diff options
Diffstat (limited to 'fs/xfs/linux-2.6')
-rw-r--r-- | fs/xfs/linux-2.6/xfs_file.c | 723 | ||||
-rw-r--r-- | fs/xfs/linux-2.6/xfs_lrw.c | 796 |
2 files changed, 723 insertions, 796 deletions
diff --git a/fs/xfs/linux-2.6/xfs_file.c b/fs/xfs/linux-2.6/xfs_file.c index 3805ada98747..51fc510828a4 100644 --- a/fs/xfs/linux-2.6/xfs_file.c +++ b/fs/xfs/linux-2.6/xfs_file.c | |||
@@ -16,6 +16,7 @@ | |||
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
17 | */ | 17 | */ |
18 | #include "xfs.h" | 18 | #include "xfs.h" |
19 | #include "xfs_fs.h" | ||
19 | #include "xfs_bit.h" | 20 | #include "xfs_bit.h" |
20 | #include "xfs_log.h" | 21 | #include "xfs_log.h" |
21 | #include "xfs_inum.h" | 22 | #include "xfs_inum.h" |
@@ -34,16 +35,738 @@ | |||
34 | #include "xfs_dir2_sf.h" | 35 | #include "xfs_dir2_sf.h" |
35 | #include "xfs_dinode.h" | 36 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | 37 | #include "xfs_inode.h" |
38 | #include "xfs_bmap.h" | ||
37 | #include "xfs_error.h" | 39 | #include "xfs_error.h" |
38 | #include "xfs_rw.h" | 40 | #include "xfs_rw.h" |
39 | #include "xfs_vnodeops.h" | 41 | #include "xfs_vnodeops.h" |
40 | #include "xfs_da_btree.h" | 42 | #include "xfs_da_btree.h" |
41 | #include "xfs_ioctl.h" | 43 | #include "xfs_ioctl.h" |
44 | #include "xfs_trace.h" | ||
42 | 45 | ||
43 | #include <linux/dcache.h> | 46 | #include <linux/dcache.h> |
44 | 47 | ||
45 | static const struct vm_operations_struct xfs_file_vm_ops; | 48 | static const struct vm_operations_struct xfs_file_vm_ops; |
46 | 49 | ||
50 | /* | ||
51 | * xfs_iozero | ||
52 | * | ||
53 | * xfs_iozero clears the specified range of buffer supplied, | ||
54 | * and marks all the affected blocks as valid and modified. If | ||
55 | * an affected block is not allocated, it will be allocated. If | ||
56 | * an affected block is not completely overwritten, and is not | ||
57 | * valid before the operation, it will be read from disk before | ||
58 | * being partially zeroed. | ||
59 | */ | ||
60 | STATIC int | ||
61 | xfs_iozero( | ||
62 | struct xfs_inode *ip, /* inode */ | ||
63 | loff_t pos, /* offset in file */ | ||
64 | size_t count) /* size of data to zero */ | ||
65 | { | ||
66 | struct page *page; | ||
67 | struct address_space *mapping; | ||
68 | int status; | ||
69 | |||
70 | mapping = VFS_I(ip)->i_mapping; | ||
71 | do { | ||
72 | unsigned offset, bytes; | ||
73 | void *fsdata; | ||
74 | |||
75 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | ||
76 | bytes = PAGE_CACHE_SIZE - offset; | ||
77 | if (bytes > count) | ||
78 | bytes = count; | ||
79 | |||
80 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | ||
81 | AOP_FLAG_UNINTERRUPTIBLE, | ||
82 | &page, &fsdata); | ||
83 | if (status) | ||
84 | break; | ||
85 | |||
86 | zero_user(page, offset, bytes); | ||
87 | |||
88 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | ||
89 | page, fsdata); | ||
90 | WARN_ON(status <= 0); /* can't return less than zero! */ | ||
91 | pos += bytes; | ||
92 | count -= bytes; | ||
93 | status = 0; | ||
94 | } while (count); | ||
95 | |||
96 | return (-status); | ||
97 | } | ||
98 | |||
99 | ssize_t /* bytes read, or (-) error */ | ||
100 | xfs_read( | ||
101 | xfs_inode_t *ip, | ||
102 | struct kiocb *iocb, | ||
103 | const struct iovec *iovp, | ||
104 | unsigned int segs, | ||
105 | loff_t *offset, | ||
106 | int ioflags) | ||
107 | { | ||
108 | struct file *file = iocb->ki_filp; | ||
109 | struct inode *inode = file->f_mapping->host; | ||
110 | xfs_mount_t *mp = ip->i_mount; | ||
111 | size_t size = 0; | ||
112 | ssize_t ret = 0; | ||
113 | xfs_fsize_t n; | ||
114 | unsigned long seg; | ||
115 | |||
116 | |||
117 | XFS_STATS_INC(xs_read_calls); | ||
118 | |||
119 | /* START copy & waste from filemap.c */ | ||
120 | for (seg = 0; seg < segs; seg++) { | ||
121 | const struct iovec *iv = &iovp[seg]; | ||
122 | |||
123 | /* | ||
124 | * If any segment has a negative length, or the cumulative | ||
125 | * length ever wraps negative then return -EINVAL. | ||
126 | */ | ||
127 | size += iv->iov_len; | ||
128 | if (unlikely((ssize_t)(size|iv->iov_len) < 0)) | ||
129 | return XFS_ERROR(-EINVAL); | ||
130 | } | ||
131 | /* END copy & waste from filemap.c */ | ||
132 | |||
133 | if (unlikely(ioflags & IO_ISDIRECT)) { | ||
134 | xfs_buftarg_t *target = | ||
135 | XFS_IS_REALTIME_INODE(ip) ? | ||
136 | mp->m_rtdev_targp : mp->m_ddev_targp; | ||
137 | if ((*offset & target->bt_smask) || | ||
138 | (size & target->bt_smask)) { | ||
139 | if (*offset == ip->i_size) { | ||
140 | return (0); | ||
141 | } | ||
142 | return -XFS_ERROR(EINVAL); | ||
143 | } | ||
144 | } | ||
145 | |||
146 | n = XFS_MAXIOFFSET(mp) - *offset; | ||
147 | if ((n <= 0) || (size == 0)) | ||
148 | return 0; | ||
149 | |||
150 | if (n < size) | ||
151 | size = n; | ||
152 | |||
153 | if (XFS_FORCED_SHUTDOWN(mp)) | ||
154 | return -EIO; | ||
155 | |||
156 | if (unlikely(ioflags & IO_ISDIRECT)) | ||
157 | mutex_lock(&inode->i_mutex); | ||
158 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | ||
159 | |||
160 | if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) { | ||
161 | int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags); | ||
162 | int iolock = XFS_IOLOCK_SHARED; | ||
163 | |||
164 | ret = -XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *offset, size, | ||
165 | dmflags, &iolock); | ||
166 | if (ret) { | ||
167 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
168 | if (unlikely(ioflags & IO_ISDIRECT)) | ||
169 | mutex_unlock(&inode->i_mutex); | ||
170 | return ret; | ||
171 | } | ||
172 | } | ||
173 | |||
174 | if (unlikely(ioflags & IO_ISDIRECT)) { | ||
175 | if (inode->i_mapping->nrpages) | ||
176 | ret = -xfs_flushinval_pages(ip, (*offset & PAGE_CACHE_MASK), | ||
177 | -1, FI_REMAPF_LOCKED); | ||
178 | mutex_unlock(&inode->i_mutex); | ||
179 | if (ret) { | ||
180 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
181 | return ret; | ||
182 | } | ||
183 | } | ||
184 | |||
185 | trace_xfs_file_read(ip, size, *offset, ioflags); | ||
186 | |||
187 | iocb->ki_pos = *offset; | ||
188 | ret = generic_file_aio_read(iocb, iovp, segs, *offset); | ||
189 | if (ret > 0) | ||
190 | XFS_STATS_ADD(xs_read_bytes, ret); | ||
191 | |||
192 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
193 | return ret; | ||
194 | } | ||
195 | |||
196 | ssize_t | ||
197 | xfs_splice_read( | ||
198 | xfs_inode_t *ip, | ||
199 | struct file *infilp, | ||
200 | loff_t *ppos, | ||
201 | struct pipe_inode_info *pipe, | ||
202 | size_t count, | ||
203 | int flags, | ||
204 | int ioflags) | ||
205 | { | ||
206 | xfs_mount_t *mp = ip->i_mount; | ||
207 | ssize_t ret; | ||
208 | |||
209 | XFS_STATS_INC(xs_read_calls); | ||
210 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | ||
211 | return -EIO; | ||
212 | |||
213 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | ||
214 | |||
215 | if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) { | ||
216 | int iolock = XFS_IOLOCK_SHARED; | ||
217 | int error; | ||
218 | |||
219 | error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *ppos, count, | ||
220 | FILP_DELAY_FLAG(infilp), &iolock); | ||
221 | if (error) { | ||
222 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
223 | return -error; | ||
224 | } | ||
225 | } | ||
226 | |||
227 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); | ||
228 | |||
229 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | ||
230 | if (ret > 0) | ||
231 | XFS_STATS_ADD(xs_read_bytes, ret); | ||
232 | |||
233 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
234 | return ret; | ||
235 | } | ||
236 | |||
237 | ssize_t | ||
238 | xfs_splice_write( | ||
239 | xfs_inode_t *ip, | ||
240 | struct pipe_inode_info *pipe, | ||
241 | struct file *outfilp, | ||
242 | loff_t *ppos, | ||
243 | size_t count, | ||
244 | int flags, | ||
245 | int ioflags) | ||
246 | { | ||
247 | xfs_mount_t *mp = ip->i_mount; | ||
248 | ssize_t ret; | ||
249 | struct inode *inode = outfilp->f_mapping->host; | ||
250 | xfs_fsize_t isize, new_size; | ||
251 | |||
252 | XFS_STATS_INC(xs_write_calls); | ||
253 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | ||
254 | return -EIO; | ||
255 | |||
256 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | ||
257 | |||
258 | if (DM_EVENT_ENABLED(ip, DM_EVENT_WRITE) && !(ioflags & IO_INVIS)) { | ||
259 | int iolock = XFS_IOLOCK_EXCL; | ||
260 | int error; | ||
261 | |||
262 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, *ppos, count, | ||
263 | FILP_DELAY_FLAG(outfilp), &iolock); | ||
264 | if (error) { | ||
265 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | ||
266 | return -error; | ||
267 | } | ||
268 | } | ||
269 | |||
270 | new_size = *ppos + count; | ||
271 | |||
272 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
273 | if (new_size > ip->i_size) | ||
274 | ip->i_new_size = new_size; | ||
275 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
276 | |||
277 | trace_xfs_file_splice_write(ip, count, *ppos, ioflags); | ||
278 | |||
279 | ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags); | ||
280 | if (ret > 0) | ||
281 | XFS_STATS_ADD(xs_write_bytes, ret); | ||
282 | |||
283 | isize = i_size_read(inode); | ||
284 | if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize)) | ||
285 | *ppos = isize; | ||
286 | |||
287 | if (*ppos > ip->i_size) { | ||
288 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
289 | if (*ppos > ip->i_size) | ||
290 | ip->i_size = *ppos; | ||
291 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
292 | } | ||
293 | |||
294 | if (ip->i_new_size) { | ||
295 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
296 | ip->i_new_size = 0; | ||
297 | if (ip->i_d.di_size > ip->i_size) | ||
298 | ip->i_d.di_size = ip->i_size; | ||
299 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
300 | } | ||
301 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | ||
302 | return ret; | ||
303 | } | ||
304 | |||
305 | /* | ||
306 | * This routine is called to handle zeroing any space in the last | ||
307 | * block of the file that is beyond the EOF. We do this since the | ||
308 | * size is being increased without writing anything to that block | ||
309 | * and we don't want anyone to read the garbage on the disk. | ||
310 | */ | ||
311 | STATIC int /* error (positive) */ | ||
312 | xfs_zero_last_block( | ||
313 | xfs_inode_t *ip, | ||
314 | xfs_fsize_t offset, | ||
315 | xfs_fsize_t isize) | ||
316 | { | ||
317 | xfs_fileoff_t last_fsb; | ||
318 | xfs_mount_t *mp = ip->i_mount; | ||
319 | int nimaps; | ||
320 | int zero_offset; | ||
321 | int zero_len; | ||
322 | int error = 0; | ||
323 | xfs_bmbt_irec_t imap; | ||
324 | |||
325 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | ||
326 | |||
327 | zero_offset = XFS_B_FSB_OFFSET(mp, isize); | ||
328 | if (zero_offset == 0) { | ||
329 | /* | ||
330 | * There are no extra bytes in the last block on disk to | ||
331 | * zero, so return. | ||
332 | */ | ||
333 | return 0; | ||
334 | } | ||
335 | |||
336 | last_fsb = XFS_B_TO_FSBT(mp, isize); | ||
337 | nimaps = 1; | ||
338 | error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap, | ||
339 | &nimaps, NULL, NULL); | ||
340 | if (error) { | ||
341 | return error; | ||
342 | } | ||
343 | ASSERT(nimaps > 0); | ||
344 | /* | ||
345 | * If the block underlying isize is just a hole, then there | ||
346 | * is nothing to zero. | ||
347 | */ | ||
348 | if (imap.br_startblock == HOLESTARTBLOCK) { | ||
349 | return 0; | ||
350 | } | ||
351 | /* | ||
352 | * Zero the part of the last block beyond the EOF, and write it | ||
353 | * out sync. We need to drop the ilock while we do this so we | ||
354 | * don't deadlock when the buffer cache calls back to us. | ||
355 | */ | ||
356 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
357 | |||
358 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | ||
359 | if (isize + zero_len > offset) | ||
360 | zero_len = offset - isize; | ||
361 | error = xfs_iozero(ip, isize, zero_len); | ||
362 | |||
363 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
364 | ASSERT(error >= 0); | ||
365 | return error; | ||
366 | } | ||
367 | |||
368 | /* | ||
369 | * Zero any on disk space between the current EOF and the new, | ||
370 | * larger EOF. This handles the normal case of zeroing the remainder | ||
371 | * of the last block in the file and the unusual case of zeroing blocks | ||
372 | * out beyond the size of the file. This second case only happens | ||
373 | * with fixed size extents and when the system crashes before the inode | ||
374 | * size was updated but after blocks were allocated. If fill is set, | ||
375 | * then any holes in the range are filled and zeroed. If not, the holes | ||
376 | * are left alone as holes. | ||
377 | */ | ||
378 | |||
379 | int /* error (positive) */ | ||
380 | xfs_zero_eof( | ||
381 | xfs_inode_t *ip, | ||
382 | xfs_off_t offset, /* starting I/O offset */ | ||
383 | xfs_fsize_t isize) /* current inode size */ | ||
384 | { | ||
385 | xfs_mount_t *mp = ip->i_mount; | ||
386 | xfs_fileoff_t start_zero_fsb; | ||
387 | xfs_fileoff_t end_zero_fsb; | ||
388 | xfs_fileoff_t zero_count_fsb; | ||
389 | xfs_fileoff_t last_fsb; | ||
390 | xfs_fileoff_t zero_off; | ||
391 | xfs_fsize_t zero_len; | ||
392 | int nimaps; | ||
393 | int error = 0; | ||
394 | xfs_bmbt_irec_t imap; | ||
395 | |||
396 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | ||
397 | ASSERT(offset > isize); | ||
398 | |||
399 | /* | ||
400 | * First handle zeroing the block on which isize resides. | ||
401 | * We only zero a part of that block so it is handled specially. | ||
402 | */ | ||
403 | error = xfs_zero_last_block(ip, offset, isize); | ||
404 | if (error) { | ||
405 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | ||
406 | return error; | ||
407 | } | ||
408 | |||
409 | /* | ||
410 | * Calculate the range between the new size and the old | ||
411 | * where blocks needing to be zeroed may exist. To get the | ||
412 | * block where the last byte in the file currently resides, | ||
413 | * we need to subtract one from the size and truncate back | ||
414 | * to a block boundary. We subtract 1 in case the size is | ||
415 | * exactly on a block boundary. | ||
416 | */ | ||
417 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | ||
418 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | ||
419 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | ||
420 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | ||
421 | if (last_fsb == end_zero_fsb) { | ||
422 | /* | ||
423 | * The size was only incremented on its last block. | ||
424 | * We took care of that above, so just return. | ||
425 | */ | ||
426 | return 0; | ||
427 | } | ||
428 | |||
429 | ASSERT(start_zero_fsb <= end_zero_fsb); | ||
430 | while (start_zero_fsb <= end_zero_fsb) { | ||
431 | nimaps = 1; | ||
432 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | ||
433 | error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb, | ||
434 | 0, NULL, 0, &imap, &nimaps, NULL, NULL); | ||
435 | if (error) { | ||
436 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | ||
437 | return error; | ||
438 | } | ||
439 | ASSERT(nimaps > 0); | ||
440 | |||
441 | if (imap.br_state == XFS_EXT_UNWRITTEN || | ||
442 | imap.br_startblock == HOLESTARTBLOCK) { | ||
443 | /* | ||
444 | * This loop handles initializing pages that were | ||
445 | * partially initialized by the code below this | ||
446 | * loop. It basically zeroes the part of the page | ||
447 | * that sits on a hole and sets the page as P_HOLE | ||
448 | * and calls remapf if it is a mapped file. | ||
449 | */ | ||
450 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | ||
451 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | ||
452 | continue; | ||
453 | } | ||
454 | |||
455 | /* | ||
456 | * There are blocks we need to zero. | ||
457 | * Drop the inode lock while we're doing the I/O. | ||
458 | * We'll still have the iolock to protect us. | ||
459 | */ | ||
460 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
461 | |||
462 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); | ||
463 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | ||
464 | |||
465 | if ((zero_off + zero_len) > offset) | ||
466 | zero_len = offset - zero_off; | ||
467 | |||
468 | error = xfs_iozero(ip, zero_off, zero_len); | ||
469 | if (error) { | ||
470 | goto out_lock; | ||
471 | } | ||
472 | |||
473 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | ||
474 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | ||
475 | |||
476 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
477 | } | ||
478 | |||
479 | return 0; | ||
480 | |||
481 | out_lock: | ||
482 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
483 | ASSERT(error >= 0); | ||
484 | return error; | ||
485 | } | ||
486 | |||
487 | ssize_t /* bytes written, or (-) error */ | ||
488 | xfs_write( | ||
489 | struct xfs_inode *xip, | ||
490 | struct kiocb *iocb, | ||
491 | const struct iovec *iovp, | ||
492 | unsigned int nsegs, | ||
493 | loff_t *offset, | ||
494 | int ioflags) | ||
495 | { | ||
496 | struct file *file = iocb->ki_filp; | ||
497 | struct address_space *mapping = file->f_mapping; | ||
498 | struct inode *inode = mapping->host; | ||
499 | unsigned long segs = nsegs; | ||
500 | xfs_mount_t *mp; | ||
501 | ssize_t ret = 0, error = 0; | ||
502 | xfs_fsize_t isize, new_size; | ||
503 | int iolock; | ||
504 | int eventsent = 0; | ||
505 | size_t ocount = 0, count; | ||
506 | loff_t pos; | ||
507 | int need_i_mutex; | ||
508 | |||
509 | XFS_STATS_INC(xs_write_calls); | ||
510 | |||
511 | error = generic_segment_checks(iovp, &segs, &ocount, VERIFY_READ); | ||
512 | if (error) | ||
513 | return error; | ||
514 | |||
515 | count = ocount; | ||
516 | pos = *offset; | ||
517 | |||
518 | if (count == 0) | ||
519 | return 0; | ||
520 | |||
521 | mp = xip->i_mount; | ||
522 | |||
523 | xfs_wait_for_freeze(mp, SB_FREEZE_WRITE); | ||
524 | |||
525 | if (XFS_FORCED_SHUTDOWN(mp)) | ||
526 | return -EIO; | ||
527 | |||
528 | relock: | ||
529 | if (ioflags & IO_ISDIRECT) { | ||
530 | iolock = XFS_IOLOCK_SHARED; | ||
531 | need_i_mutex = 0; | ||
532 | } else { | ||
533 | iolock = XFS_IOLOCK_EXCL; | ||
534 | need_i_mutex = 1; | ||
535 | mutex_lock(&inode->i_mutex); | ||
536 | } | ||
537 | |||
538 | xfs_ilock(xip, XFS_ILOCK_EXCL|iolock); | ||
539 | |||
540 | start: | ||
541 | error = -generic_write_checks(file, &pos, &count, | ||
542 | S_ISBLK(inode->i_mode)); | ||
543 | if (error) { | ||
544 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | ||
545 | goto out_unlock_mutex; | ||
546 | } | ||
547 | |||
548 | if ((DM_EVENT_ENABLED(xip, DM_EVENT_WRITE) && | ||
549 | !(ioflags & IO_INVIS) && !eventsent)) { | ||
550 | int dmflags = FILP_DELAY_FLAG(file); | ||
551 | |||
552 | if (need_i_mutex) | ||
553 | dmflags |= DM_FLAGS_IMUX; | ||
554 | |||
555 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
556 | error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, xip, | ||
557 | pos, count, dmflags, &iolock); | ||
558 | if (error) { | ||
559 | goto out_unlock_internal; | ||
560 | } | ||
561 | xfs_ilock(xip, XFS_ILOCK_EXCL); | ||
562 | eventsent = 1; | ||
563 | |||
564 | /* | ||
565 | * The iolock was dropped and reacquired in XFS_SEND_DATA | ||
566 | * so we have to recheck the size when appending. | ||
567 | * We will only "goto start;" once, since having sent the | ||
568 | * event prevents another call to XFS_SEND_DATA, which is | ||
569 | * what allows the size to change in the first place. | ||
570 | */ | ||
571 | if ((file->f_flags & O_APPEND) && pos != xip->i_size) | ||
572 | goto start; | ||
573 | } | ||
574 | |||
575 | if (ioflags & IO_ISDIRECT) { | ||
576 | xfs_buftarg_t *target = | ||
577 | XFS_IS_REALTIME_INODE(xip) ? | ||
578 | mp->m_rtdev_targp : mp->m_ddev_targp; | ||
579 | |||
580 | if ((pos & target->bt_smask) || (count & target->bt_smask)) { | ||
581 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | ||
582 | return XFS_ERROR(-EINVAL); | ||
583 | } | ||
584 | |||
585 | if (!need_i_mutex && (mapping->nrpages || pos > xip->i_size)) { | ||
586 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | ||
587 | iolock = XFS_IOLOCK_EXCL; | ||
588 | need_i_mutex = 1; | ||
589 | mutex_lock(&inode->i_mutex); | ||
590 | xfs_ilock(xip, XFS_ILOCK_EXCL|iolock); | ||
591 | goto start; | ||
592 | } | ||
593 | } | ||
594 | |||
595 | new_size = pos + count; | ||
596 | if (new_size > xip->i_size) | ||
597 | xip->i_new_size = new_size; | ||
598 | |||
599 | if (likely(!(ioflags & IO_INVIS))) | ||
600 | file_update_time(file); | ||
601 | |||
602 | /* | ||
603 | * If the offset is beyond the size of the file, we have a couple | ||
604 | * of things to do. First, if there is already space allocated | ||
605 | * we need to either create holes or zero the disk or ... | ||
606 | * | ||
607 | * If there is a page where the previous size lands, we need | ||
608 | * to zero it out up to the new size. | ||
609 | */ | ||
610 | |||
611 | if (pos > xip->i_size) { | ||
612 | error = xfs_zero_eof(xip, pos, xip->i_size); | ||
613 | if (error) { | ||
614 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
615 | goto out_unlock_internal; | ||
616 | } | ||
617 | } | ||
618 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
619 | |||
620 | /* | ||
621 | * If we're writing the file then make sure to clear the | ||
622 | * setuid and setgid bits if the process is not being run | ||
623 | * by root. This keeps people from modifying setuid and | ||
624 | * setgid binaries. | ||
625 | */ | ||
626 | error = -file_remove_suid(file); | ||
627 | if (unlikely(error)) | ||
628 | goto out_unlock_internal; | ||
629 | |||
630 | /* We can write back this queue in page reclaim */ | ||
631 | current->backing_dev_info = mapping->backing_dev_info; | ||
632 | |||
633 | if ((ioflags & IO_ISDIRECT)) { | ||
634 | if (mapping->nrpages) { | ||
635 | WARN_ON(need_i_mutex == 0); | ||
636 | error = xfs_flushinval_pages(xip, | ||
637 | (pos & PAGE_CACHE_MASK), | ||
638 | -1, FI_REMAPF_LOCKED); | ||
639 | if (error) | ||
640 | goto out_unlock_internal; | ||
641 | } | ||
642 | |||
643 | if (need_i_mutex) { | ||
644 | /* demote the lock now the cached pages are gone */ | ||
645 | xfs_ilock_demote(xip, XFS_IOLOCK_EXCL); | ||
646 | mutex_unlock(&inode->i_mutex); | ||
647 | |||
648 | iolock = XFS_IOLOCK_SHARED; | ||
649 | need_i_mutex = 0; | ||
650 | } | ||
651 | |||
652 | trace_xfs_file_direct_write(xip, count, *offset, ioflags); | ||
653 | ret = generic_file_direct_write(iocb, iovp, | ||
654 | &segs, pos, offset, count, ocount); | ||
655 | |||
656 | /* | ||
657 | * direct-io write to a hole: fall through to buffered I/O | ||
658 | * for completing the rest of the request. | ||
659 | */ | ||
660 | if (ret >= 0 && ret != count) { | ||
661 | XFS_STATS_ADD(xs_write_bytes, ret); | ||
662 | |||
663 | pos += ret; | ||
664 | count -= ret; | ||
665 | |||
666 | ioflags &= ~IO_ISDIRECT; | ||
667 | xfs_iunlock(xip, iolock); | ||
668 | goto relock; | ||
669 | } | ||
670 | } else { | ||
671 | int enospc = 0; | ||
672 | ssize_t ret2 = 0; | ||
673 | |||
674 | write_retry: | ||
675 | trace_xfs_file_buffered_write(xip, count, *offset, ioflags); | ||
676 | ret2 = generic_file_buffered_write(iocb, iovp, segs, | ||
677 | pos, offset, count, ret); | ||
678 | /* | ||
679 | * if we just got an ENOSPC, flush the inode now we | ||
680 | * aren't holding any page locks and retry *once* | ||
681 | */ | ||
682 | if (ret2 == -ENOSPC && !enospc) { | ||
683 | error = xfs_flush_pages(xip, 0, -1, 0, FI_NONE); | ||
684 | if (error) | ||
685 | goto out_unlock_internal; | ||
686 | enospc = 1; | ||
687 | goto write_retry; | ||
688 | } | ||
689 | ret = ret2; | ||
690 | } | ||
691 | |||
692 | current->backing_dev_info = NULL; | ||
693 | |||
694 | isize = i_size_read(inode); | ||
695 | if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize)) | ||
696 | *offset = isize; | ||
697 | |||
698 | if (*offset > xip->i_size) { | ||
699 | xfs_ilock(xip, XFS_ILOCK_EXCL); | ||
700 | if (*offset > xip->i_size) | ||
701 | xip->i_size = *offset; | ||
702 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
703 | } | ||
704 | |||
705 | if (ret == -ENOSPC && | ||
706 | DM_EVENT_ENABLED(xip, DM_EVENT_NOSPACE) && !(ioflags & IO_INVIS)) { | ||
707 | xfs_iunlock(xip, iolock); | ||
708 | if (need_i_mutex) | ||
709 | mutex_unlock(&inode->i_mutex); | ||
710 | error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, xip, | ||
711 | DM_RIGHT_NULL, xip, DM_RIGHT_NULL, NULL, NULL, | ||
712 | 0, 0, 0); /* Delay flag intentionally unused */ | ||
713 | if (need_i_mutex) | ||
714 | mutex_lock(&inode->i_mutex); | ||
715 | xfs_ilock(xip, iolock); | ||
716 | if (error) | ||
717 | goto out_unlock_internal; | ||
718 | goto start; | ||
719 | } | ||
720 | |||
721 | error = -ret; | ||
722 | if (ret <= 0) | ||
723 | goto out_unlock_internal; | ||
724 | |||
725 | XFS_STATS_ADD(xs_write_bytes, ret); | ||
726 | |||
727 | /* Handle various SYNC-type writes */ | ||
728 | if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) { | ||
729 | loff_t end = pos + ret - 1; | ||
730 | int error2; | ||
731 | |||
732 | xfs_iunlock(xip, iolock); | ||
733 | if (need_i_mutex) | ||
734 | mutex_unlock(&inode->i_mutex); | ||
735 | |||
736 | error2 = filemap_write_and_wait_range(mapping, pos, end); | ||
737 | if (!error) | ||
738 | error = error2; | ||
739 | if (need_i_mutex) | ||
740 | mutex_lock(&inode->i_mutex); | ||
741 | xfs_ilock(xip, iolock); | ||
742 | |||
743 | error2 = xfs_fsync(xip); | ||
744 | if (!error) | ||
745 | error = error2; | ||
746 | } | ||
747 | |||
748 | out_unlock_internal: | ||
749 | if (xip->i_new_size) { | ||
750 | xfs_ilock(xip, XFS_ILOCK_EXCL); | ||
751 | xip->i_new_size = 0; | ||
752 | /* | ||
753 | * If this was a direct or synchronous I/O that failed (such | ||
754 | * as ENOSPC) then part of the I/O may have been written to | ||
755 | * disk before the error occured. In this case the on-disk | ||
756 | * file size may have been adjusted beyond the in-memory file | ||
757 | * size and now needs to be truncated back. | ||
758 | */ | ||
759 | if (xip->i_d.di_size > xip->i_size) | ||
760 | xip->i_d.di_size = xip->i_size; | ||
761 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
762 | } | ||
763 | xfs_iunlock(xip, iolock); | ||
764 | out_unlock_mutex: | ||
765 | if (need_i_mutex) | ||
766 | mutex_unlock(&inode->i_mutex); | ||
767 | return -error; | ||
768 | } | ||
769 | |||
47 | STATIC ssize_t | 770 | STATIC ssize_t |
48 | xfs_file_aio_read( | 771 | xfs_file_aio_read( |
49 | struct kiocb *iocb, | 772 | struct kiocb *iocb, |
diff --git a/fs/xfs/linux-2.6/xfs_lrw.c b/fs/xfs/linux-2.6/xfs_lrw.c deleted file mode 100644 index eac6f80d786d..000000000000 --- a/fs/xfs/linux-2.6/xfs_lrw.c +++ /dev/null | |||
@@ -1,796 +0,0 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. | ||
3 | * All Rights Reserved. | ||
4 | * | ||
5 | * This program is free software; you can redistribute it and/or | ||
6 | * modify it under the terms of the GNU General Public License as | ||
7 | * published by the Free Software Foundation. | ||
8 | * | ||
9 | * This program is distributed in the hope that it would be useful, | ||
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
12 | * GNU General Public License for more details. | ||
13 | * | ||
14 | * You should have received a copy of the GNU General Public License | ||
15 | * along with this program; if not, write the Free Software Foundation, | ||
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | ||
17 | */ | ||
18 | #include "xfs.h" | ||
19 | #include "xfs_fs.h" | ||
20 | #include "xfs_bit.h" | ||
21 | #include "xfs_log.h" | ||
22 | #include "xfs_inum.h" | ||
23 | #include "xfs_trans.h" | ||
24 | #include "xfs_sb.h" | ||
25 | #include "xfs_ag.h" | ||
26 | #include "xfs_dir2.h" | ||
27 | #include "xfs_alloc.h" | ||
28 | #include "xfs_dmapi.h" | ||
29 | #include "xfs_quota.h" | ||
30 | #include "xfs_mount.h" | ||
31 | #include "xfs_bmap_btree.h" | ||
32 | #include "xfs_alloc_btree.h" | ||
33 | #include "xfs_ialloc_btree.h" | ||
34 | #include "xfs_dir2_sf.h" | ||
35 | #include "xfs_attr_sf.h" | ||
36 | #include "xfs_dinode.h" | ||
37 | #include "xfs_inode.h" | ||
38 | #include "xfs_bmap.h" | ||
39 | #include "xfs_btree.h" | ||
40 | #include "xfs_ialloc.h" | ||
41 | #include "xfs_rtalloc.h" | ||
42 | #include "xfs_error.h" | ||
43 | #include "xfs_itable.h" | ||
44 | #include "xfs_rw.h" | ||
45 | #include "xfs_attr.h" | ||
46 | #include "xfs_inode_item.h" | ||
47 | #include "xfs_buf_item.h" | ||
48 | #include "xfs_utils.h" | ||
49 | #include "xfs_iomap.h" | ||
50 | #include "xfs_vnodeops.h" | ||
51 | #include "xfs_trace.h" | ||
52 | |||
53 | #include <linux/capability.h> | ||
54 | #include <linux/writeback.h> | ||
55 | |||
56 | |||
57 | /* | ||
58 | * xfs_iozero | ||
59 | * | ||
60 | * xfs_iozero clears the specified range of buffer supplied, | ||
61 | * and marks all the affected blocks as valid and modified. If | ||
62 | * an affected block is not allocated, it will be allocated. If | ||
63 | * an affected block is not completely overwritten, and is not | ||
64 | * valid before the operation, it will be read from disk before | ||
65 | * being partially zeroed. | ||
66 | */ | ||
67 | STATIC int | ||
68 | xfs_iozero( | ||
69 | struct xfs_inode *ip, /* inode */ | ||
70 | loff_t pos, /* offset in file */ | ||
71 | size_t count) /* size of data to zero */ | ||
72 | { | ||
73 | struct page *page; | ||
74 | struct address_space *mapping; | ||
75 | int status; | ||
76 | |||
77 | mapping = VFS_I(ip)->i_mapping; | ||
78 | do { | ||
79 | unsigned offset, bytes; | ||
80 | void *fsdata; | ||
81 | |||
82 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | ||
83 | bytes = PAGE_CACHE_SIZE - offset; | ||
84 | if (bytes > count) | ||
85 | bytes = count; | ||
86 | |||
87 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | ||
88 | AOP_FLAG_UNINTERRUPTIBLE, | ||
89 | &page, &fsdata); | ||
90 | if (status) | ||
91 | break; | ||
92 | |||
93 | zero_user(page, offset, bytes); | ||
94 | |||
95 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | ||
96 | page, fsdata); | ||
97 | WARN_ON(status <= 0); /* can't return less than zero! */ | ||
98 | pos += bytes; | ||
99 | count -= bytes; | ||
100 | status = 0; | ||
101 | } while (count); | ||
102 | |||
103 | return (-status); | ||
104 | } | ||
105 | |||
106 | ssize_t /* bytes read, or (-) error */ | ||
107 | xfs_read( | ||
108 | xfs_inode_t *ip, | ||
109 | struct kiocb *iocb, | ||
110 | const struct iovec *iovp, | ||
111 | unsigned int segs, | ||
112 | loff_t *offset, | ||
113 | int ioflags) | ||
114 | { | ||
115 | struct file *file = iocb->ki_filp; | ||
116 | struct inode *inode = file->f_mapping->host; | ||
117 | xfs_mount_t *mp = ip->i_mount; | ||
118 | size_t size = 0; | ||
119 | ssize_t ret = 0; | ||
120 | xfs_fsize_t n; | ||
121 | unsigned long seg; | ||
122 | |||
123 | |||
124 | XFS_STATS_INC(xs_read_calls); | ||
125 | |||
126 | /* START copy & waste from filemap.c */ | ||
127 | for (seg = 0; seg < segs; seg++) { | ||
128 | const struct iovec *iv = &iovp[seg]; | ||
129 | |||
130 | /* | ||
131 | * If any segment has a negative length, or the cumulative | ||
132 | * length ever wraps negative then return -EINVAL. | ||
133 | */ | ||
134 | size += iv->iov_len; | ||
135 | if (unlikely((ssize_t)(size|iv->iov_len) < 0)) | ||
136 | return XFS_ERROR(-EINVAL); | ||
137 | } | ||
138 | /* END copy & waste from filemap.c */ | ||
139 | |||
140 | if (unlikely(ioflags & IO_ISDIRECT)) { | ||
141 | xfs_buftarg_t *target = | ||
142 | XFS_IS_REALTIME_INODE(ip) ? | ||
143 | mp->m_rtdev_targp : mp->m_ddev_targp; | ||
144 | if ((*offset & target->bt_smask) || | ||
145 | (size & target->bt_smask)) { | ||
146 | if (*offset == ip->i_size) { | ||
147 | return (0); | ||
148 | } | ||
149 | return -XFS_ERROR(EINVAL); | ||
150 | } | ||
151 | } | ||
152 | |||
153 | n = XFS_MAXIOFFSET(mp) - *offset; | ||
154 | if ((n <= 0) || (size == 0)) | ||
155 | return 0; | ||
156 | |||
157 | if (n < size) | ||
158 | size = n; | ||
159 | |||
160 | if (XFS_FORCED_SHUTDOWN(mp)) | ||
161 | return -EIO; | ||
162 | |||
163 | if (unlikely(ioflags & IO_ISDIRECT)) | ||
164 | mutex_lock(&inode->i_mutex); | ||
165 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | ||
166 | |||
167 | if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) { | ||
168 | int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags); | ||
169 | int iolock = XFS_IOLOCK_SHARED; | ||
170 | |||
171 | ret = -XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *offset, size, | ||
172 | dmflags, &iolock); | ||
173 | if (ret) { | ||
174 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
175 | if (unlikely(ioflags & IO_ISDIRECT)) | ||
176 | mutex_unlock(&inode->i_mutex); | ||
177 | return ret; | ||
178 | } | ||
179 | } | ||
180 | |||
181 | if (unlikely(ioflags & IO_ISDIRECT)) { | ||
182 | if (inode->i_mapping->nrpages) | ||
183 | ret = -xfs_flushinval_pages(ip, (*offset & PAGE_CACHE_MASK), | ||
184 | -1, FI_REMAPF_LOCKED); | ||
185 | mutex_unlock(&inode->i_mutex); | ||
186 | if (ret) { | ||
187 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
188 | return ret; | ||
189 | } | ||
190 | } | ||
191 | |||
192 | trace_xfs_file_read(ip, size, *offset, ioflags); | ||
193 | |||
194 | iocb->ki_pos = *offset; | ||
195 | ret = generic_file_aio_read(iocb, iovp, segs, *offset); | ||
196 | if (ret > 0) | ||
197 | XFS_STATS_ADD(xs_read_bytes, ret); | ||
198 | |||
199 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
200 | return ret; | ||
201 | } | ||
202 | |||
203 | ssize_t | ||
204 | xfs_splice_read( | ||
205 | xfs_inode_t *ip, | ||
206 | struct file *infilp, | ||
207 | loff_t *ppos, | ||
208 | struct pipe_inode_info *pipe, | ||
209 | size_t count, | ||
210 | int flags, | ||
211 | int ioflags) | ||
212 | { | ||
213 | xfs_mount_t *mp = ip->i_mount; | ||
214 | ssize_t ret; | ||
215 | |||
216 | XFS_STATS_INC(xs_read_calls); | ||
217 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | ||
218 | return -EIO; | ||
219 | |||
220 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | ||
221 | |||
222 | if (DM_EVENT_ENABLED(ip, DM_EVENT_READ) && !(ioflags & IO_INVIS)) { | ||
223 | int iolock = XFS_IOLOCK_SHARED; | ||
224 | int error; | ||
225 | |||
226 | error = XFS_SEND_DATA(mp, DM_EVENT_READ, ip, *ppos, count, | ||
227 | FILP_DELAY_FLAG(infilp), &iolock); | ||
228 | if (error) { | ||
229 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
230 | return -error; | ||
231 | } | ||
232 | } | ||
233 | |||
234 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); | ||
235 | |||
236 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | ||
237 | if (ret > 0) | ||
238 | XFS_STATS_ADD(xs_read_bytes, ret); | ||
239 | |||
240 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | ||
241 | return ret; | ||
242 | } | ||
243 | |||
244 | ssize_t | ||
245 | xfs_splice_write( | ||
246 | xfs_inode_t *ip, | ||
247 | struct pipe_inode_info *pipe, | ||
248 | struct file *outfilp, | ||
249 | loff_t *ppos, | ||
250 | size_t count, | ||
251 | int flags, | ||
252 | int ioflags) | ||
253 | { | ||
254 | xfs_mount_t *mp = ip->i_mount; | ||
255 | ssize_t ret; | ||
256 | struct inode *inode = outfilp->f_mapping->host; | ||
257 | xfs_fsize_t isize, new_size; | ||
258 | |||
259 | XFS_STATS_INC(xs_write_calls); | ||
260 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | ||
261 | return -EIO; | ||
262 | |||
263 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | ||
264 | |||
265 | if (DM_EVENT_ENABLED(ip, DM_EVENT_WRITE) && !(ioflags & IO_INVIS)) { | ||
266 | int iolock = XFS_IOLOCK_EXCL; | ||
267 | int error; | ||
268 | |||
269 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, *ppos, count, | ||
270 | FILP_DELAY_FLAG(outfilp), &iolock); | ||
271 | if (error) { | ||
272 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | ||
273 | return -error; | ||
274 | } | ||
275 | } | ||
276 | |||
277 | new_size = *ppos + count; | ||
278 | |||
279 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
280 | if (new_size > ip->i_size) | ||
281 | ip->i_new_size = new_size; | ||
282 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
283 | |||
284 | trace_xfs_file_splice_write(ip, count, *ppos, ioflags); | ||
285 | |||
286 | ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags); | ||
287 | if (ret > 0) | ||
288 | XFS_STATS_ADD(xs_write_bytes, ret); | ||
289 | |||
290 | isize = i_size_read(inode); | ||
291 | if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize)) | ||
292 | *ppos = isize; | ||
293 | |||
294 | if (*ppos > ip->i_size) { | ||
295 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
296 | if (*ppos > ip->i_size) | ||
297 | ip->i_size = *ppos; | ||
298 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
299 | } | ||
300 | |||
301 | if (ip->i_new_size) { | ||
302 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
303 | ip->i_new_size = 0; | ||
304 | if (ip->i_d.di_size > ip->i_size) | ||
305 | ip->i_d.di_size = ip->i_size; | ||
306 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
307 | } | ||
308 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | ||
309 | return ret; | ||
310 | } | ||
311 | |||
312 | /* | ||
313 | * This routine is called to handle zeroing any space in the last | ||
314 | * block of the file that is beyond the EOF. We do this since the | ||
315 | * size is being increased without writing anything to that block | ||
316 | * and we don't want anyone to read the garbage on the disk. | ||
317 | */ | ||
318 | STATIC int /* error (positive) */ | ||
319 | xfs_zero_last_block( | ||
320 | xfs_inode_t *ip, | ||
321 | xfs_fsize_t offset, | ||
322 | xfs_fsize_t isize) | ||
323 | { | ||
324 | xfs_fileoff_t last_fsb; | ||
325 | xfs_mount_t *mp = ip->i_mount; | ||
326 | int nimaps; | ||
327 | int zero_offset; | ||
328 | int zero_len; | ||
329 | int error = 0; | ||
330 | xfs_bmbt_irec_t imap; | ||
331 | |||
332 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | ||
333 | |||
334 | zero_offset = XFS_B_FSB_OFFSET(mp, isize); | ||
335 | if (zero_offset == 0) { | ||
336 | /* | ||
337 | * There are no extra bytes in the last block on disk to | ||
338 | * zero, so return. | ||
339 | */ | ||
340 | return 0; | ||
341 | } | ||
342 | |||
343 | last_fsb = XFS_B_TO_FSBT(mp, isize); | ||
344 | nimaps = 1; | ||
345 | error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap, | ||
346 | &nimaps, NULL, NULL); | ||
347 | if (error) { | ||
348 | return error; | ||
349 | } | ||
350 | ASSERT(nimaps > 0); | ||
351 | /* | ||
352 | * If the block underlying isize is just a hole, then there | ||
353 | * is nothing to zero. | ||
354 | */ | ||
355 | if (imap.br_startblock == HOLESTARTBLOCK) { | ||
356 | return 0; | ||
357 | } | ||
358 | /* | ||
359 | * Zero the part of the last block beyond the EOF, and write it | ||
360 | * out sync. We need to drop the ilock while we do this so we | ||
361 | * don't deadlock when the buffer cache calls back to us. | ||
362 | */ | ||
363 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
364 | |||
365 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | ||
366 | if (isize + zero_len > offset) | ||
367 | zero_len = offset - isize; | ||
368 | error = xfs_iozero(ip, isize, zero_len); | ||
369 | |||
370 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
371 | ASSERT(error >= 0); | ||
372 | return error; | ||
373 | } | ||
374 | |||
375 | /* | ||
376 | * Zero any on disk space between the current EOF and the new, | ||
377 | * larger EOF. This handles the normal case of zeroing the remainder | ||
378 | * of the last block in the file and the unusual case of zeroing blocks | ||
379 | * out beyond the size of the file. This second case only happens | ||
380 | * with fixed size extents and when the system crashes before the inode | ||
381 | * size was updated but after blocks were allocated. If fill is set, | ||
382 | * then any holes in the range are filled and zeroed. If not, the holes | ||
383 | * are left alone as holes. | ||
384 | */ | ||
385 | |||
386 | int /* error (positive) */ | ||
387 | xfs_zero_eof( | ||
388 | xfs_inode_t *ip, | ||
389 | xfs_off_t offset, /* starting I/O offset */ | ||
390 | xfs_fsize_t isize) /* current inode size */ | ||
391 | { | ||
392 | xfs_mount_t *mp = ip->i_mount; | ||
393 | xfs_fileoff_t start_zero_fsb; | ||
394 | xfs_fileoff_t end_zero_fsb; | ||
395 | xfs_fileoff_t zero_count_fsb; | ||
396 | xfs_fileoff_t last_fsb; | ||
397 | xfs_fileoff_t zero_off; | ||
398 | xfs_fsize_t zero_len; | ||
399 | int nimaps; | ||
400 | int error = 0; | ||
401 | xfs_bmbt_irec_t imap; | ||
402 | |||
403 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | ||
404 | ASSERT(offset > isize); | ||
405 | |||
406 | /* | ||
407 | * First handle zeroing the block on which isize resides. | ||
408 | * We only zero a part of that block so it is handled specially. | ||
409 | */ | ||
410 | error = xfs_zero_last_block(ip, offset, isize); | ||
411 | if (error) { | ||
412 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | ||
413 | return error; | ||
414 | } | ||
415 | |||
416 | /* | ||
417 | * Calculate the range between the new size and the old | ||
418 | * where blocks needing to be zeroed may exist. To get the | ||
419 | * block where the last byte in the file currently resides, | ||
420 | * we need to subtract one from the size and truncate back | ||
421 | * to a block boundary. We subtract 1 in case the size is | ||
422 | * exactly on a block boundary. | ||
423 | */ | ||
424 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | ||
425 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | ||
426 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | ||
427 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | ||
428 | if (last_fsb == end_zero_fsb) { | ||
429 | /* | ||
430 | * The size was only incremented on its last block. | ||
431 | * We took care of that above, so just return. | ||
432 | */ | ||
433 | return 0; | ||
434 | } | ||
435 | |||
436 | ASSERT(start_zero_fsb <= end_zero_fsb); | ||
437 | while (start_zero_fsb <= end_zero_fsb) { | ||
438 | nimaps = 1; | ||
439 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | ||
440 | error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb, | ||
441 | 0, NULL, 0, &imap, &nimaps, NULL, NULL); | ||
442 | if (error) { | ||
443 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | ||
444 | return error; | ||
445 | } | ||
446 | ASSERT(nimaps > 0); | ||
447 | |||
448 | if (imap.br_state == XFS_EXT_UNWRITTEN || | ||
449 | imap.br_startblock == HOLESTARTBLOCK) { | ||
450 | /* | ||
451 | * This loop handles initializing pages that were | ||
452 | * partially initialized by the code below this | ||
453 | * loop. It basically zeroes the part of the page | ||
454 | * that sits on a hole and sets the page as P_HOLE | ||
455 | * and calls remapf if it is a mapped file. | ||
456 | */ | ||
457 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | ||
458 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | ||
459 | continue; | ||
460 | } | ||
461 | |||
462 | /* | ||
463 | * There are blocks we need to zero. | ||
464 | * Drop the inode lock while we're doing the I/O. | ||
465 | * We'll still have the iolock to protect us. | ||
466 | */ | ||
467 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | ||
468 | |||
469 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); | ||
470 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | ||
471 | |||
472 | if ((zero_off + zero_len) > offset) | ||
473 | zero_len = offset - zero_off; | ||
474 | |||
475 | error = xfs_iozero(ip, zero_off, zero_len); | ||
476 | if (error) { | ||
477 | goto out_lock; | ||
478 | } | ||
479 | |||
480 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | ||
481 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | ||
482 | |||
483 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
484 | } | ||
485 | |||
486 | return 0; | ||
487 | |||
488 | out_lock: | ||
489 | xfs_ilock(ip, XFS_ILOCK_EXCL); | ||
490 | ASSERT(error >= 0); | ||
491 | return error; | ||
492 | } | ||
493 | |||
494 | ssize_t /* bytes written, or (-) error */ | ||
495 | xfs_write( | ||
496 | struct xfs_inode *xip, | ||
497 | struct kiocb *iocb, | ||
498 | const struct iovec *iovp, | ||
499 | unsigned int nsegs, | ||
500 | loff_t *offset, | ||
501 | int ioflags) | ||
502 | { | ||
503 | struct file *file = iocb->ki_filp; | ||
504 | struct address_space *mapping = file->f_mapping; | ||
505 | struct inode *inode = mapping->host; | ||
506 | unsigned long segs = nsegs; | ||
507 | xfs_mount_t *mp; | ||
508 | ssize_t ret = 0, error = 0; | ||
509 | xfs_fsize_t isize, new_size; | ||
510 | int iolock; | ||
511 | int eventsent = 0; | ||
512 | size_t ocount = 0, count; | ||
513 | loff_t pos; | ||
514 | int need_i_mutex; | ||
515 | |||
516 | XFS_STATS_INC(xs_write_calls); | ||
517 | |||
518 | error = generic_segment_checks(iovp, &segs, &ocount, VERIFY_READ); | ||
519 | if (error) | ||
520 | return error; | ||
521 | |||
522 | count = ocount; | ||
523 | pos = *offset; | ||
524 | |||
525 | if (count == 0) | ||
526 | return 0; | ||
527 | |||
528 | mp = xip->i_mount; | ||
529 | |||
530 | xfs_wait_for_freeze(mp, SB_FREEZE_WRITE); | ||
531 | |||
532 | if (XFS_FORCED_SHUTDOWN(mp)) | ||
533 | return -EIO; | ||
534 | |||
535 | relock: | ||
536 | if (ioflags & IO_ISDIRECT) { | ||
537 | iolock = XFS_IOLOCK_SHARED; | ||
538 | need_i_mutex = 0; | ||
539 | } else { | ||
540 | iolock = XFS_IOLOCK_EXCL; | ||
541 | need_i_mutex = 1; | ||
542 | mutex_lock(&inode->i_mutex); | ||
543 | } | ||
544 | |||
545 | xfs_ilock(xip, XFS_ILOCK_EXCL|iolock); | ||
546 | |||
547 | start: | ||
548 | error = -generic_write_checks(file, &pos, &count, | ||
549 | S_ISBLK(inode->i_mode)); | ||
550 | if (error) { | ||
551 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | ||
552 | goto out_unlock_mutex; | ||
553 | } | ||
554 | |||
555 | if ((DM_EVENT_ENABLED(xip, DM_EVENT_WRITE) && | ||
556 | !(ioflags & IO_INVIS) && !eventsent)) { | ||
557 | int dmflags = FILP_DELAY_FLAG(file); | ||
558 | |||
559 | if (need_i_mutex) | ||
560 | dmflags |= DM_FLAGS_IMUX; | ||
561 | |||
562 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
563 | error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, xip, | ||
564 | pos, count, dmflags, &iolock); | ||
565 | if (error) { | ||
566 | goto out_unlock_internal; | ||
567 | } | ||
568 | xfs_ilock(xip, XFS_ILOCK_EXCL); | ||
569 | eventsent = 1; | ||
570 | |||
571 | /* | ||
572 | * The iolock was dropped and reacquired in XFS_SEND_DATA | ||
573 | * so we have to recheck the size when appending. | ||
574 | * We will only "goto start;" once, since having sent the | ||
575 | * event prevents another call to XFS_SEND_DATA, which is | ||
576 | * what allows the size to change in the first place. | ||
577 | */ | ||
578 | if ((file->f_flags & O_APPEND) && pos != xip->i_size) | ||
579 | goto start; | ||
580 | } | ||
581 | |||
582 | if (ioflags & IO_ISDIRECT) { | ||
583 | xfs_buftarg_t *target = | ||
584 | XFS_IS_REALTIME_INODE(xip) ? | ||
585 | mp->m_rtdev_targp : mp->m_ddev_targp; | ||
586 | |||
587 | if ((pos & target->bt_smask) || (count & target->bt_smask)) { | ||
588 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | ||
589 | return XFS_ERROR(-EINVAL); | ||
590 | } | ||
591 | |||
592 | if (!need_i_mutex && (mapping->nrpages || pos > xip->i_size)) { | ||
593 | xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock); | ||
594 | iolock = XFS_IOLOCK_EXCL; | ||
595 | need_i_mutex = 1; | ||
596 | mutex_lock(&inode->i_mutex); | ||
597 | xfs_ilock(xip, XFS_ILOCK_EXCL|iolock); | ||
598 | goto start; | ||
599 | } | ||
600 | } | ||
601 | |||
602 | new_size = pos + count; | ||
603 | if (new_size > xip->i_size) | ||
604 | xip->i_new_size = new_size; | ||
605 | |||
606 | if (likely(!(ioflags & IO_INVIS))) | ||
607 | file_update_time(file); | ||
608 | |||
609 | /* | ||
610 | * If the offset is beyond the size of the file, we have a couple | ||
611 | * of things to do. First, if there is already space allocated | ||
612 | * we need to either create holes or zero the disk or ... | ||
613 | * | ||
614 | * If there is a page where the previous size lands, we need | ||
615 | * to zero it out up to the new size. | ||
616 | */ | ||
617 | |||
618 | if (pos > xip->i_size) { | ||
619 | error = xfs_zero_eof(xip, pos, xip->i_size); | ||
620 | if (error) { | ||
621 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
622 | goto out_unlock_internal; | ||
623 | } | ||
624 | } | ||
625 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
626 | |||
627 | /* | ||
628 | * If we're writing the file then make sure to clear the | ||
629 | * setuid and setgid bits if the process is not being run | ||
630 | * by root. This keeps people from modifying setuid and | ||
631 | * setgid binaries. | ||
632 | */ | ||
633 | error = -file_remove_suid(file); | ||
634 | if (unlikely(error)) | ||
635 | goto out_unlock_internal; | ||
636 | |||
637 | /* We can write back this queue in page reclaim */ | ||
638 | current->backing_dev_info = mapping->backing_dev_info; | ||
639 | |||
640 | if ((ioflags & IO_ISDIRECT)) { | ||
641 | if (mapping->nrpages) { | ||
642 | WARN_ON(need_i_mutex == 0); | ||
643 | error = xfs_flushinval_pages(xip, | ||
644 | (pos & PAGE_CACHE_MASK), | ||
645 | -1, FI_REMAPF_LOCKED); | ||
646 | if (error) | ||
647 | goto out_unlock_internal; | ||
648 | } | ||
649 | |||
650 | if (need_i_mutex) { | ||
651 | /* demote the lock now the cached pages are gone */ | ||
652 | xfs_ilock_demote(xip, XFS_IOLOCK_EXCL); | ||
653 | mutex_unlock(&inode->i_mutex); | ||
654 | |||
655 | iolock = XFS_IOLOCK_SHARED; | ||
656 | need_i_mutex = 0; | ||
657 | } | ||
658 | |||
659 | trace_xfs_file_direct_write(xip, count, *offset, ioflags); | ||
660 | ret = generic_file_direct_write(iocb, iovp, | ||
661 | &segs, pos, offset, count, ocount); | ||
662 | |||
663 | /* | ||
664 | * direct-io write to a hole: fall through to buffered I/O | ||
665 | * for completing the rest of the request. | ||
666 | */ | ||
667 | if (ret >= 0 && ret != count) { | ||
668 | XFS_STATS_ADD(xs_write_bytes, ret); | ||
669 | |||
670 | pos += ret; | ||
671 | count -= ret; | ||
672 | |||
673 | ioflags &= ~IO_ISDIRECT; | ||
674 | xfs_iunlock(xip, iolock); | ||
675 | goto relock; | ||
676 | } | ||
677 | } else { | ||
678 | int enospc = 0; | ||
679 | ssize_t ret2 = 0; | ||
680 | |||
681 | write_retry: | ||
682 | trace_xfs_file_buffered_write(xip, count, *offset, ioflags); | ||
683 | ret2 = generic_file_buffered_write(iocb, iovp, segs, | ||
684 | pos, offset, count, ret); | ||
685 | /* | ||
686 | * if we just got an ENOSPC, flush the inode now we | ||
687 | * aren't holding any page locks and retry *once* | ||
688 | */ | ||
689 | if (ret2 == -ENOSPC && !enospc) { | ||
690 | error = xfs_flush_pages(xip, 0, -1, 0, FI_NONE); | ||
691 | if (error) | ||
692 | goto out_unlock_internal; | ||
693 | enospc = 1; | ||
694 | goto write_retry; | ||
695 | } | ||
696 | ret = ret2; | ||
697 | } | ||
698 | |||
699 | current->backing_dev_info = NULL; | ||
700 | |||
701 | isize = i_size_read(inode); | ||
702 | if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize)) | ||
703 | *offset = isize; | ||
704 | |||
705 | if (*offset > xip->i_size) { | ||
706 | xfs_ilock(xip, XFS_ILOCK_EXCL); | ||
707 | if (*offset > xip->i_size) | ||
708 | xip->i_size = *offset; | ||
709 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
710 | } | ||
711 | |||
712 | if (ret == -ENOSPC && | ||
713 | DM_EVENT_ENABLED(xip, DM_EVENT_NOSPACE) && !(ioflags & IO_INVIS)) { | ||
714 | xfs_iunlock(xip, iolock); | ||
715 | if (need_i_mutex) | ||
716 | mutex_unlock(&inode->i_mutex); | ||
717 | error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, xip, | ||
718 | DM_RIGHT_NULL, xip, DM_RIGHT_NULL, NULL, NULL, | ||
719 | 0, 0, 0); /* Delay flag intentionally unused */ | ||
720 | if (need_i_mutex) | ||
721 | mutex_lock(&inode->i_mutex); | ||
722 | xfs_ilock(xip, iolock); | ||
723 | if (error) | ||
724 | goto out_unlock_internal; | ||
725 | goto start; | ||
726 | } | ||
727 | |||
728 | error = -ret; | ||
729 | if (ret <= 0) | ||
730 | goto out_unlock_internal; | ||
731 | |||
732 | XFS_STATS_ADD(xs_write_bytes, ret); | ||
733 | |||
734 | /* Handle various SYNC-type writes */ | ||
735 | if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) { | ||
736 | loff_t end = pos + ret - 1; | ||
737 | int error2; | ||
738 | |||
739 | xfs_iunlock(xip, iolock); | ||
740 | if (need_i_mutex) | ||
741 | mutex_unlock(&inode->i_mutex); | ||
742 | |||
743 | error2 = filemap_write_and_wait_range(mapping, pos, end); | ||
744 | if (!error) | ||
745 | error = error2; | ||
746 | if (need_i_mutex) | ||
747 | mutex_lock(&inode->i_mutex); | ||
748 | xfs_ilock(xip, iolock); | ||
749 | |||
750 | error2 = xfs_fsync(xip); | ||
751 | if (!error) | ||
752 | error = error2; | ||
753 | } | ||
754 | |||
755 | out_unlock_internal: | ||
756 | if (xip->i_new_size) { | ||
757 | xfs_ilock(xip, XFS_ILOCK_EXCL); | ||
758 | xip->i_new_size = 0; | ||
759 | /* | ||
760 | * If this was a direct or synchronous I/O that failed (such | ||
761 | * as ENOSPC) then part of the I/O may have been written to | ||
762 | * disk before the error occured. In this case the on-disk | ||
763 | * file size may have been adjusted beyond the in-memory file | ||
764 | * size and now needs to be truncated back. | ||
765 | */ | ||
766 | if (xip->i_d.di_size > xip->i_size) | ||
767 | xip->i_d.di_size = xip->i_size; | ||
768 | xfs_iunlock(xip, XFS_ILOCK_EXCL); | ||
769 | } | ||
770 | xfs_iunlock(xip, iolock); | ||
771 | out_unlock_mutex: | ||
772 | if (need_i_mutex) | ||
773 | mutex_unlock(&inode->i_mutex); | ||
774 | return -error; | ||
775 | } | ||
776 | |||
777 | /* | ||
778 | * If the underlying (data/log/rt) device is readonly, there are some | ||
779 | * operations that cannot proceed. | ||
780 | */ | ||
781 | int | ||
782 | xfs_dev_is_read_only( | ||
783 | xfs_mount_t *mp, | ||
784 | char *message) | ||
785 | { | ||
786 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | ||
787 | xfs_readonly_buftarg(mp->m_logdev_targp) || | ||
788 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | ||
789 | cmn_err(CE_NOTE, | ||
790 | "XFS: %s required on read-only device.", message); | ||
791 | cmn_err(CE_NOTE, | ||
792 | "XFS: write access unavailable, cannot proceed."); | ||
793 | return EROFS; | ||
794 | } | ||
795 | return 0; | ||
796 | } | ||