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
*),
47 struct exofs_sb_info
*sbi
;
49 unsigned expected_pages
;
50 struct exofs_io_state
*ios
;
56 loff_t pg_first
; /* keep 64bit also in 32-arches */
57 bool read_4_write
; /* This means two things: that the read is sync
58 * And the pages should not be unlocked.
62 static void _pcol_init(struct page_collect
*pcol
, unsigned expected_pages
,
65 struct exofs_sb_info
*sbi
= inode
->i_sb
->s_fs_info
;
69 pcol
->expected_pages
= expected_pages
;
73 pcol
->alloc_pages
= 0;
77 pcol
->read_4_write
= false;
80 static void _pcol_reset(struct page_collect
*pcol
)
82 pcol
->expected_pages
-= min(pcol
->nr_pages
, pcol
->expected_pages
);
85 pcol
->alloc_pages
= 0;
91 /* this is probably the end of the loop but in writes
92 * it might not end here. don't be left with nothing
94 if (!pcol
->expected_pages
)
95 pcol
->expected_pages
= MAX_PAGES_KMALLOC
;
98 static int pcol_try_alloc(struct page_collect
*pcol
)
100 unsigned pages
= min_t(unsigned, pcol
->expected_pages
,
103 if (!pcol
->ios
) { /* First time allocate io_state */
104 int ret
= exofs_get_io_state(&pcol
->sbi
->layout
, &pcol
->ios
);
110 /* TODO: easily support bio chaining */
111 pages
= min_t(unsigned, pages
,
112 pcol
->sbi
->layout
.group_width
* BIO_MAX_PAGES_KMALLOC
);
114 for (; pages
; pages
>>= 1) {
115 pcol
->pages
= kmalloc(pages
* sizeof(struct page
*),
117 if (likely(pcol
->pages
)) {
118 pcol
->alloc_pages
= pages
;
123 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
124 pcol
->expected_pages
);
128 static void pcol_free(struct page_collect
*pcol
)
134 exofs_put_io_state(pcol
->ios
);
139 static int pcol_add_page(struct page_collect
*pcol
, struct page
*page
,
142 if (unlikely(pcol
->nr_pages
>= pcol
->alloc_pages
))
145 pcol
->pages
[pcol
->nr_pages
++] = page
;
150 static int update_read_page(struct page
*page
, int ret
)
153 /* Everything is OK */
154 SetPageUptodate(page
);
156 ClearPageError(page
);
157 } else if (ret
== -EFAULT
) {
158 /* In this case we were trying to read something that wasn't on
159 * disk yet - return a page full of zeroes. This should be OK,
160 * because the object should be empty (if there was a write
161 * before this read, the read would be waiting with the page
163 clear_highpage(page
);
165 SetPageUptodate(page
);
167 ClearPageError(page
);
168 ret
= 0; /* recovered error */
169 EXOFS_DBGMSG("recovered read error\n");
176 static void update_write_page(struct page
*page
, int ret
)
179 mapping_set_error(page
->mapping
, ret
);
182 end_page_writeback(page
);
185 /* Called at the end of reads, to optionally unlock pages and update their
188 static int __readpages_done(struct page_collect
*pcol
, bool do_unlock
)
194 int ret
= exofs_check_io(pcol
->ios
, &resid
);
197 good_bytes
= pcol
->length
;
199 good_bytes
= pcol
->length
- resid
;
201 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
202 " length=0x%lx nr_pages=%u\n",
203 pcol
->inode
->i_ino
, _LLU(good_bytes
), pcol
->length
,
206 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
207 struct page
*page
= pcol
->pages
[i
];
208 struct inode
*inode
= page
->mapping
->host
;
211 if (inode
!= pcol
->inode
)
212 continue; /* osd might add more pages at end */
214 if (likely(length
< good_bytes
))
219 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
220 inode
->i_ino
, page
->index
,
221 page_stat
? "bad_bytes" : "good_bytes");
223 ret
= update_read_page(page
, page_stat
);
230 EXOFS_DBGMSG2("readpages_done END\n");
234 /* callback of async reads */
235 static void readpages_done(struct exofs_io_state
*ios
, void *p
)
237 struct page_collect
*pcol
= p
;
239 __readpages_done(pcol
, true);
240 atomic_dec(&pcol
->sbi
->s_curr_pending
);
244 static void _unlock_pcol_pages(struct page_collect
*pcol
, int ret
, int rw
)
248 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
249 struct page
*page
= pcol
->pages
[i
];
252 update_read_page(page
, ret
);
254 update_write_page(page
, ret
);
260 static int read_exec(struct page_collect
*pcol
, bool is_sync
)
262 struct exofs_i_info
*oi
= exofs_i(pcol
->inode
);
263 struct exofs_io_state
*ios
= pcol
->ios
;
264 struct page_collect
*pcol_copy
= NULL
;
270 /* see comment in _readpage() about sync reads */
271 WARN_ON(is_sync
&& (pcol
->nr_pages
!= 1));
273 ios
->pages
= pcol
->pages
;
274 ios
->nr_pages
= pcol
->nr_pages
;
275 ios
->length
= pcol
->length
;
276 ios
->offset
= pcol
->pg_first
<< PAGE_CACHE_SHIFT
;
279 exofs_oi_read(oi
, pcol
->ios
);
280 return __readpages_done(pcol
, false);
283 pcol_copy
= kmalloc(sizeof(*pcol_copy
), GFP_KERNEL
);
290 ios
->done
= readpages_done
;
291 ios
->private = pcol_copy
;
292 ret
= exofs_oi_read(oi
, ios
);
296 atomic_inc(&pcol
->sbi
->s_curr_pending
);
298 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
299 ios
->obj
.id
, _LLU(ios
->offset
), pcol
->length
);
301 /* pages ownership was passed to pcol_copy */
307 _unlock_pcol_pages(pcol
, ret
, READ
);
315 /* readpage_strip is called either directly from readpage() or by the VFS from
316 * within read_cache_pages(), to add one more page to be read. It will try to
317 * collect as many contiguous pages as posible. If a discontinuity is
318 * encountered, or it runs out of resources, it will submit the previous segment
319 * and will start a new collection. Eventually caller must submit the last
320 * segment if present.
322 static int readpage_strip(void *data
, struct page
*page
)
324 struct page_collect
*pcol
= data
;
325 struct inode
*inode
= pcol
->inode
;
326 struct exofs_i_info
*oi
= exofs_i(inode
);
327 loff_t i_size
= i_size_read(inode
);
328 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
332 if (PageUptodate(page
))
333 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol
->inode
->i_ino
,
336 if (page
->index
< end_index
)
337 len
= PAGE_CACHE_SIZE
;
338 else if (page
->index
== end_index
)
339 len
= i_size
& ~PAGE_CACHE_MASK
;
343 if (!len
|| !obj_created(oi
)) {
344 /* this will be out of bounds, or doesn't exist yet.
345 * Current page is cleared and the request is split
347 clear_highpage(page
);
349 SetPageUptodate(page
);
351 ClearPageError(page
);
353 if (!pcol
->read_4_write
)
355 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
356 " splitting\n", inode
->i_ino
, page
->index
);
358 return read_exec(pcol
, false);
363 if (unlikely(pcol
->pg_first
== -1)) {
364 pcol
->pg_first
= page
->index
;
365 } else if (unlikely((pcol
->pg_first
+ pcol
->nr_pages
) !=
367 /* Discontinuity detected, split the request */
368 ret
= read_exec(pcol
, false);
375 ret
= pcol_try_alloc(pcol
);
380 if (len
!= PAGE_CACHE_SIZE
)
381 zero_user(page
, len
, PAGE_CACHE_SIZE
- len
);
383 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
384 inode
->i_ino
, page
->index
, len
);
386 ret
= pcol_add_page(pcol
, page
, len
);
388 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
389 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
390 page
, len
, pcol
->nr_pages
, pcol
->length
);
392 /* split the request, and start again with current page */
393 ret
= read_exec(pcol
, false);
403 /* SetPageError(page); ??? */
408 static int exofs_readpages(struct file
*file
, struct address_space
*mapping
,
409 struct list_head
*pages
, unsigned nr_pages
)
411 struct page_collect pcol
;
414 _pcol_init(&pcol
, nr_pages
, mapping
->host
);
416 ret
= read_cache_pages(mapping
, pages
, readpage_strip
, &pcol
);
418 EXOFS_ERR("read_cache_pages => %d\n", ret
);
422 return read_exec(&pcol
, false);
425 static int _readpage(struct page
*page
, bool is_sync
)
427 struct page_collect pcol
;
430 _pcol_init(&pcol
, 1, page
->mapping
->host
);
432 /* readpage_strip might call read_exec(,is_sync==false) at several
433 * places but not if we have a single page.
435 pcol
.read_4_write
= is_sync
;
436 ret
= readpage_strip(&pcol
, page
);
438 EXOFS_ERR("_readpage => %d\n", ret
);
442 return read_exec(&pcol
, is_sync
);
446 * We don't need the file
448 static int exofs_readpage(struct file
*file
, struct page
*page
)
450 return _readpage(page
, false);
453 /* Callback for osd_write. All writes are asynchronous */
454 static void writepages_done(struct exofs_io_state
*ios
, void *p
)
456 struct page_collect
*pcol
= p
;
461 int ret
= exofs_check_io(ios
, &resid
);
463 atomic_dec(&pcol
->sbi
->s_curr_pending
);
466 good_bytes
= pcol
->length
;
468 good_bytes
= pcol
->length
- resid
;
470 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
471 " length=0x%lx nr_pages=%u\n",
472 pcol
->inode
->i_ino
, _LLU(good_bytes
), pcol
->length
,
475 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
476 struct page
*page
= pcol
->pages
[i
];
477 struct inode
*inode
= page
->mapping
->host
;
480 if (inode
!= pcol
->inode
)
481 continue; /* osd might add more pages to a bio */
483 if (likely(length
< good_bytes
))
488 update_write_page(page
, page_stat
);
490 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
491 inode
->i_ino
, page
->index
, page_stat
);
498 EXOFS_DBGMSG2("writepages_done END\n");
501 static int write_exec(struct page_collect
*pcol
)
503 struct exofs_i_info
*oi
= exofs_i(pcol
->inode
);
504 struct exofs_io_state
*ios
= pcol
->ios
;
505 struct page_collect
*pcol_copy
= NULL
;
511 pcol_copy
= kmalloc(sizeof(*pcol_copy
), GFP_KERNEL
);
513 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
520 ios
->pages
= pcol_copy
->pages
;
521 ios
->nr_pages
= pcol_copy
->nr_pages
;
522 ios
->offset
= pcol_copy
->pg_first
<< PAGE_CACHE_SHIFT
;
523 ios
->length
= pcol_copy
->length
;
524 ios
->done
= writepages_done
;
525 ios
->private = pcol_copy
;
527 ret
= exofs_oi_write(oi
, ios
);
529 EXOFS_ERR("write_exec: exofs_oi_write() Faild\n");
533 atomic_inc(&pcol
->sbi
->s_curr_pending
);
534 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
535 pcol
->inode
->i_ino
, pcol
->pg_first
, _LLU(ios
->offset
),
537 /* pages ownership was passed to pcol_copy */
542 _unlock_pcol_pages(pcol
, ret
, WRITE
);
549 /* writepage_strip is called either directly from writepage() or by the VFS from
550 * within write_cache_pages(), to add one more page to be written to storage.
551 * It will try to collect as many contiguous pages as possible. If a
552 * discontinuity is encountered or it runs out of resources it will submit the
553 * previous segment and will start a new collection.
554 * Eventually caller must submit the last segment if present.
556 static int writepage_strip(struct page
*page
,
557 struct writeback_control
*wbc_unused
, void *data
)
559 struct page_collect
*pcol
= data
;
560 struct inode
*inode
= pcol
->inode
;
561 struct exofs_i_info
*oi
= exofs_i(inode
);
562 loff_t i_size
= i_size_read(inode
);
563 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
567 BUG_ON(!PageLocked(page
));
569 ret
= wait_obj_created(oi
);
573 if (page
->index
< end_index
)
574 /* in this case, the page is within the limits of the file */
575 len
= PAGE_CACHE_SIZE
;
577 len
= i_size
& ~PAGE_CACHE_MASK
;
579 if (page
->index
> end_index
|| !len
) {
580 /* in this case, the page is outside the limits
581 * (truncate in progress)
583 ret
= write_exec(pcol
);
587 ClearPageError(page
);
589 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
590 "outside the limits\n",
591 inode
->i_ino
, page
->index
);
598 if (unlikely(pcol
->pg_first
== -1)) {
599 pcol
->pg_first
= page
->index
;
600 } else if (unlikely((pcol
->pg_first
+ pcol
->nr_pages
) !=
602 /* Discontinuity detected, split the request */
603 ret
= write_exec(pcol
);
607 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
608 inode
->i_ino
, page
->index
);
613 ret
= pcol_try_alloc(pcol
);
618 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
619 inode
->i_ino
, page
->index
, len
);
621 ret
= pcol_add_page(pcol
, page
, len
);
623 EXOFS_DBGMSG2("Failed pcol_add_page "
624 "nr_pages=%u total_length=0x%lx\n",
625 pcol
->nr_pages
, pcol
->length
);
627 /* split the request, next loop will start again */
628 ret
= write_exec(pcol
);
630 EXOFS_DBGMSG("write_exec faild => %d", ret
);
637 BUG_ON(PageWriteback(page
));
638 set_page_writeback(page
);
643 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
644 inode
->i_ino
, page
->index
, ret
);
645 set_bit(AS_EIO
, &page
->mapping
->flags
);
650 static int exofs_writepages(struct address_space
*mapping
,
651 struct writeback_control
*wbc
)
653 struct page_collect pcol
;
654 long start
, end
, expected_pages
;
657 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
658 end
= (wbc
->range_end
== LLONG_MAX
) ?
659 start
+ mapping
->nrpages
:
660 wbc
->range_end
>> PAGE_CACHE_SHIFT
;
663 expected_pages
= end
- start
+ 1;
665 expected_pages
= mapping
->nrpages
;
667 if (expected_pages
< 32L)
668 expected_pages
= 32L;
670 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
671 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
672 mapping
->host
->i_ino
, wbc
->range_start
, wbc
->range_end
,
673 mapping
->nrpages
, start
, end
, expected_pages
);
675 _pcol_init(&pcol
, expected_pages
, mapping
->host
);
677 ret
= write_cache_pages(mapping
, wbc
, writepage_strip
, &pcol
);
679 EXOFS_ERR("write_cache_pages => %d\n", ret
);
683 return write_exec(&pcol
);
686 static int exofs_writepage(struct page
*page
, struct writeback_control
*wbc
)
688 struct page_collect pcol
;
691 _pcol_init(&pcol
, 1, page
->mapping
->host
);
693 ret
= writepage_strip(page
, NULL
, &pcol
);
695 EXOFS_ERR("exofs_writepage => %d\n", ret
);
699 return write_exec(&pcol
);
702 /* i_mutex held using inode->i_size directly */
703 static void _write_failed(struct inode
*inode
, loff_t to
)
705 if (to
> inode
->i_size
)
706 truncate_pagecache(inode
, to
, inode
->i_size
);
709 int exofs_write_begin(struct file
*file
, struct address_space
*mapping
,
710 loff_t pos
, unsigned len
, unsigned flags
,
711 struct page
**pagep
, void **fsdata
)
718 ret
= simple_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
721 EXOFS_DBGMSG("simple_write_begin faild\n");
728 /* read modify write */
729 if (!PageUptodate(page
) && (len
!= PAGE_CACHE_SIZE
)) {
730 ret
= _readpage(page
, true);
732 /*SetPageError was done by _readpage. Is it ok?*/
734 EXOFS_DBGMSG("__readpage_filler faild\n");
739 _write_failed(mapping
->host
, pos
+ len
);
744 static int exofs_write_begin_export(struct file
*file
,
745 struct address_space
*mapping
,
746 loff_t pos
, unsigned len
, unsigned flags
,
747 struct page
**pagep
, void **fsdata
)
751 return exofs_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
755 static int exofs_write_end(struct file
*file
, struct address_space
*mapping
,
756 loff_t pos
, unsigned len
, unsigned copied
,
757 struct page
*page
, void *fsdata
)
759 struct inode
*inode
= mapping
->host
;
760 /* According to comment in simple_write_end i_mutex is held */
761 loff_t i_size
= inode
->i_size
;
764 ret
= simple_write_end(file
, mapping
,pos
, len
, copied
, page
, fsdata
);
766 _write_failed(inode
, pos
+ len
);
768 /* TODO: once simple_write_end marks inode dirty remove */
769 if (i_size
!= inode
->i_size
)
770 mark_inode_dirty(inode
);
774 static int exofs_releasepage(struct page
*page
, gfp_t gfp
)
776 EXOFS_DBGMSG("page 0x%lx\n", page
->index
);
781 static void exofs_invalidatepage(struct page
*page
, unsigned long offset
)
783 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page
->index
, offset
);
787 const struct address_space_operations exofs_aops
= {
788 .readpage
= exofs_readpage
,
789 .readpages
= exofs_readpages
,
790 .writepage
= exofs_writepage
,
791 .writepages
= exofs_writepages
,
792 .write_begin
= exofs_write_begin_export
,
793 .write_end
= exofs_write_end
,
794 .releasepage
= exofs_releasepage
,
795 .set_page_dirty
= __set_page_dirty_nobuffers
,
796 .invalidatepage
= exofs_invalidatepage
,
798 /* Not implemented Yet */
799 .bmap
= NULL
, /* TODO: use osd's OSD_ACT_READ_MAP */
800 .direct_IO
= NULL
, /* TODO: Should be trivial to do */
802 /* With these NULL has special meaning or default is not exported */
806 .launder_page
= NULL
,
807 .is_partially_uptodate
= NULL
,
808 .error_remove_page
= NULL
,
811 /******************************************************************************
813 *****************************************************************************/
816 * Test whether an inode is a fast symlink.
818 static inline int exofs_inode_is_fast_symlink(struct inode
*inode
)
820 struct exofs_i_info
*oi
= exofs_i(inode
);
822 return S_ISLNK(inode
->i_mode
) && (oi
->i_data
[0] != 0);
825 const struct osd_attr g_attr_logical_length
= ATTR_DEF(
826 OSD_APAGE_OBJECT_INFORMATION
, OSD_ATTR_OI_LOGICAL_LENGTH
, 8);
828 static int _do_truncate(struct inode
*inode
, loff_t newsize
)
830 struct exofs_i_info
*oi
= exofs_i(inode
);
833 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
835 ret
= exofs_oi_truncate(oi
, (u64
)newsize
);
837 truncate_setsize(inode
, newsize
);
839 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
840 inode
->i_ino
, newsize
, ret
);
845 * Set inode attributes - update size attribute on OSD if needed,
846 * otherwise just call generic functions.
848 int exofs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
850 struct inode
*inode
= dentry
->d_inode
;
853 /* if we are about to modify an object, and it hasn't been
856 error
= wait_obj_created(exofs_i(inode
));
860 error
= inode_change_ok(inode
, iattr
);
864 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
865 iattr
->ia_size
!= i_size_read(inode
)) {
866 error
= _do_truncate(inode
, iattr
->ia_size
);
871 setattr_copy(inode
, iattr
);
872 mark_inode_dirty(inode
);
876 static const struct osd_attr g_attr_inode_file_layout
= ATTR_DEF(
878 EXOFS_ATTR_INODE_FILE_LAYOUT
,
880 static const struct osd_attr g_attr_inode_dir_layout
= ATTR_DEF(
882 EXOFS_ATTR_INODE_DIR_LAYOUT
,
886 * Read the Linux inode info from the OSD, and return it as is. In exofs the
887 * inode info is in an application specific page/attribute of the osd-object.
889 static int exofs_get_inode(struct super_block
*sb
, struct exofs_i_info
*oi
,
890 struct exofs_fcb
*inode
)
892 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
893 struct osd_attr attrs
[] = {
894 [0] = g_attr_inode_data
,
895 [1] = g_attr_inode_file_layout
,
896 [2] = g_attr_inode_dir_layout
,
898 struct exofs_io_state
*ios
;
899 struct exofs_on_disk_inode_layout
*layout
;
902 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
904 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__
);
908 ios
->obj
.id
= exofs_oi_objno(oi
);
909 exofs_make_credential(oi
->i_cred
, &ios
->obj
);
910 ios
->cred
= oi
->i_cred
;
912 attrs
[1].len
= exofs_on_disk_inode_layout_size(sbi
->layout
.s_numdevs
);
913 attrs
[2].len
= exofs_on_disk_inode_layout_size(sbi
->layout
.s_numdevs
);
915 ios
->in_attr
= attrs
;
916 ios
->in_attr_len
= ARRAY_SIZE(attrs
);
918 ret
= exofs_sbi_read(ios
);
920 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
921 _LLU(ios
->obj
.id
), ret
);
922 memset(inode
, 0, sizeof(*inode
));
923 inode
->i_mode
= 0040000 | (0777 & ~022);
924 /* If object is lost on target we might as well enable it's
927 if ((ret
== -ENOENT
) || (ret
== -EINVAL
))
932 ret
= extract_attr_from_ios(ios
, &attrs
[0]);
934 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
937 WARN_ON(attrs
[0].len
!= EXOFS_INO_ATTR_SIZE
);
938 memcpy(inode
, attrs
[0].val_ptr
, EXOFS_INO_ATTR_SIZE
);
940 ret
= extract_attr_from_ios(ios
, &attrs
[1]);
942 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
946 layout
= attrs
[1].val_ptr
;
947 if (layout
->gen_func
!= cpu_to_le16(LAYOUT_MOVING_WINDOW
)) {
948 EXOFS_ERR("%s: unsupported files layout %d\n",
949 __func__
, layout
->gen_func
);
955 ret
= extract_attr_from_ios(ios
, &attrs
[2]);
957 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
961 layout
= attrs
[2].val_ptr
;
962 if (layout
->gen_func
!= cpu_to_le16(LAYOUT_MOVING_WINDOW
)) {
963 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
964 __func__
, layout
->gen_func
);
971 exofs_put_io_state(ios
);
975 static void __oi_init(struct exofs_i_info
*oi
)
977 init_waitqueue_head(&oi
->i_wq
);
981 * Fill in an inode read from the OSD and set it up for use
983 struct inode
*exofs_iget(struct super_block
*sb
, unsigned long ino
)
985 struct exofs_i_info
*oi
;
986 struct exofs_fcb fcb
;
990 inode
= iget_locked(sb
, ino
);
992 return ERR_PTR(-ENOMEM
);
993 if (!(inode
->i_state
& I_NEW
))
998 /* read the inode from the osd */
999 ret
= exofs_get_inode(sb
, oi
, &fcb
);
1003 set_obj_created(oi
);
1005 /* copy stuff from on-disk struct to in-memory struct */
1006 inode
->i_mode
= le16_to_cpu(fcb
.i_mode
);
1007 inode
->i_uid
= le32_to_cpu(fcb
.i_uid
);
1008 inode
->i_gid
= le32_to_cpu(fcb
.i_gid
);
1009 inode
->i_nlink
= le16_to_cpu(fcb
.i_links_count
);
1010 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_ctime
);
1011 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_atime
);
1012 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_mtime
);
1013 inode
->i_ctime
.tv_nsec
=
1014 inode
->i_atime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
1015 oi
->i_commit_size
= le64_to_cpu(fcb
.i_size
);
1016 i_size_write(inode
, oi
->i_commit_size
);
1017 inode
->i_blkbits
= EXOFS_BLKSHIFT
;
1018 inode
->i_generation
= le32_to_cpu(fcb
.i_generation
);
1020 oi
->i_dir_start_lookup
= 0;
1022 if ((inode
->i_nlink
== 0) && (inode
->i_mode
== 0)) {
1027 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
1030 old_decode_dev(le32_to_cpu(fcb
.i_data
[0]));
1033 new_decode_dev(le32_to_cpu(fcb
.i_data
[1]));
1035 memcpy(oi
->i_data
, fcb
.i_data
, sizeof(fcb
.i_data
));
1038 if (S_ISREG(inode
->i_mode
)) {
1039 inode
->i_op
= &exofs_file_inode_operations
;
1040 inode
->i_fop
= &exofs_file_operations
;
1041 inode
->i_mapping
->a_ops
= &exofs_aops
;
1042 } else if (S_ISDIR(inode
->i_mode
)) {
1043 inode
->i_op
= &exofs_dir_inode_operations
;
1044 inode
->i_fop
= &exofs_dir_operations
;
1045 inode
->i_mapping
->a_ops
= &exofs_aops
;
1046 } else if (S_ISLNK(inode
->i_mode
)) {
1047 if (exofs_inode_is_fast_symlink(inode
))
1048 inode
->i_op
= &exofs_fast_symlink_inode_operations
;
1050 inode
->i_op
= &exofs_symlink_inode_operations
;
1051 inode
->i_mapping
->a_ops
= &exofs_aops
;
1054 inode
->i_op
= &exofs_special_inode_operations
;
1056 init_special_inode(inode
, inode
->i_mode
,
1057 old_decode_dev(le32_to_cpu(fcb
.i_data
[0])));
1059 init_special_inode(inode
, inode
->i_mode
,
1060 new_decode_dev(le32_to_cpu(fcb
.i_data
[1])));
1063 unlock_new_inode(inode
);
1068 return ERR_PTR(ret
);
1071 int __exofs_wait_obj_created(struct exofs_i_info
*oi
)
1073 if (!obj_created(oi
)) {
1074 BUG_ON(!obj_2bcreated(oi
));
1075 wait_event(oi
->i_wq
, obj_created(oi
));
1077 return unlikely(is_bad_inode(&oi
->vfs_inode
)) ? -EIO
: 0;
1080 * Callback function from exofs_new_inode(). The important thing is that we
1081 * set the obj_created flag so that other methods know that the object exists on
1084 static void create_done(struct exofs_io_state
*ios
, void *p
)
1086 struct inode
*inode
= p
;
1087 struct exofs_i_info
*oi
= exofs_i(inode
);
1088 struct exofs_sb_info
*sbi
= inode
->i_sb
->s_fs_info
;
1091 ret
= exofs_check_io(ios
, NULL
);
1092 exofs_put_io_state(ios
);
1094 atomic_dec(&sbi
->s_curr_pending
);
1096 if (unlikely(ret
)) {
1097 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1098 _LLU(exofs_oi_objno(oi
)), _LLU(sbi
->layout
.s_pid
));
1099 /*TODO: When FS is corrupted creation can fail, object already
1100 * exist. Get rid of this asynchronous creation, if exist
1101 * increment the obj counter and try the next object. Until we
1102 * succeed. All these dangling objects will be made into lost
1103 * files by chkfs.exofs
1107 set_obj_created(oi
);
1109 atomic_dec(&inode
->i_count
);
1114 * Set up a new inode and create an object for it on the OSD
1116 struct inode
*exofs_new_inode(struct inode
*dir
, int mode
)
1118 struct super_block
*sb
;
1119 struct inode
*inode
;
1120 struct exofs_i_info
*oi
;
1121 struct exofs_sb_info
*sbi
;
1122 struct exofs_io_state
*ios
;
1126 inode
= new_inode(sb
);
1128 return ERR_PTR(-ENOMEM
);
1130 oi
= exofs_i(inode
);
1133 set_obj_2bcreated(oi
);
1135 sbi
= sb
->s_fs_info
;
1138 inode_init_owner(inode
, dir
, mode
);
1139 inode
->i_ino
= sbi
->s_nextid
++;
1140 inode
->i_blkbits
= EXOFS_BLKSHIFT
;
1141 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
1142 oi
->i_commit_size
= inode
->i_size
= 0;
1143 spin_lock(&sbi
->s_next_gen_lock
);
1144 inode
->i_generation
= sbi
->s_next_generation
++;
1145 spin_unlock(&sbi
->s_next_gen_lock
);
1146 insert_inode_hash(inode
);
1148 mark_inode_dirty(inode
);
1150 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
1151 if (unlikely(ret
)) {
1152 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1153 return ERR_PTR(ret
);
1156 ios
->obj
.id
= exofs_oi_objno(oi
);
1157 exofs_make_credential(oi
->i_cred
, &ios
->obj
);
1159 /* increment the refcount so that the inode will still be around when we
1160 * reach the callback
1162 atomic_inc(&inode
->i_count
);
1164 ios
->done
= create_done
;
1165 ios
->private = inode
;
1166 ios
->cred
= oi
->i_cred
;
1167 ret
= exofs_sbi_create(ios
);
1169 atomic_dec(&inode
->i_count
);
1170 exofs_put_io_state(ios
);
1171 return ERR_PTR(ret
);
1173 atomic_inc(&sbi
->s_curr_pending
);
1179 * struct to pass two arguments to update_inode's callback
1181 struct updatei_args
{
1182 struct exofs_sb_info
*sbi
;
1183 struct exofs_fcb fcb
;
1187 * Callback function from exofs_update_inode().
1189 static void updatei_done(struct exofs_io_state
*ios
, void *p
)
1191 struct updatei_args
*args
= p
;
1193 exofs_put_io_state(ios
);
1195 atomic_dec(&args
->sbi
->s_curr_pending
);
1201 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1202 * synchronously or asynchronously depending on the do_sync flag.
1204 static int exofs_update_inode(struct inode
*inode
, int do_sync
)
1206 struct exofs_i_info
*oi
= exofs_i(inode
);
1207 struct super_block
*sb
= inode
->i_sb
;
1208 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1209 struct exofs_io_state
*ios
;
1210 struct osd_attr attr
;
1211 struct exofs_fcb
*fcb
;
1212 struct updatei_args
*args
;
1215 args
= kzalloc(sizeof(*args
), GFP_KERNEL
);
1217 EXOFS_DBGMSG("Faild kzalloc of args\n");
1223 fcb
->i_mode
= cpu_to_le16(inode
->i_mode
);
1224 fcb
->i_uid
= cpu_to_le32(inode
->i_uid
);
1225 fcb
->i_gid
= cpu_to_le32(inode
->i_gid
);
1226 fcb
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1227 fcb
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
1228 fcb
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
1229 fcb
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
1230 oi
->i_commit_size
= i_size_read(inode
);
1231 fcb
->i_size
= cpu_to_le64(oi
->i_commit_size
);
1232 fcb
->i_generation
= cpu_to_le32(inode
->i_generation
);
1234 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
1235 if (old_valid_dev(inode
->i_rdev
)) {
1237 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
1242 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
1246 memcpy(fcb
->i_data
, oi
->i_data
, sizeof(fcb
->i_data
));
1248 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
1249 if (unlikely(ret
)) {
1250 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__
);
1254 attr
= g_attr_inode_data
;
1256 ios
->out_attr_len
= 1;
1257 ios
->out_attr
= &attr
;
1259 if (!obj_created(oi
)) {
1260 EXOFS_DBGMSG("!obj_created\n");
1261 BUG_ON(!obj_2bcreated(oi
));
1262 wait_event(oi
->i_wq
, obj_created(oi
));
1263 EXOFS_DBGMSG("wait_event done\n");
1268 ios
->done
= updatei_done
;
1269 ios
->private = args
;
1272 ret
= exofs_oi_write(oi
, ios
);
1273 if (!do_sync
&& !ret
) {
1274 atomic_inc(&sbi
->s_curr_pending
);
1275 goto out
; /* deallocation in updatei_done */
1278 exofs_put_io_state(ios
);
1282 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1283 inode
->i_ino
, do_sync
, ret
);
1287 int exofs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1289 return exofs_update_inode(inode
, wbc
->sync_mode
== WB_SYNC_ALL
);
1293 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1296 static void delete_done(struct exofs_io_state
*ios
, void *p
)
1298 struct exofs_sb_info
*sbi
= p
;
1300 exofs_put_io_state(ios
);
1302 atomic_dec(&sbi
->s_curr_pending
);
1306 * Called when the refcount of an inode reaches zero. We remove the object
1307 * from the OSD here. We make sure the object was created before we try and
1310 void exofs_evict_inode(struct inode
*inode
)
1312 struct exofs_i_info
*oi
= exofs_i(inode
);
1313 struct super_block
*sb
= inode
->i_sb
;
1314 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1315 struct exofs_io_state
*ios
;
1318 truncate_inode_pages(&inode
->i_data
, 0);
1320 /* TODO: should do better here */
1321 if (inode
->i_nlink
|| is_bad_inode(inode
))
1325 end_writeback(inode
);
1327 /* if we are deleting an obj that hasn't been created yet, wait */
1328 if (!obj_created(oi
)) {
1329 BUG_ON(!obj_2bcreated(oi
));
1330 wait_event(oi
->i_wq
, obj_created(oi
));
1331 /* ignore the error attempt a remove anyway */
1334 /* Now Remove the OSD objects */
1335 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
1336 if (unlikely(ret
)) {
1337 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__
);
1341 ios
->obj
.id
= exofs_oi_objno(oi
);
1342 ios
->done
= delete_done
;
1344 ios
->cred
= oi
->i_cred
;
1345 ret
= exofs_sbi_remove(ios
);
1347 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__
);
1348 exofs_put_io_state(ios
);
1351 atomic_inc(&sbi
->s_curr_pending
);
1356 end_writeback(inode
);