xref: /openbmc/linux/fs/gfs2/file.c (revision a09d2831)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
4  *
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.
8  */
9 
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>
17 #include <linux/mm.h>
18 #include <linux/mount.h>
19 #include <linux/fs.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/ext2_fs.h>
22 #include <linux/crc32.h>
23 #include <linux/writeback.h>
24 #include <asm/uaccess.h>
25 #include <linux/dlm.h>
26 #include <linux/dlm_plock.h>
27 
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "dir.h"
32 #include "glock.h"
33 #include "glops.h"
34 #include "inode.h"
35 #include "log.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "rgrp.h"
39 #include "trans.h"
40 #include "util.h"
41 
42 /**
43  * gfs2_llseek - seek to a location in a file
44  * @file: the file
45  * @offset: the offset
46  * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
47  *
48  * SEEK_END requires the glock for the file because it references the
49  * file's size.
50  *
51  * Returns: The new offset, or errno
52  */
53 
54 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
55 {
56 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
57 	struct gfs2_holder i_gh;
58 	loff_t error;
59 
60 	if (origin == 2) {
61 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
62 					   &i_gh);
63 		if (!error) {
64 			error = generic_file_llseek_unlocked(file, offset, origin);
65 			gfs2_glock_dq_uninit(&i_gh);
66 		}
67 	} else
68 		error = generic_file_llseek_unlocked(file, offset, origin);
69 
70 	return error;
71 }
72 
73 /**
74  * gfs2_readdir - Read directory entries from a directory
75  * @file: The directory to read from
76  * @dirent: Buffer for dirents
77  * @filldir: Function used to do the copying
78  *
79  * Returns: errno
80  */
81 
82 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
83 {
84 	struct inode *dir = file->f_mapping->host;
85 	struct gfs2_inode *dip = GFS2_I(dir);
86 	struct gfs2_holder d_gh;
87 	u64 offset = file->f_pos;
88 	int error;
89 
90 	gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
91 	error = gfs2_glock_nq(&d_gh);
92 	if (error) {
93 		gfs2_holder_uninit(&d_gh);
94 		return error;
95 	}
96 
97 	error = gfs2_dir_read(dir, &offset, dirent, filldir);
98 
99 	gfs2_glock_dq_uninit(&d_gh);
100 
101 	file->f_pos = offset;
102 
103 	return error;
104 }
105 
106 /**
107  * fsflags_cvt
108  * @table: A table of 32 u32 flags
109  * @val: a 32 bit value to convert
110  *
111  * This function can be used to convert between fsflags values and
112  * GFS2's own flags values.
113  *
114  * Returns: the converted flags
115  */
116 static u32 fsflags_cvt(const u32 *table, u32 val)
117 {
118 	u32 res = 0;
119 	while(val) {
120 		if (val & 1)
121 			res |= *table;
122 		table++;
123 		val >>= 1;
124 	}
125 	return res;
126 }
127 
128 static const u32 fsflags_to_gfs2[32] = {
129 	[3] = GFS2_DIF_SYNC,
130 	[4] = GFS2_DIF_IMMUTABLE,
131 	[5] = GFS2_DIF_APPENDONLY,
132 	[7] = GFS2_DIF_NOATIME,
133 	[12] = GFS2_DIF_EXHASH,
134 	[14] = GFS2_DIF_INHERIT_JDATA,
135 };
136 
137 static const u32 gfs2_to_fsflags[32] = {
138 	[gfs2fl_Sync] = FS_SYNC_FL,
139 	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
140 	[gfs2fl_AppendOnly] = FS_APPEND_FL,
141 	[gfs2fl_NoAtime] = FS_NOATIME_FL,
142 	[gfs2fl_ExHash] = FS_INDEX_FL,
143 	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
144 };
145 
146 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
147 {
148 	struct inode *inode = filp->f_path.dentry->d_inode;
149 	struct gfs2_inode *ip = GFS2_I(inode);
150 	struct gfs2_holder gh;
151 	int error;
152 	u32 fsflags;
153 
154 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
155 	error = gfs2_glock_nq(&gh);
156 	if (error)
157 		return error;
158 
159 	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
160 	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
161 		fsflags |= FS_JOURNAL_DATA_FL;
162 	if (put_user(fsflags, ptr))
163 		error = -EFAULT;
164 
165 	gfs2_glock_dq(&gh);
166 	gfs2_holder_uninit(&gh);
167 	return error;
168 }
169 
170 void gfs2_set_inode_flags(struct inode *inode)
171 {
172 	struct gfs2_inode *ip = GFS2_I(inode);
173 	unsigned int flags = inode->i_flags;
174 
175 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
176 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
177 		flags |= S_IMMUTABLE;
178 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
179 		flags |= S_APPEND;
180 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
181 		flags |= S_NOATIME;
182 	if (ip->i_diskflags & GFS2_DIF_SYNC)
183 		flags |= S_SYNC;
184 	inode->i_flags = flags;
185 }
186 
187 /* Flags that can be set by user space */
188 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
189 			     GFS2_DIF_IMMUTABLE|		\
190 			     GFS2_DIF_APPENDONLY|		\
191 			     GFS2_DIF_NOATIME|			\
192 			     GFS2_DIF_SYNC|			\
193 			     GFS2_DIF_SYSTEM|			\
194 			     GFS2_DIF_INHERIT_JDATA)
195 
196 /**
197  * gfs2_set_flags - set flags on an inode
198  * @inode: The inode
199  * @flags: The flags to set
200  * @mask: Indicates which flags are valid
201  *
202  */
203 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
204 {
205 	struct inode *inode = filp->f_path.dentry->d_inode;
206 	struct gfs2_inode *ip = GFS2_I(inode);
207 	struct gfs2_sbd *sdp = GFS2_SB(inode);
208 	struct buffer_head *bh;
209 	struct gfs2_holder gh;
210 	int error;
211 	u32 new_flags, flags;
212 
213 	error = mnt_want_write(filp->f_path.mnt);
214 	if (error)
215 		return error;
216 
217 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
218 	if (error)
219 		goto out_drop_write;
220 
221 	flags = ip->i_diskflags;
222 	new_flags = (flags & ~mask) | (reqflags & mask);
223 	if ((new_flags ^ flags) == 0)
224 		goto out;
225 
226 	error = -EINVAL;
227 	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
228 		goto out;
229 
230 	error = -EPERM;
231 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
232 		goto out;
233 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
234 		goto out;
235 	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
236 	    !capable(CAP_LINUX_IMMUTABLE))
237 		goto out;
238 	if (!IS_IMMUTABLE(inode)) {
239 		error = gfs2_permission(inode, MAY_WRITE);
240 		if (error)
241 			goto out;
242 	}
243 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
244 		if (flags & GFS2_DIF_JDATA)
245 			gfs2_log_flush(sdp, ip->i_gl);
246 		error = filemap_fdatawrite(inode->i_mapping);
247 		if (error)
248 			goto out;
249 		error = filemap_fdatawait(inode->i_mapping);
250 		if (error)
251 			goto out;
252 	}
253 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
254 	if (error)
255 		goto out;
256 	error = gfs2_meta_inode_buffer(ip, &bh);
257 	if (error)
258 		goto out_trans_end;
259 	gfs2_trans_add_bh(ip->i_gl, bh, 1);
260 	ip->i_diskflags = new_flags;
261 	gfs2_dinode_out(ip, bh->b_data);
262 	brelse(bh);
263 	gfs2_set_inode_flags(inode);
264 	gfs2_set_aops(inode);
265 out_trans_end:
266 	gfs2_trans_end(sdp);
267 out:
268 	gfs2_glock_dq_uninit(&gh);
269 out_drop_write:
270 	mnt_drop_write(filp->f_path.mnt);
271 	return error;
272 }
273 
274 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
275 {
276 	struct inode *inode = filp->f_path.dentry->d_inode;
277 	u32 fsflags, gfsflags;
278 	if (get_user(fsflags, ptr))
279 		return -EFAULT;
280 	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
281 	if (!S_ISDIR(inode->i_mode)) {
282 		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
283 			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
284 		return do_gfs2_set_flags(filp, gfsflags, ~0);
285 	}
286 	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
287 }
288 
289 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
290 {
291 	switch(cmd) {
292 	case FS_IOC_GETFLAGS:
293 		return gfs2_get_flags(filp, (u32 __user *)arg);
294 	case FS_IOC_SETFLAGS:
295 		return gfs2_set_flags(filp, (u32 __user *)arg);
296 	}
297 	return -ENOTTY;
298 }
299 
300 /**
301  * gfs2_allocate_page_backing - Use bmap to allocate blocks
302  * @page: The (locked) page to allocate backing for
303  *
304  * We try to allocate all the blocks required for the page in
305  * one go. This might fail for various reasons, so we keep
306  * trying until all the blocks to back this page are allocated.
307  * If some of the blocks are already allocated, thats ok too.
308  */
309 
310 static int gfs2_allocate_page_backing(struct page *page)
311 {
312 	struct inode *inode = page->mapping->host;
313 	struct buffer_head bh;
314 	unsigned long size = PAGE_CACHE_SIZE;
315 	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
316 
317 	do {
318 		bh.b_state = 0;
319 		bh.b_size = size;
320 		gfs2_block_map(inode, lblock, &bh, 1);
321 		if (!buffer_mapped(&bh))
322 			return -EIO;
323 		size -= bh.b_size;
324 		lblock += (bh.b_size >> inode->i_blkbits);
325 	} while(size > 0);
326 	return 0;
327 }
328 
329 /**
330  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
331  * @vma: The virtual memory area
332  * @page: The page which is about to become writable
333  *
334  * When the page becomes writable, we need to ensure that we have
335  * blocks allocated on disk to back that page.
336  */
337 
338 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
339 {
340 	struct page *page = vmf->page;
341 	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
342 	struct gfs2_inode *ip = GFS2_I(inode);
343 	struct gfs2_sbd *sdp = GFS2_SB(inode);
344 	unsigned long last_index;
345 	u64 pos = page->index << PAGE_CACHE_SHIFT;
346 	unsigned int data_blocks, ind_blocks, rblocks;
347 	int alloc_required = 0;
348 	struct gfs2_holder gh;
349 	struct gfs2_alloc *al;
350 	int ret;
351 
352 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
353 	ret = gfs2_glock_nq(&gh);
354 	if (ret)
355 		goto out;
356 
357 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
358 	set_bit(GIF_SW_PAGED, &ip->i_flags);
359 
360 	ret = gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE, &alloc_required);
361 	if (ret || !alloc_required)
362 		goto out_unlock;
363 	ret = -ENOMEM;
364 	al = gfs2_alloc_get(ip);
365 	if (al == NULL)
366 		goto out_unlock;
367 
368 	ret = gfs2_quota_lock_check(ip);
369 	if (ret)
370 		goto out_alloc_put;
371 	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
372 	al->al_requested = data_blocks + ind_blocks;
373 	ret = gfs2_inplace_reserve(ip);
374 	if (ret)
375 		goto out_quota_unlock;
376 
377 	rblocks = RES_DINODE + ind_blocks;
378 	if (gfs2_is_jdata(ip))
379 		rblocks += data_blocks ? data_blocks : 1;
380 	if (ind_blocks || data_blocks)
381 		rblocks += RES_STATFS + RES_QUOTA;
382 	ret = gfs2_trans_begin(sdp, rblocks, 0);
383 	if (ret)
384 		goto out_trans_fail;
385 
386 	lock_page(page);
387 	ret = -EINVAL;
388 	last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
389 	if (page->index > last_index)
390 		goto out_unlock_page;
391 	ret = 0;
392 	if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
393 		goto out_unlock_page;
394 	if (gfs2_is_stuffed(ip)) {
395 		ret = gfs2_unstuff_dinode(ip, page);
396 		if (ret)
397 			goto out_unlock_page;
398 	}
399 	ret = gfs2_allocate_page_backing(page);
400 
401 out_unlock_page:
402 	unlock_page(page);
403 	gfs2_trans_end(sdp);
404 out_trans_fail:
405 	gfs2_inplace_release(ip);
406 out_quota_unlock:
407 	gfs2_quota_unlock(ip);
408 out_alloc_put:
409 	gfs2_alloc_put(ip);
410 out_unlock:
411 	gfs2_glock_dq(&gh);
412 out:
413 	gfs2_holder_uninit(&gh);
414 	if (ret == -ENOMEM)
415 		ret = VM_FAULT_OOM;
416 	else if (ret)
417 		ret = VM_FAULT_SIGBUS;
418 	return ret;
419 }
420 
421 static const struct vm_operations_struct gfs2_vm_ops = {
422 	.fault = filemap_fault,
423 	.page_mkwrite = gfs2_page_mkwrite,
424 };
425 
426 /**
427  * gfs2_mmap -
428  * @file: The file to map
429  * @vma: The VMA which described the mapping
430  *
431  * There is no need to get a lock here unless we should be updating
432  * atime. We ignore any locking errors since the only consequence is
433  * a missed atime update (which will just be deferred until later).
434  *
435  * Returns: 0
436  */
437 
438 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
439 {
440 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
441 
442 	if (!(file->f_flags & O_NOATIME)) {
443 		struct gfs2_holder i_gh;
444 		int error;
445 
446 		gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
447 		error = gfs2_glock_nq(&i_gh);
448 		file_accessed(file);
449 		if (error == 0)
450 			gfs2_glock_dq_uninit(&i_gh);
451 	}
452 	vma->vm_ops = &gfs2_vm_ops;
453 	vma->vm_flags |= VM_CAN_NONLINEAR;
454 
455 	return 0;
456 }
457 
458 /**
459  * gfs2_open - open a file
460  * @inode: the inode to open
461  * @file: the struct file for this opening
462  *
463  * Returns: errno
464  */
465 
466 static int gfs2_open(struct inode *inode, struct file *file)
467 {
468 	struct gfs2_inode *ip = GFS2_I(inode);
469 	struct gfs2_holder i_gh;
470 	struct gfs2_file *fp;
471 	int error;
472 
473 	fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
474 	if (!fp)
475 		return -ENOMEM;
476 
477 	mutex_init(&fp->f_fl_mutex);
478 
479 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
480 	file->private_data = fp;
481 
482 	if (S_ISREG(ip->i_inode.i_mode)) {
483 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
484 					   &i_gh);
485 		if (error)
486 			goto fail;
487 
488 		if (!(file->f_flags & O_LARGEFILE) &&
489 		    ip->i_disksize > MAX_NON_LFS) {
490 			error = -EOVERFLOW;
491 			goto fail_gunlock;
492 		}
493 
494 		gfs2_glock_dq_uninit(&i_gh);
495 	}
496 
497 	return 0;
498 
499 fail_gunlock:
500 	gfs2_glock_dq_uninit(&i_gh);
501 fail:
502 	file->private_data = NULL;
503 	kfree(fp);
504 	return error;
505 }
506 
507 /**
508  * gfs2_close - called to close a struct file
509  * @inode: the inode the struct file belongs to
510  * @file: the struct file being closed
511  *
512  * Returns: errno
513  */
514 
515 static int gfs2_close(struct inode *inode, struct file *file)
516 {
517 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
518 	struct gfs2_file *fp;
519 
520 	fp = file->private_data;
521 	file->private_data = NULL;
522 
523 	if (gfs2_assert_warn(sdp, fp))
524 		return -EIO;
525 
526 	kfree(fp);
527 
528 	return 0;
529 }
530 
531 /**
532  * gfs2_fsync - sync the dirty data for a file (across the cluster)
533  * @file: the file that points to the dentry (we ignore this)
534  * @dentry: the dentry that points to the inode to sync
535  *
536  * The VFS will flush "normal" data for us. We only need to worry
537  * about metadata here. For journaled data, we just do a log flush
538  * as we can't avoid it. Otherwise we can just bale out if datasync
539  * is set. For stuffed inodes we must flush the log in order to
540  * ensure that all data is on disk.
541  *
542  * The call to write_inode_now() is there to write back metadata and
543  * the inode itself. It does also try and write the data, but thats
544  * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
545  * for us.
546  *
547  * Returns: errno
548  */
549 
550 static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
551 {
552 	struct inode *inode = dentry->d_inode;
553 	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
554 	int ret = 0;
555 
556 	if (gfs2_is_jdata(GFS2_I(inode))) {
557 		gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
558 		return 0;
559 	}
560 
561 	if (sync_state != 0) {
562 		if (!datasync)
563 			ret = write_inode_now(inode, 0);
564 
565 		if (gfs2_is_stuffed(GFS2_I(inode)))
566 			gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
567 	}
568 
569 	return ret;
570 }
571 
572 /**
573  * gfs2_file_aio_write - Perform a write to a file
574  * @iocb: The io context
575  * @iov: The data to write
576  * @nr_segs: Number of @iov segments
577  * @pos: The file position
578  *
579  * We have to do a lock/unlock here to refresh the inode size for
580  * O_APPEND writes, otherwise we can land up writing at the wrong
581  * offset. There is still a race, but provided the app is using its
582  * own file locking, this will make O_APPEND work as expected.
583  *
584  */
585 
586 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
587 				   unsigned long nr_segs, loff_t pos)
588 {
589 	struct file *file = iocb->ki_filp;
590 
591 	if (file->f_flags & O_APPEND) {
592 		struct dentry *dentry = file->f_dentry;
593 		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
594 		struct gfs2_holder gh;
595 		int ret;
596 
597 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
598 		if (ret)
599 			return ret;
600 		gfs2_glock_dq_uninit(&gh);
601 	}
602 
603 	return generic_file_aio_write(iocb, iov, nr_segs, pos);
604 }
605 
606 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
607 
608 /**
609  * gfs2_setlease - acquire/release a file lease
610  * @file: the file pointer
611  * @arg: lease type
612  * @fl: file lock
613  *
614  * We don't currently have a way to enforce a lease across the whole
615  * cluster; until we do, disable leases (by just returning -EINVAL),
616  * unless the administrator has requested purely local locking.
617  *
618  * Returns: errno
619  */
620 
621 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
622 {
623 	return -EINVAL;
624 }
625 
626 /**
627  * gfs2_lock - acquire/release a posix lock on a file
628  * @file: the file pointer
629  * @cmd: either modify or retrieve lock state, possibly wait
630  * @fl: type and range of lock
631  *
632  * Returns: errno
633  */
634 
635 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
636 {
637 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
638 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
639 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
640 
641 	if (!(fl->fl_flags & FL_POSIX))
642 		return -ENOLCK;
643 	if (__mandatory_lock(&ip->i_inode))
644 		return -ENOLCK;
645 
646 	if (cmd == F_CANCELLK) {
647 		/* Hack: */
648 		cmd = F_SETLK;
649 		fl->fl_type = F_UNLCK;
650 	}
651 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
652 		return -EIO;
653 	if (IS_GETLK(cmd))
654 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
655 	else if (fl->fl_type == F_UNLCK)
656 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
657 	else
658 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
659 }
660 
661 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
662 {
663 	struct gfs2_file *fp = file->private_data;
664 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
665 	struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
666 	struct gfs2_glock *gl;
667 	unsigned int state;
668 	int flags;
669 	int error = 0;
670 
671 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
672 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
673 
674 	mutex_lock(&fp->f_fl_mutex);
675 
676 	gl = fl_gh->gh_gl;
677 	if (gl) {
678 		if (fl_gh->gh_state == state)
679 			goto out;
680 		flock_lock_file_wait(file,
681 				     &(struct file_lock){.fl_type = F_UNLCK});
682 		gfs2_glock_dq_wait(fl_gh);
683 		gfs2_holder_reinit(state, flags, fl_gh);
684 	} else {
685 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
686 				       &gfs2_flock_glops, CREATE, &gl);
687 		if (error)
688 			goto out;
689 		gfs2_holder_init(gl, state, flags, fl_gh);
690 		gfs2_glock_put(gl);
691 	}
692 	error = gfs2_glock_nq(fl_gh);
693 	if (error) {
694 		gfs2_holder_uninit(fl_gh);
695 		if (error == GLR_TRYFAILED)
696 			error = -EAGAIN;
697 	} else {
698 		error = flock_lock_file_wait(file, fl);
699 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
700 	}
701 
702 out:
703 	mutex_unlock(&fp->f_fl_mutex);
704 	return error;
705 }
706 
707 static void do_unflock(struct file *file, struct file_lock *fl)
708 {
709 	struct gfs2_file *fp = file->private_data;
710 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
711 
712 	mutex_lock(&fp->f_fl_mutex);
713 	flock_lock_file_wait(file, fl);
714 	if (fl_gh->gh_gl)
715 		gfs2_glock_dq_uninit(fl_gh);
716 	mutex_unlock(&fp->f_fl_mutex);
717 }
718 
719 /**
720  * gfs2_flock - acquire/release a flock lock on a file
721  * @file: the file pointer
722  * @cmd: either modify or retrieve lock state, possibly wait
723  * @fl: type and range of lock
724  *
725  * Returns: errno
726  */
727 
728 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
729 {
730 	if (!(fl->fl_flags & FL_FLOCK))
731 		return -ENOLCK;
732 	if (fl->fl_type & LOCK_MAND)
733 		return -EOPNOTSUPP;
734 
735 	if (fl->fl_type == F_UNLCK) {
736 		do_unflock(file, fl);
737 		return 0;
738 	} else {
739 		return do_flock(file, cmd, fl);
740 	}
741 }
742 
743 const struct file_operations gfs2_file_fops = {
744 	.llseek		= gfs2_llseek,
745 	.read		= do_sync_read,
746 	.aio_read	= generic_file_aio_read,
747 	.write		= do_sync_write,
748 	.aio_write	= gfs2_file_aio_write,
749 	.unlocked_ioctl	= gfs2_ioctl,
750 	.mmap		= gfs2_mmap,
751 	.open		= gfs2_open,
752 	.release	= gfs2_close,
753 	.fsync		= gfs2_fsync,
754 	.lock		= gfs2_lock,
755 	.flock		= gfs2_flock,
756 	.splice_read	= generic_file_splice_read,
757 	.splice_write	= generic_file_splice_write,
758 	.setlease	= gfs2_setlease,
759 };
760 
761 const struct file_operations gfs2_dir_fops = {
762 	.readdir	= gfs2_readdir,
763 	.unlocked_ioctl	= gfs2_ioctl,
764 	.open		= gfs2_open,
765 	.release	= gfs2_close,
766 	.fsync		= gfs2_fsync,
767 	.lock		= gfs2_lock,
768 	.flock		= gfs2_flock,
769 };
770 
771 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
772 
773 const struct file_operations gfs2_file_fops_nolock = {
774 	.llseek		= gfs2_llseek,
775 	.read		= do_sync_read,
776 	.aio_read	= generic_file_aio_read,
777 	.write		= do_sync_write,
778 	.aio_write	= gfs2_file_aio_write,
779 	.unlocked_ioctl	= gfs2_ioctl,
780 	.mmap		= gfs2_mmap,
781 	.open		= gfs2_open,
782 	.release	= gfs2_close,
783 	.fsync		= gfs2_fsync,
784 	.splice_read	= generic_file_splice_read,
785 	.splice_write	= generic_file_splice_write,
786 	.setlease	= generic_setlease,
787 };
788 
789 const struct file_operations gfs2_dir_fops_nolock = {
790 	.readdir	= gfs2_readdir,
791 	.unlocked_ioctl	= gfs2_ioctl,
792 	.open		= gfs2_open,
793 	.release	= gfs2_close,
794 	.fsync		= gfs2_fsync,
795 };
796 
797