2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
7 * Copyrights for code taken from ext2:
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
13 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * This file is part of exofs.
18 * exofs is free software; you can redistribute it and/or modify
19 * it under the terms of the GNU General Public License as published by
20 * the Free Software Foundation. Since it is based on ext2, and the only
21 * valid version of GPL for the Linux kernel is version 2, the only valid
22 * version of GPL for exofs is version 2.
24 * exofs is distributed in the hope that it will be useful,
25 * but WITHOUT ANY WARRANTY; without even the implied warranty of
26 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27 * GNU General Public License for more details.
29 * You should have received a copy of the GNU General Public License
30 * along with exofs; if not, write to the Free Software
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
34 #include <linux/slab.h>
38 #define EXOFS_DBGMSG2(M...) do {} while (0)
40 enum { BIO_MAX_PAGES_KMALLOC
=
41 (PAGE_SIZE
- sizeof(struct bio
)) / sizeof(struct bio_vec
),
43 PAGE_SIZE
/ sizeof(struct page
*),
46 unsigned exofs_max_io_pages(struct ore_layout
*layout
,
47 unsigned expected_pages
)
49 unsigned pages
= min_t(unsigned, expected_pages
, MAX_PAGES_KMALLOC
);
51 /* TODO: easily support bio chaining */
52 pages
= min_t(unsigned, pages
,
53 layout
->group_width
* BIO_MAX_PAGES_KMALLOC
);
58 struct exofs_sb_info
*sbi
;
60 unsigned expected_pages
;
61 struct ore_io_state
*ios
;
67 loff_t pg_first
; /* keep 64bit also in 32-arches */
68 bool read_4_write
; /* This means two things: that the read is sync
69 * And the pages should not be unlocked.
73 static void _pcol_init(struct page_collect
*pcol
, unsigned expected_pages
,
76 struct exofs_sb_info
*sbi
= inode
->i_sb
->s_fs_info
;
80 pcol
->expected_pages
= expected_pages
;
84 pcol
->alloc_pages
= 0;
88 pcol
->read_4_write
= false;
91 static void _pcol_reset(struct page_collect
*pcol
)
93 pcol
->expected_pages
-= min(pcol
->nr_pages
, pcol
->expected_pages
);
96 pcol
->alloc_pages
= 0;
102 /* this is probably the end of the loop but in writes
103 * it might not end here. don't be left with nothing
105 if (!pcol
->expected_pages
)
106 pcol
->expected_pages
= MAX_PAGES_KMALLOC
;
109 static int pcol_try_alloc(struct page_collect
*pcol
)
113 /* TODO: easily support bio chaining */
114 pages
= exofs_max_io_pages(&pcol
->sbi
->layout
, pcol
->expected_pages
);
116 for (; pages
; pages
>>= 1) {
117 pcol
->pages
= kmalloc(pages
* sizeof(struct page
*),
119 if (likely(pcol
->pages
)) {
120 pcol
->alloc_pages
= pages
;
125 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
126 pcol
->expected_pages
);
130 static void pcol_free(struct page_collect
*pcol
)
136 ore_put_io_state(pcol
->ios
);
141 static int pcol_add_page(struct page_collect
*pcol
, struct page
*page
,
144 if (unlikely(pcol
->nr_pages
>= pcol
->alloc_pages
))
147 pcol
->pages
[pcol
->nr_pages
++] = page
;
152 enum {PAGE_WAS_NOT_IN_IO
= 17};
153 static int update_read_page(struct page
*page
, int ret
)
157 /* Everything is OK */
158 SetPageUptodate(page
);
160 ClearPageError(page
);
163 /* In this case we were trying to read something that wasn't on
164 * disk yet - return a page full of zeroes. This should be OK,
165 * because the object should be empty (if there was a write
166 * before this read, the read would be waiting with the page
168 clear_highpage(page
);
170 SetPageUptodate(page
);
172 ClearPageError(page
);
173 EXOFS_DBGMSG("recovered read error\n");
175 case PAGE_WAS_NOT_IN_IO
:
176 ret
= 0; /* recovered error */
184 static void update_write_page(struct page
*page
, int ret
)
186 if (unlikely(ret
== PAGE_WAS_NOT_IN_IO
))
187 return; /* don't pass start don't collect $200 */
190 mapping_set_error(page
->mapping
, ret
);
193 end_page_writeback(page
);
196 /* Called at the end of reads, to optionally unlock pages and update their
199 static int __readpages_done(struct page_collect
*pcol
)
205 int ret
= ore_check_io(pcol
->ios
, &resid
);
208 good_bytes
= pcol
->length
;
209 ret
= PAGE_WAS_NOT_IN_IO
;
211 good_bytes
= pcol
->length
- resid
;
213 if (good_bytes
> pcol
->ios
->length
)
214 good_bytes
= pcol
->ios
->length
;
216 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
217 " length=0x%lx nr_pages=%u\n",
218 pcol
->inode
->i_ino
, _LLU(good_bytes
), pcol
->length
,
221 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
222 struct page
*page
= pcol
->pages
[i
];
223 struct inode
*inode
= page
->mapping
->host
;
226 if (inode
!= pcol
->inode
)
227 continue; /* osd might add more pages at end */
229 if (likely(length
< good_bytes
))
234 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
235 inode
->i_ino
, page
->index
,
236 page_stat
? "bad_bytes" : "good_bytes");
238 ret
= update_read_page(page
, page_stat
);
239 if (!pcol
->read_4_write
)
245 EXOFS_DBGMSG2("readpages_done END\n");
249 /* callback of async reads */
250 static void readpages_done(struct ore_io_state
*ios
, void *p
)
252 struct page_collect
*pcol
= p
;
254 __readpages_done(pcol
);
255 atomic_dec(&pcol
->sbi
->s_curr_pending
);
259 static void _unlock_pcol_pages(struct page_collect
*pcol
, int ret
, int rw
)
263 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
264 struct page
*page
= pcol
->pages
[i
];
267 update_read_page(page
, ret
);
269 update_write_page(page
, ret
);
275 static int _maybe_not_all_in_one_io(struct ore_io_state
*ios
,
276 struct page_collect
*pcol_src
, struct page_collect
*pcol
)
278 /* length was wrong or offset was not page aligned */
279 BUG_ON(pcol_src
->nr_pages
< ios
->nr_pages
);
281 if (pcol_src
->nr_pages
> ios
->nr_pages
) {
282 struct page
**src_page
;
283 unsigned pages_less
= pcol_src
->nr_pages
- ios
->nr_pages
;
284 unsigned long len_less
= pcol_src
->length
- ios
->length
;
288 /* This IO was trimmed */
289 pcol_src
->nr_pages
= ios
->nr_pages
;
290 pcol_src
->length
= ios
->length
;
292 /* Left over pages are passed to the next io */
293 pcol
->expected_pages
+= pages_less
;
294 pcol
->nr_pages
= pages_less
;
295 pcol
->length
= len_less
;
296 src_page
= pcol_src
->pages
+ pcol_src
->nr_pages
;
297 pcol
->pg_first
= (*src_page
)->index
;
299 ret
= pcol_try_alloc(pcol
);
303 for (i
= 0; i
< pages_less
; ++i
)
304 pcol
->pages
[i
] = *src_page
++;
306 EXOFS_DBGMSG("Length was adjusted nr_pages=0x%x "
307 "pages_less=0x%x expected_pages=0x%x "
308 "next_offset=0x%llx next_len=0x%lx\n",
309 pcol_src
->nr_pages
, pages_less
, pcol
->expected_pages
,
310 pcol
->pg_first
* PAGE_SIZE
, pcol
->length
);
315 static int read_exec(struct page_collect
*pcol
)
317 struct exofs_i_info
*oi
= exofs_i(pcol
->inode
);
318 struct ore_io_state
*ios
;
319 struct page_collect
*pcol_copy
= NULL
;
326 int ret
= ore_get_rw_state(&pcol
->sbi
->layout
, &oi
->oc
, true,
327 pcol
->pg_first
<< PAGE_CACHE_SHIFT
,
328 pcol
->length
, &pcol
->ios
);
335 ios
->pages
= pcol
->pages
;
337 if (pcol
->read_4_write
) {
339 return __readpages_done(pcol
);
342 pcol_copy
= kmalloc(sizeof(*pcol_copy
), GFP_KERNEL
);
349 ios
->done
= readpages_done
;
350 ios
->private = pcol_copy
;
352 /* pages ownership was passed to pcol_copy */
355 ret
= _maybe_not_all_in_one_io(ios
, pcol_copy
, pcol
);
359 EXOFS_DBGMSG2("read_exec(0x%lx) offset=0x%llx length=0x%llx\n",
360 pcol
->inode
->i_ino
, _LLU(ios
->offset
), _LLU(ios
->length
));
366 atomic_inc(&pcol
->sbi
->s_curr_pending
);
371 if (!pcol
->read_4_write
)
372 _unlock_pcol_pages(pcol
, ret
, READ
);
380 /* readpage_strip is called either directly from readpage() or by the VFS from
381 * within read_cache_pages(), to add one more page to be read. It will try to
382 * collect as many contiguous pages as posible. If a discontinuity is
383 * encountered, or it runs out of resources, it will submit the previous segment
384 * and will start a new collection. Eventually caller must submit the last
385 * segment if present.
387 static int readpage_strip(void *data
, struct page
*page
)
389 struct page_collect
*pcol
= data
;
390 struct inode
*inode
= pcol
->inode
;
391 struct exofs_i_info
*oi
= exofs_i(inode
);
392 loff_t i_size
= i_size_read(inode
);
393 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
397 /* FIXME: Just for debugging, will be removed */
398 if (PageUptodate(page
))
399 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol
->inode
->i_ino
,
402 if (page
->index
< end_index
)
403 len
= PAGE_CACHE_SIZE
;
404 else if (page
->index
== end_index
)
405 len
= i_size
& ~PAGE_CACHE_MASK
;
409 if (!len
|| !obj_created(oi
)) {
410 /* this will be out of bounds, or doesn't exist yet.
411 * Current page is cleared and the request is split
413 clear_highpage(page
);
415 SetPageUptodate(page
);
417 ClearPageError(page
);
419 if (!pcol
->read_4_write
)
421 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
422 "read_4_write=%d index=0x%lx end_index=0x%lx "
423 "splitting\n", inode
->i_ino
, len
,
424 pcol
->read_4_write
, page
->index
, end_index
);
426 return read_exec(pcol
);
431 if (unlikely(pcol
->pg_first
== -1)) {
432 pcol
->pg_first
= page
->index
;
433 } else if (unlikely((pcol
->pg_first
+ pcol
->nr_pages
) !=
435 /* Discontinuity detected, split the request */
436 ret
= read_exec(pcol
);
443 ret
= pcol_try_alloc(pcol
);
448 if (len
!= PAGE_CACHE_SIZE
)
449 zero_user(page
, len
, PAGE_CACHE_SIZE
- len
);
451 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
452 inode
->i_ino
, page
->index
, len
);
454 ret
= pcol_add_page(pcol
, page
, len
);
456 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
457 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
458 page
, len
, pcol
->nr_pages
, pcol
->length
);
460 /* split the request, and start again with current page */
461 ret
= read_exec(pcol
);
471 /* SetPageError(page); ??? */
476 static int exofs_readpages(struct file
*file
, struct address_space
*mapping
,
477 struct list_head
*pages
, unsigned nr_pages
)
479 struct page_collect pcol
;
482 _pcol_init(&pcol
, nr_pages
, mapping
->host
);
484 ret
= read_cache_pages(mapping
, pages
, readpage_strip
, &pcol
);
486 EXOFS_ERR("read_cache_pages => %d\n", ret
);
490 ret
= read_exec(&pcol
);
494 return read_exec(&pcol
);
497 static int _readpage(struct page
*page
, bool read_4_write
)
499 struct page_collect pcol
;
502 _pcol_init(&pcol
, 1, page
->mapping
->host
);
504 pcol
.read_4_write
= read_4_write
;
505 ret
= readpage_strip(&pcol
, page
);
507 EXOFS_ERR("_readpage => %d\n", ret
);
511 return read_exec(&pcol
);
515 * We don't need the file
517 static int exofs_readpage(struct file
*file
, struct page
*page
)
519 return _readpage(page
, false);
522 /* Callback for osd_write. All writes are asynchronous */
523 static void writepages_done(struct ore_io_state
*ios
, void *p
)
525 struct page_collect
*pcol
= p
;
530 int ret
= ore_check_io(ios
, &resid
);
532 atomic_dec(&pcol
->sbi
->s_curr_pending
);
535 good_bytes
= pcol
->length
;
536 ret
= PAGE_WAS_NOT_IN_IO
;
538 good_bytes
= pcol
->length
- resid
;
540 if (good_bytes
> pcol
->ios
->length
)
541 good_bytes
= pcol
->ios
->length
;
543 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
544 " length=0x%lx nr_pages=%u\n",
545 pcol
->inode
->i_ino
, _LLU(good_bytes
), pcol
->length
,
548 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
549 struct page
*page
= pcol
->pages
[i
];
550 struct inode
*inode
= page
->mapping
->host
;
553 if (inode
!= pcol
->inode
)
554 continue; /* osd might add more pages to a bio */
556 if (likely(length
< good_bytes
))
561 update_write_page(page
, page_stat
);
563 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
564 inode
->i_ino
, page
->index
, page_stat
);
571 EXOFS_DBGMSG2("writepages_done END\n");
574 static int write_exec(struct page_collect
*pcol
)
576 struct exofs_i_info
*oi
= exofs_i(pcol
->inode
);
577 struct ore_io_state
*ios
;
578 struct page_collect
*pcol_copy
= NULL
;
585 ret
= ore_get_rw_state(&pcol
->sbi
->layout
, &oi
->oc
, false,
586 pcol
->pg_first
<< PAGE_CACHE_SHIFT
,
587 pcol
->length
, &pcol
->ios
);
591 pcol_copy
= kmalloc(sizeof(*pcol_copy
), GFP_KERNEL
);
593 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
601 ios
->pages
= pcol_copy
->pages
;
602 ios
->done
= writepages_done
;
603 ios
->private = pcol_copy
;
605 /* pages ownership was passed to pcol_copy */
608 ret
= _maybe_not_all_in_one_io(ios
, pcol_copy
, pcol
);
612 EXOFS_DBGMSG2("write_exec(0x%lx) offset=0x%llx length=0x%llx\n",
613 pcol
->inode
->i_ino
, _LLU(ios
->offset
), _LLU(ios
->length
));
615 ret
= ore_write(ios
);
617 EXOFS_ERR("write_exec: ore_write() Failed\n");
621 atomic_inc(&pcol
->sbi
->s_curr_pending
);
625 _unlock_pcol_pages(pcol
, ret
, WRITE
);
632 /* writepage_strip is called either directly from writepage() or by the VFS from
633 * within write_cache_pages(), to add one more page to be written to storage.
634 * It will try to collect as many contiguous pages as possible. If a
635 * discontinuity is encountered or it runs out of resources it will submit the
636 * previous segment and will start a new collection.
637 * Eventually caller must submit the last segment if present.
639 static int writepage_strip(struct page
*page
,
640 struct writeback_control
*wbc_unused
, void *data
)
642 struct page_collect
*pcol
= data
;
643 struct inode
*inode
= pcol
->inode
;
644 struct exofs_i_info
*oi
= exofs_i(inode
);
645 loff_t i_size
= i_size_read(inode
);
646 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
650 BUG_ON(!PageLocked(page
));
652 ret
= wait_obj_created(oi
);
656 if (page
->index
< end_index
)
657 /* in this case, the page is within the limits of the file */
658 len
= PAGE_CACHE_SIZE
;
660 len
= i_size
& ~PAGE_CACHE_MASK
;
662 if (page
->index
> end_index
|| !len
) {
663 /* in this case, the page is outside the limits
664 * (truncate in progress)
666 ret
= write_exec(pcol
);
670 ClearPageError(page
);
672 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
673 "outside the limits\n",
674 inode
->i_ino
, page
->index
);
681 if (unlikely(pcol
->pg_first
== -1)) {
682 pcol
->pg_first
= page
->index
;
683 } else if (unlikely((pcol
->pg_first
+ pcol
->nr_pages
) !=
685 /* Discontinuity detected, split the request */
686 ret
= write_exec(pcol
);
690 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
691 inode
->i_ino
, page
->index
);
696 ret
= pcol_try_alloc(pcol
);
701 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
702 inode
->i_ino
, page
->index
, len
);
704 ret
= pcol_add_page(pcol
, page
, len
);
706 EXOFS_DBGMSG2("Failed pcol_add_page "
707 "nr_pages=%u total_length=0x%lx\n",
708 pcol
->nr_pages
, pcol
->length
);
710 /* split the request, next loop will start again */
711 ret
= write_exec(pcol
);
713 EXOFS_DBGMSG("write_exec failed => %d", ret
);
720 BUG_ON(PageWriteback(page
));
721 set_page_writeback(page
);
726 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
727 inode
->i_ino
, page
->index
, ret
);
728 set_bit(AS_EIO
, &page
->mapping
->flags
);
733 static int exofs_writepages(struct address_space
*mapping
,
734 struct writeback_control
*wbc
)
736 struct page_collect pcol
;
737 long start
, end
, expected_pages
;
740 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
741 end
= (wbc
->range_end
== LLONG_MAX
) ?
742 start
+ mapping
->nrpages
:
743 wbc
->range_end
>> PAGE_CACHE_SHIFT
;
746 expected_pages
= end
- start
+ 1;
748 expected_pages
= mapping
->nrpages
;
750 if (expected_pages
< 32L)
751 expected_pages
= 32L;
753 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
754 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
755 mapping
->host
->i_ino
, wbc
->range_start
, wbc
->range_end
,
756 mapping
->nrpages
, start
, end
, expected_pages
);
758 _pcol_init(&pcol
, expected_pages
, mapping
->host
);
760 ret
= write_cache_pages(mapping
, wbc
, writepage_strip
, &pcol
);
762 EXOFS_ERR("write_cache_pages => %d\n", ret
);
766 ret
= write_exec(&pcol
);
770 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
771 return write_exec(&pcol
); /* pump the last reminder */
772 } else if (pcol
.nr_pages
) {
773 /* not SYNC let the reminder join the next writeout */
776 for (i
= 0; i
< pcol
.nr_pages
; i
++) {
777 struct page
*page
= pcol
.pages
[i
];
779 end_page_writeback(page
);
780 set_page_dirty(page
);
787 static int exofs_writepage(struct page
*page
, struct writeback_control
*wbc
)
789 struct page_collect pcol
;
792 _pcol_init(&pcol
, 1, page
->mapping
->host
);
794 ret
= writepage_strip(page
, NULL
, &pcol
);
796 EXOFS_ERR("exofs_writepage => %d\n", ret
);
800 return write_exec(&pcol
);
803 /* i_mutex held using inode->i_size directly */
804 static void _write_failed(struct inode
*inode
, loff_t to
)
806 if (to
> inode
->i_size
)
807 truncate_pagecache(inode
, to
, inode
->i_size
);
810 int exofs_write_begin(struct file
*file
, struct address_space
*mapping
,
811 loff_t pos
, unsigned len
, unsigned flags
,
812 struct page
**pagep
, void **fsdata
)
819 ret
= simple_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
822 EXOFS_DBGMSG("simple_write_begin failed\n");
829 /* read modify write */
830 if (!PageUptodate(page
) && (len
!= PAGE_CACHE_SIZE
)) {
831 loff_t i_size
= i_size_read(mapping
->host
);
832 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
835 if (page
->index
< end_index
)
836 rlen
= PAGE_CACHE_SIZE
;
837 else if (page
->index
== end_index
)
838 rlen
= i_size
& ~PAGE_CACHE_MASK
;
843 clear_highpage(page
);
844 SetPageUptodate(page
);
848 ret
= _readpage(page
, true);
850 /*SetPageError was done by _readpage. Is it ok?*/
852 EXOFS_DBGMSG("__readpage failed\n");
857 _write_failed(mapping
->host
, pos
+ len
);
862 static int exofs_write_begin_export(struct file
*file
,
863 struct address_space
*mapping
,
864 loff_t pos
, unsigned len
, unsigned flags
,
865 struct page
**pagep
, void **fsdata
)
869 return exofs_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
873 static int exofs_write_end(struct file
*file
, struct address_space
*mapping
,
874 loff_t pos
, unsigned len
, unsigned copied
,
875 struct page
*page
, void *fsdata
)
877 struct inode
*inode
= mapping
->host
;
878 /* According to comment in simple_write_end i_mutex is held */
879 loff_t i_size
= inode
->i_size
;
882 ret
= simple_write_end(file
, mapping
,pos
, len
, copied
, page
, fsdata
);
884 _write_failed(inode
, pos
+ len
);
886 /* TODO: once simple_write_end marks inode dirty remove */
887 if (i_size
!= inode
->i_size
)
888 mark_inode_dirty(inode
);
892 static int exofs_releasepage(struct page
*page
, gfp_t gfp
)
894 EXOFS_DBGMSG("page 0x%lx\n", page
->index
);
899 static void exofs_invalidatepage(struct page
*page
, unsigned long offset
)
901 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page
->index
, offset
);
905 const struct address_space_operations exofs_aops
= {
906 .readpage
= exofs_readpage
,
907 .readpages
= exofs_readpages
,
908 .writepage
= exofs_writepage
,
909 .writepages
= exofs_writepages
,
910 .write_begin
= exofs_write_begin_export
,
911 .write_end
= exofs_write_end
,
912 .releasepage
= exofs_releasepage
,
913 .set_page_dirty
= __set_page_dirty_nobuffers
,
914 .invalidatepage
= exofs_invalidatepage
,
916 /* Not implemented Yet */
917 .bmap
= NULL
, /* TODO: use osd's OSD_ACT_READ_MAP */
918 .direct_IO
= NULL
, /* TODO: Should be trivial to do */
920 /* With these NULL has special meaning or default is not exported */
923 .launder_page
= NULL
,
924 .is_partially_uptodate
= NULL
,
925 .error_remove_page
= NULL
,
928 /******************************************************************************
930 *****************************************************************************/
933 * Test whether an inode is a fast symlink.
935 static inline int exofs_inode_is_fast_symlink(struct inode
*inode
)
937 struct exofs_i_info
*oi
= exofs_i(inode
);
939 return S_ISLNK(inode
->i_mode
) && (oi
->i_data
[0] != 0);
942 static int _do_truncate(struct inode
*inode
, loff_t newsize
)
944 struct exofs_i_info
*oi
= exofs_i(inode
);
945 struct exofs_sb_info
*sbi
= inode
->i_sb
->s_fs_info
;
948 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
950 ret
= ore_truncate(&sbi
->layout
, &oi
->oc
, (u64
)newsize
);
952 truncate_setsize(inode
, newsize
);
954 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
955 inode
->i_ino
, newsize
, ret
);
960 * Set inode attributes - update size attribute on OSD if needed,
961 * otherwise just call generic functions.
963 int exofs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
965 struct inode
*inode
= dentry
->d_inode
;
968 /* if we are about to modify an object, and it hasn't been
971 error
= wait_obj_created(exofs_i(inode
));
975 error
= inode_change_ok(inode
, iattr
);
979 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
980 iattr
->ia_size
!= i_size_read(inode
)) {
981 error
= _do_truncate(inode
, iattr
->ia_size
);
986 setattr_copy(inode
, iattr
);
987 mark_inode_dirty(inode
);
991 static const struct osd_attr g_attr_inode_file_layout
= ATTR_DEF(
993 EXOFS_ATTR_INODE_FILE_LAYOUT
,
995 static const struct osd_attr g_attr_inode_dir_layout
= ATTR_DEF(
997 EXOFS_ATTR_INODE_DIR_LAYOUT
,
1001 * Read the Linux inode info from the OSD, and return it as is. In exofs the
1002 * inode info is in an application specific page/attribute of the osd-object.
1004 static int exofs_get_inode(struct super_block
*sb
, struct exofs_i_info
*oi
,
1005 struct exofs_fcb
*inode
)
1007 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1008 struct osd_attr attrs
[] = {
1009 [0] = g_attr_inode_data
,
1010 [1] = g_attr_inode_file_layout
,
1011 [2] = g_attr_inode_dir_layout
,
1013 struct ore_io_state
*ios
;
1014 struct exofs_on_disk_inode_layout
*layout
;
1017 ret
= ore_get_io_state(&sbi
->layout
, &oi
->oc
, &ios
);
1018 if (unlikely(ret
)) {
1019 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__
);
1023 attrs
[1].len
= exofs_on_disk_inode_layout_size(sbi
->oc
.numdevs
);
1024 attrs
[2].len
= exofs_on_disk_inode_layout_size(sbi
->oc
.numdevs
);
1026 ios
->in_attr
= attrs
;
1027 ios
->in_attr_len
= ARRAY_SIZE(attrs
);
1029 ret
= ore_read(ios
);
1030 if (unlikely(ret
)) {
1031 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
1032 _LLU(oi
->one_comp
.obj
.id
), ret
);
1033 memset(inode
, 0, sizeof(*inode
));
1034 inode
->i_mode
= 0040000 | (0777 & ~022);
1035 /* If object is lost on target we might as well enable it's
1038 if ((ret
== -ENOENT
) || (ret
== -EINVAL
))
1043 ret
= extract_attr_from_ios(ios
, &attrs
[0]);
1045 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
1048 WARN_ON(attrs
[0].len
!= EXOFS_INO_ATTR_SIZE
);
1049 memcpy(inode
, attrs
[0].val_ptr
, EXOFS_INO_ATTR_SIZE
);
1051 ret
= extract_attr_from_ios(ios
, &attrs
[1]);
1053 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
1057 layout
= attrs
[1].val_ptr
;
1058 if (layout
->gen_func
!= cpu_to_le16(LAYOUT_MOVING_WINDOW
)) {
1059 EXOFS_ERR("%s: unsupported files layout %d\n",
1060 __func__
, layout
->gen_func
);
1066 ret
= extract_attr_from_ios(ios
, &attrs
[2]);
1068 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
1072 layout
= attrs
[2].val_ptr
;
1073 if (layout
->gen_func
!= cpu_to_le16(LAYOUT_MOVING_WINDOW
)) {
1074 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
1075 __func__
, layout
->gen_func
);
1082 ore_put_io_state(ios
);
1086 static void __oi_init(struct exofs_i_info
*oi
)
1088 init_waitqueue_head(&oi
->i_wq
);
1092 * Fill in an inode read from the OSD and set it up for use
1094 struct inode
*exofs_iget(struct super_block
*sb
, unsigned long ino
)
1096 struct exofs_i_info
*oi
;
1097 struct exofs_fcb fcb
;
1098 struct inode
*inode
;
1101 inode
= iget_locked(sb
, ino
);
1103 return ERR_PTR(-ENOMEM
);
1104 if (!(inode
->i_state
& I_NEW
))
1106 oi
= exofs_i(inode
);
1108 exofs_init_comps(&oi
->oc
, &oi
->one_comp
, sb
->s_fs_info
,
1109 exofs_oi_objno(oi
));
1111 /* read the inode from the osd */
1112 ret
= exofs_get_inode(sb
, oi
, &fcb
);
1116 set_obj_created(oi
);
1118 /* copy stuff from on-disk struct to in-memory struct */
1119 inode
->i_mode
= le16_to_cpu(fcb
.i_mode
);
1120 inode
->i_uid
= le32_to_cpu(fcb
.i_uid
);
1121 inode
->i_gid
= le32_to_cpu(fcb
.i_gid
);
1122 inode
->i_nlink
= le16_to_cpu(fcb
.i_links_count
);
1123 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_ctime
);
1124 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_atime
);
1125 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_mtime
);
1126 inode
->i_ctime
.tv_nsec
=
1127 inode
->i_atime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
1128 oi
->i_commit_size
= le64_to_cpu(fcb
.i_size
);
1129 i_size_write(inode
, oi
->i_commit_size
);
1130 inode
->i_blkbits
= EXOFS_BLKSHIFT
;
1131 inode
->i_generation
= le32_to_cpu(fcb
.i_generation
);
1133 oi
->i_dir_start_lookup
= 0;
1135 if ((inode
->i_nlink
== 0) && (inode
->i_mode
== 0)) {
1140 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
1143 old_decode_dev(le32_to_cpu(fcb
.i_data
[0]));
1146 new_decode_dev(le32_to_cpu(fcb
.i_data
[1]));
1148 memcpy(oi
->i_data
, fcb
.i_data
, sizeof(fcb
.i_data
));
1151 inode
->i_mapping
->backing_dev_info
= sb
->s_bdi
;
1152 if (S_ISREG(inode
->i_mode
)) {
1153 inode
->i_op
= &exofs_file_inode_operations
;
1154 inode
->i_fop
= &exofs_file_operations
;
1155 inode
->i_mapping
->a_ops
= &exofs_aops
;
1156 } else if (S_ISDIR(inode
->i_mode
)) {
1157 inode
->i_op
= &exofs_dir_inode_operations
;
1158 inode
->i_fop
= &exofs_dir_operations
;
1159 inode
->i_mapping
->a_ops
= &exofs_aops
;
1160 } else if (S_ISLNK(inode
->i_mode
)) {
1161 if (exofs_inode_is_fast_symlink(inode
))
1162 inode
->i_op
= &exofs_fast_symlink_inode_operations
;
1164 inode
->i_op
= &exofs_symlink_inode_operations
;
1165 inode
->i_mapping
->a_ops
= &exofs_aops
;
1168 inode
->i_op
= &exofs_special_inode_operations
;
1170 init_special_inode(inode
, inode
->i_mode
,
1171 old_decode_dev(le32_to_cpu(fcb
.i_data
[0])));
1173 init_special_inode(inode
, inode
->i_mode
,
1174 new_decode_dev(le32_to_cpu(fcb
.i_data
[1])));
1177 unlock_new_inode(inode
);
1182 return ERR_PTR(ret
);
1185 int __exofs_wait_obj_created(struct exofs_i_info
*oi
)
1187 if (!obj_created(oi
)) {
1188 EXOFS_DBGMSG("!obj_created\n");
1189 BUG_ON(!obj_2bcreated(oi
));
1190 wait_event(oi
->i_wq
, obj_created(oi
));
1191 EXOFS_DBGMSG("wait_event done\n");
1193 return unlikely(is_bad_inode(&oi
->vfs_inode
)) ? -EIO
: 0;
1197 * Callback function from exofs_new_inode(). The important thing is that we
1198 * set the obj_created flag so that other methods know that the object exists on
1201 static void create_done(struct ore_io_state
*ios
, void *p
)
1203 struct inode
*inode
= p
;
1204 struct exofs_i_info
*oi
= exofs_i(inode
);
1205 struct exofs_sb_info
*sbi
= inode
->i_sb
->s_fs_info
;
1208 ret
= ore_check_io(ios
, NULL
);
1209 ore_put_io_state(ios
);
1211 atomic_dec(&sbi
->s_curr_pending
);
1213 if (unlikely(ret
)) {
1214 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1215 _LLU(exofs_oi_objno(oi
)),
1216 _LLU(oi
->one_comp
.obj
.partition
));
1217 /*TODO: When FS is corrupted creation can fail, object already
1218 * exist. Get rid of this asynchronous creation, if exist
1219 * increment the obj counter and try the next object. Until we
1220 * succeed. All these dangling objects will be made into lost
1221 * files by chkfs.exofs
1225 set_obj_created(oi
);
1231 * Set up a new inode and create an object for it on the OSD
1233 struct inode
*exofs_new_inode(struct inode
*dir
, int mode
)
1235 struct super_block
*sb
= dir
->i_sb
;
1236 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1237 struct inode
*inode
;
1238 struct exofs_i_info
*oi
;
1239 struct ore_io_state
*ios
;
1242 inode
= new_inode(sb
);
1244 return ERR_PTR(-ENOMEM
);
1246 oi
= exofs_i(inode
);
1249 set_obj_2bcreated(oi
);
1251 inode
->i_mapping
->backing_dev_info
= sb
->s_bdi
;
1252 inode_init_owner(inode
, dir
, mode
);
1253 inode
->i_ino
= sbi
->s_nextid
++;
1254 inode
->i_blkbits
= EXOFS_BLKSHIFT
;
1255 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
1256 oi
->i_commit_size
= inode
->i_size
= 0;
1257 spin_lock(&sbi
->s_next_gen_lock
);
1258 inode
->i_generation
= sbi
->s_next_generation
++;
1259 spin_unlock(&sbi
->s_next_gen_lock
);
1260 insert_inode_hash(inode
);
1262 exofs_init_comps(&oi
->oc
, &oi
->one_comp
, sb
->s_fs_info
,
1263 exofs_oi_objno(oi
));
1264 exofs_sbi_write_stats(sbi
); /* Make sure new sbi->s_nextid is on disk */
1266 mark_inode_dirty(inode
);
1268 ret
= ore_get_io_state(&sbi
->layout
, &oi
->oc
, &ios
);
1269 if (unlikely(ret
)) {
1270 EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n");
1271 return ERR_PTR(ret
);
1274 ios
->done
= create_done
;
1275 ios
->private = inode
;
1277 ret
= ore_create(ios
);
1279 ore_put_io_state(ios
);
1280 return ERR_PTR(ret
);
1282 atomic_inc(&sbi
->s_curr_pending
);
1288 * struct to pass two arguments to update_inode's callback
1290 struct updatei_args
{
1291 struct exofs_sb_info
*sbi
;
1292 struct exofs_fcb fcb
;
1296 * Callback function from exofs_update_inode().
1298 static void updatei_done(struct ore_io_state
*ios
, void *p
)
1300 struct updatei_args
*args
= p
;
1302 ore_put_io_state(ios
);
1304 atomic_dec(&args
->sbi
->s_curr_pending
);
1310 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1311 * synchronously or asynchronously depending on the do_sync flag.
1313 static int exofs_update_inode(struct inode
*inode
, int do_sync
)
1315 struct exofs_i_info
*oi
= exofs_i(inode
);
1316 struct super_block
*sb
= inode
->i_sb
;
1317 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1318 struct ore_io_state
*ios
;
1319 struct osd_attr attr
;
1320 struct exofs_fcb
*fcb
;
1321 struct updatei_args
*args
;
1324 args
= kzalloc(sizeof(*args
), GFP_KERNEL
);
1326 EXOFS_DBGMSG("Failed kzalloc of args\n");
1332 fcb
->i_mode
= cpu_to_le16(inode
->i_mode
);
1333 fcb
->i_uid
= cpu_to_le32(inode
->i_uid
);
1334 fcb
->i_gid
= cpu_to_le32(inode
->i_gid
);
1335 fcb
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1336 fcb
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
1337 fcb
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
1338 fcb
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
1339 oi
->i_commit_size
= i_size_read(inode
);
1340 fcb
->i_size
= cpu_to_le64(oi
->i_commit_size
);
1341 fcb
->i_generation
= cpu_to_le32(inode
->i_generation
);
1343 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
1344 if (old_valid_dev(inode
->i_rdev
)) {
1346 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
1351 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
1355 memcpy(fcb
->i_data
, oi
->i_data
, sizeof(fcb
->i_data
));
1357 ret
= ore_get_io_state(&sbi
->layout
, &oi
->oc
, &ios
);
1358 if (unlikely(ret
)) {
1359 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__
);
1363 attr
= g_attr_inode_data
;
1365 ios
->out_attr_len
= 1;
1366 ios
->out_attr
= &attr
;
1368 wait_obj_created(oi
);
1372 ios
->done
= updatei_done
;
1373 ios
->private = args
;
1376 ret
= ore_write(ios
);
1377 if (!do_sync
&& !ret
) {
1378 atomic_inc(&sbi
->s_curr_pending
);
1379 goto out
; /* deallocation in updatei_done */
1382 ore_put_io_state(ios
);
1386 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1387 inode
->i_ino
, do_sync
, ret
);
1391 int exofs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1393 /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1394 return exofs_update_inode(inode
, 1);
1398 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1401 static void delete_done(struct ore_io_state
*ios
, void *p
)
1403 struct exofs_sb_info
*sbi
= p
;
1405 ore_put_io_state(ios
);
1407 atomic_dec(&sbi
->s_curr_pending
);
1411 * Called when the refcount of an inode reaches zero. We remove the object
1412 * from the OSD here. We make sure the object was created before we try and
1415 void exofs_evict_inode(struct inode
*inode
)
1417 struct exofs_i_info
*oi
= exofs_i(inode
);
1418 struct super_block
*sb
= inode
->i_sb
;
1419 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1420 struct ore_io_state
*ios
;
1423 truncate_inode_pages(&inode
->i_data
, 0);
1425 /* TODO: should do better here */
1426 if (inode
->i_nlink
|| is_bad_inode(inode
))
1430 end_writeback(inode
);
1432 /* if we are deleting an obj that hasn't been created yet, wait.
1433 * This also makes sure that create_done cannot be called with an
1434 * already evicted inode.
1436 wait_obj_created(oi
);
1437 /* ignore the error, attempt a remove anyway */
1439 /* Now Remove the OSD objects */
1440 ret
= ore_get_io_state(&sbi
->layout
, &oi
->oc
, &ios
);
1441 if (unlikely(ret
)) {
1442 EXOFS_ERR("%s: ore_get_io_state failed\n", __func__
);
1446 ios
->done
= delete_done
;
1449 ret
= ore_remove(ios
);
1451 EXOFS_ERR("%s: ore_remove failed\n", __func__
);
1452 ore_put_io_state(ios
);
1455 atomic_inc(&sbi
->s_curr_pending
);
1460 end_writeback(inode
);