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))
173 gfs2_holder_uninit(&gh);
177 void gfs2_set_inode_flags(struct inode *inode)
179 struct gfs2_inode *ip = GFS2_I(inode);
180 unsigned int flags = inode->i_flags;
182 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
183 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
185 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
186 flags |= S_IMMUTABLE;
187 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
189 if (ip->i_diskflags & GFS2_DIF_NOATIME)
191 if (ip->i_diskflags & GFS2_DIF_SYNC)
193 inode->i_flags = flags;
196 /* Flags that can be set by user space */
197 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
198 GFS2_DIF_IMMUTABLE| \
199 GFS2_DIF_APPENDONLY| \
204 GFS2_DIF_INHERIT_JDATA)
207 * do_gfs2_set_flags - set flags on an inode
208 * @filp: file pointer
209 * @reqflags: The flags to set
210 * @mask: Indicates which flags are valid
213 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
215 struct inode *inode = file_inode(filp);
216 struct gfs2_inode *ip = GFS2_I(inode);
217 struct gfs2_sbd *sdp = GFS2_SB(inode);
218 struct buffer_head *bh;
219 struct gfs2_holder gh;
221 u32 new_flags, flags;
223 error = mnt_want_write_file(filp);
227 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
232 if (!inode_owner_or_capable(inode))
236 flags = ip->i_diskflags;
237 new_flags = (flags & ~mask) | (reqflags & mask);
238 if ((new_flags ^ flags) == 0)
242 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
246 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
248 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
250 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
251 !capable(CAP_LINUX_IMMUTABLE))
253 if (!IS_IMMUTABLE(inode)) {
254 error = gfs2_permission(inode, MAY_WRITE);
258 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
259 if (flags & GFS2_DIF_JDATA)
260 gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
261 error = filemap_fdatawrite(inode->i_mapping);
264 error = filemap_fdatawait(inode->i_mapping);
268 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
271 error = gfs2_meta_inode_buffer(ip, &bh);
274 gfs2_trans_add_meta(ip->i_gl, bh);
275 ip->i_diskflags = new_flags;
276 gfs2_dinode_out(ip, bh->b_data);
278 gfs2_set_inode_flags(inode);
279 gfs2_set_aops(inode);
283 gfs2_glock_dq_uninit(&gh);
285 mnt_drop_write_file(filp);
289 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
291 struct inode *inode = file_inode(filp);
292 u32 fsflags, gfsflags;
294 if (get_user(fsflags, ptr))
297 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
298 if (!S_ISDIR(inode->i_mode)) {
299 gfsflags &= ~GFS2_DIF_TOPDIR;
300 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
301 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
302 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_SYSTEM);
304 return do_gfs2_set_flags(filp, gfsflags, ~(GFS2_DIF_SYSTEM | GFS2_DIF_JDATA));
307 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
310 case FS_IOC_GETFLAGS:
311 return gfs2_get_flags(filp, (u32 __user *)arg);
312 case FS_IOC_SETFLAGS:
313 return gfs2_set_flags(filp, (u32 __user *)arg);
315 return gfs2_fitrim(filp, (void __user *)arg);
321 * gfs2_size_hint - Give a hint to the size of a write request
322 * @filep: The struct file
323 * @offset: The file offset of the write
324 * @size: The length of the write
326 * When we are about to do a write, this function records the total
327 * write size in order to provide a suitable hint to the lower layers
328 * about how many blocks will be required.
332 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
334 struct inode *inode = file_inode(filep);
335 struct gfs2_sbd *sdp = GFS2_SB(inode);
336 struct gfs2_inode *ip = GFS2_I(inode);
337 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
338 int hint = min_t(size_t, INT_MAX, blks);
340 if (hint > atomic_read(&ip->i_res.rs_sizehint))
341 atomic_set(&ip->i_res.rs_sizehint, hint);
345 * gfs2_allocate_page_backing - Use bmap to allocate blocks
346 * @page: The (locked) page to allocate backing for
348 * We try to allocate all the blocks required for the page in
349 * one go. This might fail for various reasons, so we keep
350 * trying until all the blocks to back this page are allocated.
351 * If some of the blocks are already allocated, thats ok too.
354 static int gfs2_allocate_page_backing(struct page *page)
356 struct inode *inode = page->mapping->host;
357 struct buffer_head bh;
358 unsigned long size = PAGE_SIZE;
359 u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits);
364 gfs2_block_map(inode, lblock, &bh, 1);
365 if (!buffer_mapped(&bh))
368 lblock += (bh.b_size >> inode->i_blkbits);
374 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
375 * @vma: The virtual memory area
376 * @vmf: The virtual memory fault containing the page to become writable
378 * When the page becomes writable, we need to ensure that we have
379 * blocks allocated on disk to back that page.
382 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
384 struct page *page = vmf->page;
385 struct inode *inode = file_inode(vma->vm_file);
386 struct gfs2_inode *ip = GFS2_I(inode);
387 struct gfs2_sbd *sdp = GFS2_SB(inode);
388 struct gfs2_alloc_parms ap = { .aflags = 0, };
389 unsigned long last_index;
390 u64 pos = page->index << PAGE_SHIFT;
391 unsigned int data_blocks, ind_blocks, rblocks;
392 struct gfs2_holder gh;
396 sb_start_pagefault(inode->i_sb);
398 /* Update file times before taking page lock */
399 file_update_time(vma->vm_file);
401 ret = gfs2_rsqa_alloc(ip);
405 gfs2_size_hint(vma->vm_file, pos, PAGE_SIZE);
407 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
408 ret = gfs2_glock_nq(&gh);
412 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
413 set_bit(GIF_SW_PAGED, &ip->i_flags);
415 if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) {
417 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
424 ret = gfs2_rindex_update(sdp);
428 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
429 ap.target = data_blocks + ind_blocks;
430 ret = gfs2_quota_lock_check(ip, &ap);
433 ret = gfs2_inplace_reserve(ip, &ap);
435 goto out_quota_unlock;
437 rblocks = RES_DINODE + ind_blocks;
438 if (gfs2_is_jdata(ip))
439 rblocks += data_blocks ? data_blocks : 1;
440 if (ind_blocks || data_blocks) {
441 rblocks += RES_STATFS + RES_QUOTA;
442 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
444 ret = gfs2_trans_begin(sdp, rblocks, 0);
450 size = i_size_read(inode);
451 last_index = (size - 1) >> PAGE_SHIFT;
452 /* Check page index against inode size */
453 if (size == 0 || (page->index > last_index))
457 /* If truncated, we must retry the operation, we may have raced
458 * with the glock demotion code.
460 if (!PageUptodate(page) || page->mapping != inode->i_mapping)
463 /* Unstuff, if required, and allocate backing blocks for page */
465 if (gfs2_is_stuffed(ip))
466 ret = gfs2_unstuff_dinode(ip, page);
468 ret = gfs2_allocate_page_backing(page);
475 gfs2_inplace_release(ip);
477 gfs2_quota_unlock(ip);
481 gfs2_holder_uninit(&gh);
483 set_page_dirty(page);
484 wait_for_stable_page(page);
487 sb_end_pagefault(inode->i_sb);
488 return block_page_mkwrite_return(ret);
491 static const struct vm_operations_struct gfs2_vm_ops = {
492 .fault = filemap_fault,
493 .map_pages = filemap_map_pages,
494 .page_mkwrite = gfs2_page_mkwrite,
499 * @file: The file to map
500 * @vma: The VMA which described the mapping
502 * There is no need to get a lock here unless we should be updating
503 * atime. We ignore any locking errors since the only consequence is
504 * a missed atime update (which will just be deferred until later).
509 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
511 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
513 if (!(file->f_flags & O_NOATIME) &&
514 !IS_NOATIME(&ip->i_inode)) {
515 struct gfs2_holder i_gh;
518 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
522 /* grab lock to update inode */
523 gfs2_glock_dq_uninit(&i_gh);
526 vma->vm_ops = &gfs2_vm_ops;
532 * gfs2_open_common - This is common to open and atomic_open
533 * @inode: The inode being opened
534 * @file: The file being opened
536 * This maybe called under a glock or not depending upon how it has
537 * been called. We must always be called under a glock for regular
538 * files, however. For other file types, it does not matter whether
539 * we hold the glock or not.
541 * Returns: Error code or 0 for success
544 int gfs2_open_common(struct inode *inode, struct file *file)
546 struct gfs2_file *fp;
549 if (S_ISREG(inode->i_mode)) {
550 ret = generic_file_open(inode, file);
555 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
559 mutex_init(&fp->f_fl_mutex);
561 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
562 file->private_data = fp;
567 * gfs2_open - open a file
568 * @inode: the inode to open
569 * @file: the struct file for this opening
571 * After atomic_open, this function is only used for opening files
572 * which are already cached. We must still get the glock for regular
573 * files to ensure that we have the file size uptodate for the large
574 * file check which is in the common code. That is only an issue for
575 * regular files though.
580 static int gfs2_open(struct inode *inode, struct file *file)
582 struct gfs2_inode *ip = GFS2_I(inode);
583 struct gfs2_holder i_gh;
585 bool need_unlock = false;
587 if (S_ISREG(ip->i_inode.i_mode)) {
588 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
595 error = gfs2_open_common(inode, file);
598 gfs2_glock_dq_uninit(&i_gh);
604 * gfs2_release - called to close a struct file
605 * @inode: the inode the struct file belongs to
606 * @file: the struct file being closed
611 static int gfs2_release(struct inode *inode, struct file *file)
613 struct gfs2_inode *ip = GFS2_I(inode);
615 kfree(file->private_data);
616 file->private_data = NULL;
618 if (!(file->f_mode & FMODE_WRITE))
621 gfs2_rsqa_delete(ip, &inode->i_writecount);
626 * gfs2_fsync - sync the dirty data for a file (across the cluster)
627 * @file: the file that points to the dentry
628 * @start: the start position in the file to sync
629 * @end: the end position in the file to sync
630 * @datasync: set if we can ignore timestamp changes
632 * We split the data flushing here so that we don't wait for the data
633 * until after we've also sent the metadata to disk. Note that for
634 * data=ordered, we will write & wait for the data at the log flush
635 * stage anyway, so this is unlikely to make much of a difference
636 * except in the data=writeback case.
638 * If the fdatawrite fails due to any reason except -EIO, we will
639 * continue the remainder of the fsync, although we'll still report
640 * the error at the end. This is to match filemap_write_and_wait_range()
646 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
649 struct address_space *mapping = file->f_mapping;
650 struct inode *inode = mapping->host;
651 int sync_state = inode->i_state & I_DIRTY_ALL;
652 struct gfs2_inode *ip = GFS2_I(inode);
653 int ret = 0, ret1 = 0;
655 if (mapping->nrpages) {
656 ret1 = filemap_fdatawrite_range(mapping, start, end);
661 if (!gfs2_is_jdata(ip))
662 sync_state &= ~I_DIRTY_PAGES;
664 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
667 ret = sync_inode_metadata(inode, 1);
670 if (gfs2_is_jdata(ip))
671 filemap_write_and_wait(mapping);
672 gfs2_ail_flush(ip->i_gl, 1);
675 if (mapping->nrpages)
676 ret = filemap_fdatawait_range(mapping, start, end);
678 return ret ? ret : ret1;
682 * gfs2_file_write_iter - Perform a write to a file
683 * @iocb: The io context
684 * @iov: The data to write
685 * @nr_segs: Number of @iov segments
686 * @pos: The file position
688 * We have to do a lock/unlock here to refresh the inode size for
689 * O_APPEND writes, otherwise we can land up writing at the wrong
690 * offset. There is still a race, but provided the app is using its
691 * own file locking, this will make O_APPEND work as expected.
695 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
697 struct file *file = iocb->ki_filp;
698 struct gfs2_inode *ip = GFS2_I(file_inode(file));
701 ret = gfs2_rsqa_alloc(ip);
705 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
707 if (iocb->ki_flags & IOCB_APPEND) {
708 struct gfs2_holder gh;
710 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
713 gfs2_glock_dq_uninit(&gh);
716 return generic_file_write_iter(iocb, from);
719 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
722 struct gfs2_inode *ip = GFS2_I(inode);
723 struct buffer_head *dibh;
725 unsigned int nr_blks;
726 sector_t lblock = offset >> inode->i_blkbits;
728 error = gfs2_meta_inode_buffer(ip, &dibh);
732 gfs2_trans_add_meta(ip->i_gl, dibh);
734 if (gfs2_is_stuffed(ip)) {
735 error = gfs2_unstuff_dinode(ip, NULL);
741 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
743 set_buffer_zeronew(&bh_map);
745 error = gfs2_block_map(inode, lblock, &bh_map, 1);
748 len -= bh_map.b_size;
749 nr_blks = bh_map.b_size >> inode->i_blkbits;
751 if (!buffer_new(&bh_map))
753 if (unlikely(!buffer_zeronew(&bh_map))) {
763 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
764 * blocks, determine how many bytes can be written.
765 * @ip: The inode in question.
766 * @len: Max cap of bytes. What we return in *len must be <= this.
767 * @data_blocks: Compute and return the number of data blocks needed
768 * @ind_blocks: Compute and return the number of indirect blocks needed
769 * @max_blocks: The total blocks available to work with.
771 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
773 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
774 unsigned int *data_blocks, unsigned int *ind_blocks,
775 unsigned int max_blocks)
778 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
779 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
781 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
782 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
786 *data_blocks = max_data;
787 *ind_blocks = max_blocks - max_data;
788 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
791 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
795 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
797 struct inode *inode = file_inode(file);
798 struct gfs2_sbd *sdp = GFS2_SB(inode);
799 struct gfs2_inode *ip = GFS2_I(inode);
800 struct gfs2_alloc_parms ap = { .aflags = 0, };
801 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
802 loff_t bytes, max_bytes, max_blks = UINT_MAX;
804 const loff_t pos = offset;
805 const loff_t count = len;
806 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
807 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
808 loff_t max_chunk_size = UINT_MAX & bsize_mask;
810 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
812 offset &= bsize_mask;
815 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
820 bytes = sdp->sd_sb.sb_bsize;
822 gfs2_size_hint(file, offset, len);
824 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
825 ap.min_target = data_blocks + ind_blocks;
830 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
836 /* We need to determine how many bytes we can actually
837 * fallocate without exceeding quota or going over the
838 * end of the fs. We start off optimistically by assuming
839 * we can write max_bytes */
840 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
842 /* Since max_bytes is most likely a theoretical max, we
843 * calculate a more realistic 'bytes' to serve as a good
844 * starting point for the number of bytes we may be able
846 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
847 ap.target = data_blocks + ind_blocks;
849 error = gfs2_quota_lock_check(ip, &ap);
852 /* ap.allowed tells us how many blocks quota will allow
853 * us to write. Check if this reduces max_blks */
854 if (ap.allowed && ap.allowed < max_blks)
855 max_blks = ap.allowed;
857 error = gfs2_inplace_reserve(ip, &ap);
861 /* check if the selected rgrp limits our max_blks further */
862 if (ap.allowed && ap.allowed < max_blks)
863 max_blks = ap.allowed;
865 /* Almost done. Calculate bytes that can be written using
866 * max_blks. We also recompute max_bytes, data_blocks and
868 calc_max_reserv(ip, &max_bytes, &data_blocks,
869 &ind_blocks, max_blks);
871 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
872 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
873 if (gfs2_is_jdata(ip))
874 rblocks += data_blocks ? data_blocks : 1;
876 error = gfs2_trans_begin(sdp, rblocks,
877 PAGE_SIZE/sdp->sd_sb.sb_bsize);
881 error = fallocate_chunk(inode, offset, max_bytes, mode);
889 gfs2_inplace_release(ip);
890 gfs2_quota_unlock(ip);
893 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) {
894 i_size_write(inode, pos + count);
895 file_update_time(file);
896 mark_inode_dirty(inode);
899 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
900 return vfs_fsync_range(file, pos, pos + count - 1,
901 (file->f_flags & __O_SYNC) ? 0 : 1);
905 gfs2_inplace_release(ip);
907 gfs2_quota_unlock(ip);
911 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
913 struct inode *inode = file_inode(file);
914 struct gfs2_inode *ip = GFS2_I(inode);
915 struct gfs2_holder gh;
918 if ((mode & ~FALLOC_FL_KEEP_SIZE) || gfs2_is_jdata(ip))
923 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
924 ret = gfs2_glock_nq(&gh);
928 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
929 (offset + len) > inode->i_size) {
930 ret = inode_newsize_ok(inode, offset + len);
935 ret = get_write_access(inode);
939 ret = gfs2_rsqa_alloc(ip);
943 ret = __gfs2_fallocate(file, mode, offset, len);
945 gfs2_rs_deltree(&ip->i_res);
948 put_write_access(inode);
952 gfs2_holder_uninit(&gh);
957 static ssize_t gfs2_file_splice_read(struct file *in, loff_t *ppos,
958 struct pipe_inode_info *pipe, size_t len,
961 struct inode *inode = in->f_mapping->host;
962 struct gfs2_inode *ip = GFS2_I(inode);
963 struct gfs2_holder gh;
968 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
974 gfs2_glock_dq_uninit(&gh);
977 return generic_file_splice_read(in, ppos, pipe, len, flags);
981 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
982 struct file *out, loff_t *ppos,
983 size_t len, unsigned int flags)
986 struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
988 error = gfs2_rsqa_alloc(ip);
990 return (ssize_t)error;
992 gfs2_size_hint(out, *ppos, len);
994 return iter_file_splice_write(pipe, out, ppos, len, flags);
997 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1000 * gfs2_lock - acquire/release a posix lock on a file
1001 * @file: the file pointer
1002 * @cmd: either modify or retrieve lock state, possibly wait
1003 * @fl: type and range of lock
1008 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1010 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1011 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1012 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1014 if (!(fl->fl_flags & FL_POSIX))
1016 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1019 if (cmd == F_CANCELLK) {
1022 fl->fl_type = F_UNLCK;
1024 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1025 if (fl->fl_type == F_UNLCK)
1026 locks_lock_file_wait(file, fl);
1030 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1031 else if (fl->fl_type == F_UNLCK)
1032 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1034 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1037 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1039 struct gfs2_file *fp = file->private_data;
1040 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1041 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1042 struct gfs2_glock *gl;
1048 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1049 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1051 mutex_lock(&fp->f_fl_mutex);
1055 if (fl_gh->gh_state == state)
1057 locks_lock_file_wait(file,
1058 &(struct file_lock) {
1060 .fl_flags = FL_FLOCK
1062 gfs2_glock_dq(fl_gh);
1063 gfs2_holder_reinit(state, flags, fl_gh);
1065 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1066 &gfs2_flock_glops, CREATE, &gl);
1069 gfs2_holder_init(gl, state, flags, fl_gh);
1072 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1073 error = gfs2_glock_nq(fl_gh);
1074 if (error != GLR_TRYFAILED)
1076 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1077 fl_gh->gh_error = 0;
1081 gfs2_holder_uninit(fl_gh);
1082 if (error == GLR_TRYFAILED)
1085 error = locks_lock_file_wait(file, fl);
1086 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1090 mutex_unlock(&fp->f_fl_mutex);
1094 static void do_unflock(struct file *file, struct file_lock *fl)
1096 struct gfs2_file *fp = file->private_data;
1097 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1099 mutex_lock(&fp->f_fl_mutex);
1100 locks_lock_file_wait(file, fl);
1102 gfs2_glock_dq(fl_gh);
1103 gfs2_holder_uninit(fl_gh);
1105 mutex_unlock(&fp->f_fl_mutex);
1109 * gfs2_flock - acquire/release a flock lock on a file
1110 * @file: the file pointer
1111 * @cmd: either modify or retrieve lock state, possibly wait
1112 * @fl: type and range of lock
1117 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1119 if (!(fl->fl_flags & FL_FLOCK))
1121 if (fl->fl_type & LOCK_MAND)
1124 if (fl->fl_type == F_UNLCK) {
1125 do_unflock(file, fl);
1128 return do_flock(file, cmd, fl);
1132 const struct file_operations gfs2_file_fops = {
1133 .llseek = gfs2_llseek,
1134 .read_iter = generic_file_read_iter,
1135 .write_iter = gfs2_file_write_iter,
1136 .unlocked_ioctl = gfs2_ioctl,
1139 .release = gfs2_release,
1140 .fsync = gfs2_fsync,
1142 .flock = gfs2_flock,
1143 .splice_read = gfs2_file_splice_read,
1144 .splice_write = gfs2_file_splice_write,
1145 .setlease = simple_nosetlease,
1146 .fallocate = gfs2_fallocate,
1149 const struct file_operations gfs2_dir_fops = {
1150 .iterate_shared = gfs2_readdir,
1151 .unlocked_ioctl = gfs2_ioctl,
1153 .release = gfs2_release,
1154 .fsync = gfs2_fsync,
1156 .flock = gfs2_flock,
1157 .llseek = default_llseek,
1160 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1162 const struct file_operations gfs2_file_fops_nolock = {
1163 .llseek = gfs2_llseek,
1164 .read_iter = generic_file_read_iter,
1165 .write_iter = gfs2_file_write_iter,
1166 .unlocked_ioctl = gfs2_ioctl,
1169 .release = gfs2_release,
1170 .fsync = gfs2_fsync,
1171 .splice_read = gfs2_file_splice_read,
1172 .splice_write = gfs2_file_splice_write,
1173 .setlease = generic_setlease,
1174 .fallocate = gfs2_fallocate,
1177 const struct file_operations gfs2_dir_fops_nolock = {
1178 .iterate_shared = gfs2_readdir,
1179 .unlocked_ioctl = gfs2_ioctl,
1181 .release = gfs2_release,
1182 .fsync = gfs2_fsync,
1183 .llseek = default_llseek,