]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
f25dfb5e | 2 | * file.c - NTFS kernel file operations. Part of the Linux-NTFS project. |
1da177e4 | 3 | * |
bfab36e8 | 4 | * Copyright (c) 2001-2007 Anton Altaparmakov |
1da177e4 LT |
5 | * |
6 | * This program/include file is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License as published | |
8 | * by the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program/include file is distributed in the hope that it will be | |
12 | * useful, but WITHOUT ANY WARRANTY; without even the implied warranty | |
13 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program (in the main directory of the Linux-NTFS | |
18 | * distribution in the file COPYING); if not, write to the Free Software | |
19 | * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | */ | |
21 | ||
1da177e4 | 22 | #include <linux/buffer_head.h> |
5a0e3ad6 | 23 | #include <linux/gfp.h> |
98b27036 AA |
24 | #include <linux/pagemap.h> |
25 | #include <linux/pagevec.h> | |
26 | #include <linux/sched.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/uio.h> | |
29 | #include <linux/writeback.h> | |
1da177e4 | 30 | |
98b27036 AA |
31 | #include <asm/page.h> |
32 | #include <asm/uaccess.h> | |
33 | ||
34 | #include "attrib.h" | |
35 | #include "bitmap.h" | |
1da177e4 LT |
36 | #include "inode.h" |
37 | #include "debug.h" | |
98b27036 AA |
38 | #include "lcnalloc.h" |
39 | #include "malloc.h" | |
40 | #include "mft.h" | |
1da177e4 LT |
41 | #include "ntfs.h" |
42 | ||
43 | /** | |
44 | * ntfs_file_open - called when an inode is about to be opened | |
45 | * @vi: inode to be opened | |
46 | * @filp: file structure describing the inode | |
47 | * | |
48 | * Limit file size to the page cache limit on architectures where unsigned long | |
49 | * is 32-bits. This is the most we can do for now without overflowing the page | |
50 | * cache page index. Doing it this way means we don't run into problems because | |
51 | * of existing too large files. It would be better to allow the user to read | |
52 | * the beginning of the file but I doubt very much anyone is going to hit this | |
53 | * check on a 32-bit architecture, so there is no point in adding the extra | |
54 | * complexity required to support this. | |
55 | * | |
56 | * On 64-bit architectures, the check is hopefully optimized away by the | |
57 | * compiler. | |
58 | * | |
59 | * After the check passes, just call generic_file_open() to do its work. | |
60 | */ | |
61 | static int ntfs_file_open(struct inode *vi, struct file *filp) | |
62 | { | |
63 | if (sizeof(unsigned long) < 8) { | |
d4b9ba7b | 64 | if (i_size_read(vi) > MAX_LFS_FILESIZE) |
a9c62a18 | 65 | return -EOVERFLOW; |
1da177e4 LT |
66 | } |
67 | return generic_file_open(vi, filp); | |
68 | } | |
69 | ||
70 | #ifdef NTFS_RW | |
71 | ||
98b27036 AA |
72 | /** |
73 | * ntfs_attr_extend_initialized - extend the initialized size of an attribute | |
74 | * @ni: ntfs inode of the attribute to extend | |
75 | * @new_init_size: requested new initialized size in bytes | |
76 | * @cached_page: store any allocated but unused page here | |
77 | * @lru_pvec: lru-buffering pagevec of the caller | |
78 | * | |
79 | * Extend the initialized size of an attribute described by the ntfs inode @ni | |
80 | * to @new_init_size bytes. This involves zeroing any non-sparse space between | |
81 | * the old initialized size and @new_init_size both in the page cache and on | |
dda65b94 AA |
82 | * disk (if relevant complete pages are already uptodate in the page cache then |
83 | * these are simply marked dirty). | |
98b27036 AA |
84 | * |
85 | * As a side-effect, the file size (vfs inode->i_size) may be incremented as, | |
86 | * in the resident attribute case, it is tied to the initialized size and, in | |
87 | * the non-resident attribute case, it may not fall below the initialized size. | |
88 | * | |
89 | * Note that if the attribute is resident, we do not need to touch the page | |
90 | * cache at all. This is because if the page cache page is not uptodate we | |
91 | * bring it uptodate later, when doing the write to the mft record since we | |
92 | * then already have the page mapped. And if the page is uptodate, the | |
93 | * non-initialized region will already have been zeroed when the page was | |
94 | * brought uptodate and the region may in fact already have been overwritten | |
95 | * with new data via mmap() based writes, so we cannot just zero it. And since | |
96 | * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped | |
97 | * is unspecified, we choose not to do zeroing and thus we do not need to touch | |
dda65b94 AA |
98 | * the page at all. For a more detailed explanation see ntfs_truncate() in |
99 | * fs/ntfs/inode.c. | |
98b27036 | 100 | * |
98b27036 AA |
101 | * Return 0 on success and -errno on error. In the case that an error is |
102 | * encountered it is possible that the initialized size will already have been | |
103 | * incremented some way towards @new_init_size but it is guaranteed that if | |
104 | * this is the case, the necessary zeroing will also have happened and that all | |
105 | * metadata is self-consistent. | |
106 | * | |
1b1dcc1b | 107 | * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be |
dda65b94 | 108 | * held by the caller. |
98b27036 | 109 | */ |
2ec93b0b | 110 | static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size) |
98b27036 AA |
111 | { |
112 | s64 old_init_size; | |
113 | loff_t old_i_size; | |
114 | pgoff_t index, end_index; | |
115 | unsigned long flags; | |
116 | struct inode *vi = VFS_I(ni); | |
117 | ntfs_inode *base_ni; | |
118 | MFT_RECORD *m = NULL; | |
119 | ATTR_RECORD *a; | |
120 | ntfs_attr_search_ctx *ctx = NULL; | |
121 | struct address_space *mapping; | |
122 | struct page *page = NULL; | |
123 | u8 *kattr; | |
124 | int err; | |
125 | u32 attr_len; | |
126 | ||
127 | read_lock_irqsave(&ni->size_lock, flags); | |
128 | old_init_size = ni->initialized_size; | |
129 | old_i_size = i_size_read(vi); | |
130 | BUG_ON(new_init_size > ni->allocated_size); | |
131 | read_unlock_irqrestore(&ni->size_lock, flags); | |
132 | ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, " | |
133 | "old_initialized_size 0x%llx, " | |
134 | "new_initialized_size 0x%llx, i_size 0x%llx.", | |
135 | vi->i_ino, (unsigned)le32_to_cpu(ni->type), | |
136 | (unsigned long long)old_init_size, | |
137 | (unsigned long long)new_init_size, old_i_size); | |
138 | if (!NInoAttr(ni)) | |
139 | base_ni = ni; | |
140 | else | |
141 | base_ni = ni->ext.base_ntfs_ino; | |
142 | /* Use goto to reduce indentation and we need the label below anyway. */ | |
143 | if (NInoNonResident(ni)) | |
144 | goto do_non_resident_extend; | |
145 | BUG_ON(old_init_size != old_i_size); | |
146 | m = map_mft_record(base_ni); | |
147 | if (IS_ERR(m)) { | |
148 | err = PTR_ERR(m); | |
149 | m = NULL; | |
150 | goto err_out; | |
151 | } | |
152 | ctx = ntfs_attr_get_search_ctx(base_ni, m); | |
153 | if (unlikely(!ctx)) { | |
154 | err = -ENOMEM; | |
155 | goto err_out; | |
156 | } | |
157 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
158 | CASE_SENSITIVE, 0, NULL, 0, ctx); | |
159 | if (unlikely(err)) { | |
160 | if (err == -ENOENT) | |
161 | err = -EIO; | |
162 | goto err_out; | |
163 | } | |
164 | m = ctx->mrec; | |
165 | a = ctx->attr; | |
166 | BUG_ON(a->non_resident); | |
167 | /* The total length of the attribute value. */ | |
168 | attr_len = le32_to_cpu(a->data.resident.value_length); | |
169 | BUG_ON(old_i_size != (loff_t)attr_len); | |
170 | /* | |
171 | * Do the zeroing in the mft record and update the attribute size in | |
172 | * the mft record. | |
173 | */ | |
174 | kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset); | |
175 | memset(kattr + attr_len, 0, new_init_size - attr_len); | |
176 | a->data.resident.value_length = cpu_to_le32((u32)new_init_size); | |
177 | /* Finally, update the sizes in the vfs and ntfs inodes. */ | |
178 | write_lock_irqsave(&ni->size_lock, flags); | |
179 | i_size_write(vi, new_init_size); | |
180 | ni->initialized_size = new_init_size; | |
181 | write_unlock_irqrestore(&ni->size_lock, flags); | |
182 | goto done; | |
183 | do_non_resident_extend: | |
184 | /* | |
185 | * If the new initialized size @new_init_size exceeds the current file | |
186 | * size (vfs inode->i_size), we need to extend the file size to the | |
187 | * new initialized size. | |
188 | */ | |
189 | if (new_init_size > old_i_size) { | |
190 | m = map_mft_record(base_ni); | |
191 | if (IS_ERR(m)) { | |
192 | err = PTR_ERR(m); | |
193 | m = NULL; | |
194 | goto err_out; | |
195 | } | |
196 | ctx = ntfs_attr_get_search_ctx(base_ni, m); | |
197 | if (unlikely(!ctx)) { | |
198 | err = -ENOMEM; | |
199 | goto err_out; | |
200 | } | |
201 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
202 | CASE_SENSITIVE, 0, NULL, 0, ctx); | |
203 | if (unlikely(err)) { | |
204 | if (err == -ENOENT) | |
205 | err = -EIO; | |
206 | goto err_out; | |
207 | } | |
208 | m = ctx->mrec; | |
209 | a = ctx->attr; | |
210 | BUG_ON(!a->non_resident); | |
211 | BUG_ON(old_i_size != (loff_t) | |
212 | sle64_to_cpu(a->data.non_resident.data_size)); | |
213 | a->data.non_resident.data_size = cpu_to_sle64(new_init_size); | |
214 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
215 | mark_mft_record_dirty(ctx->ntfs_ino); | |
216 | /* Update the file size in the vfs inode. */ | |
217 | i_size_write(vi, new_init_size); | |
218 | ntfs_attr_put_search_ctx(ctx); | |
219 | ctx = NULL; | |
220 | unmap_mft_record(base_ni); | |
221 | m = NULL; | |
222 | } | |
223 | mapping = vi->i_mapping; | |
224 | index = old_init_size >> PAGE_CACHE_SHIFT; | |
225 | end_index = (new_init_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
226 | do { | |
227 | /* | |
228 | * Read the page. If the page is not present, this will zero | |
229 | * the uninitialized regions for us. | |
230 | */ | |
090d2b18 | 231 | page = read_mapping_page(mapping, index, NULL); |
98b27036 AA |
232 | if (IS_ERR(page)) { |
233 | err = PTR_ERR(page); | |
234 | goto init_err_out; | |
235 | } | |
6fe6900e | 236 | if (unlikely(PageError(page))) { |
98b27036 AA |
237 | page_cache_release(page); |
238 | err = -EIO; | |
239 | goto init_err_out; | |
240 | } | |
241 | /* | |
242 | * Update the initialized size in the ntfs inode. This is | |
243 | * enough to make ntfs_writepage() work. | |
244 | */ | |
245 | write_lock_irqsave(&ni->size_lock, flags); | |
3c6af7fa | 246 | ni->initialized_size = (s64)(index + 1) << PAGE_CACHE_SHIFT; |
98b27036 AA |
247 | if (ni->initialized_size > new_init_size) |
248 | ni->initialized_size = new_init_size; | |
249 | write_unlock_irqrestore(&ni->size_lock, flags); | |
250 | /* Set the page dirty so it gets written out. */ | |
251 | set_page_dirty(page); | |
252 | page_cache_release(page); | |
253 | /* | |
254 | * Play nice with the vm and the rest of the system. This is | |
255 | * very much needed as we can potentially be modifying the | |
256 | * initialised size from a very small value to a really huge | |
257 | * value, e.g. | |
258 | * f = open(somefile, O_TRUNC); | |
259 | * truncate(f, 10GiB); | |
260 | * seek(f, 10GiB); | |
261 | * write(f, 1); | |
262 | * And this would mean we would be marking dirty hundreds of | |
263 | * thousands of pages or as in the above example more than | |
264 | * two and a half million pages! | |
265 | * | |
266 | * TODO: For sparse pages could optimize this workload by using | |
267 | * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This | |
268 | * would be set in readpage for sparse pages and here we would | |
269 | * not need to mark dirty any pages which have this bit set. | |
270 | * The only caveat is that we have to clear the bit everywhere | |
271 | * where we allocate any clusters that lie in the page or that | |
272 | * contain the page. | |
273 | * | |
274 | * TODO: An even greater optimization would be for us to only | |
275 | * call readpage() on pages which are not in sparse regions as | |
276 | * determined from the runlist. This would greatly reduce the | |
277 | * number of pages we read and make dirty in the case of sparse | |
278 | * files. | |
279 | */ | |
280 | balance_dirty_pages_ratelimited(mapping); | |
281 | cond_resched(); | |
282 | } while (++index < end_index); | |
283 | read_lock_irqsave(&ni->size_lock, flags); | |
284 | BUG_ON(ni->initialized_size != new_init_size); | |
285 | read_unlock_irqrestore(&ni->size_lock, flags); | |
286 | /* Now bring in sync the initialized_size in the mft record. */ | |
287 | m = map_mft_record(base_ni); | |
288 | if (IS_ERR(m)) { | |
289 | err = PTR_ERR(m); | |
290 | m = NULL; | |
291 | goto init_err_out; | |
292 | } | |
293 | ctx = ntfs_attr_get_search_ctx(base_ni, m); | |
294 | if (unlikely(!ctx)) { | |
295 | err = -ENOMEM; | |
296 | goto init_err_out; | |
297 | } | |
298 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
299 | CASE_SENSITIVE, 0, NULL, 0, ctx); | |
300 | if (unlikely(err)) { | |
301 | if (err == -ENOENT) | |
302 | err = -EIO; | |
303 | goto init_err_out; | |
304 | } | |
305 | m = ctx->mrec; | |
306 | a = ctx->attr; | |
307 | BUG_ON(!a->non_resident); | |
308 | a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size); | |
309 | done: | |
310 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
311 | mark_mft_record_dirty(ctx->ntfs_ino); | |
312 | if (ctx) | |
313 | ntfs_attr_put_search_ctx(ctx); | |
314 | if (m) | |
315 | unmap_mft_record(base_ni); | |
316 | ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.", | |
317 | (unsigned long long)new_init_size, i_size_read(vi)); | |
318 | return 0; | |
319 | init_err_out: | |
320 | write_lock_irqsave(&ni->size_lock, flags); | |
321 | ni->initialized_size = old_init_size; | |
322 | write_unlock_irqrestore(&ni->size_lock, flags); | |
323 | err_out: | |
324 | if (ctx) | |
325 | ntfs_attr_put_search_ctx(ctx); | |
326 | if (m) | |
327 | unmap_mft_record(base_ni); | |
328 | ntfs_debug("Failed. Returning error code %i.", err); | |
329 | return err; | |
330 | } | |
331 | ||
332 | /** | |
333 | * ntfs_fault_in_pages_readable - | |
334 | * | |
335 | * Fault a number of userspace pages into pagetables. | |
336 | * | |
337 | * Unlike include/linux/pagemap.h::fault_in_pages_readable(), this one copes | |
338 | * with more than two userspace pages as well as handling the single page case | |
339 | * elegantly. | |
340 | * | |
341 | * If you find this difficult to understand, then think of the while loop being | |
342 | * the following code, except that we do without the integer variable ret: | |
343 | * | |
344 | * do { | |
345 | * ret = __get_user(c, uaddr); | |
346 | * uaddr += PAGE_SIZE; | |
347 | * } while (!ret && uaddr < end); | |
348 | * | |
349 | * Note, the final __get_user() may well run out-of-bounds of the user buffer, | |
350 | * but _not_ out-of-bounds of the page the user buffer belongs to, and since | |
351 | * this is only a read and not a write, and since it is still in the same page, | |
352 | * it should not matter and this makes the code much simpler. | |
353 | */ | |
354 | static inline void ntfs_fault_in_pages_readable(const char __user *uaddr, | |
355 | int bytes) | |
356 | { | |
357 | const char __user *end; | |
358 | volatile char c; | |
359 | ||
360 | /* Set @end to the first byte outside the last page we care about. */ | |
bfab36e8 | 361 | end = (const char __user*)PAGE_ALIGN((unsigned long)uaddr + bytes); |
98b27036 AA |
362 | |
363 | while (!__get_user(c, uaddr) && (uaddr += PAGE_SIZE, uaddr < end)) | |
364 | ; | |
365 | } | |
366 | ||
367 | /** | |
368 | * ntfs_fault_in_pages_readable_iovec - | |
369 | * | |
370 | * Same as ntfs_fault_in_pages_readable() but operates on an array of iovecs. | |
371 | */ | |
372 | static inline void ntfs_fault_in_pages_readable_iovec(const struct iovec *iov, | |
373 | size_t iov_ofs, int bytes) | |
374 | { | |
375 | do { | |
376 | const char __user *buf; | |
377 | unsigned len; | |
378 | ||
379 | buf = iov->iov_base + iov_ofs; | |
380 | len = iov->iov_len - iov_ofs; | |
381 | if (len > bytes) | |
382 | len = bytes; | |
383 | ntfs_fault_in_pages_readable(buf, len); | |
384 | bytes -= len; | |
385 | iov++; | |
386 | iov_ofs = 0; | |
387 | } while (bytes); | |
388 | } | |
389 | ||
390 | /** | |
391 | * __ntfs_grab_cache_pages - obtain a number of locked pages | |
392 | * @mapping: address space mapping from which to obtain page cache pages | |
393 | * @index: starting index in @mapping at which to begin obtaining pages | |
394 | * @nr_pages: number of page cache pages to obtain | |
395 | * @pages: array of pages in which to return the obtained page cache pages | |
396 | * @cached_page: allocated but as yet unused page | |
397 | * @lru_pvec: lru-buffering pagevec of caller | |
398 | * | |
af901ca1 | 399 | * Obtain @nr_pages locked page cache pages from the mapping @mapping and |
98b27036 AA |
400 | * starting at index @index. |
401 | * | |
402 | * If a page is newly created, increment its refcount and add it to the | |
403 | * caller's lru-buffering pagevec @lru_pvec. | |
404 | * | |
405 | * This is the same as mm/filemap.c::__grab_cache_page(), except that @nr_pages | |
406 | * are obtained at once instead of just one page and that 0 is returned on | |
407 | * success and -errno on error. | |
408 | * | |
409 | * Note, the page locks are obtained in ascending page index order. | |
410 | */ | |
411 | static inline int __ntfs_grab_cache_pages(struct address_space *mapping, | |
412 | pgoff_t index, const unsigned nr_pages, struct page **pages, | |
413 | struct page **cached_page, struct pagevec *lru_pvec) | |
414 | { | |
415 | int err, nr; | |
416 | ||
417 | BUG_ON(!nr_pages); | |
418 | err = nr = 0; | |
419 | do { | |
420 | pages[nr] = find_lock_page(mapping, index); | |
421 | if (!pages[nr]) { | |
422 | if (!*cached_page) { | |
423 | *cached_page = page_cache_alloc(mapping); | |
424 | if (unlikely(!*cached_page)) { | |
425 | err = -ENOMEM; | |
426 | goto err_out; | |
427 | } | |
428 | } | |
429 | err = add_to_page_cache(*cached_page, mapping, index, | |
430 | GFP_KERNEL); | |
431 | if (unlikely(err)) { | |
432 | if (err == -EEXIST) | |
433 | continue; | |
434 | goto err_out; | |
435 | } | |
436 | pages[nr] = *cached_page; | |
437 | page_cache_get(*cached_page); | |
438 | if (unlikely(!pagevec_add(lru_pvec, *cached_page))) | |
4f98a2fe | 439 | __pagevec_lru_add_file(lru_pvec); |
98b27036 AA |
440 | *cached_page = NULL; |
441 | } | |
442 | index++; | |
443 | nr++; | |
444 | } while (nr < nr_pages); | |
445 | out: | |
446 | return err; | |
447 | err_out: | |
448 | while (nr > 0) { | |
449 | unlock_page(pages[--nr]); | |
450 | page_cache_release(pages[nr]); | |
451 | } | |
452 | goto out; | |
453 | } | |
454 | ||
455 | static inline int ntfs_submit_bh_for_read(struct buffer_head *bh) | |
456 | { | |
457 | lock_buffer(bh); | |
458 | get_bh(bh); | |
459 | bh->b_end_io = end_buffer_read_sync; | |
460 | return submit_bh(READ, bh); | |
461 | } | |
462 | ||
463 | /** | |
464 | * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data | |
465 | * @pages: array of destination pages | |
466 | * @nr_pages: number of pages in @pages | |
467 | * @pos: byte position in file at which the write begins | |
468 | * @bytes: number of bytes to be written | |
469 | * | |
470 | * This is called for non-resident attributes from ntfs_file_buffered_write() | |
1b1dcc1b | 471 | * with i_mutex held on the inode (@pages[0]->mapping->host). There are |
98b27036 AA |
472 | * @nr_pages pages in @pages which are locked but not kmap()ped. The source |
473 | * data has not yet been copied into the @pages. | |
474 | * | |
475 | * Need to fill any holes with actual clusters, allocate buffers if necessary, | |
476 | * ensure all the buffers are mapped, and bring uptodate any buffers that are | |
477 | * only partially being written to. | |
478 | * | |
479 | * If @nr_pages is greater than one, we are guaranteed that the cluster size is | |
480 | * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside | |
481 | * the same cluster and that they are the entirety of that cluster, and that | |
482 | * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole. | |
483 | * | |
484 | * i_size is not to be modified yet. | |
485 | * | |
486 | * Return 0 on success or -errno on error. | |
487 | */ | |
488 | static int ntfs_prepare_pages_for_non_resident_write(struct page **pages, | |
489 | unsigned nr_pages, s64 pos, size_t bytes) | |
490 | { | |
491 | VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend; | |
492 | LCN lcn; | |
493 | s64 bh_pos, vcn_len, end, initialized_size; | |
494 | sector_t lcn_block; | |
495 | struct page *page; | |
496 | struct inode *vi; | |
497 | ntfs_inode *ni, *base_ni = NULL; | |
498 | ntfs_volume *vol; | |
499 | runlist_element *rl, *rl2; | |
500 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
501 | ntfs_attr_search_ctx *ctx = NULL; | |
502 | MFT_RECORD *m = NULL; | |
503 | ATTR_RECORD *a = NULL; | |
504 | unsigned long flags; | |
505 | u32 attr_rec_len = 0; | |
506 | unsigned blocksize, u; | |
507 | int err, mp_size; | |
c49c3111 | 508 | bool rl_write_locked, was_hole, is_retry; |
98b27036 AA |
509 | unsigned char blocksize_bits; |
510 | struct { | |
511 | u8 runlist_merged:1; | |
512 | u8 mft_attr_mapped:1; | |
513 | u8 mp_rebuilt:1; | |
514 | u8 attr_switched:1; | |
515 | } status = { 0, 0, 0, 0 }; | |
516 | ||
517 | BUG_ON(!nr_pages); | |
518 | BUG_ON(!pages); | |
519 | BUG_ON(!*pages); | |
520 | vi = pages[0]->mapping->host; | |
521 | ni = NTFS_I(vi); | |
522 | vol = ni->vol; | |
523 | ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page " | |
d04bd1fb | 524 | "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.", |
98b27036 AA |
525 | vi->i_ino, ni->type, pages[0]->index, nr_pages, |
526 | (long long)pos, bytes); | |
78af34f0 AA |
527 | blocksize = vol->sb->s_blocksize; |
528 | blocksize_bits = vol->sb->s_blocksize_bits; | |
98b27036 AA |
529 | u = 0; |
530 | do { | |
bfab36e8 AA |
531 | page = pages[u]; |
532 | BUG_ON(!page); | |
98b27036 AA |
533 | /* |
534 | * create_empty_buffers() will create uptodate/dirty buffers if | |
535 | * the page is uptodate/dirty. | |
536 | */ | |
537 | if (!page_has_buffers(page)) { | |
538 | create_empty_buffers(page, blocksize, 0); | |
539 | if (unlikely(!page_has_buffers(page))) | |
540 | return -ENOMEM; | |
541 | } | |
542 | } while (++u < nr_pages); | |
c49c3111 | 543 | rl_write_locked = false; |
98b27036 AA |
544 | rl = NULL; |
545 | err = 0; | |
546 | vcn = lcn = -1; | |
547 | vcn_len = 0; | |
548 | lcn_block = -1; | |
c49c3111 | 549 | was_hole = false; |
98b27036 AA |
550 | cpos = pos >> vol->cluster_size_bits; |
551 | end = pos + bytes; | |
552 | cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits; | |
553 | /* | |
554 | * Loop over each page and for each page over each buffer. Use goto to | |
555 | * reduce indentation. | |
556 | */ | |
557 | u = 0; | |
558 | do_next_page: | |
559 | page = pages[u]; | |
560 | bh_pos = (s64)page->index << PAGE_CACHE_SHIFT; | |
561 | bh = head = page_buffers(page); | |
562 | do { | |
563 | VCN cdelta; | |
564 | s64 bh_end; | |
565 | unsigned bh_cofs; | |
566 | ||
567 | /* Clear buffer_new on all buffers to reinitialise state. */ | |
568 | if (buffer_new(bh)) | |
569 | clear_buffer_new(bh); | |
570 | bh_end = bh_pos + blocksize; | |
571 | bh_cpos = bh_pos >> vol->cluster_size_bits; | |
572 | bh_cofs = bh_pos & vol->cluster_size_mask; | |
573 | if (buffer_mapped(bh)) { | |
574 | /* | |
575 | * The buffer is already mapped. If it is uptodate, | |
576 | * ignore it. | |
577 | */ | |
578 | if (buffer_uptodate(bh)) | |
579 | continue; | |
580 | /* | |
581 | * The buffer is not uptodate. If the page is uptodate | |
582 | * set the buffer uptodate and otherwise ignore it. | |
583 | */ | |
584 | if (PageUptodate(page)) { | |
585 | set_buffer_uptodate(bh); | |
586 | continue; | |
587 | } | |
588 | /* | |
589 | * Neither the page nor the buffer are uptodate. If | |
590 | * the buffer is only partially being written to, we | |
591 | * need to read it in before the write, i.e. now. | |
592 | */ | |
593 | if ((bh_pos < pos && bh_end > pos) || | |
594 | (bh_pos < end && bh_end > end)) { | |
595 | /* | |
596 | * If the buffer is fully or partially within | |
597 | * the initialized size, do an actual read. | |
598 | * Otherwise, simply zero the buffer. | |
599 | */ | |
600 | read_lock_irqsave(&ni->size_lock, flags); | |
601 | initialized_size = ni->initialized_size; | |
602 | read_unlock_irqrestore(&ni->size_lock, flags); | |
603 | if (bh_pos < initialized_size) { | |
604 | ntfs_submit_bh_for_read(bh); | |
605 | *wait_bh++ = bh; | |
606 | } else { | |
eebd2aa3 CL |
607 | zero_user(page, bh_offset(bh), |
608 | blocksize); | |
98b27036 AA |
609 | set_buffer_uptodate(bh); |
610 | } | |
611 | } | |
612 | continue; | |
613 | } | |
614 | /* Unmapped buffer. Need to map it. */ | |
615 | bh->b_bdev = vol->sb->s_bdev; | |
616 | /* | |
617 | * If the current buffer is in the same clusters as the map | |
618 | * cache, there is no need to check the runlist again. The | |
619 | * map cache is made up of @vcn, which is the first cached file | |
620 | * cluster, @vcn_len which is the number of cached file | |
621 | * clusters, @lcn is the device cluster corresponding to @vcn, | |
622 | * and @lcn_block is the block number corresponding to @lcn. | |
623 | */ | |
624 | cdelta = bh_cpos - vcn; | |
625 | if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) { | |
626 | map_buffer_cached: | |
627 | BUG_ON(lcn < 0); | |
628 | bh->b_blocknr = lcn_block + | |
629 | (cdelta << (vol->cluster_size_bits - | |
630 | blocksize_bits)) + | |
631 | (bh_cofs >> blocksize_bits); | |
632 | set_buffer_mapped(bh); | |
633 | /* | |
634 | * If the page is uptodate so is the buffer. If the | |
635 | * buffer is fully outside the write, we ignore it if | |
636 | * it was already allocated and we mark it dirty so it | |
637 | * gets written out if we allocated it. On the other | |
638 | * hand, if we allocated the buffer but we are not | |
639 | * marking it dirty we set buffer_new so we can do | |
640 | * error recovery. | |
641 | */ | |
642 | if (PageUptodate(page)) { | |
643 | if (!buffer_uptodate(bh)) | |
644 | set_buffer_uptodate(bh); | |
645 | if (unlikely(was_hole)) { | |
646 | /* We allocated the buffer. */ | |
647 | unmap_underlying_metadata(bh->b_bdev, | |
648 | bh->b_blocknr); | |
649 | if (bh_end <= pos || bh_pos >= end) | |
650 | mark_buffer_dirty(bh); | |
651 | else | |
652 | set_buffer_new(bh); | |
653 | } | |
654 | continue; | |
655 | } | |
656 | /* Page is _not_ uptodate. */ | |
657 | if (likely(!was_hole)) { | |
658 | /* | |
659 | * Buffer was already allocated. If it is not | |
660 | * uptodate and is only partially being written | |
661 | * to, we need to read it in before the write, | |
662 | * i.e. now. | |
663 | */ | |
3aebf25b AA |
664 | if (!buffer_uptodate(bh) && bh_pos < end && |
665 | bh_end > pos && | |
666 | (bh_pos < pos || | |
667 | bh_end > end)) { | |
98b27036 AA |
668 | /* |
669 | * If the buffer is fully or partially | |
670 | * within the initialized size, do an | |
671 | * actual read. Otherwise, simply zero | |
672 | * the buffer. | |
673 | */ | |
674 | read_lock_irqsave(&ni->size_lock, | |
675 | flags); | |
676 | initialized_size = ni->initialized_size; | |
677 | read_unlock_irqrestore(&ni->size_lock, | |
678 | flags); | |
679 | if (bh_pos < initialized_size) { | |
680 | ntfs_submit_bh_for_read(bh); | |
681 | *wait_bh++ = bh; | |
682 | } else { | |
eebd2aa3 CL |
683 | zero_user(page, bh_offset(bh), |
684 | blocksize); | |
98b27036 AA |
685 | set_buffer_uptodate(bh); |
686 | } | |
687 | } | |
688 | continue; | |
689 | } | |
690 | /* We allocated the buffer. */ | |
691 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
692 | /* | |
693 | * If the buffer is fully outside the write, zero it, | |
694 | * set it uptodate, and mark it dirty so it gets | |
695 | * written out. If it is partially being written to, | |
696 | * zero region surrounding the write but leave it to | |
697 | * commit write to do anything else. Finally, if the | |
698 | * buffer is fully being overwritten, do nothing. | |
699 | */ | |
700 | if (bh_end <= pos || bh_pos >= end) { | |
701 | if (!buffer_uptodate(bh)) { | |
eebd2aa3 CL |
702 | zero_user(page, bh_offset(bh), |
703 | blocksize); | |
98b27036 AA |
704 | set_buffer_uptodate(bh); |
705 | } | |
706 | mark_buffer_dirty(bh); | |
707 | continue; | |
708 | } | |
709 | set_buffer_new(bh); | |
710 | if (!buffer_uptodate(bh) && | |
711 | (bh_pos < pos || bh_end > end)) { | |
712 | u8 *kaddr; | |
713 | unsigned pofs; | |
714 | ||
715 | kaddr = kmap_atomic(page, KM_USER0); | |
716 | if (bh_pos < pos) { | |
717 | pofs = bh_pos & ~PAGE_CACHE_MASK; | |
718 | memset(kaddr + pofs, 0, pos - bh_pos); | |
719 | } | |
720 | if (bh_end > end) { | |
721 | pofs = end & ~PAGE_CACHE_MASK; | |
722 | memset(kaddr + pofs, 0, bh_end - end); | |
723 | } | |
724 | kunmap_atomic(kaddr, KM_USER0); | |
725 | flush_dcache_page(page); | |
726 | } | |
727 | continue; | |
728 | } | |
729 | /* | |
730 | * Slow path: this is the first buffer in the cluster. If it | |
731 | * is outside allocated size and is not uptodate, zero it and | |
732 | * set it uptodate. | |
733 | */ | |
734 | read_lock_irqsave(&ni->size_lock, flags); | |
735 | initialized_size = ni->allocated_size; | |
736 | read_unlock_irqrestore(&ni->size_lock, flags); | |
737 | if (bh_pos > initialized_size) { | |
738 | if (PageUptodate(page)) { | |
739 | if (!buffer_uptodate(bh)) | |
740 | set_buffer_uptodate(bh); | |
741 | } else if (!buffer_uptodate(bh)) { | |
eebd2aa3 | 742 | zero_user(page, bh_offset(bh), blocksize); |
98b27036 AA |
743 | set_buffer_uptodate(bh); |
744 | } | |
745 | continue; | |
746 | } | |
c49c3111 | 747 | is_retry = false; |
98b27036 AA |
748 | if (!rl) { |
749 | down_read(&ni->runlist.lock); | |
750 | retry_remap: | |
751 | rl = ni->runlist.rl; | |
752 | } | |
753 | if (likely(rl != NULL)) { | |
754 | /* Seek to element containing target cluster. */ | |
755 | while (rl->length && rl[1].vcn <= bh_cpos) | |
756 | rl++; | |
757 | lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos); | |
758 | if (likely(lcn >= 0)) { | |
759 | /* | |
760 | * Successful remap, setup the map cache and | |
761 | * use that to deal with the buffer. | |
762 | */ | |
c49c3111 | 763 | was_hole = false; |
98b27036 AA |
764 | vcn = bh_cpos; |
765 | vcn_len = rl[1].vcn - vcn; | |
766 | lcn_block = lcn << (vol->cluster_size_bits - | |
767 | blocksize_bits); | |
d5aeaef3 | 768 | cdelta = 0; |
98b27036 | 769 | /* |
3aebf25b AA |
770 | * If the number of remaining clusters touched |
771 | * by the write is smaller or equal to the | |
772 | * number of cached clusters, unlock the | |
773 | * runlist as the map cache will be used from | |
774 | * now on. | |
98b27036 AA |
775 | */ |
776 | if (likely(vcn + vcn_len >= cend)) { | |
777 | if (rl_write_locked) { | |
778 | up_write(&ni->runlist.lock); | |
c49c3111 | 779 | rl_write_locked = false; |
98b27036 AA |
780 | } else |
781 | up_read(&ni->runlist.lock); | |
782 | rl = NULL; | |
783 | } | |
784 | goto map_buffer_cached; | |
785 | } | |
786 | } else | |
787 | lcn = LCN_RL_NOT_MAPPED; | |
788 | /* | |
789 | * If it is not a hole and not out of bounds, the runlist is | |
790 | * probably unmapped so try to map it now. | |
791 | */ | |
792 | if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) { | |
793 | if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) { | |
794 | /* Attempt to map runlist. */ | |
795 | if (!rl_write_locked) { | |
796 | /* | |
797 | * We need the runlist locked for | |
798 | * writing, so if it is locked for | |
799 | * reading relock it now and retry in | |
800 | * case it changed whilst we dropped | |
801 | * the lock. | |
802 | */ | |
803 | up_read(&ni->runlist.lock); | |
804 | down_write(&ni->runlist.lock); | |
c49c3111 | 805 | rl_write_locked = true; |
98b27036 AA |
806 | goto retry_remap; |
807 | } | |
808 | err = ntfs_map_runlist_nolock(ni, bh_cpos, | |
809 | NULL); | |
810 | if (likely(!err)) { | |
c49c3111 | 811 | is_retry = true; |
98b27036 AA |
812 | goto retry_remap; |
813 | } | |
814 | /* | |
815 | * If @vcn is out of bounds, pretend @lcn is | |
816 | * LCN_ENOENT. As long as the buffer is out | |
817 | * of bounds this will work fine. | |
818 | */ | |
819 | if (err == -ENOENT) { | |
820 | lcn = LCN_ENOENT; | |
821 | err = 0; | |
822 | goto rl_not_mapped_enoent; | |
823 | } | |
824 | } else | |
825 | err = -EIO; | |
826 | /* Failed to map the buffer, even after retrying. */ | |
827 | bh->b_blocknr = -1; | |
828 | ntfs_error(vol->sb, "Failed to write to inode 0x%lx, " | |
829 | "attribute type 0x%x, vcn 0x%llx, " | |
830 | "vcn offset 0x%x, because its " | |
831 | "location on disk could not be " | |
832 | "determined%s (error code %i).", | |
833 | ni->mft_no, ni->type, | |
834 | (unsigned long long)bh_cpos, | |
835 | (unsigned)bh_pos & | |
836 | vol->cluster_size_mask, | |
837 | is_retry ? " even after retrying" : "", | |
838 | err); | |
839 | break; | |
840 | } | |
841 | rl_not_mapped_enoent: | |
842 | /* | |
843 | * The buffer is in a hole or out of bounds. We need to fill | |
844 | * the hole, unless the buffer is in a cluster which is not | |
845 | * touched by the write, in which case we just leave the buffer | |
846 | * unmapped. This can only happen when the cluster size is | |
847 | * less than the page cache size. | |
848 | */ | |
849 | if (unlikely(vol->cluster_size < PAGE_CACHE_SIZE)) { | |
850 | bh_cend = (bh_end + vol->cluster_size - 1) >> | |
851 | vol->cluster_size_bits; | |
852 | if ((bh_cend <= cpos || bh_cpos >= cend)) { | |
853 | bh->b_blocknr = -1; | |
854 | /* | |
855 | * If the buffer is uptodate we skip it. If it | |
856 | * is not but the page is uptodate, we can set | |
857 | * the buffer uptodate. If the page is not | |
858 | * uptodate, we can clear the buffer and set it | |
859 | * uptodate. Whether this is worthwhile is | |
860 | * debatable and this could be removed. | |
861 | */ | |
862 | if (PageUptodate(page)) { | |
863 | if (!buffer_uptodate(bh)) | |
864 | set_buffer_uptodate(bh); | |
865 | } else if (!buffer_uptodate(bh)) { | |
eebd2aa3 CL |
866 | zero_user(page, bh_offset(bh), |
867 | blocksize); | |
98b27036 AA |
868 | set_buffer_uptodate(bh); |
869 | } | |
870 | continue; | |
871 | } | |
872 | } | |
873 | /* | |
874 | * Out of bounds buffer is invalid if it was not really out of | |
875 | * bounds. | |
876 | */ | |
877 | BUG_ON(lcn != LCN_HOLE); | |
878 | /* | |
879 | * We need the runlist locked for writing, so if it is locked | |
880 | * for reading relock it now and retry in case it changed | |
881 | * whilst we dropped the lock. | |
882 | */ | |
883 | BUG_ON(!rl); | |
884 | if (!rl_write_locked) { | |
885 | up_read(&ni->runlist.lock); | |
886 | down_write(&ni->runlist.lock); | |
c49c3111 | 887 | rl_write_locked = true; |
98b27036 AA |
888 | goto retry_remap; |
889 | } | |
890 | /* Find the previous last allocated cluster. */ | |
891 | BUG_ON(rl->lcn != LCN_HOLE); | |
892 | lcn = -1; | |
893 | rl2 = rl; | |
894 | while (--rl2 >= ni->runlist.rl) { | |
895 | if (rl2->lcn >= 0) { | |
896 | lcn = rl2->lcn + rl2->length; | |
897 | break; | |
898 | } | |
899 | } | |
900 | rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE, | |
c49c3111 | 901 | false); |
98b27036 AA |
902 | if (IS_ERR(rl2)) { |
903 | err = PTR_ERR(rl2); | |
904 | ntfs_debug("Failed to allocate cluster, error code %i.", | |
905 | err); | |
906 | break; | |
907 | } | |
908 | lcn = rl2->lcn; | |
909 | rl = ntfs_runlists_merge(ni->runlist.rl, rl2); | |
910 | if (IS_ERR(rl)) { | |
911 | err = PTR_ERR(rl); | |
912 | if (err != -ENOMEM) | |
913 | err = -EIO; | |
914 | if (ntfs_cluster_free_from_rl(vol, rl2)) { | |
915 | ntfs_error(vol->sb, "Failed to release " | |
916 | "allocated cluster in error " | |
917 | "code path. Run chkdsk to " | |
918 | "recover the lost cluster."); | |
919 | NVolSetErrors(vol); | |
920 | } | |
921 | ntfs_free(rl2); | |
922 | break; | |
923 | } | |
924 | ni->runlist.rl = rl; | |
925 | status.runlist_merged = 1; | |
bb8047d3 AA |
926 | ntfs_debug("Allocated cluster, lcn 0x%llx.", |
927 | (unsigned long long)lcn); | |
98b27036 AA |
928 | /* Map and lock the mft record and get the attribute record. */ |
929 | if (!NInoAttr(ni)) | |
930 | base_ni = ni; | |
931 | else | |
932 | base_ni = ni->ext.base_ntfs_ino; | |
933 | m = map_mft_record(base_ni); | |
934 | if (IS_ERR(m)) { | |
935 | err = PTR_ERR(m); | |
936 | break; | |
937 | } | |
938 | ctx = ntfs_attr_get_search_ctx(base_ni, m); | |
939 | if (unlikely(!ctx)) { | |
940 | err = -ENOMEM; | |
941 | unmap_mft_record(base_ni); | |
942 | break; | |
943 | } | |
944 | status.mft_attr_mapped = 1; | |
945 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
946 | CASE_SENSITIVE, bh_cpos, NULL, 0, ctx); | |
947 | if (unlikely(err)) { | |
948 | if (err == -ENOENT) | |
949 | err = -EIO; | |
950 | break; | |
951 | } | |
952 | m = ctx->mrec; | |
953 | a = ctx->attr; | |
954 | /* | |
955 | * Find the runlist element with which the attribute extent | |
956 | * starts. Note, we cannot use the _attr_ version because we | |
957 | * have mapped the mft record. That is ok because we know the | |
958 | * runlist fragment must be mapped already to have ever gotten | |
959 | * here, so we can just use the _rl_ version. | |
960 | */ | |
961 | vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn); | |
962 | rl2 = ntfs_rl_find_vcn_nolock(rl, vcn); | |
963 | BUG_ON(!rl2); | |
964 | BUG_ON(!rl2->length); | |
965 | BUG_ON(rl2->lcn < LCN_HOLE); | |
966 | highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn); | |
967 | /* | |
968 | * If @highest_vcn is zero, calculate the real highest_vcn | |
969 | * (which can really be zero). | |
970 | */ | |
971 | if (!highest_vcn) | |
972 | highest_vcn = (sle64_to_cpu( | |
973 | a->data.non_resident.allocated_size) >> | |
974 | vol->cluster_size_bits) - 1; | |
975 | /* | |
976 | * Determine the size of the mapping pairs array for the new | |
977 | * extent, i.e. the old extent with the hole filled. | |
978 | */ | |
979 | mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn, | |
980 | highest_vcn); | |
981 | if (unlikely(mp_size <= 0)) { | |
982 | if (!(err = mp_size)) | |
983 | err = -EIO; | |
984 | ntfs_debug("Failed to get size for mapping pairs " | |
985 | "array, error code %i.", err); | |
986 | break; | |
987 | } | |
988 | /* | |
989 | * Resize the attribute record to fit the new mapping pairs | |
990 | * array. | |
991 | */ | |
992 | attr_rec_len = le32_to_cpu(a->length); | |
993 | err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu( | |
994 | a->data.non_resident.mapping_pairs_offset)); | |
995 | if (unlikely(err)) { | |
996 | BUG_ON(err != -ENOSPC); | |
997 | // TODO: Deal with this by using the current attribute | |
998 | // and fill it with as much of the mapping pairs | |
999 | // array as possible. Then loop over each attribute | |
1000 | // extent rewriting the mapping pairs arrays as we go | |
1001 | // along and if when we reach the end we have not | |
1002 | // enough space, try to resize the last attribute | |
1003 | // extent and if even that fails, add a new attribute | |
1004 | // extent. | |
1005 | // We could also try to resize at each step in the hope | |
1006 | // that we will not need to rewrite every single extent. | |
1007 | // Note, we may need to decompress some extents to fill | |
1008 | // the runlist as we are walking the extents... | |
1009 | ntfs_error(vol->sb, "Not enough space in the mft " | |
1010 | "record for the extended attribute " | |
1011 | "record. This case is not " | |
1012 | "implemented yet."); | |
1013 | err = -EOPNOTSUPP; | |
1014 | break ; | |
1015 | } | |
1016 | status.mp_rebuilt = 1; | |
1017 | /* | |
1018 | * Generate the mapping pairs array directly into the attribute | |
1019 | * record. | |
1020 | */ | |
1021 | err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( | |
1022 | a->data.non_resident.mapping_pairs_offset), | |
1023 | mp_size, rl2, vcn, highest_vcn, NULL); | |
1024 | if (unlikely(err)) { | |
1025 | ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, " | |
1026 | "attribute type 0x%x, because building " | |
1027 | "the mapping pairs failed with error " | |
1028 | "code %i.", vi->i_ino, | |
1029 | (unsigned)le32_to_cpu(ni->type), err); | |
1030 | err = -EIO; | |
1031 | break; | |
1032 | } | |
1033 | /* Update the highest_vcn but only if it was not set. */ | |
1034 | if (unlikely(!a->data.non_resident.highest_vcn)) | |
1035 | a->data.non_resident.highest_vcn = | |
1036 | cpu_to_sle64(highest_vcn); | |
1037 | /* | |
1038 | * If the attribute is sparse/compressed, update the compressed | |
1039 | * size in the ntfs_inode structure and the attribute record. | |
1040 | */ | |
1041 | if (likely(NInoSparse(ni) || NInoCompressed(ni))) { | |
1042 | /* | |
1043 | * If we are not in the first attribute extent, switch | |
1044 | * to it, but first ensure the changes will make it to | |
1045 | * disk later. | |
1046 | */ | |
1047 | if (a->data.non_resident.lowest_vcn) { | |
1048 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
1049 | mark_mft_record_dirty(ctx->ntfs_ino); | |
1050 | ntfs_attr_reinit_search_ctx(ctx); | |
1051 | err = ntfs_attr_lookup(ni->type, ni->name, | |
1052 | ni->name_len, CASE_SENSITIVE, | |
1053 | 0, NULL, 0, ctx); | |
1054 | if (unlikely(err)) { | |
1055 | status.attr_switched = 1; | |
1056 | break; | |
1057 | } | |
1058 | /* @m is not used any more so do not set it. */ | |
1059 | a = ctx->attr; | |
1060 | } | |
1061 | write_lock_irqsave(&ni->size_lock, flags); | |
1062 | ni->itype.compressed.size += vol->cluster_size; | |
1063 | a->data.non_resident.compressed_size = | |
1064 | cpu_to_sle64(ni->itype.compressed.size); | |
1065 | write_unlock_irqrestore(&ni->size_lock, flags); | |
1066 | } | |
1067 | /* Ensure the changes make it to disk. */ | |
1068 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
1069 | mark_mft_record_dirty(ctx->ntfs_ino); | |
1070 | ntfs_attr_put_search_ctx(ctx); | |
1071 | unmap_mft_record(base_ni); | |
1072 | /* Successfully filled the hole. */ | |
1073 | status.runlist_merged = 0; | |
1074 | status.mft_attr_mapped = 0; | |
1075 | status.mp_rebuilt = 0; | |
1076 | /* Setup the map cache and use that to deal with the buffer. */ | |
c49c3111 | 1077 | was_hole = true; |
98b27036 AA |
1078 | vcn = bh_cpos; |
1079 | vcn_len = 1; | |
1080 | lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits); | |
1081 | cdelta = 0; | |
1082 | /* | |
1083 | * If the number of remaining clusters in the @pages is smaller | |
1084 | * or equal to the number of cached clusters, unlock the | |
1085 | * runlist as the map cache will be used from now on. | |
1086 | */ | |
1087 | if (likely(vcn + vcn_len >= cend)) { | |
1088 | up_write(&ni->runlist.lock); | |
c49c3111 | 1089 | rl_write_locked = false; |
98b27036 AA |
1090 | rl = NULL; |
1091 | } | |
1092 | goto map_buffer_cached; | |
1093 | } while (bh_pos += blocksize, (bh = bh->b_this_page) != head); | |
1094 | /* If there are no errors, do the next page. */ | |
1095 | if (likely(!err && ++u < nr_pages)) | |
1096 | goto do_next_page; | |
1097 | /* If there are no errors, release the runlist lock if we took it. */ | |
1098 | if (likely(!err)) { | |
1099 | if (unlikely(rl_write_locked)) { | |
1100 | up_write(&ni->runlist.lock); | |
c49c3111 | 1101 | rl_write_locked = false; |
98b27036 AA |
1102 | } else if (unlikely(rl)) |
1103 | up_read(&ni->runlist.lock); | |
1104 | rl = NULL; | |
1105 | } | |
1106 | /* If we issued read requests, let them complete. */ | |
1107 | read_lock_irqsave(&ni->size_lock, flags); | |
1108 | initialized_size = ni->initialized_size; | |
1109 | read_unlock_irqrestore(&ni->size_lock, flags); | |
1110 | while (wait_bh > wait) { | |
1111 | bh = *--wait_bh; | |
1112 | wait_on_buffer(bh); | |
1113 | if (likely(buffer_uptodate(bh))) { | |
1114 | page = bh->b_page; | |
1115 | bh_pos = ((s64)page->index << PAGE_CACHE_SHIFT) + | |
1116 | bh_offset(bh); | |
1117 | /* | |
1118 | * If the buffer overflows the initialized size, need | |
1119 | * to zero the overflowing region. | |
1120 | */ | |
1121 | if (unlikely(bh_pos + blocksize > initialized_size)) { | |
98b27036 AA |
1122 | int ofs = 0; |
1123 | ||
1124 | if (likely(bh_pos < initialized_size)) | |
1125 | ofs = initialized_size - bh_pos; | |
eebd2aa3 CL |
1126 | zero_user_segment(page, bh_offset(bh) + ofs, |
1127 | blocksize); | |
98b27036 AA |
1128 | } |
1129 | } else /* if (unlikely(!buffer_uptodate(bh))) */ | |
1130 | err = -EIO; | |
1131 | } | |
1132 | if (likely(!err)) { | |
1133 | /* Clear buffer_new on all buffers. */ | |
1134 | u = 0; | |
1135 | do { | |
1136 | bh = head = page_buffers(pages[u]); | |
1137 | do { | |
1138 | if (buffer_new(bh)) | |
1139 | clear_buffer_new(bh); | |
1140 | } while ((bh = bh->b_this_page) != head); | |
1141 | } while (++u < nr_pages); | |
1142 | ntfs_debug("Done."); | |
1143 | return err; | |
1144 | } | |
1145 | if (status.attr_switched) { | |
1146 | /* Get back to the attribute extent we modified. */ | |
1147 | ntfs_attr_reinit_search_ctx(ctx); | |
1148 | if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
1149 | CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) { | |
1150 | ntfs_error(vol->sb, "Failed to find required " | |
1151 | "attribute extent of attribute in " | |
1152 | "error code path. Run chkdsk to " | |
1153 | "recover."); | |
1154 | write_lock_irqsave(&ni->size_lock, flags); | |
1155 | ni->itype.compressed.size += vol->cluster_size; | |
1156 | write_unlock_irqrestore(&ni->size_lock, flags); | |
1157 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
1158 | mark_mft_record_dirty(ctx->ntfs_ino); | |
1159 | /* | |
1160 | * The only thing that is now wrong is the compressed | |
1161 | * size of the base attribute extent which chkdsk | |
1162 | * should be able to fix. | |
1163 | */ | |
1164 | NVolSetErrors(vol); | |
1165 | } else { | |
1166 | m = ctx->mrec; | |
1167 | a = ctx->attr; | |
1168 | status.attr_switched = 0; | |
1169 | } | |
1170 | } | |
1171 | /* | |
1172 | * If the runlist has been modified, need to restore it by punching a | |
1173 | * hole into it and we then need to deallocate the on-disk cluster as | |
1174 | * well. Note, we only modify the runlist if we are able to generate a | |
1175 | * new mapping pairs array, i.e. only when the mapped attribute extent | |
1176 | * is not switched. | |
1177 | */ | |
1178 | if (status.runlist_merged && !status.attr_switched) { | |
1179 | BUG_ON(!rl_write_locked); | |
1180 | /* Make the file cluster we allocated sparse in the runlist. */ | |
1181 | if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) { | |
1182 | ntfs_error(vol->sb, "Failed to punch hole into " | |
1183 | "attribute runlist in error code " | |
1184 | "path. Run chkdsk to recover the " | |
1185 | "lost cluster."); | |
98b27036 AA |
1186 | NVolSetErrors(vol); |
1187 | } else /* if (success) */ { | |
1188 | status.runlist_merged = 0; | |
1189 | /* | |
1190 | * Deallocate the on-disk cluster we allocated but only | |
1191 | * if we succeeded in punching its vcn out of the | |
1192 | * runlist. | |
1193 | */ | |
1194 | down_write(&vol->lcnbmp_lock); | |
1195 | if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) { | |
1196 | ntfs_error(vol->sb, "Failed to release " | |
1197 | "allocated cluster in error " | |
1198 | "code path. Run chkdsk to " | |
1199 | "recover the lost cluster."); | |
1200 | NVolSetErrors(vol); | |
1201 | } | |
1202 | up_write(&vol->lcnbmp_lock); | |
1203 | } | |
1204 | } | |
1205 | /* | |
1206 | * Resize the attribute record to its old size and rebuild the mapping | |
1207 | * pairs array. Note, we only can do this if the runlist has been | |
1208 | * restored to its old state which also implies that the mapped | |
1209 | * attribute extent is not switched. | |
1210 | */ | |
1211 | if (status.mp_rebuilt && !status.runlist_merged) { | |
1212 | if (ntfs_attr_record_resize(m, a, attr_rec_len)) { | |
1213 | ntfs_error(vol->sb, "Failed to restore attribute " | |
1214 | "record in error code path. Run " | |
1215 | "chkdsk to recover."); | |
98b27036 AA |
1216 | NVolSetErrors(vol); |
1217 | } else /* if (success) */ { | |
1218 | if (ntfs_mapping_pairs_build(vol, (u8*)a + | |
1219 | le16_to_cpu(a->data.non_resident. | |
1220 | mapping_pairs_offset), attr_rec_len - | |
1221 | le16_to_cpu(a->data.non_resident. | |
1222 | mapping_pairs_offset), ni->runlist.rl, | |
1223 | vcn, highest_vcn, NULL)) { | |
1224 | ntfs_error(vol->sb, "Failed to restore " | |
1225 | "mapping pairs array in error " | |
1226 | "code path. Run chkdsk to " | |
1227 | "recover."); | |
98b27036 AA |
1228 | NVolSetErrors(vol); |
1229 | } | |
1230 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
1231 | mark_mft_record_dirty(ctx->ntfs_ino); | |
1232 | } | |
1233 | } | |
1234 | /* Release the mft record and the attribute. */ | |
1235 | if (status.mft_attr_mapped) { | |
1236 | ntfs_attr_put_search_ctx(ctx); | |
1237 | unmap_mft_record(base_ni); | |
1238 | } | |
1239 | /* Release the runlist lock. */ | |
1240 | if (rl_write_locked) | |
1241 | up_write(&ni->runlist.lock); | |
1242 | else if (rl) | |
1243 | up_read(&ni->runlist.lock); | |
1244 | /* | |
1245 | * Zero out any newly allocated blocks to avoid exposing stale data. | |
1246 | * If BH_New is set, we know that the block was newly allocated above | |
1247 | * and that it has not been fully zeroed and marked dirty yet. | |
1248 | */ | |
1249 | nr_pages = u; | |
1250 | u = 0; | |
1251 | end = bh_cpos << vol->cluster_size_bits; | |
1252 | do { | |
1253 | page = pages[u]; | |
1254 | bh = head = page_buffers(page); | |
1255 | do { | |
1256 | if (u == nr_pages && | |
1257 | ((s64)page->index << PAGE_CACHE_SHIFT) + | |
1258 | bh_offset(bh) >= end) | |
1259 | break; | |
1260 | if (!buffer_new(bh)) | |
1261 | continue; | |
1262 | clear_buffer_new(bh); | |
1263 | if (!buffer_uptodate(bh)) { | |
1264 | if (PageUptodate(page)) | |
1265 | set_buffer_uptodate(bh); | |
1266 | else { | |
eebd2aa3 CL |
1267 | zero_user(page, bh_offset(bh), |
1268 | blocksize); | |
98b27036 AA |
1269 | set_buffer_uptodate(bh); |
1270 | } | |
1271 | } | |
1272 | mark_buffer_dirty(bh); | |
1273 | } while ((bh = bh->b_this_page) != head); | |
1274 | } while (++u <= nr_pages); | |
1275 | ntfs_error(vol->sb, "Failed. Returning error code %i.", err); | |
1276 | return err; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Copy as much as we can into the pages and return the number of bytes which | |
af901ca1 | 1281 | * were successfully copied. If a fault is encountered then clear the pages |
98b27036 AA |
1282 | * out to (ofs + bytes) and return the number of bytes which were copied. |
1283 | */ | |
1284 | static inline size_t ntfs_copy_from_user(struct page **pages, | |
1285 | unsigned nr_pages, unsigned ofs, const char __user *buf, | |
1286 | size_t bytes) | |
1287 | { | |
1288 | struct page **last_page = pages + nr_pages; | |
bfab36e8 | 1289 | char *addr; |
98b27036 AA |
1290 | size_t total = 0; |
1291 | unsigned len; | |
1292 | int left; | |
1293 | ||
1294 | do { | |
1295 | len = PAGE_CACHE_SIZE - ofs; | |
1296 | if (len > bytes) | |
1297 | len = bytes; | |
bfab36e8 AA |
1298 | addr = kmap_atomic(*pages, KM_USER0); |
1299 | left = __copy_from_user_inatomic(addr + ofs, buf, len); | |
1300 | kunmap_atomic(addr, KM_USER0); | |
98b27036 AA |
1301 | if (unlikely(left)) { |
1302 | /* Do it the slow way. */ | |
bfab36e8 AA |
1303 | addr = kmap(*pages); |
1304 | left = __copy_from_user(addr + ofs, buf, len); | |
98b27036 AA |
1305 | kunmap(*pages); |
1306 | if (unlikely(left)) | |
1307 | goto err_out; | |
1308 | } | |
1309 | total += len; | |
1310 | bytes -= len; | |
1311 | if (!bytes) | |
1312 | break; | |
1313 | buf += len; | |
1314 | ofs = 0; | |
1315 | } while (++pages < last_page); | |
1316 | out: | |
1317 | return total; | |
1318 | err_out: | |
1319 | total += len - left; | |
1320 | /* Zero the rest of the target like __copy_from_user(). */ | |
1321 | while (++pages < last_page) { | |
1322 | bytes -= len; | |
1323 | if (!bytes) | |
1324 | break; | |
1325 | len = PAGE_CACHE_SIZE; | |
1326 | if (len > bytes) | |
1327 | len = bytes; | |
eebd2aa3 | 1328 | zero_user(*pages, 0, len); |
98b27036 AA |
1329 | } |
1330 | goto out; | |
1331 | } | |
1332 | ||
01408c49 | 1333 | static size_t __ntfs_copy_from_user_iovec_inatomic(char *vaddr, |
98b27036 AA |
1334 | const struct iovec *iov, size_t iov_ofs, size_t bytes) |
1335 | { | |
1336 | size_t total = 0; | |
1337 | ||
1338 | while (1) { | |
1339 | const char __user *buf = iov->iov_base + iov_ofs; | |
1340 | unsigned len; | |
1341 | size_t left; | |
1342 | ||
1343 | len = iov->iov_len - iov_ofs; | |
1344 | if (len > bytes) | |
1345 | len = bytes; | |
1346 | left = __copy_from_user_inatomic(vaddr, buf, len); | |
1347 | total += len; | |
1348 | bytes -= len; | |
1349 | vaddr += len; | |
1350 | if (unlikely(left)) { | |
98b27036 AA |
1351 | total -= left; |
1352 | break; | |
1353 | } | |
1354 | if (!bytes) | |
1355 | break; | |
1356 | iov++; | |
1357 | iov_ofs = 0; | |
1358 | } | |
1359 | return total; | |
1360 | } | |
1361 | ||
1362 | static inline void ntfs_set_next_iovec(const struct iovec **iovp, | |
1363 | size_t *iov_ofsp, size_t bytes) | |
1364 | { | |
1365 | const struct iovec *iov = *iovp; | |
1366 | size_t iov_ofs = *iov_ofsp; | |
1367 | ||
1368 | while (bytes) { | |
1369 | unsigned len; | |
1370 | ||
1371 | len = iov->iov_len - iov_ofs; | |
1372 | if (len > bytes) | |
1373 | len = bytes; | |
1374 | bytes -= len; | |
1375 | iov_ofs += len; | |
1376 | if (iov->iov_len == iov_ofs) { | |
1377 | iov++; | |
1378 | iov_ofs = 0; | |
1379 | } | |
1380 | } | |
1381 | *iovp = iov; | |
1382 | *iov_ofsp = iov_ofs; | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * This has the same side-effects and return value as ntfs_copy_from_user(). | |
1387 | * The difference is that on a fault we need to memset the remainder of the | |
1388 | * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s | |
1389 | * single-segment behaviour. | |
1390 | * | |
01408c49 N |
1391 | * We call the same helper (__ntfs_copy_from_user_iovec_inatomic()) both |
1392 | * when atomic and when not atomic. This is ok because | |
1393 | * __ntfs_copy_from_user_iovec_inatomic() calls __copy_from_user_inatomic() | |
1394 | * and it is ok to call this when non-atomic. | |
1395 | * Infact, the only difference between __copy_from_user_inatomic() and | |
1396 | * __copy_from_user() is that the latter calls might_sleep() and the former | |
1397 | * should not zero the tail of the buffer on error. And on many | |
98b27036 AA |
1398 | * architectures __copy_from_user_inatomic() is just defined to |
1399 | * __copy_from_user() so it makes no difference at all on those architectures. | |
1400 | */ | |
1401 | static inline size_t ntfs_copy_from_user_iovec(struct page **pages, | |
1402 | unsigned nr_pages, unsigned ofs, const struct iovec **iov, | |
1403 | size_t *iov_ofs, size_t bytes) | |
1404 | { | |
1405 | struct page **last_page = pages + nr_pages; | |
bfab36e8 | 1406 | char *addr; |
98b27036 AA |
1407 | size_t copied, len, total = 0; |
1408 | ||
1409 | do { | |
1410 | len = PAGE_CACHE_SIZE - ofs; | |
1411 | if (len > bytes) | |
1412 | len = bytes; | |
bfab36e8 AA |
1413 | addr = kmap_atomic(*pages, KM_USER0); |
1414 | copied = __ntfs_copy_from_user_iovec_inatomic(addr + ofs, | |
98b27036 | 1415 | *iov, *iov_ofs, len); |
bfab36e8 | 1416 | kunmap_atomic(addr, KM_USER0); |
98b27036 AA |
1417 | if (unlikely(copied != len)) { |
1418 | /* Do it the slow way. */ | |
bfab36e8 AA |
1419 | addr = kmap(*pages); |
1420 | copied = __ntfs_copy_from_user_iovec_inatomic(addr + ofs, | |
98b27036 | 1421 | *iov, *iov_ofs, len); |
01408c49 N |
1422 | /* |
1423 | * Zero the rest of the target like __copy_from_user(). | |
1424 | */ | |
bfab36e8 | 1425 | memset(addr + ofs + copied, 0, len - copied); |
98b27036 AA |
1426 | kunmap(*pages); |
1427 | if (unlikely(copied != len)) | |
1428 | goto err_out; | |
1429 | } | |
1430 | total += len; | |
1431 | bytes -= len; | |
1432 | if (!bytes) | |
1433 | break; | |
1434 | ntfs_set_next_iovec(iov, iov_ofs, len); | |
1435 | ofs = 0; | |
1436 | } while (++pages < last_page); | |
1437 | out: | |
1438 | return total; | |
1439 | err_out: | |
1440 | total += copied; | |
1441 | /* Zero the rest of the target like __copy_from_user(). */ | |
1442 | while (++pages < last_page) { | |
1443 | bytes -= len; | |
1444 | if (!bytes) | |
1445 | break; | |
1446 | len = PAGE_CACHE_SIZE; | |
1447 | if (len > bytes) | |
1448 | len = bytes; | |
eebd2aa3 | 1449 | zero_user(*pages, 0, len); |
98b27036 AA |
1450 | } |
1451 | goto out; | |
1452 | } | |
1453 | ||
1454 | static inline void ntfs_flush_dcache_pages(struct page **pages, | |
1455 | unsigned nr_pages) | |
1456 | { | |
1457 | BUG_ON(!nr_pages); | |
f893afbe AA |
1458 | /* |
1459 | * Warning: Do not do the decrement at the same time as the call to | |
1460 | * flush_dcache_page() because it is a NULL macro on i386 and hence the | |
1461 | * decrement never happens so the loop never terminates. | |
1462 | */ | |
98b27036 | 1463 | do { |
f893afbe | 1464 | --nr_pages; |
98b27036 | 1465 | flush_dcache_page(pages[nr_pages]); |
f893afbe | 1466 | } while (nr_pages > 0); |
98b27036 AA |
1467 | } |
1468 | ||
1469 | /** | |
1470 | * ntfs_commit_pages_after_non_resident_write - commit the received data | |
1471 | * @pages: array of destination pages | |
1472 | * @nr_pages: number of pages in @pages | |
1473 | * @pos: byte position in file at which the write begins | |
1474 | * @bytes: number of bytes to be written | |
1475 | * | |
1476 | * See description of ntfs_commit_pages_after_write(), below. | |
1477 | */ | |
1478 | static inline int ntfs_commit_pages_after_non_resident_write( | |
1479 | struct page **pages, const unsigned nr_pages, | |
1480 | s64 pos, size_t bytes) | |
1481 | { | |
1482 | s64 end, initialized_size; | |
1483 | struct inode *vi; | |
1484 | ntfs_inode *ni, *base_ni; | |
1485 | struct buffer_head *bh, *head; | |
1486 | ntfs_attr_search_ctx *ctx; | |
1487 | MFT_RECORD *m; | |
1488 | ATTR_RECORD *a; | |
1489 | unsigned long flags; | |
1490 | unsigned blocksize, u; | |
1491 | int err; | |
1492 | ||
1493 | vi = pages[0]->mapping->host; | |
1494 | ni = NTFS_I(vi); | |
78af34f0 | 1495 | blocksize = vi->i_sb->s_blocksize; |
98b27036 AA |
1496 | end = pos + bytes; |
1497 | u = 0; | |
1498 | do { | |
1499 | s64 bh_pos; | |
1500 | struct page *page; | |
c49c3111 | 1501 | bool partial; |
98b27036 AA |
1502 | |
1503 | page = pages[u]; | |
1504 | bh_pos = (s64)page->index << PAGE_CACHE_SHIFT; | |
1505 | bh = head = page_buffers(page); | |
c49c3111 | 1506 | partial = false; |
98b27036 AA |
1507 | do { |
1508 | s64 bh_end; | |
1509 | ||
1510 | bh_end = bh_pos + blocksize; | |
1511 | if (bh_end <= pos || bh_pos >= end) { | |
1512 | if (!buffer_uptodate(bh)) | |
c49c3111 | 1513 | partial = true; |
98b27036 AA |
1514 | } else { |
1515 | set_buffer_uptodate(bh); | |
1516 | mark_buffer_dirty(bh); | |
1517 | } | |
1518 | } while (bh_pos += blocksize, (bh = bh->b_this_page) != head); | |
1519 | /* | |
1520 | * If all buffers are now uptodate but the page is not, set the | |
1521 | * page uptodate. | |
1522 | */ | |
1523 | if (!partial && !PageUptodate(page)) | |
1524 | SetPageUptodate(page); | |
1525 | } while (++u < nr_pages); | |
1526 | /* | |
1527 | * Finally, if we do not need to update initialized_size or i_size we | |
1528 | * are finished. | |
1529 | */ | |
1530 | read_lock_irqsave(&ni->size_lock, flags); | |
1531 | initialized_size = ni->initialized_size; | |
1532 | read_unlock_irqrestore(&ni->size_lock, flags); | |
1533 | if (end <= initialized_size) { | |
1534 | ntfs_debug("Done."); | |
1535 | return 0; | |
1536 | } | |
1537 | /* | |
1538 | * Update initialized_size/i_size as appropriate, both in the inode and | |
1539 | * the mft record. | |
1540 | */ | |
1541 | if (!NInoAttr(ni)) | |
1542 | base_ni = ni; | |
1543 | else | |
1544 | base_ni = ni->ext.base_ntfs_ino; | |
1545 | /* Map, pin, and lock the mft record. */ | |
1546 | m = map_mft_record(base_ni); | |
1547 | if (IS_ERR(m)) { | |
1548 | err = PTR_ERR(m); | |
1549 | m = NULL; | |
1550 | ctx = NULL; | |
1551 | goto err_out; | |
1552 | } | |
1553 | BUG_ON(!NInoNonResident(ni)); | |
1554 | ctx = ntfs_attr_get_search_ctx(base_ni, m); | |
1555 | if (unlikely(!ctx)) { | |
1556 | err = -ENOMEM; | |
1557 | goto err_out; | |
1558 | } | |
1559 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
1560 | CASE_SENSITIVE, 0, NULL, 0, ctx); | |
1561 | if (unlikely(err)) { | |
1562 | if (err == -ENOENT) | |
1563 | err = -EIO; | |
1564 | goto err_out; | |
1565 | } | |
1566 | a = ctx->attr; | |
1567 | BUG_ON(!a->non_resident); | |
1568 | write_lock_irqsave(&ni->size_lock, flags); | |
1569 | BUG_ON(end > ni->allocated_size); | |
1570 | ni->initialized_size = end; | |
1571 | a->data.non_resident.initialized_size = cpu_to_sle64(end); | |
1572 | if (end > i_size_read(vi)) { | |
1573 | i_size_write(vi, end); | |
1574 | a->data.non_resident.data_size = | |
1575 | a->data.non_resident.initialized_size; | |
1576 | } | |
1577 | write_unlock_irqrestore(&ni->size_lock, flags); | |
1578 | /* Mark the mft record dirty, so it gets written back. */ | |
1579 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
1580 | mark_mft_record_dirty(ctx->ntfs_ino); | |
1581 | ntfs_attr_put_search_ctx(ctx); | |
1582 | unmap_mft_record(base_ni); | |
1583 | ntfs_debug("Done."); | |
1584 | return 0; | |
1585 | err_out: | |
1586 | if (ctx) | |
1587 | ntfs_attr_put_search_ctx(ctx); | |
1588 | if (m) | |
1589 | unmap_mft_record(base_ni); | |
1590 | ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error " | |
1591 | "code %i).", err); | |
f95c4018 | 1592 | if (err != -ENOMEM) |
98b27036 | 1593 | NVolSetErrors(ni->vol); |
98b27036 AA |
1594 | return err; |
1595 | } | |
1596 | ||
1597 | /** | |
1598 | * ntfs_commit_pages_after_write - commit the received data | |
1599 | * @pages: array of destination pages | |
1600 | * @nr_pages: number of pages in @pages | |
1601 | * @pos: byte position in file at which the write begins | |
1602 | * @bytes: number of bytes to be written | |
1603 | * | |
1b1dcc1b | 1604 | * This is called from ntfs_file_buffered_write() with i_mutex held on the inode |
98b27036 AA |
1605 | * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are |
1606 | * locked but not kmap()ped. The source data has already been copied into the | |
1607 | * @page. ntfs_prepare_pages_for_non_resident_write() has been called before | |
1608 | * the data was copied (for non-resident attributes only) and it returned | |
1609 | * success. | |
1610 | * | |
1611 | * Need to set uptodate and mark dirty all buffers within the boundary of the | |
1612 | * write. If all buffers in a page are uptodate we set the page uptodate, too. | |
1613 | * | |
1614 | * Setting the buffers dirty ensures that they get written out later when | |
1615 | * ntfs_writepage() is invoked by the VM. | |
1616 | * | |
1617 | * Finally, we need to update i_size and initialized_size as appropriate both | |
1618 | * in the inode and the mft record. | |
1619 | * | |
1620 | * This is modelled after fs/buffer.c::generic_commit_write(), which marks | |
1621 | * buffers uptodate and dirty, sets the page uptodate if all buffers in the | |
1622 | * page are uptodate, and updates i_size if the end of io is beyond i_size. In | |
1623 | * that case, it also marks the inode dirty. | |
1624 | * | |
1625 | * If things have gone as outlined in | |
1626 | * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page | |
1627 | * content modifications here for non-resident attributes. For resident | |
1628 | * attributes we need to do the uptodate bringing here which we combine with | |
1629 | * the copying into the mft record which means we save one atomic kmap. | |
1630 | * | |
1631 | * Return 0 on success or -errno on error. | |
1632 | */ | |
1633 | static int ntfs_commit_pages_after_write(struct page **pages, | |
1634 | const unsigned nr_pages, s64 pos, size_t bytes) | |
1635 | { | |
1636 | s64 end, initialized_size; | |
1637 | loff_t i_size; | |
1638 | struct inode *vi; | |
1639 | ntfs_inode *ni, *base_ni; | |
1640 | struct page *page; | |
1641 | ntfs_attr_search_ctx *ctx; | |
1642 | MFT_RECORD *m; | |
1643 | ATTR_RECORD *a; | |
1644 | char *kattr, *kaddr; | |
1645 | unsigned long flags; | |
1646 | u32 attr_len; | |
1647 | int err; | |
1648 | ||
1649 | BUG_ON(!nr_pages); | |
1650 | BUG_ON(!pages); | |
1651 | page = pages[0]; | |
1652 | BUG_ON(!page); | |
1653 | vi = page->mapping->host; | |
1654 | ni = NTFS_I(vi); | |
1655 | ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page " | |
d04bd1fb | 1656 | "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.", |
98b27036 AA |
1657 | vi->i_ino, ni->type, page->index, nr_pages, |
1658 | (long long)pos, bytes); | |
1659 | if (NInoNonResident(ni)) | |
1660 | return ntfs_commit_pages_after_non_resident_write(pages, | |
1661 | nr_pages, pos, bytes); | |
1662 | BUG_ON(nr_pages > 1); | |
1663 | /* | |
1664 | * Attribute is resident, implying it is not compressed, encrypted, or | |
1665 | * sparse. | |
1666 | */ | |
1667 | if (!NInoAttr(ni)) | |
1668 | base_ni = ni; | |
1669 | else | |
1670 | base_ni = ni->ext.base_ntfs_ino; | |
1671 | BUG_ON(NInoNonResident(ni)); | |
1672 | /* Map, pin, and lock the mft record. */ | |
1673 | m = map_mft_record(base_ni); | |
1674 | if (IS_ERR(m)) { | |
1675 | err = PTR_ERR(m); | |
1676 | m = NULL; | |
1677 | ctx = NULL; | |
1678 | goto err_out; | |
1679 | } | |
1680 | ctx = ntfs_attr_get_search_ctx(base_ni, m); | |
1681 | if (unlikely(!ctx)) { | |
1682 | err = -ENOMEM; | |
1683 | goto err_out; | |
1684 | } | |
1685 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, | |
1686 | CASE_SENSITIVE, 0, NULL, 0, ctx); | |
1687 | if (unlikely(err)) { | |
1688 | if (err == -ENOENT) | |
1689 | err = -EIO; | |
1690 | goto err_out; | |
1691 | } | |
1692 | a = ctx->attr; | |
1693 | BUG_ON(a->non_resident); | |
1694 | /* The total length of the attribute value. */ | |
1695 | attr_len = le32_to_cpu(a->data.resident.value_length); | |
1696 | i_size = i_size_read(vi); | |
1697 | BUG_ON(attr_len != i_size); | |
1698 | BUG_ON(pos > attr_len); | |
1699 | end = pos + bytes; | |
1700 | BUG_ON(end > le32_to_cpu(a->length) - | |
1701 | le16_to_cpu(a->data.resident.value_offset)); | |
1702 | kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset); | |
1703 | kaddr = kmap_atomic(page, KM_USER0); | |
1704 | /* Copy the received data from the page to the mft record. */ | |
1705 | memcpy(kattr + pos, kaddr + pos, bytes); | |
1706 | /* Update the attribute length if necessary. */ | |
1707 | if (end > attr_len) { | |
1708 | attr_len = end; | |
1709 | a->data.resident.value_length = cpu_to_le32(attr_len); | |
1710 | } | |
1711 | /* | |
1712 | * If the page is not uptodate, bring the out of bounds area(s) | |
1713 | * uptodate by copying data from the mft record to the page. | |
1714 | */ | |
1715 | if (!PageUptodate(page)) { | |
1716 | if (pos > 0) | |
1717 | memcpy(kaddr, kattr, pos); | |
1718 | if (end < attr_len) | |
1719 | memcpy(kaddr + end, kattr + end, attr_len - end); | |
1720 | /* Zero the region outside the end of the attribute value. */ | |
1721 | memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len); | |
1722 | flush_dcache_page(page); | |
1723 | SetPageUptodate(page); | |
1724 | } | |
1725 | kunmap_atomic(kaddr, KM_USER0); | |
1726 | /* Update initialized_size/i_size if necessary. */ | |
1727 | read_lock_irqsave(&ni->size_lock, flags); | |
1728 | initialized_size = ni->initialized_size; | |
1729 | BUG_ON(end > ni->allocated_size); | |
1730 | read_unlock_irqrestore(&ni->size_lock, flags); | |
1731 | BUG_ON(initialized_size != i_size); | |
1732 | if (end > initialized_size) { | |
98b27036 AA |
1733 | write_lock_irqsave(&ni->size_lock, flags); |
1734 | ni->initialized_size = end; | |
1735 | i_size_write(vi, end); | |
1736 | write_unlock_irqrestore(&ni->size_lock, flags); | |
1737 | } | |
1738 | /* Mark the mft record dirty, so it gets written back. */ | |
1739 | flush_dcache_mft_record_page(ctx->ntfs_ino); | |
1740 | mark_mft_record_dirty(ctx->ntfs_ino); | |
1741 | ntfs_attr_put_search_ctx(ctx); | |
1742 | unmap_mft_record(base_ni); | |
1743 | ntfs_debug("Done."); | |
1744 | return 0; | |
1745 | err_out: | |
1746 | if (err == -ENOMEM) { | |
1747 | ntfs_warning(vi->i_sb, "Error allocating memory required to " | |
1748 | "commit the write."); | |
1749 | if (PageUptodate(page)) { | |
1750 | ntfs_warning(vi->i_sb, "Page is uptodate, setting " | |
1751 | "dirty so the write will be retried " | |
1752 | "later on by the VM."); | |
1753 | /* | |
1754 | * Put the page on mapping->dirty_pages, but leave its | |
1755 | * buffers' dirty state as-is. | |
1756 | */ | |
1757 | __set_page_dirty_nobuffers(page); | |
1758 | err = 0; | |
1759 | } else | |
1760 | ntfs_error(vi->i_sb, "Page is not uptodate. Written " | |
1761 | "data has been lost."); | |
1762 | } else { | |
1763 | ntfs_error(vi->i_sb, "Resident attribute commit write failed " | |
1764 | "with error %i.", err); | |
1765 | NVolSetErrors(ni->vol); | |
98b27036 AA |
1766 | } |
1767 | if (ctx) | |
1768 | ntfs_attr_put_search_ctx(ctx); | |
1769 | if (m) | |
1770 | unmap_mft_record(base_ni); | |
1771 | return err; | |
1772 | } | |
1773 | ||
1774 | /** | |
1775 | * ntfs_file_buffered_write - | |
1776 | * | |
1b1dcc1b | 1777 | * Locking: The vfs is holding ->i_mutex on the inode. |
98b27036 AA |
1778 | */ |
1779 | static ssize_t ntfs_file_buffered_write(struct kiocb *iocb, | |
1780 | const struct iovec *iov, unsigned long nr_segs, | |
1781 | loff_t pos, loff_t *ppos, size_t count) | |
1782 | { | |
1783 | struct file *file = iocb->ki_filp; | |
1784 | struct address_space *mapping = file->f_mapping; | |
1785 | struct inode *vi = mapping->host; | |
1786 | ntfs_inode *ni = NTFS_I(vi); | |
1787 | ntfs_volume *vol = ni->vol; | |
1788 | struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER]; | |
1789 | struct page *cached_page = NULL; | |
1790 | char __user *buf = NULL; | |
1791 | s64 end, ll; | |
1792 | VCN last_vcn; | |
1793 | LCN lcn; | |
1794 | unsigned long flags; | |
dda65b94 | 1795 | size_t bytes, iov_ofs = 0; /* Offset in the current iovec. */ |
98b27036 AA |
1796 | ssize_t status, written; |
1797 | unsigned nr_pages; | |
1798 | int err; | |
1799 | struct pagevec lru_pvec; | |
1800 | ||
1801 | ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, " | |
1802 | "pos 0x%llx, count 0x%lx.", | |
1803 | vi->i_ino, (unsigned)le32_to_cpu(ni->type), | |
1804 | (unsigned long long)pos, (unsigned long)count); | |
1805 | if (unlikely(!count)) | |
1806 | return 0; | |
1807 | BUG_ON(NInoMstProtected(ni)); | |
1808 | /* | |
1809 | * If the attribute is not an index root and it is encrypted or | |
1810 | * compressed, we cannot write to it yet. Note we need to check for | |
1811 | * AT_INDEX_ALLOCATION since this is the type of both directory and | |
1812 | * index inodes. | |
1813 | */ | |
1814 | if (ni->type != AT_INDEX_ALLOCATION) { | |
1815 | /* If file is encrypted, deny access, just like NT4. */ | |
1816 | if (NInoEncrypted(ni)) { | |
7d0ffdb2 AA |
1817 | /* |
1818 | * Reminder for later: Encrypted files are _always_ | |
1819 | * non-resident so that the content can always be | |
1820 | * encrypted. | |
1821 | */ | |
98b27036 AA |
1822 | ntfs_debug("Denying write access to encrypted file."); |
1823 | return -EACCES; | |
1824 | } | |
1825 | if (NInoCompressed(ni)) { | |
7d0ffdb2 AA |
1826 | /* Only unnamed $DATA attribute can be compressed. */ |
1827 | BUG_ON(ni->type != AT_DATA); | |
1828 | BUG_ON(ni->name_len); | |
1829 | /* | |
1830 | * Reminder for later: If resident, the data is not | |
1831 | * actually compressed. Only on the switch to non- | |
1832 | * resident does compression kick in. This is in | |
1833 | * contrast to encrypted files (see above). | |
1834 | */ | |
98b27036 AA |
1835 | ntfs_error(vi->i_sb, "Writing to compressed files is " |
1836 | "not implemented yet. Sorry."); | |
1837 | return -EOPNOTSUPP; | |
1838 | } | |
1839 | } | |
1840 | /* | |
1841 | * If a previous ntfs_truncate() failed, repeat it and abort if it | |
1842 | * fails again. | |
1843 | */ | |
1844 | if (unlikely(NInoTruncateFailed(ni))) { | |
1845 | down_write(&vi->i_alloc_sem); | |
1846 | err = ntfs_truncate(vi); | |
1847 | up_write(&vi->i_alloc_sem); | |
1848 | if (err || NInoTruncateFailed(ni)) { | |
1849 | if (!err) | |
1850 | err = -EIO; | |
1851 | ntfs_error(vol->sb, "Cannot perform write to inode " | |
1852 | "0x%lx, attribute type 0x%x, because " | |
1853 | "ntfs_truncate() failed (error code " | |
1854 | "%i).", vi->i_ino, | |
1855 | (unsigned)le32_to_cpu(ni->type), err); | |
1856 | return err; | |
1857 | } | |
1858 | } | |
1859 | /* The first byte after the write. */ | |
1860 | end = pos + count; | |
1861 | /* | |
1862 | * If the write goes beyond the allocated size, extend the allocation | |
1863 | * to cover the whole of the write, rounded up to the nearest cluster. | |
1864 | */ | |
1865 | read_lock_irqsave(&ni->size_lock, flags); | |
1866 | ll = ni->allocated_size; | |
1867 | read_unlock_irqrestore(&ni->size_lock, flags); | |
1868 | if (end > ll) { | |
1869 | /* Extend the allocation without changing the data size. */ | |
1870 | ll = ntfs_attr_extend_allocation(ni, end, -1, pos); | |
1871 | if (likely(ll >= 0)) { | |
1872 | BUG_ON(pos >= ll); | |
1873 | /* If the extension was partial truncate the write. */ | |
1874 | if (end > ll) { | |
1875 | ntfs_debug("Truncating write to inode 0x%lx, " | |
1876 | "attribute type 0x%x, because " | |
1877 | "the allocation was only " | |
1878 | "partially extended.", | |
1879 | vi->i_ino, (unsigned) | |
1880 | le32_to_cpu(ni->type)); | |
1881 | end = ll; | |
1882 | count = ll - pos; | |
1883 | } | |
1884 | } else { | |
1885 | err = ll; | |
1886 | read_lock_irqsave(&ni->size_lock, flags); | |
1887 | ll = ni->allocated_size; | |
1888 | read_unlock_irqrestore(&ni->size_lock, flags); | |
1889 | /* Perform a partial write if possible or fail. */ | |
1890 | if (pos < ll) { | |
1891 | ntfs_debug("Truncating write to inode 0x%lx, " | |
1892 | "attribute type 0x%x, because " | |
1893 | "extending the allocation " | |
1894 | "failed (error code %i).", | |
1895 | vi->i_ino, (unsigned) | |
1896 | le32_to_cpu(ni->type), err); | |
1897 | end = ll; | |
1898 | count = ll - pos; | |
1899 | } else { | |
1900 | ntfs_error(vol->sb, "Cannot perform write to " | |
1901 | "inode 0x%lx, attribute type " | |
1902 | "0x%x, because extending the " | |
1903 | "allocation failed (error " | |
1904 | "code %i).", vi->i_ino, | |
1905 | (unsigned) | |
1906 | le32_to_cpu(ni->type), err); | |
1907 | return err; | |
1908 | } | |
1909 | } | |
1910 | } | |
1911 | pagevec_init(&lru_pvec, 0); | |
1912 | written = 0; | |
1913 | /* | |
1914 | * If the write starts beyond the initialized size, extend it up to the | |
1915 | * beginning of the write and initialize all non-sparse space between | |
1916 | * the old initialized size and the new one. This automatically also | |
1917 | * increments the vfs inode->i_size to keep it above or equal to the | |
1918 | * initialized_size. | |
1919 | */ | |
1920 | read_lock_irqsave(&ni->size_lock, flags); | |
1921 | ll = ni->initialized_size; | |
1922 | read_unlock_irqrestore(&ni->size_lock, flags); | |
1923 | if (pos > ll) { | |
2ec93b0b | 1924 | err = ntfs_attr_extend_initialized(ni, pos); |
98b27036 AA |
1925 | if (err < 0) { |
1926 | ntfs_error(vol->sb, "Cannot perform write to inode " | |
1927 | "0x%lx, attribute type 0x%x, because " | |
1928 | "extending the initialized size " | |
1929 | "failed (error code %i).", vi->i_ino, | |
1930 | (unsigned)le32_to_cpu(ni->type), err); | |
1931 | status = err; | |
1932 | goto err_out; | |
1933 | } | |
1934 | } | |
1935 | /* | |
1936 | * Determine the number of pages per cluster for non-resident | |
1937 | * attributes. | |
1938 | */ | |
1939 | nr_pages = 1; | |
1940 | if (vol->cluster_size > PAGE_CACHE_SIZE && NInoNonResident(ni)) | |
1941 | nr_pages = vol->cluster_size >> PAGE_CACHE_SHIFT; | |
1942 | /* Finally, perform the actual write. */ | |
1943 | last_vcn = -1; | |
1944 | if (likely(nr_segs == 1)) | |
1945 | buf = iov->iov_base; | |
98b27036 AA |
1946 | do { |
1947 | VCN vcn; | |
1948 | pgoff_t idx, start_idx; | |
1949 | unsigned ofs, do_pages, u; | |
1950 | size_t copied; | |
1951 | ||
1952 | start_idx = idx = pos >> PAGE_CACHE_SHIFT; | |
1953 | ofs = pos & ~PAGE_CACHE_MASK; | |
1954 | bytes = PAGE_CACHE_SIZE - ofs; | |
1955 | do_pages = 1; | |
1956 | if (nr_pages > 1) { | |
1957 | vcn = pos >> vol->cluster_size_bits; | |
1958 | if (vcn != last_vcn) { | |
1959 | last_vcn = vcn; | |
1960 | /* | |
1961 | * Get the lcn of the vcn the write is in. If | |
1962 | * it is a hole, need to lock down all pages in | |
1963 | * the cluster. | |
1964 | */ | |
1965 | down_read(&ni->runlist.lock); | |
1966 | lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >> | |
c49c3111 | 1967 | vol->cluster_size_bits, false); |
98b27036 AA |
1968 | up_read(&ni->runlist.lock); |
1969 | if (unlikely(lcn < LCN_HOLE)) { | |
1970 | status = -EIO; | |
1971 | if (lcn == LCN_ENOMEM) | |
1972 | status = -ENOMEM; | |
1973 | else | |
1974 | ntfs_error(vol->sb, "Cannot " | |
1975 | "perform write to " | |
1976 | "inode 0x%lx, " | |
1977 | "attribute type 0x%x, " | |
1978 | "because the attribute " | |
1979 | "is corrupt.", | |
1980 | vi->i_ino, (unsigned) | |
1981 | le32_to_cpu(ni->type)); | |
1982 | break; | |
1983 | } | |
1984 | if (lcn == LCN_HOLE) { | |
1985 | start_idx = (pos & ~(s64) | |
1986 | vol->cluster_size_mask) | |
1987 | >> PAGE_CACHE_SHIFT; | |
1988 | bytes = vol->cluster_size - (pos & | |
1989 | vol->cluster_size_mask); | |
1990 | do_pages = nr_pages; | |
1991 | } | |
1992 | } | |
1993 | } | |
1994 | if (bytes > count) | |
1995 | bytes = count; | |
1996 | /* | |
1997 | * Bring in the user page(s) that we will copy from _first_. | |
1998 | * Otherwise there is a nasty deadlock on copying from the same | |
1999 | * page(s) as we are writing to, without it/them being marked | |
2000 | * up-to-date. Note, at present there is nothing to stop the | |
2001 | * pages being swapped out between us bringing them into memory | |
2002 | * and doing the actual copying. | |
2003 | */ | |
2004 | if (likely(nr_segs == 1)) | |
2005 | ntfs_fault_in_pages_readable(buf, bytes); | |
2006 | else | |
2007 | ntfs_fault_in_pages_readable_iovec(iov, iov_ofs, bytes); | |
2008 | /* Get and lock @do_pages starting at index @start_idx. */ | |
2009 | status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages, | |
2010 | pages, &cached_page, &lru_pvec); | |
2011 | if (unlikely(status)) | |
2012 | break; | |
2013 | /* | |
2014 | * For non-resident attributes, we need to fill any holes with | |
2015 | * actual clusters and ensure all bufferes are mapped. We also | |
2016 | * need to bring uptodate any buffers that are only partially | |
2017 | * being written to. | |
2018 | */ | |
2019 | if (NInoNonResident(ni)) { | |
2020 | status = ntfs_prepare_pages_for_non_resident_write( | |
2021 | pages, do_pages, pos, bytes); | |
2022 | if (unlikely(status)) { | |
2023 | loff_t i_size; | |
2024 | ||
2025 | do { | |
2026 | unlock_page(pages[--do_pages]); | |
2027 | page_cache_release(pages[do_pages]); | |
2028 | } while (do_pages); | |
2029 | /* | |
2030 | * The write preparation may have instantiated | |
2031 | * allocated space outside i_size. Trim this | |
2032 | * off again. We can ignore any errors in this | |
2033 | * case as we will just be waisting a bit of | |
2034 | * allocated space, which is not a disaster. | |
2035 | */ | |
2036 | i_size = i_size_read(vi); | |
2037 | if (pos + bytes > i_size) | |
2038 | vmtruncate(vi, i_size); | |
2039 | break; | |
2040 | } | |
2041 | } | |
2042 | u = (pos >> PAGE_CACHE_SHIFT) - pages[0]->index; | |
2043 | if (likely(nr_segs == 1)) { | |
2044 | copied = ntfs_copy_from_user(pages + u, do_pages - u, | |
2045 | ofs, buf, bytes); | |
2046 | buf += copied; | |
2047 | } else | |
2048 | copied = ntfs_copy_from_user_iovec(pages + u, | |
2049 | do_pages - u, ofs, &iov, &iov_ofs, | |
2050 | bytes); | |
2051 | ntfs_flush_dcache_pages(pages + u, do_pages - u); | |
2052 | status = ntfs_commit_pages_after_write(pages, do_pages, pos, | |
2053 | bytes); | |
2054 | if (likely(!status)) { | |
2055 | written += copied; | |
2056 | count -= copied; | |
2057 | pos += copied; | |
2058 | if (unlikely(copied != bytes)) | |
2059 | status = -EFAULT; | |
2060 | } | |
2061 | do { | |
2062 | unlock_page(pages[--do_pages]); | |
2063 | mark_page_accessed(pages[do_pages]); | |
2064 | page_cache_release(pages[do_pages]); | |
2065 | } while (do_pages); | |
2066 | if (unlikely(status)) | |
2067 | break; | |
2068 | balance_dirty_pages_ratelimited(mapping); | |
2069 | cond_resched(); | |
2070 | } while (count); | |
2071 | err_out: | |
2072 | *ppos = pos; | |
2073 | if (cached_page) | |
2074 | page_cache_release(cached_page); | |
4f98a2fe | 2075 | pagevec_lru_add_file(&lru_pvec); |
98b27036 AA |
2076 | ntfs_debug("Done. Returning %s (written 0x%lx, status %li).", |
2077 | written ? "written" : "status", (unsigned long)written, | |
2078 | (long)status); | |
2079 | return written ? written : status; | |
2080 | } | |
2081 | ||
2082 | /** | |
2083 | * ntfs_file_aio_write_nolock - | |
2084 | */ | |
2085 | static ssize_t ntfs_file_aio_write_nolock(struct kiocb *iocb, | |
2086 | const struct iovec *iov, unsigned long nr_segs, loff_t *ppos) | |
2087 | { | |
2088 | struct file *file = iocb->ki_filp; | |
2089 | struct address_space *mapping = file->f_mapping; | |
2090 | struct inode *inode = mapping->host; | |
2091 | loff_t pos; | |
98b27036 AA |
2092 | size_t count; /* after file limit checks */ |
2093 | ssize_t written, err; | |
2094 | ||
2095 | count = 0; | |
0ceb3314 DM |
2096 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); |
2097 | if (err) | |
2098 | return err; | |
98b27036 AA |
2099 | pos = *ppos; |
2100 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2101 | /* We can write back this queue in page reclaim. */ | |
2102 | current->backing_dev_info = mapping->backing_dev_info; | |
2103 | written = 0; | |
2104 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2105 | if (err) | |
2106 | goto out; | |
2107 | if (!count) | |
2108 | goto out; | |
2f1936b8 | 2109 | err = file_remove_suid(file); |
98b27036 AA |
2110 | if (err) |
2111 | goto out; | |
870f4817 | 2112 | file_update_time(file); |
98b27036 AA |
2113 | written = ntfs_file_buffered_write(iocb, iov, nr_segs, pos, ppos, |
2114 | count); | |
2115 | out: | |
2116 | current->backing_dev_info = NULL; | |
2117 | return written ? written : err; | |
2118 | } | |
2119 | ||
2120 | /** | |
2121 | * ntfs_file_aio_write - | |
2122 | */ | |
027445c3 BP |
2123 | static ssize_t ntfs_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2124 | unsigned long nr_segs, loff_t pos) | |
98b27036 AA |
2125 | { |
2126 | struct file *file = iocb->ki_filp; | |
2127 | struct address_space *mapping = file->f_mapping; | |
2128 | struct inode *inode = mapping->host; | |
2129 | ssize_t ret; | |
98b27036 AA |
2130 | |
2131 | BUG_ON(iocb->ki_pos != pos); | |
2132 | ||
1b1dcc1b | 2133 | mutex_lock(&inode->i_mutex); |
027445c3 | 2134 | ret = ntfs_file_aio_write_nolock(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2135 | mutex_unlock(&inode->i_mutex); |
ebbbf757 JK |
2136 | if (ret > 0) { |
2137 | int err = generic_write_sync(file, pos, ret); | |
98b27036 AA |
2138 | if (err < 0) |
2139 | ret = err; | |
2140 | } | |
2141 | return ret; | |
2142 | } | |
2143 | ||
1da177e4 LT |
2144 | /** |
2145 | * ntfs_file_fsync - sync a file to disk | |
2146 | * @filp: file to be synced | |
2147 | * @dentry: dentry describing the file to sync | |
2148 | * @datasync: if non-zero only flush user data and not metadata | |
2149 | * | |
2150 | * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync | |
2151 | * system calls. This function is inspired by fs/buffer.c::file_fsync(). | |
2152 | * | |
2153 | * If @datasync is false, write the mft record and all associated extent mft | |
2154 | * records as well as the $DATA attribute and then sync the block device. | |
2155 | * | |
2156 | * If @datasync is true and the attribute is non-resident, we skip the writing | |
2157 | * of the mft record and all associated extent mft records (this might still | |
2158 | * happen due to the write_inode_now() call). | |
2159 | * | |
2160 | * Also, if @datasync is true, we do not wait on the inode to be written out | |
2161 | * but we always wait on the page cache pages to be written out. | |
2162 | * | |
2163 | * Note: In the past @filp could be NULL so we ignore it as we don't need it | |
2164 | * anyway. | |
2165 | * | |
1b1dcc1b | 2166 | * Locking: Caller must hold i_mutex on the inode. |
1da177e4 LT |
2167 | * |
2168 | * TODO: We should probably also write all attribute/index inodes associated | |
2169 | * with this inode but since we have no simple way of getting to them we ignore | |
2170 | * this problem for now. | |
2171 | */ | |
2172 | static int ntfs_file_fsync(struct file *filp, struct dentry *dentry, | |
2173 | int datasync) | |
2174 | { | |
2175 | struct inode *vi = dentry->d_inode; | |
2176 | int err, ret = 0; | |
2177 | ||
2178 | ntfs_debug("Entering for inode 0x%lx.", vi->i_ino); | |
2179 | BUG_ON(S_ISDIR(vi->i_mode)); | |
2180 | if (!datasync || !NInoNonResident(NTFS_I(vi))) | |
a9185b41 | 2181 | ret = __ntfs_write_inode(vi, 1); |
1da177e4 | 2182 | write_inode_now(vi, !datasync); |
f25dfb5e AA |
2183 | /* |
2184 | * NOTE: If we were to use mapping->private_list (see ext2 and | |
2185 | * fs/buffer.c) for dirty blocks then we could optimize the below to be | |
2186 | * sync_mapping_buffers(vi->i_mapping). | |
2187 | */ | |
1da177e4 LT |
2188 | err = sync_blockdev(vi->i_sb->s_bdev); |
2189 | if (unlikely(err && !ret)) | |
2190 | ret = err; | |
2191 | if (likely(!ret)) | |
2192 | ntfs_debug("Done."); | |
2193 | else | |
2194 | ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx. Error " | |
2195 | "%u.", datasync ? "data" : "", vi->i_ino, -ret); | |
2196 | return ret; | |
2197 | } | |
2198 | ||
2199 | #endif /* NTFS_RW */ | |
2200 | ||
4b6f5d20 | 2201 | const struct file_operations ntfs_file_ops = { |
98b27036 | 2202 | .llseek = generic_file_llseek, /* Seek inside file. */ |
543ade1f | 2203 | .read = do_sync_read, /* Read from file. */ |
98b27036 | 2204 | .aio_read = generic_file_aio_read, /* Async read from file. */ |
1da177e4 | 2205 | #ifdef NTFS_RW |
8a9f47dd | 2206 | .write = do_sync_write, /* Write to file. */ |
98b27036 | 2207 | .aio_write = ntfs_file_aio_write, /* Async write to file. */ |
98b27036 AA |
2208 | /*.release = ,*/ /* Last file is closed. See |
2209 | fs/ext2/file.c:: | |
2210 | ext2_release_file() for | |
2211 | how to use this to discard | |
2212 | preallocated space for | |
2213 | write opened files. */ | |
2214 | .fsync = ntfs_file_fsync, /* Sync a file to disk. */ | |
2215 | /*.aio_fsync = ,*/ /* Sync all outstanding async | |
2216 | i/o operations on a | |
2217 | kiocb. */ | |
1da177e4 | 2218 | #endif /* NTFS_RW */ |
98b27036 AA |
2219 | /*.ioctl = ,*/ /* Perform function on the |
2220 | mounted filesystem. */ | |
2221 | .mmap = generic_file_mmap, /* Mmap file. */ | |
2222 | .open = ntfs_file_open, /* Open file. */ | |
5ffc4ef4 | 2223 | .splice_read = generic_file_splice_read /* Zero-copy data send with |
98b27036 AA |
2224 | the data source being on |
2225 | the ntfs partition. We do | |
2226 | not need to care about the | |
2227 | data destination. */ | |
2228 | /*.sendpage = ,*/ /* Zero-copy data send with | |
2229 | the data destination being | |
2230 | on the ntfs partition. We | |
2231 | do not need to care about | |
2232 | the data source. */ | |
1da177e4 LT |
2233 | }; |
2234 | ||
92e1d5be | 2235 | const struct inode_operations ntfs_file_inode_ops = { |
1da177e4 LT |
2236 | #ifdef NTFS_RW |
2237 | .truncate = ntfs_truncate_vfs, | |
2238 | .setattr = ntfs_setattr, | |
2239 | #endif /* NTFS_RW */ | |
2240 | }; | |
2241 | ||
4b6f5d20 | 2242 | const struct file_operations ntfs_empty_file_ops = {}; |
1da177e4 | 2243 | |
92e1d5be | 2244 | const struct inode_operations ntfs_empty_inode_ops = {}; |