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

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

Version: ~ [ linux-5.2 ] ~ [ linux-5.1.16 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.57 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.132 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.184 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.184 ] ~ [ 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-3.16.69 ] ~ [ 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-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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_shared.h"
 21 #include "xfs_format.h"
 22 #include "xfs_log_format.h"
 23 #include "xfs_trans_resv.h"
 24 #include "xfs_mount.h"
 25 #include "xfs_da_format.h"
 26 #include "xfs_da_btree.h"
 27 #include "xfs_inode.h"
 28 #include "xfs_trans.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_dir2.h"
 34 #include "xfs_dir2_priv.h"
 35 #include "xfs_ioctl.h"
 36 #include "xfs_trace.h"
 37 #include "xfs_log.h"
 38 #include "xfs_icache.h"
 39 #include "xfs_pnfs.h"
 40 
 41 #include <linux/dcache.h>
 42 #include <linux/falloc.h>
 43 #include <linux/pagevec.h>
 44 #include <linux/backing-dev.h>
 45 
 46 static const struct vm_operations_struct xfs_file_vm_ops;
 47 
 48 /*
 49  * Locking primitives for read and write IO paths to ensure we consistently use
 50  * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
 51  */
 52 static inline void
 53 xfs_rw_ilock(
 54         struct xfs_inode        *ip,
 55         int                     type)
 56 {
 57         if (type & XFS_IOLOCK_EXCL)
 58                 mutex_lock(&VFS_I(ip)->i_mutex);
 59         xfs_ilock(ip, type);
 60 }
 61 
 62 static inline void
 63 xfs_rw_iunlock(
 64         struct xfs_inode        *ip,
 65         int                     type)
 66 {
 67         xfs_iunlock(ip, type);
 68         if (type & XFS_IOLOCK_EXCL)
 69                 mutex_unlock(&VFS_I(ip)->i_mutex);
 70 }
 71 
 72 static inline void
 73 xfs_rw_ilock_demote(
 74         struct xfs_inode        *ip,
 75         int                     type)
 76 {
 77         xfs_ilock_demote(ip, type);
 78         if (type & XFS_IOLOCK_EXCL)
 79                 mutex_unlock(&VFS_I(ip)->i_mutex);
 80 }
 81 
 82 /*
 83  * xfs_iozero clears the specified range supplied via the page cache (except in
 84  * the DAX case). Writes through the page cache will allocate blocks over holes,
 85  * though the callers usually map the holes first and avoid them. If a block is
 86  * not completely zeroed, then it will be read from disk before being partially
 87  * zeroed.
 88  *
 89  * In the DAX case, we can just directly write to the underlying pages. This
 90  * will not allocate blocks, but will avoid holes and unwritten extents and so
 91  * not do unnecessary work.
 92  */
 93 int
 94 xfs_iozero(
 95         struct xfs_inode        *ip,    /* inode                        */
 96         loff_t                  pos,    /* offset in file               */
 97         size_t                  count)  /* size of data to zero         */
 98 {
 99         struct page             *page;
100         struct address_space    *mapping;
101         int                     status = 0;
102 
103 
104         mapping = VFS_I(ip)->i_mapping;
105         do {
106                 unsigned offset, bytes;
107                 void *fsdata;
108 
109                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
110                 bytes = PAGE_CACHE_SIZE - offset;
111                 if (bytes > count)
112                         bytes = count;
113 
114                 if (IS_DAX(VFS_I(ip))) {
115                         status = dax_zero_page_range(VFS_I(ip), pos, bytes,
116                                                      xfs_get_blocks_direct);
117                         if (status)
118                                 break;
119                 } else {
120                         status = pagecache_write_begin(NULL, mapping, pos, bytes,
121                                                 AOP_FLAG_UNINTERRUPTIBLE,
122                                                 &page, &fsdata);
123                         if (status)
124                                 break;
125 
126                         zero_user(page, offset, bytes);
127 
128                         status = pagecache_write_end(NULL, mapping, pos, bytes,
129                                                 bytes, page, fsdata);
130                         WARN_ON(status <= 0); /* can't return less than zero! */
131                         status = 0;
132                 }
133                 pos += bytes;
134                 count -= bytes;
135         } while (count);
136 
137         return status;
138 }
139 
140 int
141 xfs_update_prealloc_flags(
142         struct xfs_inode        *ip,
143         enum xfs_prealloc_flags flags)
144 {
145         struct xfs_trans        *tp;
146         int                     error;
147 
148         tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
149         error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
150         if (error) {
151                 xfs_trans_cancel(tp);
152                 return error;
153         }
154 
155         xfs_ilock(ip, XFS_ILOCK_EXCL);
156         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
157 
158         if (!(flags & XFS_PREALLOC_INVISIBLE)) {
159                 ip->i_d.di_mode &= ~S_ISUID;
160                 if (ip->i_d.di_mode & S_IXGRP)
161                         ip->i_d.di_mode &= ~S_ISGID;
162                 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
163         }
164 
165         if (flags & XFS_PREALLOC_SET)
166                 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
167         if (flags & XFS_PREALLOC_CLEAR)
168                 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
169 
170         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
171         if (flags & XFS_PREALLOC_SYNC)
172                 xfs_trans_set_sync(tp);
173         return xfs_trans_commit(tp);
174 }
175 
176 /*
177  * Fsync operations on directories are much simpler than on regular files,
178  * as there is no file data to flush, and thus also no need for explicit
179  * cache flush operations, and there are no non-transaction metadata updates
180  * on directories either.
181  */
182 STATIC int
183 xfs_dir_fsync(
184         struct file             *file,
185         loff_t                  start,
186         loff_t                  end,
187         int                     datasync)
188 {
189         struct xfs_inode        *ip = XFS_I(file->f_mapping->host);
190         struct xfs_mount        *mp = ip->i_mount;
191         xfs_lsn_t               lsn = 0;
192 
193         trace_xfs_dir_fsync(ip);
194 
195         xfs_ilock(ip, XFS_ILOCK_SHARED);
196         if (xfs_ipincount(ip))
197                 lsn = ip->i_itemp->ili_last_lsn;
198         xfs_iunlock(ip, XFS_ILOCK_SHARED);
199 
200         if (!lsn)
201                 return 0;
202         return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
203 }
204 
205 STATIC int
206 xfs_file_fsync(
207         struct file             *file,
208         loff_t                  start,
209         loff_t                  end,
210         int                     datasync)
211 {
212         struct inode            *inode = file->f_mapping->host;
213         struct xfs_inode        *ip = XFS_I(inode);
214         struct xfs_mount        *mp = ip->i_mount;
215         int                     error = 0;
216         int                     log_flushed = 0;
217         xfs_lsn_t               lsn = 0;
218 
219         trace_xfs_file_fsync(ip);
220 
221         error = filemap_write_and_wait_range(inode->i_mapping, start, end);
222         if (error)
223                 return error;
224 
225         if (XFS_FORCED_SHUTDOWN(mp))
226                 return -EIO;
227 
228         xfs_iflags_clear(ip, XFS_ITRUNCATED);
229 
230         if (mp->m_flags & XFS_MOUNT_BARRIER) {
231                 /*
232                  * If we have an RT and/or log subvolume we need to make sure
233                  * to flush the write cache the device used for file data
234                  * first.  This is to ensure newly written file data make
235                  * it to disk before logging the new inode size in case of
236                  * an extending write.
237                  */
238                 if (XFS_IS_REALTIME_INODE(ip))
239                         xfs_blkdev_issue_flush(mp->m_rtdev_targp);
240                 else if (mp->m_logdev_targp != mp->m_ddev_targp)
241                         xfs_blkdev_issue_flush(mp->m_ddev_targp);
242         }
243 
244         /*
245          * All metadata updates are logged, which means that we just have
246          * to flush the log up to the latest LSN that touched the inode.
247          */
248         xfs_ilock(ip, XFS_ILOCK_SHARED);
249         if (xfs_ipincount(ip)) {
250                 if (!datasync ||
251                     (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
252                         lsn = ip->i_itemp->ili_last_lsn;
253         }
254         xfs_iunlock(ip, XFS_ILOCK_SHARED);
255 
256         if (lsn)
257                 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
258 
259         /*
260          * If we only have a single device, and the log force about was
261          * a no-op we might have to flush the data device cache here.
262          * This can only happen for fdatasync/O_DSYNC if we were overwriting
263          * an already allocated file and thus do not have any metadata to
264          * commit.
265          */
266         if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
267             mp->m_logdev_targp == mp->m_ddev_targp &&
268             !XFS_IS_REALTIME_INODE(ip) &&
269             !log_flushed)
270                 xfs_blkdev_issue_flush(mp->m_ddev_targp);
271 
272         return error;
273 }
274 
275 STATIC ssize_t
276 xfs_file_read_iter(
277         struct kiocb            *iocb,
278         struct iov_iter         *to)
279 {
280         struct file             *file = iocb->ki_filp;
281         struct inode            *inode = file->f_mapping->host;
282         struct xfs_inode        *ip = XFS_I(inode);
283         struct xfs_mount        *mp = ip->i_mount;
284         size_t                  size = iov_iter_count(to);
285         ssize_t                 ret = 0;
286         int                     ioflags = 0;
287         xfs_fsize_t             n;
288         loff_t                  pos = iocb->ki_pos;
289 
290         XFS_STATS_INC(xs_read_calls);
291 
292         if (unlikely(iocb->ki_flags & IOCB_DIRECT))
293                 ioflags |= XFS_IO_ISDIRECT;
294         if (file->f_mode & FMODE_NOCMTIME)
295                 ioflags |= XFS_IO_INVIS;
296 
297         if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) {
298                 xfs_buftarg_t   *target =
299                         XFS_IS_REALTIME_INODE(ip) ?
300                                 mp->m_rtdev_targp : mp->m_ddev_targp;
301                 /* DIO must be aligned to device logical sector size */
302                 if ((pos | size) & target->bt_logical_sectormask) {
303                         if (pos == i_size_read(inode))
304                                 return 0;
305                         return -EINVAL;
306                 }
307         }
308 
309         n = mp->m_super->s_maxbytes - pos;
310         if (n <= 0 || size == 0)
311                 return 0;
312 
313         if (n < size)
314                 size = n;
315 
316         if (XFS_FORCED_SHUTDOWN(mp))
317                 return -EIO;
318 
319         /*
320          * Locking is a bit tricky here. If we take an exclusive lock for direct
321          * IO, we effectively serialise all new concurrent read IO to this file
322          * and block it behind IO that is currently in progress because IO in
323          * progress holds the IO lock shared. We only need to hold the lock
324          * exclusive to blow away the page cache, so only take lock exclusively
325          * if the page cache needs invalidation. This allows the normal direct
326          * IO case of no page cache pages to proceeed concurrently without
327          * serialisation.
328          */
329         xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
330         if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
331                 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
332                 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
333 
334                 /*
335                  * The generic dio code only flushes the range of the particular
336                  * I/O. Because we take an exclusive lock here, this whole
337                  * sequence is considerably more expensive for us. This has a
338                  * noticeable performance impact for any file with cached pages,
339                  * even when outside of the range of the particular I/O.
340                  *
341                  * Hence, amortize the cost of the lock against a full file
342                  * flush and reduce the chances of repeated iolock cycles going
343                  * forward.
344                  */
345                 if (inode->i_mapping->nrpages) {
346                         ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
347                         if (ret) {
348                                 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
349                                 return ret;
350                         }
351 
352                         /*
353                          * Invalidate whole pages. This can return an error if
354                          * we fail to invalidate a page, but this should never
355                          * happen on XFS. Warn if it does fail.
356                          */
357                         ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
358                         WARN_ON_ONCE(ret);
359                         ret = 0;
360                 }
361                 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
362         }
363 
364         trace_xfs_file_read(ip, size, pos, ioflags);
365 
366         ret = generic_file_read_iter(iocb, to);
367         if (ret > 0)
368                 XFS_STATS_ADD(xs_read_bytes, ret);
369 
370         xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
371         return ret;
372 }
373 
374 STATIC ssize_t
375 xfs_file_splice_read(
376         struct file             *infilp,
377         loff_t                  *ppos,
378         struct pipe_inode_info  *pipe,
379         size_t                  count,
380         unsigned int            flags)
381 {
382         struct xfs_inode        *ip = XFS_I(infilp->f_mapping->host);
383         int                     ioflags = 0;
384         ssize_t                 ret;
385 
386         XFS_STATS_INC(xs_read_calls);
387 
388         if (infilp->f_mode & FMODE_NOCMTIME)
389                 ioflags |= XFS_IO_INVIS;
390 
391         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
392                 return -EIO;
393 
394         xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
395 
396         trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
397 
398         /* for dax, we need to avoid the page cache */
399         if (IS_DAX(VFS_I(ip)))
400                 ret = default_file_splice_read(infilp, ppos, pipe, count, flags);
401         else
402                 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
403         if (ret > 0)
404                 XFS_STATS_ADD(xs_read_bytes, ret);
405 
406         xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
407         return ret;
408 }
409 
410 /*
411  * This routine is called to handle zeroing any space in the last block of the
412  * file that is beyond the EOF.  We do this since the size is being increased
413  * without writing anything to that block and we don't want to read the
414  * garbage on the disk.
415  */
416 STATIC int                              /* error (positive) */
417 xfs_zero_last_block(
418         struct xfs_inode        *ip,
419         xfs_fsize_t             offset,
420         xfs_fsize_t             isize,
421         bool                    *did_zeroing)
422 {
423         struct xfs_mount        *mp = ip->i_mount;
424         xfs_fileoff_t           last_fsb = XFS_B_TO_FSBT(mp, isize);
425         int                     zero_offset = XFS_B_FSB_OFFSET(mp, isize);
426         int                     zero_len;
427         int                     nimaps = 1;
428         int                     error = 0;
429         struct xfs_bmbt_irec    imap;
430 
431         xfs_ilock(ip, XFS_ILOCK_EXCL);
432         error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
433         xfs_iunlock(ip, XFS_ILOCK_EXCL);
434         if (error)
435                 return error;
436 
437         ASSERT(nimaps > 0);
438 
439         /*
440          * If the block underlying isize is just a hole, then there
441          * is nothing to zero.
442          */
443         if (imap.br_startblock == HOLESTARTBLOCK)
444                 return 0;
445 
446         zero_len = mp->m_sb.sb_blocksize - zero_offset;
447         if (isize + zero_len > offset)
448                 zero_len = offset - isize;
449         *did_zeroing = true;
450         return xfs_iozero(ip, isize, zero_len);
451 }
452 
453 /*
454  * Zero any on disk space between the current EOF and the new, larger EOF.
455  *
456  * This handles the normal case of zeroing the remainder of the last block in
457  * the file and the unusual case of zeroing blocks out beyond the size of the
458  * file.  This second case only happens with fixed size extents and when the
459  * system crashes before the inode size was updated but after blocks were
460  * allocated.
461  *
462  * Expects the iolock to be held exclusive, and will take the ilock internally.
463  */
464 int                                     /* error (positive) */
465 xfs_zero_eof(
466         struct xfs_inode        *ip,
467         xfs_off_t               offset,         /* starting I/O offset */
468         xfs_fsize_t             isize,          /* current inode size */
469         bool                    *did_zeroing)
470 {
471         struct xfs_mount        *mp = ip->i_mount;
472         xfs_fileoff_t           start_zero_fsb;
473         xfs_fileoff_t           end_zero_fsb;
474         xfs_fileoff_t           zero_count_fsb;
475         xfs_fileoff_t           last_fsb;
476         xfs_fileoff_t           zero_off;
477         xfs_fsize_t             zero_len;
478         int                     nimaps;
479         int                     error = 0;
480         struct xfs_bmbt_irec    imap;
481 
482         ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
483         ASSERT(offset > isize);
484 
485         /*
486          * First handle zeroing the block on which isize resides.
487          *
488          * We only zero a part of that block so it is handled specially.
489          */
490         if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
491                 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
492                 if (error)
493                         return error;
494         }
495 
496         /*
497          * Calculate the range between the new size and the old where blocks
498          * needing to be zeroed may exist.
499          *
500          * To get the block where the last byte in the file currently resides,
501          * we need to subtract one from the size and truncate back to a block
502          * boundary.  We subtract 1 in case the size is exactly on a block
503          * boundary.
504          */
505         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
506         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
507         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
508         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
509         if (last_fsb == end_zero_fsb) {
510                 /*
511                  * The size was only incremented on its last block.
512                  * We took care of that above, so just return.
513                  */
514                 return 0;
515         }
516 
517         ASSERT(start_zero_fsb <= end_zero_fsb);
518         while (start_zero_fsb <= end_zero_fsb) {
519                 nimaps = 1;
520                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
521 
522                 xfs_ilock(ip, XFS_ILOCK_EXCL);
523                 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
524                                           &imap, &nimaps, 0);
525                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
526                 if (error)
527                         return error;
528 
529                 ASSERT(nimaps > 0);
530 
531                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
532                     imap.br_startblock == HOLESTARTBLOCK) {
533                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
534                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
535                         continue;
536                 }
537 
538                 /*
539                  * There are blocks we need to zero.
540                  */
541                 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
542                 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
543 
544                 if ((zero_off + zero_len) > offset)
545                         zero_len = offset - zero_off;
546 
547                 error = xfs_iozero(ip, zero_off, zero_len);
548                 if (error)
549                         return error;
550 
551                 *did_zeroing = true;
552                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
553                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
554         }
555 
556         return 0;
557 }
558 
559 /*
560  * Common pre-write limit and setup checks.
561  *
562  * Called with the iolocked held either shared and exclusive according to
563  * @iolock, and returns with it held.  Might upgrade the iolock to exclusive
564  * if called for a direct write beyond i_size.
565  */
566 STATIC ssize_t
567 xfs_file_aio_write_checks(
568         struct kiocb            *iocb,
569         struct iov_iter         *from,
570         int                     *iolock)
571 {
572         struct file             *file = iocb->ki_filp;
573         struct inode            *inode = file->f_mapping->host;
574         struct xfs_inode        *ip = XFS_I(inode);
575         ssize_t                 error = 0;
576         size_t                  count = iov_iter_count(from);
577 
578 restart:
579         error = generic_write_checks(iocb, from);
580         if (error <= 0)
581                 return error;
582 
583         error = xfs_break_layouts(inode, iolock, true);
584         if (error)
585                 return error;
586 
587         /* For changing security info in file_remove_privs() we need i_mutex */
588         if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
589                 xfs_rw_iunlock(ip, *iolock);
590                 *iolock = XFS_IOLOCK_EXCL;
591                 xfs_rw_ilock(ip, *iolock);
592                 goto restart;
593         }
594         /*
595          * If the offset is beyond the size of the file, we need to zero any
596          * blocks that fall between the existing EOF and the start of this
597          * write.  If zeroing is needed and we are currently holding the
598          * iolock shared, we need to update it to exclusive which implies
599          * having to redo all checks before.
600          *
601          * We need to serialise against EOF updates that occur in IO
602          * completions here. We want to make sure that nobody is changing the
603          * size while we do this check until we have placed an IO barrier (i.e.
604          * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
605          * The spinlock effectively forms a memory barrier once we have the
606          * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
607          * and hence be able to correctly determine if we need to run zeroing.
608          */
609         spin_lock(&ip->i_flags_lock);
610         if (iocb->ki_pos > i_size_read(inode)) {
611                 bool    zero = false;
612 
613                 spin_unlock(&ip->i_flags_lock);
614                 if (*iolock == XFS_IOLOCK_SHARED) {
615                         xfs_rw_iunlock(ip, *iolock);
616                         *iolock = XFS_IOLOCK_EXCL;
617                         xfs_rw_ilock(ip, *iolock);
618                         iov_iter_reexpand(from, count);
619 
620                         /*
621                          * We now have an IO submission barrier in place, but
622                          * AIO can do EOF updates during IO completion and hence
623                          * we now need to wait for all of them to drain. Non-AIO
624                          * DIO will have drained before we are given the
625                          * XFS_IOLOCK_EXCL, and so for most cases this wait is a
626                          * no-op.
627                          */
628                         inode_dio_wait(inode);
629                         goto restart;
630                 }
631                 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
632                 if (error)
633                         return error;
634         } else
635                 spin_unlock(&ip->i_flags_lock);
636 
637         /*
638          * Updating the timestamps will grab the ilock again from
639          * xfs_fs_dirty_inode, so we have to call it after dropping the
640          * lock above.  Eventually we should look into a way to avoid
641          * the pointless lock roundtrip.
642          */
643         if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
644                 error = file_update_time(file);
645                 if (error)
646                         return error;
647         }
648 
649         /*
650          * If we're writing the file then make sure to clear the setuid and
651          * setgid bits if the process is not being run by root.  This keeps
652          * people from modifying setuid and setgid binaries.
653          */
654         if (!IS_NOSEC(inode))
655                 return file_remove_privs(file);
656         return 0;
657 }
658 
659 /*
660  * xfs_file_dio_aio_write - handle direct IO writes
661  *
662  * Lock the inode appropriately to prepare for and issue a direct IO write.
663  * By separating it from the buffered write path we remove all the tricky to
664  * follow locking changes and looping.
665  *
666  * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
667  * until we're sure the bytes at the new EOF have been zeroed and/or the cached
668  * pages are flushed out.
669  *
670  * In most cases the direct IO writes will be done holding IOLOCK_SHARED
671  * allowing them to be done in parallel with reads and other direct IO writes.
672  * However, if the IO is not aligned to filesystem blocks, the direct IO layer
673  * needs to do sub-block zeroing and that requires serialisation against other
674  * direct IOs to the same block. In this case we need to serialise the
675  * submission of the unaligned IOs so that we don't get racing block zeroing in
676  * the dio layer.  To avoid the problem with aio, we also need to wait for
677  * outstanding IOs to complete so that unwritten extent conversion is completed
678  * before we try to map the overlapping block. This is currently implemented by
679  * hitting it with a big hammer (i.e. inode_dio_wait()).
680  *
681  * Returns with locks held indicated by @iolock and errors indicated by
682  * negative return values.
683  */
684 STATIC ssize_t
685 xfs_file_dio_aio_write(
686         struct kiocb            *iocb,
687         struct iov_iter         *from)
688 {
689         struct file             *file = iocb->ki_filp;
690         struct address_space    *mapping = file->f_mapping;
691         struct inode            *inode = mapping->host;
692         struct xfs_inode        *ip = XFS_I(inode);
693         struct xfs_mount        *mp = ip->i_mount;
694         ssize_t                 ret = 0;
695         int                     unaligned_io = 0;
696         int                     iolock;
697         size_t                  count = iov_iter_count(from);
698         loff_t                  pos = iocb->ki_pos;
699         loff_t                  end;
700         struct iov_iter         data;
701         struct xfs_buftarg      *target = XFS_IS_REALTIME_INODE(ip) ?
702                                         mp->m_rtdev_targp : mp->m_ddev_targp;
703 
704         /* DIO must be aligned to device logical sector size */
705         if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask))
706                 return -EINVAL;
707 
708         /* "unaligned" here means not aligned to a filesystem block */
709         if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
710                 unaligned_io = 1;
711 
712         /*
713          * We don't need to take an exclusive lock unless there page cache needs
714          * to be invalidated or unaligned IO is being executed. We don't need to
715          * consider the EOF extension case here because
716          * xfs_file_aio_write_checks() will relock the inode as necessary for
717          * EOF zeroing cases and fill out the new inode size as appropriate.
718          */
719         if (unaligned_io || mapping->nrpages)
720                 iolock = XFS_IOLOCK_EXCL;
721         else
722                 iolock = XFS_IOLOCK_SHARED;
723         xfs_rw_ilock(ip, iolock);
724 
725         /*
726          * Recheck if there are cached pages that need invalidate after we got
727          * the iolock to protect against other threads adding new pages while
728          * we were waiting for the iolock.
729          */
730         if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
731                 xfs_rw_iunlock(ip, iolock);
732                 iolock = XFS_IOLOCK_EXCL;
733                 xfs_rw_ilock(ip, iolock);
734         }
735 
736         ret = xfs_file_aio_write_checks(iocb, from, &iolock);
737         if (ret)
738                 goto out;
739         count = iov_iter_count(from);
740         pos = iocb->ki_pos;
741         end = pos + count - 1;
742 
743         /*
744          * See xfs_file_read_iter() for why we do a full-file flush here.
745          */
746         if (mapping->nrpages) {
747                 ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
748                 if (ret)
749                         goto out;
750                 /*
751                  * Invalidate whole pages. This can return an error if we fail
752                  * to invalidate a page, but this should never happen on XFS.
753                  * Warn if it does fail.
754                  */
755                 ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
756                 WARN_ON_ONCE(ret);
757                 ret = 0;
758         }
759 
760         /*
761          * If we are doing unaligned IO, wait for all other IO to drain,
762          * otherwise demote the lock if we had to flush cached pages
763          */
764         if (unaligned_io)
765                 inode_dio_wait(inode);
766         else if (iolock == XFS_IOLOCK_EXCL) {
767                 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
768                 iolock = XFS_IOLOCK_SHARED;
769         }
770 
771         trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
772 
773         data = *from;
774         ret = mapping->a_ops->direct_IO(iocb, &data, pos);
775 
776         /* see generic_file_direct_write() for why this is necessary */
777         if (mapping->nrpages) {
778                 invalidate_inode_pages2_range(mapping,
779                                               pos >> PAGE_CACHE_SHIFT,
780                                               end >> PAGE_CACHE_SHIFT);
781         }
782 
783         if (ret > 0) {
784                 pos += ret;
785                 iov_iter_advance(from, ret);
786                 iocb->ki_pos = pos;
787         }
788 out:
789         xfs_rw_iunlock(ip, iolock);
790 
791         /*
792          * No fallback to buffered IO on errors for XFS. DAX can result in
793          * partial writes, but direct IO will either complete fully or fail.
794          */
795         ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip)));
796         return ret;
797 }
798 
799 STATIC ssize_t
800 xfs_file_buffered_aio_write(
801         struct kiocb            *iocb,
802         struct iov_iter         *from)
803 {
804         struct file             *file = iocb->ki_filp;
805         struct address_space    *mapping = file->f_mapping;
806         struct inode            *inode = mapping->host;
807         struct xfs_inode        *ip = XFS_I(inode);
808         ssize_t                 ret;
809         int                     enospc = 0;
810         int                     iolock = XFS_IOLOCK_EXCL;
811 
812         xfs_rw_ilock(ip, iolock);
813 
814         ret = xfs_file_aio_write_checks(iocb, from, &iolock);
815         if (ret)
816                 goto out;
817 
818         /* We can write back this queue in page reclaim */
819         current->backing_dev_info = inode_to_bdi(inode);
820 
821 write_retry:
822         trace_xfs_file_buffered_write(ip, iov_iter_count(from),
823                                       iocb->ki_pos, 0);
824         ret = generic_perform_write(file, from, iocb->ki_pos);
825         if (likely(ret >= 0))
826                 iocb->ki_pos += ret;
827 
828         /*
829          * If we hit a space limit, try to free up some lingering preallocated
830          * space before returning an error. In the case of ENOSPC, first try to
831          * write back all dirty inodes to free up some of the excess reserved
832          * metadata space. This reduces the chances that the eofblocks scan
833          * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
834          * also behaves as a filter to prevent too many eofblocks scans from
835          * running at the same time.
836          */
837         if (ret == -EDQUOT && !enospc) {
838                 enospc = xfs_inode_free_quota_eofblocks(ip);
839                 if (enospc)
840                         goto write_retry;
841         } else if (ret == -ENOSPC && !enospc) {
842                 struct xfs_eofblocks eofb = {0};
843 
844                 enospc = 1;
845                 xfs_flush_inodes(ip->i_mount);
846                 eofb.eof_scan_owner = ip->i_ino; /* for locking */
847                 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
848                 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
849                 goto write_retry;
850         }
851 
852         current->backing_dev_info = NULL;
853 out:
854         xfs_rw_iunlock(ip, iolock);
855         return ret;
856 }
857 
858 STATIC ssize_t
859 xfs_file_write_iter(
860         struct kiocb            *iocb,
861         struct iov_iter         *from)
862 {
863         struct file             *file = iocb->ki_filp;
864         struct address_space    *mapping = file->f_mapping;
865         struct inode            *inode = mapping->host;
866         struct xfs_inode        *ip = XFS_I(inode);
867         ssize_t                 ret;
868         size_t                  ocount = iov_iter_count(from);
869 
870         XFS_STATS_INC(xs_write_calls);
871 
872         if (ocount == 0)
873                 return 0;
874 
875         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
876                 return -EIO;
877 
878         if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode))
879                 ret = xfs_file_dio_aio_write(iocb, from);
880         else
881                 ret = xfs_file_buffered_aio_write(iocb, from);
882 
883         if (ret > 0) {
884                 ssize_t err;
885 
886                 XFS_STATS_ADD(xs_write_bytes, ret);
887 
888                 /* Handle various SYNC-type writes */
889                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
890                 if (err < 0)
891                         ret = err;
892         }
893         return ret;
894 }
895 
896 #define XFS_FALLOC_FL_SUPPORTED                                         \
897                 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
898                  FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |      \
899                  FALLOC_FL_INSERT_RANGE)
900 
901 STATIC long
902 xfs_file_fallocate(
903         struct file             *file,
904         int                     mode,
905         loff_t                  offset,
906         loff_t                  len)
907 {
908         struct inode            *inode = file_inode(file);
909         struct xfs_inode        *ip = XFS_I(inode);
910         long                    error;
911         enum xfs_prealloc_flags flags = 0;
912         uint                    iolock = XFS_IOLOCK_EXCL;
913         loff_t                  new_size = 0;
914         bool                    do_file_insert = 0;
915 
916         if (!S_ISREG(inode->i_mode))
917                 return -EINVAL;
918         if (mode & ~XFS_FALLOC_FL_SUPPORTED)
919                 return -EOPNOTSUPP;
920 
921         xfs_ilock(ip, iolock);
922         error = xfs_break_layouts(inode, &iolock, false);
923         if (error)
924                 goto out_unlock;
925 
926         xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
927         iolock |= XFS_MMAPLOCK_EXCL;
928 
929         if (mode & FALLOC_FL_PUNCH_HOLE) {
930                 error = xfs_free_file_space(ip, offset, len);
931                 if (error)
932                         goto out_unlock;
933         } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
934                 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
935 
936                 if (offset & blksize_mask || len & blksize_mask) {
937                         error = -EINVAL;
938                         goto out_unlock;
939                 }
940 
941                 /*
942                  * There is no need to overlap collapse range with EOF,
943                  * in which case it is effectively a truncate operation
944                  */
945                 if (offset + len >= i_size_read(inode)) {
946                         error = -EINVAL;
947                         goto out_unlock;
948                 }
949 
950                 new_size = i_size_read(inode) - len;
951 
952                 error = xfs_collapse_file_space(ip, offset, len);
953                 if (error)
954                         goto out_unlock;
955         } else if (mode & FALLOC_FL_INSERT_RANGE) {
956                 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
957 
958                 new_size = i_size_read(inode) + len;
959                 if (offset & blksize_mask || len & blksize_mask) {
960                         error = -EINVAL;
961                         goto out_unlock;
962                 }
963 
964                 /* check the new inode size does not wrap through zero */
965                 if (new_size > inode->i_sb->s_maxbytes) {
966                         error = -EFBIG;
967                         goto out_unlock;
968                 }
969 
970                 /* Offset should be less than i_size */
971                 if (offset >= i_size_read(inode)) {
972                         error = -EINVAL;
973                         goto out_unlock;
974                 }
975                 do_file_insert = 1;
976         } else {
977                 flags |= XFS_PREALLOC_SET;
978 
979                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
980                     offset + len > i_size_read(inode)) {
981                         new_size = offset + len;
982                         error = inode_newsize_ok(inode, new_size);
983                         if (error)
984                                 goto out_unlock;
985                 }
986 
987                 if (mode & FALLOC_FL_ZERO_RANGE)
988                         error = xfs_zero_file_space(ip, offset, len);
989                 else
990                         error = xfs_alloc_file_space(ip, offset, len,
991                                                      XFS_BMAPI_PREALLOC);
992                 if (error)
993                         goto out_unlock;
994         }
995 
996         if (file->f_flags & O_DSYNC)
997                 flags |= XFS_PREALLOC_SYNC;
998 
999         error = xfs_update_prealloc_flags(ip, flags);
1000         if (error)
1001                 goto out_unlock;
1002 
1003         /* Change file size if needed */
1004         if (new_size) {
1005                 struct iattr iattr;
1006 
1007                 iattr.ia_valid = ATTR_SIZE;
1008                 iattr.ia_size = new_size;
1009                 error = xfs_setattr_size(ip, &iattr);
1010                 if (error)
1011                         goto out_unlock;
1012         }
1013 
1014         /*
1015          * Perform hole insertion now that the file size has been
1016          * updated so that if we crash during the operation we don't
1017          * leave shifted extents past EOF and hence losing access to
1018          * the data that is contained within them.
1019          */
1020         if (do_file_insert)
1021                 error = xfs_insert_file_space(ip, offset, len);
1022 
1023 out_unlock:
1024         xfs_iunlock(ip, iolock);
1025         return error;
1026 }
1027 
1028 
1029 STATIC int
1030 xfs_file_open(
1031         struct inode    *inode,
1032         struct file     *file)
1033 {
1034         if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1035                 return -EFBIG;
1036         if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1037                 return -EIO;
1038         return 0;
1039 }
1040 
1041 STATIC int
1042 xfs_dir_open(
1043         struct inode    *inode,
1044         struct file     *file)
1045 {
1046         struct xfs_inode *ip = XFS_I(inode);
1047         int             mode;
1048         int             error;
1049 
1050         error = xfs_file_open(inode, file);
1051         if (error)
1052                 return error;
1053 
1054         /*
1055          * If there are any blocks, read-ahead block 0 as we're almost
1056          * certain to have the next operation be a read there.
1057          */
1058         mode = xfs_ilock_data_map_shared(ip);
1059         if (ip->i_d.di_nextents > 0)
1060                 xfs_dir3_data_readahead(ip, 0, -1);
1061         xfs_iunlock(ip, mode);
1062         return 0;
1063 }
1064 
1065 STATIC int
1066 xfs_file_release(
1067         struct inode    *inode,
1068         struct file     *filp)
1069 {
1070         return xfs_release(XFS_I(inode));
1071 }
1072 
1073 STATIC int
1074 xfs_file_readdir(
1075         struct file     *file,
1076         struct dir_context *ctx)
1077 {
1078         struct inode    *inode = file_inode(file);
1079         xfs_inode_t     *ip = XFS_I(inode);
1080         size_t          bufsize;
1081 
1082         /*
1083          * The Linux API doesn't pass down the total size of the buffer
1084          * we read into down to the filesystem.  With the filldir concept
1085          * it's not needed for correct information, but the XFS dir2 leaf
1086          * code wants an estimate of the buffer size to calculate it's
1087          * readahead window and size the buffers used for mapping to
1088          * physical blocks.
1089          *
1090          * Try to give it an estimate that's good enough, maybe at some
1091          * point we can change the ->readdir prototype to include the
1092          * buffer size.  For now we use the current glibc buffer size.
1093          */
1094         bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1095 
1096         return xfs_readdir(ip, ctx, bufsize);
1097 }
1098 
1099 /*
1100  * This type is designed to indicate the type of offset we would like
1101  * to search from page cache for xfs_seek_hole_data().
1102  */
1103 enum {
1104         HOLE_OFF = 0,
1105         DATA_OFF,
1106 };
1107 
1108 /*
1109  * Lookup the desired type of offset from the given page.
1110  *
1111  * On success, return true and the offset argument will point to the
1112  * start of the region that was found.  Otherwise this function will
1113  * return false and keep the offset argument unchanged.
1114  */
1115 STATIC bool
1116 xfs_lookup_buffer_offset(
1117         struct page             *page,
1118         loff_t                  *offset,
1119         unsigned int            type)
1120 {
1121         loff_t                  lastoff = page_offset(page);
1122         bool                    found = false;
1123         struct buffer_head      *bh, *head;
1124 
1125         bh = head = page_buffers(page);
1126         do {
1127                 /*
1128                  * Unwritten extents that have data in the page
1129                  * cache covering them can be identified by the
1130                  * BH_Unwritten state flag.  Pages with multiple
1131                  * buffers might have a mix of holes, data and
1132                  * unwritten extents - any buffer with valid
1133                  * data in it should have BH_Uptodate flag set
1134                  * on it.
1135                  */
1136                 if (buffer_unwritten(bh) ||
1137                     buffer_uptodate(bh)) {
1138                         if (type == DATA_OFF)
1139                                 found = true;
1140                 } else {
1141                         if (type == HOLE_OFF)
1142                                 found = true;
1143                 }
1144 
1145                 if (found) {
1146                         *offset = lastoff;
1147                         break;
1148                 }
1149                 lastoff += bh->b_size;
1150         } while ((bh = bh->b_this_page) != head);
1151 
1152         return found;
1153 }
1154 
1155 /*
1156  * This routine is called to find out and return a data or hole offset
1157  * from the page cache for unwritten extents according to the desired
1158  * type for xfs_seek_hole_data().
1159  *
1160  * The argument offset is used to tell where we start to search from the
1161  * page cache.  Map is used to figure out the end points of the range to
1162  * lookup pages.
1163  *
1164  * Return true if the desired type of offset was found, and the argument
1165  * offset is filled with that address.  Otherwise, return false and keep
1166  * offset unchanged.
1167  */
1168 STATIC bool
1169 xfs_find_get_desired_pgoff(
1170         struct inode            *inode,
1171         struct xfs_bmbt_irec    *map,
1172         unsigned int            type,
1173         loff_t                  *offset)
1174 {
1175         struct xfs_inode        *ip = XFS_I(inode);
1176         struct xfs_mount        *mp = ip->i_mount;
1177         struct pagevec          pvec;
1178         pgoff_t                 index;
1179         pgoff_t                 end;
1180         loff_t                  endoff;
1181         loff_t                  startoff = *offset;
1182         loff_t                  lastoff = startoff;
1183         bool                    found = false;
1184 
1185         pagevec_init(&pvec, 0);
1186 
1187         index = startoff >> PAGE_CACHE_SHIFT;
1188         endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1189         end = endoff >> PAGE_CACHE_SHIFT;
1190         do {
1191                 int             want;
1192                 unsigned        nr_pages;
1193                 unsigned int    i;
1194 
1195                 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1196                 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1197                                           want);
1198                 /*
1199                  * No page mapped into given range.  If we are searching holes
1200                  * and if this is the first time we got into the loop, it means
1201                  * that the given offset is landed in a hole, return it.
1202                  *
1203                  * If we have already stepped through some block buffers to find
1204                  * holes but they all contains data.  In this case, the last
1205                  * offset is already updated and pointed to the end of the last
1206                  * mapped page, if it does not reach the endpoint to search,
1207                  * that means there should be a hole between them.
1208                  */
1209                 if (nr_pages == 0) {
1210                         /* Data search found nothing */
1211                         if (type == DATA_OFF)
1212                                 break;
1213 
1214                         ASSERT(type == HOLE_OFF);
1215                         if (lastoff == startoff || lastoff < endoff) {
1216                                 found = true;
1217                                 *offset = lastoff;
1218                         }
1219                         break;
1220                 }
1221 
1222                 /*
1223                  * At lease we found one page.  If this is the first time we
1224                  * step into the loop, and if the first page index offset is
1225                  * greater than the given search offset, a hole was found.
1226                  */
1227                 if (type == HOLE_OFF && lastoff == startoff &&
1228                     lastoff < page_offset(pvec.pages[0])) {
1229                         found = true;
1230                         break;
1231                 }
1232 
1233                 for (i = 0; i < nr_pages; i++) {
1234                         struct page     *page = pvec.pages[i];
1235                         loff_t          b_offset;
1236 
1237                         /*
1238                          * At this point, the page may be truncated or
1239                          * invalidated (changing page->mapping to NULL),
1240                          * or even swizzled back from swapper_space to tmpfs
1241                          * file mapping. However, page->index will not change
1242                          * because we have a reference on the page.
1243                          *
1244                          * Searching done if the page index is out of range.
1245                          * If the current offset is not reaches the end of
1246                          * the specified search range, there should be a hole
1247                          * between them.
1248                          */
1249                         if (page->index > end) {
1250                                 if (type == HOLE_OFF && lastoff < endoff) {
1251                                         *offset = lastoff;
1252                                         found = true;
1253                                 }
1254                                 goto out;
1255                         }
1256 
1257                         lock_page(page);
1258                         /*
1259                          * Page truncated or invalidated(page->mapping == NULL).
1260                          * We can freely skip it and proceed to check the next
1261                          * page.
1262                          */
1263                         if (unlikely(page->mapping != inode->i_mapping)) {
1264                                 unlock_page(page);
1265                                 continue;
1266                         }
1267 
1268                         if (!page_has_buffers(page)) {
1269                                 unlock_page(page);
1270                                 continue;
1271                         }
1272 
1273                         found = xfs_lookup_buffer_offset(page, &b_offset, type);
1274                         if (found) {
1275                                 /*
1276                                  * The found offset may be less than the start
1277                                  * point to search if this is the first time to
1278                                  * come here.
1279                                  */
1280                                 *offset = max_t(loff_t, startoff, b_offset);
1281                                 unlock_page(page);
1282                                 goto out;
1283                         }
1284 
1285                         /*
1286                          * We either searching data but nothing was found, or
1287                          * searching hole but found a data buffer.  In either
1288                          * case, probably the next page contains the desired
1289                          * things, update the last offset to it so.
1290                          */
1291                         lastoff = page_offset(page) + PAGE_SIZE;
1292                         unlock_page(page);
1293                 }
1294 
1295                 /*
1296                  * The number of returned pages less than our desired, search
1297                  * done.  In this case, nothing was found for searching data,
1298                  * but we found a hole behind the last offset.
1299                  */
1300                 if (nr_pages < want) {
1301                         if (type == HOLE_OFF) {
1302                                 *offset = lastoff;
1303                                 found = true;
1304                         }
1305                         break;
1306                 }
1307 
1308                 index = pvec.pages[i - 1]->index + 1;
1309                 pagevec_release(&pvec);
1310         } while (index <= end);
1311 
1312 out:
1313         pagevec_release(&pvec);
1314         return found;
1315 }
1316 
1317 STATIC loff_t
1318 xfs_seek_hole_data(
1319         struct file             *file,
1320         loff_t                  start,
1321         int                     whence)
1322 {
1323         struct inode            *inode = file->f_mapping->host;
1324         struct xfs_inode        *ip = XFS_I(inode);
1325         struct xfs_mount        *mp = ip->i_mount;
1326         loff_t                  uninitialized_var(offset);
1327         xfs_fsize_t             isize;
1328         xfs_fileoff_t           fsbno;
1329         xfs_filblks_t           end;
1330         uint                    lock;
1331         int                     error;
1332 
1333         if (XFS_FORCED_SHUTDOWN(mp))
1334                 return -EIO;
1335 
1336         lock = xfs_ilock_data_map_shared(ip);
1337 
1338         isize = i_size_read(inode);
1339         if (start >= isize) {
1340                 error = -ENXIO;
1341                 goto out_unlock;
1342         }
1343 
1344         /*
1345          * Try to read extents from the first block indicated
1346          * by fsbno to the end block of the file.
1347          */
1348         fsbno = XFS_B_TO_FSBT(mp, start);
1349         end = XFS_B_TO_FSB(mp, isize);
1350 
1351         for (;;) {
1352                 struct xfs_bmbt_irec    map[2];
1353                 int                     nmap = 2;
1354                 unsigned int            i;
1355 
1356                 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1357                                        XFS_BMAPI_ENTIRE);
1358                 if (error)
1359                         goto out_unlock;
1360 
1361                 /* No extents at given offset, must be beyond EOF */
1362                 if (nmap == 0) {
1363                         error = -ENXIO;
1364                         goto out_unlock;
1365                 }
1366 
1367                 for (i = 0; i < nmap; i++) {
1368                         offset = max_t(loff_t, start,
1369                                        XFS_FSB_TO_B(mp, map[i].br_startoff));
1370 
1371                         /* Landed in the hole we wanted? */
1372                         if (whence == SEEK_HOLE &&
1373                             map[i].br_startblock == HOLESTARTBLOCK)
1374                                 goto out;
1375 
1376                         /* Landed in the data extent we wanted? */
1377                         if (whence == SEEK_DATA &&
1378                             (map[i].br_startblock == DELAYSTARTBLOCK ||
1379                              (map[i].br_state == XFS_EXT_NORM &&
1380                               !isnullstartblock(map[i].br_startblock))))
1381                                 goto out;
1382 
1383                         /*
1384                          * Landed in an unwritten extent, try to search
1385                          * for hole or data from page cache.
1386                          */
1387                         if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1388                                 if (xfs_find_get_desired_pgoff(inode, &map[i],
1389                                       whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1390                                                         &offset))
1391                                         goto out;
1392                         }
1393                 }
1394 
1395                 /*
1396                  * We only received one extent out of the two requested. This
1397                  * means we've hit EOF and didn't find what we are looking for.
1398                  */
1399                 if (nmap == 1) {
1400                         /*
1401                          * If we were looking for a hole, set offset to
1402                          * the end of the file (i.e., there is an implicit
1403                          * hole at the end of any file).
1404                          */
1405                         if (whence == SEEK_HOLE) {
1406                                 offset = isize;
1407                                 break;
1408                         }
1409                         /*
1410                          * If we were looking for data, it's nowhere to be found
1411                          */
1412                         ASSERT(whence == SEEK_DATA);
1413                         error = -ENXIO;
1414                         goto out_unlock;
1415                 }
1416 
1417                 ASSERT(i > 1);
1418 
1419                 /*
1420                  * Nothing was found, proceed to the next round of search
1421                  * if the next reading offset is not at or beyond EOF.
1422                  */
1423                 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1424                 start = XFS_FSB_TO_B(mp, fsbno);
1425                 if (start >= isize) {
1426                         if (whence == SEEK_HOLE) {
1427                                 offset = isize;
1428                                 break;
1429                         }
1430                         ASSERT(whence == SEEK_DATA);
1431                         error = -ENXIO;
1432                         goto out_unlock;
1433                 }
1434         }
1435 
1436 out:
1437         /*
1438          * If at this point we have found the hole we wanted, the returned
1439          * offset may be bigger than the file size as it may be aligned to
1440          * page boundary for unwritten extents.  We need to deal with this
1441          * situation in particular.
1442          */
1443         if (whence == SEEK_HOLE)
1444                 offset = min_t(loff_t, offset, isize);
1445         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1446 
1447 out_unlock:
1448         xfs_iunlock(ip, lock);
1449 
1450         if (error)
1451                 return error;
1452         return offset;
1453 }
1454 
1455 STATIC loff_t
1456 xfs_file_llseek(
1457         struct file     *file,
1458         loff_t          offset,
1459         int             whence)
1460 {
1461         switch (whence) {
1462         case SEEK_END:
1463         case SEEK_CUR:
1464         case SEEK_SET:
1465                 return generic_file_llseek(file, offset, whence);
1466         case SEEK_HOLE:
1467         case SEEK_DATA:
1468                 return xfs_seek_hole_data(file, offset, whence);
1469         default:
1470                 return -EINVAL;
1471         }
1472 }
1473 
1474 /*
1475  * Locking for serialisation of IO during page faults. This results in a lock
1476  * ordering of:
1477  *
1478  * mmap_sem (MM)
1479  *   sb_start_pagefault(vfs, freeze)
1480  *     i_mmap_lock (XFS - truncate serialisation)
1481  *       page_lock (MM)
1482  *         i_lock (XFS - extent map serialisation)
1483  */
1484 
1485 /*
1486  * mmap()d file has taken write protection fault and is being made writable. We
1487  * can set the page state up correctly for a writable page, which means we can
1488  * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1489  * mapping.
1490  */
1491 STATIC int
1492 xfs_filemap_page_mkwrite(
1493         struct vm_area_struct   *vma,
1494         struct vm_fault         *vmf)
1495 {
1496         struct inode            *inode = file_inode(vma->vm_file);
1497         int                     ret;
1498 
1499         trace_xfs_filemap_page_mkwrite(XFS_I(inode));
1500 
1501         sb_start_pagefault(inode->i_sb);
1502         file_update_time(vma->vm_file);
1503         xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1504 
1505         if (IS_DAX(inode)) {
1506                 ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_direct,
1507                                     xfs_end_io_dax_write);
1508         } else {
1509                 ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks);
1510                 ret = block_page_mkwrite_return(ret);
1511         }
1512 
1513         xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1514         sb_end_pagefault(inode->i_sb);
1515 
1516         return ret;
1517 }
1518 
1519 STATIC int
1520 xfs_filemap_fault(
1521         struct vm_area_struct   *vma,
1522         struct vm_fault         *vmf)
1523 {
1524         struct inode            *inode = file_inode(vma->vm_file);
1525         int                     ret;
1526 
1527         trace_xfs_filemap_fault(XFS_I(inode));
1528 
1529         /* DAX can shortcut the normal fault path on write faults! */
1530         if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
1531                 return xfs_filemap_page_mkwrite(vma, vmf);
1532 
1533         xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1534         if (IS_DAX(inode)) {
1535                 /*
1536                  * we do not want to trigger unwritten extent conversion on read
1537                  * faults - that is unnecessary overhead and would also require
1538                  * changes to xfs_get_blocks_direct() to map unwritten extent
1539                  * ioend for conversion on read-only mappings.
1540                  */
1541                 ret = __dax_fault(vma, vmf, xfs_get_blocks_direct, NULL);
1542         } else
1543                 ret = filemap_fault(vma, vmf);
1544         xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1545 
1546         return ret;
1547 }
1548 
1549 STATIC int
1550 xfs_filemap_pmd_fault(
1551         struct vm_area_struct   *vma,
1552         unsigned long           addr,
1553         pmd_t                   *pmd,
1554         unsigned int            flags)
1555 {
1556         struct inode            *inode = file_inode(vma->vm_file);
1557         struct xfs_inode        *ip = XFS_I(inode);
1558         int                     ret;
1559 
1560         if (!IS_DAX(inode))
1561                 return VM_FAULT_FALLBACK;
1562 
1563         trace_xfs_filemap_pmd_fault(ip);
1564 
1565         sb_start_pagefault(inode->i_sb);
1566         file_update_time(vma->vm_file);
1567         xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1568         ret = __dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_direct,
1569                                     xfs_end_io_dax_write);
1570         xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1571         sb_end_pagefault(inode->i_sb);
1572 
1573         return ret;
1574 }
1575 
1576 static const struct vm_operations_struct xfs_file_vm_ops = {
1577         .fault          = xfs_filemap_fault,
1578         .pmd_fault      = xfs_filemap_pmd_fault,
1579         .map_pages      = filemap_map_pages,
1580         .page_mkwrite   = xfs_filemap_page_mkwrite,
1581 };
1582 
1583 STATIC int
1584 xfs_file_mmap(
1585         struct file     *filp,
1586         struct vm_area_struct *vma)
1587 {
1588         file_accessed(filp);
1589         vma->vm_ops = &xfs_file_vm_ops;
1590         if (IS_DAX(file_inode(filp)))
1591                 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1592         return 0;
1593 }
1594 
1595 const struct file_operations xfs_file_operations = {
1596         .llseek         = xfs_file_llseek,
1597         .read_iter      = xfs_file_read_iter,
1598         .write_iter     = xfs_file_write_iter,
1599         .splice_read    = xfs_file_splice_read,
1600         .splice_write   = iter_file_splice_write,
1601         .unlocked_ioctl = xfs_file_ioctl,
1602 #ifdef CONFIG_COMPAT
1603         .compat_ioctl   = xfs_file_compat_ioctl,
1604 #endif
1605         .mmap           = xfs_file_mmap,
1606         .open           = xfs_file_open,
1607         .release        = xfs_file_release,
1608         .fsync          = xfs_file_fsync,
1609         .fallocate      = xfs_file_fallocate,
1610 };
1611 
1612 const struct file_operations xfs_dir_file_operations = {
1613         .open           = xfs_dir_open,
1614         .read           = generic_read_dir,
1615         .iterate        = xfs_file_readdir,
1616         .llseek         = generic_file_llseek,
1617         .unlocked_ioctl = xfs_file_ioctl,
1618 #ifdef CONFIG_COMPAT
1619         .compat_ioctl   = xfs_file_compat_ioctl,
1620 #endif
1621         .fsync          = xfs_dir_fsync,
1622 };
1623 

~ [ 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