blob: 621de369e6f835ae6ef186a5c742c56c5d115c1d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Anton Altaparmakovf25dfb52005-09-08 20:35:33 +01002 * file.c - NTFS kernel file operations. Part of the Linux-NTFS project.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
Anton Altaparmakov78af34f2006-02-24 10:32:33 +00004 * Copyright (c) 2001-2006 Anton Altaparmakov
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/buffer_head.h>
Anton Altaparmakov98b27032005-10-11 15:40:40 +010023#include <linux/pagemap.h>
24#include <linux/pagevec.h>
25#include <linux/sched.h>
26#include <linux/swap.h>
27#include <linux/uio.h>
28#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Anton Altaparmakov98b27032005-10-11 15:40:40 +010030#include <asm/page.h>
31#include <asm/uaccess.h>
32
33#include "attrib.h"
34#include "bitmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "inode.h"
36#include "debug.h"
Anton Altaparmakov98b27032005-10-11 15:40:40 +010037#include "lcnalloc.h"
38#include "malloc.h"
39#include "mft.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "ntfs.h"
41
42/**
43 * ntfs_file_open - called when an inode is about to be opened
44 * @vi: inode to be opened
45 * @filp: file structure describing the inode
46 *
47 * Limit file size to the page cache limit on architectures where unsigned long
48 * is 32-bits. This is the most we can do for now without overflowing the page
49 * cache page index. Doing it this way means we don't run into problems because
50 * of existing too large files. It would be better to allow the user to read
51 * the beginning of the file but I doubt very much anyone is going to hit this
52 * check on a 32-bit architecture, so there is no point in adding the extra
53 * complexity required to support this.
54 *
55 * On 64-bit architectures, the check is hopefully optimized away by the
56 * compiler.
57 *
58 * After the check passes, just call generic_file_open() to do its work.
59 */
60static int ntfs_file_open(struct inode *vi, struct file *filp)
61{
62 if (sizeof(unsigned long) < 8) {
Anton Altaparmakovd4b9ba72004-11-17 15:45:08 +000063 if (i_size_read(vi) > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 return -EFBIG;
65 }
66 return generic_file_open(vi, filp);
67}
68
69#ifdef NTFS_RW
70
71/**
Anton Altaparmakov98b27032005-10-11 15:40:40 +010072 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
73 * @ni: ntfs inode of the attribute to extend
74 * @new_init_size: requested new initialized size in bytes
75 * @cached_page: store any allocated but unused page here
76 * @lru_pvec: lru-buffering pagevec of the caller
77 *
78 * Extend the initialized size of an attribute described by the ntfs inode @ni
79 * to @new_init_size bytes. This involves zeroing any non-sparse space between
80 * the old initialized size and @new_init_size both in the page cache and on
Anton Altaparmakovdda65b92005-10-24 08:57:59 +010081 * disk (if relevant complete pages are already uptodate in the page cache then
82 * these are simply marked dirty).
Anton Altaparmakov98b27032005-10-11 15:40:40 +010083 *
84 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
85 * in the resident attribute case, it is tied to the initialized size and, in
86 * the non-resident attribute case, it may not fall below the initialized size.
87 *
88 * Note that if the attribute is resident, we do not need to touch the page
89 * cache at all. This is because if the page cache page is not uptodate we
90 * bring it uptodate later, when doing the write to the mft record since we
91 * then already have the page mapped. And if the page is uptodate, the
92 * non-initialized region will already have been zeroed when the page was
93 * brought uptodate and the region may in fact already have been overwritten
94 * with new data via mmap() based writes, so we cannot just zero it. And since
95 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
96 * is unspecified, we choose not to do zeroing and thus we do not need to touch
Anton Altaparmakovdda65b92005-10-24 08:57:59 +010097 * the page at all. For a more detailed explanation see ntfs_truncate() in
98 * fs/ntfs/inode.c.
Anton Altaparmakov98b27032005-10-11 15:40:40 +010099 *
Anton Altaparmakovdda65b92005-10-24 08:57:59 +0100100 * @cached_page and @lru_pvec are just optimizations for dealing with multiple
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100101 * pages.
102 *
103 * Return 0 on success and -errno on error. In the case that an error is
104 * encountered it is possible that the initialized size will already have been
105 * incremented some way towards @new_init_size but it is guaranteed that if
106 * this is the case, the necessary zeroing will also have happened and that all
107 * metadata is self-consistent.
108 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800109 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
Anton Altaparmakovdda65b92005-10-24 08:57:59 +0100110 * held by the caller.
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100111 */
112static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size,
113 struct page **cached_page, struct pagevec *lru_pvec)
114{
115 s64 old_init_size;
116 loff_t old_i_size;
117 pgoff_t index, end_index;
118 unsigned long flags;
119 struct inode *vi = VFS_I(ni);
120 ntfs_inode *base_ni;
121 MFT_RECORD *m = NULL;
122 ATTR_RECORD *a;
123 ntfs_attr_search_ctx *ctx = NULL;
124 struct address_space *mapping;
125 struct page *page = NULL;
126 u8 *kattr;
127 int err;
128 u32 attr_len;
129
130 read_lock_irqsave(&ni->size_lock, flags);
131 old_init_size = ni->initialized_size;
132 old_i_size = i_size_read(vi);
133 BUG_ON(new_init_size > ni->allocated_size);
134 read_unlock_irqrestore(&ni->size_lock, flags);
135 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
136 "old_initialized_size 0x%llx, "
137 "new_initialized_size 0x%llx, i_size 0x%llx.",
138 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
139 (unsigned long long)old_init_size,
140 (unsigned long long)new_init_size, old_i_size);
141 if (!NInoAttr(ni))
142 base_ni = ni;
143 else
144 base_ni = ni->ext.base_ntfs_ino;
145 /* Use goto to reduce indentation and we need the label below anyway. */
146 if (NInoNonResident(ni))
147 goto do_non_resident_extend;
148 BUG_ON(old_init_size != old_i_size);
149 m = map_mft_record(base_ni);
150 if (IS_ERR(m)) {
151 err = PTR_ERR(m);
152 m = NULL;
153 goto err_out;
154 }
155 ctx = ntfs_attr_get_search_ctx(base_ni, m);
156 if (unlikely(!ctx)) {
157 err = -ENOMEM;
158 goto err_out;
159 }
160 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
161 CASE_SENSITIVE, 0, NULL, 0, ctx);
162 if (unlikely(err)) {
163 if (err == -ENOENT)
164 err = -EIO;
165 goto err_out;
166 }
167 m = ctx->mrec;
168 a = ctx->attr;
169 BUG_ON(a->non_resident);
170 /* The total length of the attribute value. */
171 attr_len = le32_to_cpu(a->data.resident.value_length);
172 BUG_ON(old_i_size != (loff_t)attr_len);
173 /*
174 * Do the zeroing in the mft record and update the attribute size in
175 * the mft record.
176 */
177 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
178 memset(kattr + attr_len, 0, new_init_size - attr_len);
179 a->data.resident.value_length = cpu_to_le32((u32)new_init_size);
180 /* Finally, update the sizes in the vfs and ntfs inodes. */
181 write_lock_irqsave(&ni->size_lock, flags);
182 i_size_write(vi, new_init_size);
183 ni->initialized_size = new_init_size;
184 write_unlock_irqrestore(&ni->size_lock, flags);
185 goto done;
186do_non_resident_extend:
187 /*
188 * If the new initialized size @new_init_size exceeds the current file
189 * size (vfs inode->i_size), we need to extend the file size to the
190 * new initialized size.
191 */
192 if (new_init_size > old_i_size) {
193 m = map_mft_record(base_ni);
194 if (IS_ERR(m)) {
195 err = PTR_ERR(m);
196 m = NULL;
197 goto err_out;
198 }
199 ctx = ntfs_attr_get_search_ctx(base_ni, m);
200 if (unlikely(!ctx)) {
201 err = -ENOMEM;
202 goto err_out;
203 }
204 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
205 CASE_SENSITIVE, 0, NULL, 0, ctx);
206 if (unlikely(err)) {
207 if (err == -ENOENT)
208 err = -EIO;
209 goto err_out;
210 }
211 m = ctx->mrec;
212 a = ctx->attr;
213 BUG_ON(!a->non_resident);
214 BUG_ON(old_i_size != (loff_t)
215 sle64_to_cpu(a->data.non_resident.data_size));
216 a->data.non_resident.data_size = cpu_to_sle64(new_init_size);
217 flush_dcache_mft_record_page(ctx->ntfs_ino);
218 mark_mft_record_dirty(ctx->ntfs_ino);
219 /* Update the file size in the vfs inode. */
220 i_size_write(vi, new_init_size);
221 ntfs_attr_put_search_ctx(ctx);
222 ctx = NULL;
223 unmap_mft_record(base_ni);
224 m = NULL;
225 }
226 mapping = vi->i_mapping;
227 index = old_init_size >> PAGE_CACHE_SHIFT;
228 end_index = (new_init_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
229 do {
230 /*
231 * Read the page. If the page is not present, this will zero
232 * the uninitialized regions for us.
233 */
Pekka Enberg090d2b12006-06-23 02:05:08 -0700234 page = read_mapping_page(mapping, index, NULL);
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100235 if (IS_ERR(page)) {
236 err = PTR_ERR(page);
237 goto init_err_out;
238 }
Nick Piggin6fe69002007-05-06 14:49:04 -0700239 if (unlikely(PageError(page))) {
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100240 page_cache_release(page);
241 err = -EIO;
242 goto init_err_out;
243 }
244 /*
245 * Update the initialized size in the ntfs inode. This is
246 * enough to make ntfs_writepage() work.
247 */
248 write_lock_irqsave(&ni->size_lock, flags);
Anton Altaparmakov3c6af7f2005-11-24 13:41:33 +0000249 ni->initialized_size = (s64)(index + 1) << PAGE_CACHE_SHIFT;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100250 if (ni->initialized_size > new_init_size)
251 ni->initialized_size = new_init_size;
252 write_unlock_irqrestore(&ni->size_lock, flags);
253 /* Set the page dirty so it gets written out. */
254 set_page_dirty(page);
255 page_cache_release(page);
256 /*
257 * Play nice with the vm and the rest of the system. This is
258 * very much needed as we can potentially be modifying the
259 * initialised size from a very small value to a really huge
260 * value, e.g.
261 * f = open(somefile, O_TRUNC);
262 * truncate(f, 10GiB);
263 * seek(f, 10GiB);
264 * write(f, 1);
265 * And this would mean we would be marking dirty hundreds of
266 * thousands of pages or as in the above example more than
267 * two and a half million pages!
268 *
269 * TODO: For sparse pages could optimize this workload by using
270 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This
271 * would be set in readpage for sparse pages and here we would
272 * not need to mark dirty any pages which have this bit set.
273 * The only caveat is that we have to clear the bit everywhere
274 * where we allocate any clusters that lie in the page or that
275 * contain the page.
276 *
277 * TODO: An even greater optimization would be for us to only
278 * call readpage() on pages which are not in sparse regions as
279 * determined from the runlist. This would greatly reduce the
280 * number of pages we read and make dirty in the case of sparse
281 * files.
282 */
283 balance_dirty_pages_ratelimited(mapping);
284 cond_resched();
285 } while (++index < end_index);
286 read_lock_irqsave(&ni->size_lock, flags);
287 BUG_ON(ni->initialized_size != new_init_size);
288 read_unlock_irqrestore(&ni->size_lock, flags);
289 /* Now bring in sync the initialized_size in the mft record. */
290 m = map_mft_record(base_ni);
291 if (IS_ERR(m)) {
292 err = PTR_ERR(m);
293 m = NULL;
294 goto init_err_out;
295 }
296 ctx = ntfs_attr_get_search_ctx(base_ni, m);
297 if (unlikely(!ctx)) {
298 err = -ENOMEM;
299 goto init_err_out;
300 }
301 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
302 CASE_SENSITIVE, 0, NULL, 0, ctx);
303 if (unlikely(err)) {
304 if (err == -ENOENT)
305 err = -EIO;
306 goto init_err_out;
307 }
308 m = ctx->mrec;
309 a = ctx->attr;
310 BUG_ON(!a->non_resident);
311 a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size);
312done:
313 flush_dcache_mft_record_page(ctx->ntfs_ino);
314 mark_mft_record_dirty(ctx->ntfs_ino);
315 if (ctx)
316 ntfs_attr_put_search_ctx(ctx);
317 if (m)
318 unmap_mft_record(base_ni);
319 ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.",
320 (unsigned long long)new_init_size, i_size_read(vi));
321 return 0;
322init_err_out:
323 write_lock_irqsave(&ni->size_lock, flags);
324 ni->initialized_size = old_init_size;
325 write_unlock_irqrestore(&ni->size_lock, flags);
326err_out:
327 if (ctx)
328 ntfs_attr_put_search_ctx(ctx);
329 if (m)
330 unmap_mft_record(base_ni);
331 ntfs_debug("Failed. Returning error code %i.", err);
332 return err;
333}
334
335/**
336 * ntfs_fault_in_pages_readable -
337 *
338 * Fault a number of userspace pages into pagetables.
339 *
340 * Unlike include/linux/pagemap.h::fault_in_pages_readable(), this one copes
341 * with more than two userspace pages as well as handling the single page case
342 * elegantly.
343 *
344 * If you find this difficult to understand, then think of the while loop being
345 * the following code, except that we do without the integer variable ret:
346 *
347 * do {
348 * ret = __get_user(c, uaddr);
349 * uaddr += PAGE_SIZE;
350 * } while (!ret && uaddr < end);
351 *
352 * Note, the final __get_user() may well run out-of-bounds of the user buffer,
353 * but _not_ out-of-bounds of the page the user buffer belongs to, and since
354 * this is only a read and not a write, and since it is still in the same page,
355 * it should not matter and this makes the code much simpler.
356 */
357static inline void ntfs_fault_in_pages_readable(const char __user *uaddr,
358 int bytes)
359{
360 const char __user *end;
361 volatile char c;
362
363 /* Set @end to the first byte outside the last page we care about. */
364 end = (const char __user*)PAGE_ALIGN((ptrdiff_t __user)uaddr + bytes);
365
366 while (!__get_user(c, uaddr) && (uaddr += PAGE_SIZE, uaddr < end))
367 ;
368}
369
370/**
371 * ntfs_fault_in_pages_readable_iovec -
372 *
373 * Same as ntfs_fault_in_pages_readable() but operates on an array of iovecs.
374 */
375static inline void ntfs_fault_in_pages_readable_iovec(const struct iovec *iov,
376 size_t iov_ofs, int bytes)
377{
378 do {
379 const char __user *buf;
380 unsigned len;
381
382 buf = iov->iov_base + iov_ofs;
383 len = iov->iov_len - iov_ofs;
384 if (len > bytes)
385 len = bytes;
386 ntfs_fault_in_pages_readable(buf, len);
387 bytes -= len;
388 iov++;
389 iov_ofs = 0;
390 } while (bytes);
391}
392
393/**
394 * __ntfs_grab_cache_pages - obtain a number of locked pages
395 * @mapping: address space mapping from which to obtain page cache pages
396 * @index: starting index in @mapping at which to begin obtaining pages
397 * @nr_pages: number of page cache pages to obtain
398 * @pages: array of pages in which to return the obtained page cache pages
399 * @cached_page: allocated but as yet unused page
400 * @lru_pvec: lru-buffering pagevec of caller
401 *
402 * Obtain @nr_pages locked page cache pages from the mapping @maping and
403 * starting at index @index.
404 *
405 * If a page is newly created, increment its refcount and add it to the
406 * caller's lru-buffering pagevec @lru_pvec.
407 *
408 * This is the same as mm/filemap.c::__grab_cache_page(), except that @nr_pages
409 * are obtained at once instead of just one page and that 0 is returned on
410 * success and -errno on error.
411 *
412 * Note, the page locks are obtained in ascending page index order.
413 */
414static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
415 pgoff_t index, const unsigned nr_pages, struct page **pages,
416 struct page **cached_page, struct pagevec *lru_pvec)
417{
418 int err, nr;
419
420 BUG_ON(!nr_pages);
421 err = nr = 0;
422 do {
423 pages[nr] = find_lock_page(mapping, index);
424 if (!pages[nr]) {
425 if (!*cached_page) {
426 *cached_page = page_cache_alloc(mapping);
427 if (unlikely(!*cached_page)) {
428 err = -ENOMEM;
429 goto err_out;
430 }
431 }
432 err = add_to_page_cache(*cached_page, mapping, index,
433 GFP_KERNEL);
434 if (unlikely(err)) {
435 if (err == -EEXIST)
436 continue;
437 goto err_out;
438 }
439 pages[nr] = *cached_page;
440 page_cache_get(*cached_page);
441 if (unlikely(!pagevec_add(lru_pvec, *cached_page)))
442 __pagevec_lru_add(lru_pvec);
443 *cached_page = NULL;
444 }
445 index++;
446 nr++;
447 } while (nr < nr_pages);
448out:
449 return err;
450err_out:
451 while (nr > 0) {
452 unlock_page(pages[--nr]);
453 page_cache_release(pages[nr]);
454 }
455 goto out;
456}
457
458static inline int ntfs_submit_bh_for_read(struct buffer_head *bh)
459{
460 lock_buffer(bh);
461 get_bh(bh);
462 bh->b_end_io = end_buffer_read_sync;
463 return submit_bh(READ, bh);
464}
465
466/**
467 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
468 * @pages: array of destination pages
469 * @nr_pages: number of pages in @pages
470 * @pos: byte position in file at which the write begins
471 * @bytes: number of bytes to be written
472 *
473 * This is called for non-resident attributes from ntfs_file_buffered_write()
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800474 * with i_mutex held on the inode (@pages[0]->mapping->host). There are
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100475 * @nr_pages pages in @pages which are locked but not kmap()ped. The source
476 * data has not yet been copied into the @pages.
477 *
478 * Need to fill any holes with actual clusters, allocate buffers if necessary,
479 * ensure all the buffers are mapped, and bring uptodate any buffers that are
480 * only partially being written to.
481 *
482 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
483 * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside
484 * the same cluster and that they are the entirety of that cluster, and that
485 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
486 *
487 * i_size is not to be modified yet.
488 *
489 * Return 0 on success or -errno on error.
490 */
491static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
492 unsigned nr_pages, s64 pos, size_t bytes)
493{
494 VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend;
495 LCN lcn;
496 s64 bh_pos, vcn_len, end, initialized_size;
497 sector_t lcn_block;
498 struct page *page;
499 struct inode *vi;
500 ntfs_inode *ni, *base_ni = NULL;
501 ntfs_volume *vol;
502 runlist_element *rl, *rl2;
503 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
504 ntfs_attr_search_ctx *ctx = NULL;
505 MFT_RECORD *m = NULL;
506 ATTR_RECORD *a = NULL;
507 unsigned long flags;
508 u32 attr_rec_len = 0;
509 unsigned blocksize, u;
510 int err, mp_size;
Richard Knutssonc49c3112006-09-30 23:27:12 -0700511 bool rl_write_locked, was_hole, is_retry;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100512 unsigned char blocksize_bits;
513 struct {
514 u8 runlist_merged:1;
515 u8 mft_attr_mapped:1;
516 u8 mp_rebuilt:1;
517 u8 attr_switched:1;
518 } status = { 0, 0, 0, 0 };
519
520 BUG_ON(!nr_pages);
521 BUG_ON(!pages);
522 BUG_ON(!*pages);
523 vi = pages[0]->mapping->host;
524 ni = NTFS_I(vi);
525 vol = ni->vol;
526 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
Anton Altaparmakovd04bd1f2005-10-24 08:41:24 +0100527 "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100528 vi->i_ino, ni->type, pages[0]->index, nr_pages,
529 (long long)pos, bytes);
Anton Altaparmakov78af34f2006-02-24 10:32:33 +0000530 blocksize = vol->sb->s_blocksize;
531 blocksize_bits = vol->sb->s_blocksize_bits;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100532 u = 0;
533 do {
534 struct page *page = pages[u];
535 /*
536 * create_empty_buffers() will create uptodate/dirty buffers if
537 * the page is uptodate/dirty.
538 */
539 if (!page_has_buffers(page)) {
540 create_empty_buffers(page, blocksize, 0);
541 if (unlikely(!page_has_buffers(page)))
542 return -ENOMEM;
543 }
544 } while (++u < nr_pages);
Richard Knutssonc49c3112006-09-30 23:27:12 -0700545 rl_write_locked = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100546 rl = NULL;
547 err = 0;
548 vcn = lcn = -1;
549 vcn_len = 0;
550 lcn_block = -1;
Richard Knutssonc49c3112006-09-30 23:27:12 -0700551 was_hole = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100552 cpos = pos >> vol->cluster_size_bits;
553 end = pos + bytes;
554 cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
555 /*
556 * Loop over each page and for each page over each buffer. Use goto to
557 * reduce indentation.
558 */
559 u = 0;
560do_next_page:
561 page = pages[u];
562 bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
563 bh = head = page_buffers(page);
564 do {
565 VCN cdelta;
566 s64 bh_end;
567 unsigned bh_cofs;
568
569 /* Clear buffer_new on all buffers to reinitialise state. */
570 if (buffer_new(bh))
571 clear_buffer_new(bh);
572 bh_end = bh_pos + blocksize;
573 bh_cpos = bh_pos >> vol->cluster_size_bits;
574 bh_cofs = bh_pos & vol->cluster_size_mask;
575 if (buffer_mapped(bh)) {
576 /*
577 * The buffer is already mapped. If it is uptodate,
578 * ignore it.
579 */
580 if (buffer_uptodate(bh))
581 continue;
582 /*
583 * The buffer is not uptodate. If the page is uptodate
584 * set the buffer uptodate and otherwise ignore it.
585 */
586 if (PageUptodate(page)) {
587 set_buffer_uptodate(bh);
588 continue;
589 }
590 /*
591 * Neither the page nor the buffer are uptodate. If
592 * the buffer is only partially being written to, we
593 * need to read it in before the write, i.e. now.
594 */
595 if ((bh_pos < pos && bh_end > pos) ||
596 (bh_pos < end && bh_end > end)) {
597 /*
598 * If the buffer is fully or partially within
599 * the initialized size, do an actual read.
600 * Otherwise, simply zero the buffer.
601 */
602 read_lock_irqsave(&ni->size_lock, flags);
603 initialized_size = ni->initialized_size;
604 read_unlock_irqrestore(&ni->size_lock, flags);
605 if (bh_pos < initialized_size) {
606 ntfs_submit_bh_for_read(bh);
607 *wait_bh++ = bh;
608 } else {
609 u8 *kaddr = kmap_atomic(page, KM_USER0);
610 memset(kaddr + bh_offset(bh), 0,
611 blocksize);
612 kunmap_atomic(kaddr, KM_USER0);
613 flush_dcache_page(page);
614 set_buffer_uptodate(bh);
615 }
616 }
617 continue;
618 }
619 /* Unmapped buffer. Need to map it. */
620 bh->b_bdev = vol->sb->s_bdev;
621 /*
622 * If the current buffer is in the same clusters as the map
623 * cache, there is no need to check the runlist again. The
624 * map cache is made up of @vcn, which is the first cached file
625 * cluster, @vcn_len which is the number of cached file
626 * clusters, @lcn is the device cluster corresponding to @vcn,
627 * and @lcn_block is the block number corresponding to @lcn.
628 */
629 cdelta = bh_cpos - vcn;
630 if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) {
631map_buffer_cached:
632 BUG_ON(lcn < 0);
633 bh->b_blocknr = lcn_block +
634 (cdelta << (vol->cluster_size_bits -
635 blocksize_bits)) +
636 (bh_cofs >> blocksize_bits);
637 set_buffer_mapped(bh);
638 /*
639 * If the page is uptodate so is the buffer. If the
640 * buffer is fully outside the write, we ignore it if
641 * it was already allocated and we mark it dirty so it
642 * gets written out if we allocated it. On the other
643 * hand, if we allocated the buffer but we are not
644 * marking it dirty we set buffer_new so we can do
645 * error recovery.
646 */
647 if (PageUptodate(page)) {
648 if (!buffer_uptodate(bh))
649 set_buffer_uptodate(bh);
650 if (unlikely(was_hole)) {
651 /* We allocated the buffer. */
652 unmap_underlying_metadata(bh->b_bdev,
653 bh->b_blocknr);
654 if (bh_end <= pos || bh_pos >= end)
655 mark_buffer_dirty(bh);
656 else
657 set_buffer_new(bh);
658 }
659 continue;
660 }
661 /* Page is _not_ uptodate. */
662 if (likely(!was_hole)) {
663 /*
664 * Buffer was already allocated. If it is not
665 * uptodate and is only partially being written
666 * to, we need to read it in before the write,
667 * i.e. now.
668 */
Anton Altaparmakov3aebf252005-11-01 15:49:31 +0000669 if (!buffer_uptodate(bh) && bh_pos < end &&
670 bh_end > pos &&
671 (bh_pos < pos ||
672 bh_end > end)) {
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100673 /*
674 * If the buffer is fully or partially
675 * within the initialized size, do an
676 * actual read. Otherwise, simply zero
677 * the buffer.
678 */
679 read_lock_irqsave(&ni->size_lock,
680 flags);
681 initialized_size = ni->initialized_size;
682 read_unlock_irqrestore(&ni->size_lock,
683 flags);
684 if (bh_pos < initialized_size) {
685 ntfs_submit_bh_for_read(bh);
686 *wait_bh++ = bh;
687 } else {
688 u8 *kaddr = kmap_atomic(page,
689 KM_USER0);
690 memset(kaddr + bh_offset(bh),
691 0, blocksize);
692 kunmap_atomic(kaddr, KM_USER0);
693 flush_dcache_page(page);
694 set_buffer_uptodate(bh);
695 }
696 }
697 continue;
698 }
699 /* We allocated the buffer. */
700 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
701 /*
702 * If the buffer is fully outside the write, zero it,
703 * set it uptodate, and mark it dirty so it gets
704 * written out. If it is partially being written to,
705 * zero region surrounding the write but leave it to
706 * commit write to do anything else. Finally, if the
707 * buffer is fully being overwritten, do nothing.
708 */
709 if (bh_end <= pos || bh_pos >= end) {
710 if (!buffer_uptodate(bh)) {
711 u8 *kaddr = kmap_atomic(page, KM_USER0);
712 memset(kaddr + bh_offset(bh), 0,
713 blocksize);
714 kunmap_atomic(kaddr, KM_USER0);
715 flush_dcache_page(page);
716 set_buffer_uptodate(bh);
717 }
718 mark_buffer_dirty(bh);
719 continue;
720 }
721 set_buffer_new(bh);
722 if (!buffer_uptodate(bh) &&
723 (bh_pos < pos || bh_end > end)) {
724 u8 *kaddr;
725 unsigned pofs;
726
727 kaddr = kmap_atomic(page, KM_USER0);
728 if (bh_pos < pos) {
729 pofs = bh_pos & ~PAGE_CACHE_MASK;
730 memset(kaddr + pofs, 0, pos - bh_pos);
731 }
732 if (bh_end > end) {
733 pofs = end & ~PAGE_CACHE_MASK;
734 memset(kaddr + pofs, 0, bh_end - end);
735 }
736 kunmap_atomic(kaddr, KM_USER0);
737 flush_dcache_page(page);
738 }
739 continue;
740 }
741 /*
742 * Slow path: this is the first buffer in the cluster. If it
743 * is outside allocated size and is not uptodate, zero it and
744 * set it uptodate.
745 */
746 read_lock_irqsave(&ni->size_lock, flags);
747 initialized_size = ni->allocated_size;
748 read_unlock_irqrestore(&ni->size_lock, flags);
749 if (bh_pos > initialized_size) {
750 if (PageUptodate(page)) {
751 if (!buffer_uptodate(bh))
752 set_buffer_uptodate(bh);
753 } else if (!buffer_uptodate(bh)) {
754 u8 *kaddr = kmap_atomic(page, KM_USER0);
755 memset(kaddr + bh_offset(bh), 0, blocksize);
756 kunmap_atomic(kaddr, KM_USER0);
757 flush_dcache_page(page);
758 set_buffer_uptodate(bh);
759 }
760 continue;
761 }
Richard Knutssonc49c3112006-09-30 23:27:12 -0700762 is_retry = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100763 if (!rl) {
764 down_read(&ni->runlist.lock);
765retry_remap:
766 rl = ni->runlist.rl;
767 }
768 if (likely(rl != NULL)) {
769 /* Seek to element containing target cluster. */
770 while (rl->length && rl[1].vcn <= bh_cpos)
771 rl++;
772 lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos);
773 if (likely(lcn >= 0)) {
774 /*
775 * Successful remap, setup the map cache and
776 * use that to deal with the buffer.
777 */
Richard Knutssonc49c3112006-09-30 23:27:12 -0700778 was_hole = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100779 vcn = bh_cpos;
780 vcn_len = rl[1].vcn - vcn;
781 lcn_block = lcn << (vol->cluster_size_bits -
782 blocksize_bits);
Anton Altaparmakovd5aeaef2005-10-19 12:23:10 +0100783 cdelta = 0;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100784 /*
Anton Altaparmakov3aebf252005-11-01 15:49:31 +0000785 * If the number of remaining clusters touched
786 * by the write is smaller or equal to the
787 * number of cached clusters, unlock the
788 * runlist as the map cache will be used from
789 * now on.
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100790 */
791 if (likely(vcn + vcn_len >= cend)) {
792 if (rl_write_locked) {
793 up_write(&ni->runlist.lock);
Richard Knutssonc49c3112006-09-30 23:27:12 -0700794 rl_write_locked = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100795 } else
796 up_read(&ni->runlist.lock);
797 rl = NULL;
798 }
799 goto map_buffer_cached;
800 }
801 } else
802 lcn = LCN_RL_NOT_MAPPED;
803 /*
804 * If it is not a hole and not out of bounds, the runlist is
805 * probably unmapped so try to map it now.
806 */
807 if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) {
808 if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) {
809 /* Attempt to map runlist. */
810 if (!rl_write_locked) {
811 /*
812 * We need the runlist locked for
813 * writing, so if it is locked for
814 * reading relock it now and retry in
815 * case it changed whilst we dropped
816 * the lock.
817 */
818 up_read(&ni->runlist.lock);
819 down_write(&ni->runlist.lock);
Richard Knutssonc49c3112006-09-30 23:27:12 -0700820 rl_write_locked = true;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100821 goto retry_remap;
822 }
823 err = ntfs_map_runlist_nolock(ni, bh_cpos,
824 NULL);
825 if (likely(!err)) {
Richard Knutssonc49c3112006-09-30 23:27:12 -0700826 is_retry = true;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100827 goto retry_remap;
828 }
829 /*
830 * If @vcn is out of bounds, pretend @lcn is
831 * LCN_ENOENT. As long as the buffer is out
832 * of bounds this will work fine.
833 */
834 if (err == -ENOENT) {
835 lcn = LCN_ENOENT;
836 err = 0;
837 goto rl_not_mapped_enoent;
838 }
839 } else
840 err = -EIO;
841 /* Failed to map the buffer, even after retrying. */
842 bh->b_blocknr = -1;
843 ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
844 "attribute type 0x%x, vcn 0x%llx, "
845 "vcn offset 0x%x, because its "
846 "location on disk could not be "
847 "determined%s (error code %i).",
848 ni->mft_no, ni->type,
849 (unsigned long long)bh_cpos,
850 (unsigned)bh_pos &
851 vol->cluster_size_mask,
852 is_retry ? " even after retrying" : "",
853 err);
854 break;
855 }
856rl_not_mapped_enoent:
857 /*
858 * The buffer is in a hole or out of bounds. We need to fill
859 * the hole, unless the buffer is in a cluster which is not
860 * touched by the write, in which case we just leave the buffer
861 * unmapped. This can only happen when the cluster size is
862 * less than the page cache size.
863 */
864 if (unlikely(vol->cluster_size < PAGE_CACHE_SIZE)) {
865 bh_cend = (bh_end + vol->cluster_size - 1) >>
866 vol->cluster_size_bits;
867 if ((bh_cend <= cpos || bh_cpos >= cend)) {
868 bh->b_blocknr = -1;
869 /*
870 * If the buffer is uptodate we skip it. If it
871 * is not but the page is uptodate, we can set
872 * the buffer uptodate. If the page is not
873 * uptodate, we can clear the buffer and set it
874 * uptodate. Whether this is worthwhile is
875 * debatable and this could be removed.
876 */
877 if (PageUptodate(page)) {
878 if (!buffer_uptodate(bh))
879 set_buffer_uptodate(bh);
880 } else if (!buffer_uptodate(bh)) {
881 u8 *kaddr = kmap_atomic(page, KM_USER0);
882 memset(kaddr + bh_offset(bh), 0,
883 blocksize);
884 kunmap_atomic(kaddr, KM_USER0);
885 flush_dcache_page(page);
886 set_buffer_uptodate(bh);
887 }
888 continue;
889 }
890 }
891 /*
892 * Out of bounds buffer is invalid if it was not really out of
893 * bounds.
894 */
895 BUG_ON(lcn != LCN_HOLE);
896 /*
897 * We need the runlist locked for writing, so if it is locked
898 * for reading relock it now and retry in case it changed
899 * whilst we dropped the lock.
900 */
901 BUG_ON(!rl);
902 if (!rl_write_locked) {
903 up_read(&ni->runlist.lock);
904 down_write(&ni->runlist.lock);
Richard Knutssonc49c3112006-09-30 23:27:12 -0700905 rl_write_locked = true;
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100906 goto retry_remap;
907 }
908 /* Find the previous last allocated cluster. */
909 BUG_ON(rl->lcn != LCN_HOLE);
910 lcn = -1;
911 rl2 = rl;
912 while (--rl2 >= ni->runlist.rl) {
913 if (rl2->lcn >= 0) {
914 lcn = rl2->lcn + rl2->length;
915 break;
916 }
917 }
918 rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
Richard Knutssonc49c3112006-09-30 23:27:12 -0700919 false);
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100920 if (IS_ERR(rl2)) {
921 err = PTR_ERR(rl2);
922 ntfs_debug("Failed to allocate cluster, error code %i.",
923 err);
924 break;
925 }
926 lcn = rl2->lcn;
927 rl = ntfs_runlists_merge(ni->runlist.rl, rl2);
928 if (IS_ERR(rl)) {
929 err = PTR_ERR(rl);
930 if (err != -ENOMEM)
931 err = -EIO;
932 if (ntfs_cluster_free_from_rl(vol, rl2)) {
933 ntfs_error(vol->sb, "Failed to release "
934 "allocated cluster in error "
935 "code path. Run chkdsk to "
936 "recover the lost cluster.");
937 NVolSetErrors(vol);
938 }
939 ntfs_free(rl2);
940 break;
941 }
942 ni->runlist.rl = rl;
943 status.runlist_merged = 1;
Anton Altaparmakovbb8047d2006-03-07 11:53:46 +0000944 ntfs_debug("Allocated cluster, lcn 0x%llx.",
945 (unsigned long long)lcn);
Anton Altaparmakov98b27032005-10-11 15:40:40 +0100946 /* Map and lock the mft record and get the attribute record. */
947 if (!NInoAttr(ni))
948 base_ni = ni;
949 else
950 base_ni = ni->ext.base_ntfs_ino;
951 m = map_mft_record(base_ni);
952 if (IS_ERR(m)) {
953 err = PTR_ERR(m);
954 break;
955 }
956 ctx = ntfs_attr_get_search_ctx(base_ni, m);
957 if (unlikely(!ctx)) {
958 err = -ENOMEM;
959 unmap_mft_record(base_ni);
960 break;
961 }
962 status.mft_attr_mapped = 1;
963 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
964 CASE_SENSITIVE, bh_cpos, NULL, 0, ctx);
965 if (unlikely(err)) {
966 if (err == -ENOENT)
967 err = -EIO;
968 break;
969 }
970 m = ctx->mrec;
971 a = ctx->attr;
972 /*
973 * Find the runlist element with which the attribute extent
974 * starts. Note, we cannot use the _attr_ version because we
975 * have mapped the mft record. That is ok because we know the
976 * runlist fragment must be mapped already to have ever gotten
977 * here, so we can just use the _rl_ version.
978 */
979 vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn);
980 rl2 = ntfs_rl_find_vcn_nolock(rl, vcn);
981 BUG_ON(!rl2);
982 BUG_ON(!rl2->length);
983 BUG_ON(rl2->lcn < LCN_HOLE);
984 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
985 /*
986 * If @highest_vcn is zero, calculate the real highest_vcn
987 * (which can really be zero).
988 */
989 if (!highest_vcn)
990 highest_vcn = (sle64_to_cpu(
991 a->data.non_resident.allocated_size) >>
992 vol->cluster_size_bits) - 1;
993 /*
994 * Determine the size of the mapping pairs array for the new
995 * extent, i.e. the old extent with the hole filled.
996 */
997 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn,
998 highest_vcn);
999 if (unlikely(mp_size <= 0)) {
1000 if (!(err = mp_size))
1001 err = -EIO;
1002 ntfs_debug("Failed to get size for mapping pairs "
1003 "array, error code %i.", err);
1004 break;
1005 }
1006 /*
1007 * Resize the attribute record to fit the new mapping pairs
1008 * array.
1009 */
1010 attr_rec_len = le32_to_cpu(a->length);
1011 err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu(
1012 a->data.non_resident.mapping_pairs_offset));
1013 if (unlikely(err)) {
1014 BUG_ON(err != -ENOSPC);
1015 // TODO: Deal with this by using the current attribute
1016 // and fill it with as much of the mapping pairs
1017 // array as possible. Then loop over each attribute
1018 // extent rewriting the mapping pairs arrays as we go
1019 // along and if when we reach the end we have not
1020 // enough space, try to resize the last attribute
1021 // extent and if even that fails, add a new attribute
1022 // extent.
1023 // We could also try to resize at each step in the hope
1024 // that we will not need to rewrite every single extent.
1025 // Note, we may need to decompress some extents to fill
1026 // the runlist as we are walking the extents...
1027 ntfs_error(vol->sb, "Not enough space in the mft "
1028 "record for the extended attribute "
1029 "record. This case is not "
1030 "implemented yet.");
1031 err = -EOPNOTSUPP;
1032 break ;
1033 }
1034 status.mp_rebuilt = 1;
1035 /*
1036 * Generate the mapping pairs array directly into the attribute
1037 * record.
1038 */
1039 err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1040 a->data.non_resident.mapping_pairs_offset),
1041 mp_size, rl2, vcn, highest_vcn, NULL);
1042 if (unlikely(err)) {
1043 ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, "
1044 "attribute type 0x%x, because building "
1045 "the mapping pairs failed with error "
1046 "code %i.", vi->i_ino,
1047 (unsigned)le32_to_cpu(ni->type), err);
1048 err = -EIO;
1049 break;
1050 }
1051 /* Update the highest_vcn but only if it was not set. */
1052 if (unlikely(!a->data.non_resident.highest_vcn))
1053 a->data.non_resident.highest_vcn =
1054 cpu_to_sle64(highest_vcn);
1055 /*
1056 * If the attribute is sparse/compressed, update the compressed
1057 * size in the ntfs_inode structure and the attribute record.
1058 */
1059 if (likely(NInoSparse(ni) || NInoCompressed(ni))) {
1060 /*
1061 * If we are not in the first attribute extent, switch
1062 * to it, but first ensure the changes will make it to
1063 * disk later.
1064 */
1065 if (a->data.non_resident.lowest_vcn) {
1066 flush_dcache_mft_record_page(ctx->ntfs_ino);
1067 mark_mft_record_dirty(ctx->ntfs_ino);
1068 ntfs_attr_reinit_search_ctx(ctx);
1069 err = ntfs_attr_lookup(ni->type, ni->name,
1070 ni->name_len, CASE_SENSITIVE,
1071 0, NULL, 0, ctx);
1072 if (unlikely(err)) {
1073 status.attr_switched = 1;
1074 break;
1075 }
1076 /* @m is not used any more so do not set it. */
1077 a = ctx->attr;
1078 }
1079 write_lock_irqsave(&ni->size_lock, flags);
1080 ni->itype.compressed.size += vol->cluster_size;
1081 a->data.non_resident.compressed_size =
1082 cpu_to_sle64(ni->itype.compressed.size);
1083 write_unlock_irqrestore(&ni->size_lock, flags);
1084 }
1085 /* Ensure the changes make it to disk. */
1086 flush_dcache_mft_record_page(ctx->ntfs_ino);
1087 mark_mft_record_dirty(ctx->ntfs_ino);
1088 ntfs_attr_put_search_ctx(ctx);
1089 unmap_mft_record(base_ni);
1090 /* Successfully filled the hole. */
1091 status.runlist_merged = 0;
1092 status.mft_attr_mapped = 0;
1093 status.mp_rebuilt = 0;
1094 /* Setup the map cache and use that to deal with the buffer. */
Richard Knutssonc49c3112006-09-30 23:27:12 -07001095 was_hole = true;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001096 vcn = bh_cpos;
1097 vcn_len = 1;
1098 lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
1099 cdelta = 0;
1100 /*
1101 * If the number of remaining clusters in the @pages is smaller
1102 * or equal to the number of cached clusters, unlock the
1103 * runlist as the map cache will be used from now on.
1104 */
1105 if (likely(vcn + vcn_len >= cend)) {
1106 up_write(&ni->runlist.lock);
Richard Knutssonc49c3112006-09-30 23:27:12 -07001107 rl_write_locked = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001108 rl = NULL;
1109 }
1110 goto map_buffer_cached;
1111 } while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1112 /* If there are no errors, do the next page. */
1113 if (likely(!err && ++u < nr_pages))
1114 goto do_next_page;
1115 /* If there are no errors, release the runlist lock if we took it. */
1116 if (likely(!err)) {
1117 if (unlikely(rl_write_locked)) {
1118 up_write(&ni->runlist.lock);
Richard Knutssonc49c3112006-09-30 23:27:12 -07001119 rl_write_locked = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001120 } else if (unlikely(rl))
1121 up_read(&ni->runlist.lock);
1122 rl = NULL;
1123 }
1124 /* If we issued read requests, let them complete. */
1125 read_lock_irqsave(&ni->size_lock, flags);
1126 initialized_size = ni->initialized_size;
1127 read_unlock_irqrestore(&ni->size_lock, flags);
1128 while (wait_bh > wait) {
1129 bh = *--wait_bh;
1130 wait_on_buffer(bh);
1131 if (likely(buffer_uptodate(bh))) {
1132 page = bh->b_page;
1133 bh_pos = ((s64)page->index << PAGE_CACHE_SHIFT) +
1134 bh_offset(bh);
1135 /*
1136 * If the buffer overflows the initialized size, need
1137 * to zero the overflowing region.
1138 */
1139 if (unlikely(bh_pos + blocksize > initialized_size)) {
1140 u8 *kaddr;
1141 int ofs = 0;
1142
1143 if (likely(bh_pos < initialized_size))
1144 ofs = initialized_size - bh_pos;
1145 kaddr = kmap_atomic(page, KM_USER0);
1146 memset(kaddr + bh_offset(bh) + ofs, 0,
1147 blocksize - ofs);
1148 kunmap_atomic(kaddr, KM_USER0);
1149 flush_dcache_page(page);
1150 }
1151 } else /* if (unlikely(!buffer_uptodate(bh))) */
1152 err = -EIO;
1153 }
1154 if (likely(!err)) {
1155 /* Clear buffer_new on all buffers. */
1156 u = 0;
1157 do {
1158 bh = head = page_buffers(pages[u]);
1159 do {
1160 if (buffer_new(bh))
1161 clear_buffer_new(bh);
1162 } while ((bh = bh->b_this_page) != head);
1163 } while (++u < nr_pages);
1164 ntfs_debug("Done.");
1165 return err;
1166 }
1167 if (status.attr_switched) {
1168 /* Get back to the attribute extent we modified. */
1169 ntfs_attr_reinit_search_ctx(ctx);
1170 if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1171 CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) {
1172 ntfs_error(vol->sb, "Failed to find required "
1173 "attribute extent of attribute in "
1174 "error code path. Run chkdsk to "
1175 "recover.");
1176 write_lock_irqsave(&ni->size_lock, flags);
1177 ni->itype.compressed.size += vol->cluster_size;
1178 write_unlock_irqrestore(&ni->size_lock, flags);
1179 flush_dcache_mft_record_page(ctx->ntfs_ino);
1180 mark_mft_record_dirty(ctx->ntfs_ino);
1181 /*
1182 * The only thing that is now wrong is the compressed
1183 * size of the base attribute extent which chkdsk
1184 * should be able to fix.
1185 */
1186 NVolSetErrors(vol);
1187 } else {
1188 m = ctx->mrec;
1189 a = ctx->attr;
1190 status.attr_switched = 0;
1191 }
1192 }
1193 /*
1194 * If the runlist has been modified, need to restore it by punching a
1195 * hole into it and we then need to deallocate the on-disk cluster as
1196 * well. Note, we only modify the runlist if we are able to generate a
1197 * new mapping pairs array, i.e. only when the mapped attribute extent
1198 * is not switched.
1199 */
1200 if (status.runlist_merged && !status.attr_switched) {
1201 BUG_ON(!rl_write_locked);
1202 /* Make the file cluster we allocated sparse in the runlist. */
1203 if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) {
1204 ntfs_error(vol->sb, "Failed to punch hole into "
1205 "attribute runlist in error code "
1206 "path. Run chkdsk to recover the "
1207 "lost cluster.");
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001208 NVolSetErrors(vol);
1209 } else /* if (success) */ {
1210 status.runlist_merged = 0;
1211 /*
1212 * Deallocate the on-disk cluster we allocated but only
1213 * if we succeeded in punching its vcn out of the
1214 * runlist.
1215 */
1216 down_write(&vol->lcnbmp_lock);
1217 if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1218 ntfs_error(vol->sb, "Failed to release "
1219 "allocated cluster in error "
1220 "code path. Run chkdsk to "
1221 "recover the lost cluster.");
1222 NVolSetErrors(vol);
1223 }
1224 up_write(&vol->lcnbmp_lock);
1225 }
1226 }
1227 /*
1228 * Resize the attribute record to its old size and rebuild the mapping
1229 * pairs array. Note, we only can do this if the runlist has been
1230 * restored to its old state which also implies that the mapped
1231 * attribute extent is not switched.
1232 */
1233 if (status.mp_rebuilt && !status.runlist_merged) {
1234 if (ntfs_attr_record_resize(m, a, attr_rec_len)) {
1235 ntfs_error(vol->sb, "Failed to restore attribute "
1236 "record in error code path. Run "
1237 "chkdsk to recover.");
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001238 NVolSetErrors(vol);
1239 } else /* if (success) */ {
1240 if (ntfs_mapping_pairs_build(vol, (u8*)a +
1241 le16_to_cpu(a->data.non_resident.
1242 mapping_pairs_offset), attr_rec_len -
1243 le16_to_cpu(a->data.non_resident.
1244 mapping_pairs_offset), ni->runlist.rl,
1245 vcn, highest_vcn, NULL)) {
1246 ntfs_error(vol->sb, "Failed to restore "
1247 "mapping pairs array in error "
1248 "code path. Run chkdsk to "
1249 "recover.");
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001250 NVolSetErrors(vol);
1251 }
1252 flush_dcache_mft_record_page(ctx->ntfs_ino);
1253 mark_mft_record_dirty(ctx->ntfs_ino);
1254 }
1255 }
1256 /* Release the mft record and the attribute. */
1257 if (status.mft_attr_mapped) {
1258 ntfs_attr_put_search_ctx(ctx);
1259 unmap_mft_record(base_ni);
1260 }
1261 /* Release the runlist lock. */
1262 if (rl_write_locked)
1263 up_write(&ni->runlist.lock);
1264 else if (rl)
1265 up_read(&ni->runlist.lock);
1266 /*
1267 * Zero out any newly allocated blocks to avoid exposing stale data.
1268 * If BH_New is set, we know that the block was newly allocated above
1269 * and that it has not been fully zeroed and marked dirty yet.
1270 */
1271 nr_pages = u;
1272 u = 0;
1273 end = bh_cpos << vol->cluster_size_bits;
1274 do {
1275 page = pages[u];
1276 bh = head = page_buffers(page);
1277 do {
1278 if (u == nr_pages &&
1279 ((s64)page->index << PAGE_CACHE_SHIFT) +
1280 bh_offset(bh) >= end)
1281 break;
1282 if (!buffer_new(bh))
1283 continue;
1284 clear_buffer_new(bh);
1285 if (!buffer_uptodate(bh)) {
1286 if (PageUptodate(page))
1287 set_buffer_uptodate(bh);
1288 else {
1289 u8 *kaddr = kmap_atomic(page, KM_USER0);
1290 memset(kaddr + bh_offset(bh), 0,
1291 blocksize);
1292 kunmap_atomic(kaddr, KM_USER0);
1293 flush_dcache_page(page);
1294 set_buffer_uptodate(bh);
1295 }
1296 }
1297 mark_buffer_dirty(bh);
1298 } while ((bh = bh->b_this_page) != head);
1299 } while (++u <= nr_pages);
1300 ntfs_error(vol->sb, "Failed. Returning error code %i.", err);
1301 return err;
1302}
1303
1304/*
1305 * Copy as much as we can into the pages and return the number of bytes which
1306 * were sucessfully copied. If a fault is encountered then clear the pages
1307 * out to (ofs + bytes) and return the number of bytes which were copied.
1308 */
1309static inline size_t ntfs_copy_from_user(struct page **pages,
1310 unsigned nr_pages, unsigned ofs, const char __user *buf,
1311 size_t bytes)
1312{
1313 struct page **last_page = pages + nr_pages;
1314 char *kaddr;
1315 size_t total = 0;
1316 unsigned len;
1317 int left;
1318
1319 do {
1320 len = PAGE_CACHE_SIZE - ofs;
1321 if (len > bytes)
1322 len = bytes;
1323 kaddr = kmap_atomic(*pages, KM_USER0);
1324 left = __copy_from_user_inatomic(kaddr + ofs, buf, len);
1325 kunmap_atomic(kaddr, KM_USER0);
1326 if (unlikely(left)) {
1327 /* Do it the slow way. */
1328 kaddr = kmap(*pages);
1329 left = __copy_from_user(kaddr + ofs, buf, len);
1330 kunmap(*pages);
1331 if (unlikely(left))
1332 goto err_out;
1333 }
1334 total += len;
1335 bytes -= len;
1336 if (!bytes)
1337 break;
1338 buf += len;
1339 ofs = 0;
1340 } while (++pages < last_page);
1341out:
1342 return total;
1343err_out:
1344 total += len - left;
1345 /* Zero the rest of the target like __copy_from_user(). */
1346 while (++pages < last_page) {
1347 bytes -= len;
1348 if (!bytes)
1349 break;
1350 len = PAGE_CACHE_SIZE;
1351 if (len > bytes)
1352 len = bytes;
1353 kaddr = kmap_atomic(*pages, KM_USER0);
1354 memset(kaddr, 0, len);
1355 kunmap_atomic(kaddr, KM_USER0);
1356 }
1357 goto out;
1358}
1359
NeilBrown01408c42006-06-25 05:47:58 -07001360static size_t __ntfs_copy_from_user_iovec_inatomic(char *vaddr,
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001361 const struct iovec *iov, size_t iov_ofs, size_t bytes)
1362{
1363 size_t total = 0;
1364
1365 while (1) {
1366 const char __user *buf = iov->iov_base + iov_ofs;
1367 unsigned len;
1368 size_t left;
1369
1370 len = iov->iov_len - iov_ofs;
1371 if (len > bytes)
1372 len = bytes;
1373 left = __copy_from_user_inatomic(vaddr, buf, len);
1374 total += len;
1375 bytes -= len;
1376 vaddr += len;
1377 if (unlikely(left)) {
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001378 total -= left;
1379 break;
1380 }
1381 if (!bytes)
1382 break;
1383 iov++;
1384 iov_ofs = 0;
1385 }
1386 return total;
1387}
1388
1389static inline void ntfs_set_next_iovec(const struct iovec **iovp,
1390 size_t *iov_ofsp, size_t bytes)
1391{
1392 const struct iovec *iov = *iovp;
1393 size_t iov_ofs = *iov_ofsp;
1394
1395 while (bytes) {
1396 unsigned len;
1397
1398 len = iov->iov_len - iov_ofs;
1399 if (len > bytes)
1400 len = bytes;
1401 bytes -= len;
1402 iov_ofs += len;
1403 if (iov->iov_len == iov_ofs) {
1404 iov++;
1405 iov_ofs = 0;
1406 }
1407 }
1408 *iovp = iov;
1409 *iov_ofsp = iov_ofs;
1410}
1411
1412/*
1413 * This has the same side-effects and return value as ntfs_copy_from_user().
1414 * The difference is that on a fault we need to memset the remainder of the
1415 * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s
1416 * single-segment behaviour.
1417 *
NeilBrown01408c42006-06-25 05:47:58 -07001418 * We call the same helper (__ntfs_copy_from_user_iovec_inatomic()) both
1419 * when atomic and when not atomic. This is ok because
1420 * __ntfs_copy_from_user_iovec_inatomic() calls __copy_from_user_inatomic()
1421 * and it is ok to call this when non-atomic.
1422 * Infact, the only difference between __copy_from_user_inatomic() and
1423 * __copy_from_user() is that the latter calls might_sleep() and the former
1424 * should not zero the tail of the buffer on error. And on many
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001425 * architectures __copy_from_user_inatomic() is just defined to
1426 * __copy_from_user() so it makes no difference at all on those architectures.
1427 */
1428static inline size_t ntfs_copy_from_user_iovec(struct page **pages,
1429 unsigned nr_pages, unsigned ofs, const struct iovec **iov,
1430 size_t *iov_ofs, size_t bytes)
1431{
1432 struct page **last_page = pages + nr_pages;
1433 char *kaddr;
1434 size_t copied, len, total = 0;
1435
1436 do {
1437 len = PAGE_CACHE_SIZE - ofs;
1438 if (len > bytes)
1439 len = bytes;
1440 kaddr = kmap_atomic(*pages, KM_USER0);
NeilBrown01408c42006-06-25 05:47:58 -07001441 copied = __ntfs_copy_from_user_iovec_inatomic(kaddr + ofs,
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001442 *iov, *iov_ofs, len);
1443 kunmap_atomic(kaddr, KM_USER0);
1444 if (unlikely(copied != len)) {
1445 /* Do it the slow way. */
1446 kaddr = kmap(*pages);
NeilBrown01408c42006-06-25 05:47:58 -07001447 copied = __ntfs_copy_from_user_iovec_inatomic(kaddr + ofs,
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001448 *iov, *iov_ofs, len);
NeilBrown01408c42006-06-25 05:47:58 -07001449 /*
1450 * Zero the rest of the target like __copy_from_user().
1451 */
1452 memset(kaddr + ofs + copied, 0, len - copied);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001453 kunmap(*pages);
1454 if (unlikely(copied != len))
1455 goto err_out;
1456 }
1457 total += len;
1458 bytes -= len;
1459 if (!bytes)
1460 break;
1461 ntfs_set_next_iovec(iov, iov_ofs, len);
1462 ofs = 0;
1463 } while (++pages < last_page);
1464out:
1465 return total;
1466err_out:
1467 total += copied;
1468 /* Zero the rest of the target like __copy_from_user(). */
1469 while (++pages < last_page) {
1470 bytes -= len;
1471 if (!bytes)
1472 break;
1473 len = PAGE_CACHE_SIZE;
1474 if (len > bytes)
1475 len = bytes;
1476 kaddr = kmap_atomic(*pages, KM_USER0);
1477 memset(kaddr, 0, len);
1478 kunmap_atomic(kaddr, KM_USER0);
1479 }
1480 goto out;
1481}
1482
1483static inline void ntfs_flush_dcache_pages(struct page **pages,
1484 unsigned nr_pages)
1485{
1486 BUG_ON(!nr_pages);
Anton Altaparmakovf893afb2006-06-22 14:47:15 -07001487 /*
1488 * Warning: Do not do the decrement at the same time as the call to
1489 * flush_dcache_page() because it is a NULL macro on i386 and hence the
1490 * decrement never happens so the loop never terminates.
1491 */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001492 do {
Anton Altaparmakovf893afb2006-06-22 14:47:15 -07001493 --nr_pages;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001494 flush_dcache_page(pages[nr_pages]);
Anton Altaparmakovf893afb2006-06-22 14:47:15 -07001495 } while (nr_pages > 0);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001496}
1497
1498/**
1499 * ntfs_commit_pages_after_non_resident_write - commit the received data
1500 * @pages: array of destination pages
1501 * @nr_pages: number of pages in @pages
1502 * @pos: byte position in file at which the write begins
1503 * @bytes: number of bytes to be written
1504 *
1505 * See description of ntfs_commit_pages_after_write(), below.
1506 */
1507static inline int ntfs_commit_pages_after_non_resident_write(
1508 struct page **pages, const unsigned nr_pages,
1509 s64 pos, size_t bytes)
1510{
1511 s64 end, initialized_size;
1512 struct inode *vi;
1513 ntfs_inode *ni, *base_ni;
1514 struct buffer_head *bh, *head;
1515 ntfs_attr_search_ctx *ctx;
1516 MFT_RECORD *m;
1517 ATTR_RECORD *a;
1518 unsigned long flags;
1519 unsigned blocksize, u;
1520 int err;
1521
1522 vi = pages[0]->mapping->host;
1523 ni = NTFS_I(vi);
Anton Altaparmakov78af34f2006-02-24 10:32:33 +00001524 blocksize = vi->i_sb->s_blocksize;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001525 end = pos + bytes;
1526 u = 0;
1527 do {
1528 s64 bh_pos;
1529 struct page *page;
Richard Knutssonc49c3112006-09-30 23:27:12 -07001530 bool partial;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001531
1532 page = pages[u];
1533 bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
1534 bh = head = page_buffers(page);
Richard Knutssonc49c3112006-09-30 23:27:12 -07001535 partial = false;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001536 do {
1537 s64 bh_end;
1538
1539 bh_end = bh_pos + blocksize;
1540 if (bh_end <= pos || bh_pos >= end) {
1541 if (!buffer_uptodate(bh))
Richard Knutssonc49c3112006-09-30 23:27:12 -07001542 partial = true;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001543 } else {
1544 set_buffer_uptodate(bh);
1545 mark_buffer_dirty(bh);
1546 }
1547 } while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1548 /*
1549 * If all buffers are now uptodate but the page is not, set the
1550 * page uptodate.
1551 */
1552 if (!partial && !PageUptodate(page))
1553 SetPageUptodate(page);
1554 } while (++u < nr_pages);
1555 /*
1556 * Finally, if we do not need to update initialized_size or i_size we
1557 * are finished.
1558 */
1559 read_lock_irqsave(&ni->size_lock, flags);
1560 initialized_size = ni->initialized_size;
1561 read_unlock_irqrestore(&ni->size_lock, flags);
1562 if (end <= initialized_size) {
1563 ntfs_debug("Done.");
1564 return 0;
1565 }
1566 /*
1567 * Update initialized_size/i_size as appropriate, both in the inode and
1568 * the mft record.
1569 */
1570 if (!NInoAttr(ni))
1571 base_ni = ni;
1572 else
1573 base_ni = ni->ext.base_ntfs_ino;
1574 /* Map, pin, and lock the mft record. */
1575 m = map_mft_record(base_ni);
1576 if (IS_ERR(m)) {
1577 err = PTR_ERR(m);
1578 m = NULL;
1579 ctx = NULL;
1580 goto err_out;
1581 }
1582 BUG_ON(!NInoNonResident(ni));
1583 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1584 if (unlikely(!ctx)) {
1585 err = -ENOMEM;
1586 goto err_out;
1587 }
1588 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1589 CASE_SENSITIVE, 0, NULL, 0, ctx);
1590 if (unlikely(err)) {
1591 if (err == -ENOENT)
1592 err = -EIO;
1593 goto err_out;
1594 }
1595 a = ctx->attr;
1596 BUG_ON(!a->non_resident);
1597 write_lock_irqsave(&ni->size_lock, flags);
1598 BUG_ON(end > ni->allocated_size);
1599 ni->initialized_size = end;
1600 a->data.non_resident.initialized_size = cpu_to_sle64(end);
1601 if (end > i_size_read(vi)) {
1602 i_size_write(vi, end);
1603 a->data.non_resident.data_size =
1604 a->data.non_resident.initialized_size;
1605 }
1606 write_unlock_irqrestore(&ni->size_lock, flags);
1607 /* Mark the mft record dirty, so it gets written back. */
1608 flush_dcache_mft_record_page(ctx->ntfs_ino);
1609 mark_mft_record_dirty(ctx->ntfs_ino);
1610 ntfs_attr_put_search_ctx(ctx);
1611 unmap_mft_record(base_ni);
1612 ntfs_debug("Done.");
1613 return 0;
1614err_out:
1615 if (ctx)
1616 ntfs_attr_put_search_ctx(ctx);
1617 if (m)
1618 unmap_mft_record(base_ni);
1619 ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
1620 "code %i).", err);
Anton Altaparmakovf95c4012006-03-23 15:59:32 +00001621 if (err != -ENOMEM)
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001622 NVolSetErrors(ni->vol);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001623 return err;
1624}
1625
1626/**
1627 * ntfs_commit_pages_after_write - commit the received data
1628 * @pages: array of destination pages
1629 * @nr_pages: number of pages in @pages
1630 * @pos: byte position in file at which the write begins
1631 * @bytes: number of bytes to be written
1632 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001633 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001634 * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are
1635 * locked but not kmap()ped. The source data has already been copied into the
1636 * @page. ntfs_prepare_pages_for_non_resident_write() has been called before
1637 * the data was copied (for non-resident attributes only) and it returned
1638 * success.
1639 *
1640 * Need to set uptodate and mark dirty all buffers within the boundary of the
1641 * write. If all buffers in a page are uptodate we set the page uptodate, too.
1642 *
1643 * Setting the buffers dirty ensures that they get written out later when
1644 * ntfs_writepage() is invoked by the VM.
1645 *
1646 * Finally, we need to update i_size and initialized_size as appropriate both
1647 * in the inode and the mft record.
1648 *
1649 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1650 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1651 * page are uptodate, and updates i_size if the end of io is beyond i_size. In
1652 * that case, it also marks the inode dirty.
1653 *
1654 * If things have gone as outlined in
1655 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1656 * content modifications here for non-resident attributes. For resident
1657 * attributes we need to do the uptodate bringing here which we combine with
1658 * the copying into the mft record which means we save one atomic kmap.
1659 *
1660 * Return 0 on success or -errno on error.
1661 */
1662static int ntfs_commit_pages_after_write(struct page **pages,
1663 const unsigned nr_pages, s64 pos, size_t bytes)
1664{
1665 s64 end, initialized_size;
1666 loff_t i_size;
1667 struct inode *vi;
1668 ntfs_inode *ni, *base_ni;
1669 struct page *page;
1670 ntfs_attr_search_ctx *ctx;
1671 MFT_RECORD *m;
1672 ATTR_RECORD *a;
1673 char *kattr, *kaddr;
1674 unsigned long flags;
1675 u32 attr_len;
1676 int err;
1677
1678 BUG_ON(!nr_pages);
1679 BUG_ON(!pages);
1680 page = pages[0];
1681 BUG_ON(!page);
1682 vi = page->mapping->host;
1683 ni = NTFS_I(vi);
1684 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
Anton Altaparmakovd04bd1f2005-10-24 08:41:24 +01001685 "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001686 vi->i_ino, ni->type, page->index, nr_pages,
1687 (long long)pos, bytes);
1688 if (NInoNonResident(ni))
1689 return ntfs_commit_pages_after_non_resident_write(pages,
1690 nr_pages, pos, bytes);
1691 BUG_ON(nr_pages > 1);
1692 /*
1693 * Attribute is resident, implying it is not compressed, encrypted, or
1694 * sparse.
1695 */
1696 if (!NInoAttr(ni))
1697 base_ni = ni;
1698 else
1699 base_ni = ni->ext.base_ntfs_ino;
1700 BUG_ON(NInoNonResident(ni));
1701 /* Map, pin, and lock the mft record. */
1702 m = map_mft_record(base_ni);
1703 if (IS_ERR(m)) {
1704 err = PTR_ERR(m);
1705 m = NULL;
1706 ctx = NULL;
1707 goto err_out;
1708 }
1709 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1710 if (unlikely(!ctx)) {
1711 err = -ENOMEM;
1712 goto err_out;
1713 }
1714 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1715 CASE_SENSITIVE, 0, NULL, 0, ctx);
1716 if (unlikely(err)) {
1717 if (err == -ENOENT)
1718 err = -EIO;
1719 goto err_out;
1720 }
1721 a = ctx->attr;
1722 BUG_ON(a->non_resident);
1723 /* The total length of the attribute value. */
1724 attr_len = le32_to_cpu(a->data.resident.value_length);
1725 i_size = i_size_read(vi);
1726 BUG_ON(attr_len != i_size);
1727 BUG_ON(pos > attr_len);
1728 end = pos + bytes;
1729 BUG_ON(end > le32_to_cpu(a->length) -
1730 le16_to_cpu(a->data.resident.value_offset));
1731 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
1732 kaddr = kmap_atomic(page, KM_USER0);
1733 /* Copy the received data from the page to the mft record. */
1734 memcpy(kattr + pos, kaddr + pos, bytes);
1735 /* Update the attribute length if necessary. */
1736 if (end > attr_len) {
1737 attr_len = end;
1738 a->data.resident.value_length = cpu_to_le32(attr_len);
1739 }
1740 /*
1741 * If the page is not uptodate, bring the out of bounds area(s)
1742 * uptodate by copying data from the mft record to the page.
1743 */
1744 if (!PageUptodate(page)) {
1745 if (pos > 0)
1746 memcpy(kaddr, kattr, pos);
1747 if (end < attr_len)
1748 memcpy(kaddr + end, kattr + end, attr_len - end);
1749 /* Zero the region outside the end of the attribute value. */
1750 memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len);
1751 flush_dcache_page(page);
1752 SetPageUptodate(page);
1753 }
1754 kunmap_atomic(kaddr, KM_USER0);
1755 /* Update initialized_size/i_size if necessary. */
1756 read_lock_irqsave(&ni->size_lock, flags);
1757 initialized_size = ni->initialized_size;
1758 BUG_ON(end > ni->allocated_size);
1759 read_unlock_irqrestore(&ni->size_lock, flags);
1760 BUG_ON(initialized_size != i_size);
1761 if (end > initialized_size) {
1762 unsigned long flags;
1763
1764 write_lock_irqsave(&ni->size_lock, flags);
1765 ni->initialized_size = end;
1766 i_size_write(vi, end);
1767 write_unlock_irqrestore(&ni->size_lock, flags);
1768 }
1769 /* Mark the mft record dirty, so it gets written back. */
1770 flush_dcache_mft_record_page(ctx->ntfs_ino);
1771 mark_mft_record_dirty(ctx->ntfs_ino);
1772 ntfs_attr_put_search_ctx(ctx);
1773 unmap_mft_record(base_ni);
1774 ntfs_debug("Done.");
1775 return 0;
1776err_out:
1777 if (err == -ENOMEM) {
1778 ntfs_warning(vi->i_sb, "Error allocating memory required to "
1779 "commit the write.");
1780 if (PageUptodate(page)) {
1781 ntfs_warning(vi->i_sb, "Page is uptodate, setting "
1782 "dirty so the write will be retried "
1783 "later on by the VM.");
1784 /*
1785 * Put the page on mapping->dirty_pages, but leave its
1786 * buffers' dirty state as-is.
1787 */
1788 __set_page_dirty_nobuffers(page);
1789 err = 0;
1790 } else
1791 ntfs_error(vi->i_sb, "Page is not uptodate. Written "
1792 "data has been lost.");
1793 } else {
1794 ntfs_error(vi->i_sb, "Resident attribute commit write failed "
1795 "with error %i.", err);
1796 NVolSetErrors(ni->vol);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001797 }
1798 if (ctx)
1799 ntfs_attr_put_search_ctx(ctx);
1800 if (m)
1801 unmap_mft_record(base_ni);
1802 return err;
1803}
1804
1805/**
1806 * ntfs_file_buffered_write -
1807 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001808 * Locking: The vfs is holding ->i_mutex on the inode.
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001809 */
1810static ssize_t ntfs_file_buffered_write(struct kiocb *iocb,
1811 const struct iovec *iov, unsigned long nr_segs,
1812 loff_t pos, loff_t *ppos, size_t count)
1813{
1814 struct file *file = iocb->ki_filp;
1815 struct address_space *mapping = file->f_mapping;
1816 struct inode *vi = mapping->host;
1817 ntfs_inode *ni = NTFS_I(vi);
1818 ntfs_volume *vol = ni->vol;
1819 struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER];
1820 struct page *cached_page = NULL;
1821 char __user *buf = NULL;
1822 s64 end, ll;
1823 VCN last_vcn;
1824 LCN lcn;
1825 unsigned long flags;
Anton Altaparmakovdda65b92005-10-24 08:57:59 +01001826 size_t bytes, iov_ofs = 0; /* Offset in the current iovec. */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001827 ssize_t status, written;
1828 unsigned nr_pages;
1829 int err;
1830 struct pagevec lru_pvec;
1831
1832 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
1833 "pos 0x%llx, count 0x%lx.",
1834 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
1835 (unsigned long long)pos, (unsigned long)count);
1836 if (unlikely(!count))
1837 return 0;
1838 BUG_ON(NInoMstProtected(ni));
1839 /*
1840 * If the attribute is not an index root and it is encrypted or
1841 * compressed, we cannot write to it yet. Note we need to check for
1842 * AT_INDEX_ALLOCATION since this is the type of both directory and
1843 * index inodes.
1844 */
1845 if (ni->type != AT_INDEX_ALLOCATION) {
1846 /* If file is encrypted, deny access, just like NT4. */
1847 if (NInoEncrypted(ni)) {
Anton Altaparmakov7d0ffdb2005-10-19 12:21:19 +01001848 /*
1849 * Reminder for later: Encrypted files are _always_
1850 * non-resident so that the content can always be
1851 * encrypted.
1852 */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001853 ntfs_debug("Denying write access to encrypted file.");
1854 return -EACCES;
1855 }
1856 if (NInoCompressed(ni)) {
Anton Altaparmakov7d0ffdb2005-10-19 12:21:19 +01001857 /* Only unnamed $DATA attribute can be compressed. */
1858 BUG_ON(ni->type != AT_DATA);
1859 BUG_ON(ni->name_len);
1860 /*
1861 * Reminder for later: If resident, the data is not
1862 * actually compressed. Only on the switch to non-
1863 * resident does compression kick in. This is in
1864 * contrast to encrypted files (see above).
1865 */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001866 ntfs_error(vi->i_sb, "Writing to compressed files is "
1867 "not implemented yet. Sorry.");
1868 return -EOPNOTSUPP;
1869 }
1870 }
1871 /*
1872 * If a previous ntfs_truncate() failed, repeat it and abort if it
1873 * fails again.
1874 */
1875 if (unlikely(NInoTruncateFailed(ni))) {
1876 down_write(&vi->i_alloc_sem);
1877 err = ntfs_truncate(vi);
1878 up_write(&vi->i_alloc_sem);
1879 if (err || NInoTruncateFailed(ni)) {
1880 if (!err)
1881 err = -EIO;
1882 ntfs_error(vol->sb, "Cannot perform write to inode "
1883 "0x%lx, attribute type 0x%x, because "
1884 "ntfs_truncate() failed (error code "
1885 "%i).", vi->i_ino,
1886 (unsigned)le32_to_cpu(ni->type), err);
1887 return err;
1888 }
1889 }
1890 /* The first byte after the write. */
1891 end = pos + count;
1892 /*
1893 * If the write goes beyond the allocated size, extend the allocation
1894 * to cover the whole of the write, rounded up to the nearest cluster.
1895 */
1896 read_lock_irqsave(&ni->size_lock, flags);
1897 ll = ni->allocated_size;
1898 read_unlock_irqrestore(&ni->size_lock, flags);
1899 if (end > ll) {
1900 /* Extend the allocation without changing the data size. */
1901 ll = ntfs_attr_extend_allocation(ni, end, -1, pos);
1902 if (likely(ll >= 0)) {
1903 BUG_ON(pos >= ll);
1904 /* If the extension was partial truncate the write. */
1905 if (end > ll) {
1906 ntfs_debug("Truncating write to inode 0x%lx, "
1907 "attribute type 0x%x, because "
1908 "the allocation was only "
1909 "partially extended.",
1910 vi->i_ino, (unsigned)
1911 le32_to_cpu(ni->type));
1912 end = ll;
1913 count = ll - pos;
1914 }
1915 } else {
1916 err = ll;
1917 read_lock_irqsave(&ni->size_lock, flags);
1918 ll = ni->allocated_size;
1919 read_unlock_irqrestore(&ni->size_lock, flags);
1920 /* Perform a partial write if possible or fail. */
1921 if (pos < ll) {
1922 ntfs_debug("Truncating write to inode 0x%lx, "
1923 "attribute type 0x%x, because "
1924 "extending the allocation "
1925 "failed (error code %i).",
1926 vi->i_ino, (unsigned)
1927 le32_to_cpu(ni->type), err);
1928 end = ll;
1929 count = ll - pos;
1930 } else {
1931 ntfs_error(vol->sb, "Cannot perform write to "
1932 "inode 0x%lx, attribute type "
1933 "0x%x, because extending the "
1934 "allocation failed (error "
1935 "code %i).", vi->i_ino,
1936 (unsigned)
1937 le32_to_cpu(ni->type), err);
1938 return err;
1939 }
1940 }
1941 }
1942 pagevec_init(&lru_pvec, 0);
1943 written = 0;
1944 /*
1945 * If the write starts beyond the initialized size, extend it up to the
1946 * beginning of the write and initialize all non-sparse space between
1947 * the old initialized size and the new one. This automatically also
1948 * increments the vfs inode->i_size to keep it above or equal to the
1949 * initialized_size.
1950 */
1951 read_lock_irqsave(&ni->size_lock, flags);
1952 ll = ni->initialized_size;
1953 read_unlock_irqrestore(&ni->size_lock, flags);
1954 if (pos > ll) {
1955 err = ntfs_attr_extend_initialized(ni, pos, &cached_page,
1956 &lru_pvec);
1957 if (err < 0) {
1958 ntfs_error(vol->sb, "Cannot perform write to inode "
1959 "0x%lx, attribute type 0x%x, because "
1960 "extending the initialized size "
1961 "failed (error code %i).", vi->i_ino,
1962 (unsigned)le32_to_cpu(ni->type), err);
1963 status = err;
1964 goto err_out;
1965 }
1966 }
1967 /*
1968 * Determine the number of pages per cluster for non-resident
1969 * attributes.
1970 */
1971 nr_pages = 1;
1972 if (vol->cluster_size > PAGE_CACHE_SIZE && NInoNonResident(ni))
1973 nr_pages = vol->cluster_size >> PAGE_CACHE_SHIFT;
1974 /* Finally, perform the actual write. */
1975 last_vcn = -1;
1976 if (likely(nr_segs == 1))
1977 buf = iov->iov_base;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01001978 do {
1979 VCN vcn;
1980 pgoff_t idx, start_idx;
1981 unsigned ofs, do_pages, u;
1982 size_t copied;
1983
1984 start_idx = idx = pos >> PAGE_CACHE_SHIFT;
1985 ofs = pos & ~PAGE_CACHE_MASK;
1986 bytes = PAGE_CACHE_SIZE - ofs;
1987 do_pages = 1;
1988 if (nr_pages > 1) {
1989 vcn = pos >> vol->cluster_size_bits;
1990 if (vcn != last_vcn) {
1991 last_vcn = vcn;
1992 /*
1993 * Get the lcn of the vcn the write is in. If
1994 * it is a hole, need to lock down all pages in
1995 * the cluster.
1996 */
1997 down_read(&ni->runlist.lock);
1998 lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >>
Richard Knutssonc49c3112006-09-30 23:27:12 -07001999 vol->cluster_size_bits, false);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002000 up_read(&ni->runlist.lock);
2001 if (unlikely(lcn < LCN_HOLE)) {
2002 status = -EIO;
2003 if (lcn == LCN_ENOMEM)
2004 status = -ENOMEM;
2005 else
2006 ntfs_error(vol->sb, "Cannot "
2007 "perform write to "
2008 "inode 0x%lx, "
2009 "attribute type 0x%x, "
2010 "because the attribute "
2011 "is corrupt.",
2012 vi->i_ino, (unsigned)
2013 le32_to_cpu(ni->type));
2014 break;
2015 }
2016 if (lcn == LCN_HOLE) {
2017 start_idx = (pos & ~(s64)
2018 vol->cluster_size_mask)
2019 >> PAGE_CACHE_SHIFT;
2020 bytes = vol->cluster_size - (pos &
2021 vol->cluster_size_mask);
2022 do_pages = nr_pages;
2023 }
2024 }
2025 }
2026 if (bytes > count)
2027 bytes = count;
2028 /*
2029 * Bring in the user page(s) that we will copy from _first_.
2030 * Otherwise there is a nasty deadlock on copying from the same
2031 * page(s) as we are writing to, without it/them being marked
2032 * up-to-date. Note, at present there is nothing to stop the
2033 * pages being swapped out between us bringing them into memory
2034 * and doing the actual copying.
2035 */
2036 if (likely(nr_segs == 1))
2037 ntfs_fault_in_pages_readable(buf, bytes);
2038 else
2039 ntfs_fault_in_pages_readable_iovec(iov, iov_ofs, bytes);
2040 /* Get and lock @do_pages starting at index @start_idx. */
2041 status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages,
2042 pages, &cached_page, &lru_pvec);
2043 if (unlikely(status))
2044 break;
2045 /*
2046 * For non-resident attributes, we need to fill any holes with
2047 * actual clusters and ensure all bufferes are mapped. We also
2048 * need to bring uptodate any buffers that are only partially
2049 * being written to.
2050 */
2051 if (NInoNonResident(ni)) {
2052 status = ntfs_prepare_pages_for_non_resident_write(
2053 pages, do_pages, pos, bytes);
2054 if (unlikely(status)) {
2055 loff_t i_size;
2056
2057 do {
2058 unlock_page(pages[--do_pages]);
2059 page_cache_release(pages[do_pages]);
2060 } while (do_pages);
2061 /*
2062 * The write preparation may have instantiated
2063 * allocated space outside i_size. Trim this
2064 * off again. We can ignore any errors in this
2065 * case as we will just be waisting a bit of
2066 * allocated space, which is not a disaster.
2067 */
2068 i_size = i_size_read(vi);
2069 if (pos + bytes > i_size)
2070 vmtruncate(vi, i_size);
2071 break;
2072 }
2073 }
2074 u = (pos >> PAGE_CACHE_SHIFT) - pages[0]->index;
2075 if (likely(nr_segs == 1)) {
2076 copied = ntfs_copy_from_user(pages + u, do_pages - u,
2077 ofs, buf, bytes);
2078 buf += copied;
2079 } else
2080 copied = ntfs_copy_from_user_iovec(pages + u,
2081 do_pages - u, ofs, &iov, &iov_ofs,
2082 bytes);
2083 ntfs_flush_dcache_pages(pages + u, do_pages - u);
2084 status = ntfs_commit_pages_after_write(pages, do_pages, pos,
2085 bytes);
2086 if (likely(!status)) {
2087 written += copied;
2088 count -= copied;
2089 pos += copied;
2090 if (unlikely(copied != bytes))
2091 status = -EFAULT;
2092 }
2093 do {
2094 unlock_page(pages[--do_pages]);
2095 mark_page_accessed(pages[do_pages]);
2096 page_cache_release(pages[do_pages]);
2097 } while (do_pages);
2098 if (unlikely(status))
2099 break;
2100 balance_dirty_pages_ratelimited(mapping);
2101 cond_resched();
2102 } while (count);
2103err_out:
2104 *ppos = pos;
2105 if (cached_page)
2106 page_cache_release(cached_page);
2107 /* For now, when the user asks for O_SYNC, we actually give O_DSYNC. */
2108 if (likely(!status)) {
2109 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(vi))) {
2110 if (!mapping->a_ops->writepage || !is_sync_kiocb(iocb))
2111 status = generic_osync_inode(vi, mapping,
2112 OSYNC_METADATA|OSYNC_DATA);
2113 }
2114 }
2115 pagevec_lru_add(&lru_pvec);
2116 ntfs_debug("Done. Returning %s (written 0x%lx, status %li).",
2117 written ? "written" : "status", (unsigned long)written,
2118 (long)status);
2119 return written ? written : status;
2120}
2121
2122/**
2123 * ntfs_file_aio_write_nolock -
2124 */
2125static ssize_t ntfs_file_aio_write_nolock(struct kiocb *iocb,
2126 const struct iovec *iov, unsigned long nr_segs, loff_t *ppos)
2127{
2128 struct file *file = iocb->ki_filp;
2129 struct address_space *mapping = file->f_mapping;
2130 struct inode *inode = mapping->host;
2131 loff_t pos;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002132 size_t count; /* after file limit checks */
2133 ssize_t written, err;
2134
2135 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002136 err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ);
2137 if (err)
2138 return err;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002139 pos = *ppos;
2140 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2141 /* We can write back this queue in page reclaim. */
2142 current->backing_dev_info = mapping->backing_dev_info;
2143 written = 0;
2144 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2145 if (err)
2146 goto out;
2147 if (!count)
2148 goto out;
Josef "Jeff" Sipek20d29372006-12-08 02:36:41 -08002149 err = remove_suid(file->f_path.dentry);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002150 if (err)
2151 goto out;
Christoph Hellwig870f4812006-01-09 20:52:01 -08002152 file_update_time(file);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002153 written = ntfs_file_buffered_write(iocb, iov, nr_segs, pos, ppos,
2154 count);
2155out:
2156 current->backing_dev_info = NULL;
2157 return written ? written : err;
2158}
2159
2160/**
2161 * ntfs_file_aio_write -
2162 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002163static ssize_t ntfs_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2164 unsigned long nr_segs, loff_t pos)
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002165{
2166 struct file *file = iocb->ki_filp;
2167 struct address_space *mapping = file->f_mapping;
2168 struct inode *inode = mapping->host;
2169 ssize_t ret;
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002170
2171 BUG_ON(iocb->ki_pos != pos);
2172
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002173 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002174 ret = ntfs_file_aio_write_nolock(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002175 mutex_unlock(&inode->i_mutex);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002176 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2177 int err = sync_page_range(inode, mapping, pos, ret);
2178 if (err < 0)
2179 ret = err;
2180 }
2181 return ret;
2182}
2183
2184/**
2185 * ntfs_file_writev -
2186 *
2187 * Basically the same as generic_file_writev() except that it ends up calling
2188 * ntfs_file_aio_write_nolock() instead of __generic_file_aio_write_nolock().
2189 */
2190static ssize_t ntfs_file_writev(struct file *file, const struct iovec *iov,
2191 unsigned long nr_segs, loff_t *ppos)
2192{
2193 struct address_space *mapping = file->f_mapping;
2194 struct inode *inode = mapping->host;
2195 struct kiocb kiocb;
2196 ssize_t ret;
2197
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002198 mutex_lock(&inode->i_mutex);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002199 init_sync_kiocb(&kiocb, file);
2200 ret = ntfs_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2201 if (ret == -EIOCBQUEUED)
2202 ret = wait_on_sync_kiocb(&kiocb);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002203 mutex_unlock(&inode->i_mutex);
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002204 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2205 int err = sync_page_range(inode, mapping, *ppos - ret, ret);
2206 if (err < 0)
2207 ret = err;
2208 }
2209 return ret;
2210}
2211
2212/**
2213 * ntfs_file_write - simple wrapper for ntfs_file_writev()
2214 */
2215static ssize_t ntfs_file_write(struct file *file, const char __user *buf,
2216 size_t count, loff_t *ppos)
2217{
2218 struct iovec local_iov = { .iov_base = (void __user *)buf,
2219 .iov_len = count };
2220
2221 return ntfs_file_writev(file, &local_iov, 1, ppos);
2222}
2223
2224/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 * ntfs_file_fsync - sync a file to disk
2226 * @filp: file to be synced
2227 * @dentry: dentry describing the file to sync
2228 * @datasync: if non-zero only flush user data and not metadata
2229 *
2230 * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync
2231 * system calls. This function is inspired by fs/buffer.c::file_fsync().
2232 *
2233 * If @datasync is false, write the mft record and all associated extent mft
2234 * records as well as the $DATA attribute and then sync the block device.
2235 *
2236 * If @datasync is true and the attribute is non-resident, we skip the writing
2237 * of the mft record and all associated extent mft records (this might still
2238 * happen due to the write_inode_now() call).
2239 *
2240 * Also, if @datasync is true, we do not wait on the inode to be written out
2241 * but we always wait on the page cache pages to be written out.
2242 *
2243 * Note: In the past @filp could be NULL so we ignore it as we don't need it
2244 * anyway.
2245 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002246 * Locking: Caller must hold i_mutex on the inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 *
2248 * TODO: We should probably also write all attribute/index inodes associated
2249 * with this inode but since we have no simple way of getting to them we ignore
2250 * this problem for now.
2251 */
2252static int ntfs_file_fsync(struct file *filp, struct dentry *dentry,
2253 int datasync)
2254{
2255 struct inode *vi = dentry->d_inode;
2256 int err, ret = 0;
2257
2258 ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
2259 BUG_ON(S_ISDIR(vi->i_mode));
2260 if (!datasync || !NInoNonResident(NTFS_I(vi)))
2261 ret = ntfs_write_inode(vi, 1);
2262 write_inode_now(vi, !datasync);
Anton Altaparmakovf25dfb52005-09-08 20:35:33 +01002263 /*
2264 * NOTE: If we were to use mapping->private_list (see ext2 and
2265 * fs/buffer.c) for dirty blocks then we could optimize the below to be
2266 * sync_mapping_buffers(vi->i_mapping).
2267 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 err = sync_blockdev(vi->i_sb->s_bdev);
2269 if (unlikely(err && !ret))
2270 ret = err;
2271 if (likely(!ret))
2272 ntfs_debug("Done.");
2273 else
2274 ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx. Error "
2275 "%u.", datasync ? "data" : "", vi->i_ino, -ret);
2276 return ret;
2277}
2278
2279#endif /* NTFS_RW */
2280
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002281const struct file_operations ntfs_file_ops = {
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002282 .llseek = generic_file_llseek, /* Seek inside file. */
Badari Pulavarty543ade12006-09-30 23:28:48 -07002283 .read = do_sync_read, /* Read from file. */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002284 .aio_read = generic_file_aio_read, /* Async read from file. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285#ifdef NTFS_RW
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002286 .write = ntfs_file_write, /* Write to file. */
2287 .aio_write = ntfs_file_aio_write, /* Async write to file. */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002288 /*.release = ,*/ /* Last file is closed. See
2289 fs/ext2/file.c::
2290 ext2_release_file() for
2291 how to use this to discard
2292 preallocated space for
2293 write opened files. */
2294 .fsync = ntfs_file_fsync, /* Sync a file to disk. */
2295 /*.aio_fsync = ,*/ /* Sync all outstanding async
2296 i/o operations on a
2297 kiocb. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298#endif /* NTFS_RW */
Anton Altaparmakov98b27032005-10-11 15:40:40 +01002299 /*.ioctl = ,*/ /* Perform function on the
2300 mounted filesystem. */
2301 .mmap = generic_file_mmap, /* Mmap file. */
2302 .open = ntfs_file_open, /* Open file. */
2303 .sendfile = generic_file_sendfile, /* Zero-copy data send with
2304 the data source being on
2305 the ntfs partition. We do
2306 not need to care about the
2307 data destination. */
2308 /*.sendpage = ,*/ /* Zero-copy data send with
2309 the data destination being
2310 on the ntfs partition. We
2311 do not need to care about
2312 the data source. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313};
2314
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002315const struct inode_operations ntfs_file_inode_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316#ifdef NTFS_RW
2317 .truncate = ntfs_truncate_vfs,
2318 .setattr = ntfs_setattr,
2319#endif /* NTFS_RW */
2320};
2321
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002322const struct file_operations ntfs_empty_file_ops = {};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002324const struct inode_operations ntfs_empty_inode_ops = {};