->write_super lock_super pushdown
[safe/jmp/linux-2.6] / fs / exofs / super.c
1 /*
2  * Copyright (C) 2005, 2006
3  * Avishay Traeger (avishay@gmail.com) (avishay@il.ibm.com)
4  * Copyright (C) 2005, 2006
5  * International Business Machines
6  * Copyright (C) 2008, 2009
7  * Boaz Harrosh <bharrosh@panasas.com>
8  *
9  * Copyrights for code taken from ext2:
10  *     Copyright (C) 1992, 1993, 1994, 1995
11  *     Remy Card (card@masi.ibp.fr)
12  *     Laboratoire MASI - Institut Blaise Pascal
13  *     Universite Pierre et Marie Curie (Paris VI)
14  *     from
15  *     linux/fs/minix/inode.c
16  *     Copyright (C) 1991, 1992  Linus Torvalds
17  *
18  * This file is part of exofs.
19  *
20  * exofs is free software; you can redistribute it and/or modify
21  * it under the terms of the GNU General Public License as published by
22  * the Free Software Foundation.  Since it is based on ext2, and the only
23  * valid version of GPL for the Linux kernel is version 2, the only valid
24  * version of GPL for exofs is version 2.
25  *
26  * exofs is distributed in the hope that it will be useful,
27  * but WITHOUT ANY WARRANTY; without even the implied warranty of
28  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29  * GNU General Public License for more details.
30  *
31  * You should have received a copy of the GNU General Public License
32  * along with exofs; if not, write to the Free Software
33  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
34  */
35
36 #include <linux/string.h>
37 #include <linux/parser.h>
38 #include <linux/vfs.h>
39 #include <linux/random.h>
40 #include <linux/exportfs.h>
41
42 #include "exofs.h"
43
44 /******************************************************************************
45  * MOUNT OPTIONS
46  *****************************************************************************/
47
48 /*
49  * struct to hold what we get from mount options
50  */
51 struct exofs_mountopt {
52         const char *dev_name;
53         uint64_t pid;
54         int timeout;
55 };
56
57 /*
58  * exofs-specific mount-time options.
59  */
60 enum { Opt_pid, Opt_to, Opt_mkfs, Opt_format, Opt_err };
61
62 /*
63  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
64  * kernel's parsing functions do not currently support that.  32-bit should be
65  * sufficient for most applications now.
66  */
67 static match_table_t tokens = {
68         {Opt_pid, "pid=%u"},
69         {Opt_to, "to=%u"},
70         {Opt_err, NULL}
71 };
72
73 /*
74  * The main option parsing method.  Also makes sure that all of the mandatory
75  * mount options were set.
76  */
77 static int parse_options(char *options, struct exofs_mountopt *opts)
78 {
79         char *p;
80         substring_t args[MAX_OPT_ARGS];
81         int option;
82         bool s_pid = false;
83
84         EXOFS_DBGMSG("parse_options %s\n", options);
85         /* defaults */
86         memset(opts, 0, sizeof(*opts));
87         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
88
89         while ((p = strsep(&options, ",")) != NULL) {
90                 int token;
91                 char str[32];
92
93                 if (!*p)
94                         continue;
95
96                 token = match_token(p, tokens, args);
97                 switch (token) {
98                 case Opt_pid:
99                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
100                                 return -EINVAL;
101                         opts->pid = simple_strtoull(str, NULL, 0);
102                         if (opts->pid < EXOFS_MIN_PID) {
103                                 EXOFS_ERR("Partition ID must be >= %u",
104                                           EXOFS_MIN_PID);
105                                 return -EINVAL;
106                         }
107                         s_pid = 1;
108                         break;
109                 case Opt_to:
110                         if (match_int(&args[0], &option))
111                                 return -EINVAL;
112                         if (option <= 0) {
113                                 EXOFS_ERR("Timout must be > 0");
114                                 return -EINVAL;
115                         }
116                         opts->timeout = option * HZ;
117                         break;
118                 }
119         }
120
121         if (!s_pid) {
122                 EXOFS_ERR("Need to specify the following options:\n");
123                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
124                 return -EINVAL;
125         }
126
127         return 0;
128 }
129
130 /******************************************************************************
131  * INODE CACHE
132  *****************************************************************************/
133
134 /*
135  * Our inode cache.  Isn't it pretty?
136  */
137 static struct kmem_cache *exofs_inode_cachep;
138
139 /*
140  * Allocate an inode in the cache
141  */
142 static struct inode *exofs_alloc_inode(struct super_block *sb)
143 {
144         struct exofs_i_info *oi;
145
146         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
147         if (!oi)
148                 return NULL;
149
150         oi->vfs_inode.i_version = 1;
151         return &oi->vfs_inode;
152 }
153
154 /*
155  * Remove an inode from the cache
156  */
157 static void exofs_destroy_inode(struct inode *inode)
158 {
159         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
160 }
161
162 /*
163  * Initialize the inode
164  */
165 static void exofs_init_once(void *foo)
166 {
167         struct exofs_i_info *oi = foo;
168
169         inode_init_once(&oi->vfs_inode);
170 }
171
172 /*
173  * Create and initialize the inode cache
174  */
175 static int init_inodecache(void)
176 {
177         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
178                                 sizeof(struct exofs_i_info), 0,
179                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
180                                 exofs_init_once);
181         if (exofs_inode_cachep == NULL)
182                 return -ENOMEM;
183         return 0;
184 }
185
186 /*
187  * Destroy the inode cache
188  */
189 static void destroy_inodecache(void)
190 {
191         kmem_cache_destroy(exofs_inode_cachep);
192 }
193
194 /******************************************************************************
195  * SUPERBLOCK FUNCTIONS
196  *****************************************************************************/
197 static const struct super_operations exofs_sops;
198 static const struct export_operations exofs_export_ops;
199
200 /*
201  * Write the superblock to the OSD
202  */
203 static void exofs_write_super(struct super_block *sb)
204 {
205         struct exofs_sb_info *sbi;
206         struct exofs_fscb *fscb;
207         struct osd_request *or;
208         struct osd_obj_id obj;
209         int ret;
210
211         fscb = kzalloc(sizeof(struct exofs_fscb), GFP_KERNEL);
212         if (!fscb) {
213                 EXOFS_ERR("exofs_write_super: memory allocation failed.\n");
214                 return;
215         }
216
217         lock_super(sb);
218         lock_kernel();
219         sbi = sb->s_fs_info;
220         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
221         fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
222         fscb->s_magic = cpu_to_le16(sb->s_magic);
223         fscb->s_newfs = 0;
224
225         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
226         if (unlikely(!or)) {
227                 EXOFS_ERR("exofs_write_super: osd_start_request failed.\n");
228                 goto out;
229         }
230
231         obj.partition = sbi->s_pid;
232         obj.id = EXOFS_SUPER_ID;
233         ret = osd_req_write_kern(or, &obj, 0, fscb, sizeof(*fscb));
234         if (unlikely(ret)) {
235                 EXOFS_ERR("exofs_write_super: osd_req_write_kern failed.\n");
236                 goto out;
237         }
238
239         ret = exofs_sync_op(or, sbi->s_timeout, sbi->s_cred);
240         if (unlikely(ret)) {
241                 EXOFS_ERR("exofs_write_super: exofs_sync_op failed.\n");
242                 goto out;
243         }
244         sb->s_dirt = 0;
245
246 out:
247         if (or)
248                 osd_end_request(or);
249         unlock_kernel();
250         unlock_super(sb);
251         kfree(fscb);
252 }
253
254 /*
255  * This function is called when the vfs is freeing the superblock.  We just
256  * need to free our own part.
257  */
258 static void exofs_put_super(struct super_block *sb)
259 {
260         int num_pend;
261         struct exofs_sb_info *sbi = sb->s_fs_info;
262
263         lock_kernel();
264
265         if (sb->s_dirt)
266                 exofs_write_super(sb);
267
268         /* make sure there are no pending commands */
269         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
270              num_pend = atomic_read(&sbi->s_curr_pending)) {
271                 wait_queue_head_t wq;
272                 init_waitqueue_head(&wq);
273                 wait_event_timeout(wq,
274                                   (atomic_read(&sbi->s_curr_pending) == 0),
275                                   msecs_to_jiffies(100));
276         }
277
278         osduld_put_device(sbi->s_dev);
279         kfree(sb->s_fs_info);
280         sb->s_fs_info = NULL;
281
282         unlock_kernel();
283 }
284
285 /*
286  * Read the superblock from the OSD and fill in the fields
287  */
288 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
289 {
290         struct inode *root;
291         struct exofs_mountopt *opts = data;
292         struct exofs_sb_info *sbi;      /*extended info                  */
293         struct exofs_fscb fscb;         /*on-disk superblock info        */
294         struct osd_request *or = NULL;
295         struct osd_obj_id obj;
296         int ret;
297
298         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
299         if (!sbi)
300                 return -ENOMEM;
301         sb->s_fs_info = sbi;
302
303         /* use mount options to fill superblock */
304         sbi->s_dev = osduld_path_lookup(opts->dev_name);
305         if (IS_ERR(sbi->s_dev)) {
306                 ret = PTR_ERR(sbi->s_dev);
307                 sbi->s_dev = NULL;
308                 goto free_sbi;
309         }
310
311         sbi->s_pid = opts->pid;
312         sbi->s_timeout = opts->timeout;
313
314         /* fill in some other data by hand */
315         memset(sb->s_id, 0, sizeof(sb->s_id));
316         strcpy(sb->s_id, "exofs");
317         sb->s_blocksize = EXOFS_BLKSIZE;
318         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
319         sb->s_maxbytes = MAX_LFS_FILESIZE;
320         atomic_set(&sbi->s_curr_pending, 0);
321         sb->s_bdev = NULL;
322         sb->s_dev = 0;
323
324         /* read data from on-disk superblock object */
325         obj.partition = sbi->s_pid;
326         obj.id = EXOFS_SUPER_ID;
327         exofs_make_credential(sbi->s_cred, &obj);
328
329         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
330         if (unlikely(!or)) {
331                 if (!silent)
332                         EXOFS_ERR(
333                                "exofs_fill_super: osd_start_request failed.\n");
334                 ret = -ENOMEM;
335                 goto free_sbi;
336         }
337         ret = osd_req_read_kern(or, &obj, 0, &fscb, sizeof(fscb));
338         if (unlikely(ret)) {
339                 if (!silent)
340                         EXOFS_ERR(
341                                "exofs_fill_super: osd_req_read_kern failed.\n");
342                 ret = -ENOMEM;
343                 goto free_sbi;
344         }
345
346         ret = exofs_sync_op(or, sbi->s_timeout, sbi->s_cred);
347         if (unlikely(ret)) {
348                 if (!silent)
349                         EXOFS_ERR("exofs_fill_super: exofs_sync_op failed.\n");
350                 ret = -EIO;
351                 goto free_sbi;
352         }
353
354         sb->s_magic = le16_to_cpu(fscb.s_magic);
355         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
356         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
357
358         /* make sure what we read from the object store is correct */
359         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
360                 if (!silent)
361                         EXOFS_ERR("ERROR: Bad magic value\n");
362                 ret = -EINVAL;
363                 goto free_sbi;
364         }
365
366         /* start generation numbers from a random point */
367         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
368         spin_lock_init(&sbi->s_next_gen_lock);
369
370         /* set up operation vectors */
371         sb->s_op = &exofs_sops;
372         sb->s_export_op = &exofs_export_ops;
373         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
374         if (IS_ERR(root)) {
375                 EXOFS_ERR("ERROR: exofs_iget failed\n");
376                 ret = PTR_ERR(root);
377                 goto free_sbi;
378         }
379         sb->s_root = d_alloc_root(root);
380         if (!sb->s_root) {
381                 iput(root);
382                 EXOFS_ERR("ERROR: get root inode failed\n");
383                 ret = -ENOMEM;
384                 goto free_sbi;
385         }
386
387         if (!S_ISDIR(root->i_mode)) {
388                 dput(sb->s_root);
389                 sb->s_root = NULL;
390                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
391                        root->i_mode);
392                 ret = -EINVAL;
393                 goto free_sbi;
394         }
395
396         ret = 0;
397 out:
398         if (or)
399                 osd_end_request(or);
400         return ret;
401
402 free_sbi:
403         osduld_put_device(sbi->s_dev); /* NULL safe */
404         kfree(sbi);
405         goto out;
406 }
407
408 /*
409  * Set up the superblock (calls exofs_fill_super eventually)
410  */
411 static int exofs_get_sb(struct file_system_type *type,
412                           int flags, const char *dev_name,
413                           void *data, struct vfsmount *mnt)
414 {
415         struct exofs_mountopt opts;
416         int ret;
417
418         ret = parse_options(data, &opts);
419         if (ret)
420                 return ret;
421
422         opts.dev_name = dev_name;
423         return get_sb_nodev(type, flags, &opts, exofs_fill_super, mnt);
424 }
425
426 /*
427  * Return information about the file system state in the buffer.  This is used
428  * by the 'df' command, for example.
429  */
430 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
431 {
432         struct super_block *sb = dentry->d_sb;
433         struct exofs_sb_info *sbi = sb->s_fs_info;
434         struct osd_obj_id obj = {sbi->s_pid, 0};
435         struct osd_attr attrs[] = {
436                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
437                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
438                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
439                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
440         };
441         uint64_t capacity = ULLONG_MAX;
442         uint64_t used = ULLONG_MAX;
443         struct osd_request *or;
444         uint8_t cred_a[OSD_CAP_LEN];
445         int ret;
446
447         /* get used/capacity attributes */
448         exofs_make_credential(cred_a, &obj);
449
450         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
451         if (unlikely(!or)) {
452                 EXOFS_DBGMSG("exofs_statfs: osd_start_request failed.\n");
453                 return -ENOMEM;
454         }
455
456         osd_req_get_attributes(or, &obj);
457         osd_req_add_get_attr_list(or, attrs, ARRAY_SIZE(attrs));
458         ret = exofs_sync_op(or, sbi->s_timeout, cred_a);
459         if (unlikely(ret))
460                 goto out;
461
462         ret = extract_attr_from_req(or, &attrs[0]);
463         if (likely(!ret))
464                 capacity = get_unaligned_be64(attrs[0].val_ptr);
465         else
466                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
467
468         ret = extract_attr_from_req(or, &attrs[1]);
469         if (likely(!ret))
470                 used = get_unaligned_be64(attrs[1].val_ptr);
471         else
472                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
473
474         /* fill in the stats buffer */
475         buf->f_type = EXOFS_SUPER_MAGIC;
476         buf->f_bsize = EXOFS_BLKSIZE;
477         buf->f_blocks = (capacity >> EXOFS_BLKSHIFT);
478         buf->f_bfree = ((capacity - used) >> EXOFS_BLKSHIFT);
479         buf->f_bavail = buf->f_bfree;
480         buf->f_files = sbi->s_numfiles;
481         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
482         buf->f_namelen = EXOFS_NAME_LEN;
483
484 out:
485         osd_end_request(or);
486         return ret;
487 }
488
489 static const struct super_operations exofs_sops = {
490         .alloc_inode    = exofs_alloc_inode,
491         .destroy_inode  = exofs_destroy_inode,
492         .write_inode    = exofs_write_inode,
493         .delete_inode   = exofs_delete_inode,
494         .put_super      = exofs_put_super,
495         .write_super    = exofs_write_super,
496         .statfs         = exofs_statfs,
497 };
498
499 /******************************************************************************
500  * EXPORT OPERATIONS
501  *****************************************************************************/
502
503 struct dentry *exofs_get_parent(struct dentry *child)
504 {
505         unsigned long ino = exofs_parent_ino(child);
506
507         if (!ino)
508                 return NULL;
509
510         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
511 }
512
513 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
514                 u64 ino, u32 generation)
515 {
516         struct inode *inode;
517
518         inode = exofs_iget(sb, ino);
519         if (IS_ERR(inode))
520                 return ERR_CAST(inode);
521         if (generation && inode->i_generation != generation) {
522                 /* we didn't find the right inode.. */
523                 iput(inode);
524                 return ERR_PTR(-ESTALE);
525         }
526         return inode;
527 }
528
529 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
530                                 struct fid *fid, int fh_len, int fh_type)
531 {
532         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
533                                     exofs_nfs_get_inode);
534 }
535
536 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
537                                 struct fid *fid, int fh_len, int fh_type)
538 {
539         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
540                                     exofs_nfs_get_inode);
541 }
542
543 static const struct export_operations exofs_export_ops = {
544         .fh_to_dentry = exofs_fh_to_dentry,
545         .fh_to_parent = exofs_fh_to_parent,
546         .get_parent = exofs_get_parent,
547 };
548
549 /******************************************************************************
550  * INSMOD/RMMOD
551  *****************************************************************************/
552
553 /*
554  * struct that describes this file system
555  */
556 static struct file_system_type exofs_type = {
557         .owner          = THIS_MODULE,
558         .name           = "exofs",
559         .get_sb         = exofs_get_sb,
560         .kill_sb        = generic_shutdown_super,
561 };
562
563 static int __init init_exofs(void)
564 {
565         int err;
566
567         err = init_inodecache();
568         if (err)
569                 goto out;
570
571         err = register_filesystem(&exofs_type);
572         if (err)
573                 goto out_d;
574
575         return 0;
576 out_d:
577         destroy_inodecache();
578 out:
579         return err;
580 }
581
582 static void __exit exit_exofs(void)
583 {
584         unregister_filesystem(&exofs_type);
585         destroy_inodecache();
586 }
587
588 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
589 MODULE_DESCRIPTION("exofs");
590 MODULE_LICENSE("GPL");
591
592 module_init(init_exofs)
593 module_exit(exit_exofs)