mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / afs / file.c
blob510cba15fa560549de1227be97baa0770d783762
1 /* AFS filesystem file handling
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/fs.h>
16 #include <linux/pagemap.h>
17 #include <linux/writeback.h>
18 #include <linux/gfp.h>
19 #include <linux/task_io_accounting_ops.h>
20 #include "internal.h"
22 static int afs_readpage(struct file *file, struct page *page);
23 static void afs_invalidatepage(struct page *page, unsigned int offset,
24 unsigned int length);
25 static int afs_releasepage(struct page *page, gfp_t gfp_flags);
26 static int afs_launder_page(struct page *page);
28 static int afs_readpages(struct file *filp, struct address_space *mapping,
29 struct list_head *pages, unsigned nr_pages);
31 const struct file_operations afs_file_operations = {
32 .open = afs_open,
33 .flush = afs_flush,
34 .release = afs_release,
35 .llseek = generic_file_llseek,
36 .read_iter = generic_file_read_iter,
37 .write_iter = afs_file_write,
38 .mmap = generic_file_readonly_mmap,
39 .splice_read = generic_file_splice_read,
40 .fsync = afs_fsync,
41 .lock = afs_lock,
42 .flock = afs_flock,
45 const struct inode_operations afs_file_inode_operations = {
46 .getattr = afs_getattr,
47 .setattr = afs_setattr,
48 .permission = afs_permission,
49 .listxattr = afs_listxattr,
52 const struct address_space_operations afs_fs_aops = {
53 .readpage = afs_readpage,
54 .readpages = afs_readpages,
55 .set_page_dirty = afs_set_page_dirty,
56 .launder_page = afs_launder_page,
57 .releasepage = afs_releasepage,
58 .invalidatepage = afs_invalidatepage,
59 .write_begin = afs_write_begin,
60 .write_end = afs_write_end,
61 .writepage = afs_writepage,
62 .writepages = afs_writepages,
66 * open an AFS file or directory and attach a key to it
68 int afs_open(struct inode *inode, struct file *file)
70 struct afs_vnode *vnode = AFS_FS_I(inode);
71 struct key *key;
72 int ret;
74 _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
76 key = afs_request_key(vnode->volume->cell);
77 if (IS_ERR(key)) {
78 _leave(" = %ld [key]", PTR_ERR(key));
79 return PTR_ERR(key);
82 ret = afs_validate(vnode, key);
83 if (ret < 0) {
84 _leave(" = %d [val]", ret);
85 return ret;
88 file->private_data = key;
89 _leave(" = 0");
90 return 0;
94 * release an AFS file or directory and discard its key
96 int afs_release(struct inode *inode, struct file *file)
98 struct afs_vnode *vnode = AFS_FS_I(inode);
100 _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode);
102 key_put(file->private_data);
103 _leave(" = 0");
104 return 0;
108 * Dispose of a ref to a read record.
110 void afs_put_read(struct afs_read *req)
112 int i;
114 if (atomic_dec_and_test(&req->usage)) {
115 for (i = 0; i < req->nr_pages; i++)
116 if (req->pages[i])
117 put_page(req->pages[i]);
118 kfree(req);
122 #ifdef CONFIG_AFS_FSCACHE
124 * deal with notification that a page was read from the cache
126 static void afs_file_readpage_read_complete(struct page *page,
127 void *data,
128 int error)
130 _enter("%p,%p,%d", page, data, error);
132 /* if the read completes with an error, we just unlock the page and let
133 * the VM reissue the readpage */
134 if (!error)
135 SetPageUptodate(page);
136 unlock_page(page);
138 #endif
141 * read page from file, directory or symlink, given a key to use
143 int afs_page_filler(void *data, struct page *page)
145 struct inode *inode = page->mapping->host;
146 struct afs_vnode *vnode = AFS_FS_I(inode);
147 struct afs_read *req;
148 struct key *key = data;
149 int ret;
151 _enter("{%x},{%lu},{%lu}", key_serial(key), inode->i_ino, page->index);
153 BUG_ON(!PageLocked(page));
155 ret = -ESTALE;
156 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
157 goto error;
159 /* is it cached? */
160 #ifdef CONFIG_AFS_FSCACHE
161 ret = fscache_read_or_alloc_page(vnode->cache,
162 page,
163 afs_file_readpage_read_complete,
164 NULL,
165 GFP_KERNEL);
166 #else
167 ret = -ENOBUFS;
168 #endif
169 switch (ret) {
170 /* read BIO submitted (page in cache) */
171 case 0:
172 break;
174 /* page not yet cached */
175 case -ENODATA:
176 _debug("cache said ENODATA");
177 goto go_on;
179 /* page will not be cached */
180 case -ENOBUFS:
181 _debug("cache said ENOBUFS");
182 default:
183 go_on:
184 req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *),
185 GFP_KERNEL);
186 if (!req)
187 goto enomem;
189 /* We request a full page. If the page is a partial one at the
190 * end of the file, the server will return a short read and the
191 * unmarshalling code will clear the unfilled space.
193 atomic_set(&req->usage, 1);
194 req->pos = (loff_t)page->index << PAGE_SHIFT;
195 req->len = PAGE_SIZE;
196 req->nr_pages = 1;
197 req->pages[0] = page;
198 get_page(page);
200 /* read the contents of the file from the server into the
201 * page */
202 ret = afs_vnode_fetch_data(vnode, key, req);
203 afs_put_read(req);
204 if (ret < 0) {
205 if (ret == -ENOENT) {
206 _debug("got NOENT from server"
207 " - marking file deleted and stale");
208 set_bit(AFS_VNODE_DELETED, &vnode->flags);
209 ret = -ESTALE;
212 #ifdef CONFIG_AFS_FSCACHE
213 fscache_uncache_page(vnode->cache, page);
214 #endif
215 BUG_ON(PageFsCache(page));
217 if (ret == -EINTR ||
218 ret == -ENOMEM ||
219 ret == -ERESTARTSYS ||
220 ret == -EAGAIN)
221 goto error;
222 goto io_error;
225 SetPageUptodate(page);
227 /* send the page to the cache */
228 #ifdef CONFIG_AFS_FSCACHE
229 if (PageFsCache(page) &&
230 fscache_write_page(vnode->cache, page, GFP_KERNEL) != 0) {
231 fscache_uncache_page(vnode->cache, page);
232 BUG_ON(PageFsCache(page));
234 #endif
235 unlock_page(page);
238 _leave(" = 0");
239 return 0;
241 io_error:
242 SetPageError(page);
243 goto error;
244 enomem:
245 ret = -ENOMEM;
246 error:
247 unlock_page(page);
248 _leave(" = %d", ret);
249 return ret;
253 * read page from file, directory or symlink, given a file to nominate the key
254 * to be used
256 static int afs_readpage(struct file *file, struct page *page)
258 struct key *key;
259 int ret;
261 if (file) {
262 key = file->private_data;
263 ASSERT(key != NULL);
264 ret = afs_page_filler(key, page);
265 } else {
266 struct inode *inode = page->mapping->host;
267 key = afs_request_key(AFS_FS_S(inode->i_sb)->volume->cell);
268 if (IS_ERR(key)) {
269 ret = PTR_ERR(key);
270 } else {
271 ret = afs_page_filler(key, page);
272 key_put(key);
275 return ret;
279 * Make pages available as they're filled.
281 static void afs_readpages_page_done(struct afs_call *call, struct afs_read *req)
283 #ifdef CONFIG_AFS_FSCACHE
284 struct afs_vnode *vnode = call->reply;
285 #endif
286 struct page *page = req->pages[req->index];
288 req->pages[req->index] = NULL;
289 SetPageUptodate(page);
291 /* send the page to the cache */
292 #ifdef CONFIG_AFS_FSCACHE
293 if (PageFsCache(page) &&
294 fscache_write_page(vnode->cache, page, GFP_KERNEL) != 0) {
295 fscache_uncache_page(vnode->cache, page);
296 BUG_ON(PageFsCache(page));
298 #endif
299 unlock_page(page);
300 put_page(page);
304 * Read a contiguous set of pages.
306 static int afs_readpages_one(struct file *file, struct address_space *mapping,
307 struct list_head *pages)
309 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
310 struct afs_read *req;
311 struct list_head *p;
312 struct page *first, *page;
313 struct key *key = file->private_data;
314 pgoff_t index;
315 int ret, n, i;
317 /* Count the number of contiguous pages at the front of the list. Note
318 * that the list goes prev-wards rather than next-wards.
320 first = list_entry(pages->prev, struct page, lru);
321 index = first->index + 1;
322 n = 1;
323 for (p = first->lru.prev; p != pages; p = p->prev) {
324 page = list_entry(p, struct page, lru);
325 if (page->index != index)
326 break;
327 index++;
328 n++;
331 req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *) * n,
332 GFP_NOFS);
333 if (!req)
334 return -ENOMEM;
336 atomic_set(&req->usage, 1);
337 req->page_done = afs_readpages_page_done;
338 req->pos = first->index;
339 req->pos <<= PAGE_SHIFT;
341 /* Transfer the pages to the request. We add them in until one fails
342 * to add to the LRU and then we stop (as that'll make a hole in the
343 * contiguous run.
345 * Note that it's possible for the file size to change whilst we're
346 * doing this, but we rely on the server returning less than we asked
347 * for if the file shrank. We also rely on this to deal with a partial
348 * page at the end of the file.
350 do {
351 page = list_entry(pages->prev, struct page, lru);
352 list_del(&page->lru);
353 index = page->index;
354 if (add_to_page_cache_lru(page, mapping, index,
355 readahead_gfp_mask(mapping))) {
356 #ifdef CONFIG_AFS_FSCACHE
357 fscache_uncache_page(vnode->cache, page);
358 #endif
359 put_page(page);
360 break;
363 req->pages[req->nr_pages++] = page;
364 req->len += PAGE_SIZE;
365 } while (req->nr_pages < n);
367 if (req->nr_pages == 0) {
368 kfree(req);
369 return 0;
372 ret = afs_vnode_fetch_data(vnode, key, req);
373 if (ret < 0)
374 goto error;
376 task_io_account_read(PAGE_SIZE * req->nr_pages);
377 afs_put_read(req);
378 return 0;
380 error:
381 if (ret == -ENOENT) {
382 _debug("got NOENT from server"
383 " - marking file deleted and stale");
384 set_bit(AFS_VNODE_DELETED, &vnode->flags);
385 ret = -ESTALE;
388 for (i = 0; i < req->nr_pages; i++) {
389 page = req->pages[i];
390 if (page) {
391 #ifdef CONFIG_AFS_FSCACHE
392 fscache_uncache_page(vnode->cache, page);
393 #endif
394 SetPageError(page);
395 unlock_page(page);
399 afs_put_read(req);
400 return ret;
404 * read a set of pages
406 static int afs_readpages(struct file *file, struct address_space *mapping,
407 struct list_head *pages, unsigned nr_pages)
409 struct key *key = file->private_data;
410 struct afs_vnode *vnode;
411 int ret = 0;
413 _enter("{%d},{%lu},,%d",
414 key_serial(key), mapping->host->i_ino, nr_pages);
416 ASSERT(key != NULL);
418 vnode = AFS_FS_I(mapping->host);
419 if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
420 _leave(" = -ESTALE");
421 return -ESTALE;
424 /* attempt to read as many of the pages as possible */
425 #ifdef CONFIG_AFS_FSCACHE
426 ret = fscache_read_or_alloc_pages(vnode->cache,
427 mapping,
428 pages,
429 &nr_pages,
430 afs_file_readpage_read_complete,
431 NULL,
432 mapping_gfp_mask(mapping));
433 #else
434 ret = -ENOBUFS;
435 #endif
437 switch (ret) {
438 /* all pages are being read from the cache */
439 case 0:
440 BUG_ON(!list_empty(pages));
441 BUG_ON(nr_pages != 0);
442 _leave(" = 0 [reading all]");
443 return 0;
445 /* there were pages that couldn't be read from the cache */
446 case -ENODATA:
447 case -ENOBUFS:
448 break;
450 /* other error */
451 default:
452 _leave(" = %d", ret);
453 return ret;
456 while (!list_empty(pages)) {
457 ret = afs_readpages_one(file, mapping, pages);
458 if (ret < 0)
459 break;
462 _leave(" = %d [netting]", ret);
463 return ret;
467 * write back a dirty page
469 static int afs_launder_page(struct page *page)
471 _enter("{%lu}", page->index);
473 return 0;
477 * invalidate part or all of a page
478 * - release a page and clean up its private data if offset is 0 (indicating
479 * the entire page)
481 static void afs_invalidatepage(struct page *page, unsigned int offset,
482 unsigned int length)
484 struct afs_writeback *wb = (struct afs_writeback *) page_private(page);
486 _enter("{%lu},%u,%u", page->index, offset, length);
488 BUG_ON(!PageLocked(page));
490 /* we clean up only if the entire page is being invalidated */
491 if (offset == 0 && length == PAGE_SIZE) {
492 #ifdef CONFIG_AFS_FSCACHE
493 if (PageFsCache(page)) {
494 struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
495 fscache_wait_on_page_write(vnode->cache, page);
496 fscache_uncache_page(vnode->cache, page);
498 #endif
500 if (PagePrivate(page)) {
501 if (wb && !PageWriteback(page)) {
502 set_page_private(page, 0);
503 afs_put_writeback(wb);
506 if (!page_private(page))
507 ClearPagePrivate(page);
511 _leave("");
515 * release a page and clean up its private state if it's not busy
516 * - return true if the page can now be released, false if not
518 static int afs_releasepage(struct page *page, gfp_t gfp_flags)
520 struct afs_writeback *wb = (struct afs_writeback *) page_private(page);
521 struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
523 _enter("{{%x:%u}[%lu],%lx},%x",
524 vnode->fid.vid, vnode->fid.vnode, page->index, page->flags,
525 gfp_flags);
527 /* deny if page is being written to the cache and the caller hasn't
528 * elected to wait */
529 #ifdef CONFIG_AFS_FSCACHE
530 if (!fscache_maybe_release_page(vnode->cache, page, gfp_flags)) {
531 _leave(" = F [cache busy]");
532 return 0;
534 #endif
536 if (PagePrivate(page)) {
537 if (wb) {
538 set_page_private(page, 0);
539 afs_put_writeback(wb);
541 ClearPagePrivate(page);
544 /* indicate that the page can be released */
545 _leave(" = T");
546 return 1;