xref: /linux/fs/gfs2/file.c (revision 4bdc0d676a643140bdf17dbf7eafedee3d496a3c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
5  */
6 
7 #include <linux/slab.h>
8 #include <linux/spinlock.h>
9 #include <linux/compat.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/pagemap.h>
13 #include <linux/uio.h>
14 #include <linux/blkdev.h>
15 #include <linux/mm.h>
16 #include <linux/mount.h>
17 #include <linux/fs.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/falloc.h>
20 #include <linux/swap.h>
21 #include <linux/crc32.h>
22 #include <linux/writeback.h>
23 #include <linux/uaccess.h>
24 #include <linux/dlm.h>
25 #include <linux/dlm_plock.h>
26 #include <linux/delay.h>
27 #include <linux/backing-dev.h>
28 
29 #include "gfs2.h"
30 #include "incore.h"
31 #include "bmap.h"
32 #include "aops.h"
33 #include "dir.h"
34 #include "glock.h"
35 #include "glops.h"
36 #include "inode.h"
37 #include "log.h"
38 #include "meta_io.h"
39 #include "quota.h"
40 #include "rgrp.h"
41 #include "trans.h"
42 #include "util.h"
43 
44 /**
45  * gfs2_llseek - seek to a location in a file
46  * @file: the file
47  * @offset: the offset
48  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
49  *
50  * SEEK_END requires the glock for the file because it references the
51  * file's size.
52  *
53  * Returns: The new offset, or errno
54  */
55 
56 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
57 {
58 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 	struct gfs2_holder i_gh;
60 	loff_t error;
61 
62 	switch (whence) {
63 	case SEEK_END:
64 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
65 					   &i_gh);
66 		if (!error) {
67 			error = generic_file_llseek(file, offset, whence);
68 			gfs2_glock_dq_uninit(&i_gh);
69 		}
70 		break;
71 
72 	case SEEK_DATA:
73 		error = gfs2_seek_data(file, offset);
74 		break;
75 
76 	case SEEK_HOLE:
77 		error = gfs2_seek_hole(file, offset);
78 		break;
79 
80 	case SEEK_CUR:
81 	case SEEK_SET:
82 		/*
83 		 * These don't reference inode->i_size and don't depend on the
84 		 * block mapping, so we don't need the glock.
85 		 */
86 		error = generic_file_llseek(file, offset, whence);
87 		break;
88 	default:
89 		error = -EINVAL;
90 	}
91 
92 	return error;
93 }
94 
95 /**
96  * gfs2_readdir - Iterator for a directory
97  * @file: The directory to read from
98  * @ctx: What to feed directory entries to
99  *
100  * Returns: errno
101  */
102 
103 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
104 {
105 	struct inode *dir = file->f_mapping->host;
106 	struct gfs2_inode *dip = GFS2_I(dir);
107 	struct gfs2_holder d_gh;
108 	int error;
109 
110 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
111 	if (error)
112 		return error;
113 
114 	error = gfs2_dir_read(dir, ctx, &file->f_ra);
115 
116 	gfs2_glock_dq_uninit(&d_gh);
117 
118 	return error;
119 }
120 
121 /**
122  * fsflag_gfs2flag
123  *
124  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
125  * and to GFS2_DIF_JDATA for non-directories.
126  */
127 static struct {
128 	u32 fsflag;
129 	u32 gfsflag;
130 } fsflag_gfs2flag[] = {
131 	{FS_SYNC_FL, GFS2_DIF_SYNC},
132 	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
133 	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
134 	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
135 	{FS_INDEX_FL, GFS2_DIF_EXHASH},
136 	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
137 	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
138 };
139 
140 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
141 {
142 	int i;
143 	u32 fsflags = 0;
144 
145 	if (S_ISDIR(inode->i_mode))
146 		gfsflags &= ~GFS2_DIF_JDATA;
147 	else
148 		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
149 
150 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
151 		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
152 			fsflags |= fsflag_gfs2flag[i].fsflag;
153 	return fsflags;
154 }
155 
156 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
157 {
158 	struct inode *inode = file_inode(filp);
159 	struct gfs2_inode *ip = GFS2_I(inode);
160 	struct gfs2_holder gh;
161 	int error;
162 	u32 fsflags;
163 
164 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
165 	error = gfs2_glock_nq(&gh);
166 	if (error)
167 		goto out_uninit;
168 
169 	fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
170 
171 	if (put_user(fsflags, ptr))
172 		error = -EFAULT;
173 
174 	gfs2_glock_dq(&gh);
175 out_uninit:
176 	gfs2_holder_uninit(&gh);
177 	return error;
178 }
179 
180 void gfs2_set_inode_flags(struct inode *inode)
181 {
182 	struct gfs2_inode *ip = GFS2_I(inode);
183 	unsigned int flags = inode->i_flags;
184 
185 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
186 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
187 		flags |= S_NOSEC;
188 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
189 		flags |= S_IMMUTABLE;
190 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
191 		flags |= S_APPEND;
192 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
193 		flags |= S_NOATIME;
194 	if (ip->i_diskflags & GFS2_DIF_SYNC)
195 		flags |= S_SYNC;
196 	inode->i_flags = flags;
197 }
198 
199 /* Flags that can be set by user space */
200 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
201 			     GFS2_DIF_IMMUTABLE|		\
202 			     GFS2_DIF_APPENDONLY|		\
203 			     GFS2_DIF_NOATIME|			\
204 			     GFS2_DIF_SYNC|			\
205 			     GFS2_DIF_TOPDIR|			\
206 			     GFS2_DIF_INHERIT_JDATA)
207 
208 /**
209  * do_gfs2_set_flags - set flags on an inode
210  * @filp: file pointer
211  * @reqflags: The flags to set
212  * @mask: Indicates which flags are valid
213  * @fsflags: The FS_* inode flags passed in
214  *
215  */
216 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask,
217 			     const u32 fsflags)
218 {
219 	struct inode *inode = file_inode(filp);
220 	struct gfs2_inode *ip = GFS2_I(inode);
221 	struct gfs2_sbd *sdp = GFS2_SB(inode);
222 	struct buffer_head *bh;
223 	struct gfs2_holder gh;
224 	int error;
225 	u32 new_flags, flags, oldflags;
226 
227 	error = mnt_want_write_file(filp);
228 	if (error)
229 		return error;
230 
231 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
232 	if (error)
233 		goto out_drop_write;
234 
235 	oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
236 	error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags);
237 	if (error)
238 		goto out;
239 
240 	error = -EACCES;
241 	if (!inode_owner_or_capable(inode))
242 		goto out;
243 
244 	error = 0;
245 	flags = ip->i_diskflags;
246 	new_flags = (flags & ~mask) | (reqflags & mask);
247 	if ((new_flags ^ flags) == 0)
248 		goto out;
249 
250 	error = -EPERM;
251 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
252 		goto out;
253 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
254 		goto out;
255 	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
256 	    !capable(CAP_LINUX_IMMUTABLE))
257 		goto out;
258 	if (!IS_IMMUTABLE(inode)) {
259 		error = gfs2_permission(inode, MAY_WRITE);
260 		if (error)
261 			goto out;
262 	}
263 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
264 		if (new_flags & GFS2_DIF_JDATA)
265 			gfs2_log_flush(sdp, ip->i_gl,
266 				       GFS2_LOG_HEAD_FLUSH_NORMAL |
267 				       GFS2_LFC_SET_FLAGS);
268 		error = filemap_fdatawrite(inode->i_mapping);
269 		if (error)
270 			goto out;
271 		error = filemap_fdatawait(inode->i_mapping);
272 		if (error)
273 			goto out;
274 		if (new_flags & GFS2_DIF_JDATA)
275 			gfs2_ordered_del_inode(ip);
276 	}
277 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
278 	if (error)
279 		goto out;
280 	error = gfs2_meta_inode_buffer(ip, &bh);
281 	if (error)
282 		goto out_trans_end;
283 	inode->i_ctime = current_time(inode);
284 	gfs2_trans_add_meta(ip->i_gl, bh);
285 	ip->i_diskflags = new_flags;
286 	gfs2_dinode_out(ip, bh->b_data);
287 	brelse(bh);
288 	gfs2_set_inode_flags(inode);
289 	gfs2_set_aops(inode);
290 out_trans_end:
291 	gfs2_trans_end(sdp);
292 out:
293 	gfs2_glock_dq_uninit(&gh);
294 out_drop_write:
295 	mnt_drop_write_file(filp);
296 	return error;
297 }
298 
299 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
300 {
301 	struct inode *inode = file_inode(filp);
302 	u32 fsflags, gfsflags = 0;
303 	u32 mask;
304 	int i;
305 
306 	if (get_user(fsflags, ptr))
307 		return -EFAULT;
308 
309 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
310 		if (fsflags & fsflag_gfs2flag[i].fsflag) {
311 			fsflags &= ~fsflag_gfs2flag[i].fsflag;
312 			gfsflags |= fsflag_gfs2flag[i].gfsflag;
313 		}
314 	}
315 	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
316 		return -EINVAL;
317 
318 	mask = GFS2_FLAGS_USER_SET;
319 	if (S_ISDIR(inode->i_mode)) {
320 		mask &= ~GFS2_DIF_JDATA;
321 	} else {
322 		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
323 		if (gfsflags & GFS2_DIF_TOPDIR)
324 			return -EINVAL;
325 		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
326 	}
327 
328 	return do_gfs2_set_flags(filp, gfsflags, mask, fsflags);
329 }
330 
331 static int gfs2_getlabel(struct file *filp, char __user *label)
332 {
333 	struct inode *inode = file_inode(filp);
334 	struct gfs2_sbd *sdp = GFS2_SB(inode);
335 
336 	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
337 		return -EFAULT;
338 
339 	return 0;
340 }
341 
342 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
343 {
344 	switch(cmd) {
345 	case FS_IOC_GETFLAGS:
346 		return gfs2_get_flags(filp, (u32 __user *)arg);
347 	case FS_IOC_SETFLAGS:
348 		return gfs2_set_flags(filp, (u32 __user *)arg);
349 	case FITRIM:
350 		return gfs2_fitrim(filp, (void __user *)arg);
351 	case FS_IOC_GETFSLABEL:
352 		return gfs2_getlabel(filp, (char __user *)arg);
353 	}
354 
355 	return -ENOTTY;
356 }
357 
358 #ifdef CONFIG_COMPAT
359 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
360 {
361 	switch(cmd) {
362 	/* These are just misnamed, they actually get/put from/to user an int */
363 	case FS_IOC32_GETFLAGS:
364 		cmd = FS_IOC_GETFLAGS;
365 		break;
366 	case FS_IOC32_SETFLAGS:
367 		cmd = FS_IOC_SETFLAGS;
368 		break;
369 	/* Keep this list in sync with gfs2_ioctl */
370 	case FITRIM:
371 	case FS_IOC_GETFSLABEL:
372 		break;
373 	default:
374 		return -ENOIOCTLCMD;
375 	}
376 
377 	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
378 }
379 #else
380 #define gfs2_compat_ioctl NULL
381 #endif
382 
383 /**
384  * gfs2_size_hint - Give a hint to the size of a write request
385  * @filep: The struct file
386  * @offset: The file offset of the write
387  * @size: The length of the write
388  *
389  * When we are about to do a write, this function records the total
390  * write size in order to provide a suitable hint to the lower layers
391  * about how many blocks will be required.
392  *
393  */
394 
395 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
396 {
397 	struct inode *inode = file_inode(filep);
398 	struct gfs2_sbd *sdp = GFS2_SB(inode);
399 	struct gfs2_inode *ip = GFS2_I(inode);
400 	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
401 	int hint = min_t(size_t, INT_MAX, blks);
402 
403 	if (hint > atomic_read(&ip->i_sizehint))
404 		atomic_set(&ip->i_sizehint, hint);
405 }
406 
407 /**
408  * gfs2_allocate_page_backing - Allocate blocks for a write fault
409  * @page: The (locked) page to allocate backing for
410  * @length: Size of the allocation
411  *
412  * We try to allocate all the blocks required for the page in one go.  This
413  * might fail for various reasons, so we keep trying until all the blocks to
414  * back this page are allocated.  If some of the blocks are already allocated,
415  * that is ok too.
416  */
417 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
418 {
419 	u64 pos = page_offset(page);
420 
421 	do {
422 		struct iomap iomap = { };
423 
424 		if (gfs2_iomap_get_alloc(page->mapping->host, pos, length, &iomap))
425 			return -EIO;
426 
427 		if (length < iomap.length)
428 			iomap.length = length;
429 		length -= iomap.length;
430 		pos += iomap.length;
431 	} while (length > 0);
432 
433 	return 0;
434 }
435 
436 /**
437  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
438  * @vma: The virtual memory area
439  * @vmf: The virtual memory fault containing the page to become writable
440  *
441  * When the page becomes writable, we need to ensure that we have
442  * blocks allocated on disk to back that page.
443  */
444 
445 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
446 {
447 	struct page *page = vmf->page;
448 	struct inode *inode = file_inode(vmf->vma->vm_file);
449 	struct gfs2_inode *ip = GFS2_I(inode);
450 	struct gfs2_sbd *sdp = GFS2_SB(inode);
451 	struct gfs2_alloc_parms ap = { .aflags = 0, };
452 	u64 offset = page_offset(page);
453 	unsigned int data_blocks, ind_blocks, rblocks;
454 	struct gfs2_holder gh;
455 	unsigned int length;
456 	loff_t size;
457 	int ret;
458 
459 	sb_start_pagefault(inode->i_sb);
460 
461 	ret = gfs2_rsqa_alloc(ip);
462 	if (ret)
463 		goto out;
464 
465 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
466 	ret = gfs2_glock_nq(&gh);
467 	if (ret)
468 		goto out_uninit;
469 
470 	/* Check page index against inode size */
471 	size = i_size_read(inode);
472 	if (offset >= size) {
473 		ret = -EINVAL;
474 		goto out_unlock;
475 	}
476 
477 	/* Update file times before taking page lock */
478 	file_update_time(vmf->vma->vm_file);
479 
480 	/* page is wholly or partially inside EOF */
481 	if (offset > size - PAGE_SIZE)
482 		length = offset_in_page(size);
483 	else
484 		length = PAGE_SIZE;
485 
486 	gfs2_size_hint(vmf->vma->vm_file, offset, length);
487 
488 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
489 	set_bit(GIF_SW_PAGED, &ip->i_flags);
490 
491 	/*
492 	 * iomap_writepage / iomap_writepages currently don't support inline
493 	 * files, so always unstuff here.
494 	 */
495 
496 	if (!gfs2_is_stuffed(ip) &&
497 	    !gfs2_write_alloc_required(ip, offset, length)) {
498 		lock_page(page);
499 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
500 			ret = -EAGAIN;
501 			unlock_page(page);
502 		}
503 		goto out_unlock;
504 	}
505 
506 	ret = gfs2_rindex_update(sdp);
507 	if (ret)
508 		goto out_unlock;
509 
510 	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
511 	ap.target = data_blocks + ind_blocks;
512 	ret = gfs2_quota_lock_check(ip, &ap);
513 	if (ret)
514 		goto out_unlock;
515 	ret = gfs2_inplace_reserve(ip, &ap);
516 	if (ret)
517 		goto out_quota_unlock;
518 
519 	rblocks = RES_DINODE + ind_blocks;
520 	if (gfs2_is_jdata(ip))
521 		rblocks += data_blocks ? data_blocks : 1;
522 	if (ind_blocks || data_blocks) {
523 		rblocks += RES_STATFS + RES_QUOTA;
524 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
525 	}
526 	ret = gfs2_trans_begin(sdp, rblocks, 0);
527 	if (ret)
528 		goto out_trans_fail;
529 
530 	lock_page(page);
531 	ret = -EAGAIN;
532 	/* If truncated, we must retry the operation, we may have raced
533 	 * with the glock demotion code.
534 	 */
535 	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
536 		goto out_trans_end;
537 
538 	/* Unstuff, if required, and allocate backing blocks for page */
539 	ret = 0;
540 	if (gfs2_is_stuffed(ip))
541 		ret = gfs2_unstuff_dinode(ip, page);
542 	if (ret == 0)
543 		ret = gfs2_allocate_page_backing(page, length);
544 
545 out_trans_end:
546 	if (ret)
547 		unlock_page(page);
548 	gfs2_trans_end(sdp);
549 out_trans_fail:
550 	gfs2_inplace_release(ip);
551 out_quota_unlock:
552 	gfs2_quota_unlock(ip);
553 out_unlock:
554 	gfs2_glock_dq(&gh);
555 out_uninit:
556 	gfs2_holder_uninit(&gh);
557 	if (ret == 0) {
558 		set_page_dirty(page);
559 		wait_for_stable_page(page);
560 	}
561 out:
562 	sb_end_pagefault(inode->i_sb);
563 	return block_page_mkwrite_return(ret);
564 }
565 
566 static const struct vm_operations_struct gfs2_vm_ops = {
567 	.fault = filemap_fault,
568 	.map_pages = filemap_map_pages,
569 	.page_mkwrite = gfs2_page_mkwrite,
570 };
571 
572 /**
573  * gfs2_mmap -
574  * @file: The file to map
575  * @vma: The VMA which described the mapping
576  *
577  * There is no need to get a lock here unless we should be updating
578  * atime. We ignore any locking errors since the only consequence is
579  * a missed atime update (which will just be deferred until later).
580  *
581  * Returns: 0
582  */
583 
584 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
585 {
586 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
587 
588 	if (!(file->f_flags & O_NOATIME) &&
589 	    !IS_NOATIME(&ip->i_inode)) {
590 		struct gfs2_holder i_gh;
591 		int error;
592 
593 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
594 					   &i_gh);
595 		if (error)
596 			return error;
597 		/* grab lock to update inode */
598 		gfs2_glock_dq_uninit(&i_gh);
599 		file_accessed(file);
600 	}
601 	vma->vm_ops = &gfs2_vm_ops;
602 
603 	return 0;
604 }
605 
606 /**
607  * gfs2_open_common - This is common to open and atomic_open
608  * @inode: The inode being opened
609  * @file: The file being opened
610  *
611  * This maybe called under a glock or not depending upon how it has
612  * been called. We must always be called under a glock for regular
613  * files, however. For other file types, it does not matter whether
614  * we hold the glock or not.
615  *
616  * Returns: Error code or 0 for success
617  */
618 
619 int gfs2_open_common(struct inode *inode, struct file *file)
620 {
621 	struct gfs2_file *fp;
622 	int ret;
623 
624 	if (S_ISREG(inode->i_mode)) {
625 		ret = generic_file_open(inode, file);
626 		if (ret)
627 			return ret;
628 	}
629 
630 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
631 	if (!fp)
632 		return -ENOMEM;
633 
634 	mutex_init(&fp->f_fl_mutex);
635 
636 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
637 	file->private_data = fp;
638 	return 0;
639 }
640 
641 /**
642  * gfs2_open - open a file
643  * @inode: the inode to open
644  * @file: the struct file for this opening
645  *
646  * After atomic_open, this function is only used for opening files
647  * which are already cached. We must still get the glock for regular
648  * files to ensure that we have the file size uptodate for the large
649  * file check which is in the common code. That is only an issue for
650  * regular files though.
651  *
652  * Returns: errno
653  */
654 
655 static int gfs2_open(struct inode *inode, struct file *file)
656 {
657 	struct gfs2_inode *ip = GFS2_I(inode);
658 	struct gfs2_holder i_gh;
659 	int error;
660 	bool need_unlock = false;
661 
662 	if (S_ISREG(ip->i_inode.i_mode)) {
663 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
664 					   &i_gh);
665 		if (error)
666 			return error;
667 		need_unlock = true;
668 	}
669 
670 	error = gfs2_open_common(inode, file);
671 
672 	if (need_unlock)
673 		gfs2_glock_dq_uninit(&i_gh);
674 
675 	return error;
676 }
677 
678 /**
679  * gfs2_release - called to close a struct file
680  * @inode: the inode the struct file belongs to
681  * @file: the struct file being closed
682  *
683  * Returns: errno
684  */
685 
686 static int gfs2_release(struct inode *inode, struct file *file)
687 {
688 	struct gfs2_inode *ip = GFS2_I(inode);
689 
690 	kfree(file->private_data);
691 	file->private_data = NULL;
692 
693 	if (!(file->f_mode & FMODE_WRITE))
694 		return 0;
695 
696 	gfs2_rsqa_delete(ip, &inode->i_writecount);
697 	return 0;
698 }
699 
700 /**
701  * gfs2_fsync - sync the dirty data for a file (across the cluster)
702  * @file: the file that points to the dentry
703  * @start: the start position in the file to sync
704  * @end: the end position in the file to sync
705  * @datasync: set if we can ignore timestamp changes
706  *
707  * We split the data flushing here so that we don't wait for the data
708  * until after we've also sent the metadata to disk. Note that for
709  * data=ordered, we will write & wait for the data at the log flush
710  * stage anyway, so this is unlikely to make much of a difference
711  * except in the data=writeback case.
712  *
713  * If the fdatawrite fails due to any reason except -EIO, we will
714  * continue the remainder of the fsync, although we'll still report
715  * the error at the end. This is to match filemap_write_and_wait_range()
716  * behaviour.
717  *
718  * Returns: errno
719  */
720 
721 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
722 		      int datasync)
723 {
724 	struct address_space *mapping = file->f_mapping;
725 	struct inode *inode = mapping->host;
726 	int sync_state = inode->i_state & I_DIRTY_ALL;
727 	struct gfs2_inode *ip = GFS2_I(inode);
728 	int ret = 0, ret1 = 0;
729 
730 	if (mapping->nrpages) {
731 		ret1 = filemap_fdatawrite_range(mapping, start, end);
732 		if (ret1 == -EIO)
733 			return ret1;
734 	}
735 
736 	if (!gfs2_is_jdata(ip))
737 		sync_state &= ~I_DIRTY_PAGES;
738 	if (datasync)
739 		sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
740 
741 	if (sync_state) {
742 		ret = sync_inode_metadata(inode, 1);
743 		if (ret)
744 			return ret;
745 		if (gfs2_is_jdata(ip))
746 			ret = file_write_and_wait(file);
747 		if (ret)
748 			return ret;
749 		gfs2_ail_flush(ip->i_gl, 1);
750 	}
751 
752 	if (mapping->nrpages)
753 		ret = file_fdatawait_range(file, start, end);
754 
755 	return ret ? ret : ret1;
756 }
757 
758 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to)
759 {
760 	struct file *file = iocb->ki_filp;
761 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
762 	size_t count = iov_iter_count(to);
763 	struct gfs2_holder gh;
764 	ssize_t ret;
765 
766 	if (!count)
767 		return 0; /* skip atime */
768 
769 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
770 	ret = gfs2_glock_nq(&gh);
771 	if (ret)
772 		goto out_uninit;
773 
774 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
775 			   is_sync_kiocb(iocb));
776 
777 	gfs2_glock_dq(&gh);
778 out_uninit:
779 	gfs2_holder_uninit(&gh);
780 	return ret;
781 }
782 
783 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
784 {
785 	struct file *file = iocb->ki_filp;
786 	struct inode *inode = file->f_mapping->host;
787 	struct gfs2_inode *ip = GFS2_I(inode);
788 	size_t len = iov_iter_count(from);
789 	loff_t offset = iocb->ki_pos;
790 	struct gfs2_holder gh;
791 	ssize_t ret;
792 
793 	/*
794 	 * Deferred lock, even if its a write, since we do no allocation on
795 	 * this path. All we need to change is the atime, and this lock mode
796 	 * ensures that other nodes have flushed their buffered read caches
797 	 * (i.e. their page cache entries for this inode). We do not,
798 	 * unfortunately, have the option of only flushing a range like the
799 	 * VFS does.
800 	 */
801 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
802 	ret = gfs2_glock_nq(&gh);
803 	if (ret)
804 		goto out_uninit;
805 
806 	/* Silently fall back to buffered I/O when writing beyond EOF */
807 	if (offset + len > i_size_read(&ip->i_inode))
808 		goto out;
809 
810 	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
811 			   is_sync_kiocb(iocb));
812 
813 out:
814 	gfs2_glock_dq(&gh);
815 out_uninit:
816 	gfs2_holder_uninit(&gh);
817 	return ret;
818 }
819 
820 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
821 {
822 	ssize_t ret;
823 
824 	if (iocb->ki_flags & IOCB_DIRECT) {
825 		ret = gfs2_file_direct_read(iocb, to);
826 		if (likely(ret != -ENOTBLK))
827 			return ret;
828 		iocb->ki_flags &= ~IOCB_DIRECT;
829 	}
830 	return generic_file_read_iter(iocb, to);
831 }
832 
833 /**
834  * gfs2_file_write_iter - Perform a write to a file
835  * @iocb: The io context
836  * @from: The data to write
837  *
838  * We have to do a lock/unlock here to refresh the inode size for
839  * O_APPEND writes, otherwise we can land up writing at the wrong
840  * offset. There is still a race, but provided the app is using its
841  * own file locking, this will make O_APPEND work as expected.
842  *
843  */
844 
845 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
846 {
847 	struct file *file = iocb->ki_filp;
848 	struct inode *inode = file_inode(file);
849 	struct gfs2_inode *ip = GFS2_I(inode);
850 	ssize_t written = 0, ret;
851 
852 	ret = gfs2_rsqa_alloc(ip);
853 	if (ret)
854 		return ret;
855 
856 	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
857 
858 	if (iocb->ki_flags & IOCB_APPEND) {
859 		struct gfs2_holder gh;
860 
861 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
862 		if (ret)
863 			return ret;
864 		gfs2_glock_dq_uninit(&gh);
865 	}
866 
867 	inode_lock(inode);
868 	ret = generic_write_checks(iocb, from);
869 	if (ret <= 0)
870 		goto out;
871 
872 	/* We can write back this queue in page reclaim */
873 	current->backing_dev_info = inode_to_bdi(inode);
874 
875 	ret = file_remove_privs(file);
876 	if (ret)
877 		goto out2;
878 
879 	ret = file_update_time(file);
880 	if (ret)
881 		goto out2;
882 
883 	if (iocb->ki_flags & IOCB_DIRECT) {
884 		struct address_space *mapping = file->f_mapping;
885 		loff_t pos, endbyte;
886 		ssize_t buffered;
887 
888 		written = gfs2_file_direct_write(iocb, from);
889 		if (written < 0 || !iov_iter_count(from))
890 			goto out2;
891 
892 		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
893 		if (unlikely(ret < 0))
894 			goto out2;
895 		buffered = ret;
896 
897 		/*
898 		 * We need to ensure that the page cache pages are written to
899 		 * disk and invalidated to preserve the expected O_DIRECT
900 		 * semantics.
901 		 */
902 		pos = iocb->ki_pos;
903 		endbyte = pos + buffered - 1;
904 		ret = filemap_write_and_wait_range(mapping, pos, endbyte);
905 		if (!ret) {
906 			iocb->ki_pos += buffered;
907 			written += buffered;
908 			invalidate_mapping_pages(mapping,
909 						 pos >> PAGE_SHIFT,
910 						 endbyte >> PAGE_SHIFT);
911 		} else {
912 			/*
913 			 * We don't know how much we wrote, so just return
914 			 * the number of bytes which were direct-written
915 			 */
916 		}
917 	} else {
918 		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
919 		if (likely(ret > 0))
920 			iocb->ki_pos += ret;
921 	}
922 
923 out2:
924 	current->backing_dev_info = NULL;
925 out:
926 	inode_unlock(inode);
927 	if (likely(ret > 0)) {
928 		/* Handle various SYNC-type writes */
929 		ret = generic_write_sync(iocb, ret);
930 	}
931 	return written ? written : ret;
932 }
933 
934 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
935 			   int mode)
936 {
937 	struct super_block *sb = inode->i_sb;
938 	struct gfs2_inode *ip = GFS2_I(inode);
939 	loff_t end = offset + len;
940 	struct buffer_head *dibh;
941 	int error;
942 
943 	error = gfs2_meta_inode_buffer(ip, &dibh);
944 	if (unlikely(error))
945 		return error;
946 
947 	gfs2_trans_add_meta(ip->i_gl, dibh);
948 
949 	if (gfs2_is_stuffed(ip)) {
950 		error = gfs2_unstuff_dinode(ip, NULL);
951 		if (unlikely(error))
952 			goto out;
953 	}
954 
955 	while (offset < end) {
956 		struct iomap iomap = { };
957 
958 		error = gfs2_iomap_get_alloc(inode, offset, end - offset,
959 					     &iomap);
960 		if (error)
961 			goto out;
962 		offset = iomap.offset + iomap.length;
963 		if (!(iomap.flags & IOMAP_F_NEW))
964 			continue;
965 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
966 					 iomap.length >> inode->i_blkbits,
967 					 GFP_NOFS);
968 		if (error) {
969 			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
970 			goto out;
971 		}
972 	}
973 out:
974 	brelse(dibh);
975 	return error;
976 }
977 
978 /**
979  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
980  *                     blocks, determine how many bytes can be written.
981  * @ip:          The inode in question.
982  * @len:         Max cap of bytes. What we return in *len must be <= this.
983  * @data_blocks: Compute and return the number of data blocks needed
984  * @ind_blocks:  Compute and return the number of indirect blocks needed
985  * @max_blocks:  The total blocks available to work with.
986  *
987  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
988  */
989 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
990 			    unsigned int *data_blocks, unsigned int *ind_blocks,
991 			    unsigned int max_blocks)
992 {
993 	loff_t max = *len;
994 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
995 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
996 
997 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
998 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
999 		max_data -= tmp;
1000 	}
1001 
1002 	*data_blocks = max_data;
1003 	*ind_blocks = max_blocks - max_data;
1004 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1005 	if (*len > max) {
1006 		*len = max;
1007 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1008 	}
1009 }
1010 
1011 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1012 {
1013 	struct inode *inode = file_inode(file);
1014 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1015 	struct gfs2_inode *ip = GFS2_I(inode);
1016 	struct gfs2_alloc_parms ap = { .aflags = 0, };
1017 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1018 	loff_t bytes, max_bytes, max_blks;
1019 	int error;
1020 	const loff_t pos = offset;
1021 	const loff_t count = len;
1022 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1023 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1024 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1025 
1026 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1027 
1028 	offset &= bsize_mask;
1029 
1030 	len = next - offset;
1031 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1032 	if (!bytes)
1033 		bytes = UINT_MAX;
1034 	bytes &= bsize_mask;
1035 	if (bytes == 0)
1036 		bytes = sdp->sd_sb.sb_bsize;
1037 
1038 	gfs2_size_hint(file, offset, len);
1039 
1040 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1041 	ap.min_target = data_blocks + ind_blocks;
1042 
1043 	while (len > 0) {
1044 		if (len < bytes)
1045 			bytes = len;
1046 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1047 			len -= bytes;
1048 			offset += bytes;
1049 			continue;
1050 		}
1051 
1052 		/* We need to determine how many bytes we can actually
1053 		 * fallocate without exceeding quota or going over the
1054 		 * end of the fs. We start off optimistically by assuming
1055 		 * we can write max_bytes */
1056 		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1057 
1058 		/* Since max_bytes is most likely a theoretical max, we
1059 		 * calculate a more realistic 'bytes' to serve as a good
1060 		 * starting point for the number of bytes we may be able
1061 		 * to write */
1062 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1063 		ap.target = data_blocks + ind_blocks;
1064 
1065 		error = gfs2_quota_lock_check(ip, &ap);
1066 		if (error)
1067 			return error;
1068 		/* ap.allowed tells us how many blocks quota will allow
1069 		 * us to write. Check if this reduces max_blks */
1070 		max_blks = UINT_MAX;
1071 		if (ap.allowed)
1072 			max_blks = ap.allowed;
1073 
1074 		error = gfs2_inplace_reserve(ip, &ap);
1075 		if (error)
1076 			goto out_qunlock;
1077 
1078 		/* check if the selected rgrp limits our max_blks further */
1079 		if (ap.allowed && ap.allowed < max_blks)
1080 			max_blks = ap.allowed;
1081 
1082 		/* Almost done. Calculate bytes that can be written using
1083 		 * max_blks. We also recompute max_bytes, data_blocks and
1084 		 * ind_blocks */
1085 		calc_max_reserv(ip, &max_bytes, &data_blocks,
1086 				&ind_blocks, max_blks);
1087 
1088 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1089 			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1090 		if (gfs2_is_jdata(ip))
1091 			rblocks += data_blocks ? data_blocks : 1;
1092 
1093 		error = gfs2_trans_begin(sdp, rblocks,
1094 					 PAGE_SIZE >> inode->i_blkbits);
1095 		if (error)
1096 			goto out_trans_fail;
1097 
1098 		error = fallocate_chunk(inode, offset, max_bytes, mode);
1099 		gfs2_trans_end(sdp);
1100 
1101 		if (error)
1102 			goto out_trans_fail;
1103 
1104 		len -= max_bytes;
1105 		offset += max_bytes;
1106 		gfs2_inplace_release(ip);
1107 		gfs2_quota_unlock(ip);
1108 	}
1109 
1110 	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1111 		i_size_write(inode, pos + count);
1112 	file_update_time(file);
1113 	mark_inode_dirty(inode);
1114 
1115 	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1116 		return vfs_fsync_range(file, pos, pos + count - 1,
1117 			       (file->f_flags & __O_SYNC) ? 0 : 1);
1118 	return 0;
1119 
1120 out_trans_fail:
1121 	gfs2_inplace_release(ip);
1122 out_qunlock:
1123 	gfs2_quota_unlock(ip);
1124 	return error;
1125 }
1126 
1127 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1128 {
1129 	struct inode *inode = file_inode(file);
1130 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1131 	struct gfs2_inode *ip = GFS2_I(inode);
1132 	struct gfs2_holder gh;
1133 	int ret;
1134 
1135 	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1136 		return -EOPNOTSUPP;
1137 	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1138 	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1139 		return -EOPNOTSUPP;
1140 
1141 	inode_lock(inode);
1142 
1143 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1144 	ret = gfs2_glock_nq(&gh);
1145 	if (ret)
1146 		goto out_uninit;
1147 
1148 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1149 	    (offset + len) > inode->i_size) {
1150 		ret = inode_newsize_ok(inode, offset + len);
1151 		if (ret)
1152 			goto out_unlock;
1153 	}
1154 
1155 	ret = get_write_access(inode);
1156 	if (ret)
1157 		goto out_unlock;
1158 
1159 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1160 		ret = __gfs2_punch_hole(file, offset, len);
1161 	} else {
1162 		ret = gfs2_rsqa_alloc(ip);
1163 		if (ret)
1164 			goto out_putw;
1165 
1166 		ret = __gfs2_fallocate(file, mode, offset, len);
1167 
1168 		if (ret)
1169 			gfs2_rs_deltree(&ip->i_res);
1170 	}
1171 
1172 out_putw:
1173 	put_write_access(inode);
1174 out_unlock:
1175 	gfs2_glock_dq(&gh);
1176 out_uninit:
1177 	gfs2_holder_uninit(&gh);
1178 	inode_unlock(inode);
1179 	return ret;
1180 }
1181 
1182 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1183 				      struct file *out, loff_t *ppos,
1184 				      size_t len, unsigned int flags)
1185 {
1186 	int error;
1187 	struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
1188 
1189 	error = gfs2_rsqa_alloc(ip);
1190 	if (error)
1191 		return (ssize_t)error;
1192 
1193 	gfs2_size_hint(out, *ppos, len);
1194 
1195 	return iter_file_splice_write(pipe, out, ppos, len, flags);
1196 }
1197 
1198 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1199 
1200 /**
1201  * gfs2_lock - acquire/release a posix lock on a file
1202  * @file: the file pointer
1203  * @cmd: either modify or retrieve lock state, possibly wait
1204  * @fl: type and range of lock
1205  *
1206  * Returns: errno
1207  */
1208 
1209 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1210 {
1211 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1212 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1213 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1214 
1215 	if (!(fl->fl_flags & FL_POSIX))
1216 		return -ENOLCK;
1217 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1218 		return -ENOLCK;
1219 
1220 	if (cmd == F_CANCELLK) {
1221 		/* Hack: */
1222 		cmd = F_SETLK;
1223 		fl->fl_type = F_UNLCK;
1224 	}
1225 	if (unlikely(gfs2_withdrawn(sdp))) {
1226 		if (fl->fl_type == F_UNLCK)
1227 			locks_lock_file_wait(file, fl);
1228 		return -EIO;
1229 	}
1230 	if (IS_GETLK(cmd))
1231 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1232 	else if (fl->fl_type == F_UNLCK)
1233 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1234 	else
1235 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1236 }
1237 
1238 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1239 {
1240 	struct gfs2_file *fp = file->private_data;
1241 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1242 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1243 	struct gfs2_glock *gl;
1244 	unsigned int state;
1245 	u16 flags;
1246 	int error = 0;
1247 	int sleeptime;
1248 
1249 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1250 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1251 
1252 	mutex_lock(&fp->f_fl_mutex);
1253 
1254 	if (gfs2_holder_initialized(fl_gh)) {
1255 		struct file_lock request;
1256 		if (fl_gh->gh_state == state)
1257 			goto out;
1258 		locks_init_lock(&request);
1259 		request.fl_type = F_UNLCK;
1260 		request.fl_flags = FL_FLOCK;
1261 		locks_lock_file_wait(file, &request);
1262 		gfs2_glock_dq(fl_gh);
1263 		gfs2_holder_reinit(state, flags, fl_gh);
1264 	} else {
1265 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1266 				       &gfs2_flock_glops, CREATE, &gl);
1267 		if (error)
1268 			goto out;
1269 		gfs2_holder_init(gl, state, flags, fl_gh);
1270 		gfs2_glock_put(gl);
1271 	}
1272 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1273 		error = gfs2_glock_nq(fl_gh);
1274 		if (error != GLR_TRYFAILED)
1275 			break;
1276 		fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1277 		fl_gh->gh_error = 0;
1278 		msleep(sleeptime);
1279 	}
1280 	if (error) {
1281 		gfs2_holder_uninit(fl_gh);
1282 		if (error == GLR_TRYFAILED)
1283 			error = -EAGAIN;
1284 	} else {
1285 		error = locks_lock_file_wait(file, fl);
1286 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1287 	}
1288 
1289 out:
1290 	mutex_unlock(&fp->f_fl_mutex);
1291 	return error;
1292 }
1293 
1294 static void do_unflock(struct file *file, struct file_lock *fl)
1295 {
1296 	struct gfs2_file *fp = file->private_data;
1297 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1298 
1299 	mutex_lock(&fp->f_fl_mutex);
1300 	locks_lock_file_wait(file, fl);
1301 	if (gfs2_holder_initialized(fl_gh)) {
1302 		gfs2_glock_dq(fl_gh);
1303 		gfs2_holder_uninit(fl_gh);
1304 	}
1305 	mutex_unlock(&fp->f_fl_mutex);
1306 }
1307 
1308 /**
1309  * gfs2_flock - acquire/release a flock lock on a file
1310  * @file: the file pointer
1311  * @cmd: either modify or retrieve lock state, possibly wait
1312  * @fl: type and range of lock
1313  *
1314  * Returns: errno
1315  */
1316 
1317 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1318 {
1319 	if (!(fl->fl_flags & FL_FLOCK))
1320 		return -ENOLCK;
1321 	if (fl->fl_type & LOCK_MAND)
1322 		return -EOPNOTSUPP;
1323 
1324 	if (fl->fl_type == F_UNLCK) {
1325 		do_unflock(file, fl);
1326 		return 0;
1327 	} else {
1328 		return do_flock(file, cmd, fl);
1329 	}
1330 }
1331 
1332 const struct file_operations gfs2_file_fops = {
1333 	.llseek		= gfs2_llseek,
1334 	.read_iter	= gfs2_file_read_iter,
1335 	.write_iter	= gfs2_file_write_iter,
1336 	.iopoll		= iomap_dio_iopoll,
1337 	.unlocked_ioctl	= gfs2_ioctl,
1338 	.compat_ioctl	= gfs2_compat_ioctl,
1339 	.mmap		= gfs2_mmap,
1340 	.open		= gfs2_open,
1341 	.release	= gfs2_release,
1342 	.fsync		= gfs2_fsync,
1343 	.lock		= gfs2_lock,
1344 	.flock		= gfs2_flock,
1345 	.splice_read	= generic_file_splice_read,
1346 	.splice_write	= gfs2_file_splice_write,
1347 	.setlease	= simple_nosetlease,
1348 	.fallocate	= gfs2_fallocate,
1349 };
1350 
1351 const struct file_operations gfs2_dir_fops = {
1352 	.iterate_shared	= gfs2_readdir,
1353 	.unlocked_ioctl	= gfs2_ioctl,
1354 	.compat_ioctl	= gfs2_compat_ioctl,
1355 	.open		= gfs2_open,
1356 	.release	= gfs2_release,
1357 	.fsync		= gfs2_fsync,
1358 	.lock		= gfs2_lock,
1359 	.flock		= gfs2_flock,
1360 	.llseek		= default_llseek,
1361 };
1362 
1363 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1364 
1365 const struct file_operations gfs2_file_fops_nolock = {
1366 	.llseek		= gfs2_llseek,
1367 	.read_iter	= gfs2_file_read_iter,
1368 	.write_iter	= gfs2_file_write_iter,
1369 	.iopoll		= iomap_dio_iopoll,
1370 	.unlocked_ioctl	= gfs2_ioctl,
1371 	.compat_ioctl	= gfs2_compat_ioctl,
1372 	.mmap		= gfs2_mmap,
1373 	.open		= gfs2_open,
1374 	.release	= gfs2_release,
1375 	.fsync		= gfs2_fsync,
1376 	.splice_read	= generic_file_splice_read,
1377 	.splice_write	= gfs2_file_splice_write,
1378 	.setlease	= generic_setlease,
1379 	.fallocate	= gfs2_fallocate,
1380 };
1381 
1382 const struct file_operations gfs2_dir_fops_nolock = {
1383 	.iterate_shared	= gfs2_readdir,
1384 	.unlocked_ioctl	= gfs2_ioctl,
1385 	.compat_ioctl	= gfs2_compat_ioctl,
1386 	.open		= gfs2_open,
1387 	.release	= gfs2_release,
1388 	.fsync		= gfs2_fsync,
1389 	.llseek		= default_llseek,
1390 };
1391 
1392