Whamcloud - gitweb
- KML fixes and updates due to SMFS plugin API.
[fs/lustre-release.git] / lustre / mds / mds_fs.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  *  mds/mds_fs.c
5  *  Lustre Metadata Server (MDS) filesystem interface code
6  *
7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
8  *   Author: Andreas Dilger <adilger@clusterfs.com>
9  *
10  *   This file is part of Lustre, http://www.lustre.org.
11  *
12  *   Lustre is free software; you can redistribute it and/or
13  *   modify it under the terms of version 2 of the GNU General Public
14  *   License as published by the Free Software Foundation.
15  *
16  *   Lustre is distributed in the hope that it will be useful,
17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  *   GNU General Public License for more details.
20  *
21  *   You should have received a copy of the GNU General Public License
22  *   along with Lustre; if not, write to the Free Software
23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24  */
25
26 #ifndef EXPORT_SYMTAB
27 # define EXPORT_SYMTAB
28 #endif
29 #define DEBUG_SUBSYSTEM S_MDS
30
31 #include <linux/module.h>
32 #include <linux/kmod.h>
33 #include <linux/version.h>
34 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
35 #include <linux/mount.h>
36 #endif
37 #include <linux/lustre_mds.h>
38 #include <linux/obd_class.h>
39 #include <linux/obd_support.h>
40 #include <linux/lustre_lib.h>
41 #include <linux/lustre_fsfilt.h>
42 #include <libcfs/list.h>
43
44 #include <linux/lustre_smfs.h>
45 #include "mds_internal.h"
46
47 /* This limit is arbitrary, but for now we fit it in 1 page (32k clients) */
48 #define MDS_MAX_CLIENTS (PAGE_SIZE * 8)
49
50 #define LAST_RCVD "last_rcvd"
51 #define LOV_OBJID "lov_objid"
52 #define LAST_FID  "last_fid"
53 #define VIRT_FID  "virt_fid"
54
55 /* Add client data to the MDS.  We use a bitmap to locate a free space
56  * in the last_rcvd file if cl_off is -1 (i.e. a new client).
57  * Otherwise, we have just read the data from the last_rcvd file and
58  * we know its offset.
59  */
60 int mds_client_add(struct obd_device *obd, struct mds_obd *mds,
61                    struct mds_export_data *med, int cl_idx)
62 {
63         unsigned long *bitmap = mds->mds_client_bitmap;
64         int new_client = (cl_idx == -1);
65         ENTRY;
66
67         LASSERT(bitmap != NULL);
68
69         /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
70         if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
71                 RETURN(0);
72
73         /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
74          * there's no need for extra complication here
75          */
76         if (new_client) {
77                 cl_idx = find_first_zero_bit(bitmap, MDS_MAX_CLIENTS);
78         repeat:
79                 if (cl_idx >= MDS_MAX_CLIENTS) {
80                         CERROR("no room for clients - fix MDS_MAX_CLIENTS\n");
81                         return -ENOMEM;
82                 }
83                 if (test_and_set_bit(cl_idx, bitmap)) {
84                         cl_idx = find_next_zero_bit(bitmap, MDS_MAX_CLIENTS,
85                                                     cl_idx);
86                         goto repeat;
87                 }
88         } else {
89                 if (test_and_set_bit(cl_idx, bitmap)) {
90                         CERROR("MDS client %d: bit already set in bitmap!!\n",
91                                cl_idx);
92                         LBUG();
93                 }
94         }
95
96         CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
97                cl_idx, med->med_mcd->mcd_uuid);
98
99         med->med_idx = cl_idx;
100         med->med_off = le32_to_cpu(mds->mds_server_data->msd_client_start) +
101                 (cl_idx * le16_to_cpu(mds->mds_server_data->msd_client_size));
102
103         if (new_client) {
104                 struct file *file = mds->mds_rcvd_filp;
105                 struct lvfs_run_ctxt saved;
106                 loff_t off = med->med_off;
107                 int rc;
108
109                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
110                 rc = fsfilt_write_record(obd, file, med->med_mcd,
111                                          sizeof(*med->med_mcd), &off, 1);
112                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
113
114                 if (rc)
115                         return rc;
116                 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
117                        med->med_idx, med->med_off,
118                        (unsigned int)sizeof(*med->med_mcd));
119         }
120         return 0;
121 }
122
123 int mds_client_free(struct obd_export *exp, int clear_client)
124 {
125         struct mds_export_data *med = &exp->exp_mds_data;
126         struct mds_obd *mds = &exp->exp_obd->u.mds;
127         unsigned long *bitmap = mds->mds_client_bitmap;
128         struct obd_device *obd = exp->exp_obd;
129         struct mds_client_data zero_mcd;
130         struct lvfs_run_ctxt saved;
131         int rc;
132
133         if (!med->med_mcd)
134                 RETURN(0);
135
136         /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
137         if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
138                 GOTO(free_and_out, 0);
139
140         CDEBUG(D_INFO, "freeing client at idx %u (%lld)with UUID '%s'\n",
141                med->med_idx, med->med_off, med->med_mcd->mcd_uuid);
142
143         LASSERT(bitmap);
144
145         /* Clear the bit _after_ zeroing out the client so we don't
146            race with mds_client_add and zero out new clients.*/
147         if (!test_bit(med->med_idx, bitmap)) {
148                 CERROR("MDS client %u: bit already clear in bitmap!!\n",
149                        med->med_idx);
150                 LBUG();
151         }
152
153         if (clear_client) {
154                 memset(&zero_mcd, 0, sizeof zero_mcd);
155                 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
156                 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
157                                          sizeof(zero_mcd), &med->med_off, 1);
158                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
159
160                 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
161                        "zeroing out client %s idx %u in %s rc %d\n",
162                        med->med_mcd->mcd_uuid, med->med_idx, LAST_RCVD, rc);
163         }
164
165         if (!test_and_clear_bit(med->med_idx, bitmap)) {
166                 CERROR("MDS client %u: bit already clear in bitmap!!\n",
167                        med->med_idx);
168                 LBUG();
169         }
170
171
172         /* Make sure the server's last_transno is up to date. Do this
173          * after the client is freed so we know all the client's
174          * transactions have been committed. */
175         mds_update_server_data(exp->exp_obd, 1);
176
177 free_and_out:
178         OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
179         med->med_mcd = NULL;
180         return 0;
181 }
182
183 static int mds_server_free_data(struct mds_obd *mds)
184 {
185         OBD_FREE(mds->mds_client_bitmap, MDS_MAX_CLIENTS / 8);
186         OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
187         mds->mds_server_data = NULL;
188
189         return 0;
190 }
191
192 static int mds_read_last_fid(struct obd_device *obd, struct file *file)
193 {
194         int rc = 0;
195         loff_t off = 0;
196         struct mds_obd *mds = &obd->u.mds;
197         unsigned long last_fid_size = file->f_dentry->d_inode->i_size;
198         ENTRY;
199
200         if (last_fid_size == 0) {
201                 CWARN("%s: initializing new %s\n", obd->obd_name,
202                       file->f_dentry->d_name.name);
203
204                 /* 
205                  * as fid is used for forming res_id for locking, it should not
206                  * be zero. This will keep us out of lots possible problems,
207                  * asserts, etc.
208                  */
209                 mds_set_last_fid(obd, 0);
210         } else {
211                 __u64 lastfid;
212                 
213                 rc = fsfilt_read_record(obd, file, &lastfid,
214                                         sizeof(lastfid), &off);
215                 if (rc) {
216                         CERROR("error reading MDS %s: rc = %d\n",
217                                file->f_dentry->d_name.name, rc);
218                         RETURN(rc);
219                 }
220
221                 /* 
222                  * make sure, that fid is up-to-date.
223                  */
224                 mds_set_last_fid(obd, lastfid);
225         }
226
227         CDEBUG(D_INODE, "%s: server last_fid: "LPU64"\n",
228                obd->obd_name, mds->mds_last_fid);
229
230         rc = mds_update_last_fid(obd, NULL, 1);
231         RETURN(rc);
232 }
233
234 static int mds_read_last_rcvd(struct obd_device *obd, struct file *file)
235 {
236         unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
237         struct mds_obd *mds = &obd->u.mds;
238         struct mds_server_data *msd = NULL;
239         struct mds_client_data *mcd = NULL;
240         loff_t off = 0;
241         __u64 mount_count;
242         int cl_idx, rc = 0;
243         ENTRY;
244
245         /* ensure padding in the struct is the correct size */
246         LASSERT(offsetof(struct mds_server_data, msd_padding) +
247                 sizeof(msd->msd_padding) == MDS_LR_SERVER_SIZE);
248         LASSERT(offsetof(struct mds_client_data, mcd_padding) +
249                 sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE);
250
251         OBD_ALLOC_WAIT(msd, sizeof(*msd));
252         if (!msd)
253                 RETURN(-ENOMEM);
254
255         OBD_ALLOC_WAIT(mds->mds_client_bitmap, MDS_MAX_CLIENTS / 8);
256         if (!mds->mds_client_bitmap) {
257                 OBD_FREE(msd, sizeof(*msd));
258                 RETURN(-ENOMEM);
259         }
260
261         mds->mds_server_data = msd;
262
263         if (last_rcvd_size == 0) {
264                 CWARN("%s: initializing new %s\n", obd->obd_name,
265                       file->f_dentry->d_name.name);
266
267                 memcpy(msd->msd_uuid, obd->obd_uuid.uuid,sizeof(msd->msd_uuid));
268                 msd->msd_last_transno = 0;
269                 mount_count = msd->msd_mount_count = 0;
270                 msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE);
271                 msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START);
272                 msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE);
273                 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
274         } else {
275                 rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off);
276                 if (rc) {
277                         CERROR("error reading MDS %s: rc = %d\n",
278                                file->f_dentry->d_name.name, rc);
279                         GOTO(err_msd, rc);
280                 }
281                 if (strcmp(msd->msd_uuid, obd->obd_uuid.uuid) != 0) {
282                         CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
283                                obd->obd_uuid.uuid, msd->msd_uuid);
284                         GOTO(err_msd, rc = -EINVAL);
285                 }
286                 mount_count = le64_to_cpu(msd->msd_mount_count);
287         }
288         if (msd->msd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) {
289                 CERROR("unsupported incompat feature %x\n",
290                        le32_to_cpu(msd->msd_feature_incompat) &
291                        ~MDS_INCOMPAT_SUPP);
292                 GOTO(err_msd, rc = -EINVAL);
293         }
294         /* XXX updating existing b_devel fs only, can be removed in future */
295         msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
296         if (msd->msd_feature_rocompat & ~cpu_to_le32(MDS_ROCOMPAT_SUPP)) {
297                 CERROR("unsupported read-only feature %x\n",
298                        le32_to_cpu(msd->msd_feature_rocompat) &
299                        ~MDS_ROCOMPAT_SUPP);
300                 /* Do something like remount filesystem read-only */
301                 GOTO(err_msd, rc = -EINVAL);
302         }
303
304         mds->mds_last_transno = le64_to_cpu(msd->msd_last_transno);
305
306         CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
307                obd->obd_name, mds->mds_last_transno);
308         CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
309                obd->obd_name, mount_count + 1);
310         CDEBUG(D_INODE, "%s: server data size: %u\n",
311                obd->obd_name, le32_to_cpu(msd->msd_server_size));
312         CDEBUG(D_INODE, "%s: per-client data start: %u\n",
313                obd->obd_name, le32_to_cpu(msd->msd_client_start));
314         CDEBUG(D_INODE, "%s: per-client data size: %u\n",
315                obd->obd_name, le32_to_cpu(msd->msd_client_size));
316         CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
317                obd->obd_name, last_rcvd_size);
318         CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
319                last_rcvd_size <= le32_to_cpu(msd->msd_client_start) ? 0 :
320                (last_rcvd_size - le32_to_cpu(msd->msd_client_start)) /
321                 le16_to_cpu(msd->msd_client_size));
322
323         /* When we do a clean MDS shutdown, we save the last_transno into
324          * the header.  If we find clients with higher last_transno values
325          * then those clients may need recovery done. */
326         for (cl_idx = 0, off = le32_to_cpu(msd->msd_client_start);
327              off < last_rcvd_size; cl_idx++) {
328                 __u64 last_transno;
329                 struct obd_export *exp;
330                 struct mds_export_data *med;
331
332                 if (!mcd) {
333                         OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
334                         if (!mcd)
335                                 GOTO(err_client, rc = -ENOMEM);
336                 }
337
338                 /* Don't assume off is incremented properly by
339                  * fsfilt_read_record(), in case sizeof(*mcd)
340                  * isn't the same as msd->msd_client_size.  */
341                 off = le32_to_cpu(msd->msd_client_start) +
342                         cl_idx * le16_to_cpu(msd->msd_client_size);
343                 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
344                 if (rc) {
345                         CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
346                                file->f_dentry->d_name.name, cl_idx, off, rc);
347                         break; /* read error shouldn't cause startup to fail */
348                 }
349
350                 if (mcd->mcd_uuid[0] == '\0') {
351                         CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
352                                cl_idx);
353                         continue;
354                 }
355
356                 last_transno = le64_to_cpu(mcd->mcd_last_transno) >
357                                le64_to_cpu(mcd->mcd_last_close_transno) ?
358                                le64_to_cpu(mcd->mcd_last_transno) :
359                                le64_to_cpu(mcd->mcd_last_close_transno);
360
361                 /* These exports are cleaned up by mds_disconnect(), so they
362                  * need to be set up like real exports as mds_connect() does.
363                  */
364                 CDEBUG(D_HA|D_WARNING,"RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
365                        " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
366                        last_transno, le64_to_cpu(msd->msd_last_transno),
367                        mcd->mcd_last_xid);
368
369                 exp = class_new_export(obd);
370                 if (exp == NULL)
371                         GOTO(err_client, rc = -ENOMEM);
372
373                 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
374                        sizeof exp->exp_client_uuid.uuid);
375                 med = &exp->exp_mds_data;
376                 med->med_mcd = mcd;
377                 mds_client_add(obd, mds, med, cl_idx);
378                 /* create helper if export init gets more complex */
379                 INIT_LIST_HEAD(&med->med_open_head);
380                 spin_lock_init(&med->med_open_lock);
381
382                 mcd = NULL;
383                 exp->exp_connected = 0;
384                 exp->exp_req_replay_needed = 1;
385                 obd->obd_recoverable_clients++;
386                 obd->obd_max_recoverable_clients++;
387
388                 /* track clients to separate req replay
389                  * from lock replay. bug 6063 */
390                 atomic_inc(&obd->obd_req_replay_clients);
391                 exp->exp_req_replay_needed = 1;
392                 atomic_inc(&obd->obd_lock_replay_clients);
393                 exp->exp_lock_replay_needed = 1;
394                 
395                 class_export_put(exp);
396
397                 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
398                        cl_idx, last_transno);
399
400                 if (last_transno > mds->mds_last_transno)
401                        mds->mds_last_transno = last_transno;
402         }
403         if (mcd)
404                 OBD_FREE(mcd, sizeof(*mcd));
405         obd->obd_last_committed = mds->mds_last_transno;
406         if (obd->obd_recoverable_clients) {
407                 CWARN("RECOVERY: service %s, %d recoverable clients, "
408                       "last_transno "LPU64"\n", obd->obd_name,
409                       obd->obd_recoverable_clients, mds->mds_last_transno);
410                 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
411                 target_start_recovery_thread(obd, mds_handle);
412                 obd->obd_recovery_start = LTIME_S(CURRENT_TIME);
413         }
414         
415         mds->mds_mount_count = mount_count + 1;
416         msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
417
418         /* save it, so mount count and last_transno is current */
419         rc = mds_update_server_data(obd, 1);
420         if (rc)
421                 GOTO(err_client, rc);
422
423         RETURN(0);
424
425 err_client:
426         class_disconnect_exports(obd, 0);
427 err_msd:
428         mds_server_free_data(mds);
429         RETURN(rc);
430 }
431
432 static int mds_fs_post_setup(struct obd_device *obd)
433 {
434         struct mds_obd *mds = &obd->u.mds;
435         struct dentry *dentry;
436         int rc = 0;
437         ENTRY;
438        
439         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
440         if (IS_ERR(dentry)) {
441                 CERROR("Can't find ROOT, err = %d\n",
442                        (int)PTR_ERR(dentry));
443                 RETURN(PTR_ERR(dentry));
444         }
445         
446         rc = fsfilt_post_setup(obd, dentry);
447         if (rc)
448                 goto out_dentry;
449
450         LASSERT(dentry->d_inode != NULL);
451         
452         fsfilt_set_fs_flags(obd, dentry->d_inode, 
453                             SM_DO_REC | SM_DO_COW);
454         
455         fsfilt_set_fs_flags(obd, mds->mds_pending_dir->d_inode, 
456                             SM_DO_REC | SM_DO_COW);
457         
458         fsfilt_set_mds_flags(obd, mds->mds_sb);
459
460 out_dentry:
461         l_dput(dentry);
462         RETURN(rc); 
463 }
464
465 /*
466  * sets up root inode lustre_id. It tries to read it first from root inode and
467  * if it is not there, new rootid is allocated and saved there.
468  */
469 int mds_fs_setup_rootid(struct obd_device *obd)
470 {
471         int rc = 0;
472         void *handle;
473         struct inode *inode;
474         struct dentry *dentry;
475         struct mds_obd *mds = &obd->u.mds;
476         ENTRY;
477
478         /* getting root directory and setup its fid. */
479         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
480         if (IS_ERR(dentry)) {
481                 CERROR("Can't find ROOT by "DLID4", err = %d\n",
482                        OLID4(&mds->mds_rootid), (int)PTR_ERR(dentry));
483                 RETURN(PTR_ERR(dentry));
484         }
485
486         inode = dentry->d_inode;
487         LASSERT(dentry->d_inode);
488
489         rc = mds_pack_inode2id(obd, &mds->mds_rootid, inode, 1);
490         if (rc < 0) {
491                 if (rc != -ENODATA)
492                         GOTO(out_dentry, rc);
493         } else {
494                 /*
495                  * rootid is filled by mds_read_inode_sid(), so we do not need
496                  * to allocate it and update. The only thing we need to check is
497                  * mds_num.
498                  */
499                 LASSERT(id_group(&mds->mds_rootid) == mds->mds_num);
500                 mds_set_last_fid(obd, id_fid(&mds->mds_rootid));
501                 GOTO(out_dentry, rc);
502         }
503
504         /* allocating new one, as it is not found in root inode. */
505         handle = fsfilt_start(obd, inode,
506                               FSFILT_OP_SETATTR, NULL);
507         
508         if (IS_ERR(handle)) {
509                 rc = PTR_ERR(handle);
510                 CERROR("fsfilt_start() failed, rc = %d\n", rc);
511                 GOTO(out_dentry, rc);
512         }
513         
514         down(&inode->i_sem);
515         rc = mds_alloc_inode_sid(obd, inode, handle, &mds->mds_rootid);
516         up(&inode->i_sem);
517         
518         if (rc) {
519                 CERROR("mds_alloc_inode_sid() failed, rc = %d\n",
520                        rc);
521                 GOTO(out_dentry, rc);
522         }
523
524         rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
525         if (rc)
526                 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
527
528         EXIT;
529 out_dentry:
530         l_dput(dentry);
531         if (rc == 0)
532                 CWARN("%s: rootid: "DLID4"\n", obd->obd_name,
533                       OLID4(&mds->mds_rootid));
534         return rc;
535 }
536
537 static int mds_update_virtid_fid(struct obd_device *obd,
538                                  void *handle, int force_sync)
539 {
540         struct mds_obd *mds = &obd->u.mds;
541         struct file *filp = mds->mds_virtid_filp;
542         struct lvfs_run_ctxt saved;
543         loff_t off = 0;
544         int rc = 0;
545         ENTRY;
546
547         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
548         rc = fsfilt_write_record(obd, filp, &mds->mds_virtid_fid,
549                                  sizeof(mds->mds_virtid_fid),
550                                  &off, force_sync);
551         if (rc) {
552                 CERROR("error writing MDS virtid_fid #"LPU64
553                        ", err = %d\n", mds->mds_virtid_fid, rc);
554         }
555                 
556         CDEBUG(D_SUPER, "wrote virtid fid #"LPU64" at idx "
557                "%llu: err = %d\n", mds->mds_virtid_fid,
558                off, rc);
559         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
560
561         RETURN(rc);
562 }
563
564 static int mds_read_virtid_fid(struct obd_device *obd,
565                                struct file *file)
566 {
567         int rc = 0;
568         loff_t off = 0;
569         struct mds_obd *mds = &obd->u.mds;
570         unsigned long virtid_fid_size = file->f_dentry->d_inode->i_size;
571         ENTRY;
572
573         if (virtid_fid_size == 0) {
574                 mds->mds_virtid_fid = mds_alloc_fid(obd);
575         } else {
576                 rc = fsfilt_read_record(obd, file, &mds->mds_virtid_fid,
577                                         sizeof(mds->mds_virtid_fid), &off);
578                 if (rc) {
579                         CERROR("error reading MDS %s: rc = %d\n",
580                                file->f_dentry->d_name.name, rc);
581                         RETURN(rc);
582                 }
583         }
584         rc = mds_update_virtid_fid(obd, NULL, 1);
585
586         RETURN(rc);
587 }
588
589 /*
590  * initializes lustre_id for virtual id directory, it is needed sometimes, as it
591  * is possible that it will be the parent for object an operations is going to
592  * be performed on.
593  */
594 int mds_fs_setup_virtid(struct obd_device *obd)
595 {
596         int rc = 0;
597         void *handle;
598         struct lustre_id sid;
599         struct mds_obd *mds = &obd->u.mds;
600         struct inode *inode = mds->mds_id_dir->d_inode;
601         ENTRY;
602
603         handle = fsfilt_start(obd, inode,
604                               FSFILT_OP_SETATTR, NULL);
605         
606         if (IS_ERR(handle)) {
607                 rc = PTR_ERR(handle);
608                 CERROR("fsfilt_start() failed, rc = %d\n", rc);
609                 RETURN(rc);
610         }
611
612         id_group(&sid) = mds->mds_num;
613         id_fid(&sid) = mds->mds_virtid_fid;
614
615         id_ino(&sid) = inode->i_ino;
616         id_gen(&sid) = inode->i_generation;
617         id_type(&sid) = (S_IFMT & inode->i_mode);
618
619         down(&inode->i_sem);
620         rc = mds_update_inode_sid(obd, inode, handle, &sid);
621         up(&inode->i_sem);
622
623         if (rc) {
624                 CERROR("mds_update_inode_sid() failed, rc = %d\n",
625                        rc);
626                 RETURN(rc);
627         }
628
629         rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
630         if (rc) {
631                 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
632                 RETURN(rc);
633         }
634
635         RETURN(rc);
636 }
637
638 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
639 {
640         struct mds_obd *mds = &obd->u.mds;
641         struct lvfs_run_ctxt saved;
642         struct dentry *dentry;
643         struct file *file;
644         int rc;
645         ENTRY;
646
647         rc = cleanup_group_info();
648         if (rc)
649                 RETURN(rc);
650
651         mds->mds_vfsmnt = mnt;
652         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
653
654         fsfilt_setup(obd, mds->mds_sb);
655
656         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
657         obd->obd_lvfs_ctxt.pwdmnt = mnt;
658         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
659         obd->obd_lvfs_ctxt.fs = get_ds();
660         obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
661
662         /* setup the directory tree */
663         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
664         dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
665         if (IS_ERR(dentry)) {
666                 rc = PTR_ERR(dentry);
667                 CERROR("cannot create ROOT directory: rc = %d\n", rc);
668                 GOTO(err_pop, rc);
669         }
670
671         mdc_pack_id(&mds->mds_rootid, dentry->d_inode->i_ino,
672                     dentry->d_inode->i_generation, S_IFDIR, 0, 0);
673
674         dput(dentry);
675         
676         dentry = lookup_one_len("__iopen__", current->fs->pwd,
677                                 strlen("__iopen__"));
678         if (IS_ERR(dentry)) {
679                 rc = PTR_ERR(dentry);
680                 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
681                 GOTO(err_pop, rc);
682         }
683         mds->mds_id_de = dentry;
684         if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
685                 rc = -ENOENT;
686                 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
687                 GOTO(err_id_de, rc);
688         }
689
690         dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
691         if (IS_ERR(dentry)) {
692                 rc = PTR_ERR(dentry);
693                 CERROR("cannot create PENDING directory: rc = %d\n", rc);
694                 GOTO(err_id_de, rc);
695         }
696         mds->mds_pending_dir = dentry;
697       
698         dentry = simple_mkdir(current->fs->pwd, "LOGS", 0777, 1);
699         if (IS_ERR(dentry)) {
700                 rc = PTR_ERR(dentry);
701                 CERROR("cannot create LOGS directory: rc = %d\n", rc);
702                 GOTO(err_pending, rc);
703         }
704         mds->mds_logs_dir = dentry;
705
706         dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
707         if (IS_ERR(dentry)) {
708                 rc = PTR_ERR(dentry);
709                 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
710                 GOTO(err_logs, rc);
711         }
712         mds->mds_objects_dir = dentry;
713
714         dentry = simple_mkdir(current->fs->pwd, "FIDS", 0777, 1);
715         if (IS_ERR(dentry)) {
716                 rc = PTR_ERR(dentry);
717                 CERROR("cannot create FIDS directory: rc = %d\n", rc);
718                 GOTO(err_objects, rc);
719         }
720         mds->mds_id_dir = dentry;
721
722         dentry = simple_mkdir(current->fs->pwd, "UNNAMED", 0777, 1);
723         if (IS_ERR(dentry)) {
724                 rc = PTR_ERR(dentry);
725                 CERROR("cannot create UNNAMED directory: rc = %d\n", rc);
726                 GOTO(err_unnamed, rc);
727         }
728         mds->mds_unnamed_dir = dentry;
729
730         /* open and test the last rcvd file */
731         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
732         if (IS_ERR(file)) {
733                 rc = PTR_ERR(file);
734                 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
735                 GOTO(err_id_dir, rc = PTR_ERR(file));
736         }
737         mds->mds_rcvd_filp = file;
738         
739         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
740                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
741                        file->f_dentry->d_inode->i_mode);
742                 GOTO(err_last_rcvd, rc = -ENOENT);
743         }
744
745         rc = mds_read_last_rcvd(obd, file);
746         if (rc) {
747                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
748                 GOTO(err_last_rcvd, rc);
749         }
750
751         /* open and test last fid file */
752         file = filp_open(LAST_FID, O_RDWR | O_CREAT, 0644);
753         if (IS_ERR(file)) {
754                 rc = PTR_ERR(file);
755                 CERROR("cannot open/create %s file: rc = %d\n",
756                        LAST_FID, rc);
757                 GOTO(err_client, rc = PTR_ERR(file));
758         }
759         mds->mds_fid_filp = file;
760         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
761                 CERROR("%s is not a regular file!: mode = %o\n",
762                        LAST_FID, file->f_dentry->d_inode->i_mode);
763                 GOTO(err_last_fid, rc = -ENOENT);
764         }
765
766         rc = mds_read_last_fid(obd, file);
767         if (rc) {
768                 CERROR("cannot read %s: rc = %d\n", LAST_FID, rc);
769                 GOTO(err_last_fid, rc);
770         }
771
772         /* open and test virtid fid file */
773         file = filp_open(VIRT_FID, O_RDWR | O_CREAT, 0644);
774         if (IS_ERR(file)) {
775                 rc = PTR_ERR(file);
776                 CERROR("cannot open/create %s file: rc = %d\n",
777                        VIRT_FID, rc);
778                 GOTO(err_last_fid, rc = PTR_ERR(file));
779         }
780         mds->mds_virtid_filp = file;
781         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
782                 CERROR("%s is not a regular file!: mode = %o\n",
783                        VIRT_FID, file->f_dentry->d_inode->i_mode);
784                 GOTO(err_virtid_fid, rc = -ENOENT);
785         }
786
787         rc = mds_read_virtid_fid(obd, file);
788         if (rc) {
789                 CERROR("cannot read %s: rc = %d\n", VIRT_FID, rc);
790                 GOTO(err_virtid_fid, rc);
791         }
792         
793         /* open and test the lov objid file */
794         file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
795         if (IS_ERR(file)) {
796                 rc = PTR_ERR(file);
797                 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
798                 GOTO(err_last_fid, rc = PTR_ERR(file));
799         }
800         mds->mds_dt_objid_filp = file;
801         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
802                 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
803                        file->f_dentry->d_inode->i_mode);
804                 GOTO(err_lov_objid, rc = -ENOENT);
805         }
806 err_pop:
807         if (!rc) {
808                 rc = mds_fs_post_setup(obd);
809                 if (rc)
810                         CERROR("can not post setup fsfilt\n");        
811         }
812         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
813         return rc;
814
815 err_lov_objid:
816         if (mds->mds_dt_objid_filp && filp_close(mds->mds_dt_objid_filp, 0))
817                 CERROR("can't close %s after error\n", LOV_OBJID);
818 err_virtid_fid:
819         if (mds->mds_virtid_filp && filp_close(mds->mds_virtid_filp, 0))
820                 CERROR("can't close %s after error\n", VIRT_FID);
821 err_last_fid:
822         if (mds->mds_fid_filp && filp_close(mds->mds_fid_filp, 0))
823                 CERROR("can't close %s after error\n", LAST_FID);
824 err_client:
825         class_disconnect_exports(obd, 0);
826 err_last_rcvd:
827         if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
828                 CERROR("can't close %s after error\n", LAST_RCVD);
829 err_unnamed:
830         dput(mds->mds_unnamed_dir);
831 err_id_dir:
832         dput(mds->mds_id_dir);
833 err_objects:
834         dput(mds->mds_objects_dir);
835 err_logs:
836         dput(mds->mds_logs_dir);
837 err_pending:
838         dput(mds->mds_pending_dir);
839 err_id_de:
840         dput(mds->mds_id_de);
841         goto err_pop;
842 }
843
844 static int  mds_fs_post_cleanup(struct obd_device *obd)
845 {
846         int    rc = 0;
847         rc = fsfilt_post_cleanup(obd);
848         return rc; 
849 }
850
851 int mds_fs_cleanup(struct obd_device *obd, int flags)
852 {
853         struct mds_obd *mds = &obd->u.mds;
854         struct lvfs_run_ctxt saved;
855         int rc = 0;
856
857         if (flags & OBD_OPT_FAILOVER)
858                 CERROR("%s: shutting down for failover; client state will"
859                        " be preserved.\n", obd->obd_name);
860
861         class_disconnect_exports(obd, flags); /* cleans up client info too */
862         target_cleanup_recovery(obd);
863         mds_server_free_data(mds);
864
865         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
866         if (mds->mds_virtid_filp) {
867                 rc = filp_close(mds->mds_virtid_filp, 0);
868                 mds->mds_virtid_filp = NULL;
869                 if (rc)
870                         CERROR("%s file won't close, rc = %d\n", VIRT_FID, rc);
871         }
872         if (mds->mds_fid_filp) {
873                 rc = filp_close(mds->mds_fid_filp, 0);
874                 mds->mds_fid_filp = NULL;
875                 if (rc)
876                         CERROR("%s file won't close, rc = %d\n", LAST_FID, rc);
877         }
878         if (mds->mds_rcvd_filp) {
879                 rc = filp_close(mds->mds_rcvd_filp, 0);
880                 mds->mds_rcvd_filp = NULL;
881                 if (rc)
882                         CERROR("%s file won't close, rc = %d\n", LAST_RCVD, rc);
883         }
884         if (mds->mds_dt_objid_filp) {
885                 rc = filp_close(mds->mds_dt_objid_filp, 0);
886                 mds->mds_dt_objid_filp = NULL;
887                 if (rc)
888                         CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
889         }
890         if (mds->mds_unnamed_dir != NULL) {
891                 l_dput(mds->mds_unnamed_dir);
892                 mds->mds_unnamed_dir = NULL;
893         }
894         if (mds->mds_id_dir != NULL) {
895                 l_dput(mds->mds_id_dir);
896                 mds->mds_id_dir = NULL;
897         }
898         if (mds->mds_objects_dir != NULL) {
899                 l_dput(mds->mds_objects_dir);
900                 mds->mds_objects_dir = NULL;
901         }
902         if (mds->mds_logs_dir) {
903                 l_dput(mds->mds_logs_dir);
904                 mds->mds_logs_dir = NULL;
905         }
906         if (mds->mds_pending_dir) {
907                 l_dput(mds->mds_pending_dir);
908                 mds->mds_pending_dir = NULL;
909         }
910         rc = mds_fs_post_cleanup(obd);
911         
912         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
913         shrink_dcache_parent(mds->mds_id_de);
914         dput(mds->mds_id_de);
915
916         return rc;
917 }
918
919 /* Creates an object with the same name as its id.  Because this is not at all
920  * performance sensitive, it is accomplished by creating a file, checking the
921  * id, and renaming it. */
922 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
923                    void *acl, int acl_size,
924                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
925 {
926         struct mds_obd *mds = &exp->exp_obd->u.mds;
927         struct inode *parent_inode = mds->mds_objects_dir->d_inode;
928         struct file *filp;
929         struct dentry *dchild;
930         struct lvfs_run_ctxt saved;
931         char idname[LL_ID_NAMELEN];
932         int rc = 0, err, idlen;
933         void *handle;
934         ENTRY;
935
936         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
937         down(&parent_inode->i_sem);
938         if (oa->o_id) {
939                 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
940                 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
941                 if (IS_ERR(dchild))
942                         GOTO(out_pop, rc = PTR_ERR(dchild));
943
944                 if (dchild->d_inode == NULL) {
945                         struct dentry_params dp;
946                         struct inode *inode;
947
948                         CWARN("creating log with ID "LPU64"\n", oa->o_id);
949                         
950                         dchild->d_fsdata = (void *) &dp;
951                         dp.p_ptr = NULL;
952                         dp.p_inum = oa->o_id;
953                         rc = ll_vfs_create(parent_inode, dchild, S_IFREG, NULL);
954                         if (dchild->d_fsdata == (void *)(unsigned long)oa->o_id)
955                                 dchild->d_fsdata = NULL;
956                         if (rc) {
957                                 CDEBUG(D_INODE, "err during create: %d\n", rc);
958                                 dput(dchild);
959                                 GOTO(out_pop, rc);
960                         }
961                         inode = dchild->d_inode;
962                         LASSERT(inode->i_ino == oa->o_id);
963                         inode->i_generation = oa->o_generation;
964                         CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
965                                inode->i_ino, inode->i_generation);
966                         mark_inode_dirty(inode);
967                 } else {
968                         CWARN("it should be here!\n");
969                 }
970                 GOTO(out_pop, rc);
971         }
972
973         sprintf(idname, "OBJECTS/%u.%u", ll_insecure_random_int(), current->pid);
974         filp = filp_open(idname, O_CREAT | O_EXCL, 0644);
975         if (IS_ERR(filp)) {
976                 rc = PTR_ERR(filp);
977                 if (rc == -EEXIST) {
978                         CERROR("impossible object name collision %s\n",
979                                idname);
980                         LBUG();
981                 }
982                 CERROR("error creating tmp object %s: rc %d\n", 
983                        idname, rc);
984                 GOTO(out_pop, rc);
985         }
986
987         LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
988
989         oa->o_id = filp->f_dentry->d_inode->i_ino;
990         oa->o_generation = filp->f_dentry->d_inode->i_generation;
991         idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
992         
993         CWARN("created log anonymous "LPU64"/%u\n",
994               oa->o_id, oa->o_generation);
995
996         dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
997         if (IS_ERR(dchild)) {
998                 CERROR("getting neg dentry for obj rename: %d\n", rc);
999                 GOTO(out_close, rc = PTR_ERR(dchild));
1000         }
1001         if (dchild->d_inode != NULL) {
1002                 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
1003                        oa->o_id, oa->o_generation);
1004                 LBUG();
1005         }
1006
1007         handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
1008                               FSFILT_OP_RENAME, NULL);
1009         if (IS_ERR(handle))
1010                 GOTO(out_dput, rc = PTR_ERR(handle));
1011
1012         lock_kernel();
1013         rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
1014                         mds->mds_objects_dir->d_inode, dchild);
1015         unlock_kernel();
1016         if (rc)
1017                 CERROR("error renaming new object "LPU64":%u: rc %d\n",
1018                        oa->o_id, oa->o_generation, rc);
1019
1020         err = fsfilt_commit(exp->exp_obd, mds->mds_sb, 
1021                             mds->mds_objects_dir->d_inode, handle, 0);
1022         if (!err) {
1023                 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
1024                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLGROUP;
1025         } else if (!rc)
1026                 rc = err;
1027 out_dput:
1028         dput(dchild);
1029 out_close:
1030         err = filp_close(filp, 0);
1031         if (err) {
1032                 CERROR("closing tmpfile %s: rc %d\n", idname, rc);
1033                 if (!rc)
1034                         rc = err;
1035         }
1036 out_pop:
1037         up(&parent_inode->i_sem);
1038         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1039         RETURN(rc);
1040 }
1041
1042 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
1043                     struct lov_stripe_md *ea, struct obd_trans_info *oti)
1044 {
1045         struct mds_obd *mds = &exp->exp_obd->u.mds;
1046         struct inode *parent_inode = mds->mds_objects_dir->d_inode;
1047         struct obd_device *obd = exp->exp_obd;
1048         struct lvfs_run_ctxt saved;
1049         char idname[LL_ID_NAMELEN];
1050         struct dentry *de;
1051         void *handle;
1052         int err, idlen, rc = 0;
1053         ENTRY;
1054
1055         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1056
1057         idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1058
1059         down(&parent_inode->i_sem);
1060         de = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1061         if (IS_ERR(de) || de->d_inode == NULL) {
1062                 rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
1063                 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
1064                        oa->o_id, idname, rc);
1065                 GOTO(out_dput, rc);
1066         }
1067         /* Stripe count is 1 here since this is some MDS specific stuff
1068            that is unlinked, not spanned across multiple OSTs */
1069         handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
1070                                   FSFILT_OP_UNLINK, oti, 1);
1071
1072         if (IS_ERR(handle))
1073                 GOTO(out_dput, rc = PTR_ERR(handle));
1074         
1075         rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
1076         if (rc) 
1077                 CERROR("error destroying object "LPU64":%u: rc %d\n",
1078                        oa->o_id, oa->o_generation, rc);
1079         
1080         err = fsfilt_commit(obd, mds->mds_sb, mds->mds_objects_dir->d_inode, 
1081                             handle, exp->exp_sync);
1082         if (err && !rc)
1083                 rc = err;
1084 out_dput:
1085         if (de != NULL)
1086                 l_dput(de);
1087         up(&parent_inode->i_sem);
1088         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1089         RETURN(rc);
1090 }