Whamcloud - gitweb
b=6427
[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((char *)med->med_mcd->mcd_uuid, (char *)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((char *)med->med_mcd->mcd_uuid, (char *)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((char *)msd->msd_uuid, (char *)obd->obd_uuid.uuid)) {
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
448         l_dput(dentry);
449         RETURN(rc); 
450 }
451
452 /*
453  * sets up root inode lustre_id. It tries to read it first from root inode and
454  * if it is not there, new rootid is allocated and saved there.
455  */
456 int mds_fs_setup_rootid(struct obd_device *obd)
457 {
458         int rc = 0;
459         void *handle;
460         struct inode *inode;
461         struct dentry *dentry;
462         struct mds_obd *mds = &obd->u.mds;
463         ENTRY;
464
465         /* getting root directory and setup its fid. */
466         dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
467         if (IS_ERR(dentry)) {
468                 CERROR("Can't find ROOT by "DLID4", err = %d\n",
469                        OLID4(&mds->mds_rootid), (int)PTR_ERR(dentry));
470                 RETURN(PTR_ERR(dentry));
471         }
472
473         inode = dentry->d_inode;
474         LASSERT(dentry->d_inode);
475
476         rc = mds_pack_inode2id(obd, &mds->mds_rootid, inode, 1);
477         if (rc < 0) {
478                 if (rc != -ENODATA)
479                         GOTO(out_dentry, rc);
480         } else {
481                 /*
482                  * rootid is filled by mds_read_inode_sid(), so we do not need
483                  * to allocate it and update. The only thing we need to check is
484                  * mds_num.
485                  */
486                 LASSERT(id_group(&mds->mds_rootid) == mds->mds_num);
487                 mds_set_last_fid(obd, id_fid(&mds->mds_rootid));
488                 GOTO(out_dentry, rc);
489         }
490
491         /* allocating new one, as it is not found in root inode. */
492         handle = fsfilt_start(obd, inode,
493                               FSFILT_OP_SETATTR, NULL);
494         
495         if (IS_ERR(handle)) {
496                 rc = PTR_ERR(handle);
497                 CERROR("fsfilt_start() failed, rc = %d\n", rc);
498                 GOTO(out_dentry, rc);
499         }
500         
501         down(&inode->i_sem);
502         rc = mds_alloc_inode_sid(obd, inode, handle, &mds->mds_rootid);
503         up(&inode->i_sem);
504         
505         if (rc) {
506                 CERROR("mds_alloc_inode_sid() failed, rc = %d\n",
507                        rc);
508                 GOTO(out_dentry, rc);
509         }
510
511         rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
512         if (rc)
513                 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
514
515         EXIT;
516 out_dentry:
517         l_dput(dentry);
518         if (rc == 0)
519                 CWARN("%s: rootid: "DLID4"\n", obd->obd_name,
520                       OLID4(&mds->mds_rootid));
521         return rc;
522 }
523
524 static int mds_update_virtid_fid(struct obd_device *obd,
525                                  void *handle, int force_sync)
526 {
527         struct mds_obd *mds = &obd->u.mds;
528         struct file *filp = mds->mds_virtid_filp;
529         struct lvfs_run_ctxt saved;
530         loff_t off = 0;
531         int rc = 0;
532         ENTRY;
533
534         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
535         rc = fsfilt_write_record(obd, filp, &mds->mds_virtid_fid,
536                                  sizeof(mds->mds_virtid_fid),
537                                  &off, force_sync);
538         if (rc) {
539                 CERROR("error writing MDS virtid_fid #"LPU64
540                        ", err = %d\n", mds->mds_virtid_fid, rc);
541         }
542                 
543         CDEBUG(D_SUPER, "wrote virtid fid #"LPU64" at idx "
544                "%llu: err = %d\n", mds->mds_virtid_fid,
545                off, rc);
546         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
547
548         RETURN(rc);
549 }
550
551 static int mds_read_virtid_fid(struct obd_device *obd,
552                                struct file *file)
553 {
554         int rc = 0;
555         loff_t off = 0;
556         struct mds_obd *mds = &obd->u.mds;
557         unsigned long virtid_fid_size = file->f_dentry->d_inode->i_size;
558         ENTRY;
559
560         if (virtid_fid_size == 0) {
561                 mds->mds_virtid_fid = mds_alloc_fid(obd);
562         } else {
563                 rc = fsfilt_read_record(obd, file, &mds->mds_virtid_fid,
564                                         sizeof(mds->mds_virtid_fid), &off);
565                 if (rc) {
566                         CERROR("error reading MDS %s: rc = %d\n",
567                                file->f_dentry->d_name.name, rc);
568                         RETURN(rc);
569                 }
570         }
571         rc = mds_update_virtid_fid(obd, NULL, 1);
572
573         RETURN(rc);
574 }
575
576 /*
577  * initializes lustre_id for virtual id directory, it is needed sometimes, as it
578  * is possible that it will be the parent for object an operations is going to
579  * be performed on.
580  */
581 int mds_fs_setup_virtid(struct obd_device *obd)
582 {
583         int rc = 0;
584         void *handle;
585         struct lustre_id sid;
586         struct mds_obd *mds = &obd->u.mds;
587         struct inode *inode = mds->mds_id_dir->d_inode;
588         ENTRY;
589
590         handle = fsfilt_start(obd, inode,
591                               FSFILT_OP_SETATTR, NULL);
592         
593         if (IS_ERR(handle)) {
594                 rc = PTR_ERR(handle);
595                 CERROR("fsfilt_start() failed, rc = %d\n", rc);
596                 RETURN(rc);
597         }
598
599         id_group(&sid) = mds->mds_num;
600         id_fid(&sid) = mds->mds_virtid_fid;
601
602         id_ino(&sid) = inode->i_ino;
603         id_gen(&sid) = inode->i_generation;
604         id_type(&sid) = (S_IFMT & inode->i_mode);
605
606         down(&inode->i_sem);
607         rc = mds_update_inode_sid(obd, inode, handle, &sid);
608         up(&inode->i_sem);
609
610         if (rc) {
611                 CERROR("mds_update_inode_sid() failed, rc = %d\n",
612                        rc);
613                 RETURN(rc);
614         }
615
616         rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
617         if (rc) {
618                 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
619                 RETURN(rc);
620         }
621
622         RETURN(rc);
623 }
624
625 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
626 {
627         struct mds_obd *mds = &obd->u.mds;
628         struct lvfs_run_ctxt saved;
629         struct dentry *dentry;
630         struct file *file;
631         int rc;
632         ENTRY;
633
634         rc = cleanup_group_info();
635         if (rc)
636                 RETURN(rc);
637
638         mds->mds_vfsmnt = mnt;
639         mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
640
641         fsfilt_setup(obd, mds->mds_sb);
642
643         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
644         obd->obd_lvfs_ctxt.pwdmnt = mnt;
645         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
646         obd->obd_lvfs_ctxt.fs = get_ds();
647         obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
648
649         /* setup the directory tree */
650         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
651         dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
652         if (IS_ERR(dentry)) {
653                 rc = PTR_ERR(dentry);
654                 CERROR("cannot create ROOT directory: rc = %d\n", rc);
655                 GOTO(err_pop, rc);
656         }
657
658         mdc_pack_id(&mds->mds_rootid, dentry->d_inode->i_ino,
659                     dentry->d_inode->i_generation, S_IFDIR, 0, 0);
660
661         dput(dentry);
662         
663         dentry = lookup_one_len("__iopen__", current->fs->pwd,
664                                 strlen("__iopen__"));
665         if (IS_ERR(dentry)) {
666                 rc = PTR_ERR(dentry);
667                 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
668                 GOTO(err_pop, rc);
669         }
670         mds->mds_id_de = dentry;
671         if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
672                 rc = -ENOENT;
673                 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
674                 GOTO(err_id_de, rc);
675         }
676
677         dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
678         if (IS_ERR(dentry)) {
679                 rc = PTR_ERR(dentry);
680                 CERROR("cannot create PENDING directory: rc = %d\n", rc);
681                 GOTO(err_id_de, rc);
682         }
683         mds->mds_pending_dir = dentry;
684       
685         dentry = simple_mkdir(current->fs->pwd, "LOGS", 0777, 1);
686         if (IS_ERR(dentry)) {
687                 rc = PTR_ERR(dentry);
688                 CERROR("cannot create LOGS directory: rc = %d\n", rc);
689                 GOTO(err_pending, rc);
690         }
691         mds->mds_logs_dir = dentry;
692
693         dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
694         if (IS_ERR(dentry)) {
695                 rc = PTR_ERR(dentry);
696                 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
697                 GOTO(err_logs, rc);
698         }
699         mds->mds_objects_dir = dentry;
700
701         dentry = simple_mkdir(current->fs->pwd, "FIDS", 0777, 1);
702         if (IS_ERR(dentry)) {
703                 rc = PTR_ERR(dentry);
704                 CERROR("cannot create FIDS directory: rc = %d\n", rc);
705                 GOTO(err_objects, rc);
706         }
707         mds->mds_id_dir = dentry;
708
709         dentry = simple_mkdir(current->fs->pwd, "UNNAMED", 0777, 1);
710         if (IS_ERR(dentry)) {
711                 rc = PTR_ERR(dentry);
712                 CERROR("cannot create UNNAMED directory: rc = %d\n", rc);
713                 GOTO(err_unnamed, rc);
714         }
715         mds->mds_unnamed_dir = dentry;
716
717         /* open and test the last rcvd file */
718         file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
719         if (IS_ERR(file)) {
720                 rc = PTR_ERR(file);
721                 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
722                 GOTO(err_id_dir, rc = PTR_ERR(file));
723         }
724         mds->mds_rcvd_filp = file;
725         
726         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
727                 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
728                        file->f_dentry->d_inode->i_mode);
729                 GOTO(err_last_rcvd, rc = -ENOENT);
730         }
731
732         rc = mds_read_last_rcvd(obd, file);
733         if (rc) {
734                 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
735                 GOTO(err_last_rcvd, rc);
736         }
737
738         /* open and test last fid file */
739         file = filp_open(LAST_FID, O_RDWR | O_CREAT, 0644);
740         if (IS_ERR(file)) {
741                 rc = PTR_ERR(file);
742                 CERROR("cannot open/create %s file: rc = %d\n",
743                        LAST_FID, rc);
744                 GOTO(err_client, rc = PTR_ERR(file));
745         }
746         mds->mds_fid_filp = file;
747         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
748                 CERROR("%s is not a regular file!: mode = %o\n",
749                        LAST_FID, file->f_dentry->d_inode->i_mode);
750                 GOTO(err_last_fid, rc = -ENOENT);
751         }
752
753         rc = mds_read_last_fid(obd, file);
754         if (rc) {
755                 CERROR("cannot read %s: rc = %d\n", LAST_FID, rc);
756                 GOTO(err_last_fid, rc);
757         }
758
759         /* open and test virtid fid file */
760         file = filp_open(VIRT_FID, O_RDWR | O_CREAT, 0644);
761         if (IS_ERR(file)) {
762                 rc = PTR_ERR(file);
763                 CERROR("cannot open/create %s file: rc = %d\n",
764                        VIRT_FID, rc);
765                 GOTO(err_last_fid, rc = PTR_ERR(file));
766         }
767         mds->mds_virtid_filp = file;
768         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
769                 CERROR("%s is not a regular file!: mode = %o\n",
770                        VIRT_FID, file->f_dentry->d_inode->i_mode);
771                 GOTO(err_virtid_fid, rc = -ENOENT);
772         }
773
774         rc = mds_read_virtid_fid(obd, file);
775         if (rc) {
776                 CERROR("cannot read %s: rc = %d\n", VIRT_FID, rc);
777                 GOTO(err_virtid_fid, rc);
778         }
779         
780         /* open and test the lov objid file */
781         file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
782         if (IS_ERR(file)) {
783                 rc = PTR_ERR(file);
784                 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
785                 GOTO(err_last_fid, rc = PTR_ERR(file));
786         }
787         mds->mds_dt_objid_filp = file;
788         if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
789                 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
790                        file->f_dentry->d_inode->i_mode);
791                 GOTO(err_lov_objid, rc = -ENOENT);
792         }
793 err_pop:
794         if (!rc) {
795                 rc = mds_fs_post_setup(obd);
796                 if (rc)
797                         CERROR("can not post setup fsfilt\n");        
798         }
799         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
800         return rc;
801
802 err_lov_objid:
803         if (mds->mds_dt_objid_filp && filp_close(mds->mds_dt_objid_filp, 0))
804                 CERROR("can't close %s after error\n", LOV_OBJID);
805 err_virtid_fid:
806         if (mds->mds_virtid_filp && filp_close(mds->mds_virtid_filp, 0))
807                 CERROR("can't close %s after error\n", VIRT_FID);
808 err_last_fid:
809         if (mds->mds_fid_filp && filp_close(mds->mds_fid_filp, 0))
810                 CERROR("can't close %s after error\n", LAST_FID);
811 err_client:
812         class_disconnect_exports(obd, 0);
813 err_last_rcvd:
814         if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
815                 CERROR("can't close %s after error\n", LAST_RCVD);
816 err_unnamed:
817         dput(mds->mds_unnamed_dir);
818 err_id_dir:
819         dput(mds->mds_id_dir);
820 err_objects:
821         dput(mds->mds_objects_dir);
822 err_logs:
823         dput(mds->mds_logs_dir);
824 err_pending:
825         dput(mds->mds_pending_dir);
826 err_id_de:
827         dput(mds->mds_id_de);
828         goto err_pop;
829 }
830
831 static int  mds_fs_post_cleanup(struct obd_device *obd)
832 {
833         int    rc = 0;
834         rc = fsfilt_post_cleanup(obd);
835         return rc; 
836 }
837
838 int mds_fs_cleanup(struct obd_device *obd, int flags)
839 {
840         struct mds_obd *mds = &obd->u.mds;
841         struct lvfs_run_ctxt saved;
842         int rc = 0;
843
844         if (flags & OBD_OPT_FAILOVER)
845                 CERROR("%s: shutting down for failover; client state will"
846                        " be preserved.\n", obd->obd_name);
847
848         class_disconnect_exports(obd, flags); /* cleans up client info too */
849         target_cleanup_recovery(obd);
850         mds_server_free_data(mds);
851
852         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
853         if (mds->mds_virtid_filp) {
854                 rc = filp_close(mds->mds_virtid_filp, 0);
855                 mds->mds_virtid_filp = NULL;
856                 if (rc)
857                         CERROR("%s file won't close, rc = %d\n", VIRT_FID, rc);
858         }
859         if (mds->mds_fid_filp) {
860                 rc = filp_close(mds->mds_fid_filp, 0);
861                 mds->mds_fid_filp = NULL;
862                 if (rc)
863                         CERROR("%s file won't close, rc = %d\n", LAST_FID, rc);
864         }
865         if (mds->mds_rcvd_filp) {
866                 rc = filp_close(mds->mds_rcvd_filp, 0);
867                 mds->mds_rcvd_filp = NULL;
868                 if (rc)
869                         CERROR("%s file won't close, rc = %d\n", LAST_RCVD, rc);
870         }
871         if (mds->mds_dt_objid_filp) {
872                 rc = filp_close(mds->mds_dt_objid_filp, 0);
873                 mds->mds_dt_objid_filp = NULL;
874                 if (rc)
875                         CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
876         }
877         if (mds->mds_unnamed_dir != NULL) {
878                 l_dput(mds->mds_unnamed_dir);
879                 mds->mds_unnamed_dir = NULL;
880         }
881         if (mds->mds_id_dir != NULL) {
882                 l_dput(mds->mds_id_dir);
883                 mds->mds_id_dir = NULL;
884         }
885         if (mds->mds_objects_dir != NULL) {
886                 l_dput(mds->mds_objects_dir);
887                 mds->mds_objects_dir = NULL;
888         }
889         if (mds->mds_logs_dir) {
890                 l_dput(mds->mds_logs_dir);
891                 mds->mds_logs_dir = NULL;
892         }
893         if (mds->mds_pending_dir) {
894                 l_dput(mds->mds_pending_dir);
895                 mds->mds_pending_dir = NULL;
896         }
897         rc = mds_fs_post_cleanup(obd);
898         
899         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
900         shrink_dcache_parent(mds->mds_id_de);
901         dput(mds->mds_id_de);
902
903         return rc;
904 }
905
906 /* Creates an object with the same name as its id.  Because this is not at all
907  * performance sensitive, it is accomplished by creating a file, checking the
908  * id, and renaming it. */
909 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
910                    void *acl, int acl_size,
911                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
912 {
913         struct mds_obd *mds = &exp->exp_obd->u.mds;
914         struct inode *parent_inode = mds->mds_objects_dir->d_inode;
915         struct file *filp;
916         struct dentry *dchild;
917         struct lvfs_run_ctxt saved;
918         char idname[LL_ID_NAMELEN];
919         int rc = 0, err, idlen;
920         void *handle;
921         ENTRY;
922
923         push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
924         down(&parent_inode->i_sem);
925         if (oa->o_id) {
926                 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
927                 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
928                 if (IS_ERR(dchild))
929                         GOTO(out_pop, rc = PTR_ERR(dchild));
930
931                 if (dchild->d_inode == NULL) {
932                         struct dentry_params dp;
933                         struct inode *inode;
934
935                         CWARN("creating log with ID "LPU64"\n", oa->o_id);
936                         
937                         dchild->d_fsdata = (void *) &dp;
938                         dp.p_ptr = NULL;
939                         dp.p_inum = oa->o_id;
940                         rc = ll_vfs_create(parent_inode, dchild, S_IFREG, NULL);
941                         if (dchild->d_fsdata == (void *)(unsigned long)oa->o_id)
942                                 dchild->d_fsdata = NULL;
943                         if (rc) {
944                                 CDEBUG(D_INODE, "err during create: %d\n", rc);
945                                 dput(dchild);
946                                 GOTO(out_pop, rc);
947                         }
948                         inode = dchild->d_inode;
949                         LASSERT(inode->i_ino == oa->o_id);
950                         inode->i_generation = oa->o_generation;
951                         CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
952                                inode->i_ino, inode->i_generation);
953                         mark_inode_dirty(inode);
954                 } else {
955                         CWARN("it should be here!\n");
956                 }
957                 GOTO(out_pop, rc);
958         }
959
960         sprintf(idname, "OBJECTS/%u.%u", ll_insecure_random_int(), current->pid);
961         filp = filp_open(idname, O_CREAT | O_EXCL, 0644);
962         if (IS_ERR(filp)) {
963                 rc = PTR_ERR(filp);
964                 if (rc == -EEXIST) {
965                         CERROR("impossible object name collision %s\n",
966                                idname);
967                         LBUG();
968                 }
969                 CERROR("error creating tmp object %s: rc %d\n", 
970                        idname, rc);
971                 GOTO(out_pop, rc);
972         }
973
974         LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
975
976         oa->o_id = filp->f_dentry->d_inode->i_ino;
977         oa->o_generation = filp->f_dentry->d_inode->i_generation;
978         idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
979         
980         CWARN("created log anonymous "LPU64"/%u\n",
981               oa->o_id, oa->o_generation);
982
983         dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
984         if (IS_ERR(dchild)) {
985                 CERROR("getting neg dentry for obj rename: %d\n", rc);
986                 GOTO(out_close, rc = PTR_ERR(dchild));
987         }
988         if (dchild->d_inode != NULL) {
989                 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
990                        oa->o_id, oa->o_generation);
991                 LBUG();
992         }
993
994         handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
995                               FSFILT_OP_RENAME, NULL);
996         if (IS_ERR(handle))
997                 GOTO(out_dput, rc = PTR_ERR(handle));
998
999         lock_kernel();
1000         rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
1001                         mds->mds_objects_dir->d_inode, dchild);
1002         unlock_kernel();
1003         if (rc)
1004                 CERROR("error renaming new object "LPU64":%u: rc %d\n",
1005                        oa->o_id, oa->o_generation, rc);
1006
1007         err = fsfilt_commit(exp->exp_obd, mds->mds_sb, 
1008                             mds->mds_objects_dir->d_inode, handle, 0);
1009         if (!err) {
1010                 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
1011                 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLGROUP;
1012         } else if (!rc)
1013                 rc = err;
1014 out_dput:
1015         dput(dchild);
1016 out_close:
1017         err = filp_close(filp, 0);
1018         if (err) {
1019                 CERROR("closing tmpfile %s: rc %d\n", idname, rc);
1020                 if (!rc)
1021                         rc = err;
1022         }
1023 out_pop:
1024         up(&parent_inode->i_sem);
1025         pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1026         RETURN(rc);
1027 }
1028
1029 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
1030                     struct lov_stripe_md *ea, struct obd_trans_info *oti)
1031 {
1032         struct mds_obd *mds = &exp->exp_obd->u.mds;
1033         struct inode *parent_inode = mds->mds_objects_dir->d_inode;
1034         struct obd_device *obd = exp->exp_obd;
1035         struct lvfs_run_ctxt saved;
1036         char idname[LL_ID_NAMELEN];
1037         struct dentry *de;
1038         void *handle;
1039         int err, idlen, rc = 0;
1040         ENTRY;
1041
1042         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1043
1044         idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1045
1046         down(&parent_inode->i_sem);
1047         de = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1048         if (IS_ERR(de) || de->d_inode == NULL) {
1049                 rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
1050                 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
1051                        oa->o_id, idname, rc);
1052                 GOTO(out_dput, rc);
1053         }
1054         /* Stripe count is 1 here since this is some MDS specific stuff
1055            that is unlinked, not spanned across multiple OSTs */
1056         handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
1057                                   FSFILT_OP_UNLINK, oti, 1);
1058
1059         if (IS_ERR(handle))
1060                 GOTO(out_dput, rc = PTR_ERR(handle));
1061         
1062         rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
1063         if (rc) 
1064                 CERROR("error destroying object "LPU64":%u: rc %d\n",
1065                        oa->o_id, oa->o_generation, rc);
1066         
1067         err = fsfilt_commit(obd, mds->mds_sb, mds->mds_objects_dir->d_inode, 
1068                             handle, exp->exp_sync);
1069         if (err && !rc)
1070                 rc = err;
1071 out_dput:
1072         if (de != NULL)
1073                 l_dput(de);
1074         up(&parent_inode->i_sem);
1075         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1076         RETURN(rc);
1077 }