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
)
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
);
224 if (!pcol
->read_4_write
)
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
);
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
)
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 ios
->pages
= pcol
->pages
;
271 ios
->nr_pages
= pcol
->nr_pages
;
272 ios
->length
= pcol
->length
;
273 ios
->offset
= pcol
->pg_first
<< PAGE_CACHE_SHIFT
;
275 if (pcol
->read_4_write
) {
276 exofs_oi_read(oi
, pcol
->ios
);
277 return __readpages_done(pcol
);
280 pcol_copy
= kmalloc(sizeof(*pcol_copy
), GFP_KERNEL
);
287 ios
->done
= readpages_done
;
288 ios
->private = pcol_copy
;
289 ret
= exofs_oi_read(oi
, ios
);
293 atomic_inc(&pcol
->sbi
->s_curr_pending
);
295 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
296 ios
->obj
.id
, _LLU(ios
->offset
), pcol
->length
);
298 /* pages ownership was passed to pcol_copy */
303 if (!pcol
->read_4_write
)
304 _unlock_pcol_pages(pcol
, ret
, READ
);
312 /* readpage_strip is called either directly from readpage() or by the VFS from
313 * within read_cache_pages(), to add one more page to be read. It will try to
314 * collect as many contiguous pages as posible. If a discontinuity is
315 * encountered, or it runs out of resources, it will submit the previous segment
316 * and will start a new collection. Eventually caller must submit the last
317 * segment if present.
319 static int readpage_strip(void *data
, struct page
*page
)
321 struct page_collect
*pcol
= data
;
322 struct inode
*inode
= pcol
->inode
;
323 struct exofs_i_info
*oi
= exofs_i(inode
);
324 loff_t i_size
= i_size_read(inode
);
325 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
329 /* FIXME: Just for debugging, will be removed */
330 if (PageUptodate(page
))
331 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol
->inode
->i_ino
,
334 if (page
->index
< end_index
)
335 len
= PAGE_CACHE_SIZE
;
336 else if (page
->index
== end_index
)
337 len
= i_size
& ~PAGE_CACHE_MASK
;
341 if (!len
|| !obj_created(oi
)) {
342 /* this will be out of bounds, or doesn't exist yet.
343 * Current page is cleared and the request is split
345 clear_highpage(page
);
347 SetPageUptodate(page
);
349 ClearPageError(page
);
351 if (!pcol
->read_4_write
)
353 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
354 "read_4_write=%d index=0x%lx end_index=0x%lx "
355 "splitting\n", inode
->i_ino
, len
,
356 pcol
->read_4_write
, page
->index
, end_index
);
358 return read_exec(pcol
);
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
);
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
);
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
);
425 static int _readpage(struct page
*page
, bool read_4_write
)
427 struct page_collect pcol
;
430 _pcol_init(&pcol
, 1, page
->mapping
->host
);
432 pcol
.read_4_write
= read_4_write
;
433 ret
= readpage_strip(&pcol
, page
);
435 EXOFS_ERR("_readpage => %d\n", ret
);
439 return read_exec(&pcol
);
443 * We don't need the file
445 static int exofs_readpage(struct file
*file
, struct page
*page
)
447 return _readpage(page
, false);
450 /* Callback for osd_write. All writes are asynchronous */
451 static void writepages_done(struct exofs_io_state
*ios
, void *p
)
453 struct page_collect
*pcol
= p
;
458 int ret
= exofs_check_io(ios
, &resid
);
460 atomic_dec(&pcol
->sbi
->s_curr_pending
);
463 good_bytes
= pcol
->length
;
465 good_bytes
= pcol
->length
- resid
;
467 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
468 " length=0x%lx nr_pages=%u\n",
469 pcol
->inode
->i_ino
, _LLU(good_bytes
), pcol
->length
,
472 for (i
= 0; i
< pcol
->nr_pages
; i
++) {
473 struct page
*page
= pcol
->pages
[i
];
474 struct inode
*inode
= page
->mapping
->host
;
477 if (inode
!= pcol
->inode
)
478 continue; /* osd might add more pages to a bio */
480 if (likely(length
< good_bytes
))
485 update_write_page(page
, page_stat
);
487 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
488 inode
->i_ino
, page
->index
, page_stat
);
495 EXOFS_DBGMSG2("writepages_done END\n");
498 static int write_exec(struct page_collect
*pcol
)
500 struct exofs_i_info
*oi
= exofs_i(pcol
->inode
);
501 struct exofs_io_state
*ios
= pcol
->ios
;
502 struct page_collect
*pcol_copy
= NULL
;
508 pcol_copy
= kmalloc(sizeof(*pcol_copy
), GFP_KERNEL
);
510 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
517 ios
->pages
= pcol_copy
->pages
;
518 ios
->nr_pages
= pcol_copy
->nr_pages
;
519 ios
->offset
= pcol_copy
->pg_first
<< PAGE_CACHE_SHIFT
;
520 ios
->length
= pcol_copy
->length
;
521 ios
->done
= writepages_done
;
522 ios
->private = pcol_copy
;
524 ret
= exofs_oi_write(oi
, ios
);
526 EXOFS_ERR("write_exec: exofs_oi_write() Failed\n");
530 atomic_inc(&pcol
->sbi
->s_curr_pending
);
531 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
532 pcol
->inode
->i_ino
, pcol
->pg_first
, _LLU(ios
->offset
),
534 /* pages ownership was passed to pcol_copy */
539 _unlock_pcol_pages(pcol
, ret
, WRITE
);
546 /* writepage_strip is called either directly from writepage() or by the VFS from
547 * within write_cache_pages(), to add one more page to be written to storage.
548 * It will try to collect as many contiguous pages as possible. If a
549 * discontinuity is encountered or it runs out of resources it will submit the
550 * previous segment and will start a new collection.
551 * Eventually caller must submit the last segment if present.
553 static int writepage_strip(struct page
*page
,
554 struct writeback_control
*wbc_unused
, void *data
)
556 struct page_collect
*pcol
= data
;
557 struct inode
*inode
= pcol
->inode
;
558 struct exofs_i_info
*oi
= exofs_i(inode
);
559 loff_t i_size
= i_size_read(inode
);
560 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
564 BUG_ON(!PageLocked(page
));
566 ret
= wait_obj_created(oi
);
570 if (page
->index
< end_index
)
571 /* in this case, the page is within the limits of the file */
572 len
= PAGE_CACHE_SIZE
;
574 len
= i_size
& ~PAGE_CACHE_MASK
;
576 if (page
->index
> end_index
|| !len
) {
577 /* in this case, the page is outside the limits
578 * (truncate in progress)
580 ret
= write_exec(pcol
);
584 ClearPageError(page
);
586 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
587 "outside the limits\n",
588 inode
->i_ino
, page
->index
);
595 if (unlikely(pcol
->pg_first
== -1)) {
596 pcol
->pg_first
= page
->index
;
597 } else if (unlikely((pcol
->pg_first
+ pcol
->nr_pages
) !=
599 /* Discontinuity detected, split the request */
600 ret
= write_exec(pcol
);
604 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
605 inode
->i_ino
, page
->index
);
610 ret
= pcol_try_alloc(pcol
);
615 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
616 inode
->i_ino
, page
->index
, len
);
618 ret
= pcol_add_page(pcol
, page
, len
);
620 EXOFS_DBGMSG2("Failed pcol_add_page "
621 "nr_pages=%u total_length=0x%lx\n",
622 pcol
->nr_pages
, pcol
->length
);
624 /* split the request, next loop will start again */
625 ret
= write_exec(pcol
);
627 EXOFS_DBGMSG("write_exec failed => %d", ret
);
634 BUG_ON(PageWriteback(page
));
635 set_page_writeback(page
);
640 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
641 inode
->i_ino
, page
->index
, ret
);
642 set_bit(AS_EIO
, &page
->mapping
->flags
);
647 static int exofs_writepages(struct address_space
*mapping
,
648 struct writeback_control
*wbc
)
650 struct page_collect pcol
;
651 long start
, end
, expected_pages
;
654 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
655 end
= (wbc
->range_end
== LLONG_MAX
) ?
656 start
+ mapping
->nrpages
:
657 wbc
->range_end
>> PAGE_CACHE_SHIFT
;
660 expected_pages
= end
- start
+ 1;
662 expected_pages
= mapping
->nrpages
;
664 if (expected_pages
< 32L)
665 expected_pages
= 32L;
667 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
668 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
669 mapping
->host
->i_ino
, wbc
->range_start
, wbc
->range_end
,
670 mapping
->nrpages
, start
, end
, expected_pages
);
672 _pcol_init(&pcol
, expected_pages
, mapping
->host
);
674 ret
= write_cache_pages(mapping
, wbc
, writepage_strip
, &pcol
);
676 EXOFS_ERR("write_cache_pages => %d\n", ret
);
680 return write_exec(&pcol
);
683 static int exofs_writepage(struct page
*page
, struct writeback_control
*wbc
)
685 struct page_collect pcol
;
688 _pcol_init(&pcol
, 1, page
->mapping
->host
);
690 ret
= writepage_strip(page
, NULL
, &pcol
);
692 EXOFS_ERR("exofs_writepage => %d\n", ret
);
696 return write_exec(&pcol
);
699 /* i_mutex held using inode->i_size directly */
700 static void _write_failed(struct inode
*inode
, loff_t to
)
702 if (to
> inode
->i_size
)
703 truncate_pagecache(inode
, to
, inode
->i_size
);
706 int exofs_write_begin(struct file
*file
, struct address_space
*mapping
,
707 loff_t pos
, unsigned len
, unsigned flags
,
708 struct page
**pagep
, void **fsdata
)
715 ret
= simple_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
718 EXOFS_DBGMSG("simple_write_begin failed\n");
725 /* read modify write */
726 if (!PageUptodate(page
) && (len
!= PAGE_CACHE_SIZE
)) {
727 loff_t i_size
= i_size_read(mapping
->host
);
728 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
731 if (page
->index
< end_index
)
732 rlen
= PAGE_CACHE_SIZE
;
733 else if (page
->index
== end_index
)
734 rlen
= i_size
& ~PAGE_CACHE_MASK
;
739 clear_highpage(page
);
740 SetPageUptodate(page
);
744 ret
= _readpage(page
, true);
746 /*SetPageError was done by _readpage. Is it ok?*/
748 EXOFS_DBGMSG("__readpage failed\n");
753 _write_failed(mapping
->host
, pos
+ len
);
758 static int exofs_write_begin_export(struct file
*file
,
759 struct address_space
*mapping
,
760 loff_t pos
, unsigned len
, unsigned flags
,
761 struct page
**pagep
, void **fsdata
)
765 return exofs_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
769 static int exofs_write_end(struct file
*file
, struct address_space
*mapping
,
770 loff_t pos
, unsigned len
, unsigned copied
,
771 struct page
*page
, void *fsdata
)
773 struct inode
*inode
= mapping
->host
;
774 /* According to comment in simple_write_end i_mutex is held */
775 loff_t i_size
= inode
->i_size
;
778 ret
= simple_write_end(file
, mapping
,pos
, len
, copied
, page
, fsdata
);
780 _write_failed(inode
, pos
+ len
);
782 /* TODO: once simple_write_end marks inode dirty remove */
783 if (i_size
!= inode
->i_size
)
784 mark_inode_dirty(inode
);
788 static int exofs_releasepage(struct page
*page
, gfp_t gfp
)
790 EXOFS_DBGMSG("page 0x%lx\n", page
->index
);
795 static void exofs_invalidatepage(struct page
*page
, unsigned long offset
)
797 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page
->index
, offset
);
801 const struct address_space_operations exofs_aops
= {
802 .readpage
= exofs_readpage
,
803 .readpages
= exofs_readpages
,
804 .writepage
= exofs_writepage
,
805 .writepages
= exofs_writepages
,
806 .write_begin
= exofs_write_begin_export
,
807 .write_end
= exofs_write_end
,
808 .releasepage
= exofs_releasepage
,
809 .set_page_dirty
= __set_page_dirty_nobuffers
,
810 .invalidatepage
= exofs_invalidatepage
,
812 /* Not implemented Yet */
813 .bmap
= NULL
, /* TODO: use osd's OSD_ACT_READ_MAP */
814 .direct_IO
= NULL
, /* TODO: Should be trivial to do */
816 /* With these NULL has special meaning or default is not exported */
820 .launder_page
= NULL
,
821 .is_partially_uptodate
= NULL
,
822 .error_remove_page
= NULL
,
825 /******************************************************************************
827 *****************************************************************************/
830 * Test whether an inode is a fast symlink.
832 static inline int exofs_inode_is_fast_symlink(struct inode
*inode
)
834 struct exofs_i_info
*oi
= exofs_i(inode
);
836 return S_ISLNK(inode
->i_mode
) && (oi
->i_data
[0] != 0);
839 const struct osd_attr g_attr_logical_length
= ATTR_DEF(
840 OSD_APAGE_OBJECT_INFORMATION
, OSD_ATTR_OI_LOGICAL_LENGTH
, 8);
842 static int _do_truncate(struct inode
*inode
, loff_t newsize
)
844 struct exofs_i_info
*oi
= exofs_i(inode
);
847 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
849 ret
= exofs_oi_truncate(oi
, (u64
)newsize
);
851 truncate_setsize(inode
, newsize
);
853 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
854 inode
->i_ino
, newsize
, ret
);
859 * Set inode attributes - update size attribute on OSD if needed,
860 * otherwise just call generic functions.
862 int exofs_setattr(struct dentry
*dentry
, struct iattr
*iattr
)
864 struct inode
*inode
= dentry
->d_inode
;
867 /* if we are about to modify an object, and it hasn't been
870 error
= wait_obj_created(exofs_i(inode
));
874 error
= inode_change_ok(inode
, iattr
);
878 if ((iattr
->ia_valid
& ATTR_SIZE
) &&
879 iattr
->ia_size
!= i_size_read(inode
)) {
880 error
= _do_truncate(inode
, iattr
->ia_size
);
885 setattr_copy(inode
, iattr
);
886 mark_inode_dirty(inode
);
890 static const struct osd_attr g_attr_inode_file_layout
= ATTR_DEF(
892 EXOFS_ATTR_INODE_FILE_LAYOUT
,
894 static const struct osd_attr g_attr_inode_dir_layout
= ATTR_DEF(
896 EXOFS_ATTR_INODE_DIR_LAYOUT
,
900 * Read the Linux inode info from the OSD, and return it as is. In exofs the
901 * inode info is in an application specific page/attribute of the osd-object.
903 static int exofs_get_inode(struct super_block
*sb
, struct exofs_i_info
*oi
,
904 struct exofs_fcb
*inode
)
906 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
907 struct osd_attr attrs
[] = {
908 [0] = g_attr_inode_data
,
909 [1] = g_attr_inode_file_layout
,
910 [2] = g_attr_inode_dir_layout
,
912 struct exofs_io_state
*ios
;
913 struct exofs_on_disk_inode_layout
*layout
;
916 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
918 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__
);
922 ios
->obj
.id
= exofs_oi_objno(oi
);
923 exofs_make_credential(oi
->i_cred
, &ios
->obj
);
924 ios
->cred
= oi
->i_cred
;
926 attrs
[1].len
= exofs_on_disk_inode_layout_size(sbi
->layout
.s_numdevs
);
927 attrs
[2].len
= exofs_on_disk_inode_layout_size(sbi
->layout
.s_numdevs
);
929 ios
->in_attr
= attrs
;
930 ios
->in_attr_len
= ARRAY_SIZE(attrs
);
932 ret
= exofs_sbi_read(ios
);
934 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
935 _LLU(ios
->obj
.id
), ret
);
936 memset(inode
, 0, sizeof(*inode
));
937 inode
->i_mode
= 0040000 | (0777 & ~022);
938 /* If object is lost on target we might as well enable it's
941 if ((ret
== -ENOENT
) || (ret
== -EINVAL
))
946 ret
= extract_attr_from_ios(ios
, &attrs
[0]);
948 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
951 WARN_ON(attrs
[0].len
!= EXOFS_INO_ATTR_SIZE
);
952 memcpy(inode
, attrs
[0].val_ptr
, EXOFS_INO_ATTR_SIZE
);
954 ret
= extract_attr_from_ios(ios
, &attrs
[1]);
956 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
960 layout
= attrs
[1].val_ptr
;
961 if (layout
->gen_func
!= cpu_to_le16(LAYOUT_MOVING_WINDOW
)) {
962 EXOFS_ERR("%s: unsupported files layout %d\n",
963 __func__
, layout
->gen_func
);
969 ret
= extract_attr_from_ios(ios
, &attrs
[2]);
971 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__
);
975 layout
= attrs
[2].val_ptr
;
976 if (layout
->gen_func
!= cpu_to_le16(LAYOUT_MOVING_WINDOW
)) {
977 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
978 __func__
, layout
->gen_func
);
985 exofs_put_io_state(ios
);
989 static void __oi_init(struct exofs_i_info
*oi
)
991 init_waitqueue_head(&oi
->i_wq
);
995 * Fill in an inode read from the OSD and set it up for use
997 struct inode
*exofs_iget(struct super_block
*sb
, unsigned long ino
)
999 struct exofs_i_info
*oi
;
1000 struct exofs_fcb fcb
;
1001 struct inode
*inode
;
1004 inode
= iget_locked(sb
, ino
);
1006 return ERR_PTR(-ENOMEM
);
1007 if (!(inode
->i_state
& I_NEW
))
1009 oi
= exofs_i(inode
);
1012 /* read the inode from the osd */
1013 ret
= exofs_get_inode(sb
, oi
, &fcb
);
1017 set_obj_created(oi
);
1019 /* copy stuff from on-disk struct to in-memory struct */
1020 inode
->i_mode
= le16_to_cpu(fcb
.i_mode
);
1021 inode
->i_uid
= le32_to_cpu(fcb
.i_uid
);
1022 inode
->i_gid
= le32_to_cpu(fcb
.i_gid
);
1023 inode
->i_nlink
= le16_to_cpu(fcb
.i_links_count
);
1024 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_ctime
);
1025 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_atime
);
1026 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(fcb
.i_mtime
);
1027 inode
->i_ctime
.tv_nsec
=
1028 inode
->i_atime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
1029 oi
->i_commit_size
= le64_to_cpu(fcb
.i_size
);
1030 i_size_write(inode
, oi
->i_commit_size
);
1031 inode
->i_blkbits
= EXOFS_BLKSHIFT
;
1032 inode
->i_generation
= le32_to_cpu(fcb
.i_generation
);
1034 oi
->i_dir_start_lookup
= 0;
1036 if ((inode
->i_nlink
== 0) && (inode
->i_mode
== 0)) {
1041 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
1044 old_decode_dev(le32_to_cpu(fcb
.i_data
[0]));
1047 new_decode_dev(le32_to_cpu(fcb
.i_data
[1]));
1049 memcpy(oi
->i_data
, fcb
.i_data
, sizeof(fcb
.i_data
));
1052 if (S_ISREG(inode
->i_mode
)) {
1053 inode
->i_op
= &exofs_file_inode_operations
;
1054 inode
->i_fop
= &exofs_file_operations
;
1055 inode
->i_mapping
->a_ops
= &exofs_aops
;
1056 } else if (S_ISDIR(inode
->i_mode
)) {
1057 inode
->i_op
= &exofs_dir_inode_operations
;
1058 inode
->i_fop
= &exofs_dir_operations
;
1059 inode
->i_mapping
->a_ops
= &exofs_aops
;
1060 } else if (S_ISLNK(inode
->i_mode
)) {
1061 if (exofs_inode_is_fast_symlink(inode
))
1062 inode
->i_op
= &exofs_fast_symlink_inode_operations
;
1064 inode
->i_op
= &exofs_symlink_inode_operations
;
1065 inode
->i_mapping
->a_ops
= &exofs_aops
;
1068 inode
->i_op
= &exofs_special_inode_operations
;
1070 init_special_inode(inode
, inode
->i_mode
,
1071 old_decode_dev(le32_to_cpu(fcb
.i_data
[0])));
1073 init_special_inode(inode
, inode
->i_mode
,
1074 new_decode_dev(le32_to_cpu(fcb
.i_data
[1])));
1077 unlock_new_inode(inode
);
1082 return ERR_PTR(ret
);
1085 int __exofs_wait_obj_created(struct exofs_i_info
*oi
)
1087 if (!obj_created(oi
)) {
1088 EXOFS_DBGMSG("!obj_created\n");
1089 BUG_ON(!obj_2bcreated(oi
));
1090 wait_event(oi
->i_wq
, obj_created(oi
));
1091 EXOFS_DBGMSG("wait_event done\n");
1093 return unlikely(is_bad_inode(&oi
->vfs_inode
)) ? -EIO
: 0;
1096 * Callback function from exofs_new_inode(). The important thing is that we
1097 * set the obj_created flag so that other methods know that the object exists on
1100 static void create_done(struct exofs_io_state
*ios
, void *p
)
1102 struct inode
*inode
= p
;
1103 struct exofs_i_info
*oi
= exofs_i(inode
);
1104 struct exofs_sb_info
*sbi
= inode
->i_sb
->s_fs_info
;
1107 ret
= exofs_check_io(ios
, NULL
);
1108 exofs_put_io_state(ios
);
1110 atomic_dec(&sbi
->s_curr_pending
);
1112 if (unlikely(ret
)) {
1113 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1114 _LLU(exofs_oi_objno(oi
)), _LLU(sbi
->layout
.s_pid
));
1115 /*TODO: When FS is corrupted creation can fail, object already
1116 * exist. Get rid of this asynchronous creation, if exist
1117 * increment the obj counter and try the next object. Until we
1118 * succeed. All these dangling objects will be made into lost
1119 * files by chkfs.exofs
1123 set_obj_created(oi
);
1129 * Set up a new inode and create an object for it on the OSD
1131 struct inode
*exofs_new_inode(struct inode
*dir
, int mode
)
1133 struct super_block
*sb
;
1134 struct inode
*inode
;
1135 struct exofs_i_info
*oi
;
1136 struct exofs_sb_info
*sbi
;
1137 struct exofs_io_state
*ios
;
1141 inode
= new_inode(sb
);
1143 return ERR_PTR(-ENOMEM
);
1145 oi
= exofs_i(inode
);
1148 set_obj_2bcreated(oi
);
1150 sbi
= sb
->s_fs_info
;
1153 inode_init_owner(inode
, dir
, mode
);
1154 inode
->i_ino
= sbi
->s_nextid
++;
1155 inode
->i_blkbits
= EXOFS_BLKSHIFT
;
1156 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
1157 oi
->i_commit_size
= inode
->i_size
= 0;
1158 spin_lock(&sbi
->s_next_gen_lock
);
1159 inode
->i_generation
= sbi
->s_next_generation
++;
1160 spin_unlock(&sbi
->s_next_gen_lock
);
1161 insert_inode_hash(inode
);
1163 mark_inode_dirty(inode
);
1165 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
1166 if (unlikely(ret
)) {
1167 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1168 return ERR_PTR(ret
);
1171 ios
->obj
.id
= exofs_oi_objno(oi
);
1172 exofs_make_credential(oi
->i_cred
, &ios
->obj
);
1174 ios
->done
= create_done
;
1175 ios
->private = inode
;
1176 ios
->cred
= oi
->i_cred
;
1177 ret
= exofs_sbi_create(ios
);
1179 exofs_put_io_state(ios
);
1180 return ERR_PTR(ret
);
1182 atomic_inc(&sbi
->s_curr_pending
);
1188 * struct to pass two arguments to update_inode's callback
1190 struct updatei_args
{
1191 struct exofs_sb_info
*sbi
;
1192 struct exofs_fcb fcb
;
1196 * Callback function from exofs_update_inode().
1198 static void updatei_done(struct exofs_io_state
*ios
, void *p
)
1200 struct updatei_args
*args
= p
;
1202 exofs_put_io_state(ios
);
1204 atomic_dec(&args
->sbi
->s_curr_pending
);
1210 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1211 * synchronously or asynchronously depending on the do_sync flag.
1213 static int exofs_update_inode(struct inode
*inode
, int do_sync
)
1215 struct exofs_i_info
*oi
= exofs_i(inode
);
1216 struct super_block
*sb
= inode
->i_sb
;
1217 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1218 struct exofs_io_state
*ios
;
1219 struct osd_attr attr
;
1220 struct exofs_fcb
*fcb
;
1221 struct updatei_args
*args
;
1224 args
= kzalloc(sizeof(*args
), GFP_KERNEL
);
1226 EXOFS_DBGMSG("Failed kzalloc of args\n");
1232 fcb
->i_mode
= cpu_to_le16(inode
->i_mode
);
1233 fcb
->i_uid
= cpu_to_le32(inode
->i_uid
);
1234 fcb
->i_gid
= cpu_to_le32(inode
->i_gid
);
1235 fcb
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
1236 fcb
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
1237 fcb
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
1238 fcb
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
1239 oi
->i_commit_size
= i_size_read(inode
);
1240 fcb
->i_size
= cpu_to_le64(oi
->i_commit_size
);
1241 fcb
->i_generation
= cpu_to_le32(inode
->i_generation
);
1243 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
1244 if (old_valid_dev(inode
->i_rdev
)) {
1246 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
1251 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
1255 memcpy(fcb
->i_data
, oi
->i_data
, sizeof(fcb
->i_data
));
1257 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
1258 if (unlikely(ret
)) {
1259 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__
);
1263 attr
= g_attr_inode_data
;
1265 ios
->out_attr_len
= 1;
1266 ios
->out_attr
= &attr
;
1268 wait_obj_created(oi
);
1272 ios
->done
= updatei_done
;
1273 ios
->private = args
;
1276 ret
= exofs_oi_write(oi
, ios
);
1277 if (!do_sync
&& !ret
) {
1278 atomic_inc(&sbi
->s_curr_pending
);
1279 goto out
; /* deallocation in updatei_done */
1282 exofs_put_io_state(ios
);
1286 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1287 inode
->i_ino
, do_sync
, ret
);
1291 int exofs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1293 return exofs_update_inode(inode
, wbc
->sync_mode
== WB_SYNC_ALL
);
1297 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1300 static void delete_done(struct exofs_io_state
*ios
, void *p
)
1302 struct exofs_sb_info
*sbi
= p
;
1304 exofs_put_io_state(ios
);
1306 atomic_dec(&sbi
->s_curr_pending
);
1310 * Called when the refcount of an inode reaches zero. We remove the object
1311 * from the OSD here. We make sure the object was created before we try and
1314 void exofs_evict_inode(struct inode
*inode
)
1316 struct exofs_i_info
*oi
= exofs_i(inode
);
1317 struct super_block
*sb
= inode
->i_sb
;
1318 struct exofs_sb_info
*sbi
= sb
->s_fs_info
;
1319 struct exofs_io_state
*ios
;
1322 truncate_inode_pages(&inode
->i_data
, 0);
1324 /* TODO: should do better here */
1325 if (inode
->i_nlink
|| is_bad_inode(inode
))
1329 end_writeback(inode
);
1331 /* if we are deleting an obj that hasn't been created yet, wait.
1332 * This also makes sure that create_done cannot be called with an
1333 * already evicted inode.
1335 wait_obj_created(oi
);
1336 /* ignore the error, attempt a remove anyway */
1338 /* Now Remove the OSD objects */
1339 ret
= exofs_get_io_state(&sbi
->layout
, &ios
);
1340 if (unlikely(ret
)) {
1341 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__
);
1345 ios
->obj
.id
= exofs_oi_objno(oi
);
1346 ios
->done
= delete_done
;
1348 ios
->cred
= oi
->i_cred
;
1349 ret
= exofs_sbi_remove(ios
);
1351 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__
);
1352 exofs_put_io_state(ios
);
1355 atomic_inc(&sbi
->s_curr_pending
);
1360 end_writeback(inode
);