?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'move-drivers' of git://git.kernel.org/pub/scm/linux/kernel/git/davej...
[linux-2.6/next.git] / drivers / staging / pohmelfs / inode.c
blobc93ef207b0b442120400621ce36dd7e873ec966f
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 #if 0
237 dprintk("%s: %u/%u added trans: %p, gen: %u, page: %p, [High: %d], size: %lu, idx: %lu.\n",
238 __func__, i, trans->page_num, trans, trans->gen, page,
239 !!PageHighMem(page), page_private(page), page->index);
240 #endif
241 wbc->nr_to_write--;
242
243 if (wbc->nr_to_write <= 0)
244 done = 1;
245
246 continue;
247 out_continue:
248 unlock_page(page);
249 trans->pages[i] = NULL;
250 }
251
252 err = netfs_trans_finish(trans, psb);
253 if (err)
254 break;
255
256 continue;
257
258 err_out_reset:
259 trans->result = err;
260 netfs_trans_reset(trans);
261 netfs_trans_put(trans);
262 break;
263 }
264
265 if (!scanned && !done) {
266 /*
267 * We hit the last page and there is more work to be done: wrap
268 * back to the start of the file
269 */
270 scanned = 1;
271 index = 0;
272 goto retry;
273 }
274
275 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
276 mapping->writeback_index = index;
277
278 return err;
279 }
280
281 /*
282 * Inode writeback creation completion callback.
283 * Only invoked for just created inodes, which do not have pages attached,
284 * like dirs and empty files.
285 */
286 static int pohmelfs_write_inode_complete(struct page **pages, unsigned int page_num,
287 void *private, int err)
288 {
289 struct inode *inode = private;
290 struct pohmelfs_inode *pi = POHMELFS_I(inode);
291
292 if (inode) {
293 if (err) {
294 mark_inode_dirty(inode);
295 clear_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
296 } else {
297 set_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state);
298 }
299
300 pohmelfs_put_inode(pi);
301 }
302
303 return err;
304 }
305
306 int pohmelfs_write_create_inode(struct pohmelfs_inode *pi)
307 {
308 struct netfs_trans *t;
309 struct inode *inode = &pi->vfs_inode;
310 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
311 int err;
312
313 if (test_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state))
314 return 0;
315
316 dprintk("%s: started ino: %llu.\n", __func__, pi->ino);
317
318 err = pohmelfs_path_length(pi);
319 if (err < 0)
320 goto err_out_exit;
321
322 t = netfs_trans_alloc(psb, err + 1, 0, 0);
323 if (!t) {
324 err = -ENOMEM;
325 goto err_out_exit;
326 }
327 t->complete = pohmelfs_write_inode_complete;
328 t->private = igrab(inode);
329 if (!t->private) {
330 err = -ENOENT;
331 goto err_out_put;
332 }
333
334 err = pohmelfs_write_inode_create(inode, t);
335 if (err)
336 goto err_out_put;
337
338 netfs_trans_finish(t, POHMELFS_SB(inode->i_sb));
339
340 return 0;
341
342 err_out_put:
343 t->result = err;
344 netfs_trans_put(t);
345 err_out_exit:
346 return err;
347 }
348
349 /*
350 * Sync all not-yet-created children in given directory to the server.
351 */
352 static int pohmelfs_write_inode_create_children(struct inode *inode)
353 {
354 struct pohmelfs_inode *parent = POHMELFS_I(inode);
355 struct super_block *sb = inode->i_sb;
356 struct pohmelfs_name *n;
357
358 while (!list_empty(&parent->sync_create_list)) {
359 n = NULL;
360 mutex_lock(&parent->offset_lock);
361 if (!list_empty(&parent->sync_create_list)) {
362 n = list_first_entry(&parent->sync_create_list,
363 struct pohmelfs_name, sync_create_entry);
364 list_del_init(&n->sync_create_entry);
365 }
366 mutex_unlock(&parent->offset_lock);
367
368 if (!n)
369 break;
370
371 inode = ilookup(sb, n->ino);
372
373 dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n",
374 __func__, parent->ino, n->ino, inode);
375
376 if (inode && (inode->i_state & I_DIRTY)) {
377 struct pohmelfs_inode *pi = POHMELFS_I(inode);
378 pohmelfs_write_create_inode(pi);
379 /* pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0); */
380 iput(inode);
381 }
382 }
383
384 return 0;
385 }
386
387 /*
388 * Removes given child from given inode on server.
389 */
390 int pohmelfs_remove_child(struct pohmelfs_inode *pi, struct pohmelfs_name *n)
391 {
392 return pohmelfs_meta_command_data(pi, pi->ino, NETFS_REMOVE, NULL, 0, NULL, NULL, 0);
393 }
394
395 /*
396 * Writeback for given inode.
397 */
398 static int pohmelfs_write_inode(struct inode *inode,
399 struct writeback_control *wbc)
400 {
401 struct pohmelfs_inode *pi = POHMELFS_I(inode);
402
403 pohmelfs_write_create_inode(pi);
404 pohmelfs_write_inode_create_children(inode);
405
406 return 0;
407 }
408
409 /*
410 * It is not exported, sorry...
411 */
412 static inline wait_queue_head_t *page_waitqueue(struct page *page)
413 {
414 const struct zone *zone = page_zone(page);
415
416 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
417 }
418
419 static int pohmelfs_wait_on_page_locked(struct page *page)
420 {
421 struct pohmelfs_sb *psb = POHMELFS_SB(page->mapping->host->i_sb);
422 long ret = psb->wait_on_page_timeout;
423 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
424 int err = 0;
425
426 if (!PageLocked(page))
427 return 0;
428
429 for (;;) {
430 prepare_to_wait(page_waitqueue(page),
431 &wait.wait, TASK_INTERRUPTIBLE);
432
433 dprintk("%s: page: %p, locked: %d, uptodate: %d, error: %d, flags: %lx.\n",
434 __func__, page, PageLocked(page), PageUptodate(page),
435 PageError(page), page->flags);
436
437 if (!PageLocked(page))
438 break;
439
440 if (!signal_pending(current)) {
441 ret = schedule_timeout(ret);
442 if (!ret)
443 break;
444 continue;
445 }
446 ret = -ERESTARTSYS;
447 break;
448 }
449 finish_wait(page_waitqueue(page), &wait.wait);
450
451 if (!ret)
452 err = -ETIMEDOUT;
453
454
455 if (!err)
456 SetPageUptodate(page);
457
458 if (err)
459 printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n",
460 __func__, page, PageUptodate(page), PageLocked(page), err);
461
462 return err;
463 }
464
465 static int pohmelfs_read_page_complete(struct page **pages, unsigned int page_num,
466 void *private, int err)
467 {
468 struct page *page = private;
469
470 if (PageChecked(page))
471 return err;
472
473 if (err < 0) {
474 dprintk("%s: page: %p, err: %d.\n", __func__, page, err);
475 SetPageError(page);
476 }
477
478 unlock_page(page);
479
480 return err;
481 }
482
483 /*
484 * Read a page from remote server.
485 * Function will wait until page is unlocked.
486 */
487 static int pohmelfs_readpage(struct file *file, struct page *page)
488 {
489 struct inode *inode = page->mapping->host;
490 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
491 struct pohmelfs_inode *pi = POHMELFS_I(inode);
492 struct netfs_trans *t;
493 struct netfs_cmd *cmd;
494 int err, path_len;
495 void *data;
496 u64 isize;
497
498 err = pohmelfs_data_lock(pi, page->index << PAGE_CACHE_SHIFT,
499 PAGE_SIZE, POHMELFS_READ_LOCK);
500 if (err)
501 goto err_out_exit;
502
503 isize = i_size_read(inode);
504 if (isize <= page->index << PAGE_CACHE_SHIFT) {
505 SetPageUptodate(page);
506 unlock_page(page);
507 return 0;
508 }
509
510 path_len = pohmelfs_path_length(pi);
511 if (path_len < 0) {
512 err = path_len;
513 goto err_out_exit;
514 }
515
516 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
517 if (!t) {
518 err = -ENOMEM;
519 goto err_out_exit;
520 }
521
522 t->complete = pohmelfs_read_page_complete;
523 t->private = page;
524
525 cmd = netfs_trans_current(t);
526 data = (void *)(cmd + 1);
527
528 err = pohmelfs_construct_path_string(pi, data, path_len);
529 if (err < 0)
530 goto err_out_free;
531
532 path_len = err;
533
534 cmd->id = pi->ino;
535 cmd->start = page->index;
536 cmd->start <<= PAGE_CACHE_SHIFT;
537 cmd->size = PAGE_CACHE_SIZE + path_len;
538 cmd->cmd = NETFS_READ_PAGE;
539 cmd->ext = path_len;
540
541 dprintk("%s: path: '%s', page: %p, ino: %llu, start: %llu, size: %lu.\n",
542 __func__, (char *)data, page, pi->ino, cmd->start, PAGE_CACHE_SIZE);
543
544 netfs_convert_cmd(cmd);
545 netfs_trans_update(cmd, t, path_len);
546
547 err = netfs_trans_finish(t, psb);
548 if (err)
549 goto err_out_return;
550
551 return pohmelfs_wait_on_page_locked(page);
552
553 err_out_free:
554 t->result = err;
555 netfs_trans_put(t);
556 err_out_exit:
557 SetPageError(page);
558 if (PageLocked(page))
559 unlock_page(page);
560 err_out_return:
561 printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n",
562 __func__, page, page->index << PAGE_CACHE_SHIFT, PAGE_CACHE_SIZE, err);
563
564 return err;
565 }
566
567 /*
568 * Write begin/end magic.
569 * Allocates a page and writes inode if it was not synced to server before.
570 */
571 static int pohmelfs_write_begin(struct file *file, struct address_space *mapping,
572 loff_t pos, unsigned len, unsigned flags,
573 struct page **pagep, void **fsdata)
574 {
575 struct inode *inode = mapping->host;
576 struct page *page;
577 pgoff_t index;
578 unsigned start, end;
579 int err;
580
581 *pagep = NULL;
582
583 index = pos >> PAGE_CACHE_SHIFT;
584 start = pos & (PAGE_CACHE_SIZE - 1);
585 end = start + len;
586
587 page = grab_cache_page(mapping, index);
588 #if 0
589 dprintk("%s: page: %p pos: %llu, len: %u, index: %lu, start: %u, end: %u, uptodate: %d.\n",
590 __func__, page, pos, len, index, start, end, PageUptodate(page));
591 #endif
592 if (!page) {
593 err = -ENOMEM;
594 goto err_out_exit;
595 }
596
597 while (!PageUptodate(page)) {
598 if (start && test_bit(NETFS_INODE_REMOTE_SYNCED, &POHMELFS_I(inode)->state)) {
599 err = pohmelfs_readpage(file, page);
600 if (err)
601 goto err_out_exit;
602
603 lock_page(page);
604 continue;
605 }
606
607 if (len != PAGE_CACHE_SIZE) {
608 void *kaddr = kmap_atomic(page, KM_USER0);
609
610 memset(kaddr + start, 0, PAGE_CACHE_SIZE - start);
611 flush_dcache_page(page);
612 kunmap_atomic(kaddr, KM_USER0);
613 }
614 SetPageUptodate(page);
615 }
616
617 set_page_private(page, end);
618
619 *pagep = page;
620
621 return 0;
622
623 err_out_exit:
624 page_cache_release(page);
625 *pagep = NULL;
626
627 return err;
628 }
629
630 static int pohmelfs_write_end(struct file *file, struct address_space *mapping,
631 loff_t pos, unsigned len, unsigned copied,
632 struct page *page, void *fsdata)
633 {
634 struct inode *inode = mapping->host;
635
636 if (copied != len) {
637 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
638 void *kaddr = kmap_atomic(page, KM_USER0);
639
640 memset(kaddr + from + copied, 0, len - copied);
641 flush_dcache_page(page);
642 kunmap_atomic(kaddr, KM_USER0);
643 }
644
645 SetPageUptodate(page);
646 set_page_dirty(page);
647 #if 0
648 dprintk("%s: page: %p [U: %d, D: %d, L: %d], pos: %llu, len: %u, copied: %u.\n",
649 __func__, page,
650 PageUptodate(page), PageDirty(page), PageLocked(page),
651 pos, len, copied);
652 #endif
653 flush_dcache_page(page);
654
655 unlock_page(page);
656 page_cache_release(page);
657
658 if (pos + copied > inode->i_size) {
659 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
660
661 psb->avail_size -= pos + copied - inode->i_size;
662
663 i_size_write(inode, pos + copied);
664 }
665
666 return copied;
667 }
668
669 static int pohmelfs_readpages_trans_complete(struct page **__pages, unsigned int page_num,
670 void *private, int err)
671 {
672 struct pohmelfs_inode *pi = private;
673 unsigned int i, num;
674 struct page **pages, *page = (struct page *)__pages;
675 loff_t index = page->index;
676
677 pages = kzalloc(sizeof(void *) * page_num, GFP_NOIO);
678 if (!pages)
679 return -ENOMEM;
680
681 num = find_get_pages_contig(pi->vfs_inode.i_mapping, index, page_num, pages);
682 if (num <= 0) {
683 err = num;
684 goto err_out_free;
685 }
686
687 for (i = 0; i < num; ++i) {
688 page = pages[i];
689
690 if (err)
691 printk("%s: %u/%u: page: %p, index: %lu, uptodate: %d, locked: %d, err: %d.\n",
692 __func__, i, num, page, page->index,
693 PageUptodate(page), PageLocked(page), err);
694
695 if (!PageChecked(page)) {
696 if (err < 0)
697 SetPageError(page);
698 unlock_page(page);
699 }
700 page_cache_release(page);
701 page_cache_release(page);
702 }
703
704 err_out_free:
705 kfree(pages);
706 return err;
707 }
708
709 static int pohmelfs_send_readpages(struct pohmelfs_inode *pi, struct page *first, unsigned int num)
710 {
711 struct netfs_trans *t;
712 struct netfs_cmd *cmd;
713 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
714 int err, path_len;
715 void *data;
716
717 err = pohmelfs_data_lock(pi, first->index << PAGE_CACHE_SHIFT,
718 num * PAGE_SIZE, POHMELFS_READ_LOCK);
719 if (err)
720 goto err_out_exit;
721
722 path_len = pohmelfs_path_length(pi);
723 if (path_len < 0) {
724 err = path_len;
725 goto err_out_exit;
726 }
727
728 t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0);
729 if (!t) {
730 err = -ENOMEM;
731 goto err_out_exit;
732 }
733
734 cmd = netfs_trans_current(t);
735 data = (void *)(cmd + 1);
736
737 t->complete = pohmelfs_readpages_trans_complete;
738 t->private = pi;
739 t->page_num = num;
740 t->pages = (struct page **)first;
741
742 err = pohmelfs_construct_path_string(pi, data, path_len);
743 if (err < 0)
744 goto err_out_put;
745
746 path_len = err;
747
748 cmd->cmd = NETFS_READ_PAGES;
749 cmd->start = first->index;
750 cmd->start <<= PAGE_CACHE_SHIFT;
751 cmd->size = (num << 8 | PAGE_CACHE_SHIFT);
752 cmd->id = pi->ino;
753 cmd->ext = path_len;
754
755 dprintk("%s: t: %p, gen: %u, path: '%s', path_len: %u, "
756 "start: %lu, num: %u.\n",
757 __func__, t, t->gen, (char *)data, path_len,
758 first->index, num);
759
760 netfs_convert_cmd(cmd);
761 netfs_trans_update(cmd, t, path_len);
762
763 return netfs_trans_finish(t, psb);
764
765 err_out_put:
766 netfs_trans_free(t);
767 err_out_exit:
768 pohmelfs_readpages_trans_complete((struct page **)first, num, pi, err);
769 return err;
770 }
771
772 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
773
774 static int pohmelfs_readpages(struct file *file, struct address_space *mapping,
775 struct list_head *pages, unsigned nr_pages)
776 {
777 unsigned int page_idx, num = 0;
778 struct page *page = NULL, *first = NULL;
779
780 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
781 page = list_to_page(pages);
782
783 prefetchw(&page->flags);
784 list_del(&page->lru);
785
786 if (!add_to_page_cache_lru(page, mapping,
787 page->index, GFP_KERNEL)) {
788
789 if (!num) {
790 num = 1;
791 first = page;
792 continue;
793 }
794
795 dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n",
796 __func__, page, page->index, first->index);
797
798 if (unlikely(first->index + num != page->index) || (num > 500)) {
799 pohmelfs_send_readpages(POHMELFS_I(mapping->host),
800 first, num);
801 first = page;
802 num = 0;
803 }
804
805 num++;
806 }
807 }
808 pohmelfs_send_readpages(POHMELFS_I(mapping->host), first, num);
809
810 /*
811 * This will be sync read, so when last page is processed,
812 * all previous are alerady unlocked and ready to be used.
813 */
814 return 0;
815 }
816
817 /*
818 * Small address space operations for POHMELFS.
819 */
820 const struct address_space_operations pohmelfs_aops = {
821 .readpage = pohmelfs_readpage,
822 .readpages = pohmelfs_readpages,
823 .writepages = pohmelfs_writepages,
824 .write_begin = pohmelfs_write_begin,
825 .write_end = pohmelfs_write_end,
826 .set_page_dirty = __set_page_dirty_nobuffers,
827 };
828
829 static void pohmelfs_i_callback(struct rcu_head *head)
830 {
831 struct inode *inode = container_of(head, struct inode, i_rcu);
832 INIT_LIST_HEAD(&inode->i_dentry);
833 kmem_cache_free(pohmelfs_inode_cache, POHMELFS_I(inode));
834 }
835
836 /*
837 * ->destroy_inode() callback. Deletes inode from the caches
838 * and frees private data.
839 */
840 static void pohmelfs_destroy_inode(struct inode *inode)
841 {
842 struct super_block *sb = inode->i_sb;
843 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
844 struct pohmelfs_inode *pi = POHMELFS_I(inode);
845
846 /* pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK); */
847
848 pohmelfs_inode_del_inode(psb, pi);
849
850 dprintk("%s: pi: %p, inode: %p, ino: %llu.\n",
851 __func__, pi, &pi->vfs_inode, pi->ino);
852 atomic_long_dec(&psb->total_inodes);
853 call_rcu(&inode->i_rcu, pohmelfs_i_callback);
854 }
855
856 /*
857 * ->alloc_inode() callback. Allocates inode and initializes private data.
858 */
859 static struct inode *pohmelfs_alloc_inode(struct super_block *sb)
860 {
861 struct pohmelfs_inode *pi;
862
863 pi = kmem_cache_alloc(pohmelfs_inode_cache, GFP_NOIO);
864 if (!pi)
865 return NULL;
866
867 pi->hash_root = RB_ROOT;
868 mutex_init(&pi->offset_lock);
869
870 INIT_LIST_HEAD(&pi->sync_create_list);
871
872 INIT_LIST_HEAD(&pi->inode_entry);
873
874 pi->lock_type = 0;
875 pi->state = 0;
876 pi->total_len = 0;
877 pi->drop_count = 0;
878
879 dprintk("%s: pi: %p, inode: %p.\n", __func__, pi, &pi->vfs_inode);
880
881 atomic_long_inc(&POHMELFS_SB(sb)->total_inodes);
882
883 return &pi->vfs_inode;
884 }
885
886 /*
887 * We want fsync() to work on POHMELFS.
888 */
889 static int pohmelfs_fsync(struct file *file, int datasync)
890 {
891 struct inode *inode = file->f_mapping->host;
892
893 return sync_inode_metadata(inode, 1);
894 }
895
896 ssize_t pohmelfs_write(struct file *file, const char __user *buf,
897 size_t len, loff_t *ppos)
898 {
899 struct address_space *mapping = file->f_mapping;
900 struct inode *inode = mapping->host;
901 struct pohmelfs_inode *pi = POHMELFS_I(inode);
902 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
903 struct kiocb kiocb;
904 ssize_t ret;
905 loff_t pos = *ppos;
906
907 init_sync_kiocb(&kiocb, file);
908 kiocb.ki_pos = pos;
909 kiocb.ki_left = len;
910
911 dprintk("%s: len: %zu, pos: %llu.\n", __func__, len, pos);
912
913 mutex_lock(&inode->i_mutex);
914 ret = pohmelfs_data_lock(pi, pos, len, POHMELFS_WRITE_LOCK);
915 if (ret)
916 goto err_out_unlock;
917
918 ret = __generic_file_aio_write(&kiocb, &iov, 1, &kiocb.ki_pos);
919 *ppos = kiocb.ki_pos;
920
921 mutex_unlock(&inode->i_mutex);
922 WARN_ON(ret < 0);
923
924 if (ret > 0) {
925 ssize_t err;
926
927 err = generic_write_sync(file, pos, ret);
928 if (err < 0)
929 ret = err;
930 WARN_ON(ret < 0);
931 }
932
933 return ret;
934
935 err_out_unlock:
936 mutex_unlock(&inode->i_mutex);
937 return ret;
938 }
939
940 static const struct file_operations pohmelfs_file_ops = {
941 .open = generic_file_open,
942 .fsync = pohmelfs_fsync,
943
944 .llseek = generic_file_llseek,
945
946 .read = do_sync_read,
947 .aio_read = generic_file_aio_read,
948
949 .mmap = generic_file_mmap,
950
951 .splice_read = generic_file_splice_read,
952 .splice_write = generic_file_splice_write,
953
954 .write = pohmelfs_write,
955 .aio_write = generic_file_aio_write,
956 };
957
958 const struct inode_operations pohmelfs_symlink_inode_operations = {
959 .readlink = generic_readlink,
960 .follow_link = page_follow_link_light,
961 .put_link = page_put_link,
962 };
963
964 int pohmelfs_setattr_raw(struct inode *inode, struct iattr *attr)
965 {
966 int err;
967
968 err = inode_change_ok(inode, attr);
969 if (err) {
970 dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__, POHMELFS_I(inode)->ino);
971 goto err_out_exit;
972 }
973
974 if ((attr->ia_valid & ATTR_SIZE) &&
975 attr->ia_size != i_size_read(inode)) {
976 err = vmtruncate(inode, attr->ia_size);
977 if (err) {
978 dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__, POHMELFS_I(inode)->ino);
979 goto err_out_exit;
980 }
981 }
982
983 setattr_copy(inode, attr);
984 mark_inode_dirty(inode);
985
986 dprintk("%s: ino: %llu, mode: %o -> %o, uid: %u -> %u, gid: %u -> %u, size: %llu -> %llu.\n",
987 __func__, POHMELFS_I(inode)->ino, inode->i_mode, attr->ia_mode,
988 inode->i_uid, attr->ia_uid, inode->i_gid, attr->ia_gid, inode->i_size, attr->ia_size);
989
990 return 0;
991
992 err_out_exit:
993 return err;
994 }
995
996 int pohmelfs_setattr(struct dentry *dentry, struct iattr *attr)
997 {
998 struct inode *inode = dentry->d_inode;
999 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1000 int err;
1001
1002 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1003 if (err)
1004 goto err_out_exit;
1005
1006 err = security_inode_setattr(dentry, attr);
1007 if (err)
1008 goto err_out_exit;
1009
1010 err = pohmelfs_setattr_raw(inode, attr);
1011 if (err)
1012 goto err_out_exit;
1013
1014 return 0;
1015
1016 err_out_exit:
1017 return err;
1018 }
1019
1020 static int pohmelfs_send_xattr_req(struct pohmelfs_inode *pi, u64 id, u64 start,
1021 const char *name, const void *value, size_t attrsize, int command)
1022 {
1023 struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb);
1024 int err, path_len, namelen = strlen(name) + 1; /* 0-byte */
1025 struct netfs_trans *t;
1026 struct netfs_cmd *cmd;
1027 void *data;
1028
1029 dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n",
1030 __func__, id, start, name, attrsize, command);
1031
1032 path_len = pohmelfs_path_length(pi);
1033 if (path_len < 0) {
1034 err = path_len;
1035 goto err_out_exit;
1036 }
1037
1038 t = netfs_trans_alloc(psb, namelen + path_len + attrsize, 0, 0);
1039 if (!t) {
1040 err = -ENOMEM;
1041 goto err_out_exit;
1042 }
1043
1044 cmd = netfs_trans_current(t);
1045 data = cmd + 1;
1046
1047 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1048 if (path_len < 0) {
1049 err = path_len;
1050 goto err_out_put;
1051 }
1052 data += path_len;
1053
1054 /*
1055 * 'name' is a NUL-terminated string already and
1056 * 'namelen' includes 0-byte.
1057 */
1058 memcpy(data, name, namelen);
1059 data += namelen;
1060
1061 memcpy(data, value, attrsize);
1062
1063 cmd->cmd = command;
1064 cmd->id = id;
1065 cmd->start = start;
1066 cmd->size = attrsize + namelen + path_len;
1067 cmd->ext = path_len;
1068 cmd->csize = 0;
1069 cmd->cpad = 0;
1070
1071 netfs_convert_cmd(cmd);
1072 netfs_trans_update(cmd, t, namelen + path_len + attrsize);
1073
1074 return netfs_trans_finish(t, psb);
1075
1076 err_out_put:
1077 t->result = err;
1078 netfs_trans_put(t);
1079 err_out_exit:
1080 return err;
1081 }
1082
1083 static int pohmelfs_setxattr(struct dentry *dentry, const char *name,
1084 const void *value, size_t attrsize, int flags)
1085 {
1086 struct inode *inode = dentry->d_inode;
1087 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1088 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1089
1090 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1091 return -EOPNOTSUPP;
1092
1093 return pohmelfs_send_xattr_req(pi, flags, attrsize, name,
1094 value, attrsize, NETFS_XATTR_SET);
1095 }
1096
1097 static ssize_t pohmelfs_getxattr(struct dentry *dentry, const char *name,
1098 void *value, size_t attrsize)
1099 {
1100 struct inode *inode = dentry->d_inode;
1101 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1102 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1103 struct pohmelfs_mcache *m;
1104 int err;
1105 long timeout = psb->mcache_timeout;
1106
1107 if (!(psb->state_flags & POHMELFS_FLAGS_XATTR))
1108 return -EOPNOTSUPP;
1109
1110 m = pohmelfs_mcache_alloc(psb, 0, attrsize, value);
1111 if (IS_ERR(m))
1112 return PTR_ERR(m);
1113
1114 dprintk("%s: ino: %llu, name: '%s', size: %zu.\n",
1115 __func__, pi->ino, name, attrsize);
1116
1117 err = pohmelfs_send_xattr_req(pi, m->gen, attrsize, name, value, 0, NETFS_XATTR_GET);
1118 if (err)
1119 goto err_out_put;
1120
1121 do {
1122 err = wait_for_completion_timeout(&m->complete, timeout);
1123 if (err) {
1124 err = m->err;
1125 break;
1126 }
1127
1128 /*
1129 * This loop is a bit ugly, since it waits until reference counter
1130 * hits 1 and then puts the object here. Main goal is to prevent race with
1131 * the network thread, when it can start processing the given request, i.e.
1132 * increase its reference counter but yet not complete it, while
1133 * we will exit from ->getxattr() with timeout, and although request
1134 * will not be freed (its reference counter was increased by network
1135 * thread), data pointer provided by user may be released, so we will
1136 * overwrite an already freed area in the network thread.
1137 *
1138 * Now after timeout we remove request from the cache, so it can not be
1139 * found by network thread, and wait for its reference counter to hit 1,
1140 * i.e. if network thread already started to process this request, we wait
1141 * for it to finish, and then free object locally. If reference counter is
1142 * already 1, i.e. request is not used by anyone else, we can free it without
1143 * problem.
1144 */
1145 err = -ETIMEDOUT;
1146 timeout = HZ;
1147
1148 pohmelfs_mcache_remove_locked(psb, m);
1149 } while (atomic_read(&m->refcnt) != 1);
1150
1151 pohmelfs_mcache_put(psb, m);
1152
1153 dprintk("%s: ino: %llu, err: %d.\n", __func__, pi->ino, err);
1154
1155 return err;
1156
1157 err_out_put:
1158 pohmelfs_mcache_put(psb, m);
1159 return err;
1160 }
1161
1162 static int pohmelfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1163 {
1164 struct inode *inode = dentry->d_inode;
1165 #if 0
1166 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1167 int err;
1168
1169 err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_READ_LOCK);
1170 if (err)
1171 return err;
1172 dprintk("%s: ino: %llu, mode: %o, uid: %u, gid: %u, size: %llu.\n",
1173 __func__, pi->ino, inode->i_mode, inode->i_uid,
1174 inode->i_gid, inode->i_size);
1175 #endif
1176
1177 generic_fillattr(inode, stat);
1178 return 0;
1179 }
1180
1181 const struct inode_operations pohmelfs_file_inode_operations = {
1182 .setattr = pohmelfs_setattr,
1183 .getattr = pohmelfs_getattr,
1184 .setxattr = pohmelfs_setxattr,
1185 .getxattr = pohmelfs_getxattr,
1186 };
1187
1188 /*
1189 * Fill inode data: mode, size, operation callbacks and so on...
1190 */
1191 void pohmelfs_fill_inode(struct inode *inode, struct netfs_inode_info *info)
1192 {
1193 inode->i_mode = info->mode;
1194 inode->i_nlink = info->nlink;
1195 inode->i_uid = info->uid;
1196 inode->i_gid = info->gid;
1197 inode->i_blocks = info->blocks;
1198 inode->i_rdev = info->rdev;
1199 inode->i_size = info->size;
1200 inode->i_version = info->version;
1201 inode->i_blkbits = ffs(info->blocksize);
1202
1203 dprintk("%s: inode: %p, num: %lu/%llu inode is regular: %d, dir: %d, link: %d, mode: %o, size: %llu.\n",
1204 __func__, inode, inode->i_ino, info->ino,
1205 S_ISREG(inode->i_mode), S_ISDIR(inode->i_mode),
1206 S_ISLNK(inode->i_mode), inode->i_mode, inode->i_size);
1207
1208 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1209
1210 /*
1211 * i_mapping is a pointer to i_data during inode initialization.
1212 */
1213 inode->i_data.a_ops = &pohmelfs_aops;
1214
1215 if (S_ISREG(inode->i_mode)) {
1216 inode->i_fop = &pohmelfs_file_ops;
1217 inode->i_op = &pohmelfs_file_inode_operations;
1218 } else if (S_ISDIR(inode->i_mode)) {
1219 inode->i_fop = &pohmelfs_dir_fops;
1220 inode->i_op = &pohmelfs_dir_inode_ops;
1221 } else if (S_ISLNK(inode->i_mode)) {
1222 inode->i_op = &pohmelfs_symlink_inode_operations;
1223 inode->i_fop = &pohmelfs_file_ops;
1224 } else {
1225 inode->i_fop = &generic_ro_fops;
1226 }
1227 }
1228
1229 static int pohmelfs_drop_inode(struct inode *inode)
1230 {
1231 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1232 struct pohmelfs_inode *pi = POHMELFS_I(inode);
1233
1234 spin_lock(&psb->ino_lock);
1235 list_del_init(&pi->inode_entry);
1236 spin_unlock(&psb->ino_lock);
1237
1238 return generic_drop_inode(inode);
1239 }
1240
1241 static struct pohmelfs_inode *pohmelfs_get_inode_from_list(struct pohmelfs_sb *psb,
1242 struct list_head *head, unsigned int *count)
1243 {
1244 struct pohmelfs_inode *pi = NULL;
1245
1246 spin_lock(&psb->ino_lock);
1247 if (!list_empty(head)) {
1248 pi = list_entry(head->next, struct pohmelfs_inode,
1249 inode_entry);
1250 list_del_init(&pi->inode_entry);
1251 *count = pi->drop_count;
1252 pi->drop_count = 0;
1253 }
1254 spin_unlock(&psb->ino_lock);
1255
1256 return pi;
1257 }
1258
1259 static void pohmelfs_flush_transactions(struct pohmelfs_sb *psb)
1260 {
1261 struct pohmelfs_config *c;
1262
1263 mutex_lock(&psb->state_lock);
1264 list_for_each_entry(c, &psb->state_list, config_entry) {
1265 pohmelfs_state_flush_transactions(&c->state);
1266 }
1267 mutex_unlock(&psb->state_lock);
1268 }
1269
1270 /*
1271 * ->put_super() callback. Invoked before superblock is destroyed,
1272 * so it has to clean all private data.
1273 */
1274 static void pohmelfs_put_super(struct super_block *sb)
1275 {
1276 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1277 struct pohmelfs_inode *pi;
1278 unsigned int count = 0;
1279 unsigned int in_drop_list = 0;
1280 struct inode *inode, *tmp;
1281
1282 dprintk("%s.\n", __func__);
1283
1284 /*
1285 * Kill pending transactions, which could affect inodes in-flight.
1286 */
1287 pohmelfs_flush_transactions(psb);
1288
1289 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) {
1290 inode = &pi->vfs_inode;
1291
1292 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1293 __func__, pi->ino, pi, inode, count);
1294
1295 if (atomic_read(&inode->i_count) != count) {
1296 printk("%s: ino: %llu, pi: %p, inode: %p, count: %u, i_count: %d.\n",
1297 __func__, pi->ino, pi, inode, count,
1298 atomic_read(&inode->i_count));
1299 count = atomic_read(&inode->i_count);
1300 in_drop_list++;
1301 }
1302
1303 while (count--)
1304 iput(&pi->vfs_inode);
1305 }
1306
1307 list_for_each_entry_safe(inode, tmp, &sb->s_inodes, i_sb_list) {
1308 pi = POHMELFS_I(inode);
1309
1310 dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n",
1311 __func__, pi->ino, pi, inode, atomic_read(&inode->i_count));
1312
1313 /*
1314 * These are special inodes, they were created during
1315 * directory reading or lookup, and were not bound to dentry,
1316 * so they live here with reference counter being 1 and prevent
1317 * umount from succeed since it believes that they are busy.
1318 */
1319 count = atomic_read(&inode->i_count);
1320 if (count) {
1321 list_del_init(&inode->i_sb_list);
1322 while (count--)
1323 iput(&pi->vfs_inode);
1324 }
1325 }
1326
1327 psb->trans_scan_timeout = psb->drop_scan_timeout = 0;
1328 cancel_delayed_work_sync(&psb->dwork);
1329 cancel_delayed_work_sync(&psb->drop_dwork);
1330 flush_scheduled_work();
1331
1332 dprintk("%s: stopped workqueues.\n", __func__);
1333
1334 pohmelfs_crypto_exit(psb);
1335 pohmelfs_state_exit(psb);
1336
1337 bdi_destroy(&psb->bdi);
1338
1339 kfree(psb);
1340 sb->s_fs_info = NULL;
1341 }
1342
1343 static int pohmelfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1344 {
1345 struct super_block *sb = dentry->d_sb;
1346 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1347
1348 /*
1349 * There are no filesystem size limits yet.
1350 */
1351 memset(buf, 0, sizeof(struct kstatfs));
1352
1353 buf->f_type = POHMELFS_MAGIC_NUM; /* 'POH.' */
1354 buf->f_bsize = sb->s_blocksize;
1355 buf->f_files = psb->ino;
1356 buf->f_namelen = 255;
1357 buf->f_files = atomic_long_read(&psb->total_inodes);
1358 buf->f_bfree = buf->f_bavail = psb->avail_size >> PAGE_SHIFT;
1359 buf->f_blocks = psb->total_size >> PAGE_SHIFT;
1360
1361 dprintk("%s: total: %llu, avail: %llu, inodes: %llu, bsize: %lu.\n",
1362 __func__, psb->total_size, psb->avail_size, buf->f_files, sb->s_blocksize);
1363
1364 return 0;
1365 }
1366
1367 static int pohmelfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
1368 {
1369 struct pohmelfs_sb *psb = POHMELFS_SB(vfs->mnt_sb);
1370
1371 seq_printf(seq, ",idx=%u", psb->idx);
1372 seq_printf(seq, ",trans_scan_timeout=%u", jiffies_to_msecs(psb->trans_scan_timeout));
1373 seq_printf(seq, ",drop_scan_timeout=%u", jiffies_to_msecs(psb->drop_scan_timeout));
1374 seq_printf(seq, ",wait_on_page_timeout=%u", jiffies_to_msecs(psb->wait_on_page_timeout));
1375 seq_printf(seq, ",trans_retries=%u", psb->trans_retries);
1376 seq_printf(seq, ",crypto_thread_num=%u", psb->crypto_thread_num);
1377 seq_printf(seq, ",trans_max_pages=%u", psb->trans_max_pages);
1378 seq_printf(seq, ",mcache_timeout=%u", jiffies_to_msecs(psb->mcache_timeout));
1379 if (psb->crypto_fail_unsupported)
1380 seq_printf(seq, ",crypto_fail_unsupported");
1381
1382 return 0;
1383 }
1384
1385 enum {
1386 pohmelfs_opt_idx,
1387 pohmelfs_opt_crypto_thread_num,
1388 pohmelfs_opt_trans_max_pages,
1389 pohmelfs_opt_crypto_fail_unsupported,
1390
1391 /* Remountable options */
1392 pohmelfs_opt_trans_scan_timeout,
1393 pohmelfs_opt_drop_scan_timeout,
1394 pohmelfs_opt_wait_on_page_timeout,
1395 pohmelfs_opt_trans_retries,
1396 pohmelfs_opt_mcache_timeout,
1397 };
1398
1399 static struct match_token pohmelfs_tokens[] = {
1400 {pohmelfs_opt_idx, "idx=%u"},
1401 {pohmelfs_opt_crypto_thread_num, "crypto_thread_num=%u"},
1402 {pohmelfs_opt_trans_max_pages, "trans_max_pages=%u"},
1403 {pohmelfs_opt_crypto_fail_unsupported, "crypto_fail_unsupported"},
1404 {pohmelfs_opt_trans_scan_timeout, "trans_scan_timeout=%u"},
1405 {pohmelfs_opt_drop_scan_timeout, "drop_scan_timeout=%u"},
1406 {pohmelfs_opt_wait_on_page_timeout, "wait_on_page_timeout=%u"},
1407 {pohmelfs_opt_trans_retries, "trans_retries=%u"},
1408 {pohmelfs_opt_mcache_timeout, "mcache_timeout=%u"},
1409 };
1410
1411 static int pohmelfs_parse_options(char *options, struct pohmelfs_sb *psb, int remount)
1412 {
1413 char *p;
1414 substring_t args[MAX_OPT_ARGS];
1415 int option, err;
1416
1417 if (!options)
1418 return 0;
1419
1420 while ((p = strsep(&options, ",")) != NULL) {
1421 int token;
1422 if (!*p)
1423 continue;
1424
1425 token = match_token(p, pohmelfs_tokens, args);
1426
1427 err = match_int(&args[0], &option);
1428 if (err)
1429 return err;
1430
1431 if (remount && token <= pohmelfs_opt_crypto_fail_unsupported)
1432 continue;
1433
1434 switch (token) {
1435 case pohmelfs_opt_idx:
1436 psb->idx = option;
1437 break;
1438 case pohmelfs_opt_trans_scan_timeout:
1439 psb->trans_scan_timeout = msecs_to_jiffies(option);
1440 break;
1441 case pohmelfs_opt_drop_scan_timeout:
1442 psb->drop_scan_timeout = msecs_to_jiffies(option);
1443 break;
1444 case pohmelfs_opt_wait_on_page_timeout:
1445 psb->wait_on_page_timeout = msecs_to_jiffies(option);
1446 break;
1447 case pohmelfs_opt_mcache_timeout:
1448 psb->mcache_timeout = msecs_to_jiffies(option);
1449 break;
1450 case pohmelfs_opt_trans_retries:
1451 psb->trans_retries = option;
1452 break;
1453 case pohmelfs_opt_crypto_thread_num:
1454 psb->crypto_thread_num = option;
1455 break;
1456 case pohmelfs_opt_trans_max_pages:
1457 psb->trans_max_pages = option;
1458 break;
1459 case pohmelfs_opt_crypto_fail_unsupported:
1460 psb->crypto_fail_unsupported = 1;
1461 break;
1462 default:
1463 return -EINVAL;
1464 }
1465 }
1466
1467 return 0;
1468 }
1469
1470 static int pohmelfs_remount(struct super_block *sb, int *flags, char *data)
1471 {
1472 int err;
1473 struct pohmelfs_sb *psb = POHMELFS_SB(sb);
1474 unsigned long old_sb_flags = sb->s_flags;
1475
1476 err = pohmelfs_parse_options(data, psb, 1);
1477 if (err)
1478 goto err_out_restore;
1479
1480 if (!(*flags & MS_RDONLY))
1481 sb->s_flags &= ~MS_RDONLY;
1482 return 0;
1483
1484 err_out_restore:
1485 sb->s_flags = old_sb_flags;
1486 return err;
1487 }
1488
1489 static void pohmelfs_flush_inode(struct pohmelfs_inode *pi, unsigned int count)
1490 {
1491 struct inode *inode = &pi->vfs_inode;
1492
1493 dprintk("%s: %p: ino: %llu, owned: %d.\n",
1494 __func__, inode, pi->ino, test_bit(NETFS_INODE_OWNED, &pi->state));
1495
1496 mutex_lock(&inode->i_mutex);
1497 if (test_and_clear_bit(NETFS_INODE_OWNED, &pi->state)) {
1498 filemap_fdatawrite(inode->i_mapping);
1499 inode->i_sb->s_op->write_inode(inode, 0);
1500 }
1501
1502 #ifdef POHMELFS_TRUNCATE_ON_INODE_FLUSH
1503 truncate_inode_pages(inode->i_mapping, 0);
1504 #endif
1505
1506 pohmelfs_data_unlock(pi, 0, ~0, POHMELFS_WRITE_LOCK);
1507 mutex_unlock(&inode->i_mutex);
1508 }
1509
1510 static void pohmelfs_put_inode_count(struct pohmelfs_inode *pi, unsigned int count)
1511 {
1512 dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n",
1513 __func__, pi->ino, pi, &pi->vfs_inode, count);
1514
1515 if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH, &pi->state))
1516 pohmelfs_flush_inode(pi, count);
1517
1518 while (count--)
1519 iput(&pi->vfs_inode);
1520 }
1521
1522 static void pohmelfs_drop_scan(struct work_struct *work)
1523 {
1524 struct pohmelfs_sb *psb =
1525 container_of(work, struct pohmelfs_sb, drop_dwork.work);
1526 struct pohmelfs_inode *pi;
1527 unsigned int count = 0;
1528
1529 while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count)))
1530 pohmelfs_put_inode_count(pi, count);
1531
1532 pohmelfs_check_states(psb);
1533
1534 if (psb->drop_scan_timeout)
1535 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1536 }
1537
1538 /*
1539 * Run through all transactions starting from the oldest,
1540 * drop transaction from current state and try to send it
1541 * to all remote nodes, which are currently installed.
1542 */
1543 static void pohmelfs_trans_scan_state(struct netfs_state *st)
1544 {
1545 struct rb_node *rb_node;
1546 struct netfs_trans_dst *dst;
1547 struct pohmelfs_sb *psb = st->psb;
1548 unsigned int timeout = psb->trans_scan_timeout;
1549 struct netfs_trans *t;
1550 int err;
1551
1552 mutex_lock(&st->trans_lock);
1553 for (rb_node = rb_first(&st->trans_root); rb_node; ) {
1554 dst = rb_entry(rb_node, struct netfs_trans_dst, state_entry);
1555 t = dst->trans;
1556
1557 if (timeout && time_after(dst->send_time + timeout, jiffies)
1558 && dst->retries == 0)
1559 break;
1560
1561 dprintk("%s: t: %p, gen: %u, st: %p, retries: %u, max: %u.\n",
1562 __func__, t, t->gen, st, dst->retries, psb->trans_retries);
1563 netfs_trans_get(t);
1564
1565 rb_node = rb_next(rb_node);
1566
1567 err = -ETIMEDOUT;
1568 if (timeout && (++dst->retries < psb->trans_retries))
1569 err = netfs_trans_resend(t, psb);
1570
1571 if (err || (t->flags & NETFS_TRANS_SINGLE_DST)) {
1572 if (netfs_trans_remove_nolock(dst, st))
1573 netfs_trans_drop_dst_nostate(dst);
1574 }
1575
1576 t->result = err;
1577 netfs_trans_put(t);
1578 }
1579 mutex_unlock(&st->trans_lock);
1580 }
1581
1582 /*
1583 * Walk through all installed network states and resend all
1584 * transactions, which are old enough.
1585 */
1586 static void pohmelfs_trans_scan(struct work_struct *work)
1587 {
1588 struct pohmelfs_sb *psb =
1589 container_of(work, struct pohmelfs_sb, dwork.work);
1590 struct netfs_state *st;
1591 struct pohmelfs_config *c;
1592
1593 mutex_lock(&psb->state_lock);
1594 list_for_each_entry(c, &psb->state_list, config_entry) {
1595 st = &c->state;
1596
1597 pohmelfs_trans_scan_state(st);
1598 }
1599 mutex_unlock(&psb->state_lock);
1600
1601 /*
1602 * If no timeout specified then system is in the middle of umount process,
1603 * so no need to reschedule scanning process again.
1604 */
1605 if (psb->trans_scan_timeout)
1606 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1607 }
1608
1609 int pohmelfs_meta_command_data(struct pohmelfs_inode *pi, u64 id, unsigned int cmd_op, char *addon,
1610 unsigned int flags, netfs_trans_complete_t complete, void *priv, u64 start)
1611 {
1612 struct inode *inode = &pi->vfs_inode;
1613 struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb);
1614 int err = 0, sz;
1615 struct netfs_trans *t;
1616 int path_len, addon_len = 0;
1617 void *data;
1618 struct netfs_inode_info *info;
1619 struct netfs_cmd *cmd;
1620
1621 dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__, pi->ino, cmd_op, addon);
1622
1623 path_len = pohmelfs_path_length(pi);
1624 if (path_len < 0) {
1625 err = path_len;
1626 goto err_out_exit;
1627 }
1628
1629 if (addon)
1630 addon_len = strlen(addon) + 1; /* 0-byte */
1631 sz = addon_len;
1632
1633 if (cmd_op == NETFS_INODE_INFO)
1634 sz += sizeof(struct netfs_inode_info);
1635
1636 t = netfs_trans_alloc(psb, sz + path_len, flags, 0);
1637 if (!t) {
1638 err = -ENOMEM;
1639 goto err_out_exit;
1640 }
1641 t->complete = complete;
1642 t->private = priv;
1643
1644 cmd = netfs_trans_current(t);
1645 data = (void *)(cmd + 1);
1646
1647 if (cmd_op == NETFS_INODE_INFO) {
1648 info = (struct netfs_inode_info *)(cmd + 1);
1649 data = (void *)(info + 1);
1650
1651 /*
1652 * We are under i_mutex, can read and change whatever we want...
1653 */
1654 info->mode = inode->i_mode;
1655 info->nlink = inode->i_nlink;
1656 info->uid = inode->i_uid;
1657 info->gid = inode->i_gid;
1658 info->blocks = inode->i_blocks;
1659 info->rdev = inode->i_rdev;
1660 info->size = inode->i_size;
1661 info->version = inode->i_version;
1662
1663 netfs_convert_inode_info(info);
1664 }
1665
1666 path_len = pohmelfs_construct_path_string(pi, data, path_len);
1667 if (path_len < 0)
1668 goto err_out_free;
1669
1670 dprintk("%s: path_len: %d.\n", __func__, path_len);
1671
1672 if (addon) {
1673 path_len--; /* Do not place null-byte before the addon */
1674 path_len += sprintf(data + path_len, "/%s", addon) + 1; /* 0 - byte */
1675 }
1676
1677 sz += path_len;
1678
1679 cmd->cmd = cmd_op;
1680 cmd->ext = path_len;
1681 cmd->size = sz;
1682 cmd->id = id;
1683 cmd->start = start;
1684
1685 netfs_convert_cmd(cmd);
1686 netfs_trans_update(cmd, t, sz);
1687
1688 /*
1689 * Note, that it is possible to leak error here: transaction callback will not
1690 * be invoked for allocation path failure.
1691 */
1692 return netfs_trans_finish(t, psb);
1693
1694 err_out_free:
1695 netfs_trans_free(t);
1696 err_out_exit:
1697 if (complete)
1698 complete(NULL, 0, priv, err);
1699 return err;
1700 }
1701
1702 int pohmelfs_meta_command(struct pohmelfs_inode *pi, unsigned int cmd_op, unsigned int flags,
1703 netfs_trans_complete_t complete, void *priv, u64 start)
1704 {
1705 return pohmelfs_meta_command_data(pi, pi->ino, cmd_op, NULL, flags, complete, priv, start);
1706 }
1707
1708 /*
1709 * Send request and wait for POHMELFS root capabilities response,
1710 * which will update server's informaion about size of the export,
1711 * permissions, number of objects, available size and so on.
1712 */
1713 static int pohmelfs_root_handshake(struct pohmelfs_sb *psb)
1714 {
1715 struct netfs_trans *t;
1716 struct netfs_cmd *cmd;
1717 int err = -ENOMEM;
1718
1719 t = netfs_trans_alloc(psb, 0, 0, 0);
1720 if (!t)
1721 goto err_out_exit;
1722
1723 cmd = netfs_trans_current(t);
1724
1725 cmd->cmd = NETFS_CAPABILITIES;
1726 cmd->id = POHMELFS_ROOT_CAPABILITIES;
1727 cmd->size = 0;
1728 cmd->start = 0;
1729 cmd->ext = 0;
1730 cmd->csize = 0;
1731
1732 netfs_convert_cmd(cmd);
1733 netfs_trans_update(cmd, t, 0);
1734
1735 err = netfs_trans_finish(t, psb);
1736 if (err)
1737 goto err_out_exit;
1738
1739 psb->flags = ~0;
1740 err = wait_event_interruptible_timeout(psb->wait,
1741 (psb->flags != ~0),
1742 psb->wait_on_page_timeout);
1743 if (!err)
1744 err = -ETIMEDOUT;
1745 else if (err > 0)
1746 err = -psb->flags;
1747
1748 if (err)
1749 goto err_out_exit;
1750
1751 return 0;
1752
1753 err_out_exit:
1754 return err;
1755 }
1756
1757 static int pohmelfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
1758 {
1759 struct netfs_state *st;
1760 struct pohmelfs_ctl *ctl;
1761 struct pohmelfs_sb *psb = POHMELFS_SB(mnt->mnt_sb);
1762 struct pohmelfs_config *c;
1763
1764 mutex_lock(&psb->state_lock);
1765
1766 seq_printf(m, "\nidx addr(:port) socket_type protocol active priority permissions\n");
1767
1768 list_for_each_entry(c, &psb->state_list, config_entry) {
1769 st = &c->state;
1770 ctl = &st->ctl;
1771
1772 seq_printf(m, "%u ", ctl->idx);
1773 if (ctl->addr.sa_family == AF_INET) {
1774 struct sockaddr_in *sin = (struct sockaddr_in *)&st->ctl.addr;
1775 seq_printf(m, "%pI4:%u", &sin->sin_addr.s_addr, ntohs(sin->sin_port));
1776 } else if (ctl->addr.sa_family == AF_INET6) {
1777 struct sockaddr_in6 *sin = (struct sockaddr_in6 *)&st->ctl.addr;
1778 seq_printf(m, "%pi6:%u", &sin->sin6_addr, ntohs(sin->sin6_port));
1779 } else {
1780 unsigned int i;
1781 for (i = 0; i < ctl->addrlen; ++i)
1782 seq_printf(m, "%02x.", ctl->addr.addr[i]);
1783 }
1784
1785 seq_printf(m, " %u %u %d %u %x\n",
1786 ctl->type, ctl->proto,
1787 st->socket != NULL,
1788 ctl->prio, ctl->perm);
1789 }
1790 mutex_unlock(&psb->state_lock);
1791
1792 return 0;
1793 }
1794
1795 static const struct super_operations pohmelfs_sb_ops = {
1796 .alloc_inode = pohmelfs_alloc_inode,
1797 .destroy_inode = pohmelfs_destroy_inode,
1798 .drop_inode = pohmelfs_drop_inode,
1799 .write_inode = pohmelfs_write_inode,
1800 .put_super = pohmelfs_put_super,
1801 .remount_fs = pohmelfs_remount,
1802 .statfs = pohmelfs_statfs,
1803 .show_options = pohmelfs_show_options,
1804 .show_stats = pohmelfs_show_stats,
1805 };
1806
1807 /*
1808 * Allocate private superblock and create root dir.
1809 */
1810 static int pohmelfs_fill_super(struct super_block *sb, void *data, int silent)
1811 {
1812 struct pohmelfs_sb *psb;
1813 int err = -ENOMEM;
1814 struct inode *root;
1815 struct pohmelfs_inode *npi;
1816 struct qstr str;
1817
1818 psb = kzalloc(sizeof(struct pohmelfs_sb), GFP_KERNEL);
1819 if (!psb)
1820 goto err_out_exit;
1821
1822 err = bdi_init(&psb->bdi);
1823 if (err)
1824 goto err_out_free_sb;
1825
1826 err = bdi_register(&psb->bdi, NULL, "pfs-%d", atomic_inc_return(&psb_bdi_num));
1827 if (err) {
1828 bdi_destroy(&psb->bdi);
1829 goto err_out_free_sb;
1830 }
1831
1832 sb->s_fs_info = psb;
1833 sb->s_op = &pohmelfs_sb_ops;
1834 sb->s_magic = POHMELFS_MAGIC_NUM;
1835 sb->s_maxbytes = MAX_LFS_FILESIZE;
1836 sb->s_blocksize = PAGE_SIZE;
1837 sb->s_bdi = &psb->bdi;
1838
1839 psb->sb = sb;
1840
1841 psb->ino = 2;
1842 psb->idx = 0;
1843 psb->active_state = NULL;
1844 psb->trans_retries = 5;
1845 psb->trans_data_size = PAGE_SIZE;
1846 psb->drop_scan_timeout = msecs_to_jiffies(1000);
1847 psb->trans_scan_timeout = msecs_to_jiffies(5000);
1848 psb->wait_on_page_timeout = msecs_to_jiffies(5000);
1849 init_waitqueue_head(&psb->wait);
1850
1851 spin_lock_init(&psb->ino_lock);
1852
1853 INIT_LIST_HEAD(&psb->drop_list);
1854
1855 mutex_init(&psb->mcache_lock);
1856 psb->mcache_root = RB_ROOT;
1857 psb->mcache_timeout = msecs_to_jiffies(5000);
1858 atomic_long_set(&psb->mcache_gen, 0);
1859
1860 psb->trans_max_pages = 100;
1861
1862 psb->crypto_align_size = 16;
1863 psb->crypto_attached_size = 0;
1864 psb->hash_strlen = 0;
1865 psb->cipher_strlen = 0;
1866 psb->perform_crypto = 0;
1867 psb->crypto_thread_num = 2;
1868 psb->crypto_fail_unsupported = 0;
1869 mutex_init(&psb->crypto_thread_lock);
1870 INIT_LIST_HEAD(&psb->crypto_ready_list);
1871 INIT_LIST_HEAD(&psb->crypto_active_list);
1872
1873 atomic_set(&psb->trans_gen, 1);
1874 atomic_long_set(&psb->total_inodes, 0);
1875
1876 mutex_init(&psb->state_lock);
1877 INIT_LIST_HEAD(&psb->state_list);
1878
1879 err = pohmelfs_parse_options((char *) data, psb, 0);
1880 if (err)
1881 goto err_out_free_bdi;
1882
1883 err = pohmelfs_copy_crypto(psb);
1884 if (err)
1885 goto err_out_free_bdi;
1886
1887 err = pohmelfs_state_init(psb);
1888 if (err)
1889 goto err_out_free_strings;
1890
1891 err = pohmelfs_crypto_init(psb);
1892 if (err)
1893 goto err_out_state_exit;
1894
1895 err = pohmelfs_root_handshake(psb);
1896 if (err)
1897 goto err_out_crypto_exit;
1898
1899 str.name = "/";
1900 str.hash = jhash("/", 1, 0);
1901 str.len = 1;
1902
1903 npi = pohmelfs_create_entry_local(psb, NULL, &str, 0, 0755|S_IFDIR);
1904 if (IS_ERR(npi)) {
1905 err = PTR_ERR(npi);
1906 goto err_out_crypto_exit;
1907 }
1908 set_bit(NETFS_INODE_REMOTE_SYNCED, &npi->state);
1909 clear_bit(NETFS_INODE_OWNED, &npi->state);
1910
1911 root = &npi->vfs_inode;
1912
1913 sb->s_root = d_alloc_root(root);
1914 if (!sb->s_root)
1915 goto err_out_put_root;
1916
1917 INIT_DELAYED_WORK(&psb->drop_dwork, pohmelfs_drop_scan);
1918 schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout);
1919
1920 INIT_DELAYED_WORK(&psb->dwork, pohmelfs_trans_scan);
1921 schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout);
1922
1923 return 0;
1924
1925 err_out_put_root:
1926 iput(root);
1927 err_out_crypto_exit:
1928 pohmelfs_crypto_exit(psb);
1929 err_out_state_exit:
1930 pohmelfs_state_exit(psb);
1931 err_out_free_strings:
1932 kfree(psb->cipher_string);
1933 kfree(psb->hash_string);
1934 err_out_free_bdi:
1935 bdi_destroy(&psb->bdi);
1936 err_out_free_sb:
1937 kfree(psb);
1938 err_out_exit:
1939
1940 dprintk("%s: err: %d.\n", __func__, err);
1941 return err;
1942 }
1943
1944 /*
1945 * Some VFS magic here...
1946 */
1947 static struct dentry *pohmelfs_mount(struct file_system_type *fs_type,
1948 int flags, const char *dev_name, void *data)
1949 {
1950 return mount_nodev(fs_type, flags, data, pohmelfs_fill_super);
1951 }
1952
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 */
1958 static void pohmelfs_kill_super(struct super_block *sb)
1959 {
1960 sync_inodes_sb(sb);
1961 kill_anon_super(sb);
1962 }
1963
1964 static struct file_system_type pohmel_fs_type = {
1965 .owner = THIS_MODULE,
1966 .name = "pohmel",
1967 .mount = pohmelfs_mount,
1968 .kill_sb = pohmelfs_kill_super,
1969 };
1970
1971 /*
1972 * Cache and module initializations and freeing routings.
1973 */
1974 static void pohmelfs_init_once(void *data)
1975 {
1976 struct pohmelfs_inode *pi = data;
1977
1978 inode_init_once(&pi->vfs_inode);
1979 }
1980
1981 static 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
1993 static void pohmelfs_destroy_inodecache(void)
1994 {
1995 kmem_cache_destroy(pohmelfs_inode_cache);
1996 }
1997
1998 static 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
2024 err_out_trans:
2025 netfs_trans_exit();
2026 err_out_mcache_exit:
2027 pohmelfs_mcache_exit();
2028 err_out_destroy:
2029 pohmelfs_destroy_inodecache();
2030 err_out_config_exit:
2031 pohmelfs_config_exit();
2032 err_out_exit:
2033 return err;
2034 }
2035
2036 static void __exit exit_pohmel_fs(void)
2037 {
2038 unregister_filesystem(&pohmel_fs_type);
2039 pohmelfs_destroy_inodecache();
2040 pohmelfs_mcache_exit();
2041 pohmelfs_config_exit();
2042 netfs_trans_exit();
2043 }
2044
2045 module_init(init_pohmel_fs);
2046 module_exit(exit_pohmel_fs);
2047
2048 MODULE_LICENSE("GPL");
2049 MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
2050 MODULE_DESCRIPTION("Pohmel filesystem");
linux-next