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;
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);
146 struct backing_dev_info *bdi = mapping->backing_dev_info;
151 pgoff_t end; /* Inclusive */
155 if (wbc->nonblocking && bdi_write_congested(bdi)) {
156 wbc->encountered_congestion = 1;
160 if (wbc->range_cyclic) {
161 index = mapping->writeback_index; /* Start from prev offset */
164 index = wbc->range_start >> PAGE_CACHE_SHIFT;
165 end = wbc->range_end >> PAGE_CACHE_SHIFT;
166 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
171 while (!done && (index <= end)) {
172 unsigned int i = min(end - index, (pgoff_t)psb->trans_max_pages);
174 struct netfs_trans *trans;
176 err = pohmelfs_inode_has_dirty_pages(mapping, index);
180 err = pohmelfs_path_length(pi);
191 trans = netfs_trans_alloc(psb, path_len, 0, i);
196 trans->complete = &pohmelfs_write_trans_complete;
198 trans->page_num = nr_pages = find_get_pages_tag(mapping, &index,
199 PAGECACHE_TAG_DIRTY, trans->page_num,
202 dprintk("%s: t: %p, nr_pages: %u, end: %lu, index: %lu, max: %u.\n",
203 __func__, trans, nr_pages, end, index, trans->page_num);
208 err = pohmelfs_write_inode_create(inode, trans);
215 for (i = 0; i < trans->page_num; i++) {
216 struct page *page = trans->pages[i];
220 if (unlikely(page->mapping != mapping))
223 if (!wbc->range_cyclic && page->index > end) {
228 if (wbc->sync_mode != WB_SYNC_NONE)
229 wait_on_page_writeback(page);
231 if (PageWriteback(page) ||
232 !clear_page_dirty_for_io(page)) {
233 dprintk("%s: not clear for io page: %p, writeback: %d.\n",
234 __func__, page, PageWriteback(page));
238 set_page_writeback(page);
240 trans->attached_size += page_private(page);
241 trans->attached_pages++;
243 dprintk("%s: %u/%u added trans: %p, gen: %u, page: %p, [High: %d], size: %lu, idx: %lu.\n",
244 __func__, i, trans->page_num, trans, trans->gen, page,
245 !!PageHighMem(page), page_private(page), page->index);
249 if (wbc->nr_to_write <= 0)
251 if (wbc->nonblocking && bdi_write_congested(bdi)) {
252 wbc->encountered_congestion = 1;
259 trans->pages[i] = NULL;
262 err = netfs_trans_finish(trans, psb);
270 netfs_trans_reset(trans);
271 netfs_trans_put(trans);
275 if (!scanned && !done) {
277 * We hit the last page and there is more work to be done: wrap
278 * back to the start of the file
285 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
286 mapping->writeback_index = index;
292 * Inode writeback creation completion callback.
293 * Only invoked for just created inodes, which do not have pages attached,
294 * like dirs and empty files.
296 static int pohmelfs_write_inode_complete(struct page **pages, unsigned int page_num,
297 void *private, int err)
299 struct inode *inode = private;
300 struct pohmelfs_inode *pi = POHMELFS_I(inode);
304 mark_inode_dirty(inode);
305 clear_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
307 set_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
310 pohmelfs_put_inode(pi);
316 int pohmelfs_write_create_inode(struct pohmelfs_inode *pi)
318 struct netfs_trans *t;
319 struct inode *inode = &pi->vfs_inode;
320 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
323 if (test_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state))
326 dprintk("%s: started ino: %llu.\n", __func__, pi->ino);
328 err = pohmelfs_path_length(pi);
332 t = netfs_trans_alloc(psb, err + 1, 0, 0);
337 t->complete = pohmelfs_write_inode_complete;
338 t->private = igrab(inode);
344 err = pohmelfs_write_inode_create(inode, t);
348 netfs_trans_finish(t, POHMELFS_SB(inode->i_sb));
360 * Sync all not-yet-created children in given directory to the server.
362 static int pohmelfs_write_inode_create_children(struct inode *inode)
364 struct pohmelfs_inode *parent = POHMELFS_I(inode);
365 struct super_block *sb = inode->i_sb;
366 struct pohmelfs_name *n;
368 while (!list_empty(&parent->sync_create_list)) {
370 mutex_lock(&parent->offset_lock);
371 if (!list_empty(&parent->sync_create_list)) {
372 n = list_first_entry(&parent->sync_create_list,
373 struct pohmelfs_name, sync_create_entry);
374 list_del_init(&n->sync_create_entry);
376 mutex_unlock(&parent->offset_lock);
381 inode = ilookup(sb, n->ino);
383 dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n",
384 __func__, parent->ino, n->ino, inode);
386 if (inode && (inode->i_state & I_DIRTY)) {
387 struct pohmelfs_inode *pi = POHMELFS_I(inode);
388 pohmelfs_write_create_inode(pi);
389 //pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0);
398 * Removes given child from given inode on server.
400 int pohmelfs_remove_child(struct pohmelfs_inode *pi, struct pohmelfs_name *n)
402 return pohmelfs_meta_command_data(pi, pi->ino, NETFS_REMOVE, NULL, 0, NULL, NULL, 0);
406 * Writeback for given inode.
408 static int pohmelfs_write_inode(struct inode *inode, int sync)
410 struct pohmelfs_inode *pi = POHMELFS_I(inode);
412 pohmelfs_write_create_inode(pi);
413 pohmelfs_write_inode_create_children(inode);
419 * It is not exported, sorry...
421 static inline wait_queue_head_t *page_waitqueue(struct page *page)
423 const struct zone *zone = page_zone(page);
425 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
428 static int pohmelfs_wait_on_page_locked(struct page *page)
430 struct pohmelfs_sb *psb = POHMELFS_SB(page->mapping->host->i_sb);
431 long ret = psb->wait_on_page_timeout;
432 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
435 if (!PageLocked(page))
439 prepare_to_wait(page_waitqueue(page),
440 &wait.wait, TASK_INTERRUPTIBLE);
442 dprintk("%s: page: %p, locked: %d, uptodate: %d, error: %d, flags: %lx.\n",
443 __func__, page, PageLocked(page), PageUptodate(page),
444 PageError(page), page->flags);
446 if (!PageLocked(page))
449 if (!signal_pending(current)) {
450 ret = schedule_timeout(ret);
458 finish_wait(page_waitqueue(page), &wait.wait);
465 SetPageUptodate(page);
468 printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n",
469 __func__, page, PageUptodate(page), PageLocked(page), err);
474 static int pohmelfs_read_page_complete(struct page **pages, unsigned int page_num,
475 void *private, int err)
477 struct page *page = private;
479 if (PageChecked(page))
483 dprintk("%s: page: %p, err: %d.\n", __func__, page, err);
493 * Read a page from remote server.
494 * Function will wait until page is unlocked.
496 static int pohmelfs_readpage(struct file *file, struct page *page)
498 struct inode *inode = page->mapping->host;
499 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
500 struct pohmelfs_inode *pi = POHMELFS_I(inode);
501 struct netfs_trans *t;
502 struct netfs_cmd *cmd;
507 err = pohmelfs_data_lock(pi, page->index << PAGE_CACHE_SHIFT,
508 PAGE_SIZE, POHMELFS_READ_LOCK);
512 isize = i_size_read(inode);
513 if (isize <= page->index << PAGE_CACHE_SHIFT) {
514 SetPageUptodate(page);
519 path_len = pohmelfs_path_length(pi);
525 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
531 t->complete = pohmelfs_read_page_complete;
534 cmd = netfs_trans_current(t);
535 data = (void *)(cmd + 1);
537 err = pohmelfs_construct_path_string(pi, data, path_len);
544 cmd->start = page->index;
545 cmd->start <<= PAGE_CACHE_SHIFT;
546 cmd->size = PAGE_CACHE_SIZE + path_len;
547 cmd->cmd = NETFS_READ_PAGE;
550 dprintk("%s: path: '%s', page: %p, ino: %llu, start: %llu, size: %lu.\n",
551 __func__, (char *)data, page, pi->ino, cmd->start, PAGE_CACHE_SIZE);
553 netfs_convert_cmd(cmd);
554 netfs_trans_update(cmd, t, path_len);
556 err = netfs_trans_finish(t, psb);
560 return pohmelfs_wait_on_page_locked(page);
567 if (PageLocked(page))
570 printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n",
571 __func__, page, page->index << PAGE_CACHE_SHIFT, PAGE_CACHE_SIZE, err);
577 * Write begin/end magic.
578 * Allocates a page and writes inode if it was not synced to server before.
580 static int pohmelfs_write_begin(struct file *file, struct address_space *mapping,
581 loff_t pos, unsigned len, unsigned flags,
582 struct page **pagep, void **fsdata)
584 struct inode *inode = mapping->host;
592 index = pos >> PAGE_CACHE_SHIFT;
593 start = pos & (PAGE_CACHE_SIZE - 1);
596 page = grab_cache_page(mapping, index);
598 dprintk("%s: page: %p pos: %llu, len: %u, index: %lu, start: %u, end: %u, uptodate: %d.\n",
599 __func__, page, pos, len, index, start, end, PageUptodate(page));
606 while (!PageUptodate(page)) {
607 if (start && test_bit(NETFS_INODE_REMOTE_SYNCED, &POHMELFS_I(inode)->state)) {
608 err = pohmelfs_readpage(file, page);
616 if (len != PAGE_CACHE_SIZE) {
617 void *kaddr = kmap_atomic(page, KM_USER0);
619 memset(kaddr + start, 0, PAGE_CACHE_SIZE - start);
620 flush_dcache_page(page);
621 kunmap_atomic(kaddr, KM_USER0);
623 SetPageUptodate(page);
626 set_page_private(page, end);
633 page_cache_release(page);
639 static int pohmelfs_write_end(struct file *file, struct address_space *mapping,
640 loff_t pos, unsigned len, unsigned copied,
641 struct page *page, void *fsdata)
643 struct inode *inode = mapping->host;
646 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
647 void *kaddr = kmap_atomic(page, KM_USER0);
649 memset(kaddr + from + copied, 0, len - copied);
650 flush_dcache_page(page);
651 kunmap_atomic(kaddr, KM_USER0);
654 SetPageUptodate(page);
655 set_page_dirty(page);
657 dprintk("%s: page: %p [U: %d, D: %d, L: %d], pos: %llu, len: %u, copied: %u.\n",
659 PageUptodate(page), PageDirty(page), PageLocked(page),
662 flush_dcache_page(page);
665 page_cache_release(page);
667 if (pos + copied > inode->i_size) {
668 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
670 psb->avail_size -= pos + copied - inode->i_size;
672 i_size_write(inode, pos + copied);
678 static int pohmelfs_readpages_trans_complete(struct page **__pages, unsigned int page_num,
679 void *private, int err)
681 struct pohmelfs_inode *pi = private;
683 struct page **pages, *page = (struct page *)__pages;
684 loff_t index = page->index;
686 pages = kzalloc(sizeof(void *) * page_num, GFP_NOIO);
690 num = find_get_pages_contig(pi->vfs_inode.i_mapping, index, page_num, pages);
696 for (i=0; i<num; ++i) {
700 printk("%s: %u/%u: page: %p, index: %lu, uptodate: %d, locked: %d, err: %d.\n",
701 __func__, i, num, page, page->index,
702 PageUptodate(page), PageLocked(page), err);
704 if (!PageChecked(page)) {
709 page_cache_release(page);
710 page_cache_release(page);
718 static int pohmelfs_send_readpages(struct pohmelfs_inode *pi, struct page *first, unsigned int num)
720 struct netfs_trans *t;
721 struct netfs_cmd *cmd;
722 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
726 err = pohmelfs_data_lock(pi, first->index << PAGE_CACHE_SHIFT,
727 num * PAGE_SIZE, POHMELFS_READ_LOCK);
731 path_len = pohmelfs_path_length(pi);
737 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
743 cmd = netfs_trans_current(t);
744 data = (void *)(cmd + 1);
746 t->complete = pohmelfs_readpages_trans_complete;
749 t->pages = (struct page **)first;
751 err = pohmelfs_construct_path_string(pi, data, path_len);
757 cmd->cmd = NETFS_READ_PAGES;
758 cmd->start = first->index;
759 cmd->start <<= PAGE_CACHE_SHIFT;
760 cmd->size = (num << 8 | PAGE_CACHE_SHIFT);
764 dprintk("%s: t: %p, gen: %u, path: '%s', path_len: %u, "
765 "start: %lu, num: %u.\n",
766 __func__, t, t->gen, (char *)data, path_len,
769 netfs_convert_cmd(cmd);
770 netfs_trans_update(cmd, t, path_len);
772 return netfs_trans_finish(t, psb);
777 pohmelfs_readpages_trans_complete((struct page **)first, num, pi, err);
781 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
783 static int pohmelfs_readpages(struct file *file, struct address_space *mapping,
784 struct list_head *pages, unsigned nr_pages)
786 unsigned int page_idx, num = 0;
787 struct page *page = NULL, *first = NULL;
789 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
790 page = list_to_page(pages);
792 prefetchw(&page->flags);
793 list_del(&page->lru);
795 if (!add_to_page_cache_lru(page, mapping,
796 page->index, GFP_KERNEL)) {
804 dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n",
805 __func__, page, page->index, first->index);
807 if (unlikely(first->index + num != page->index) || (num > 500)) {
808 pohmelfs_send_readpages(POHMELFS_I(mapping->host),
817 pohmelfs_send_readpages(POHMELFS_I(mapping->host), first, num);
820 * This will be sync read, so when last page is processed,
821 * all previous are alerady unlocked and ready to be used.
827 * Small addres space operations for POHMELFS.
829 const struct address_space_operations pohmelfs_aops = {
830 .readpage = pohmelfs_readpage,
831 .readpages = pohmelfs_readpages,
832 .writepages = pohmelfs_writepages,
833 .write_begin = pohmelfs_write_begin,
834 .write_end = pohmelfs_write_end,
835 .set_page_dirty = __set_page_dirty_nobuffers,
839 * ->detroy_inode() callback. Deletes inode from the caches
840 * and frees private data.
842 static void pohmelfs_destroy_inode(struct inode *inode)
844 struct super_block *sb = inode->i_sb;
845 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
846 struct pohmelfs_inode *pi = POHMELFS_I(inode);
848 //pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK);
850 pohmelfs_inode_del_inode(psb, pi);
852 dprintk("%s: pi: %p, inode: %p, ino: %llu.\n",
853 __func__, pi, &pi->vfs_inode, pi->ino);
854 kmem_cache_free(pohmelfs_inode_cache, pi);
855 atomic_long_dec(&psb->total_inodes);
859 * ->alloc_inode() callback. Allocates inode and initilizes private data.
861 static struct inode *pohmelfs_alloc_inode(struct super_block *sb)
863 struct pohmelfs_inode *pi;
865 pi = kmem_cache_alloc(pohmelfs_inode_cache, GFP_NOIO);
869 pi->hash_root = RB_ROOT;
870 mutex_init(&pi->offset_lock);
872 INIT_LIST_HEAD(&pi->sync_create_list);
874 INIT_LIST_HEAD(&pi->inode_entry);
881 dprintk("%s: pi: %p, inode: %p.\n", __func__, pi, &pi->vfs_inode);
883 atomic_long_inc(&POHMELFS_SB(sb)->total_inodes);
885 return &pi->vfs_inode;
889 * We want fsync() to work on POHMELFS.
891 static int pohmelfs_fsync(struct file *file, struct dentry *dentry, int datasync)
893 struct inode *inode = file->f_mapping->host;
894 struct writeback_control wbc = {
895 .sync_mode = WB_SYNC_ALL,
896 .nr_to_write = 0, /* sys_fsync did this */
899 return sync_inode(inode, &wbc);
902 ssize_t pohmelfs_write(struct file *file, const char __user *buf,
903 size_t len, loff_t *ppos)
905 struct address_space *mapping = file->f_mapping;
906 struct inode *inode = mapping->host;
907 struct pohmelfs_inode *pi = POHMELFS_I(inode);
908 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
913 init_sync_kiocb(&kiocb, file);
917 dprintk("%s: len: %zu, pos: %llu.\n", __func__, len, pos);
919 mutex_lock(&inode->i_mutex);
920 ret = pohmelfs_data_lock(pi, pos, len, POHMELFS_WRITE_LOCK);
924 ret = generic_file_aio_write_nolock(&kiocb, &iov, 1, pos);
925 *ppos = kiocb.ki_pos;
927 mutex_unlock(&inode->i_mutex);
930 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
933 err = sync_page_range(inode, mapping, pos, ret);
942 mutex_unlock(&inode->i_mutex);
946 const static struct file_operations pohmelfs_file_ops = {
947 .open = generic_file_open,
948 .fsync = pohmelfs_fsync,
950 .llseek = generic_file_llseek,
952 .read = do_sync_read,
953 .aio_read = generic_file_aio_read,
955 .mmap = generic_file_mmap,
957 .splice_read = generic_file_splice_read,
958 .splice_write = generic_file_splice_write,
960 .write = pohmelfs_write,
961 .aio_write = generic_file_aio_write,
964 const struct inode_operations pohmelfs_symlink_inode_operations = {
965 .readlink = generic_readlink,
966 .follow_link = page_follow_link_light,
967 .put_link = page_put_link,
970 int pohmelfs_setattr_raw(struct inode *inode, struct iattr *attr)
974 err = inode_change_ok(inode, attr);
976 dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__, POHMELFS_I(inode)->ino);
980 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
981 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
982 err = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
987 err = inode_setattr(inode, attr);
989 dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__, POHMELFS_I(inode)->ino);
993 dprintk("%s: ino: %llu, mode: %o -> %o, uid: %u -> %u, gid: %u -> %u, size: %llu -> %llu.\n",
994 __func__, POHMELFS_I(inode)->ino, inode->i_mode, attr->ia_mode,
995 inode->i_uid, attr->ia_uid, inode->i_gid, attr->ia_gid, inode->i_size, attr->ia_size);
1003 int pohmelfs_setattr(struct dentry *dentry, struct iattr *attr)
1005 struct inode *inode = dentry->d_inode;
1006 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1009 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1013 err = security_inode_setattr(dentry, attr);
1017 err = pohmelfs_setattr_raw(inode, attr);
1027 static int pohmelfs_send_xattr_req(struct pohmelfs_inode *pi, u64 id, u64 start,
1028 const char *name, const void *value, size_t attrsize, int command)
1030 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
1031 int err, path_len, namelen = strlen(name) + 1; /* 0-byte */
1032 struct netfs_trans *t;
1033 struct netfs_cmd *cmd;
1036 dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n",
1037 __func__, id, start, name, attrsize, command);
1039 path_len = pohmelfs_path_length(pi);
1045 t = netfs_trans_alloc(psb, namelen + path_len + attrsize, 0, 0);
1051 cmd = netfs_trans_current(t);
1054 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1062 * 'name' is a NUL-terminated string already and
1063 * 'namelen' includes 0-byte.
1065 memcpy(data, name, namelen);
1068 memcpy(data, value, attrsize);
1073 cmd->size = attrsize + namelen + path_len;
1074 cmd->ext = path_len;
1078 netfs_convert_cmd(cmd);
1079 netfs_trans_update(cmd, t, namelen + path_len + attrsize);
1081 return netfs_trans_finish(t, psb);
1090 static int pohmelfs_setxattr(struct dentry *dentry, const char *name,
1091 const void *value, size_t attrsize, int flags)
1093 struct inode *inode = dentry->d_inode;
1094 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1095 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1097 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1100 return pohmelfs_send_xattr_req(pi, flags, attrsize, name,
1101 value, attrsize, NETFS_XATTR_SET);
1104 static ssize_t pohmelfs_getxattr(struct dentry *dentry, const char *name,
1105 void *value, size_t attrsize)
1107 struct inode *inode = dentry->d_inode;
1108 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1109 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1110 struct pohmelfs_mcache *m;
1112 long timeout = psb->mcache_timeout;
1114 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1117 m = pohmelfs_mcache_alloc(psb, 0, attrsize, value);
1121 dprintk("%s: ino: %llu, name: '%s', size: %zu.\n",
1122 __func__, pi->ino, name, attrsize);
1124 err = pohmelfs_send_xattr_req(pi, m->gen, attrsize, name, value, 0, NETFS_XATTR_GET);
1129 err = wait_for_completion_timeout(&m->complete, timeout);
1136 * This loop is a bit ugly, since it waits until reference counter
1137 * hits 1 and then put object here. Main goal is to prevent race with
1138 * network thread, when it can start processing given request, i.e.
1139 * increase its reference counter but yet not complete it, while
1140 * we will exit from ->getxattr() with timeout, and although request
1141 * will not be freed (its reference counter was increased by network
1142 * thread), data pointer provided by user may be released, so we will
1143 * overwrite already freed area in network thread.
1145 * Now after timeout we remove request from the cache, so it can not be
1146 * found by network thread, and wait for its reference counter to hit 1,
1147 * i.e. if network thread already started to process this request, we wait
1148 * it to finish, and then free object locally. If reference counter is
1149 * already 1, i.e. request is not used by anyone else, we can free it without
1155 pohmelfs_mcache_remove_locked(psb, m);
1156 } while (atomic_read(&m->refcnt) != 1);
1158 pohmelfs_mcache_put(psb, m);
1160 dprintk("%s: ino: %llu, err: %d.\n", __func__, pi->ino, err);
1165 pohmelfs_mcache_put(psb, m);
1169 static int pohmelfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1171 struct inode *inode = dentry->d_inode;
1173 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1176 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_READ_LOCK);
1179 dprintk("%s: ino: %llu, mode: %o, uid: %u, gid: %u, size: %llu.\n",
1180 __func__, pi->ino, inode->i_mode, inode->i_uid,
1181 inode->i_gid, inode->i_size);
1184 generic_fillattr(inode, stat);
1188 const struct inode_operations pohmelfs_file_inode_operations = {
1189 .setattr = pohmelfs_setattr,
1190 .getattr = pohmelfs_getattr,
1191 .setxattr = pohmelfs_setxattr,
1192 .getxattr = pohmelfs_getxattr,
1196 * Fill inode data: mode, size, operation callbacks and so on...
1198 void pohmelfs_fill_inode(struct inode *inode, struct netfs_inode_info *info)
1200 inode->i_mode = info->mode;
1201 inode->i_nlink = info->nlink;
1202 inode->i_uid = info->uid;
1203 inode->i_gid = info->gid;
1204 inode->i_blocks = info->blocks;
1205 inode->i_rdev = info->rdev;
1206 inode->i_size = info->size;
1207 inode->i_version = info->version;
1208 inode->i_blkbits = ffs(info->blocksize);
1210 dprintk("%s: inode: %p, num: %lu/%llu inode is regular: %d, dir: %d, link: %d, mode: %o, size: %llu.\n",
1211 __func__, inode, inode->i_ino, info->ino,
1212 S_ISREG(inode->i_mode), S_ISDIR(inode->i_mode),
1213 S_ISLNK(inode->i_mode), inode->i_mode, inode->i_size);
1215 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1218 * i_mapping is a pointer to i_data during inode initialization.
1220 inode->i_data.a_ops = &pohmelfs_aops;
1222 if (S_ISREG(inode->i_mode)) {
1223 inode->i_fop = &pohmelfs_file_ops;
1224 inode->i_op = &pohmelfs_file_inode_operations;
1225 } else if (S_ISDIR(inode->i_mode)) {
1226 inode->i_fop = &pohmelfs_dir_fops;
1227 inode->i_op = &pohmelfs_dir_inode_ops;
1228 } else if (S_ISLNK(inode->i_mode)) {
1229 inode->i_op = &pohmelfs_symlink_inode_operations;
1230 inode->i_fop = &pohmelfs_file_ops;
1232 inode->i_fop = &generic_ro_fops;
1236 static void pohmelfs_drop_inode(struct inode *inode)
1238 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1239 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1241 spin_lock(&psb->ino_lock);
1242 list_del_init(&pi->inode_entry);
1243 spin_unlock(&psb->ino_lock);
1245 generic_drop_inode(inode);
1248 static struct pohmelfs_inode *pohmelfs_get_inode_from_list(struct pohmelfs_sb *psb,
1249 struct list_head *head, unsigned int *count)
1251 struct pohmelfs_inode *pi = NULL;
1253 spin_lock(&psb->ino_lock);
1254 if (!list_empty(head)) {
1255 pi = list_entry(head->next, struct pohmelfs_inode,
1257 list_del_init(&pi->inode_entry);
1258 *count = pi->drop_count;
1261 spin_unlock(&psb->ino_lock);
1266 static void pohmelfs_flush_transactions(struct pohmelfs_sb *psb)
1268 struct pohmelfs_config *c;
1270 mutex_lock(&psb->state_lock);
1271 list_for_each_entry(c, &psb->state_list, config_entry) {
1272 pohmelfs_state_flush_transactions(&c->state);
1274 mutex_unlock(&psb->state_lock);
1278 * ->put_super() callback. Invoked before superblock is destroyed,
1279 * so it has to clean all private data.
1281 static void pohmelfs_put_super(struct super_block *sb)
1283 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1284 struct pohmelfs_inode *pi;
1286 unsigned int in_drop_list = 0;
1287 struct inode *inode, *tmp;
1289 dprintk("%s.\n", __func__);
1292 * Kill pending transactions, which could affect inodes in-flight.
1294 pohmelfs_flush_transactions(psb);
1296 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) {
1297 inode = &pi->vfs_inode;
1299 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1300 __func__, pi->ino, pi, inode, count);
1302 if (atomic_read(&inode->i_count) != count) {
1303 printk("%s: ino: %llu, pi: %p, inode: %p, count: %u, i_count: %d.\n",
1304 __func__, pi->ino, pi, inode, count,
1305 atomic_read(&inode->i_count));
1306 count = atomic_read(&inode->i_count);
1311 iput(&pi->vfs_inode);
1314 list_for_each_entry_safe(inode, tmp, &sb->s_inodes, i_sb_list) {
1315 pi = POHMELFS_I(inode);
1317 dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n",
1318 __func__, pi->ino, pi, inode, atomic_read(&inode->i_count));
1321 * These are special inodes, they were created during
1322 * directory reading or lookup, and were not bound to dentry,
1323 * so they live here with reference counter being 1 and prevent
1324 * umount from succeed since it believes that they are busy.
1326 count = atomic_read(&inode->i_count);
1328 list_del_init(&inode->i_sb_list);
1330 iput(&pi->vfs_inode);
1334 psb->trans_scan_timeout = psb->drop_scan_timeout = 0;
1335 cancel_rearming_delayed_work(&psb->dwork);
1336 cancel_rearming_delayed_work(&psb->drop_dwork);
1337 flush_scheduled_work();
1339 dprintk("%s: stopped workqueues.\n", __func__);
1341 pohmelfs_crypto_exit(psb);
1342 pohmelfs_state_exit(psb);
1345 sb->s_fs_info = NULL;
1348 static int pohmelfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1350 struct super_block *sb = dentry->d_sb;
1351 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1354 * There are no filesystem size limits yet.
1356 memset(buf, 0, sizeof(struct kstatfs));
1358 buf->f_type = POHMELFS_MAGIC_NUM; /* 'POH.' */
1359 buf->f_bsize = sb->s_blocksize;
1360 buf->f_files = psb->ino;
1361 buf->f_namelen = 255;
1362 buf->f_files = atomic_long_read(&psb->total_inodes);
1363 buf->f_bfree = buf->f_bavail = psb->avail_size >> PAGE_SHIFT;
1364 buf->f_blocks = psb->total_size >> PAGE_SHIFT;
1366 dprintk("%s: total: %llu, avail: %llu, inodes: %llu, bsize: %lu.\n",
1367 __func__, psb->total_size, psb->avail_size, buf->f_files, sb->s_blocksize);
1372 static int pohmelfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
1374 struct pohmelfs_sb *psb = POHMELFS_SB(vfs->mnt_sb);
1376 seq_printf(seq, ",idx=%u", psb->idx);
1377 seq_printf(seq, ",trans_scan_timeout=%u", jiffies_to_msecs(psb->trans_scan_timeout));
1378 seq_printf(seq, ",drop_scan_timeout=%u", jiffies_to_msecs(psb->drop_scan_timeout));
1379 seq_printf(seq, ",wait_on_page_timeout=%u", jiffies_to_msecs(psb->wait_on_page_timeout));
1380 seq_printf(seq, ",trans_retries=%u", psb->trans_retries);
1381 seq_printf(seq, ",crypto_thread_num=%u", psb->crypto_thread_num);
1382 seq_printf(seq, ",trans_max_pages=%u", psb->trans_max_pages);
1383 seq_printf(seq, ",mcache_timeout=%u", jiffies_to_msecs(psb->mcache_timeout));
1384 if (psb->crypto_fail_unsupported)
1385 seq_printf(seq, ",crypto_fail_unsupported");
1392 pohmelfs_opt_crypto_thread_num,
1393 pohmelfs_opt_trans_max_pages,
1394 pohmelfs_opt_crypto_fail_unsupported,
1396 /* Remountable options */
1397 pohmelfs_opt_trans_scan_timeout,
1398 pohmelfs_opt_drop_scan_timeout,
1399 pohmelfs_opt_wait_on_page_timeout,
1400 pohmelfs_opt_trans_retries,
1401 pohmelfs_opt_mcache_timeout,
1404 static struct match_token pohmelfs_tokens[] = {
1405 {pohmelfs_opt_idx, "idx=%u"},
1406 {pohmelfs_opt_crypto_thread_num, "crypto_thread_num=%u"},
1407 {pohmelfs_opt_trans_max_pages, "trans_max_pages=%u"},
1408 {pohmelfs_opt_crypto_fail_unsupported, "crypto_fail_unsupported"},
1409 {pohmelfs_opt_trans_scan_timeout, "trans_scan_timeout=%u"},
1410 {pohmelfs_opt_drop_scan_timeout, "drop_scan_timeout=%u"},
1411 {pohmelfs_opt_wait_on_page_timeout, "wait_on_page_timeout=%u"},
1412 {pohmelfs_opt_trans_retries, "trans_retries=%u"},
1413 {pohmelfs_opt_mcache_timeout, "mcache_timeout=%u"},
1416 static int pohmelfs_parse_options(char *options, struct pohmelfs_sb *psb, int remount)
1419 substring_t args[MAX_OPT_ARGS];
1425 while ((p = strsep(&options, ",")) != NULL) {
1430 token = match_token(p, pohmelfs_tokens, args);
1432 err = match_int(&args[0], &option);
1436 if (remount && token <= pohmelfs_opt_crypto_fail_unsupported)
1440 case pohmelfs_opt_idx:
1443 case pohmelfs_opt_trans_scan_timeout:
1444 psb->trans_scan_timeout = msecs_to_jiffies(option);
1446 case pohmelfs_opt_drop_scan_timeout:
1447 psb->drop_scan_timeout = msecs_to_jiffies(option);
1449 case pohmelfs_opt_wait_on_page_timeout:
1450 psb->wait_on_page_timeout = msecs_to_jiffies(option);
1452 case pohmelfs_opt_mcache_timeout:
1453 psb->mcache_timeout = msecs_to_jiffies(option);
1455 case pohmelfs_opt_trans_retries:
1456 psb->trans_retries = option;
1458 case pohmelfs_opt_crypto_thread_num:
1459 psb->crypto_thread_num = option;
1461 case pohmelfs_opt_trans_max_pages:
1462 psb->trans_max_pages = option;
1464 case pohmelfs_opt_crypto_fail_unsupported:
1465 psb->crypto_fail_unsupported = 1;
1475 static int pohmelfs_remount(struct super_block *sb, int *flags, char *data)
1478 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1479 unsigned long old_sb_flags = sb->s_flags;
1481 err = pohmelfs_parse_options(data, psb, 1);
1483 goto err_out_restore;
1485 if (!(*flags & MS_RDONLY))
1486 sb->s_flags &= ~MS_RDONLY;
1490 sb->s_flags = old_sb_flags;
1494 static void pohmelfs_flush_inode(struct pohmelfs_inode *pi, unsigned int count)
1496 struct inode *inode = &pi->vfs_inode;
1498 dprintk("%s: %p: ino: %llu, owned: %d.\n",
1499 __func__, inode, pi->ino, test_bit(NETFS_INODE_OWNED, &pi->state));
1501 mutex_lock(&inode->i_mutex);
1502 if (test_and_clear_bit(NETFS_INODE_OWNED, &pi->state)) {
1503 filemap_fdatawrite(inode->i_mapping);
1504 inode->i_sb->s_op->write_inode(inode, 0);
1507 truncate_inode_pages(inode->i_mapping, 0);
1509 pohmelfs_data_unlock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1510 mutex_unlock(&inode->i_mutex);
1513 static void pohmelfs_put_inode_count(struct pohmelfs_inode *pi, unsigned int count)
1515 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1516 __func__, pi->ino, pi, &pi->vfs_inode, count);
1518 if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH, &pi->state))
1519 pohmelfs_flush_inode(pi, count);
1522 iput(&pi->vfs_inode);
1525 static void pohmelfs_drop_scan(struct work_struct *work)
1527 struct pohmelfs_sb *psb =
1528 container_of(work, struct pohmelfs_sb, drop_dwork.work);
1529 struct pohmelfs_inode *pi;
1530 unsigned int count = 0;
1532 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) {
1533 pohmelfs_put_inode_count(pi, count);
1535 pohmelfs_check_states(psb);
1537 if (psb->drop_scan_timeout)
1538 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1542 * Run through all transactions starting from the oldest,
1543 * drop transaction from current state and try to send it
1544 * to all remote nodes, which are currently installed.
1546 static void pohmelfs_trans_scan_state(struct netfs_state *st)
1548 struct rb_node *rb_node;
1549 struct netfs_trans_dst *dst;
1550 struct pohmelfs_sb *psb = st->psb;
1551 unsigned int timeout = psb->trans_scan_timeout;
1552 struct netfs_trans *t;
1555 mutex_lock(&st->trans_lock);
1556 for (rb_node = rb_first(&st->trans_root); rb_node; ) {
1557 dst = rb_entry(rb_node, struct netfs_trans_dst, state_entry);
1560 if (timeout && time_after(dst->send_time + timeout, jiffies)
1561 && dst->retries == 0)
1564 dprintk("%s: t: %p, gen: %u, st: %p, retries: %u, max: %u.\n",
1565 __func__, t, t->gen, st, dst->retries, psb->trans_retries);
1568 rb_node = rb_next(rb_node);
1571 if (timeout && (++dst->retries < psb->trans_retries)) {
1572 err = netfs_trans_resend(t, psb);
1575 if (err || (t->flags & NETFS_TRANS_SINGLE_DST)) {
1576 if (netfs_trans_remove_nolock(dst, st))
1577 netfs_trans_drop_dst_nostate(dst);
1583 mutex_unlock(&st->trans_lock);
1587 * Walk through all installed network states and resend all
1588 * transactions, which are old enough.
1590 static void pohmelfs_trans_scan(struct work_struct *work)
1592 struct pohmelfs_sb *psb =
1593 container_of(work, struct pohmelfs_sb, dwork.work);
1594 struct netfs_state *st;
1595 struct pohmelfs_config *c;
1597 mutex_lock(&psb->state_lock);
1598 list_for_each_entry(c, &psb->state_list, config_entry) {
1601 pohmelfs_trans_scan_state(st);
1603 mutex_unlock(&psb->state_lock);
1606 * If no timeout specified then system is in the middle of umount process,
1607 * so no need to reschedule scanning process again.
1609 if (psb->trans_scan_timeout)
1610 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1613 int pohmelfs_meta_command_data(struct pohmelfs_inode *pi, u64 id, unsigned int cmd_op, char *addon,
1614 unsigned int flags, netfs_trans_complete_t complete, void *priv, u64 start)
1616 struct inode *inode = &pi->vfs_inode;
1617 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1619 struct netfs_trans *t;
1620 int path_len, addon_len = 0;
1622 struct netfs_inode_info *info;
1623 struct netfs_cmd *cmd;
1625 dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__, pi->ino, cmd_op, addon);
1627 path_len = pohmelfs_path_length(pi);
1634 addon_len = strlen(addon) + 1; /* 0-byte */
1637 if (cmd_op == NETFS_INODE_INFO)
1638 sz += sizeof(struct netfs_inode_info);
1640 t = netfs_trans_alloc(psb, sz + path_len, flags, 0);
1645 t->complete = complete;
1648 cmd = netfs_trans_current(t);
1649 data = (void *)(cmd + 1);
1651 if (cmd_op == NETFS_INODE_INFO) {
1652 info = (struct netfs_inode_info *)(cmd + 1);
1653 data = (void *)(info + 1);
1656 * We are under i_mutex, can read and change whatever we want...
1658 info->mode = inode->i_mode;
1659 info->nlink = inode->i_nlink;
1660 info->uid = inode->i_uid;
1661 info->gid = inode->i_gid;
1662 info->blocks = inode->i_blocks;
1663 info->rdev = inode->i_rdev;
1664 info->size = inode->i_size;
1665 info->version = inode->i_version;
1667 netfs_convert_inode_info(info);
1670 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1674 dprintk("%s: path_len: %d.\n", __func__, path_len);
1677 path_len--; /* Do not place null-byte before the addon */
1678 path_len += sprintf(data + path_len, "/%s", addon) + 1; /* 0 - byte */
1684 cmd->ext = path_len;
1689 netfs_convert_cmd(cmd);
1690 netfs_trans_update(cmd, t, sz);
1693 * Note, that it is possible to leak error here: transaction callback will not
1694 * be invoked for allocation path failure.
1696 return netfs_trans_finish(t, psb);
1699 netfs_trans_free(t);
1702 complete(NULL, 0, priv, err);
1706 int pohmelfs_meta_command(struct pohmelfs_inode *pi, unsigned int cmd_op, unsigned int flags,
1707 netfs_trans_complete_t complete, void *priv, u64 start)
1709 return pohmelfs_meta_command_data(pi, pi->ino, cmd_op, NULL, flags, complete, priv, start);
1713 * Send request and wait for POHMELFS root capabilities response,
1714 * which will update server's informaion about size of the export,
1715 * permissions, number of objects, available size and so on.
1717 static int pohmelfs_root_handshake(struct pohmelfs_sb *psb)
1719 struct netfs_trans *t;
1720 struct netfs_cmd *cmd;
1723 t = netfs_trans_alloc(psb, 0, 0, 0);
1727 cmd = netfs_trans_current(t);
1729 cmd->cmd = NETFS_CAPABILITIES;
1730 cmd->id = POHMELFS_ROOT_CAPABILITIES;
1736 netfs_convert_cmd(cmd);
1737 netfs_trans_update(cmd, t, 0);
1739 err = netfs_trans_finish(t, psb);
1744 err = wait_event_interruptible_timeout(psb->wait,
1746 psb->wait_on_page_timeout);
1762 static int pohmelfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
1764 struct netfs_state *st;
1765 struct pohmelfs_ctl *ctl;
1766 struct pohmelfs_sb *psb = POHMELFS_SB(mnt->mnt_sb);
1767 struct pohmelfs_config *c;
1769 mutex_lock(&psb->state_lock);
1771 seq_printf(m, "\nidx addr(:port) socket_type protocol active priority permissions\n");
1773 list_for_each_entry(c, &psb->state_list, config_entry) {
1777 seq_printf(m, "%u ", ctl->idx);
1778 if (ctl->addr.sa_family == AF_INET) {
1779 struct sockaddr_in *sin = (struct sockaddr_in *)&st->ctl.addr;
1780 //seq_printf(m, "%pi4:%u", &sin->sin_addr.s_addr, ntohs(sin->sin_port));
1781 seq_printf(m, "%u.%u.%u.%u:%u", NIPQUAD(sin->sin_addr.s_addr), ntohs(sin->sin_port));
1782 } else if (ctl->addr.sa_family == AF_INET6) {
1783 struct sockaddr_in6 *sin = (struct sockaddr_in6 *)&st->ctl.addr;
1784 seq_printf(m, "%pi6:%u", &sin->sin6_addr, ntohs(sin->sin6_port));
1787 for (i=0; i<ctl->addrlen; ++i)
1788 seq_printf(m, "%02x.", ctl->addr.addr[i]);
1791 seq_printf(m, " %u %u %d %u %x\n",
1792 ctl->type, ctl->proto,
1794 ctl->prio, ctl->perm);
1796 mutex_unlock(&psb->state_lock);
1801 static const struct super_operations pohmelfs_sb_ops = {
1802 .alloc_inode = pohmelfs_alloc_inode,
1803 .destroy_inode = pohmelfs_destroy_inode,
1804 .drop_inode = pohmelfs_drop_inode,
1805 .write_inode = pohmelfs_write_inode,
1806 .put_super = pohmelfs_put_super,
1807 .remount_fs = pohmelfs_remount,
1808 .statfs = pohmelfs_statfs,
1809 .show_options = pohmelfs_show_options,
1810 .show_stats = pohmelfs_show_stats,
1814 * Allocate private superblock and create root dir.
1816 static int pohmelfs_fill_super(struct super_block *sb, void *data, int silent)
1818 struct pohmelfs_sb *psb;
1821 struct pohmelfs_inode *npi;
1824 psb = kzalloc(sizeof(struct pohmelfs_sb), GFP_KERNEL);
1828 sb->s_fs_info = psb;
1829 sb->s_op = &pohmelfs_sb_ops;
1830 sb->s_magic = POHMELFS_MAGIC_NUM;
1831 sb->s_maxbytes = MAX_LFS_FILESIZE;
1832 sb->s_blocksize = PAGE_SIZE;
1838 psb->active_state = NULL;
1839 psb->trans_retries = 5;
1840 psb->trans_data_size = PAGE_SIZE;
1841 psb->drop_scan_timeout = msecs_to_jiffies(1000);
1842 psb->trans_scan_timeout = msecs_to_jiffies(5000);
1843 psb->wait_on_page_timeout = msecs_to_jiffies(5000);
1844 init_waitqueue_head(&psb->wait);
1846 spin_lock_init(&psb->ino_lock);
1848 INIT_LIST_HEAD(&psb->drop_list);
1850 mutex_init(&psb->mcache_lock);
1851 psb->mcache_root = RB_ROOT;
1852 psb->mcache_timeout = msecs_to_jiffies(5000);
1853 atomic_long_set(&psb->mcache_gen, 0);
1855 psb->trans_max_pages = 100;
1857 psb->crypto_align_size = 16;
1858 psb->crypto_attached_size = 0;
1859 psb->hash_strlen = 0;
1860 psb->cipher_strlen = 0;
1861 psb->perform_crypto = 0;
1862 psb->crypto_thread_num = 2;
1863 psb->crypto_fail_unsupported = 0;
1864 mutex_init(&psb->crypto_thread_lock);
1865 INIT_LIST_HEAD(&psb->crypto_ready_list);
1866 INIT_LIST_HEAD(&psb->crypto_active_list);
1868 atomic_set(&psb->trans_gen, 1);
1869 atomic_set(&psb->total_inodes, 0);
1871 mutex_init(&psb->state_lock);
1872 INIT_LIST_HEAD(&psb->state_list);
1874 err = pohmelfs_parse_options((char *) data, psb, 0);
1876 goto err_out_free_sb;
1878 err = pohmelfs_copy_crypto(psb);
1880 goto err_out_free_sb;
1882 err = pohmelfs_state_init(psb);
1884 goto err_out_free_strings;
1886 err = pohmelfs_crypto_init(psb);
1888 goto err_out_state_exit;
1890 err = pohmelfs_root_handshake(psb);
1892 goto err_out_crypto_exit;
1895 str.hash = jhash("/", 1, 0);
1898 npi = pohmelfs_create_entry_local(psb, NULL, &str, 0, 0755|S_IFDIR);
1901 goto err_out_crypto_exit;
1903 set_bit(NETFS_INODE_REMOTE_SYNCED, &npi->state);
1904 clear_bit(NETFS_INODE_OWNED, &npi->state);
1906 root = &npi->vfs_inode;
1908 sb->s_root = d_alloc_root(root);
1910 goto err_out_put_root;
1912 INIT_DELAYED_WORK(&psb->drop_dwork, pohmelfs_drop_scan);
1913 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1915 INIT_DELAYED_WORK(&psb->dwork, pohmelfs_trans_scan);
1916 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1922 err_out_crypto_exit:
1923 pohmelfs_crypto_exit(psb);
1925 pohmelfs_state_exit(psb);
1926 err_out_free_strings:
1927 kfree(psb->cipher_string);
1928 kfree(psb->hash_string);
1933 dprintk("%s: err: %d.\n", __func__, err);
1938 * Some VFS magic here...
1940 static int pohmelfs_get_sb(struct file_system_type *fs_type,
1941 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1943 return get_sb_nodev(fs_type, flags, data, pohmelfs_fill_super,
1948 * We need this to sync all inodes earlier, since when writeback
1949 * is invoked from the umount/mntput path dcache is already shrunk,
1950 * see generic_shutdown_super(), and no inodes can access the path.
1952 static void pohmelfs_kill_super(struct super_block *sb)
1954 struct writeback_control wbc = {
1955 .sync_mode = WB_SYNC_ALL,
1957 .range_end = LLONG_MAX,
1958 .nr_to_write = LONG_MAX,
1960 generic_sync_sb_inodes(sb, &wbc);
1962 kill_anon_super(sb);
1965 static struct file_system_type pohmel_fs_type = {
1966 .owner = THIS_MODULE,
1968 .get_sb = pohmelfs_get_sb,
1969 .kill_sb = pohmelfs_kill_super,
1973 * Cache and module initializations and freeing routings.
1975 static void pohmelfs_init_once(void *data)
1977 struct pohmelfs_inode *pi = data;
1979 inode_init_once(&pi->vfs_inode);
1982 static int __init pohmelfs_init_inodecache(void)
1984 pohmelfs_inode_cache = kmem_cache_create("pohmelfs_inode_cache",
1985 sizeof(struct pohmelfs_inode),
1986 0, (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1987 pohmelfs_init_once);
1988 if (!pohmelfs_inode_cache)
1994 static void pohmelfs_destroy_inodecache(void)
1996 kmem_cache_destroy(pohmelfs_inode_cache);
1999 static int __init init_pohmel_fs(void)
2003 err = pohmelfs_config_init();
2007 err = pohmelfs_init_inodecache();
2009 goto err_out_config_exit;
2011 err = pohmelfs_mcache_init();
2013 goto err_out_destroy;
2015 err = netfs_trans_init();
2017 goto err_out_mcache_exit;
2019 err = register_filesystem(&pohmel_fs_type);
2027 err_out_mcache_exit:
2028 pohmelfs_mcache_exit();
2030 pohmelfs_destroy_inodecache();
2031 err_out_config_exit:
2032 pohmelfs_config_exit();
2037 static void __exit exit_pohmel_fs(void)
2039 unregister_filesystem(&pohmel_fs_type);
2040 pohmelfs_destroy_inodecache();
2041 pohmelfs_mcache_exit();
2042 pohmelfs_config_exit();
2046 module_init(init_pohmel_fs);
2047 module_exit(exit_pohmel_fs);
2049 MODULE_LICENSE("GPL");
2050 MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
2051 MODULE_DESCRIPTION("Pohmel filesystem");