2 * 2007+ Copyright (c) Evgeniy Polyakov <zbr@ioremap.net>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
16 #include <linux/module.h>
17 #include <linux/backing-dev.h>
18 #include <linux/crypto.h>
20 #include <linux/jhash.h>
21 #include <linux/hash.h>
22 #include <linux/ktime.h>
24 #include <linux/mount.h>
25 #include <linux/pagemap.h>
26 #include <linux/pagevec.h>
27 #include <linux/parser.h>
28 #include <linux/swap.h>
29 #include <linux/slab.h>
30 #include <linux/statfs.h>
31 #include <linux/writeback.h>
32 #include <linux/quotaops.h>
36 #define POHMELFS_MAGIC_NUM 0x504f482e
38 static struct kmem_cache
*pohmelfs_inode_cache
;
39 static atomic_t psb_bdi_num
= ATOMIC_INIT(0);
42 * Removes inode from all trees, drops local name cache and removes all queued
43 * requests for object removal.
45 void pohmelfs_inode_del_inode(struct pohmelfs_sb
*psb
, struct pohmelfs_inode
*pi
)
47 mutex_lock(&pi
->offset_lock
);
48 pohmelfs_free_names(pi
);
49 mutex_unlock(&pi
->offset_lock
);
51 dprintk("%s: deleted stuff in ino: %llu.\n", __func__
, pi
->ino
);
55 * Sync inode to server.
56 * Returns zero in success and negative error value otherwise.
57 * It will gather path to root directory into structures containing
58 * creation mode, permissions and names, so that the whole path
59 * to given inode could be created using only single network command.
61 int pohmelfs_write_inode_create(struct inode
*inode
, struct netfs_trans
*trans
)
63 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
64 int err
= -ENOMEM
, size
;
65 struct netfs_cmd
*cmd
;
67 int cur_len
= netfs_trans_cur_len(trans
);
69 if (unlikely(cur_len
< 0))
72 cmd
= netfs_trans_current(trans
);
73 cur_len
-= sizeof(struct netfs_cmd
);
75 data
= (void *)(cmd
+ 1);
77 err
= pohmelfs_construct_path_string(pi
, data
, cur_len
);
83 cmd
->start
= i_size_read(inode
);
84 cmd
->cmd
= NETFS_CREATE
;
87 cmd
->ext
= inode
->i_mode
;
89 netfs_convert_cmd(cmd
);
91 netfs_trans_update(cmd
, trans
, size
);
96 printk("%s: completed ino: %llu, err: %d.\n", __func__
, pi
->ino
, err
);
100 static int pohmelfs_write_trans_complete(struct page
**pages
, unsigned int page_num
,
101 void *private, int err
)
105 dprintk("%s: pages: %lu-%lu, page_num: %u, err: %d.\n",
106 __func__
, pages
[0]->index
, pages
[page_num
-1]->index
,
109 for (i
= 0; i
< page_num
; i
++) {
110 struct page
*page
= pages
[i
];
115 end_page_writeback(page
);
119 set_page_dirty(page
);
123 page_cache_release(page
);
125 /* dprintk("%s: %3u/%u: page: %p.\n", __func__, i, page_num, page); */
130 static int pohmelfs_inode_has_dirty_pages(struct address_space
*mapping
, pgoff_t index
)
136 ret
= radix_tree_gang_lookup_tag(&mapping
->page_tree
,
137 (void **)&page
, index
, 1, PAGECACHE_TAG_DIRTY
);
142 static int pohmelfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
144 struct inode
*inode
= mapping
->host
;
145 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
146 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
151 pgoff_t end
; /* Inclusive */
155 if (wbc
->range_cyclic
) {
156 index
= mapping
->writeback_index
; /* Start from prev offset */
159 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
160 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
161 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
166 while (!done
&& (index
<= end
)) {
167 unsigned int i
= min(end
- index
, (pgoff_t
)psb
->trans_max_pages
);
169 struct netfs_trans
*trans
;
171 err
= pohmelfs_inode_has_dirty_pages(mapping
, index
);
175 err
= pohmelfs_path_length(pi
);
186 trans
= netfs_trans_alloc(psb
, path_len
, 0, i
);
191 trans
->complete
= &pohmelfs_write_trans_complete
;
193 trans
->page_num
= nr_pages
= find_get_pages_tag(mapping
, &index
,
194 PAGECACHE_TAG_DIRTY
, trans
->page_num
,
197 dprintk("%s: t: %p, nr_pages: %u, end: %lu, index: %lu, max: %u.\n",
198 __func__
, trans
, nr_pages
, end
, index
, trans
->page_num
);
203 err
= pohmelfs_write_inode_create(inode
, trans
);
210 for (i
= 0; i
< trans
->page_num
; i
++) {
211 struct page
*page
= trans
->pages
[i
];
215 if (unlikely(page
->mapping
!= mapping
))
218 if (!wbc
->range_cyclic
&& page
->index
> end
) {
223 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
224 wait_on_page_writeback(page
);
226 if (PageWriteback(page
) ||
227 !clear_page_dirty_for_io(page
)) {
228 dprintk("%s: not clear for io page: %p, writeback: %d.\n",
229 __func__
, page
, PageWriteback(page
));
233 set_page_writeback(page
);
235 trans
->attached_size
+= page_private(page
);
236 trans
->attached_pages
++;
238 dprintk("%s: %u/%u added trans: %p, gen: %u, page: %p, [High: %d], size: %lu, idx: %lu.\n",
239 __func__
, i
, trans
->page_num
, trans
, trans
->gen
, page
,
240 !!PageHighMem(page
), page_private(page
), page
->index
);
244 if (wbc
->nr_to_write
<= 0)
250 trans
->pages
[i
] = NULL
;
253 err
= netfs_trans_finish(trans
, psb
);
261 netfs_trans_reset(trans
);
262 netfs_trans_put(trans
);
266 if (!scanned
&& !done
) {
268 * We hit the last page and there is more work to be done: wrap
269 * back to the start of the file
276 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
277 mapping
->writeback_index
= index
;
283 * Inode writeback creation completion callback.
284 * Only invoked for just created inodes, which do not have pages attached,
285 * like dirs and empty files.
287 static int pohmelfs_write_inode_complete(struct page
**pages
, unsigned int page_num
,
288 void *private, int err
)
290 struct inode
*inode
= private;
291 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
295 mark_inode_dirty(inode
);
296 clear_bit(NETFS_INODE_REMOTE_SYNCED
, &pi
->state
);
298 set_bit(NETFS_INODE_REMOTE_SYNCED
, &pi
->state
);
301 pohmelfs_put_inode(pi
);
307 int pohmelfs_write_create_inode(struct pohmelfs_inode
*pi
)
309 struct netfs_trans
*t
;
310 struct inode
*inode
= &pi
->vfs_inode
;
311 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
314 if (test_bit(NETFS_INODE_REMOTE_SYNCED
, &pi
->state
))
317 dprintk("%s: started ino: %llu.\n", __func__
, pi
->ino
);
319 err
= pohmelfs_path_length(pi
);
323 t
= netfs_trans_alloc(psb
, err
+ 1, 0, 0);
328 t
->complete
= pohmelfs_write_inode_complete
;
329 t
->private = igrab(inode
);
335 err
= pohmelfs_write_inode_create(inode
, t
);
339 netfs_trans_finish(t
, POHMELFS_SB(inode
->i_sb
));
351 * Sync all not-yet-created children in given directory to the server.
353 static int pohmelfs_write_inode_create_children(struct inode
*inode
)
355 struct pohmelfs_inode
*parent
= POHMELFS_I(inode
);
356 struct super_block
*sb
= inode
->i_sb
;
357 struct pohmelfs_name
*n
;
359 while (!list_empty(&parent
->sync_create_list
)) {
361 mutex_lock(&parent
->offset_lock
);
362 if (!list_empty(&parent
->sync_create_list
)) {
363 n
= list_first_entry(&parent
->sync_create_list
,
364 struct pohmelfs_name
, sync_create_entry
);
365 list_del_init(&n
->sync_create_entry
);
367 mutex_unlock(&parent
->offset_lock
);
372 inode
= ilookup(sb
, n
->ino
);
374 dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n",
375 __func__
, parent
->ino
, n
->ino
, inode
);
377 if (inode
&& (inode
->i_state
& I_DIRTY
)) {
378 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
379 pohmelfs_write_create_inode(pi
);
380 /* pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0); */
389 * Removes given child from given inode on server.
391 int pohmelfs_remove_child(struct pohmelfs_inode
*pi
, struct pohmelfs_name
*n
)
393 return pohmelfs_meta_command_data(pi
, pi
->ino
, NETFS_REMOVE
, NULL
, 0, NULL
, NULL
, 0);
397 * Writeback for given inode.
399 static int pohmelfs_write_inode(struct inode
*inode
,
400 struct writeback_control
*wbc
)
402 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
404 pohmelfs_write_create_inode(pi
);
405 pohmelfs_write_inode_create_children(inode
);
411 * It is not exported, sorry...
413 static inline wait_queue_head_t
*page_waitqueue(struct page
*page
)
415 const struct zone
*zone
= page_zone(page
);
417 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
420 static int pohmelfs_wait_on_page_locked(struct page
*page
)
422 struct pohmelfs_sb
*psb
= POHMELFS_SB(page
->mapping
->host
->i_sb
);
423 long ret
= psb
->wait_on_page_timeout
;
424 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
427 if (!PageLocked(page
))
431 prepare_to_wait(page_waitqueue(page
),
432 &wait
.wait
, TASK_INTERRUPTIBLE
);
434 dprintk("%s: page: %p, locked: %d, uptodate: %d, error: %d, flags: %lx.\n",
435 __func__
, page
, PageLocked(page
), PageUptodate(page
),
436 PageError(page
), page
->flags
);
438 if (!PageLocked(page
))
441 if (!signal_pending(current
)) {
442 ret
= schedule_timeout(ret
);
450 finish_wait(page_waitqueue(page
), &wait
.wait
);
457 SetPageUptodate(page
);
460 printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n",
461 __func__
, page
, PageUptodate(page
), PageLocked(page
), err
);
466 static int pohmelfs_read_page_complete(struct page
**pages
, unsigned int page_num
,
467 void *private, int err
)
469 struct page
*page
= private;
471 if (PageChecked(page
))
475 dprintk("%s: page: %p, err: %d.\n", __func__
, page
, err
);
485 * Read a page from remote server.
486 * Function will wait until page is unlocked.
488 static int pohmelfs_readpage(struct file
*file
, struct page
*page
)
490 struct inode
*inode
= page
->mapping
->host
;
491 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
492 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
493 struct netfs_trans
*t
;
494 struct netfs_cmd
*cmd
;
499 err
= pohmelfs_data_lock(pi
, page
->index
<< PAGE_CACHE_SHIFT
,
500 PAGE_SIZE
, POHMELFS_READ_LOCK
);
504 isize
= i_size_read(inode
);
505 if (isize
<= page
->index
<< PAGE_CACHE_SHIFT
) {
506 SetPageUptodate(page
);
511 path_len
= pohmelfs_path_length(pi
);
517 t
= netfs_trans_alloc(psb
, path_len
, NETFS_TRANS_SINGLE_DST
, 0);
523 t
->complete
= pohmelfs_read_page_complete
;
526 cmd
= netfs_trans_current(t
);
527 data
= (void *)(cmd
+ 1);
529 err
= pohmelfs_construct_path_string(pi
, data
, path_len
);
536 cmd
->start
= page
->index
;
537 cmd
->start
<<= PAGE_CACHE_SHIFT
;
538 cmd
->size
= PAGE_CACHE_SIZE
+ path_len
;
539 cmd
->cmd
= NETFS_READ_PAGE
;
542 dprintk("%s: path: '%s', page: %p, ino: %llu, start: %llu, size: %lu.\n",
543 __func__
, (char *)data
, page
, pi
->ino
, cmd
->start
, PAGE_CACHE_SIZE
);
545 netfs_convert_cmd(cmd
);
546 netfs_trans_update(cmd
, t
, path_len
);
548 err
= netfs_trans_finish(t
, psb
);
552 return pohmelfs_wait_on_page_locked(page
);
559 if (PageLocked(page
))
562 printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n",
563 __func__
, page
, page
->index
<< PAGE_CACHE_SHIFT
, PAGE_CACHE_SIZE
, err
);
569 * Write begin/end magic.
570 * Allocates a page and writes inode if it was not synced to server before.
572 static int pohmelfs_write_begin(struct file
*file
, struct address_space
*mapping
,
573 loff_t pos
, unsigned len
, unsigned flags
,
574 struct page
**pagep
, void **fsdata
)
576 struct inode
*inode
= mapping
->host
;
584 index
= pos
>> PAGE_CACHE_SHIFT
;
585 start
= pos
& (PAGE_CACHE_SIZE
- 1);
588 page
= grab_cache_page(mapping
, index
);
590 dprintk("%s: page: %p pos: %llu, len: %u, index: %lu, start: %u, end: %u, uptodate: %d.\n",
591 __func__
, page
, pos
, len
, index
, start
, end
, PageUptodate(page
));
598 while (!PageUptodate(page
)) {
599 if (start
&& test_bit(NETFS_INODE_REMOTE_SYNCED
, &POHMELFS_I(inode
)->state
)) {
600 err
= pohmelfs_readpage(file
, page
);
608 if (len
!= PAGE_CACHE_SIZE
) {
609 void *kaddr
= kmap_atomic(page
, KM_USER0
);
611 memset(kaddr
+ start
, 0, PAGE_CACHE_SIZE
- start
);
612 flush_dcache_page(page
);
613 kunmap_atomic(kaddr
, KM_USER0
);
615 SetPageUptodate(page
);
618 set_page_private(page
, end
);
625 page_cache_release(page
);
631 static int pohmelfs_write_end(struct file
*file
, struct address_space
*mapping
,
632 loff_t pos
, unsigned len
, unsigned copied
,
633 struct page
*page
, void *fsdata
)
635 struct inode
*inode
= mapping
->host
;
638 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
639 void *kaddr
= kmap_atomic(page
, KM_USER0
);
641 memset(kaddr
+ from
+ copied
, 0, len
- copied
);
642 flush_dcache_page(page
);
643 kunmap_atomic(kaddr
, KM_USER0
);
646 SetPageUptodate(page
);
647 set_page_dirty(page
);
649 dprintk("%s: page: %p [U: %d, D: %d, L: %d], pos: %llu, len: %u, copied: %u.\n",
651 PageUptodate(page
), PageDirty(page
), PageLocked(page
),
654 flush_dcache_page(page
);
657 page_cache_release(page
);
659 if (pos
+ copied
> inode
->i_size
) {
660 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
662 psb
->avail_size
-= pos
+ copied
- inode
->i_size
;
664 i_size_write(inode
, pos
+ copied
);
670 static int pohmelfs_readpages_trans_complete(struct page
**__pages
, unsigned int page_num
,
671 void *private, int err
)
673 struct pohmelfs_inode
*pi
= private;
675 struct page
**pages
, *page
= (struct page
*)__pages
;
676 loff_t index
= page
->index
;
678 pages
= kzalloc(sizeof(void *) * page_num
, GFP_NOIO
);
682 num
= find_get_pages_contig(pi
->vfs_inode
.i_mapping
, index
, page_num
, pages
);
688 for (i
=0; i
<num
; ++i
) {
692 printk("%s: %u/%u: page: %p, index: %lu, uptodate: %d, locked: %d, err: %d.\n",
693 __func__
, i
, num
, page
, page
->index
,
694 PageUptodate(page
), PageLocked(page
), err
);
696 if (!PageChecked(page
)) {
701 page_cache_release(page
);
702 page_cache_release(page
);
710 static int pohmelfs_send_readpages(struct pohmelfs_inode
*pi
, struct page
*first
, unsigned int num
)
712 struct netfs_trans
*t
;
713 struct netfs_cmd
*cmd
;
714 struct pohmelfs_sb
*psb
= POHMELFS_SB(pi
->vfs_inode
.i_sb
);
718 err
= pohmelfs_data_lock(pi
, first
->index
<< PAGE_CACHE_SHIFT
,
719 num
* PAGE_SIZE
, POHMELFS_READ_LOCK
);
723 path_len
= pohmelfs_path_length(pi
);
729 t
= netfs_trans_alloc(psb
, path_len
, NETFS_TRANS_SINGLE_DST
, 0);
735 cmd
= netfs_trans_current(t
);
736 data
= (void *)(cmd
+ 1);
738 t
->complete
= pohmelfs_readpages_trans_complete
;
741 t
->pages
= (struct page
**)first
;
743 err
= pohmelfs_construct_path_string(pi
, data
, path_len
);
749 cmd
->cmd
= NETFS_READ_PAGES
;
750 cmd
->start
= first
->index
;
751 cmd
->start
<<= PAGE_CACHE_SHIFT
;
752 cmd
->size
= (num
<< 8 | PAGE_CACHE_SHIFT
);
756 dprintk("%s: t: %p, gen: %u, path: '%s', path_len: %u, "
757 "start: %lu, num: %u.\n",
758 __func__
, t
, t
->gen
, (char *)data
, path_len
,
761 netfs_convert_cmd(cmd
);
762 netfs_trans_update(cmd
, t
, path_len
);
764 return netfs_trans_finish(t
, psb
);
769 pohmelfs_readpages_trans_complete((struct page
**)first
, num
, pi
, err
);
773 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
775 static int pohmelfs_readpages(struct file
*file
, struct address_space
*mapping
,
776 struct list_head
*pages
, unsigned nr_pages
)
778 unsigned int page_idx
, num
= 0;
779 struct page
*page
= NULL
, *first
= NULL
;
781 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
782 page
= list_to_page(pages
);
784 prefetchw(&page
->flags
);
785 list_del(&page
->lru
);
787 if (!add_to_page_cache_lru(page
, mapping
,
788 page
->index
, GFP_KERNEL
)) {
796 dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n",
797 __func__
, page
, page
->index
, first
->index
);
799 if (unlikely(first
->index
+ num
!= page
->index
) || (num
> 500)) {
800 pohmelfs_send_readpages(POHMELFS_I(mapping
->host
),
809 pohmelfs_send_readpages(POHMELFS_I(mapping
->host
), first
, num
);
812 * This will be sync read, so when last page is processed,
813 * all previous are alerady unlocked and ready to be used.
819 * Small addres space operations for POHMELFS.
821 const struct address_space_operations pohmelfs_aops
= {
822 .readpage
= pohmelfs_readpage
,
823 .readpages
= pohmelfs_readpages
,
824 .writepages
= pohmelfs_writepages
,
825 .write_begin
= pohmelfs_write_begin
,
826 .write_end
= pohmelfs_write_end
,
827 .set_page_dirty
= __set_page_dirty_nobuffers
,
831 * ->detroy_inode() callback. Deletes inode from the caches
832 * and frees private data.
834 static void pohmelfs_destroy_inode(struct inode
*inode
)
836 struct super_block
*sb
= inode
->i_sb
;
837 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
838 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
840 /* pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK); */
842 pohmelfs_inode_del_inode(psb
, pi
);
844 dprintk("%s: pi: %p, inode: %p, ino: %llu.\n",
845 __func__
, pi
, &pi
->vfs_inode
, pi
->ino
);
846 kmem_cache_free(pohmelfs_inode_cache
, pi
);
847 atomic_long_dec(&psb
->total_inodes
);
851 * ->alloc_inode() callback. Allocates inode and initializes private data.
853 static struct inode
*pohmelfs_alloc_inode(struct super_block
*sb
)
855 struct pohmelfs_inode
*pi
;
857 pi
= kmem_cache_alloc(pohmelfs_inode_cache
, GFP_NOIO
);
861 pi
->hash_root
= RB_ROOT
;
862 mutex_init(&pi
->offset_lock
);
864 INIT_LIST_HEAD(&pi
->sync_create_list
);
866 INIT_LIST_HEAD(&pi
->inode_entry
);
873 dprintk("%s: pi: %p, inode: %p.\n", __func__
, pi
, &pi
->vfs_inode
);
875 atomic_long_inc(&POHMELFS_SB(sb
)->total_inodes
);
877 return &pi
->vfs_inode
;
881 * We want fsync() to work on POHMELFS.
883 static int pohmelfs_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
885 struct inode
*inode
= file
->f_mapping
->host
;
886 struct writeback_control wbc
= {
887 .sync_mode
= WB_SYNC_ALL
,
888 .nr_to_write
= 0, /* sys_fsync did this */
891 return sync_inode(inode
, &wbc
);
894 ssize_t
pohmelfs_write(struct file
*file
, const char __user
*buf
,
895 size_t len
, loff_t
*ppos
)
897 struct address_space
*mapping
= file
->f_mapping
;
898 struct inode
*inode
= mapping
->host
;
899 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
900 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
905 init_sync_kiocb(&kiocb
, file
);
909 dprintk("%s: len: %zu, pos: %llu.\n", __func__
, len
, pos
);
911 mutex_lock(&inode
->i_mutex
);
912 ret
= pohmelfs_data_lock(pi
, pos
, len
, POHMELFS_WRITE_LOCK
);
916 ret
= __generic_file_aio_write(&kiocb
, &iov
, 1, &kiocb
.ki_pos
);
917 *ppos
= kiocb
.ki_pos
;
919 mutex_unlock(&inode
->i_mutex
);
925 err
= generic_write_sync(file
, pos
, ret
);
934 mutex_unlock(&inode
->i_mutex
);
938 static const struct file_operations pohmelfs_file_ops
= {
939 .open
= generic_file_open
,
940 .fsync
= pohmelfs_fsync
,
942 .llseek
= generic_file_llseek
,
944 .read
= do_sync_read
,
945 .aio_read
= generic_file_aio_read
,
947 .mmap
= generic_file_mmap
,
949 .splice_read
= generic_file_splice_read
,
950 .splice_write
= generic_file_splice_write
,
952 .write
= pohmelfs_write
,
953 .aio_write
= generic_file_aio_write
,
956 const struct inode_operations pohmelfs_symlink_inode_operations
= {
957 .readlink
= generic_readlink
,
958 .follow_link
= page_follow_link_light
,
959 .put_link
= page_put_link
,
962 int pohmelfs_setattr_raw(struct inode
*inode
, struct iattr
*attr
)
966 err
= inode_change_ok(inode
, attr
);
968 dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__
, POHMELFS_I(inode
)->ino
);
972 if ((attr
->ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
973 (attr
->ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
974 err
= dquot_transfer(inode
, attr
);
979 err
= inode_setattr(inode
, attr
);
981 dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__
, POHMELFS_I(inode
)->ino
);
985 dprintk("%s: ino: %llu, mode: %o -> %o, uid: %u -> %u, gid: %u -> %u, size: %llu -> %llu.\n",
986 __func__
, POHMELFS_I(inode
)->ino
, inode
->i_mode
, attr
->ia_mode
,
987 inode
->i_uid
, attr
->ia_uid
, inode
->i_gid
, attr
->ia_gid
, inode
->i_size
, attr
->ia_size
);
995 int pohmelfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
997 struct inode
*inode
= dentry
->d_inode
;
998 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1001 err
= pohmelfs_data_lock(pi
, 0, ~0, POHMELFS_WRITE_LOCK
);
1005 err
= security_inode_setattr(dentry
, attr
);
1009 err
= pohmelfs_setattr_raw(inode
, attr
);
1019 static int pohmelfs_send_xattr_req(struct pohmelfs_inode
*pi
, u64 id
, u64 start
,
1020 const char *name
, const void *value
, size_t attrsize
, int command
)
1022 struct pohmelfs_sb
*psb
= POHMELFS_SB(pi
->vfs_inode
.i_sb
);
1023 int err
, path_len
, namelen
= strlen(name
) + 1; /* 0-byte */
1024 struct netfs_trans
*t
;
1025 struct netfs_cmd
*cmd
;
1028 dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n",
1029 __func__
, id
, start
, name
, attrsize
, command
);
1031 path_len
= pohmelfs_path_length(pi
);
1037 t
= netfs_trans_alloc(psb
, namelen
+ path_len
+ attrsize
, 0, 0);
1043 cmd
= netfs_trans_current(t
);
1046 path_len
= pohmelfs_construct_path_string(pi
, data
, path_len
);
1054 * 'name' is a NUL-terminated string already and
1055 * 'namelen' includes 0-byte.
1057 memcpy(data
, name
, namelen
);
1060 memcpy(data
, value
, attrsize
);
1065 cmd
->size
= attrsize
+ namelen
+ path_len
;
1066 cmd
->ext
= path_len
;
1070 netfs_convert_cmd(cmd
);
1071 netfs_trans_update(cmd
, t
, namelen
+ path_len
+ attrsize
);
1073 return netfs_trans_finish(t
, psb
);
1082 static int pohmelfs_setxattr(struct dentry
*dentry
, const char *name
,
1083 const void *value
, size_t attrsize
, int flags
)
1085 struct inode
*inode
= dentry
->d_inode
;
1086 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1087 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1089 if (!(psb
->state_flags
& POHMELFS_FLAGS_XATTR
))
1092 return pohmelfs_send_xattr_req(pi
, flags
, attrsize
, name
,
1093 value
, attrsize
, NETFS_XATTR_SET
);
1096 static ssize_t
pohmelfs_getxattr(struct dentry
*dentry
, const char *name
,
1097 void *value
, size_t attrsize
)
1099 struct inode
*inode
= dentry
->d_inode
;
1100 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1101 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1102 struct pohmelfs_mcache
*m
;
1104 long timeout
= psb
->mcache_timeout
;
1106 if (!(psb
->state_flags
& POHMELFS_FLAGS_XATTR
))
1109 m
= pohmelfs_mcache_alloc(psb
, 0, attrsize
, value
);
1113 dprintk("%s: ino: %llu, name: '%s', size: %zu.\n",
1114 __func__
, pi
->ino
, name
, attrsize
);
1116 err
= pohmelfs_send_xattr_req(pi
, m
->gen
, attrsize
, name
, value
, 0, NETFS_XATTR_GET
);
1121 err
= wait_for_completion_timeout(&m
->complete
, timeout
);
1128 * This loop is a bit ugly, since it waits until reference counter
1129 * hits 1 and then put object here. Main goal is to prevent race with
1130 * network thread, when it can start processing given request, i.e.
1131 * increase its reference counter but yet not complete it, while
1132 * we will exit from ->getxattr() with timeout, and although request
1133 * will not be freed (its reference counter was increased by network
1134 * thread), data pointer provided by user may be released, so we will
1135 * overwrite already freed area in network thread.
1137 * Now after timeout we remove request from the cache, so it can not be
1138 * found by network thread, and wait for its reference counter to hit 1,
1139 * i.e. if network thread already started to process this request, we wait
1140 * it to finish, and then free object locally. If reference counter is
1141 * already 1, i.e. request is not used by anyone else, we can free it without
1147 pohmelfs_mcache_remove_locked(psb
, m
);
1148 } while (atomic_read(&m
->refcnt
) != 1);
1150 pohmelfs_mcache_put(psb
, m
);
1152 dprintk("%s: ino: %llu, err: %d.\n", __func__
, pi
->ino
, err
);
1157 pohmelfs_mcache_put(psb
, m
);
1161 static int pohmelfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
1163 struct inode
*inode
= dentry
->d_inode
;
1165 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1168 err
= pohmelfs_data_lock(pi
, 0, ~0, POHMELFS_READ_LOCK
);
1171 dprintk("%s: ino: %llu, mode: %o, uid: %u, gid: %u, size: %llu.\n",
1172 __func__
, pi
->ino
, inode
->i_mode
, inode
->i_uid
,
1173 inode
->i_gid
, inode
->i_size
);
1176 generic_fillattr(inode
, stat
);
1180 const struct inode_operations pohmelfs_file_inode_operations
= {
1181 .setattr
= pohmelfs_setattr
,
1182 .getattr
= pohmelfs_getattr
,
1183 .setxattr
= pohmelfs_setxattr
,
1184 .getxattr
= pohmelfs_getxattr
,
1188 * Fill inode data: mode, size, operation callbacks and so on...
1190 void pohmelfs_fill_inode(struct inode
*inode
, struct netfs_inode_info
*info
)
1192 inode
->i_mode
= info
->mode
;
1193 inode
->i_nlink
= info
->nlink
;
1194 inode
->i_uid
= info
->uid
;
1195 inode
->i_gid
= info
->gid
;
1196 inode
->i_blocks
= info
->blocks
;
1197 inode
->i_rdev
= info
->rdev
;
1198 inode
->i_size
= info
->size
;
1199 inode
->i_version
= info
->version
;
1200 inode
->i_blkbits
= ffs(info
->blocksize
);
1202 dprintk("%s: inode: %p, num: %lu/%llu inode is regular: %d, dir: %d, link: %d, mode: %o, size: %llu.\n",
1203 __func__
, inode
, inode
->i_ino
, info
->ino
,
1204 S_ISREG(inode
->i_mode
), S_ISDIR(inode
->i_mode
),
1205 S_ISLNK(inode
->i_mode
), inode
->i_mode
, inode
->i_size
);
1207 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
1210 * i_mapping is a pointer to i_data during inode initialization.
1212 inode
->i_data
.a_ops
= &pohmelfs_aops
;
1214 if (S_ISREG(inode
->i_mode
)) {
1215 inode
->i_fop
= &pohmelfs_file_ops
;
1216 inode
->i_op
= &pohmelfs_file_inode_operations
;
1217 } else if (S_ISDIR(inode
->i_mode
)) {
1218 inode
->i_fop
= &pohmelfs_dir_fops
;
1219 inode
->i_op
= &pohmelfs_dir_inode_ops
;
1220 } else if (S_ISLNK(inode
->i_mode
)) {
1221 inode
->i_op
= &pohmelfs_symlink_inode_operations
;
1222 inode
->i_fop
= &pohmelfs_file_ops
;
1224 inode
->i_fop
= &generic_ro_fops
;
1228 static void pohmelfs_drop_inode(struct inode
*inode
)
1230 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1231 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1233 spin_lock(&psb
->ino_lock
);
1234 list_del_init(&pi
->inode_entry
);
1235 spin_unlock(&psb
->ino_lock
);
1237 generic_drop_inode(inode
);
1240 static struct pohmelfs_inode
*pohmelfs_get_inode_from_list(struct pohmelfs_sb
*psb
,
1241 struct list_head
*head
, unsigned int *count
)
1243 struct pohmelfs_inode
*pi
= NULL
;
1245 spin_lock(&psb
->ino_lock
);
1246 if (!list_empty(head
)) {
1247 pi
= list_entry(head
->next
, struct pohmelfs_inode
,
1249 list_del_init(&pi
->inode_entry
);
1250 *count
= pi
->drop_count
;
1253 spin_unlock(&psb
->ino_lock
);
1258 static void pohmelfs_flush_transactions(struct pohmelfs_sb
*psb
)
1260 struct pohmelfs_config
*c
;
1262 mutex_lock(&psb
->state_lock
);
1263 list_for_each_entry(c
, &psb
->state_list
, config_entry
) {
1264 pohmelfs_state_flush_transactions(&c
->state
);
1266 mutex_unlock(&psb
->state_lock
);
1270 * ->put_super() callback. Invoked before superblock is destroyed,
1271 * so it has to clean all private data.
1273 static void pohmelfs_put_super(struct super_block
*sb
)
1275 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
1276 struct pohmelfs_inode
*pi
;
1278 unsigned int in_drop_list
= 0;
1279 struct inode
*inode
, *tmp
;
1281 dprintk("%s.\n", __func__
);
1284 * Kill pending transactions, which could affect inodes in-flight.
1286 pohmelfs_flush_transactions(psb
);
1288 while ((pi
= pohmelfs_get_inode_from_list(psb
, &psb
->drop_list
, &count
))) {
1289 inode
= &pi
->vfs_inode
;
1291 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1292 __func__
, pi
->ino
, pi
, inode
, count
);
1294 if (atomic_read(&inode
->i_count
) != count
) {
1295 printk("%s: ino: %llu, pi: %p, inode: %p, count: %u, i_count: %d.\n",
1296 __func__
, pi
->ino
, pi
, inode
, count
,
1297 atomic_read(&inode
->i_count
));
1298 count
= atomic_read(&inode
->i_count
);
1303 iput(&pi
->vfs_inode
);
1306 list_for_each_entry_safe(inode
, tmp
, &sb
->s_inodes
, i_sb_list
) {
1307 pi
= POHMELFS_I(inode
);
1309 dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n",
1310 __func__
, pi
->ino
, pi
, inode
, atomic_read(&inode
->i_count
));
1313 * These are special inodes, they were created during
1314 * directory reading or lookup, and were not bound to dentry,
1315 * so they live here with reference counter being 1 and prevent
1316 * umount from succeed since it believes that they are busy.
1318 count
= atomic_read(&inode
->i_count
);
1320 list_del_init(&inode
->i_sb_list
);
1322 iput(&pi
->vfs_inode
);
1326 psb
->trans_scan_timeout
= psb
->drop_scan_timeout
= 0;
1327 cancel_rearming_delayed_work(&psb
->dwork
);
1328 cancel_rearming_delayed_work(&psb
->drop_dwork
);
1329 flush_scheduled_work();
1331 dprintk("%s: stopped workqueues.\n", __func__
);
1333 pohmelfs_crypto_exit(psb
);
1334 pohmelfs_state_exit(psb
);
1336 bdi_destroy(&psb
->bdi
);
1339 sb
->s_fs_info
= NULL
;
1342 static int pohmelfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1344 struct super_block
*sb
= dentry
->d_sb
;
1345 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
1348 * There are no filesystem size limits yet.
1350 memset(buf
, 0, sizeof(struct kstatfs
));
1352 buf
->f_type
= POHMELFS_MAGIC_NUM
; /* 'POH.' */
1353 buf
->f_bsize
= sb
->s_blocksize
;
1354 buf
->f_files
= psb
->ino
;
1355 buf
->f_namelen
= 255;
1356 buf
->f_files
= atomic_long_read(&psb
->total_inodes
);
1357 buf
->f_bfree
= buf
->f_bavail
= psb
->avail_size
>> PAGE_SHIFT
;
1358 buf
->f_blocks
= psb
->total_size
>> PAGE_SHIFT
;
1360 dprintk("%s: total: %llu, avail: %llu, inodes: %llu, bsize: %lu.\n",
1361 __func__
, psb
->total_size
, psb
->avail_size
, buf
->f_files
, sb
->s_blocksize
);
1366 static int pohmelfs_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1368 struct pohmelfs_sb
*psb
= POHMELFS_SB(vfs
->mnt_sb
);
1370 seq_printf(seq
, ",idx=%u", psb
->idx
);
1371 seq_printf(seq
, ",trans_scan_timeout=%u", jiffies_to_msecs(psb
->trans_scan_timeout
));
1372 seq_printf(seq
, ",drop_scan_timeout=%u", jiffies_to_msecs(psb
->drop_scan_timeout
));
1373 seq_printf(seq
, ",wait_on_page_timeout=%u", jiffies_to_msecs(psb
->wait_on_page_timeout
));
1374 seq_printf(seq
, ",trans_retries=%u", psb
->trans_retries
);
1375 seq_printf(seq
, ",crypto_thread_num=%u", psb
->crypto_thread_num
);
1376 seq_printf(seq
, ",trans_max_pages=%u", psb
->trans_max_pages
);
1377 seq_printf(seq
, ",mcache_timeout=%u", jiffies_to_msecs(psb
->mcache_timeout
));
1378 if (psb
->crypto_fail_unsupported
)
1379 seq_printf(seq
, ",crypto_fail_unsupported");
1386 pohmelfs_opt_crypto_thread_num
,
1387 pohmelfs_opt_trans_max_pages
,
1388 pohmelfs_opt_crypto_fail_unsupported
,
1390 /* Remountable options */
1391 pohmelfs_opt_trans_scan_timeout
,
1392 pohmelfs_opt_drop_scan_timeout
,
1393 pohmelfs_opt_wait_on_page_timeout
,
1394 pohmelfs_opt_trans_retries
,
1395 pohmelfs_opt_mcache_timeout
,
1398 static struct match_token pohmelfs_tokens
[] = {
1399 {pohmelfs_opt_idx
, "idx=%u"},
1400 {pohmelfs_opt_crypto_thread_num
, "crypto_thread_num=%u"},
1401 {pohmelfs_opt_trans_max_pages
, "trans_max_pages=%u"},
1402 {pohmelfs_opt_crypto_fail_unsupported
, "crypto_fail_unsupported"},
1403 {pohmelfs_opt_trans_scan_timeout
, "trans_scan_timeout=%u"},
1404 {pohmelfs_opt_drop_scan_timeout
, "drop_scan_timeout=%u"},
1405 {pohmelfs_opt_wait_on_page_timeout
, "wait_on_page_timeout=%u"},
1406 {pohmelfs_opt_trans_retries
, "trans_retries=%u"},
1407 {pohmelfs_opt_mcache_timeout
, "mcache_timeout=%u"},
1410 static int pohmelfs_parse_options(char *options
, struct pohmelfs_sb
*psb
, int remount
)
1413 substring_t args
[MAX_OPT_ARGS
];
1419 while ((p
= strsep(&options
, ",")) != NULL
) {
1424 token
= match_token(p
, pohmelfs_tokens
, args
);
1426 err
= match_int(&args
[0], &option
);
1430 if (remount
&& token
<= pohmelfs_opt_crypto_fail_unsupported
)
1434 case pohmelfs_opt_idx
:
1437 case pohmelfs_opt_trans_scan_timeout
:
1438 psb
->trans_scan_timeout
= msecs_to_jiffies(option
);
1440 case pohmelfs_opt_drop_scan_timeout
:
1441 psb
->drop_scan_timeout
= msecs_to_jiffies(option
);
1443 case pohmelfs_opt_wait_on_page_timeout
:
1444 psb
->wait_on_page_timeout
= msecs_to_jiffies(option
);
1446 case pohmelfs_opt_mcache_timeout
:
1447 psb
->mcache_timeout
= msecs_to_jiffies(option
);
1449 case pohmelfs_opt_trans_retries
:
1450 psb
->trans_retries
= option
;
1452 case pohmelfs_opt_crypto_thread_num
:
1453 psb
->crypto_thread_num
= option
;
1455 case pohmelfs_opt_trans_max_pages
:
1456 psb
->trans_max_pages
= option
;
1458 case pohmelfs_opt_crypto_fail_unsupported
:
1459 psb
->crypto_fail_unsupported
= 1;
1469 static int pohmelfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1472 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
1473 unsigned long old_sb_flags
= sb
->s_flags
;
1475 err
= pohmelfs_parse_options(data
, psb
, 1);
1477 goto err_out_restore
;
1479 if (!(*flags
& MS_RDONLY
))
1480 sb
->s_flags
&= ~MS_RDONLY
;
1484 sb
->s_flags
= old_sb_flags
;
1488 static void pohmelfs_flush_inode(struct pohmelfs_inode
*pi
, unsigned int count
)
1490 struct inode
*inode
= &pi
->vfs_inode
;
1492 dprintk("%s: %p: ino: %llu, owned: %d.\n",
1493 __func__
, inode
, pi
->ino
, test_bit(NETFS_INODE_OWNED
, &pi
->state
));
1495 mutex_lock(&inode
->i_mutex
);
1496 if (test_and_clear_bit(NETFS_INODE_OWNED
, &pi
->state
)) {
1497 filemap_fdatawrite(inode
->i_mapping
);
1498 inode
->i_sb
->s_op
->write_inode(inode
, 0);
1501 #ifdef POHMELFS_TRUNCATE_ON_INODE_FLUSH
1502 truncate_inode_pages(inode
->i_mapping
, 0);
1505 pohmelfs_data_unlock(pi
, 0, ~0, POHMELFS_WRITE_LOCK
);
1506 mutex_unlock(&inode
->i_mutex
);
1509 static void pohmelfs_put_inode_count(struct pohmelfs_inode
*pi
, unsigned int count
)
1511 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1512 __func__
, pi
->ino
, pi
, &pi
->vfs_inode
, count
);
1514 if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH
, &pi
->state
))
1515 pohmelfs_flush_inode(pi
, count
);
1518 iput(&pi
->vfs_inode
);
1521 static void pohmelfs_drop_scan(struct work_struct
*work
)
1523 struct pohmelfs_sb
*psb
=
1524 container_of(work
, struct pohmelfs_sb
, drop_dwork
.work
);
1525 struct pohmelfs_inode
*pi
;
1526 unsigned int count
= 0;
1528 while ((pi
= pohmelfs_get_inode_from_list(psb
, &psb
->drop_list
, &count
)))
1529 pohmelfs_put_inode_count(pi
, count
);
1531 pohmelfs_check_states(psb
);
1533 if (psb
->drop_scan_timeout
)
1534 schedule_delayed_work(&psb
->drop_dwork
, psb
->drop_scan_timeout
);
1538 * Run through all transactions starting from the oldest,
1539 * drop transaction from current state and try to send it
1540 * to all remote nodes, which are currently installed.
1542 static void pohmelfs_trans_scan_state(struct netfs_state
*st
)
1544 struct rb_node
*rb_node
;
1545 struct netfs_trans_dst
*dst
;
1546 struct pohmelfs_sb
*psb
= st
->psb
;
1547 unsigned int timeout
= psb
->trans_scan_timeout
;
1548 struct netfs_trans
*t
;
1551 mutex_lock(&st
->trans_lock
);
1552 for (rb_node
= rb_first(&st
->trans_root
); rb_node
; ) {
1553 dst
= rb_entry(rb_node
, struct netfs_trans_dst
, state_entry
);
1556 if (timeout
&& time_after(dst
->send_time
+ timeout
, jiffies
)
1557 && dst
->retries
== 0)
1560 dprintk("%s: t: %p, gen: %u, st: %p, retries: %u, max: %u.\n",
1561 __func__
, t
, t
->gen
, st
, dst
->retries
, psb
->trans_retries
);
1564 rb_node
= rb_next(rb_node
);
1567 if (timeout
&& (++dst
->retries
< psb
->trans_retries
))
1568 err
= netfs_trans_resend(t
, psb
);
1570 if (err
|| (t
->flags
& NETFS_TRANS_SINGLE_DST
)) {
1571 if (netfs_trans_remove_nolock(dst
, st
))
1572 netfs_trans_drop_dst_nostate(dst
);
1578 mutex_unlock(&st
->trans_lock
);
1582 * Walk through all installed network states and resend all
1583 * transactions, which are old enough.
1585 static void pohmelfs_trans_scan(struct work_struct
*work
)
1587 struct pohmelfs_sb
*psb
=
1588 container_of(work
, struct pohmelfs_sb
, dwork
.work
);
1589 struct netfs_state
*st
;
1590 struct pohmelfs_config
*c
;
1592 mutex_lock(&psb
->state_lock
);
1593 list_for_each_entry(c
, &psb
->state_list
, config_entry
) {
1596 pohmelfs_trans_scan_state(st
);
1598 mutex_unlock(&psb
->state_lock
);
1601 * If no timeout specified then system is in the middle of umount process,
1602 * so no need to reschedule scanning process again.
1604 if (psb
->trans_scan_timeout
)
1605 schedule_delayed_work(&psb
->dwork
, psb
->trans_scan_timeout
);
1608 int pohmelfs_meta_command_data(struct pohmelfs_inode
*pi
, u64 id
, unsigned int cmd_op
, char *addon
,
1609 unsigned int flags
, netfs_trans_complete_t complete
, void *priv
, u64 start
)
1611 struct inode
*inode
= &pi
->vfs_inode
;
1612 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1614 struct netfs_trans
*t
;
1615 int path_len
, addon_len
= 0;
1617 struct netfs_inode_info
*info
;
1618 struct netfs_cmd
*cmd
;
1620 dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__
, pi
->ino
, cmd_op
, addon
);
1622 path_len
= pohmelfs_path_length(pi
);
1629 addon_len
= strlen(addon
) + 1; /* 0-byte */
1632 if (cmd_op
== NETFS_INODE_INFO
)
1633 sz
+= sizeof(struct netfs_inode_info
);
1635 t
= netfs_trans_alloc(psb
, sz
+ path_len
, flags
, 0);
1640 t
->complete
= complete
;
1643 cmd
= netfs_trans_current(t
);
1644 data
= (void *)(cmd
+ 1);
1646 if (cmd_op
== NETFS_INODE_INFO
) {
1647 info
= (struct netfs_inode_info
*)(cmd
+ 1);
1648 data
= (void *)(info
+ 1);
1651 * We are under i_mutex, can read and change whatever we want...
1653 info
->mode
= inode
->i_mode
;
1654 info
->nlink
= inode
->i_nlink
;
1655 info
->uid
= inode
->i_uid
;
1656 info
->gid
= inode
->i_gid
;
1657 info
->blocks
= inode
->i_blocks
;
1658 info
->rdev
= inode
->i_rdev
;
1659 info
->size
= inode
->i_size
;
1660 info
->version
= inode
->i_version
;
1662 netfs_convert_inode_info(info
);
1665 path_len
= pohmelfs_construct_path_string(pi
, data
, path_len
);
1669 dprintk("%s: path_len: %d.\n", __func__
, path_len
);
1672 path_len
--; /* Do not place null-byte before the addon */
1673 path_len
+= sprintf(data
+ path_len
, "/%s", addon
) + 1; /* 0 - byte */
1679 cmd
->ext
= path_len
;
1684 netfs_convert_cmd(cmd
);
1685 netfs_trans_update(cmd
, t
, sz
);
1688 * Note, that it is possible to leak error here: transaction callback will not
1689 * be invoked for allocation path failure.
1691 return netfs_trans_finish(t
, psb
);
1694 netfs_trans_free(t
);
1697 complete(NULL
, 0, priv
, err
);
1701 int pohmelfs_meta_command(struct pohmelfs_inode
*pi
, unsigned int cmd_op
, unsigned int flags
,
1702 netfs_trans_complete_t complete
, void *priv
, u64 start
)
1704 return pohmelfs_meta_command_data(pi
, pi
->ino
, cmd_op
, NULL
, flags
, complete
, priv
, start
);
1708 * Send request and wait for POHMELFS root capabilities response,
1709 * which will update server's informaion about size of the export,
1710 * permissions, number of objects, available size and so on.
1712 static int pohmelfs_root_handshake(struct pohmelfs_sb
*psb
)
1714 struct netfs_trans
*t
;
1715 struct netfs_cmd
*cmd
;
1718 t
= netfs_trans_alloc(psb
, 0, 0, 0);
1722 cmd
= netfs_trans_current(t
);
1724 cmd
->cmd
= NETFS_CAPABILITIES
;
1725 cmd
->id
= POHMELFS_ROOT_CAPABILITIES
;
1731 netfs_convert_cmd(cmd
);
1732 netfs_trans_update(cmd
, t
, 0);
1734 err
= netfs_trans_finish(t
, psb
);
1739 err
= wait_event_interruptible_timeout(psb
->wait
,
1741 psb
->wait_on_page_timeout
);
1756 static int pohmelfs_show_stats(struct seq_file
*m
, struct vfsmount
*mnt
)
1758 struct netfs_state
*st
;
1759 struct pohmelfs_ctl
*ctl
;
1760 struct pohmelfs_sb
*psb
= POHMELFS_SB(mnt
->mnt_sb
);
1761 struct pohmelfs_config
*c
;
1763 mutex_lock(&psb
->state_lock
);
1765 seq_printf(m
, "\nidx addr(:port) socket_type protocol active priority permissions\n");
1767 list_for_each_entry(c
, &psb
->state_list
, config_entry
) {
1771 seq_printf(m
, "%u ", ctl
->idx
);
1772 if (ctl
->addr
.sa_family
== AF_INET
) {
1773 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&st
->ctl
.addr
;
1774 seq_printf(m
, "%pI4:%u", &sin
->sin_addr
.s_addr
, ntohs(sin
->sin_port
));
1775 } else if (ctl
->addr
.sa_family
== AF_INET6
) {
1776 struct sockaddr_in6
*sin
= (struct sockaddr_in6
*)&st
->ctl
.addr
;
1777 seq_printf(m
, "%pi6:%u", &sin
->sin6_addr
, ntohs(sin
->sin6_port
));
1780 for (i
=0; i
<ctl
->addrlen
; ++i
)
1781 seq_printf(m
, "%02x.", ctl
->addr
.addr
[i
]);
1784 seq_printf(m
, " %u %u %d %u %x\n",
1785 ctl
->type
, ctl
->proto
,
1787 ctl
->prio
, ctl
->perm
);
1789 mutex_unlock(&psb
->state_lock
);
1794 static const struct super_operations pohmelfs_sb_ops
= {
1795 .alloc_inode
= pohmelfs_alloc_inode
,
1796 .destroy_inode
= pohmelfs_destroy_inode
,
1797 .drop_inode
= pohmelfs_drop_inode
,
1798 .write_inode
= pohmelfs_write_inode
,
1799 .put_super
= pohmelfs_put_super
,
1800 .remount_fs
= pohmelfs_remount
,
1801 .statfs
= pohmelfs_statfs
,
1802 .show_options
= pohmelfs_show_options
,
1803 .show_stats
= pohmelfs_show_stats
,
1807 * Allocate private superblock and create root dir.
1809 static int pohmelfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1811 struct pohmelfs_sb
*psb
;
1814 struct pohmelfs_inode
*npi
;
1817 psb
= kzalloc(sizeof(struct pohmelfs_sb
), GFP_KERNEL
);
1821 err
= bdi_init(&psb
->bdi
);
1823 goto err_out_free_sb
;
1825 err
= bdi_register(&psb
->bdi
, NULL
, "pfs-%d", atomic_inc_return(&psb_bdi_num
));
1827 bdi_destroy(&psb
->bdi
);
1828 goto err_out_free_sb
;
1831 sb
->s_fs_info
= psb
;
1832 sb
->s_op
= &pohmelfs_sb_ops
;
1833 sb
->s_magic
= POHMELFS_MAGIC_NUM
;
1834 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1835 sb
->s_blocksize
= PAGE_SIZE
;
1836 sb
->s_bdi
= &psb
->bdi
;
1842 psb
->active_state
= NULL
;
1843 psb
->trans_retries
= 5;
1844 psb
->trans_data_size
= PAGE_SIZE
;
1845 psb
->drop_scan_timeout
= msecs_to_jiffies(1000);
1846 psb
->trans_scan_timeout
= msecs_to_jiffies(5000);
1847 psb
->wait_on_page_timeout
= msecs_to_jiffies(5000);
1848 init_waitqueue_head(&psb
->wait
);
1850 spin_lock_init(&psb
->ino_lock
);
1852 INIT_LIST_HEAD(&psb
->drop_list
);
1854 mutex_init(&psb
->mcache_lock
);
1855 psb
->mcache_root
= RB_ROOT
;
1856 psb
->mcache_timeout
= msecs_to_jiffies(5000);
1857 atomic_long_set(&psb
->mcache_gen
, 0);
1859 psb
->trans_max_pages
= 100;
1861 psb
->crypto_align_size
= 16;
1862 psb
->crypto_attached_size
= 0;
1863 psb
->hash_strlen
= 0;
1864 psb
->cipher_strlen
= 0;
1865 psb
->perform_crypto
= 0;
1866 psb
->crypto_thread_num
= 2;
1867 psb
->crypto_fail_unsupported
= 0;
1868 mutex_init(&psb
->crypto_thread_lock
);
1869 INIT_LIST_HEAD(&psb
->crypto_ready_list
);
1870 INIT_LIST_HEAD(&psb
->crypto_active_list
);
1872 atomic_set(&psb
->trans_gen
, 1);
1873 atomic_long_set(&psb
->total_inodes
, 0);
1875 mutex_init(&psb
->state_lock
);
1876 INIT_LIST_HEAD(&psb
->state_list
);
1878 err
= pohmelfs_parse_options((char *) data
, psb
, 0);
1880 goto err_out_free_bdi
;
1882 err
= pohmelfs_copy_crypto(psb
);
1884 goto err_out_free_bdi
;
1886 err
= pohmelfs_state_init(psb
);
1888 goto err_out_free_strings
;
1890 err
= pohmelfs_crypto_init(psb
);
1892 goto err_out_state_exit
;
1894 err
= pohmelfs_root_handshake(psb
);
1896 goto err_out_crypto_exit
;
1899 str
.hash
= jhash("/", 1, 0);
1902 npi
= pohmelfs_create_entry_local(psb
, NULL
, &str
, 0, 0755|S_IFDIR
);
1905 goto err_out_crypto_exit
;
1907 set_bit(NETFS_INODE_REMOTE_SYNCED
, &npi
->state
);
1908 clear_bit(NETFS_INODE_OWNED
, &npi
->state
);
1910 root
= &npi
->vfs_inode
;
1912 sb
->s_root
= d_alloc_root(root
);
1914 goto err_out_put_root
;
1916 INIT_DELAYED_WORK(&psb
->drop_dwork
, pohmelfs_drop_scan
);
1917 schedule_delayed_work(&psb
->drop_dwork
, psb
->drop_scan_timeout
);
1919 INIT_DELAYED_WORK(&psb
->dwork
, pohmelfs_trans_scan
);
1920 schedule_delayed_work(&psb
->dwork
, psb
->trans_scan_timeout
);
1926 err_out_crypto_exit
:
1927 pohmelfs_crypto_exit(psb
);
1929 pohmelfs_state_exit(psb
);
1930 err_out_free_strings
:
1931 kfree(psb
->cipher_string
);
1932 kfree(psb
->hash_string
);
1934 bdi_destroy(&psb
->bdi
);
1939 dprintk("%s: err: %d.\n", __func__
, err
);
1944 * Some VFS magic here...
1946 static int pohmelfs_get_sb(struct file_system_type
*fs_type
,
1947 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
1949 return get_sb_nodev(fs_type
, flags
, data
, pohmelfs_fill_super
,
1954 * We need this to sync all inodes earlier, since when writeback
1955 * is invoked from the umount/mntput path dcache is already shrunk,
1956 * see generic_shutdown_super(), and no inodes can access the path.
1958 static void pohmelfs_kill_super(struct super_block
*sb
)
1961 kill_anon_super(sb
);
1964 static struct file_system_type pohmel_fs_type
= {
1965 .owner
= THIS_MODULE
,
1967 .get_sb
= pohmelfs_get_sb
,
1968 .kill_sb
= pohmelfs_kill_super
,
1972 * Cache and module initializations and freeing routings.
1974 static void pohmelfs_init_once(void *data
)
1976 struct pohmelfs_inode
*pi
= data
;
1978 inode_init_once(&pi
->vfs_inode
);
1981 static int __init
pohmelfs_init_inodecache(void)
1983 pohmelfs_inode_cache
= kmem_cache_create("pohmelfs_inode_cache",
1984 sizeof(struct pohmelfs_inode
),
1985 0, (SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
),
1986 pohmelfs_init_once
);
1987 if (!pohmelfs_inode_cache
)
1993 static void pohmelfs_destroy_inodecache(void)
1995 kmem_cache_destroy(pohmelfs_inode_cache
);
1998 static int __init
init_pohmel_fs(void)
2002 err
= pohmelfs_config_init();
2006 err
= pohmelfs_init_inodecache();
2008 goto err_out_config_exit
;
2010 err
= pohmelfs_mcache_init();
2012 goto err_out_destroy
;
2014 err
= netfs_trans_init();
2016 goto err_out_mcache_exit
;
2018 err
= register_filesystem(&pohmel_fs_type
);
2026 err_out_mcache_exit
:
2027 pohmelfs_mcache_exit();
2029 pohmelfs_destroy_inodecache();
2030 err_out_config_exit
:
2031 pohmelfs_config_exit();
2036 static void __exit
exit_pohmel_fs(void)
2038 unregister_filesystem(&pohmel_fs_type
);
2039 pohmelfs_destroy_inodecache();
2040 pohmelfs_mcache_exit();
2041 pohmelfs_config_exit();
2045 module_init(init_pohmel_fs
);
2046 module_exit(exit_pohmel_fs
);
2048 MODULE_LICENSE("GPL");
2049 MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
2050 MODULE_DESCRIPTION("Pohmel filesystem");