Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / llite / llite_lib.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Lustre Light Super operations
5  *
6  *  Copyright (c) 2002-2005 Cluster File Systems, Inc.
7  *
8  *   This file is part of Lustre, http://www.lustre.org.
9  *
10  *   Lustre is free software; you can redistribute it and/or
11  *   modify it under the terms of version 2 of the GNU General Public
12  *   License as published by the Free Software Foundation.
13  *
14  *   Lustre is distributed in the hope that it will be useful,
15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  *   GNU General Public License for more details.
18  *
19  *   You should have received a copy of the GNU General Public License
20  *   along with Lustre; if not, write to the Free Software
21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22  */
23
24 #define DEBUG_SUBSYSTEM S_LLITE
25
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
30
31 #include <lustre_lite.h>
32 #include <lustre_ha.h>
33 #include <lustre_dlm.h>
34 #include <lprocfs_status.h>
35 #include <lustre_disk.h>
36 #include <lustre_param.h>
37 #include <lustre_log.h>
38 #include "llite_internal.h"
39
40 cfs_mem_cache_t *ll_file_data_slab;
41
42 LIST_HEAD(ll_super_blocks);
43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
44
45 extern struct address_space_operations ll_aops;
46 extern struct address_space_operations ll_dir_aops;
47
48 #ifndef log2
49 #define log2(n) ffz(~(n))
50 #endif
51
52
53 static struct ll_sb_info *ll_init_sbi(void)
54 {
55         struct ll_sb_info *sbi = NULL;
56         class_uuid_t uuid;
57         int i;
58         ENTRY;
59
60         OBD_ALLOC(sbi, sizeof(*sbi));
61         if (!sbi)
62                 RETURN(NULL);
63
64         spin_lock_init(&sbi->ll_lock);
65         spin_lock_init(&sbi->ll_lco.lco_lock);
66         spin_lock_init(&sbi->ll_pp_extent_lock);
67         spin_lock_init(&sbi->ll_process_lock);
68         sbi->ll_rw_stats_on = 0;
69         INIT_LIST_HEAD(&sbi->ll_pglist);
70         if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
71                 sbi->ll_async_page_max = num_physpages / 2;
72         else
73                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
74         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
75                                            SBI_DEFAULT_READAHEAD_MAX);
76         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
77                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
78
79         INIT_LIST_HEAD(&sbi->ll_conn_chain);
80         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
81
82         ll_generate_random_uuid(uuid);
83         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
84         CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
85
86         spin_lock(&ll_sb_lock);
87         list_add_tail(&sbi->ll_list, &ll_super_blocks);
88         spin_unlock(&ll_sb_lock);
89
90 #ifdef ENABLE_LLITE_CHECKSUM
91         sbi->ll_flags |= LL_SBI_CHECKSUM;
92 #endif
93
94 #ifdef HAVE_LRU_RESIZE_SUPPORT
95         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
96 #endif
97
98 #ifdef HAVE_EXPORT___IGET
99         INIT_LIST_HEAD(&sbi->ll_deathrow);
100         spin_lock_init(&sbi->ll_deathrow_lock);
101 #endif
102         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
103                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
104                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
105         }
106
107         RETURN(sbi);
108 }
109
110 void ll_free_sbi(struct super_block *sb)
111 {
112         struct ll_sb_info *sbi = ll_s2sbi(sb);
113         ENTRY;
114
115         if (sbi != NULL) {
116                 spin_lock(&ll_sb_lock);
117                 list_del(&sbi->ll_list);
118                 spin_unlock(&ll_sb_lock);
119                 OBD_FREE(sbi, sizeof(*sbi));
120         }
121         EXIT;
122 }
123
124 static struct dentry_operations ll_d_root_ops = {
125 #ifdef LUSTRE_KERNEL_VERSION
126         .d_compare = ll_dcompare,
127 #endif
128 };
129
130 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
131  * us to make MDS RPCs with large enough reply buffers to hold the
132  * maximum-sized (= maximum striped) EA and cookie without having to
133  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
134 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
135 {
136         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
137         __u32 valsize = sizeof(struct lov_desc);
138         int rc, easize, def_easize, cookiesize;
139         struct lov_desc desc;
140         __u32 stripes;
141         ENTRY;
142
143         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
144                           &valsize, &desc);
145         if (rc)
146                 RETURN(rc);
147
148         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
149         lsm.lsm_stripe_count = stripes;
150         easize = obd_size_diskmd(dt_exp, &lsm);
151
152         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
153         def_easize = obd_size_diskmd(dt_exp, &lsm);
154
155         cookiesize = stripes * sizeof(struct llog_cookie);
156
157         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
158                easize, cookiesize);
159
160         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
161         RETURN(rc);
162 }
163
164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
165                                     uid_t nllu, gid_t nllg)
166 {
167         struct inode *root = 0;
168         struct ll_sb_info *sbi = ll_s2sbi(sb);
169         struct obd_device *obd;
170         struct lu_fid rootfid;
171         struct obd_capa *oc = NULL;
172         struct obd_statfs osfs;
173         struct ptlrpc_request *request = NULL;
174         struct lustre_handle dt_conn = {0, };
175         struct lustre_handle md_conn = {0, };
176         struct obd_connect_data *data = NULL;
177         struct lustre_md lmd;
178         obd_valid valid;
179         int size, err, checksum;
180         ENTRY;
181
182         obd = class_name2obd(md);
183         if (!obd) {
184                 CERROR("MD %s: not setup or attached\n", md);
185                 RETURN(-EINVAL);
186         }
187
188         OBD_ALLOC_PTR(data);
189         if (data == NULL)
190                 RETURN(-ENOMEM);
191
192         if (proc_lustre_fs_root) {
193                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
194                                                   dt, md);
195                 if (err < 0)
196                         CERROR("could not register mount in /proc/lustre");
197         }
198
199         /* indicate the features supported by this client */
200         data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201                                   OBD_CONNECT_JOIN |
202                                   OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
203                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204                                   OBD_CONNECT_CANCELSET;
205 #ifdef HAVE_LRU_RESIZE_SUPPORT
206         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
207                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
208 #endif
209 #ifdef CONFIG_FS_POSIX_ACL
210         data->ocd_connect_flags |= OBD_CONNECT_ACL;
211 #endif
212         data->ocd_ibits_known = MDS_INODELOCK_FULL;
213         data->ocd_version = LUSTRE_VERSION_CODE;
214
215         if (sb->s_flags & MS_RDONLY)
216                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
217         if (sbi->ll_flags & LL_SBI_USER_XATTR)
218                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
219
220 #ifdef HAVE_MS_FLOCK_LOCK
221         /* force vfs to use lustre handler for flock() calls - bug 10743 */
222         sb->s_flags |= MS_FLOCK_LOCK;
223 #endif
224         
225         if (sbi->ll_flags & LL_SBI_FLOCK)
226                 sbi->ll_fop = &ll_file_operations_flock;
227         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
228                 sbi->ll_fop = &ll_file_operations;
229         else
230                 sbi->ll_fop = &ll_file_operations_noflock;
231
232         /* real client */
233         data->ocd_connect_flags |= OBD_CONNECT_REAL;
234         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
235                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
236                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
237         } else {
238                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
239                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
240         }
241         data->ocd_nllu = nllu;
242         data->ocd_nllg = nllg;
243
244         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
245         if (err == -EBUSY) {
246                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
247                                    "recovery, of which this client is not a "
248                                    "part. Please wait for recovery to complete,"
249                                    " abort, or time out.\n", md);
250                 GOTO(out, err);
251         } else if (err) {
252                 CERROR("cannot connect to %s: rc = %d\n", md, err);
253                 GOTO(out, err);
254         }
255         sbi->ll_md_exp = class_conn2export(&md_conn);
256
257         err = obd_fid_init(sbi->ll_md_exp);
258         if (err) {
259                 CERROR("Can't init metadata layer FID infrastructure, "
260                        "rc %d\n", err);
261                 GOTO(out_md, err);
262         }
263
264         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
265         if (err)
266                 GOTO(out_md_fid, err);
267
268         size = sizeof(*data);
269         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
270                            KEY_CONN_DATA,  &size, data);
271         if (err) {
272                 CERROR("Get connect data failed: %d \n", err);
273                 GOTO(out_md, err);
274         }
275
276         LASSERT(osfs.os_bsize);
277         sb->s_blocksize = osfs.os_bsize;
278         sb->s_blocksize_bits = log2(osfs.os_bsize);
279         sb->s_magic = LL_SUPER_MAGIC;
280
281         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
282          *         retval = in_file->f_op->sendfile(...);
283          *         if (*ppos > max)
284          *                 retval = -EOVERFLOW;
285          *
286          * it will check if *ppos is greater than max. However, max equals to
287          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
288          * has been defined as a signed long long ineger in linux kernel. */
289 #if BITS_PER_LONG == 64
290         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
291 #else
292         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
293 #endif
294         sbi->ll_namelen = osfs.os_namelen;
295         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
296
297         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
298             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
299                 LCONSOLE_INFO("Disabling user_xattr feature because "
300                               "it is not supported on the server\n");
301                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
302         }
303
304         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
305 #ifdef MS_POSIXACL
306                 sb->s_flags |= MS_POSIXACL;
307 #endif
308                 sbi->ll_flags |= LL_SBI_ACL;
309         } else {
310                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
311 #ifdef MS_POSIXACL
312                 sb->s_flags &= ~MS_POSIXACL;
313 #endif
314                 sbi->ll_flags &= ~LL_SBI_ACL;
315         }
316
317         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
318                 sbi->ll_flags |= LL_SBI_JOIN;
319
320         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
321                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
322                         /* sometimes local client claims to be remote, but mdt
323                          * will disagree when client gss not applied. */
324                         LCONSOLE_INFO("client claims to be remote, but server "
325                                       "rejected, forced to be local.\n");
326                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
327                 }
328         } else {
329                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
330                         /* with gss applied, remote client can not claim to be
331                          * local, so mdt maybe force client to be remote. */
332                         LCONSOLE_INFO("client claims to be local, but server "
333                                       "rejected, forced to be remote.\n");
334                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
335                 }
336         }
337
338         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
339                 LCONSOLE_INFO("client enabled MDS capability!\n");
340                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
341         }
342
343         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
344                 LCONSOLE_INFO("client enabled OSS capability!\n");
345                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
346         }
347
348         sbi->ll_sdev_orig = sb->s_dev;
349 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
350         /* We set sb->s_dev equal on all lustre clients in order to support
351          * NFS export clustering.  NFSD requires that the FSID be the same
352          * on all clients. */
353         /* s_dev is also used in lt_compare() to compare two fs, but that is
354          * only a node-local comparison. */
355
356         /* XXX: this will not work with LMV */
357         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
358                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
359 #endif
360
361         obd = class_name2obd(dt);
362         if (!obd) {
363                 CERROR("DT %s: not setup or attached\n", dt);
364                 GOTO(out_md_fid, err = -ENODEV);
365         }
366
367         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
368                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
369                                   OBD_CONNECT_CANCELSET;
370         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
371                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
372
373 #ifdef HAVE_LRU_RESIZE_SUPPORT
374         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
375 #endif
376         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
377                "ocd_grant: %d\n", data->ocd_connect_flags,
378                data->ocd_version, data->ocd_grant);
379
380         obd->obd_upcall.onu_owner = &sbi->ll_lco;
381         obd->obd_upcall.onu_upcall = ll_ocd_update;
382         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
383
384         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
385         if (err == -EBUSY) {
386                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
387                                    "recovery, of which this client is not a "
388                                    "part.  Please wait for recovery to "
389                                    "complete, abort, or time out.\n", dt);
390                 GOTO(out_md_fid, err);
391         } else if (err) {
392                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
393                 GOTO(out_md_fid, err);
394         }
395
396         sbi->ll_dt_exp = class_conn2export(&dt_conn);
397
398         err = obd_fid_init(sbi->ll_dt_exp);
399         if (err) {
400                 CERROR("Can't init data layer FID infrastructure, "
401                        "rc %d\n", err);
402                 GOTO(out_dt, err);
403         }
404         
405         spin_lock(&sbi->ll_lco.lco_lock);
406         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
407         spin_unlock(&sbi->ll_lco.lco_lock);
408
409         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
410
411         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
412                                   0, NULL, NULL, NULL);
413         if (err < 0) {
414                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
415                                    "filesystem. There must be at least one "
416                                    "active OST for a client to start.\n");
417                 GOTO(out_dt_fid, err);
418         }
419
420         if (!ll_async_page_slab) {
421                 ll_async_page_slab_size =
422                         size_round(sizeof(struct ll_async_page)) + err;
423                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
424                                                         ll_async_page_slab_size,
425                                                            0, 0);
426                 if (!ll_async_page_slab)
427                         GOTO(out_dt_fid, err = -ENOMEM);
428         }
429
430         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
431         if (err) {
432                 CERROR("cannot mds_connect: rc = %d\n", err);
433                 GOTO(out_dt_fid, err);
434         }
435         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
436         sbi->ll_root_fid = rootfid;
437
438         sb->s_op = &lustre_super_operations;
439         sb->s_export_op = &lustre_export_operations;
440
441         /* make root inode
442          * XXX: move this to after cbd setup? */
443         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
444         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
445                 valid |= OBD_MD_FLRMTPERM;
446         else if (sbi->ll_flags & LL_SBI_ACL)
447                 valid |= OBD_MD_FLACL;
448
449         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
450         if (oc)
451                 free_capa(oc);
452         if (err) {
453                 CERROR("md_getattr failed for root: rc = %d\n", err);
454                 GOTO(out_dt_fid, err);
455         }
456         memset(&lmd, 0, sizeof(lmd));
457         err = md_get_lustre_md(sbi->ll_md_exp, request, 
458                                REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp, 
459                                &lmd);
460         if (err) {
461                 CERROR("failed to understand root inode md: rc = %d\n", err);
462                 ptlrpc_req_finished (request);
463                 GOTO(out_dt_fid, err);
464         }
465
466         LASSERT(fid_is_sane(&sbi->ll_root_fid));
467         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
468         md_free_lustre_md(sbi->ll_md_exp, &lmd);
469         ptlrpc_req_finished(request);
470
471         if (root == NULL || is_bad_inode(root)) {
472                 if (lmd.lsm)
473                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
474 #ifdef CONFIG_FS_POSIX_ACL
475                 if (lmd.posix_acl) {
476                         posix_acl_release(lmd.posix_acl);
477                         lmd.posix_acl = NULL;
478                 }
479 #endif
480                 CERROR("lustre_lite: bad iget4 for root\n");
481                 GOTO(out_root, err = -EBADF);
482         }
483
484         err = ll_close_thread_start(&sbi->ll_lcq);
485         if (err) {
486                 CERROR("cannot start close thread: rc %d\n", err);
487                 GOTO(out_root, err);
488         }
489
490         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
491         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
492                                  sizeof(checksum), &checksum, NULL);
493
494         sb->s_root = d_alloc_root(root);
495         if (data != NULL)
496                 OBD_FREE(data, sizeof(*data));
497         sb->s_root->d_op = &ll_d_root_ops;
498         RETURN(err);
499 out_root:
500         if (root)
501                 iput(root);
502 out_dt_fid:
503         obd_fid_fini(sbi->ll_dt_exp);
504 out_dt:
505         obd_disconnect(sbi->ll_dt_exp);
506         sbi->ll_dt_exp = NULL;
507 out_md_fid:
508         obd_fid_fini(sbi->ll_md_exp);
509 out_md:
510         obd_disconnect(sbi->ll_md_exp);
511         sbi->ll_md_exp = NULL;
512 out:
513         if (data != NULL)
514                 OBD_FREE_PTR(data);
515         lprocfs_unregister_mountpoint(sbi);
516         return err;
517 }
518
519 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
520 {
521         int size, rc;
522
523         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
524         size = sizeof(int);
525         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
526                           &size, lmmsize);
527         if (rc)
528                 CERROR("Get max mdsize error rc %d \n", rc);
529
530         RETURN(rc);
531 }
532
533 void ll_dump_inode(struct inode *inode)
534 {
535         struct list_head *tmp;
536         int dentry_count = 0;
537
538         LASSERT(inode != NULL);
539
540         list_for_each(tmp, &inode->i_dentry)
541                 dentry_count++;
542
543         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
544                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
545                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
546 }
547
548 void lustre_dump_dentry(struct dentry *dentry, int recur)
549 {
550         struct list_head *tmp;
551         int subdirs = 0;
552
553         LASSERT(dentry != NULL);
554
555         list_for_each(tmp, &dentry->d_subdirs)
556                 subdirs++;
557
558         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
559                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
560                dentry->d_name.len, dentry->d_name.name,
561                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
562                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
563                dentry->d_flags, dentry->d_fsdata, subdirs);
564         if (dentry->d_inode != NULL)
565                 ll_dump_inode(dentry->d_inode);
566
567         if (recur == 0)
568                 return;
569
570         list_for_each(tmp, &dentry->d_subdirs) {
571                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
572                 lustre_dump_dentry(d, recur - 1);
573         }
574 }
575
576 #ifdef HAVE_EXPORT___IGET
577 static void prune_dir_dentries(struct inode *inode)
578 {
579         struct dentry *dentry, *prev = NULL;
580
581         /* due to lustre specific logic, a directory
582          * can have few dentries - a bug from VFS POV */
583 restart:
584         spin_lock(&dcache_lock);
585         if (!list_empty(&inode->i_dentry)) {
586                 dentry = list_entry(inode->i_dentry.prev,
587                                     struct dentry, d_alias);
588                 /* in order to prevent infinite loops we
589                  * break if previous dentry is busy */
590                 if (dentry != prev) {
591                         prev = dentry;
592                         dget_locked(dentry);
593                         spin_unlock(&dcache_lock);
594
595                         /* try to kill all child dentries */
596                         lock_dentry(dentry);
597                         shrink_dcache_parent(dentry);
598                         unlock_dentry(dentry);
599                         dput(dentry);
600
601                         /* now try to get rid of current dentry */
602                         d_prune_aliases(inode);
603                         goto restart;
604                 }
605         }
606         spin_unlock(&dcache_lock);
607 }
608
609 static void prune_deathrow_one(struct ll_inode_info *lli)
610 {
611         struct inode *inode = ll_info2i(lli);
612
613         /* first, try to drop any dentries - they hold a ref on the inode */
614         if (S_ISDIR(inode->i_mode))
615                 prune_dir_dentries(inode);
616         else
617                 d_prune_aliases(inode);
618
619
620         /* if somebody still uses it, leave it */
621         LASSERT(atomic_read(&inode->i_count) > 0);
622         if (atomic_read(&inode->i_count) > 1)
623                 goto out;
624
625         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
626                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
627
628         /* seems nobody uses it anymore */
629         inode->i_nlink = 0;
630
631 out:
632         iput(inode);
633         return;
634 }
635
636 static void prune_deathrow(struct ll_sb_info *sbi, int try)
637 {
638         struct ll_inode_info *lli;
639         int empty;
640
641         do {
642                 if (need_resched() && try)
643                         break;
644
645                 if (try) {
646                         if (!spin_trylock(&sbi->ll_deathrow_lock))
647                                 break;
648                 } else {
649                         spin_lock(&sbi->ll_deathrow_lock);
650                 }
651
652                 empty = 1;
653                 lli = NULL;
654                 if (!list_empty(&sbi->ll_deathrow)) {
655                         lli = list_entry(sbi->ll_deathrow.next,
656                                          struct ll_inode_info,
657                                          lli_dead_list);
658                         list_del_init(&lli->lli_dead_list);
659                         if (!list_empty(&sbi->ll_deathrow))
660                                 empty = 0;
661                 }
662                 spin_unlock(&sbi->ll_deathrow_lock);
663
664                 if (lli)
665                         prune_deathrow_one(lli);
666
667         } while (empty == 0);
668 }
669 #else /* !HAVE_EXPORT___IGET */
670 #define prune_deathrow(sbi, try) do {} while (0)
671 #endif /* HAVE_EXPORT___IGET */
672
673 void client_common_put_super(struct super_block *sb)
674 {
675         struct ll_sb_info *sbi = ll_s2sbi(sb);
676         ENTRY;
677
678         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
679
680         ll_close_thread_shutdown(sbi->ll_lcq);
681
682         /* destroy inodes in deathrow */
683         prune_deathrow(sbi, 0);
684
685         list_del(&sbi->ll_conn_chain);
686
687         obd_fid_fini(sbi->ll_dt_exp);
688         obd_disconnect(sbi->ll_dt_exp);
689         sbi->ll_dt_exp = NULL;
690
691         lprocfs_unregister_mountpoint(sbi);
692
693         obd_fid_fini(sbi->ll_md_exp);
694         obd_disconnect(sbi->ll_md_exp);
695         sbi->ll_md_exp = NULL;
696
697         lustre_throw_orphan_dentries(sb);
698
699         EXIT;
700 }
701
702 void ll_kill_super(struct super_block *sb)
703 {
704         struct ll_sb_info *sbi;
705
706         ENTRY;
707
708         /* not init sb ?*/
709         if (!(sb->s_flags & MS_ACTIVE))
710                 return;
711
712         sbi = ll_s2sbi(sb);
713         /* we need restore s_dev from changed for clustred NFS before put_super
714          * because new kernels have cached s_dev and change sb->s_dev in
715          * put_super not affected real removing devices */
716         if (sbi)
717                 sb->s_dev = sbi->ll_sdev_orig;
718         EXIT;
719 }
720
721 char *ll_read_opt(const char *opt, char *data)
722 {
723         char *value;
724         char *retval;
725         ENTRY;
726
727         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
728         if (strncmp(opt, data, strlen(opt)))
729                 RETURN(NULL);
730         if ((value = strchr(data, '=')) == NULL)
731                 RETURN(NULL);
732
733         value++;
734         OBD_ALLOC(retval, strlen(value) + 1);
735         if (!retval) {
736                 CERROR("out of memory!\n");
737                 RETURN(NULL);
738         }
739
740         memcpy(retval, value, strlen(value)+1);
741         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
742         RETURN(retval);
743 }
744
745 static inline int ll_set_opt(const char *opt, char *data, int fl)
746 {
747         if (strncmp(opt, data, strlen(opt)) != 0)
748                 return(0);
749         else
750                 return(fl);
751 }
752
753 /* non-client-specific mount options are parsed in lmd_parse */
754 static int ll_options(char *options, int *flags)
755 {
756         int tmp;
757         char *s1 = options, *s2;
758         ENTRY;
759
760         if (!options) 
761                 RETURN(0);
762
763         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
764
765         while (*s1) {
766                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
767                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
768                 if (tmp) {
769                         *flags |= tmp;
770                         goto next;
771                 }
772                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
773                 if (tmp) {
774                         *flags |= tmp;
775                         goto next;
776                 }
777                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
778                 if (tmp) {
779                         *flags |= tmp;
780                         goto next;
781                 }
782                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
783                 if (tmp) {
784                         *flags &= ~tmp;
785                         goto next;
786                 }
787                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
788                 if (tmp) {
789                         *flags |= tmp;
790                         goto next;
791                 }
792                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
793                 if (tmp) {
794                         *flags &= ~tmp;
795                         goto next;
796                 }
797                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
798                 if (tmp) {
799                         /* Ignore deprecated mount option.  The client will
800                          * always try to mount with ACL support, whether this
801                          * is used depends on whether server supports it. */
802                         goto next;
803                 }
804                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
805                 if (tmp) {
806                         goto next;
807                 }
808                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
809                 if (tmp) {
810                         *flags |= tmp;
811                         goto next;
812                 }
813
814                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
815                 if (tmp) {
816                         *flags |= tmp;
817                         goto next;
818                 }
819                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
820                 if (tmp) {
821                         *flags &= ~tmp;
822                         goto next;
823                 }
824                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
825                 if (tmp) {
826                         *flags |= tmp;
827                         goto next;
828                 }
829                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
830                 if (tmp) {
831                         *flags &= ~tmp;
832                         goto next;
833                 }
834
835                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
836                                    s1);
837                 RETURN(-EINVAL);
838
839 next:
840                 /* Find next opt */
841                 s2 = strchr(s1, ',');
842                 if (s2 == NULL)
843                         break;
844                 s1 = s2 + 1;
845         }
846         RETURN(0);
847 }
848
849 void ll_lli_init(struct ll_inode_info *lli)
850 {
851         lli->lli_inode_magic = LLI_INODE_MAGIC;
852         sema_init(&lli->lli_size_sem, 1);
853         sema_init(&lli->lli_write_sem, 1);
854         lli->lli_flags = 0;
855         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
856         spin_lock_init(&lli->lli_lock);
857         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
858         INIT_LIST_HEAD(&lli->lli_close_list);
859         lli->lli_inode_magic = LLI_INODE_MAGIC;
860         sema_init(&lli->lli_och_sem, 1);
861         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
862         lli->lli_mds_exec_och = NULL;
863         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
864         lli->lli_open_fd_exec_count = 0;
865         INIT_LIST_HEAD(&lli->lli_dead_list);
866         lli->lli_remote_perms = NULL;
867         lli->lli_rmtperm_utime = 0;
868         sema_init(&lli->lli_rmtperm_sem, 1);
869         INIT_LIST_HEAD(&lli->lli_oss_capas);
870 }
871
872 int ll_fill_super(struct super_block *sb)
873 {
874         struct lustre_profile *lprof;
875         struct lustre_sb_info *lsi = s2lsi(sb);
876         struct ll_sb_info *sbi;
877         char  *dt = NULL, *md = NULL;
878         char  *profilenm = get_profile_name(sb);
879         struct config_llog_instance cfg = {0, };
880         char   ll_instance[sizeof(sb) * 2 + 1];
881         int    err;
882         ENTRY;
883
884         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
885
886         cfs_module_get();
887
888         sb->s_type->fs_flags |= FS_ODD_RENAME;
889         /* client additional sb info */
890         lsi->lsi_llsbi = sbi = ll_init_sbi();
891         if (!sbi) {
892                 cfs_module_put();
893                 RETURN(-ENOMEM);
894         }
895
896         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
897         if (err) 
898                 GOTO(out_free, err);
899
900         /* Generate a string unique to this super, in case some joker tries
901            to mount the same fs at two mount points.
902            Use the address of the super itself.*/
903         sprintf(ll_instance, "%p", sb);
904         cfg.cfg_instance = ll_instance;
905         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
906
907         /* set up client obds */
908         err = lustre_process_log(sb, profilenm, &cfg);
909         if (err < 0) {
910                 CERROR("Unable to process log: %d\n", err);
911                 GOTO(out_free, err);
912         }
913
914         lprof = class_get_profile(profilenm);
915         if (lprof == NULL) {
916                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
917                                    " read from the MGS.  Does that filesystem "
918                                    "exist?\n", profilenm);
919                 GOTO(out_free, err = -EINVAL);
920         }
921         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
922                lprof->lp_md, lprof->lp_dt);
923
924         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
925                   strlen(ll_instance) + 2);
926         if (!dt)
927                 GOTO(out_free, err = -ENOMEM);
928         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
929
930         OBD_ALLOC(md, strlen(lprof->lp_md) +
931                   strlen(ll_instance) + 2);
932         if (!md)
933                 GOTO(out_free, err = -ENOMEM);
934         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
935
936         /* connections, registrations, sb setup */
937         err = client_common_fill_super(sb, md, dt,
938                                        lsi->lsi_lmd->lmd_nllu,
939                                        lsi->lsi_lmd->lmd_nllg);
940
941 out_free:
942         if (md)
943                 OBD_FREE(md, strlen(md) + 1);
944         if (dt)
945                 OBD_FREE(dt, strlen(dt) + 1);
946         if (err) 
947                 ll_put_super(sb);
948         else
949                 LCONSOLE_WARN("Client %s has started\n", profilenm);        
950
951         RETURN(err);
952 } /* ll_fill_super */
953
954
955 void ll_put_super(struct super_block *sb)
956 {
957         struct config_llog_instance cfg;
958         char   ll_instance[sizeof(sb) * 2 + 1];
959         struct obd_device *obd;
960         struct lustre_sb_info *lsi = s2lsi(sb);
961         struct ll_sb_info *sbi = ll_s2sbi(sb);
962         char *profilenm = get_profile_name(sb);
963         int force = 1, next;
964         ENTRY;
965
966         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
967
968         ll_print_capa_stat(sbi);
969
970         sprintf(ll_instance, "%p", sb);
971         cfg.cfg_instance = ll_instance;
972         lustre_end_log(sb, NULL, &cfg);
973         
974         if (sbi->ll_md_exp) {
975                 obd = class_exp2obd(sbi->ll_md_exp);
976                 if (obd) 
977                         force = obd->obd_force;
978         }
979         
980         /* We need to set force before the lov_disconnect in 
981            lustre_common_put_super, since l_d cleans up osc's as well. */
982         if (force) {
983                 next = 0;
984                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
985                                                      &next)) != NULL) {
986                         obd->obd_force = force;
987                 }
988         }                       
989
990         if (sbi->ll_lcq) {
991                 /* Only if client_common_fill_super succeeded */
992                 client_common_put_super(sb);
993         }
994         next = 0;
995         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
996                 class_manual_cleanup(obd);
997         }
998
999         if (profilenm)
1000                 class_del_profile(profilenm);
1001
1002         ll_free_sbi(sb);
1003         lsi->lsi_llsbi = NULL;
1004
1005         lustre_common_put_super(sb);
1006
1007         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1008         
1009         cfs_module_put();
1010
1011         EXIT;
1012 } /* client_put_super */
1013
1014 #ifdef HAVE_REGISTER_CACHE
1015 #include <linux/cache_def.h>
1016 #ifdef HAVE_CACHE_RETURN_INT
1017 static int
1018 #else
1019 static void
1020 #endif
1021 ll_shrink_cache(int priority, unsigned int gfp_mask)
1022 {
1023         struct ll_sb_info *sbi;
1024         int count = 0;
1025
1026         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1027                 count += llap_shrink_cache(sbi, priority);
1028
1029 #ifdef HAVE_CACHE_RETURN_INT
1030         return count;
1031 #endif
1032 }
1033
1034 struct cache_definition ll_cache_definition = {
1035         .name = "llap_cache",
1036         .shrink = ll_shrink_cache
1037 };
1038 #endif /* HAVE_REGISTER_CACHE */
1039
1040 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1041 {
1042         struct inode *inode = NULL;
1043         /* NOTE: we depend on atomic igrab() -bzzz */
1044         lock_res_and_lock(lock);
1045         if (lock->l_ast_data) {
1046                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1047                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1048                         inode = igrab(lock->l_ast_data);
1049                 } else {
1050                         inode = lock->l_ast_data;
1051                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1052                                                 D_INFO : D_WARNING,
1053                                         lock, __FILE__, __func__, __LINE__,
1054                                         "l_ast_data %p is bogus: magic %08x",
1055                                         lock->l_ast_data, lli->lli_inode_magic);
1056                         inode = NULL;
1057                 }
1058         }
1059         unlock_res_and_lock(lock);
1060         return inode;
1061 }
1062
1063 static int null_if_equal(struct ldlm_lock *lock, void *data)
1064 {
1065         if (data == lock->l_ast_data) {
1066                 lock->l_ast_data = NULL;
1067
1068                 if (lock->l_req_mode != lock->l_granted_mode)
1069                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1070         }
1071
1072         return LDLM_ITER_CONTINUE;
1073 }
1074
1075 void ll_clear_inode(struct inode *inode)
1076 {
1077         struct ll_inode_info *lli = ll_i2info(inode);
1078         struct ll_sb_info *sbi = ll_i2sbi(inode);
1079         ENTRY;
1080
1081         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1082                inode->i_generation, inode);
1083
1084         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1085         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1086                          null_if_equal, inode);
1087
1088         LASSERT(!lli->lli_open_fd_write_count);
1089         LASSERT(!lli->lli_open_fd_read_count);
1090         LASSERT(!lli->lli_open_fd_exec_count);
1091
1092         if (lli->lli_mds_write_och)
1093                 ll_md_real_close(inode, FMODE_WRITE);
1094         if (lli->lli_mds_exec_och)
1095                 ll_md_real_close(inode, FMODE_EXEC);
1096         if (lli->lli_mds_read_och)
1097                 ll_md_real_close(inode, FMODE_READ);
1098
1099         if (lli->lli_smd) {
1100                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1101                                   null_if_equal, inode);
1102
1103                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1104                 lli->lli_smd = NULL;
1105         }
1106
1107         if (lli->lli_symlink_name) {
1108                 OBD_FREE(lli->lli_symlink_name,
1109                          strlen(lli->lli_symlink_name) + 1);
1110                 lli->lli_symlink_name = NULL;
1111         }
1112
1113         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1114                 LASSERT(lli->lli_posix_acl == NULL);
1115                 if (lli->lli_remote_perms) {
1116                         free_rmtperm_hash(lli->lli_remote_perms);
1117                         lli->lli_remote_perms = NULL;
1118                 }
1119         }
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         else if (lli->lli_posix_acl) {
1122                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1123                 LASSERT(lli->lli_remote_perms == NULL);
1124                 posix_acl_release(lli->lli_posix_acl);
1125                 lli->lli_posix_acl = NULL;
1126         }
1127 #endif
1128         lli->lli_inode_magic = LLI_INODE_DEAD;
1129
1130 #ifdef HAVE_EXPORT___IGET
1131         spin_lock(&sbi->ll_deathrow_lock);
1132         list_del_init(&lli->lli_dead_list);
1133         spin_unlock(&sbi->ll_deathrow_lock);
1134 #endif
1135         ll_clear_inode_capas(inode);
1136
1137         EXIT;
1138 }
1139
1140 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1141                   struct md_open_data **mod)
1142 {
1143         struct lustre_md md;
1144         struct ll_sb_info *sbi = ll_i2sbi(inode);
1145         struct ptlrpc_request *request = NULL;
1146         int rc;
1147         ENTRY;
1148         
1149         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, 
1150                                      LUSTRE_OPC_ANY, NULL);
1151         if (IS_ERR(op_data))
1152                 RETURN(PTR_ERR(op_data));
1153
1154         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0, 
1155                         &request, mod);
1156         if (rc) {
1157                 ptlrpc_req_finished(request);
1158                 if (rc == -ENOENT) {
1159                         inode->i_nlink = 0;
1160                         /* Unlinked special device node? Or just a race?
1161                          * Pretend we done everything. */
1162                         if (!S_ISREG(inode->i_mode) &&
1163                             !S_ISDIR(inode->i_mode))
1164                                 rc = inode_setattr(inode, &op_data->op_attr);
1165                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1166                         CERROR("md_setattr fails: rc = %d\n", rc);
1167                 }
1168                 RETURN(rc);
1169         }
1170
1171         rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1172                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1173         if (rc) {
1174                 ptlrpc_req_finished(request);
1175                 RETURN(rc);
1176         }
1177
1178         /* We call inode_setattr to adjust timestamps.
1179          * If there is at least some data in file, we cleared ATTR_SIZE
1180          * above to avoid invoking vmtruncate, otherwise it is important
1181          * to call vmtruncate in inode_setattr to update inode->i_size
1182          * (bug 6196) */
1183         rc = inode_setattr(inode, &op_data->op_attr);
1184
1185         /* Extract epoch data if obtained. */
1186         op_data->op_handle = md.body->handle;
1187         op_data->op_ioepoch = md.body->ioepoch;
1188
1189         ll_update_inode(inode, &md);
1190         ptlrpc_req_finished(request);
1191
1192         RETURN(rc);
1193 }
1194
1195 /* Close IO epoch and send Size-on-MDS attribute update. */
1196 static int ll_setattr_done_writing(struct inode *inode,
1197                                    struct md_op_data *op_data,
1198                                    struct md_open_data *mod)
1199 {
1200         struct ll_inode_info *lli = ll_i2info(inode);
1201         int rc = 0;
1202         ENTRY;
1203         
1204         LASSERT(op_data != NULL);
1205         if (!S_ISREG(inode->i_mode))
1206                 RETURN(0);
1207
1208         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1209                op_data->op_ioepoch, PFID(&lli->lli_fid));
1210
1211         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1212         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1213         if (rc == -EAGAIN) {
1214                 /* MDS has instructed us to obtain Size-on-MDS attribute
1215                  * from OSTs and send setattr to back to MDS. */
1216                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1217                                          op_data->op_ioepoch);
1218         } else if (rc) {
1219                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1220                        inode->i_ino, rc);
1221         }
1222         RETURN(rc);
1223 }
1224
1225 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1226  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1227  * keep these values until such a time that objects are allocated for it.
1228  * We do the MDS operations first, as it is checking permissions for us.
1229  * We don't to the MDS RPC if there is nothing that we want to store there,
1230  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1231  * going to do an RPC anyways.
1232  *
1233  * If we are doing a truncate, we will send the mtime and ctime updates
1234  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1235  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1236  * at the same time.
1237  */
1238 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1239 {
1240         struct ll_inode_info *lli = ll_i2info(inode);
1241         struct lov_stripe_md *lsm = lli->lli_smd;
1242         struct ll_sb_info *sbi = ll_i2sbi(inode);
1243         struct md_op_data *op_data = NULL;
1244         struct md_open_data *mod = NULL;
1245         int ia_valid = attr->ia_valid;
1246         int rc = 0, rc1 = 0;
1247         ENTRY;
1248
1249         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1250                attr->ia_valid);
1251         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1252
1253         if (ia_valid & ATTR_SIZE) {
1254                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1255                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1256                                attr->ia_size, ll_file_maxbytes(inode));
1257                         RETURN(-EFBIG);
1258                 }
1259
1260                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1261         }
1262
1263         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1264         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1265                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1266                         RETURN(-EPERM);
1267         }
1268
1269         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1270         if (attr->ia_valid & ATTR_CTIME) {
1271                 attr->ia_ctime = CURRENT_TIME;
1272                 attr->ia_valid |= ATTR_CTIME_SET;
1273         }
1274         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1275                 attr->ia_atime = CURRENT_TIME;
1276                 attr->ia_valid |= ATTR_ATIME_SET;
1277         }
1278         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1279                 attr->ia_mtime = CURRENT_TIME;
1280                 attr->ia_valid |= ATTR_MTIME_SET;
1281         }
1282         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1283                 /* To avoid stale mtime on mds, obtain it from ost and send 
1284                    to mds. */
1285                 rc = ll_glimpse_size(inode, 0);
1286                 if (rc) 
1287                         RETURN(rc);
1288                 
1289                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1290                 attr->ia_mtime = inode->i_mtime;
1291         }
1292
1293         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1294                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1295                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1296                        CURRENT_SECONDS);
1297
1298         /* NB: ATTR_SIZE will only be set after this point if the size
1299          * resides on the MDS, ie, this file has no objects. */
1300         if (lsm)
1301                 attr->ia_valid &= ~ATTR_SIZE;
1302
1303         /* We always do an MDS RPC, even if we're only changing the size;
1304          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1305
1306         OBD_ALLOC_PTR(op_data);
1307         if (op_data == NULL)
1308                 RETURN(-ENOMEM);
1309         
1310         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1311
1312         /* Open epoch for truncate. */
1313         if (ia_valid & ATTR_SIZE)
1314                 op_data->op_flags = MF_EPOCH_OPEN;
1315         
1316         rc = ll_md_setattr(inode, op_data, &mod);
1317         if (rc)
1318                 GOTO(out, rc);
1319
1320         if (op_data->op_ioepoch)
1321                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1322                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1323
1324         if (!lsm || !S_ISREG(inode->i_mode)) {
1325                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1326                 GOTO(out, rc = 0);
1327         }
1328
1329         /* We really need to get our PW lock before we change inode->i_size.
1330          * If we don't we can race with other i_size updaters on our node, like
1331          * ll_file_read.  We can also race with i_size propogation to other
1332          * nodes through dirtying and writeback of final cached pages.  This
1333          * last one is especially bad for racing o_append users on other
1334          * nodes. */
1335         if (ia_valid & ATTR_SIZE) {
1336                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1337                                                            OBD_OBJECT_EOF } };
1338                 struct lustre_handle lockh = { 0 };
1339                 int err, ast_flags = 0;
1340                 /* XXX when we fix the AST intents to pass the discard-range
1341                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1342                  * XXX here. */
1343                 if (attr->ia_size == 0)
1344                         ast_flags = LDLM_AST_DISCARD_DATA;
1345
1346                 UNLOCK_INODE_MUTEX(inode);
1347                 UP_WRITE_I_ALLOC_SEM(inode);
1348                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1349                                     ast_flags);
1350                 LOCK_INODE_MUTEX(inode);
1351                 DOWN_WRITE_I_ALLOC_SEM(inode);
1352
1353                 if (rc != 0)
1354                         GOTO(out, rc);
1355
1356                 /* Only ll_inode_size_lock is taken at this level.
1357                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1358                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1359                  * ll_truncate dropped ll_inode_size_lock() */
1360                 ll_inode_size_lock(inode, 0);
1361                 rc = vmtruncate(inode, attr->ia_size);
1362                 if (rc != 0) {
1363                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1364                         ll_inode_size_unlock(inode, 0);
1365                 }
1366
1367                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1368                 if (err) {
1369                         CERROR("ll_extent_unlock failed: %d\n", err);
1370                         if (!rc)
1371                                 rc = err;
1372                 }
1373         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1374                 obd_flag flags;
1375                 struct obd_info oinfo = { { { 0 } } };
1376                 struct obdo *oa;
1377
1378                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1379                        inode->i_ino, LTIME_S(attr->ia_mtime));
1380
1381                 OBDO_ALLOC(oa);
1382                 if (oa) {
1383                         oa->o_id = lsm->lsm_object_id;
1384                         oa->o_gr = lsm->lsm_object_gr;
1385                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1386
1387                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1388                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1389                                 OBD_MD_FLFID | OBD_MD_FLGENER | 
1390                                 OBD_MD_FLGROUP;
1391
1392                         obdo_from_inode(oa, inode, flags);
1393
1394                         oinfo.oi_oa = oa;
1395                         oinfo.oi_md = lsm;
1396                         oinfo.oi_capa = ll_mdscapa_get(inode);
1397
1398                         /* XXX: this looks unnecessary now. */
1399                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1400                         capa_put(oinfo.oi_capa);
1401                         if (rc)
1402                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1403                         OBDO_FREE(oa);
1404                 } else {
1405                         rc = -ENOMEM;
1406                 }
1407         }
1408         EXIT;
1409 out:
1410         if (op_data) {
1411                 if (op_data->op_ioepoch)
1412                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1413                 ll_finish_md_op_data(op_data);
1414         }
1415         return rc ? rc : rc1;
1416 }
1417
1418 int ll_setattr(struct dentry *de, struct iattr *attr)
1419 {
1420         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1421             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1422                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1423
1424         return ll_setattr_raw(de->d_inode, attr);
1425 }
1426
1427 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1428                        __u64 max_age)
1429 {
1430         struct ll_sb_info *sbi = ll_s2sbi(sb);
1431         struct obd_statfs obd_osfs;
1432         int rc;
1433         ENTRY;
1434
1435         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1436         if (rc) {
1437                 CERROR("md_statfs fails: rc = %d\n", rc);
1438                 RETURN(rc);
1439         }
1440
1441         osfs->os_type = sb->s_magic;
1442
1443         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1444                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1445
1446         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1447                               &obd_osfs, max_age);
1448         if (rc) {
1449                 CERROR("obd_statfs fails: rc = %d\n", rc);
1450                 RETURN(rc);
1451         }
1452
1453         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1454                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1455                obd_osfs.os_files);
1456
1457         osfs->os_bsize = obd_osfs.os_bsize;
1458         osfs->os_blocks = obd_osfs.os_blocks;
1459         osfs->os_bfree = obd_osfs.os_bfree;
1460         osfs->os_bavail = obd_osfs.os_bavail;
1461
1462         /* If we don't have as many objects free on the OST as inodes
1463          * on the MDS, we reduce the total number of inodes to
1464          * compensate, so that the "inodes in use" number is correct.
1465          */
1466         if (obd_osfs.os_ffree < osfs->os_ffree) {
1467                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1468                         obd_osfs.os_ffree;
1469                 osfs->os_ffree = obd_osfs.os_ffree;
1470         }
1471
1472         RETURN(rc);
1473 }
1474 #ifndef HAVE_STATFS_DENTRY_PARAM
1475 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1476 {
1477 #else
1478 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1479 {
1480         struct super_block *sb = de->d_sb;
1481 #endif
1482         struct obd_statfs osfs;
1483         int rc;
1484
1485         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1486         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1487
1488         /* For now we will always get up-to-date statfs values, but in the
1489          * future we may allow some amount of caching on the client (e.g.
1490          * from QOS or lprocfs updates). */
1491         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1492         if (rc)
1493                 return rc;
1494
1495         statfs_unpack(sfs, &osfs);
1496
1497         /* We need to downshift for all 32-bit kernels, because we can't
1498          * tell if the kernel is being called via sys_statfs64() or not.
1499          * Stop before overflowing f_bsize - in which case it is better
1500          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1501         if (sizeof(long) < 8) {
1502                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1503                         sfs->f_bsize <<= 1;
1504
1505                         osfs.os_blocks >>= 1;
1506                         osfs.os_bfree >>= 1;
1507                         osfs.os_bavail >>= 1;
1508                 }
1509         }
1510
1511         sfs->f_blocks = osfs.os_blocks;
1512         sfs->f_bfree = osfs.os_bfree;
1513         sfs->f_bavail = osfs.os_bavail;
1514
1515         return 0;
1516 }
1517
1518 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1519 {
1520         struct ll_inode_info *lli;
1521         struct lov_stripe_md *lsm;
1522
1523         lli = ll_i2info(inode);
1524         LASSERT(lli->lli_size_sem_owner != current);
1525         down(&lli->lli_size_sem);
1526         LASSERT(lli->lli_size_sem_owner == NULL);
1527         lli->lli_size_sem_owner = current;
1528         lsm = lli->lli_smd;
1529         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1530                  lsm, lock_lsm);
1531         if (lock_lsm)
1532                 lov_stripe_lock(lsm);
1533 }
1534
1535 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1536 {
1537         struct ll_inode_info *lli;
1538         struct lov_stripe_md *lsm;
1539
1540         lli = ll_i2info(inode);
1541         lsm = lli->lli_smd;
1542         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1543                  lsm, unlock_lsm);
1544         if (unlock_lsm)
1545                 lov_stripe_unlock(lsm);
1546         LASSERT(lli->lli_size_sem_owner == current);
1547         lli->lli_size_sem_owner = NULL;
1548         up(&lli->lli_size_sem);
1549 }
1550
1551 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1552 {
1553         struct ll_inode_info *lli = ll_i2info(inode);
1554
1555         dump_lsm(D_INODE, lsm);
1556         dump_lsm(D_INODE, lli->lli_smd);
1557         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1558                  "lsm must be joined lsm %p\n", lsm);
1559         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1560         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1561                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1562         lli->lli_smd = lsm;
1563         lli->lli_maxbytes = lsm->lsm_maxbytes;
1564         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1565                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1566 }
1567
1568 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1569 {
1570         struct ll_inode_info *lli = ll_i2info(inode);
1571         struct mdt_body *body = md->body;
1572         struct lov_stripe_md *lsm = md->lsm;
1573         struct ll_sb_info *sbi = ll_i2sbi(inode);
1574
1575         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1576         if (lsm != NULL) {
1577                 if (lli->lli_smd == NULL) {
1578                         if (lsm->lsm_magic != LOV_MAGIC &&
1579                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1580                                 dump_lsm(D_ERROR, lsm);
1581                                 LBUG();
1582                         }
1583                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1584                                lsm, inode->i_ino, inode->i_generation, inode);
1585                         /* ll_inode_size_lock() requires it is only called
1586                          * with lli_smd != NULL or lock_lsm == 0 or we can
1587                          * race between lock/unlock.  bug 9547 */
1588                         lli->lli_smd = lsm;
1589                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1590                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1591                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1592                 } else {
1593                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1594                              lli->lli_smd->lsm_stripe_count ==
1595                                         lsm->lsm_stripe_count) {
1596                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1597                                         CERROR("lsm mismatch for inode %ld\n",
1598                                                 inode->i_ino);
1599                                         CERROR("lli_smd:\n");
1600                                         dump_lsm(D_ERROR, lli->lli_smd);
1601                                         CERROR("lsm:\n");
1602                                         dump_lsm(D_ERROR, lsm);
1603                                         LBUG();
1604                                 }
1605                         } else
1606                                 ll_replace_lsm(inode, lsm);
1607                 }
1608                 if (lli->lli_smd != lsm)
1609                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1610         }
1611
1612         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1613                 if (body->valid & OBD_MD_FLRMTPERM)
1614                         ll_update_remote_perm(inode, md->remote_perm);
1615         }
1616 #ifdef CONFIG_FS_POSIX_ACL
1617         else if (body->valid & OBD_MD_FLACL) {
1618                 spin_lock(&lli->lli_lock);
1619                 if (lli->lli_posix_acl)
1620                         posix_acl_release(lli->lli_posix_acl);
1621                 lli->lli_posix_acl = md->posix_acl;
1622                 spin_unlock(&lli->lli_lock);
1623         }
1624 #endif
1625         if (body->valid & OBD_MD_FLATIME &&
1626             body->atime > LTIME_S(inode->i_atime))
1627                 LTIME_S(inode->i_atime) = body->atime;
1628         
1629         /* mtime is always updated with ctime, but can be set in past.
1630            As write and utime(2) may happen within 1 second, and utime's
1631            mtime has a priority over write's one, so take mtime from mds 
1632            for the same ctimes. */
1633         if (body->valid & OBD_MD_FLCTIME &&
1634             body->ctime >= LTIME_S(inode->i_ctime)) {
1635                 LTIME_S(inode->i_ctime) = body->ctime;
1636                 if (body->valid & OBD_MD_FLMTIME) {
1637                         CDEBUG(D_INODE, "setting ino %lu mtime "
1638                                "from %lu to "LPU64"\n", inode->i_ino, 
1639                                LTIME_S(inode->i_mtime), body->mtime);
1640                         LTIME_S(inode->i_mtime) = body->mtime;
1641                 }
1642         }
1643         if (body->valid & OBD_MD_FLMODE)
1644                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1645         if (body->valid & OBD_MD_FLTYPE)
1646                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1647         if (S_ISREG(inode->i_mode)) {
1648                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1649         } else {
1650                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1651         }
1652 #ifdef HAVE_INODE_BLKSIZE
1653         inode->i_blksize = 1<<inode->i_blkbits;
1654 #endif
1655         if (body->valid & OBD_MD_FLUID)
1656                 inode->i_uid = body->uid;
1657         if (body->valid & OBD_MD_FLGID)
1658                 inode->i_gid = body->gid;
1659         if (body->valid & OBD_MD_FLFLAGS)
1660                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1661         if (body->valid & OBD_MD_FLNLINK)
1662                 inode->i_nlink = body->nlink;
1663         if (body->valid & OBD_MD_FLRDEV)
1664                 inode->i_rdev = old_decode_dev(body->rdev);
1665         if (body->valid & OBD_MD_FLSIZE) {
1666                 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1667                         if (lli->lli_flags & (LLIF_DONE_WRITING |
1668                                               LLIF_EPOCH_PENDING |
1669                                               LLIF_SOM_DIRTY))
1670                           CWARN("ino %lu flags %lu still has size authority!"
1671                                 "do not trust the size got from MDS\n", 
1672                                 inode->i_ino, lli->lli_flags);
1673                         else {
1674                                 i_size_write(inode, body->size);
1675                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1676                         }
1677                 } else {
1678                         i_size_write(inode, body->size);
1679                 }
1680
1681                 if (body->valid & OBD_MD_FLBLOCKS)
1682                         inode->i_blocks = body->blocks;
1683         }
1684
1685         if (body->valid & OBD_MD_FLID) {
1686                 /* FID shouldn't be changed! */
1687                 if (fid_is_sane(&lli->lli_fid)) {
1688                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1689                                  "Trying to change FID "DFID
1690                                  " to the "DFID", inode %lu/%u(%p)\n",
1691                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1692                                  inode->i_ino, inode->i_generation, inode);
1693                 } else 
1694                         lli->lli_fid = body->fid1;
1695         }
1696
1697         LASSERT(fid_seq(&lli->lli_fid) != 0);
1698
1699         if (body->valid & OBD_MD_FLMDSCAPA) {
1700                 LASSERT(md->mds_capa);
1701                 ll_add_capa(inode, md->mds_capa);
1702         }
1703         if (body->valid & OBD_MD_FLOSSCAPA) {
1704                 LASSERT(md->oss_capa);
1705                 ll_add_capa(inode, md->oss_capa);
1706         }
1707 }
1708
1709 static struct backing_dev_info ll_backing_dev_info = {
1710         .ra_pages       = 0,    /* No readahead */
1711 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1712         .capabilities   = 0,    /* Does contribute to dirty memory */
1713 #else
1714         .memory_backed  = 0,    /* Does contribute to dirty memory */
1715 #endif
1716 };
1717
1718 void ll_read_inode2(struct inode *inode, void *opaque)
1719 {
1720         struct lustre_md *md = opaque;
1721         struct ll_inode_info *lli = ll_i2info(inode);
1722         ENTRY;
1723
1724         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1725                inode->i_ino, inode->i_generation, inode);
1726
1727         ll_lli_init(lli);
1728
1729         LASSERT(!lli->lli_smd);
1730
1731         /* Core attributes from the MDS first.  This is a new inode, and
1732          * the VFS doesn't zero times in the core inode so we have to do
1733          * it ourselves.  They will be overwritten by either MDS or OST
1734          * attributes - we just need to make sure they aren't newer. */
1735         LTIME_S(inode->i_mtime) = 0;
1736         LTIME_S(inode->i_atime) = 0;
1737         LTIME_S(inode->i_ctime) = 0;
1738         inode->i_rdev = 0;
1739         ll_update_inode(inode, md);
1740
1741         /* OIDEBUG(inode); */
1742
1743         if (S_ISREG(inode->i_mode)) {
1744                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1745                 inode->i_op = &ll_file_inode_operations;
1746                 inode->i_fop = sbi->ll_fop;
1747                 inode->i_mapping->a_ops = &ll_aops;
1748                 EXIT;
1749         } else if (S_ISDIR(inode->i_mode)) {
1750                 inode->i_op = &ll_dir_inode_operations;
1751                 inode->i_fop = &ll_dir_operations;
1752                 inode->i_mapping->a_ops = &ll_dir_aops;
1753                 EXIT;
1754         } else if (S_ISLNK(inode->i_mode)) {
1755                 inode->i_op = &ll_fast_symlink_inode_operations;
1756                 EXIT;
1757         } else {
1758                 inode->i_op = &ll_special_inode_operations;
1759
1760                 init_special_inode(inode, inode->i_mode,
1761                                    kdev_t_to_nr(inode->i_rdev));
1762
1763                 /* initializing backing dev info. */
1764                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1765
1766                 EXIT;
1767         }
1768 }
1769
1770 void ll_delete_inode(struct inode *inode)
1771 {
1772         struct ll_sb_info *sbi = ll_i2sbi(inode);
1773         int rc;
1774         ENTRY;
1775
1776         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1777         if (rc) {
1778                 CERROR("fid_delete() failed, rc %d\n", rc);
1779         }
1780         clear_inode(inode);
1781
1782         EXIT;
1783 }
1784
1785 int ll_iocontrol(struct inode *inode, struct file *file,
1786                  unsigned int cmd, unsigned long arg)
1787 {
1788         struct ll_sb_info *sbi = ll_i2sbi(inode);
1789         struct ptlrpc_request *req = NULL;
1790         int rc, flags = 0;
1791         ENTRY;
1792
1793         switch(cmd) {
1794         case EXT3_IOC_GETFLAGS: {
1795                 struct mdt_body *body;
1796                 struct obd_capa *oc;
1797
1798                 oc = ll_mdscapa_get(inode);
1799                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1800                                 OBD_MD_FLFLAGS, 0, &req);
1801                 capa_put(oc);
1802                 if (rc) {
1803                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1804                         RETURN(-abs(rc));
1805                 }
1806
1807                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1808                                       sizeof(*body));
1809
1810                 /*Now the ext3 will be packed directly back to client,
1811                  *no need convert here*/
1812                 flags = body->flags;
1813
1814                 ptlrpc_req_finished (req);
1815
1816                 RETURN(put_user(flags, (int *)arg));
1817         }
1818         case EXT3_IOC_SETFLAGS: {
1819                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1820                 struct obd_info oinfo = { { { 0 } } };
1821                 struct md_op_data *op_data;
1822
1823                 if (get_user(flags, (int *)arg))
1824                         RETURN(-EFAULT);
1825
1826                 oinfo.oi_md = lsm;
1827                 OBDO_ALLOC(oinfo.oi_oa);
1828                 if (!oinfo.oi_oa)
1829                         RETURN(-ENOMEM);
1830
1831                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1832                                              LUSTRE_OPC_ANY, NULL);
1833                 if (IS_ERR(op_data))
1834                         RETURN(PTR_ERR(op_data));
1835
1836                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1837                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1838                 rc = md_setattr(sbi->ll_md_exp, op_data,
1839                                 NULL, 0, NULL, 0, &req, NULL);
1840                 ll_finish_md_op_data(op_data);
1841                 ptlrpc_req_finished(req);
1842                 if (rc || lsm == NULL) {
1843                         OBDO_FREE(oinfo.oi_oa);
1844                         RETURN(rc);
1845                 }
1846
1847                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1848                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1849                 oinfo.oi_oa->o_flags = flags;
1850                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | 
1851                                        OBD_MD_FLGROUP;
1852                 oinfo.oi_capa = ll_mdscapa_get(inode);
1853
1854                 obdo_from_inode(oinfo.oi_oa, inode,
1855                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1856                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1857                 capa_put(oinfo.oi_capa);
1858                 OBDO_FREE(oinfo.oi_oa);
1859                 if (rc) {
1860                         if (rc != -EPERM && rc != -EACCES)
1861                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1862                         RETURN(rc);
1863                 }
1864
1865                 inode->i_flags = ll_ext_to_inode_flags(flags |
1866                                                        MDS_BFLAG_EXT_FLAGS);
1867                 RETURN(0);
1868         }
1869         default:
1870                 RETURN(-ENOSYS);
1871         }
1872
1873         RETURN(0);
1874 }
1875
1876 int ll_flush_ctx(struct inode *inode)
1877 {
1878         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1879
1880         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1881
1882         obd_set_info_async(sbi->ll_md_exp,
1883                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1884                            0, NULL, NULL);
1885         obd_set_info_async(sbi->ll_dt_exp,
1886                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1887                            0, NULL, NULL);
1888         return 0;
1889 }
1890
1891 /* umount -f client means force down, don't save state */
1892 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1893 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1894 {
1895         struct super_block *sb = vfsmnt->mnt_sb;
1896 #else
1897 void ll_umount_begin(struct super_block *sb)
1898 {
1899 #endif
1900         struct lustre_sb_info *lsi = s2lsi(sb);
1901         struct ll_sb_info *sbi = ll_s2sbi(sb);
1902         struct obd_device *obd;
1903         struct obd_ioctl_data ioc_data = { 0 };
1904         ENTRY;
1905
1906 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1907         if (!(flags & MNT_FORCE)) {
1908                 EXIT;
1909                 return;
1910         }
1911 #endif
1912
1913         /* Tell the MGC we got umount -f */
1914         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1915
1916         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1917                sb->s_count, atomic_read(&sb->s_active));
1918
1919         obd = class_exp2obd(sbi->ll_md_exp);
1920         if (obd == NULL) {
1921                 CERROR("Invalid MDC connection handle "LPX64"\n",
1922                        sbi->ll_md_exp->exp_handle.h_cookie);
1923                 EXIT;
1924                 return;
1925         }
1926         obd->obd_force = 1;
1927         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1928                       &ioc_data, NULL);
1929
1930         obd = class_exp2obd(sbi->ll_dt_exp);
1931         if (obd == NULL) {
1932                 CERROR("Invalid LOV connection handle "LPX64"\n",
1933                        sbi->ll_dt_exp->exp_handle.h_cookie);
1934                 EXIT;
1935                 return;
1936         }
1937
1938         obd->obd_force = 1;
1939         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1940                       &ioc_data, NULL);
1941
1942         /* Really, we'd like to wait until there are no requests outstanding,
1943          * and then continue.  For now, we just invalidate the requests,
1944          * schedule, and hope.
1945          */
1946         schedule();
1947
1948         EXIT;
1949 }
1950
1951 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1952 {
1953         struct ll_sb_info *sbi = ll_s2sbi(sb);
1954         int err;
1955         __u32 read_only;
1956
1957         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1958                 read_only = *flags & MS_RDONLY;
1959                 err = obd_set_info_async(sbi->ll_md_exp,
1960                                          sizeof(KEY_READ_ONLY) - 1,
1961                                          KEY_READ_ONLY, sizeof(read_only),
1962                                          &read_only, NULL);
1963                 if (err) {
1964                         CERROR("Failed to change the read-only flag during "
1965                                "remount: %d\n", err);
1966                         return err;
1967                 }
1968
1969                 if (read_only)
1970                         sb->s_flags |= MS_RDONLY;
1971                 else
1972                         sb->s_flags &= ~MS_RDONLY;
1973         }
1974         return 0;
1975 }
1976
1977 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1978                   int offset, struct super_block *sb)
1979 {
1980         struct ll_sb_info *sbi = NULL;
1981         struct lustre_md md;
1982         int rc = 0;
1983         ENTRY;
1984
1985         LASSERT(*inode || sb);
1986         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1987         prune_deathrow(sbi, 1);
1988         memset(&md, 0, sizeof(struct lustre_md));
1989
1990         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
1991                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1992         if (rc)
1993                 RETURN(rc);
1994
1995         if (*inode) {
1996                 ll_update_inode(*inode, &md);
1997         } else {
1998                 LASSERT(sb != NULL);
1999
2000                 /*
2001                  * At this point server returns to client's same fid as client
2002                  * generated for creating. So using ->fid1 is okay here.
2003                  */
2004                 LASSERT(fid_is_sane(&md.body->fid1));
2005
2006                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2007                 if (*inode == NULL || is_bad_inode(*inode)) {
2008                         if (md.lsm)
2009                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2010 #ifdef CONFIG_FS_POSIX_ACL
2011                         if (md.posix_acl) {
2012                                 posix_acl_release(md.posix_acl);
2013                                 md.posix_acl = NULL;
2014                         }
2015 #endif
2016                         rc = -ENOMEM;
2017                         CERROR("new_inode -fatal: rc %d\n", rc);
2018                         GOTO(out, rc);
2019                 }
2020         }
2021
2022         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2023                          ll_i2info(*inode)->lli_smd);
2024 out:
2025         md_free_lustre_md(sbi->ll_md_exp, &md);
2026         RETURN(rc);
2027 }
2028
2029 char *llap_origins[] = {
2030         [LLAP_ORIGIN_UNKNOWN] = "--",
2031         [LLAP_ORIGIN_READPAGE] = "rp",
2032         [LLAP_ORIGIN_READAHEAD] = "ra",
2033         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2034         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2035 };
2036
2037 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2038                                              struct list_head *list)
2039 {
2040         struct ll_async_page *llap;
2041         struct list_head *pos;
2042
2043         list_for_each(pos, list) {
2044                 if (pos == &sbi->ll_pglist)
2045                         return NULL;
2046                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2047                 if (llap->llap_page == NULL)
2048                         continue;
2049                 return llap;
2050         }
2051         LBUG();
2052         return NULL;
2053 }
2054
2055 int ll_obd_statfs(struct inode *inode, void *arg)
2056 {
2057         struct ll_sb_info *sbi = NULL;
2058         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2059         struct lov_obd *lov = NULL;
2060         struct obd_statfs stat_buf = {0};
2061         char *buf = NULL;
2062         struct obd_ioctl_data *data = NULL;
2063         __u32 type, index;
2064         int len = 0, rc;
2065
2066         if (!inode || !(sbi = ll_i2sbi(inode)))
2067                 GOTO(out_statfs, rc = -EINVAL);
2068
2069         rc = obd_ioctl_getdata(&buf, &len, arg);
2070         if (rc)
2071                 GOTO(out_statfs, rc);
2072
2073         data = (void*)buf;
2074         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2075             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2076                 GOTO(out_statfs, rc = -EINVAL);
2077
2078         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2079         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2080
2081         if (type == LL_STATFS_MDC) {
2082                 if (index > 0)
2083                         GOTO(out_statfs, rc = -ENODEV);
2084                 client_obd = class_exp2obd(sbi->ll_md_exp);
2085         } else if (type == LL_STATFS_LOV) {
2086                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2087                 lov = &lov_obd->u.lov;
2088
2089                 if ((index >= lov->desc.ld_tgt_count))
2090                         GOTO(out_statfs, rc = -ENODEV);
2091                 if (!lov->lov_tgts[index])
2092                         /* Try again with the next index */
2093                         GOTO(out_statfs, rc = -EAGAIN);
2094
2095                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2096                 if (!lov->lov_tgts[index]->ltd_active)
2097                         GOTO(out_uuid, rc = -ENODATA);
2098         }
2099
2100         if (!client_obd)
2101                 GOTO(out_statfs, rc = -EINVAL);
2102
2103         rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2104         if (rc)
2105                 GOTO(out_statfs, rc);
2106
2107         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2108                 GOTO(out_statfs, rc = -EFAULT);
2109
2110 out_uuid:
2111         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2112                          data->ioc_plen2))
2113                 rc = -EFAULT;
2114
2115 out_statfs:
2116         if (buf)
2117                 obd_ioctl_freedata(buf, len);
2118         return rc;
2119 }
2120
2121 int ll_process_config(struct lustre_cfg *lcfg)
2122 {
2123         char *ptr;
2124         void *sb;
2125         struct lprocfs_static_vars lvars;
2126         unsigned long x; 
2127         int rc = 0;
2128
2129         lprocfs_init_vars(llite, &lvars);
2130
2131         /* The instance name contains the sb: lustre-client-aacfe000 */
2132         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2133         if (!ptr || !*(++ptr)) 
2134                 return -EINVAL;
2135         if (sscanf(ptr, "%lx", &x) != 1)
2136                 return -EINVAL;
2137         sb = (void *)x;
2138         /* This better be a real Lustre superblock! */
2139         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2140
2141         /* Note we have not called client_common_fill_super yet, so 
2142            proc fns must be able to handle that! */
2143         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2144                                       lcfg, sb);
2145         return(rc);
2146 }
2147
2148 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2149 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2150                                        struct inode *i1, struct inode *i2,
2151                                        const char *name, int namelen,
2152                                        int mode, __u32 opc, void *data)
2153 {
2154         LASSERT(i1 != NULL);
2155
2156         if (namelen > ll_i2sbi(i1)->ll_namelen)
2157                 return ERR_PTR(-ENAMETOOLONG);
2158         
2159         if (op_data == NULL)
2160                 OBD_ALLOC_PTR(op_data);
2161         
2162         if (op_data == NULL)
2163                 return ERR_PTR(-ENOMEM);
2164
2165         ll_i2gids(op_data->op_suppgids, i1, i2);
2166         op_data->op_fid1 = *ll_inode2fid(i1);
2167         op_data->op_capa1 = ll_mdscapa_get(i1);
2168
2169         if (i2) {
2170                 op_data->op_fid2 = *ll_inode2fid(i2);
2171                 op_data->op_capa2 = ll_mdscapa_get(i2);
2172         } else {
2173                 fid_zero(&op_data->op_fid2);
2174         }
2175
2176         op_data->op_name = name;
2177         op_data->op_namelen = namelen;
2178         op_data->op_mode = mode;
2179         op_data->op_mod_time = CURRENT_SECONDS;
2180         op_data->op_fsuid = current->fsuid;
2181         op_data->op_fsgid = current->fsgid;
2182         op_data->op_cap = current->cap_effective;
2183         op_data->op_bias = MDS_CHECK_SPLIT;
2184         op_data->op_opc = opc;
2185         op_data->op_mds = 0;
2186         op_data->op_data = data;
2187
2188         return op_data;
2189 }
2190
2191 void ll_finish_md_op_data(struct md_op_data *op_data)
2192 {
2193         capa_put(op_data->op_capa1);
2194         capa_put(op_data->op_capa2);
2195         OBD_FREE_PTR(op_data);
2196 }
2197
2198 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2199 {
2200         struct ll_sb_info *sbi = ll_i2sbi(inode);
2201         struct ptlrpc_request *req = NULL;
2202         struct mdt_body *body;
2203         char *cmd, *buf;
2204         struct obd_capa *oc;
2205         int rc, buflen;
2206         ENTRY;
2207
2208         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2209                 RETURN(-EBADE);
2210
2211         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2212
2213         OBD_ALLOC(cmd, ioc->cmd_len);
2214         if (!cmd)
2215                 RETURN(-ENOMEM);
2216         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2217                 GOTO(out, rc = -EFAULT);
2218
2219         oc = ll_mdscapa_get(inode);
2220         rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2221                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2222                          ioc->cmd_len, ioc->res_len, 0, &req);
2223         capa_put(oc);
2224         if (rc < 0) {
2225                 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2226                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2227                 GOTO(out, rc);
2228         }
2229
2230         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2231         LASSERT(body);
2232
2233         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2234         LASSERT(buflen <= ioc->res_len);
2235         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2236         LASSERT(buf);
2237         if (copy_to_user(ioc->res, buf, buflen))
2238                 GOTO(out, rc = -EFAULT);
2239         EXIT;
2240 out:
2241         if (req)
2242                 ptlrpc_req_finished(req);
2243         OBD_FREE(cmd, ioc->cmd_len);
2244         return rc;
2245 }
2246
2247 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2248 {
2249         struct ll_sb_info *sbi = ll_i2sbi(inode);
2250         struct ptlrpc_request *req = NULL;
2251         char *cmd, *buf;
2252         struct obd_capa *oc;
2253         int buflen, rc;
2254         ENTRY;
2255
2256         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2257                 RETURN(-EBADE);
2258
2259         if (!(sbi->ll_flags & LL_SBI_ACL)) 
2260                 RETURN(-EOPNOTSUPP);
2261
2262         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2263
2264         OBD_ALLOC(cmd, ioc->cmd_len);
2265         if (!cmd)
2266                 RETURN(-ENOMEM);
2267         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2268                 GOTO(out, rc = -EFAULT);
2269
2270         oc = ll_mdscapa_get(inode);
2271         rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2272                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2273                          ioc->cmd_len, ioc->res_len, 0, &req);
2274         capa_put(oc);
2275         if (rc) {
2276                 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2277                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2278                 GOTO(out, rc);
2279         }
2280
2281         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2282         LASSERT(buflen <= ioc->res_len);
2283         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2284         LASSERT(buf);
2285         if (copy_to_user(ioc->res, buf, buflen))
2286                 GOTO(out, rc = -EFAULT);
2287         EXIT;
2288 out:
2289         if (req)
2290                 ptlrpc_req_finished(req);
2291         OBD_FREE(cmd, ioc->cmd_len);
2292         return rc;
2293 }