Whamcloud - gitweb
LU-12631 llite: report latency for filesystem ops
[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, 2017, 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/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
46 #include <linux/mm.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
50 #endif
51 #include <linux/security.h>
52
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #ifdef HAVE_UAPI_LINUX_MOUNT_H
55 #include <uapi/linux/mount.h>
56 #endif
57
58 #include <lustre_ha.h>
59 #include <lustre_dlm.h>
60 #include <lprocfs_status.h>
61 #include <lustre_disk.h>
62 #include <uapi/linux/lustre/lustre_param.h>
63 #include <lustre_log.h>
64 #include <cl_object.h>
65 #include <obd_cksum.h>
66 #include "llite_internal.h"
67
68 struct kmem_cache *ll_file_data_slab;
69
70 #ifndef log2
71 #define log2(n) ffz(~(n))
72 #endif
73
74 static struct ll_sb_info *ll_init_sbi(void)
75 {
76         struct ll_sb_info *sbi = NULL;
77         unsigned long pages;
78         unsigned long lru_page_max;
79         struct sysinfo si;
80         int rc;
81         int i;
82
83         ENTRY;
84
85         OBD_ALLOC_PTR(sbi);
86         if (sbi == NULL)
87                 RETURN(ERR_PTR(-ENOMEM));
88
89         rc = pcc_super_init(&sbi->ll_pcc_super);
90         if (rc < 0)
91                 GOTO(out_sbi, rc);
92
93         spin_lock_init(&sbi->ll_lock);
94         mutex_init(&sbi->ll_lco.lco_lock);
95         spin_lock_init(&sbi->ll_pp_extent_lock);
96         spin_lock_init(&sbi->ll_process_lock);
97         sbi->ll_rw_stats_on = 0;
98         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
99
100         si_meminfo(&si);
101         pages = si.totalram - si.totalhigh;
102         lru_page_max = pages / 2;
103
104         sbi->ll_ra_info.ra_async_max_active = 0;
105         sbi->ll_ra_info.ll_readahead_wq =
106                 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
107                                 sbi->ll_ra_info.ra_async_max_active);
108         if (!sbi->ll_ra_info.ll_readahead_wq)
109                 GOTO(out_pcc, rc = -ENOMEM);
110
111         /* initialize ll_cache data */
112         sbi->ll_cache = cl_cache_init(lru_page_max);
113         if (sbi->ll_cache == NULL)
114                 GOTO(out_destroy_ra, rc = -ENOMEM);
115
116         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
117                                            SBI_DEFAULT_READAHEAD_MAX);
118         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
119                                 sbi->ll_ra_info.ra_max_pages_per_file;
120         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
121         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
122
123         sbi->ll_flags |= LL_SBI_VERBOSE;
124 #ifdef ENABLE_CHECKSUM
125         sbi->ll_flags |= LL_SBI_CHECKSUM;
126 #endif
127 #ifdef ENABLE_FLOCK
128         sbi->ll_flags |= LL_SBI_FLOCK;
129 #endif
130
131 #ifdef HAVE_LRU_RESIZE_SUPPORT
132         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
133 #endif
134         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
135
136         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
138                                pp_r_hist.oh_lock);
139                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
140                                pp_w_hist.oh_lock);
141         }
142
143         /* metadata statahead is enabled by default */
144         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
145         sbi->ll_sa_max = LL_SA_RPC_DEF;
146         atomic_set(&sbi->ll_sa_total, 0);
147         atomic_set(&sbi->ll_sa_wrong, 0);
148         atomic_set(&sbi->ll_sa_running, 0);
149         atomic_set(&sbi->ll_agl_total, 0);
150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151         sbi->ll_flags |= LL_SBI_FAST_READ;
152         sbi->ll_flags |= LL_SBI_TINY_WRITE;
153
154         /* root squash */
155         sbi->ll_squash.rsi_uid = 0;
156         sbi->ll_squash.rsi_gid = 0;
157         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
158         spin_lock_init(&sbi->ll_squash.rsi_lock);
159
160         /* Per-filesystem file heat */
161         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
162         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
163         RETURN(sbi);
164 out_destroy_ra:
165         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
166 out_pcc:
167         pcc_super_fini(&sbi->ll_pcc_super);
168 out_sbi:
169         OBD_FREE_PTR(sbi);
170         RETURN(ERR_PTR(rc));
171 }
172
173 static void ll_free_sbi(struct super_block *sb)
174 {
175         struct ll_sb_info *sbi = ll_s2sbi(sb);
176         ENTRY;
177
178         if (sbi != NULL) {
179                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
180                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
181                 if (sbi->ll_ra_info.ll_readahead_wq)
182                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
183                 if (sbi->ll_cache != NULL) {
184                         cl_cache_decref(sbi->ll_cache);
185                         sbi->ll_cache = NULL;
186                 }
187                 pcc_super_fini(&sbi->ll_pcc_super);
188                 OBD_FREE(sbi, sizeof(*sbi));
189         }
190         EXIT;
191 }
192
193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
194                                     struct vfsmount *mnt)
195 {
196         struct inode *root = NULL;
197         struct ll_sb_info *sbi = ll_s2sbi(sb);
198         struct obd_statfs *osfs = NULL;
199         struct ptlrpc_request *request = NULL;
200         struct obd_connect_data *data = NULL;
201         struct obd_uuid *uuid;
202         struct md_op_data *op_data;
203         struct lustre_md lmd;
204         u64 valid;
205         int size, err, checksum;
206
207         ENTRY;
208         sbi->ll_md_obd = class_name2obd(md);
209         if (!sbi->ll_md_obd) {
210                 CERROR("MD %s: not setup or attached\n", md);
211                 RETURN(-EINVAL);
212         }
213
214         OBD_ALLOC_PTR(data);
215         if (data == NULL)
216                 RETURN(-ENOMEM);
217
218         OBD_ALLOC_PTR(osfs);
219         if (osfs == NULL) {
220                 OBD_FREE_PTR(data);
221                 RETURN(-ENOMEM);
222         }
223
224         /* pass client page size via ocd_grant_blkbits, the server should report
225          * back its backend blocksize for grant calculation purpose */
226         data->ocd_grant_blkbits = PAGE_SHIFT;
227
228         /* indicate MDT features supported by this client */
229         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
230                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
231                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
232                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
233                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
234                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
235                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
236                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
237                                   OBD_CONNECT_64BITHASH |
238                                   OBD_CONNECT_EINPROGRESS |
239                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
240                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
241                                   OBD_CONNECT_MAX_EASIZE |
242                                   OBD_CONNECT_FLOCK_DEAD |
243                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
244                                   OBD_CONNECT_OPEN_BY_FID |
245                                   OBD_CONNECT_DIR_STRIPE |
246                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
247                                   OBD_CONNECT_SUBTREE |
248                                   OBD_CONNECT_MULTIMODRPCS |
249                                   OBD_CONNECT_GRANT_PARAM |
250                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
251
252         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
253                                    OBD_CONNECT2_SUM_STATFS |
254                                    OBD_CONNECT2_OVERSTRIPING |
255                                    OBD_CONNECT2_FLR |
256                                    OBD_CONNECT2_LOCK_CONVERT |
257                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
258                                    OBD_CONNECT2_INC_XID |
259                                    OBD_CONNECT2_LSOM |
260                                    OBD_CONNECT2_ASYNC_DISCARD |
261                                    OBD_CONNECT2_PCC;
262
263 #ifdef HAVE_LRU_RESIZE_SUPPORT
264         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
265                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
266 #endif
267 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
268         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
269                                    OBD_CONNECT_LARGE_ACL;
270 #endif
271
272         data->ocd_cksum_types = obd_cksum_types_supported_client();
273
274         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
275                 /* flag mdc connection as lightweight, only used for test
276                  * purpose, use with care */
277                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
278
279         data->ocd_ibits_known = MDS_INODELOCK_FULL;
280         data->ocd_version = LUSTRE_VERSION_CODE;
281
282         if (sb->s_flags & SB_RDONLY)
283                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
284         if (sbi->ll_flags & LL_SBI_USER_XATTR)
285                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
286
287 #ifdef SB_NOSEC
288         /* Setting this indicates we correctly support S_NOSEC (See kernel
289          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
290          */
291         sb->s_flags |= SB_NOSEC;
292 #endif
293
294         if (sbi->ll_flags & LL_SBI_FLOCK)
295                 sbi->ll_fop = &ll_file_operations_flock;
296         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
297                 sbi->ll_fop = &ll_file_operations;
298         else
299                 sbi->ll_fop = &ll_file_operations_noflock;
300
301         /* always ping even if server suppress_pings */
302         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
303                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
304
305         obd_connect_set_secctx(data);
306
307 #if defined(CONFIG_SECURITY)
308         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
309 #endif
310
311         data->ocd_brw_size = MD_MAX_BRW_SIZE;
312
313         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
314                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
315         if (err == -EBUSY) {
316                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
317                                    "recovery, of which this client is not a "
318                                    "part. Please wait for recovery to complete,"
319                                    " abort, or time out.\n", md);
320                 GOTO(out, err);
321         } else if (err) {
322                 CERROR("cannot connect to %s: rc = %d\n", md, err);
323                 GOTO(out, err);
324         }
325
326         sbi->ll_md_exp->exp_connect_data = *data;
327
328         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
329                            LUSTRE_SEQ_METADATA);
330         if (err) {
331                 CERROR("%s: Can't init metadata layer FID infrastructure, "
332                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
333                 GOTO(out_md, err);
334         }
335
336         /* For mount, we only need fs info from MDT0, and also in DNE, it
337          * can make sure the client can be mounted as long as MDT0 is
338          * avaible */
339         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
340                         ktime_get_seconds() - sbi->ll_statfs_max_age,
341                         OBD_STATFS_FOR_MDT0);
342         if (err)
343                 GOTO(out_md_fid, err);
344
345         /* This needs to be after statfs to ensure connect has finished.
346          * Note that "data" does NOT contain the valid connect reply.
347          * If connecting to a 1.8 server there will be no LMV device, so
348          * we can access the MDC export directly and exp_connect_flags will
349          * be non-zero, but if accessing an upgraded 2.1 server it will
350          * have the correct flags filled in.
351          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
352         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
353         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
354             valid != CLIENT_CONNECT_MDT_REQD) {
355                 char *buf;
356
357                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
358                 obd_connect_flags2str(buf, PAGE_SIZE,
359                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
360                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
361                                    "feature(s) needed for correct operation "
362                                    "of this client (%s). Please upgrade "
363                                    "server or downgrade client.\n",
364                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
365                 OBD_FREE(buf, PAGE_SIZE);
366                 GOTO(out_md_fid, err = -EPROTO);
367         }
368
369         size = sizeof(*data);
370         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
371                            KEY_CONN_DATA,  &size, data);
372         if (err) {
373                 CERROR("%s: Get connect data failed: rc = %d\n",
374                        sbi->ll_md_exp->exp_obd->obd_name, err);
375                 GOTO(out_md_fid, err);
376         }
377
378         LASSERT(osfs->os_bsize);
379         sb->s_blocksize = osfs->os_bsize;
380         sb->s_blocksize_bits = log2(osfs->os_bsize);
381         sb->s_magic = LL_SUPER_MAGIC;
382         sb->s_maxbytes = MAX_LFS_FILESIZE;
383         sbi->ll_namelen = osfs->os_namelen;
384         sbi->ll_mnt.mnt = current->fs->root.mnt;
385
386         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
387             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
388                 LCONSOLE_INFO("Disabling user_xattr feature because "
389                               "it is not supported on the server\n");
390                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
391         }
392
393         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
394 #ifdef SB_POSIXACL
395                 sb->s_flags |= SB_POSIXACL;
396 #endif
397                 sbi->ll_flags |= LL_SBI_ACL;
398         } else {
399                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
400 #ifdef SB_POSIXACL
401                 sb->s_flags &= ~SB_POSIXACL;
402 #endif
403                 sbi->ll_flags &= ~LL_SBI_ACL;
404         }
405
406         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
407                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
408
409         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
410                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
411
412         if (obd_connect_has_secctx(data))
413                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
414
415         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
416                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
417                         LCONSOLE_INFO("%s: disabling xattr cache due to "
418                                       "unknown maximum xattr size.\n", dt);
419                 } else if (!sbi->ll_xattr_cache_set) {
420                         /* If xattr_cache is already set (no matter 0 or 1)
421                          * during processing llog, it won't be enabled here. */
422                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
423                         sbi->ll_xattr_cache_enabled = 1;
424                 }
425         }
426
427         sbi->ll_dt_obd = class_name2obd(dt);
428         if (!sbi->ll_dt_obd) {
429                 CERROR("DT %s: not setup or attached\n", dt);
430                 GOTO(out_md_fid, err = -ENODEV);
431         }
432
433         /* pass client page size via ocd_grant_blkbits, the server should report
434          * back its backend blocksize for grant calculation purpose */
435         data->ocd_grant_blkbits = PAGE_SHIFT;
436
437         /* indicate OST features supported by this client */
438         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
439                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
440                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
441                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
442                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
443                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
444                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
445                                   OBD_CONNECT_EINPROGRESS |
446                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
447                                   OBD_CONNECT_LAYOUTLOCK |
448                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
449                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
450                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
451
452 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
453  * can interoperate with an older version of lockahead which was released prior
454  * to landing in master. This support will be dropped when 2.13 development
455  * starts.  At the point, we should not just drop the connect flag (below), we
456  * should also remove the support in the code.
457  *
458  * Removing it means a few things:
459  * 1. Remove this section here
460  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
461  * 3. Remove function exp_connect_lockahead_old
462  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
463  * */
464 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
465         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
466 #endif
467
468         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
469                                    OBD_CONNECT2_INC_XID;
470
471         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
472                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
473
474         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
475          * disabled by default, because it can still be enabled on the
476          * fly via /sys. As a consequence, we still need to come to an
477          * agreement on the supported algorithms at connect time
478          */
479         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
480
481         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
482                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
483         else
484                 data->ocd_cksum_types = obd_cksum_types_supported_client();
485
486 #ifdef HAVE_LRU_RESIZE_SUPPORT
487         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
488 #endif
489         /* always ping even if server suppress_pings */
490         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
491                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
492
493         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
494                "ocd_grant: %d\n", data->ocd_connect_flags,
495                data->ocd_version, data->ocd_grant);
496
497         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
498         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
499
500         data->ocd_brw_size = DT_MAX_BRW_SIZE;
501
502         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
503                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
504         if (err == -EBUSY) {
505                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
506                                    "recovery, of which this client is not a "
507                                    "part.  Please wait for recovery to "
508                                    "complete, abort, or time out.\n", dt);
509                 GOTO(out_md, err);
510         } else if (err) {
511                 CERROR("%s: Cannot connect to %s: rc = %d\n",
512                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
513                 GOTO(out_md, err);
514         }
515
516         sbi->ll_dt_exp->exp_connect_data = *data;
517
518         /* Don't change value if it was specified in the config log */
519         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
520                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
521                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
522                               (data->ocd_brw_size >> PAGE_SHIFT));
523                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
524                     sbi->ll_ra_info.ra_max_pages_per_file)
525                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
526                                 sbi->ll_ra_info.ra_max_pages_per_file;
527         }
528
529         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
530                            LUSTRE_SEQ_METADATA);
531         if (err) {
532                 CERROR("%s: Can't init data layer FID infrastructure, "
533                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
534                 GOTO(out_dt, err);
535         }
536
537         mutex_lock(&sbi->ll_lco.lco_lock);
538         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
539         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
540         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
541         mutex_unlock(&sbi->ll_lco.lco_lock);
542
543         fid_zero(&sbi->ll_root_fid);
544         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
545                            &sbi->ll_root_fid);
546         if (err) {
547                 CERROR("cannot mds_connect: rc = %d\n", err);
548                 GOTO(out_lock_cn_cb, err);
549         }
550         if (!fid_is_sane(&sbi->ll_root_fid)) {
551                 CERROR("%s: Invalid root fid "DFID" during mount\n",
552                        sbi->ll_md_exp->exp_obd->obd_name,
553                        PFID(&sbi->ll_root_fid));
554                 GOTO(out_lock_cn_cb, err = -EINVAL);
555         }
556         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
557
558         sb->s_op = &lustre_super_operations;
559         sb->s_xattr = ll_xattr_handlers;
560 #if THREAD_SIZE >= 8192 /*b=17630*/
561         sb->s_export_op = &lustre_export_operations;
562 #endif
563
564         /* make root inode
565          * XXX: move this to after cbd setup? */
566         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
567         if (sbi->ll_flags & LL_SBI_ACL)
568                 valid |= OBD_MD_FLACL;
569
570         OBD_ALLOC_PTR(op_data);
571         if (op_data == NULL)
572                 GOTO(out_lock_cn_cb, err = -ENOMEM);
573
574         op_data->op_fid1 = sbi->ll_root_fid;
575         op_data->op_mode = 0;
576         op_data->op_valid = valid;
577
578         err = md_getattr(sbi->ll_md_exp, op_data, &request);
579
580         OBD_FREE_PTR(op_data);
581         if (err) {
582                 CERROR("%s: md_getattr failed for root: rc = %d\n",
583                        sbi->ll_md_exp->exp_obd->obd_name, err);
584                 GOTO(out_lock_cn_cb, err);
585         }
586
587         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
588                                sbi->ll_md_exp, &lmd);
589         if (err) {
590                 CERROR("failed to understand root inode md: rc = %d\n", err);
591                 ptlrpc_req_finished(request);
592                 GOTO(out_lock_cn_cb, err);
593         }
594
595         LASSERT(fid_is_sane(&sbi->ll_root_fid));
596         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
597                                             sbi->ll_flags & LL_SBI_32BIT_API),
598                        &lmd);
599         md_free_lustre_md(sbi->ll_md_exp, &lmd);
600         ptlrpc_req_finished(request);
601
602         if (IS_ERR(root)) {
603 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
604                 if (lmd.posix_acl) {
605                         posix_acl_release(lmd.posix_acl);
606                         lmd.posix_acl = NULL;
607                 }
608 #endif
609                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
610                 root = NULL;
611                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
612                        sbi->ll_fsname, err);
613                 GOTO(out_root, err);
614         }
615
616         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
617         if (sbi->ll_checksum_set) {
618                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
619                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
620                                          sizeof(checksum), &checksum, NULL);
621                 if (err) {
622                         CERROR("%s: Set checksum failed: rc = %d\n",
623                                sbi->ll_dt_exp->exp_obd->obd_name, err);
624                         GOTO(out_root, err);
625                 }
626         }
627         cl_sb_init(sb);
628
629         sb->s_root = d_make_root(root);
630         if (sb->s_root == NULL) {
631                 err = -ENOMEM;
632                 CERROR("%s: can't make root dentry: rc = %d\n",
633                        sbi->ll_fsname, err);
634                 GOTO(out_root, err);
635         }
636
637         sbi->ll_sdev_orig = sb->s_dev;
638
639         /* We set sb->s_dev equal on all lustre clients in order to support
640          * NFS export clustering.  NFSD requires that the FSID be the same
641          * on all clients. */
642         /* s_dev is also used in lt_compare() to compare two fs, but that is
643          * only a node-local comparison. */
644         uuid = obd_get_uuid(sbi->ll_md_exp);
645         if (uuid != NULL)
646                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
647
648         if (data != NULL)
649                 OBD_FREE_PTR(data);
650         if (osfs != NULL)
651                 OBD_FREE_PTR(osfs);
652
653         if (sbi->ll_dt_obd) {
654                 err = sysfs_create_link(&sbi->ll_kset.kobj,
655                                         &sbi->ll_dt_obd->obd_kset.kobj,
656                                         sbi->ll_dt_obd->obd_type->typ_name);
657                 if (err < 0) {
658                         CERROR("%s: could not register %s in llite: rc = %d\n",
659                                dt, sbi->ll_fsname, err);
660                         err = 0;
661                 }
662         }
663
664         if (sbi->ll_md_obd) {
665                 err = sysfs_create_link(&sbi->ll_kset.kobj,
666                                         &sbi->ll_md_obd->obd_kset.kobj,
667                                         sbi->ll_md_obd->obd_type->typ_name);
668                 if (err < 0) {
669                         CERROR("%s: could not register %s in llite: rc = %d\n",
670                                md, sbi->ll_fsname, err);
671                         err = 0;
672                 }
673         }
674
675         RETURN(err);
676 out_root:
677         if (root)
678                 iput(root);
679 out_lock_cn_cb:
680         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
681 out_dt:
682         obd_disconnect(sbi->ll_dt_exp);
683         sbi->ll_dt_exp = NULL;
684         sbi->ll_dt_obd = NULL;
685 out_md_fid:
686         obd_fid_fini(sbi->ll_md_exp->exp_obd);
687 out_md:
688         obd_disconnect(sbi->ll_md_exp);
689         sbi->ll_md_exp = NULL;
690         sbi->ll_md_obd = NULL;
691 out:
692         if (data != NULL)
693                 OBD_FREE_PTR(data);
694         if (osfs != NULL)
695                 OBD_FREE_PTR(osfs);
696         return err;
697 }
698
699 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
700 {
701         int size, rc;
702
703         size = sizeof(*lmmsize);
704         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
705                           KEY_MAX_EASIZE, &size, lmmsize);
706         if (rc != 0) {
707                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
708                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
709                 RETURN(rc);
710         }
711
712         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
713
714         size = sizeof(int);
715         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
716                           KEY_MAX_EASIZE, &size, lmmsize);
717         if (rc)
718                 CERROR("Get max mdsize error rc %d\n", rc);
719
720         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
721
722         RETURN(rc);
723 }
724
725 /**
726  * Get the value of the default_easize parameter.
727  *
728  * \see client_obd::cl_default_mds_easize
729  *
730  * \param[in] sbi       superblock info for this filesystem
731  * \param[out] lmmsize  pointer to storage location for value
732  *
733  * \retval 0            on success
734  * \retval negative     negated errno on failure
735  */
736 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
737 {
738         int size, rc;
739
740         size = sizeof(int);
741         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
742                          KEY_DEFAULT_EASIZE, &size, lmmsize);
743         if (rc)
744                 CERROR("Get default mdsize error rc %d\n", rc);
745
746         RETURN(rc);
747 }
748
749 /**
750  * Set the default_easize parameter to the given value.
751  *
752  * \see client_obd::cl_default_mds_easize
753  *
754  * \param[in] sbi       superblock info for this filesystem
755  * \param[in] lmmsize   the size to set
756  *
757  * \retval 0            on success
758  * \retval negative     negated errno on failure
759  */
760 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
761 {
762         int rc;
763
764         if (lmmsize < sizeof(struct lov_mds_md) ||
765             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
766                 return -EINVAL;
767
768         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
769                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
770                                 sizeof(int), &lmmsize, NULL);
771
772         RETURN(rc);
773 }
774
775 static void client_common_put_super(struct super_block *sb)
776 {
777         struct ll_sb_info *sbi = ll_s2sbi(sb);
778         ENTRY;
779
780         cl_sb_fini(sb);
781
782         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
783         obd_disconnect(sbi->ll_dt_exp);
784         sbi->ll_dt_exp = NULL;
785
786         ll_debugfs_unregister_super(sb);
787
788         obd_fid_fini(sbi->ll_md_exp->exp_obd);
789         obd_disconnect(sbi->ll_md_exp);
790         sbi->ll_md_exp = NULL;
791
792         EXIT;
793 }
794
795 void ll_kill_super(struct super_block *sb)
796 {
797         struct ll_sb_info *sbi;
798         ENTRY;
799
800         /* not init sb ?*/
801         if (!(sb->s_flags & SB_ACTIVE))
802                 return;
803
804         sbi = ll_s2sbi(sb);
805         /* we need restore s_dev from changed for clustred NFS before put_super
806          * because new kernels have cached s_dev and change sb->s_dev in
807          * put_super not affected real removing devices */
808         if (sbi) {
809                 sb->s_dev = sbi->ll_sdev_orig;
810
811                 /* wait running statahead threads to quit */
812                 while (atomic_read(&sbi->ll_sa_running) > 0) {
813                         set_current_state(TASK_UNINTERRUPTIBLE);
814                         schedule_timeout(cfs_time_seconds(1) >> 3);
815                 }
816         }
817
818         EXIT;
819 }
820
821 static inline int ll_set_opt(const char *opt, char *data, int fl)
822 {
823         if (strncmp(opt, data, strlen(opt)) != 0)
824                 return 0;
825         else
826                 return fl;
827 }
828
829 /* non-client-specific mount options are parsed in lmd_parse */
830 static int ll_options(char *options, struct ll_sb_info *sbi)
831 {
832         int tmp;
833         char *s1 = options, *s2;
834         int *flags = &sbi->ll_flags;
835         ENTRY;
836
837         if (!options)
838                 RETURN(0);
839
840         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
841
842         while (*s1) {
843                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
844                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
845                 if (tmp) {
846                         *flags |= tmp;
847                         goto next;
848                 }
849                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
850                 if (tmp) {
851                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
852                         goto next;
853                 }
854                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
855                 if (tmp) {
856                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
857                         goto next;
858                 }
859                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
860                 if (tmp) {
861                         *flags &= ~tmp;
862                         goto next;
863                 }
864                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
865                 if (tmp) {
866                         *flags |= tmp;
867                         goto next;
868                 }
869                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
870                 if (tmp) {
871                         *flags &= ~tmp;
872                         goto next;
873                 }
874                 tmp = ll_set_opt("context", s1, 1);
875                 if (tmp)
876                         goto next;
877                 tmp = ll_set_opt("fscontext", s1, 1);
878                 if (tmp)
879                         goto next;
880                 tmp = ll_set_opt("defcontext", s1, 1);
881                 if (tmp)
882                         goto next;
883                 tmp = ll_set_opt("rootcontext", s1, 1);
884                 if (tmp)
885                         goto next;
886                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
887                 if (tmp) {
888                         *flags |= tmp;
889                         goto next;
890                 }
891                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
892                 if (tmp) {
893                         *flags &= ~tmp;
894                         goto next;
895                 }
896
897                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
898                 if (tmp) {
899                         *flags |= tmp;
900                         sbi->ll_checksum_set = 1;
901                         goto next;
902                 }
903                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
904                 if (tmp) {
905                         *flags &= ~tmp;
906                         sbi->ll_checksum_set = 1;
907                         goto next;
908                 }
909                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
910                 if (tmp) {
911                         *flags |= tmp;
912                         goto next;
913                 }
914                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
915                 if (tmp) {
916                         *flags &= ~tmp;
917                         goto next;
918                 }
919                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
920                 if (tmp) {
921                         *flags |= tmp;
922                         goto next;
923                 }
924                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
925                 if (tmp) {
926                         *flags &= ~tmp;
927                         goto next;
928                 }
929                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
930                 if (tmp) {
931                         *flags |= tmp;
932                         goto next;
933                 }
934                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
935                 if (tmp) {
936                         *flags |= tmp;
937                         goto next;
938                 }
939                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
940                 if (tmp) {
941                         *flags &= ~tmp;
942                         goto next;
943                 }
944                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
945                 if (tmp) {
946                         *flags |= tmp;
947                         goto next;
948                 }
949                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
950                                    s1);
951                 RETURN(-EINVAL);
952
953 next:
954                 /* Find next opt */
955                 s2 = strchr(s1, ',');
956                 if (s2 == NULL)
957                         break;
958                 s1 = s2 + 1;
959         }
960         RETURN(0);
961 }
962
963 void ll_lli_init(struct ll_inode_info *lli)
964 {
965         lli->lli_inode_magic = LLI_INODE_MAGIC;
966         lli->lli_flags = 0;
967         spin_lock_init(&lli->lli_lock);
968         lli->lli_posix_acl = NULL;
969         /* Do not set lli_fid, it has been initialized already. */
970         fid_zero(&lli->lli_pfid);
971         lli->lli_mds_read_och = NULL;
972         lli->lli_mds_write_och = NULL;
973         lli->lli_mds_exec_och = NULL;
974         lli->lli_open_fd_read_count = 0;
975         lli->lli_open_fd_write_count = 0;
976         lli->lli_open_fd_exec_count = 0;
977         mutex_init(&lli->lli_och_mutex);
978         spin_lock_init(&lli->lli_agl_lock);
979         spin_lock_init(&lli->lli_layout_lock);
980         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
981         lli->lli_clob = NULL;
982
983         init_rwsem(&lli->lli_xattrs_list_rwsem);
984         mutex_init(&lli->lli_xattrs_enq_lock);
985
986         LASSERT(lli->lli_vfs_inode.i_mode != 0);
987         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
988                 lli->lli_opendir_key = NULL;
989                 lli->lli_sai = NULL;
990                 spin_lock_init(&lli->lli_sa_lock);
991                 lli->lli_opendir_pid = 0;
992                 lli->lli_sa_enabled = 0;
993                 init_rwsem(&lli->lli_lsm_sem);
994         } else {
995                 mutex_init(&lli->lli_size_mutex);
996                 lli->lli_symlink_name = NULL;
997                 init_rwsem(&lli->lli_trunc_sem);
998                 range_lock_tree_init(&lli->lli_write_tree);
999                 init_rwsem(&lli->lli_glimpse_sem);
1000                 lli->lli_glimpse_time = ktime_set(0, 0);
1001                 INIT_LIST_HEAD(&lli->lli_agl_list);
1002                 lli->lli_agl_index = 0;
1003                 lli->lli_async_rc = 0;
1004                 spin_lock_init(&lli->lli_heat_lock);
1005                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1006                 lli->lli_heat_flags = 0;
1007                 mutex_init(&lli->lli_pcc_lock);
1008                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1009                 lli->lli_pcc_inode = NULL;
1010                 mutex_init(&lli->lli_group_mutex);
1011                 lli->lli_group_users = 0;
1012                 lli->lli_group_gid = 0;
1013         }
1014         mutex_init(&lli->lli_layout_mutex);
1015         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1016 }
1017
1018 #define MAX_STRING_SIZE 128
1019
1020 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1021
1022 #define LSI_BDI_INITIALIZED     0x00400000
1023
1024 #ifndef HAVE_BDI_CAP_MAP_COPY
1025 # define BDI_CAP_MAP_COPY       0
1026 #endif
1027
1028 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1029 {
1030         struct  lustre_sb_info *lsi = s2lsi(sb);
1031         char buf[MAX_STRING_SIZE];
1032         va_list args;
1033         int err;
1034
1035         err = bdi_init(&lsi->lsi_bdi);
1036         if (err)
1037                 return err;
1038
1039         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1040         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1041         lsi->lsi_bdi.name = "lustre";
1042         va_start(args, fmt);
1043         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1044         va_end(args);
1045         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1046         va_end(args);
1047         if (!err)
1048                 sb->s_bdi = &lsi->lsi_bdi;
1049
1050         return err;
1051 }
1052 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1053
1054 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1055 {
1056         struct  lustre_profile *lprof = NULL;
1057         struct  lustre_sb_info *lsi = s2lsi(sb);
1058         struct  ll_sb_info *sbi = NULL;
1059         char    *dt = NULL, *md = NULL;
1060         char    *profilenm = get_profile_name(sb);
1061         struct config_llog_instance *cfg;
1062         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1063         const int instlen = 16 + 2;
1064         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1065         char name[MAX_STRING_SIZE];
1066         int md_len = 0;
1067         int dt_len = 0;
1068         uuid_t uuid;
1069         char *ptr;
1070         int len;
1071         int err;
1072
1073         ENTRY;
1074         /* for ASLR, to map between cfg_instance and hashed ptr */
1075         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1076                profilenm, cfg_instance, sb);
1077
1078         try_module_get(THIS_MODULE);
1079
1080         OBD_ALLOC_PTR(cfg);
1081         if (cfg == NULL)
1082                 GOTO(out_free_cfg, err = -ENOMEM);
1083
1084         /* client additional sb info */
1085         lsi->lsi_llsbi = sbi = ll_init_sbi();
1086         if (IS_ERR(sbi))
1087                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1088
1089         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1090         if (err)
1091                 GOTO(out_free_cfg, err);
1092
1093         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1094         sb->s_d_op = &ll_d_ops;
1095
1096         /* UUID handling */
1097         generate_random_uuid(uuid.b);
1098         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1099
1100         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1101
1102         /* Get fsname */
1103         len = strlen(profilenm);
1104         ptr = strrchr(profilenm, '-');
1105         if (ptr && (strcmp(ptr, "-client") == 0))
1106                 len -= 7;
1107
1108         if (len > LUSTRE_MAXFSNAME) {
1109                 if (unlikely(len >= MAX_STRING_SIZE))
1110                         len = MAX_STRING_SIZE - 1;
1111                 strncpy(name, profilenm, len);
1112                 name[len] = '\0';
1113                 err = -ENAMETOOLONG;
1114                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1115                        name, LUSTRE_MAXFSNAME, err);
1116                 GOTO(out_free_cfg, err);
1117         }
1118         strncpy(sbi->ll_fsname, profilenm, len);
1119         sbi->ll_fsname[len] = '\0';
1120
1121         /* Mount info */
1122         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1123                  profilenm, cfg_instance);
1124
1125         err = super_setup_bdi_name(sb, "%s", name);
1126         if (err)
1127                 GOTO(out_free_cfg, err);
1128
1129         /* Call ll_debugfs_register_super() before lustre_process_log()
1130          * so that "llite.*.*" params can be processed correctly.
1131          */
1132         err = ll_debugfs_register_super(sb, name);
1133         if (err < 0) {
1134                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1135                        sbi->ll_fsname, err);
1136                 err = 0;
1137         }
1138
1139         /* The cfg_instance is a value unique to this super, in case some
1140          * joker tries to mount the same fs at two mount points.
1141          */
1142         cfg->cfg_instance = cfg_instance;
1143         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1144         cfg->cfg_callback = class_config_llog_handler;
1145         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1146         /* set up client obds */
1147         err = lustre_process_log(sb, profilenm, cfg);
1148         if (err < 0)
1149                 GOTO(out_debugfs, err);
1150
1151         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1152         lprof = class_get_profile(profilenm);
1153         if (lprof == NULL) {
1154                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1155                                    " read from the MGS.  Does that filesystem "
1156                                    "exist?\n", profilenm);
1157                 GOTO(out_debugfs, err = -EINVAL);
1158         }
1159         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1160                lprof->lp_md, lprof->lp_dt);
1161
1162         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1163         OBD_ALLOC(dt, dt_len);
1164         if (!dt)
1165                 GOTO(out_profile, err = -ENOMEM);
1166         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1167
1168         md_len = strlen(lprof->lp_md) + instlen + 2;
1169         OBD_ALLOC(md, md_len);
1170         if (!md)
1171                 GOTO(out_free_dt, err = -ENOMEM);
1172         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1173
1174         /* connections, registrations, sb setup */
1175         err = client_common_fill_super(sb, md, dt, mnt);
1176         if (err < 0)
1177                 GOTO(out_free_md, err);
1178
1179         sbi->ll_client_common_fill_super_succeeded = 1;
1180
1181 out_free_md:
1182         if (md)
1183                 OBD_FREE(md, md_len);
1184 out_free_dt:
1185         if (dt)
1186                 OBD_FREE(dt, dt_len);
1187 out_profile:
1188         if (lprof)
1189                 class_put_profile(lprof);
1190 out_debugfs:
1191         if (err < 0)
1192                 ll_debugfs_unregister_super(sb);
1193 out_free_cfg:
1194         if (cfg)
1195                 OBD_FREE_PTR(cfg);
1196
1197         if (err)
1198                 ll_put_super(sb);
1199         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1200                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1201         RETURN(err);
1202 } /* ll_fill_super */
1203
1204 void ll_put_super(struct super_block *sb)
1205 {
1206         struct config_llog_instance cfg, params_cfg;
1207         struct obd_device *obd;
1208         struct lustre_sb_info *lsi = s2lsi(sb);
1209         struct ll_sb_info *sbi = ll_s2sbi(sb);
1210         char *profilenm = get_profile_name(sb);
1211         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1212         long ccc_count;
1213         int next, force = 1, rc = 0;
1214         ENTRY;
1215
1216         if (IS_ERR(sbi))
1217                 GOTO(out_no_sbi, 0);
1218
1219         /* Should replace instance_id with something better for ASLR */
1220         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1221                profilenm, cfg_instance, sb);
1222
1223         cfg.cfg_instance = cfg_instance;
1224         lustre_end_log(sb, profilenm, &cfg);
1225
1226         params_cfg.cfg_instance = cfg_instance;
1227         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1228
1229         if (sbi->ll_md_exp) {
1230                 obd = class_exp2obd(sbi->ll_md_exp);
1231                 if (obd)
1232                         force = obd->obd_force;
1233         }
1234
1235         /* Wait for unstable pages to be committed to stable storage */
1236         if (force == 0) {
1237                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1238                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1239                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1240                         &lwi);
1241         }
1242
1243         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1244         if (force == 0 && rc != -EINTR)
1245                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1246
1247         /* We need to set force before the lov_disconnect in
1248            lustre_common_put_super, since l_d cleans up osc's as well. */
1249         if (force) {
1250                 next = 0;
1251                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1252                                                      &next)) != NULL) {
1253                         obd->obd_force = force;
1254                 }
1255         }
1256
1257         if (sbi->ll_client_common_fill_super_succeeded) {
1258                 /* Only if client_common_fill_super succeeded */
1259                 client_common_put_super(sb);
1260         }
1261
1262         next = 0;
1263         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1264                 class_manual_cleanup(obd);
1265         }
1266
1267         if (sbi->ll_flags & LL_SBI_VERBOSE)
1268                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1269
1270         if (profilenm)
1271                 class_del_profile(profilenm);
1272
1273 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1274         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1275                 bdi_destroy(&lsi->lsi_bdi);
1276                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1277         }
1278 #endif
1279
1280         ll_free_sbi(sb);
1281         lsi->lsi_llsbi = NULL;
1282 out_no_sbi:
1283         lustre_common_put_super(sb);
1284
1285         cl_env_cache_purge(~0);
1286
1287         module_put(THIS_MODULE);
1288
1289         EXIT;
1290 } /* client_put_super */
1291
1292 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1293 {
1294         struct inode *inode = NULL;
1295
1296         /* NOTE: we depend on atomic igrab() -bzzz */
1297         lock_res_and_lock(lock);
1298         if (lock->l_resource->lr_lvb_inode) {
1299                 struct ll_inode_info * lli;
1300                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1301                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1302                         inode = igrab(lock->l_resource->lr_lvb_inode);
1303                 } else {
1304                         inode = lock->l_resource->lr_lvb_inode;
1305                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1306                                          D_WARNING, lock, "lr_lvb_inode %p is "
1307                                          "bogus: magic %08x",
1308                                          lock->l_resource->lr_lvb_inode,
1309                                          lli->lli_inode_magic);
1310                         inode = NULL;
1311                 }
1312         }
1313         unlock_res_and_lock(lock);
1314         return inode;
1315 }
1316
1317 void ll_dir_clear_lsm_md(struct inode *inode)
1318 {
1319         struct ll_inode_info *lli = ll_i2info(inode);
1320
1321         LASSERT(S_ISDIR(inode->i_mode));
1322
1323         if (lli->lli_lsm_md) {
1324                 lmv_free_memmd(lli->lli_lsm_md);
1325                 lli->lli_lsm_md = NULL;
1326         }
1327
1328         if (lli->lli_default_lsm_md) {
1329                 lmv_free_memmd(lli->lli_default_lsm_md);
1330                 lli->lli_default_lsm_md = NULL;
1331         }
1332 }
1333
1334 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1335                                       const struct lu_fid *fid,
1336                                       struct lustre_md *md)
1337 {
1338         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1339         struct mdt_body         *body = md->body;
1340         struct inode            *inode;
1341         ino_t                   ino;
1342         ENTRY;
1343
1344         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1345         inode = iget_locked(sb, ino);
1346         if (inode == NULL) {
1347                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1348                        sbi->ll_fsname, PFID(fid));
1349                 RETURN(ERR_PTR(-ENOENT));
1350         }
1351
1352         if (inode->i_state & I_NEW) {
1353                 struct ll_inode_info *lli = ll_i2info(inode);
1354                 struct lmv_stripe_md *lsm = md->lmv;
1355
1356                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1357                                 (body->mbo_mode & S_IFMT);
1358                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1359                          PFID(fid));
1360
1361                 inode->i_mtime.tv_sec = 0;
1362                 inode->i_atime.tv_sec = 0;
1363                 inode->i_ctime.tv_sec = 0;
1364                 inode->i_rdev = 0;
1365
1366 #ifdef HAVE_BACKING_DEV_INFO
1367                 /* initializing backing dev info. */
1368                 inode->i_mapping->backing_dev_info =
1369                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1370 #endif
1371                 inode->i_op = &ll_dir_inode_operations;
1372                 inode->i_fop = &ll_dir_operations;
1373                 lli->lli_fid = *fid;
1374                 ll_lli_init(lli);
1375
1376                 LASSERT(lsm != NULL);
1377                 /* master object FID */
1378                 lli->lli_pfid = body->mbo_fid1;
1379                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1380                        lli, PFID(fid), PFID(&lli->lli_pfid));
1381                 unlock_new_inode(inode);
1382         }
1383
1384         RETURN(inode);
1385 }
1386
1387 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1388 {
1389         struct lu_fid *fid;
1390         struct lmv_stripe_md *lsm = md->lmv;
1391         struct ll_inode_info *lli = ll_i2info(inode);
1392         int i;
1393
1394         LASSERT(lsm != NULL);
1395
1396         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1397                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1398         lsm_md_dump(D_INODE, lsm);
1399
1400         if (!lmv_dir_striped(lsm))
1401                 goto out;
1402
1403         /* XXX sigh, this lsm_root initialization should be in
1404          * LMV layer, but it needs ll_iget right now, so we
1405          * put this here right now. */
1406         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1407                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1408                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1409
1410                 if (!fid_is_sane(fid))
1411                         continue;
1412
1413                 /* Unfortunately ll_iget will call ll_update_inode,
1414                  * where the initialization of slave inode is slightly
1415                  * different, so it reset lsm_md to NULL to avoid
1416                  * initializing lsm for slave inode. */
1417                 lsm->lsm_md_oinfo[i].lmo_root =
1418                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1419                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1420                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1421
1422                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1423                         while (i-- > 0) {
1424                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1425                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1426                         }
1427                         return rc;
1428                 }
1429         }
1430 out:
1431         lli->lli_lsm_md = lsm;
1432
1433         return 0;
1434 }
1435
1436 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1437 {
1438         struct ll_inode_info *lli = ll_i2info(inode);
1439
1440         if (!md->default_lmv) {
1441                 /* clear default lsm */
1442                 if (lli->lli_default_lsm_md) {
1443                         down_write(&lli->lli_lsm_sem);
1444                         if (lli->lli_default_lsm_md) {
1445                                 lmv_free_memmd(lli->lli_default_lsm_md);
1446                                 lli->lli_default_lsm_md = NULL;
1447                         }
1448                         up_write(&lli->lli_lsm_sem);
1449                 }
1450         } else if (lli->lli_default_lsm_md) {
1451                 /* update default lsm if it changes */
1452                 down_read(&lli->lli_lsm_sem);
1453                 if (lli->lli_default_lsm_md &&
1454                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1455                         up_read(&lli->lli_lsm_sem);
1456                         down_write(&lli->lli_lsm_sem);
1457                         if (lli->lli_default_lsm_md)
1458                                 lmv_free_memmd(lli->lli_default_lsm_md);
1459                         lli->lli_default_lsm_md = md->default_lmv;
1460                         lsm_md_dump(D_INODE, md->default_lmv);
1461                         md->default_lmv = NULL;
1462                         up_write(&lli->lli_lsm_sem);
1463                 } else {
1464                         up_read(&lli->lli_lsm_sem);
1465                 }
1466         } else {
1467                 /* init default lsm */
1468                 down_write(&lli->lli_lsm_sem);
1469                 lli->lli_default_lsm_md = md->default_lmv;
1470                 lsm_md_dump(D_INODE, md->default_lmv);
1471                 md->default_lmv = NULL;
1472                 up_write(&lli->lli_lsm_sem);
1473         }
1474 }
1475
1476 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1477 {
1478         struct ll_inode_info *lli = ll_i2info(inode);
1479         struct lmv_stripe_md *lsm = md->lmv;
1480         int rc = 0;
1481
1482         ENTRY;
1483
1484         LASSERT(S_ISDIR(inode->i_mode));
1485         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1486                PFID(ll_inode2fid(inode)));
1487
1488         /* update default LMV */
1489         if (md->default_lmv)
1490                 ll_update_default_lsm_md(inode, md);
1491
1492         /*
1493          * no striped information from request, lustre_md from req does not
1494          * include stripeEA, see ll_md_setattr()
1495          */
1496         if (!lsm)
1497                 RETURN(0);
1498
1499         /*
1500          * normally dir layout doesn't change, only take read lock to check
1501          * that to avoid blocking other MD operations.
1502          */
1503         if (lli->lli_lsm_md)
1504                 down_read(&lli->lli_lsm_sem);
1505         else
1506                 down_write(&lli->lli_lsm_sem);
1507
1508         /*
1509          * if dir layout mismatch, check whether version is increased, which
1510          * means layout is changed, this happens in dir migration and lfsck.
1511          *
1512          * foreign LMV should not change.
1513          */
1514         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1515                 if (lmv_dir_striped(lli->lli_lsm_md) &&
1516                     lsm->lsm_md_layout_version <=
1517                     lli->lli_lsm_md->lsm_md_layout_version) {
1518                         CERROR("%s: "DFID" dir layout mismatch:\n",
1519                                ll_i2sbi(inode)->ll_fsname,
1520                                PFID(&lli->lli_fid));
1521                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1522                         lsm_md_dump(D_ERROR, lsm);
1523                         GOTO(unlock, rc = -EINVAL);
1524                 }
1525
1526                 /* layout changed, switch to write lock */
1527                 up_read(&lli->lli_lsm_sem);
1528                 down_write(&lli->lli_lsm_sem);
1529                 ll_dir_clear_lsm_md(inode);
1530         }
1531
1532         /* set directory layout */
1533         if (!lli->lli_lsm_md) {
1534                 struct cl_attr  *attr;
1535
1536                 rc = ll_init_lsm_md(inode, md);
1537                 up_write(&lli->lli_lsm_sem);
1538                 if (rc != 0)
1539                         RETURN(rc);
1540
1541                 /* set md->lmv to NULL, so the following free lustre_md
1542                  * will not free this lsm */
1543                 md->lmv = NULL;
1544
1545                 /*
1546                  * md_merge_attr() may take long, since lsm is already set,
1547                  * switch to read lock.
1548                  */
1549                 down_read(&lli->lli_lsm_sem);
1550
1551                 if (!lmv_dir_striped(lli->lli_lsm_md))
1552                         GOTO(unlock, rc);
1553
1554                 OBD_ALLOC_PTR(attr);
1555                 if (attr == NULL)
1556                         GOTO(unlock, rc = -ENOMEM);
1557
1558                 /* validate the lsm */
1559                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1560                                    ll_md_blocking_ast);
1561                 if (rc != 0) {
1562                         OBD_FREE_PTR(attr);
1563                         GOTO(unlock, rc);
1564                 }
1565
1566                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1567                         md->body->mbo_nlink = attr->cat_nlink;
1568                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1569                         md->body->mbo_size = attr->cat_size;
1570                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1571                         md->body->mbo_atime = attr->cat_atime;
1572                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1573                         md->body->mbo_ctime = attr->cat_ctime;
1574                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1575                         md->body->mbo_mtime = attr->cat_mtime;
1576
1577                 OBD_FREE_PTR(attr);
1578         }
1579 unlock:
1580         up_read(&lli->lli_lsm_sem);
1581
1582         RETURN(rc);
1583 }
1584
1585 void ll_clear_inode(struct inode *inode)
1586 {
1587         struct ll_inode_info *lli = ll_i2info(inode);
1588         struct ll_sb_info *sbi = ll_i2sbi(inode);
1589
1590         ENTRY;
1591
1592         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1593                PFID(ll_inode2fid(inode)), inode);
1594
1595         if (S_ISDIR(inode->i_mode)) {
1596                 /* these should have been cleared in ll_file_release */
1597                 LASSERT(lli->lli_opendir_key == NULL);
1598                 LASSERT(lli->lli_sai == NULL);
1599                 LASSERT(lli->lli_opendir_pid == 0);
1600         } else {
1601                 pcc_inode_free(inode);
1602         }
1603
1604         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1605
1606         LASSERT(!lli->lli_open_fd_write_count);
1607         LASSERT(!lli->lli_open_fd_read_count);
1608         LASSERT(!lli->lli_open_fd_exec_count);
1609
1610         if (lli->lli_mds_write_och)
1611                 ll_md_real_close(inode, FMODE_WRITE);
1612         if (lli->lli_mds_exec_och)
1613                 ll_md_real_close(inode, FMODE_EXEC);
1614         if (lli->lli_mds_read_och)
1615                 ll_md_real_close(inode, FMODE_READ);
1616
1617         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1618                 OBD_FREE(lli->lli_symlink_name,
1619                          strlen(lli->lli_symlink_name) + 1);
1620                 lli->lli_symlink_name = NULL;
1621         }
1622
1623         ll_xattr_cache_destroy(inode);
1624
1625 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1626         forget_all_cached_acls(inode);
1627         if (lli->lli_posix_acl) {
1628                 posix_acl_release(lli->lli_posix_acl);
1629                 lli->lli_posix_acl = NULL;
1630         }
1631 #endif
1632         lli->lli_inode_magic = LLI_INODE_DEAD;
1633
1634         if (S_ISDIR(inode->i_mode))
1635                 ll_dir_clear_lsm_md(inode);
1636         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1637                 LASSERT(list_empty(&lli->lli_agl_list));
1638
1639         /*
1640          * XXX This has to be done before lsm is freed below, because
1641          * cl_object still uses inode lsm.
1642          */
1643         cl_inode_fini(inode);
1644
1645         EXIT;
1646 }
1647
1648 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1649 {
1650         struct lustre_md md;
1651         struct inode *inode = dentry->d_inode;
1652         struct ll_sb_info *sbi = ll_i2sbi(inode);
1653         struct ptlrpc_request *request = NULL;
1654         int rc, ia_valid;
1655         ENTRY;
1656
1657         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1658                                      LUSTRE_OPC_ANY, NULL);
1659         if (IS_ERR(op_data))
1660                 RETURN(PTR_ERR(op_data));
1661
1662         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1663         if (rc) {
1664                 ptlrpc_req_finished(request);
1665                 if (rc == -ENOENT) {
1666                         clear_nlink(inode);
1667                         /* Unlinked special device node? Or just a race?
1668                          * Pretend we done everything. */
1669                         if (!S_ISREG(inode->i_mode) &&
1670                             !S_ISDIR(inode->i_mode)) {
1671                                 ia_valid = op_data->op_attr.ia_valid;
1672                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1673                                 rc = simple_setattr(dentry, &op_data->op_attr);
1674                                 op_data->op_attr.ia_valid = ia_valid;
1675                         }
1676                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1677                         CERROR("md_setattr fails: rc = %d\n", rc);
1678                 }
1679                 RETURN(rc);
1680         }
1681
1682         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1683                               sbi->ll_md_exp, &md);
1684         if (rc) {
1685                 ptlrpc_req_finished(request);
1686                 RETURN(rc);
1687         }
1688
1689         ia_valid = op_data->op_attr.ia_valid;
1690         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1691          * cache is not cleared yet. */
1692         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1693         if (S_ISREG(inode->i_mode))
1694                 inode_lock(inode);
1695         rc = simple_setattr(dentry, &op_data->op_attr);
1696         if (S_ISREG(inode->i_mode))
1697                 inode_unlock(inode);
1698         op_data->op_attr.ia_valid = ia_valid;
1699
1700         rc = ll_update_inode(inode, &md);
1701         ptlrpc_req_finished(request);
1702
1703         RETURN(rc);
1704 }
1705
1706 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1707  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1708  * keep these values until such a time that objects are allocated for it.
1709  * We do the MDS operations first, as it is checking permissions for us.
1710  * We don't to the MDS RPC if there is nothing that we want to store there,
1711  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1712  * going to do an RPC anyways.
1713  *
1714  * If we are doing a truncate, we will send the mtime and ctime updates
1715  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1716  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1717  * at the same time.
1718  *
1719  * In case of HSMimport, we only set attr on MDS.
1720  */
1721 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1722                    enum op_xvalid xvalid, bool hsm_import)
1723 {
1724         struct inode *inode = dentry->d_inode;
1725         struct ll_inode_info *lli = ll_i2info(inode);
1726         struct md_op_data *op_data = NULL;
1727         ktime_t kstart = ktime_get();
1728         int rc = 0;
1729
1730         ENTRY;
1731
1732         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1733                "valid %x, hsm_import %d\n",
1734                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1735                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1736                hsm_import);
1737
1738         if (attr->ia_valid & ATTR_SIZE) {
1739                 /* Check new size against VFS/VM file size limit and rlimit */
1740                 rc = inode_newsize_ok(inode, attr->ia_size);
1741                 if (rc)
1742                         RETURN(rc);
1743
1744                 /* The maximum Lustre file size is variable, based on the
1745                  * OST maximum object size and number of stripes.  This
1746                  * needs another check in addition to the VFS check above. */
1747                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1748                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1749                                PFID(&lli->lli_fid), attr->ia_size,
1750                                ll_file_maxbytes(inode));
1751                         RETURN(-EFBIG);
1752                 }
1753
1754                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1755         }
1756
1757         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1758         if (attr->ia_valid & TIMES_SET_FLAGS) {
1759                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1760                     !cfs_capable(CFS_CAP_FOWNER))
1761                         RETURN(-EPERM);
1762         }
1763
1764         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1765         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1766              (attr->ia_valid & ATTR_CTIME)) {
1767                 attr->ia_ctime = current_time(inode);
1768                 xvalid |= OP_XVALID_CTIME_SET;
1769         }
1770         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1771             (attr->ia_valid & ATTR_ATIME)) {
1772                 attr->ia_atime = current_time(inode);
1773                 attr->ia_valid |= ATTR_ATIME_SET;
1774         }
1775         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1776             (attr->ia_valid & ATTR_MTIME)) {
1777                 attr->ia_mtime = current_time(inode);
1778                 attr->ia_valid |= ATTR_MTIME_SET;
1779         }
1780
1781         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1782                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1783                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1784                        ktime_get_real_seconds());
1785
1786         if (S_ISREG(inode->i_mode)) {
1787                 if (attr->ia_valid & ATTR_SIZE)
1788                         inode_dio_write_done(inode);
1789                 inode_unlock(inode);
1790         }
1791
1792         /* We always do an MDS RPC, even if we're only changing the size;
1793          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1794
1795         OBD_ALLOC_PTR(op_data);
1796         if (op_data == NULL)
1797                 GOTO(out, rc = -ENOMEM);
1798
1799         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1800                 /* If we are changing file size, file content is
1801                  * modified, flag it.
1802                  */
1803                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1804                 op_data->op_bias |= MDS_DATA_MODIFIED;
1805                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1806         }
1807
1808         if (attr->ia_valid & ATTR_FILE) {
1809                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1810
1811                 if (fd->fd_lease_och)
1812                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1813         }
1814
1815         op_data->op_attr = *attr;
1816         op_data->op_xvalid = xvalid;
1817
1818         rc = ll_md_setattr(dentry, op_data);
1819         if (rc)
1820                 GOTO(out, rc);
1821
1822         if (!S_ISREG(inode->i_mode) || hsm_import)
1823                 GOTO(out, rc = 0);
1824
1825         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1826                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1827             xvalid & OP_XVALID_CTIME_SET) {
1828                 bool cached = false;
1829
1830                 rc = pcc_inode_setattr(inode, attr, &cached);
1831                 if (cached) {
1832                         if (rc) {
1833                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1834                                        "rc = %d\n",
1835                                        ll_i2sbi(inode)->ll_fsname,
1836                                        PFID(&lli->lli_fid), rc);
1837                                 GOTO(out, rc);
1838                         }
1839                 } else {
1840                         /* For truncate and utimes sending attributes to OSTs,
1841                          * setting mtime/atime to the past will be performed
1842                          * under PW [0:EOF] extent lock (new_size:EOF for
1843                          * truncate). It may seem excessive to send mtime/atime
1844                          * updates to OSTs when not setting times to past, but
1845                          * it is necessary due to possible time
1846                          * de-synchronization between MDT inode and OST objects
1847                          */
1848                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1849                 }
1850         }
1851
1852         /* If the file was restored, it needs to set dirty flag.
1853          *
1854          * We've already sent MDS_DATA_MODIFIED flag in
1855          * ll_md_setattr() for truncate. However, the MDT refuses to
1856          * set the HS_DIRTY flag on released files, so we have to set
1857          * it again if the file has been restored. Please check how
1858          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1859          *
1860          * Please notice that if the file is not released, the previous
1861          * MDS_DATA_MODIFIED has taken effect and usually
1862          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1863          * This way we can save an RPC for common open + trunc
1864          * operation. */
1865         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1866                 struct hsm_state_set hss = {
1867                         .hss_valid = HSS_SETMASK,
1868                         .hss_setmask = HS_DIRTY,
1869                 };
1870                 int rc2;
1871
1872                 rc2 = ll_hsm_state_set(inode, &hss);
1873                 /* truncate and write can happen at the same time, so that
1874                  * the file can be set modified even though the file is not
1875                  * restored from released state, and ll_hsm_state_set() is
1876                  * not applicable for the file, and rc2 < 0 is normal in this
1877                  * case. */
1878                 if (rc2 < 0)
1879                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1880                                PFID(ll_inode2fid(inode)), rc2);
1881         }
1882
1883         EXIT;
1884 out:
1885         if (op_data != NULL)
1886                 ll_finish_md_op_data(op_data);
1887
1888         if (S_ISREG(inode->i_mode)) {
1889                 inode_lock(inode);
1890                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1891                         inode_dio_wait(inode);
1892                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1893                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1894                  * inode flags, so there is a gap where S_NOSEC is not set.
1895                  * This can cause a writer to take the i_mutex unnecessarily,
1896                  * but this is safe to do and should be rare. */
1897                 inode_has_no_xattr(inode);
1898         }
1899
1900         if (!rc)
1901                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1902                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1903                                    ktime_us_delta(ktime_get(), kstart));
1904
1905         return rc;
1906 }
1907
1908 int ll_setattr(struct dentry *de, struct iattr *attr)
1909 {
1910         int mode = de->d_inode->i_mode;
1911         enum op_xvalid xvalid = 0;
1912
1913         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1914                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1915                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1916
1917         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1918                                (ATTR_SIZE|ATTR_MODE)) &&
1919             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1920              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1921               !(attr->ia_mode & S_ISGID))))
1922                 attr->ia_valid |= ATTR_FORCE;
1923
1924         if ((attr->ia_valid & ATTR_MODE) &&
1925             (mode & S_ISUID) &&
1926             !(attr->ia_mode & S_ISUID) &&
1927             !(attr->ia_valid & ATTR_KILL_SUID))
1928                 attr->ia_valid |= ATTR_KILL_SUID;
1929
1930         if ((attr->ia_valid & ATTR_MODE) &&
1931             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1932             !(attr->ia_mode & S_ISGID) &&
1933             !(attr->ia_valid & ATTR_KILL_SGID))
1934                 attr->ia_valid |= ATTR_KILL_SGID;
1935
1936         return ll_setattr_raw(de, attr, xvalid, false);
1937 }
1938
1939 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1940                        u32 flags)
1941 {
1942         struct obd_statfs obd_osfs = { 0 };
1943         time64_t max_age;
1944         int rc;
1945
1946         ENTRY;
1947         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1948
1949         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1950         if (rc)
1951                 RETURN(rc);
1952
1953         osfs->os_type = LL_SUPER_MAGIC;
1954
1955         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1956               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1957
1958         if (osfs->os_state & OS_STATE_SUM)
1959                 GOTO(out, rc);
1960
1961         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1962                 flags |= OBD_STATFS_NODELAY;
1963
1964         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1965         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1966                 GOTO(out, rc = 0);
1967
1968         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1969                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1970                obd_osfs.os_files);
1971
1972         osfs->os_bsize = obd_osfs.os_bsize;
1973         osfs->os_blocks = obd_osfs.os_blocks;
1974         osfs->os_bfree = obd_osfs.os_bfree;
1975         osfs->os_bavail = obd_osfs.os_bavail;
1976
1977         /* If we have _some_ OSTs, but don't have as many free objects on the
1978          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1979          * to compensate, so that the "inodes in use" number is correct.
1980          * This should be kept in sync with lod_statfs() behaviour.
1981          */
1982         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1983                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1984                                  obd_osfs.os_ffree;
1985                 osfs->os_ffree = obd_osfs.os_ffree;
1986         }
1987
1988 out:
1989         RETURN(rc);
1990 }
1991
1992 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1993 {
1994         struct super_block *sb = de->d_sb;
1995         struct obd_statfs osfs;
1996         __u64 fsid = huge_encode_dev(sb->s_dev);
1997         ktime_t kstart = ktime_get();
1998         int rc;
1999
2000         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2001
2002         /* Some amount of caching on the client is allowed */
2003         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2004         if (rc)
2005                 return rc;
2006
2007         statfs_unpack(sfs, &osfs);
2008
2009         /* We need to downshift for all 32-bit kernels, because we can't
2010          * tell if the kernel is being called via sys_statfs64() or not.
2011          * Stop before overflowing f_bsize - in which case it is better
2012          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2013         if (sizeof(long) < 8) {
2014                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2015                         sfs->f_bsize <<= 1;
2016
2017                         osfs.os_blocks >>= 1;
2018                         osfs.os_bfree >>= 1;
2019                         osfs.os_bavail >>= 1;
2020                 }
2021         }
2022
2023         sfs->f_blocks = osfs.os_blocks;
2024         sfs->f_bfree = osfs.os_bfree;
2025         sfs->f_bavail = osfs.os_bavail;
2026         sfs->f_fsid.val[0] = (__u32)fsid;
2027         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2028
2029         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2030                            ktime_us_delta(ktime_get(), kstart));
2031
2032         return 0;
2033 }
2034
2035 void ll_inode_size_lock(struct inode *inode)
2036 {
2037         struct ll_inode_info *lli;
2038
2039         LASSERT(!S_ISDIR(inode->i_mode));
2040
2041         lli = ll_i2info(inode);
2042         mutex_lock(&lli->lli_size_mutex);
2043 }
2044
2045 void ll_inode_size_unlock(struct inode *inode)
2046 {
2047         struct ll_inode_info *lli;
2048
2049         lli = ll_i2info(inode);
2050         mutex_unlock(&lli->lli_size_mutex);
2051 }
2052
2053 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2054 {
2055         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2056         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2057                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2058         else
2059                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2060 }
2061
2062 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2063 {
2064         struct ll_inode_info *lli = ll_i2info(inode);
2065         struct mdt_body *body = md->body;
2066         struct ll_sb_info *sbi = ll_i2sbi(inode);
2067         int rc = 0;
2068
2069         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2070                 rc = cl_file_inode_init(inode, md);
2071                 if (rc)
2072                         return rc;
2073         }
2074
2075         if (S_ISDIR(inode->i_mode)) {
2076                 rc = ll_update_lsm_md(inode, md);
2077                 if (rc != 0)
2078                         return rc;
2079         }
2080
2081 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2082         if (body->mbo_valid & OBD_MD_FLACL) {
2083                 spin_lock(&lli->lli_lock);
2084                 if (lli->lli_posix_acl)
2085                         posix_acl_release(lli->lli_posix_acl);
2086                 lli->lli_posix_acl = md->posix_acl;
2087                 spin_unlock(&lli->lli_lock);
2088         }
2089 #endif
2090         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2091                                         sbi->ll_flags & LL_SBI_32BIT_API);
2092         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2093
2094         if (body->mbo_valid & OBD_MD_FLATIME) {
2095                 if (body->mbo_atime > inode->i_atime.tv_sec)
2096                         inode->i_atime.tv_sec = body->mbo_atime;
2097                 lli->lli_atime = body->mbo_atime;
2098         }
2099
2100         if (body->mbo_valid & OBD_MD_FLMTIME) {
2101                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2102                         CDEBUG(D_INODE,
2103                                "setting ino %lu mtime from %lld to %llu\n",
2104                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2105                                body->mbo_mtime);
2106                         inode->i_mtime.tv_sec = body->mbo_mtime;
2107                 }
2108                 lli->lli_mtime = body->mbo_mtime;
2109         }
2110
2111         if (body->mbo_valid & OBD_MD_FLCTIME) {
2112                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2113                         inode->i_ctime.tv_sec = body->mbo_ctime;
2114                 lli->lli_ctime = body->mbo_ctime;
2115         }
2116
2117         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2118         if (body->mbo_valid & OBD_MD_FLFLAGS)
2119                 ll_update_inode_flags(inode, body->mbo_flags);
2120         if (body->mbo_valid & OBD_MD_FLMODE)
2121                 inode->i_mode = (inode->i_mode & S_IFMT) |
2122                                 (body->mbo_mode & ~S_IFMT);
2123
2124         if (body->mbo_valid & OBD_MD_FLTYPE)
2125                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2126                                 (body->mbo_mode & S_IFMT);
2127
2128         LASSERT(inode->i_mode != 0);
2129         if (S_ISREG(inode->i_mode))
2130                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2131                                        LL_MAX_BLKSIZE_BITS);
2132         else
2133                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2134
2135         if (body->mbo_valid & OBD_MD_FLUID)
2136                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2137         if (body->mbo_valid & OBD_MD_FLGID)
2138                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2139         if (body->mbo_valid & OBD_MD_FLPROJID)
2140                 lli->lli_projid = body->mbo_projid;
2141         if (body->mbo_valid & OBD_MD_FLNLINK)
2142                 set_nlink(inode, body->mbo_nlink);
2143         if (body->mbo_valid & OBD_MD_FLRDEV)
2144                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2145
2146         if (body->mbo_valid & OBD_MD_FLID) {
2147                 /* FID shouldn't be changed! */
2148                 if (fid_is_sane(&lli->lli_fid)) {
2149                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2150                                  "Trying to change FID "DFID
2151                                  " to the "DFID", inode "DFID"(%p)\n",
2152                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2153                                  PFID(ll_inode2fid(inode)), inode);
2154                 } else {
2155                         lli->lli_fid = body->mbo_fid1;
2156                 }
2157         }
2158
2159         LASSERT(fid_seq(&lli->lli_fid) != 0);
2160
2161         if (body->mbo_valid & OBD_MD_FLSIZE) {
2162                 i_size_write(inode, body->mbo_size);
2163
2164                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2165                        PFID(ll_inode2fid(inode)),
2166                        (unsigned long long)body->mbo_size);
2167
2168                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2169                         inode->i_blocks = body->mbo_blocks;
2170         }
2171
2172         if (body->mbo_valid & OBD_MD_TSTATE) {
2173                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2174                  * clear it when done to ensure to start again
2175                  * glimpsing updated attrs
2176                  */
2177                 if (body->mbo_t_state & MS_RESTORE)
2178                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2179                 else
2180                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2181         }
2182
2183         return 0;
2184 }
2185
2186 int ll_read_inode2(struct inode *inode, void *opaque)
2187 {
2188         struct lustre_md *md = opaque;
2189         struct ll_inode_info *lli = ll_i2info(inode);
2190         int     rc;
2191         ENTRY;
2192
2193         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2194                PFID(&lli->lli_fid), inode);
2195
2196         /* Core attributes from the MDS first.  This is a new inode, and
2197          * the VFS doesn't zero times in the core inode so we have to do
2198          * it ourselves.  They will be overwritten by either MDS or OST
2199          * attributes - we just need to make sure they aren't newer.
2200          */
2201         inode->i_mtime.tv_sec = 0;
2202         inode->i_atime.tv_sec = 0;
2203         inode->i_ctime.tv_sec = 0;
2204         inode->i_rdev = 0;
2205         rc = ll_update_inode(inode, md);
2206         if (rc != 0)
2207                 RETURN(rc);
2208
2209         /* OIDEBUG(inode); */
2210
2211 #ifdef HAVE_BACKING_DEV_INFO
2212         /* initializing backing dev info. */
2213         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2214 #endif
2215         if (S_ISREG(inode->i_mode)) {
2216                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2217                 inode->i_op = &ll_file_inode_operations;
2218                 inode->i_fop = sbi->ll_fop;
2219                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2220                 EXIT;
2221         } else if (S_ISDIR(inode->i_mode)) {
2222                 inode->i_op = &ll_dir_inode_operations;
2223                 inode->i_fop = &ll_dir_operations;
2224                 EXIT;
2225         } else if (S_ISLNK(inode->i_mode)) {
2226                 inode->i_op = &ll_fast_symlink_inode_operations;
2227                 EXIT;
2228         } else {
2229                 inode->i_op = &ll_special_inode_operations;
2230
2231                 init_special_inode(inode, inode->i_mode,
2232                                    inode->i_rdev);
2233
2234                 EXIT;
2235         }
2236
2237         return 0;
2238 }
2239
2240 void ll_delete_inode(struct inode *inode)
2241 {
2242         struct ll_inode_info *lli = ll_i2info(inode);
2243         struct address_space *mapping = &inode->i_data;
2244         unsigned long nrpages;
2245         unsigned long flags;
2246
2247         ENTRY;
2248
2249         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2250                 /* It is last chance to write out dirty pages,
2251                  * otherwise we may lose data while umount.
2252                  *
2253                  * If i_nlink is 0 then just discard data. This is safe because
2254                  * local inode gets i_nlink 0 from server only for the last
2255                  * unlink, so that file is not opened somewhere else
2256                  */
2257                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2258                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2259         }
2260         truncate_inode_pages_final(mapping);
2261
2262         /* Workaround for LU-118: Note nrpages may not be totally updated when
2263          * truncate_inode_pages() returns, as there can be a page in the process
2264          * of deletion (inside __delete_from_page_cache()) in the specified
2265          * range. Thus mapping->nrpages can be non-zero when this function
2266          * returns even after truncation of the whole mapping.  Only do this if
2267          * npages isn't already zero.
2268          */
2269         nrpages = mapping->nrpages;
2270         if (nrpages) {
2271                 xa_lock_irqsave(&mapping->i_pages, flags);
2272                 nrpages = mapping->nrpages;
2273                 xa_unlock_irqrestore(&mapping->i_pages, flags);
2274         } /* Workaround end */
2275
2276         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2277                  "see https://jira.whamcloud.com/browse/LU-118\n",
2278                  ll_i2sbi(inode)->ll_fsname,
2279                  PFID(ll_inode2fid(inode)), inode, nrpages);
2280
2281         ll_clear_inode(inode);
2282         clear_inode(inode);
2283
2284         EXIT;
2285 }
2286
2287 int ll_iocontrol(struct inode *inode, struct file *file,
2288                  unsigned int cmd, unsigned long arg)
2289 {
2290         struct ll_sb_info *sbi = ll_i2sbi(inode);
2291         struct ptlrpc_request *req = NULL;
2292         int rc, flags = 0;
2293         ENTRY;
2294
2295         switch (cmd) {
2296         case FS_IOC_GETFLAGS: {
2297                 struct mdt_body *body;
2298                 struct md_op_data *op_data;
2299
2300                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2301                                              0, 0, LUSTRE_OPC_ANY,
2302                                              NULL);
2303                 if (IS_ERR(op_data))
2304                         RETURN(PTR_ERR(op_data));
2305
2306                 op_data->op_valid = OBD_MD_FLFLAGS;
2307                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2308                 ll_finish_md_op_data(op_data);
2309                 if (rc) {
2310                         CERROR("%s: failure inode "DFID": rc = %d\n",
2311                                sbi->ll_md_exp->exp_obd->obd_name,
2312                                PFID(ll_inode2fid(inode)), rc);
2313                         RETURN(-abs(rc));
2314                 }
2315
2316                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2317
2318                 flags = body->mbo_flags;
2319
2320                 ptlrpc_req_finished(req);
2321
2322                 RETURN(put_user(flags, (int __user *)arg));
2323         }
2324         case FS_IOC_SETFLAGS: {
2325                 struct iattr *attr;
2326                 struct md_op_data *op_data;
2327                 struct cl_object *obj;
2328                 struct fsxattr fa = { 0 };
2329
2330                 if (get_user(flags, (int __user *)arg))
2331                         RETURN(-EFAULT);
2332
2333                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2334                 if (flags & LUSTRE_PROJINHERIT_FL)
2335                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2336
2337                 rc = ll_ioctl_check_project(inode, &fa);
2338                 if (rc)
2339                         RETURN(rc);
2340
2341                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2342                                              LUSTRE_OPC_ANY, NULL);
2343                 if (IS_ERR(op_data))
2344                         RETURN(PTR_ERR(op_data));
2345
2346                 op_data->op_attr_flags = flags;
2347                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2348                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2349                 ll_finish_md_op_data(op_data);
2350                 ptlrpc_req_finished(req);
2351                 if (rc)
2352                         RETURN(rc);
2353
2354                 ll_update_inode_flags(inode, flags);
2355
2356                 obj = ll_i2info(inode)->lli_clob;
2357                 if (obj == NULL)
2358                         RETURN(0);
2359
2360                 OBD_ALLOC_PTR(attr);
2361                 if (attr == NULL)
2362                         RETURN(-ENOMEM);
2363
2364                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2365
2366                 OBD_FREE_PTR(attr);
2367                 RETURN(rc);
2368         }
2369         default:
2370                 RETURN(-ENOSYS);
2371         }
2372
2373         RETURN(0);
2374 }
2375
2376 int ll_flush_ctx(struct inode *inode)
2377 {
2378         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2379
2380         CDEBUG(D_SEC, "flush context for user %d\n",
2381                from_kuid(&init_user_ns, current_uid()));
2382
2383         obd_set_info_async(NULL, sbi->ll_md_exp,
2384                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2385                            0, NULL, NULL);
2386         obd_set_info_async(NULL, sbi->ll_dt_exp,
2387                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2388                            0, NULL, NULL);
2389         return 0;
2390 }
2391
2392 /* umount -f client means force down, don't save state */
2393 void ll_umount_begin(struct super_block *sb)
2394 {
2395         struct ll_sb_info *sbi = ll_s2sbi(sb);
2396         struct obd_device *obd;
2397         struct obd_ioctl_data *ioc_data;
2398         struct l_wait_info lwi;
2399         wait_queue_head_t waitq;
2400         ENTRY;
2401
2402         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2403                sb->s_count, atomic_read(&sb->s_active));
2404
2405         obd = class_exp2obd(sbi->ll_md_exp);
2406         if (obd == NULL) {
2407                 CERROR("Invalid MDC connection handle %#llx\n",
2408                        sbi->ll_md_exp->exp_handle.h_cookie);
2409                 EXIT;
2410                 return;
2411         }
2412         obd->obd_force = 1;
2413
2414         obd = class_exp2obd(sbi->ll_dt_exp);
2415         if (obd == NULL) {
2416                 CERROR("Invalid LOV connection handle %#llx\n",
2417                        sbi->ll_dt_exp->exp_handle.h_cookie);
2418                 EXIT;
2419                 return;
2420         }
2421         obd->obd_force = 1;
2422
2423         OBD_ALLOC_PTR(ioc_data);
2424         if (ioc_data) {
2425                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2426                               sizeof *ioc_data, ioc_data, NULL);
2427
2428                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2429                               sizeof *ioc_data, ioc_data, NULL);
2430
2431                 OBD_FREE_PTR(ioc_data);
2432         }
2433
2434         /* Really, we'd like to wait until there are no requests outstanding,
2435          * and then continue.  For now, we just periodically checking for vfs
2436          * to decrement mnt_cnt and hope to finish it within 10sec.
2437          */
2438         init_waitqueue_head(&waitq);
2439         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2440                                    cfs_time_seconds(1), NULL, NULL);
2441         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2442
2443         EXIT;
2444 }
2445
2446 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2447 {
2448         struct ll_sb_info *sbi = ll_s2sbi(sb);
2449         char *profilenm = get_profile_name(sb);
2450         int err;
2451         __u32 read_only;
2452
2453         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2454                 read_only = *flags & MS_RDONLY;
2455                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2456                                          sizeof(KEY_READ_ONLY),
2457                                          KEY_READ_ONLY, sizeof(read_only),
2458                                          &read_only, NULL);
2459                 if (err) {
2460                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2461                                       profilenm, read_only ?
2462                                       "read-only" : "read-write", err);
2463                         return err;
2464                 }
2465
2466                 if (read_only)
2467                         sb->s_flags |= SB_RDONLY;
2468                 else
2469                         sb->s_flags &= ~SB_RDONLY;
2470
2471                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2472                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2473                                       read_only ?  "read-only" : "read-write");
2474         }
2475         return 0;
2476 }
2477
2478 /**
2479  * Cleanup the open handle that is cached on MDT-side.
2480  *
2481  * For open case, the client side open handling thread may hit error
2482  * after the MDT grant the open. Under such case, the client should
2483  * send close RPC to the MDT as cleanup; otherwise, the open handle
2484  * on the MDT will be leaked there until the client umount or evicted.
2485  *
2486  * In further, if someone unlinked the file, because the open handle
2487  * holds the reference on such file/object, then it will block the
2488  * subsequent threads that want to locate such object via FID.
2489  *
2490  * \param[in] sb        super block for this file-system
2491  * \param[in] open_req  pointer to the original open request
2492  */
2493 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2494 {
2495         struct mdt_body                 *body;
2496         struct md_op_data               *op_data;
2497         struct ptlrpc_request           *close_req = NULL;
2498         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2499         ENTRY;
2500
2501         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2502         OBD_ALLOC_PTR(op_data);
2503         if (op_data == NULL) {
2504                 CWARN("%s: cannot allocate op_data to release open handle for "
2505                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2506
2507                 RETURN_EXIT;
2508         }
2509
2510         op_data->op_fid1 = body->mbo_fid1;
2511         op_data->op_open_handle = body->mbo_open_handle;
2512         op_data->op_mod_time = ktime_get_real_seconds();
2513         md_close(exp, op_data, NULL, &close_req);
2514         ptlrpc_req_finished(close_req);
2515         ll_finish_md_op_data(op_data);
2516
2517         EXIT;
2518 }
2519
2520 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2521                   struct super_block *sb, struct lookup_intent *it)
2522 {
2523         struct ll_sb_info *sbi = NULL;
2524         struct lustre_md md = { NULL };
2525         bool default_lmv_deleted = false;
2526         int rc;
2527
2528         ENTRY;
2529
2530         LASSERT(*inode || sb);
2531         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2532         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2533                               sbi->ll_md_exp, &md);
2534         if (rc != 0)
2535                 GOTO(out, rc);
2536
2537         /*
2538          * clear default_lmv only if intent_getattr reply doesn't contain it.
2539          * but it needs to be done after iget, check this early because
2540          * ll_update_lsm_md() may change md.
2541          */
2542         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2543             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2544                 default_lmv_deleted = true;
2545
2546         if (*inode) {
2547                 rc = ll_update_inode(*inode, &md);
2548                 if (rc != 0)
2549                         GOTO(out, rc);
2550         } else {
2551                 LASSERT(sb != NULL);
2552
2553                 /*
2554                  * At this point server returns to client's same fid as client
2555                  * generated for creating. So using ->fid1 is okay here.
2556                  */
2557                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2558                         CERROR("%s: Fid is insane "DFID"\n",
2559                                 sbi->ll_fsname,
2560                                 PFID(&md.body->mbo_fid1));
2561                         GOTO(out, rc = -EINVAL);
2562                 }
2563
2564                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2565                                              sbi->ll_flags & LL_SBI_32BIT_API),
2566                                  &md);
2567                 if (IS_ERR(*inode)) {
2568 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2569                         if (md.posix_acl) {
2570                                 posix_acl_release(md.posix_acl);
2571                                 md.posix_acl = NULL;
2572                         }
2573 #endif
2574                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2575                         *inode = NULL;
2576                         CERROR("new_inode -fatal: rc %d\n", rc);
2577                         GOTO(out, rc);
2578                 }
2579         }
2580
2581         /* Handling piggyback layout lock.
2582          * Layout lock can be piggybacked by getattr and open request.
2583          * The lsm can be applied to inode only if it comes with a layout lock
2584          * otherwise correct layout may be overwritten, for example:
2585          * 1. proc1: mdt returns a lsm but not granting layout
2586          * 2. layout was changed by another client
2587          * 3. proc2: refresh layout and layout lock granted
2588          * 4. proc1: to apply a stale layout */
2589         if (it != NULL && it->it_lock_mode != 0) {
2590                 struct lustre_handle lockh;
2591                 struct ldlm_lock *lock;
2592
2593                 lockh.cookie = it->it_lock_handle;
2594                 lock = ldlm_handle2lock(&lockh);
2595                 LASSERT(lock != NULL);
2596                 if (ldlm_has_layout(lock)) {
2597                         struct cl_object_conf conf;
2598
2599                         memset(&conf, 0, sizeof(conf));
2600                         conf.coc_opc = OBJECT_CONF_SET;
2601                         conf.coc_inode = *inode;
2602                         conf.coc_lock = lock;
2603                         conf.u.coc_layout = md.layout;
2604                         (void)ll_layout_conf(*inode, &conf);
2605                 }
2606                 LDLM_LOCK_PUT(lock);
2607         }
2608
2609         if (default_lmv_deleted)
2610                 ll_update_default_lsm_md(*inode, &md);
2611
2612         GOTO(out, rc = 0);
2613
2614 out:
2615         /* cleanup will be done if necessary */
2616         md_free_lustre_md(sbi->ll_md_exp, &md);
2617
2618         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2619                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2620
2621         return rc;
2622 }
2623
2624 int ll_obd_statfs(struct inode *inode, void __user *arg)
2625 {
2626         struct ll_sb_info *sbi = NULL;
2627         struct obd_export *exp;
2628         char *buf = NULL;
2629         struct obd_ioctl_data *data = NULL;
2630         __u32 type;
2631         int len = 0, rc;
2632
2633         if (!inode || !(sbi = ll_i2sbi(inode)))
2634                 GOTO(out_statfs, rc = -EINVAL);
2635
2636         rc = obd_ioctl_getdata(&buf, &len, arg);
2637         if (rc)
2638                 GOTO(out_statfs, rc);
2639
2640         data = (void*)buf;
2641         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2642             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2643                 GOTO(out_statfs, rc = -EINVAL);
2644
2645         if (data->ioc_inllen1 != sizeof(__u32) ||
2646             data->ioc_inllen2 != sizeof(__u32) ||
2647             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2648             data->ioc_plen2 != sizeof(struct obd_uuid))
2649                 GOTO(out_statfs, rc = -EINVAL);
2650
2651         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2652         if (type & LL_STATFS_LMV)
2653                 exp = sbi->ll_md_exp;
2654         else if (type & LL_STATFS_LOV)
2655                 exp = sbi->ll_dt_exp;
2656         else
2657                 GOTO(out_statfs, rc = -ENODEV);
2658
2659         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2660         if (rc)
2661                 GOTO(out_statfs, rc);
2662 out_statfs:
2663         OBD_FREE_LARGE(buf, len);
2664         return rc;
2665 }
2666
2667 /*
2668  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2669  * be called early to avoid deadlock.
2670  */
2671 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2672 {
2673         if (op_data->op_mea2_sem) {
2674                 up_read(op_data->op_mea2_sem);
2675                 op_data->op_mea2_sem = NULL;
2676         }
2677
2678         if (op_data->op_mea1_sem) {
2679                 up_read(op_data->op_mea1_sem);
2680                 op_data->op_mea1_sem = NULL;
2681         }
2682 }
2683
2684 /* this function prepares md_op_data hint for passing it down to MD stack. */
2685 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2686                                       struct inode *i1, struct inode *i2,
2687                                       const char *name, size_t namelen,
2688                                       __u32 mode, enum md_op_code opc,
2689                                       void *data)
2690 {
2691         LASSERT(i1 != NULL);
2692
2693         if (name == NULL) {
2694                 /* Do not reuse namelen for something else. */
2695                 if (namelen != 0)
2696                         return ERR_PTR(-EINVAL);
2697         } else {
2698                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2699                         return ERR_PTR(-ENAMETOOLONG);
2700
2701                 if (!lu_name_is_valid_2(name, namelen))
2702                         return ERR_PTR(-EINVAL);
2703         }
2704
2705         if (op_data == NULL)
2706                 OBD_ALLOC_PTR(op_data);
2707
2708         if (op_data == NULL)
2709                 return ERR_PTR(-ENOMEM);
2710
2711         ll_i2gids(op_data->op_suppgids, i1, i2);
2712         op_data->op_fid1 = *ll_inode2fid(i1);
2713         op_data->op_code = opc;
2714
2715         if (S_ISDIR(i1->i_mode)) {
2716                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2717                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2718                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2719                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2720         }
2721
2722         if (i2) {
2723                 op_data->op_fid2 = *ll_inode2fid(i2);
2724                 if (S_ISDIR(i2->i_mode)) {
2725                         if (i2 != i1) {
2726                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2727                                 op_data->op_mea2_sem =
2728                                                 &ll_i2info(i2)->lli_lsm_sem;
2729                         }
2730                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2731                 }
2732         } else {
2733                 fid_zero(&op_data->op_fid2);
2734         }
2735
2736         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2737                 op_data->op_cli_flags |= CLI_HASH64;
2738
2739         if (ll_need_32bit_api(ll_i2sbi(i1)))
2740                 op_data->op_cli_flags |= CLI_API32;
2741
2742         op_data->op_name = name;
2743         op_data->op_namelen = namelen;
2744         op_data->op_mode = mode;
2745         op_data->op_mod_time = ktime_get_real_seconds();
2746         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2747         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2748         op_data->op_cap = cfs_curproc_cap_pack();
2749         op_data->op_mds = 0;
2750         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2751              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2752                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2753         }
2754         op_data->op_data = data;
2755
2756         return op_data;
2757 }
2758
2759 void ll_finish_md_op_data(struct md_op_data *op_data)
2760 {
2761         ll_unlock_md_op_lsm(op_data);
2762         security_release_secctx(op_data->op_file_secctx,
2763                                 op_data->op_file_secctx_size);
2764         OBD_FREE_PTR(op_data);
2765 }
2766
2767 #ifdef HAVE_SUPEROPS_USE_DENTRY
2768 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2769 #else
2770 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2771 #endif
2772 {
2773         struct ll_sb_info *sbi;
2774
2775 #ifdef HAVE_SUPEROPS_USE_DENTRY
2776         LASSERT((seq != NULL) && (dentry != NULL));
2777         sbi = ll_s2sbi(dentry->d_sb);
2778 #else
2779         LASSERT((seq != NULL) && (vfs != NULL));
2780         sbi = ll_s2sbi(vfs->mnt_sb);
2781 #endif
2782
2783         if (sbi->ll_flags & LL_SBI_NOLCK)
2784                 seq_puts(seq, ",nolock");
2785
2786         /* "flock" is the default since 2.13, but it wasn't for many years,
2787          * so it is still useful to print this to show it is enabled.
2788          * Start to print "noflock" so it is now clear when flock is disabled.
2789          */
2790         if (sbi->ll_flags & LL_SBI_FLOCK)
2791                 seq_puts(seq, ",flock");
2792         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2793                 seq_puts(seq, ",localflock");
2794         else
2795                 seq_puts(seq, ",noflock");
2796
2797         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2798                 seq_puts(seq, ",user_xattr");
2799
2800         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2801                 seq_puts(seq, ",lazystatfs");
2802
2803         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2804                 seq_puts(seq, ",user_fid2path");
2805
2806         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2807                 seq_puts(seq, ",always_ping");
2808
2809         RETURN(0);
2810 }
2811
2812 /**
2813  * Get obd name by cmd, and copy out to user space
2814  */
2815 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2816 {
2817         struct ll_sb_info *sbi = ll_i2sbi(inode);
2818         struct obd_device *obd;
2819         ENTRY;
2820
2821         if (cmd == OBD_IOC_GETDTNAME)
2822                 obd = class_exp2obd(sbi->ll_dt_exp);
2823         else if (cmd == OBD_IOC_GETMDNAME)
2824                 obd = class_exp2obd(sbi->ll_md_exp);
2825         else
2826                 RETURN(-EINVAL);
2827
2828         if (!obd)
2829                 RETURN(-ENOENT);
2830
2831         if (copy_to_user((void __user *)arg, obd->obd_name,
2832                          strlen(obd->obd_name) + 1))
2833                 RETURN(-EFAULT);
2834
2835         RETURN(0);
2836 }
2837
2838 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2839 {
2840         char *path = NULL;
2841
2842         struct path p;
2843
2844         p.dentry = dentry;
2845         p.mnt = current->fs->root.mnt;
2846         path_get(&p);
2847         path = d_path(&p, buf, bufsize);
2848         path_put(&p);
2849         return path;
2850 }
2851
2852 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2853 {
2854         char *buf, *path = NULL;
2855         struct dentry *dentry = NULL;
2856         struct inode *inode = page->mapping->host;
2857
2858         /* this can be called inside spin lock so use GFP_ATOMIC. */
2859         buf = (char *)__get_free_page(GFP_ATOMIC);
2860         if (buf != NULL) {
2861                 dentry = d_find_alias(page->mapping->host);
2862                 if (dentry != NULL)
2863                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2864         }
2865
2866         /* The below message is checked in recovery-small.sh test_24b */
2867         CDEBUG(D_WARNING,
2868                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2869                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2870                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2871                PFID(ll_inode2fid(inode)),
2872                (path && !IS_ERR(path)) ? path : "", ioret);
2873
2874         if (dentry != NULL)
2875                 dput(dentry);
2876
2877         if (buf != NULL)
2878                 free_page((unsigned long)buf);
2879 }
2880
2881 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2882                         struct lov_user_md **kbuf)
2883 {
2884         struct lov_user_md      lum;
2885         ssize_t                 lum_size;
2886         ENTRY;
2887
2888         if (copy_from_user(&lum, md, sizeof(lum)))
2889                 RETURN(-EFAULT);
2890
2891         lum_size = ll_lov_user_md_size(&lum);
2892         if (lum_size < 0)
2893                 RETURN(lum_size);
2894
2895         OBD_ALLOC_LARGE(*kbuf, lum_size);
2896         if (*kbuf == NULL)
2897                 RETURN(-ENOMEM);
2898
2899         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2900                 OBD_FREE_LARGE(*kbuf, lum_size);
2901                 RETURN(-EFAULT);
2902         }
2903
2904         RETURN(lum_size);
2905 }
2906
2907 /*
2908  * Compute llite root squash state after a change of root squash
2909  * configuration setting or add/remove of a lnet nid
2910  */
2911 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2912 {
2913         struct root_squash_info *squash = &sbi->ll_squash;
2914         int i;
2915         bool matched;
2916         struct lnet_process_id id;
2917
2918         /* Update norootsquash flag */
2919         spin_lock(&squash->rsi_lock);
2920         if (list_empty(&squash->rsi_nosquash_nids))
2921                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2922         else {
2923                 /* Do not apply root squash as soon as one of our NIDs is
2924                  * in the nosquash_nids list */
2925                 matched = false;
2926                 i = 0;
2927                 while (LNetGetId(i++, &id) != -ENOENT) {
2928                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2929                                 continue;
2930                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2931                                 matched = true;
2932                                 break;
2933                         }
2934                 }
2935                 if (matched)
2936                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2937                 else
2938                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2939         }
2940         spin_unlock(&squash->rsi_lock);
2941 }
2942
2943 /**
2944  * Parse linkea content to extract information about a given hardlink
2945  *
2946  * \param[in]   ldata      - Initialized linkea data
2947  * \param[in]   linkno     - Link identifier
2948  * \param[out]  parent_fid - The entry's parent FID
2949  * \param[out]  ln         - Entry name destination buffer
2950  *
2951  * \retval 0 on success
2952  * \retval Appropriate negative error code on failure
2953  */
2954 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2955                             struct lu_fid *parent_fid, struct lu_name *ln)
2956 {
2957         unsigned int    idx;
2958         int             rc;
2959         ENTRY;
2960
2961         rc = linkea_init_with_rec(ldata);
2962         if (rc < 0)
2963                 RETURN(rc);
2964
2965         if (linkno >= ldata->ld_leh->leh_reccount)
2966                 /* beyond last link */
2967                 RETURN(-ENODATA);
2968
2969         linkea_first_entry(ldata);
2970         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2971                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2972                                     parent_fid);
2973                 if (idx == linkno)
2974                         break;
2975
2976                 linkea_next_entry(ldata);
2977         }
2978
2979         if (idx < linkno)
2980                 RETURN(-ENODATA);
2981
2982         RETURN(0);
2983 }
2984
2985 /**
2986  * Get parent FID and name of an identified link. Operation is performed for
2987  * a given link number, letting the caller iterate over linkno to list one or
2988  * all links of an entry.
2989  *
2990  * \param[in]     file - File descriptor against which to perform the operation
2991  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2992  *                       on and the available size. It is eventually filled with
2993  *                       the requested information or left untouched on error
2994  *
2995  * \retval - 0 on success
2996  * \retval - Appropriate negative error code on failure
2997  */
2998 int ll_getparent(struct file *file, struct getparent __user *arg)
2999 {
3000         struct inode            *inode = file_inode(file);
3001         struct linkea_data      *ldata;
3002         struct lu_buf            buf = LU_BUF_NULL;
3003         struct lu_name           ln;
3004         struct lu_fid            parent_fid;
3005         __u32                    linkno;
3006         __u32                    name_size;
3007         int                      rc;
3008
3009         ENTRY;
3010
3011         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3012             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3013                 RETURN(-EPERM);
3014
3015         if (get_user(name_size, &arg->gp_name_size))
3016                 RETURN(-EFAULT);
3017
3018         if (get_user(linkno, &arg->gp_linkno))
3019                 RETURN(-EFAULT);
3020
3021         if (name_size > PATH_MAX)
3022                 RETURN(-EINVAL);
3023
3024         OBD_ALLOC(ldata, sizeof(*ldata));
3025         if (ldata == NULL)
3026                 RETURN(-ENOMEM);
3027
3028         rc = linkea_data_new(ldata, &buf);
3029         if (rc < 0)
3030                 GOTO(ldata_free, rc);
3031
3032         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3033                            buf.lb_len, OBD_MD_FLXATTR);
3034         if (rc < 0)
3035                 GOTO(lb_free, rc);
3036
3037         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3038         if (rc < 0)
3039                 GOTO(lb_free, rc);
3040
3041         if (ln.ln_namelen >= name_size)
3042                 GOTO(lb_free, rc = -EOVERFLOW);
3043
3044         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3045                 GOTO(lb_free, rc = -EFAULT);
3046
3047         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3048                 GOTO(lb_free, rc = -EFAULT);
3049
3050         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3051                 GOTO(lb_free, rc = -EFAULT);
3052
3053 lb_free:
3054         lu_buf_free(&buf);
3055 ldata_free:
3056         OBD_FREE(ldata, sizeof(*ldata));
3057
3058         RETURN(rc);
3059 }