Whamcloud - gitweb
llite and MDC statfs operations.
[fs/lustre-release.git] / lustre / llite / super.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  * This code is issued under the GNU General Public License.
7  * See the file COPYING in this distribution
8  *
9  * Copryright (C) 1996 Peter J. Braam <braam@stelias.com>
10  * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
11  * Copryright (C) 1999 Seagate Technology Inc.
12  * Copryright (C) 2001 Mountain View Data, Inc.
13  * Copryright (C) 2002 Cluster File Systems, Inc.
14  *
15  */
16
17 #define DEBUG_SUBSYSTEM S_LLITE
18
19 #include <linux/module.h>
20 #include <linux/lustre_lite.h>
21 #include <linux/lustre_ha.h>
22 #include <linux/lustre_dlm.h>
23
24 kmem_cache_t *ll_file_data_slab;
25 extern struct address_space_operations ll_aops;
26 extern struct address_space_operations ll_dir_aops;
27 struct super_operations ll_super_operations;
28
29 extern int ll_recover(struct ptlrpc_client *);
30 extern int ll_commitcbd_setup(struct ll_sb_info *);
31 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
32
33 static char *ll_read_opt(const char *opt, char *data)
34 {
35         char *value;
36         char *retval;
37         ENTRY;
38
39         CDEBUG(D_INFO, "option: %s, data %s\n", opt, data);
40         if ( strncmp(opt, data, strlen(opt)) )
41                 RETURN(NULL);
42         if ( (value = strchr(data, '=')) == NULL )
43                 RETURN(NULL);
44
45         value++;
46         OBD_ALLOC(retval, strlen(value) + 1);
47         if ( !retval ) {
48                 CERROR("out of memory!\n");
49                 RETURN(NULL);
50         }
51         
52         memcpy(retval, value, strlen(value)+1);
53         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
54         RETURN(retval);
55 }
56
57 static void ll_options(char *options, char **ost, char **mds)
58 {
59         char *this_char;
60         ENTRY;
61
62         if (!options) {
63                 EXIT;
64                 return;
65         }
66
67         for (this_char = strtok (options, ",");
68              this_char != NULL;
69              this_char = strtok (NULL, ",")) {
70                 CDEBUG(D_INFO, "this_char %s\n", this_char);
71                 if ( (!*ost && (*ost = ll_read_opt("ost", this_char)))||
72                      (!*mds && (*mds = ll_read_opt("mds", this_char))) )
73                         continue;
74         }
75         EXIT;
76 }
77
78 static struct super_block * ll_read_super(struct super_block *sb,
79                                           void *data, int silent)
80 {
81         struct inode *root = 0;
82         struct ll_sb_info *sbi;
83         char *ost = NULL;
84         char *mds = NULL;
85         int devno;
86         int err;
87         struct ll_fid rootfid;
88         __u64 last_committed, last_rcvd;
89         __u32 last_xid;
90         struct ptlrpc_request *request = NULL;
91         struct ll_inode_md md;
92
93         ENTRY;
94         MOD_INC_USE_COUNT;
95
96         OBD_ALLOC(sbi, sizeof(*sbi));
97         if (!sbi) {
98                 MOD_DEC_USE_COUNT;
99                 RETURN(NULL);
100         }
101
102         sb->u.generic_sbp = sbi;
103
104         ll_options(data, &ost, &mds);
105
106         if (!ost) {
107                 CERROR("no ost\n");
108                 GOTO(out_free, sb = NULL);
109         }
110
111         if (!mds) {
112                 CERROR("no mds\n");
113                 GOTO(out_free, sb = NULL);
114         }
115
116         devno = simple_strtoul(ost, NULL, 0);
117         if (devno >= MAX_OBD_DEVICES) {
118                 CERROR("devno of %s too high\n", ost);
119                 GOTO(out_free, sb = NULL);
120         }
121
122         sbi->ll_osc_conn.oc_dev = &obd_dev[devno];
123         err = obd_connect(&sbi->ll_osc_conn);
124         if (err) {
125                 CERROR("cannot connect to %s: rc = %d\n", ost, err);
126                 GOTO(out_free, sb = NULL);
127         }
128
129         devno = simple_strtoul(mds, NULL, 0);
130         if (devno >= MAX_OBD_DEVICES) {
131                 CERROR("devno of %s too high\n", mds);
132                 GOTO(out_free, sb = NULL);
133         }
134
135         sbi->ll_mdc_conn.oc_dev = &obd_dev[devno];
136         err = obd_connect(&sbi->ll_mdc_conn);
137         if (err) {
138                 CERROR("cannot connect to %s: rc = %d\n", mds, err);
139                 GOTO(out_free, sb = NULL);
140         }
141
142 #if 0
143         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
144         if (err) {
145                 CERROR("cannot connect to MDS: rc = %d\n", err);
146                 GOTO(out_rpc, sb = NULL);
147         }
148 #endif 
149
150         sbi2mdc(sbi)->mdc_conn->c_level = LUSTRE_CONN_FULL;
151
152         /* XXX: need to store the last_* values somewhere */
153         err = mdc_getstatus(&sbi->ll_mdc_conn,
154                             &rootfid, &last_committed, 
155                             &last_rcvd, &last_xid,
156                             &request);
157         if (err) {
158                 CERROR("cannot mds_connect: rc = %d\n", err);
159                 GOTO(out_disc, sb = NULL);
160         }
161         CDEBUG(D_SUPER, "rootfid %ld\n", (unsigned long)rootfid.id);
162         sbi->ll_rootino = rootfid.id;
163
164         sb->s_maxbytes = 1ULL << 36;
165         /* XXX get this with a get_info call (like we have in OBDFS),
166            this info call should return the blocksize of the MDS (statfs?) */
167         sb->s_blocksize = 4096;
168         sb->s_blocksize_bits = 12;
169         sb->s_magic = LL_SUPER_MAGIC;
170         sb->s_op = &ll_super_operations;
171
172         /* make root inode */
173         err = mdc_getattr(&sbi->ll_mdc_conn,
174                           sbi->ll_rootino, S_IFDIR,
175                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
176         if (err) {
177                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
178                 GOTO(out_mdc, sb = NULL);
179         }
180
181         /* initialize committed transaction callback daemon */
182         spin_lock_init(&sbi->ll_commitcbd_lock); 
183         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
184         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
185         sbi->ll_commitcbd_flags = 0;
186         err = ll_commitcbd_setup(sbi);
187         if (err) {
188                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
189                 GOTO(out_mdc, sb = NULL);
190         }
191
192         md.body = lustre_msg_buf(request->rq_repmsg, 0);
193         md.obdo = NULL;
194         root = iget4(sb, sbi->ll_rootino, NULL, &md);
195                      
196         if (root) {
197                 sb->s_root = d_alloc_root(root);
198         } else {
199                 CERROR("lustre_lite: bad iget4 for root\n");
200                 GOTO(out_cdb, sb = NULL);
201         }
202
203         ptlrpc_free_req(request);
204
205 out_dev:
206         if (mds)
207                 OBD_FREE(mds, strlen(mds) + 1);
208         if (ost)
209                 OBD_FREE(ost, strlen(ost) + 1);
210
211         RETURN(sb);
212
213 out_cdb:
214         ll_commitcbd_cleanup(sbi);
215 out_mdc:
216         obd_disconnect(&sbi->ll_mdc_conn);
217         obd_disconnect(&sbi->ll_osc_conn);
218 out_disc:
219         ptlrpc_free_req(request);
220         obd_disconnect(&sbi->ll_osc_conn);
221 out_free:
222         OBD_FREE(sbi, sizeof(*sbi));
223
224         MOD_DEC_USE_COUNT;
225         goto out_dev;
226 } /* ll_read_super */
227
228 static void ll_put_super(struct super_block *sb)
229 {
230         struct ll_sb_info *sbi = ll_s2sbi(sb);
231         ENTRY;
232         ll_commitcbd_cleanup(sbi);
233         obd_disconnect(&sbi->ll_osc_conn);
234         obd_disconnect(&sbi->ll_mdc_conn);
235         OBD_FREE(sbi, sizeof(*sbi));
236
237         MOD_DEC_USE_COUNT;
238         EXIT;
239 } /* ll_put_super */
240
241 static void ll_clear_inode(struct inode *inode)
242 {
243         if (atomic_read(&inode->i_count) == 0) {
244                 struct obdo *oa = ll_i2info(inode)->lli_obdo;
245                 if (oa) {
246                         obdo_free(oa);
247                         ll_i2info(inode)->lli_obdo = NULL;
248                 }
249                 if (ll_i2info(inode)->lli_symlink_name) {
250                         OBD_FREE(ll_i2info(inode)->lli_symlink_name,
251                                  strlen(ll_i2info(inode)->lli_symlink_name)+ 1);
252                         ll_i2info(inode)->lli_symlink_name = NULL;
253                 }
254         }
255 }
256
257 static void ll_delete_inode(struct inode *inode)
258 {
259         if (S_ISREG(inode->i_mode)) { 
260                 int err;
261                 struct obdo *oa;
262                 oa = ll_i2info(inode)->lli_obdo;
263
264                 if (!oa)
265                         GOTO(out, -EINVAL);
266
267                 if (oa->o_id == 0)
268                         /* No obdo was ever created */
269                         GOTO(out, 0);
270
271                 err = obd_destroy(ll_i2obdconn(inode), oa);
272                 CDEBUG(D_INODE, "obd destroy of %Ld error %d\n",
273                        (unsigned long long)oa->o_id, err);
274         }
275 out:
276         clear_inode(inode);
277 }
278
279 /* like inode_setattr, but doesn't mark the inode dirty */
280 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
281 {
282         unsigned int ia_valid = attr->ia_valid;
283         int error = 0;
284
285         if ((ia_valid & ATTR_SIZE) && trunc) {
286                 error = vmtruncate(inode, attr->ia_size);
287                 if (error)
288                         goto out;
289         } else if (ia_valid & ATTR_SIZE)
290                 inode->i_size = attr->ia_size;
291
292         if (ia_valid & ATTR_UID)
293                 inode->i_uid = attr->ia_uid;
294         if (ia_valid & ATTR_GID)
295                 inode->i_gid = attr->ia_gid;
296         if (ia_valid & ATTR_ATIME)
297                 inode->i_atime = attr->ia_atime;
298         if (ia_valid & ATTR_MTIME)
299                 inode->i_mtime = attr->ia_mtime;
300         if (ia_valid & ATTR_CTIME)
301                 inode->i_ctime = attr->ia_ctime;
302         if (ia_valid & ATTR_MODE) {
303                 inode->i_mode = attr->ia_mode;
304                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
305                         inode->i_mode &= ~S_ISGID;
306         }
307 out:
308         return error;
309 }
310
311 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
312 {
313         struct ptlrpc_request *request = NULL;
314         struct ll_sb_info *sbi = ll_i2sbi(inode);
315         int err;
316
317         ENTRY;
318
319         /* change incore inode */
320         ll_attr2inode(inode, attr, do_trunc);
321
322         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr,
323                           &request);
324         if (err)
325                 CERROR("mdc_setattr fails (%d)\n", err);
326
327         ptlrpc_req_finished(request);
328
329         RETURN(err);
330 }
331
332 int ll_setattr(struct dentry *de, struct iattr *attr)
333 {
334         int rc = inode_change_ok(de->d_inode, attr);
335
336         if (rc)
337                 return rc;
338
339         return ll_inode_setattr(de->d_inode, attr, 1);
340 }
341
342 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
343 {
344         struct ptlrpc_request *request = NULL;
345         struct ll_sb_info *sbi = ll_s2sbi(sb);
346         int err;
347         ENTRY;
348
349         memset(sfs, 0, sizeof(*sfs));
350         err = mdc_statfs(&sbi->ll_mdc_conn, sfs, &request);
351         if (err)
352                 CERROR("obd_statfs fails (%d)\n", err);
353         else
354                 CDEBUG(D_SUPER,
355                        "statfs returns blocks %ld/%ld objects %ld/%ld\n",
356                        sfs->f_bavail, sfs->f_blocks, sfs->f_files,sfs->f_ffree);
357
358         ptlrpc_req_finished(request);
359
360         RETURN(err);
361 }
362
363 static void ll_to_inode(struct inode *dst, struct ll_inode_md *md)
364 {
365         struct mds_body *body = md->body;
366         struct ll_inode_info *ii = ll_i2info(dst);
367
368         /* core attributes first */
369         if (body->valid & OBD_MD_FLID)
370                 dst->i_ino = body->ino;
371         if (body->valid & OBD_MD_FLATIME)
372                 dst->i_atime = body->atime;
373         if (body->valid & OBD_MD_FLMTIME)
374                 dst->i_mtime = body->mtime;
375         if (body->valid & OBD_MD_FLCTIME)
376                 dst->i_ctime = body->ctime;
377         if (body->valid & OBD_MD_FLSIZE)
378                 dst->i_size = body->size;
379         if (body->valid & OBD_MD_FLMODE)
380                 dst->i_mode = body->mode;
381         if (body->valid & OBD_MD_FLUID)
382                 dst->i_uid = body->uid;
383         if (body->valid & OBD_MD_FLGID)
384                 dst->i_gid = body->gid;
385         if (body->valid & OBD_MD_FLFLAGS)
386                 dst->i_flags = body->flags;
387         if (body->valid & OBD_MD_FLNLINK)
388                 dst->i_nlink = body->nlink;
389         if (body->valid & OBD_MD_FLGENER)
390                 dst->i_generation = body->generation;
391
392         /* this will become more elaborate for striping etc */
393         if (md->obdo != NULL && md->obdo->o_valid != 0) {
394                 ii->lli_obdo = obdo_alloc();
395                 memcpy(ii->lli_obdo, md->obdo, sizeof(*md->obdo));
396         }
397 #if 0
398
399         if (obdo_has_inline(oa)) {
400                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
401                     S_ISFIFO(inode->i_mode)) {
402                         obd_rdev rdev = *((obd_rdev *)oa->o_inline);
403                         CDEBUG(D_INODE,
404                                "copying device %x from obdo to inode\n", rdev);
405                         init_special_inode(inode, inode->i_mode, rdev);
406                 } else {
407                         CDEBUG(D_INFO, "copying inline from obdo to inode\n");
408                         memcpy(oinfo->lli_inline, oa->o_inline, OBD_INLINESZ);
409                 }
410                 oinfo->lli_flags |= OBD_FL_INLINEDATA;
411         }
412 #endif 
413 } /* ll_to_inode */
414
415 static void ll_read_inode2(struct inode *inode, void *opaque)
416 {
417         struct ll_inode_md *md = opaque;
418
419         ENTRY;
420         ll_to_inode(inode, md);
421
422         /* OIDEBUG(inode); */
423
424         if (S_ISREG(inode->i_mode)) {
425                 inode->i_op = &ll_file_inode_operations;
426                 inode->i_fop = &ll_file_operations;
427                 inode->i_mapping->a_ops = &ll_aops;
428                 EXIT;
429         } else if (S_ISDIR(inode->i_mode)) {
430                 inode->i_op = &ll_dir_inode_operations;
431                 inode->i_fop = &ll_dir_operations;
432                 inode->i_mapping->a_ops = &ll_dir_aops;
433                 EXIT;
434         } else if (S_ISLNK(inode->i_mode)) {
435                 inode->i_op = &ll_fast_symlink_inode_operations;
436                 EXIT;
437         } else {
438                 init_special_inode(inode, inode->i_mode,
439                                    ((int *)ll_i2info(inode)->lli_inline)[0]);
440                 EXIT;
441         }
442
443         return;
444 }
445
446 /* exported operations */
447 struct super_operations ll_super_operations =
448 {
449         read_inode2: ll_read_inode2,
450         clear_inode: ll_clear_inode,
451         delete_inode: ll_delete_inode,
452         put_super: ll_put_super,
453         statfs: ll_statfs
454 };
455
456 struct file_system_type lustre_lite_fs_type = {
457         "lustre_lite", 0, ll_read_super, NULL
458 };
459
460 static int __init init_lustre_lite(void)
461 {
462         printk(KERN_INFO "Lustre Lite 0.0.1, braam@clusterfs.com\n");
463         ll_file_data_slab = kmem_cache_create("ll_file_data",
464                                               sizeof(struct ll_file_data), 0,
465                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
466         if (ll_file_data_slab == NULL)
467                 return -ENOMEM;
468         return register_filesystem(&lustre_lite_fs_type);
469 }
470
471 static void __exit exit_lustre_lite(void)
472 {
473         unregister_filesystem(&lustre_lite_fs_type);
474         kmem_cache_destroy(ll_file_data_slab);
475 }
476
477 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
478 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
479 MODULE_LICENSE("GPL");
480
481 module_init(init_lustre_lite);
482 module_exit(exit_lustre_lite);