blob: 9f1985e857e254c5315928e7daf84461baa1a3ca [file] [log] [blame]
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001/*
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>
Boaz Harrosh8cf74b32009-03-22 12:47:26 +020040#include <linux/exportfs.h>
Boaz Harroshba9e5e92008-10-28 16:11:41 +020041
42#include "exofs.h"
43
44/******************************************************************************
45 * MOUNT OPTIONS
46 *****************************************************************************/
47
48/*
49 * struct to hold what we get from mount options
50 */
51struct exofs_mountopt {
52 const char *dev_name;
53 uint64_t pid;
54 int timeout;
55};
56
57/*
58 * exofs-specific mount-time options.
59 */
60enum { 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 */
67static 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 */
77static 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 */
137static struct kmem_cache *exofs_inode_cachep;
138
139/*
140 * Allocate an inode in the cache
141 */
142static 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 */
157static 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 */
165static 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 */
175static 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 */
189static void destroy_inodecache(void)
190{
191 kmem_cache_destroy(exofs_inode_cachep);
192}
193
194/******************************************************************************
195 * SUPERBLOCK FUNCTIONS
196 *****************************************************************************/
197static const struct super_operations exofs_sops;
Boaz Harrosh8cf74b32009-03-22 12:47:26 +0200198static const struct export_operations exofs_export_ops;
Boaz Harroshba9e5e92008-10-28 16:11:41 +0200199
200/*
201 * Write the superblock to the OSD
202 */
203static 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_kernel();
218 sbi = sb->s_fs_info;
219 fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
220 fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
221 fscb->s_magic = cpu_to_le16(sb->s_magic);
222 fscb->s_newfs = 0;
223
224 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
225 if (unlikely(!or)) {
226 EXOFS_ERR("exofs_write_super: osd_start_request failed.\n");
227 goto out;
228 }
229
230 obj.partition = sbi->s_pid;
231 obj.id = EXOFS_SUPER_ID;
232 ret = osd_req_write_kern(or, &obj, 0, fscb, sizeof(*fscb));
233 if (unlikely(ret)) {
234 EXOFS_ERR("exofs_write_super: osd_req_write_kern failed.\n");
235 goto out;
236 }
237
238 ret = exofs_sync_op(or, sbi->s_timeout, sbi->s_cred);
239 if (unlikely(ret)) {
240 EXOFS_ERR("exofs_write_super: exofs_sync_op failed.\n");
241 goto out;
242 }
243 sb->s_dirt = 0;
244
245out:
246 if (or)
247 osd_end_request(or);
248 unlock_kernel();
249 kfree(fscb);
250}
251
252/*
253 * This function is called when the vfs is freeing the superblock. We just
254 * need to free our own part.
255 */
256static void exofs_put_super(struct super_block *sb)
257{
258 int num_pend;
259 struct exofs_sb_info *sbi = sb->s_fs_info;
260
261 /* make sure there are no pending commands */
262 for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
263 num_pend = atomic_read(&sbi->s_curr_pending)) {
264 wait_queue_head_t wq;
265 init_waitqueue_head(&wq);
266 wait_event_timeout(wq,
267 (atomic_read(&sbi->s_curr_pending) == 0),
268 msecs_to_jiffies(100));
269 }
270
271 osduld_put_device(sbi->s_dev);
272 kfree(sb->s_fs_info);
273 sb->s_fs_info = NULL;
274}
275
276/*
277 * Read the superblock from the OSD and fill in the fields
278 */
279static int exofs_fill_super(struct super_block *sb, void *data, int silent)
280{
281 struct inode *root;
282 struct exofs_mountopt *opts = data;
283 struct exofs_sb_info *sbi; /*extended info */
284 struct exofs_fscb fscb; /*on-disk superblock info */
285 struct osd_request *or = NULL;
286 struct osd_obj_id obj;
287 int ret;
288
289 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
290 if (!sbi)
291 return -ENOMEM;
292 sb->s_fs_info = sbi;
293
294 /* use mount options to fill superblock */
295 sbi->s_dev = osduld_path_lookup(opts->dev_name);
296 if (IS_ERR(sbi->s_dev)) {
297 ret = PTR_ERR(sbi->s_dev);
298 sbi->s_dev = NULL;
299 goto free_sbi;
300 }
301
302 sbi->s_pid = opts->pid;
303 sbi->s_timeout = opts->timeout;
304
305 /* fill in some other data by hand */
306 memset(sb->s_id, 0, sizeof(sb->s_id));
307 strcpy(sb->s_id, "exofs");
308 sb->s_blocksize = EXOFS_BLKSIZE;
309 sb->s_blocksize_bits = EXOFS_BLKSHIFT;
310 sb->s_maxbytes = MAX_LFS_FILESIZE;
311 atomic_set(&sbi->s_curr_pending, 0);
312 sb->s_bdev = NULL;
313 sb->s_dev = 0;
314
315 /* read data from on-disk superblock object */
316 obj.partition = sbi->s_pid;
317 obj.id = EXOFS_SUPER_ID;
318 exofs_make_credential(sbi->s_cred, &obj);
319
320 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
321 if (unlikely(!or)) {
322 if (!silent)
323 EXOFS_ERR(
324 "exofs_fill_super: osd_start_request failed.\n");
325 ret = -ENOMEM;
326 goto free_sbi;
327 }
328 ret = osd_req_read_kern(or, &obj, 0, &fscb, sizeof(fscb));
329 if (unlikely(ret)) {
330 if (!silent)
331 EXOFS_ERR(
332 "exofs_fill_super: osd_req_read_kern failed.\n");
333 ret = -ENOMEM;
334 goto free_sbi;
335 }
336
337 ret = exofs_sync_op(or, sbi->s_timeout, sbi->s_cred);
338 if (unlikely(ret)) {
339 if (!silent)
340 EXOFS_ERR("exofs_fill_super: exofs_sync_op failed.\n");
341 ret = -EIO;
342 goto free_sbi;
343 }
344
345 sb->s_magic = le16_to_cpu(fscb.s_magic);
346 sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
347 sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
348
349 /* make sure what we read from the object store is correct */
350 if (sb->s_magic != EXOFS_SUPER_MAGIC) {
351 if (!silent)
352 EXOFS_ERR("ERROR: Bad magic value\n");
353 ret = -EINVAL;
354 goto free_sbi;
355 }
356
357 /* start generation numbers from a random point */
358 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
359 spin_lock_init(&sbi->s_next_gen_lock);
360
361 /* set up operation vectors */
362 sb->s_op = &exofs_sops;
Boaz Harrosh8cf74b32009-03-22 12:47:26 +0200363 sb->s_export_op = &exofs_export_ops;
Boaz Harroshba9e5e92008-10-28 16:11:41 +0200364 root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
365 if (IS_ERR(root)) {
366 EXOFS_ERR("ERROR: exofs_iget failed\n");
367 ret = PTR_ERR(root);
368 goto free_sbi;
369 }
370 sb->s_root = d_alloc_root(root);
371 if (!sb->s_root) {
372 iput(root);
373 EXOFS_ERR("ERROR: get root inode failed\n");
374 ret = -ENOMEM;
375 goto free_sbi;
376 }
377
378 if (!S_ISDIR(root->i_mode)) {
379 dput(sb->s_root);
380 sb->s_root = NULL;
381 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
382 root->i_mode);
383 ret = -EINVAL;
384 goto free_sbi;
385 }
386
387 ret = 0;
388out:
389 if (or)
390 osd_end_request(or);
391 return ret;
392
393free_sbi:
394 osduld_put_device(sbi->s_dev); /* NULL safe */
395 kfree(sbi);
396 goto out;
397}
398
399/*
400 * Set up the superblock (calls exofs_fill_super eventually)
401 */
402static int exofs_get_sb(struct file_system_type *type,
403 int flags, const char *dev_name,
404 void *data, struct vfsmount *mnt)
405{
406 struct exofs_mountopt opts;
407 int ret;
408
409 ret = parse_options(data, &opts);
410 if (ret)
411 return ret;
412
413 opts.dev_name = dev_name;
414 return get_sb_nodev(type, flags, &opts, exofs_fill_super, mnt);
415}
416
417/*
418 * Return information about the file system state in the buffer. This is used
419 * by the 'df' command, for example.
420 */
421static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
422{
423 struct super_block *sb = dentry->d_sb;
424 struct exofs_sb_info *sbi = sb->s_fs_info;
425 struct osd_obj_id obj = {sbi->s_pid, 0};
426 struct osd_attr attrs[] = {
427 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
428 OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
429 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
430 OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
431 };
432 uint64_t capacity = ULLONG_MAX;
433 uint64_t used = ULLONG_MAX;
434 struct osd_request *or;
435 uint8_t cred_a[OSD_CAP_LEN];
436 int ret;
437
438 /* get used/capacity attributes */
439 exofs_make_credential(cred_a, &obj);
440
441 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
442 if (unlikely(!or)) {
443 EXOFS_DBGMSG("exofs_statfs: osd_start_request failed.\n");
444 return -ENOMEM;
445 }
446
447 osd_req_get_attributes(or, &obj);
448 osd_req_add_get_attr_list(or, attrs, ARRAY_SIZE(attrs));
449 ret = exofs_sync_op(or, sbi->s_timeout, cred_a);
450 if (unlikely(ret))
451 goto out;
452
453 ret = extract_attr_from_req(or, &attrs[0]);
454 if (likely(!ret))
455 capacity = get_unaligned_be64(attrs[0].val_ptr);
456 else
457 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
458
459 ret = extract_attr_from_req(or, &attrs[1]);
460 if (likely(!ret))
461 used = get_unaligned_be64(attrs[1].val_ptr);
462 else
463 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
464
465 /* fill in the stats buffer */
466 buf->f_type = EXOFS_SUPER_MAGIC;
467 buf->f_bsize = EXOFS_BLKSIZE;
468 buf->f_blocks = (capacity >> EXOFS_BLKSHIFT);
469 buf->f_bfree = ((capacity - used) >> EXOFS_BLKSHIFT);
470 buf->f_bavail = buf->f_bfree;
471 buf->f_files = sbi->s_numfiles;
472 buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
473 buf->f_namelen = EXOFS_NAME_LEN;
474
475out:
476 osd_end_request(or);
477 return ret;
478}
479
480static const struct super_operations exofs_sops = {
481 .alloc_inode = exofs_alloc_inode,
482 .destroy_inode = exofs_destroy_inode,
483 .write_inode = exofs_write_inode,
484 .delete_inode = exofs_delete_inode,
485 .put_super = exofs_put_super,
486 .write_super = exofs_write_super,
487 .statfs = exofs_statfs,
488};
489
490/******************************************************************************
Boaz Harrosh8cf74b32009-03-22 12:47:26 +0200491 * EXPORT OPERATIONS
492 *****************************************************************************/
493
494struct dentry *exofs_get_parent(struct dentry *child)
495{
496 unsigned long ino = exofs_parent_ino(child);
497
498 if (!ino)
499 return NULL;
500
501 return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
502}
503
504static struct inode *exofs_nfs_get_inode(struct super_block *sb,
505 u64 ino, u32 generation)
506{
507 struct inode *inode;
508
509 inode = exofs_iget(sb, ino);
510 if (IS_ERR(inode))
511 return ERR_CAST(inode);
512 if (generation && inode->i_generation != generation) {
513 /* we didn't find the right inode.. */
514 iput(inode);
515 return ERR_PTR(-ESTALE);
516 }
517 return inode;
518}
519
520static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
521 struct fid *fid, int fh_len, int fh_type)
522{
523 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
524 exofs_nfs_get_inode);
525}
526
527static struct dentry *exofs_fh_to_parent(struct super_block *sb,
528 struct fid *fid, int fh_len, int fh_type)
529{
530 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
531 exofs_nfs_get_inode);
532}
533
534static const struct export_operations exofs_export_ops = {
535 .fh_to_dentry = exofs_fh_to_dentry,
536 .fh_to_parent = exofs_fh_to_parent,
537 .get_parent = exofs_get_parent,
538};
539
540/******************************************************************************
Boaz Harroshba9e5e92008-10-28 16:11:41 +0200541 * INSMOD/RMMOD
542 *****************************************************************************/
543
544/*
545 * struct that describes this file system
546 */
547static struct file_system_type exofs_type = {
548 .owner = THIS_MODULE,
549 .name = "exofs",
550 .get_sb = exofs_get_sb,
551 .kill_sb = generic_shutdown_super,
552};
553
554static int __init init_exofs(void)
555{
556 int err;
557
558 err = init_inodecache();
559 if (err)
560 goto out;
561
562 err = register_filesystem(&exofs_type);
563 if (err)
564 goto out_d;
565
566 return 0;
567out_d:
568 destroy_inodecache();
569out:
570 return err;
571}
572
573static void __exit exit_exofs(void)
574{
575 unregister_filesystem(&exofs_type);
576 destroy_inodecache();
577}
578
579MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
580MODULE_DESCRIPTION("exofs");
581MODULE_LICENSE("GPL");
582
583module_init(init_exofs)
584module_exit(exit_exofs)