~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/xfs/xfs_file.c

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (c) 2000-2005 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_log.h"
 21 #include "xfs_sb.h"
 22 #include "xfs_ag.h"
 23 #include "xfs_trans.h"
 24 #include "xfs_mount.h"
 25 #include "xfs_bmap_btree.h"
 26 #include "xfs_alloc.h"
 27 #include "xfs_dinode.h"
 28 #include "xfs_inode.h"
 29 #include "xfs_inode_item.h"
 30 #include "xfs_bmap.h"
 31 #include "xfs_bmap_util.h"
 32 #include "xfs_error.h"
 33 #include "xfs_da_btree.h"
 34 #include "xfs_dir2_format.h"
 35 #include "xfs_dir2.h"
 36 #include "xfs_dir2_priv.h"
 37 #include "xfs_ioctl.h"
 38 #include "xfs_trace.h"
 39 
 40 #include <linux/aio.h>
 41 #include <linux/dcache.h>
 42 #include <linux/falloc.h>
 43 #include <linux/pagevec.h>
 44 
 45 static const struct vm_operations_struct xfs_file_vm_ops;
 46 
 47 /*
 48  * Locking primitives for read and write IO paths to ensure we consistently use
 49  * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
 50  */
 51 static inline void
 52 xfs_rw_ilock(
 53         struct xfs_inode        *ip,
 54         int                     type)
 55 {
 56         if (type & XFS_IOLOCK_EXCL)
 57                 mutex_lock(&VFS_I(ip)->i_mutex);
 58         xfs_ilock(ip, type);
 59 }
 60 
 61 static inline void
 62 xfs_rw_iunlock(
 63         struct xfs_inode        *ip,
 64         int                     type)
 65 {
 66         xfs_iunlock(ip, type);
 67         if (type & XFS_IOLOCK_EXCL)
 68                 mutex_unlock(&VFS_I(ip)->i_mutex);
 69 }
 70 
 71 static inline void
 72 xfs_rw_ilock_demote(
 73         struct xfs_inode        *ip,
 74         int                     type)
 75 {
 76         xfs_ilock_demote(ip, type);
 77         if (type & XFS_IOLOCK_EXCL)
 78                 mutex_unlock(&VFS_I(ip)->i_mutex);
 79 }
 80 
 81 /*
 82  *      xfs_iozero
 83  *
 84  *      xfs_iozero clears the specified range of buffer supplied,
 85  *      and marks all the affected blocks as valid and modified.  If
 86  *      an affected block is not allocated, it will be allocated.  If
 87  *      an affected block is not completely overwritten, and is not
 88  *      valid before the operation, it will be read from disk before
 89  *      being partially zeroed.
 90  */
 91 int
 92 xfs_iozero(
 93         struct xfs_inode        *ip,    /* inode                        */
 94         loff_t                  pos,    /* offset in file               */
 95         size_t                  count)  /* size of data to zero         */
 96 {
 97         struct page             *page;
 98         struct address_space    *mapping;
 99         int                     status;
100 
101         mapping = VFS_I(ip)->i_mapping;
102         do {
103                 unsigned offset, bytes;
104                 void *fsdata;
105 
106                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
107                 bytes = PAGE_CACHE_SIZE - offset;
108                 if (bytes > count)
109                         bytes = count;
110 
111                 status = pagecache_write_begin(NULL, mapping, pos, bytes,
112                                         AOP_FLAG_UNINTERRUPTIBLE,
113                                         &page, &fsdata);
114                 if (status)
115                         break;
116 
117                 zero_user(page, offset, bytes);
118 
119                 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
120                                         page, fsdata);
121                 WARN_ON(status <= 0); /* can't return less than zero! */
122                 pos += bytes;
123                 count -= bytes;
124                 status = 0;
125         } while (count);
126 
127         return (-status);
128 }
129 
130 /*
131  * Fsync operations on directories are much simpler than on regular files,
132  * as there is no file data to flush, and thus also no need for explicit
133  * cache flush operations, and there are no non-transaction metadata updates
134  * on directories either.
135  */
136 STATIC int
137 xfs_dir_fsync(
138         struct file             *file,
139         loff_t                  start,
140         loff_t                  end,
141         int                     datasync)
142 {
143         struct xfs_inode        *ip = XFS_I(file->f_mapping->host);
144         struct xfs_mount        *mp = ip->i_mount;
145         xfs_lsn_t               lsn = 0;
146 
147         trace_xfs_dir_fsync(ip);
148 
149         xfs_ilock(ip, XFS_ILOCK_SHARED);
150         if (xfs_ipincount(ip))
151                 lsn = ip->i_itemp->ili_last_lsn;
152         xfs_iunlock(ip, XFS_ILOCK_SHARED);
153 
154         if (!lsn)
155                 return 0;
156         return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
157 }
158 
159 STATIC int
160 xfs_file_fsync(
161         struct file             *file,
162         loff_t                  start,
163         loff_t                  end,
164         int                     datasync)
165 {
166         struct inode            *inode = file->f_mapping->host;
167         struct xfs_inode        *ip = XFS_I(inode);
168         struct xfs_mount        *mp = ip->i_mount;
169         int                     error = 0;
170         int                     log_flushed = 0;
171         xfs_lsn_t               lsn = 0;
172 
173         trace_xfs_file_fsync(ip);
174 
175         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
176         if (error)
177                 return error;
178 
179         if (XFS_FORCED_SHUTDOWN(mp))
180                 return -XFS_ERROR(EIO);
181 
182         xfs_iflags_clear(ip, XFS_ITRUNCATED);
183 
184         if (mp->m_flags & XFS_MOUNT_BARRIER) {
185                 /*
186                  * If we have an RT and/or log subvolume we need to make sure
187                  * to flush the write cache the device used for file data
188                  * first.  This is to ensure newly written file data make
189                  * it to disk before logging the new inode size in case of
190                  * an extending write.
191                  */
192                 if (XFS_IS_REALTIME_INODE(ip))
193                         xfs_blkdev_issue_flush(mp->m_rtdev_targp);
194                 else if (mp->m_logdev_targp != mp->m_ddev_targp)
195                         xfs_blkdev_issue_flush(mp->m_ddev_targp);
196         }
197 
198         /*
199          * All metadata updates are logged, which means that we just have
200          * to flush the log up to the latest LSN that touched the inode.
201          */
202         xfs_ilock(ip, XFS_ILOCK_SHARED);
203         if (xfs_ipincount(ip)) {
204                 if (!datasync ||
205                     (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
206                         lsn = ip->i_itemp->ili_last_lsn;
207         }
208         xfs_iunlock(ip, XFS_ILOCK_SHARED);
209 
210         if (lsn)
211                 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
212 
213         /*
214          * If we only have a single device, and the log force about was
215          * a no-op we might have to flush the data device cache here.
216          * This can only happen for fdatasync/O_DSYNC if we were overwriting
217          * an already allocated file and thus do not have any metadata to
218          * commit.
219          */
220         if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
221             mp->m_logdev_targp == mp->m_ddev_targp &&
222             !XFS_IS_REALTIME_INODE(ip) &&
223             !log_flushed)
224                 xfs_blkdev_issue_flush(mp->m_ddev_targp);
225 
226         return -error;
227 }
228 
229 STATIC ssize_t
230 xfs_file_aio_read(
231         struct kiocb            *iocb,
232         const struct iovec      *iovp,
233         unsigned long           nr_segs,
234         loff_t                  pos)
235 {
236         struct file             *file = iocb->ki_filp;
237         struct inode            *inode = file->f_mapping->host;
238         struct xfs_inode        *ip = XFS_I(inode);
239         struct xfs_mount        *mp = ip->i_mount;
240         size_t                  size = 0;
241         ssize_t                 ret = 0;
242         int                     ioflags = 0;
243         xfs_fsize_t             n;
244 
245         XFS_STATS_INC(xs_read_calls);
246 
247         BUG_ON(iocb->ki_pos != pos);
248 
249         if (unlikely(file->f_flags & O_DIRECT))
250                 ioflags |= IO_ISDIRECT;
251         if (file->f_mode & FMODE_NOCMTIME)
252                 ioflags |= IO_INVIS;
253 
254         ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
255         if (ret < 0)
256                 return ret;
257 
258         if (unlikely(ioflags & IO_ISDIRECT)) {
259                 xfs_buftarg_t   *target =
260                         XFS_IS_REALTIME_INODE(ip) ?
261                                 mp->m_rtdev_targp : mp->m_ddev_targp;
262                 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
263                         if (pos == i_size_read(inode))
264                                 return 0;
265                         return -XFS_ERROR(EINVAL);
266                 }
267         }
268 
269         n = mp->m_super->s_maxbytes - pos;
270         if (n <= 0 || size == 0)
271                 return 0;
272 
273         if (n < size)
274                 size = n;
275 
276         if (XFS_FORCED_SHUTDOWN(mp))
277                 return -EIO;
278 
279         /*
280          * Locking is a bit tricky here. If we take an exclusive lock
281          * for direct IO, we effectively serialise all new concurrent
282          * read IO to this file and block it behind IO that is currently in
283          * progress because IO in progress holds the IO lock shared. We only
284          * need to hold the lock exclusive to blow away the page cache, so
285          * only take lock exclusively if the page cache needs invalidation.
286          * This allows the normal direct IO case of no page cache pages to
287          * proceeed concurrently without serialisation.
288          */
289         xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
290         if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
291                 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
292                 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
293 
294                 if (inode->i_mapping->nrpages) {
295                         ret = -filemap_write_and_wait_range(
296                                                         VFS_I(ip)->i_mapping,
297                                                         pos, -1);
298                         if (ret) {
299                                 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
300                                 return ret;
301                         }
302 
303                         /*
304                          * Invalidate whole pages. This can return an error if
305                          * we fail to invalidate a page, but this should never
306                          * happen on XFS. Warn if it does fail.
307                          */
308                         ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
309                                                 pos >> PAGE_CACHE_SHIFT, -1);
310                         WARN_ON_ONCE(ret);
311                         ret = 0;
312                 }
313                 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
314         }
315 
316         trace_xfs_file_read(ip, size, pos, ioflags);
317 
318         ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
319         if (ret > 0)
320                 XFS_STATS_ADD(xs_read_bytes, ret);
321 
322         xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
323         return ret;
324 }
325 
326 STATIC ssize_t
327 xfs_file_splice_read(
328         struct file             *infilp,
329         loff_t                  *ppos,
330         struct pipe_inode_info  *pipe,
331         size_t                  count,
332         unsigned int            flags)
333 {
334         struct xfs_inode        *ip = XFS_I(infilp->f_mapping->host);
335         int                     ioflags = 0;
336         ssize_t                 ret;
337 
338         XFS_STATS_INC(xs_read_calls);
339 
340         if (infilp->f_mode & FMODE_NOCMTIME)
341                 ioflags |= IO_INVIS;
342 
343         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
344                 return -EIO;
345 
346         xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
347 
348         trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
349 
350         ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
351         if (ret > 0)
352                 XFS_STATS_ADD(xs_read_bytes, ret);
353 
354         xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
355         return ret;
356 }
357 
358 /*
359  * xfs_file_splice_write() does not use xfs_rw_ilock() because
360  * generic_file_splice_write() takes the i_mutex itself. This, in theory,
361  * couuld cause lock inversions between the aio_write path and the splice path
362  * if someone is doing concurrent splice(2) based writes and write(2) based
363  * writes to the same inode. The only real way to fix this is to re-implement
364  * the generic code here with correct locking orders.
365  */
366 STATIC ssize_t
367 xfs_file_splice_write(
368         struct pipe_inode_info  *pipe,
369         struct file             *outfilp,
370         loff_t                  *ppos,
371         size_t                  count,
372         unsigned int            flags)
373 {
374         struct inode            *inode = outfilp->f_mapping->host;
375         struct xfs_inode        *ip = XFS_I(inode);
376         int                     ioflags = 0;
377         ssize_t                 ret;
378 
379         XFS_STATS_INC(xs_write_calls);
380 
381         if (outfilp->f_mode & FMODE_NOCMTIME)
382                 ioflags |= IO_INVIS;
383 
384         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
385                 return -EIO;
386 
387         xfs_ilock(ip, XFS_IOLOCK_EXCL);
388 
389         trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
390 
391         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
392         if (ret > 0)
393                 XFS_STATS_ADD(xs_write_bytes, ret);
394 
395         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
396         return ret;
397 }
398 
399 /*
400  * This routine is called to handle zeroing any space in the last block of the
401  * file that is beyond the EOF.  We do this since the size is being increased
402  * without writing anything to that block and we don't want to read the
403  * garbage on the disk.
404  */
405 STATIC int                              /* error (positive) */
406 xfs_zero_last_block(
407         struct xfs_inode        *ip,
408         xfs_fsize_t             offset,
409         xfs_fsize_t             isize)
410 {
411         struct xfs_mount        *mp = ip->i_mount;
412         xfs_fileoff_t           last_fsb = XFS_B_TO_FSBT(mp, isize);
413         int                     zero_offset = XFS_B_FSB_OFFSET(mp, isize);
414         int                     zero_len;
415         int                     nimaps = 1;
416         int                     error = 0;
417         struct xfs_bmbt_irec    imap;
418 
419         xfs_ilock(ip, XFS_ILOCK_EXCL);
420         error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
421         xfs_iunlock(ip, XFS_ILOCK_EXCL);
422         if (error)
423                 return error;
424 
425         ASSERT(nimaps > 0);
426 
427         /*
428          * If the block underlying isize is just a hole, then there
429          * is nothing to zero.
430          */
431         if (imap.br_startblock == HOLESTARTBLOCK)
432                 return 0;
433 
434         zero_len = mp->m_sb.sb_blocksize - zero_offset;
435         if (isize + zero_len > offset)
436                 zero_len = offset - isize;
437         return xfs_iozero(ip, isize, zero_len);
438 }
439 
440 /*
441  * Zero any on disk space between the current EOF and the new, larger EOF.
442  *
443  * This handles the normal case of zeroing the remainder of the last block in
444  * the file and the unusual case of zeroing blocks out beyond the size of the
445  * file.  This second case only happens with fixed size extents and when the
446  * system crashes before the inode size was updated but after blocks were
447  * allocated.
448  *
449  * Expects the iolock to be held exclusive, and will take the ilock internally.
450  */
451 int                                     /* error (positive) */
452 xfs_zero_eof(
453         struct xfs_inode        *ip,
454         xfs_off_t               offset,         /* starting I/O offset */
455         xfs_fsize_t             isize)          /* current inode size */
456 {
457         struct xfs_mount        *mp = ip->i_mount;
458         xfs_fileoff_t           start_zero_fsb;
459         xfs_fileoff_t           end_zero_fsb;
460         xfs_fileoff_t           zero_count_fsb;
461         xfs_fileoff_t           last_fsb;
462         xfs_fileoff_t           zero_off;
463         xfs_fsize_t             zero_len;
464         int                     nimaps;
465         int                     error = 0;
466         struct xfs_bmbt_irec    imap;
467 
468         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
469         ASSERT(offset > isize);
470 
471         /*
472          * First handle zeroing the block on which isize resides.
473          *
474          * We only zero a part of that block so it is handled specially.
475          */
476         if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
477                 error = xfs_zero_last_block(ip, offset, isize);
478                 if (error)
479                         return error;
480         }
481 
482         /*
483          * Calculate the range between the new size and the old where blocks
484          * needing to be zeroed may exist.
485          *
486          * To get the block where the last byte in the file currently resides,
487          * we need to subtract one from the size and truncate back to a block
488          * boundary.  We subtract 1 in case the size is exactly on a block
489          * boundary.
490          */
491         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
492         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
493         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
494         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
495         if (last_fsb == end_zero_fsb) {
496                 /*
497                  * The size was only incremented on its last block.
498                  * We took care of that above, so just return.
499                  */
500                 return 0;
501         }
502 
503         ASSERT(start_zero_fsb <= end_zero_fsb);
504         while (start_zero_fsb <= end_zero_fsb) {
505                 nimaps = 1;
506                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
507 
508                 xfs_ilock(ip, XFS_ILOCK_EXCL);
509                 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
510                                           &imap, &nimaps, 0);
511                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
512                 if (error)
513                         return error;
514 
515                 ASSERT(nimaps > 0);
516 
517                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
518                     imap.br_startblock == HOLESTARTBLOCK) {
519                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
520                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
521                         continue;
522                 }
523 
524                 /*
525                  * There are blocks we need to zero.
526                  */
527                 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
528                 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
529 
530                 if ((zero_off + zero_len) > offset)
531                         zero_len = offset - zero_off;
532 
533                 error = xfs_iozero(ip, zero_off, zero_len);
534                 if (error)
535                         return error;
536 
537                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
538                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
539         }
540 
541         return 0;
542 }
543 
544 /*
545  * Common pre-write limit and setup checks.
546  *
547  * Called with the iolocked held either shared and exclusive according to
548  * @iolock, and returns with it held.  Might upgrade the iolock to exclusive
549  * if called for a direct write beyond i_size.
550  */
551 STATIC ssize_t
552 xfs_file_aio_write_checks(
553         struct file             *file,
554         loff_t                  *pos,
555         size_t                  *count,
556         int                     *iolock)
557 {
558         struct inode            *inode = file->f_mapping->host;
559         struct xfs_inode        *ip = XFS_I(inode);
560         int                     error = 0;
561 
562 restart:
563         error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
564         if (error)
565                 return error;
566 
567         /*
568          * If the offset is beyond the size of the file, we need to zero any
569          * blocks that fall between the existing EOF and the start of this
570          * write.  If zeroing is needed and we are currently holding the
571          * iolock shared, we need to update it to exclusive which implies
572          * having to redo all checks before.
573          */
574         if (*pos > i_size_read(inode)) {
575                 if (*iolock == XFS_IOLOCK_SHARED) {
576                         xfs_rw_iunlock(ip, *iolock);
577                         *iolock = XFS_IOLOCK_EXCL;
578                         xfs_rw_ilock(ip, *iolock);
579                         goto restart;
580                 }
581                 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
582                 if (error)
583                         return error;
584         }
585 
586         /*
587          * Updating the timestamps will grab the ilock again from
588          * xfs_fs_dirty_inode, so we have to call it after dropping the
589          * lock above.  Eventually we should look into a way to avoid
590          * the pointless lock roundtrip.
591          */
592         if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
593                 error = file_update_time(file);
594                 if (error)
595                         return error;
596         }
597 
598         /*
599          * If we're writing the file then make sure to clear the setuid and
600          * setgid bits if the process is not being run by root.  This keeps
601          * people from modifying setuid and setgid binaries.
602          */
603         return file_remove_suid(file);
604 }
605 
606 /*
607  * xfs_file_dio_aio_write - handle direct IO writes
608  *
609  * Lock the inode appropriately to prepare for and issue a direct IO write.
610  * By separating it from the buffered write path we remove all the tricky to
611  * follow locking changes and looping.
612  *
613  * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
614  * until we're sure the bytes at the new EOF have been zeroed and/or the cached
615  * pages are flushed out.
616  *
617  * In most cases the direct IO writes will be done holding IOLOCK_SHARED
618  * allowing them to be done in parallel with reads and other direct IO writes.
619  * However, if the IO is not aligned to filesystem blocks, the direct IO layer
620  * needs to do sub-block zeroing and that requires serialisation against other
621  * direct IOs to the same block. In this case we need to serialise the
622  * submission of the unaligned IOs so that we don't get racing block zeroing in
623  * the dio layer.  To avoid the problem with aio, we also need to wait for
624  * outstanding IOs to complete so that unwritten extent conversion is completed
625  * before we try to map the overlapping block. This is currently implemented by
626  * hitting it with a big hammer (i.e. inode_dio_wait()).
627  *
628  * Returns with locks held indicated by @iolock and errors indicated by
629  * negative return values.
630  */
631 STATIC ssize_t
632 xfs_file_dio_aio_write(
633         struct kiocb            *iocb,
634         const struct iovec      *iovp,
635         unsigned long           nr_segs,
636         loff_t                  pos,
637         size_t                  ocount)
638 {
639         struct file             *file = iocb->ki_filp;
640         struct address_space    *mapping = file->f_mapping;
641         struct inode            *inode = mapping->host;
642         struct xfs_inode        *ip = XFS_I(inode);
643         struct xfs_mount        *mp = ip->i_mount;
644         ssize_t                 ret = 0;
645         size_t                  count = ocount;
646         int                     unaligned_io = 0;
647         int                     iolock;
648         struct xfs_buftarg      *target = XFS_IS_REALTIME_INODE(ip) ?
649                                         mp->m_rtdev_targp : mp->m_ddev_targp;
650 
651         if ((pos & target->bt_smask) || (count & target->bt_smask))
652                 return -XFS_ERROR(EINVAL);
653 
654         if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
655                 unaligned_io = 1;
656 
657         /*
658          * We don't need to take an exclusive lock unless there page cache needs
659          * to be invalidated or unaligned IO is being executed. We don't need to
660          * consider the EOF extension case here because
661          * xfs_file_aio_write_checks() will relock the inode as necessary for
662          * EOF zeroing cases and fill out the new inode size as appropriate.
663          */
664         if (unaligned_io || mapping->nrpages)
665                 iolock = XFS_IOLOCK_EXCL;
666         else
667                 iolock = XFS_IOLOCK_SHARED;
668         xfs_rw_ilock(ip, iolock);
669 
670         /*
671          * Recheck if there are cached pages that need invalidate after we got
672          * the iolock to protect against other threads adding new pages while
673          * we were waiting for the iolock.
674          */
675         if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
676                 xfs_rw_iunlock(ip, iolock);
677                 iolock = XFS_IOLOCK_EXCL;
678                 xfs_rw_ilock(ip, iolock);
679         }
680 
681         ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
682         if (ret)
683                 goto out;
684 
685         if (mapping->nrpages) {
686                 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
687                                                     pos, -1);
688                 if (ret)
689                         goto out;
690                 /*
691                  * Invalidate whole pages. This can return an error if
692                  * we fail to invalidate a page, but this should never
693                  * happen on XFS. Warn if it does fail.
694                  */
695                 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
696                                                 pos >> PAGE_CACHE_SHIFT, -1);
697                 WARN_ON_ONCE(ret);
698                 ret = 0;
699         }
700 
701         /*
702          * If we are doing unaligned IO, wait for all other IO to drain,
703          * otherwise demote the lock if we had to flush cached pages
704          */
705         if (unaligned_io)
706                 inode_dio_wait(inode);
707         else if (iolock == XFS_IOLOCK_EXCL) {
708                 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
709                 iolock = XFS_IOLOCK_SHARED;
710         }
711 
712         trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
713         ret = generic_file_direct_write(iocb, iovp,
714                         &nr_segs, pos, &iocb->ki_pos, count, ocount);
715 
716 out:
717         xfs_rw_iunlock(ip, iolock);
718 
719         /* No fallback to buffered IO on errors for XFS. */
720         ASSERT(ret < 0 || ret == count);
721         return ret;
722 }
723 
724 STATIC ssize_t
725 xfs_file_buffered_aio_write(
726         struct kiocb            *iocb,
727         const struct iovec      *iovp,
728         unsigned long           nr_segs,
729         loff_t                  pos,
730         size_t                  ocount)
731 {
732         struct file             *file = iocb->ki_filp;
733         struct address_space    *mapping = file->f_mapping;
734         struct inode            *inode = mapping->host;
735         struct xfs_inode        *ip = XFS_I(inode);
736         ssize_t                 ret;
737         int                     enospc = 0;
738         int                     iolock = XFS_IOLOCK_EXCL;
739         size_t                  count = ocount;
740 
741         xfs_rw_ilock(ip, iolock);
742 
743         ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
744         if (ret)
745                 goto out;
746 
747         /* We can write back this queue in page reclaim */
748         current->backing_dev_info = mapping->backing_dev_info;
749 
750 write_retry:
751         trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
752         ret = generic_file_buffered_write(iocb, iovp, nr_segs,
753                         pos, &iocb->ki_pos, count, 0);
754 
755         /*
756          * If we just got an ENOSPC, try to write back all dirty inodes to
757          * convert delalloc space to free up some of the excess reserved
758          * metadata space.
759          */
760         if (ret == -ENOSPC && !enospc) {
761                 enospc = 1;
762                 xfs_flush_inodes(ip->i_mount);
763                 goto write_retry;
764         }
765 
766         current->backing_dev_info = NULL;
767 out:
768         xfs_rw_iunlock(ip, iolock);
769         return ret;
770 }
771 
772 STATIC ssize_t
773 xfs_file_aio_write(
774         struct kiocb            *iocb,
775         const struct iovec      *iovp,
776         unsigned long           nr_segs,
777         loff_t                  pos)
778 {
779         struct file             *file = iocb->ki_filp;
780         struct address_space    *mapping = file->f_mapping;
781         struct inode            *inode = mapping->host;
782         struct xfs_inode        *ip = XFS_I(inode);
783         ssize_t                 ret;
784         size_t                  ocount = 0;
785 
786         XFS_STATS_INC(xs_write_calls);
787 
788         BUG_ON(iocb->ki_pos != pos);
789 
790         ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
791         if (ret)
792                 return ret;
793 
794         if (ocount == 0)
795                 return 0;
796 
797         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
798                 ret = -EIO;
799                 goto out;
800         }
801 
802         if (unlikely(file->f_flags & O_DIRECT))
803                 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
804         else
805                 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
806                                                   ocount);
807 
808         if (ret > 0) {
809                 ssize_t err;
810 
811                 XFS_STATS_ADD(xs_write_bytes, ret);
812 
813                 /* Handle various SYNC-type writes */
814                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
815                 if (err < 0)
816                         ret = err;
817         }
818 
819 out:
820         return ret;
821 }
822 
823 STATIC long
824 xfs_file_fallocate(
825         struct file     *file,
826         int             mode,
827         loff_t          offset,
828         loff_t          len)
829 {
830         struct inode    *inode = file_inode(file);
831         long            error;
832         loff_t          new_size = 0;
833         xfs_flock64_t   bf;
834         xfs_inode_t     *ip = XFS_I(inode);
835         int             cmd = XFS_IOC_RESVSP;
836         int             attr_flags = XFS_ATTR_NOLOCK;
837 
838         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
839                 return -EOPNOTSUPP;
840 
841         bf.l_whence = 0;
842         bf.l_start = offset;
843         bf.l_len = len;
844 
845         xfs_ilock(ip, XFS_IOLOCK_EXCL);
846 
847         if (mode & FALLOC_FL_PUNCH_HOLE)
848                 cmd = XFS_IOC_UNRESVSP;
849 
850         /* check the new inode size is valid before allocating */
851         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
852             offset + len > i_size_read(inode)) {
853                 new_size = offset + len;
854                 error = inode_newsize_ok(inode, new_size);
855                 if (error)
856                         goto out_unlock;
857         }
858 
859         if (file->f_flags & O_DSYNC)
860                 attr_flags |= XFS_ATTR_SYNC;
861 
862         error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
863         if (error)
864                 goto out_unlock;
865 
866         /* Change file size if needed */
867         if (new_size) {
868                 struct iattr iattr;
869 
870                 iattr.ia_valid = ATTR_SIZE;
871                 iattr.ia_size = new_size;
872                 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
873         }
874 
875 out_unlock:
876         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
877         return error;
878 }
879 
880 
881 STATIC int
882 xfs_file_open(
883         struct inode    *inode,
884         struct file     *file)
885 {
886         if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
887                 return -EFBIG;
888         if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
889                 return -EIO;
890         return 0;
891 }
892 
893 STATIC int
894 xfs_dir_open(
895         struct inode    *inode,
896         struct file     *file)
897 {
898         struct xfs_inode *ip = XFS_I(inode);
899         int             mode;
900         int             error;
901 
902         error = xfs_file_open(inode, file);
903         if (error)
904                 return error;
905 
906         /*
907          * If there are any blocks, read-ahead block 0 as we're almost
908          * certain to have the next operation be a read there.
909          */
910         mode = xfs_ilock_map_shared(ip);
911         if (ip->i_d.di_nextents > 0)
912                 xfs_dir3_data_readahead(NULL, ip, 0, -1);
913         xfs_iunlock(ip, mode);
914         return 0;
915 }
916 
917 STATIC int
918 xfs_file_release(
919         struct inode    *inode,
920         struct file     *filp)
921 {
922         return -xfs_release(XFS_I(inode));
923 }
924 
925 STATIC int
926 xfs_file_readdir(
927         struct file     *file,
928         struct dir_context *ctx)
929 {
930         struct inode    *inode = file_inode(file);
931         xfs_inode_t     *ip = XFS_I(inode);
932         int             error;
933         size_t          bufsize;
934 
935         /*
936          * The Linux API doesn't pass down the total size of the buffer
937          * we read into down to the filesystem.  With the filldir concept
938          * it's not needed for correct information, but the XFS dir2 leaf
939          * code wants an estimate of the buffer size to calculate it's
940          * readahead window and size the buffers used for mapping to
941          * physical blocks.
942          *
943          * Try to give it an estimate that's good enough, maybe at some
944          * point we can change the ->readdir prototype to include the
945          * buffer size.  For now we use the current glibc buffer size.
946          */
947         bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
948 
949         error = xfs_readdir(ip, ctx, bufsize);
950         if (error)
951                 return -error;
952         return 0;
953 }
954 
955 STATIC int
956 xfs_file_mmap(
957         struct file     *filp,
958         struct vm_area_struct *vma)
959 {
960         vma->vm_ops = &xfs_file_vm_ops;
961 
962         file_accessed(filp);
963         return 0;
964 }
965 
966 /*
967  * mmap()d file has taken write protection fault and is being made
968  * writable. We can set the page state up correctly for a writable
969  * page, which means we can do correct delalloc accounting (ENOSPC
970  * checking!) and unwritten extent mapping.
971  */
972 STATIC int
973 xfs_vm_page_mkwrite(
974         struct vm_area_struct   *vma,
975         struct vm_fault         *vmf)
976 {
977         return block_page_mkwrite(vma, vmf, xfs_get_blocks);
978 }
979 
980 /*
981  * This type is designed to indicate the type of offset we would like
982  * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
983  */
984 enum {
985         HOLE_OFF = 0,
986         DATA_OFF,
987 };
988 
989 /*
990  * Lookup the desired type of offset from the given page.
991  *
992  * On success, return true and the offset argument will point to the
993  * start of the region that was found.  Otherwise this function will
994  * return false and keep the offset argument unchanged.
995  */
996 STATIC bool
997 xfs_lookup_buffer_offset(
998         struct page             *page,
999         loff_t                  *offset,
1000         unsigned int            type)
1001 {
1002         loff_t                  lastoff = page_offset(page);
1003         bool                    found = false;
1004         struct buffer_head      *bh, *head;
1005 
1006         bh = head = page_buffers(page);
1007         do {
1008                 /*
1009                  * Unwritten extents that have data in the page
1010                  * cache covering them can be identified by the
1011                  * BH_Unwritten state flag.  Pages with multiple
1012                  * buffers might have a mix of holes, data and
1013                  * unwritten extents - any buffer with valid
1014                  * data in it should have BH_Uptodate flag set
1015                  * on it.
1016                  */
1017                 if (buffer_unwritten(bh) ||
1018                     buffer_uptodate(bh)) {
1019                         if (type == DATA_OFF)
1020                                 found = true;
1021                 } else {
1022                         if (type == HOLE_OFF)
1023                                 found = true;
1024                 }
1025 
1026                 if (found) {
1027                         *offset = lastoff;
1028                         break;
1029                 }
1030                 lastoff += bh->b_size;
1031         } while ((bh = bh->b_this_page) != head);
1032 
1033         return found;
1034 }
1035 
1036 /*
1037  * This routine is called to find out and return a data or hole offset
1038  * from the page cache for unwritten extents according to the desired
1039  * type for xfs_seek_data() or xfs_seek_hole().
1040  *
1041  * The argument offset is used to tell where we start to search from the
1042  * page cache.  Map is used to figure out the end points of the range to
1043  * lookup pages.
1044  *
1045  * Return true if the desired type of offset was found, and the argument
1046  * offset is filled with that address.  Otherwise, return false and keep
1047  * offset unchanged.
1048  */
1049 STATIC bool
1050 xfs_find_get_desired_pgoff(
1051         struct inode            *inode,
1052         struct xfs_bmbt_irec    *map,
1053         unsigned int            type,
1054         loff_t                  *offset)
1055 {
1056         struct xfs_inode        *ip = XFS_I(inode);
1057         struct xfs_mount        *mp = ip->i_mount;
1058         struct pagevec          pvec;
1059         pgoff_t                 index;
1060         pgoff_t                 end;
1061         loff_t                  endoff;
1062         loff_t                  startoff = *offset;
1063         loff_t                  lastoff = startoff;
1064         bool                    found = false;
1065 
1066         pagevec_init(&pvec, 0);
1067 
1068         index = startoff >> PAGE_CACHE_SHIFT;
1069         endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1070         end = endoff >> PAGE_CACHE_SHIFT;
1071         do {
1072                 int             want;
1073                 unsigned        nr_pages;
1074                 unsigned int    i;
1075 
1076                 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1077                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1078                                           want);
1079                 /*
1080                  * No page mapped into given range.  If we are searching holes
1081                  * and if this is the first time we got into the loop, it means
1082                  * that the given offset is landed in a hole, return it.
1083                  *
1084                  * If we have already stepped through some block buffers to find
1085                  * holes but they all contains data.  In this case, the last
1086                  * offset is already updated and pointed to the end of the last
1087                  * mapped page, if it does not reach the endpoint to search,
1088                  * that means there should be a hole between them.
1089                  */
1090                 if (nr_pages == 0) {
1091                         /* Data search found nothing */
1092                         if (type == DATA_OFF)
1093                                 break;
1094 
1095                         ASSERT(type == HOLE_OFF);
1096                         if (lastoff == startoff || lastoff < endoff) {
1097                                 found = true;
1098                                 *offset = lastoff;
1099                         }
1100                         break;
1101                 }
1102 
1103                 /*
1104                  * At lease we found one page.  If this is the first time we
1105                  * step into the loop, and if the first page index offset is
1106                  * greater than the given search offset, a hole was found.
1107                  */
1108                 if (type == HOLE_OFF && lastoff == startoff &&
1109                     lastoff < page_offset(pvec.pages[0])) {
1110                         found = true;
1111                         break;
1112                 }
1113 
1114                 for (i = 0; i < nr_pages; i++) {
1115                         struct page     *page = pvec.pages[i];
1116                         loff_t          b_offset;
1117 
1118                         /*
1119                          * At this point, the page may be truncated or
1120                          * invalidated (changing page->mapping to NULL),
1121                          * or even swizzled back from swapper_space to tmpfs
1122                          * file mapping. However, page->index will not change
1123                          * because we have a reference on the page.
1124                          *
1125                          * Searching done if the page index is out of range.
1126                          * If the current offset is not reaches the end of
1127                          * the specified search range, there should be a hole
1128                          * between them.
1129                          */
1130                         if (page->index > end) {
1131                                 if (type == HOLE_OFF && lastoff < endoff) {
1132                                         *offset = lastoff;
1133                                         found = true;
1134                                 }
1135                                 goto out;
1136                         }
1137 
1138                         lock_page(page);
1139                         /*
1140                          * Page truncated or invalidated(page->mapping == NULL).
1141                          * We can freely skip it and proceed to check the next
1142                          * page.
1143                          */
1144                         if (unlikely(page->mapping != inode->i_mapping)) {
1145                                 unlock_page(page);
1146                                 continue;
1147                         }
1148 
1149                         if (!page_has_buffers(page)) {
1150                                 unlock_page(page);
1151                                 continue;
1152                         }
1153 
1154                         found = xfs_lookup_buffer_offset(page, &b_offset, type);
1155                         if (found) {
1156                                 /*
1157                                  * The found offset may be less than the start
1158                                  * point to search if this is the first time to
1159                                  * come here.
1160                                  */
1161                                 *offset = max_t(loff_t, startoff, b_offset);
1162                                 unlock_page(page);
1163                                 goto out;
1164                         }
1165 
1166                         /*
1167                          * We either searching data but nothing was found, or
1168                          * searching hole but found a data buffer.  In either
1169                          * case, probably the next page contains the desired
1170                          * things, update the last offset to it so.
1171                          */
1172                         lastoff = page_offset(page) + PAGE_SIZE;
1173                         unlock_page(page);
1174                 }
1175 
1176                 /*
1177                  * The number of returned pages less than our desired, search
1178                  * done.  In this case, nothing was found for searching data,
1179                  * but we found a hole behind the last offset.
1180                  */
1181                 if (nr_pages < want) {
1182                         if (type == HOLE_OFF) {
1183                                 *offset = lastoff;
1184                                 found = true;
1185                         }
1186                         break;
1187                 }
1188 
1189                 index = pvec.pages[i - 1]->index + 1;
1190                 pagevec_release(&pvec);
1191         } while (index <= end);
1192 
1193 out:
1194         pagevec_release(&pvec);
1195         return found;
1196 }
1197 
1198 STATIC loff_t
1199 xfs_seek_data(
1200         struct file             *file,
1201         loff_t                  start)
1202 {
1203         struct inode            *inode = file->f_mapping->host;
1204         struct xfs_inode        *ip = XFS_I(inode);
1205         struct xfs_mount        *mp = ip->i_mount;
1206         loff_t                  uninitialized_var(offset);
1207         xfs_fsize_t             isize;
1208         xfs_fileoff_t           fsbno;
1209         xfs_filblks_t           end;
1210         uint                    lock;
1211         int                     error;
1212 
1213         lock = xfs_ilock_map_shared(ip);
1214 
1215         isize = i_size_read(inode);
1216         if (start >= isize) {
1217                 error = ENXIO;
1218                 goto out_unlock;
1219         }
1220 
1221         /*
1222          * Try to read extents from the first block indicated
1223          * by fsbno to the end block of the file.
1224          */
1225         fsbno = XFS_B_TO_FSBT(mp, start);
1226         end = XFS_B_TO_FSB(mp, isize);
1227         for (;;) {
1228                 struct xfs_bmbt_irec    map[2];
1229                 int                     nmap = 2;
1230                 unsigned int            i;
1231 
1232                 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1233                                        XFS_BMAPI_ENTIRE);
1234                 if (error)
1235                         goto out_unlock;
1236 
1237                 /* No extents at given offset, must be beyond EOF */
1238                 if (nmap == 0) {
1239                         error = ENXIO;
1240                         goto out_unlock;
1241                 }
1242 
1243                 for (i = 0; i < nmap; i++) {
1244                         offset = max_t(loff_t, start,
1245                                        XFS_FSB_TO_B(mp, map[i].br_startoff));
1246 
1247                         /* Landed in a data extent */
1248                         if (map[i].br_startblock == DELAYSTARTBLOCK ||
1249                             (map[i].br_state == XFS_EXT_NORM &&
1250                              !isnullstartblock(map[i].br_startblock)))
1251                                 goto out;
1252 
1253                         /*
1254                          * Landed in an unwritten extent, try to search data
1255                          * from page cache.
1256                          */
1257                         if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1258                                 if (xfs_find_get_desired_pgoff(inode, &map[i],
1259                                                         DATA_OFF, &offset))
1260                                         goto out;
1261                         }
1262                 }
1263 
1264                 /*
1265                  * map[0] is hole or its an unwritten extent but
1266                  * without data in page cache.  Probably means that
1267                  * we are reading after EOF if nothing in map[1].
1268                  */
1269                 if (nmap == 1) {
1270                         error = ENXIO;
1271                         goto out_unlock;
1272                 }
1273 
1274                 ASSERT(i > 1);
1275 
1276                 /*
1277                  * Nothing was found, proceed to the next round of search
1278                  * if reading offset not beyond or hit EOF.
1279                  */
1280                 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1281                 start = XFS_FSB_TO_B(mp, fsbno);
1282                 if (start >= isize) {
1283                         error = ENXIO;
1284                         goto out_unlock;
1285                 }
1286         }
1287 
1288 out:
1289         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1290 
1291 out_unlock:
1292         xfs_iunlock_map_shared(ip, lock);
1293 
1294         if (error)
1295                 return -error;
1296         return offset;
1297 }
1298 
1299 STATIC loff_t
1300 xfs_seek_hole(
1301         struct file             *file,
1302         loff_t                  start)
1303 {
1304         struct inode            *inode = file->f_mapping->host;
1305         struct xfs_inode        *ip = XFS_I(inode);
1306         struct xfs_mount        *mp = ip->i_mount;
1307         loff_t                  uninitialized_var(offset);
1308         xfs_fsize_t             isize;
1309         xfs_fileoff_t           fsbno;
1310         xfs_filblks_t           end;
1311         uint                    lock;
1312         int                     error;
1313 
1314         if (XFS_FORCED_SHUTDOWN(mp))
1315                 return -XFS_ERROR(EIO);
1316 
1317         lock = xfs_ilock_map_shared(ip);
1318 
1319         isize = i_size_read(inode);
1320         if (start >= isize) {
1321                 error = ENXIO;
1322                 goto out_unlock;
1323         }
1324 
1325         fsbno = XFS_B_TO_FSBT(mp, start);
1326         end = XFS_B_TO_FSB(mp, isize);
1327 
1328         for (;;) {
1329                 struct xfs_bmbt_irec    map[2];
1330                 int                     nmap = 2;
1331                 unsigned int            i;
1332 
1333                 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1334                                        XFS_BMAPI_ENTIRE);
1335                 if (error)
1336                         goto out_unlock;
1337 
1338                 /* No extents at given offset, must be beyond EOF */
1339                 if (nmap == 0) {
1340                         error = ENXIO;
1341                         goto out_unlock;
1342                 }
1343 
1344                 for (i = 0; i < nmap; i++) {
1345                         offset = max_t(loff_t, start,
1346                                        XFS_FSB_TO_B(mp, map[i].br_startoff));
1347 
1348                         /* Landed in a hole */
1349                         if (map[i].br_startblock == HOLESTARTBLOCK)
1350                                 goto out;
1351 
1352                         /*
1353                          * Landed in an unwritten extent, try to search hole
1354                          * from page cache.
1355                          */
1356                         if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1357                                 if (xfs_find_get_desired_pgoff(inode, &map[i],
1358                                                         HOLE_OFF, &offset))
1359                                         goto out;
1360                         }
1361                 }
1362 
1363                 /*
1364                  * map[0] contains data or its unwritten but contains
1365                  * data in page cache, probably means that we are
1366                  * reading after EOF.  We should fix offset to point
1367                  * to the end of the file(i.e., there is an implicit
1368                  * hole at the end of any file).
1369                  */
1370                 if (nmap == 1) {
1371                         offset = isize;
1372                         break;
1373                 }
1374 
1375                 ASSERT(i > 1);
1376 
1377                 /*
1378                  * Both mappings contains data, proceed to the next round of
1379                  * search if the current reading offset not beyond or hit EOF.
1380                  */
1381                 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1382                 start = XFS_FSB_TO_B(mp, fsbno);
1383                 if (start >= isize) {
1384                         offset = isize;
1385                         break;
1386                 }
1387         }
1388 
1389 out:
1390         /*
1391          * At this point, we must have found a hole.  However, the returned
1392          * offset may be bigger than the file size as it may be aligned to
1393          * page boundary for unwritten extents, we need to deal with this
1394          * situation in particular.
1395          */
1396         offset = min_t(loff_t, offset, isize);
1397         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1398 
1399 out_unlock:
1400         xfs_iunlock_map_shared(ip, lock);
1401 
1402         if (error)
1403                 return -error;
1404         return offset;
1405 }
1406 
1407 STATIC loff_t
1408 xfs_file_llseek(
1409         struct file     *file,
1410         loff_t          offset,
1411         int             origin)
1412 {
1413         switch (origin) {
1414         case SEEK_END:
1415         case SEEK_CUR:
1416         case SEEK_SET:
1417                 return generic_file_llseek(file, offset, origin);
1418         case SEEK_DATA:
1419                 return xfs_seek_data(file, offset);
1420         case SEEK_HOLE:
1421                 return xfs_seek_hole(file, offset);
1422         default:
1423                 return -EINVAL;
1424         }
1425 }
1426 
1427 const struct file_operations xfs_file_operations = {
1428         .llseek         = xfs_file_llseek,
1429         .read           = do_sync_read,
1430         .write          = do_sync_write,
1431         .aio_read       = xfs_file_aio_read,
1432         .aio_write      = xfs_file_aio_write,
1433         .splice_read    = xfs_file_splice_read,
1434         .splice_write   = xfs_file_splice_write,
1435         .unlocked_ioctl = xfs_file_ioctl,
1436 #ifdef CONFIG_COMPAT
1437         .compat_ioctl   = xfs_file_compat_ioctl,
1438 #endif
1439         .mmap           = xfs_file_mmap,
1440         .open           = xfs_file_open,
1441         .release        = xfs_file_release,
1442         .fsync          = xfs_file_fsync,
1443         .fallocate      = xfs_file_fallocate,
1444 };
1445 
1446 const struct file_operations xfs_dir_file_operations = {
1447         .open           = xfs_dir_open,
1448         .read           = generic_read_dir,
1449         .iterate        = xfs_file_readdir,
1450         .llseek         = generic_file_llseek,
1451         .unlocked_ioctl = xfs_file_ioctl,
1452 #ifdef CONFIG_COMPAT
1453         .compat_ioctl   = xfs_file_compat_ioctl,
1454 #endif
1455         .fsync          = xfs_dir_fsync,
1456 };
1457 
1458 static const struct vm_operations_struct xfs_file_vm_ops = {
1459         .fault          = filemap_fault,
1460         .page_mkwrite   = xfs_vm_page_mkwrite,
1461         .remap_pages    = generic_file_remap_pages,
1462 };
1463 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp