2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/pagemap.h>
15 #include <linux/uio.h>
16 #include <linux/blkdev.h>
18 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/falloc.h>
22 #include <linux/swap.h>
23 #include <linux/crc32.h>
24 #include <linux/writeback.h>
25 #include <asm/uaccess.h>
26 #include <linux/dlm.h>
27 #include <linux/dlm_plock.h>
28 #include <linux/delay.h>
45 * gfs2_llseek - seek to a location in a file
48 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
50 * SEEK_END requires the glock for the file because it references the
53 * Returns: The new offset, or errno
56 static loff_t
gfs2_llseek(struct file
*file
, loff_t offset
, int whence
)
58 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
59 struct gfs2_holder i_gh
;
63 case SEEK_END
: /* These reference inode->i_size */
66 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
69 error
= generic_file_llseek(file
, offset
, whence
);
70 gfs2_glock_dq_uninit(&i_gh
);
75 error
= generic_file_llseek(file
, offset
, whence
);
85 * gfs2_readdir - Iterator for a directory
86 * @file: The directory to read from
87 * @ctx: What to feed directory entries to
92 static int gfs2_readdir(struct file
*file
, struct dir_context
*ctx
)
94 struct inode
*dir
= file
->f_mapping
->host
;
95 struct gfs2_inode
*dip
= GFS2_I(dir
);
96 struct gfs2_holder d_gh
;
99 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, &d_gh
);
103 error
= gfs2_dir_read(dir
, ctx
, &file
->f_ra
);
105 gfs2_glock_dq_uninit(&d_gh
);
112 * @table: A table of 32 u32 flags
113 * @val: a 32 bit value to convert
115 * This function can be used to convert between fsflags values and
116 * GFS2's own flags values.
118 * Returns: the converted flags
120 static u32
fsflags_cvt(const u32
*table
, u32 val
)
132 static const u32 fsflags_to_gfs2
[32] = {
134 [4] = GFS2_DIF_IMMUTABLE
,
135 [5] = GFS2_DIF_APPENDONLY
,
136 [7] = GFS2_DIF_NOATIME
,
137 [12] = GFS2_DIF_EXHASH
,
138 [14] = GFS2_DIF_INHERIT_JDATA
,
139 [17] = GFS2_DIF_TOPDIR
,
142 static const u32 gfs2_to_fsflags
[32] = {
143 [gfs2fl_Sync
] = FS_SYNC_FL
,
144 [gfs2fl_Immutable
] = FS_IMMUTABLE_FL
,
145 [gfs2fl_AppendOnly
] = FS_APPEND_FL
,
146 [gfs2fl_NoAtime
] = FS_NOATIME_FL
,
147 [gfs2fl_ExHash
] = FS_INDEX_FL
,
148 [gfs2fl_TopLevel
] = FS_TOPDIR_FL
,
149 [gfs2fl_InheritJdata
] = FS_JOURNAL_DATA_FL
,
152 static int gfs2_get_flags(struct file
*filp
, u32 __user
*ptr
)
154 struct inode
*inode
= file_inode(filp
);
155 struct gfs2_inode
*ip
= GFS2_I(inode
);
156 struct gfs2_holder gh
;
160 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
161 error
= gfs2_glock_nq(&gh
);
165 fsflags
= fsflags_cvt(gfs2_to_fsflags
, ip
->i_diskflags
);
166 if (!S_ISDIR(inode
->i_mode
) && ip
->i_diskflags
& GFS2_DIF_JDATA
)
167 fsflags
|= FS_JOURNAL_DATA_FL
;
168 if (put_user(fsflags
, ptr
))
172 gfs2_holder_uninit(&gh
);
176 void gfs2_set_inode_flags(struct inode
*inode
)
178 struct gfs2_inode
*ip
= GFS2_I(inode
);
179 unsigned int flags
= inode
->i_flags
;
181 flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_NOSEC
);
182 if ((ip
->i_eattr
== 0) && !is_sxid(inode
->i_mode
))
184 if (ip
->i_diskflags
& GFS2_DIF_IMMUTABLE
)
185 flags
|= S_IMMUTABLE
;
186 if (ip
->i_diskflags
& GFS2_DIF_APPENDONLY
)
188 if (ip
->i_diskflags
& GFS2_DIF_NOATIME
)
190 if (ip
->i_diskflags
& GFS2_DIF_SYNC
)
192 inode
->i_flags
= flags
;
195 /* Flags that can be set by user space */
196 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
197 GFS2_DIF_IMMUTABLE| \
198 GFS2_DIF_APPENDONLY| \
203 GFS2_DIF_INHERIT_JDATA)
206 * do_gfs2_set_flags - set flags on an inode
207 * @filp: file pointer
208 * @reqflags: The flags to set
209 * @mask: Indicates which flags are valid
212 static int do_gfs2_set_flags(struct file
*filp
, u32 reqflags
, u32 mask
)
214 struct inode
*inode
= file_inode(filp
);
215 struct gfs2_inode
*ip
= GFS2_I(inode
);
216 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
217 struct buffer_head
*bh
;
218 struct gfs2_holder gh
;
220 u32 new_flags
, flags
;
222 error
= mnt_want_write_file(filp
);
226 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
231 if (!inode_owner_or_capable(inode
))
235 flags
= ip
->i_diskflags
;
236 new_flags
= (flags
& ~mask
) | (reqflags
& mask
);
237 if ((new_flags
^ flags
) == 0)
241 if ((new_flags
^ flags
) & ~GFS2_FLAGS_USER_SET
)
245 if (IS_IMMUTABLE(inode
) && (new_flags
& GFS2_DIF_IMMUTABLE
))
247 if (IS_APPEND(inode
) && (new_flags
& GFS2_DIF_APPENDONLY
))
249 if (((new_flags
^ flags
) & GFS2_DIF_IMMUTABLE
) &&
250 !capable(CAP_LINUX_IMMUTABLE
))
252 if (!IS_IMMUTABLE(inode
)) {
253 error
= gfs2_permission(inode
, MAY_WRITE
);
257 if ((flags
^ new_flags
) & GFS2_DIF_JDATA
) {
258 if (new_flags
& GFS2_DIF_JDATA
)
259 gfs2_log_flush(sdp
, ip
->i_gl
, NORMAL_FLUSH
);
260 error
= filemap_fdatawrite(inode
->i_mapping
);
263 error
= filemap_fdatawait(inode
->i_mapping
);
266 if (new_flags
& GFS2_DIF_JDATA
)
267 gfs2_ordered_del_inode(ip
);
269 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
272 error
= gfs2_meta_inode_buffer(ip
, &bh
);
275 gfs2_trans_add_meta(ip
->i_gl
, bh
);
276 ip
->i_diskflags
= new_flags
;
277 gfs2_dinode_out(ip
, bh
->b_data
);
279 gfs2_set_inode_flags(inode
);
280 gfs2_set_aops(inode
);
284 gfs2_glock_dq_uninit(&gh
);
286 mnt_drop_write_file(filp
);
290 static int gfs2_set_flags(struct file
*filp
, u32 __user
*ptr
)
292 struct inode
*inode
= file_inode(filp
);
293 u32 fsflags
, gfsflags
;
295 if (get_user(fsflags
, ptr
))
298 gfsflags
= fsflags_cvt(fsflags_to_gfs2
, fsflags
);
299 if (!S_ISDIR(inode
->i_mode
)) {
300 gfsflags
&= ~GFS2_DIF_TOPDIR
;
301 if (gfsflags
& GFS2_DIF_INHERIT_JDATA
)
302 gfsflags
^= (GFS2_DIF_JDATA
| GFS2_DIF_INHERIT_JDATA
);
303 return do_gfs2_set_flags(filp
, gfsflags
, ~0);
305 return do_gfs2_set_flags(filp
, gfsflags
, ~GFS2_DIF_JDATA
);
308 static long gfs2_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
311 case FS_IOC_GETFLAGS
:
312 return gfs2_get_flags(filp
, (u32 __user
*)arg
);
313 case FS_IOC_SETFLAGS
:
314 return gfs2_set_flags(filp
, (u32 __user
*)arg
);
316 return gfs2_fitrim(filp
, (void __user
*)arg
);
322 * gfs2_size_hint - Give a hint to the size of a write request
323 * @filep: The struct file
324 * @offset: The file offset of the write
325 * @size: The length of the write
327 * When we are about to do a write, this function records the total
328 * write size in order to provide a suitable hint to the lower layers
329 * about how many blocks will be required.
333 static void gfs2_size_hint(struct file
*filep
, loff_t offset
, size_t size
)
335 struct inode
*inode
= file_inode(filep
);
336 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
337 struct gfs2_inode
*ip
= GFS2_I(inode
);
338 size_t blks
= (size
+ sdp
->sd_sb
.sb_bsize
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
339 int hint
= min_t(size_t, INT_MAX
, blks
);
341 if (hint
> atomic_read(&ip
->i_res
->rs_sizehint
))
342 atomic_set(&ip
->i_res
->rs_sizehint
, hint
);
346 * gfs2_allocate_page_backing - Use bmap to allocate blocks
347 * @page: The (locked) page to allocate backing for
349 * We try to allocate all the blocks required for the page in
350 * one go. This might fail for various reasons, so we keep
351 * trying until all the blocks to back this page are allocated.
352 * If some of the blocks are already allocated, thats ok too.
355 static int gfs2_allocate_page_backing(struct page
*page
)
357 struct inode
*inode
= page
->mapping
->host
;
358 struct buffer_head bh
;
359 unsigned long size
= PAGE_CACHE_SIZE
;
360 u64 lblock
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
365 gfs2_block_map(inode
, lblock
, &bh
, 1);
366 if (!buffer_mapped(&bh
))
369 lblock
+= (bh
.b_size
>> inode
->i_blkbits
);
375 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
376 * @vma: The virtual memory area
377 * @vmf: The virtual memory fault containing the page to become writable
379 * When the page becomes writable, we need to ensure that we have
380 * blocks allocated on disk to back that page.
383 static int gfs2_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
385 struct page
*page
= vmf
->page
;
386 struct inode
*inode
= file_inode(vma
->vm_file
);
387 struct gfs2_inode
*ip
= GFS2_I(inode
);
388 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
389 struct gfs2_alloc_parms ap
= { .aflags
= 0, };
390 unsigned long last_index
;
391 u64 pos
= page
->index
<< PAGE_CACHE_SHIFT
;
392 unsigned int data_blocks
, ind_blocks
, rblocks
;
393 struct gfs2_holder gh
;
397 sb_start_pagefault(inode
->i_sb
);
399 /* Update file times before taking page lock */
400 file_update_time(vma
->vm_file
);
402 ret
= get_write_access(inode
);
406 ret
= gfs2_rs_alloc(ip
);
408 goto out_write_access
;
410 gfs2_size_hint(vma
->vm_file
, pos
, PAGE_CACHE_SIZE
);
412 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
413 ret
= gfs2_glock_nq(&gh
);
417 set_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
);
418 set_bit(GIF_SW_PAGED
, &ip
->i_flags
);
420 if (!gfs2_write_alloc_required(ip
, pos
, PAGE_CACHE_SIZE
)) {
422 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
) {
429 ret
= gfs2_rindex_update(sdp
);
433 gfs2_write_calc_reserv(ip
, PAGE_CACHE_SIZE
, &data_blocks
, &ind_blocks
);
434 ap
.target
= data_blocks
+ ind_blocks
;
435 ret
= gfs2_quota_lock_check(ip
, &ap
);
438 ret
= gfs2_inplace_reserve(ip
, &ap
);
440 goto out_quota_unlock
;
442 rblocks
= RES_DINODE
+ ind_blocks
;
443 if (gfs2_is_jdata(ip
))
444 rblocks
+= data_blocks
? data_blocks
: 1;
445 if (ind_blocks
|| data_blocks
) {
446 rblocks
+= RES_STATFS
+ RES_QUOTA
;
447 rblocks
+= gfs2_rg_blocks(ip
, data_blocks
+ ind_blocks
);
449 ret
= gfs2_trans_begin(sdp
, rblocks
, 0);
455 size
= i_size_read(inode
);
456 last_index
= (size
- 1) >> PAGE_CACHE_SHIFT
;
457 /* Check page index against inode size */
458 if (size
== 0 || (page
->index
> last_index
))
462 /* If truncated, we must retry the operation, we may have raced
463 * with the glock demotion code.
465 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
)
468 /* Unstuff, if required, and allocate backing blocks for page */
470 if (gfs2_is_stuffed(ip
))
471 ret
= gfs2_unstuff_dinode(ip
, page
);
473 ret
= gfs2_allocate_page_backing(page
);
480 gfs2_inplace_release(ip
);
482 gfs2_quota_unlock(ip
);
486 gfs2_holder_uninit(&gh
);
488 set_page_dirty(page
);
489 wait_for_stable_page(page
);
492 put_write_access(inode
);
494 sb_end_pagefault(inode
->i_sb
);
495 return block_page_mkwrite_return(ret
);
498 static const struct vm_operations_struct gfs2_vm_ops
= {
499 .fault
= filemap_fault
,
500 .map_pages
= filemap_map_pages
,
501 .page_mkwrite
= gfs2_page_mkwrite
,
506 * @file: The file to map
507 * @vma: The VMA which described the mapping
509 * There is no need to get a lock here unless we should be updating
510 * atime. We ignore any locking errors since the only consequence is
511 * a missed atime update (which will just be deferred until later).
516 static int gfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
518 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
520 if (!(file
->f_flags
& O_NOATIME
) &&
521 !IS_NOATIME(&ip
->i_inode
)) {
522 struct gfs2_holder i_gh
;
525 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
529 /* grab lock to update inode */
530 gfs2_glock_dq_uninit(&i_gh
);
533 vma
->vm_ops
= &gfs2_vm_ops
;
539 * gfs2_open_common - This is common to open and atomic_open
540 * @inode: The inode being opened
541 * @file: The file being opened
543 * This maybe called under a glock or not depending upon how it has
544 * been called. We must always be called under a glock for regular
545 * files, however. For other file types, it does not matter whether
546 * we hold the glock or not.
548 * Returns: Error code or 0 for success
551 int gfs2_open_common(struct inode
*inode
, struct file
*file
)
553 struct gfs2_file
*fp
;
556 if (S_ISREG(inode
->i_mode
)) {
557 ret
= generic_file_open(inode
, file
);
562 fp
= kzalloc(sizeof(struct gfs2_file
), GFP_NOFS
);
566 mutex_init(&fp
->f_fl_mutex
);
568 gfs2_assert_warn(GFS2_SB(inode
), !file
->private_data
);
569 file
->private_data
= fp
;
574 * gfs2_open - open a file
575 * @inode: the inode to open
576 * @file: the struct file for this opening
578 * After atomic_open, this function is only used for opening files
579 * which are already cached. We must still get the glock for regular
580 * files to ensure that we have the file size uptodate for the large
581 * file check which is in the common code. That is only an issue for
582 * regular files though.
587 static int gfs2_open(struct inode
*inode
, struct file
*file
)
589 struct gfs2_inode
*ip
= GFS2_I(inode
);
590 struct gfs2_holder i_gh
;
592 bool need_unlock
= false;
594 if (S_ISREG(ip
->i_inode
.i_mode
)) {
595 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
602 error
= gfs2_open_common(inode
, file
);
605 gfs2_glock_dq_uninit(&i_gh
);
611 * gfs2_release - called to close a struct file
612 * @inode: the inode the struct file belongs to
613 * @file: the struct file being closed
618 static int gfs2_release(struct inode
*inode
, struct file
*file
)
620 struct gfs2_inode
*ip
= GFS2_I(inode
);
622 kfree(file
->private_data
);
623 file
->private_data
= NULL
;
625 if (!(file
->f_mode
& FMODE_WRITE
))
628 gfs2_rs_delete(ip
, &inode
->i_writecount
);
633 * gfs2_fsync - sync the dirty data for a file (across the cluster)
634 * @file: the file that points to the dentry
635 * @start: the start position in the file to sync
636 * @end: the end position in the file to sync
637 * @datasync: set if we can ignore timestamp changes
639 * We split the data flushing here so that we don't wait for the data
640 * until after we've also sent the metadata to disk. Note that for
641 * data=ordered, we will write & wait for the data at the log flush
642 * stage anyway, so this is unlikely to make much of a difference
643 * except in the data=writeback case.
645 * If the fdatawrite fails due to any reason except -EIO, we will
646 * continue the remainder of the fsync, although we'll still report
647 * the error at the end. This is to match filemap_write_and_wait_range()
653 static int gfs2_fsync(struct file
*file
, loff_t start
, loff_t end
,
656 struct address_space
*mapping
= file
->f_mapping
;
657 struct inode
*inode
= mapping
->host
;
658 int sync_state
= inode
->i_state
& I_DIRTY_ALL
;
659 struct gfs2_inode
*ip
= GFS2_I(inode
);
660 int ret
= 0, ret1
= 0;
662 if (mapping
->nrpages
) {
663 ret1
= filemap_fdatawrite_range(mapping
, start
, end
);
668 if (!gfs2_is_jdata(ip
))
669 sync_state
&= ~I_DIRTY_PAGES
;
671 sync_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_TIME
);
674 ret
= sync_inode_metadata(inode
, 1);
677 if (gfs2_is_jdata(ip
))
678 filemap_write_and_wait(mapping
);
679 gfs2_ail_flush(ip
->i_gl
, 1);
682 if (mapping
->nrpages
)
683 ret
= filemap_fdatawait_range(mapping
, start
, end
);
685 return ret
? ret
: ret1
;
689 * gfs2_file_write_iter - Perform a write to a file
690 * @iocb: The io context
691 * @iov: The data to write
692 * @nr_segs: Number of @iov segments
693 * @pos: The file position
695 * We have to do a lock/unlock here to refresh the inode size for
696 * O_APPEND writes, otherwise we can land up writing at the wrong
697 * offset. There is still a race, but provided the app is using its
698 * own file locking, this will make O_APPEND work as expected.
702 static ssize_t
gfs2_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
704 struct file
*file
= iocb
->ki_filp
;
705 struct gfs2_inode
*ip
= GFS2_I(file_inode(file
));
708 ret
= gfs2_rs_alloc(ip
);
712 gfs2_size_hint(file
, iocb
->ki_pos
, iov_iter_count(from
));
714 if (iocb
->ki_flags
& IOCB_APPEND
) {
715 struct gfs2_holder gh
;
717 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
720 gfs2_glock_dq_uninit(&gh
);
723 return generic_file_write_iter(iocb
, from
);
726 static int fallocate_chunk(struct inode
*inode
, loff_t offset
, loff_t len
,
729 struct gfs2_inode
*ip
= GFS2_I(inode
);
730 struct buffer_head
*dibh
;
732 unsigned int nr_blks
;
733 sector_t lblock
= offset
>> inode
->i_blkbits
;
735 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
739 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
741 if (gfs2_is_stuffed(ip
)) {
742 error
= gfs2_unstuff_dinode(ip
, NULL
);
748 struct buffer_head bh_map
= { .b_state
= 0, .b_blocknr
= 0 };
750 set_buffer_zeronew(&bh_map
);
752 error
= gfs2_block_map(inode
, lblock
, &bh_map
, 1);
755 len
-= bh_map
.b_size
;
756 nr_blks
= bh_map
.b_size
>> inode
->i_blkbits
;
758 if (!buffer_new(&bh_map
))
760 if (unlikely(!buffer_zeronew(&bh_map
))) {
770 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
771 * blocks, determine how many bytes can be written.
772 * @ip: The inode in question.
773 * @len: Max cap of bytes. What we return in *len must be <= this.
774 * @data_blocks: Compute and return the number of data blocks needed
775 * @ind_blocks: Compute and return the number of indirect blocks needed
776 * @max_blocks: The total blocks available to work with.
778 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
780 static void calc_max_reserv(struct gfs2_inode
*ip
, loff_t
*len
,
781 unsigned int *data_blocks
, unsigned int *ind_blocks
,
782 unsigned int max_blocks
)
785 const struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
786 unsigned int tmp
, max_data
= max_blocks
- 3 * (sdp
->sd_max_height
- 1);
788 for (tmp
= max_data
; tmp
> sdp
->sd_diptrs
;) {
789 tmp
= DIV_ROUND_UP(tmp
, sdp
->sd_inptrs
);
793 *data_blocks
= max_data
;
794 *ind_blocks
= max_blocks
- max_data
;
795 *len
= ((loff_t
)max_data
- 3) << sdp
->sd_sb
.sb_bsize_shift
;
798 gfs2_write_calc_reserv(ip
, max
, data_blocks
, ind_blocks
);
802 static long __gfs2_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
804 struct inode
*inode
= file_inode(file
);
805 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
806 struct gfs2_inode
*ip
= GFS2_I(inode
);
807 struct gfs2_alloc_parms ap
= { .aflags
= 0, };
808 unsigned int data_blocks
= 0, ind_blocks
= 0, rblocks
;
809 loff_t bytes
, max_bytes
, max_blks
;
811 const loff_t pos
= offset
;
812 const loff_t count
= len
;
813 loff_t bsize_mask
= ~((loff_t
)sdp
->sd_sb
.sb_bsize
- 1);
814 loff_t next
= (offset
+ len
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
815 loff_t max_chunk_size
= UINT_MAX
& bsize_mask
;
817 next
= (next
+ 1) << sdp
->sd_sb
.sb_bsize_shift
;
819 offset
&= bsize_mask
;
822 bytes
= sdp
->sd_max_rg_data
* sdp
->sd_sb
.sb_bsize
/ 2;
827 bytes
= sdp
->sd_sb
.sb_bsize
;
829 gfs2_size_hint(file
, offset
, len
);
831 gfs2_write_calc_reserv(ip
, PAGE_SIZE
, &data_blocks
, &ind_blocks
);
832 ap
.min_target
= data_blocks
+ ind_blocks
;
837 if (!gfs2_write_alloc_required(ip
, offset
, bytes
)) {
843 /* We need to determine how many bytes we can actually
844 * fallocate without exceeding quota or going over the
845 * end of the fs. We start off optimistically by assuming
846 * we can write max_bytes */
847 max_bytes
= (len
> max_chunk_size
) ? max_chunk_size
: len
;
849 /* Since max_bytes is most likely a theoretical max, we
850 * calculate a more realistic 'bytes' to serve as a good
851 * starting point for the number of bytes we may be able
853 gfs2_write_calc_reserv(ip
, bytes
, &data_blocks
, &ind_blocks
);
854 ap
.target
= data_blocks
+ ind_blocks
;
856 error
= gfs2_quota_lock_check(ip
, &ap
);
859 /* ap.allowed tells us how many blocks quota will allow
860 * us to write. Check if this reduces max_blks */
863 max_blks
= ap
.allowed
;
865 error
= gfs2_inplace_reserve(ip
, &ap
);
869 /* check if the selected rgrp limits our max_blks further */
870 if (ap
.allowed
&& ap
.allowed
< max_blks
)
871 max_blks
= ap
.allowed
;
873 /* Almost done. Calculate bytes that can be written using
874 * max_blks. We also recompute max_bytes, data_blocks and
876 calc_max_reserv(ip
, &max_bytes
, &data_blocks
,
877 &ind_blocks
, max_blks
);
879 rblocks
= RES_DINODE
+ ind_blocks
+ RES_STATFS
+ RES_QUOTA
+
880 RES_RG_HDR
+ gfs2_rg_blocks(ip
, data_blocks
+ ind_blocks
);
881 if (gfs2_is_jdata(ip
))
882 rblocks
+= data_blocks
? data_blocks
: 1;
884 error
= gfs2_trans_begin(sdp
, rblocks
,
885 PAGE_CACHE_SIZE
/sdp
->sd_sb
.sb_bsize
);
889 error
= fallocate_chunk(inode
, offset
, max_bytes
, mode
);
897 gfs2_inplace_release(ip
);
898 gfs2_quota_unlock(ip
);
901 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && (pos
+ count
) > inode
->i_size
) {
902 i_size_write(inode
, pos
+ count
);
903 file_update_time(file
);
904 mark_inode_dirty(inode
);
907 return generic_write_sync(file
, pos
, count
);
910 gfs2_inplace_release(ip
);
912 gfs2_quota_unlock(ip
);
916 static long gfs2_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
918 struct inode
*inode
= file_inode(file
);
919 struct gfs2_inode
*ip
= GFS2_I(inode
);
920 struct gfs2_holder gh
;
923 if ((mode
& ~FALLOC_FL_KEEP_SIZE
) || gfs2_is_jdata(ip
))
926 mutex_lock(&inode
->i_mutex
);
928 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
929 ret
= gfs2_glock_nq(&gh
);
933 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
934 (offset
+ len
) > inode
->i_size
) {
935 ret
= inode_newsize_ok(inode
, offset
+ len
);
940 ret
= get_write_access(inode
);
944 ret
= gfs2_rs_alloc(ip
);
948 ret
= __gfs2_fallocate(file
, mode
, offset
, len
);
950 gfs2_rs_deltree(ip
->i_res
);
952 put_write_access(inode
);
956 gfs2_holder_uninit(&gh
);
957 mutex_unlock(&inode
->i_mutex
);
961 static ssize_t
gfs2_file_splice_write(struct pipe_inode_info
*pipe
,
962 struct file
*out
, loff_t
*ppos
,
963 size_t len
, unsigned int flags
)
966 struct gfs2_inode
*ip
= GFS2_I(out
->f_mapping
->host
);
968 error
= gfs2_rs_alloc(ip
);
970 return (ssize_t
)error
;
972 gfs2_size_hint(out
, *ppos
, len
);
974 return iter_file_splice_write(pipe
, out
, ppos
, len
, flags
);
977 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
980 * gfs2_lock - acquire/release a posix lock on a file
981 * @file: the file pointer
982 * @cmd: either modify or retrieve lock state, possibly wait
983 * @fl: type and range of lock
988 static int gfs2_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
990 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
991 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
992 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
994 if (!(fl
->fl_flags
& FL_POSIX
))
996 if (__mandatory_lock(&ip
->i_inode
) && fl
->fl_type
!= F_UNLCK
)
999 if (cmd
== F_CANCELLK
) {
1002 fl
->fl_type
= F_UNLCK
;
1004 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
1005 if (fl
->fl_type
== F_UNLCK
)
1006 locks_lock_file_wait(file
, fl
);
1010 return dlm_posix_get(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
1011 else if (fl
->fl_type
== F_UNLCK
)
1012 return dlm_posix_unlock(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
1014 return dlm_posix_lock(ls
->ls_dlm
, ip
->i_no_addr
, file
, cmd
, fl
);
1017 static int do_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1019 struct gfs2_file
*fp
= file
->private_data
;
1020 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1021 struct gfs2_inode
*ip
= GFS2_I(file_inode(file
));
1022 struct gfs2_glock
*gl
;
1028 state
= (fl
->fl_type
== F_WRLCK
) ? LM_ST_EXCLUSIVE
: LM_ST_SHARED
;
1029 flags
= (IS_SETLKW(cmd
) ? 0 : LM_FLAG_TRY_1CB
) | GL_EXACT
;
1031 mutex_lock(&fp
->f_fl_mutex
);
1035 if (fl_gh
->gh_state
== state
)
1037 locks_lock_file_wait(file
,
1038 &(struct file_lock
){.fl_type
= F_UNLCK
});
1039 gfs2_glock_dq(fl_gh
);
1040 gfs2_holder_reinit(state
, flags
, fl_gh
);
1042 error
= gfs2_glock_get(GFS2_SB(&ip
->i_inode
), ip
->i_no_addr
,
1043 &gfs2_flock_glops
, CREATE
, &gl
);
1046 gfs2_holder_init(gl
, state
, flags
, fl_gh
);
1049 for (sleeptime
= 1; sleeptime
<= 4; sleeptime
<<= 1) {
1050 error
= gfs2_glock_nq(fl_gh
);
1051 if (error
!= GLR_TRYFAILED
)
1053 fl_gh
->gh_flags
= LM_FLAG_TRY
| GL_EXACT
;
1054 fl_gh
->gh_error
= 0;
1058 gfs2_holder_uninit(fl_gh
);
1059 if (error
== GLR_TRYFAILED
)
1062 error
= locks_lock_file_wait(file
, fl
);
1063 gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), !error
);
1067 mutex_unlock(&fp
->f_fl_mutex
);
1071 static void do_unflock(struct file
*file
, struct file_lock
*fl
)
1073 struct gfs2_file
*fp
= file
->private_data
;
1074 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1076 mutex_lock(&fp
->f_fl_mutex
);
1077 locks_lock_file_wait(file
, fl
);
1079 gfs2_glock_dq(fl_gh
);
1080 gfs2_holder_uninit(fl_gh
);
1082 mutex_unlock(&fp
->f_fl_mutex
);
1086 * gfs2_flock - acquire/release a flock lock on a file
1087 * @file: the file pointer
1088 * @cmd: either modify or retrieve lock state, possibly wait
1089 * @fl: type and range of lock
1094 static int gfs2_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1096 if (!(fl
->fl_flags
& FL_FLOCK
))
1098 if (fl
->fl_type
& LOCK_MAND
)
1101 if (fl
->fl_type
== F_UNLCK
) {
1102 do_unflock(file
, fl
);
1105 return do_flock(file
, cmd
, fl
);
1109 const struct file_operations gfs2_file_fops
= {
1110 .llseek
= gfs2_llseek
,
1111 .read_iter
= generic_file_read_iter
,
1112 .write_iter
= gfs2_file_write_iter
,
1113 .unlocked_ioctl
= gfs2_ioctl
,
1116 .release
= gfs2_release
,
1117 .fsync
= gfs2_fsync
,
1119 .flock
= gfs2_flock
,
1120 .splice_read
= generic_file_splice_read
,
1121 .splice_write
= gfs2_file_splice_write
,
1122 .setlease
= simple_nosetlease
,
1123 .fallocate
= gfs2_fallocate
,
1126 const struct file_operations gfs2_dir_fops
= {
1127 .iterate
= gfs2_readdir
,
1128 .unlocked_ioctl
= gfs2_ioctl
,
1130 .release
= gfs2_release
,
1131 .fsync
= gfs2_fsync
,
1133 .flock
= gfs2_flock
,
1134 .llseek
= default_llseek
,
1137 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1139 const struct file_operations gfs2_file_fops_nolock
= {
1140 .llseek
= gfs2_llseek
,
1141 .read_iter
= generic_file_read_iter
,
1142 .write_iter
= gfs2_file_write_iter
,
1143 .unlocked_ioctl
= gfs2_ioctl
,
1146 .release
= gfs2_release
,
1147 .fsync
= gfs2_fsync
,
1148 .splice_read
= generic_file_splice_read
,
1149 .splice_write
= gfs2_file_splice_write
,
1150 .setlease
= generic_setlease
,
1151 .fallocate
= gfs2_fallocate
,
1154 const struct file_operations gfs2_dir_fops_nolock
= {
1155 .iterate
= gfs2_readdir
,
1156 .unlocked_ioctl
= gfs2_ioctl
,
1158 .release
= gfs2_release
,
1159 .fsync
= gfs2_fsync
,
1160 .llseek
= default_llseek
,