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>
35 #define POHMELFS_MAGIC_NUM 0x504f482e
37 static struct kmem_cache
*pohmelfs_inode_cache
;
38 static atomic_t psb_bdi_num
= ATOMIC_INIT(0);
41 * Removes inode from all trees, drops local name cache and removes all queued
42 * requests for object removal.
44 void pohmelfs_inode_del_inode(struct pohmelfs_sb
*psb
, struct pohmelfs_inode
*pi
)
46 mutex_lock(&pi
->offset_lock
);
47 pohmelfs_free_names(pi
);
48 mutex_unlock(&pi
->offset_lock
);
50 dprintk("%s: deleted stuff in ino: %llu.\n", __func__
, pi
->ino
);
54 * Sync inode to server.
55 * Returns zero in success and negative error value otherwise.
56 * It will gather path to root directory into structures containing
57 * creation mode, permissions and names, so that the whole path
58 * to given inode could be created using only single network command.
60 int pohmelfs_write_inode_create(struct inode
*inode
, struct netfs_trans
*trans
)
62 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
63 int err
= -ENOMEM
, size
;
64 struct netfs_cmd
*cmd
;
66 int cur_len
= netfs_trans_cur_len(trans
);
68 if (unlikely(cur_len
< 0))
71 cmd
= netfs_trans_current(trans
);
72 cur_len
-= sizeof(struct netfs_cmd
);
74 data
= (void *)(cmd
+ 1);
76 err
= pohmelfs_construct_path_string(pi
, data
, cur_len
);
82 cmd
->start
= i_size_read(inode
);
83 cmd
->cmd
= NETFS_CREATE
;
86 cmd
->ext
= inode
->i_mode
;
88 netfs_convert_cmd(cmd
);
90 netfs_trans_update(cmd
, trans
, size
);
95 printk("%s: completed ino: %llu, err: %d.\n", __func__
, pi
->ino
, err
);
99 static int pohmelfs_write_trans_complete(struct page
**pages
, unsigned int page_num
,
100 void *private, int err
)
104 dprintk("%s: pages: %lu-%lu, page_num: %u, err: %d.\n",
105 __func__
, pages
[0]->index
, pages
[page_num
-1]->index
,
108 for (i
= 0; i
< page_num
; i
++) {
109 struct page
*page
= pages
[i
];
114 end_page_writeback(page
);
118 set_page_dirty(page
);
122 page_cache_release(page
);
124 /* dprintk("%s: %3u/%u: page: %p.\n", __func__, i, page_num, page); */
129 static int pohmelfs_inode_has_dirty_pages(struct address_space
*mapping
, pgoff_t index
)
135 ret
= radix_tree_gang_lookup_tag(&mapping
->page_tree
,
136 (void **)&page
, index
, 1, PAGECACHE_TAG_DIRTY
);
141 static int pohmelfs_writepages(struct address_space
*mapping
, struct writeback_control
*wbc
)
143 struct inode
*inode
= mapping
->host
;
144 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
145 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
150 pgoff_t end
; /* Inclusive */
154 if (wbc
->range_cyclic
) {
155 index
= mapping
->writeback_index
; /* Start from prev offset */
158 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
159 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
160 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
165 while (!done
&& (index
<= end
)) {
166 unsigned int i
= min(end
- index
, (pgoff_t
)psb
->trans_max_pages
);
168 struct netfs_trans
*trans
;
170 err
= pohmelfs_inode_has_dirty_pages(mapping
, index
);
174 err
= pohmelfs_path_length(pi
);
185 trans
= netfs_trans_alloc(psb
, path_len
, 0, i
);
190 trans
->complete
= &pohmelfs_write_trans_complete
;
192 trans
->page_num
= nr_pages
= find_get_pages_tag(mapping
, &index
,
193 PAGECACHE_TAG_DIRTY
, trans
->page_num
,
196 dprintk("%s: t: %p, nr_pages: %u, end: %lu, index: %lu, max: %u.\n",
197 __func__
, trans
, nr_pages
, end
, index
, trans
->page_num
);
202 err
= pohmelfs_write_inode_create(inode
, trans
);
209 for (i
= 0; i
< trans
->page_num
; i
++) {
210 struct page
*page
= trans
->pages
[i
];
214 if (unlikely(page
->mapping
!= mapping
))
217 if (!wbc
->range_cyclic
&& page
->index
> end
) {
222 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
223 wait_on_page_writeback(page
);
225 if (PageWriteback(page
) ||
226 !clear_page_dirty_for_io(page
)) {
227 dprintk("%s: not clear for io page: %p, writeback: %d.\n",
228 __func__
, page
, PageWriteback(page
));
232 set_page_writeback(page
);
234 trans
->attached_size
+= page_private(page
);
235 trans
->attached_pages
++;
237 dprintk("%s: %u/%u added trans: %p, gen: %u, page: %p, [High: %d], size: %lu, idx: %lu.\n",
238 __func__
, i
, trans
->page_num
, trans
, trans
->gen
, page
,
239 !!PageHighMem(page
), page_private(page
), page
->index
);
243 if (wbc
->nr_to_write
<= 0)
249 trans
->pages
[i
] = NULL
;
252 err
= netfs_trans_finish(trans
, psb
);
260 netfs_trans_reset(trans
);
261 netfs_trans_put(trans
);
265 if (!scanned
&& !done
) {
267 * We hit the last page and there is more work to be done: wrap
268 * back to the start of the file
275 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
276 mapping
->writeback_index
= index
;
282 * Inode writeback creation completion callback.
283 * Only invoked for just created inodes, which do not have pages attached,
284 * like dirs and empty files.
286 static int pohmelfs_write_inode_complete(struct page
**pages
, unsigned int page_num
,
287 void *private, int err
)
289 struct inode
*inode
= private;
290 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
294 mark_inode_dirty(inode
);
295 clear_bit(NETFS_INODE_REMOTE_SYNCED
, &pi
->state
);
297 set_bit(NETFS_INODE_REMOTE_SYNCED
, &pi
->state
);
300 pohmelfs_put_inode(pi
);
306 int pohmelfs_write_create_inode(struct pohmelfs_inode
*pi
)
308 struct netfs_trans
*t
;
309 struct inode
*inode
= &pi
->vfs_inode
;
310 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
313 if (test_bit(NETFS_INODE_REMOTE_SYNCED
, &pi
->state
))
316 dprintk("%s: started ino: %llu.\n", __func__
, pi
->ino
);
318 err
= pohmelfs_path_length(pi
);
322 t
= netfs_trans_alloc(psb
, err
+ 1, 0, 0);
327 t
->complete
= pohmelfs_write_inode_complete
;
328 t
->private = igrab(inode
);
334 err
= pohmelfs_write_inode_create(inode
, t
);
338 netfs_trans_finish(t
, POHMELFS_SB(inode
->i_sb
));
350 * Sync all not-yet-created children in given directory to the server.
352 static int pohmelfs_write_inode_create_children(struct inode
*inode
)
354 struct pohmelfs_inode
*parent
= POHMELFS_I(inode
);
355 struct super_block
*sb
= inode
->i_sb
;
356 struct pohmelfs_name
*n
;
358 while (!list_empty(&parent
->sync_create_list
)) {
360 mutex_lock(&parent
->offset_lock
);
361 if (!list_empty(&parent
->sync_create_list
)) {
362 n
= list_first_entry(&parent
->sync_create_list
,
363 struct pohmelfs_name
, sync_create_entry
);
364 list_del_init(&n
->sync_create_entry
);
366 mutex_unlock(&parent
->offset_lock
);
371 inode
= ilookup(sb
, n
->ino
);
373 dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n",
374 __func__
, parent
->ino
, n
->ino
, inode
);
376 if (inode
&& (inode
->i_state
& I_DIRTY
)) {
377 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
378 pohmelfs_write_create_inode(pi
);
379 /* pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0); */
388 * Removes given child from given inode on server.
390 int pohmelfs_remove_child(struct pohmelfs_inode
*pi
, struct pohmelfs_name
*n
)
392 return pohmelfs_meta_command_data(pi
, pi
->ino
, NETFS_REMOVE
, NULL
, 0, NULL
, NULL
, 0);
396 * Writeback for given inode.
398 static int pohmelfs_write_inode(struct inode
*inode
,
399 struct writeback_control
*wbc
)
401 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
403 pohmelfs_write_create_inode(pi
);
404 pohmelfs_write_inode_create_children(inode
);
410 * It is not exported, sorry...
412 static inline wait_queue_head_t
*page_waitqueue(struct page
*page
)
414 const struct zone
*zone
= page_zone(page
);
416 return &zone
->wait_table
[hash_ptr(page
, zone
->wait_table_bits
)];
419 static int pohmelfs_wait_on_page_locked(struct page
*page
)
421 struct pohmelfs_sb
*psb
= POHMELFS_SB(page
->mapping
->host
->i_sb
);
422 long ret
= psb
->wait_on_page_timeout
;
423 DEFINE_WAIT_BIT(wait
, &page
->flags
, PG_locked
);
426 if (!PageLocked(page
))
430 prepare_to_wait(page_waitqueue(page
),
431 &wait
.wait
, TASK_INTERRUPTIBLE
);
433 dprintk("%s: page: %p, locked: %d, uptodate: %d, error: %d, flags: %lx.\n",
434 __func__
, page
, PageLocked(page
), PageUptodate(page
),
435 PageError(page
), page
->flags
);
437 if (!PageLocked(page
))
440 if (!signal_pending(current
)) {
441 ret
= schedule_timeout(ret
);
449 finish_wait(page_waitqueue(page
), &wait
.wait
);
456 SetPageUptodate(page
);
459 printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n",
460 __func__
, page
, PageUptodate(page
), PageLocked(page
), err
);
465 static int pohmelfs_read_page_complete(struct page
**pages
, unsigned int page_num
,
466 void *private, int err
)
468 struct page
*page
= private;
470 if (PageChecked(page
))
474 dprintk("%s: page: %p, err: %d.\n", __func__
, page
, err
);
484 * Read a page from remote server.
485 * Function will wait until page is unlocked.
487 static int pohmelfs_readpage(struct file
*file
, struct page
*page
)
489 struct inode
*inode
= page
->mapping
->host
;
490 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
491 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
492 struct netfs_trans
*t
;
493 struct netfs_cmd
*cmd
;
498 err
= pohmelfs_data_lock(pi
, page
->index
<< PAGE_CACHE_SHIFT
,
499 PAGE_SIZE
, POHMELFS_READ_LOCK
);
503 isize
= i_size_read(inode
);
504 if (isize
<= page
->index
<< PAGE_CACHE_SHIFT
) {
505 SetPageUptodate(page
);
510 path_len
= pohmelfs_path_length(pi
);
516 t
= netfs_trans_alloc(psb
, path_len
, NETFS_TRANS_SINGLE_DST
, 0);
522 t
->complete
= pohmelfs_read_page_complete
;
525 cmd
= netfs_trans_current(t
);
526 data
= (void *)(cmd
+ 1);
528 err
= pohmelfs_construct_path_string(pi
, data
, path_len
);
535 cmd
->start
= page
->index
;
536 cmd
->start
<<= PAGE_CACHE_SHIFT
;
537 cmd
->size
= PAGE_CACHE_SIZE
+ path_len
;
538 cmd
->cmd
= NETFS_READ_PAGE
;
541 dprintk("%s: path: '%s', page: %p, ino: %llu, start: %llu, size: %lu.\n",
542 __func__
, (char *)data
, page
, pi
->ino
, cmd
->start
, PAGE_CACHE_SIZE
);
544 netfs_convert_cmd(cmd
);
545 netfs_trans_update(cmd
, t
, path_len
);
547 err
= netfs_trans_finish(t
, psb
);
551 return pohmelfs_wait_on_page_locked(page
);
558 if (PageLocked(page
))
561 printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n",
562 __func__
, page
, page
->index
<< PAGE_CACHE_SHIFT
, PAGE_CACHE_SIZE
, err
);
568 * Write begin/end magic.
569 * Allocates a page and writes inode if it was not synced to server before.
571 static int pohmelfs_write_begin(struct file
*file
, struct address_space
*mapping
,
572 loff_t pos
, unsigned len
, unsigned flags
,
573 struct page
**pagep
, void **fsdata
)
575 struct inode
*inode
= mapping
->host
;
583 index
= pos
>> PAGE_CACHE_SHIFT
;
584 start
= pos
& (PAGE_CACHE_SIZE
- 1);
587 page
= grab_cache_page(mapping
, index
);
589 dprintk("%s: page: %p pos: %llu, len: %u, index: %lu, start: %u, end: %u, uptodate: %d.\n",
590 __func__
, page
, pos
, len
, index
, start
, end
, PageUptodate(page
));
597 while (!PageUptodate(page
)) {
598 if (start
&& test_bit(NETFS_INODE_REMOTE_SYNCED
, &POHMELFS_I(inode
)->state
)) {
599 err
= pohmelfs_readpage(file
, page
);
607 if (len
!= PAGE_CACHE_SIZE
) {
608 void *kaddr
= kmap_atomic(page
, KM_USER0
);
610 memset(kaddr
+ start
, 0, PAGE_CACHE_SIZE
- start
);
611 flush_dcache_page(page
);
612 kunmap_atomic(kaddr
, KM_USER0
);
614 SetPageUptodate(page
);
617 set_page_private(page
, end
);
624 page_cache_release(page
);
630 static int pohmelfs_write_end(struct file
*file
, struct address_space
*mapping
,
631 loff_t pos
, unsigned len
, unsigned copied
,
632 struct page
*page
, void *fsdata
)
634 struct inode
*inode
= mapping
->host
;
637 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
638 void *kaddr
= kmap_atomic(page
, KM_USER0
);
640 memset(kaddr
+ from
+ copied
, 0, len
- copied
);
641 flush_dcache_page(page
);
642 kunmap_atomic(kaddr
, KM_USER0
);
645 SetPageUptodate(page
);
646 set_page_dirty(page
);
648 dprintk("%s: page: %p [U: %d, D: %d, L: %d], pos: %llu, len: %u, copied: %u.\n",
650 PageUptodate(page
), PageDirty(page
), PageLocked(page
),
653 flush_dcache_page(page
);
656 page_cache_release(page
);
658 if (pos
+ copied
> inode
->i_size
) {
659 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
661 psb
->avail_size
-= pos
+ copied
- inode
->i_size
;
663 i_size_write(inode
, pos
+ copied
);
669 static int pohmelfs_readpages_trans_complete(struct page
**__pages
, unsigned int page_num
,
670 void *private, int err
)
672 struct pohmelfs_inode
*pi
= private;
674 struct page
**pages
, *page
= (struct page
*)__pages
;
675 loff_t index
= page
->index
;
677 pages
= kzalloc(sizeof(void *) * page_num
, GFP_NOIO
);
681 num
= find_get_pages_contig(pi
->vfs_inode
.i_mapping
, index
, page_num
, pages
);
687 for (i
= 0; i
< num
; ++i
) {
691 printk("%s: %u/%u: page: %p, index: %lu, uptodate: %d, locked: %d, err: %d.\n",
692 __func__
, i
, num
, page
, page
->index
,
693 PageUptodate(page
), PageLocked(page
), err
);
695 if (!PageChecked(page
)) {
700 page_cache_release(page
);
701 page_cache_release(page
);
709 static int pohmelfs_send_readpages(struct pohmelfs_inode
*pi
, struct page
*first
, unsigned int num
)
711 struct netfs_trans
*t
;
712 struct netfs_cmd
*cmd
;
713 struct pohmelfs_sb
*psb
= POHMELFS_SB(pi
->vfs_inode
.i_sb
);
717 err
= pohmelfs_data_lock(pi
, first
->index
<< PAGE_CACHE_SHIFT
,
718 num
* PAGE_SIZE
, POHMELFS_READ_LOCK
);
722 path_len
= pohmelfs_path_length(pi
);
728 t
= netfs_trans_alloc(psb
, path_len
, NETFS_TRANS_SINGLE_DST
, 0);
734 cmd
= netfs_trans_current(t
);
735 data
= (void *)(cmd
+ 1);
737 t
->complete
= pohmelfs_readpages_trans_complete
;
740 t
->pages
= (struct page
**)first
;
742 err
= pohmelfs_construct_path_string(pi
, data
, path_len
);
748 cmd
->cmd
= NETFS_READ_PAGES
;
749 cmd
->start
= first
->index
;
750 cmd
->start
<<= PAGE_CACHE_SHIFT
;
751 cmd
->size
= (num
<< 8 | PAGE_CACHE_SHIFT
);
755 dprintk("%s: t: %p, gen: %u, path: '%s', path_len: %u, "
756 "start: %lu, num: %u.\n",
757 __func__
, t
, t
->gen
, (char *)data
, path_len
,
760 netfs_convert_cmd(cmd
);
761 netfs_trans_update(cmd
, t
, path_len
);
763 return netfs_trans_finish(t
, psb
);
768 pohmelfs_readpages_trans_complete((struct page
**)first
, num
, pi
, err
);
772 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
774 static int pohmelfs_readpages(struct file
*file
, struct address_space
*mapping
,
775 struct list_head
*pages
, unsigned nr_pages
)
777 unsigned int page_idx
, num
= 0;
778 struct page
*page
= NULL
, *first
= NULL
;
780 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
781 page
= list_to_page(pages
);
783 prefetchw(&page
->flags
);
784 list_del(&page
->lru
);
786 if (!add_to_page_cache_lru(page
, mapping
,
787 page
->index
, GFP_KERNEL
)) {
795 dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n",
796 __func__
, page
, page
->index
, first
->index
);
798 if (unlikely(first
->index
+ num
!= page
->index
) || (num
> 500)) {
799 pohmelfs_send_readpages(POHMELFS_I(mapping
->host
),
808 pohmelfs_send_readpages(POHMELFS_I(mapping
->host
), first
, num
);
811 * This will be sync read, so when last page is processed,
812 * all previous are alerady unlocked and ready to be used.
818 * Small address space operations for POHMELFS.
820 const struct address_space_operations pohmelfs_aops
= {
821 .readpage
= pohmelfs_readpage
,
822 .readpages
= pohmelfs_readpages
,
823 .writepages
= pohmelfs_writepages
,
824 .write_begin
= pohmelfs_write_begin
,
825 .write_end
= pohmelfs_write_end
,
826 .set_page_dirty
= __set_page_dirty_nobuffers
,
830 * ->detroy_inode() callback. Deletes inode from the caches
831 * and frees private data.
833 static void pohmelfs_destroy_inode(struct inode
*inode
)
835 struct super_block
*sb
= inode
->i_sb
;
836 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
837 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
839 /* pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK); */
841 pohmelfs_inode_del_inode(psb
, pi
);
843 dprintk("%s: pi: %p, inode: %p, ino: %llu.\n",
844 __func__
, pi
, &pi
->vfs_inode
, pi
->ino
);
845 kmem_cache_free(pohmelfs_inode_cache
, pi
);
846 atomic_long_dec(&psb
->total_inodes
);
850 * ->alloc_inode() callback. Allocates inode and initializes private data.
852 static struct inode
*pohmelfs_alloc_inode(struct super_block
*sb
)
854 struct pohmelfs_inode
*pi
;
856 pi
= kmem_cache_alloc(pohmelfs_inode_cache
, GFP_NOIO
);
860 pi
->hash_root
= RB_ROOT
;
861 mutex_init(&pi
->offset_lock
);
863 INIT_LIST_HEAD(&pi
->sync_create_list
);
865 INIT_LIST_HEAD(&pi
->inode_entry
);
872 dprintk("%s: pi: %p, inode: %p.\n", __func__
, pi
, &pi
->vfs_inode
);
874 atomic_long_inc(&POHMELFS_SB(sb
)->total_inodes
);
876 return &pi
->vfs_inode
;
880 * We want fsync() to work on POHMELFS.
882 static int pohmelfs_fsync(struct file
*file
, int datasync
)
884 struct inode
*inode
= file
->f_mapping
->host
;
885 struct writeback_control wbc
= {
886 .sync_mode
= WB_SYNC_ALL
,
887 .nr_to_write
= 0, /* sys_fsync did this */
890 return sync_inode(inode
, &wbc
);
893 ssize_t
pohmelfs_write(struct file
*file
, const char __user
*buf
,
894 size_t len
, loff_t
*ppos
)
896 struct address_space
*mapping
= file
->f_mapping
;
897 struct inode
*inode
= mapping
->host
;
898 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
899 struct iovec iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= len
};
904 init_sync_kiocb(&kiocb
, file
);
908 dprintk("%s: len: %zu, pos: %llu.\n", __func__
, len
, pos
);
910 mutex_lock(&inode
->i_mutex
);
911 ret
= pohmelfs_data_lock(pi
, pos
, len
, POHMELFS_WRITE_LOCK
);
915 ret
= __generic_file_aio_write(&kiocb
, &iov
, 1, &kiocb
.ki_pos
);
916 *ppos
= kiocb
.ki_pos
;
918 mutex_unlock(&inode
->i_mutex
);
924 err
= generic_write_sync(file
, pos
, ret
);
933 mutex_unlock(&inode
->i_mutex
);
937 static const struct file_operations pohmelfs_file_ops
= {
938 .open
= generic_file_open
,
939 .fsync
= pohmelfs_fsync
,
941 .llseek
= generic_file_llseek
,
943 .read
= do_sync_read
,
944 .aio_read
= generic_file_aio_read
,
946 .mmap
= generic_file_mmap
,
948 .splice_read
= generic_file_splice_read
,
949 .splice_write
= generic_file_splice_write
,
951 .write
= pohmelfs_write
,
952 .aio_write
= generic_file_aio_write
,
955 const struct inode_operations pohmelfs_symlink_inode_operations
= {
956 .readlink
= generic_readlink
,
957 .follow_link
= page_follow_link_light
,
958 .put_link
= page_put_link
,
961 int pohmelfs_setattr_raw(struct inode
*inode
, struct iattr
*attr
)
965 err
= inode_change_ok(inode
, attr
);
967 dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__
, POHMELFS_I(inode
)->ino
);
971 if ((attr
->ia_valid
& ATTR_SIZE
) &&
972 attr
->ia_size
!= i_size_read(inode
)) {
973 err
= vmtruncate(inode
, attr
->ia_size
);
975 dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__
, POHMELFS_I(inode
)->ino
);
980 setattr_copy(inode
, attr
);
981 mark_inode_dirty(inode
);
983 dprintk("%s: ino: %llu, mode: %o -> %o, uid: %u -> %u, gid: %u -> %u, size: %llu -> %llu.\n",
984 __func__
, POHMELFS_I(inode
)->ino
, inode
->i_mode
, attr
->ia_mode
,
985 inode
->i_uid
, attr
->ia_uid
, inode
->i_gid
, attr
->ia_gid
, inode
->i_size
, attr
->ia_size
);
993 int pohmelfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
995 struct inode
*inode
= dentry
->d_inode
;
996 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
999 err
= pohmelfs_data_lock(pi
, 0, ~0, POHMELFS_WRITE_LOCK
);
1003 err
= security_inode_setattr(dentry
, attr
);
1007 err
= pohmelfs_setattr_raw(inode
, attr
);
1017 static int pohmelfs_send_xattr_req(struct pohmelfs_inode
*pi
, u64 id
, u64 start
,
1018 const char *name
, const void *value
, size_t attrsize
, int command
)
1020 struct pohmelfs_sb
*psb
= POHMELFS_SB(pi
->vfs_inode
.i_sb
);
1021 int err
, path_len
, namelen
= strlen(name
) + 1; /* 0-byte */
1022 struct netfs_trans
*t
;
1023 struct netfs_cmd
*cmd
;
1026 dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n",
1027 __func__
, id
, start
, name
, attrsize
, command
);
1029 path_len
= pohmelfs_path_length(pi
);
1035 t
= netfs_trans_alloc(psb
, namelen
+ path_len
+ attrsize
, 0, 0);
1041 cmd
= netfs_trans_current(t
);
1044 path_len
= pohmelfs_construct_path_string(pi
, data
, path_len
);
1052 * 'name' is a NUL-terminated string already and
1053 * 'namelen' includes 0-byte.
1055 memcpy(data
, name
, namelen
);
1058 memcpy(data
, value
, attrsize
);
1063 cmd
->size
= attrsize
+ namelen
+ path_len
;
1064 cmd
->ext
= path_len
;
1068 netfs_convert_cmd(cmd
);
1069 netfs_trans_update(cmd
, t
, namelen
+ path_len
+ attrsize
);
1071 return netfs_trans_finish(t
, psb
);
1080 static int pohmelfs_setxattr(struct dentry
*dentry
, const char *name
,
1081 const void *value
, size_t attrsize
, int flags
)
1083 struct inode
*inode
= dentry
->d_inode
;
1084 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1085 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1087 if (!(psb
->state_flags
& POHMELFS_FLAGS_XATTR
))
1090 return pohmelfs_send_xattr_req(pi
, flags
, attrsize
, name
,
1091 value
, attrsize
, NETFS_XATTR_SET
);
1094 static ssize_t
pohmelfs_getxattr(struct dentry
*dentry
, const char *name
,
1095 void *value
, size_t attrsize
)
1097 struct inode
*inode
= dentry
->d_inode
;
1098 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1099 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1100 struct pohmelfs_mcache
*m
;
1102 long timeout
= psb
->mcache_timeout
;
1104 if (!(psb
->state_flags
& POHMELFS_FLAGS_XATTR
))
1107 m
= pohmelfs_mcache_alloc(psb
, 0, attrsize
, value
);
1111 dprintk("%s: ino: %llu, name: '%s', size: %zu.\n",
1112 __func__
, pi
->ino
, name
, attrsize
);
1114 err
= pohmelfs_send_xattr_req(pi
, m
->gen
, attrsize
, name
, value
, 0, NETFS_XATTR_GET
);
1119 err
= wait_for_completion_timeout(&m
->complete
, timeout
);
1126 * This loop is a bit ugly, since it waits until reference counter
1127 * hits 1 and then put object here. Main goal is to prevent race with
1128 * network thread, when it can start processing given request, i.e.
1129 * increase its reference counter but yet not complete it, while
1130 * we will exit from ->getxattr() with timeout, and although request
1131 * will not be freed (its reference counter was increased by network
1132 * thread), data pointer provided by user may be released, so we will
1133 * overwrite already freed area in network thread.
1135 * Now after timeout we remove request from the cache, so it can not be
1136 * found by network thread, and wait for its reference counter to hit 1,
1137 * i.e. if network thread already started to process this request, we wait
1138 * it to finish, and then free object locally. If reference counter is
1139 * already 1, i.e. request is not used by anyone else, we can free it without
1145 pohmelfs_mcache_remove_locked(psb
, m
);
1146 } while (atomic_read(&m
->refcnt
) != 1);
1148 pohmelfs_mcache_put(psb
, m
);
1150 dprintk("%s: ino: %llu, err: %d.\n", __func__
, pi
->ino
, err
);
1155 pohmelfs_mcache_put(psb
, m
);
1159 static int pohmelfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
1161 struct inode
*inode
= dentry
->d_inode
;
1163 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1166 err
= pohmelfs_data_lock(pi
, 0, ~0, POHMELFS_READ_LOCK
);
1169 dprintk("%s: ino: %llu, mode: %o, uid: %u, gid: %u, size: %llu.\n",
1170 __func__
, pi
->ino
, inode
->i_mode
, inode
->i_uid
,
1171 inode
->i_gid
, inode
->i_size
);
1174 generic_fillattr(inode
, stat
);
1178 const struct inode_operations pohmelfs_file_inode_operations
= {
1179 .setattr
= pohmelfs_setattr
,
1180 .getattr
= pohmelfs_getattr
,
1181 .setxattr
= pohmelfs_setxattr
,
1182 .getxattr
= pohmelfs_getxattr
,
1186 * Fill inode data: mode, size, operation callbacks and so on...
1188 void pohmelfs_fill_inode(struct inode
*inode
, struct netfs_inode_info
*info
)
1190 inode
->i_mode
= info
->mode
;
1191 inode
->i_nlink
= info
->nlink
;
1192 inode
->i_uid
= info
->uid
;
1193 inode
->i_gid
= info
->gid
;
1194 inode
->i_blocks
= info
->blocks
;
1195 inode
->i_rdev
= info
->rdev
;
1196 inode
->i_size
= info
->size
;
1197 inode
->i_version
= info
->version
;
1198 inode
->i_blkbits
= ffs(info
->blocksize
);
1200 dprintk("%s: inode: %p, num: %lu/%llu inode is regular: %d, dir: %d, link: %d, mode: %o, size: %llu.\n",
1201 __func__
, inode
, inode
->i_ino
, info
->ino
,
1202 S_ISREG(inode
->i_mode
), S_ISDIR(inode
->i_mode
),
1203 S_ISLNK(inode
->i_mode
), inode
->i_mode
, inode
->i_size
);
1205 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
1208 * i_mapping is a pointer to i_data during inode initialization.
1210 inode
->i_data
.a_ops
= &pohmelfs_aops
;
1212 if (S_ISREG(inode
->i_mode
)) {
1213 inode
->i_fop
= &pohmelfs_file_ops
;
1214 inode
->i_op
= &pohmelfs_file_inode_operations
;
1215 } else if (S_ISDIR(inode
->i_mode
)) {
1216 inode
->i_fop
= &pohmelfs_dir_fops
;
1217 inode
->i_op
= &pohmelfs_dir_inode_ops
;
1218 } else if (S_ISLNK(inode
->i_mode
)) {
1219 inode
->i_op
= &pohmelfs_symlink_inode_operations
;
1220 inode
->i_fop
= &pohmelfs_file_ops
;
1222 inode
->i_fop
= &generic_ro_fops
;
1226 static int pohmelfs_drop_inode(struct inode
*inode
)
1228 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1229 struct pohmelfs_inode
*pi
= POHMELFS_I(inode
);
1231 spin_lock(&psb
->ino_lock
);
1232 list_del_init(&pi
->inode_entry
);
1233 spin_unlock(&psb
->ino_lock
);
1235 return generic_drop_inode(inode
);
1238 static struct pohmelfs_inode
*pohmelfs_get_inode_from_list(struct pohmelfs_sb
*psb
,
1239 struct list_head
*head
, unsigned int *count
)
1241 struct pohmelfs_inode
*pi
= NULL
;
1243 spin_lock(&psb
->ino_lock
);
1244 if (!list_empty(head
)) {
1245 pi
= list_entry(head
->next
, struct pohmelfs_inode
,
1247 list_del_init(&pi
->inode_entry
);
1248 *count
= pi
->drop_count
;
1251 spin_unlock(&psb
->ino_lock
);
1256 static void pohmelfs_flush_transactions(struct pohmelfs_sb
*psb
)
1258 struct pohmelfs_config
*c
;
1260 mutex_lock(&psb
->state_lock
);
1261 list_for_each_entry(c
, &psb
->state_list
, config_entry
) {
1262 pohmelfs_state_flush_transactions(&c
->state
);
1264 mutex_unlock(&psb
->state_lock
);
1268 * ->put_super() callback. Invoked before superblock is destroyed,
1269 * so it has to clean all private data.
1271 static void pohmelfs_put_super(struct super_block
*sb
)
1273 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
1274 struct pohmelfs_inode
*pi
;
1275 unsigned int count
= 0;
1276 unsigned int in_drop_list
= 0;
1277 struct inode
*inode
, *tmp
;
1279 dprintk("%s.\n", __func__
);
1282 * Kill pending transactions, which could affect inodes in-flight.
1284 pohmelfs_flush_transactions(psb
);
1286 while ((pi
= pohmelfs_get_inode_from_list(psb
, &psb
->drop_list
, &count
))) {
1287 inode
= &pi
->vfs_inode
;
1289 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1290 __func__
, pi
->ino
, pi
, inode
, count
);
1292 if (atomic_read(&inode
->i_count
) != count
) {
1293 printk("%s: ino: %llu, pi: %p, inode: %p, count: %u, i_count: %d.\n",
1294 __func__
, pi
->ino
, pi
, inode
, count
,
1295 atomic_read(&inode
->i_count
));
1296 count
= atomic_read(&inode
->i_count
);
1301 iput(&pi
->vfs_inode
);
1304 list_for_each_entry_safe(inode
, tmp
, &sb
->s_inodes
, i_sb_list
) {
1305 pi
= POHMELFS_I(inode
);
1307 dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n",
1308 __func__
, pi
->ino
, pi
, inode
, atomic_read(&inode
->i_count
));
1311 * These are special inodes, they were created during
1312 * directory reading or lookup, and were not bound to dentry,
1313 * so they live here with reference counter being 1 and prevent
1314 * umount from succeed since it believes that they are busy.
1316 count
= atomic_read(&inode
->i_count
);
1318 list_del_init(&inode
->i_sb_list
);
1320 iput(&pi
->vfs_inode
);
1324 psb
->trans_scan_timeout
= psb
->drop_scan_timeout
= 0;
1325 cancel_rearming_delayed_work(&psb
->dwork
);
1326 cancel_rearming_delayed_work(&psb
->drop_dwork
);
1327 flush_scheduled_work();
1329 dprintk("%s: stopped workqueues.\n", __func__
);
1331 pohmelfs_crypto_exit(psb
);
1332 pohmelfs_state_exit(psb
);
1334 bdi_destroy(&psb
->bdi
);
1337 sb
->s_fs_info
= NULL
;
1340 static int pohmelfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1342 struct super_block
*sb
= dentry
->d_sb
;
1343 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
1346 * There are no filesystem size limits yet.
1348 memset(buf
, 0, sizeof(struct kstatfs
));
1350 buf
->f_type
= POHMELFS_MAGIC_NUM
; /* 'POH.' */
1351 buf
->f_bsize
= sb
->s_blocksize
;
1352 buf
->f_files
= psb
->ino
;
1353 buf
->f_namelen
= 255;
1354 buf
->f_files
= atomic_long_read(&psb
->total_inodes
);
1355 buf
->f_bfree
= buf
->f_bavail
= psb
->avail_size
>> PAGE_SHIFT
;
1356 buf
->f_blocks
= psb
->total_size
>> PAGE_SHIFT
;
1358 dprintk("%s: total: %llu, avail: %llu, inodes: %llu, bsize: %lu.\n",
1359 __func__
, psb
->total_size
, psb
->avail_size
, buf
->f_files
, sb
->s_blocksize
);
1364 static int pohmelfs_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1366 struct pohmelfs_sb
*psb
= POHMELFS_SB(vfs
->mnt_sb
);
1368 seq_printf(seq
, ",idx=%u", psb
->idx
);
1369 seq_printf(seq
, ",trans_scan_timeout=%u", jiffies_to_msecs(psb
->trans_scan_timeout
));
1370 seq_printf(seq
, ",drop_scan_timeout=%u", jiffies_to_msecs(psb
->drop_scan_timeout
));
1371 seq_printf(seq
, ",wait_on_page_timeout=%u", jiffies_to_msecs(psb
->wait_on_page_timeout
));
1372 seq_printf(seq
, ",trans_retries=%u", psb
->trans_retries
);
1373 seq_printf(seq
, ",crypto_thread_num=%u", psb
->crypto_thread_num
);
1374 seq_printf(seq
, ",trans_max_pages=%u", psb
->trans_max_pages
);
1375 seq_printf(seq
, ",mcache_timeout=%u", jiffies_to_msecs(psb
->mcache_timeout
));
1376 if (psb
->crypto_fail_unsupported
)
1377 seq_printf(seq
, ",crypto_fail_unsupported");
1384 pohmelfs_opt_crypto_thread_num
,
1385 pohmelfs_opt_trans_max_pages
,
1386 pohmelfs_opt_crypto_fail_unsupported
,
1388 /* Remountable options */
1389 pohmelfs_opt_trans_scan_timeout
,
1390 pohmelfs_opt_drop_scan_timeout
,
1391 pohmelfs_opt_wait_on_page_timeout
,
1392 pohmelfs_opt_trans_retries
,
1393 pohmelfs_opt_mcache_timeout
,
1396 static struct match_token pohmelfs_tokens
[] = {
1397 {pohmelfs_opt_idx
, "idx=%u"},
1398 {pohmelfs_opt_crypto_thread_num
, "crypto_thread_num=%u"},
1399 {pohmelfs_opt_trans_max_pages
, "trans_max_pages=%u"},
1400 {pohmelfs_opt_crypto_fail_unsupported
, "crypto_fail_unsupported"},
1401 {pohmelfs_opt_trans_scan_timeout
, "trans_scan_timeout=%u"},
1402 {pohmelfs_opt_drop_scan_timeout
, "drop_scan_timeout=%u"},
1403 {pohmelfs_opt_wait_on_page_timeout
, "wait_on_page_timeout=%u"},
1404 {pohmelfs_opt_trans_retries
, "trans_retries=%u"},
1405 {pohmelfs_opt_mcache_timeout
, "mcache_timeout=%u"},
1408 static int pohmelfs_parse_options(char *options
, struct pohmelfs_sb
*psb
, int remount
)
1411 substring_t args
[MAX_OPT_ARGS
];
1417 while ((p
= strsep(&options
, ",")) != NULL
) {
1422 token
= match_token(p
, pohmelfs_tokens
, args
);
1424 err
= match_int(&args
[0], &option
);
1428 if (remount
&& token
<= pohmelfs_opt_crypto_fail_unsupported
)
1432 case pohmelfs_opt_idx
:
1435 case pohmelfs_opt_trans_scan_timeout
:
1436 psb
->trans_scan_timeout
= msecs_to_jiffies(option
);
1438 case pohmelfs_opt_drop_scan_timeout
:
1439 psb
->drop_scan_timeout
= msecs_to_jiffies(option
);
1441 case pohmelfs_opt_wait_on_page_timeout
:
1442 psb
->wait_on_page_timeout
= msecs_to_jiffies(option
);
1444 case pohmelfs_opt_mcache_timeout
:
1445 psb
->mcache_timeout
= msecs_to_jiffies(option
);
1447 case pohmelfs_opt_trans_retries
:
1448 psb
->trans_retries
= option
;
1450 case pohmelfs_opt_crypto_thread_num
:
1451 psb
->crypto_thread_num
= option
;
1453 case pohmelfs_opt_trans_max_pages
:
1454 psb
->trans_max_pages
= option
;
1456 case pohmelfs_opt_crypto_fail_unsupported
:
1457 psb
->crypto_fail_unsupported
= 1;
1467 static int pohmelfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1470 struct pohmelfs_sb
*psb
= POHMELFS_SB(sb
);
1471 unsigned long old_sb_flags
= sb
->s_flags
;
1473 err
= pohmelfs_parse_options(data
, psb
, 1);
1475 goto err_out_restore
;
1477 if (!(*flags
& MS_RDONLY
))
1478 sb
->s_flags
&= ~MS_RDONLY
;
1482 sb
->s_flags
= old_sb_flags
;
1486 static void pohmelfs_flush_inode(struct pohmelfs_inode
*pi
, unsigned int count
)
1488 struct inode
*inode
= &pi
->vfs_inode
;
1490 dprintk("%s: %p: ino: %llu, owned: %d.\n",
1491 __func__
, inode
, pi
->ino
, test_bit(NETFS_INODE_OWNED
, &pi
->state
));
1493 mutex_lock(&inode
->i_mutex
);
1494 if (test_and_clear_bit(NETFS_INODE_OWNED
, &pi
->state
)) {
1495 filemap_fdatawrite(inode
->i_mapping
);
1496 inode
->i_sb
->s_op
->write_inode(inode
, 0);
1499 #ifdef POHMELFS_TRUNCATE_ON_INODE_FLUSH
1500 truncate_inode_pages(inode
->i_mapping
, 0);
1503 pohmelfs_data_unlock(pi
, 0, ~0, POHMELFS_WRITE_LOCK
);
1504 mutex_unlock(&inode
->i_mutex
);
1507 static void pohmelfs_put_inode_count(struct pohmelfs_inode
*pi
, unsigned int count
)
1509 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1510 __func__
, pi
->ino
, pi
, &pi
->vfs_inode
, count
);
1512 if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH
, &pi
->state
))
1513 pohmelfs_flush_inode(pi
, count
);
1516 iput(&pi
->vfs_inode
);
1519 static void pohmelfs_drop_scan(struct work_struct
*work
)
1521 struct pohmelfs_sb
*psb
=
1522 container_of(work
, struct pohmelfs_sb
, drop_dwork
.work
);
1523 struct pohmelfs_inode
*pi
;
1524 unsigned int count
= 0;
1526 while ((pi
= pohmelfs_get_inode_from_list(psb
, &psb
->drop_list
, &count
)))
1527 pohmelfs_put_inode_count(pi
, count
);
1529 pohmelfs_check_states(psb
);
1531 if (psb
->drop_scan_timeout
)
1532 schedule_delayed_work(&psb
->drop_dwork
, psb
->drop_scan_timeout
);
1536 * Run through all transactions starting from the oldest,
1537 * drop transaction from current state and try to send it
1538 * to all remote nodes, which are currently installed.
1540 static void pohmelfs_trans_scan_state(struct netfs_state
*st
)
1542 struct rb_node
*rb_node
;
1543 struct netfs_trans_dst
*dst
;
1544 struct pohmelfs_sb
*psb
= st
->psb
;
1545 unsigned int timeout
= psb
->trans_scan_timeout
;
1546 struct netfs_trans
*t
;
1549 mutex_lock(&st
->trans_lock
);
1550 for (rb_node
= rb_first(&st
->trans_root
); rb_node
; ) {
1551 dst
= rb_entry(rb_node
, struct netfs_trans_dst
, state_entry
);
1554 if (timeout
&& time_after(dst
->send_time
+ timeout
, jiffies
)
1555 && dst
->retries
== 0)
1558 dprintk("%s: t: %p, gen: %u, st: %p, retries: %u, max: %u.\n",
1559 __func__
, t
, t
->gen
, st
, dst
->retries
, psb
->trans_retries
);
1562 rb_node
= rb_next(rb_node
);
1565 if (timeout
&& (++dst
->retries
< psb
->trans_retries
))
1566 err
= netfs_trans_resend(t
, psb
);
1568 if (err
|| (t
->flags
& NETFS_TRANS_SINGLE_DST
)) {
1569 if (netfs_trans_remove_nolock(dst
, st
))
1570 netfs_trans_drop_dst_nostate(dst
);
1576 mutex_unlock(&st
->trans_lock
);
1580 * Walk through all installed network states and resend all
1581 * transactions, which are old enough.
1583 static void pohmelfs_trans_scan(struct work_struct
*work
)
1585 struct pohmelfs_sb
*psb
=
1586 container_of(work
, struct pohmelfs_sb
, dwork
.work
);
1587 struct netfs_state
*st
;
1588 struct pohmelfs_config
*c
;
1590 mutex_lock(&psb
->state_lock
);
1591 list_for_each_entry(c
, &psb
->state_list
, config_entry
) {
1594 pohmelfs_trans_scan_state(st
);
1596 mutex_unlock(&psb
->state_lock
);
1599 * If no timeout specified then system is in the middle of umount process,
1600 * so no need to reschedule scanning process again.
1602 if (psb
->trans_scan_timeout
)
1603 schedule_delayed_work(&psb
->dwork
, psb
->trans_scan_timeout
);
1606 int pohmelfs_meta_command_data(struct pohmelfs_inode
*pi
, u64 id
, unsigned int cmd_op
, char *addon
,
1607 unsigned int flags
, netfs_trans_complete_t complete
, void *priv
, u64 start
)
1609 struct inode
*inode
= &pi
->vfs_inode
;
1610 struct pohmelfs_sb
*psb
= POHMELFS_SB(inode
->i_sb
);
1612 struct netfs_trans
*t
;
1613 int path_len
, addon_len
= 0;
1615 struct netfs_inode_info
*info
;
1616 struct netfs_cmd
*cmd
;
1618 dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__
, pi
->ino
, cmd_op
, addon
);
1620 path_len
= pohmelfs_path_length(pi
);
1627 addon_len
= strlen(addon
) + 1; /* 0-byte */
1630 if (cmd_op
== NETFS_INODE_INFO
)
1631 sz
+= sizeof(struct netfs_inode_info
);
1633 t
= netfs_trans_alloc(psb
, sz
+ path_len
, flags
, 0);
1638 t
->complete
= complete
;
1641 cmd
= netfs_trans_current(t
);
1642 data
= (void *)(cmd
+ 1);
1644 if (cmd_op
== NETFS_INODE_INFO
) {
1645 info
= (struct netfs_inode_info
*)(cmd
+ 1);
1646 data
= (void *)(info
+ 1);
1649 * We are under i_mutex, can read and change whatever we want...
1651 info
->mode
= inode
->i_mode
;
1652 info
->nlink
= inode
->i_nlink
;
1653 info
->uid
= inode
->i_uid
;
1654 info
->gid
= inode
->i_gid
;
1655 info
->blocks
= inode
->i_blocks
;
1656 info
->rdev
= inode
->i_rdev
;
1657 info
->size
= inode
->i_size
;
1658 info
->version
= inode
->i_version
;
1660 netfs_convert_inode_info(info
);
1663 path_len
= pohmelfs_construct_path_string(pi
, data
, path_len
);
1667 dprintk("%s: path_len: %d.\n", __func__
, path_len
);
1670 path_len
--; /* Do not place null-byte before the addon */
1671 path_len
+= sprintf(data
+ path_len
, "/%s", addon
) + 1; /* 0 - byte */
1677 cmd
->ext
= path_len
;
1682 netfs_convert_cmd(cmd
);
1683 netfs_trans_update(cmd
, t
, sz
);
1686 * Note, that it is possible to leak error here: transaction callback will not
1687 * be invoked for allocation path failure.
1689 return netfs_trans_finish(t
, psb
);
1692 netfs_trans_free(t
);
1695 complete(NULL
, 0, priv
, err
);
1699 int pohmelfs_meta_command(struct pohmelfs_inode
*pi
, unsigned int cmd_op
, unsigned int flags
,
1700 netfs_trans_complete_t complete
, void *priv
, u64 start
)
1702 return pohmelfs_meta_command_data(pi
, pi
->ino
, cmd_op
, NULL
, flags
, complete
, priv
, start
);
1706 * Send request and wait for POHMELFS root capabilities response,
1707 * which will update server's informaion about size of the export,
1708 * permissions, number of objects, available size and so on.
1710 static int pohmelfs_root_handshake(struct pohmelfs_sb
*psb
)
1712 struct netfs_trans
*t
;
1713 struct netfs_cmd
*cmd
;
1716 t
= netfs_trans_alloc(psb
, 0, 0, 0);
1720 cmd
= netfs_trans_current(t
);
1722 cmd
->cmd
= NETFS_CAPABILITIES
;
1723 cmd
->id
= POHMELFS_ROOT_CAPABILITIES
;
1729 netfs_convert_cmd(cmd
);
1730 netfs_trans_update(cmd
, t
, 0);
1732 err
= netfs_trans_finish(t
, psb
);
1737 err
= wait_event_interruptible_timeout(psb
->wait
,
1739 psb
->wait_on_page_timeout
);
1754 static int pohmelfs_show_stats(struct seq_file
*m
, struct vfsmount
*mnt
)
1756 struct netfs_state
*st
;
1757 struct pohmelfs_ctl
*ctl
;
1758 struct pohmelfs_sb
*psb
= POHMELFS_SB(mnt
->mnt_sb
);
1759 struct pohmelfs_config
*c
;
1761 mutex_lock(&psb
->state_lock
);
1763 seq_printf(m
, "\nidx addr(:port) socket_type protocol active priority permissions\n");
1765 list_for_each_entry(c
, &psb
->state_list
, config_entry
) {
1769 seq_printf(m
, "%u ", ctl
->idx
);
1770 if (ctl
->addr
.sa_family
== AF_INET
) {
1771 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&st
->ctl
.addr
;
1772 seq_printf(m
, "%pI4:%u", &sin
->sin_addr
.s_addr
, ntohs(sin
->sin_port
));
1773 } else if (ctl
->addr
.sa_family
== AF_INET6
) {
1774 struct sockaddr_in6
*sin
= (struct sockaddr_in6
*)&st
->ctl
.addr
;
1775 seq_printf(m
, "%pi6:%u", &sin
->sin6_addr
, ntohs(sin
->sin6_port
));
1778 for (i
= 0; i
< ctl
->addrlen
; ++i
)
1779 seq_printf(m
, "%02x.", ctl
->addr
.addr
[i
]);
1782 seq_printf(m
, " %u %u %d %u %x\n",
1783 ctl
->type
, ctl
->proto
,
1785 ctl
->prio
, ctl
->perm
);
1787 mutex_unlock(&psb
->state_lock
);
1792 static const struct super_operations pohmelfs_sb_ops
= {
1793 .alloc_inode
= pohmelfs_alloc_inode
,
1794 .destroy_inode
= pohmelfs_destroy_inode
,
1795 .drop_inode
= pohmelfs_drop_inode
,
1796 .write_inode
= pohmelfs_write_inode
,
1797 .put_super
= pohmelfs_put_super
,
1798 .remount_fs
= pohmelfs_remount
,
1799 .statfs
= pohmelfs_statfs
,
1800 .show_options
= pohmelfs_show_options
,
1801 .show_stats
= pohmelfs_show_stats
,
1805 * Allocate private superblock and create root dir.
1807 static int pohmelfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1809 struct pohmelfs_sb
*psb
;
1812 struct pohmelfs_inode
*npi
;
1815 psb
= kzalloc(sizeof(struct pohmelfs_sb
), GFP_KERNEL
);
1819 err
= bdi_init(&psb
->bdi
);
1821 goto err_out_free_sb
;
1823 err
= bdi_register(&psb
->bdi
, NULL
, "pfs-%d", atomic_inc_return(&psb_bdi_num
));
1825 bdi_destroy(&psb
->bdi
);
1826 goto err_out_free_sb
;
1829 sb
->s_fs_info
= psb
;
1830 sb
->s_op
= &pohmelfs_sb_ops
;
1831 sb
->s_magic
= POHMELFS_MAGIC_NUM
;
1832 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1833 sb
->s_blocksize
= PAGE_SIZE
;
1834 sb
->s_bdi
= &psb
->bdi
;
1840 psb
->active_state
= NULL
;
1841 psb
->trans_retries
= 5;
1842 psb
->trans_data_size
= PAGE_SIZE
;
1843 psb
->drop_scan_timeout
= msecs_to_jiffies(1000);
1844 psb
->trans_scan_timeout
= msecs_to_jiffies(5000);
1845 psb
->wait_on_page_timeout
= msecs_to_jiffies(5000);
1846 init_waitqueue_head(&psb
->wait
);
1848 spin_lock_init(&psb
->ino_lock
);
1850 INIT_LIST_HEAD(&psb
->drop_list
);
1852 mutex_init(&psb
->mcache_lock
);
1853 psb
->mcache_root
= RB_ROOT
;
1854 psb
->mcache_timeout
= msecs_to_jiffies(5000);
1855 atomic_long_set(&psb
->mcache_gen
, 0);
1857 psb
->trans_max_pages
= 100;
1859 psb
->crypto_align_size
= 16;
1860 psb
->crypto_attached_size
= 0;
1861 psb
->hash_strlen
= 0;
1862 psb
->cipher_strlen
= 0;
1863 psb
->perform_crypto
= 0;
1864 psb
->crypto_thread_num
= 2;
1865 psb
->crypto_fail_unsupported
= 0;
1866 mutex_init(&psb
->crypto_thread_lock
);
1867 INIT_LIST_HEAD(&psb
->crypto_ready_list
);
1868 INIT_LIST_HEAD(&psb
->crypto_active_list
);
1870 atomic_set(&psb
->trans_gen
, 1);
1871 atomic_long_set(&psb
->total_inodes
, 0);
1873 mutex_init(&psb
->state_lock
);
1874 INIT_LIST_HEAD(&psb
->state_list
);
1876 err
= pohmelfs_parse_options((char *) data
, psb
, 0);
1878 goto err_out_free_bdi
;
1880 err
= pohmelfs_copy_crypto(psb
);
1882 goto err_out_free_bdi
;
1884 err
= pohmelfs_state_init(psb
);
1886 goto err_out_free_strings
;
1888 err
= pohmelfs_crypto_init(psb
);
1890 goto err_out_state_exit
;
1892 err
= pohmelfs_root_handshake(psb
);
1894 goto err_out_crypto_exit
;
1897 str
.hash
= jhash("/", 1, 0);
1900 npi
= pohmelfs_create_entry_local(psb
, NULL
, &str
, 0, 0755|S_IFDIR
);
1903 goto err_out_crypto_exit
;
1905 set_bit(NETFS_INODE_REMOTE_SYNCED
, &npi
->state
);
1906 clear_bit(NETFS_INODE_OWNED
, &npi
->state
);
1908 root
= &npi
->vfs_inode
;
1910 sb
->s_root
= d_alloc_root(root
);
1912 goto err_out_put_root
;
1914 INIT_DELAYED_WORK(&psb
->drop_dwork
, pohmelfs_drop_scan
);
1915 schedule_delayed_work(&psb
->drop_dwork
, psb
->drop_scan_timeout
);
1917 INIT_DELAYED_WORK(&psb
->dwork
, pohmelfs_trans_scan
);
1918 schedule_delayed_work(&psb
->dwork
, psb
->trans_scan_timeout
);
1924 err_out_crypto_exit
:
1925 pohmelfs_crypto_exit(psb
);
1927 pohmelfs_state_exit(psb
);
1928 err_out_free_strings
:
1929 kfree(psb
->cipher_string
);
1930 kfree(psb
->hash_string
);
1932 bdi_destroy(&psb
->bdi
);
1937 dprintk("%s: err: %d.\n", __func__
, err
);
1942 * Some VFS magic here...
1944 static int pohmelfs_get_sb(struct file_system_type
*fs_type
,
1945 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
1947 return get_sb_nodev(fs_type
, flags
, data
, pohmelfs_fill_super
,
1952 * We need this to sync all inodes earlier, since when writeback
1953 * is invoked from the umount/mntput path dcache is already shrunk,
1954 * see generic_shutdown_super(), and no inodes can access the path.
1956 static void pohmelfs_kill_super(struct super_block
*sb
)
1959 kill_anon_super(sb
);
1962 static struct file_system_type pohmel_fs_type
= {
1963 .owner
= THIS_MODULE
,
1965 .get_sb
= pohmelfs_get_sb
,
1966 .kill_sb
= pohmelfs_kill_super
,
1970 * Cache and module initializations and freeing routings.
1972 static void pohmelfs_init_once(void *data
)
1974 struct pohmelfs_inode
*pi
= data
;
1976 inode_init_once(&pi
->vfs_inode
);
1979 static int __init
pohmelfs_init_inodecache(void)
1981 pohmelfs_inode_cache
= kmem_cache_create("pohmelfs_inode_cache",
1982 sizeof(struct pohmelfs_inode
),
1983 0, (SLAB_RECLAIM_ACCOUNT
|SLAB_MEM_SPREAD
),
1984 pohmelfs_init_once
);
1985 if (!pohmelfs_inode_cache
)
1991 static void pohmelfs_destroy_inodecache(void)
1993 kmem_cache_destroy(pohmelfs_inode_cache
);
1996 static int __init
init_pohmel_fs(void)
2000 err
= pohmelfs_config_init();
2004 err
= pohmelfs_init_inodecache();
2006 goto err_out_config_exit
;
2008 err
= pohmelfs_mcache_init();
2010 goto err_out_destroy
;
2012 err
= netfs_trans_init();
2014 goto err_out_mcache_exit
;
2016 err
= register_filesystem(&pohmel_fs_type
);
2024 err_out_mcache_exit
:
2025 pohmelfs_mcache_exit();
2027 pohmelfs_destroy_inodecache();
2028 err_out_config_exit
:
2029 pohmelfs_config_exit();
2034 static void __exit
exit_pohmel_fs(void)
2036 unregister_filesystem(&pohmel_fs_type
);
2037 pohmelfs_destroy_inodecache();
2038 pohmelfs_mcache_exit();
2039 pohmelfs_config_exit();
2043 module_init(init_pohmel_fs
);
2044 module_exit(exit_pohmel_fs
);
2046 MODULE_LICENSE("GPL");
2047 MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
2048 MODULE_DESCRIPTION("Pohmel filesystem");