Whamcloud - gitweb
e20a0c767b7ac38d635083e91229208b9da369d3
[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  *
31  * lustre/llite/llite_lib.c
32  *
33  * Lustre Light Super operations
34  */
35
36 #define DEBUG_SUBSYSTEM S_LLITE
37
38 #include <linux/cpu.h>
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 #include <linux/delay.h>
49 #include <linux/uidgid.h>
50 #include <linux/fs_struct.h>
51
52 #ifndef HAVE_CPUS_READ_LOCK
53 #include <libcfs/linux/linux-cpu.h>
54 #endif
55 #include <uapi/linux/lustre/lustre_ioctl.h>
56 #ifdef HAVE_UAPI_LINUX_MOUNT_H
57 #include <uapi/linux/mount.h>
58 #endif
59
60 #include <lustre_ha.h>
61 #include <lustre_dlm.h>
62 #include <lprocfs_status.h>
63 #include <lustre_disk.h>
64 #include <uapi/linux/lustre/lustre_param.h>
65 #include <lustre_log.h>
66 #include <cl_object.h>
67 #include <obd_cksum.h>
68 #include "llite_internal.h"
69
70 struct kmem_cache *ll_file_data_slab;
71
72 #ifndef log2
73 #define log2(n) ffz(~(n))
74 #endif
75
76 /**
77  * If there is only one number of core visible to Lustre,
78  * async readahead will be disabled, to avoid massive over
79  * subscription, we use 1/2 of active cores as default max
80  * async readahead requests.
81  */
82 static inline unsigned int ll_get_ra_async_max_active(void)
83 {
84         return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
85 }
86
87 static struct ll_sb_info *ll_init_sbi(void)
88 {
89         struct ll_sb_info *sbi = NULL;
90         unsigned long pages;
91         unsigned long lru_page_max;
92         struct sysinfo si;
93         int rc;
94         int i;
95
96         ENTRY;
97
98         OBD_ALLOC_PTR(sbi);
99         if (sbi == NULL)
100                 RETURN(ERR_PTR(-ENOMEM));
101
102         rc = pcc_super_init(&sbi->ll_pcc_super);
103         if (rc < 0)
104                 GOTO(out_sbi, rc);
105
106         spin_lock_init(&sbi->ll_lock);
107         mutex_init(&sbi->ll_lco.lco_lock);
108         spin_lock_init(&sbi->ll_pp_extent_lock);
109         spin_lock_init(&sbi->ll_process_lock);
110         sbi->ll_rw_stats_on = 0;
111         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
112
113         si_meminfo(&si);
114         pages = si.totalram - si.totalhigh;
115         lru_page_max = pages / 2;
116
117         sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
118         sbi->ll_ra_info.ll_readahead_wq =
119                 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
120                                        0, CFS_CPT_ANY,
121                                        sbi->ll_ra_info.ra_async_max_active);
122         if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
123                 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
124
125         /* initialize ll_cache data */
126         sbi->ll_cache = cl_cache_init(lru_page_max);
127         if (sbi->ll_cache == NULL)
128                 GOTO(out_destroy_ra, rc = -ENOMEM);
129
130         /* initialize foreign symlink prefix path */
131         OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
132         if (sbi->ll_foreign_symlink_prefix == NULL)
133                 GOTO(out_destroy_ra, rc = -ENOMEM);
134         memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
135         sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
136
137         /* initialize foreign symlink upcall path, none by default */
138         OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
139         if (sbi->ll_foreign_symlink_upcall == NULL)
140                 GOTO(out_destroy_ra, rc = -ENOMEM);
141         memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
142         sbi->ll_foreign_symlink_upcall_items = NULL;
143         sbi->ll_foreign_symlink_upcall_nb_items = 0;
144         init_rwsem(&sbi->ll_foreign_symlink_sem);
145         /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
146          * not enabled by default
147          */
148
149         sbi->ll_ra_info.ra_max_pages =
150                 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
151         sbi->ll_ra_info.ra_max_pages_per_file =
152                 min(sbi->ll_ra_info.ra_max_pages / 4,
153                     SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
154         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
155                                 sbi->ll_ra_info.ra_max_pages_per_file;
156         sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
157         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
158         atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
159
160         sbi->ll_flags |= LL_SBI_VERBOSE;
161 #ifdef ENABLE_CHECKSUM
162         sbi->ll_flags |= LL_SBI_CHECKSUM;
163 #endif
164 #ifdef ENABLE_FLOCK
165         sbi->ll_flags |= LL_SBI_FLOCK;
166 #endif
167
168 #ifdef HAVE_LRU_RESIZE_SUPPORT
169         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
170 #endif
171         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
172
173         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
174                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
175                                pp_r_hist.oh_lock);
176                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
177                                pp_w_hist.oh_lock);
178         }
179
180         /* metadata statahead is enabled by default */
181         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
182         sbi->ll_sa_max = LL_SA_RPC_DEF;
183         atomic_set(&sbi->ll_sa_total, 0);
184         atomic_set(&sbi->ll_sa_wrong, 0);
185         atomic_set(&sbi->ll_sa_running, 0);
186         atomic_set(&sbi->ll_agl_total, 0);
187         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
188         sbi->ll_flags |= LL_SBI_FAST_READ;
189         sbi->ll_flags |= LL_SBI_TINY_WRITE;
190         sbi->ll_flags |= LL_SBI_PARALLEL_DIO;
191         ll_sbi_set_encrypt(sbi, true);
192
193         /* root squash */
194         sbi->ll_squash.rsi_uid = 0;
195         sbi->ll_squash.rsi_gid = 0;
196         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
197         spin_lock_init(&sbi->ll_squash.rsi_lock);
198
199         /* Per-filesystem file heat */
200         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
201         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
202
203         /* Per-fs open heat level before requesting open lock */
204         sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
205         sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
206         sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
207         RETURN(sbi);
208 out_destroy_ra:
209         if (sbi->ll_foreign_symlink_prefix)
210                 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
211         if (sbi->ll_cache) {
212                 cl_cache_decref(sbi->ll_cache);
213                 sbi->ll_cache = NULL;
214         }
215         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
216 out_pcc:
217         pcc_super_fini(&sbi->ll_pcc_super);
218 out_sbi:
219         OBD_FREE_PTR(sbi);
220         RETURN(ERR_PTR(rc));
221 }
222
223 static void ll_free_sbi(struct super_block *sb)
224 {
225         struct ll_sb_info *sbi = ll_s2sbi(sb);
226         ENTRY;
227
228         if (sbi != NULL) {
229                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
230                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
231                 if (sbi->ll_ra_info.ll_readahead_wq)
232                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
233                 if (sbi->ll_cache != NULL) {
234                         cl_cache_decref(sbi->ll_cache);
235                         sbi->ll_cache = NULL;
236                 }
237                 if (sbi->ll_foreign_symlink_prefix) {
238                         OBD_FREE(sbi->ll_foreign_symlink_prefix,
239                                  sbi->ll_foreign_symlink_prefix_size);
240                         sbi->ll_foreign_symlink_prefix = NULL;
241                 }
242                 if (sbi->ll_foreign_symlink_upcall) {
243                         OBD_FREE(sbi->ll_foreign_symlink_upcall,
244                                  strlen(sbi->ll_foreign_symlink_upcall) +
245                                        1);
246                         sbi->ll_foreign_symlink_upcall = NULL;
247                 }
248                 if (sbi->ll_foreign_symlink_upcall_items) {
249                         int i;
250                         int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
251                         struct ll_foreign_symlink_upcall_item *items =
252                                 sbi->ll_foreign_symlink_upcall_items;
253
254                         for (i = 0 ; i < nb_items; i++)
255                                 if (items[i].type == STRING_TYPE)
256                                         OBD_FREE(items[i].string,
257                                                        items[i].size);
258
259                         OBD_FREE_LARGE(items, nb_items *
260                                 sizeof(struct ll_foreign_symlink_upcall_item));
261                         sbi->ll_foreign_symlink_upcall_items = NULL;
262                 }
263                 pcc_super_fini(&sbi->ll_pcc_super);
264                 OBD_FREE(sbi, sizeof(*sbi));
265         }
266         EXIT;
267 }
268
269 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
270 {
271         struct inode *root = NULL;
272         struct ll_sb_info *sbi = ll_s2sbi(sb);
273         struct obd_statfs *osfs = NULL;
274         struct ptlrpc_request *request = NULL;
275         struct obd_connect_data *data = NULL;
276         struct obd_uuid *uuid;
277         struct md_op_data *op_data;
278         struct lustre_md lmd;
279         u64 valid;
280         int size, err, checksum;
281
282         ENTRY;
283         sbi->ll_md_obd = class_name2obd(md);
284         if (!sbi->ll_md_obd) {
285                 CERROR("MD %s: not setup or attached\n", md);
286                 RETURN(-EINVAL);
287         }
288
289         OBD_ALLOC_PTR(data);
290         if (data == NULL)
291                 RETURN(-ENOMEM);
292
293         OBD_ALLOC_PTR(osfs);
294         if (osfs == NULL) {
295                 OBD_FREE_PTR(data);
296                 RETURN(-ENOMEM);
297         }
298
299         /* pass client page size via ocd_grant_blkbits, the server should report
300          * back its backend blocksize for grant calculation purpose */
301         data->ocd_grant_blkbits = PAGE_SHIFT;
302
303         /* indicate MDT features supported by this client */
304         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
305                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
306                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
307                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
308                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
309                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
310                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
311                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
312                                   OBD_CONNECT_64BITHASH |
313                                   OBD_CONNECT_EINPROGRESS |
314                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
315                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
316                                   OBD_CONNECT_MAX_EASIZE |
317                                   OBD_CONNECT_FLOCK_DEAD |
318                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
319                                   OBD_CONNECT_OPEN_BY_FID |
320                                   OBD_CONNECT_DIR_STRIPE |
321                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
322                                   OBD_CONNECT_SUBTREE |
323                                   OBD_CONNECT_MULTIMODRPCS |
324                                   OBD_CONNECT_GRANT_PARAM |
325                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
326
327         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
328                                    OBD_CONNECT2_SUM_STATFS |
329                                    OBD_CONNECT2_OVERSTRIPING |
330                                    OBD_CONNECT2_FLR |
331                                    OBD_CONNECT2_LOCK_CONVERT |
332                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
333                                    OBD_CONNECT2_INC_XID |
334                                    OBD_CONNECT2_LSOM |
335                                    OBD_CONNECT2_ASYNC_DISCARD |
336                                    OBD_CONNECT2_PCC |
337                                    OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
338                                    OBD_CONNECT2_GETATTR_PFID |
339                                    OBD_CONNECT2_DOM_LVB |
340                                    OBD_CONNECT2_REP_MBITS |
341                                    OBD_CONNECT2_ATOMIC_OPEN_LOCK;
342
343 #ifdef HAVE_LRU_RESIZE_SUPPORT
344         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
345                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
346 #endif
347         data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
348
349         data->ocd_cksum_types = obd_cksum_types_supported_client();
350
351         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
352                 /* flag mdc connection as lightweight, only used for test
353                  * purpose, use with care */
354                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
355
356         data->ocd_ibits_known = MDS_INODELOCK_FULL;
357         data->ocd_version = LUSTRE_VERSION_CODE;
358
359         if (sb->s_flags & SB_RDONLY)
360                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
361         if (sbi->ll_flags & LL_SBI_USER_XATTR)
362                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
363
364 #ifdef SB_NOSEC
365         /* Setting this indicates we correctly support S_NOSEC (See kernel
366          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
367          */
368         sb->s_flags |= SB_NOSEC;
369 #endif
370         sbi->ll_fop = ll_select_file_operations(sbi);
371
372         /* always ping even if server suppress_pings */
373         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
374                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
375
376         obd_connect_set_secctx(data);
377         if (ll_sbi_has_encrypt(sbi))
378                 obd_connect_set_enc(data);
379
380 #if defined(CONFIG_SECURITY)
381         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
382 #endif
383
384         data->ocd_brw_size = MD_MAX_BRW_SIZE;
385
386         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
387                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
388         if (err == -EBUSY) {
389                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
390                                    "recovery, of which this client is not a "
391                                    "part. Please wait for recovery to complete,"
392                                    " abort, or time out.\n", md);
393                 GOTO(out, err);
394         } else if (err) {
395                 CERROR("cannot connect to %s: rc = %d\n", md, err);
396                 GOTO(out, err);
397         }
398
399         sbi->ll_md_exp->exp_connect_data = *data;
400
401         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
402                            LUSTRE_SEQ_METADATA);
403         if (err) {
404                 CERROR("%s: Can't init metadata layer FID infrastructure, "
405                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
406                 GOTO(out_md, err);
407         }
408
409         /* For mount, we only need fs info from MDT0, and also in DNE, it
410          * can make sure the client can be mounted as long as MDT0 is
411          * avaible */
412         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
413                         ktime_get_seconds() - sbi->ll_statfs_max_age,
414                         OBD_STATFS_FOR_MDT0);
415         if (err)
416                 GOTO(out_md_fid, err);
417
418         /* This needs to be after statfs to ensure connect has finished.
419          * Note that "data" does NOT contain the valid connect reply.
420          * If connecting to a 1.8 server there will be no LMV device, so
421          * we can access the MDC export directly and exp_connect_flags will
422          * be non-zero, but if accessing an upgraded 2.1 server it will
423          * have the correct flags filled in.
424          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
425         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
426         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
427             valid != CLIENT_CONNECT_MDT_REQD) {
428                 char *buf;
429
430                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
431                 obd_connect_flags2str(buf, PAGE_SIZE,
432                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
433                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
434                                    "feature(s) needed for correct operation "
435                                    "of this client (%s). Please upgrade "
436                                    "server or downgrade client.\n",
437                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
438                 OBD_FREE(buf, PAGE_SIZE);
439                 GOTO(out_md_fid, err = -EPROTO);
440         }
441
442         size = sizeof(*data);
443         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
444                            KEY_CONN_DATA,  &size, data);
445         if (err) {
446                 CERROR("%s: Get connect data failed: rc = %d\n",
447                        sbi->ll_md_exp->exp_obd->obd_name, err);
448                 GOTO(out_md_fid, err);
449         }
450
451         LASSERT(osfs->os_bsize);
452         sb->s_blocksize = osfs->os_bsize;
453         sb->s_blocksize_bits = log2(osfs->os_bsize);
454         sb->s_magic = LL_SUPER_MAGIC;
455         sb->s_maxbytes = MAX_LFS_FILESIZE;
456         sbi->ll_namelen = osfs->os_namelen;
457         sbi->ll_mnt.mnt = current->fs->root.mnt;
458
459         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
460             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
461                 LCONSOLE_INFO("Disabling user_xattr feature because "
462                               "it is not supported on the server\n");
463                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
464         }
465
466         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
467 #ifdef SB_POSIXACL
468                 sb->s_flags |= SB_POSIXACL;
469 #endif
470                 sbi->ll_flags |= LL_SBI_ACL;
471         } else {
472                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
473 #ifdef SB_POSIXACL
474                 sb->s_flags &= ~SB_POSIXACL;
475 #endif
476                 sbi->ll_flags &= ~LL_SBI_ACL;
477         }
478
479         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
480                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
481
482         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
483                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
484
485         if (obd_connect_has_secctx(data))
486                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
487
488         if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
489                 if (ll_sbi_has_test_dummy_encryption(sbi))
490                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
491                                       sbi->ll_fsname,
492                                       sbi->ll_md_exp->exp_obd->obd_name);
493                 ll_sbi_set_encrypt(sbi, false);
494         }
495
496         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
497                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
498                         LCONSOLE_INFO("%s: disabling xattr cache due to "
499                                       "unknown maximum xattr size.\n", dt);
500                 } else if (!sbi->ll_xattr_cache_set) {
501                         /* If xattr_cache is already set (no matter 0 or 1)
502                          * during processing llog, it won't be enabled here. */
503                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
504                         sbi->ll_xattr_cache_enabled = 1;
505                 }
506         }
507
508         sbi->ll_dt_obd = class_name2obd(dt);
509         if (!sbi->ll_dt_obd) {
510                 CERROR("DT %s: not setup or attached\n", dt);
511                 GOTO(out_md_fid, err = -ENODEV);
512         }
513
514         /* pass client page size via ocd_grant_blkbits, the server should report
515          * back its backend blocksize for grant calculation purpose */
516         data->ocd_grant_blkbits = PAGE_SHIFT;
517
518         /* indicate OST features supported by this client */
519         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
520                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
521                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
522                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
523                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
524                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
525                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
526                                   OBD_CONNECT_EINPROGRESS |
527                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
528                                   OBD_CONNECT_LAYOUTLOCK |
529                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
530                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
531                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
532         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
533                                    OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
534                                    OBD_CONNECT2_REP_MBITS;
535
536         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
537                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
538
539         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
540          * disabled by default, because it can still be enabled on the
541          * fly via /sys. As a consequence, we still need to come to an
542          * agreement on the supported algorithms at connect time
543          */
544         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
545
546         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
547                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
548         else
549                 data->ocd_cksum_types = obd_cksum_types_supported_client();
550
551 #ifdef HAVE_LRU_RESIZE_SUPPORT
552         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
553 #endif
554         /* always ping even if server suppress_pings */
555         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
556                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
557
558         if (ll_sbi_has_encrypt(sbi))
559                 obd_connect_set_enc(data);
560
561         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
562                "ocd_grant: %d\n", data->ocd_connect_flags,
563                data->ocd_version, data->ocd_grant);
564
565         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
566         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
567
568         data->ocd_brw_size = DT_MAX_BRW_SIZE;
569
570         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
571                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
572         if (err == -EBUSY) {
573                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
574                                    "recovery, of which this client is not a "
575                                    "part.  Please wait for recovery to "
576                                    "complete, abort, or time out.\n", dt);
577                 GOTO(out_md, err);
578         } else if (err) {
579                 CERROR("%s: Cannot connect to %s: rc = %d\n",
580                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
581                 GOTO(out_md, err);
582         }
583
584         if (ll_sbi_has_encrypt(sbi) &&
585             !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
586                 if (ll_sbi_has_test_dummy_encryption(sbi))
587                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
588                                       sbi->ll_fsname, dt);
589                 ll_sbi_set_encrypt(sbi, false);
590         } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
591                 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
592         }
593
594         sbi->ll_dt_exp->exp_connect_data = *data;
595
596         /* Don't change value if it was specified in the config log */
597         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
598                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
599                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
600                               (data->ocd_brw_size >> PAGE_SHIFT));
601                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
602                     sbi->ll_ra_info.ra_max_pages_per_file)
603                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
604                                 sbi->ll_ra_info.ra_max_pages_per_file;
605         }
606
607         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
608                            LUSTRE_SEQ_METADATA);
609         if (err) {
610                 CERROR("%s: Can't init data layer FID infrastructure, "
611                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
612                 GOTO(out_dt, err);
613         }
614
615         mutex_lock(&sbi->ll_lco.lco_lock);
616         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
617         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
618         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
619         mutex_unlock(&sbi->ll_lco.lco_lock);
620
621         fid_zero(&sbi->ll_root_fid);
622         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
623                            &sbi->ll_root_fid);
624         if (err) {
625                 CERROR("cannot mds_connect: rc = %d\n", err);
626                 GOTO(out_lock_cn_cb, err);
627         }
628         if (!fid_is_sane(&sbi->ll_root_fid)) {
629                 CERROR("%s: Invalid root fid "DFID" during mount\n",
630                        sbi->ll_md_exp->exp_obd->obd_name,
631                        PFID(&sbi->ll_root_fid));
632                 GOTO(out_lock_cn_cb, err = -EINVAL);
633         }
634         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
635
636         sb->s_op = &lustre_super_operations;
637         sb->s_xattr = ll_xattr_handlers;
638 #if THREAD_SIZE >= 8192 /*b=17630*/
639         sb->s_export_op = &lustre_export_operations;
640 #endif
641 #ifdef HAVE_LUSTRE_CRYPTO
642         llcrypt_set_ops(sb, &lustre_cryptops);
643 #endif
644
645         /* make root inode
646          * XXX: move this to after cbd setup? */
647         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
648         if (sbi->ll_flags & LL_SBI_ACL)
649                 valid |= OBD_MD_FLACL;
650
651         OBD_ALLOC_PTR(op_data);
652         if (op_data == NULL)
653                 GOTO(out_lock_cn_cb, err = -ENOMEM);
654
655         op_data->op_fid1 = sbi->ll_root_fid;
656         op_data->op_mode = 0;
657         op_data->op_valid = valid;
658
659         err = md_getattr(sbi->ll_md_exp, op_data, &request);
660
661         OBD_FREE_PTR(op_data);
662         if (err) {
663                 CERROR("%s: md_getattr failed for root: rc = %d\n",
664                        sbi->ll_md_exp->exp_obd->obd_name, err);
665                 GOTO(out_lock_cn_cb, err);
666         }
667
668         err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
669                                sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
670         if (err) {
671                 CERROR("failed to understand root inode md: rc = %d\n", err);
672                 ptlrpc_req_finished(request);
673                 GOTO(out_lock_cn_cb, err);
674         }
675
676         LASSERT(fid_is_sane(&sbi->ll_root_fid));
677         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
678                                             sbi->ll_flags & LL_SBI_32BIT_API),
679                        &lmd);
680         md_free_lustre_md(sbi->ll_md_exp, &lmd);
681         ptlrpc_req_finished(request);
682
683         if (IS_ERR(root)) {
684                 lmd_clear_acl(&lmd);
685                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
686                 root = NULL;
687                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
688                        sbi->ll_fsname, err);
689                 GOTO(out_root, err);
690         }
691
692         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
693         if (sbi->ll_checksum_set) {
694                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
695                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
696                                          sizeof(checksum), &checksum, NULL);
697                 if (err) {
698                         CERROR("%s: Set checksum failed: rc = %d\n",
699                                sbi->ll_dt_exp->exp_obd->obd_name, err);
700                         GOTO(out_root, err);
701                 }
702         }
703         cl_sb_init(sb);
704
705         sb->s_root = d_make_root(root);
706         if (sb->s_root == NULL) {
707                 err = -ENOMEM;
708                 CERROR("%s: can't make root dentry: rc = %d\n",
709                        sbi->ll_fsname, err);
710                 GOTO(out_root, err);
711         }
712
713         sbi->ll_sdev_orig = sb->s_dev;
714
715         /* We set sb->s_dev equal on all lustre clients in order to support
716          * NFS export clustering.  NFSD requires that the FSID be the same
717          * on all clients. */
718         /* s_dev is also used in lt_compare() to compare two fs, but that is
719          * only a node-local comparison. */
720         uuid = obd_get_uuid(sbi->ll_md_exp);
721         if (uuid != NULL)
722                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
723
724         if (data != NULL)
725                 OBD_FREE_PTR(data);
726         if (osfs != NULL)
727                 OBD_FREE_PTR(osfs);
728
729         if (sbi->ll_dt_obd) {
730                 err = sysfs_create_link(&sbi->ll_kset.kobj,
731                                         &sbi->ll_dt_obd->obd_kset.kobj,
732                                         sbi->ll_dt_obd->obd_type->typ_name);
733                 if (err < 0) {
734                         CERROR("%s: could not register %s in llite: rc = %d\n",
735                                dt, sbi->ll_fsname, err);
736                         err = 0;
737                 }
738         }
739
740         if (sbi->ll_md_obd) {
741                 err = sysfs_create_link(&sbi->ll_kset.kobj,
742                                         &sbi->ll_md_obd->obd_kset.kobj,
743                                         sbi->ll_md_obd->obd_type->typ_name);
744                 if (err < 0) {
745                         CERROR("%s: could not register %s in llite: rc = %d\n",
746                                md, sbi->ll_fsname, err);
747                         err = 0;
748                 }
749         }
750
751         RETURN(err);
752 out_root:
753         iput(root);
754 out_lock_cn_cb:
755         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
756 out_dt:
757         obd_disconnect(sbi->ll_dt_exp);
758         sbi->ll_dt_exp = NULL;
759         sbi->ll_dt_obd = NULL;
760 out_md_fid:
761         obd_fid_fini(sbi->ll_md_exp->exp_obd);
762 out_md:
763         obd_disconnect(sbi->ll_md_exp);
764         sbi->ll_md_exp = NULL;
765         sbi->ll_md_obd = NULL;
766 out:
767         if (data != NULL)
768                 OBD_FREE_PTR(data);
769         if (osfs != NULL)
770                 OBD_FREE_PTR(osfs);
771         return err;
772 }
773
774 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
775 {
776         int size, rc;
777
778         size = sizeof(*lmmsize);
779         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
780                           KEY_MAX_EASIZE, &size, lmmsize);
781         if (rc != 0) {
782                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
783                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
784                 RETURN(rc);
785         }
786
787         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
788
789         size = sizeof(int);
790         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
791                           KEY_MAX_EASIZE, &size, lmmsize);
792         if (rc)
793                 CERROR("Get max mdsize error rc %d\n", rc);
794
795         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
796
797         RETURN(rc);
798 }
799
800 /**
801  * Get the value of the default_easize parameter.
802  *
803  * \see client_obd::cl_default_mds_easize
804  *
805  * \param[in] sbi       superblock info for this filesystem
806  * \param[out] lmmsize  pointer to storage location for value
807  *
808  * \retval 0            on success
809  * \retval negative     negated errno on failure
810  */
811 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
812 {
813         int size, rc;
814
815         size = sizeof(int);
816         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
817                          KEY_DEFAULT_EASIZE, &size, lmmsize);
818         if (rc)
819                 CERROR("Get default mdsize error rc %d\n", rc);
820
821         RETURN(rc);
822 }
823
824 /**
825  * Set the default_easize parameter to the given value.
826  *
827  * \see client_obd::cl_default_mds_easize
828  *
829  * \param[in] sbi       superblock info for this filesystem
830  * \param[in] lmmsize   the size to set
831  *
832  * \retval 0            on success
833  * \retval negative     negated errno on failure
834  */
835 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
836 {
837         int rc;
838
839         if (lmmsize < sizeof(struct lov_mds_md) ||
840             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
841                 return -EINVAL;
842
843         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
844                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
845                                 sizeof(int), &lmmsize, NULL);
846
847         RETURN(rc);
848 }
849
850 static void client_common_put_super(struct super_block *sb)
851 {
852         struct ll_sb_info *sbi = ll_s2sbi(sb);
853         ENTRY;
854
855         cl_sb_fini(sb);
856
857         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
858         obd_disconnect(sbi->ll_dt_exp);
859         sbi->ll_dt_exp = NULL;
860
861         ll_debugfs_unregister_super(sb);
862
863         obd_fid_fini(sbi->ll_md_exp->exp_obd);
864         obd_disconnect(sbi->ll_md_exp);
865         sbi->ll_md_exp = NULL;
866
867         EXIT;
868 }
869
870 void ll_kill_super(struct super_block *sb)
871 {
872         struct ll_sb_info *sbi;
873         ENTRY;
874
875         /* not init sb ?*/
876         if (!(sb->s_flags & SB_ACTIVE))
877                 return;
878
879         sbi = ll_s2sbi(sb);
880         /* we need restore s_dev from changed for clustred NFS before put_super
881          * because new kernels have cached s_dev and change sb->s_dev in
882          * put_super not affected real removing devices */
883         if (sbi) {
884                 sb->s_dev = sbi->ll_sdev_orig;
885
886                 /* wait running statahead threads to quit */
887                 while (atomic_read(&sbi->ll_sa_running) > 0)
888                         schedule_timeout_uninterruptible(
889                                 cfs_time_seconds(1) >> 3);
890         }
891
892         EXIT;
893 }
894
895 static inline int ll_set_opt(const char *opt, char *data, int fl)
896 {
897         if (strncmp(opt, data, strlen(opt)) != 0)
898                 return 0;
899         else
900                 return fl;
901 }
902
903 /* non-client-specific mount options are parsed in lmd_parse */
904 static int ll_options(char *options, struct ll_sb_info *sbi)
905 {
906         int tmp;
907         char *s1 = options, *s2;
908         int *flags = &sbi->ll_flags;
909         ENTRY;
910
911         if (!options)
912                 RETURN(0);
913
914         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
915
916         while (*s1) {
917                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
918                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
919                 if (tmp) {
920                         *flags |= tmp;
921                         goto next;
922                 }
923                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
924                 if (tmp) {
925                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
926                         goto next;
927                 }
928                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
929                 if (tmp) {
930                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
931                         goto next;
932                 }
933                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
934                 if (tmp) {
935                         *flags &= ~tmp;
936                         goto next;
937                 }
938                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
939                 if (tmp) {
940                         *flags |= tmp;
941                         goto next;
942                 }
943                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
944                 if (tmp) {
945                         *flags &= ~tmp;
946                         goto next;
947                 }
948                 tmp = ll_set_opt("context", s1, 1);
949                 if (tmp)
950                         goto next;
951                 tmp = ll_set_opt("fscontext", s1, 1);
952                 if (tmp)
953                         goto next;
954                 tmp = ll_set_opt("defcontext", s1, 1);
955                 if (tmp)
956                         goto next;
957                 tmp = ll_set_opt("rootcontext", s1, 1);
958                 if (tmp)
959                         goto next;
960                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
961                 if (tmp) {
962                         *flags |= tmp;
963                         goto next;
964                 }
965                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
966                 if (tmp) {
967                         *flags &= ~tmp;
968                         goto next;
969                 }
970
971                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
972                 if (tmp) {
973                         *flags |= tmp;
974                         sbi->ll_checksum_set = 1;
975                         goto next;
976                 }
977                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
978                 if (tmp) {
979                         *flags &= ~tmp;
980                         sbi->ll_checksum_set = 1;
981                         goto next;
982                 }
983                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
984                 if (tmp) {
985                         *flags |= tmp;
986                         goto next;
987                 }
988                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
989                 if (tmp) {
990                         *flags &= ~tmp;
991                         goto next;
992                 }
993                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
994                 if (tmp) {
995                         *flags |= tmp;
996                         goto next;
997                 }
998                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
999                 if (tmp) {
1000                         *flags &= ~tmp;
1001                         goto next;
1002                 }
1003                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
1004                 if (tmp) {
1005                         *flags |= tmp;
1006                         goto next;
1007                 }
1008                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
1009                 if (tmp) {
1010                         *flags |= tmp;
1011                         goto next;
1012                 }
1013                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
1014                 if (tmp) {
1015                         *flags &= ~tmp;
1016                         goto next;
1017                 }
1018                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
1019                 if (tmp) {
1020                         *flags |= tmp;
1021                         goto next;
1022                 }
1023                 tmp = ll_set_opt("test_dummy_encryption", s1,
1024                                  LL_SBI_TEST_DUMMY_ENCRYPTION);
1025                 if (tmp) {
1026 #ifdef HAVE_LUSTRE_CRYPTO
1027                         *flags |= tmp;
1028 #else
1029                         LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1030 #endif
1031                         goto next;
1032                 }
1033                 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
1034                 if (tmp) {
1035 #ifdef HAVE_LUSTRE_CRYPTO
1036                         *flags &= ~tmp;
1037 #else
1038                         LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
1039 #endif
1040                         goto next;
1041                 }
1042                 tmp = ll_set_opt("foreign_symlink", s1, LL_SBI_FOREIGN_SYMLINK);
1043                 if (tmp) {
1044                         int prefix_pos = sizeof("foreign_symlink=") - 1;
1045                         int equal_pos = sizeof("foreign_symlink=") - 2;
1046
1047                         /* non-default prefix provided ? */
1048                         if (strlen(s1) >= sizeof("foreign_symlink=") &&
1049                             *(s1 + equal_pos) == '=') {
1050                                 char *old = sbi->ll_foreign_symlink_prefix;
1051                                 size_t old_len =
1052                                         sbi->ll_foreign_symlink_prefix_size;
1053
1054                                 /* path must be absolute */
1055                                 if (*(s1 + sizeof("foreign_symlink=")
1056                                       - 1) != '/') {
1057                                         LCONSOLE_ERROR_MSG(0x152,
1058                                                 "foreign prefix '%s' must be an absolute path\n",
1059                                                 s1 + prefix_pos);
1060                                         RETURN(-EINVAL);
1061                                 }
1062                                 /* last option ? */
1063                                 s2 = strchrnul(s1 + prefix_pos, ',');
1064
1065                                 if (sbi->ll_foreign_symlink_prefix) {
1066                                         sbi->ll_foreign_symlink_prefix = NULL;
1067                                         sbi->ll_foreign_symlink_prefix_size = 0;
1068                                 }
1069                                 /* alloc for path length and '\0' */
1070                                 OBD_ALLOC(sbi->ll_foreign_symlink_prefix,
1071                                                 s2 - (s1 + prefix_pos) + 1);
1072                                 if (!sbi->ll_foreign_symlink_prefix) {
1073                                         /* restore previous */
1074                                         sbi->ll_foreign_symlink_prefix = old;
1075                                         sbi->ll_foreign_symlink_prefix_size =
1076                                                 old_len;
1077                                         RETURN(-ENOMEM);
1078                                 }
1079                                 if (old)
1080                                         OBD_FREE(old, old_len);
1081                                 strncpy(sbi->ll_foreign_symlink_prefix,
1082                                         s1 + prefix_pos,
1083                                         s2 - (s1 + prefix_pos));
1084                                 sbi->ll_foreign_symlink_prefix_size =
1085                                         s2 - (s1 + prefix_pos) + 1;
1086                         } else {
1087                                 LCONSOLE_ERROR_MSG(0x152,
1088                                                    "invalid %s option\n", s1);
1089                         }
1090                         /* enable foreign symlink support */
1091                         *flags |= tmp;
1092                         goto next;
1093                 }
1094                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1095                                    s1);
1096                 RETURN(-EINVAL);
1097
1098 next:
1099                 /* Find next opt */
1100                 s2 = strchr(s1, ',');
1101                 if (s2 == NULL)
1102                         break;
1103                 s1 = s2 + 1;
1104         }
1105         RETURN(0);
1106 }
1107
1108 void ll_lli_init(struct ll_inode_info *lli)
1109 {
1110         lli->lli_inode_magic = LLI_INODE_MAGIC;
1111         lli->lli_flags = 0;
1112         spin_lock_init(&lli->lli_lock);
1113         lli->lli_posix_acl = NULL;
1114         /* Do not set lli_fid, it has been initialized already. */
1115         fid_zero(&lli->lli_pfid);
1116         lli->lli_mds_read_och = NULL;
1117         lli->lli_mds_write_och = NULL;
1118         lli->lli_mds_exec_och = NULL;
1119         lli->lli_open_fd_read_count = 0;
1120         lli->lli_open_fd_write_count = 0;
1121         lli->lli_open_fd_exec_count = 0;
1122         mutex_init(&lli->lli_och_mutex);
1123         spin_lock_init(&lli->lli_agl_lock);
1124         spin_lock_init(&lli->lli_layout_lock);
1125         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1126         lli->lli_clob = NULL;
1127
1128         init_rwsem(&lli->lli_xattrs_list_rwsem);
1129         mutex_init(&lli->lli_xattrs_enq_lock);
1130
1131         LASSERT(lli->lli_vfs_inode.i_mode != 0);
1132         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1133                 lli->lli_opendir_key = NULL;
1134                 lli->lli_sai = NULL;
1135                 spin_lock_init(&lli->lli_sa_lock);
1136                 lli->lli_opendir_pid = 0;
1137                 lli->lli_sa_enabled = 0;
1138                 init_rwsem(&lli->lli_lsm_sem);
1139         } else {
1140                 mutex_init(&lli->lli_size_mutex);
1141                 mutex_init(&lli->lli_setattr_mutex);
1142                 lli->lli_symlink_name = NULL;
1143                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1144                 range_lock_tree_init(&lli->lli_write_tree);
1145                 init_rwsem(&lli->lli_glimpse_sem);
1146                 lli->lli_glimpse_time = ktime_set(0, 0);
1147                 INIT_LIST_HEAD(&lli->lli_agl_list);
1148                 lli->lli_agl_index = 0;
1149                 lli->lli_async_rc = 0;
1150                 spin_lock_init(&lli->lli_heat_lock);
1151                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1152                 lli->lli_heat_flags = 0;
1153                 mutex_init(&lli->lli_pcc_lock);
1154                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1155                 lli->lli_pcc_inode = NULL;
1156                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1157                 lli->lli_pcc_generation = 0;
1158                 mutex_init(&lli->lli_group_mutex);
1159                 lli->lli_group_users = 0;
1160                 lli->lli_group_gid = 0;
1161         }
1162         mutex_init(&lli->lli_layout_mutex);
1163         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1164 }
1165
1166 #define MAX_STRING_SIZE 128
1167
1168 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1169
1170 #define LSI_BDI_INITIALIZED     0x00400000
1171
1172 #ifndef HAVE_BDI_CAP_MAP_COPY
1173 # define BDI_CAP_MAP_COPY       0
1174 #endif
1175
1176 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1177 {
1178         struct  lustre_sb_info *lsi = s2lsi(sb);
1179         char buf[MAX_STRING_SIZE];
1180         va_list args;
1181         int err;
1182
1183         err = bdi_init(&lsi->lsi_bdi);
1184         if (err)
1185                 return err;
1186
1187         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1188         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1189         lsi->lsi_bdi.name = "lustre";
1190         va_start(args, fmt);
1191         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1192         va_end(args);
1193         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1194         va_end(args);
1195         if (!err)
1196                 sb->s_bdi = &lsi->lsi_bdi;
1197
1198         return err;
1199 }
1200 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1201
1202 int ll_fill_super(struct super_block *sb)
1203 {
1204         struct  lustre_profile *lprof = NULL;
1205         struct  lustre_sb_info *lsi = s2lsi(sb);
1206         struct  ll_sb_info *sbi = NULL;
1207         char    *dt = NULL, *md = NULL;
1208         char    *profilenm = get_profile_name(sb);
1209         struct config_llog_instance *cfg;
1210         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1211         const int instlen = LUSTRE_MAXINSTANCE + 2;
1212         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1213         char name[MAX_STRING_SIZE];
1214         int md_len = 0;
1215         int dt_len = 0;
1216         uuid_t uuid;
1217         char *ptr;
1218         int len;
1219         int err;
1220
1221         ENTRY;
1222         /* for ASLR, to map between cfg_instance and hashed ptr */
1223         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1224                profilenm, cfg_instance, sb);
1225
1226         OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1227
1228         OBD_ALLOC_PTR(cfg);
1229         if (cfg == NULL)
1230                 GOTO(out_free_cfg, err = -ENOMEM);
1231
1232         /* client additional sb info */
1233         lsi->lsi_llsbi = sbi = ll_init_sbi();
1234         if (IS_ERR(sbi))
1235                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1236
1237         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1238         if (err)
1239                 GOTO(out_free_cfg, err);
1240
1241         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1242         sb->s_d_op = &ll_d_ops;
1243
1244         /* UUID handling */
1245         generate_random_uuid(uuid.b);
1246         snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1247
1248         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1249
1250         /* Get fsname */
1251         len = strlen(profilenm);
1252         ptr = strrchr(profilenm, '-');
1253         if (ptr && (strcmp(ptr, "-client") == 0))
1254                 len -= 7;
1255
1256         if (len > LUSTRE_MAXFSNAME) {
1257                 if (unlikely(len >= MAX_STRING_SIZE))
1258                         len = MAX_STRING_SIZE - 1;
1259                 strncpy(name, profilenm, len);
1260                 name[len] = '\0';
1261                 err = -ENAMETOOLONG;
1262                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1263                        name, LUSTRE_MAXFSNAME, err);
1264                 GOTO(out_free_cfg, err);
1265         }
1266         strncpy(sbi->ll_fsname, profilenm, len);
1267         sbi->ll_fsname[len] = '\0';
1268
1269         /* Mount info */
1270         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1271                  profilenm, cfg_instance);
1272
1273         err = super_setup_bdi_name(sb, "%s", name);
1274         if (err)
1275                 GOTO(out_free_cfg, err);
1276
1277         /* Call ll_debugfs_register_super() before lustre_process_log()
1278          * so that "llite.*.*" params can be processed correctly.
1279          */
1280         err = ll_debugfs_register_super(sb, name);
1281         if (err < 0) {
1282                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1283                        sbi->ll_fsname, err);
1284                 err = 0;
1285         }
1286
1287         /* The cfg_instance is a value unique to this super, in case some
1288          * joker tries to mount the same fs at two mount points.
1289          */
1290         cfg->cfg_instance = cfg_instance;
1291         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1292         cfg->cfg_callback = class_config_llog_handler;
1293         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1294         /* set up client obds */
1295         err = lustre_process_log(sb, profilenm, cfg);
1296         if (err < 0)
1297                 GOTO(out_debugfs, err);
1298
1299         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1300         lprof = class_get_profile(profilenm);
1301         if (lprof == NULL) {
1302                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1303                                    " read from the MGS.  Does that filesystem "
1304                                    "exist?\n", profilenm);
1305                 GOTO(out_debugfs, err = -EINVAL);
1306         }
1307         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1308                lprof->lp_md, lprof->lp_dt);
1309
1310         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1311         OBD_ALLOC(dt, dt_len);
1312         if (!dt)
1313                 GOTO(out_profile, err = -ENOMEM);
1314         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1315
1316         md_len = strlen(lprof->lp_md) + instlen + 2;
1317         OBD_ALLOC(md, md_len);
1318         if (!md)
1319                 GOTO(out_free_dt, err = -ENOMEM);
1320         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1321
1322         /* connections, registrations, sb setup */
1323         err = client_common_fill_super(sb, md, dt);
1324         if (err < 0)
1325                 GOTO(out_free_md, err);
1326
1327         sbi->ll_client_common_fill_super_succeeded = 1;
1328
1329 out_free_md:
1330         if (md)
1331                 OBD_FREE(md, md_len);
1332 out_free_dt:
1333         if (dt)
1334                 OBD_FREE(dt, dt_len);
1335 out_profile:
1336         if (lprof)
1337                 class_put_profile(lprof);
1338 out_debugfs:
1339         if (err < 0)
1340                 ll_debugfs_unregister_super(sb);
1341 out_free_cfg:
1342         if (cfg)
1343                 OBD_FREE_PTR(cfg);
1344
1345         if (err)
1346                 ll_put_super(sb);
1347         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1348                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1349         RETURN(err);
1350 } /* ll_fill_super */
1351
1352 void ll_put_super(struct super_block *sb)
1353 {
1354         struct config_llog_instance cfg, params_cfg;
1355         struct obd_device *obd;
1356         struct lustre_sb_info *lsi = s2lsi(sb);
1357         struct ll_sb_info *sbi = ll_s2sbi(sb);
1358         char *profilenm = get_profile_name(sb);
1359         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1360         long ccc_count;
1361         int next, force = 1, rc = 0;
1362         ENTRY;
1363
1364         if (IS_ERR(sbi))
1365                 GOTO(out_no_sbi, 0);
1366
1367         /* Should replace instance_id with something better for ASLR */
1368         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1369                profilenm, cfg_instance, sb);
1370
1371         cfg.cfg_instance = cfg_instance;
1372         lustre_end_log(sb, profilenm, &cfg);
1373
1374         params_cfg.cfg_instance = cfg_instance;
1375         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1376
1377         if (sbi->ll_md_exp) {
1378                 obd = class_exp2obd(sbi->ll_md_exp);
1379                 if (obd)
1380                         force = obd->obd_force;
1381         }
1382
1383         /* Wait for unstable pages to be committed to stable storage */
1384         if (force == 0) {
1385                 rc = l_wait_event_abortable(
1386                         sbi->ll_cache->ccc_unstable_waitq,
1387                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1388         }
1389
1390         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1391         if (force == 0 && rc != -ERESTARTSYS)
1392                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1393
1394         /* We need to set force before the lov_disconnect in
1395          * lustre_common_put_super, since l_d cleans up osc's as well.
1396          */
1397         if (force) {
1398                 next = 0;
1399                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1400                                                      &next)) != NULL) {
1401                         obd->obd_force = force;
1402                 }
1403         }
1404
1405         if (sbi->ll_client_common_fill_super_succeeded) {
1406                 /* Only if client_common_fill_super succeeded */
1407                 client_common_put_super(sb);
1408         }
1409
1410         next = 0;
1411         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1412                 class_manual_cleanup(obd);
1413
1414         if (sbi->ll_flags & LL_SBI_VERBOSE)
1415                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1416
1417         if (profilenm)
1418                 class_del_profile(profilenm);
1419
1420 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1421         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1422                 bdi_destroy(&lsi->lsi_bdi);
1423                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1424         }
1425 #endif
1426
1427         ll_free_sbi(sb);
1428         lsi->lsi_llsbi = NULL;
1429 out_no_sbi:
1430         lustre_common_put_super(sb);
1431
1432         cl_env_cache_purge(~0);
1433
1434         EXIT;
1435 } /* client_put_super */
1436
1437 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1438 {
1439         struct inode *inode = NULL;
1440
1441         /* NOTE: we depend on atomic igrab() -bzzz */
1442         lock_res_and_lock(lock);
1443         if (lock->l_resource->lr_lvb_inode) {
1444                 struct ll_inode_info * lli;
1445                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1446                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1447                         inode = igrab(lock->l_resource->lr_lvb_inode);
1448                 } else {
1449                         inode = lock->l_resource->lr_lvb_inode;
1450                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1451                                          D_WARNING, lock, "lr_lvb_inode %p is "
1452                                          "bogus: magic %08x",
1453                                          lock->l_resource->lr_lvb_inode,
1454                                          lli->lli_inode_magic);
1455                         inode = NULL;
1456                 }
1457         }
1458         unlock_res_and_lock(lock);
1459         return inode;
1460 }
1461
1462 void ll_dir_clear_lsm_md(struct inode *inode)
1463 {
1464         struct ll_inode_info *lli = ll_i2info(inode);
1465
1466         LASSERT(S_ISDIR(inode->i_mode));
1467
1468         if (lli->lli_lsm_md) {
1469                 lmv_free_memmd(lli->lli_lsm_md);
1470                 lli->lli_lsm_md = NULL;
1471         }
1472
1473         if (lli->lli_default_lsm_md) {
1474                 lmv_free_memmd(lli->lli_default_lsm_md);
1475                 lli->lli_default_lsm_md = NULL;
1476         }
1477 }
1478
1479 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1480                                       const struct lu_fid *fid,
1481                                       struct lustre_md *md)
1482 {
1483         struct ll_sb_info *sbi = ll_s2sbi(sb);
1484         struct ll_inode_info *lli;
1485         struct mdt_body *body = md->body;
1486         struct inode *inode;
1487         ino_t ino;
1488
1489         ENTRY;
1490
1491         LASSERT(md->lmv);
1492         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1493         inode = iget_locked(sb, ino);
1494         if (inode == NULL) {
1495                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1496                        sbi->ll_fsname, PFID(fid));
1497                 RETURN(ERR_PTR(-ENOENT));
1498         }
1499
1500         lli = ll_i2info(inode);
1501         if (inode->i_state & I_NEW) {
1502                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1503                                 (body->mbo_mode & S_IFMT);
1504                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1505                          PFID(fid));
1506
1507                 inode->i_mtime.tv_sec = 0;
1508                 inode->i_atime.tv_sec = 0;
1509                 inode->i_ctime.tv_sec = 0;
1510                 inode->i_rdev = 0;
1511
1512 #ifdef HAVE_BACKING_DEV_INFO
1513                 /* initializing backing dev info. */
1514                 inode->i_mapping->backing_dev_info =
1515                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1516 #endif
1517                 inode->i_op = &ll_dir_inode_operations;
1518                 inode->i_fop = &ll_dir_operations;
1519                 lli->lli_fid = *fid;
1520                 ll_lli_init(lli);
1521
1522                 /* master object FID */
1523                 lli->lli_pfid = body->mbo_fid1;
1524                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1525                        lli, PFID(fid), PFID(&lli->lli_pfid));
1526                 unlock_new_inode(inode);
1527         } else {
1528                 /* in directory restripe/auto-split, a directory will be
1529                  * transformed to a stripe if it's plain, set its pfid here,
1530                  * otherwise ll_lock_cancel_bits() can't find the master inode.
1531                  */
1532                 lli->lli_pfid = body->mbo_fid1;
1533         }
1534
1535         RETURN(inode);
1536 }
1537
1538 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1539 {
1540         struct lu_fid *fid;
1541         struct lmv_stripe_md *lsm = md->lmv;
1542         struct ll_inode_info *lli = ll_i2info(inode);
1543         int i;
1544
1545         LASSERT(lsm != NULL);
1546
1547         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1548                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1549         lsm_md_dump(D_INODE, lsm);
1550
1551         if (!lmv_dir_striped(lsm))
1552                 goto out;
1553
1554         /* XXX sigh, this lsm_root initialization should be in
1555          * LMV layer, but it needs ll_iget right now, so we
1556          * put this here right now. */
1557         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1558                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1559                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1560
1561                 if (!fid_is_sane(fid))
1562                         continue;
1563
1564                 /* Unfortunately ll_iget will call ll_update_inode,
1565                  * where the initialization of slave inode is slightly
1566                  * different, so it reset lsm_md to NULL to avoid
1567                  * initializing lsm for slave inode. */
1568                 lsm->lsm_md_oinfo[i].lmo_root =
1569                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1570                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1571                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1572
1573                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1574                         while (i-- > 0) {
1575                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1576                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1577                         }
1578                         return rc;
1579                 }
1580         }
1581 out:
1582         lli->lli_lsm_md = lsm;
1583
1584         return 0;
1585 }
1586
1587 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1588 {
1589         struct ll_inode_info *lli = ll_i2info(inode);
1590
1591         if (!md->default_lmv) {
1592                 /* clear default lsm */
1593                 if (lli->lli_default_lsm_md) {
1594                         down_write(&lli->lli_lsm_sem);
1595                         if (lli->lli_default_lsm_md) {
1596                                 lmv_free_memmd(lli->lli_default_lsm_md);
1597                                 lli->lli_default_lsm_md = NULL;
1598                         }
1599                         up_write(&lli->lli_lsm_sem);
1600                 }
1601                 return;
1602         }
1603
1604         if (lli->lli_default_lsm_md) {
1605                 /* do nonthing if default lsm isn't changed */
1606                 down_read(&lli->lli_lsm_sem);
1607                 if (lli->lli_default_lsm_md &&
1608                     lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1609                         up_read(&lli->lli_lsm_sem);
1610                         return;
1611                 }
1612                 up_read(&lli->lli_lsm_sem);
1613         }
1614
1615         down_write(&lli->lli_lsm_sem);
1616         if (lli->lli_default_lsm_md)
1617                 lmv_free_memmd(lli->lli_default_lsm_md);
1618         lli->lli_default_lsm_md = md->default_lmv;
1619         lsm_md_dump(D_INODE, md->default_lmv);
1620         md->default_lmv = NULL;
1621         up_write(&lli->lli_lsm_sem);
1622 }
1623
1624 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1625 {
1626         struct ll_inode_info *lli = ll_i2info(inode);
1627         struct lmv_stripe_md *lsm = md->lmv;
1628         struct cl_attr  *attr;
1629         int rc = 0;
1630
1631         ENTRY;
1632
1633         LASSERT(S_ISDIR(inode->i_mode));
1634         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1635                PFID(ll_inode2fid(inode)));
1636
1637         /* update default LMV */
1638         if (md->default_lmv)
1639                 ll_update_default_lsm_md(inode, md);
1640
1641         /* after dir migration/restripe, a stripe may be turned into a
1642          * directory, in this case, zero out its lli_pfid.
1643          */
1644         if (unlikely(fid_is_norm(&lli->lli_pfid)))
1645                 fid_zero(&lli->lli_pfid);
1646
1647         /*
1648          * no striped information from request, lustre_md from req does not
1649          * include stripeEA, see ll_md_setattr()
1650          */
1651         if (!lsm)
1652                 RETURN(0);
1653
1654         /*
1655          * normally dir layout doesn't change, only take read lock to check
1656          * that to avoid blocking other MD operations.
1657          */
1658         down_read(&lli->lli_lsm_sem);
1659
1660         /* some current lookup initialized lsm, and unchanged */
1661         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1662                 GOTO(unlock, rc = 0);
1663
1664         /* if dir layout doesn't match, check whether version is increased,
1665          * which means layout is changed, this happens in dir split/merge and
1666          * lfsck.
1667          *
1668          * foreign LMV should not change.
1669          */
1670         if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1671             lsm->lsm_md_layout_version <=
1672             lli->lli_lsm_md->lsm_md_layout_version) {
1673                 CERROR("%s: "DFID" dir layout mismatch:\n",
1674                        ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1675                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1676                 lsm_md_dump(D_ERROR, lsm);
1677                 GOTO(unlock, rc = -EINVAL);
1678         }
1679
1680         up_read(&lli->lli_lsm_sem);
1681         down_write(&lli->lli_lsm_sem);
1682         /* clear existing lsm */
1683         if (lli->lli_lsm_md) {
1684                 lmv_free_memmd(lli->lli_lsm_md);
1685                 lli->lli_lsm_md = NULL;
1686         }
1687
1688         rc = ll_init_lsm_md(inode, md);
1689         up_write(&lli->lli_lsm_sem);
1690
1691         if (rc)
1692                 RETURN(rc);
1693
1694         /* set md->lmv to NULL, so the following free lustre_md will not free
1695          * this lsm.
1696          */
1697         md->lmv = NULL;
1698
1699         /* md_merge_attr() may take long, since lsm is already set, switch to
1700          * read lock.
1701          */
1702         down_read(&lli->lli_lsm_sem);
1703
1704         if (!lmv_dir_striped(lli->lli_lsm_md))
1705                 GOTO(unlock, rc = 0);
1706
1707         OBD_ALLOC_PTR(attr);
1708         if (!attr)
1709                 GOTO(unlock, rc = -ENOMEM);
1710
1711         /* validate the lsm */
1712         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1713                            ll_md_blocking_ast);
1714         if (!rc) {
1715                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1716                         md->body->mbo_nlink = attr->cat_nlink;
1717                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1718                         md->body->mbo_size = attr->cat_size;
1719                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1720                         md->body->mbo_atime = attr->cat_atime;
1721                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1722                         md->body->mbo_ctime = attr->cat_ctime;
1723                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1724                         md->body->mbo_mtime = attr->cat_mtime;
1725         }
1726
1727         OBD_FREE_PTR(attr);
1728         GOTO(unlock, rc);
1729 unlock:
1730         up_read(&lli->lli_lsm_sem);
1731
1732         return rc;
1733 }
1734
1735 void ll_clear_inode(struct inode *inode)
1736 {
1737         struct ll_inode_info *lli = ll_i2info(inode);
1738         struct ll_sb_info *sbi = ll_i2sbi(inode);
1739
1740         ENTRY;
1741
1742         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1743                PFID(ll_inode2fid(inode)), inode);
1744
1745         if (S_ISDIR(inode->i_mode)) {
1746                 /* these should have been cleared in ll_file_release */
1747                 LASSERT(lli->lli_opendir_key == NULL);
1748                 LASSERT(lli->lli_sai == NULL);
1749                 LASSERT(lli->lli_opendir_pid == 0);
1750         } else {
1751                 pcc_inode_free(inode);
1752         }
1753
1754         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1755
1756         LASSERT(!lli->lli_open_fd_write_count);
1757         LASSERT(!lli->lli_open_fd_read_count);
1758         LASSERT(!lli->lli_open_fd_exec_count);
1759
1760         if (lli->lli_mds_write_och)
1761                 ll_md_real_close(inode, FMODE_WRITE);
1762         if (lli->lli_mds_exec_och)
1763                 ll_md_real_close(inode, FMODE_EXEC);
1764         if (lli->lli_mds_read_och)
1765                 ll_md_real_close(inode, FMODE_READ);
1766
1767         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1768                 OBD_FREE(lli->lli_symlink_name,
1769                          strlen(lli->lli_symlink_name) + 1);
1770                 lli->lli_symlink_name = NULL;
1771         }
1772
1773         ll_xattr_cache_destroy(inode);
1774
1775         forget_all_cached_acls(inode);
1776         lli_clear_acl(lli);
1777         lli->lli_inode_magic = LLI_INODE_DEAD;
1778
1779         if (S_ISDIR(inode->i_mode))
1780                 ll_dir_clear_lsm_md(inode);
1781         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1782                 LASSERT(list_empty(&lli->lli_agl_list));
1783
1784         /*
1785          * XXX This has to be done before lsm is freed below, because
1786          * cl_object still uses inode lsm.
1787          */
1788         cl_inode_fini(inode);
1789
1790         llcrypt_put_encryption_info(inode);
1791
1792         EXIT;
1793 }
1794
1795 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1796 {
1797         struct lustre_md md;
1798         struct inode *inode = dentry->d_inode;
1799         struct ll_sb_info *sbi = ll_i2sbi(inode);
1800         struct ptlrpc_request *request = NULL;
1801         int rc, ia_valid;
1802         ENTRY;
1803
1804         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1805                                      LUSTRE_OPC_ANY, NULL);
1806         if (IS_ERR(op_data))
1807                 RETURN(PTR_ERR(op_data));
1808
1809         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1810         if (rc) {
1811                 ptlrpc_req_finished(request);
1812                 if (rc == -ENOENT) {
1813                         clear_nlink(inode);
1814                         /* Unlinked special device node? Or just a race?
1815                          * Pretend we done everything. */
1816                         if (!S_ISREG(inode->i_mode) &&
1817                             !S_ISDIR(inode->i_mode)) {
1818                                 ia_valid = op_data->op_attr.ia_valid;
1819                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1820                                 rc = simple_setattr(dentry, &op_data->op_attr);
1821                                 op_data->op_attr.ia_valid = ia_valid;
1822                         }
1823                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1824                         CERROR("md_setattr fails: rc = %d\n", rc);
1825                 }
1826                 RETURN(rc);
1827         }
1828
1829         rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1830                               sbi->ll_md_exp, &md);
1831         if (rc) {
1832                 ptlrpc_req_finished(request);
1833                 RETURN(rc);
1834         }
1835
1836         ia_valid = op_data->op_attr.ia_valid;
1837         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1838          * cache is not cleared yet. */
1839         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1840         if (S_ISREG(inode->i_mode))
1841                 inode_lock(inode);
1842         rc = simple_setattr(dentry, &op_data->op_attr);
1843         if (S_ISREG(inode->i_mode))
1844                 inode_unlock(inode);
1845         op_data->op_attr.ia_valid = ia_valid;
1846
1847         rc = ll_update_inode(inode, &md);
1848         ptlrpc_req_finished(request);
1849
1850         RETURN(rc);
1851 }
1852
1853 /**
1854  * Zero portion of page that is part of @inode.
1855  * This implies, if necessary:
1856  * - taking cl_lock on range corresponding to concerned page
1857  * - grabbing vm page
1858  * - associating cl_page
1859  * - proceeding to clio read
1860  * - zeroing range in page
1861  * - proceeding to cl_page flush
1862  * - releasing cl_lock
1863  *
1864  * \param[in] inode     inode
1865  * \param[in] index     page index
1866  * \param[in] offset    offset in page to start zero from
1867  * \param[in] len       len to zero
1868  *
1869  * \retval 0            on success
1870  * \retval negative     errno on failure
1871  */
1872 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1873                     unsigned len)
1874 {
1875         struct ll_inode_info *lli = ll_i2info(inode);
1876         struct cl_object *clob = lli->lli_clob;
1877         __u16 refcheck;
1878         struct lu_env *env = NULL;
1879         struct cl_io *io = NULL;
1880         struct cl_page *clpage = NULL;
1881         struct page *vmpage = NULL;
1882         unsigned from = index << PAGE_SHIFT;
1883         struct cl_lock *lock = NULL;
1884         struct cl_lock_descr *descr = NULL;
1885         struct cl_2queue *queue = NULL;
1886         struct cl_sync_io *anchor = NULL;
1887         bool holdinglock = false;
1888         bool lockedbymyself = true;
1889         int rc;
1890
1891         ENTRY;
1892
1893         env = cl_env_get(&refcheck);
1894         if (IS_ERR(env))
1895                 RETURN(PTR_ERR(env));
1896
1897         io = vvp_env_thread_io(env);
1898         io->ci_obj = clob;
1899         rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1900         if (rc)
1901                 GOTO(putenv, rc);
1902
1903         lock = vvp_env_lock(env);
1904         descr = &lock->cll_descr;
1905         descr->cld_obj   = io->ci_obj;
1906         descr->cld_start = cl_index(io->ci_obj, from);
1907         descr->cld_end   = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1908         descr->cld_mode  = CLM_WRITE;
1909         descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1910
1911         /* request lock for page */
1912         rc = cl_lock_request(env, io, lock);
1913         /* -ECANCELED indicates a matching lock with a different extent
1914          * was already present, and -EEXIST indicates a matching lock
1915          * on exactly the same extent was already present.
1916          * In both cases it means we are covered.
1917          */
1918         if (rc == -ECANCELED || rc == -EEXIST)
1919                 rc = 0;
1920         else if (rc < 0)
1921                 GOTO(iofini, rc);
1922         else
1923                 holdinglock = true;
1924
1925         /* grab page */
1926         vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1927         if (vmpage == NULL)
1928                 GOTO(rellock, rc = -EOPNOTSUPP);
1929
1930         if (!PageDirty(vmpage)) {
1931                 /* associate cl_page */
1932                 clpage = cl_page_find(env, clob, vmpage->index,
1933                                       vmpage, CPT_CACHEABLE);
1934                 if (IS_ERR(clpage))
1935                         GOTO(pagefini, rc = PTR_ERR(clpage));
1936
1937                 cl_page_assume(env, io, clpage);
1938         }
1939
1940         if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1941             !PageWriteback(vmpage)) {
1942                 /* read page */
1943                 /* set PagePrivate2 to detect special case of empty page
1944                  * in osc_brw_fini_request()
1945                  */
1946                 SetPagePrivate2(vmpage);
1947                 rc = ll_io_read_page(env, io, clpage, NULL);
1948                 if (!PagePrivate2(vmpage))
1949                         /* PagePrivate2 was cleared in osc_brw_fini_request()
1950                          * meaning we read an empty page. In this case, in order
1951                          * to avoid allocating unnecessary block in truncated
1952                          * file, we must not zero and write as below. Subsequent
1953                          * server-side truncate will handle things correctly.
1954                          */
1955                         GOTO(clpfini, rc = 0);
1956                 ClearPagePrivate2(vmpage);
1957                 if (rc)
1958                         GOTO(clpfini, rc);
1959                 lockedbymyself = trylock_page(vmpage);
1960                 cl_page_assume(env, io, clpage);
1961         }
1962
1963         /* zero range in page */
1964         zero_user(vmpage, offset, len);
1965
1966         if (holdinglock && clpage) {
1967                 /* explicitly write newly modified page */
1968                 queue = &io->ci_queue;
1969                 cl_2queue_init(queue);
1970                 anchor = &vvp_env_info(env)->vti_anchor;
1971                 cl_sync_io_init(anchor, 1);
1972                 clpage->cp_sync_io = anchor;
1973                 cl_2queue_add(queue, clpage, true);
1974                 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1975                 if (rc)
1976                         GOTO(queuefini1, rc);
1977                 rc = cl_sync_io_wait(env, anchor, 0);
1978                 if (rc)
1979                         GOTO(queuefini2, rc);
1980                 cl_page_assume(env, io, clpage);
1981
1982 queuefini2:
1983                 cl_2queue_discard(env, io, queue);
1984 queuefini1:
1985                 cl_2queue_disown(env, io, queue);
1986                 cl_2queue_fini(env, queue);
1987         }
1988
1989 clpfini:
1990         if (clpage)
1991                 cl_page_put(env, clpage);
1992 pagefini:
1993         if (lockedbymyself) {
1994                 unlock_page(vmpage);
1995                 put_page(vmpage);
1996         }
1997 rellock:
1998         if (holdinglock)
1999                 cl_lock_release(env, lock);
2000 iofini:
2001         cl_io_fini(env, io);
2002 putenv:
2003         if (env)
2004                 cl_env_put(env, &refcheck);
2005
2006         RETURN(rc);
2007 }
2008
2009 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2010  * object(s) determine the file size and mtime.  Otherwise, the MDS will
2011  * keep these values until such a time that objects are allocated for it.
2012  * We do the MDS operations first, as it is checking permissions for us.
2013  * We don't to the MDS RPC if there is nothing that we want to store there,
2014  * otherwise there is no harm in updating mtime/atime on the MDS if we are
2015  * going to do an RPC anyways.
2016  *
2017  * If we are doing a truncate, we will send the mtime and ctime updates
2018  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2019  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2020  * at the same time.
2021  *
2022  * In case of HSMimport, we only set attr on MDS.
2023  */
2024 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2025                    enum op_xvalid xvalid, bool hsm_import)
2026 {
2027         struct inode *inode = dentry->d_inode;
2028         struct ll_inode_info *lli = ll_i2info(inode);
2029         struct md_op_data *op_data = NULL;
2030         ktime_t kstart = ktime_get();
2031         int rc = 0;
2032
2033         ENTRY;
2034
2035         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2036                "valid %x, hsm_import %d\n",
2037                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2038                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2039                hsm_import);
2040
2041         if (attr->ia_valid & ATTR_SIZE) {
2042                 /* Check new size against VFS/VM file size limit and rlimit */
2043                 rc = inode_newsize_ok(inode, attr->ia_size);
2044                 if (rc)
2045                         RETURN(rc);
2046
2047                 /* The maximum Lustre file size is variable, based on the
2048                  * OST maximum object size and number of stripes.  This
2049                  * needs another check in addition to the VFS check above. */
2050                 if (attr->ia_size > ll_file_maxbytes(inode)) {
2051                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2052                                PFID(&lli->lli_fid), attr->ia_size,
2053                                ll_file_maxbytes(inode));
2054                         RETURN(-EFBIG);
2055                 }
2056
2057                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2058         }
2059
2060         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2061         if (attr->ia_valid & TIMES_SET_FLAGS) {
2062                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2063                     !capable(CAP_FOWNER))
2064                         RETURN(-EPERM);
2065         }
2066
2067         /* We mark all of the fields "set" so MDS/OST does not re-set them */
2068         if (!(xvalid & OP_XVALID_CTIME_SET) &&
2069              (attr->ia_valid & ATTR_CTIME)) {
2070                 attr->ia_ctime = current_time(inode);
2071                 xvalid |= OP_XVALID_CTIME_SET;
2072         }
2073         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2074             (attr->ia_valid & ATTR_ATIME)) {
2075                 attr->ia_atime = current_time(inode);
2076                 attr->ia_valid |= ATTR_ATIME_SET;
2077         }
2078         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2079             (attr->ia_valid & ATTR_MTIME)) {
2080                 attr->ia_mtime = current_time(inode);
2081                 attr->ia_valid |= ATTR_MTIME_SET;
2082         }
2083
2084         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2085                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2086                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2087                        ktime_get_real_seconds());
2088
2089         if (S_ISREG(inode->i_mode))
2090                 inode_unlock(inode);
2091
2092         /* We always do an MDS RPC, even if we're only changing the size;
2093          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2094
2095         OBD_ALLOC_PTR(op_data);
2096         if (op_data == NULL)
2097                 GOTO(out, rc = -ENOMEM);
2098
2099         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2100                 /* If we are changing file size, file content is
2101                  * modified, flag it.
2102                  */
2103                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2104                 op_data->op_bias |= MDS_DATA_MODIFIED;
2105                 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2106         }
2107
2108         if (attr->ia_valid & ATTR_FILE) {
2109                 struct ll_file_data *fd = attr->ia_file->private_data;
2110
2111                 if (fd->fd_lease_och)
2112                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2113         }
2114
2115         op_data->op_attr = *attr;
2116         op_data->op_xvalid = xvalid;
2117
2118         rc = ll_md_setattr(dentry, op_data);
2119         if (rc)
2120                 GOTO(out, rc);
2121
2122         if (!S_ISREG(inode->i_mode) || hsm_import)
2123                 GOTO(out, rc = 0);
2124
2125         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2126                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2127             xvalid & OP_XVALID_CTIME_SET) {
2128                 bool cached = false;
2129
2130                 rc = pcc_inode_setattr(inode, attr, &cached);
2131                 if (cached) {
2132                         if (rc) {
2133                                 CERROR("%s: PCC inode "DFID" setattr failed: "
2134                                        "rc = %d\n",
2135                                        ll_i2sbi(inode)->ll_fsname,
2136                                        PFID(&lli->lli_fid), rc);
2137                                 GOTO(out, rc);
2138                         }
2139                 } else {
2140                         unsigned int flags = 0;
2141
2142                         /* For truncate and utimes sending attributes to OSTs,
2143                          * setting mtime/atime to the past will be performed
2144                          * under PW [0:EOF] extent lock (new_size:EOF for
2145                          * truncate). It may seem excessive to send mtime/atime
2146                          * updates to OSTs when not setting times to past, but
2147                          * it is necessary due to possible time
2148                          * de-synchronization between MDT inode and OST objects
2149                          */
2150                         if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2151                                 xvalid |= OP_XVALID_FLAGS;
2152                                 flags = LUSTRE_ENCRYPT_FL;
2153                                 /* Call to ll_io_zero_page is not necessary if
2154                                  * truncating on PAGE_SIZE boundary, because
2155                                  * whole pages will be wiped.
2156                                  * In case of Direct IO, all we need is to set
2157                                  * new size.
2158                                  */
2159                                 if (attr->ia_valid & ATTR_SIZE &&
2160                                     attr->ia_size & ~PAGE_MASK &&
2161                                     !(attr->ia_valid & ATTR_FILE &&
2162                                       attr->ia_file->f_flags & O_DIRECT)) {
2163                                         pgoff_t offset =
2164                                                 attr->ia_size & (PAGE_SIZE - 1);
2165
2166                                         rc = ll_io_zero_page(inode,
2167                                                     attr->ia_size >> PAGE_SHIFT,
2168                                                     offset, PAGE_SIZE - offset);
2169                                         if (rc)
2170                                                 GOTO(out, rc);
2171                                 }
2172                         }
2173                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2174                 }
2175         }
2176
2177         /* If the file was restored, it needs to set dirty flag.
2178          *
2179          * We've already sent MDS_DATA_MODIFIED flag in
2180          * ll_md_setattr() for truncate. However, the MDT refuses to
2181          * set the HS_DIRTY flag on released files, so we have to set
2182          * it again if the file has been restored. Please check how
2183          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2184          *
2185          * Please notice that if the file is not released, the previous
2186          * MDS_DATA_MODIFIED has taken effect and usually
2187          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2188          * This way we can save an RPC for common open + trunc
2189          * operation. */
2190         if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2191                 struct hsm_state_set hss = {
2192                         .hss_valid = HSS_SETMASK,
2193                         .hss_setmask = HS_DIRTY,
2194                 };
2195                 int rc2;
2196
2197                 rc2 = ll_hsm_state_set(inode, &hss);
2198                 /* truncate and write can happen at the same time, so that
2199                  * the file can be set modified even though the file is not
2200                  * restored from released state, and ll_hsm_state_set() is
2201                  * not applicable for the file, and rc2 < 0 is normal in this
2202                  * case. */
2203                 if (rc2 < 0)
2204                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2205                                PFID(ll_inode2fid(inode)), rc2);
2206         }
2207
2208         EXIT;
2209 out:
2210         if (op_data != NULL)
2211                 ll_finish_md_op_data(op_data);
2212
2213         if (S_ISREG(inode->i_mode)) {
2214                 inode_lock(inode);
2215                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2216                         inode_dio_wait(inode);
2217                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2218                  * flag.  ll_update_inode (called from ll_md_setattr), clears
2219                  * inode flags, so there is a gap where S_NOSEC is not set.
2220                  * This can cause a writer to take the i_mutex unnecessarily,
2221                  * but this is safe to do and should be rare. */
2222                 inode_has_no_xattr(inode);
2223         }
2224
2225         if (!rc)
2226                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2227                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2228                                    ktime_us_delta(ktime_get(), kstart));
2229
2230         return rc;
2231 }
2232
2233 int ll_setattr(struct dentry *de, struct iattr *attr)
2234 {
2235         int mode = de->d_inode->i_mode;
2236         enum op_xvalid xvalid = 0;
2237         int rc;
2238
2239         rc = llcrypt_prepare_setattr(de, attr);
2240         if (rc)
2241                 return rc;
2242
2243         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2244                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2245                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2246
2247         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2248                                (ATTR_SIZE|ATTR_MODE)) &&
2249             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2250              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2251               !(attr->ia_mode & S_ISGID))))
2252                 attr->ia_valid |= ATTR_FORCE;
2253
2254         if ((attr->ia_valid & ATTR_MODE) &&
2255             (mode & S_ISUID) &&
2256             !(attr->ia_mode & S_ISUID) &&
2257             !(attr->ia_valid & ATTR_KILL_SUID))
2258                 attr->ia_valid |= ATTR_KILL_SUID;
2259
2260         if ((attr->ia_valid & ATTR_MODE) &&
2261             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2262             !(attr->ia_mode & S_ISGID) &&
2263             !(attr->ia_valid & ATTR_KILL_SGID))
2264                 attr->ia_valid |= ATTR_KILL_SGID;
2265
2266         return ll_setattr_raw(de, attr, xvalid, false);
2267 }
2268
2269 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2270                        u32 flags)
2271 {
2272         struct obd_statfs obd_osfs = { 0 };
2273         time64_t max_age;
2274         int rc;
2275
2276         ENTRY;
2277         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2278
2279         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2280                 flags |= OBD_STATFS_NODELAY;
2281
2282         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2283         if (rc)
2284                 RETURN(rc);
2285
2286         osfs->os_type = LL_SUPER_MAGIC;
2287
2288         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2289               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2290
2291         if (osfs->os_state & OS_STATFS_SUM)
2292                 GOTO(out, rc);
2293
2294         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2295         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
2296                 GOTO(out, rc = 0);
2297
2298         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2299                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2300                obd_osfs.os_files);
2301
2302         osfs->os_bsize = obd_osfs.os_bsize;
2303         osfs->os_blocks = obd_osfs.os_blocks;
2304         osfs->os_bfree = obd_osfs.os_bfree;
2305         osfs->os_bavail = obd_osfs.os_bavail;
2306
2307         /* If we have _some_ OSTs, but don't have as many free objects on the
2308          * OSTs as inodes on the MDTs, reduce the reported number of inodes
2309          * to compensate, so that the "inodes in use" number is correct.
2310          * This should be kept in sync with lod_statfs() behaviour.
2311          */
2312         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2313                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2314                                  obd_osfs.os_ffree;
2315                 osfs->os_ffree = obd_osfs.os_ffree;
2316         }
2317
2318 out:
2319         RETURN(rc);
2320 }
2321
2322 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2323 {
2324         struct if_quotactl qctl = {
2325                 .qc_cmd = LUSTRE_Q_GETQUOTA,
2326                 .qc_type = PRJQUOTA,
2327                 .qc_valid = QC_GENERAL,
2328         };
2329         u64 limit, curblock;
2330         int ret;
2331
2332         qctl.qc_id = ll_i2info(inode)->lli_projid;
2333         ret = quotactl_ioctl(ll_i2sbi(inode), &qctl);
2334         if (ret) {
2335                 /* ignore errors if project ID does not have
2336                  * a quota limit or feature unsupported.
2337                  */
2338                 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2339                         ret = 0;
2340                 return ret;
2341         }
2342
2343         limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2344                  qctl.qc_dqblk.dqb_bsoftlimit :
2345                  qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2346         if (limit && sfs->f_blocks > limit) {
2347                 curblock = (qctl.qc_dqblk.dqb_curspace +
2348                                 sfs->f_bsize - 1) / sfs->f_bsize;
2349                 sfs->f_blocks = limit;
2350                 sfs->f_bfree = sfs->f_bavail =
2351                         (sfs->f_blocks > curblock) ?
2352                         (sfs->f_blocks - curblock) : 0;
2353         }
2354
2355         limit = qctl.qc_dqblk.dqb_isoftlimit ?
2356                 qctl.qc_dqblk.dqb_isoftlimit :
2357                 qctl.qc_dqblk.dqb_ihardlimit;
2358         if (limit && sfs->f_files > limit) {
2359                 sfs->f_files = limit;
2360                 sfs->f_ffree = (sfs->f_files >
2361                         qctl.qc_dqblk.dqb_curinodes) ?
2362                         (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2363         }
2364
2365         return 0;
2366 }
2367
2368 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2369 {
2370         struct super_block *sb = de->d_sb;
2371         struct obd_statfs osfs;
2372         __u64 fsid = huge_encode_dev(sb->s_dev);
2373         ktime_t kstart = ktime_get();
2374         int rc;
2375
2376         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2377
2378         /* Some amount of caching on the client is allowed */
2379         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2380         if (rc)
2381                 return rc;
2382
2383         statfs_unpack(sfs, &osfs);
2384
2385         /* We need to downshift for all 32-bit kernels, because we can't
2386          * tell if the kernel is being called via sys_statfs64() or not.
2387          * Stop before overflowing f_bsize - in which case it is better
2388          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2389         if (sizeof(long) < 8) {
2390                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2391                         sfs->f_bsize <<= 1;
2392
2393                         osfs.os_blocks >>= 1;
2394                         osfs.os_bfree >>= 1;
2395                         osfs.os_bavail >>= 1;
2396                 }
2397         }
2398
2399         sfs->f_blocks = osfs.os_blocks;
2400         sfs->f_bfree = osfs.os_bfree;
2401         sfs->f_bavail = osfs.os_bavail;
2402         sfs->f_fsid.val[0] = (__u32)fsid;
2403         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2404         if (ll_i2info(de->d_inode)->lli_projid)
2405                 return ll_statfs_project(de->d_inode, sfs);
2406
2407         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2408                            ktime_us_delta(ktime_get(), kstart));
2409
2410         return 0;
2411 }
2412
2413 void ll_inode_size_lock(struct inode *inode)
2414 {
2415         struct ll_inode_info *lli;
2416
2417         LASSERT(!S_ISDIR(inode->i_mode));
2418
2419         lli = ll_i2info(inode);
2420         mutex_lock(&lli->lli_size_mutex);
2421 }
2422
2423 void ll_inode_size_unlock(struct inode *inode)
2424 {
2425         struct ll_inode_info *lli;
2426
2427         lli = ll_i2info(inode);
2428         mutex_unlock(&lli->lli_size_mutex);
2429 }
2430
2431 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2432 {
2433         /* do not clear encryption flag */
2434         ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2435         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2436         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2437                 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2438         else
2439                 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2440 }
2441
2442 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2443 {
2444         struct ll_inode_info *lli = ll_i2info(inode);
2445         struct mdt_body *body = md->body;
2446         struct ll_sb_info *sbi = ll_i2sbi(inode);
2447         int rc = 0;
2448
2449         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2450                 rc = cl_file_inode_init(inode, md);
2451                 if (rc)
2452                         return rc;
2453         }
2454
2455         if (S_ISDIR(inode->i_mode)) {
2456                 rc = ll_update_lsm_md(inode, md);
2457                 if (rc != 0)
2458                         return rc;
2459         }
2460
2461         if (body->mbo_valid & OBD_MD_FLACL)
2462                 lli_replace_acl(lli, md);
2463
2464         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2465                                         sbi->ll_flags & LL_SBI_32BIT_API);
2466         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2467
2468         if (body->mbo_valid & OBD_MD_FLATIME) {
2469                 if (body->mbo_atime > inode->i_atime.tv_sec)
2470                         inode->i_atime.tv_sec = body->mbo_atime;
2471                 lli->lli_atime = body->mbo_atime;
2472         }
2473
2474         if (body->mbo_valid & OBD_MD_FLMTIME) {
2475                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2476                         CDEBUG(D_INODE,
2477                                "setting ino %lu mtime from %lld to %llu\n",
2478                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2479                                body->mbo_mtime);
2480                         inode->i_mtime.tv_sec = body->mbo_mtime;
2481                 }
2482                 lli->lli_mtime = body->mbo_mtime;
2483         }
2484
2485         if (body->mbo_valid & OBD_MD_FLCTIME) {
2486                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2487                         inode->i_ctime.tv_sec = body->mbo_ctime;
2488                 lli->lli_ctime = body->mbo_ctime;
2489         }
2490
2491         if (body->mbo_valid & OBD_MD_FLBTIME)
2492                 lli->lli_btime = body->mbo_btime;
2493
2494         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2495         if (body->mbo_valid & OBD_MD_FLFLAGS)
2496                 ll_update_inode_flags(inode, body->mbo_flags);
2497         if (body->mbo_valid & OBD_MD_FLMODE)
2498                 inode->i_mode = (inode->i_mode & S_IFMT) |
2499                                 (body->mbo_mode & ~S_IFMT);
2500
2501         if (body->mbo_valid & OBD_MD_FLTYPE)
2502                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2503                                 (body->mbo_mode & S_IFMT);
2504
2505         LASSERT(inode->i_mode != 0);
2506         if (body->mbo_valid & OBD_MD_FLUID)
2507                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2508         if (body->mbo_valid & OBD_MD_FLGID)
2509                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2510         if (body->mbo_valid & OBD_MD_FLPROJID)
2511                 lli->lli_projid = body->mbo_projid;
2512         if (body->mbo_valid & OBD_MD_FLNLINK)
2513                 set_nlink(inode, body->mbo_nlink);
2514         if (body->mbo_valid & OBD_MD_FLRDEV)
2515                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2516
2517         if (body->mbo_valid & OBD_MD_FLID) {
2518                 /* FID shouldn't be changed! */
2519                 if (fid_is_sane(&lli->lli_fid)) {
2520                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2521                                  "Trying to change FID "DFID
2522                                  " to the "DFID", inode "DFID"(%p)\n",
2523                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2524                                  PFID(ll_inode2fid(inode)), inode);
2525                 } else {
2526                         lli->lli_fid = body->mbo_fid1;
2527                 }
2528         }
2529
2530         LASSERT(fid_seq(&lli->lli_fid) != 0);
2531
2532         lli->lli_attr_valid = body->mbo_valid;
2533         if (body->mbo_valid & OBD_MD_FLSIZE) {
2534                 i_size_write(inode, body->mbo_size);
2535
2536                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2537                        PFID(ll_inode2fid(inode)),
2538                        (unsigned long long)body->mbo_size);
2539
2540                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2541                         inode->i_blocks = body->mbo_blocks;
2542         } else {
2543                 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2544                         lli->lli_lazysize = body->mbo_size;
2545                 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2546                         lli->lli_lazyblocks = body->mbo_blocks;
2547         }
2548
2549         if (body->mbo_valid & OBD_MD_TSTATE) {
2550                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2551                  * clear it when done to ensure to start again
2552                  * glimpsing updated attrs
2553                  */
2554                 if (body->mbo_t_state & MS_RESTORE)
2555                         set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2556                 else
2557                         clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2558         }
2559
2560         return 0;
2561 }
2562
2563 void ll_truncate_inode_pages_final(struct inode *inode)
2564 {
2565         struct address_space *mapping = &inode->i_data;
2566         unsigned long nrpages;
2567         unsigned long flags;
2568
2569         truncate_inode_pages_final(mapping);
2570
2571         /* Workaround for LU-118: Note nrpages may not be totally updated when
2572          * truncate_inode_pages() returns, as there can be a page in the process
2573          * of deletion (inside __delete_from_page_cache()) in the specified
2574          * range. Thus mapping->nrpages can be non-zero when this function
2575          * returns even after truncation of the whole mapping.  Only do this if
2576          * npages isn't already zero.
2577          */
2578         nrpages = mapping->nrpages;
2579         if (nrpages) {
2580                 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2581                 nrpages = mapping->nrpages;
2582                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2583         } /* Workaround end */
2584
2585         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2586                  "see https://jira.whamcloud.com/browse/LU-118\n",
2587                  ll_i2sbi(inode)->ll_fsname,
2588                  PFID(ll_inode2fid(inode)), inode, nrpages);
2589 }
2590
2591 int ll_read_inode2(struct inode *inode, void *opaque)
2592 {
2593         struct lustre_md *md = opaque;
2594         struct ll_inode_info *lli = ll_i2info(inode);
2595         int     rc;
2596         ENTRY;
2597
2598         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2599                PFID(&lli->lli_fid), inode);
2600
2601         /* Core attributes from the MDS first.  This is a new inode, and
2602          * the VFS doesn't zero times in the core inode so we have to do
2603          * it ourselves.  They will be overwritten by either MDS or OST
2604          * attributes - we just need to make sure they aren't newer.
2605          */
2606         inode->i_mtime.tv_sec = 0;
2607         inode->i_atime.tv_sec = 0;
2608         inode->i_ctime.tv_sec = 0;
2609         inode->i_rdev = 0;
2610         rc = ll_update_inode(inode, md);
2611         if (rc != 0)
2612                 RETURN(rc);
2613
2614         /* OIDEBUG(inode); */
2615
2616 #ifdef HAVE_BACKING_DEV_INFO
2617         /* initializing backing dev info. */
2618         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2619 #endif
2620         if (S_ISREG(inode->i_mode)) {
2621                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2622                 inode->i_op = &ll_file_inode_operations;
2623                 inode->i_fop = sbi->ll_fop;
2624                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2625                 EXIT;
2626         } else if (S_ISDIR(inode->i_mode)) {
2627                 inode->i_op = &ll_dir_inode_operations;
2628                 inode->i_fop = &ll_dir_operations;
2629                 EXIT;
2630         } else if (S_ISLNK(inode->i_mode)) {
2631                 inode->i_op = &ll_fast_symlink_inode_operations;
2632                 EXIT;
2633         } else {
2634                 inode->i_op = &ll_special_inode_operations;
2635
2636                 init_special_inode(inode, inode->i_mode,
2637                                    inode->i_rdev);
2638
2639                 EXIT;
2640         }
2641
2642         return 0;
2643 }
2644
2645 void ll_delete_inode(struct inode *inode)
2646 {
2647         struct ll_inode_info *lli = ll_i2info(inode);
2648         ENTRY;
2649
2650         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2651                 /* It is last chance to write out dirty pages,
2652                  * otherwise we may lose data while umount.
2653                  *
2654                  * If i_nlink is 0 then just discard data. This is safe because
2655                  * local inode gets i_nlink 0 from server only for the last
2656                  * unlink, so that file is not opened somewhere else
2657                  */
2658                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2659                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2660         }
2661
2662         ll_truncate_inode_pages_final(inode);
2663         ll_clear_inode(inode);
2664         clear_inode(inode);
2665
2666         EXIT;
2667 }
2668
2669 int ll_iocontrol(struct inode *inode, struct file *file,
2670                  unsigned int cmd, unsigned long arg)
2671 {
2672         struct ll_sb_info *sbi = ll_i2sbi(inode);
2673         struct ptlrpc_request *req = NULL;
2674         int rc, flags = 0;
2675         ENTRY;
2676
2677         switch (cmd) {
2678         case FS_IOC_GETFLAGS: {
2679                 struct mdt_body *body;
2680                 struct md_op_data *op_data;
2681
2682                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2683                                              0, 0, LUSTRE_OPC_ANY,
2684                                              NULL);
2685                 if (IS_ERR(op_data))
2686                         RETURN(PTR_ERR(op_data));
2687
2688                 op_data->op_valid = OBD_MD_FLFLAGS;
2689                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2690                 ll_finish_md_op_data(op_data);
2691                 if (rc) {
2692                         CERROR("%s: failure inode "DFID": rc = %d\n",
2693                                sbi->ll_md_exp->exp_obd->obd_name,
2694                                PFID(ll_inode2fid(inode)), rc);
2695                         RETURN(-abs(rc));
2696                 }
2697
2698                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2699
2700                 flags = body->mbo_flags;
2701
2702                 ptlrpc_req_finished(req);
2703
2704                 RETURN(put_user(flags, (int __user *)arg));
2705         }
2706         case FS_IOC_SETFLAGS: {
2707                 struct iattr *attr;
2708                 struct md_op_data *op_data;
2709                 struct cl_object *obj;
2710                 struct fsxattr fa = { 0 };
2711
2712                 if (get_user(flags, (int __user *)arg))
2713                         RETURN(-EFAULT);
2714
2715                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2716                 if (flags & LUSTRE_PROJINHERIT_FL)
2717                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2718
2719                 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2720                                             fa.fsx_projid);
2721                 if (rc)
2722                         RETURN(rc);
2723
2724                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2725                                              LUSTRE_OPC_ANY, NULL);
2726                 if (IS_ERR(op_data))
2727                         RETURN(PTR_ERR(op_data));
2728
2729                 op_data->op_attr_flags = flags;
2730                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2731                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2732                 ll_finish_md_op_data(op_data);
2733                 ptlrpc_req_finished(req);
2734                 if (rc)
2735                         RETURN(rc);
2736
2737                 ll_update_inode_flags(inode, flags);
2738
2739                 obj = ll_i2info(inode)->lli_clob;
2740                 if (obj == NULL)
2741                         RETURN(0);
2742
2743                 OBD_ALLOC_PTR(attr);
2744                 if (attr == NULL)
2745                         RETURN(-ENOMEM);
2746
2747                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2748
2749                 OBD_FREE_PTR(attr);
2750                 RETURN(rc);
2751         }
2752         default:
2753                 RETURN(-ENOSYS);
2754         }
2755
2756         RETURN(0);
2757 }
2758
2759 int ll_flush_ctx(struct inode *inode)
2760 {
2761         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2762
2763         CDEBUG(D_SEC, "flush context for user %d\n",
2764                from_kuid(&init_user_ns, current_uid()));
2765
2766         obd_set_info_async(NULL, sbi->ll_md_exp,
2767                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2768                            0, NULL, NULL);
2769         obd_set_info_async(NULL, sbi->ll_dt_exp,
2770                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2771                            0, NULL, NULL);
2772         return 0;
2773 }
2774
2775 /* umount -f client means force down, don't save state */
2776 void ll_umount_begin(struct super_block *sb)
2777 {
2778         struct ll_sb_info *sbi = ll_s2sbi(sb);
2779         struct obd_device *obd;
2780         struct obd_ioctl_data *ioc_data;
2781         int cnt;
2782         ENTRY;
2783
2784         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2785                sb->s_count, atomic_read(&sb->s_active));
2786
2787         obd = class_exp2obd(sbi->ll_md_exp);
2788         if (obd == NULL) {
2789                 CERROR("Invalid MDC connection handle %#llx\n",
2790                        sbi->ll_md_exp->exp_handle.h_cookie);
2791                 EXIT;
2792                 return;
2793         }
2794         obd->obd_force = 1;
2795
2796         obd = class_exp2obd(sbi->ll_dt_exp);
2797         if (obd == NULL) {
2798                 CERROR("Invalid LOV connection handle %#llx\n",
2799                        sbi->ll_dt_exp->exp_handle.h_cookie);
2800                 EXIT;
2801                 return;
2802         }
2803         obd->obd_force = 1;
2804
2805         OBD_ALLOC_PTR(ioc_data);
2806         if (ioc_data) {
2807                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2808                               sizeof *ioc_data, ioc_data, NULL);
2809
2810                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2811                               sizeof *ioc_data, ioc_data, NULL);
2812
2813                 OBD_FREE_PTR(ioc_data);
2814         }
2815
2816         /* Really, we'd like to wait until there are no requests outstanding,
2817          * and then continue.  For now, we just periodically checking for vfs
2818          * to decrement mnt_cnt and hope to finish it within 10sec.
2819          */
2820         cnt = 10;
2821         while (cnt > 0 &&
2822                !may_umount(sbi->ll_mnt.mnt)) {
2823                 ssleep(1);
2824                 cnt -= 1;
2825         }
2826
2827         EXIT;
2828 }
2829
2830 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2831 {
2832         struct ll_sb_info *sbi = ll_s2sbi(sb);
2833         char *profilenm = get_profile_name(sb);
2834         int err;
2835         __u32 read_only;
2836
2837         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2838                 read_only = *flags & MS_RDONLY;
2839                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2840                                          sizeof(KEY_READ_ONLY),
2841                                          KEY_READ_ONLY, sizeof(read_only),
2842                                          &read_only, NULL);
2843                 if (err) {
2844                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2845                                       profilenm, read_only ?
2846                                       "read-only" : "read-write", err);
2847                         return err;
2848                 }
2849
2850                 if (read_only)
2851                         sb->s_flags |= SB_RDONLY;
2852                 else
2853                         sb->s_flags &= ~SB_RDONLY;
2854
2855                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2856                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2857                                       read_only ?  "read-only" : "read-write");
2858         }
2859         return 0;
2860 }
2861
2862 /**
2863  * Cleanup the open handle that is cached on MDT-side.
2864  *
2865  * For open case, the client side open handling thread may hit error
2866  * after the MDT grant the open. Under such case, the client should
2867  * send close RPC to the MDT as cleanup; otherwise, the open handle
2868  * on the MDT will be leaked there until the client umount or evicted.
2869  *
2870  * In further, if someone unlinked the file, because the open handle
2871  * holds the reference on such file/object, then it will block the
2872  * subsequent threads that want to locate such object via FID.
2873  *
2874  * \param[in] sb        super block for this file-system
2875  * \param[in] open_req  pointer to the original open request
2876  */
2877 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
2878 {
2879         struct mdt_body                 *body;
2880         struct md_op_data               *op_data;
2881         struct ptlrpc_request           *close_req = NULL;
2882         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2883         ENTRY;
2884
2885         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
2886         OBD_ALLOC_PTR(op_data);
2887         if (op_data == NULL) {
2888                 CWARN("%s: cannot allocate op_data to release open handle for "
2889                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2890
2891                 RETURN_EXIT;
2892         }
2893
2894         op_data->op_fid1 = body->mbo_fid1;
2895         op_data->op_open_handle = body->mbo_open_handle;
2896         op_data->op_mod_time = ktime_get_real_seconds();
2897         md_close(exp, op_data, NULL, &close_req);
2898         ptlrpc_req_finished(close_req);
2899         ll_finish_md_op_data(op_data);
2900
2901         EXIT;
2902 }
2903
2904 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
2905                   struct super_block *sb, struct lookup_intent *it)
2906 {
2907         struct ll_sb_info *sbi = NULL;
2908         struct lustre_md md = { NULL };
2909         bool default_lmv_deleted = false;
2910         int rc;
2911
2912         ENTRY;
2913
2914         LASSERT(*inode || sb);
2915         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2916         rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
2917                               sbi->ll_md_exp, &md);
2918         if (rc != 0)
2919                 GOTO(out, rc);
2920
2921         /*
2922          * clear default_lmv only if intent_getattr reply doesn't contain it.
2923          * but it needs to be done after iget, check this early because
2924          * ll_update_lsm_md() may change md.
2925          */
2926         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2927             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2928                 default_lmv_deleted = true;
2929
2930         if (*inode) {
2931                 rc = ll_update_inode(*inode, &md);
2932                 if (rc != 0)
2933                         GOTO(out, rc);
2934         } else {
2935                 LASSERT(sb != NULL);
2936
2937                 /*
2938                  * At this point server returns to client's same fid as client
2939                  * generated for creating. So using ->fid1 is okay here.
2940                  */
2941                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2942                         CERROR("%s: Fid is insane "DFID"\n",
2943                                 sbi->ll_fsname,
2944                                 PFID(&md.body->mbo_fid1));
2945                         GOTO(out, rc = -EINVAL);
2946                 }
2947
2948                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2949                                              sbi->ll_flags & LL_SBI_32BIT_API),
2950                                  &md);
2951                 if (IS_ERR(*inode)) {
2952                         lmd_clear_acl(&md);
2953                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2954                         *inode = NULL;
2955                         CERROR("new_inode -fatal: rc %d\n", rc);
2956                         GOTO(out, rc);
2957                 }
2958         }
2959
2960         /* Handling piggyback layout lock.
2961          * Layout lock can be piggybacked by getattr and open request.
2962          * The lsm can be applied to inode only if it comes with a layout lock
2963          * otherwise correct layout may be overwritten, for example:
2964          * 1. proc1: mdt returns a lsm but not granting layout
2965          * 2. layout was changed by another client
2966          * 3. proc2: refresh layout and layout lock granted
2967          * 4. proc1: to apply a stale layout */
2968         if (it != NULL && it->it_lock_mode != 0) {
2969                 struct lustre_handle lockh;
2970                 struct ldlm_lock *lock;
2971
2972                 lockh.cookie = it->it_lock_handle;
2973                 lock = ldlm_handle2lock(&lockh);
2974                 LASSERT(lock != NULL);
2975                 if (ldlm_has_layout(lock)) {
2976                         struct cl_object_conf conf;
2977
2978                         memset(&conf, 0, sizeof(conf));
2979                         conf.coc_opc = OBJECT_CONF_SET;
2980                         conf.coc_inode = *inode;
2981                         conf.coc_lock = lock;
2982                         conf.u.coc_layout = md.layout;
2983                         (void)ll_layout_conf(*inode, &conf);
2984                 }
2985                 LDLM_LOCK_PUT(lock);
2986         }
2987
2988         if (default_lmv_deleted)
2989                 ll_update_default_lsm_md(*inode, &md);
2990
2991         /* we may want to apply some policy for foreign file/dir */
2992         if (ll_sbi_has_foreign_symlink(sbi)) {
2993                 rc = ll_manage_foreign(*inode, &md);
2994                 if (rc < 0)
2995                         GOTO(out, rc);
2996         }
2997
2998         GOTO(out, rc = 0);
2999
3000 out:
3001         /* cleanup will be done if necessary */
3002         md_free_lustre_md(sbi->ll_md_exp, &md);
3003
3004         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3005                 ll_intent_drop_lock(it);
3006                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3007         }
3008
3009         return rc;
3010 }
3011
3012 int ll_obd_statfs(struct inode *inode, void __user *arg)
3013 {
3014         struct ll_sb_info *sbi = NULL;
3015         struct obd_export *exp;
3016         struct obd_ioctl_data *data = NULL;
3017         __u32 type;
3018         int len = 0, rc;
3019
3020         if (inode)
3021                 sbi = ll_i2sbi(inode);
3022         if (!sbi)
3023                 GOTO(out_statfs, rc = -EINVAL);
3024
3025         rc = obd_ioctl_getdata(&data, &len, arg);
3026         if (rc)
3027                 GOTO(out_statfs, rc);
3028
3029         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3030             !data->ioc_pbuf1 || !data->ioc_pbuf2)
3031                 GOTO(out_statfs, rc = -EINVAL);
3032
3033         if (data->ioc_inllen1 != sizeof(__u32) ||
3034             data->ioc_inllen2 != sizeof(__u32) ||
3035             data->ioc_plen1 != sizeof(struct obd_statfs) ||
3036             data->ioc_plen2 != sizeof(struct obd_uuid))
3037                 GOTO(out_statfs, rc = -EINVAL);
3038
3039         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3040         if (type & LL_STATFS_LMV)
3041                 exp = sbi->ll_md_exp;
3042         else if (type & LL_STATFS_LOV)
3043                 exp = sbi->ll_dt_exp;
3044         else
3045                 GOTO(out_statfs, rc = -ENODEV);
3046
3047         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3048         if (rc)
3049                 GOTO(out_statfs, rc);
3050 out_statfs:
3051         OBD_FREE_LARGE(data, len);
3052         return rc;
3053 }
3054
3055 /*
3056  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3057  * be called early to avoid deadlock.
3058  */
3059 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3060 {
3061         if (op_data->op_mea2_sem) {
3062                 up_read_non_owner(op_data->op_mea2_sem);
3063                 op_data->op_mea2_sem = NULL;
3064         }
3065
3066         if (op_data->op_mea1_sem) {
3067                 up_read_non_owner(op_data->op_mea1_sem);
3068                 op_data->op_mea1_sem = NULL;
3069         }
3070 }
3071
3072 /* this function prepares md_op_data hint for passing it down to MD stack. */
3073 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3074                                       struct inode *i1, struct inode *i2,
3075                                       const char *name, size_t namelen,
3076                                       __u32 mode, enum md_op_code opc,
3077                                       void *data)
3078 {
3079         LASSERT(i1 != NULL);
3080
3081         if (name == NULL) {
3082                 /* Do not reuse namelen for something else. */
3083                 if (namelen != 0)
3084                         return ERR_PTR(-EINVAL);
3085         } else {
3086                 if (namelen > ll_i2sbi(i1)->ll_namelen)
3087                         return ERR_PTR(-ENAMETOOLONG);
3088
3089                 /* "/" is not valid name, but it's allowed */
3090                 if (!lu_name_is_valid_2(name, namelen) &&
3091                     strncmp("/", name, namelen) != 0)
3092                         return ERR_PTR(-EINVAL);
3093         }
3094
3095         if (op_data == NULL)
3096                 OBD_ALLOC_PTR(op_data);
3097
3098         if (op_data == NULL)
3099                 return ERR_PTR(-ENOMEM);
3100
3101         ll_i2gids(op_data->op_suppgids, i1, i2);
3102         op_data->op_fid1 = *ll_inode2fid(i1);
3103         op_data->op_code = opc;
3104
3105         if (S_ISDIR(i1->i_mode)) {
3106                 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3107                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3108                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3109                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3110         }
3111
3112         if (i2) {
3113                 op_data->op_fid2 = *ll_inode2fid(i2);
3114                 if (S_ISDIR(i2->i_mode)) {
3115                         if (i2 != i1) {
3116                                 /* i2 is typically a child of i1, and MUST be
3117                                  * further from the root to avoid deadlocks.
3118                                  */
3119                                 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3120                                 op_data->op_mea2_sem =
3121                                                 &ll_i2info(i2)->lli_lsm_sem;
3122                         }
3123                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3124                 }
3125         } else {
3126                 fid_zero(&op_data->op_fid2);
3127         }
3128
3129         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
3130                 op_data->op_cli_flags |= CLI_HASH64;
3131
3132         if (ll_need_32bit_api(ll_i2sbi(i1)))
3133                 op_data->op_cli_flags |= CLI_API32;
3134
3135         op_data->op_name = name;
3136         op_data->op_namelen = namelen;
3137         op_data->op_mode = mode;
3138         op_data->op_mod_time = ktime_get_real_seconds();
3139         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3140         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3141         op_data->op_cap = cfs_curproc_cap_pack();
3142         op_data->op_mds = 0;
3143         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3144              filename_is_volatile(name, namelen, &op_data->op_mds)) {
3145                 op_data->op_bias |= MDS_CREATE_VOLATILE;
3146         }
3147         op_data->op_data = data;
3148
3149         return op_data;
3150 }
3151
3152 void ll_finish_md_op_data(struct md_op_data *op_data)
3153 {
3154         ll_unlock_md_op_lsm(op_data);
3155         ll_security_release_secctx(op_data->op_file_secctx,
3156                                    op_data->op_file_secctx_size);
3157         llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3158         OBD_FREE_PTR(op_data);
3159 }
3160
3161 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3162 {
3163         struct ll_sb_info *sbi;
3164
3165         LASSERT(seq && dentry);
3166         sbi = ll_s2sbi(dentry->d_sb);
3167
3168         if (sbi->ll_flags & LL_SBI_NOLCK)
3169                 seq_puts(seq, ",nolock");
3170
3171         /* "flock" is the default since 2.13, but it wasn't for many years,
3172          * so it is still useful to print this to show it is enabled.
3173          * Start to print "noflock" so it is now clear when flock is disabled.
3174          */
3175         if (sbi->ll_flags & LL_SBI_FLOCK)
3176                 seq_puts(seq, ",flock");
3177         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3178                 seq_puts(seq, ",localflock");
3179         else
3180                 seq_puts(seq, ",noflock");
3181
3182         if (sbi->ll_flags & LL_SBI_USER_XATTR)
3183                 seq_puts(seq, ",user_xattr");
3184
3185         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3186                 seq_puts(seq, ",lazystatfs");
3187
3188         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3189                 seq_puts(seq, ",user_fid2path");
3190
3191         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3192                 seq_puts(seq, ",always_ping");