?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / drivers / staging / pohmelfs / inode.c
blob687e5d3974d4f2573ae27f79754d38e5ee4d572f
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
33 #include "netfs.h"
34
35 #define POHMELFS_MAGIC_NUM 0x504f482e
36
37 static struct kmem_cache *pohmelfs_inode_cache;
38 static atomic_t psb_bdi_num = ATOMIC_INIT(0);
39
40 /*
41 * Removes inode from all trees, drops local name cache and removes all queued
42 * requests for object removal.
43 */
44 void pohmelfs_inode_del_inode(struct pohmelfs_sb *psb, struct pohmelfs_inode *pi)
45 {
46 mutex_lock(&pi->offset_lock);
47 pohmelfs_free_names(pi);
48 mutex_unlock(&pi->offset_lock);
49
50 dprintk("%s: deleted stuff in ino: %llu.\n", __func__, pi->ino);
51 }
52
53 /*
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.
59 */
60 int pohmelfs_write_inode_create(struct inode *inode, struct netfs_trans *trans)
61 {
62 struct pohmelfs_inode *pi = POHMELFS_I(inode);
63 int err = -ENOMEM, size;
64 struct netfs_cmd *cmd;
65 void *data;
66 int cur_len = netfs_trans_cur_len(trans);
67
68 if (unlikely(cur_len < 0))
69 return -ETOOSMALL;
70
71 cmd = netfs_trans_current(trans);
72 cur_len -= sizeof(struct netfs_cmd);
73
74 data = (void *)(cmd + 1);
75
76 err = pohmelfs_construct_path_string(pi, data, cur_len);
77 if (err < 0)
78 goto err_out_exit;
79
80 size = err;
81
82 cmd->start = i_size_read(inode);
83 cmd->cmd = NETFS_CREATE;
84 cmd->size = size;
85 cmd->id = pi->ino;
86 cmd->ext = inode->i_mode;
87
88 netfs_convert_cmd(cmd);
89
90 netfs_trans_update(cmd, trans, size);
91
92 return 0;
93
94 err_out_exit:
95 printk("%s: completed ino: %llu, err: %d.\n", __func__, pi->ino, err);
96 return err;
97 }
98
99 static int pohmelfs_write_trans_complete(struct page **pages, unsigned int page_num,
100 void *private, int err)
101 {
102 unsigned i;
103
104 dprintk("%s: pages: %lu-%lu, page_num: %u, err: %d.\n",
105 __func__, pages[0]->index, pages[page_num-1]->index,
106 page_num, err);
107
108 for (i = 0; i < page_num; i++) {
109 struct page *page = pages[i];
110
111 if (!page)
112 continue;
113
114 end_page_writeback(page);
115
116 if (err < 0) {
117 SetPageError(page);
118 set_page_dirty(page);
119 }
120
121 unlock_page(page);
122 page_cache_release(page);
123
124 /* dprintk("%s: %3u/%u: page: %p.\n", __func__, i, page_num, page); */
125 }
126 return err;
127 }
128
129 static int pohmelfs_inode_has_dirty_pages(struct address_space *mapping, pgoff_t index)
130 {
131 int ret;
132 struct page *page;
133
134 rcu_read_lock();
135 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
136 (void **)&page, index, 1, PAGECACHE_TAG_DIRTY);
137 rcu_read_unlock();
138 return ret;
139 }
140
141 static int pohmelfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
142 {
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 int err = 0;
147 int done = 0;
148 int nr_pages;
149 pgoff_t index;
150 pgoff_t end; /* Inclusive */
151 int scanned = 0;
152 int range_whole = 0;
153
154 if (wbc->range_cyclic) {
155 index = mapping->writeback_index; /* Start from prev offset */
156 end = -1;
157 } else {
158 index = wbc->range_start >> PAGE_CACHE_SHIFT;
159 end = wbc->range_end >> PAGE_CACHE_SHIFT;
160 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
161 range_whole = 1;
162 scanned = 1;
163 }
164 retry:
165 while (!done && (index <= end)) {
166 unsigned int i = min(end - index, (pgoff_t)psb->trans_max_pages);
167 int path_len;
168 struct netfs_trans *trans;
169
170 err = pohmelfs_inode_has_dirty_pages(mapping, index);
171 if (!err)
172 break;
173
174 err = pohmelfs_path_length(pi);
175 if (err < 0)
176 break;
177
178 path_len = err;
179
180 if (path_len <= 2) {
181 err = -ENOENT;
182 break;
183 }
184
185 trans = netfs_trans_alloc(psb, path_len, 0, i);
186 if (!trans) {
187 err = -ENOMEM;
188 break;
189 }
190 trans->complete = &pohmelfs_write_trans_complete;
191
192 trans->page_num = nr_pages = find_get_pages_tag(mapping, &index,
193 PAGECACHE_TAG_DIRTY, trans->page_num,
194 trans->pages);
195
196 dprintk("%s: t: %p, nr_pages: %u, end: %lu, index: %lu, max: %u.\n",
197 __func__, trans, nr_pages, end, index, trans->page_num);
198
199 if (!nr_pages)
200 goto err_out_reset;
201
202 err = pohmelfs_write_inode_create(inode, trans);
203 if (err)
204 goto err_out_reset;
205
206 err = 0;
207 scanned = 1;
208
209 for (i = 0; i < trans->page_num; i++) {
210 struct page *page = trans->pages[i];
211
212 lock_page(page);
213
214 if (unlikely(page->mapping != mapping))
215 goto out_continue;
216
217 if (!wbc->range_cyclic && page->index > end) {
218 done = 1;
219 goto out_continue;
220 }
221
222 if (wbc->sync_mode != WB_SYNC_NONE)
223 wait_on_page_writeback(page);
224
225 if (PageWriteback(page) ||
226 !clear_page_dirty_for_io(page)) {
227 dprintk("%s: not clear for io page: %p, writeback: %d.\n",
228 __func__, page, PageWriteback(page));
229 goto out_continue;
230 }
231
232 set_page_writeback(page);
233
234 trans->attached_size += page_private(page);
235 trans->attached_pages++;
236 wbc->nr_to_write--;
237
238 if (wbc->nr_to_write <= 0)
239 done = 1;
240
241 continue;
242 out_continue:
243 unlock_page(page);
244 trans->pages[i] = NULL;
245 }
246
247 err = netfs_trans_finish(trans, psb);
248 if (err)
249 break;
250
251 continue;
252
253 err_out_reset:
254 trans->result = err;
255 netfs_trans_reset(trans);
256 netfs_trans_put(trans);
257 break;
258 }
259
260 if (!scanned && !done) {
261 /*
262 * We hit the last page and there is more work to be done: wrap
263 * back to the start of the file
264 */
265 scanned = 1;
266 index = 0;
267 goto retry;
268 }
269
270 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
271 mapping->writeback_index = index;
272
273 return err;
274 }
275
276 /*
277 * Inode writeback creation completion callback.
278 * Only invoked for just created inodes, which do not have pages attached,
279 * like dirs and empty files.
280 */
281 static int pohmelfs_write_inode_complete(struct page **pages, unsigned int page_num,
282 void *private, int err)
283 {
284 struct inode *inode = private;
285 struct pohmelfs_inode *pi = POHMELFS_I(inode);
286
287 if (inode) {
288 if (err) {
289 mark_inode_dirty(inode);
290 clear_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
291 } else {
292 set_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
293 }
294
295 pohmelfs_put_inode(pi);
296 }
297
298 return err;
299 }
300
301 int pohmelfs_write_create_inode(struct pohmelfs_inode *pi)
302 {
303 struct netfs_trans *t;
304 struct inode *inode = &pi->vfs_inode;
305 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
306 int err;
307
308 if (test_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state))
309 return 0;
310
311 dprintk("%s: started ino: %llu.\n", __func__, pi->ino);
312
313 err = pohmelfs_path_length(pi);
314 if (err < 0)
315 goto err_out_exit;
316
317 t = netfs_trans_alloc(psb, err + 1, 0, 0);
318 if (!t) {
319 err = -ENOMEM;
320 goto err_out_exit;
321 }
322 t->complete = pohmelfs_write_inode_complete;
323 t->private = igrab(inode);
324 if (!t->private) {
325 err = -ENOENT;
326 goto err_out_put;
327 }
328
329 err = pohmelfs_write_inode_create(inode, t);
330 if (err)
331 goto err_out_put;
332
333 netfs_trans_finish(t, POHMELFS_SB(inode->i_sb));
334
335 return 0;
336
337 err_out_put:
338 t->result = err;
339 netfs_trans_put(t);
340 err_out_exit:
341 return err;
342 }
343
344 /*
345 * Sync all not-yet-created children in given directory to the server.
346 */
347 static int pohmelfs_write_inode_create_children(struct inode *inode)
348 {
349 struct pohmelfs_inode *parent = POHMELFS_I(inode);
350 struct super_block *sb = inode->i_sb;
351 struct pohmelfs_name *n;
352
353 while (!list_empty(&parent->sync_create_list)) {
354 n = NULL;
355 mutex_lock(&parent->offset_lock);
356 if (!list_empty(&parent->sync_create_list)) {
357 n = list_first_entry(&parent->sync_create_list,
358 struct pohmelfs_name, sync_create_entry);
359 list_del_init(&n->sync_create_entry);
360 }
361 mutex_unlock(&parent->offset_lock);
362
363 if (!n)
364 break;
365
366 inode = ilookup(sb, n->ino);
367
368 dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n",
369 __func__, parent->ino, n->ino, inode);
370
371 if (inode && (inode->i_state & I_DIRTY)) {
372 struct pohmelfs_inode *pi = POHMELFS_I(inode);
373 pohmelfs_write_create_inode(pi);
374 /* pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0); */
375 iput(inode);
376 }
377 }
378
379 return 0;
380 }
381
382 /*
383 * Removes given child from given inode on server.
384 */
385 int pohmelfs_remove_child(struct pohmelfs_inode *pi, struct pohmelfs_name *n)
386 {
387 return pohmelfs_meta_command_data(pi, pi->ino, NETFS_REMOVE, NULL, 0, NULL, NULL, 0);
388 }
389
390 /*
391 * Writeback for given inode.
392 */
393 static int pohmelfs_write_inode(struct inode *inode,
394 struct writeback_control *wbc)
395 {
396 struct pohmelfs_inode *pi = POHMELFS_I(inode);
397
398 pohmelfs_write_create_inode(pi);
399 pohmelfs_write_inode_create_children(inode);
400
401 return 0;
402 }
403
404 /*
405 * It is not exported, sorry...
406 */
407 static inline wait_queue_head_t *page_waitqueue(struct page *page)
408 {
409 const struct zone *zone = page_zone(page);
410
411 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
412 }
413
414 static int pohmelfs_wait_on_page_locked(struct page *page)
415 {
416 struct pohmelfs_sb *psb = POHMELFS_SB(page->mapping->host->i_sb);
417 long ret = psb->wait_on_page_timeout;
418 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
419 int err = 0;
420
421 if (!PageLocked(page))
422 return 0;
423
424 for (;;) {
425 prepare_to_wait(page_waitqueue(page),
426 &wait.wait, TASK_INTERRUPTIBLE);
427
428 dprintk("%s: page: %p, locked: %d, uptodate: %d, error: %d, flags: %lx.\n",
429 __func__, page, PageLocked(page), PageUptodate(page),
430 PageError(page), page->flags);
431
432 if (!PageLocked(page))
433 break;
434
435 if (!signal_pending(current)) {
436 ret = schedule_timeout(ret);
437 if (!ret)
438 break;
439 continue;
440 }
441 ret = -ERESTARTSYS;
442 break;
443 }
444 finish_wait(page_waitqueue(page), &wait.wait);
445
446 if (!ret)
447 err = -ETIMEDOUT;
448
449
450 if (!err)
451 SetPageUptodate(page);
452
453 if (err)
454 printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n",
455 __func__, page, PageUptodate(page), PageLocked(page), err);
456
457 return err;
458 }
459
460 static int pohmelfs_read_page_complete(struct page **pages, unsigned int page_num,
461 void *private, int err)
462 {
463 struct page *page = private;
464
465 if (PageChecked(page))
466 return err;
467
468 if (err < 0) {
469 dprintk("%s: page: %p, err: %d.\n", __func__, page, err);
470 SetPageError(page);
471 }
472
473 unlock_page(page);
474
475 return err;
476 }
477
478 /*
479 * Read a page from remote server.
480 * Function will wait until page is unlocked.
481 */
482 static int pohmelfs_readpage(struct file *file, struct page *page)
483 {
484 struct inode *inode = page->mapping->host;
485 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
486 struct pohmelfs_inode *pi = POHMELFS_I(inode);
487 struct netfs_trans *t;
488 struct netfs_cmd *cmd;
489 int err, path_len;
490 void *data;
491 u64 isize;
492
493 err = pohmelfs_data_lock(pi, page->index << PAGE_CACHE_SHIFT,
494 PAGE_SIZE, POHMELFS_READ_LOCK);
495 if (err)
496 goto err_out_exit;
497
498 isize = i_size_read(inode);
499 if (isize <= page->index << PAGE_CACHE_SHIFT) {
500 SetPageUptodate(page);
501 unlock_page(page);
502 return 0;
503 }
504
505 path_len = pohmelfs_path_length(pi);
506 if (path_len < 0) {
507 err = path_len;
508 goto err_out_exit;
509 }
510
511 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
512 if (!t) {
513 err = -ENOMEM;
514 goto err_out_exit;
515 }
516
517 t->complete = pohmelfs_read_page_complete;
518 t->private = page;
519
520 cmd = netfs_trans_current(t);
521 data = (void *)(cmd + 1);
522
523 err = pohmelfs_construct_path_string(pi, data, path_len);
524 if (err < 0)
525 goto err_out_free;
526
527 path_len = err;
528
529 cmd->id = pi->ino;
530 cmd->start = page->index;
531 cmd->start <<= PAGE_CACHE_SHIFT;
532 cmd->size = PAGE_CACHE_SIZE + path_len;
533 cmd->cmd = NETFS_READ_PAGE;
534 cmd->ext = path_len;
535
536 dprintk("%s: path: '%s', page: %p, ino: %llu, start: %llu, size: %lu.\n",
537 __func__, (char *)data, page, pi->ino, cmd->start, PAGE_CACHE_SIZE);
538
539 netfs_convert_cmd(cmd);
540 netfs_trans_update(cmd, t, path_len);
541
542 err = netfs_trans_finish(t, psb);
543 if (err)
544 goto err_out_return;
545
546 return pohmelfs_wait_on_page_locked(page);
547
548 err_out_free:
549 t->result = err;
550 netfs_trans_put(t);
551 err_out_exit:
552 SetPageError(page);
553 if (PageLocked(page))
554 unlock_page(page);
555 err_out_return:
556 printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n",
557 __func__, page, page->index << PAGE_CACHE_SHIFT, PAGE_CACHE_SIZE, err);
558
559 return err;
560 }
561
562 /*
563 * Write begin/end magic.
564 * Allocates a page and writes inode if it was not synced to server before.
565 */
566 static int pohmelfs_write_begin(struct file *file, struct address_space *mapping,
567 loff_t pos, unsigned len, unsigned flags,
568 struct page **pagep, void **fsdata)
569 {
570 struct inode *inode = mapping->host;
571 struct page *page;
572 pgoff_t index;
573 unsigned start, end;
574 int err;
575
576 *pagep = NULL;
577
578 index = pos >> PAGE_CACHE_SHIFT;
579 start = pos & (PAGE_CACHE_SIZE - 1);
580 end = start + len;
581
582 page = grab_cache_page(mapping, index);
583 if (!page) {
584 err = -ENOMEM;
585 goto err_out_exit;
586 }
587
588 while (!PageUptodate(page)) {
589 if (start && test_bit(NETFS_INODE_REMOTE_SYNCED, &POHMELFS_I(inode)->state)) {
590 err = pohmelfs_readpage(file, page);
591 if (err)
592 goto err_out_exit;
593
594 lock_page(page);
595 continue;
596 }
597
598 if (len != PAGE_CACHE_SIZE) {
599 void *kaddr = kmap_atomic(page, KM_USER0);
600
601 memset(kaddr + start, 0, PAGE_CACHE_SIZE - start);
602 flush_dcache_page(page);
603 kunmap_atomic(kaddr, KM_USER0);
604 }
605 SetPageUptodate(page);
606 }
607
608 set_page_private(page, end);
609
610 *pagep = page;
611
612 return 0;
613
614 err_out_exit:
615 page_cache_release(page);
616 *pagep = NULL;
617
618 return err;
619 }
620
621 static int pohmelfs_write_end(struct file *file, struct address_space *mapping,
622 loff_t pos, unsigned len, unsigned copied,
623 struct page *page, void *fsdata)
624 {
625 struct inode *inode = mapping->host;
626
627 if (copied != len) {
628 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
629 void *kaddr = kmap_atomic(page, KM_USER0);
630
631 memset(kaddr + from + copied, 0, len - copied);
632 flush_dcache_page(page);
633 kunmap_atomic(kaddr, KM_USER0);
634 }
635
636 SetPageUptodate(page);
637 set_page_dirty(page);
638 flush_dcache_page(page);
639
640 unlock_page(page);
641 page_cache_release(page);
642
643 if (pos + copied > inode->i_size) {
644 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
645
646 psb->avail_size -= pos + copied - inode->i_size;
647
648 i_size_write(inode, pos + copied);
649 }
650
651 return copied;
652 }
653
654 static int pohmelfs_readpages_trans_complete(struct page **__pages, unsigned int page_num,
655 void *private, int err)
656 {
657 struct pohmelfs_inode *pi = private;
658 unsigned int i, num;
659 struct page **pages, *page = (struct page *)__pages;
660 loff_t index = page->index;
661
662 pages = kzalloc(sizeof(void *) * page_num, GFP_NOIO);
663 if (!pages)
664 return -ENOMEM;
665
666 num = find_get_pages_contig(pi->vfs_inode.i_mapping, index, page_num, pages);
667 if (num <= 0) {
668 err = num;
669 goto err_out_free;
670 }
671
672 for (i = 0; i < num; ++i) {
673 page = pages[i];
674
675 if (err)
676 printk("%s: %u/%u: page: %p, index: %lu, uptodate: %d, locked: %d, err: %d.\n",
677 __func__, i, num, page, page->index,
678 PageUptodate(page), PageLocked(page), err);
679
680 if (!PageChecked(page)) {
681 if (err < 0)
682 SetPageError(page);
683 unlock_page(page);
684 }
685 page_cache_release(page);
686 page_cache_release(page);
687 }
688
689 err_out_free:
690 kfree(pages);
691 return err;
692 }
693
694 static int pohmelfs_send_readpages(struct pohmelfs_inode *pi, struct page *first, unsigned int num)
695 {
696 struct netfs_trans *t;
697 struct netfs_cmd *cmd;
698 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
699 int err, path_len;
700 void *data;
701
702 err = pohmelfs_data_lock(pi, first->index << PAGE_CACHE_SHIFT,
703 num * PAGE_SIZE, POHMELFS_READ_LOCK);
704 if (err)
705 goto err_out_exit;
706
707 path_len = pohmelfs_path_length(pi);
708 if (path_len < 0) {
709 err = path_len;
710 goto err_out_exit;
711 }
712
713 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
714 if (!t) {
715 err = -ENOMEM;
716 goto err_out_exit;
717 }
718
719 cmd = netfs_trans_current(t);
720 data = (void *)(cmd + 1);
721
722 t->complete = pohmelfs_readpages_trans_complete;
723 t->private = pi;
724 t->page_num = num;
725 t->pages = (struct page **)first;
726
727 err = pohmelfs_construct_path_string(pi, data, path_len);
728 if (err < 0)
729 goto err_out_put;
730
731 path_len = err;
732
733 cmd->cmd = NETFS_READ_PAGES;
734 cmd->start = first->index;
735 cmd->start <<= PAGE_CACHE_SHIFT;
736 cmd->size = (num << 8 | PAGE_CACHE_SHIFT);
737 cmd->id = pi->ino;
738 cmd->ext = path_len;
739
740 dprintk("%s: t: %p, gen: %u, path: '%s', path_len: %u, "
741 "start: %lu, num: %u.\n",
742 __func__, t, t->gen, (char *)data, path_len,
743 first->index, num);
744
745 netfs_convert_cmd(cmd);
746 netfs_trans_update(cmd, t, path_len);
747
748 return netfs_trans_finish(t, psb);
749
750 err_out_put:
751 netfs_trans_free(t);
752 err_out_exit:
753 pohmelfs_readpages_trans_complete((struct page **)first, num, pi, err);
754 return err;
755 }
756
757 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
758
759 static int pohmelfs_readpages(struct file *file, struct address_space *mapping,
760 struct list_head *pages, unsigned nr_pages)
761 {
762 unsigned int page_idx, num = 0;
763 struct page *page = NULL, *first = NULL;
764
765 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
766 page = list_to_page(pages);
767
768 prefetchw(&page->flags);
769 list_del(&page->lru);
770
771 if (!add_to_page_cache_lru(page, mapping,
772 page->index, GFP_KERNEL)) {
773
774 if (!num) {
775 num = 1;
776 first = page;
777 continue;
778 }
779
780 dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n",
781 __func__, page, page->index, first->index);
782
783 if (unlikely(first->index + num != page->index) || (num > 500)) {
784 pohmelfs_send_readpages(POHMELFS_I(mapping->host),
785 first, num);
786 first = page;
787 num = 0;
788 }
789
790 num++;
791 }
792 }
793 pohmelfs_send_readpages(POHMELFS_I(mapping->host), first, num);
794
795 /*
796 * This will be sync read, so when last page is processed,
797 * all previous are alerady unlocked and ready to be used.
798 */
799 return 0;
800 }
801
802 /*
803 * Small address space operations for POHMELFS.
804 */
805 const struct address_space_operations pohmelfs_aops = {
806 .readpage = pohmelfs_readpage,
807 .readpages = pohmelfs_readpages,
808 .writepages = pohmelfs_writepages,
809 .write_begin = pohmelfs_write_begin,
810 .write_end = pohmelfs_write_end,
811 .set_page_dirty = __set_page_dirty_nobuffers,
812 };
813
814 /*
815 * ->detroy_inode() callback. Deletes inode from the caches
816 * and frees private data.
817 */
818 static void pohmelfs_destroy_inode(struct inode *inode)
819 {
820 struct super_block *sb = inode->i_sb;
821 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
822 struct pohmelfs_inode *pi = POHMELFS_I(inode);
823
824 /* pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK); */
825
826 pohmelfs_inode_del_inode(psb, pi);
827
828 dprintk("%s: pi: %p, inode: %p, ino: %llu.\n",
829 __func__, pi, &pi->vfs_inode, pi->ino);
830 kmem_cache_free(pohmelfs_inode_cache, pi);
831 atomic_long_dec(&psb->total_inodes);
832 }
833
834 /*
835 * ->alloc_inode() callback. Allocates inode and initializes private data.
836 */
837 static struct inode *pohmelfs_alloc_inode(struct super_block *sb)
838 {
839 struct pohmelfs_inode *pi;
840
841 pi = kmem_cache_alloc(pohmelfs_inode_cache, GFP_NOIO);
842 if (!pi)
843 return NULL;
844
845 pi->hash_root = RB_ROOT;
846 mutex_init(&pi->offset_lock);
847
848 INIT_LIST_HEAD(&pi->sync_create_list);
849
850 INIT_LIST_HEAD(&pi->inode_entry);
851
852 pi->lock_type = 0;
853 pi->state = 0;
854 pi->total_len = 0;
855 pi->drop_count = 0;
856
857 dprintk("%s: pi: %p, inode: %p.\n", __func__, pi, &pi->vfs_inode);
858
859 atomic_long_inc(&POHMELFS_SB(sb)->total_inodes);
860
861 return &pi->vfs_inode;
862 }
863
864 /*
865 * We want fsync() to work on POHMELFS.
866 */
867 static int pohmelfs_fsync(struct file *file, int datasync)
868 {
869 struct inode *inode = file->f_mapping->host;
870 struct writeback_control wbc = {
871 .sync_mode = WB_SYNC_ALL,
872 .nr_to_write = 0, /* sys_fsync did this */
873 };
874
875 return sync_inode(inode, &wbc);
876 }
877
878 ssize_t pohmelfs_write(struct file *file, const char __user *buf,
879 size_t len, loff_t *ppos)
880 {
881 struct address_space *mapping = file->f_mapping;
882 struct inode *inode = mapping->host;
883 struct pohmelfs_inode *pi = POHMELFS_I(inode);
884 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
885 struct kiocb kiocb;
886 ssize_t ret;
887 loff_t pos = *ppos;
888
889 init_sync_kiocb(&kiocb, file);
890 kiocb.ki_pos = pos;
891 kiocb.ki_left = len;
892
893 dprintk("%s: len: %zu, pos: %llu.\n", __func__, len, pos);
894
895 mutex_lock(&inode->i_mutex);
896 ret = pohmelfs_data_lock(pi, pos, len, POHMELFS_WRITE_LOCK);
897 if (ret)
898 goto err_out_unlock;
899
900 ret = __generic_file_aio_write(&kiocb, &iov, 1, &kiocb.ki_pos);
901 *ppos = kiocb.ki_pos;
902
903 mutex_unlock(&inode->i_mutex);
904 WARN_ON(ret < 0);
905
906 if (ret > 0) {
907 ssize_t err;
908
909 err = generic_write_sync(file, pos, ret);
910 if (err < 0)
911 ret = err;
912 WARN_ON(ret < 0);
913 }
914
915 return ret;
916
917 err_out_unlock:
918 mutex_unlock(&inode->i_mutex);
919 return ret;
920 }
921
922 static const struct file_operations pohmelfs_file_ops = {
923 .open = generic_file_open,
924 .fsync = pohmelfs_fsync,
925
926 .llseek = generic_file_llseek,
927
928 .read = do_sync_read,
929 .aio_read = generic_file_aio_read,
930
931 .mmap = generic_file_mmap,
932
933 .splice_read = generic_file_splice_read,
934 .splice_write = generic_file_splice_write,
935
936 .write = pohmelfs_write,
937 .aio_write = generic_file_aio_write,
938 };
939
940 const struct inode_operations pohmelfs_symlink_inode_operations = {
941 .readlink = generic_readlink,
942 .follow_link = page_follow_link_light,
943 .put_link = page_put_link,
944 };
945
946 int pohmelfs_setattr_raw(struct inode *inode, struct iattr *attr)
947 {
948 int err;
949
950 err = inode_change_ok(inode, attr);
951 if (err) {
952 dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__, POHMELFS_I(inode)->ino);
953 goto err_out_exit;
954 }
955
956 if ((attr->ia_valid & ATTR_SIZE) &&
957 attr->ia_size != i_size_read(inode)) {
958 err = vmtruncate(inode, attr->ia_size);
959 if (err) {
960 dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__, POHMELFS_I(inode)->ino);
961 goto err_out_exit;
962 }
963 }
964
965 setattr_copy(inode, attr);
966 mark_inode_dirty(inode);
967
968 dprintk("%s: ino: %llu, mode: %o -> %o, uid: %u -> %u, gid: %u -> %u, size: %llu -> %llu.\n",
969 __func__, POHMELFS_I(inode)->ino, inode->i_mode, attr->ia_mode,
970 inode->i_uid, attr->ia_uid, inode->i_gid, attr->ia_gid, inode->i_size, attr->ia_size);
971
972 return 0;
973
974 err_out_exit:
975 return err;
976 }
977
978 int pohmelfs_setattr(struct dentry *dentry, struct iattr *attr)
979 {
980 struct inode *inode = dentry->d_inode;
981 struct pohmelfs_inode *pi = POHMELFS_I(inode);
982 int err;
983
984 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
985 if (err)
986 goto err_out_exit;
987
988 err = security_inode_setattr(dentry, attr);
989 if (err)
990 goto err_out_exit;
991
992 err = pohmelfs_setattr_raw(inode, attr);
993 if (err)
994 goto err_out_exit;
995
996 return 0;
997
998 err_out_exit:
999 return err;
1000 }
1001
1002 static int pohmelfs_send_xattr_req(struct pohmelfs_inode *pi, u64 id, u64 start,
1003 const char *name, const void *value, size_t attrsize, int command)
1004 {
1005 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
1006 int err, path_len, namelen = strlen(name) + 1; /* 0-byte */
1007 struct netfs_trans *t;
1008 struct netfs_cmd *cmd;
1009 void *data;
1010
1011 dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n",
1012 __func__, id, start, name, attrsize, command);
1013
1014 path_len = pohmelfs_path_length(pi);
1015 if (path_len < 0) {
1016 err = path_len;
1017 goto err_out_exit;
1018 }
1019
1020 t = netfs_trans_alloc(psb, namelen + path_len + attrsize, 0, 0);
1021 if (!t) {
1022 err = -ENOMEM;
1023 goto err_out_exit;
1024 }
1025
1026 cmd = netfs_trans_current(t);
1027 data = cmd + 1;
1028
1029 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1030 if (path_len < 0) {
1031 err = path_len;
1032 goto err_out_put;
1033 }
1034 data += path_len;
1035
1036 /*
1037 * 'name' is a NUL-terminated string already and
1038 * 'namelen' includes 0-byte.
1039 */
1040 memcpy(data, name, namelen);
1041 data += namelen;
1042
1043 memcpy(data, value, attrsize);
1044
1045 cmd->cmd = command;
1046 cmd->id = id;
1047 cmd->start = start;
1048 cmd->size = attrsize + namelen + path_len;
1049 cmd->ext = path_len;
1050 cmd->csize = 0;
1051 cmd->cpad = 0;
1052
1053 netfs_convert_cmd(cmd);
1054 netfs_trans_update(cmd, t, namelen + path_len + attrsize);
1055
1056 return netfs_trans_finish(t, psb);
1057
1058 err_out_put:
1059 t->result = err;
1060 netfs_trans_put(t);
1061 err_out_exit:
1062 return err;
1063 }
1064
1065 static int pohmelfs_setxattr(struct dentry *dentry, const char *name,
1066 const void *value, size_t attrsize, int flags)
1067 {
1068 struct inode *inode = dentry->d_inode;
1069 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1070 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1071
1072 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1073 return -EOPNOTSUPP;
1074
1075 return pohmelfs_send_xattr_req(pi, flags, attrsize, name,
1076 value, attrsize, NETFS_XATTR_SET);
1077 }
1078
1079 static ssize_t pohmelfs_getxattr(struct dentry *dentry, const char *name,
1080 void *value, size_t attrsize)
1081 {
1082 struct inode *inode = dentry->d_inode;
1083 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1084 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1085 struct pohmelfs_mcache *m;
1086 int err;
1087 long timeout = psb->mcache_timeout;
1088
1089 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1090 return -EOPNOTSUPP;
1091
1092 m = pohmelfs_mcache_alloc(psb, 0, attrsize, value);
1093 if (IS_ERR(m))
1094 return PTR_ERR(m);
1095
1096 dprintk("%s: ino: %llu, name: '%s', size: %zu.\n",
1097 __func__, pi->ino, name, attrsize);
1098
1099 err = pohmelfs_send_xattr_req(pi, m->gen, attrsize, name, value, 0, NETFS_XATTR_GET);
1100 if (err)
1101 goto err_out_put;
1102
1103 do {
1104 err = wait_for_completion_timeout(&m->complete, timeout);
1105 if (err) {
1106 err = m->err;
1107 break;
1108 }
1109
1110 /*
1111 * This loop is a bit ugly, since it waits until reference counter
1112 * hits 1 and then put object here. Main goal is to prevent race with
1113 * network thread, when it can start processing given request, i.e.
1114 * increase its reference counter but yet not complete it, while
1115 * we will exit from ->getxattr() with timeout, and although request
1116 * will not be freed (its reference counter was increased by network
1117 * thread), data pointer provided by user may be released, so we will
1118 * overwrite already freed area in network thread.
1119 *
1120 * Now after timeout we remove request from the cache, so it can not be
1121 * found by network thread, and wait for its reference counter to hit 1,
1122 * i.e. if network thread already started to process this request, we wait
1123 * it to finish, and then free object locally. If reference counter is
1124 * already 1, i.e. request is not used by anyone else, we can free it without
1125 * problem.
1126 */
1127 err = -ETIMEDOUT;
1128 timeout = HZ;
1129
1130 pohmelfs_mcache_remove_locked(psb, m);
1131 } while (atomic_read(&m->refcnt) != 1);
1132
1133 pohmelfs_mcache_put(psb, m);
1134
1135 dprintk("%s: ino: %llu, err: %d.\n", __func__, pi->ino, err);
1136
1137 return err;
1138
1139 err_out_put:
1140 pohmelfs_mcache_put(psb, m);
1141 return err;
1142 }
1143
1144 static int pohmelfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1145 {
1146 struct inode *inode = dentry->d_inode;
1147
1148 generic_fillattr(inode, stat);
1149 return 0;
1150 }
1151
1152 const struct inode_operations pohmelfs_file_inode_operations = {
1153 .setattr = pohmelfs_setattr,
1154 .getattr = pohmelfs_getattr,
1155 .setxattr = pohmelfs_setxattr,
1156 .getxattr = pohmelfs_getxattr,
1157 };
1158
1159 /*
1160 * Fill inode data: mode, size, operation callbacks and so on...
1161 */
1162 void pohmelfs_fill_inode(struct inode *inode, struct netfs_inode_info *info)
1163 {
1164 inode->i_mode = info->mode;
1165 inode->i_nlink = info->nlink;
1166 inode->i_uid = info->uid;
1167 inode->i_gid = info->gid;
1168 inode->i_blocks = info->blocks;
1169 inode->i_rdev = info->rdev;
1170 inode->i_size = info->size;
1171 inode->i_version = info->version;
1172 inode->i_blkbits = ffs(info->blocksize);
1173
1174 dprintk("%s: inode: %p, num: %lu/%llu inode is regular: %d, dir: %d, link: %d, mode: %o, size: %llu.\n",
1175 __func__, inode, inode->i_ino, info->ino,
1176 S_ISREG(inode->i_mode), S_ISDIR(inode->i_mode),
1177 S_ISLNK(inode->i_mode), inode->i_mode, inode->i_size);
1178
1179 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1180
1181 /*
1182 * i_mapping is a pointer to i_data during inode initialization.
1183 */
1184 inode->i_data.a_ops = &pohmelfs_aops;
1185
1186 if (S_ISREG(inode->i_mode)) {
1187 inode->i_fop = &pohmelfs_file_ops;
1188 inode->i_op = &pohmelfs_file_inode_operations;
1189 } else if (S_ISDIR(inode->i_mode)) {
1190 inode->i_fop = &pohmelfs_dir_fops;
1191 inode->i_op = &pohmelfs_dir_inode_ops;
1192 } else if (S_ISLNK(inode->i_mode)) {
1193 inode->i_op = &pohmelfs_symlink_inode_operations;
1194 inode->i_fop = &pohmelfs_file_ops;
1195 } else {
1196 inode->i_fop = &generic_ro_fops;
1197 }
1198 }
1199
1200 static int pohmelfs_drop_inode(struct inode *inode)
1201 {
1202 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1203 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1204
1205 spin_lock(&psb->ino_lock);
1206 list_del_init(&pi->inode_entry);
1207 spin_unlock(&psb->ino_lock);
1208
1209 return generic_drop_inode(inode);
1210 }
1211
1212 static struct pohmelfs_inode *pohmelfs_get_inode_from_list(struct pohmelfs_sb *psb,
1213 struct list_head *head, unsigned int *count)
1214 {
1215 struct pohmelfs_inode *pi = NULL;
1216
1217 spin_lock(&psb->ino_lock);
1218 if (!list_empty(head)) {
1219 pi = list_entry(head->next, struct pohmelfs_inode,
1220 inode_entry);
1221 list_del_init(&pi->inode_entry);
1222 *count = pi->drop_count;
1223 pi->drop_count = 0;
1224 }
1225 spin_unlock(&psb->ino_lock);
1226
1227 return pi;
1228 }
1229
1230 static void pohmelfs_flush_transactions(struct pohmelfs_sb *psb)
1231 {
1232 struct pohmelfs_config *c;
1233
1234 mutex_lock(&psb->state_lock);
1235 list_for_each_entry(c, &psb->state_list, config_entry) {
1236 pohmelfs_state_flush_transactions(&c->state);
1237 }
1238 mutex_unlock(&psb->state_lock);
1239 }
1240
1241 /*
1242 * ->put_super() callback. Invoked before superblock is destroyed,
1243 * so it has to clean all private data.
1244 */
1245 static void pohmelfs_put_super(struct super_block *sb)
1246 {
1247 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1248 struct pohmelfs_inode *pi;
1249 unsigned int count = 0;
1250 unsigned int in_drop_list = 0;
1251 struct inode *inode, *tmp;
1252
1253 dprintk("%s.\n", __func__);
1254
1255 /*
1256 * Kill pending transactions, which could affect inodes in-flight.
1257 */
1258 pohmelfs_flush_transactions(psb);
1259
1260 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) {
1261 inode = &pi->vfs_inode;
1262
1263 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1264 __func__, pi->ino, pi, inode, count);
1265
1266 if (atomic_read(&inode->i_count) != count) {
1267 printk("%s: ino: %llu, pi: %p, inode: %p, count: %u, i_count: %d.\n",
1268 __func__, pi->ino, pi, inode, count,
1269 atomic_read(&inode->i_count));
1270 count = atomic_read(&inode->i_count);
1271 in_drop_list++;
1272 }
1273
1274 while (count--)
1275 iput(&pi->vfs_inode);
1276 }
1277
1278 list_for_each_entry_safe(inode, tmp, &sb->s_inodes, i_sb_list) {
1279 pi = POHMELFS_I(inode);
1280
1281 dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n",
1282 __func__, pi->ino, pi, inode, atomic_read(&inode->i_count));
1283
1284 /*
1285 * These are special inodes, they were created during
1286 * directory reading or lookup, and were not bound to dentry,
1287 * so they live here with reference counter being 1 and prevent
1288 * umount from succeed since it believes that they are busy.
1289 */
1290 count = atomic_read(&inode->i_count);
1291 if (count) {
1292 list_del_init(&inode->i_sb_list);
1293 while (count--)
1294 iput(&pi->vfs_inode);
1295 }
1296 }
1297
1298 psb->trans_scan_timeout = psb->drop_scan_timeout = 0;
1299 cancel_rearming_delayed_work(&psb->dwork);
1300 cancel_rearming_delayed_work(&psb->drop_dwork);
1301 flush_scheduled_work();
1302
1303 dprintk("%s: stopped workqueues.\n", __func__);
1304
1305 pohmelfs_crypto_exit(psb);
1306 pohmelfs_state_exit(psb);
1307
1308 bdi_destroy(&psb->bdi);
1309
1310 kfree(psb);
1311 sb->s_fs_info = NULL;
1312 }
1313
1314 static int pohmelfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1315 {
1316 struct super_block *sb = dentry->d_sb;
1317 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1318
1319 /*
1320 * There are no filesystem size limits yet.
1321 */
1322 memset(buf, 0, sizeof(struct kstatfs));
1323
1324 buf->f_type = POHMELFS_MAGIC_NUM; /* 'POH.' */
1325 buf->f_bsize = sb->s_blocksize;
1326 buf->f_files = psb->ino;
1327 buf->f_namelen = 255;
1328 buf->f_files = atomic_long_read(&psb->total_inodes);
1329 buf->f_bfree = buf->f_bavail = psb->avail_size >> PAGE_SHIFT;
1330 buf->f_blocks = psb->total_size >> PAGE_SHIFT;
1331
1332 dprintk("%s: total: %llu, avail: %llu, inodes: %llu, bsize: %lu.\n",
1333 __func__, psb->total_size, psb->avail_size, buf->f_files, sb->s_blocksize);
1334
1335 return 0;
1336 }
1337
1338 static int pohmelfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
1339 {
1340 struct pohmelfs_sb *psb = POHMELFS_SB(vfs->mnt_sb);
1341
1342 seq_printf(seq, ",idx=%u", psb->idx);
1343 seq_printf(seq, ",trans_scan_timeout=%u", jiffies_to_msecs(psb->trans_scan_timeout));
1344 seq_printf(seq, ",drop_scan_timeout=%u", jiffies_to_msecs(psb->drop_scan_timeout));
1345 seq_printf(seq, ",wait_on_page_timeout=%u", jiffies_to_msecs(psb->wait_on_page_timeout));
1346 seq_printf(seq, ",trans_retries=%u", psb->trans_retries);
1347 seq_printf(seq, ",crypto_thread_num=%u", psb->crypto_thread_num);
1348 seq_printf(seq, ",trans_max_pages=%u", psb->trans_max_pages);
1349 seq_printf(seq, ",mcache_timeout=%u", jiffies_to_msecs(psb->mcache_timeout));
1350 if (psb->crypto_fail_unsupported)
1351 seq_printf(seq, ",crypto_fail_unsupported");
1352
1353 return 0;
1354 }
1355
1356 enum {
1357 pohmelfs_opt_idx,
1358 pohmelfs_opt_crypto_thread_num,
1359 pohmelfs_opt_trans_max_pages,
1360 pohmelfs_opt_crypto_fail_unsupported,
1361
1362 /* Remountable options */
1363 pohmelfs_opt_trans_scan_timeout,
1364 pohmelfs_opt_drop_scan_timeout,
1365 pohmelfs_opt_wait_on_page_timeout,
1366 pohmelfs_opt_trans_retries,
1367 pohmelfs_opt_mcache_timeout,
1368 };
1369
1370 static struct match_token pohmelfs_tokens[] = {
1371 {pohmelfs_opt_idx, "idx=%u"},
1372 {pohmelfs_opt_crypto_thread_num, "crypto_thread_num=%u"},
1373 {pohmelfs_opt_trans_max_pages, "trans_max_pages=%u"},
1374 {pohmelfs_opt_crypto_fail_unsupported, "crypto_fail_unsupported"},
1375 {pohmelfs_opt_trans_scan_timeout, "trans_scan_timeout=%u"},
1376 {pohmelfs_opt_drop_scan_timeout, "drop_scan_timeout=%u"},
1377 {pohmelfs_opt_wait_on_page_timeout, "wait_on_page_timeout=%u"},
1378 {pohmelfs_opt_trans_retries, "trans_retries=%u"},
1379 {pohmelfs_opt_mcache_timeout, "mcache_timeout=%u"},
1380 };
1381
1382 static int pohmelfs_parse_options(char *options, struct pohmelfs_sb *psb, int remount)
1383 {
1384 char *p;
1385 substring_t args[MAX_OPT_ARGS];
1386 int option, err;
1387
1388 if (!options)
1389 return 0;
1390
1391 while ((p = strsep(&options, ",")) != NULL) {
1392 int token;
1393 if (!*p)
1394 continue;
1395
1396 token = match_token(p, pohmelfs_tokens, args);
1397
1398 err = match_int(&args[0], &option);
1399 if (err)
1400 return err;
1401
1402 if (remount && token <= pohmelfs_opt_crypto_fail_unsupported)
1403 continue;
1404
1405 switch (token) {
1406 case pohmelfs_opt_idx:
1407 psb->idx = option;
1408 break;
1409 case pohmelfs_opt_trans_scan_timeout:
1410 psb->trans_scan_timeout = msecs_to_jiffies(option);
1411 break;
1412 case pohmelfs_opt_drop_scan_timeout:
1413 psb->drop_scan_timeout = msecs_to_jiffies(option);
1414 break;
1415 case pohmelfs_opt_wait_on_page_timeout:
1416 psb->wait_on_page_timeout = msecs_to_jiffies(option);
1417 break;
1418 case pohmelfs_opt_mcache_timeout:
1419 psb->mcache_timeout = msecs_to_jiffies(option);
1420 break;
1421 case pohmelfs_opt_trans_retries:
1422 psb->trans_retries = option;
1423 break;
1424 case pohmelfs_opt_crypto_thread_num:
1425 psb->crypto_thread_num = option;
1426 break;
1427 case pohmelfs_opt_trans_max_pages:
1428 psb->trans_max_pages = option;
1429 break;
1430 case pohmelfs_opt_crypto_fail_unsupported:
1431 psb->crypto_fail_unsupported = 1;
1432 break;
1433 default:
1434 return -EINVAL;
1435 }
1436 }
1437
1438 return 0;
1439 }
1440
1441 static int pohmelfs_remount(struct super_block *sb, int *flags, char *data)
1442 {
1443 int err;
1444 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1445 unsigned long old_sb_flags = sb->s_flags;
1446
1447 err = pohmelfs_parse_options(data, psb, 1);
1448 if (err)
1449 goto err_out_restore;
1450
1451 if (!(*flags & MS_RDONLY))
1452 sb->s_flags &= ~MS_RDONLY;
1453 return 0;
1454
1455 err_out_restore:
1456 sb->s_flags = old_sb_flags;
1457 return err;
1458 }
1459
1460 static void pohmelfs_flush_inode(struct pohmelfs_inode *pi, unsigned int count)
1461 {
1462 struct inode *inode = &pi->vfs_inode;
1463
1464 dprintk("%s: %p: ino: %llu, owned: %d.\n",
1465 __func__, inode, pi->ino, test_bit(NETFS_INODE_OWNED, &pi->state));
1466
1467 mutex_lock(&inode->i_mutex);
1468 if (test_and_clear_bit(NETFS_INODE_OWNED, &pi->state)) {
1469 filemap_fdatawrite(inode->i_mapping);
1470 inode->i_sb->s_op->write_inode(inode, 0);
1471 }
1472
1473 #ifdef POHMELFS_TRUNCATE_ON_INODE_FLUSH
1474 truncate_inode_pages(inode->i_mapping, 0);
1475 #endif
1476
1477 pohmelfs_data_unlock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1478 mutex_unlock(&inode->i_mutex);
1479 }
1480
1481 static void pohmelfs_put_inode_count(struct pohmelfs_inode *pi, unsigned int count)
1482 {
1483 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1484 __func__, pi->ino, pi, &pi->vfs_inode, count);
1485
1486 if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH, &pi->state))
1487 pohmelfs_flush_inode(pi, count);
1488
1489 while (count--)
1490 iput(&pi->vfs_inode);
1491 }
1492
1493 static void pohmelfs_drop_scan(struct work_struct *work)
1494 {
1495 struct pohmelfs_sb *psb =
1496 container_of(work, struct pohmelfs_sb, drop_dwork.work);
1497 struct pohmelfs_inode *pi;
1498 unsigned int count = 0;
1499
1500 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count)))
1501 pohmelfs_put_inode_count(pi, count);
1502
1503 pohmelfs_check_states(psb);
1504
1505 if (psb->drop_scan_timeout)
1506 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1507 }
1508
1509 /*
1510 * Run through all transactions starting from the oldest,
1511 * drop transaction from current state and try to send it
1512 * to all remote nodes, which are currently installed.
1513 */
1514 static void pohmelfs_trans_scan_state(struct netfs_state *st)
1515 {
1516 struct rb_node *rb_node;
1517 struct netfs_trans_dst *dst;
1518 struct pohmelfs_sb *psb = st->psb;
1519 unsigned int timeout = psb->trans_scan_timeout;
1520 struct netfs_trans *t;
1521 int err;
1522
1523 mutex_lock(&st->trans_lock);
1524 for (rb_node = rb_first(&st->trans_root); rb_node; ) {
1525 dst = rb_entry(rb_node, struct netfs_trans_dst, state_entry);
1526 t = dst->trans;
1527
1528 if (timeout && time_after(dst->send_time + timeout, jiffies)
1529 && dst->retries == 0)
1530 break;
1531
1532 dprintk("%s: t: %p, gen: %u, st: %p, retries: %u, max: %u.\n",
1533 __func__, t, t->gen, st, dst->retries, psb->trans_retries);
1534 netfs_trans_get(t);
1535
1536 rb_node = rb_next(rb_node);
1537
1538 err = -ETIMEDOUT;
1539 if (timeout && (++dst->retries < psb->trans_retries))
1540 err = netfs_trans_resend(t, psb);
1541
1542 if (err || (t->flags & NETFS_TRANS_SINGLE_DST)) {
1543 if (netfs_trans_remove_nolock(dst, st))
1544 netfs_trans_drop_dst_nostate(dst);
1545 }
1546
1547 t->result = err;
1548 netfs_trans_put(t);
1549 }
1550 mutex_unlock(&st->trans_lock);
1551 }
1552
1553 /*
1554 * Walk through all installed network states and resend all
1555 * transactions, which are old enough.
1556 */
1557 static void pohmelfs_trans_scan(struct work_struct *work)
1558 {
1559 struct pohmelfs_sb *psb =
1560 container_of(work, struct pohmelfs_sb, dwork.work);
1561 struct netfs_state *st;
1562 struct pohmelfs_config *c;
1563
1564 mutex_lock(&psb->state_lock);
1565 list_for_each_entry(c, &psb->state_list, config_entry) {
1566 st = &c->state;
1567
1568 pohmelfs_trans_scan_state(st);
1569 }
1570 mutex_unlock(&psb->state_lock);
1571
1572 /*
1573 * If no timeout specified then system is in the middle of umount process,
1574 * so no need to reschedule scanning process again.
1575 */
1576 if (psb->trans_scan_timeout)
1577 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1578 }
1579
1580 int pohmelfs_meta_command_data(struct pohmelfs_inode *pi, u64 id, unsigned int cmd_op, char *addon,
1581 unsigned int flags, netfs_trans_complete_t complete, void *priv, u64 start)
1582 {
1583 struct inode *inode = &pi->vfs_inode;
1584 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1585 int err = 0, sz;
1586 struct netfs_trans *t;
1587 int path_len, addon_len = 0;
1588 void *data;
1589 struct netfs_inode_info *info;
1590 struct netfs_cmd *cmd;
1591
1592 dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__, pi->ino, cmd_op, addon);
1593
1594 path_len = pohmelfs_path_length(pi);
1595 if (path_len < 0) {
1596 err = path_len;
1597 goto err_out_exit;
1598 }
1599
1600 if (addon)
1601 addon_len = strlen(addon) + 1; /* 0-byte */
1602 sz = addon_len;
1603
1604 if (cmd_op == NETFS_INODE_INFO)
1605 sz += sizeof(struct netfs_inode_info);
1606
1607 t = netfs_trans_alloc(psb, sz + path_len, flags, 0);
1608 if (!t) {
1609 err = -ENOMEM;
1610 goto err_out_exit;
1611 }
1612 t->complete = complete;
1613 t->private = priv;
1614
1615 cmd = netfs_trans_current(t);
1616 data = (void *)(cmd + 1);
1617
1618 if (cmd_op == NETFS_INODE_INFO) {
1619 info = (struct netfs_inode_info *)(cmd + 1);
1620 data = (void *)(info + 1);
1621
1622 /*
1623 * We are under i_mutex, can read and change whatever we want...
1624 */
1625 info->mode = inode->i_mode;
1626 info->nlink = inode->i_nlink;
1627 info->uid = inode->i_uid;
1628 info->gid = inode->i_gid;
1629 info->blocks = inode->i_blocks;
1630 info->rdev = inode->i_rdev;
1631 info->size = inode->i_size;
1632 info->version = inode->i_version;
1633
1634 netfs_convert_inode_info(info);
1635 }
1636
1637 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1638 if (path_len < 0)
1639 goto err_out_free;
1640
1641 dprintk("%s: path_len: %d.\n", __func__, path_len);
1642
1643 if (addon) {
1644 path_len--; /* Do not place null-byte before the addon */
1645 path_len += sprintf(data + path_len, "/%s", addon) + 1; /* 0 - byte */
1646 }
1647
1648 sz += path_len;
1649
1650 cmd->cmd = cmd_op;
1651 cmd->ext = path_len;
1652 cmd->size = sz;
1653 cmd->id = id;
1654 cmd->start = start;
1655
1656 netfs_convert_cmd(cmd);
1657 netfs_trans_update(cmd, t, sz);
1658
1659 /*
1660 * Note, that it is possible to leak error here: transaction callback will not
1661 * be invoked for allocation path failure.
1662 */
1663 return netfs_trans_finish(t, psb);
1664
1665 err_out_free:
1666 netfs_trans_free(t);
1667 err_out_exit:
1668 if (complete)
1669 complete(NULL, 0, priv, err);
1670 return err;
1671 }
1672
1673 int pohmelfs_meta_command(struct pohmelfs_inode *pi, unsigned int cmd_op, unsigned int flags,
1674 netfs_trans_complete_t complete, void *priv, u64 start)
1675 {
1676 return pohmelfs_meta_command_data(pi, pi->ino, cmd_op, NULL, flags, complete, priv, start);
1677 }
1678
1679 /*
1680 * Send request and wait for POHMELFS root capabilities response,
1681 * which will update server's informaion about size of the export,
1682 * permissions, number of objects, available size and so on.
1683 */
1684 static int pohmelfs_root_handshake(struct pohmelfs_sb *psb)
1685 {
1686 struct netfs_trans *t;
1687 struct netfs_cmd *cmd;
1688 int err = -ENOMEM;
1689
1690 t = netfs_trans_alloc(psb, 0, 0, 0);
1691 if (!t)
1692 goto err_out_exit;
1693
1694 cmd = netfs_trans_current(t);
1695
1696 cmd->cmd = NETFS_CAPABILITIES;
1697 cmd->id = POHMELFS_ROOT_CAPABILITIES;
1698 cmd->size = 0;
1699 cmd->start = 0;
1700 cmd->ext = 0;
1701 cmd->csize = 0;
1702
1703 netfs_convert_cmd(cmd);
1704 netfs_trans_update(cmd, t, 0);
1705
1706 err = netfs_trans_finish(t, psb);
1707 if (err)
1708 goto err_out_exit;
1709
1710 psb->flags = ~0;
1711 err = wait_event_interruptible_timeout(psb->wait,
1712 (psb->flags != ~0),
1713 psb->wait_on_page_timeout);
1714 if (!err)
1715 err = -ETIMEDOUT;
1716 else if (err > 0)
1717 err = -psb->flags;
1718
1719 if (err)
1720 goto err_out_exit;
1721
1722 return 0;
1723
1724 err_out_exit:
1725 return err;
1726 }
1727
1728 static int pohmelfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
1729 {
1730 struct netfs_state *st;
1731 struct pohmelfs_ctl *ctl;
1732 struct pohmelfs_sb *psb = POHMELFS_SB(mnt->mnt_sb);
1733 struct pohmelfs_config *c;
1734
1735 mutex_lock(&psb->state_lock);
1736
1737 seq_printf(m, "\nidx addr(:port) socket_type protocol active priority permissions\n");
1738
1739 list_for_each_entry(c, &psb->state_list, config_entry) {
1740 st = &c->state;
1741 ctl = &st->ctl;
1742
1743 seq_printf(m, "%u ", ctl->idx);
1744 if (ctl->addr.sa_family == AF_INET) {
1745 struct sockaddr_in *sin = (struct sockaddr_in *)&st->ctl.addr;
1746 seq_printf(m, "%pI4:%u", &sin->sin_addr.s_addr, ntohs(sin->sin_port));
1747 } else if (ctl->addr.sa_family == AF_INET6) {
1748 struct sockaddr_in6 *sin = (struct sockaddr_in6 *)&st->ctl.addr;
1749 seq_printf(m, "%pi6:%u", &sin->sin6_addr, ntohs(sin->sin6_port));
1750 } else {
1751 unsigned int i;
1752 for (i = 0; i < ctl->addrlen; ++i)
1753 seq_printf(m, "%02x.", ctl->addr.addr[i]);
1754 }
1755
1756 seq_printf(m, " %u %u %d %u %x\n",
1757 ctl->type, ctl->proto,
1758 st->socket != NULL,
1759 ctl->prio, ctl->perm);
1760 }
1761 mutex_unlock(&psb->state_lock);
1762
1763 return 0;
1764 }
1765
1766 static const struct super_operations pohmelfs_sb_ops = {
1767 .alloc_inode = pohmelfs_alloc_inode,
1768 .destroy_inode = pohmelfs_destroy_inode,
1769 .drop_inode = pohmelfs_drop_inode,
1770 .write_inode = pohmelfs_write_inode,
1771 .put_super = pohmelfs_put_super,
1772 .remount_fs = pohmelfs_remount,
1773 .statfs = pohmelfs_statfs,
1774 .show_options = pohmelfs_show_options,
1775 .show_stats = pohmelfs_show_stats,
1776 };
1777
1778 /*
1779 * Allocate private superblock and create root dir.
1780 */
1781 static int pohmelfs_fill_super(struct super_block *sb, void *data, int silent)
1782 {
1783 struct pohmelfs_sb *psb;
1784 int err = -ENOMEM;
1785 struct inode *root;
1786 struct pohmelfs_inode *npi;
1787 struct qstr str;
1788
1789 psb = kzalloc(sizeof(struct pohmelfs_sb), GFP_KERNEL);
1790 if (!psb)
1791 goto err_out_exit;
1792
1793 err = bdi_init(&psb->bdi);
1794 if (err)
1795 goto err_out_free_sb;
1796
1797 err = bdi_register(&psb->bdi, NULL, "pfs-%d", atomic_inc_return(&psb_bdi_num));
1798 if (err) {
1799 bdi_destroy(&psb->bdi);
1800 goto err_out_free_sb;
1801 }
1802
1803 sb->s_fs_info = psb;
1804 sb->s_op = &pohmelfs_sb_ops;
1805 sb->s_magic = POHMELFS_MAGIC_NUM;
1806 sb->s_maxbytes = MAX_LFS_FILESIZE;
1807 sb->s_blocksize = PAGE_SIZE;
1808 sb->s_bdi = &psb->bdi;
1809
1810 psb->sb = sb;
1811
1812 psb->ino = 2;
1813 psb->idx = 0;
1814 psb->active_state = NULL;
1815 psb->trans_retries = 5;
1816 psb->trans_data_size = PAGE_SIZE;
1817 psb->drop_scan_timeout = msecs_to_jiffies(1000);
1818 psb->trans_scan_timeout = msecs_to_jiffies(5000);
1819 psb->wait_on_page_timeout = msecs_to_jiffies(5000);
1820 init_waitqueue_head(&psb->wait);
1821
1822 spin_lock_init(&psb->ino_lock);
1823
1824 INIT_LIST_HEAD(&psb->drop_list);
1825
1826 mutex_init(&psb->mcache_lock);
1827 psb->mcache_root = RB_ROOT;
1828 psb->mcache_timeout = msecs_to_jiffies(5000);
1829 atomic_long_set(&psb->mcache_gen, 0);
1830
1831 psb->trans_max_pages = 100;
1832
1833 psb->crypto_align_size = 16;
1834 psb->crypto_attached_size = 0;
1835 psb->hash_strlen = 0;
1836 psb->cipher_strlen = 0;
1837 psb->perform_crypto = 0;
1838 psb->crypto_thread_num = 2;
1839 psb->crypto_fail_unsupported = 0;
1840 mutex_init(&psb->crypto_thread_lock);
1841 INIT_LIST_HEAD(&psb->crypto_ready_list);
1842 INIT_LIST_HEAD(&psb->crypto_active_list);
1843
1844 atomic_set(&psb->trans_gen, 1);
1845 atomic_long_set(&psb->total_inodes, 0);
1846
1847 mutex_init(&psb->state_lock);
1848 INIT_LIST_HEAD(&psb->state_list);
1849
1850 err = pohmelfs_parse_options((char *) data, psb, 0);
1851 if (err)
1852 goto err_out_free_bdi;
1853
1854 err = pohmelfs_copy_crypto(psb);
1855 if (err)
1856 goto err_out_free_bdi;
1857
1858 err = pohmelfs_state_init(psb);
1859 if (err)
1860 goto err_out_free_strings;
1861
1862 err = pohmelfs_crypto_init(psb);
1863 if (err)
1864 goto err_out_state_exit;
1865
1866 err = pohmelfs_root_handshake(psb);
1867 if (err)
1868 goto err_out_crypto_exit;
1869
1870 str.name = "/";
1871 str.hash = jhash("/", 1, 0);
1872 str.len = 1;
1873
1874 npi = pohmelfs_create_entry_local(psb, NULL, &str, 0, 0755|S_IFDIR);
1875 if (IS_ERR(npi)) {
1876 err = PTR_ERR(npi);
1877 goto err_out_crypto_exit;
1878 }
1879 set_bit(NETFS_INODE_REMOTE_SYNCED, &npi->state);
1880 clear_bit(NETFS_INODE_OWNED, &npi->state);
1881
1882 root = &npi->vfs_inode;
1883
1884 sb->s_root = d_alloc_root(root);
1885 if (!sb->s_root)
1886 goto err_out_put_root;
1887
1888 INIT_DELAYED_WORK(&psb->drop_dwork, pohmelfs_drop_scan);
1889 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1890
1891 INIT_DELAYED_WORK(&psb->dwork, pohmelfs_trans_scan);
1892 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1893
1894 return 0;
1895
1896 err_out_put_root:
1897 iput(root);
1898 err_out_crypto_exit:
1899 pohmelfs_crypto_exit(psb);
1900 err_out_state_exit:
1901 pohmelfs_state_exit(psb);
1902 err_out_free_strings:
1903 kfree(psb->cipher_string);
1904 kfree(psb->hash_string);
1905 err_out_free_bdi:
1906 bdi_destroy(&psb->bdi);
1907 err_out_free_sb:
1908 kfree(psb);
1909 err_out_exit:
1910
1911 dprintk("%s: err: %d.\n", __func__, err);
1912 return err;
1913 }
1914
1915 /*
1916 * Some VFS magic here...
1917 */
1918 static int pohmelfs_get_sb(struct file_system_type *fs_type,
1919 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1920 {
1921 return get_sb_nodev(fs_type, flags, data, pohmelfs_fill_super,
1922 mnt);
1923 }
1924
1925 /*
1926 * We need this to sync all inodes earlier, since when writeback
1927 * is invoked from the umount/mntput path dcache is already shrunk,
1928 * see generic_shutdown_super(), and no inodes can access the path.
1929 */
1930 static void pohmelfs_kill_super(struct super_block *sb)
1931 {
1932 sync_inodes_sb(sb);
1933 kill_anon_super(sb);
1934 }
1935
1936 static struct file_system_type pohmel_fs_type = {
1937 .owner = THIS_MODULE,
1938 .name = "pohmel",
1939 .get_sb = pohmelfs_get_sb,
1940 .kill_sb = pohmelfs_kill_super,
1941 };
1942
1943 /*
1944 * Cache and module initializations and freeing routings.
1945 */
1946 static void pohmelfs_init_once(void *data)
1947 {
1948 struct pohmelfs_inode *pi = data;
1949
1950 inode_init_once(&pi->vfs_inode);
1951 }
1952
1953 static int __init pohmelfs_init_inodecache(void)
1954 {
1955 pohmelfs_inode_cache = kmem_cache_create("pohmelfs_inode_cache",
1956 sizeof(struct pohmelfs_inode),
1957 0, (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1958 pohmelfs_init_once);
1959 if (!pohmelfs_inode_cache)
1960 return -ENOMEM;
1961
1962 return 0;
1963 }
1964
1965 static void pohmelfs_destroy_inodecache(void)
1966 {
1967 kmem_cache_destroy(pohmelfs_inode_cache);
1968 }
1969
1970 static int __init init_pohmel_fs(void)
1971 {
1972 int err;
1973
1974 err = pohmelfs_config_init();
1975 if (err)
1976 goto err_out_exit;
1977
1978 err = pohmelfs_init_inodecache();
1979 if (err)
1980 goto err_out_config_exit;
1981
1982 err = pohmelfs_mcache_init();
1983 if (err)
1984 goto err_out_destroy;
1985
1986 err = netfs_trans_init();
1987 if (err)
1988 goto err_out_mcache_exit;
1989
1990 err = register_filesystem(&pohmel_fs_type);
1991 if (err)
1992 goto err_out_trans;
1993
1994 return 0;
1995
1996 err_out_trans:
1997 netfs_trans_exit();
1998 err_out_mcache_exit:
1999 pohmelfs_mcache_exit();
2000 err_out_destroy:
2001 pohmelfs_destroy_inodecache();
2002 err_out_config_exit:
2003 pohmelfs_config_exit();
2004 err_out_exit:
2005 return err;
2006 }
2007
2008 static void __exit exit_pohmel_fs(void)
2009 {
2010 unregister_filesystem(&pohmel_fs_type);
2011 pohmelfs_destroy_inodecache();
2012 pohmelfs_mcache_exit();
2013 pohmelfs_config_exit();
2014 netfs_trans_exit();
2015 }
2016
2017 module_init(init_pohmel_fs);
2018 module_exit(exit_pohmel_fs);
2019
2020 MODULE_LICENSE("GPL");
2021 MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
2022 MODULE_DESCRIPTION("Pohmel filesystem");
Tomato firmware and modifications.