]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/staging/pohmelfs/inode.c
fs: Add missing mutex_unlock
[net-next-2.6.git] / drivers / staging / pohmelfs / inode.c
CommitLineData
b3f08cad
EP
1/*
2 * 2007+ Copyright (c) Evgeniy Polyakov <zbr@ioremap.net>
3 * All rights reserved.
4 *
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.
9 *
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.
14 */
15
16#include <linux/module.h>
17#include <linux/backing-dev.h>
18#include <linux/crypto.h>
19#include <linux/fs.h>
20#include <linux/jhash.h>
21#include <linux/hash.h>
22#include <linux/ktime.h>
23#include <linux/mm.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>
33
34#include "netfs.h"
35
36#define POHMELFS_MAGIC_NUM 0x504f482e
37
38static struct kmem_cache *pohmelfs_inode_cache;
182374a0 39static atomic_t psb_bdi_num = ATOMIC_INIT(0);
b3f08cad
EP
40
41/*
42 * Removes inode from all trees, drops local name cache and removes all queued
43 * requests for object removal.
44 */
45void pohmelfs_inode_del_inode(struct pohmelfs_sb *psb, struct pohmelfs_inode *pi)
46{
47 mutex_lock(&pi->offset_lock);
48 pohmelfs_free_names(pi);
49 mutex_unlock(&pi->offset_lock);
50
51 dprintk("%s: deleted stuff in ino: %llu.\n", __func__, pi->ino);
52}
53
54/*
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.
60 */
61int pohmelfs_write_inode_create(struct inode *inode, struct netfs_trans *trans)
62{
63 struct pohmelfs_inode *pi = POHMELFS_I(inode);
64 int err = -ENOMEM, size;
65 struct netfs_cmd *cmd;
66 void *data;
67 int cur_len = netfs_trans_cur_len(trans);
68
69 if (unlikely(cur_len < 0))
70 return -ETOOSMALL;
71
72 cmd = netfs_trans_current(trans);
73 cur_len -= sizeof(struct netfs_cmd);
74
75 data = (void *)(cmd + 1);
76
77 err = pohmelfs_construct_path_string(pi, data, cur_len);
78 if (err < 0)
79 goto err_out_exit;
80
81 size = err;
82
83 cmd->start = i_size_read(inode);
84 cmd->cmd = NETFS_CREATE;
85 cmd->size = size;
86 cmd->id = pi->ino;
87 cmd->ext = inode->i_mode;
88
89 netfs_convert_cmd(cmd);
90
91 netfs_trans_update(cmd, trans, size);
92
93 return 0;
94
95err_out_exit:
96 printk("%s: completed ino: %llu, err: %d.\n", __func__, pi->ino, err);
97 return err;
98}
99
100static int pohmelfs_write_trans_complete(struct page **pages, unsigned int page_num,
101 void *private, int err)
102{
103 unsigned i;
104
105 dprintk("%s: pages: %lu-%lu, page_num: %u, err: %d.\n",
106 __func__, pages[0]->index, pages[page_num-1]->index,
107 page_num, err);
108
109 for (i = 0; i < page_num; i++) {
110 struct page *page = pages[i];
111
112 if (!page)
113 continue;
114
115 end_page_writeback(page);
116
117 if (err < 0) {
118 SetPageError(page);
119 set_page_dirty(page);
120 }
121
122 unlock_page(page);
123 page_cache_release(page);
124
125 /* dprintk("%s: %3u/%u: page: %p.\n", __func__, i, page_num, page); */
126 }
127 return err;
128}
129
130static int pohmelfs_inode_has_dirty_pages(struct address_space *mapping, pgoff_t index)
131{
132 int ret;
133 struct page *page;
134
135 rcu_read_lock();
136 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
137 (void **)&page, index, 1, PAGECACHE_TAG_DIRTY);
138 rcu_read_unlock();
139 return ret;
140}
141
142static int pohmelfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
143{
144 struct inode *inode = mapping->host;
145 struct pohmelfs_inode *pi = POHMELFS_I(inode);
146 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
b3f08cad
EP
147 int err = 0;
148 int done = 0;
149 int nr_pages;
150 pgoff_t index;
151 pgoff_t end; /* Inclusive */
152 int scanned = 0;
153 int range_whole = 0;
154
b3f08cad
EP
155 if (wbc->range_cyclic) {
156 index = mapping->writeback_index; /* Start from prev offset */
157 end = -1;
158 } else {
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)
162 range_whole = 1;
163 scanned = 1;
164 }
165retry:
166 while (!done && (index <= end)) {
167 unsigned int i = min(end - index, (pgoff_t)psb->trans_max_pages);
168 int path_len;
169 struct netfs_trans *trans;
170
171 err = pohmelfs_inode_has_dirty_pages(mapping, index);
172 if (!err)
173 break;
174
175 err = pohmelfs_path_length(pi);
176 if (err < 0)
177 break;
178
179 path_len = err;
180
181 if (path_len <= 2) {
182 err = -ENOENT;
183 break;
184 }
185
186 trans = netfs_trans_alloc(psb, path_len, 0, i);
187 if (!trans) {
188 err = -ENOMEM;
189 break;
190 }
191 trans->complete = &pohmelfs_write_trans_complete;
192
193 trans->page_num = nr_pages = find_get_pages_tag(mapping, &index,
194 PAGECACHE_TAG_DIRTY, trans->page_num,
195 trans->pages);
196
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);
199
200 if (!nr_pages)
201 goto err_out_reset;
202
203 err = pohmelfs_write_inode_create(inode, trans);
204 if (err)
205 goto err_out_reset;
206
207 err = 0;
208 scanned = 1;
209
210 for (i = 0; i < trans->page_num; i++) {
211 struct page *page = trans->pages[i];
212
213 lock_page(page);
214
215 if (unlikely(page->mapping != mapping))
216 goto out_continue;
217
218 if (!wbc->range_cyclic && page->index > end) {
219 done = 1;
220 goto out_continue;
221 }
222
223 if (wbc->sync_mode != WB_SYNC_NONE)
224 wait_on_page_writeback(page);
225
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));
230 goto out_continue;
231 }
232
233 set_page_writeback(page);
234
235 trans->attached_size += page_private(page);
236 trans->attached_pages++;
237#if 0
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);
241#endif
242 wbc->nr_to_write--;
243
244 if (wbc->nr_to_write <= 0)
245 done = 1;
b3f08cad
EP
246
247 continue;
248out_continue:
249 unlock_page(page);
250 trans->pages[i] = NULL;
251 }
252
253 err = netfs_trans_finish(trans, psb);
254 if (err)
255 break;
256
257 continue;
258
259err_out_reset:
260 trans->result = err;
261 netfs_trans_reset(trans);
262 netfs_trans_put(trans);
263 break;
264 }
265
266 if (!scanned && !done) {
267 /*
268 * We hit the last page and there is more work to be done: wrap
269 * back to the start of the file
270 */
271 scanned = 1;
272 index = 0;
273 goto retry;
274 }
275
276 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
277 mapping->writeback_index = index;
278
279 return err;
280}
281
282/*
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.
286 */
287static int pohmelfs_write_inode_complete(struct page **pages, unsigned int page_num,
288 void *private, int err)
289{
290 struct inode *inode = private;
291 struct pohmelfs_inode *pi = POHMELFS_I(inode);
292
293 if (inode) {
294 if (err) {
295 mark_inode_dirty(inode);
296 clear_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
297 } else {
298 set_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
299 }
300
301 pohmelfs_put_inode(pi);
302 }
303
304 return err;
305}
306
307int pohmelfs_write_create_inode(struct pohmelfs_inode *pi)
308{
309 struct netfs_trans *t;
310 struct inode *inode = &pi->vfs_inode;
311 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
312 int err;
313
314 if (test_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state))
315 return 0;
316
317 dprintk("%s: started ino: %llu.\n", __func__, pi->ino);
318
319 err = pohmelfs_path_length(pi);
320 if (err < 0)
321 goto err_out_exit;
322
323 t = netfs_trans_alloc(psb, err + 1, 0, 0);
324 if (!t) {
325 err = -ENOMEM;
e3c0acf4 326 goto err_out_exit;
b3f08cad
EP
327 }
328 t->complete = pohmelfs_write_inode_complete;
329 t->private = igrab(inode);
330 if (!t->private) {
331 err = -ENOENT;
332 goto err_out_put;
333 }
334
335 err = pohmelfs_write_inode_create(inode, t);
336 if (err)
337 goto err_out_put;
338
339 netfs_trans_finish(t, POHMELFS_SB(inode->i_sb));
340
341 return 0;
342
343err_out_put:
344 t->result = err;
345 netfs_trans_put(t);
346err_out_exit:
347 return err;
348}
349
350/*
351 * Sync all not-yet-created children in given directory to the server.
352 */
353static int pohmelfs_write_inode_create_children(struct inode *inode)
354{
355 struct pohmelfs_inode *parent = POHMELFS_I(inode);
356 struct super_block *sb = inode->i_sb;
357 struct pohmelfs_name *n;
358
359 while (!list_empty(&parent->sync_create_list)) {
360 n = NULL;
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);
366 }
367 mutex_unlock(&parent->offset_lock);
368
369 if (!n)
370 break;
371
372 inode = ilookup(sb, n->ino);
373
374 dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n",
375 __func__, parent->ino, n->ino, inode);
376
377 if (inode && (inode->i_state & I_DIRTY)) {
378 struct pohmelfs_inode *pi = POHMELFS_I(inode);
379 pohmelfs_write_create_inode(pi);
3bafeab7 380 /* pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0); */
b3f08cad
EP
381 iput(inode);
382 }
383 }
384
385 return 0;
386}
387
388/*
389 * Removes given child from given inode on server.
390 */
391int pohmelfs_remove_child(struct pohmelfs_inode *pi, struct pohmelfs_name *n)
392{
393 return pohmelfs_meta_command_data(pi, pi->ino, NETFS_REMOVE, NULL, 0, NULL, NULL, 0);
394}
395
396/*
397 * Writeback for given inode.
398 */
582de7ca
RD
399static int pohmelfs_write_inode(struct inode *inode,
400 struct writeback_control *wbc)
b3f08cad
EP
401{
402 struct pohmelfs_inode *pi = POHMELFS_I(inode);
403
404 pohmelfs_write_create_inode(pi);
405 pohmelfs_write_inode_create_children(inode);
406
407 return 0;
408}
409
410/*
411 * It is not exported, sorry...
412 */
413static inline wait_queue_head_t *page_waitqueue(struct page *page)
414{
415 const struct zone *zone = page_zone(page);
416
417 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
418}
419
420static int pohmelfs_wait_on_page_locked(struct page *page)
421{
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);
425 int err = 0;
426
427 if (!PageLocked(page))
428 return 0;
429
430 for (;;) {
431 prepare_to_wait(page_waitqueue(page),
432 &wait.wait, TASK_INTERRUPTIBLE);
433
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);
437
438 if (!PageLocked(page))
439 break;
440
441 if (!signal_pending(current)) {
442 ret = schedule_timeout(ret);
443 if (!ret)
444 break;
445 continue;
446 }
447 ret = -ERESTARTSYS;
448 break;
449 }
450 finish_wait(page_waitqueue(page), &wait.wait);
451
452 if (!ret)
453 err = -ETIMEDOUT;
454
455
456 if (!err)
457 SetPageUptodate(page);
458
459 if (err)
460 printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n",
461 __func__, page, PageUptodate(page), PageLocked(page), err);
462
463 return err;
464}
465
466static int pohmelfs_read_page_complete(struct page **pages, unsigned int page_num,
467 void *private, int err)
468{
469 struct page *page = private;
470
471 if (PageChecked(page))
472 return err;
473
474 if (err < 0) {
475 dprintk("%s: page: %p, err: %d.\n", __func__, page, err);
476 SetPageError(page);
477 }
478
479 unlock_page(page);
480
481 return err;
482}
483
484/*
485 * Read a page from remote server.
486 * Function will wait until page is unlocked.
487 */
488static int pohmelfs_readpage(struct file *file, struct page *page)
489{
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;
495 int err, path_len;
496 void *data;
497 u64 isize;
498
499 err = pohmelfs_data_lock(pi, page->index << PAGE_CACHE_SHIFT,
500 PAGE_SIZE, POHMELFS_READ_LOCK);
501 if (err)
502 goto err_out_exit;
503
504 isize = i_size_read(inode);
505 if (isize <= page->index << PAGE_CACHE_SHIFT) {
506 SetPageUptodate(page);
507 unlock_page(page);
508 return 0;
509 }
510
511 path_len = pohmelfs_path_length(pi);
512 if (path_len < 0) {
513 err = path_len;
514 goto err_out_exit;
515 }
516
517 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
518 if (!t) {
519 err = -ENOMEM;
520 goto err_out_exit;
521 }
522
523 t->complete = pohmelfs_read_page_complete;
524 t->private = page;
525
526 cmd = netfs_trans_current(t);
527 data = (void *)(cmd + 1);
528
529 err = pohmelfs_construct_path_string(pi, data, path_len);
530 if (err < 0)
531 goto err_out_free;
532
533 path_len = err;
534
535 cmd->id = pi->ino;
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;
540 cmd->ext = path_len;
541
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);
544
545 netfs_convert_cmd(cmd);
546 netfs_trans_update(cmd, t, path_len);
547
548 err = netfs_trans_finish(t, psb);
549 if (err)
550 goto err_out_return;
551
552 return pohmelfs_wait_on_page_locked(page);
553
554err_out_free:
555 t->result = err;
556 netfs_trans_put(t);
557err_out_exit:
558 SetPageError(page);
559 if (PageLocked(page))
560 unlock_page(page);
561err_out_return:
562 printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n",
563 __func__, page, page->index << PAGE_CACHE_SHIFT, PAGE_CACHE_SIZE, err);
564
565 return err;
566}
567
568/*
569 * Write begin/end magic.
570 * Allocates a page and writes inode if it was not synced to server before.
571 */
572static 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)
575{
576 struct inode *inode = mapping->host;
577 struct page *page;
578 pgoff_t index;
579 unsigned start, end;
580 int err;
581
582 *pagep = NULL;
583
584 index = pos >> PAGE_CACHE_SHIFT;
585 start = pos & (PAGE_CACHE_SIZE - 1);
586 end = start + len;
587
588 page = grab_cache_page(mapping, index);
589#if 0
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));
592#endif
593 if (!page) {
594 err = -ENOMEM;
595 goto err_out_exit;
596 }
597
598 while (!PageUptodate(page)) {
599 if (start && test_bit(NETFS_INODE_REMOTE_SYNCED, &POHMELFS_I(inode)->state)) {
600 err = pohmelfs_readpage(file, page);
601 if (err)
602 goto err_out_exit;
603
604 lock_page(page);
605 continue;
606 }
607
608 if (len != PAGE_CACHE_SIZE) {
609 void *kaddr = kmap_atomic(page, KM_USER0);
610
611 memset(kaddr + start, 0, PAGE_CACHE_SIZE - start);
612 flush_dcache_page(page);
613 kunmap_atomic(kaddr, KM_USER0);
614 }
615 SetPageUptodate(page);
616 }
617
618 set_page_private(page, end);
619
620 *pagep = page;
621
622 return 0;
623
624err_out_exit:
625 page_cache_release(page);
626 *pagep = NULL;
627
628 return err;
629}
630
631static 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)
634{
635 struct inode *inode = mapping->host;
636
637 if (copied != len) {
638 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
639 void *kaddr = kmap_atomic(page, KM_USER0);
640
641 memset(kaddr + from + copied, 0, len - copied);
642 flush_dcache_page(page);
643 kunmap_atomic(kaddr, KM_USER0);
644 }
645
646 SetPageUptodate(page);
647 set_page_dirty(page);
648#if 0
649 dprintk("%s: page: %p [U: %d, D: %d, L: %d], pos: %llu, len: %u, copied: %u.\n",
650 __func__, page,
651 PageUptodate(page), PageDirty(page), PageLocked(page),
652 pos, len, copied);
653#endif
654 flush_dcache_page(page);
655
656 unlock_page(page);
657 page_cache_release(page);
658
659 if (pos + copied > inode->i_size) {
660 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
661
662 psb->avail_size -= pos + copied - inode->i_size;
663
664 i_size_write(inode, pos + copied);
665 }
666
667 return copied;
668}
669
670static int pohmelfs_readpages_trans_complete(struct page **__pages, unsigned int page_num,
671 void *private, int err)
672{
673 struct pohmelfs_inode *pi = private;
674 unsigned int i, num;
675 struct page **pages, *page = (struct page *)__pages;
676 loff_t index = page->index;
677
678 pages = kzalloc(sizeof(void *) * page_num, GFP_NOIO);
679 if (!pages)
680 return -ENOMEM;
681
682 num = find_get_pages_contig(pi->vfs_inode.i_mapping, index, page_num, pages);
683 if (num <= 0) {
684 err = num;
685 goto err_out_free;
686 }
687
5d6892b3 688 for (i = 0; i < num; ++i) {
b3f08cad
EP
689 page = pages[i];
690
691 if (err)
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);
695
696 if (!PageChecked(page)) {
697 if (err < 0)
698 SetPageError(page);
699 unlock_page(page);
700 }
701 page_cache_release(page);
702 page_cache_release(page);
703 }
704
705err_out_free:
706 kfree(pages);
707 return err;
708}
709
710static int pohmelfs_send_readpages(struct pohmelfs_inode *pi, struct page *first, unsigned int num)
711{
712 struct netfs_trans *t;
713 struct netfs_cmd *cmd;
714 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
715 int err, path_len;
716 void *data;
717
718 err = pohmelfs_data_lock(pi, first->index << PAGE_CACHE_SHIFT,
719 num * PAGE_SIZE, POHMELFS_READ_LOCK);
720 if (err)
721 goto err_out_exit;
722
723 path_len = pohmelfs_path_length(pi);
724 if (path_len < 0) {
725 err = path_len;
726 goto err_out_exit;
727 }
728
729 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
730 if (!t) {
731 err = -ENOMEM;
732 goto err_out_exit;
733 }
734
735 cmd = netfs_trans_current(t);
736 data = (void *)(cmd + 1);
737
738 t->complete = pohmelfs_readpages_trans_complete;
739 t->private = pi;
740 t->page_num = num;
741 t->pages = (struct page **)first;
742
743 err = pohmelfs_construct_path_string(pi, data, path_len);
744 if (err < 0)
745 goto err_out_put;
746
747 path_len = err;
748
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);
753 cmd->id = pi->ino;
754 cmd->ext = path_len;
755
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,
759 first->index, num);
760
761 netfs_convert_cmd(cmd);
762 netfs_trans_update(cmd, t, path_len);
763
764 return netfs_trans_finish(t, psb);
765
766err_out_put:
767 netfs_trans_free(t);
768err_out_exit:
769 pohmelfs_readpages_trans_complete((struct page **)first, num, pi, err);
770 return err;
771}
772
773#define list_to_page(head) (list_entry((head)->prev, struct page, lru))
774
775static int pohmelfs_readpages(struct file *file, struct address_space *mapping,
776 struct list_head *pages, unsigned nr_pages)
777{
778 unsigned int page_idx, num = 0;
779 struct page *page = NULL, *first = NULL;
780
781 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
782 page = list_to_page(pages);
783
784 prefetchw(&page->flags);
785 list_del(&page->lru);
786
787 if (!add_to_page_cache_lru(page, mapping,
788 page->index, GFP_KERNEL)) {
789
790 if (!num) {
791 num = 1;
792 first = page;
793 continue;
794 }
795
796 dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n",
797 __func__, page, page->index, first->index);
798
799 if (unlikely(first->index + num != page->index) || (num > 500)) {
800 pohmelfs_send_readpages(POHMELFS_I(mapping->host),
801 first, num);
802 first = page;
803 num = 0;
804 }
805
806 num++;
807 }
808 }
809 pohmelfs_send_readpages(POHMELFS_I(mapping->host), first, num);
810
811 /*
812 * This will be sync read, so when last page is processed,
813 * all previous are alerady unlocked and ready to be used.
814 */
815 return 0;
816}
817
818/*
819 * Small addres space operations for POHMELFS.
820 */
821const 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,
828};
829
830/*
831 * ->detroy_inode() callback. Deletes inode from the caches
832 * and frees private data.
833 */
834static void pohmelfs_destroy_inode(struct inode *inode)
835{
836 struct super_block *sb = inode->i_sb;
837 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
838 struct pohmelfs_inode *pi = POHMELFS_I(inode);
839
3bafeab7 840 /* pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK); */
b3f08cad
EP
841
842 pohmelfs_inode_del_inode(psb, pi);
843
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);
848}
849
850/*
851 * ->alloc_inode() callback. Allocates inode and initilizes private data.
852 */
853static struct inode *pohmelfs_alloc_inode(struct super_block *sb)
854{
855 struct pohmelfs_inode *pi;
856
857 pi = kmem_cache_alloc(pohmelfs_inode_cache, GFP_NOIO);
858 if (!pi)
859 return NULL;
860
861 pi->hash_root = RB_ROOT;
862 mutex_init(&pi->offset_lock);
863
864 INIT_LIST_HEAD(&pi->sync_create_list);
865
866 INIT_LIST_HEAD(&pi->inode_entry);
867
868 pi->lock_type = 0;
869 pi->state = 0;
870 pi->total_len = 0;
871 pi->drop_count = 0;
872
873 dprintk("%s: pi: %p, inode: %p.\n", __func__, pi, &pi->vfs_inode);
874
875 atomic_long_inc(&POHMELFS_SB(sb)->total_inodes);
876
877 return &pi->vfs_inode;
878}
879
880/*
881 * We want fsync() to work on POHMELFS.
882 */
883static int pohmelfs_fsync(struct file *file, struct dentry *dentry, int datasync)
884{
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 */
889 };
890
891 return sync_inode(inode, &wbc);
892}
893
894ssize_t pohmelfs_write(struct file *file, const char __user *buf,
895 size_t len, loff_t *ppos)
896{
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 };
901 struct kiocb kiocb;
902 ssize_t ret;
903 loff_t pos = *ppos;
904
905 init_sync_kiocb(&kiocb, file);
906 kiocb.ki_pos = pos;
907 kiocb.ki_left = len;
908
02d84ca5 909 dprintk("%s: len: %zu, pos: %llu.\n", __func__, len, pos);
b3f08cad
EP
910
911 mutex_lock(&inode->i_mutex);
912 ret = pohmelfs_data_lock(pi, pos, len, POHMELFS_WRITE_LOCK);
913 if (ret)
914 goto err_out_unlock;
915
b04f9321 916 ret = __generic_file_aio_write(&kiocb, &iov, 1, &kiocb.ki_pos);
b3f08cad
EP
917 *ppos = kiocb.ki_pos;
918
919 mutex_unlock(&inode->i_mutex);
920 WARN_ON(ret < 0);
921
aa3caafe 922 if (ret > 0) {
b3f08cad
EP
923 ssize_t err;
924
aa3caafe 925 err = generic_write_sync(file, pos, ret);
b3f08cad
EP
926 if (err < 0)
927 ret = err;
928 WARN_ON(ret < 0);
929 }
930
931 return ret;
932
933err_out_unlock:
934 mutex_unlock(&inode->i_mutex);
935 return ret;
936}
937
80fe3e5d 938static const struct file_operations pohmelfs_file_ops = {
b3f08cad
EP
939 .open = generic_file_open,
940 .fsync = pohmelfs_fsync,
941
942 .llseek = generic_file_llseek,
943
944 .read = do_sync_read,
945 .aio_read = generic_file_aio_read,
946
947 .mmap = generic_file_mmap,
948
949 .splice_read = generic_file_splice_read,
950 .splice_write = generic_file_splice_write,
951
952 .write = pohmelfs_write,
953 .aio_write = generic_file_aio_write,
954};
955
956const struct inode_operations pohmelfs_symlink_inode_operations = {
957 .readlink = generic_readlink,
958 .follow_link = page_follow_link_light,
959 .put_link = page_put_link,
960};
961
962int pohmelfs_setattr_raw(struct inode *inode, struct iattr *attr)
963{
964 int err;
965
966 err = inode_change_ok(inode, attr);
967 if (err) {
968 dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__, POHMELFS_I(inode)->ino);
969 goto err_out_exit;
970 }
971
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)) {
b43fa828 974 err = dquot_transfer(inode, attr);
b3f08cad
EP
975 if (err)
976 goto err_out_exit;
977 }
978
979 err = inode_setattr(inode, attr);
980 if (err) {
981 dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__, POHMELFS_I(inode)->ino);
982 goto err_out_exit;
983 }
984
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);
988
989 return 0;
990
991err_out_exit:
992 return err;
993}
994
995int pohmelfs_setattr(struct dentry *dentry, struct iattr *attr)
996{
997 struct inode *inode = dentry->d_inode;
998 struct pohmelfs_inode *pi = POHMELFS_I(inode);
999 int err;
1000
1001 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1002 if (err)
1003 goto err_out_exit;
1004
1005 err = security_inode_setattr(dentry, attr);
1006 if (err)
1007 goto err_out_exit;
1008
1009 err = pohmelfs_setattr_raw(inode, attr);
1010 if (err)
1011 goto err_out_exit;
1012
1013 return 0;
1014
1015err_out_exit:
1016 return err;
1017}
1018
1019static 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)
1021{
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;
1026 void *data;
1027
02d84ca5 1028 dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n",
b3f08cad
EP
1029 __func__, id, start, name, attrsize, command);
1030
1031 path_len = pohmelfs_path_length(pi);
1032 if (path_len < 0) {
1033 err = path_len;
1034 goto err_out_exit;
1035 }
1036
1037 t = netfs_trans_alloc(psb, namelen + path_len + attrsize, 0, 0);
1038 if (!t) {
1039 err = -ENOMEM;
1040 goto err_out_exit;
1041 }
1042
1043 cmd = netfs_trans_current(t);
1044 data = cmd + 1;
1045
1046 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1047 if (path_len < 0) {
1048 err = path_len;
1049 goto err_out_put;
1050 }
1051 data += path_len;
1052
1053 /*
1054 * 'name' is a NUL-terminated string already and
1055 * 'namelen' includes 0-byte.
1056 */
1057 memcpy(data, name, namelen);
1058 data += namelen;
1059
1060 memcpy(data, value, attrsize);
1061
1062 cmd->cmd = command;
1063 cmd->id = id;
1064 cmd->start = start;
1065 cmd->size = attrsize + namelen + path_len;
1066 cmd->ext = path_len;
1067 cmd->csize = 0;
1068 cmd->cpad = 0;
1069
1070 netfs_convert_cmd(cmd);
1071 netfs_trans_update(cmd, t, namelen + path_len + attrsize);
1072
1073 return netfs_trans_finish(t, psb);
1074
1075err_out_put:
1076 t->result = err;
1077 netfs_trans_put(t);
1078err_out_exit:
1079 return err;
1080}
1081
1082static int pohmelfs_setxattr(struct dentry *dentry, const char *name,
1083 const void *value, size_t attrsize, int flags)
1084{
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);
1088
1089 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1090 return -EOPNOTSUPP;
1091
1092 return pohmelfs_send_xattr_req(pi, flags, attrsize, name,
1093 value, attrsize, NETFS_XATTR_SET);
1094}
1095
1096static ssize_t pohmelfs_getxattr(struct dentry *dentry, const char *name,
1097 void *value, size_t attrsize)
1098{
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;
1103 int err;
1104 long timeout = psb->mcache_timeout;
1105
1106 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1107 return -EOPNOTSUPP;
1108
1109 m = pohmelfs_mcache_alloc(psb, 0, attrsize, value);
1110 if (IS_ERR(m))
1111 return PTR_ERR(m);
1112
1113 dprintk("%s: ino: %llu, name: '%s', size: %zu.\n",
1114 __func__, pi->ino, name, attrsize);
1115
1116 err = pohmelfs_send_xattr_req(pi, m->gen, attrsize, name, value, 0, NETFS_XATTR_GET);
1117 if (err)
1118 goto err_out_put;
1119
1120 do {
1121 err = wait_for_completion_timeout(&m->complete, timeout);
1122 if (err) {
1123 err = m->err;
1124 break;
1125 }
1126
1127 /*
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.
1136 *
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
1142 * problem.
1143 */
1144 err = -ETIMEDOUT;
1145 timeout = HZ;
1146
1147 pohmelfs_mcache_remove_locked(psb, m);
1148 } while (atomic_read(&m->refcnt) != 1);
1149
1150 pohmelfs_mcache_put(psb, m);
1151
1152 dprintk("%s: ino: %llu, err: %d.\n", __func__, pi->ino, err);
1153
1154 return err;
1155
1156err_out_put:
1157 pohmelfs_mcache_put(psb, m);
1158 return err;
1159}
1160
1161static int pohmelfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1162{
1163 struct inode *inode = dentry->d_inode;
e5043424 1164#if 0
b3f08cad
EP
1165 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1166 int err;
e5043424 1167
b3f08cad
EP
1168 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_READ_LOCK);
1169 if (err)
1170 return err;
b3f08cad
EP
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);
e5043424 1174#endif
b3f08cad
EP
1175
1176 generic_fillattr(inode, stat);
1177 return 0;
1178}
1179
1180const struct inode_operations pohmelfs_file_inode_operations = {
1181 .setattr = pohmelfs_setattr,
1182 .getattr = pohmelfs_getattr,
1183 .setxattr = pohmelfs_setxattr,
1184 .getxattr = pohmelfs_getxattr,
1185};
1186
1187/*
1188 * Fill inode data: mode, size, operation callbacks and so on...
1189 */
1190void pohmelfs_fill_inode(struct inode *inode, struct netfs_inode_info *info)
1191{
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);
1201
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);
1206
1207 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1208
1209 /*
1210 * i_mapping is a pointer to i_data during inode initialization.
1211 */
1212 inode->i_data.a_ops = &pohmelfs_aops;
1213
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;
1223 } else {
1224 inode->i_fop = &generic_ro_fops;
1225 }
1226}
1227
1228static void pohmelfs_drop_inode(struct inode *inode)
1229{
1230 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1231 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1232
1233 spin_lock(&psb->ino_lock);
1234 list_del_init(&pi->inode_entry);
1235 spin_unlock(&psb->ino_lock);
1236
1237 generic_drop_inode(inode);
1238}
1239
1240static struct pohmelfs_inode *pohmelfs_get_inode_from_list(struct pohmelfs_sb *psb,
1241 struct list_head *head, unsigned int *count)
1242{
1243 struct pohmelfs_inode *pi = NULL;
1244
1245 spin_lock(&psb->ino_lock);
1246 if (!list_empty(head)) {
1247 pi = list_entry(head->next, struct pohmelfs_inode,
1248 inode_entry);
1249 list_del_init(&pi->inode_entry);
1250 *count = pi->drop_count;
1251 pi->drop_count = 0;
1252 }
1253 spin_unlock(&psb->ino_lock);
1254
1255 return pi;
1256}
1257
1258static void pohmelfs_flush_transactions(struct pohmelfs_sb *psb)
1259{
1260 struct pohmelfs_config *c;
1261
1262 mutex_lock(&psb->state_lock);
1263 list_for_each_entry(c, &psb->state_list, config_entry) {
1264 pohmelfs_state_flush_transactions(&c->state);
1265 }
1266 mutex_unlock(&psb->state_lock);
1267}
1268
1269/*
1270 * ->put_super() callback. Invoked before superblock is destroyed,
1271 * so it has to clean all private data.
1272 */
1273static void pohmelfs_put_super(struct super_block *sb)
1274{
1275 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1276 struct pohmelfs_inode *pi;
1277 unsigned int count;
1278 unsigned int in_drop_list = 0;
1279 struct inode *inode, *tmp;
1280
1281 dprintk("%s.\n", __func__);
1282
1283 /*
1284 * Kill pending transactions, which could affect inodes in-flight.
1285 */
1286 pohmelfs_flush_transactions(psb);
1287
1288 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) {
1289 inode = &pi->vfs_inode;
1290
1291 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1292 __func__, pi->ino, pi, inode, count);
1293
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);
1299 in_drop_list++;
1300 }
1301
1302 while (count--)
1303 iput(&pi->vfs_inode);
1304 }
1305
1306 list_for_each_entry_safe(inode, tmp, &sb->s_inodes, i_sb_list) {
1307 pi = POHMELFS_I(inode);
1308
1309 dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n",
1310 __func__, pi->ino, pi, inode, atomic_read(&inode->i_count));
1311
1312 /*
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.
1317 */
1318 count = atomic_read(&inode->i_count);
1319 if (count) {
1320 list_del_init(&inode->i_sb_list);
1321 while (count--)
1322 iput(&pi->vfs_inode);
1323 }
1324 }
1325
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();
1330
1331 dprintk("%s: stopped workqueues.\n", __func__);
1332
1333 pohmelfs_crypto_exit(psb);
1334 pohmelfs_state_exit(psb);
1335
182374a0
EP
1336 bdi_destroy(&psb->bdi);
1337
b3f08cad
EP
1338 kfree(psb);
1339 sb->s_fs_info = NULL;
b3f08cad
EP
1340}
1341
b3f08cad
EP
1342static int pohmelfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1343{
1344 struct super_block *sb = dentry->d_sb;
1345 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1346
1347 /*
1348 * There are no filesystem size limits yet.
1349 */
1350 memset(buf, 0, sizeof(struct kstatfs));
1351
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;
1359
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);
1362
1363 return 0;
1364}
1365
1366static int pohmelfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
1367{
1368 struct pohmelfs_sb *psb = POHMELFS_SB(vfs->mnt_sb);
1369
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");
1380
1381 return 0;
1382}
1383
b3f08cad
EP
1384enum {
1385 pohmelfs_opt_idx,
e5043424
EP
1386 pohmelfs_opt_crypto_thread_num,
1387 pohmelfs_opt_trans_max_pages,
1388 pohmelfs_opt_crypto_fail_unsupported,
1389
1390 /* Remountable options */
b3f08cad
EP
1391 pohmelfs_opt_trans_scan_timeout,
1392 pohmelfs_opt_drop_scan_timeout,
1393 pohmelfs_opt_wait_on_page_timeout,
1394 pohmelfs_opt_trans_retries,
b3f08cad
EP
1395 pohmelfs_opt_mcache_timeout,
1396};
1397
1398static struct match_token pohmelfs_tokens[] = {
1399 {pohmelfs_opt_idx, "idx=%u"},
e5043424
EP
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"},
b3f08cad
EP
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"},
b3f08cad
EP
1407 {pohmelfs_opt_mcache_timeout, "mcache_timeout=%u"},
1408};
1409
e5043424 1410static int pohmelfs_parse_options(char *options, struct pohmelfs_sb *psb, int remount)
b3f08cad
EP
1411{
1412 char *p;
1413 substring_t args[MAX_OPT_ARGS];
1414 int option, err;
1415
1416 if (!options)
1417 return 0;
1418
1419 while ((p = strsep(&options, ",")) != NULL) {
1420 int token;
1421 if (!*p)
1422 continue;
1423
1424 token = match_token(p, pohmelfs_tokens, args);
1425
1426 err = match_int(&args[0], &option);
1427 if (err)
1428 return err;
1429
e5043424
EP
1430 if (remount && token <= pohmelfs_opt_crypto_fail_unsupported)
1431 continue;
1432
b3f08cad 1433 switch (token) {
5d6892b3
RP
1434 case pohmelfs_opt_idx:
1435 psb->idx = option;
1436 break;
1437 case pohmelfs_opt_trans_scan_timeout:
1438 psb->trans_scan_timeout = msecs_to_jiffies(option);
1439 break;
1440 case pohmelfs_opt_drop_scan_timeout:
1441 psb->drop_scan_timeout = msecs_to_jiffies(option);
1442 break;
1443 case pohmelfs_opt_wait_on_page_timeout:
1444 psb->wait_on_page_timeout = msecs_to_jiffies(option);
1445 break;
1446 case pohmelfs_opt_mcache_timeout:
1447 psb->mcache_timeout = msecs_to_jiffies(option);
1448 break;
1449 case pohmelfs_opt_trans_retries:
1450 psb->trans_retries = option;
1451 break;
1452 case pohmelfs_opt_crypto_thread_num:
1453 psb->crypto_thread_num = option;
1454 break;
1455 case pohmelfs_opt_trans_max_pages:
1456 psb->trans_max_pages = option;
1457 break;
1458 case pohmelfs_opt_crypto_fail_unsupported:
1459 psb->crypto_fail_unsupported = 1;
1460 break;
1461 default:
1462 return -EINVAL;
b3f08cad
EP
1463 }
1464 }
1465
1466 return 0;
1467}
1468
e5043424
EP
1469static int pohmelfs_remount(struct super_block *sb, int *flags, char *data)
1470{
1471 int err;
1472 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1473 unsigned long old_sb_flags = sb->s_flags;
1474
1475 err = pohmelfs_parse_options(data, psb, 1);
1476 if (err)
1477 goto err_out_restore;
1478
1479 if (!(*flags & MS_RDONLY))
1480 sb->s_flags &= ~MS_RDONLY;
1481 return 0;
1482
1483err_out_restore:
1484 sb->s_flags = old_sb_flags;
1485 return err;
1486}
1487
b3f08cad
EP
1488static void pohmelfs_flush_inode(struct pohmelfs_inode *pi, unsigned int count)
1489{
1490 struct inode *inode = &pi->vfs_inode;
1491
1492 dprintk("%s: %p: ino: %llu, owned: %d.\n",
1493 __func__, inode, pi->ino, test_bit(NETFS_INODE_OWNED, &pi->state));
1494
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);
1499 }
1500
2d7cf8ef 1501#ifdef POHMELFS_TRUNCATE_ON_INODE_FLUSH
b3f08cad 1502 truncate_inode_pages(inode->i_mapping, 0);
2d7cf8ef 1503#endif
b3f08cad
EP
1504
1505 pohmelfs_data_unlock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1506 mutex_unlock(&inode->i_mutex);
1507}
1508
1509static void pohmelfs_put_inode_count(struct pohmelfs_inode *pi, unsigned int count)
1510{
1511 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1512 __func__, pi->ino, pi, &pi->vfs_inode, count);
1513
1514 if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH, &pi->state))
1515 pohmelfs_flush_inode(pi, count);
1516
1517 while (count--)
1518 iput(&pi->vfs_inode);
1519}
1520
1521static void pohmelfs_drop_scan(struct work_struct *work)
1522{
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;
1527
dc828461 1528 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count)))
b3f08cad 1529 pohmelfs_put_inode_count(pi, count);
dc828461 1530
b3f08cad
EP
1531 pohmelfs_check_states(psb);
1532
1533 if (psb->drop_scan_timeout)
1534 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1535}
1536
1537/*
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.
1541 */
1542static void pohmelfs_trans_scan_state(struct netfs_state *st)
1543{
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;
1549 int err;
1550
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);
1554 t = dst->trans;
1555
1556 if (timeout && time_after(dst->send_time + timeout, jiffies)
1557 && dst->retries == 0)
1558 break;
1559
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);
1562 netfs_trans_get(t);
1563
1564 rb_node = rb_next(rb_node);
1565
1566 err = -ETIMEDOUT;
dc828461 1567 if (timeout && (++dst->retries < psb->trans_retries))
b3f08cad 1568 err = netfs_trans_resend(t, psb);
b3f08cad
EP
1569
1570 if (err || (t->flags & NETFS_TRANS_SINGLE_DST)) {
1571 if (netfs_trans_remove_nolock(dst, st))
1572 netfs_trans_drop_dst_nostate(dst);
1573 }
1574
1575 t->result = err;
1576 netfs_trans_put(t);
1577 }
1578 mutex_unlock(&st->trans_lock);
1579}
1580
1581/*
1582 * Walk through all installed network states and resend all
1583 * transactions, which are old enough.
1584 */
1585static void pohmelfs_trans_scan(struct work_struct *work)
1586{
1587 struct pohmelfs_sb *psb =
1588 container_of(work, struct pohmelfs_sb, dwork.work);
1589 struct netfs_state *st;
1590 struct pohmelfs_config *c;
1591
1592 mutex_lock(&psb->state_lock);
1593 list_for_each_entry(c, &psb->state_list, config_entry) {
1594 st = &c->state;
1595
1596 pohmelfs_trans_scan_state(st);
1597 }
1598 mutex_unlock(&psb->state_lock);
1599
1600 /*
1601 * If no timeout specified then system is in the middle of umount process,
1602 * so no need to reschedule scanning process again.
1603 */
1604 if (psb->trans_scan_timeout)
1605 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1606}
1607
1608int 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)
1610{
1611 struct inode *inode = &pi->vfs_inode;
1612 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1613 int err = 0, sz;
1614 struct netfs_trans *t;
1615 int path_len, addon_len = 0;
1616 void *data;
1617 struct netfs_inode_info *info;
1618 struct netfs_cmd *cmd;
1619
1620 dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__, pi->ino, cmd_op, addon);
1621
1622 path_len = pohmelfs_path_length(pi);
1623 if (path_len < 0) {
1624 err = path_len;
1625 goto err_out_exit;
1626 }
1627
1628 if (addon)
1629 addon_len = strlen(addon) + 1; /* 0-byte */
1630 sz = addon_len;
1631
1632 if (cmd_op == NETFS_INODE_INFO)
1633 sz += sizeof(struct netfs_inode_info);
1634
1635 t = netfs_trans_alloc(psb, sz + path_len, flags, 0);
1636 if (!t) {
1637 err = -ENOMEM;
1638 goto err_out_exit;
1639 }
1640 t->complete = complete;
1641 t->private = priv;
1642
1643 cmd = netfs_trans_current(t);
1644 data = (void *)(cmd + 1);
1645
1646 if (cmd_op == NETFS_INODE_INFO) {
1647 info = (struct netfs_inode_info *)(cmd + 1);
1648 data = (void *)(info + 1);
1649
1650 /*
1651 * We are under i_mutex, can read and change whatever we want...
1652 */
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;
1661
1662 netfs_convert_inode_info(info);
1663 }
1664
1665 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1666 if (path_len < 0)
1667 goto err_out_free;
1668
1669 dprintk("%s: path_len: %d.\n", __func__, path_len);
1670
1671 if (addon) {
1672 path_len--; /* Do not place null-byte before the addon */
1673 path_len += sprintf(data + path_len, "/%s", addon) + 1; /* 0 - byte */
1674 }
1675
1676 sz += path_len;
1677
1678 cmd->cmd = cmd_op;
1679 cmd->ext = path_len;
1680 cmd->size = sz;
1681 cmd->id = id;
1682 cmd->start = start;
1683
1684 netfs_convert_cmd(cmd);
1685 netfs_trans_update(cmd, t, sz);
1686
1687 /*
1688 * Note, that it is possible to leak error here: transaction callback will not
1689 * be invoked for allocation path failure.
1690 */
1691 return netfs_trans_finish(t, psb);
1692
1693err_out_free:
1694 netfs_trans_free(t);
1695err_out_exit:
1696 if (complete)
1697 complete(NULL, 0, priv, err);
1698 return err;
1699}
1700
1701int pohmelfs_meta_command(struct pohmelfs_inode *pi, unsigned int cmd_op, unsigned int flags,
1702 netfs_trans_complete_t complete, void *priv, u64 start)
1703{
1704 return pohmelfs_meta_command_data(pi, pi->ino, cmd_op, NULL, flags, complete, priv, start);
1705}
1706
1707/*
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.
1711 */
1712static int pohmelfs_root_handshake(struct pohmelfs_sb *psb)
1713{
1714 struct netfs_trans *t;
1715 struct netfs_cmd *cmd;
1716 int err = -ENOMEM;
1717
1718 t = netfs_trans_alloc(psb, 0, 0, 0);
1719 if (!t)
1720 goto err_out_exit;
1721
1722 cmd = netfs_trans_current(t);
1723
1724 cmd->cmd = NETFS_CAPABILITIES;
1725 cmd->id = POHMELFS_ROOT_CAPABILITIES;
1726 cmd->size = 0;
1727 cmd->start = 0;
1728 cmd->ext = 0;
1729 cmd->csize = 0;
1730
1731 netfs_convert_cmd(cmd);
1732 netfs_trans_update(cmd, t, 0);
1733
1734 err = netfs_trans_finish(t, psb);
1735 if (err)
1736 goto err_out_exit;
1737
1738 psb->flags = ~0;
1739 err = wait_event_interruptible_timeout(psb->wait,
1740 (psb->flags != ~0),
1741 psb->wait_on_page_timeout);
2d7cf8ef 1742 if (!err)
b3f08cad 1743 err = -ETIMEDOUT;
2d7cf8ef 1744 else if (err > 0)
b3f08cad 1745 err = -psb->flags;
b3f08cad
EP
1746
1747 if (err)
1748 goto err_out_exit;
1749
1750 return 0;
1751
1752err_out_exit:
1753 return err;
1754}
1755
f2739de1
EP
1756static int pohmelfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
1757{
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;
1762
1763 mutex_lock(&psb->state_lock);
1764
1765 seq_printf(m, "\nidx addr(:port) socket_type protocol active priority permissions\n");
1766
1767 list_for_each_entry(c, &psb->state_list, config_entry) {
1768 st = &c->state;
1769 ctl = &st->ctl;
1770
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;
0c14c06c 1774 seq_printf(m, "%pI4:%u", &sin->sin_addr.s_addr, ntohs(sin->sin_port));
f2739de1
EP
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));
1778 } else {
1779 unsigned int i;
5d6892b3 1780 for (i = 0; i < ctl->addrlen; ++i)
f2739de1
EP
1781 seq_printf(m, "%02x.", ctl->addr.addr[i]);
1782 }
1783
1784 seq_printf(m, " %u %u %d %u %x\n",
1785 ctl->type, ctl->proto,
1786 st->socket != NULL,
1787 ctl->prio, ctl->perm);
1788 }
1789 mutex_unlock(&psb->state_lock);
1790
1791 return 0;
1792}
1793
e5043424
EP
1794static 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,
f2739de1 1803 .show_stats = pohmelfs_show_stats,
e5043424
EP
1804};
1805
b3f08cad
EP
1806/*
1807 * Allocate private superblock and create root dir.
1808 */
1809static int pohmelfs_fill_super(struct super_block *sb, void *data, int silent)
1810{
1811 struct pohmelfs_sb *psb;
1812 int err = -ENOMEM;
1813 struct inode *root;
1814 struct pohmelfs_inode *npi;
1815 struct qstr str;
1816
b3f08cad
EP
1817 psb = kzalloc(sizeof(struct pohmelfs_sb), GFP_KERNEL);
1818 if (!psb)
1819 goto err_out_exit;
1820
182374a0
EP
1821 err = bdi_init(&psb->bdi);
1822 if (err)
1823 goto err_out_free_sb;
1824
1825 err = bdi_register(&psb->bdi, NULL, "pfs-%d", atomic_inc_return(&psb_bdi_num));
1826 if (err) {
1827 bdi_destroy(&psb->bdi);
1828 goto err_out_free_sb;
1829 }
1830
b3f08cad
EP
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;
182374a0 1836 sb->s_bdi = &psb->bdi;
b3f08cad
EP
1837
1838 psb->sb = sb;
1839
1840 psb->ino = 2;
1841 psb->idx = 0;
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);
1849
1850 spin_lock_init(&psb->ino_lock);
1851
1852 INIT_LIST_HEAD(&psb->drop_list);
1853
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);
1858
1859 psb->trans_max_pages = 100;
1860
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);
1871
1872 atomic_set(&psb->trans_gen, 1);
76efa5e3 1873 atomic_long_set(&psb->total_inodes, 0);
b3f08cad
EP
1874
1875 mutex_init(&psb->state_lock);
1876 INIT_LIST_HEAD(&psb->state_list);
1877
e5043424 1878 err = pohmelfs_parse_options((char *) data, psb, 0);
b3f08cad 1879 if (err)
182374a0 1880 goto err_out_free_bdi;
b3f08cad
EP
1881
1882 err = pohmelfs_copy_crypto(psb);
1883 if (err)
182374a0 1884 goto err_out_free_bdi;
b3f08cad
EP
1885
1886 err = pohmelfs_state_init(psb);
1887 if (err)
1888 goto err_out_free_strings;
1889
1890 err = pohmelfs_crypto_init(psb);
1891 if (err)
1892 goto err_out_state_exit;
1893
1894 err = pohmelfs_root_handshake(psb);
1895 if (err)
1896 goto err_out_crypto_exit;
1897
1898 str.name = "/";
1899 str.hash = jhash("/", 1, 0);
1900 str.len = 1;
1901
1902 npi = pohmelfs_create_entry_local(psb, NULL, &str, 0, 0755|S_IFDIR);
1903 if (IS_ERR(npi)) {
1904 err = PTR_ERR(npi);
1905 goto err_out_crypto_exit;
1906 }
872dc5e5
EP
1907 set_bit(NETFS_INODE_REMOTE_SYNCED, &npi->state);
1908 clear_bit(NETFS_INODE_OWNED, &npi->state);
b3f08cad
EP
1909
1910 root = &npi->vfs_inode;
1911
1912 sb->s_root = d_alloc_root(root);
1913 if (!sb->s_root)
1914 goto err_out_put_root;
1915
1916 INIT_DELAYED_WORK(&psb->drop_dwork, pohmelfs_drop_scan);
1917 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1918
1919 INIT_DELAYED_WORK(&psb->dwork, pohmelfs_trans_scan);
1920 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1921
1922 return 0;
1923
1924err_out_put_root:
1925 iput(root);
1926err_out_crypto_exit:
1927 pohmelfs_crypto_exit(psb);
1928err_out_state_exit:
1929 pohmelfs_state_exit(psb);
1930err_out_free_strings:
1931 kfree(psb->cipher_string);
1932 kfree(psb->hash_string);
182374a0
EP
1933err_out_free_bdi:
1934 bdi_destroy(&psb->bdi);
b3f08cad
EP
1935err_out_free_sb:
1936 kfree(psb);
1937err_out_exit:
1938
1939 dprintk("%s: err: %d.\n", __func__, err);
1940 return err;
1941}
1942
1943/*
1944 * Some VFS magic here...
1945 */
1946static int pohmelfs_get_sb(struct file_system_type *fs_type,
1947 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1948{
1949 return get_sb_nodev(fs_type, flags, data, pohmelfs_fill_super,
1950 mnt);
1951}
1952
cee9bb2e
EP
1953/*
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.
1957 */
1958static void pohmelfs_kill_super(struct super_block *sb)
1959{
d8a8559c 1960 sync_inodes_sb(sb);
cee9bb2e
EP
1961 kill_anon_super(sb);
1962}
1963
b3f08cad
EP
1964static struct file_system_type pohmel_fs_type = {
1965 .owner = THIS_MODULE,
1966 .name = "pohmel",
1967 .get_sb = pohmelfs_get_sb,
cee9bb2e 1968 .kill_sb = pohmelfs_kill_super,
b3f08cad
EP
1969};
1970
1971/*
1972 * Cache and module initializations and freeing routings.
1973 */
1974static void pohmelfs_init_once(void *data)
1975{
1976 struct pohmelfs_inode *pi = data;
1977
1978 inode_init_once(&pi->vfs_inode);
1979}
1980
1981static int __init pohmelfs_init_inodecache(void)
1982{
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)
1988 return -ENOMEM;
1989
1990 return 0;
1991}
1992
1993static void pohmelfs_destroy_inodecache(void)
1994{
1995 kmem_cache_destroy(pohmelfs_inode_cache);
1996}
1997
1998static int __init init_pohmel_fs(void)
1999{
2000 int err;
2001
2002 err = pohmelfs_config_init();
2003 if (err)
2004 goto err_out_exit;
2005
2006 err = pohmelfs_init_inodecache();
2007 if (err)
2008 goto err_out_config_exit;
2009
2010 err = pohmelfs_mcache_init();
2011 if (err)
2012 goto err_out_destroy;
2013
2014 err = netfs_trans_init();
2015 if (err)
2016 goto err_out_mcache_exit;
2017
2018 err = register_filesystem(&pohmel_fs_type);
2019 if (err)
2020 goto err_out_trans;
2021
2022 return 0;
2023
2024err_out_trans:
2025 netfs_trans_exit();
2026err_out_mcache_exit:
2027 pohmelfs_mcache_exit();
2028err_out_destroy:
2029 pohmelfs_destroy_inodecache();
2030err_out_config_exit:
2031 pohmelfs_config_exit();
2032err_out_exit:
2033 return err;
2034}
2035
2036static void __exit exit_pohmel_fs(void)
2037{
5d6892b3 2038 unregister_filesystem(&pohmel_fs_type);
b3f08cad
EP
2039 pohmelfs_destroy_inodecache();
2040 pohmelfs_mcache_exit();
2041 pohmelfs_config_exit();
2042 netfs_trans_exit();
2043}
2044
2045module_init(init_pohmel_fs);
2046module_exit(exit_pohmel_fs);
2047
2048MODULE_LICENSE("GPL");
2049MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
2050MODULE_DESCRIPTION("Pohmel filesystem");