0b5c9b6774c86b13bc030aa9f9a51c9a7fcc717c
[linux-drm-fsl-dcu.git] / fs / xfs / linux-2.6 / xfs_lrw.c
1 /*
2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_bit.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_bmap.h"
39 #include "xfs_btree.h"
40 #include "xfs_ialloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
44 #include "xfs_rw.h"
45 #include "xfs_acl.h"
46 #include "xfs_attr.h"
47 #include "xfs_inode_item.h"
48 #include "xfs_buf_item.h"
49 #include "xfs_utils.h"
50 #include "xfs_iomap.h"
51
52 #include <linux/capability.h>
53 #include <linux/writeback.h>
54
55
56 #if defined(XFS_RW_TRACE)
57 void
58 xfs_rw_enter_trace(
59         int                     tag,
60         xfs_iocore_t            *io,
61         void                    *data,
62         size_t                  segs,
63         loff_t                  offset,
64         int                     ioflags)
65 {
66         xfs_inode_t     *ip = XFS_IO_INODE(io);
67
68         if (ip->i_rwtrace == NULL)
69                 return;
70         ktrace_enter(ip->i_rwtrace,
71                 (void *)(unsigned long)tag,
72                 (void *)ip,
73                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
74                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
75                 (void *)data,
76                 (void *)((unsigned long)segs),
77                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
78                 (void *)((unsigned long)(offset & 0xffffffff)),
79                 (void *)((unsigned long)ioflags),
80                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
81                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
82                 (void *)((unsigned long)current_pid()),
83                 (void *)NULL,
84                 (void *)NULL,
85                 (void *)NULL,
86                 (void *)NULL);
87 }
88
89 void
90 xfs_inval_cached_trace(
91         xfs_iocore_t    *io,
92         xfs_off_t       offset,
93         xfs_off_t       len,
94         xfs_off_t       first,
95         xfs_off_t       last)
96 {
97         xfs_inode_t     *ip = XFS_IO_INODE(io);
98
99         if (ip->i_rwtrace == NULL)
100                 return;
101         ktrace_enter(ip->i_rwtrace,
102                 (void *)(__psint_t)XFS_INVAL_CACHED,
103                 (void *)ip,
104                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
105                 (void *)((unsigned long)(offset & 0xffffffff)),
106                 (void *)((unsigned long)((len >> 32) & 0xffffffff)),
107                 (void *)((unsigned long)(len & 0xffffffff)),
108                 (void *)((unsigned long)((first >> 32) & 0xffffffff)),
109                 (void *)((unsigned long)(first & 0xffffffff)),
110                 (void *)((unsigned long)((last >> 32) & 0xffffffff)),
111                 (void *)((unsigned long)(last & 0xffffffff)),
112                 (void *)((unsigned long)current_pid()),
113                 (void *)NULL,
114                 (void *)NULL,
115                 (void *)NULL,
116                 (void *)NULL,
117                 (void *)NULL);
118 }
119 #endif
120
121 /*
122  *      xfs_iozero
123  *
124  *      xfs_iozero clears the specified range of buffer supplied,
125  *      and marks all the affected blocks as valid and modified.  If
126  *      an affected block is not allocated, it will be allocated.  If
127  *      an affected block is not completely overwritten, and is not
128  *      valid before the operation, it will be read from disk before
129  *      being partially zeroed.
130  */
131 STATIC int
132 xfs_iozero(
133         struct inode            *ip,    /* inode                        */
134         loff_t                  pos,    /* offset in file               */
135         size_t                  count)  /* size of data to zero         */
136 {
137         unsigned                bytes;
138         struct page             *page;
139         struct address_space    *mapping;
140         char                    *kaddr;
141         int                     status;
142
143         mapping = ip->i_mapping;
144         do {
145                 unsigned long index, offset;
146
147                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
148                 index = pos >> PAGE_CACHE_SHIFT;
149                 bytes = PAGE_CACHE_SIZE - offset;
150                 if (bytes > count)
151                         bytes = count;
152
153                 status = -ENOMEM;
154                 page = grab_cache_page(mapping, index);
155                 if (!page)
156                         break;
157
158                 kaddr = kmap(page);
159                 status = mapping->a_ops->prepare_write(NULL, page, offset,
160                                                         offset + bytes);
161                 if (status) {
162                         goto unlock;
163                 }
164
165                 memset((void *) (kaddr + offset), 0, bytes);
166                 flush_dcache_page(page);
167                 status = mapping->a_ops->commit_write(NULL, page, offset,
168                                                         offset + bytes);
169                 if (!status) {
170                         pos += bytes;
171                         count -= bytes;
172                 }
173
174 unlock:
175                 kunmap(page);
176                 unlock_page(page);
177                 page_cache_release(page);
178                 if (status)
179                         break;
180         } while (count);
181
182         return (-status);
183 }
184
185 ssize_t                 /* bytes read, or (-)  error */
186 xfs_read(
187         bhv_desc_t              *bdp,
188         struct kiocb            *iocb,
189         const struct iovec      *iovp,
190         unsigned int            segs,
191         loff_t                  *offset,
192         int                     ioflags,
193         cred_t                  *credp)
194 {
195         struct file             *file = iocb->ki_filp;
196         struct inode            *inode = file->f_mapping->host;
197         size_t                  size = 0;
198         ssize_t                 ret;
199         xfs_fsize_t             n;
200         xfs_inode_t             *ip;
201         xfs_mount_t             *mp;
202         bhv_vnode_t             *vp;
203         unsigned long           seg;
204
205         ip = XFS_BHVTOI(bdp);
206         vp = BHV_TO_VNODE(bdp);
207         mp = ip->i_mount;
208
209         XFS_STATS_INC(xs_read_calls);
210
211         /* START copy & waste from filemap.c */
212         for (seg = 0; seg < segs; seg++) {
213                 const struct iovec *iv = &iovp[seg];
214
215                 /*
216                  * If any segment has a negative length, or the cumulative
217                  * length ever wraps negative then return -EINVAL.
218                  */
219                 size += iv->iov_len;
220                 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
221                         return XFS_ERROR(-EINVAL);
222         }
223         /* END copy & waste from filemap.c */
224
225         if (unlikely(ioflags & IO_ISDIRECT)) {
226                 xfs_buftarg_t   *target =
227                         (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
228                                 mp->m_rtdev_targp : mp->m_ddev_targp;
229                 if ((*offset & target->bt_smask) ||
230                     (size & target->bt_smask)) {
231                         if (*offset == ip->i_d.di_size) {
232                                 return (0);
233                         }
234                         return -XFS_ERROR(EINVAL);
235                 }
236         }
237
238         n = XFS_MAXIOFFSET(mp) - *offset;
239         if ((n <= 0) || (size == 0))
240                 return 0;
241
242         if (n < size)
243                 size = n;
244
245         if (XFS_FORCED_SHUTDOWN(mp))
246                 return -EIO;
247
248         if (unlikely(ioflags & IO_ISDIRECT))
249                 mutex_lock(&inode->i_mutex);
250         xfs_ilock(ip, XFS_IOLOCK_SHARED);
251
252         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
253             !(ioflags & IO_INVIS)) {
254                 bhv_vrwlock_t locktype = VRWLOCK_READ;
255                 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
256
257                 ret = -XFS_SEND_DATA(mp, DM_EVENT_READ,
258                                         BHV_TO_VNODE(bdp), *offset, size,
259                                         dmflags, &locktype);
260                 if (ret) {
261                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
262                         if (unlikely(ioflags & IO_ISDIRECT))
263                                 mutex_unlock(&inode->i_mutex);
264                         return ret;
265                 }
266         }
267
268         if (unlikely(ioflags & IO_ISDIRECT)) {
269                 if (VN_CACHED(vp))
270                         bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
271                                                  -1, FI_REMAPF_LOCKED);
272                 mutex_unlock(&inode->i_mutex);
273         }
274
275         xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
276                                 (void *)iovp, segs, *offset, ioflags);
277
278         iocb->ki_pos = *offset;
279         ret = generic_file_aio_read(iocb, iovp, segs, *offset);
280         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
281                 ret = wait_on_sync_kiocb(iocb);
282         if (ret > 0)
283                 XFS_STATS_ADD(xs_read_bytes, ret);
284
285         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
286         return ret;
287 }
288
289 ssize_t
290 xfs_sendfile(
291         bhv_desc_t              *bdp,
292         struct file             *filp,
293         loff_t                  *offset,
294         int                     ioflags,
295         size_t                  count,
296         read_actor_t            actor,
297         void                    *target,
298         cred_t                  *credp)
299 {
300         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
301         xfs_mount_t             *mp = ip->i_mount;
302         ssize_t                 ret;
303
304         XFS_STATS_INC(xs_read_calls);
305         if (XFS_FORCED_SHUTDOWN(mp))
306                 return -EIO;
307
308         xfs_ilock(ip, XFS_IOLOCK_SHARED);
309
310         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
311             (!(ioflags & IO_INVIS))) {
312                 bhv_vrwlock_t locktype = VRWLOCK_READ;
313                 int error;
314
315                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
316                                       *offset, count,
317                                       FILP_DELAY_FLAG(filp), &locktype);
318                 if (error) {
319                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
320                         return -error;
321                 }
322         }
323         xfs_rw_enter_trace(XFS_SENDFILE_ENTER, &ip->i_iocore,
324                    (void *)(unsigned long)target, count, *offset, ioflags);
325         ret = generic_file_sendfile(filp, offset, count, actor, target);
326         if (ret > 0)
327                 XFS_STATS_ADD(xs_read_bytes, ret);
328
329         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
330         return ret;
331 }
332
333 ssize_t
334 xfs_splice_read(
335         bhv_desc_t              *bdp,
336         struct file             *infilp,
337         loff_t                  *ppos,
338         struct pipe_inode_info  *pipe,
339         size_t                  count,
340         int                     flags,
341         int                     ioflags,
342         cred_t                  *credp)
343 {
344         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
345         xfs_mount_t             *mp = ip->i_mount;
346         ssize_t                 ret;
347
348         XFS_STATS_INC(xs_read_calls);
349         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
350                 return -EIO;
351
352         xfs_ilock(ip, XFS_IOLOCK_SHARED);
353
354         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
355             (!(ioflags & IO_INVIS))) {
356                 bhv_vrwlock_t locktype = VRWLOCK_READ;
357                 int error;
358
359                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
360                                         *ppos, count,
361                                         FILP_DELAY_FLAG(infilp), &locktype);
362                 if (error) {
363                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
364                         return -error;
365                 }
366         }
367         xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER, &ip->i_iocore,
368                            pipe, count, *ppos, ioflags);
369         ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
370         if (ret > 0)
371                 XFS_STATS_ADD(xs_read_bytes, ret);
372
373         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
374         return ret;
375 }
376
377 ssize_t
378 xfs_splice_write(
379         bhv_desc_t              *bdp,
380         struct pipe_inode_info  *pipe,
381         struct file             *outfilp,
382         loff_t                  *ppos,
383         size_t                  count,
384         int                     flags,
385         int                     ioflags,
386         cred_t                  *credp)
387 {
388         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
389         xfs_mount_t             *mp = ip->i_mount;
390         ssize_t                 ret;
391         struct inode            *inode = outfilp->f_mapping->host;
392         xfs_fsize_t             isize;
393
394         XFS_STATS_INC(xs_write_calls);
395         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
396                 return -EIO;
397
398         xfs_ilock(ip, XFS_IOLOCK_EXCL);
399
400         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_WRITE) &&
401             (!(ioflags & IO_INVIS))) {
402                 bhv_vrwlock_t locktype = VRWLOCK_WRITE;
403                 int error;
404
405                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, BHV_TO_VNODE(bdp),
406                                         *ppos, count,
407                                         FILP_DELAY_FLAG(outfilp), &locktype);
408                 if (error) {
409                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
410                         return -error;
411                 }
412         }
413         xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER, &ip->i_iocore,
414                            pipe, count, *ppos, ioflags);
415         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
416         if (ret > 0)
417                 XFS_STATS_ADD(xs_write_bytes, ret);
418
419         isize = i_size_read(inode);
420         if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize))
421                 *ppos = isize;
422
423         if (*ppos > ip->i_d.di_size) {
424                 xfs_ilock(ip, XFS_ILOCK_EXCL);
425                 if (*ppos > ip->i_d.di_size) {
426                         ip->i_d.di_size = *ppos;
427                         i_size_write(inode, *ppos);
428                         ip->i_update_core = 1;
429                         ip->i_update_size = 1;
430                 }
431                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
432         }
433         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
434         return ret;
435 }
436
437 /*
438  * This routine is called to handle zeroing any space in the last
439  * block of the file that is beyond the EOF.  We do this since the
440  * size is being increased without writing anything to that block
441  * and we don't want anyone to read the garbage on the disk.
442  */
443 STATIC int                              /* error (positive) */
444 xfs_zero_last_block(
445         struct inode    *ip,
446         xfs_iocore_t    *io,
447         xfs_fsize_t     offset,
448         xfs_fsize_t     isize)
449 {
450         xfs_fileoff_t   last_fsb;
451         xfs_mount_t     *mp = io->io_mount;
452         int             nimaps;
453         int             zero_offset;
454         int             zero_len;
455         int             error = 0;
456         xfs_bmbt_irec_t imap;
457
458         ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
459
460         zero_offset = XFS_B_FSB_OFFSET(mp, isize);
461         if (zero_offset == 0) {
462                 /*
463                  * There are no extra bytes in the last block on disk to
464                  * zero, so return.
465                  */
466                 return 0;
467         }
468
469         last_fsb = XFS_B_TO_FSBT(mp, isize);
470         nimaps = 1;
471         error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
472                           &nimaps, NULL, NULL);
473         if (error) {
474                 return error;
475         }
476         ASSERT(nimaps > 0);
477         /*
478          * If the block underlying isize is just a hole, then there
479          * is nothing to zero.
480          */
481         if (imap.br_startblock == HOLESTARTBLOCK) {
482                 return 0;
483         }
484         /*
485          * Zero the part of the last block beyond the EOF, and write it
486          * out sync.  We need to drop the ilock while we do this so we
487          * don't deadlock when the buffer cache calls back to us.
488          */
489         XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
490
491         zero_len = mp->m_sb.sb_blocksize - zero_offset;
492         if (isize + zero_len > offset)
493                 zero_len = offset - isize;
494         error = xfs_iozero(ip, isize, zero_len);
495
496         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
497         ASSERT(error >= 0);
498         return error;
499 }
500
501 /*
502  * Zero any on disk space between the current EOF and the new,
503  * larger EOF.  This handles the normal case of zeroing the remainder
504  * of the last block in the file and the unusual case of zeroing blocks
505  * out beyond the size of the file.  This second case only happens
506  * with fixed size extents and when the system crashes before the inode
507  * size was updated but after blocks were allocated.  If fill is set,
508  * then any holes in the range are filled and zeroed.  If not, the holes
509  * are left alone as holes.
510  */
511
512 int                                     /* error (positive) */
513 xfs_zero_eof(
514         bhv_vnode_t     *vp,
515         xfs_iocore_t    *io,
516         xfs_off_t       offset,         /* starting I/O offset */
517         xfs_fsize_t     isize)          /* current inode size */
518 {
519         struct inode    *ip = vn_to_inode(vp);
520         xfs_fileoff_t   start_zero_fsb;
521         xfs_fileoff_t   end_zero_fsb;
522         xfs_fileoff_t   zero_count_fsb;
523         xfs_fileoff_t   last_fsb;
524         xfs_fileoff_t   zero_off;
525         xfs_fsize_t     zero_len;
526         xfs_mount_t     *mp = io->io_mount;
527         int             nimaps;
528         int             error = 0;
529         xfs_bmbt_irec_t imap;
530
531         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
532         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
533         ASSERT(offset > isize);
534
535         /*
536          * First handle zeroing the block on which isize resides.
537          * We only zero a part of that block so it is handled specially.
538          */
539         error = xfs_zero_last_block(ip, io, offset, isize);
540         if (error) {
541                 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
542                 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
543                 return error;
544         }
545
546         /*
547          * Calculate the range between the new size and the old
548          * where blocks needing to be zeroed may exist.  To get the
549          * block where the last byte in the file currently resides,
550          * we need to subtract one from the size and truncate back
551          * to a block boundary.  We subtract 1 in case the size is
552          * exactly on a block boundary.
553          */
554         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
555         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
556         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
557         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
558         if (last_fsb == end_zero_fsb) {
559                 /*
560                  * The size was only incremented on its last block.
561                  * We took care of that above, so just return.
562                  */
563                 return 0;
564         }
565
566         ASSERT(start_zero_fsb <= end_zero_fsb);
567         while (start_zero_fsb <= end_zero_fsb) {
568                 nimaps = 1;
569                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
570                 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
571                                   0, NULL, 0, &imap, &nimaps, NULL, NULL);
572                 if (error) {
573                         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
574                         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
575                         return error;
576                 }
577                 ASSERT(nimaps > 0);
578
579                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
580                     imap.br_startblock == HOLESTARTBLOCK) {
581                         /*
582                          * This loop handles initializing pages that were
583                          * partially initialized by the code below this
584                          * loop. It basically zeroes the part of the page
585                          * that sits on a hole and sets the page as P_HOLE
586                          * and calls remapf if it is a mapped file.
587                          */
588                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
589                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
590                         continue;
591                 }
592
593                 /*
594                  * There are blocks we need to zero.
595                  * Drop the inode lock while we're doing the I/O.
596                  * We'll still have the iolock to protect us.
597                  */
598                 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
599
600                 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
601                 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
602
603                 if ((zero_off + zero_len) > offset)
604                         zero_len = offset - zero_off;
605
606                 error = xfs_iozero(ip, zero_off, zero_len);
607                 if (error) {
608                         goto out_lock;
609                 }
610
611                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
612                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
613
614                 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
615         }
616
617         return 0;
618
619 out_lock:
620
621         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
622         ASSERT(error >= 0);
623         return error;
624 }
625
626 ssize_t                         /* bytes written, or (-) error */
627 xfs_write(
628         bhv_desc_t              *bdp,
629         struct kiocb            *iocb,
630         const struct iovec      *iovp,
631         unsigned int            nsegs,
632         loff_t                  *offset,
633         int                     ioflags,
634         cred_t                  *credp)
635 {
636         struct file             *file = iocb->ki_filp;
637         struct address_space    *mapping = file->f_mapping;
638         struct inode            *inode = mapping->host;
639         unsigned long           segs = nsegs;
640         xfs_inode_t             *xip;
641         xfs_mount_t             *mp;
642         ssize_t                 ret = 0, error = 0;
643         xfs_fsize_t             isize, new_size;
644         xfs_iocore_t            *io;
645         bhv_vnode_t             *vp;
646         unsigned long           seg;
647         int                     iolock;
648         int                     eventsent = 0;
649         bhv_vrwlock_t           locktype;
650         size_t                  ocount = 0, count;
651         loff_t                  pos;
652         int                     need_i_mutex = 1, need_flush = 0;
653
654         XFS_STATS_INC(xs_write_calls);
655
656         vp = BHV_TO_VNODE(bdp);
657         xip = XFS_BHVTOI(bdp);
658
659         for (seg = 0; seg < segs; seg++) {
660                 const struct iovec *iv = &iovp[seg];
661
662                 /*
663                  * If any segment has a negative length, or the cumulative
664                  * length ever wraps negative then return -EINVAL.
665                  */
666                 ocount += iv->iov_len;
667                 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
668                         return -EINVAL;
669                 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
670                         continue;
671                 if (seg == 0)
672                         return -EFAULT;
673                 segs = seg;
674                 ocount -= iv->iov_len;  /* This segment is no good */
675                 break;
676         }
677
678         count = ocount;
679         pos = *offset;
680
681         if (count == 0)
682                 return 0;
683
684         io = &xip->i_iocore;
685         mp = io->io_mount;
686
687         vfs_wait_for_freeze(vp->v_vfsp, SB_FREEZE_WRITE);
688
689         if (XFS_FORCED_SHUTDOWN(mp))
690                 return -EIO;
691
692         if (ioflags & IO_ISDIRECT) {
693                 xfs_buftarg_t   *target =
694                         (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
695                                 mp->m_rtdev_targp : mp->m_ddev_targp;
696
697                 if ((pos & target->bt_smask) || (count & target->bt_smask))
698                         return XFS_ERROR(-EINVAL);
699
700                 if (!VN_CACHED(vp) && pos < i_size_read(inode))
701                         need_i_mutex = 0;
702
703                 if (VN_CACHED(vp))
704                         need_flush = 1;
705         }
706
707 relock:
708         if (need_i_mutex) {
709                 iolock = XFS_IOLOCK_EXCL;
710                 locktype = VRWLOCK_WRITE;
711
712                 mutex_lock(&inode->i_mutex);
713         } else {
714                 iolock = XFS_IOLOCK_SHARED;
715                 locktype = VRWLOCK_WRITE_DIRECT;
716         }
717
718         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
719
720         isize = i_size_read(inode);
721
722         if (file->f_flags & O_APPEND)
723                 *offset = isize;
724
725 start:
726         error = -generic_write_checks(file, &pos, &count,
727                                         S_ISBLK(inode->i_mode));
728         if (error) {
729                 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
730                 goto out_unlock_mutex;
731         }
732
733         new_size = pos + count;
734         if (new_size > isize)
735                 io->io_new_size = new_size;
736
737         if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
738             !(ioflags & IO_INVIS) && !eventsent)) {
739                 loff_t          savedsize = pos;
740                 int             dmflags = FILP_DELAY_FLAG(file);
741
742                 if (need_i_mutex)
743                         dmflags |= DM_FLAGS_IMUX;
744
745                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
746                 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
747                                       pos, count,
748                                       dmflags, &locktype);
749                 if (error) {
750                         xfs_iunlock(xip, iolock);
751                         goto out_unlock_mutex;
752                 }
753                 xfs_ilock(xip, XFS_ILOCK_EXCL);
754                 eventsent = 1;
755
756                 /*
757                  * The iolock was dropped and reacquired in XFS_SEND_DATA
758                  * so we have to recheck the size when appending.
759                  * We will only "goto start;" once, since having sent the
760                  * event prevents another call to XFS_SEND_DATA, which is
761                  * what allows the size to change in the first place.
762                  */
763                 if ((file->f_flags & O_APPEND) && savedsize != isize) {
764                         pos = isize = xip->i_d.di_size;
765                         goto start;
766                 }
767         }
768
769         if (likely(!(ioflags & IO_INVIS))) {
770                 file_update_time(file);
771                 xfs_ichgtime_fast(xip, inode,
772                                   XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
773         }
774
775         /*
776          * If the offset is beyond the size of the file, we have a couple
777          * of things to do. First, if there is already space allocated
778          * we need to either create holes or zero the disk or ...
779          *
780          * If there is a page where the previous size lands, we need
781          * to zero it out up to the new size.
782          */
783
784         if (pos > isize) {
785                 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, pos, isize);
786                 if (error) {
787                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
788                         goto out_unlock_mutex;
789                 }
790         }
791         xfs_iunlock(xip, XFS_ILOCK_EXCL);
792
793         /*
794          * If we're writing the file then make sure to clear the
795          * setuid and setgid bits if the process is not being run
796          * by root.  This keeps people from modifying setuid and
797          * setgid binaries.
798          */
799
800         if (((xip->i_d.di_mode & S_ISUID) ||
801             ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
802                 (S_ISGID | S_IXGRP))) &&
803              !capable(CAP_FSETID)) {
804                 error = xfs_write_clear_setuid(xip);
805                 if (likely(!error))
806                         error = -remove_suid(file->f_path.dentry);
807                 if (unlikely(error)) {
808                         xfs_iunlock(xip, iolock);
809                         goto out_unlock_mutex;
810                 }
811         }
812
813 retry:
814         /* We can write back this queue in page reclaim */
815         current->backing_dev_info = mapping->backing_dev_info;
816
817         if ((ioflags & IO_ISDIRECT)) {
818                 if (need_flush) {
819                         xfs_inval_cached_trace(io, pos, -1,
820                                         ctooff(offtoct(pos)), -1);
821                         bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
822                                         -1, FI_REMAPF_LOCKED);
823                 }
824
825                 if (need_i_mutex) {
826                         /* demote the lock now the cached pages are gone */
827                         XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
828                         mutex_unlock(&inode->i_mutex);
829
830                         iolock = XFS_IOLOCK_SHARED;
831                         locktype = VRWLOCK_WRITE_DIRECT;
832                         need_i_mutex = 0;
833                 }
834
835                 xfs_rw_enter_trace(XFS_DIOWR_ENTER, io, (void *)iovp, segs,
836                                 *offset, ioflags);
837                 ret = generic_file_direct_write(iocb, iovp,
838                                 &segs, pos, offset, count, ocount);
839
840                 /*
841                  * direct-io write to a hole: fall through to buffered I/O
842                  * for completing the rest of the request.
843                  */
844                 if (ret >= 0 && ret != count) {
845                         XFS_STATS_ADD(xs_write_bytes, ret);
846
847                         pos += ret;
848                         count -= ret;
849
850                         need_i_mutex = 1;
851                         ioflags &= ~IO_ISDIRECT;
852                         xfs_iunlock(xip, iolock);
853                         goto relock;
854                 }
855         } else {
856                 xfs_rw_enter_trace(XFS_WRITE_ENTER, io, (void *)iovp, segs,
857                                 *offset, ioflags);
858                 ret = generic_file_buffered_write(iocb, iovp, segs,
859                                 pos, offset, count, ret);
860         }
861
862         current->backing_dev_info = NULL;
863
864         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
865                 ret = wait_on_sync_kiocb(iocb);
866
867         if ((ret == -ENOSPC) &&
868             DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
869             !(ioflags & IO_INVIS)) {
870
871                 xfs_rwunlock(bdp, locktype);
872                 if (need_i_mutex)
873                         mutex_unlock(&inode->i_mutex);
874                 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
875                                 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
876                                 0, 0, 0); /* Delay flag intentionally  unused */
877                 if (error)
878                         goto out_nounlocks;
879                 if (need_i_mutex)
880                         mutex_lock(&inode->i_mutex);
881                 xfs_rwlock(bdp, locktype);
882                 pos = xip->i_d.di_size;
883                 ret = 0;
884                 goto retry;
885         }
886
887         isize = i_size_read(inode);
888         if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
889                 *offset = isize;
890
891         if (*offset > xip->i_d.di_size) {
892                 xfs_ilock(xip, XFS_ILOCK_EXCL);
893                 if (*offset > xip->i_d.di_size) {
894                         xip->i_d.di_size = *offset;
895                         i_size_write(inode, *offset);
896                         xip->i_update_core = 1;
897                         xip->i_update_size = 1;
898                 }
899                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
900         }
901
902         error = -ret;
903         if (ret <= 0)
904                 goto out_unlock_internal;
905
906         XFS_STATS_ADD(xs_write_bytes, ret);
907
908         /* Handle various SYNC-type writes */
909         if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
910                 error = xfs_write_sync_logforce(mp, xip);
911                 if (error)
912                         goto out_unlock_internal;
913
914                 xfs_rwunlock(bdp, locktype);
915                 if (need_i_mutex)
916                         mutex_unlock(&inode->i_mutex);
917
918                 error = sync_page_range(inode, mapping, pos, ret);
919                 if (!error)
920                         error = ret;
921                 return error;
922         }
923
924  out_unlock_internal:
925         xfs_rwunlock(bdp, locktype);
926  out_unlock_mutex:
927         if (need_i_mutex)
928                 mutex_unlock(&inode->i_mutex);
929  out_nounlocks:
930         return -error;
931 }
932
933 /*
934  * All xfs metadata buffers except log state machine buffers
935  * get this attached as their b_bdstrat callback function.
936  * This is so that we can catch a buffer
937  * after prematurely unpinning it to forcibly shutdown the filesystem.
938  */
939 int
940 xfs_bdstrat_cb(struct xfs_buf *bp)
941 {
942         xfs_mount_t     *mp;
943
944         mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
945         if (!XFS_FORCED_SHUTDOWN(mp)) {
946                 xfs_buf_iorequest(bp);
947                 return 0;
948         } else {
949                 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
950                 /*
951                  * Metadata write that didn't get logged but
952                  * written delayed anyway. These aren't associated
953                  * with a transaction, and can be ignored.
954                  */
955                 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
956                     (XFS_BUF_ISREAD(bp)) == 0)
957                         return (xfs_bioerror_relse(bp));
958                 else
959                         return (xfs_bioerror(bp));
960         }
961 }
962
963
964 int
965 xfs_bmap(bhv_desc_t     *bdp,
966         xfs_off_t       offset,
967         ssize_t         count,
968         int             flags,
969         xfs_iomap_t     *iomapp,
970         int             *niomaps)
971 {
972         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
973         xfs_iocore_t    *io = &ip->i_iocore;
974
975         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
976         ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
977                ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
978
979         return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
980 }
981
982 /*
983  * Wrapper around bdstrat so that we can stop data
984  * from going to disk in case we are shutting down the filesystem.
985  * Typically user data goes thru this path; one of the exceptions
986  * is the superblock.
987  */
988 int
989 xfsbdstrat(
990         struct xfs_mount        *mp,
991         struct xfs_buf          *bp)
992 {
993         ASSERT(mp);
994         if (!XFS_FORCED_SHUTDOWN(mp)) {
995                 /* Grio redirection would go here
996                  * if (XFS_BUF_IS_GRIO(bp)) {
997                  */
998
999                 xfs_buf_iorequest(bp);
1000                 return 0;
1001         }
1002
1003         xfs_buftrace("XFSBDSTRAT IOERROR", bp);
1004         return (xfs_bioerror_relse(bp));
1005 }
1006
1007 /*
1008  * If the underlying (data/log/rt) device is readonly, there are some
1009  * operations that cannot proceed.
1010  */
1011 int
1012 xfs_dev_is_read_only(
1013         xfs_mount_t             *mp,
1014         char                    *message)
1015 {
1016         if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1017             xfs_readonly_buftarg(mp->m_logdev_targp) ||
1018             (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1019                 cmn_err(CE_NOTE,
1020                         "XFS: %s required on read-only device.", message);
1021                 cmn_err(CE_NOTE,
1022                         "XFS: write access unavailable, cannot proceed.");
1023                 return EROFS;
1024         }
1025         return 0;
1026 }