = sb->s_magic;
1742
1743         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1744                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1745
1746         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1747                 flags |= OBD_STATFS_NODELAY;
1748
1749         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1750         if (rc) {
1751                 CERROR("obd_statfs fails: rc = %d\n", rc);
1752                 RETURN(rc);
1753         }
1754
1755         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1756                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1757                obd_osfs.os_files);
1758
1759         osfs->os_bsize = obd_osfs.os_bsize;
1760         osfs->os_blocks = obd_osfs.os_blocks;
1761         osfs->os_bfree = obd_osfs.os_bfree;
1762         osfs->os_bavail = obd_osfs.os_bavail;
1763
1764         /* If we don't have as many objects free on the OST as inodes
1765          * on the MDS, we reduce the total number of inodes to
1766          * compensate, so that the "inodes in use" number is correct.
1767          */
1768         if (obd_osfs.os_ffree < osfs->os_ffree) {
1769                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1770                         obd_osfs.os_ffree;
1771                 osfs->os_ffree = obd_osfs.os_ffree;
1772         }
1773
1774         RETURN(rc);
1775 }
1776 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1777 {
1778         struct super_block *sb = de->d_sb;
1779         struct obd_statfs osfs;
1780         __u64 fsid = huge_encode_dev(sb->s_dev);
1781         int rc;
1782
1783         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1784         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1785
1786         /* Some amount of caching on the client is allowed */
1787         rc = ll_statfs_internal(sb, &osfs,
1788                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1789                                 0);
1790         if (rc)
1791                 return rc;
1792
1793         statfs_unpack(sfs, &osfs);
1794
1795         /* We need to downshift for all 32-bit kernels, because we can't
1796          * tell if the kernel is being called via sys_statfs64() or not.
1797          * Stop before overflowing f_bsize - in which case it is better
1798          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1799         if (sizeof(long) < 8) {
1800                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1801                         sfs->f_bsize <<= 1;
1802
1803                         osfs.os_blocks >>= 1;
1804                         osfs.os_bfree >>= 1;
1805                         osfs.os_bavail >>= 1;
1806                 }
1807         }
1808
1809         sfs->f_blocks = osfs.os_blocks;
1810         sfs->f_bfree = osfs.os_bfree;
1811         sfs->f_bavail = osfs.os_bavail;
1812         sfs->f_fsid.val[0] = (__u32)fsid;
1813         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1814         return 0;
1815 }
1816
1817 void ll_inode_size_lock(struct inode *inode)
1818 {
1819         struct ll_inode_info *lli;
1820
1821         LASSERT(!S_ISDIR(inode->i_mode));
1822
1823         lli = ll_i2info(inode);
1824         mutex_lock(&lli->lli_size_mutex);
1825 }
1826
1827 void ll_inode_size_unlock(struct inode *inode)
1828 {
1829         struct ll_inode_info *lli;
1830
1831         lli = ll_i2info(inode);
1832         mutex_unlock(&lli->lli_size_mutex);
1833 }
1834
1835 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1836 {
1837         struct ll_inode_info *lli = ll_i2info(inode);
1838         struct mdt_body *body = md->body;
1839         struct ll_sb_info *sbi = ll_i2sbi(inode);
1840
1841         if (body->mbo_valid & OBD_MD_FLEASIZE)
1842                 cl_file_inode_init(inode, md);
1843
1844         if (S_ISDIR(inode->i_mode)) {
1845                 int     rc;
1846
1847                 rc = ll_update_lsm_md(inode, md);
1848                 if (rc != 0)
1849                         return rc;
1850         }
1851
1852 #ifdef CONFIG_FS_POSIX_ACL
1853         if (body->mbo_valid & OBD_MD_FLACL) {
1854                 spin_lock(&lli->lli_lock);
1855                 if (lli->lli_posix_acl)
1856                         posix_acl_release(lli->lli_posix_acl);
1857                 lli->lli_posix_acl = md->posix_acl;
1858                 spin_unlock(&lli->lli_lock);
1859         }
1860 #endif
1861         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1862                                         sbi->ll_flags & LL_SBI_32BIT_API);
1863         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1864
1865         if (body->mbo_valid & OBD_MD_FLATIME) {
1866                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1867                         LTIME_S(inode->i_atime) = body->mbo_atime;
1868                 lli->lli_atime = body->mbo_atime;
1869         }
1870
1871         if (body->mbo_valid & OBD_MD_FLMTIME) {
1872                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1873                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1874                                "to %llu\n", inode->i_ino,
1875                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1876                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1877                 }
1878                 lli->lli_mtime = body->mbo_mtime;
1879         }
1880
1881         if (body->mbo_valid & OBD_MD_FLCTIME) {
1882                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1883                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1884                 lli->lli_ctime = body->mbo_ctime;
1885         }
1886
1887         if (body->mbo_valid & OBD_MD_FLMODE)
1888                 inode->i_mode = (inode->i_mode & S_IFMT) |
1889                                 (body->mbo_mode & ~S_IFMT);
1890
1891         if (body->mbo_valid & OBD_MD_FLTYPE)
1892                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1893                                 (body->mbo_mode & S_IFMT);
1894
1895         LASSERT(inode->i_mode != 0);
1896         if (S_ISREG(inode->i_mode))
1897                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1898                                        LL_MAX_BLKSIZE_BITS);
1899         else
1900                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1901
1902         if (body->mbo_valid & OBD_MD_FLUID)
1903                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1904         if (body->mbo_valid & OBD_MD_FLGID)
1905                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1906         if (body->mbo_valid & OBD_MD_FLFLAGS)
1907                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1908         if (body->mbo_valid & OBD_MD_FLNLINK)
1909                 set_nlink(inode, body->mbo_nlink);
1910         if (body->mbo_valid & OBD_MD_FLRDEV)
1911                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1912
1913         if (body->mbo_valid & OBD_MD_FLID) {
1914                 /* FID shouldn't be changed! */
1915                 if (fid_is_sane(&lli->lli_fid)) {
1916                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1917                                  "Trying to change FID "DFID
1918                                  " to the "DFID", inode "DFID"(%p)\n",
1919                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1920                                  PFID(ll_inode2fid(inode)), inode);
1921                 } else {
1922                         lli->lli_fid = body->mbo_fid1;
1923                 }
1924         }
1925
1926         LASSERT(fid_seq(&lli->lli_fid) != 0);
1927
1928         if (body->mbo_valid & OBD_MD_FLSIZE) {
1929                 i_size_write(inode, body->mbo_size);
1930
1931                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1932                        PFID(ll_inode2fid(inode)),
1933                        (unsigned long long)body->mbo_size);
1934
1935                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1936                         inode->i_blocks = body->mbo_blocks;
1937         }
1938
1939         if (body->mbo_valid & OBD_MD_TSTATE) {
1940                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1941                  * clear it when done to ensure to start again
1942                  * glimpsing updated attrs
1943                  */
1944                 if (body->mbo_t_state & MS_RESTORE)
1945                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1946                 else
1947                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1948         }
1949
1950         return 0;
1951 }
1952
1953 int ll_read_inode2(struct inode *inode, void *opaque)
1954 {
1955         struct lustre_md *md = opaque;
1956         struct ll_inode_info *lli = ll_i2info(inode);
1957         int     rc;
1958         ENTRY;
1959
1960         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1961                PFID(&lli->lli_fid), inode);
1962
1963         /* Core attributes from the MDS first.  This is a new inode, and
1964          * the VFS doesn't zero times in the core inode so we have to do
1965          * it ourselves.  They will be overwritten by either MDS or OST
1966          * attributes - we just need to make sure they aren't newer. */
1967         LTIME_S(inode->i_mtime) = 0;
1968         LTIME_S(inode->i_atime) = 0;
1969         LTIME_S(inode->i_ctime) = 0;
1970         inode->i_rdev = 0;
1971         rc = ll_update_inode(inode, md);
1972         if (rc != 0)
1973                 RETURN(rc);
1974
1975         /* OIDEBUG(inode); */
1976
1977 #ifdef HAVE_BACKING_DEV_INFO
1978         /* initializing backing dev info. */
1979         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1980 #endif
1981         if (S_ISREG(inode->i_mode)) {
1982                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1983                 inode->i_op = &ll_file_inode_operations;
1984                 inode->i_fop = sbi->ll_fop;
1985                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1986                 EXIT;
1987         } else if (S_ISDIR(inode->i_mode)) {
1988                 inode->i_op = &ll_dir_inode_operations;
1989                 inode->i_fop = &ll_dir_operations;
1990                 EXIT;
1991         } else if (S_ISLNK(inode->i_mode)) {
1992                 inode->i_op = &ll_fast_symlink_inode_operations;
1993                 EXIT;
1994         } else {
1995                 inode->i_op = &ll_special_inode_operations;
1996
1997                 init_special_inode(inode, inode->i_mode,
1998                                    inode->i_rdev);
1999
2000                 EXIT;
2001         }
2002
2003         return 0;
2004 }
2005
2006 void ll_delete_inode(struct inode *inode)
2007 {
2008         struct ll_inode_info *lli = ll_i2info(inode);
2009         ENTRY;
2010
2011         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2012                 /* It is last chance to write out dirty pages,
2013                  * otherwise we may lose data while umount */
2014                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2015
2016         truncate_inode_pages_final(&inode->i_data);
2017
2018         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2019                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2020                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2021
2022 #ifdef HAVE_SBOPS_EVICT_INODE
2023         ll_clear_inode(inode);
2024 #endif
2025         clear_inode(inode);
2026
2027         EXIT;
2028 }
2029
2030 int ll_iocontrol(struct inode *inode, struct file *file,
2031                  unsigned int cmd, unsigned long arg)
2032 {
2033         struct ll_sb_info *sbi = ll_i2sbi(inode);
2034         struct ptlrpc_request *req = NULL;
2035         int rc, flags = 0;
2036         ENTRY;
2037
2038         switch(cmd) {
2039         case FSFILT_IOC_GETFLAGS: {
2040                 struct mdt_body *body;
2041                 struct md_op_data *op_data;
2042
2043                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2044                                              0, 0, LUSTRE_OPC_ANY,
2045                                              NULL);
2046                 if (IS_ERR(op_data))
2047                         RETURN(PTR_ERR(op_data));
2048
2049                 op_data->op_valid = OBD_MD_FLFLAGS;
2050                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2051                 ll_finish_md_op_data(op_data);
2052                 if (rc) {
2053                         CERROR("%s: failure inode "DFID": rc = %d\n",
2054                                sbi->ll_md_exp->exp_obd->obd_name,
2055                                PFID(ll_inode2fid(inode)), rc);
2056                         RETURN(-abs(rc));
2057                 }
2058
2059                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2060
2061                 flags = body->mbo_flags;
2062
2063                 ptlrpc_req_finished(req);
2064
2065                 RETURN(put_user(flags, (int __user *)arg));
2066         }
2067         case FSFILT_IOC_SETFLAGS: {
2068                 struct iattr *attr;
2069                 struct md_op_data *op_data;
2070                 struct cl_object *obj;
2071
2072                 if (get_user(flags, (int __user *)arg))
2073                         RETURN(-EFAULT);
2074
2075                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2076                                              LUSTRE_OPC_ANY, NULL);
2077                 if (IS_ERR(op_data))
2078                         RETURN(PTR_ERR(op_data));
2079
2080                 op_data->op_attr_flags = flags;
2081                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2082                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2083                 ll_finish_md_op_data(op_data);
2084                 ptlrpc_req_finished(req);
2085                 if (rc)
2086                         RETURN(rc);
2087
2088                 inode->i_flags = ll_ext_to_inode_flags(flags);
2089
2090                 obj = ll_i2info(inode)->lli_clob;
2091                 if (obj == NULL)
2092                         RETURN(0);
2093
2094                 OBD_ALLOC_PTR(attr);
2095                 if (attr == NULL)
2096                         RETURN(-ENOMEM);
2097
2098                 attr->ia_valid = ATTR_ATTR_FLAG;
2099                 rc = cl_setattr_ost(obj, attr, flags);
2100
2101                 OBD_FREE_PTR(attr);
2102                 RETURN(rc);
2103         }
2104         default:
2105                 RETURN(-ENOSYS);
2106         }
2107
2108         RETURN(0);
2109 }
2110
2111 int ll_flush_ctx(struct inode *inode)
2112 {
2113         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2114
2115         CDEBUG(D_SEC, "flush context for user %d\n",
2116                from_kuid(&init_user_ns, current_uid()));
2117
2118         obd_set_info_async(NULL, sbi->ll_md_exp,
2119                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2120                            0, NULL, NULL);
2121         obd_set_info_async(NULL, sbi->ll_dt_exp,
2122                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2123                            0, NULL, NULL);
2124         return 0;
2125 }
2126
2127 /* umount -f client means force down, don't save state */
2128 void ll_umount_begin(struct super_block *sb)
2129 {
2130         struct ll_sb_info *sbi = ll_s2sbi(sb);
2131         struct obd_device *obd;
2132         struct obd_ioctl_data *ioc_data;
2133         struct l_wait_info lwi;
2134         wait_queue_head_t waitq;
2135         ENTRY;
2136
2137         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2138                sb->s_count, atomic_read(&sb->s_active));
2139
2140         obd = class_exp2obd(sbi->ll_md_exp);
2141         if (obd == NULL) {
2142                 CERROR("Invalid MDC connection handle %#llx\n",
2143                        sbi->ll_md_exp->exp_handle.h_cookie);
2144                 EXIT;
2145                 return;
2146         }
2147         obd->obd_force = 1;
2148
2149         obd = class_exp2obd(sbi->ll_dt_exp);
2150         if (obd == NULL) {
2151                 CERROR("Invalid LOV connection handle %#llx\n",
2152                        sbi->ll_dt_exp->exp_handle.h_cookie);
2153                 EXIT;
2154                 return;
2155         }
2156         obd->obd_force = 1;
2157
2158         OBD_ALLOC_PTR(ioc_data);
2159         if (ioc_data) {
2160                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2161                               sizeof *ioc_data, ioc_data, NULL);
2162
2163                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2164                               sizeof *ioc_data, ioc_data, NULL);
2165
2166                 OBD_FREE_PTR(ioc_data);
2167         }
2168
2169         /* Really, we'd like to wait until there are no requests outstanding,
2170          * and then continue.  For now, we just periodically checking for vfs
2171          * to decrement mnt_cnt and hope to finish it within 10sec.
2172          */
2173         init_waitqueue_head(&waitq);
2174         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2175                                    cfs_time_seconds(1), NULL, NULL);
2176         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2177
2178         EXIT;
2179 }
2180
2181 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2182 {
2183         struct ll_sb_info *sbi = ll_s2sbi(sb);
2184         char *profilenm = get_profile_name(sb);
2185         int err;
2186         __u32 read_only;
2187
2188         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2189                 read_only = *flags & MS_RDONLY;
2190                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2191                                          sizeof(KEY_READ_ONLY),
2192                                          KEY_READ_ONLY, sizeof(read_only),
2193                                          &read_only, NULL);
2194                 if (err) {
2195                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2196                                       profilenm, read_only ?
2197                                       "read-only" : "read-write", err);
2198                         return err;
2199                 }
2200
2201                 if (read_only)
2202                         sb->s_flags |= MS_RDONLY;
2203                 else
2204                         sb->s_flags &= ~MS_RDONLY;
2205
2206                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2207                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2208                                       read_only ?  "read-only" : "read-write");
2209         }
2210         return 0;
2211 }
2212
2213 /**
2214  * Cleanup the open handle that is cached on MDT-side.
2215  *
2216  * For open case, the client side open handling thread may hit error
2217  * after the MDT grant the open. Under such case, the client should
2218  * send close RPC to the MDT as cleanup; otherwise, the open handle
2219  * on the MDT will be leaked there until the client umount or evicted.
2220  *
2221  * In further, if someone unlinked the file, because the open handle
2222  * holds the reference on such file/object, then it will block the
2223  * subsequent threads that want to locate such object via FID.
2224  *
2225  * \param[in] sb        super block for this file-system
2226  * \param[in] open_req  pointer to the original open request
2227  */
2228 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2229 {
2230         struct mdt_body                 *body;
2231         struct md_op_data               *op_data;
2232         struct ptlrpc_request           *close_req = NULL;
2233         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2234         ENTRY;
2235
2236         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2237         OBD_ALLOC_PTR(op_data);
2238         if (op_data == NULL) {
2239                 CWARN("%s: cannot allocate op_data to release open handle for "
2240                       DFID"\n",
2241                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2242
2243                 RETURN_EXIT;
2244         }
2245
2246         op_data->op_fid1 = body->mbo_fid1;
2247         op_data->op_handle = body->mbo_handle;
2248         op_data->op_mod_time = cfs_time_current_sec();
2249         md_close(exp, op_data, NULL, &close_req);
2250         ptlrpc_req_finished(close_req);
2251         ll_finish_md_op_data(op_data);
2252
2253         EXIT;
2254 }
2255
2256 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2257                   struct super_block *sb, struct lookup_intent *it)
2258 {
2259         struct ll_sb_info *sbi = NULL;
2260         struct lustre_md md = { NULL };
2261         int rc;
2262         ENTRY;
2263
2264         LASSERT(*inode || sb);
2265         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2266         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2267                               sbi->ll_md_exp, &md);
2268         if (rc != 0)
2269                 GOTO(cleanup, rc);
2270
2271         if (*inode) {
2272                 rc = ll_update_inode(*inode, &md);
2273                 if (rc != 0)
2274                         GOTO(out, rc);
2275         } else {
2276                 LASSERT(sb != NULL);
2277
2278                 /*
2279                  * At this point server returns to client's same fid as client
2280                  * generated for creating. So using ->fid1 is okay here.
2281                  */
2282                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2283                         CERROR("%s: Fid is insane "DFID"\n",
2284                                 ll_get_fsname(sb, NULL, 0),
2285                                 PFID(&md.body->mbo_fid1));
2286                         GOTO(out, rc = -EINVAL);
2287                 }
2288
2289                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2290                                              sbi->ll_flags & LL_SBI_32BIT_API),
2291                                  &md);
2292                 if (IS_ERR(*inode)) {
2293 #ifdef CONFIG_FS_POSIX_ACL
2294                         if (md.posix_acl) {
2295                                 posix_acl_release(md.posix_acl);
2296                                 md.posix_acl = NULL;
2297                         }
2298 #endif
2299                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2300                         *inode = NULL;
2301                         CERROR("new_inode -fatal: rc %d\n", rc);
2302                         GOTO(out, rc);
2303                 }
2304         }
2305
2306         /* Handling piggyback layout lock.
2307          * Layout lock can be piggybacked by getattr and open request.
2308          * The lsm can be applied to inode only if it comes with a layout lock
2309          * otherwise correct layout may be overwritten, for example:
2310          * 1. proc1: mdt returns a lsm but not granting layout
2311          * 2. layout was changed by another client
2312          * 3. proc2: refresh layout and layout lock granted
2313          * 4. proc1: to apply a stale layout */
2314         if (it != NULL && it->it_lock_mode != 0) {
2315                 struct lustre_handle lockh;
2316                 struct ldlm_lock *lock;
2317
2318                 lockh.cookie = it->it_lock_handle;
2319                 lock = ldlm_handle2lock(&lockh);
2320                 LASSERT(lock != NULL);
2321                 if (ldlm_has_layout(lock)) {
2322                         struct cl_object_conf conf;
2323
2324                         memset(&conf, 0, sizeof(conf));
2325                         conf.coc_opc = OBJECT_CONF_SET;
2326                         conf.coc_inode = *inode;
2327                         conf.coc_lock = lock;
2328                         conf.u.coc_layout = md.layout;
2329                         (void)ll_layout_conf(*inode, &conf);
2330                 }
2331                 LDLM_LOCK_PUT(lock);
2332         }
2333
2334         GOTO(out, rc = 0);
2335
2336 out:
2337         md_free_lustre_md(sbi->ll_md_exp, &md);
2338
2339 cleanup:
2340         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2341                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2342
2343         return rc;
2344 }
2345
2346 int ll_obd_statfs(struct inode *inode, void __user *arg)
2347 {
2348         struct ll_sb_info *sbi = NULL;
2349         struct obd_export *exp;
2350         char *buf = NULL;
2351         struct obd_ioctl_data *data = NULL;
2352         __u32 type;
2353         int len = 0, rc;
2354
2355         if (!inode || !(sbi = ll_i2sbi(inode)))
2356                 GOTO(out_statfs, rc = -EINVAL);
2357
2358         rc = obd_ioctl_getdata(&buf, &len, arg);
2359         if (rc)
2360                 GOTO(out_statfs, rc);
2361
2362         data = (void*)buf;
2363         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2364             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2365                 GOTO(out_statfs, rc = -EINVAL);
2366
2367         if (data->ioc_inllen1 != sizeof(__u32) ||
2368             data->ioc_inllen2 != sizeof(__u32) ||
2369             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2370             data->ioc_plen2 != sizeof(struct obd_uuid))
2371                 GOTO(out_statfs, rc = -EINVAL);
2372
2373         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2374         if (type & LL_STATFS_LMV)
2375                 exp = sbi->ll_md_exp;
2376         else if (type & LL_STATFS_LOV)
2377                 exp = sbi->ll_dt_exp;
2378         else
2379                 GOTO(out_statfs, rc = -ENODEV);
2380
2381         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2382         if (rc)
2383                 GOTO(out_statfs, rc);
2384 out_statfs:
2385         if (buf)
2386                 obd_ioctl_freedata(buf, len);
2387         return rc;
2388 }
2389
2390 int ll_process_config(struct lustre_cfg *lcfg)
2391 {
2392         struct super_block *sb;
2393         unsigned long x;
2394         int rc = 0;
2395         char *ptr;
2396
2397         /* The instance name contains the sb: lustre-client-aacfe000 */
2398         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2399         if (!ptr || !*(++ptr))
2400                 return -EINVAL;
2401         if (sscanf(ptr, "%lx", &x) != 1)
2402                 return -EINVAL;
2403         sb = (struct super_block *)x;
2404         /* This better be a real Lustre superblock! */
2405         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2406
2407         /* Note we have not called client_common_fill_super yet, so
2408            proc fns must be able to handle that! */
2409         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2410                                       lcfg, sb);
2411         if (rc > 0)
2412                 rc = 0;
2413         return rc;
2414 }
2415
2416 /* this function prepares md_op_data hint for passing it down to MD stack. */
2417 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2418                                       struct inode *i1, struct inode *i2,
2419                                       const char *name, size_t namelen,
2420                                       __u32 mode, __u32 opc, void *data)
2421 {
2422         LASSERT(i1 != NULL);
2423
2424         if (name == NULL) {
2425                 /* Do not reuse namelen for something else. */
2426                 if (namelen != 0)
2427                         return ERR_PTR(-EINVAL);
2428         } else {
2429                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2430                         return ERR_PTR(-ENAMETOOLONG);
2431
2432                 if (!lu_name_is_valid_2(name, namelen))
2433                         return ERR_PTR(-EINVAL);
2434         }
2435
2436         if (op_data == NULL)
2437                 OBD_ALLOC_PTR(op_data);
2438
2439         if (op_data == NULL)
2440                 return ERR_PTR(-ENOMEM);
2441
2442         ll_i2gids(op_data->op_suppgids, i1, i2);
2443         op_data->op_fid1 = *ll_inode2fid(i1);
2444         op_data->op_default_stripe_offset = -1;
2445         if (S_ISDIR(i1->i_mode)) {
2446                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2447                 if (opc == LUSTRE_OPC_MKDIR)
2448                         op_data->op_default_stripe_offset =
2449                                    ll_i2info(i1)->lli_def_stripe_offset;
2450         }
2451
2452         if (i2) {
2453                 op_data->op_fid2 = *ll_inode2fid(i2);
2454                 if (S_ISDIR(i2->i_mode))
2455                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2456         } else {
2457                 fid_zero(&op_data->op_fid2);
2458         }
2459
2460         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2461                 op_data->op_cli_flags |= CLI_HASH64;
2462
2463         if (ll_need_32bit_api(ll_i2sbi(i1)))
2464                 op_data->op_cli_flags |= CLI_API32;
2465
2466         op_data->op_name = name;
2467         op_data->op_namelen = namelen;
2468         op_data->op_mode = mode;
2469         op_data->op_mod_time = cfs_time_current_sec();
2470         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2471         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2472         op_data->op_cap = cfs_curproc_cap_pack();
2473         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2474              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2475                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2476         } else {
2477                 op_data->op_mds = 0;
2478         }
2479         op_data->op_data = data;
2480
2481         return op_data;
2482 }
2483
2484 void ll_finish_md_op_data(struct md_op_data *op_data)
2485 {
2486         security_release_secctx(op_data->op_file_secctx,
2487                                 op_data->op_file_secctx_size);
2488         OBD_FREE_PTR(op_data);
2489 }
2490
2491 #ifdef HAVE_SUPEROPS_USE_DENTRY
2492 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2493 #else
2494 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2495 #endif
2496 {
2497         struct ll_sb_info *sbi;
2498
2499 #ifdef HAVE_SUPEROPS_USE_DENTRY
2500         LASSERT((seq != NULL) && (dentry != NULL));
2501         sbi = ll_s2sbi(dentry->d_sb);
2502 #else
2503         LASSERT((seq != NULL) && (vfs != NULL));
2504         sbi = ll_s2sbi(vfs->mnt_sb);
2505 #endif
2506
2507         if (sbi->ll_flags & LL_SBI_NOLCK)
2508                 seq_puts(seq, ",nolock");
2509
2510         if (sbi->ll_flags & LL_SBI_FLOCK)
2511                 seq_puts(seq, ",flock");
2512
2513         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2514                 seq_puts(seq, ",localflock");
2515
2516         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2517                 seq_puts(seq, ",user_xattr");
2518
2519         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2520                 seq_puts(seq, ",lazystatfs");
2521
2522         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2523                 seq_puts(seq, ",user_fid2path");
2524
2525         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2526                 seq_puts(seq, ",always_ping");
2527
2528         RETURN(0);
2529 }
2530
2531 /**
2532  * Get obd name by cmd, and copy out to user space
2533  */
2534 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2535 {
2536         struct ll_sb_info *sbi = ll_i2sbi(inode);
2537         struct obd_device *obd;
2538         ENTRY;
2539
2540         if (cmd == OBD_IOC_GETDTNAME)
2541                 obd = class_exp2obd(sbi->ll_dt_exp);
2542         else if (cmd == OBD_IOC_GETMDNAME)
2543                 obd = class_exp2obd(sbi->ll_md_exp);
2544         else
2545                 RETURN(-EINVAL);
2546
2547         if (!obd)
2548                 RETURN(-ENOENT);
2549
2550         if (copy_to_user((void __user *)arg, obd->obd_name,
2551                          strlen(obd->obd_name) + 1))
2552                 RETURN(-EFAULT);
2553
2554         RETURN(0);
2555 }
2556
2557 /**
2558  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2559  * fsname will be returned in this buffer; otherwise, a static buffer will be
2560  * used to store the fsname and returned to caller.
2561  */
2562 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2563 {
2564         static char fsname_static[MTI_NAME_MAXLEN];
2565         struct lustre_sb_info *lsi = s2lsi(sb);
2566         char *ptr;
2567         int len;
2568
2569         if (buf == NULL) {
2570                 /* this means the caller wants to use static buffer
2571                  * and it doesn't care about race. Usually this is
2572                  * in error reporting path */
2573                 buf = fsname_static;
2574                 buflen = sizeof(fsname_static);
2575         }
2576
2577         len = strlen(lsi->lsi_lmd->lmd_profile);
2578         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2579         if (ptr && (strcmp(ptr, "-client") == 0))
2580                 len -= 7;
2581
2582         if (unlikely(len >= buflen))
2583                 len = buflen - 1;
2584         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2585         buf[len] = '\0';
2586
2587         return buf;
2588 }
2589
2590 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2591 {
2592         char *path = NULL;
2593
2594         struct path p;
2595
2596         p.dentry = dentry;
2597         p.mnt = current->fs->root.mnt;
2598         path_get(&p);
2599         path = d_path(&p, buf, bufsize);
2600         path_put(&p);
2601         return path;
2602 }
2603
2604 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2605 {
2606         char *buf, *path = NULL;
2607         struct dentry *dentry = NULL;
2608         struct inode *inode = page->mapping->host;
2609
2610         /* this can be called inside spin lock so use GFP_ATOMIC. */
2611         buf = (char *)__get_free_page(GFP_ATOMIC);
2612         if (buf != NULL) {
2613                 dentry = d_find_alias(page->mapping->host);
2614                 if (dentry != NULL)
2615                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2616         }
2617
2618         CDEBUG(D_WARNING,
2619                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2620                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2621                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2622                PFID(ll_inode2fid(inode)),
2623                (path && !IS_ERR(path)) ? path : "", ioret);
2624
2625         if (dentry != NULL)
2626                 dput(dentry);
2627
2628         if (buf != NULL)
2629                 free_page((unsigned long)buf);
2630 }
2631
2632 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2633                         struct lov_user_md **kbuf)
2634 {
2635         struct lov_user_md      lum;
2636         ssize_t                 lum_size;
2637         ENTRY;
2638
2639         if (copy_from_user(&lum, md, sizeof(lum)))
2640                 RETURN(-EFAULT);
2641
2642         lum_size = ll_lov_user_md_size(&lum);
2643         if (lum_size < 0)
2644                 RETURN(lum_size);
2645
2646         OBD_ALLOC(*kbuf, lum_size);
2647         if (*kbuf == NULL)
2648                 RETURN(-ENOMEM);
2649
2650         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2651                 OBD_FREE(*kbuf, lum_size);
2652                 RETURN(-EFAULT);
2653         }
2654
2655         RETURN(lum_size);
2656 }
2657
2658 /*
2659  * Compute llite root squash state after a change of root squash
2660  * configuration setting or add/remove of a lnet nid
2661  */
2662 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2663 {
2664         struct root_squash_info *squash = &sbi->ll_squash;
2665         int i;
2666         bool matched;
2667         struct lnet_process_id id;
2668
2669         /* Update norootsquash flag */
2670         down_write(&squash->rsi_sem);
2671         if (list_empty(&squash->rsi_nosquash_nids))
2672                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2673         else {
2674                 /* Do not apply root squash as soon as one of our NIDs is
2675                  * in the nosquash_nids list */
2676                 matched = false;
2677                 i = 0;
2678                 while (LNetGetId(i++, &id) != -ENOENT) {
2679                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2680                                 continue;
2681                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2682                                 matched = true;
2683                                 break;
2684                         }
2685                 }
2686                 if (matched)
2687                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2688                 else
2689                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2690         }
2691         up_write(&squash->rsi_sem);
2692 }
2693
2694 /**
2695  * Parse linkea content to extract information about a given hardlink
2696  *
2697  * \param[in]   ldata      - Initialized linkea data
2698  * \param[in]   linkno     - Link identifier
2699  * \param[out]  parent_fid - The entry's parent FID
2700  * \param[out]  ln         - Entry name destination buffer
2701  *
2702  * \retval 0 on success
2703  * \retval Appropriate negative error code on failure
2704  */
2705 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2706                             struct lu_fid *parent_fid, struct lu_name *ln)
2707 {
2708         unsigned int    idx;
2709         int             rc;
2710         ENTRY;
2711
2712         rc = linkea_init_with_rec(ldata);
2713         if (rc < 0)
2714                 RETURN(rc);
2715
2716         if (linkno >= ldata->ld_leh->leh_reccount)
2717                 /* beyond last link */
2718                 RETURN(-ENODATA);
2719
2720         linkea_first_entry(ldata);
2721         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2722                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2723                                     parent_fid);
2724                 if (idx == linkno)
2725                         break;
2726
2727                 linkea_next_entry(ldata);
2728         }
2729
2730         if (idx < linkno)
2731                 RETURN(-ENODATA);
2732
2733         RETURN(0);
2734 }
2735
2736 /**
2737  * Get parent FID and name of an identified link. Operation is performed for
2738  * a given link number, letting the caller iterate over linkno to list one or
2739  * all links of an entry.
2740  *
2741  * \param[in]     file - File descriptor against which to perform the operation
2742  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2743  *                       on and the available size. It is eventually filled with
2744  *                       the requested information or left untouched on error
2745  *
2746  * \retval - 0 on success
2747  * \retval - Appropriate negative error code on failure
2748  */
2749 int ll_getparent(struct file *file, struct getparent __user *arg)
2750 {
2751         struct dentry           *dentry = file_dentry(file);
2752         struct inode            *inode = file_inode(file);
2753         struct linkea_data      *ldata;
2754         struct lu_buf            buf = LU_BUF_NULL;
2755         struct lu_name           ln;
2756         struct lu_fid            parent_fid;
2757         __u32                    linkno;
2758         __u32                    name_size;
2759         int                      rc;
2760
2761         ENTRY;
2762
2763         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2764             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2765                 RETURN(-EPERM);
2766
2767         if (get_user(name_size, &arg->gp_name_size))
2768                 RETURN(-EFAULT);
2769
2770         if (get_user(linkno, &arg->gp_linkno))
2771                 RETURN(-EFAULT);
2772
2773         if (name_size > PATH_MAX)
2774                 RETURN(-EINVAL);
2775
2776         OBD_ALLOC(ldata, sizeof(*ldata));
2777         if (ldata == NULL)
2778                 RETURN(-ENOMEM);
2779
2780         rc = linkea_data_new(ldata, &buf);
2781         if (rc < 0)
2782                 GOTO(ldata_free, rc);
2783
2784         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2785         if (rc < 0)
2786                 GOTO(lb_free, rc);
2787
2788         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2789         if (rc < 0)
2790                 GOTO(lb_free, rc);
2791
2792         if (ln.ln_namelen >= name_size)
2793                 GOTO(lb_free, rc = -EOVERFLOW);
2794
2795         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2796                 GOTO(lb_free, rc = -EFAULT);
2797
2798         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2799                 GOTO(lb_free, rc = -EFAULT);
2800
2801         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2802                 GOTO(lb_free, rc = -EFAULT);
2803
2804 lb_free:
2805         lu_buf_free(&buf);
2806 ldata_free:
2807         OBD_FREE(ldata, sizeof(*ldata));
2808
2809         RETURN(rc);
2810 }