Automatic merge of /spare/repo/linux-2.6/.git branch HEAD
[linux-2.6] / fs / xfs / linux-2.6 / xfs_aops.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32
33 #include "xfs.h"
34 #include "xfs_inum.h"
35 #include "xfs_log.h"
36 #include "xfs_sb.h"
37 #include "xfs_dir.h"
38 #include "xfs_dir2.h"
39 #include "xfs_trans.h"
40 #include "xfs_dmapi.h"
41 #include "xfs_mount.h"
42 #include "xfs_bmap_btree.h"
43 #include "xfs_alloc_btree.h"
44 #include "xfs_ialloc_btree.h"
45 #include "xfs_alloc.h"
46 #include "xfs_btree.h"
47 #include "xfs_attr_sf.h"
48 #include "xfs_dir_sf.h"
49 #include "xfs_dir2_sf.h"
50 #include "xfs_dinode.h"
51 #include "xfs_inode.h"
52 #include "xfs_error.h"
53 #include "xfs_rw.h"
54 #include "xfs_iomap.h"
55 #include <linux/mpage.h>
56 #include <linux/writeback.h>
57
58 STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
59 STATIC void xfs_convert_page(struct inode *, struct page *, xfs_iomap_t *,
60                 struct writeback_control *wbc, void *, int, int);
61
62 #if defined(XFS_RW_TRACE)
63 void
64 xfs_page_trace(
65         int             tag,
66         struct inode    *inode,
67         struct page     *page,
68         int             mask)
69 {
70         xfs_inode_t     *ip;
71         bhv_desc_t      *bdp;
72         vnode_t         *vp = LINVFS_GET_VP(inode);
73         loff_t          isize = i_size_read(inode);
74         loff_t          offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
75         int             delalloc = -1, unmapped = -1, unwritten = -1;
76
77         if (page_has_buffers(page))
78                 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
79
80         bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops);
81         ip = XFS_BHVTOI(bdp);
82         if (!ip->i_rwtrace)
83                 return;
84
85         ktrace_enter(ip->i_rwtrace,
86                 (void *)((unsigned long)tag),
87                 (void *)ip,
88                 (void *)inode,
89                 (void *)page,
90                 (void *)((unsigned long)mask),
91                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
92                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
93                 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
94                 (void *)((unsigned long)(isize & 0xffffffff)),
95                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
96                 (void *)((unsigned long)(offset & 0xffffffff)),
97                 (void *)((unsigned long)delalloc),
98                 (void *)((unsigned long)unmapped),
99                 (void *)((unsigned long)unwritten),
100                 (void *)NULL,
101                 (void *)NULL);
102 }
103 #else
104 #define xfs_page_trace(tag, inode, page, mask)
105 #endif
106
107 void
108 linvfs_unwritten_done(
109         struct buffer_head      *bh,
110         int                     uptodate)
111 {
112         xfs_buf_t               *pb = (xfs_buf_t *)bh->b_private;
113
114         ASSERT(buffer_unwritten(bh));
115         bh->b_end_io = NULL;
116         clear_buffer_unwritten(bh);
117         if (!uptodate)
118                 pagebuf_ioerror(pb, EIO);
119         if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
120                 pagebuf_iodone(pb, 1, 1);
121         }
122         end_buffer_async_write(bh, uptodate);
123 }
124
125 /*
126  * Issue transactions to convert a buffer range from unwritten
127  * to written extents (buffered IO).
128  */
129 STATIC void
130 linvfs_unwritten_convert(
131         xfs_buf_t       *bp)
132 {
133         vnode_t         *vp = XFS_BUF_FSPRIVATE(bp, vnode_t *);
134         int             error;
135
136         BUG_ON(atomic_read(&bp->pb_hold) < 1);
137         VOP_BMAP(vp, XFS_BUF_OFFSET(bp), XFS_BUF_SIZE(bp),
138                         BMAPI_UNWRITTEN, NULL, NULL, error);
139         XFS_BUF_SET_FSPRIVATE(bp, NULL);
140         XFS_BUF_CLR_IODONE_FUNC(bp);
141         XFS_BUF_UNDATAIO(bp);
142         iput(LINVFS_GET_IP(vp));
143         pagebuf_iodone(bp, 0, 0);
144 }
145
146 /*
147  * Issue transactions to convert a buffer range from unwritten
148  * to written extents (direct IO).
149  */
150 STATIC void
151 linvfs_unwritten_convert_direct(
152         struct inode    *inode,
153         loff_t          offset,
154         ssize_t         size,
155         void            *private)
156 {
157         ASSERT(!private || inode == (struct inode *)private);
158
159         /* private indicates an unwritten extent lay beneath this IO */
160         if (private && size > 0) {
161                 vnode_t *vp = LINVFS_GET_VP(inode);
162                 int     error;
163
164                 VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
165         }
166 }
167
168 STATIC int
169 xfs_map_blocks(
170         struct inode            *inode,
171         loff_t                  offset,
172         ssize_t                 count,
173         xfs_iomap_t             *mapp,
174         int                     flags)
175 {
176         vnode_t                 *vp = LINVFS_GET_VP(inode);
177         int                     error, nmaps = 1;
178
179         VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error);
180         if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
181                 VMODIFY(vp);
182         return -error;
183 }
184
185 /*
186  * Finds the corresponding mapping in block @map array of the
187  * given @offset within a @page.
188  */
189 STATIC xfs_iomap_t *
190 xfs_offset_to_map(
191         struct page             *page,
192         xfs_iomap_t             *iomapp,
193         unsigned long           offset)
194 {
195         loff_t                  full_offset;    /* offset from start of file */
196
197         ASSERT(offset < PAGE_CACHE_SIZE);
198
199         full_offset = page->index;              /* NB: using 64bit number */
200         full_offset <<= PAGE_CACHE_SHIFT;       /* offset from file start */
201         full_offset += offset;                  /* offset from page start */
202
203         if (full_offset < iomapp->iomap_offset)
204                 return NULL;
205         if (iomapp->iomap_offset + (iomapp->iomap_bsize -1) >= full_offset)
206                 return iomapp;
207         return NULL;
208 }
209
210 STATIC void
211 xfs_map_at_offset(
212         struct page             *page,
213         struct buffer_head      *bh,
214         unsigned long           offset,
215         int                     block_bits,
216         xfs_iomap_t             *iomapp)
217 {
218         xfs_daddr_t             bn;
219         loff_t                  delta;
220         int                     sector_shift;
221
222         ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
223         ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
224         ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL);
225
226         delta = page->index;
227         delta <<= PAGE_CACHE_SHIFT;
228         delta += offset;
229         delta -= iomapp->iomap_offset;
230         delta >>= block_bits;
231
232         sector_shift = block_bits - BBSHIFT;
233         bn = iomapp->iomap_bn >> sector_shift;
234         bn += delta;
235         BUG_ON(!bn && !(iomapp->iomap_flags & IOMAP_REALTIME));
236         ASSERT((bn << sector_shift) >= iomapp->iomap_bn);
237
238         lock_buffer(bh);
239         bh->b_blocknr = bn;
240         bh->b_bdev = iomapp->iomap_target->pbr_bdev;
241         set_buffer_mapped(bh);
242         clear_buffer_delay(bh);
243 }
244
245 /*
246  * Look for a page at index which is unlocked and contains our
247  * unwritten extent flagged buffers at its head.  Returns page
248  * locked and with an extra reference count, and length of the
249  * unwritten extent component on this page that we can write,
250  * in units of filesystem blocks.
251  */
252 STATIC struct page *
253 xfs_probe_unwritten_page(
254         struct address_space    *mapping,
255         pgoff_t                 index,
256         xfs_iomap_t             *iomapp,
257         xfs_buf_t               *pb,
258         unsigned long           max_offset,
259         unsigned long           *fsbs,
260         unsigned int            bbits)
261 {
262         struct page             *page;
263
264         page = find_trylock_page(mapping, index);
265         if (!page)
266                 return NULL;
267         if (PageWriteback(page))
268                 goto out;
269
270         if (page->mapping && page_has_buffers(page)) {
271                 struct buffer_head      *bh, *head;
272                 unsigned long           p_offset = 0;
273
274                 *fsbs = 0;
275                 bh = head = page_buffers(page);
276                 do {
277                         if (!buffer_unwritten(bh) || !buffer_uptodate(bh))
278                                 break;
279                         if (!xfs_offset_to_map(page, iomapp, p_offset))
280                                 break;
281                         if (p_offset >= max_offset)
282                                 break;
283                         xfs_map_at_offset(page, bh, p_offset, bbits, iomapp);
284                         set_buffer_unwritten_io(bh);
285                         bh->b_private = pb;
286                         p_offset += bh->b_size;
287                         (*fsbs)++;
288                 } while ((bh = bh->b_this_page) != head);
289
290                 if (p_offset)
291                         return page;
292         }
293
294 out:
295         unlock_page(page);
296         return NULL;
297 }
298
299 /*
300  * Look for a page at index which is unlocked and not mapped
301  * yet - clustering for mmap write case.
302  */
303 STATIC unsigned int
304 xfs_probe_unmapped_page(
305         struct address_space    *mapping,
306         pgoff_t                 index,
307         unsigned int            pg_offset)
308 {
309         struct page             *page;
310         int                     ret = 0;
311
312         page = find_trylock_page(mapping, index);
313         if (!page)
314                 return 0;
315         if (PageWriteback(page))
316                 goto out;
317
318         if (page->mapping && PageDirty(page)) {
319                 if (page_has_buffers(page)) {
320                         struct buffer_head      *bh, *head;
321
322                         bh = head = page_buffers(page);
323                         do {
324                                 if (buffer_mapped(bh) || !buffer_uptodate(bh))
325                                         break;
326                                 ret += bh->b_size;
327                                 if (ret >= pg_offset)
328                                         break;
329                         } while ((bh = bh->b_this_page) != head);
330                 } else
331                         ret = PAGE_CACHE_SIZE;
332         }
333
334 out:
335         unlock_page(page);
336         return ret;
337 }
338
339 STATIC unsigned int
340 xfs_probe_unmapped_cluster(
341         struct inode            *inode,
342         struct page             *startpage,
343         struct buffer_head      *bh,
344         struct buffer_head      *head)
345 {
346         pgoff_t                 tindex, tlast, tloff;
347         unsigned int            pg_offset, len, total = 0;
348         struct address_space    *mapping = inode->i_mapping;
349
350         /* First sum forwards in this page */
351         do {
352                 if (buffer_mapped(bh))
353                         break;
354                 total += bh->b_size;
355         } while ((bh = bh->b_this_page) != head);
356
357         /* If we reached the end of the page, sum forwards in
358          * following pages.
359          */
360         if (bh == head) {
361                 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
362                 /* Prune this back to avoid pathological behavior */
363                 tloff = min(tlast, startpage->index + 64);
364                 for (tindex = startpage->index + 1; tindex < tloff; tindex++) {
365                         len = xfs_probe_unmapped_page(mapping, tindex,
366                                                         PAGE_CACHE_SIZE);
367                         if (!len)
368                                 return total;
369                         total += len;
370                 }
371                 if (tindex == tlast &&
372                     (pg_offset = i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
373                         total += xfs_probe_unmapped_page(mapping,
374                                                         tindex, pg_offset);
375                 }
376         }
377         return total;
378 }
379
380 /*
381  * Probe for a given page (index) in the inode and test if it is delayed
382  * and without unwritten buffers.  Returns page locked and with an extra
383  * reference count.
384  */
385 STATIC struct page *
386 xfs_probe_delalloc_page(
387         struct inode            *inode,
388         pgoff_t                 index)
389 {
390         struct page             *page;
391
392         page = find_trylock_page(inode->i_mapping, index);
393         if (!page)
394                 return NULL;
395         if (PageWriteback(page))
396                 goto out;
397
398         if (page->mapping && page_has_buffers(page)) {
399                 struct buffer_head      *bh, *head;
400                 int                     acceptable = 0;
401
402                 bh = head = page_buffers(page);
403                 do {
404                         if (buffer_unwritten(bh)) {
405                                 acceptable = 0;
406                                 break;
407                         } else if (buffer_delay(bh)) {
408                                 acceptable = 1;
409                         }
410                 } while ((bh = bh->b_this_page) != head);
411
412                 if (acceptable)
413                         return page;
414         }
415
416 out:
417         unlock_page(page);
418         return NULL;
419 }
420
421 STATIC int
422 xfs_map_unwritten(
423         struct inode            *inode,
424         struct page             *start_page,
425         struct buffer_head      *head,
426         struct buffer_head      *curr,
427         unsigned long           p_offset,
428         int                     block_bits,
429         xfs_iomap_t             *iomapp,
430         struct writeback_control *wbc,
431         int                     startio,
432         int                     all_bh)
433 {
434         struct buffer_head      *bh = curr;
435         xfs_iomap_t             *tmp;
436         xfs_buf_t               *pb;
437         loff_t                  offset, size;
438         unsigned long           nblocks = 0;
439
440         offset = start_page->index;
441         offset <<= PAGE_CACHE_SHIFT;
442         offset += p_offset;
443
444         /* get an "empty" pagebuf to manage IO completion
445          * Proper values will be set before returning */
446         pb = pagebuf_lookup(iomapp->iomap_target, 0, 0, 0);
447         if (!pb)
448                 return -EAGAIN;
449
450         /* Take a reference to the inode to prevent it from
451          * being reclaimed while we have outstanding unwritten
452          * extent IO on it.
453          */
454         if ((igrab(inode)) != inode) {
455                 pagebuf_free(pb);
456                 return -EAGAIN;
457         }
458
459         /* Set the count to 1 initially, this will stop an I/O
460          * completion callout which happens before we have started
461          * all the I/O from calling pagebuf_iodone too early.
462          */
463         atomic_set(&pb->pb_io_remaining, 1);
464
465         /* First map forwards in the page consecutive buffers
466          * covering this unwritten extent
467          */
468         do {
469                 if (!buffer_unwritten(bh))
470                         break;
471                 tmp = xfs_offset_to_map(start_page, iomapp, p_offset);
472                 if (!tmp)
473                         break;
474                 xfs_map_at_offset(start_page, bh, p_offset, block_bits, iomapp);
475                 set_buffer_unwritten_io(bh);
476                 bh->b_private = pb;
477                 p_offset += bh->b_size;
478                 nblocks++;
479         } while ((bh = bh->b_this_page) != head);
480
481         atomic_add(nblocks, &pb->pb_io_remaining);
482
483         /* If we reached the end of the page, map forwards in any
484          * following pages which are also covered by this extent.
485          */
486         if (bh == head) {
487                 struct address_space    *mapping = inode->i_mapping;
488                 pgoff_t                 tindex, tloff, tlast;
489                 unsigned long           bs;
490                 unsigned int            pg_offset, bbits = inode->i_blkbits;
491                 struct page             *page;
492
493                 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
494                 tloff = (iomapp->iomap_offset + iomapp->iomap_bsize) >> PAGE_CACHE_SHIFT;
495                 tloff = min(tlast, tloff);
496                 for (tindex = start_page->index + 1; tindex < tloff; tindex++) {
497                         page = xfs_probe_unwritten_page(mapping,
498                                                 tindex, iomapp, pb,
499                                                 PAGE_CACHE_SIZE, &bs, bbits);
500                         if (!page)
501                                 break;
502                         nblocks += bs;
503                         atomic_add(bs, &pb->pb_io_remaining);
504                         xfs_convert_page(inode, page, iomapp, wbc, pb,
505                                                         startio, all_bh);
506                         /* stop if converting the next page might add
507                          * enough blocks that the corresponding byte
508                          * count won't fit in our ulong page buf length */
509                         if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
510                                 goto enough;
511                 }
512
513                 if (tindex == tlast &&
514                     (pg_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1)))) {
515                         page = xfs_probe_unwritten_page(mapping,
516                                                         tindex, iomapp, pb,
517                                                         pg_offset, &bs, bbits);
518                         if (page) {
519                                 nblocks += bs;
520                                 atomic_add(bs, &pb->pb_io_remaining);
521                                 xfs_convert_page(inode, page, iomapp, wbc, pb,
522                                                         startio, all_bh);
523                                 if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
524                                         goto enough;
525                         }
526                 }
527         }
528
529 enough:
530         size = nblocks;         /* NB: using 64bit number here */
531         size <<= block_bits;    /* convert fsb's to byte range */
532
533         XFS_BUF_DATAIO(pb);
534         XFS_BUF_ASYNC(pb);
535         XFS_BUF_SET_SIZE(pb, size);
536         XFS_BUF_SET_COUNT(pb, size);
537         XFS_BUF_SET_OFFSET(pb, offset);
538         XFS_BUF_SET_FSPRIVATE(pb, LINVFS_GET_VP(inode));
539         XFS_BUF_SET_IODONE_FUNC(pb, linvfs_unwritten_convert);
540
541         if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
542                 pagebuf_iodone(pb, 1, 1);
543         }
544
545         return 0;
546 }
547
548 STATIC void
549 xfs_submit_page(
550         struct page             *page,
551         struct writeback_control *wbc,
552         struct buffer_head      *bh_arr[],
553         int                     bh_count,
554         int                     probed_page,
555         int                     clear_dirty)
556 {
557         struct buffer_head      *bh;
558         int                     i;
559
560         BUG_ON(PageWriteback(page));
561         if (bh_count)
562                 set_page_writeback(page);
563         if (clear_dirty)
564                 clear_page_dirty(page);
565         unlock_page(page);
566
567         if (bh_count) {
568                 for (i = 0; i < bh_count; i++) {
569                         bh = bh_arr[i];
570                         mark_buffer_async_write(bh);
571                         if (buffer_unwritten(bh))
572                                 set_buffer_unwritten_io(bh);
573                         set_buffer_uptodate(bh);
574                         clear_buffer_dirty(bh);
575                 }
576
577                 for (i = 0; i < bh_count; i++)
578                         submit_bh(WRITE, bh_arr[i]);
579
580                 if (probed_page && clear_dirty)
581                         wbc->nr_to_write--;     /* Wrote an "extra" page */
582         }
583 }
584
585 /*
586  * Allocate & map buffers for page given the extent map. Write it out.
587  * except for the original page of a writepage, this is called on
588  * delalloc/unwritten pages only, for the original page it is possible
589  * that the page has no mapping at all.
590  */
591 STATIC void
592 xfs_convert_page(
593         struct inode            *inode,
594         struct page             *page,
595         xfs_iomap_t             *iomapp,
596         struct writeback_control *wbc,
597         void                    *private,
598         int                     startio,
599         int                     all_bh)
600 {
601         struct buffer_head      *bh_arr[MAX_BUF_PER_PAGE], *bh, *head;
602         xfs_iomap_t             *mp = iomapp, *tmp;
603         unsigned long           offset, end_offset;
604         int                     index = 0;
605         int                     bbits = inode->i_blkbits;
606         int                     len, page_dirty;
607
608         end_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1));
609
610         /*
611          * page_dirty is initially a count of buffers on the page before
612          * EOF and is decrememted as we move each into a cleanable state.
613          */
614         len = 1 << inode->i_blkbits;
615         end_offset = max(end_offset, PAGE_CACHE_SIZE);
616         end_offset = roundup(end_offset, len);
617         page_dirty = end_offset / len;
618
619         offset = 0;
620         bh = head = page_buffers(page);
621         do {
622                 if (offset >= end_offset)
623                         break;
624                 if (!(PageUptodate(page) || buffer_uptodate(bh)))
625                         continue;
626                 if (buffer_mapped(bh) && all_bh &&
627                     !(buffer_unwritten(bh) || buffer_delay(bh))) {
628                         if (startio) {
629                                 lock_buffer(bh);
630                                 bh_arr[index++] = bh;
631                                 page_dirty--;
632                         }
633                         continue;
634                 }
635                 tmp = xfs_offset_to_map(page, mp, offset);
636                 if (!tmp)
637                         continue;
638                 ASSERT(!(tmp->iomap_flags & IOMAP_HOLE));
639                 ASSERT(!(tmp->iomap_flags & IOMAP_DELAY));
640
641                 /* If this is a new unwritten extent buffer (i.e. one
642                  * that we haven't passed in private data for, we must
643                  * now map this buffer too.
644                  */
645                 if (buffer_unwritten(bh) && !bh->b_end_io) {
646                         ASSERT(tmp->iomap_flags & IOMAP_UNWRITTEN);
647                         xfs_map_unwritten(inode, page, head, bh, offset,
648                                         bbits, tmp, wbc, startio, all_bh);
649                 } else if (! (buffer_unwritten(bh) && buffer_locked(bh))) {
650                         xfs_map_at_offset(page, bh, offset, bbits, tmp);
651                         if (buffer_unwritten(bh)) {
652                                 set_buffer_unwritten_io(bh);
653                                 bh->b_private = private;
654                                 ASSERT(private);
655                         }
656                 }
657                 if (startio) {
658                         bh_arr[index++] = bh;
659                 } else {
660                         set_buffer_dirty(bh);
661                         unlock_buffer(bh);
662                         mark_buffer_dirty(bh);
663                 }
664                 page_dirty--;
665         } while (offset += len, (bh = bh->b_this_page) != head);
666
667         if (startio && index) {
668                 xfs_submit_page(page, wbc, bh_arr, index, 1, !page_dirty);
669         } else {
670                 unlock_page(page);
671         }
672 }
673
674 /*
675  * Convert & write out a cluster of pages in the same extent as defined
676  * by mp and following the start page.
677  */
678 STATIC void
679 xfs_cluster_write(
680         struct inode            *inode,
681         pgoff_t                 tindex,
682         xfs_iomap_t             *iomapp,
683         struct writeback_control *wbc,
684         int                     startio,
685         int                     all_bh,
686         pgoff_t                 tlast)
687 {
688         struct page             *page;
689
690         for (; tindex <= tlast; tindex++) {
691                 page = xfs_probe_delalloc_page(inode, tindex);
692                 if (!page)
693                         break;
694                 xfs_convert_page(inode, page, iomapp, wbc, NULL,
695                                 startio, all_bh);
696         }
697 }
698
699 /*
700  * Calling this without startio set means we are being asked to make a dirty
701  * page ready for freeing it's buffers.  When called with startio set then
702  * we are coming from writepage.
703  *
704  * When called with startio set it is important that we write the WHOLE
705  * page if possible.
706  * The bh->b_state's cannot know if any of the blocks or which block for
707  * that matter are dirty due to mmap writes, and therefore bh uptodate is
708  * only vaild if the page itself isn't completely uptodate.  Some layers
709  * may clear the page dirty flag prior to calling write page, under the
710  * assumption the entire page will be written out; by not writing out the
711  * whole page the page can be reused before all valid dirty data is
712  * written out.  Note: in the case of a page that has been dirty'd by
713  * mapwrite and but partially setup by block_prepare_write the
714  * bh->b_states's will not agree and only ones setup by BPW/BCW will have
715  * valid state, thus the whole page must be written out thing.
716  */
717
718 STATIC int
719 xfs_page_state_convert(
720         struct inode    *inode,
721         struct page     *page,
722         struct writeback_control *wbc,
723         int             startio,
724         int             unmapped) /* also implies page uptodate */
725 {
726         struct buffer_head      *bh_arr[MAX_BUF_PER_PAGE], *bh, *head;
727         xfs_iomap_t             *iomp, iomap;
728         loff_t                  offset;
729         unsigned long           p_offset = 0;
730         __uint64_t              end_offset;
731         pgoff_t                 end_index, last_index, tlast;
732         int                     len, err, i, cnt = 0, uptodate = 1;
733         int                     flags;
734         int                     page_dirty;
735
736         /* wait for other IO threads? */
737         flags = (startio && wbc->sync_mode != WB_SYNC_NONE) ? 0 : BMAPI_TRYLOCK;
738
739         /* Is this page beyond the end of the file? */
740         offset = i_size_read(inode);
741         end_index = offset >> PAGE_CACHE_SHIFT;
742         last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
743         if (page->index >= end_index) {
744                 if ((page->index >= end_index + 1) ||
745                     !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
746                         err = -EIO;
747                         goto error;
748                 }
749         }
750
751         end_offset = min_t(unsigned long long,
752                         (loff_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
753         offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
754
755         /*
756          * page_dirty is initially a count of buffers on the page before
757          * EOF and is decrememted as we move each into a cleanable state.
758          */
759         len = 1 << inode->i_blkbits;
760         p_offset = max(p_offset, PAGE_CACHE_SIZE);
761         p_offset = roundup(p_offset, len);
762         page_dirty = p_offset / len;
763
764         iomp = NULL;
765         p_offset = 0;
766         bh = head = page_buffers(page);
767
768         do {
769                 if (offset >= end_offset)
770                         break;
771                 if (!buffer_uptodate(bh))
772                         uptodate = 0;
773                 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio)
774                         continue;
775
776                 if (iomp) {
777                         iomp = xfs_offset_to_map(page, &iomap, p_offset);
778                 }
779
780                 /*
781                  * First case, map an unwritten extent and prepare for
782                  * extent state conversion transaction on completion.
783                  */
784                 if (buffer_unwritten(bh)) {
785                         if (!startio)
786                                 continue;
787                         if (!iomp) {
788                                 err = xfs_map_blocks(inode, offset, len, &iomap,
789                                                 BMAPI_READ|BMAPI_IGNSTATE);
790                                 if (err) {
791                                         goto error;
792                                 }
793                                 iomp = xfs_offset_to_map(page, &iomap,
794                                                                 p_offset);
795                         }
796                         if (iomp) {
797                                 if (!bh->b_end_io) {
798                                         err = xfs_map_unwritten(inode, page,
799                                                         head, bh, p_offset,
800                                                         inode->i_blkbits, iomp,
801                                                         wbc, startio, unmapped);
802                                         if (err) {
803                                                 goto error;
804                                         }
805                                 } else {
806                                         set_bit(BH_Lock, &bh->b_state);
807                                 }
808                                 BUG_ON(!buffer_locked(bh));
809                                 bh_arr[cnt++] = bh;
810                                 page_dirty--;
811                         }
812                 /*
813                  * Second case, allocate space for a delalloc buffer.
814                  * We can return EAGAIN here in the release page case.
815                  */
816                 } else if (buffer_delay(bh)) {
817                         if (!iomp) {
818                                 err = xfs_map_blocks(inode, offset, len, &iomap,
819                                                 BMAPI_ALLOCATE | flags);
820                                 if (err) {
821                                         goto error;
822                                 }
823                                 iomp = xfs_offset_to_map(page, &iomap,
824                                                                 p_offset);
825                         }
826                         if (iomp) {
827                                 xfs_map_at_offset(page, bh, p_offset,
828                                                 inode->i_blkbits, iomp);
829                                 if (startio) {
830                                         bh_arr[cnt++] = bh;
831                                 } else {
832                                         set_buffer_dirty(bh);
833                                         unlock_buffer(bh);
834                                         mark_buffer_dirty(bh);
835                                 }
836                                 page_dirty--;
837                         }
838                 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
839                            (unmapped || startio)) {
840
841                         if (!buffer_mapped(bh)) {
842                                 int     size;
843
844                                 /*
845                                  * Getting here implies an unmapped buffer
846                                  * was found, and we are in a path where we
847                                  * need to write the whole page out.
848                                  */
849                                 if (!iomp) {
850                                         size = xfs_probe_unmapped_cluster(
851                                                         inode, page, bh, head);
852                                         err = xfs_map_blocks(inode, offset,
853                                                         size, &iomap,
854                                                         BMAPI_WRITE|BMAPI_MMAP);
855                                         if (err) {
856                                                 goto error;
857                                         }
858                                         iomp = xfs_offset_to_map(page, &iomap,
859                                                                      p_offset);
860                                 }
861                                 if (iomp) {
862                                         xfs_map_at_offset(page,
863                                                         bh, p_offset,
864                                                         inode->i_blkbits, iomp);
865                                         if (startio) {
866                                                 bh_arr[cnt++] = bh;
867                                         } else {
868                                                 set_buffer_dirty(bh);
869                                                 unlock_buffer(bh);
870                                                 mark_buffer_dirty(bh);
871                                         }
872                                         page_dirty--;
873                                 }
874                         } else if (startio) {
875                                 if (buffer_uptodate(bh) &&
876                                     !test_and_set_bit(BH_Lock, &bh->b_state)) {
877                                         bh_arr[cnt++] = bh;
878                                         page_dirty--;
879                                 }
880                         }
881                 }
882         } while (offset += len, p_offset += len,
883                 ((bh = bh->b_this_page) != head));
884
885         if (uptodate && bh == head)
886                 SetPageUptodate(page);
887
888         if (startio) {
889                 xfs_submit_page(page, wbc, bh_arr, cnt, 0, !page_dirty);
890         }
891
892         if (iomp) {
893                 offset = (iomp->iomap_offset + iomp->iomap_bsize - 1) >>
894                                         PAGE_CACHE_SHIFT;
895                 tlast = min_t(pgoff_t, offset, last_index);
896                 xfs_cluster_write(inode, page->index + 1, iomp, wbc,
897                                         startio, unmapped, tlast);
898         }
899
900         return page_dirty;
901
902 error:
903         for (i = 0; i < cnt; i++) {
904                 unlock_buffer(bh_arr[i]);
905         }
906
907         /*
908          * If it's delalloc and we have nowhere to put it,
909          * throw it away, unless the lower layers told
910          * us to try again.
911          */
912         if (err != -EAGAIN) {
913                 if (!unmapped) {
914                         block_invalidatepage(page, 0);
915                 }
916                 ClearPageUptodate(page);
917         }
918         return err;
919 }
920
921 STATIC int
922 __linvfs_get_block(
923         struct inode            *inode,
924         sector_t                iblock,
925         unsigned long           blocks,
926         struct buffer_head      *bh_result,
927         int                     create,
928         int                     direct,
929         bmapi_flags_t           flags)
930 {
931         vnode_t                 *vp = LINVFS_GET_VP(inode);
932         xfs_iomap_t             iomap;
933         int                     retpbbm = 1;
934         int                     error;
935         ssize_t                 size;
936         loff_t                  offset = (loff_t)iblock << inode->i_blkbits;
937
938         if (blocks)
939                 size = blocks << inode->i_blkbits;
940         else
941                 size = 1 << inode->i_blkbits;
942
943         VOP_BMAP(vp, offset, size,
944                 create ? flags : BMAPI_READ, &iomap, &retpbbm, error);
945         if (error)
946                 return -error;
947
948         if (retpbbm == 0)
949                 return 0;
950
951         if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
952                 xfs_daddr_t             bn;
953                 loff_t                  delta;
954
955                 /* For unwritten extents do not report a disk address on
956                  * the read case (treat as if we're reading into a hole).
957                  */
958                 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
959                         delta = offset - iomap.iomap_offset;
960                         delta >>= inode->i_blkbits;
961
962                         bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT);
963                         bn += delta;
964                         BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME));
965                         bh_result->b_blocknr = bn;
966                         set_buffer_mapped(bh_result);
967                 }
968                 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
969                         if (direct)
970                                 bh_result->b_private = inode;
971                         set_buffer_unwritten(bh_result);
972                         set_buffer_delay(bh_result);
973                 }
974         }
975
976         /* If this is a realtime file, data might be on a new device */
977         bh_result->b_bdev = iomap.iomap_target->pbr_bdev;
978
979         /* If we previously allocated a block out beyond eof and
980          * we are now coming back to use it then we will need to
981          * flag it as new even if it has a disk address.
982          */
983         if (create &&
984             ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
985              (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) {
986                 set_buffer_new(bh_result);
987         }
988
989         if (iomap.iomap_flags & IOMAP_DELAY) {
990                 BUG_ON(direct);
991                 if (create) {
992                         set_buffer_uptodate(bh_result);
993                         set_buffer_mapped(bh_result);
994                         set_buffer_delay(bh_result);
995                 }
996         }
997
998         if (blocks) {
999                 bh_result->b_size = (ssize_t)min(
1000                         (loff_t)(iomap.iomap_bsize - iomap.iomap_delta),
1001                         (loff_t)(blocks << inode->i_blkbits));
1002         }
1003
1004         return 0;
1005 }
1006
1007 int
1008 linvfs_get_block(
1009         struct inode            *inode,
1010         sector_t                iblock,
1011         struct buffer_head      *bh_result,
1012         int                     create)
1013 {
1014         return __linvfs_get_block(inode, iblock, 0, bh_result,
1015                                         create, 0, BMAPI_WRITE);
1016 }
1017
1018 STATIC int
1019 linvfs_get_blocks_direct(
1020         struct inode            *inode,
1021         sector_t                iblock,
1022         unsigned long           max_blocks,
1023         struct buffer_head      *bh_result,
1024         int                     create)
1025 {
1026         return __linvfs_get_block(inode, iblock, max_blocks, bh_result,
1027                                         create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1028 }
1029
1030 STATIC ssize_t
1031 linvfs_direct_IO(
1032         int                     rw,
1033         struct kiocb            *iocb,
1034         const struct iovec      *iov,
1035         loff_t                  offset,
1036         unsigned long           nr_segs)
1037 {
1038         struct file     *file = iocb->ki_filp;
1039         struct inode    *inode = file->f_mapping->host;
1040         vnode_t         *vp = LINVFS_GET_VP(inode);
1041         xfs_iomap_t     iomap;
1042         int             maps = 1;
1043         int             error;
1044
1045         VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
1046         if (error)
1047                 return -error;
1048
1049         return blockdev_direct_IO_own_locking(rw, iocb, inode,
1050                 iomap.iomap_target->pbr_bdev,
1051                 iov, offset, nr_segs,
1052                 linvfs_get_blocks_direct,
1053                 linvfs_unwritten_convert_direct);
1054 }
1055
1056
1057 STATIC sector_t
1058 linvfs_bmap(
1059         struct address_space    *mapping,
1060         sector_t                block)
1061 {
1062         struct inode            *inode = (struct inode *)mapping->host;
1063         vnode_t                 *vp = LINVFS_GET_VP(inode);
1064         int                     error;
1065
1066         vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address);
1067
1068         VOP_RWLOCK(vp, VRWLOCK_READ);
1069         VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
1070         VOP_RWUNLOCK(vp, VRWLOCK_READ);
1071         return generic_block_bmap(mapping, block, linvfs_get_block);
1072 }
1073
1074 STATIC int
1075 linvfs_readpage(
1076         struct file             *unused,
1077         struct page             *page)
1078 {
1079         return mpage_readpage(page, linvfs_get_block);
1080 }
1081
1082 STATIC int
1083 linvfs_readpages(
1084         struct file             *unused,
1085         struct address_space    *mapping,
1086         struct list_head        *pages,
1087         unsigned                nr_pages)
1088 {
1089         return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block);
1090 }
1091
1092 STATIC void
1093 xfs_count_page_state(
1094         struct page             *page,
1095         int                     *delalloc,
1096         int                     *unmapped,
1097         int                     *unwritten)
1098 {
1099         struct buffer_head      *bh, *head;
1100
1101         *delalloc = *unmapped = *unwritten = 0;
1102
1103         bh = head = page_buffers(page);
1104         do {
1105                 if (buffer_uptodate(bh) && !buffer_mapped(bh))
1106                         (*unmapped) = 1;
1107                 else if (buffer_unwritten(bh) && !buffer_delay(bh))
1108                         clear_buffer_unwritten(bh);
1109                 else if (buffer_unwritten(bh))
1110                         (*unwritten) = 1;
1111                 else if (buffer_delay(bh))
1112                         (*delalloc) = 1;
1113         } while ((bh = bh->b_this_page) != head);
1114 }
1115
1116
1117 /*
1118  * writepage: Called from one of two places:
1119  *
1120  * 1. we are flushing a delalloc buffer head.
1121  *
1122  * 2. we are writing out a dirty page. Typically the page dirty
1123  *    state is cleared before we get here. In this case is it
1124  *    conceivable we have no buffer heads.
1125  *
1126  * For delalloc space on the page we need to allocate space and
1127  * flush it. For unmapped buffer heads on the page we should
1128  * allocate space if the page is uptodate. For any other dirty
1129  * buffer heads on the page we should flush them.
1130  *
1131  * If we detect that a transaction would be required to flush
1132  * the page, we have to check the process flags first, if we
1133  * are already in a transaction or disk I/O during allocations
1134  * is off, we need to fail the writepage and redirty the page.
1135  */
1136
1137 STATIC int
1138 linvfs_writepage(
1139         struct page             *page,
1140         struct writeback_control *wbc)
1141 {
1142         int                     error;
1143         int                     need_trans;
1144         int                     delalloc, unmapped, unwritten;
1145         struct inode            *inode = page->mapping->host;
1146
1147         xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1148
1149         /*
1150          * We need a transaction if:
1151          *  1. There are delalloc buffers on the page
1152          *  2. The page is uptodate and we have unmapped buffers
1153          *  3. The page is uptodate and we have no buffers
1154          *  4. There are unwritten buffers on the page
1155          */
1156
1157         if (!page_has_buffers(page)) {
1158                 unmapped = 1;
1159                 need_trans = 1;
1160         } else {
1161                 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1162                 if (!PageUptodate(page))
1163                         unmapped = 0;
1164                 need_trans = delalloc + unmapped + unwritten;
1165         }
1166
1167         /*
1168          * If we need a transaction and the process flags say
1169          * we are already in a transaction, or no IO is allowed
1170          * then mark the page dirty again and leave the page
1171          * as is.
1172          */
1173         if (PFLAGS_TEST_FSTRANS() && need_trans)
1174                 goto out_fail;
1175
1176         /*
1177          * Delay hooking up buffer heads until we have
1178          * made our go/no-go decision.
1179          */
1180         if (!page_has_buffers(page))
1181                 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1182
1183         /*
1184          * Convert delayed allocate, unwritten or unmapped space
1185          * to real space and flush out to disk.
1186          */
1187         error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1188         if (error == -EAGAIN)
1189                 goto out_fail;
1190         if (unlikely(error < 0))
1191                 goto out_unlock;
1192
1193         return 0;
1194
1195 out_fail:
1196         redirty_page_for_writepage(wbc, page);
1197         unlock_page(page);
1198         return 0;
1199 out_unlock:
1200         unlock_page(page);
1201         return error;
1202 }
1203
1204 /*
1205  * Called to move a page into cleanable state - and from there
1206  * to be released. Possibly the page is already clean. We always
1207  * have buffer heads in this call.
1208  *
1209  * Returns 0 if the page is ok to release, 1 otherwise.
1210  *
1211  * Possible scenarios are:
1212  *
1213  * 1. We are being called to release a page which has been written
1214  *    to via regular I/O. buffer heads will be dirty and possibly
1215  *    delalloc. If no delalloc buffer heads in this case then we
1216  *    can just return zero.
1217  *
1218  * 2. We are called to release a page which has been written via
1219  *    mmap, all we need to do is ensure there is no delalloc
1220  *    state in the buffer heads, if not we can let the caller
1221  *    free them and we should come back later via writepage.
1222  */
1223 STATIC int
1224 linvfs_release_page(
1225         struct page             *page,
1226         int                     gfp_mask)
1227 {
1228         struct inode            *inode = page->mapping->host;
1229         int                     dirty, delalloc, unmapped, unwritten;
1230         struct writeback_control wbc = {
1231                 .sync_mode = WB_SYNC_ALL,
1232                 .nr_to_write = 1,
1233         };
1234
1235         xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
1236
1237         xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1238         if (!delalloc && !unwritten)
1239                 goto free_buffers;
1240
1241         if (!(gfp_mask & __GFP_FS))
1242                 return 0;
1243
1244         /* If we are already inside a transaction or the thread cannot
1245          * do I/O, we cannot release this page.
1246          */
1247         if (PFLAGS_TEST_FSTRANS())
1248                 return 0;
1249
1250         /*
1251          * Convert delalloc space to real space, do not flush the
1252          * data out to disk, that will be done by the caller.
1253          * Never need to allocate space here - we will always
1254          * come back to writepage in that case.
1255          */
1256         dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1257         if (dirty == 0 && !unwritten)
1258                 goto free_buffers;
1259         return 0;
1260
1261 free_buffers:
1262         return try_to_free_buffers(page);
1263 }
1264
1265 STATIC int
1266 linvfs_prepare_write(
1267         struct file             *file,
1268         struct page             *page,
1269         unsigned int            from,
1270         unsigned int            to)
1271 {
1272         return block_prepare_write(page, from, to, linvfs_get_block);
1273 }
1274
1275 struct address_space_operations linvfs_aops = {
1276         .readpage               = linvfs_readpage,
1277         .readpages              = linvfs_readpages,
1278         .writepage              = linvfs_writepage,
1279         .sync_page              = block_sync_page,
1280         .releasepage            = linvfs_release_page,
1281         .prepare_write          = linvfs_prepare_write,
1282         .commit_write           = generic_commit_write,
1283         .bmap                   = linvfs_bmap,
1284         .direct_IO              = linvfs_direct_IO,
1285 };