1/*
2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
6 *
7 * Copyrights for code taken from ext2:
8 *     Copyright (C) 1992, 1993, 1994, 1995
9 *     Remy Card (card@masi.ibp.fr)
10 *     Laboratoire MASI - Institut Blaise Pascal
11 *     Universite Pierre et Marie Curie (Paris VI)
12 *     from
13 *     linux/fs/minix/inode.c
14 *     Copyright (C) 1991, 1992  Linus Torvalds
15 *
16 * This file is part of exofs.
17 *
18 * exofs is free software; you can redistribute it and/or modify
19 * it under the terms of the GNU General Public License as published by
20 * the Free Software Foundation.  Since it is based on ext2, and the only
21 * valid version of GPL for the Linux kernel is version 2, the only valid
22 * version of GPL for exofs is version 2.
23 *
24 * exofs is distributed in the hope that it will be useful,
25 * but WITHOUT ANY WARRANTY; without even the implied warranty of
26 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27 * GNU General Public License for more details.
28 *
29 * You should have received a copy of the GNU General Public License
30 * along with exofs; if not, write to the Free Software
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
32 */
33
34#include <linux/string.h>
35#include <linux/parser.h>
36#include <linux/vfs.h>
37#include <linux/random.h>
38#include <linux/exportfs.h>
39#include <linux/slab.h>
40
41#include "exofs.h"
42
43/******************************************************************************
44 * MOUNT OPTIONS
45 *****************************************************************************/
46
47/*
48 * struct to hold what we get from mount options
49 */
50struct exofs_mountopt {
51	const char *dev_name;
52	uint64_t pid;
53	int timeout;
54};
55
56/*
57 * exofs-specific mount-time options.
58 */
59enum { Opt_pid, Opt_to, Opt_mkfs, Opt_format, Opt_err };
60
61/*
62 * Our mount-time options.  These should ideally be 64-bit unsigned, but the
63 * kernel's parsing functions do not currently support that.  32-bit should be
64 * sufficient for most applications now.
65 */
66static match_table_t tokens = {
67	{Opt_pid, "pid=%u"},
68	{Opt_to, "to=%u"},
69	{Opt_err, NULL}
70};
71
72/*
73 * The main option parsing method.  Also makes sure that all of the mandatory
74 * mount options were set.
75 */
76static int parse_options(char *options, struct exofs_mountopt *opts)
77{
78	char *p;
79	substring_t args[MAX_OPT_ARGS];
80	int option;
81	bool s_pid = false;
82
83	EXOFS_DBGMSG("parse_options %s\n", options);
84	/* defaults */
85	memset(opts, 0, sizeof(*opts));
86	opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
87
88	while ((p = strsep(&options, ",")) != NULL) {
89		int token;
90		char str[32];
91
92		if (!*p)
93			continue;
94
95		token = match_token(p, tokens, args);
96		switch (token) {
97		case Opt_pid:
98			if (0 == match_strlcpy(str, &args[0], sizeof(str)))
99				return -EINVAL;
100			opts->pid = simple_strtoull(str, NULL, 0);
101			if (opts->pid < EXOFS_MIN_PID) {
102				EXOFS_ERR("Partition ID must be >= %u",
103					  EXOFS_MIN_PID);
104				return -EINVAL;
105			}
106			s_pid = 1;
107			break;
108		case Opt_to:
109			if (match_int(&args[0], &option))
110				return -EINVAL;
111			if (option <= 0) {
112				EXOFS_ERR("Timout must be > 0");
113				return -EINVAL;
114			}
115			opts->timeout = option * HZ;
116			break;
117		}
118	}
119
120	if (!s_pid) {
121		EXOFS_ERR("Need to specify the following options:\n");
122		EXOFS_ERR("    -o pid=pid_no_to_use\n");
123		return -EINVAL;
124	}
125
126	return 0;
127}
128
129/******************************************************************************
130 * INODE CACHE
131 *****************************************************************************/
132
133/*
134 * Our inode cache.  Isn't it pretty?
135 */
136static struct kmem_cache *exofs_inode_cachep;
137
138/*
139 * Allocate an inode in the cache
140 */
141static struct inode *exofs_alloc_inode(struct super_block *sb)
142{
143	struct exofs_i_info *oi;
144
145	oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
146	if (!oi)
147		return NULL;
148
149	oi->vfs_inode.i_version = 1;
150	return &oi->vfs_inode;
151}
152
153/*
154 * Remove an inode from the cache
155 */
156static void exofs_destroy_inode(struct inode *inode)
157{
158	kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
159}
160
161/*
162 * Initialize the inode
163 */
164static void exofs_init_once(void *foo)
165{
166	struct exofs_i_info *oi = foo;
167
168	inode_init_once(&oi->vfs_inode);
169}
170
171/*
172 * Create and initialize the inode cache
173 */
174static int init_inodecache(void)
175{
176	exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
177				sizeof(struct exofs_i_info), 0,
178				SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
179				exofs_init_once);
180	if (exofs_inode_cachep == NULL)
181		return -ENOMEM;
182	return 0;
183}
184
185/*
186 * Destroy the inode cache
187 */
188static void destroy_inodecache(void)
189{
190	kmem_cache_destroy(exofs_inode_cachep);
191}
192
193/******************************************************************************
194 * SUPERBLOCK FUNCTIONS
195 *****************************************************************************/
196static const struct super_operations exofs_sops;
197static const struct export_operations exofs_export_ops;
198
199/*
200 * Write the superblock to the OSD
201 */
202int exofs_sync_fs(struct super_block *sb, int wait)
203{
204	struct exofs_sb_info *sbi;
205	struct exofs_fscb *fscb;
206	struct exofs_io_state *ios;
207	int ret = -ENOMEM;
208
209	lock_super(sb);
210	sbi = sb->s_fs_info;
211	fscb = &sbi->s_fscb;
212
213	ret = exofs_get_io_state(&sbi->layout, &ios);
214	if (ret)
215		goto out;
216
217	/* Note: We only write the changing part of the fscb. .i.e upto the
218	 *       the fscb->s_dev_table_oid member. There is no read-modify-write
219	 *       here.
220	 */
221	ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
222	memset(fscb, 0, ios->length);
223	fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
224	fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
225	fscb->s_magic = cpu_to_le16(sb->s_magic);
226	fscb->s_newfs = 0;
227	fscb->s_version = EXOFS_FSCB_VER;
228
229	ios->obj.id = EXOFS_SUPER_ID;
230	ios->offset = 0;
231	ios->kern_buff = fscb;
232	ios->cred = sbi->s_cred;
233
234	ret = exofs_sbi_write(ios);
235	if (unlikely(ret)) {
236		EXOFS_ERR("%s: exofs_sbi_write failed.\n", __func__);
237		goto out;
238	}
239	sb->s_dirt = 0;
240
241out:
242	EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
243	exofs_put_io_state(ios);
244	unlock_super(sb);
245	return ret;
246}
247
248static void exofs_write_super(struct super_block *sb)
249{
250	if (!(sb->s_flags & MS_RDONLY))
251		exofs_sync_fs(sb, 1);
252	else
253		sb->s_dirt = 0;
254}
255
256static void _exofs_print_device(const char *msg, const char *dev_path,
257				struct osd_dev *od, u64 pid)
258{
259	const struct osd_dev_info *odi = osduld_device_info(od);
260
261	printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
262		msg, dev_path ?: "", odi->osdname, _LLU(pid));
263}
264
265void exofs_free_sbi(struct exofs_sb_info *sbi)
266{
267	while (sbi->layout.s_numdevs) {
268		int i = --sbi->layout.s_numdevs;
269		struct osd_dev *od = sbi->layout.s_ods[i];
270
271		if (od) {
272			sbi->layout.s_ods[i] = NULL;
273			osduld_put_device(od);
274		}
275	}
276	kfree(sbi);
277}
278
279/*
280 * This function is called when the vfs is freeing the superblock.  We just
281 * need to free our own part.
282 */
283static void exofs_put_super(struct super_block *sb)
284{
285	int num_pend;
286	struct exofs_sb_info *sbi = sb->s_fs_info;
287
288	if (sb->s_dirt)
289		exofs_write_super(sb);
290
291	/* make sure there are no pending commands */
292	for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
293	     num_pend = atomic_read(&sbi->s_curr_pending)) {
294		wait_queue_head_t wq;
295		init_waitqueue_head(&wq);
296		wait_event_timeout(wq,
297				  (atomic_read(&sbi->s_curr_pending) == 0),
298				  msecs_to_jiffies(100));
299	}
300
301	_exofs_print_device("Unmounting", NULL, sbi->layout.s_ods[0],
302			    sbi->layout.s_pid);
303
304	bdi_destroy(&sbi->bdi);
305	exofs_free_sbi(sbi);
306	sb->s_fs_info = NULL;
307}
308
309static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
310				    struct exofs_device_table *dt)
311{
312	u64 stripe_length;
313
314	sbi->data_map.odm_num_comps   =
315				le32_to_cpu(dt->dt_data_map.cb_num_comps);
316	sbi->data_map.odm_stripe_unit =
317				le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
318	sbi->data_map.odm_group_width =
319				le32_to_cpu(dt->dt_data_map.cb_group_width);
320	sbi->data_map.odm_group_depth =
321				le32_to_cpu(dt->dt_data_map.cb_group_depth);
322	sbi->data_map.odm_mirror_cnt  =
323				le32_to_cpu(dt->dt_data_map.cb_mirror_cnt);
324	sbi->data_map.odm_raid_algorithm  =
325				le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
326
327	if (sbi->data_map.odm_num_comps != numdevs) {
328		EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
329			  sbi->data_map.odm_num_comps, numdevs);
330		return -EINVAL;
331	}
332	if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
333		EXOFS_ERR("Only RAID_0 for now\n");
334		return -EINVAL;
335	}
336	if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
337		EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
338			  numdevs, sbi->data_map.odm_mirror_cnt);
339		return -EINVAL;
340	}
341
342	if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
343		EXOFS_ERR("Stripe Unit(0x%llx)"
344			  " must be Multples of PAGE_SIZE(0x%lx)\n",
345			  _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
346		return -EINVAL;
347	}
348
349	sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
350	sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
351
352	if (sbi->data_map.odm_group_width) {
353		sbi->layout.group_width = sbi->data_map.odm_group_width;
354		sbi->layout.group_depth = sbi->data_map.odm_group_depth;
355		if (!sbi->layout.group_depth) {
356			EXOFS_ERR("group_depth == 0 && group_width != 0\n");
357			return -EINVAL;
358		}
359		sbi->layout.group_count = sbi->data_map.odm_num_comps /
360						sbi->layout.mirrors_p1 /
361						sbi->data_map.odm_group_width;
362	} else {
363		if (sbi->data_map.odm_group_depth) {
364			printk(KERN_NOTICE "Warning: group_depth ignored "
365				"group_width == 0 && group_depth == %d\n",
366				sbi->data_map.odm_group_depth);
367			sbi->data_map.odm_group_depth = 0;
368		}
369		sbi->layout.group_width = sbi->data_map.odm_num_comps /
370							sbi->layout.mirrors_p1;
371		sbi->layout.group_depth = -1;
372		sbi->layout.group_count = 1;
373	}
374
375	stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
376	if (stripe_length >= (1ULL << 32)) {
377		EXOFS_ERR("Total Stripe length(0x%llx)"
378			  " >= 32bit is not supported\n", _LLU(stripe_length));
379		return -EINVAL;
380	}
381
382	return 0;
383}
384
385/* @odi is valid only as long as @fscb_dev is valid */
386static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
387			     struct osd_dev_info *odi)
388{
389	odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
390	memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
391
392	odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
393	odi->osdname = dt_dev->osdname;
394
395	if (dt_dev->long_name_offset)
396		return -EINVAL;
397
398	/* Make sure osdname is printable!
399	 * mkexofs should give us space for a null-terminator else the
400	 * device-table is invalid.
401	 */
402	if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
403		odi->osdname_len = sizeof(dt_dev->osdname) - 1;
404	dt_dev->osdname[odi->osdname_len] = 0;
405
406	/* If it's all zeros something is bad we read past end-of-obj */
407	return !(odi->systemid_len || odi->osdname_len);
408}
409
410static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
411				       unsigned table_count)
412{
413	struct exofs_sb_info *sbi = *psbi;
414	struct osd_dev *fscb_od;
415	struct osd_obj_id obj = {.partition = sbi->layout.s_pid,
416				 .id = EXOFS_DEVTABLE_ID};
417	struct exofs_device_table *dt;
418	unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
419					     sizeof(*dt);
420	unsigned numdevs, i;
421	int ret;
422
423	dt = kmalloc(table_bytes, GFP_KERNEL);
424	if (unlikely(!dt)) {
425		EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
426			  table_bytes);
427		return -ENOMEM;
428	}
429
430	fscb_od = sbi->layout.s_ods[0];
431	sbi->layout.s_ods[0] = NULL;
432	sbi->layout.s_numdevs = 0;
433	ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes);
434	if (unlikely(ret)) {
435		EXOFS_ERR("ERROR: reading device table\n");
436		goto out;
437	}
438
439	numdevs = le64_to_cpu(dt->dt_num_devices);
440	if (unlikely(!numdevs)) {
441		ret = -EINVAL;
442		goto out;
443	}
444	WARN_ON(table_count != numdevs);
445
446	ret = _read_and_match_data_map(sbi, numdevs, dt);
447	if (unlikely(ret))
448		goto out;
449
450	if (likely(numdevs > 1)) {
451		unsigned size = numdevs * sizeof(sbi->layout.s_ods[0]);
452
453		sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL);
454		if (unlikely(!sbi)) {
455			ret = -ENOMEM;
456			goto out;
457		}
458		memset(&sbi->layout.s_ods[1], 0,
459		       size - sizeof(sbi->layout.s_ods[0]));
460		*psbi = sbi;
461	}
462
463	for (i = 0; i < numdevs; i++) {
464		struct exofs_fscb fscb;
465		struct osd_dev_info odi;
466		struct osd_dev *od;
467
468		if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
469			EXOFS_ERR("ERROR: Read all-zeros device entry\n");
470			ret = -EINVAL;
471			goto out;
472		}
473
474		printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
475		       i, odi.osdname);
476
477		/* On all devices the device table is identical. The user can
478		 * specify any one of the participating devices on the command
479		 * line. We always keep them in device-table order.
480		 */
481		if (fscb_od && osduld_device_same(fscb_od, &odi)) {
482			sbi->layout.s_ods[i] = fscb_od;
483			++sbi->layout.s_numdevs;
484			fscb_od = NULL;
485			continue;
486		}
487
488		od = osduld_info_lookup(&odi);
489		if (unlikely(IS_ERR(od))) {
490			ret = PTR_ERR(od);
491			EXOFS_ERR("ERROR: device requested is not found "
492				  "osd_name-%s =>%d\n", odi.osdname, ret);
493			goto out;
494		}
495
496		sbi->layout.s_ods[i] = od;
497		++sbi->layout.s_numdevs;
498
499		/* Read the fscb of the other devices to make sure the FS
500		 * partition is there.
501		 */
502		ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb,
503				      sizeof(fscb));
504		if (unlikely(ret)) {
505			EXOFS_ERR("ERROR: Malformed participating device "
506				  "error reading fscb osd_name-%s\n",
507				  odi.osdname);
508			goto out;
509		}
510
511		/* TODO: verify other information is correct and FS-uuid
512		 *	 matches. Benny what did you say about device table
513		 *	 generation and old devices?
514		 */
515	}
516
517out:
518	kfree(dt);
519	if (unlikely(!ret && fscb_od)) {
520		EXOFS_ERR(
521		      "ERROR: Bad device-table container device not present\n");
522		osduld_put_device(fscb_od);
523		ret = -EINVAL;
524	}
525
526	return ret;
527}
528
529/*
530 * Read the superblock from the OSD and fill in the fields
531 */
532static int exofs_fill_super(struct super_block *sb, void *data, int silent)
533{
534	struct inode *root;
535	struct exofs_mountopt *opts = data;
536	struct exofs_sb_info *sbi;	/*extended info                  */
537	struct osd_dev *od;		/* Master device                 */
538	struct exofs_fscb fscb;		/*on-disk superblock info        */
539	struct osd_obj_id obj;
540	unsigned table_count;
541	int ret;
542
543	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
544	if (!sbi)
545		return -ENOMEM;
546
547	ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
548	if (ret)
549		goto free_bdi;
550
551	/* use mount options to fill superblock */
552	od = osduld_path_lookup(opts->dev_name);
553	if (IS_ERR(od)) {
554		ret = PTR_ERR(od);
555		goto free_sbi;
556	}
557
558	/* Default layout in case we do not have a device-table */
559	sbi->layout.stripe_unit = PAGE_SIZE;
560	sbi->layout.mirrors_p1 = 1;
561	sbi->layout.group_width = 1;
562	sbi->layout.group_depth = -1;
563	sbi->layout.group_count = 1;
564	sbi->layout.s_ods[0] = od;
565	sbi->layout.s_numdevs = 1;
566	sbi->layout.s_pid = opts->pid;
567	sbi->s_timeout = opts->timeout;
568
569	/* fill in some other data by hand */
570	memset(sb->s_id, 0, sizeof(sb->s_id));
571	strcpy(sb->s_id, "exofs");
572	sb->s_blocksize = EXOFS_BLKSIZE;
573	sb->s_blocksize_bits = EXOFS_BLKSHIFT;
574	sb->s_maxbytes = MAX_LFS_FILESIZE;
575	atomic_set(&sbi->s_curr_pending, 0);
576	sb->s_bdev = NULL;
577	sb->s_dev = 0;
578
579	obj.partition = sbi->layout.s_pid;
580	obj.id = EXOFS_SUPER_ID;
581	exofs_make_credential(sbi->s_cred, &obj);
582
583	ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb, sizeof(fscb));
584	if (unlikely(ret))
585		goto free_sbi;
586
587	sb->s_magic = le16_to_cpu(fscb.s_magic);
588	sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
589	sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
590
591	/* make sure what we read from the object store is correct */
592	if (sb->s_magic != EXOFS_SUPER_MAGIC) {
593		if (!silent)
594			EXOFS_ERR("ERROR: Bad magic value\n");
595		ret = -EINVAL;
596		goto free_sbi;
597	}
598	if (le32_to_cpu(fscb.s_version) != EXOFS_FSCB_VER) {
599		EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
600			  EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
601		ret = -EINVAL;
602		goto free_sbi;
603	}
604
605	/* start generation numbers from a random point */
606	get_random_bytes(&sbi->s_next_generation, sizeof(u32));
607	spin_lock_init(&sbi->s_next_gen_lock);
608
609	table_count = le64_to_cpu(fscb.s_dev_table_count);
610	if (table_count) {
611		ret = exofs_read_lookup_dev_table(&sbi, table_count);
612		if (unlikely(ret))
613			goto free_sbi;
614	}
615
616	/* set up operation vectors */
617	sb->s_bdi = &sbi->bdi;
618	sb->s_fs_info = sbi;
619	sb->s_op = &exofs_sops;
620	sb->s_export_op = &exofs_export_ops;
621	root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
622	if (IS_ERR(root)) {
623		EXOFS_ERR("ERROR: exofs_iget failed\n");
624		ret = PTR_ERR(root);
625		goto free_sbi;
626	}
627	sb->s_root = d_alloc_root(root);
628	if (!sb->s_root) {
629		iput(root);
630		EXOFS_ERR("ERROR: get root inode failed\n");
631		ret = -ENOMEM;
632		goto free_sbi;
633	}
634
635	if (!S_ISDIR(root->i_mode)) {
636		dput(sb->s_root);
637		sb->s_root = NULL;
638		EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
639		       root->i_mode);
640		ret = -EINVAL;
641		goto free_sbi;
642	}
643
644	_exofs_print_device("Mounting", opts->dev_name, sbi->layout.s_ods[0],
645			    sbi->layout.s_pid);
646	return 0;
647
648free_sbi:
649	bdi_destroy(&sbi->bdi);
650free_bdi:
651	EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
652		  opts->dev_name, sbi->layout.s_pid, ret);
653	exofs_free_sbi(sbi);
654	return ret;
655}
656
657/*
658 * Set up the superblock (calls exofs_fill_super eventually)
659 */
660static int exofs_get_sb(struct file_system_type *type,
661			  int flags, const char *dev_name,
662			  void *data, struct vfsmount *mnt)
663{
664	struct exofs_mountopt opts;
665	int ret;
666
667	ret = parse_options(data, &opts);
668	if (ret)
669		return ret;
670
671	opts.dev_name = dev_name;
672	return get_sb_nodev(type, flags, &opts, exofs_fill_super, mnt);
673}
674
675/*
676 * Return information about the file system state in the buffer.  This is used
677 * by the 'df' command, for example.
678 */
679static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
680{
681	struct super_block *sb = dentry->d_sb;
682	struct exofs_sb_info *sbi = sb->s_fs_info;
683	struct exofs_io_state *ios;
684	struct osd_attr attrs[] = {
685		ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
686			OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
687		ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
688			OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
689	};
690	uint64_t capacity = ULLONG_MAX;
691	uint64_t used = ULLONG_MAX;
692	uint8_t cred_a[OSD_CAP_LEN];
693	int ret;
694
695	ret = exofs_get_io_state(&sbi->layout, &ios);
696	if (ret) {
697		EXOFS_DBGMSG("exofs_get_io_state failed.\n");
698		return ret;
699	}
700
701	exofs_make_credential(cred_a, &ios->obj);
702	ios->cred = sbi->s_cred;
703	ios->in_attr = attrs;
704	ios->in_attr_len = ARRAY_SIZE(attrs);
705
706	ret = exofs_sbi_read(ios);
707	if (unlikely(ret))
708		goto out;
709
710	ret = extract_attr_from_ios(ios, &attrs[0]);
711	if (likely(!ret)) {
712		capacity = get_unaligned_be64(attrs[0].val_ptr);
713		if (unlikely(!capacity))
714			capacity = ULLONG_MAX;
715	} else
716		EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
717
718	ret = extract_attr_from_ios(ios, &attrs[1]);
719	if (likely(!ret))
720		used = get_unaligned_be64(attrs[1].val_ptr);
721	else
722		EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
723
724	/* fill in the stats buffer */
725	buf->f_type = EXOFS_SUPER_MAGIC;
726	buf->f_bsize = EXOFS_BLKSIZE;
727	buf->f_blocks = capacity >> 9;
728	buf->f_bfree = (capacity - used) >> 9;
729	buf->f_bavail = buf->f_bfree;
730	buf->f_files = sbi->s_numfiles;
731	buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
732	buf->f_namelen = EXOFS_NAME_LEN;
733
734out:
735	exofs_put_io_state(ios);
736	return ret;
737}
738
739static const struct super_operations exofs_sops = {
740	.alloc_inode    = exofs_alloc_inode,
741	.destroy_inode  = exofs_destroy_inode,
742	.write_inode    = exofs_write_inode,
743	.evict_inode    = exofs_evict_inode,
744	.put_super      = exofs_put_super,
745	.write_super    = exofs_write_super,
746	.sync_fs	= exofs_sync_fs,
747	.statfs         = exofs_statfs,
748};
749
750/******************************************************************************
751 * EXPORT OPERATIONS
752 *****************************************************************************/
753
754struct dentry *exofs_get_parent(struct dentry *child)
755{
756	unsigned long ino = exofs_parent_ino(child);
757
758	if (!ino)
759		return NULL;
760
761	return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
762}
763
764static struct inode *exofs_nfs_get_inode(struct super_block *sb,
765		u64 ino, u32 generation)
766{
767	struct inode *inode;
768
769	inode = exofs_iget(sb, ino);
770	if (IS_ERR(inode))
771		return ERR_CAST(inode);
772	if (generation && inode->i_generation != generation) {
773		/* we didn't find the right inode.. */
774		iput(inode);
775		return ERR_PTR(-ESTALE);
776	}
777	return inode;
778}
779
780static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
781				struct fid *fid, int fh_len, int fh_type)
782{
783	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
784				    exofs_nfs_get_inode);
785}
786
787static struct dentry *exofs_fh_to_parent(struct super_block *sb,
788				struct fid *fid, int fh_len, int fh_type)
789{
790	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
791				    exofs_nfs_get_inode);
792}
793
794static const struct export_operations exofs_export_ops = {
795	.fh_to_dentry = exofs_fh_to_dentry,
796	.fh_to_parent = exofs_fh_to_parent,
797	.get_parent = exofs_get_parent,
798};
799
800/******************************************************************************
801 * INSMOD/RMMOD
802 *****************************************************************************/
803
804/*
805 * struct that describes this file system
806 */
807static struct file_system_type exofs_type = {
808	.owner          = THIS_MODULE,
809	.name           = "exofs",
810	.get_sb         = exofs_get_sb,
811	.kill_sb        = generic_shutdown_super,
812};
813
814static int __init init_exofs(void)
815{
816	int err;
817
818	err = init_inodecache();
819	if (err)
820		goto out;
821
822	err = register_filesystem(&exofs_type);
823	if (err)
824		goto out_d;
825
826	return 0;
827out_d:
828	destroy_inodecache();
829out:
830	return err;
831}
832
833static void __exit exit_exofs(void)
834{
835	unregister_filesystem(&exofs_type);
836	destroy_inodecache();
837}
838
839MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
840MODULE_DESCRIPTION("exofs");
841MODULE_LICENSE("GPL");
842
843module_init(init_exofs)
844module_exit(exit_exofs)
845