Commit | Line | Data |
---|---|---|
b3f08cad EP |
1 | /* |
2 | * 2007+ Copyright (c) Evgeniy Polyakov <zbr@ioremap.net> | |
3 | * All rights reserved. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | */ | |
15 | ||
16 | #include <linux/module.h> | |
17 | #include <linux/backing-dev.h> | |
18 | #include <linux/crypto.h> | |
19 | #include <linux/fs.h> | |
20 | #include <linux/jhash.h> | |
21 | #include <linux/hash.h> | |
22 | #include <linux/ktime.h> | |
23 | #include <linux/mm.h> | |
24 | #include <linux/mount.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/pagevec.h> | |
27 | #include <linux/parser.h> | |
28 | #include <linux/swap.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/statfs.h> | |
31 | #include <linux/writeback.h> | |
268bb0ce | 32 | #include <linux/prefetch.h> |
b3f08cad EP |
33 | |
34 | #include "netfs.h" | |
35 | ||
36 | #define POHMELFS_MAGIC_NUM 0x504f482e | |
37 | ||
38 | static struct kmem_cache *pohmelfs_inode_cache; | |
182374a0 | 39 | static atomic_t psb_bdi_num = ATOMIC_INIT(0); |
b3f08cad EP |
40 | |
41 | /* | |
42 | * Removes inode from all trees, drops local name cache and removes all queued | |
43 | * requests for object removal. | |
44 | */ | |
45 | void pohmelfs_inode_del_inode(struct pohmelfs_sb *psb, struct pohmelfs_inode *pi) | |
46 | { | |
47 | mutex_lock(&pi->offset_lock); | |
48 | pohmelfs_free_names(pi); | |
49 | mutex_unlock(&pi->offset_lock); | |
50 | ||
51 | dprintk("%s: deleted stuff in ino: %llu.\n", __func__, pi->ino); | |
52 | } | |
53 | ||
54 | /* | |
55 | * Sync inode to server. | |
56 | * Returns zero in success and negative error value otherwise. | |
57 | * It will gather path to root directory into structures containing | |
58 | * creation mode, permissions and names, so that the whole path | |
59 | * to given inode could be created using only single network command. | |
60 | */ | |
61 | int pohmelfs_write_inode_create(struct inode *inode, struct netfs_trans *trans) | |
62 | { | |
63 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
64 | int err = -ENOMEM, size; | |
65 | struct netfs_cmd *cmd; | |
66 | void *data; | |
67 | int cur_len = netfs_trans_cur_len(trans); | |
68 | ||
69 | if (unlikely(cur_len < 0)) | |
70 | return -ETOOSMALL; | |
71 | ||
72 | cmd = netfs_trans_current(trans); | |
73 | cur_len -= sizeof(struct netfs_cmd); | |
74 | ||
75 | data = (void *)(cmd + 1); | |
76 | ||
77 | err = pohmelfs_construct_path_string(pi, data, cur_len); | |
78 | if (err < 0) | |
79 | goto err_out_exit; | |
80 | ||
81 | size = err; | |
82 | ||
83 | cmd->start = i_size_read(inode); | |
84 | cmd->cmd = NETFS_CREATE; | |
85 | cmd->size = size; | |
86 | cmd->id = pi->ino; | |
87 | cmd->ext = inode->i_mode; | |
88 | ||
89 | netfs_convert_cmd(cmd); | |
90 | ||
91 | netfs_trans_update(cmd, trans, size); | |
92 | ||
93 | return 0; | |
94 | ||
95 | err_out_exit: | |
96 | printk("%s: completed ino: %llu, err: %d.\n", __func__, pi->ino, err); | |
97 | return err; | |
98 | } | |
99 | ||
100 | static int pohmelfs_write_trans_complete(struct page **pages, unsigned int page_num, | |
101 | void *private, int err) | |
102 | { | |
103 | unsigned i; | |
104 | ||
105 | dprintk("%s: pages: %lu-%lu, page_num: %u, err: %d.\n", | |
106 | __func__, pages[0]->index, pages[page_num-1]->index, | |
107 | page_num, err); | |
108 | ||
109 | for (i = 0; i < page_num; i++) { | |
110 | struct page *page = pages[i]; | |
111 | ||
112 | if (!page) | |
113 | continue; | |
114 | ||
115 | end_page_writeback(page); | |
116 | ||
117 | if (err < 0) { | |
118 | SetPageError(page); | |
119 | set_page_dirty(page); | |
120 | } | |
121 | ||
122 | unlock_page(page); | |
123 | page_cache_release(page); | |
124 | ||
125 | /* dprintk("%s: %3u/%u: page: %p.\n", __func__, i, page_num, page); */ | |
126 | } | |
127 | return err; | |
128 | } | |
129 | ||
130 | static int pohmelfs_inode_has_dirty_pages(struct address_space *mapping, pgoff_t index) | |
131 | { | |
132 | int ret; | |
133 | struct page *page; | |
134 | ||
135 | rcu_read_lock(); | |
136 | ret = radix_tree_gang_lookup_tag(&mapping->page_tree, | |
137 | (void **)&page, index, 1, PAGECACHE_TAG_DIRTY); | |
138 | rcu_read_unlock(); | |
139 | return ret; | |
140 | } | |
141 | ||
142 | static int pohmelfs_writepages(struct address_space *mapping, struct writeback_control *wbc) | |
143 | { | |
144 | struct inode *inode = mapping->host; | |
145 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
146 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
b3f08cad EP |
147 | int err = 0; |
148 | int done = 0; | |
149 | int nr_pages; | |
150 | pgoff_t index; | |
151 | pgoff_t end; /* Inclusive */ | |
152 | int scanned = 0; | |
153 | int range_whole = 0; | |
154 | ||
b3f08cad EP |
155 | if (wbc->range_cyclic) { |
156 | index = mapping->writeback_index; /* Start from prev offset */ | |
157 | end = -1; | |
158 | } else { | |
159 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
160 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
161 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
162 | range_whole = 1; | |
163 | scanned = 1; | |
164 | } | |
165 | retry: | |
166 | while (!done && (index <= end)) { | |
167 | unsigned int i = min(end - index, (pgoff_t)psb->trans_max_pages); | |
168 | int path_len; | |
169 | struct netfs_trans *trans; | |
170 | ||
171 | err = pohmelfs_inode_has_dirty_pages(mapping, index); | |
172 | if (!err) | |
173 | break; | |
174 | ||
175 | err = pohmelfs_path_length(pi); | |
176 | if (err < 0) | |
177 | break; | |
178 | ||
179 | path_len = err; | |
180 | ||
181 | if (path_len <= 2) { | |
182 | err = -ENOENT; | |
183 | break; | |
184 | } | |
185 | ||
186 | trans = netfs_trans_alloc(psb, path_len, 0, i); | |
187 | if (!trans) { | |
188 | err = -ENOMEM; | |
189 | break; | |
190 | } | |
191 | trans->complete = &pohmelfs_write_trans_complete; | |
192 | ||
193 | trans->page_num = nr_pages = find_get_pages_tag(mapping, &index, | |
194 | PAGECACHE_TAG_DIRTY, trans->page_num, | |
195 | trans->pages); | |
196 | ||
197 | dprintk("%s: t: %p, nr_pages: %u, end: %lu, index: %lu, max: %u.\n", | |
198 | __func__, trans, nr_pages, end, index, trans->page_num); | |
199 | ||
200 | if (!nr_pages) | |
201 | goto err_out_reset; | |
202 | ||
203 | err = pohmelfs_write_inode_create(inode, trans); | |
204 | if (err) | |
205 | goto err_out_reset; | |
206 | ||
207 | err = 0; | |
208 | scanned = 1; | |
209 | ||
210 | for (i = 0; i < trans->page_num; i++) { | |
211 | struct page *page = trans->pages[i]; | |
212 | ||
213 | lock_page(page); | |
214 | ||
215 | if (unlikely(page->mapping != mapping)) | |
216 | goto out_continue; | |
217 | ||
218 | if (!wbc->range_cyclic && page->index > end) { | |
219 | done = 1; | |
220 | goto out_continue; | |
221 | } | |
222 | ||
223 | if (wbc->sync_mode != WB_SYNC_NONE) | |
224 | wait_on_page_writeback(page); | |
225 | ||
226 | if (PageWriteback(page) || | |
227 | !clear_page_dirty_for_io(page)) { | |
228 | dprintk("%s: not clear for io page: %p, writeback: %d.\n", | |
229 | __func__, page, PageWriteback(page)); | |
230 | goto out_continue; | |
231 | } | |
232 | ||
233 | set_page_writeback(page); | |
234 | ||
235 | trans->attached_size += page_private(page); | |
236 | trans->attached_pages++; | |
237 | #if 0 | |
238 | dprintk("%s: %u/%u added trans: %p, gen: %u, page: %p, [High: %d], size: %lu, idx: %lu.\n", | |
239 | __func__, i, trans->page_num, trans, trans->gen, page, | |
240 | !!PageHighMem(page), page_private(page), page->index); | |
241 | #endif | |
242 | wbc->nr_to_write--; | |
243 | ||
244 | if (wbc->nr_to_write <= 0) | |
245 | done = 1; | |
b3f08cad EP |
246 | |
247 | continue; | |
248 | out_continue: | |
249 | unlock_page(page); | |
250 | trans->pages[i] = NULL; | |
251 | } | |
252 | ||
253 | err = netfs_trans_finish(trans, psb); | |
254 | if (err) | |
255 | break; | |
256 | ||
257 | continue; | |
258 | ||
259 | err_out_reset: | |
260 | trans->result = err; | |
261 | netfs_trans_reset(trans); | |
262 | netfs_trans_put(trans); | |
263 | break; | |
264 | } | |
265 | ||
266 | if (!scanned && !done) { | |
267 | /* | |
268 | * We hit the last page and there is more work to be done: wrap | |
269 | * back to the start of the file | |
270 | */ | |
271 | scanned = 1; | |
272 | index = 0; | |
273 | goto retry; | |
274 | } | |
275 | ||
276 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
277 | mapping->writeback_index = index; | |
278 | ||
279 | return err; | |
280 | } | |
281 | ||
282 | /* | |
283 | * Inode writeback creation completion callback. | |
284 | * Only invoked for just created inodes, which do not have pages attached, | |
285 | * like dirs and empty files. | |
286 | */ | |
287 | static int pohmelfs_write_inode_complete(struct page **pages, unsigned int page_num, | |
288 | void *private, int err) | |
289 | { | |
290 | struct inode *inode = private; | |
291 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
292 | ||
293 | if (inode) { | |
294 | if (err) { | |
295 | mark_inode_dirty(inode); | |
296 | clear_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state); | |
297 | } else { | |
298 | set_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state); | |
299 | } | |
300 | ||
301 | pohmelfs_put_inode(pi); | |
302 | } | |
303 | ||
304 | return err; | |
305 | } | |
306 | ||
307 | int pohmelfs_write_create_inode(struct pohmelfs_inode *pi) | |
308 | { | |
309 | struct netfs_trans *t; | |
310 | struct inode *inode = &pi->vfs_inode; | |
311 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
312 | int err; | |
313 | ||
314 | if (test_bit(NETFS_INODE_REMOTE_SYNCED, &pi->state)) | |
315 | return 0; | |
316 | ||
317 | dprintk("%s: started ino: %llu.\n", __func__, pi->ino); | |
318 | ||
319 | err = pohmelfs_path_length(pi); | |
320 | if (err < 0) | |
321 | goto err_out_exit; | |
322 | ||
323 | t = netfs_trans_alloc(psb, err + 1, 0, 0); | |
324 | if (!t) { | |
325 | err = -ENOMEM; | |
e3c0acf4 | 326 | goto err_out_exit; |
b3f08cad EP |
327 | } |
328 | t->complete = pohmelfs_write_inode_complete; | |
329 | t->private = igrab(inode); | |
330 | if (!t->private) { | |
331 | err = -ENOENT; | |
332 | goto err_out_put; | |
333 | } | |
334 | ||
335 | err = pohmelfs_write_inode_create(inode, t); | |
336 | if (err) | |
337 | goto err_out_put; | |
338 | ||
339 | netfs_trans_finish(t, POHMELFS_SB(inode->i_sb)); | |
340 | ||
341 | return 0; | |
342 | ||
343 | err_out_put: | |
344 | t->result = err; | |
345 | netfs_trans_put(t); | |
346 | err_out_exit: | |
347 | return err; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Sync all not-yet-created children in given directory to the server. | |
352 | */ | |
353 | static int pohmelfs_write_inode_create_children(struct inode *inode) | |
354 | { | |
355 | struct pohmelfs_inode *parent = POHMELFS_I(inode); | |
356 | struct super_block *sb = inode->i_sb; | |
357 | struct pohmelfs_name *n; | |
358 | ||
359 | while (!list_empty(&parent->sync_create_list)) { | |
360 | n = NULL; | |
361 | mutex_lock(&parent->offset_lock); | |
362 | if (!list_empty(&parent->sync_create_list)) { | |
363 | n = list_first_entry(&parent->sync_create_list, | |
364 | struct pohmelfs_name, sync_create_entry); | |
365 | list_del_init(&n->sync_create_entry); | |
366 | } | |
367 | mutex_unlock(&parent->offset_lock); | |
368 | ||
369 | if (!n) | |
370 | break; | |
371 | ||
372 | inode = ilookup(sb, n->ino); | |
373 | ||
374 | dprintk("%s: parent: %llu, ino: %llu, inode: %p.\n", | |
375 | __func__, parent->ino, n->ino, inode); | |
376 | ||
377 | if (inode && (inode->i_state & I_DIRTY)) { | |
378 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
379 | pohmelfs_write_create_inode(pi); | |
3bafeab7 | 380 | /* pohmelfs_meta_command(pi, NETFS_INODE_INFO, 0, NULL, NULL, 0); */ |
b3f08cad EP |
381 | iput(inode); |
382 | } | |
383 | } | |
384 | ||
385 | return 0; | |
386 | } | |
387 | ||
388 | /* | |
389 | * Removes given child from given inode on server. | |
390 | */ | |
391 | int pohmelfs_remove_child(struct pohmelfs_inode *pi, struct pohmelfs_name *n) | |
392 | { | |
393 | return pohmelfs_meta_command_data(pi, pi->ino, NETFS_REMOVE, NULL, 0, NULL, NULL, 0); | |
394 | } | |
395 | ||
396 | /* | |
397 | * Writeback for given inode. | |
398 | */ | |
582de7ca RD |
399 | static int pohmelfs_write_inode(struct inode *inode, |
400 | struct writeback_control *wbc) | |
b3f08cad EP |
401 | { |
402 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
403 | ||
404 | pohmelfs_write_create_inode(pi); | |
405 | pohmelfs_write_inode_create_children(inode); | |
406 | ||
407 | return 0; | |
408 | } | |
409 | ||
410 | /* | |
411 | * It is not exported, sorry... | |
412 | */ | |
413 | static inline wait_queue_head_t *page_waitqueue(struct page *page) | |
414 | { | |
415 | const struct zone *zone = page_zone(page); | |
416 | ||
417 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
418 | } | |
419 | ||
420 | static int pohmelfs_wait_on_page_locked(struct page *page) | |
421 | { | |
422 | struct pohmelfs_sb *psb = POHMELFS_SB(page->mapping->host->i_sb); | |
423 | long ret = psb->wait_on_page_timeout; | |
424 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
425 | int err = 0; | |
426 | ||
427 | if (!PageLocked(page)) | |
428 | return 0; | |
429 | ||
430 | for (;;) { | |
431 | prepare_to_wait(page_waitqueue(page), | |
432 | &wait.wait, TASK_INTERRUPTIBLE); | |
433 | ||
434 | dprintk("%s: page: %p, locked: %d, uptodate: %d, error: %d, flags: %lx.\n", | |
435 | __func__, page, PageLocked(page), PageUptodate(page), | |
436 | PageError(page), page->flags); | |
437 | ||
438 | if (!PageLocked(page)) | |
439 | break; | |
440 | ||
441 | if (!signal_pending(current)) { | |
442 | ret = schedule_timeout(ret); | |
443 | if (!ret) | |
444 | break; | |
445 | continue; | |
446 | } | |
447 | ret = -ERESTARTSYS; | |
448 | break; | |
449 | } | |
450 | finish_wait(page_waitqueue(page), &wait.wait); | |
451 | ||
452 | if (!ret) | |
453 | err = -ETIMEDOUT; | |
454 | ||
455 | ||
456 | if (!err) | |
457 | SetPageUptodate(page); | |
458 | ||
459 | if (err) | |
460 | printk("%s: page: %p, uptodate: %d, locked: %d, err: %d.\n", | |
461 | __func__, page, PageUptodate(page), PageLocked(page), err); | |
462 | ||
463 | return err; | |
464 | } | |
465 | ||
466 | static int pohmelfs_read_page_complete(struct page **pages, unsigned int page_num, | |
467 | void *private, int err) | |
468 | { | |
469 | struct page *page = private; | |
470 | ||
471 | if (PageChecked(page)) | |
472 | return err; | |
473 | ||
474 | if (err < 0) { | |
475 | dprintk("%s: page: %p, err: %d.\n", __func__, page, err); | |
476 | SetPageError(page); | |
477 | } | |
478 | ||
479 | unlock_page(page); | |
480 | ||
481 | return err; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Read a page from remote server. | |
486 | * Function will wait until page is unlocked. | |
487 | */ | |
488 | static int pohmelfs_readpage(struct file *file, struct page *page) | |
489 | { | |
490 | struct inode *inode = page->mapping->host; | |
491 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
492 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
493 | struct netfs_trans *t; | |
494 | struct netfs_cmd *cmd; | |
495 | int err, path_len; | |
496 | void *data; | |
497 | u64 isize; | |
498 | ||
499 | err = pohmelfs_data_lock(pi, page->index << PAGE_CACHE_SHIFT, | |
500 | PAGE_SIZE, POHMELFS_READ_LOCK); | |
501 | if (err) | |
502 | goto err_out_exit; | |
503 | ||
504 | isize = i_size_read(inode); | |
505 | if (isize <= page->index << PAGE_CACHE_SHIFT) { | |
506 | SetPageUptodate(page); | |
507 | unlock_page(page); | |
508 | return 0; | |
509 | } | |
510 | ||
511 | path_len = pohmelfs_path_length(pi); | |
512 | if (path_len < 0) { | |
513 | err = path_len; | |
514 | goto err_out_exit; | |
515 | } | |
516 | ||
517 | t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0); | |
518 | if (!t) { | |
519 | err = -ENOMEM; | |
520 | goto err_out_exit; | |
521 | } | |
522 | ||
523 | t->complete = pohmelfs_read_page_complete; | |
524 | t->private = page; | |
525 | ||
526 | cmd = netfs_trans_current(t); | |
527 | data = (void *)(cmd + 1); | |
528 | ||
529 | err = pohmelfs_construct_path_string(pi, data, path_len); | |
530 | if (err < 0) | |
531 | goto err_out_free; | |
532 | ||
533 | path_len = err; | |
534 | ||
535 | cmd->id = pi->ino; | |
536 | cmd->start = page->index; | |
537 | cmd->start <<= PAGE_CACHE_SHIFT; | |
538 | cmd->size = PAGE_CACHE_SIZE + path_len; | |
539 | cmd->cmd = NETFS_READ_PAGE; | |
540 | cmd->ext = path_len; | |
541 | ||
542 | dprintk("%s: path: '%s', page: %p, ino: %llu, start: %llu, size: %lu.\n", | |
543 | __func__, (char *)data, page, pi->ino, cmd->start, PAGE_CACHE_SIZE); | |
544 | ||
545 | netfs_convert_cmd(cmd); | |
546 | netfs_trans_update(cmd, t, path_len); | |
547 | ||
548 | err = netfs_trans_finish(t, psb); | |
549 | if (err) | |
550 | goto err_out_return; | |
551 | ||
552 | return pohmelfs_wait_on_page_locked(page); | |
553 | ||
554 | err_out_free: | |
555 | t->result = err; | |
556 | netfs_trans_put(t); | |
557 | err_out_exit: | |
558 | SetPageError(page); | |
559 | if (PageLocked(page)) | |
560 | unlock_page(page); | |
561 | err_out_return: | |
562 | printk("%s: page: %p, start: %lu, size: %lu, err: %d.\n", | |
563 | __func__, page, page->index << PAGE_CACHE_SHIFT, PAGE_CACHE_SIZE, err); | |
564 | ||
565 | return err; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Write begin/end magic. | |
570 | * Allocates a page and writes inode if it was not synced to server before. | |
571 | */ | |
572 | static int pohmelfs_write_begin(struct file *file, struct address_space *mapping, | |
573 | loff_t pos, unsigned len, unsigned flags, | |
574 | struct page **pagep, void **fsdata) | |
575 | { | |
576 | struct inode *inode = mapping->host; | |
577 | struct page *page; | |
578 | pgoff_t index; | |
579 | unsigned start, end; | |
580 | int err; | |
581 | ||
582 | *pagep = NULL; | |
583 | ||
584 | index = pos >> PAGE_CACHE_SHIFT; | |
585 | start = pos & (PAGE_CACHE_SIZE - 1); | |
586 | end = start + len; | |
587 | ||
588 | page = grab_cache_page(mapping, index); | |
589 | #if 0 | |
590 | dprintk("%s: page: %p pos: %llu, len: %u, index: %lu, start: %u, end: %u, uptodate: %d.\n", | |
591 | __func__, page, pos, len, index, start, end, PageUptodate(page)); | |
592 | #endif | |
593 | if (!page) { | |
594 | err = -ENOMEM; | |
595 | goto err_out_exit; | |
596 | } | |
597 | ||
598 | while (!PageUptodate(page)) { | |
599 | if (start && test_bit(NETFS_INODE_REMOTE_SYNCED, &POHMELFS_I(inode)->state)) { | |
600 | err = pohmelfs_readpage(file, page); | |
601 | if (err) | |
602 | goto err_out_exit; | |
603 | ||
604 | lock_page(page); | |
605 | continue; | |
606 | } | |
607 | ||
608 | if (len != PAGE_CACHE_SIZE) { | |
609 | void *kaddr = kmap_atomic(page, KM_USER0); | |
610 | ||
611 | memset(kaddr + start, 0, PAGE_CACHE_SIZE - start); | |
612 | flush_dcache_page(page); | |
613 | kunmap_atomic(kaddr, KM_USER0); | |
614 | } | |
615 | SetPageUptodate(page); | |
616 | } | |
617 | ||
618 | set_page_private(page, end); | |
619 | ||
620 | *pagep = page; | |
621 | ||
622 | return 0; | |
623 | ||
624 | err_out_exit: | |
625 | page_cache_release(page); | |
626 | *pagep = NULL; | |
627 | ||
628 | return err; | |
629 | } | |
630 | ||
631 | static int pohmelfs_write_end(struct file *file, struct address_space *mapping, | |
632 | loff_t pos, unsigned len, unsigned copied, | |
633 | struct page *page, void *fsdata) | |
634 | { | |
635 | struct inode *inode = mapping->host; | |
636 | ||
637 | if (copied != len) { | |
638 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
639 | void *kaddr = kmap_atomic(page, KM_USER0); | |
640 | ||
641 | memset(kaddr + from + copied, 0, len - copied); | |
642 | flush_dcache_page(page); | |
643 | kunmap_atomic(kaddr, KM_USER0); | |
644 | } | |
645 | ||
646 | SetPageUptodate(page); | |
647 | set_page_dirty(page); | |
648 | #if 0 | |
649 | dprintk("%s: page: %p [U: %d, D: %d, L: %d], pos: %llu, len: %u, copied: %u.\n", | |
650 | __func__, page, | |
651 | PageUptodate(page), PageDirty(page), PageLocked(page), | |
652 | pos, len, copied); | |
653 | #endif | |
654 | flush_dcache_page(page); | |
655 | ||
656 | unlock_page(page); | |
657 | page_cache_release(page); | |
658 | ||
659 | if (pos + copied > inode->i_size) { | |
660 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
661 | ||
662 | psb->avail_size -= pos + copied - inode->i_size; | |
663 | ||
664 | i_size_write(inode, pos + copied); | |
665 | } | |
666 | ||
667 | return copied; | |
668 | } | |
669 | ||
670 | static int pohmelfs_readpages_trans_complete(struct page **__pages, unsigned int page_num, | |
671 | void *private, int err) | |
672 | { | |
673 | struct pohmelfs_inode *pi = private; | |
674 | unsigned int i, num; | |
675 | struct page **pages, *page = (struct page *)__pages; | |
676 | loff_t index = page->index; | |
677 | ||
678 | pages = kzalloc(sizeof(void *) * page_num, GFP_NOIO); | |
679 | if (!pages) | |
680 | return -ENOMEM; | |
681 | ||
682 | num = find_get_pages_contig(pi->vfs_inode.i_mapping, index, page_num, pages); | |
683 | if (num <= 0) { | |
684 | err = num; | |
685 | goto err_out_free; | |
686 | } | |
687 | ||
5d6892b3 | 688 | for (i = 0; i < num; ++i) { |
b3f08cad EP |
689 | page = pages[i]; |
690 | ||
691 | if (err) | |
692 | printk("%s: %u/%u: page: %p, index: %lu, uptodate: %d, locked: %d, err: %d.\n", | |
693 | __func__, i, num, page, page->index, | |
694 | PageUptodate(page), PageLocked(page), err); | |
695 | ||
696 | if (!PageChecked(page)) { | |
697 | if (err < 0) | |
698 | SetPageError(page); | |
699 | unlock_page(page); | |
700 | } | |
701 | page_cache_release(page); | |
702 | page_cache_release(page); | |
703 | } | |
704 | ||
705 | err_out_free: | |
706 | kfree(pages); | |
707 | return err; | |
708 | } | |
709 | ||
710 | static int pohmelfs_send_readpages(struct pohmelfs_inode *pi, struct page *first, unsigned int num) | |
711 | { | |
712 | struct netfs_trans *t; | |
713 | struct netfs_cmd *cmd; | |
714 | struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb); | |
715 | int err, path_len; | |
716 | void *data; | |
717 | ||
718 | err = pohmelfs_data_lock(pi, first->index << PAGE_CACHE_SHIFT, | |
719 | num * PAGE_SIZE, POHMELFS_READ_LOCK); | |
720 | if (err) | |
721 | goto err_out_exit; | |
722 | ||
723 | path_len = pohmelfs_path_length(pi); | |
724 | if (path_len < 0) { | |
725 | err = path_len; | |
726 | goto err_out_exit; | |
727 | } | |
728 | ||
729 | t = netfs_trans_alloc(psb, path_len, NETFS_TRANS_SINGLE_DST, 0); | |
730 | if (!t) { | |
731 | err = -ENOMEM; | |
732 | goto err_out_exit; | |
733 | } | |
734 | ||
735 | cmd = netfs_trans_current(t); | |
736 | data = (void *)(cmd + 1); | |
737 | ||
738 | t->complete = pohmelfs_readpages_trans_complete; | |
739 | t->private = pi; | |
740 | t->page_num = num; | |
741 | t->pages = (struct page **)first; | |
742 | ||
743 | err = pohmelfs_construct_path_string(pi, data, path_len); | |
744 | if (err < 0) | |
745 | goto err_out_put; | |
746 | ||
747 | path_len = err; | |
748 | ||
749 | cmd->cmd = NETFS_READ_PAGES; | |
750 | cmd->start = first->index; | |
751 | cmd->start <<= PAGE_CACHE_SHIFT; | |
752 | cmd->size = (num << 8 | PAGE_CACHE_SHIFT); | |
753 | cmd->id = pi->ino; | |
754 | cmd->ext = path_len; | |
755 | ||
756 | dprintk("%s: t: %p, gen: %u, path: '%s', path_len: %u, " | |
757 | "start: %lu, num: %u.\n", | |
758 | __func__, t, t->gen, (char *)data, path_len, | |
759 | first->index, num); | |
760 | ||
761 | netfs_convert_cmd(cmd); | |
762 | netfs_trans_update(cmd, t, path_len); | |
763 | ||
764 | return netfs_trans_finish(t, psb); | |
765 | ||
766 | err_out_put: | |
767 | netfs_trans_free(t); | |
768 | err_out_exit: | |
769 | pohmelfs_readpages_trans_complete((struct page **)first, num, pi, err); | |
770 | return err; | |
771 | } | |
772 | ||
773 | #define list_to_page(head) (list_entry((head)->prev, struct page, lru)) | |
774 | ||
775 | static int pohmelfs_readpages(struct file *file, struct address_space *mapping, | |
776 | struct list_head *pages, unsigned nr_pages) | |
777 | { | |
778 | unsigned int page_idx, num = 0; | |
779 | struct page *page = NULL, *first = NULL; | |
780 | ||
781 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
782 | page = list_to_page(pages); | |
783 | ||
784 | prefetchw(&page->flags); | |
785 | list_del(&page->lru); | |
786 | ||
787 | if (!add_to_page_cache_lru(page, mapping, | |
788 | page->index, GFP_KERNEL)) { | |
789 | ||
790 | if (!num) { | |
791 | num = 1; | |
792 | first = page; | |
793 | continue; | |
794 | } | |
795 | ||
796 | dprintk("%s: added to lru page: %p, page_index: %lu, first_index: %lu.\n", | |
797 | __func__, page, page->index, first->index); | |
798 | ||
799 | if (unlikely(first->index + num != page->index) || (num > 500)) { | |
800 | pohmelfs_send_readpages(POHMELFS_I(mapping->host), | |
801 | first, num); | |
802 | first = page; | |
803 | num = 0; | |
804 | } | |
805 | ||
806 | num++; | |
807 | } | |
808 | } | |
809 | pohmelfs_send_readpages(POHMELFS_I(mapping->host), first, num); | |
810 | ||
811 | /* | |
812 | * This will be sync read, so when last page is processed, | |
813 | * all previous are alerady unlocked and ready to be used. | |
814 | */ | |
815 | return 0; | |
816 | } | |
817 | ||
818 | /* | |
f69b0d64 | 819 | * Small address space operations for POHMELFS. |
b3f08cad EP |
820 | */ |
821 | const struct address_space_operations pohmelfs_aops = { | |
822 | .readpage = pohmelfs_readpage, | |
823 | .readpages = pohmelfs_readpages, | |
824 | .writepages = pohmelfs_writepages, | |
825 | .write_begin = pohmelfs_write_begin, | |
826 | .write_end = pohmelfs_write_end, | |
827 | .set_page_dirty = __set_page_dirty_nobuffers, | |
828 | }; | |
829 | ||
fa0d7e3d NP |
830 | static void pohmelfs_i_callback(struct rcu_head *head) |
831 | { | |
832 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
833 | kmem_cache_free(pohmelfs_inode_cache, POHMELFS_I(inode)); |
834 | } | |
835 | ||
b3f08cad | 836 | /* |
c69ab1a2 | 837 | * ->destroy_inode() callback. Deletes inode from the caches |
b3f08cad EP |
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 | ||
3bafeab7 | 846 | /* pohmelfs_data_unlock(pi, 0, inode->i_size, POHMELFS_READ_LOCK); */ |
b3f08cad EP |
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); | |
b3f08cad | 852 | atomic_long_dec(&psb->total_inodes); |
fa0d7e3d | 853 | call_rcu(&inode->i_rcu, pohmelfs_i_callback); |
b3f08cad EP |
854 | } |
855 | ||
856 | /* | |
37523e84 | 857 | * ->alloc_inode() callback. Allocates inode and initializes private data. |
b3f08cad EP |
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 | */ | |
02c24a82 | 889 | static int pohmelfs_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
b3f08cad EP |
890 | { |
891 | struct inode *inode = file->f_mapping->host; | |
02c24a82 JB |
892 | int err = filemap_write_and_wait_range(inode->i_mapping, start, end); |
893 | if (!err) { | |
894 | mutex_lock(&inode->i_mutex); | |
895 | err = sync_inode_metadata(inode, 1); | |
896 | mutex_unlock(&inode->i_mutex); | |
897 | } | |
898 | return err; | |
b3f08cad EP |
899 | } |
900 | ||
901 | ssize_t pohmelfs_write(struct file *file, const char __user *buf, | |
902 | size_t len, loff_t *ppos) | |
903 | { | |
904 | struct address_space *mapping = file->f_mapping; | |
905 | struct inode *inode = mapping->host; | |
906 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
907 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len }; | |
908 | struct kiocb kiocb; | |
909 | ssize_t ret; | |
910 | loff_t pos = *ppos; | |
911 | ||
912 | init_sync_kiocb(&kiocb, file); | |
913 | kiocb.ki_pos = pos; | |
914 | kiocb.ki_left = len; | |
915 | ||
02d84ca5 | 916 | dprintk("%s: len: %zu, pos: %llu.\n", __func__, len, pos); |
b3f08cad EP |
917 | |
918 | mutex_lock(&inode->i_mutex); | |
919 | ret = pohmelfs_data_lock(pi, pos, len, POHMELFS_WRITE_LOCK); | |
920 | if (ret) | |
921 | goto err_out_unlock; | |
922 | ||
b04f9321 | 923 | ret = __generic_file_aio_write(&kiocb, &iov, 1, &kiocb.ki_pos); |
b3f08cad EP |
924 | *ppos = kiocb.ki_pos; |
925 | ||
926 | mutex_unlock(&inode->i_mutex); | |
927 | WARN_ON(ret < 0); | |
928 | ||
aa3caafe | 929 | if (ret > 0) { |
b3f08cad EP |
930 | ssize_t err; |
931 | ||
aa3caafe | 932 | err = generic_write_sync(file, pos, ret); |
b3f08cad EP |
933 | if (err < 0) |
934 | ret = err; | |
935 | WARN_ON(ret < 0); | |
936 | } | |
937 | ||
938 | return ret; | |
939 | ||
940 | err_out_unlock: | |
941 | mutex_unlock(&inode->i_mutex); | |
942 | return ret; | |
943 | } | |
944 | ||
80fe3e5d | 945 | static const struct file_operations pohmelfs_file_ops = { |
b3f08cad EP |
946 | .open = generic_file_open, |
947 | .fsync = pohmelfs_fsync, | |
948 | ||
949 | .llseek = generic_file_llseek, | |
950 | ||
951 | .read = do_sync_read, | |
952 | .aio_read = generic_file_aio_read, | |
953 | ||
954 | .mmap = generic_file_mmap, | |
955 | ||
956 | .splice_read = generic_file_splice_read, | |
957 | .splice_write = generic_file_splice_write, | |
958 | ||
959 | .write = pohmelfs_write, | |
960 | .aio_write = generic_file_aio_write, | |
961 | }; | |
962 | ||
963 | const struct inode_operations pohmelfs_symlink_inode_operations = { | |
964 | .readlink = generic_readlink, | |
965 | .follow_link = page_follow_link_light, | |
966 | .put_link = page_put_link, | |
967 | }; | |
968 | ||
969 | int pohmelfs_setattr_raw(struct inode *inode, struct iattr *attr) | |
970 | { | |
971 | int err; | |
972 | ||
973 | err = inode_change_ok(inode, attr); | |
974 | if (err) { | |
975 | dprintk("%s: ino: %llu, inode changes are not allowed.\n", __func__, POHMELFS_I(inode)->ino); | |
976 | goto err_out_exit; | |
977 | } | |
978 | ||
1025774c CH |
979 | if ((attr->ia_valid & ATTR_SIZE) && |
980 | attr->ia_size != i_size_read(inode)) { | |
981 | err = vmtruncate(inode, attr->ia_size); | |
982 | if (err) { | |
983 | dprintk("%s: ino: %llu, failed to set the attributes.\n", __func__, POHMELFS_I(inode)->ino); | |
984 | goto err_out_exit; | |
985 | } | |
b3f08cad EP |
986 | } |
987 | ||
1025774c CH |
988 | setattr_copy(inode, attr); |
989 | mark_inode_dirty(inode); | |
990 | ||
b3f08cad EP |
991 | dprintk("%s: ino: %llu, mode: %o -> %o, uid: %u -> %u, gid: %u -> %u, size: %llu -> %llu.\n", |
992 | __func__, POHMELFS_I(inode)->ino, inode->i_mode, attr->ia_mode, | |
993 | inode->i_uid, attr->ia_uid, inode->i_gid, attr->ia_gid, inode->i_size, attr->ia_size); | |
994 | ||
995 | return 0; | |
996 | ||
997 | err_out_exit: | |
998 | return err; | |
999 | } | |
1000 | ||
1001 | int pohmelfs_setattr(struct dentry *dentry, struct iattr *attr) | |
1002 | { | |
1003 | struct inode *inode = dentry->d_inode; | |
1004 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
1005 | int err; | |
1006 | ||
1007 | err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_WRITE_LOCK); | |
1008 | if (err) | |
1009 | goto err_out_exit; | |
1010 | ||
1011 | err = security_inode_setattr(dentry, attr); | |
1012 | if (err) | |
1013 | goto err_out_exit; | |
1014 | ||
1015 | err = pohmelfs_setattr_raw(inode, attr); | |
1016 | if (err) | |
1017 | goto err_out_exit; | |
1018 | ||
1019 | return 0; | |
1020 | ||
1021 | err_out_exit: | |
1022 | return err; | |
1023 | } | |
1024 | ||
1025 | static int pohmelfs_send_xattr_req(struct pohmelfs_inode *pi, u64 id, u64 start, | |
1026 | const char *name, const void *value, size_t attrsize, int command) | |
1027 | { | |
1028 | struct pohmelfs_sb *psb = POHMELFS_SB(pi->vfs_inode.i_sb); | |
1029 | int err, path_len, namelen = strlen(name) + 1; /* 0-byte */ | |
1030 | struct netfs_trans *t; | |
1031 | struct netfs_cmd *cmd; | |
1032 | void *data; | |
1033 | ||
02d84ca5 | 1034 | dprintk("%s: id: %llu, start: %llu, name: '%s', attrsize: %zu, cmd: %d.\n", |
b3f08cad EP |
1035 | __func__, id, start, name, attrsize, command); |
1036 | ||
1037 | path_len = pohmelfs_path_length(pi); | |
1038 | if (path_len < 0) { | |
1039 | err = path_len; | |
1040 | goto err_out_exit; | |
1041 | } | |
1042 | ||
1043 | t = netfs_trans_alloc(psb, namelen + path_len + attrsize, 0, 0); | |
1044 | if (!t) { | |
1045 | err = -ENOMEM; | |
1046 | goto err_out_exit; | |
1047 | } | |
1048 | ||
1049 | cmd = netfs_trans_current(t); | |
1050 | data = cmd + 1; | |
1051 | ||
1052 | path_len = pohmelfs_construct_path_string(pi, data, path_len); | |
1053 | if (path_len < 0) { | |
1054 | err = path_len; | |
1055 | goto err_out_put; | |
1056 | } | |
1057 | data += path_len; | |
1058 | ||
1059 | /* | |
1060 | * 'name' is a NUL-terminated string already and | |
1061 | * 'namelen' includes 0-byte. | |
1062 | */ | |
1063 | memcpy(data, name, namelen); | |
1064 | data += namelen; | |
1065 | ||
1066 | memcpy(data, value, attrsize); | |
1067 | ||
1068 | cmd->cmd = command; | |
1069 | cmd->id = id; | |
1070 | cmd->start = start; | |
1071 | cmd->size = attrsize + namelen + path_len; | |
1072 | cmd->ext = path_len; | |
1073 | cmd->csize = 0; | |
1074 | cmd->cpad = 0; | |
1075 | ||
1076 | netfs_convert_cmd(cmd); | |
1077 | netfs_trans_update(cmd, t, namelen + path_len + attrsize); | |
1078 | ||
1079 | return netfs_trans_finish(t, psb); | |
1080 | ||
1081 | err_out_put: | |
1082 | t->result = err; | |
1083 | netfs_trans_put(t); | |
1084 | err_out_exit: | |
1085 | return err; | |
1086 | } | |
1087 | ||
1088 | static int pohmelfs_setxattr(struct dentry *dentry, const char *name, | |
1089 | const void *value, size_t attrsize, int flags) | |
1090 | { | |
1091 | struct inode *inode = dentry->d_inode; | |
1092 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
1093 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
1094 | ||
1095 | if (!(psb->state_flags & POHMELFS_FLAGS_XATTR)) | |
1096 | return -EOPNOTSUPP; | |
1097 | ||
1098 | return pohmelfs_send_xattr_req(pi, flags, attrsize, name, | |
1099 | value, attrsize, NETFS_XATTR_SET); | |
1100 | } | |
1101 | ||
1102 | static ssize_t pohmelfs_getxattr(struct dentry *dentry, const char *name, | |
1103 | void *value, size_t attrsize) | |
1104 | { | |
1105 | struct inode *inode = dentry->d_inode; | |
1106 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
1107 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
1108 | struct pohmelfs_mcache *m; | |
1109 | int err; | |
1110 | long timeout = psb->mcache_timeout; | |
1111 | ||
1112 | if (!(psb->state_flags & POHMELFS_FLAGS_XATTR)) | |
1113 | return -EOPNOTSUPP; | |
1114 | ||
1115 | m = pohmelfs_mcache_alloc(psb, 0, attrsize, value); | |
1116 | if (IS_ERR(m)) | |
1117 | return PTR_ERR(m); | |
1118 | ||
1119 | dprintk("%s: ino: %llu, name: '%s', size: %zu.\n", | |
1120 | __func__, pi->ino, name, attrsize); | |
1121 | ||
1122 | err = pohmelfs_send_xattr_req(pi, m->gen, attrsize, name, value, 0, NETFS_XATTR_GET); | |
1123 | if (err) | |
1124 | goto err_out_put; | |
1125 | ||
1126 | do { | |
1127 | err = wait_for_completion_timeout(&m->complete, timeout); | |
1128 | if (err) { | |
1129 | err = m->err; | |
1130 | break; | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * This loop is a bit ugly, since it waits until reference counter | |
c69ab1a2 JM |
1135 | * hits 1 and then puts the object here. Main goal is to prevent race with |
1136 | * the network thread, when it can start processing the given request, i.e. | |
b3f08cad EP |
1137 | * increase its reference counter but yet not complete it, while |
1138 | * we will exit from ->getxattr() with timeout, and although request | |
1139 | * will not be freed (its reference counter was increased by network | |
1140 | * thread), data pointer provided by user may be released, so we will | |
c69ab1a2 | 1141 | * overwrite an already freed area in the network thread. |
b3f08cad EP |
1142 | * |
1143 | * Now after timeout we remove request from the cache, so it can not be | |
1144 | * found by network thread, and wait for its reference counter to hit 1, | |
1145 | * i.e. if network thread already started to process this request, we wait | |
c69ab1a2 | 1146 | * for it to finish, and then free object locally. If reference counter is |
b3f08cad EP |
1147 | * already 1, i.e. request is not used by anyone else, we can free it without |
1148 | * problem. | |
1149 | */ | |
1150 | err = -ETIMEDOUT; | |
1151 | timeout = HZ; | |
1152 | ||
1153 | pohmelfs_mcache_remove_locked(psb, m); | |
1154 | } while (atomic_read(&m->refcnt) != 1); | |
1155 | ||
1156 | pohmelfs_mcache_put(psb, m); | |
1157 | ||
1158 | dprintk("%s: ino: %llu, err: %d.\n", __func__, pi->ino, err); | |
1159 | ||
1160 | return err; | |
1161 | ||
1162 | err_out_put: | |
1163 | pohmelfs_mcache_put(psb, m); | |
1164 | return err; | |
1165 | } | |
1166 | ||
1167 | static int pohmelfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) | |
1168 | { | |
1169 | struct inode *inode = dentry->d_inode; | |
e5043424 | 1170 | #if 0 |
b3f08cad EP |
1171 | struct pohmelfs_inode *pi = POHMELFS_I(inode); |
1172 | int err; | |
e5043424 | 1173 | |
b3f08cad EP |
1174 | err = pohmelfs_data_lock(pi, 0, ~0, POHMELFS_READ_LOCK); |
1175 | if (err) | |
1176 | return err; | |
b3f08cad EP |
1177 | dprintk("%s: ino: %llu, mode: %o, uid: %u, gid: %u, size: %llu.\n", |
1178 | __func__, pi->ino, inode->i_mode, inode->i_uid, | |
1179 | inode->i_gid, inode->i_size); | |
e5043424 | 1180 | #endif |
b3f08cad EP |
1181 | |
1182 | generic_fillattr(inode, stat); | |
1183 | return 0; | |
1184 | } | |
1185 | ||
1186 | const struct inode_operations pohmelfs_file_inode_operations = { | |
1187 | .setattr = pohmelfs_setattr, | |
1188 | .getattr = pohmelfs_getattr, | |
1189 | .setxattr = pohmelfs_setxattr, | |
1190 | .getxattr = pohmelfs_getxattr, | |
1191 | }; | |
1192 | ||
1193 | /* | |
1194 | * Fill inode data: mode, size, operation callbacks and so on... | |
1195 | */ | |
1196 | void pohmelfs_fill_inode(struct inode *inode, struct netfs_inode_info *info) | |
1197 | { | |
1198 | inode->i_mode = info->mode; | |
bfe86848 | 1199 | set_nlink(inode, info->nlink); |
b3f08cad EP |
1200 | inode->i_uid = info->uid; |
1201 | inode->i_gid = info->gid; | |
1202 | inode->i_blocks = info->blocks; | |
1203 | inode->i_rdev = info->rdev; | |
1204 | inode->i_size = info->size; | |
1205 | inode->i_version = info->version; | |
1206 | inode->i_blkbits = ffs(info->blocksize); | |
1207 | ||
1208 | dprintk("%s: inode: %p, num: %lu/%llu inode is regular: %d, dir: %d, link: %d, mode: %o, size: %llu.\n", | |
1209 | __func__, inode, inode->i_ino, info->ino, | |
1210 | S_ISREG(inode->i_mode), S_ISDIR(inode->i_mode), | |
1211 | S_ISLNK(inode->i_mode), inode->i_mode, inode->i_size); | |
1212 | ||
1213 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC; | |
1214 | ||
1215 | /* | |
1216 | * i_mapping is a pointer to i_data during inode initialization. | |
1217 | */ | |
1218 | inode->i_data.a_ops = &pohmelfs_aops; | |
1219 | ||
1220 | if (S_ISREG(inode->i_mode)) { | |
1221 | inode->i_fop = &pohmelfs_file_ops; | |
1222 | inode->i_op = &pohmelfs_file_inode_operations; | |
1223 | } else if (S_ISDIR(inode->i_mode)) { | |
1224 | inode->i_fop = &pohmelfs_dir_fops; | |
1225 | inode->i_op = &pohmelfs_dir_inode_ops; | |
1226 | } else if (S_ISLNK(inode->i_mode)) { | |
1227 | inode->i_op = &pohmelfs_symlink_inode_operations; | |
1228 | inode->i_fop = &pohmelfs_file_ops; | |
1229 | } else { | |
1230 | inode->i_fop = &generic_ro_fops; | |
1231 | } | |
1232 | } | |
1233 | ||
45321ac5 | 1234 | static int pohmelfs_drop_inode(struct inode *inode) |
b3f08cad EP |
1235 | { |
1236 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
1237 | struct pohmelfs_inode *pi = POHMELFS_I(inode); | |
1238 | ||
1239 | spin_lock(&psb->ino_lock); | |
1240 | list_del_init(&pi->inode_entry); | |
1241 | spin_unlock(&psb->ino_lock); | |
1242 | ||
45321ac5 | 1243 | return generic_drop_inode(inode); |
b3f08cad EP |
1244 | } |
1245 | ||
1246 | static struct pohmelfs_inode *pohmelfs_get_inode_from_list(struct pohmelfs_sb *psb, | |
1247 | struct list_head *head, unsigned int *count) | |
1248 | { | |
1249 | struct pohmelfs_inode *pi = NULL; | |
1250 | ||
1251 | spin_lock(&psb->ino_lock); | |
1252 | if (!list_empty(head)) { | |
1253 | pi = list_entry(head->next, struct pohmelfs_inode, | |
1254 | inode_entry); | |
1255 | list_del_init(&pi->inode_entry); | |
1256 | *count = pi->drop_count; | |
1257 | pi->drop_count = 0; | |
1258 | } | |
1259 | spin_unlock(&psb->ino_lock); | |
1260 | ||
1261 | return pi; | |
1262 | } | |
1263 | ||
1264 | static void pohmelfs_flush_transactions(struct pohmelfs_sb *psb) | |
1265 | { | |
1266 | struct pohmelfs_config *c; | |
1267 | ||
1268 | mutex_lock(&psb->state_lock); | |
1269 | list_for_each_entry(c, &psb->state_list, config_entry) { | |
1270 | pohmelfs_state_flush_transactions(&c->state); | |
1271 | } | |
1272 | mutex_unlock(&psb->state_lock); | |
1273 | } | |
1274 | ||
1275 | /* | |
1276 | * ->put_super() callback. Invoked before superblock is destroyed, | |
1277 | * so it has to clean all private data. | |
1278 | */ | |
1279 | static void pohmelfs_put_super(struct super_block *sb) | |
1280 | { | |
1281 | struct pohmelfs_sb *psb = POHMELFS_SB(sb); | |
1282 | struct pohmelfs_inode *pi; | |
ab17a4ac | 1283 | unsigned int count = 0; |
b3f08cad EP |
1284 | unsigned int in_drop_list = 0; |
1285 | struct inode *inode, *tmp; | |
1286 | ||
1287 | dprintk("%s.\n", __func__); | |
1288 | ||
1289 | /* | |
1290 | * Kill pending transactions, which could affect inodes in-flight. | |
1291 | */ | |
1292 | pohmelfs_flush_transactions(psb); | |
1293 | ||
1294 | while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) { | |
1295 | inode = &pi->vfs_inode; | |
1296 | ||
1297 | dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n", | |
1298 | __func__, pi->ino, pi, inode, count); | |
1299 | ||
1300 | if (atomic_read(&inode->i_count) != count) { | |
1301 | printk("%s: ino: %llu, pi: %p, inode: %p, count: %u, i_count: %d.\n", | |
1302 | __func__, pi->ino, pi, inode, count, | |
1303 | atomic_read(&inode->i_count)); | |
1304 | count = atomic_read(&inode->i_count); | |
1305 | in_drop_list++; | |
1306 | } | |
1307 | ||
1308 | while (count--) | |
1309 | iput(&pi->vfs_inode); | |
1310 | } | |
1311 | ||
1312 | list_for_each_entry_safe(inode, tmp, &sb->s_inodes, i_sb_list) { | |
1313 | pi = POHMELFS_I(inode); | |
1314 | ||
1315 | dprintk("%s: ino: %llu, pi: %p, inode: %p, i_count: %u.\n", | |
1316 | __func__, pi->ino, pi, inode, atomic_read(&inode->i_count)); | |
1317 | ||
1318 | /* | |
1319 | * These are special inodes, they were created during | |
1320 | * directory reading or lookup, and were not bound to dentry, | |
1321 | * so they live here with reference counter being 1 and prevent | |
1322 | * umount from succeed since it believes that they are busy. | |
1323 | */ | |
1324 | count = atomic_read(&inode->i_count); | |
1325 | if (count) { | |
1326 | list_del_init(&inode->i_sb_list); | |
1327 | while (count--) | |
1328 | iput(&pi->vfs_inode); | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | psb->trans_scan_timeout = psb->drop_scan_timeout = 0; | |
afe2c511 TH |
1333 | cancel_delayed_work_sync(&psb->dwork); |
1334 | cancel_delayed_work_sync(&psb->drop_dwork); | |
b3f08cad EP |
1335 | flush_scheduled_work(); |
1336 | ||
1337 | dprintk("%s: stopped workqueues.\n", __func__); | |
1338 | ||
1339 | pohmelfs_crypto_exit(psb); | |
1340 | pohmelfs_state_exit(psb); | |
1341 | ||
182374a0 EP |
1342 | bdi_destroy(&psb->bdi); |
1343 | ||
b3f08cad EP |
1344 | kfree(psb); |
1345 | sb->s_fs_info = NULL; | |
b3f08cad EP |
1346 | } |
1347 | ||
b3f08cad EP |
1348 | static int pohmelfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1349 | { | |
1350 | struct super_block *sb = dentry->d_sb; | |
1351 | struct pohmelfs_sb *psb = POHMELFS_SB(sb); | |
1352 | ||
1353 | /* | |
1354 | * There are no filesystem size limits yet. | |
1355 | */ | |
1356 | memset(buf, 0, sizeof(struct kstatfs)); | |
1357 | ||
1358 | buf->f_type = POHMELFS_MAGIC_NUM; /* 'POH.' */ | |
1359 | buf->f_bsize = sb->s_blocksize; | |
1360 | buf->f_files = psb->ino; | |
1361 | buf->f_namelen = 255; | |
1362 | buf->f_files = atomic_long_read(&psb->total_inodes); | |
1363 | buf->f_bfree = buf->f_bavail = psb->avail_size >> PAGE_SHIFT; | |
1364 | buf->f_blocks = psb->total_size >> PAGE_SHIFT; | |
1365 | ||
1366 | dprintk("%s: total: %llu, avail: %llu, inodes: %llu, bsize: %lu.\n", | |
1367 | __func__, psb->total_size, psb->avail_size, buf->f_files, sb->s_blocksize); | |
1368 | ||
1369 | return 0; | |
1370 | } | |
1371 | ||
34c80b1d | 1372 | static int pohmelfs_show_options(struct seq_file *seq, struct dentry *root) |
b3f08cad | 1373 | { |
34c80b1d | 1374 | struct pohmelfs_sb *psb = POHMELFS_SB(root->d_sb); |
b3f08cad EP |
1375 | |
1376 | seq_printf(seq, ",idx=%u", psb->idx); | |
1377 | seq_printf(seq, ",trans_scan_timeout=%u", jiffies_to_msecs(psb->trans_scan_timeout)); | |
1378 | seq_printf(seq, ",drop_scan_timeout=%u", jiffies_to_msecs(psb->drop_scan_timeout)); | |
1379 | seq_printf(seq, ",wait_on_page_timeout=%u", jiffies_to_msecs(psb->wait_on_page_timeout)); | |
1380 | seq_printf(seq, ",trans_retries=%u", psb->trans_retries); | |
1381 | seq_printf(seq, ",crypto_thread_num=%u", psb->crypto_thread_num); | |
1382 | seq_printf(seq, ",trans_max_pages=%u", psb->trans_max_pages); | |
1383 | seq_printf(seq, ",mcache_timeout=%u", jiffies_to_msecs(psb->mcache_timeout)); | |
1384 | if (psb->crypto_fail_unsupported) | |
1385 | seq_printf(seq, ",crypto_fail_unsupported"); | |
1386 | ||
1387 | return 0; | |
1388 | } | |
1389 | ||
b3f08cad EP |
1390 | enum { |
1391 | pohmelfs_opt_idx, | |
e5043424 EP |
1392 | pohmelfs_opt_crypto_thread_num, |
1393 | pohmelfs_opt_trans_max_pages, | |
1394 | pohmelfs_opt_crypto_fail_unsupported, | |
1395 | ||
1396 | /* Remountable options */ | |
b3f08cad EP |
1397 | pohmelfs_opt_trans_scan_timeout, |
1398 | pohmelfs_opt_drop_scan_timeout, | |
1399 | pohmelfs_opt_wait_on_page_timeout, | |
1400 | pohmelfs_opt_trans_retries, | |
b3f08cad EP |
1401 | pohmelfs_opt_mcache_timeout, |
1402 | }; | |
1403 | ||
1404 | static struct match_token pohmelfs_tokens[] = { | |
1405 | {pohmelfs_opt_idx, "idx=%u"}, | |
e5043424 EP |
1406 | {pohmelfs_opt_crypto_thread_num, "crypto_thread_num=%u"}, |
1407 | {pohmelfs_opt_trans_max_pages, "trans_max_pages=%u"}, | |
1408 | {pohmelfs_opt_crypto_fail_unsupported, "crypto_fail_unsupported"}, | |
b3f08cad EP |
1409 | {pohmelfs_opt_trans_scan_timeout, "trans_scan_timeout=%u"}, |
1410 | {pohmelfs_opt_drop_scan_timeout, "drop_scan_timeout=%u"}, | |
1411 | {pohmelfs_opt_wait_on_page_timeout, "wait_on_page_timeout=%u"}, | |
1412 | {pohmelfs_opt_trans_retries, "trans_retries=%u"}, | |
b3f08cad EP |
1413 | {pohmelfs_opt_mcache_timeout, "mcache_timeout=%u"}, |
1414 | }; | |
1415 | ||
e5043424 | 1416 | static int pohmelfs_parse_options(char *options, struct pohmelfs_sb *psb, int remount) |
b3f08cad EP |
1417 | { |
1418 | char *p; | |
1419 | substring_t args[MAX_OPT_ARGS]; | |
1420 | int option, err; | |
1421 | ||
1422 | if (!options) | |
1423 | return 0; | |
1424 | ||
1425 | while ((p = strsep(&options, ",")) != NULL) { | |
1426 | int token; | |
1427 | if (!*p) | |
1428 | continue; | |
1429 | ||
1430 | token = match_token(p, pohmelfs_tokens, args); | |
1431 | ||
1432 | err = match_int(&args[0], &option); | |
1433 | if (err) | |
1434 | return err; | |
1435 | ||
e5043424 EP |
1436 | if (remount && token <= pohmelfs_opt_crypto_fail_unsupported) |
1437 | continue; | |
1438 | ||
b3f08cad | 1439 | switch (token) { |
5d6892b3 RP |
1440 | case pohmelfs_opt_idx: |
1441 | psb->idx = option; | |
1442 | break; | |
1443 | case pohmelfs_opt_trans_scan_timeout: | |
1444 | psb->trans_scan_timeout = msecs_to_jiffies(option); | |
1445 | break; | |
1446 | case pohmelfs_opt_drop_scan_timeout: | |
1447 | psb->drop_scan_timeout = msecs_to_jiffies(option); | |
1448 | break; | |
1449 | case pohmelfs_opt_wait_on_page_timeout: | |
1450 | psb->wait_on_page_timeout = msecs_to_jiffies(option); | |
1451 | break; | |
1452 | case pohmelfs_opt_mcache_timeout: | |
1453 | psb->mcache_timeout = msecs_to_jiffies(option); | |
1454 | break; | |
1455 | case pohmelfs_opt_trans_retries: | |
1456 | psb->trans_retries = option; | |
1457 | break; | |
1458 | case pohmelfs_opt_crypto_thread_num: | |
1459 | psb->crypto_thread_num = option; | |
1460 | break; | |
1461 | case pohmelfs_opt_trans_max_pages: | |
1462 | psb->trans_max_pages = option; | |
1463 | break; | |
1464 | case pohmelfs_opt_crypto_fail_unsupported: | |
1465 | psb->crypto_fail_unsupported = 1; | |
1466 | break; | |
1467 | default: | |
1468 | return -EINVAL; | |
b3f08cad EP |
1469 | } |
1470 | } | |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
e5043424 EP |
1475 | static int pohmelfs_remount(struct super_block *sb, int *flags, char *data) |
1476 | { | |
1477 | int err; | |
1478 | struct pohmelfs_sb *psb = POHMELFS_SB(sb); | |
1479 | unsigned long old_sb_flags = sb->s_flags; | |
1480 | ||
1481 | err = pohmelfs_parse_options(data, psb, 1); | |
1482 | if (err) | |
1483 | goto err_out_restore; | |
1484 | ||
1485 | if (!(*flags & MS_RDONLY)) | |
1486 | sb->s_flags &= ~MS_RDONLY; | |
1487 | return 0; | |
1488 | ||
1489 | err_out_restore: | |
1490 | sb->s_flags = old_sb_flags; | |
1491 | return err; | |
1492 | } | |
1493 | ||
b3f08cad EP |
1494 | static void pohmelfs_flush_inode(struct pohmelfs_inode *pi, unsigned int count) |
1495 | { | |
1496 | struct inode *inode = &pi->vfs_inode; | |
1497 | ||
1498 | dprintk("%s: %p: ino: %llu, owned: %d.\n", | |
1499 | __func__, inode, pi->ino, test_bit(NETFS_INODE_OWNED, &pi->state)); | |
1500 | ||
1501 | mutex_lock(&inode->i_mutex); | |
1502 | if (test_and_clear_bit(NETFS_INODE_OWNED, &pi->state)) { | |
1503 | filemap_fdatawrite(inode->i_mapping); | |
1504 | inode->i_sb->s_op->write_inode(inode, 0); | |
1505 | } | |
1506 | ||
2d7cf8ef | 1507 | #ifdef POHMELFS_TRUNCATE_ON_INODE_FLUSH |
b3f08cad | 1508 | truncate_inode_pages(inode->i_mapping, 0); |
2d7cf8ef | 1509 | #endif |
b3f08cad EP |
1510 | |
1511 | pohmelfs_data_unlock(pi, 0, ~0, POHMELFS_WRITE_LOCK); | |
1512 | mutex_unlock(&inode->i_mutex); | |
1513 | } | |
1514 | ||
1515 | static void pohmelfs_put_inode_count(struct pohmelfs_inode *pi, unsigned int count) | |
1516 | { | |
1517 | dprintk("%s: ino: %llu, pi: %p, inode: %p, count: %u.\n", | |
1518 | __func__, pi->ino, pi, &pi->vfs_inode, count); | |
1519 | ||
1520 | if (test_and_clear_bit(NETFS_INODE_NEED_FLUSH, &pi->state)) | |
1521 | pohmelfs_flush_inode(pi, count); | |
1522 | ||
1523 | while (count--) | |
1524 | iput(&pi->vfs_inode); | |
1525 | } | |
1526 | ||
1527 | static void pohmelfs_drop_scan(struct work_struct *work) | |
1528 | { | |
1529 | struct pohmelfs_sb *psb = | |
1530 | container_of(work, struct pohmelfs_sb, drop_dwork.work); | |
1531 | struct pohmelfs_inode *pi; | |
1532 | unsigned int count = 0; | |
1533 | ||
dc828461 | 1534 | while ((pi = pohmelfs_get_inode_from_list(psb, &psb->drop_list, &count))) |
b3f08cad | 1535 | pohmelfs_put_inode_count(pi, count); |
dc828461 | 1536 | |
b3f08cad EP |
1537 | pohmelfs_check_states(psb); |
1538 | ||
1539 | if (psb->drop_scan_timeout) | |
1540 | schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout); | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * Run through all transactions starting from the oldest, | |
1545 | * drop transaction from current state and try to send it | |
1546 | * to all remote nodes, which are currently installed. | |
1547 | */ | |
1548 | static void pohmelfs_trans_scan_state(struct netfs_state *st) | |
1549 | { | |
1550 | struct rb_node *rb_node; | |
1551 | struct netfs_trans_dst *dst; | |
1552 | struct pohmelfs_sb *psb = st->psb; | |
1553 | unsigned int timeout = psb->trans_scan_timeout; | |
1554 | struct netfs_trans *t; | |
1555 | int err; | |
1556 | ||
1557 | mutex_lock(&st->trans_lock); | |
1558 | for (rb_node = rb_first(&st->trans_root); rb_node; ) { | |
1559 | dst = rb_entry(rb_node, struct netfs_trans_dst, state_entry); | |
1560 | t = dst->trans; | |
1561 | ||
1562 | if (timeout && time_after(dst->send_time + timeout, jiffies) | |
1563 | && dst->retries == 0) | |
1564 | break; | |
1565 | ||
1566 | dprintk("%s: t: %p, gen: %u, st: %p, retries: %u, max: %u.\n", | |
1567 | __func__, t, t->gen, st, dst->retries, psb->trans_retries); | |
1568 | netfs_trans_get(t); | |
1569 | ||
1570 | rb_node = rb_next(rb_node); | |
1571 | ||
1572 | err = -ETIMEDOUT; | |
dc828461 | 1573 | if (timeout && (++dst->retries < psb->trans_retries)) |
b3f08cad | 1574 | err = netfs_trans_resend(t, psb); |
b3f08cad EP |
1575 | |
1576 | if (err || (t->flags & NETFS_TRANS_SINGLE_DST)) { | |
1577 | if (netfs_trans_remove_nolock(dst, st)) | |
1578 | netfs_trans_drop_dst_nostate(dst); | |
1579 | } | |
1580 | ||
1581 | t->result = err; | |
1582 | netfs_trans_put(t); | |
1583 | } | |
1584 | mutex_unlock(&st->trans_lock); | |
1585 | } | |
1586 | ||
1587 | /* | |
1588 | * Walk through all installed network states and resend all | |
1589 | * transactions, which are old enough. | |
1590 | */ | |
1591 | static void pohmelfs_trans_scan(struct work_struct *work) | |
1592 | { | |
1593 | struct pohmelfs_sb *psb = | |
1594 | container_of(work, struct pohmelfs_sb, dwork.work); | |
1595 | struct netfs_state *st; | |
1596 | struct pohmelfs_config *c; | |
1597 | ||
1598 | mutex_lock(&psb->state_lock); | |
1599 | list_for_each_entry(c, &psb->state_list, config_entry) { | |
1600 | st = &c->state; | |
1601 | ||
1602 | pohmelfs_trans_scan_state(st); | |
1603 | } | |
1604 | mutex_unlock(&psb->state_lock); | |
1605 | ||
1606 | /* | |
1607 | * If no timeout specified then system is in the middle of umount process, | |
1608 | * so no need to reschedule scanning process again. | |
1609 | */ | |
1610 | if (psb->trans_scan_timeout) | |
1611 | schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout); | |
1612 | } | |
1613 | ||
1614 | int pohmelfs_meta_command_data(struct pohmelfs_inode *pi, u64 id, unsigned int cmd_op, char *addon, | |
1615 | unsigned int flags, netfs_trans_complete_t complete, void *priv, u64 start) | |
1616 | { | |
1617 | struct inode *inode = &pi->vfs_inode; | |
1618 | struct pohmelfs_sb *psb = POHMELFS_SB(inode->i_sb); | |
1619 | int err = 0, sz; | |
1620 | struct netfs_trans *t; | |
1621 | int path_len, addon_len = 0; | |
1622 | void *data; | |
1623 | struct netfs_inode_info *info; | |
1624 | struct netfs_cmd *cmd; | |
1625 | ||
1626 | dprintk("%s: ino: %llu, cmd: %u, addon: %p.\n", __func__, pi->ino, cmd_op, addon); | |
1627 | ||
1628 | path_len = pohmelfs_path_length(pi); | |
1629 | if (path_len < 0) { | |
1630 | err = path_len; | |
1631 | goto err_out_exit; | |
1632 | } | |
1633 | ||
1634 | if (addon) | |
1635 | addon_len = strlen(addon) + 1; /* 0-byte */ | |
1636 | sz = addon_len; | |
1637 | ||
1638 | if (cmd_op == NETFS_INODE_INFO) | |
1639 | sz += sizeof(struct netfs_inode_info); | |
1640 | ||
1641 | t = netfs_trans_alloc(psb, sz + path_len, flags, 0); | |
1642 | if (!t) { | |
1643 | err = -ENOMEM; | |
1644 | goto err_out_exit; | |
1645 | } | |
1646 | t->complete = complete; | |
1647 | t->private = priv; | |
1648 | ||
1649 | cmd = netfs_trans_current(t); | |
1650 | data = (void *)(cmd + 1); | |
1651 | ||
1652 | if (cmd_op == NETFS_INODE_INFO) { | |
1653 | info = (struct netfs_inode_info *)(cmd + 1); | |
1654 | data = (void *)(info + 1); | |
1655 | ||
1656 | /* | |
1657 | * We are under i_mutex, can read and change whatever we want... | |
1658 | */ | |
1659 | info->mode = inode->i_mode; | |
1660 | info->nlink = inode->i_nlink; | |
1661 | info->uid = inode->i_uid; | |
1662 | info->gid = inode->i_gid; | |
1663 | info->blocks = inode->i_blocks; | |
1664 | info->rdev = inode->i_rdev; | |
1665 | info->size = inode->i_size; | |
1666 | info->version = inode->i_version; | |
1667 | ||
1668 | netfs_convert_inode_info(info); | |
1669 | } | |
1670 | ||
1671 | path_len = pohmelfs_construct_path_string(pi, data, path_len); | |
1672 | if (path_len < 0) | |
1673 | goto err_out_free; | |
1674 | ||
1675 | dprintk("%s: path_len: %d.\n", __func__, path_len); | |
1676 | ||
1677 | if (addon) { | |
1678 | path_len--; /* Do not place null-byte before the addon */ | |
1679 | path_len += sprintf(data + path_len, "/%s", addon) + 1; /* 0 - byte */ | |
1680 | } | |
1681 | ||
1682 | sz += path_len; | |
1683 | ||
1684 | cmd->cmd = cmd_op; | |
1685 | cmd->ext = path_len; | |
1686 | cmd->size = sz; | |
1687 | cmd->id = id; | |
1688 | cmd->start = start; | |
1689 | ||
1690 | netfs_convert_cmd(cmd); | |
1691 | netfs_trans_update(cmd, t, sz); | |
1692 | ||
1693 | /* | |
1694 | * Note, that it is possible to leak error here: transaction callback will not | |
1695 | * be invoked for allocation path failure. | |
1696 | */ | |
1697 | return netfs_trans_finish(t, psb); | |
1698 | ||
1699 | err_out_free: | |
1700 | netfs_trans_free(t); | |
1701 | err_out_exit: | |
1702 | if (complete) | |
1703 | complete(NULL, 0, priv, err); | |
1704 | return err; | |
1705 | } | |
1706 | ||
1707 | int pohmelfs_meta_command(struct pohmelfs_inode *pi, unsigned int cmd_op, unsigned int flags, | |
1708 | netfs_trans_complete_t complete, void *priv, u64 start) | |
1709 | { | |
1710 | return pohmelfs_meta_command_data(pi, pi->ino, cmd_op, NULL, flags, complete, priv, start); | |
1711 | } | |
1712 | ||
1713 | /* | |
1714 | * Send request and wait for POHMELFS root capabilities response, | |
1715 | * which will update server's informaion about size of the export, | |
1716 | * permissions, number of objects, available size and so on. | |
1717 | */ | |
1718 | static int pohmelfs_root_handshake(struct pohmelfs_sb *psb) | |
1719 | { | |
1720 | struct netfs_trans *t; | |
1721 | struct netfs_cmd *cmd; | |
1722 | int err = -ENOMEM; | |
1723 | ||
1724 | t = netfs_trans_alloc(psb, 0, 0, 0); | |
1725 | if (!t) | |
1726 | goto err_out_exit; | |
1727 | ||
1728 | cmd = netfs_trans_current(t); | |
1729 | ||
1730 | cmd->cmd = NETFS_CAPABILITIES; | |
1731 | cmd->id = POHMELFS_ROOT_CAPABILITIES; | |
1732 | cmd->size = 0; | |
1733 | cmd->start = 0; | |
1734 | cmd->ext = 0; | |
1735 | cmd->csize = 0; | |
1736 | ||
1737 | netfs_convert_cmd(cmd); | |
1738 | netfs_trans_update(cmd, t, 0); | |
1739 | ||
1740 | err = netfs_trans_finish(t, psb); | |
1741 | if (err) | |
1742 | goto err_out_exit; | |
1743 | ||
1744 | psb->flags = ~0; | |
1745 | err = wait_event_interruptible_timeout(psb->wait, | |
1746 | (psb->flags != ~0), | |
1747 | psb->wait_on_page_timeout); | |
2d7cf8ef | 1748 | if (!err) |
b3f08cad | 1749 | err = -ETIMEDOUT; |
2d7cf8ef | 1750 | else if (err > 0) |
b3f08cad | 1751 | err = -psb->flags; |
b3f08cad EP |
1752 | |
1753 | if (err) | |
1754 | goto err_out_exit; | |
1755 | ||
1756 | return 0; | |
1757 | ||
1758 | err_out_exit: | |
1759 | return err; | |
1760 | } | |
1761 | ||
64132379 | 1762 | static int pohmelfs_show_stats(struct seq_file *m, struct dentry *root) |
f2739de1 EP |
1763 | { |
1764 | struct netfs_state *st; | |
1765 | struct pohmelfs_ctl *ctl; | |
64132379 | 1766 | struct pohmelfs_sb *psb = POHMELFS_SB(root->d_sb); |
f2739de1 EP |
1767 | struct pohmelfs_config *c; |
1768 | ||
1769 | mutex_lock(&psb->state_lock); | |
1770 | ||
1771 | seq_printf(m, "\nidx addr(:port) socket_type protocol active priority permissions\n"); | |
1772 | ||
1773 | list_for_each_entry(c, &psb->state_list, config_entry) { | |
1774 | st = &c->state; | |
1775 | ctl = &st->ctl; | |
1776 | ||
1777 | seq_printf(m, "%u ", ctl->idx); | |
1778 | if (ctl->addr.sa_family == AF_INET) { | |
1779 | struct sockaddr_in *sin = (struct sockaddr_in *)&st->ctl.addr; | |
0c14c06c | 1780 | seq_printf(m, "%pI4:%u", &sin->sin_addr.s_addr, ntohs(sin->sin_port)); |
f2739de1 EP |
1781 | } else if (ctl->addr.sa_family == AF_INET6) { |
1782 | struct sockaddr_in6 *sin = (struct sockaddr_in6 *)&st->ctl.addr; | |
1783 | seq_printf(m, "%pi6:%u", &sin->sin6_addr, ntohs(sin->sin6_port)); | |
1784 | } else { | |
1785 | unsigned int i; | |
5d6892b3 | 1786 | for (i = 0; i < ctl->addrlen; ++i) |
f2739de1 EP |
1787 | seq_printf(m, "%02x.", ctl->addr.addr[i]); |
1788 | } | |
1789 | ||
1790 | seq_printf(m, " %u %u %d %u %x\n", | |
1791 | ctl->type, ctl->proto, | |
1792 | st->socket != NULL, | |
1793 | ctl->prio, ctl->perm); | |
1794 | } | |
1795 | mutex_unlock(&psb->state_lock); | |
1796 | ||
1797 | return 0; | |
1798 | } | |
1799 | ||
e5043424 EP |
1800 | static const struct super_operations pohmelfs_sb_ops = { |
1801 | .alloc_inode = pohmelfs_alloc_inode, | |
1802 | .destroy_inode = pohmelfs_destroy_inode, | |
1803 | .drop_inode = pohmelfs_drop_inode, | |
1804 | .write_inode = pohmelfs_write_inode, | |
1805 | .put_super = pohmelfs_put_super, | |
1806 | .remount_fs = pohmelfs_remount, | |
1807 | .statfs = pohmelfs_statfs, | |
1808 | .show_options = pohmelfs_show_options, | |
f2739de1 | 1809 | .show_stats = pohmelfs_show_stats, |
e5043424 EP |
1810 | }; |
1811 | ||
b3f08cad EP |
1812 | /* |
1813 | * Allocate private superblock and create root dir. | |
1814 | */ | |
1815 | static int pohmelfs_fill_super(struct super_block *sb, void *data, int silent) | |
1816 | { | |
1817 | struct pohmelfs_sb *psb; | |
1818 | int err = -ENOMEM; | |
1819 | struct inode *root; | |
1820 | struct pohmelfs_inode *npi; | |
1821 | struct qstr str; | |
1822 | ||
b3f08cad EP |
1823 | psb = kzalloc(sizeof(struct pohmelfs_sb), GFP_KERNEL); |
1824 | if (!psb) | |
1825 | goto err_out_exit; | |
1826 | ||
182374a0 EP |
1827 | err = bdi_init(&psb->bdi); |
1828 | if (err) | |
1829 | goto err_out_free_sb; | |
1830 | ||
1831 | err = bdi_register(&psb->bdi, NULL, "pfs-%d", atomic_inc_return(&psb_bdi_num)); | |
1832 | if (err) { | |
1833 | bdi_destroy(&psb->bdi); | |
1834 | goto err_out_free_sb; | |
1835 | } | |
1836 | ||
b3f08cad EP |
1837 | sb->s_fs_info = psb; |
1838 | sb->s_op = &pohmelfs_sb_ops; | |
1839 | sb->s_magic = POHMELFS_MAGIC_NUM; | |
1840 | sb->s_maxbytes = MAX_LFS_FILESIZE; | |
1841 | sb->s_blocksize = PAGE_SIZE; | |
182374a0 | 1842 | sb->s_bdi = &psb->bdi; |
b3f08cad EP |
1843 | |
1844 | psb->sb = sb; | |
1845 | ||
1846 | psb->ino = 2; | |
1847 | psb->idx = 0; | |
1848 | psb->active_state = NULL; | |
1849 | psb->trans_retries = 5; | |
1850 | psb->trans_data_size = PAGE_SIZE; | |
1851 | psb->drop_scan_timeout = msecs_to_jiffies(1000); | |
1852 | psb->trans_scan_timeout = msecs_to_jiffies(5000); | |
1853 | psb->wait_on_page_timeout = msecs_to_jiffies(5000); | |
1854 | init_waitqueue_head(&psb->wait); | |
1855 | ||
1856 | spin_lock_init(&psb->ino_lock); | |
1857 | ||
1858 | INIT_LIST_HEAD(&psb->drop_list); | |
1859 | ||
1860 | mutex_init(&psb->mcache_lock); | |
1861 | psb->mcache_root = RB_ROOT; | |
1862 | psb->mcache_timeout = msecs_to_jiffies(5000); | |
1863 | atomic_long_set(&psb->mcache_gen, 0); | |
1864 | ||
1865 | psb->trans_max_pages = 100; | |
1866 | ||
1867 | psb->crypto_align_size = 16; | |
1868 | psb->crypto_attached_size = 0; | |
1869 | psb->hash_strlen = 0; | |
1870 | psb->cipher_strlen = 0; | |
1871 | psb->perform_crypto = 0; | |
1872 | psb->crypto_thread_num = 2; | |
1873 | psb->crypto_fail_unsupported = 0; | |
1874 | mutex_init(&psb->crypto_thread_lock); | |
1875 | INIT_LIST_HEAD(&psb->crypto_ready_list); | |
1876 | INIT_LIST_HEAD(&psb->crypto_active_list); | |
1877 | ||
1878 | atomic_set(&psb->trans_gen, 1); | |
76efa5e3 | 1879 | atomic_long_set(&psb->total_inodes, 0); |
b3f08cad EP |
1880 | |
1881 | mutex_init(&psb->state_lock); | |
1882 | INIT_LIST_HEAD(&psb->state_list); | |
1883 | ||
e5043424 | 1884 | err = pohmelfs_parse_options((char *) data, psb, 0); |
b3f08cad | 1885 | if (err) |
182374a0 | 1886 | goto err_out_free_bdi; |
b3f08cad EP |
1887 | |
1888 | err = pohmelfs_copy_crypto(psb); | |
1889 | if (err) | |
182374a0 | 1890 | goto err_out_free_bdi; |
b3f08cad EP |
1891 | |
1892 | err = pohmelfs_state_init(psb); | |
1893 | if (err) | |
1894 | goto err_out_free_strings; | |
1895 | ||
1896 | err = pohmelfs_crypto_init(psb); | |
1897 | if (err) | |
1898 | goto err_out_state_exit; | |
1899 | ||
1900 | err = pohmelfs_root_handshake(psb); | |
1901 | if (err) | |
1902 | goto err_out_crypto_exit; | |
1903 | ||
1904 | str.name = "/"; | |
1905 | str.hash = jhash("/", 1, 0); | |
1906 | str.len = 1; | |
1907 | ||
1908 | npi = pohmelfs_create_entry_local(psb, NULL, &str, 0, 0755|S_IFDIR); | |
1909 | if (IS_ERR(npi)) { | |
1910 | err = PTR_ERR(npi); | |
1911 | goto err_out_crypto_exit; | |
1912 | } | |
872dc5e5 EP |
1913 | set_bit(NETFS_INODE_REMOTE_SYNCED, &npi->state); |
1914 | clear_bit(NETFS_INODE_OWNED, &npi->state); | |
b3f08cad EP |
1915 | |
1916 | root = &npi->vfs_inode; | |
1917 | ||
1918 | sb->s_root = d_alloc_root(root); | |
1919 | if (!sb->s_root) | |
1920 | goto err_out_put_root; | |
1921 | ||
1922 | INIT_DELAYED_WORK(&psb->drop_dwork, pohmelfs_drop_scan); | |
1923 | schedule_delayed_work(&psb->drop_dwork, psb->drop_scan_timeout); | |
1924 | ||
1925 | INIT_DELAYED_WORK(&psb->dwork, pohmelfs_trans_scan); | |
1926 | schedule_delayed_work(&psb->dwork, psb->trans_scan_timeout); | |
1927 | ||
1928 | return 0; | |
1929 | ||
1930 | err_out_put_root: | |
1931 | iput(root); | |
1932 | err_out_crypto_exit: | |
1933 | pohmelfs_crypto_exit(psb); | |
1934 | err_out_state_exit: | |
1935 | pohmelfs_state_exit(psb); | |
1936 | err_out_free_strings: | |
1937 | kfree(psb->cipher_string); | |
1938 | kfree(psb->hash_string); | |
182374a0 EP |
1939 | err_out_free_bdi: |
1940 | bdi_destroy(&psb->bdi); | |
b3f08cad EP |
1941 | err_out_free_sb: |
1942 | kfree(psb); | |
1943 | err_out_exit: | |
1944 | ||
1945 | dprintk("%s: err: %d.\n", __func__, err); | |
1946 | return err; | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * Some VFS magic here... | |
1951 | */ | |
3c26ff6e AV |
1952 | static struct dentry *pohmelfs_mount(struct file_system_type *fs_type, |
1953 | int flags, const char *dev_name, void *data) | |
b3f08cad | 1954 | { |
3c26ff6e | 1955 | return mount_nodev(fs_type, flags, data, pohmelfs_fill_super); |
b3f08cad EP |
1956 | } |
1957 | ||
cee9bb2e EP |
1958 | /* |
1959 | * We need this to sync all inodes earlier, since when writeback | |
1960 | * is invoked from the umount/mntput path dcache is already shrunk, | |
1961 | * see generic_shutdown_super(), and no inodes can access the path. | |
1962 | */ | |
1963 | static void pohmelfs_kill_super(struct super_block *sb) | |
1964 | { | |
d8a8559c | 1965 | sync_inodes_sb(sb); |
cee9bb2e EP |
1966 | kill_anon_super(sb); |
1967 | } | |
1968 | ||
b3f08cad EP |
1969 | static struct file_system_type pohmel_fs_type = { |
1970 | .owner = THIS_MODULE, | |
1971 | .name = "pohmel", | |
3c26ff6e | 1972 | .mount = pohmelfs_mount, |
cee9bb2e | 1973 | .kill_sb = pohmelfs_kill_super, |
b3f08cad EP |
1974 | }; |
1975 | ||
1976 | /* | |
1977 | * Cache and module initializations and freeing routings. | |
1978 | */ | |
1979 | static void pohmelfs_init_once(void *data) | |
1980 | { | |
1981 | struct pohmelfs_inode *pi = data; | |
1982 | ||
1983 | inode_init_once(&pi->vfs_inode); | |
1984 | } | |
1985 | ||
1986 | static int __init pohmelfs_init_inodecache(void) | |
1987 | { | |
1988 | pohmelfs_inode_cache = kmem_cache_create("pohmelfs_inode_cache", | |
1989 | sizeof(struct pohmelfs_inode), | |
1990 | 0, (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD), | |
1991 | pohmelfs_init_once); | |
1992 | if (!pohmelfs_inode_cache) | |
1993 | return -ENOMEM; | |
1994 | ||
1995 | return 0; | |
1996 | } | |
1997 | ||
1998 | static void pohmelfs_destroy_inodecache(void) | |
1999 | { | |
2000 | kmem_cache_destroy(pohmelfs_inode_cache); | |
2001 | } | |
2002 | ||
2003 | static int __init init_pohmel_fs(void) | |
2004 | { | |
2005 | int err; | |
2006 | ||
2007 | err = pohmelfs_config_init(); | |
2008 | if (err) | |
2009 | goto err_out_exit; | |
2010 | ||
2011 | err = pohmelfs_init_inodecache(); | |
2012 | if (err) | |
2013 | goto err_out_config_exit; | |
2014 | ||
2015 | err = pohmelfs_mcache_init(); | |
2016 | if (err) | |
2017 | goto err_out_destroy; | |
2018 | ||
2019 | err = netfs_trans_init(); | |
2020 | if (err) | |
2021 | goto err_out_mcache_exit; | |
2022 | ||
2023 | err = register_filesystem(&pohmel_fs_type); | |
2024 | if (err) | |
2025 | goto err_out_trans; | |
2026 | ||
2027 | return 0; | |
2028 | ||
2029 | err_out_trans: | |
2030 | netfs_trans_exit(); | |
2031 | err_out_mcache_exit: | |
2032 | pohmelfs_mcache_exit(); | |
2033 | err_out_destroy: | |
2034 | pohmelfs_destroy_inodecache(); | |
2035 | err_out_config_exit: | |
2036 | pohmelfs_config_exit(); | |
2037 | err_out_exit: | |
2038 | return err; | |
2039 | } | |
2040 | ||
2041 | static void __exit exit_pohmel_fs(void) | |
2042 | { | |
5d6892b3 | 2043 | unregister_filesystem(&pohmel_fs_type); |
b3f08cad EP |
2044 | pohmelfs_destroy_inodecache(); |
2045 | pohmelfs_mcache_exit(); | |
2046 | pohmelfs_config_exit(); | |
2047 | netfs_trans_exit(); | |
2048 | } | |
2049 | ||
2050 | module_init(init_pohmel_fs); | |
2051 | module_exit(exit_pohmel_fs); | |
2052 | ||
2053 | MODULE_LICENSE("GPL"); | |
2054 | MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>"); | |
2055 | MODULE_DESCRIPTION("Pohmel filesystem"); |