xref: /openbmc/linux/fs/gfs2/file.c (revision c4f7ac64)
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 #include <linux/fileattr.h>
29 
30 #include "gfs2.h"
31 #include "incore.h"
32 #include "bmap.h"
33 #include "aops.h"
34 #include "dir.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "log.h"
39 #include "meta_io.h"
40 #include "quota.h"
41 #include "rgrp.h"
42 #include "trans.h"
43 #include "util.h"
44 
45 /**
46  * gfs2_llseek - seek to a location in a file
47  * @file: the file
48  * @offset: the offset
49  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
50  *
51  * SEEK_END requires the glock for the file because it references the
52  * file's size.
53  *
54  * Returns: The new offset, or errno
55  */
56 
57 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
58 {
59 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
60 	struct gfs2_holder i_gh;
61 	loff_t error;
62 
63 	switch (whence) {
64 	case SEEK_END:
65 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
66 					   &i_gh);
67 		if (!error) {
68 			error = generic_file_llseek(file, offset, whence);
69 			gfs2_glock_dq_uninit(&i_gh);
70 		}
71 		break;
72 
73 	case SEEK_DATA:
74 		error = gfs2_seek_data(file, offset);
75 		break;
76 
77 	case SEEK_HOLE:
78 		error = gfs2_seek_hole(file, offset);
79 		break;
80 
81 	case SEEK_CUR:
82 	case SEEK_SET:
83 		/*
84 		 * These don't reference inode->i_size and don't depend on the
85 		 * block mapping, so we don't need the glock.
86 		 */
87 		error = generic_file_llseek(file, offset, whence);
88 		break;
89 	default:
90 		error = -EINVAL;
91 	}
92 
93 	return error;
94 }
95 
96 /**
97  * gfs2_readdir - Iterator for a directory
98  * @file: The directory to read from
99  * @ctx: What to feed directory entries to
100  *
101  * Returns: errno
102  */
103 
104 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
105 {
106 	struct inode *dir = file->f_mapping->host;
107 	struct gfs2_inode *dip = GFS2_I(dir);
108 	struct gfs2_holder d_gh;
109 	int error;
110 
111 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
112 	if (error)
113 		return error;
114 
115 	error = gfs2_dir_read(dir, ctx, &file->f_ra);
116 
117 	gfs2_glock_dq_uninit(&d_gh);
118 
119 	return error;
120 }
121 
122 /*
123  * struct fsflag_gfs2flag
124  *
125  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126  * and to GFS2_DIF_JDATA for non-directories.
127  */
128 static struct {
129 	u32 fsflag;
130 	u32 gfsflag;
131 } fsflag_gfs2flag[] = {
132 	{FS_SYNC_FL, GFS2_DIF_SYNC},
133 	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
134 	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
135 	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
136 	{FS_INDEX_FL, GFS2_DIF_EXHASH},
137 	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
138 	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
139 };
140 
141 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
142 {
143 	int i;
144 	u32 fsflags = 0;
145 
146 	if (S_ISDIR(inode->i_mode))
147 		gfsflags &= ~GFS2_DIF_JDATA;
148 	else
149 		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
150 
151 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
152 		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
153 			fsflags |= fsflag_gfs2flag[i].fsflag;
154 	return fsflags;
155 }
156 
157 int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
158 {
159 	struct inode *inode = d_inode(dentry);
160 	struct gfs2_inode *ip = GFS2_I(inode);
161 	struct gfs2_holder gh;
162 	int error;
163 	u32 fsflags;
164 
165 	if (d_is_special(dentry))
166 		return -ENOTTY;
167 
168 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
169 	error = gfs2_glock_nq(&gh);
170 	if (error)
171 		goto out_uninit;
172 
173 	fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
174 
175 	fileattr_fill_flags(fa, fsflags);
176 
177 	gfs2_glock_dq(&gh);
178 out_uninit:
179 	gfs2_holder_uninit(&gh);
180 	return error;
181 }
182 
183 void gfs2_set_inode_flags(struct inode *inode)
184 {
185 	struct gfs2_inode *ip = GFS2_I(inode);
186 	unsigned int flags = inode->i_flags;
187 
188 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
189 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
190 		flags |= S_NOSEC;
191 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
192 		flags |= S_IMMUTABLE;
193 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
194 		flags |= S_APPEND;
195 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
196 		flags |= S_NOATIME;
197 	if (ip->i_diskflags & GFS2_DIF_SYNC)
198 		flags |= S_SYNC;
199 	inode->i_flags = flags;
200 }
201 
202 /* Flags that can be set by user space */
203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
204 			     GFS2_DIF_IMMUTABLE|		\
205 			     GFS2_DIF_APPENDONLY|		\
206 			     GFS2_DIF_NOATIME|			\
207 			     GFS2_DIF_SYNC|			\
208 			     GFS2_DIF_TOPDIR|			\
209 			     GFS2_DIF_INHERIT_JDATA)
210 
211 /**
212  * do_gfs2_set_flags - set flags on an inode
213  * @filp: file pointer
214  * @reqflags: The flags to set
215  * @mask: Indicates which flags are valid
216  * @fsflags: The FS_* inode flags passed in
217  *
218  */
219 static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask,
220 			     const u32 fsflags)
221 {
222 	struct gfs2_inode *ip = GFS2_I(inode);
223 	struct gfs2_sbd *sdp = GFS2_SB(inode);
224 	struct buffer_head *bh;
225 	struct gfs2_holder gh;
226 	int error;
227 	u32 new_flags, flags;
228 
229 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
230 	if (error)
231 		return error;
232 
233 	error = 0;
234 	flags = ip->i_diskflags;
235 	new_flags = (flags & ~mask) | (reqflags & mask);
236 	if ((new_flags ^ flags) == 0)
237 		goto out;
238 
239 	error = -EPERM;
240 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
241 		goto out;
242 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
243 		goto out;
244 	if (!IS_IMMUTABLE(inode)) {
245 		error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
246 		if (error)
247 			goto out;
248 	}
249 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
250 		if (new_flags & GFS2_DIF_JDATA)
251 			gfs2_log_flush(sdp, ip->i_gl,
252 				       GFS2_LOG_HEAD_FLUSH_NORMAL |
253 				       GFS2_LFC_SET_FLAGS);
254 		error = filemap_fdatawrite(inode->i_mapping);
255 		if (error)
256 			goto out;
257 		error = filemap_fdatawait(inode->i_mapping);
258 		if (error)
259 			goto out;
260 		if (new_flags & GFS2_DIF_JDATA)
261 			gfs2_ordered_del_inode(ip);
262 	}
263 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
264 	if (error)
265 		goto out;
266 	error = gfs2_meta_inode_buffer(ip, &bh);
267 	if (error)
268 		goto out_trans_end;
269 	inode->i_ctime = current_time(inode);
270 	gfs2_trans_add_meta(ip->i_gl, bh);
271 	ip->i_diskflags = new_flags;
272 	gfs2_dinode_out(ip, bh->b_data);
273 	brelse(bh);
274 	gfs2_set_inode_flags(inode);
275 	gfs2_set_aops(inode);
276 out_trans_end:
277 	gfs2_trans_end(sdp);
278 out:
279 	gfs2_glock_dq_uninit(&gh);
280 	return error;
281 }
282 
283 int gfs2_fileattr_set(struct user_namespace *mnt_userns,
284 		      struct dentry *dentry, struct fileattr *fa)
285 {
286 	struct inode *inode = d_inode(dentry);
287 	u32 fsflags = fa->flags, gfsflags = 0;
288 	u32 mask;
289 	int i;
290 
291 	if (d_is_special(dentry))
292 		return -ENOTTY;
293 
294 	if (fileattr_has_fsx(fa))
295 		return -EOPNOTSUPP;
296 
297 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
298 		if (fsflags & fsflag_gfs2flag[i].fsflag) {
299 			fsflags &= ~fsflag_gfs2flag[i].fsflag;
300 			gfsflags |= fsflag_gfs2flag[i].gfsflag;
301 		}
302 	}
303 	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
304 		return -EINVAL;
305 
306 	mask = GFS2_FLAGS_USER_SET;
307 	if (S_ISDIR(inode->i_mode)) {
308 		mask &= ~GFS2_DIF_JDATA;
309 	} else {
310 		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
311 		if (gfsflags & GFS2_DIF_TOPDIR)
312 			return -EINVAL;
313 		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
314 	}
315 
316 	return do_gfs2_set_flags(inode, gfsflags, mask, fsflags);
317 }
318 
319 static int gfs2_getlabel(struct file *filp, char __user *label)
320 {
321 	struct inode *inode = file_inode(filp);
322 	struct gfs2_sbd *sdp = GFS2_SB(inode);
323 
324 	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
325 		return -EFAULT;
326 
327 	return 0;
328 }
329 
330 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
331 {
332 	switch(cmd) {
333 	case FITRIM:
334 		return gfs2_fitrim(filp, (void __user *)arg);
335 	case FS_IOC_GETFSLABEL:
336 		return gfs2_getlabel(filp, (char __user *)arg);
337 	}
338 
339 	return -ENOTTY;
340 }
341 
342 #ifdef CONFIG_COMPAT
343 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
344 {
345 	switch(cmd) {
346 	/* Keep this list in sync with gfs2_ioctl */
347 	case FITRIM:
348 	case FS_IOC_GETFSLABEL:
349 		break;
350 	default:
351 		return -ENOIOCTLCMD;
352 	}
353 
354 	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
355 }
356 #else
357 #define gfs2_compat_ioctl NULL
358 #endif
359 
360 /**
361  * gfs2_size_hint - Give a hint to the size of a write request
362  * @filep: The struct file
363  * @offset: The file offset of the write
364  * @size: The length of the write
365  *
366  * When we are about to do a write, this function records the total
367  * write size in order to provide a suitable hint to the lower layers
368  * about how many blocks will be required.
369  *
370  */
371 
372 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
373 {
374 	struct inode *inode = file_inode(filep);
375 	struct gfs2_sbd *sdp = GFS2_SB(inode);
376 	struct gfs2_inode *ip = GFS2_I(inode);
377 	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
378 	int hint = min_t(size_t, INT_MAX, blks);
379 
380 	if (hint > atomic_read(&ip->i_sizehint))
381 		atomic_set(&ip->i_sizehint, hint);
382 }
383 
384 /**
385  * gfs2_allocate_page_backing - Allocate blocks for a write fault
386  * @page: The (locked) page to allocate backing for
387  * @length: Size of the allocation
388  *
389  * We try to allocate all the blocks required for the page in one go.  This
390  * might fail for various reasons, so we keep trying until all the blocks to
391  * back this page are allocated.  If some of the blocks are already allocated,
392  * that is ok too.
393  */
394 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
395 {
396 	u64 pos = page_offset(page);
397 
398 	do {
399 		struct iomap iomap = { };
400 
401 		if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
402 			return -EIO;
403 
404 		if (length < iomap.length)
405 			iomap.length = length;
406 		length -= iomap.length;
407 		pos += iomap.length;
408 	} while (length > 0);
409 
410 	return 0;
411 }
412 
413 /**
414  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
415  * @vmf: The virtual memory fault containing the page to become writable
416  *
417  * When the page becomes writable, we need to ensure that we have
418  * blocks allocated on disk to back that page.
419  */
420 
421 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
422 {
423 	struct page *page = vmf->page;
424 	struct inode *inode = file_inode(vmf->vma->vm_file);
425 	struct gfs2_inode *ip = GFS2_I(inode);
426 	struct gfs2_sbd *sdp = GFS2_SB(inode);
427 	struct gfs2_alloc_parms ap = { .aflags = 0, };
428 	u64 offset = page_offset(page);
429 	unsigned int data_blocks, ind_blocks, rblocks;
430 	struct gfs2_holder gh;
431 	unsigned int length;
432 	loff_t size;
433 	int ret;
434 
435 	sb_start_pagefault(inode->i_sb);
436 
437 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
438 	ret = gfs2_glock_nq(&gh);
439 	if (ret)
440 		goto out_uninit;
441 
442 	/* Check page index against inode size */
443 	size = i_size_read(inode);
444 	if (offset >= size) {
445 		ret = -EINVAL;
446 		goto out_unlock;
447 	}
448 
449 	/* Update file times before taking page lock */
450 	file_update_time(vmf->vma->vm_file);
451 
452 	/* page is wholly or partially inside EOF */
453 	if (offset > size - PAGE_SIZE)
454 		length = offset_in_page(size);
455 	else
456 		length = PAGE_SIZE;
457 
458 	gfs2_size_hint(vmf->vma->vm_file, offset, length);
459 
460 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
461 	set_bit(GIF_SW_PAGED, &ip->i_flags);
462 
463 	/*
464 	 * iomap_writepage / iomap_writepages currently don't support inline
465 	 * files, so always unstuff here.
466 	 */
467 
468 	if (!gfs2_is_stuffed(ip) &&
469 	    !gfs2_write_alloc_required(ip, offset, length)) {
470 		lock_page(page);
471 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
472 			ret = -EAGAIN;
473 			unlock_page(page);
474 		}
475 		goto out_unlock;
476 	}
477 
478 	ret = gfs2_rindex_update(sdp);
479 	if (ret)
480 		goto out_unlock;
481 
482 	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
483 	ap.target = data_blocks + ind_blocks;
484 	ret = gfs2_quota_lock_check(ip, &ap);
485 	if (ret)
486 		goto out_unlock;
487 	ret = gfs2_inplace_reserve(ip, &ap);
488 	if (ret)
489 		goto out_quota_unlock;
490 
491 	rblocks = RES_DINODE + ind_blocks;
492 	if (gfs2_is_jdata(ip))
493 		rblocks += data_blocks ? data_blocks : 1;
494 	if (ind_blocks || data_blocks) {
495 		rblocks += RES_STATFS + RES_QUOTA;
496 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
497 	}
498 	ret = gfs2_trans_begin(sdp, rblocks, 0);
499 	if (ret)
500 		goto out_trans_fail;
501 
502 	lock_page(page);
503 	ret = -EAGAIN;
504 	/* If truncated, we must retry the operation, we may have raced
505 	 * with the glock demotion code.
506 	 */
507 	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
508 		goto out_trans_end;
509 
510 	/* Unstuff, if required, and allocate backing blocks for page */
511 	ret = 0;
512 	if (gfs2_is_stuffed(ip))
513 		ret = gfs2_unstuff_dinode(ip, page);
514 	if (ret == 0)
515 		ret = gfs2_allocate_page_backing(page, length);
516 
517 out_trans_end:
518 	if (ret)
519 		unlock_page(page);
520 	gfs2_trans_end(sdp);
521 out_trans_fail:
522 	gfs2_inplace_release(ip);
523 out_quota_unlock:
524 	gfs2_quota_unlock(ip);
525 out_unlock:
526 	gfs2_glock_dq(&gh);
527 out_uninit:
528 	gfs2_holder_uninit(&gh);
529 	if (ret == 0) {
530 		set_page_dirty(page);
531 		wait_for_stable_page(page);
532 	}
533 	sb_end_pagefault(inode->i_sb);
534 	return block_page_mkwrite_return(ret);
535 }
536 
537 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
538 {
539 	struct inode *inode = file_inode(vmf->vma->vm_file);
540 	struct gfs2_inode *ip = GFS2_I(inode);
541 	struct gfs2_holder gh;
542 	vm_fault_t ret;
543 	int err;
544 
545 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
546 	err = gfs2_glock_nq(&gh);
547 	if (err) {
548 		ret = block_page_mkwrite_return(err);
549 		goto out_uninit;
550 	}
551 	ret = filemap_fault(vmf);
552 	gfs2_glock_dq(&gh);
553 out_uninit:
554 	gfs2_holder_uninit(&gh);
555 	return ret;
556 }
557 
558 static const struct vm_operations_struct gfs2_vm_ops = {
559 	.fault = gfs2_fault,
560 	.map_pages = filemap_map_pages,
561 	.page_mkwrite = gfs2_page_mkwrite,
562 };
563 
564 /**
565  * gfs2_mmap
566  * @file: The file to map
567  * @vma: The VMA which described the mapping
568  *
569  * There is no need to get a lock here unless we should be updating
570  * atime. We ignore any locking errors since the only consequence is
571  * a missed atime update (which will just be deferred until later).
572  *
573  * Returns: 0
574  */
575 
576 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
577 {
578 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
579 
580 	if (!(file->f_flags & O_NOATIME) &&
581 	    !IS_NOATIME(&ip->i_inode)) {
582 		struct gfs2_holder i_gh;
583 		int error;
584 
585 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
586 					   &i_gh);
587 		if (error)
588 			return error;
589 		/* grab lock to update inode */
590 		gfs2_glock_dq_uninit(&i_gh);
591 		file_accessed(file);
592 	}
593 	vma->vm_ops = &gfs2_vm_ops;
594 
595 	return 0;
596 }
597 
598 /**
599  * gfs2_open_common - This is common to open and atomic_open
600  * @inode: The inode being opened
601  * @file: The file being opened
602  *
603  * This maybe called under a glock or not depending upon how it has
604  * been called. We must always be called under a glock for regular
605  * files, however. For other file types, it does not matter whether
606  * we hold the glock or not.
607  *
608  * Returns: Error code or 0 for success
609  */
610 
611 int gfs2_open_common(struct inode *inode, struct file *file)
612 {
613 	struct gfs2_file *fp;
614 	int ret;
615 
616 	if (S_ISREG(inode->i_mode)) {
617 		ret = generic_file_open(inode, file);
618 		if (ret)
619 			return ret;
620 	}
621 
622 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
623 	if (!fp)
624 		return -ENOMEM;
625 
626 	mutex_init(&fp->f_fl_mutex);
627 
628 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
629 	file->private_data = fp;
630 	if (file->f_mode & FMODE_WRITE) {
631 		ret = gfs2_qa_get(GFS2_I(inode));
632 		if (ret)
633 			goto fail;
634 	}
635 	return 0;
636 
637 fail:
638 	kfree(file->private_data);
639 	file->private_data = NULL;
640 	return ret;
641 }
642 
643 /**
644  * gfs2_open - open a file
645  * @inode: the inode to open
646  * @file: the struct file for this opening
647  *
648  * After atomic_open, this function is only used for opening files
649  * which are already cached. We must still get the glock for regular
650  * files to ensure that we have the file size uptodate for the large
651  * file check which is in the common code. That is only an issue for
652  * regular files though.
653  *
654  * Returns: errno
655  */
656 
657 static int gfs2_open(struct inode *inode, struct file *file)
658 {
659 	struct gfs2_inode *ip = GFS2_I(inode);
660 	struct gfs2_holder i_gh;
661 	int error;
662 	bool need_unlock = false;
663 
664 	if (S_ISREG(ip->i_inode.i_mode)) {
665 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
666 					   &i_gh);
667 		if (error)
668 			return error;
669 		need_unlock = true;
670 	}
671 
672 	error = gfs2_open_common(inode, file);
673 
674 	if (need_unlock)
675 		gfs2_glock_dq_uninit(&i_gh);
676 
677 	return error;
678 }
679 
680 /**
681  * gfs2_release - called to close a struct file
682  * @inode: the inode the struct file belongs to
683  * @file: the struct file being closed
684  *
685  * Returns: errno
686  */
687 
688 static int gfs2_release(struct inode *inode, struct file *file)
689 {
690 	struct gfs2_inode *ip = GFS2_I(inode);
691 
692 	kfree(file->private_data);
693 	file->private_data = NULL;
694 
695 	if (gfs2_rs_active(&ip->i_res))
696 		gfs2_rs_delete(ip, &inode->i_writecount);
697 	if (file->f_mode & FMODE_WRITE)
698 		gfs2_qa_put(ip);
699 	return 0;
700 }
701 
702 /**
703  * gfs2_fsync - sync the dirty data for a file (across the cluster)
704  * @file: the file that points to the dentry
705  * @start: the start position in the file to sync
706  * @end: the end position in the file to sync
707  * @datasync: set if we can ignore timestamp changes
708  *
709  * We split the data flushing here so that we don't wait for the data
710  * until after we've also sent the metadata to disk. Note that for
711  * data=ordered, we will write & wait for the data at the log flush
712  * stage anyway, so this is unlikely to make much of a difference
713  * except in the data=writeback case.
714  *
715  * If the fdatawrite fails due to any reason except -EIO, we will
716  * continue the remainder of the fsync, although we'll still report
717  * the error at the end. This is to match filemap_write_and_wait_range()
718  * behaviour.
719  *
720  * Returns: errno
721  */
722 
723 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
724 		      int datasync)
725 {
726 	struct address_space *mapping = file->f_mapping;
727 	struct inode *inode = mapping->host;
728 	int sync_state = inode->i_state & I_DIRTY;
729 	struct gfs2_inode *ip = GFS2_I(inode);
730 	int ret = 0, ret1 = 0;
731 
732 	if (mapping->nrpages) {
733 		ret1 = filemap_fdatawrite_range(mapping, start, end);
734 		if (ret1 == -EIO)
735 			return ret1;
736 	}
737 
738 	if (!gfs2_is_jdata(ip))
739 		sync_state &= ~I_DIRTY_PAGES;
740 	if (datasync)
741 		sync_state &= ~I_DIRTY_SYNC;
742 
743 	if (sync_state) {
744 		ret = sync_inode_metadata(inode, 1);
745 		if (ret)
746 			return ret;
747 		if (gfs2_is_jdata(ip))
748 			ret = file_write_and_wait(file);
749 		if (ret)
750 			return ret;
751 		gfs2_ail_flush(ip->i_gl, 1);
752 	}
753 
754 	if (mapping->nrpages)
755 		ret = file_fdatawait_range(file, start, end);
756 
757 	return ret ? ret : ret1;
758 }
759 
760 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
761 				     struct gfs2_holder *gh)
762 {
763 	struct file *file = iocb->ki_filp;
764 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
765 	size_t count = iov_iter_count(to);
766 	ssize_t ret;
767 
768 	if (!count)
769 		return 0; /* skip atime */
770 
771 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
772 	ret = gfs2_glock_nq(gh);
773 	if (ret)
774 		goto out_uninit;
775 
776 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, 0);
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 				      struct gfs2_holder *gh)
785 {
786 	struct file *file = iocb->ki_filp;
787 	struct inode *inode = file->f_mapping->host;
788 	struct gfs2_inode *ip = GFS2_I(inode);
789 	size_t len = iov_iter_count(from);
790 	loff_t offset = iocb->ki_pos;
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, 0);
811 	if (ret == -ENOTBLK)
812 		ret = 0;
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 	struct gfs2_inode *ip;
823 	struct gfs2_holder gh;
824 	size_t written = 0;
825 	ssize_t ret;
826 
827 	if (iocb->ki_flags & IOCB_DIRECT) {
828 		ret = gfs2_file_direct_read(iocb, to, &gh);
829 		if (likely(ret != -ENOTBLK))
830 			return ret;
831 		iocb->ki_flags &= ~IOCB_DIRECT;
832 	}
833 	iocb->ki_flags |= IOCB_NOIO;
834 	ret = generic_file_read_iter(iocb, to);
835 	iocb->ki_flags &= ~IOCB_NOIO;
836 	if (ret >= 0) {
837 		if (!iov_iter_count(to))
838 			return ret;
839 		written = ret;
840 	} else {
841 		if (ret != -EAGAIN)
842 			return ret;
843 		if (iocb->ki_flags & IOCB_NOWAIT)
844 			return ret;
845 	}
846 	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
847 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
848 	ret = gfs2_glock_nq(&gh);
849 	if (ret)
850 		goto out_uninit;
851 	ret = generic_file_read_iter(iocb, to);
852 	if (ret > 0)
853 		written += ret;
854 	gfs2_glock_dq(&gh);
855 out_uninit:
856 	gfs2_holder_uninit(&gh);
857 	return written ? written : ret;
858 }
859 
860 /**
861  * gfs2_file_write_iter - Perform a write to a file
862  * @iocb: The io context
863  * @from: The data to write
864  *
865  * We have to do a lock/unlock here to refresh the inode size for
866  * O_APPEND writes, otherwise we can land up writing at the wrong
867  * offset. There is still a race, but provided the app is using its
868  * own file locking, this will make O_APPEND work as expected.
869  *
870  */
871 
872 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
873 {
874 	struct file *file = iocb->ki_filp;
875 	struct inode *inode = file_inode(file);
876 	struct gfs2_inode *ip = GFS2_I(inode);
877 	struct gfs2_holder gh;
878 	ssize_t ret;
879 
880 	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
881 
882 	if (iocb->ki_flags & IOCB_APPEND) {
883 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
884 		if (ret)
885 			return ret;
886 		gfs2_glock_dq_uninit(&gh);
887 	}
888 
889 	inode_lock(inode);
890 	ret = generic_write_checks(iocb, from);
891 	if (ret <= 0)
892 		goto out_unlock;
893 
894 	ret = file_remove_privs(file);
895 	if (ret)
896 		goto out_unlock;
897 
898 	ret = file_update_time(file);
899 	if (ret)
900 		goto out_unlock;
901 
902 	if (iocb->ki_flags & IOCB_DIRECT) {
903 		struct address_space *mapping = file->f_mapping;
904 		ssize_t buffered, ret2;
905 
906 		ret = gfs2_file_direct_write(iocb, from, &gh);
907 		if (ret < 0 || !iov_iter_count(from))
908 			goto out_unlock;
909 
910 		iocb->ki_flags |= IOCB_DSYNC;
911 		current->backing_dev_info = inode_to_bdi(inode);
912 		buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
913 		current->backing_dev_info = NULL;
914 		if (unlikely(buffered <= 0))
915 			goto out_unlock;
916 
917 		/*
918 		 * We need to ensure that the page cache pages are written to
919 		 * disk and invalidated to preserve the expected O_DIRECT
920 		 * semantics.  If the writeback or invalidate fails, only report
921 		 * the direct I/O range as we don't know if the buffered pages
922 		 * made it to disk.
923 		 */
924 		iocb->ki_pos += buffered;
925 		ret2 = generic_write_sync(iocb, buffered);
926 		invalidate_mapping_pages(mapping,
927 				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
928 				(iocb->ki_pos - 1) >> PAGE_SHIFT);
929 		if (!ret || ret2 > 0)
930 			ret += ret2;
931 	} else {
932 		current->backing_dev_info = inode_to_bdi(inode);
933 		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
934 		current->backing_dev_info = NULL;
935 		if (likely(ret > 0)) {
936 			iocb->ki_pos += ret;
937 			ret = generic_write_sync(iocb, ret);
938 		}
939 	}
940 
941 out_unlock:
942 	inode_unlock(inode);
943 	return ret;
944 }
945 
946 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
947 			   int mode)
948 {
949 	struct super_block *sb = inode->i_sb;
950 	struct gfs2_inode *ip = GFS2_I(inode);
951 	loff_t end = offset + len;
952 	struct buffer_head *dibh;
953 	int error;
954 
955 	error = gfs2_meta_inode_buffer(ip, &dibh);
956 	if (unlikely(error))
957 		return error;
958 
959 	gfs2_trans_add_meta(ip->i_gl, dibh);
960 
961 	if (gfs2_is_stuffed(ip)) {
962 		error = gfs2_unstuff_dinode(ip, NULL);
963 		if (unlikely(error))
964 			goto out;
965 	}
966 
967 	while (offset < end) {
968 		struct iomap iomap = { };
969 
970 		error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
971 		if (error)
972 			goto out;
973 		offset = iomap.offset + iomap.length;
974 		if (!(iomap.flags & IOMAP_F_NEW))
975 			continue;
976 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
977 					 iomap.length >> inode->i_blkbits,
978 					 GFP_NOFS);
979 		if (error) {
980 			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
981 			goto out;
982 		}
983 	}
984 out:
985 	brelse(dibh);
986 	return error;
987 }
988 
989 /**
990  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
991  *                     blocks, determine how many bytes can be written.
992  * @ip:          The inode in question.
993  * @len:         Max cap of bytes. What we return in *len must be <= this.
994  * @data_blocks: Compute and return the number of data blocks needed
995  * @ind_blocks:  Compute and return the number of indirect blocks needed
996  * @max_blocks:  The total blocks available to work with.
997  *
998  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
999  */
1000 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1001 			    unsigned int *data_blocks, unsigned int *ind_blocks,
1002 			    unsigned int max_blocks)
1003 {
1004 	loff_t max = *len;
1005 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1006 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1007 
1008 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1009 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1010 		max_data -= tmp;
1011 	}
1012 
1013 	*data_blocks = max_data;
1014 	*ind_blocks = max_blocks - max_data;
1015 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1016 	if (*len > max) {
1017 		*len = max;
1018 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1019 	}
1020 }
1021 
1022 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1023 {
1024 	struct inode *inode = file_inode(file);
1025 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1026 	struct gfs2_inode *ip = GFS2_I(inode);
1027 	struct gfs2_alloc_parms ap = { .aflags = 0, };
1028 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1029 	loff_t bytes, max_bytes, max_blks;
1030 	int error;
1031 	const loff_t pos = offset;
1032 	const loff_t count = len;
1033 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1034 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1035 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1036 
1037 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1038 
1039 	offset &= bsize_mask;
1040 
1041 	len = next - offset;
1042 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1043 	if (!bytes)
1044 		bytes = UINT_MAX;
1045 	bytes &= bsize_mask;
1046 	if (bytes == 0)
1047 		bytes = sdp->sd_sb.sb_bsize;
1048 
1049 	gfs2_size_hint(file, offset, len);
1050 
1051 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1052 	ap.min_target = data_blocks + ind_blocks;
1053 
1054 	while (len > 0) {
1055 		if (len < bytes)
1056 			bytes = len;
1057 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1058 			len -= bytes;
1059 			offset += bytes;
1060 			continue;
1061 		}
1062 
1063 		/* We need to determine how many bytes we can actually
1064 		 * fallocate without exceeding quota or going over the
1065 		 * end of the fs. We start off optimistically by assuming
1066 		 * we can write max_bytes */
1067 		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1068 
1069 		/* Since max_bytes is most likely a theoretical max, we
1070 		 * calculate a more realistic 'bytes' to serve as a good
1071 		 * starting point for the number of bytes we may be able
1072 		 * to write */
1073 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1074 		ap.target = data_blocks + ind_blocks;
1075 
1076 		error = gfs2_quota_lock_check(ip, &ap);
1077 		if (error)
1078 			return error;
1079 		/* ap.allowed tells us how many blocks quota will allow
1080 		 * us to write. Check if this reduces max_blks */
1081 		max_blks = UINT_MAX;
1082 		if (ap.allowed)
1083 			max_blks = ap.allowed;
1084 
1085 		error = gfs2_inplace_reserve(ip, &ap);
1086 		if (error)
1087 			goto out_qunlock;
1088 
1089 		/* check if the selected rgrp limits our max_blks further */
1090 		if (ip->i_res.rs_reserved < max_blks)
1091 			max_blks = ip->i_res.rs_reserved;
1092 
1093 		/* Almost done. Calculate bytes that can be written using
1094 		 * max_blks. We also recompute max_bytes, data_blocks and
1095 		 * ind_blocks */
1096 		calc_max_reserv(ip, &max_bytes, &data_blocks,
1097 				&ind_blocks, max_blks);
1098 
1099 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1100 			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1101 		if (gfs2_is_jdata(ip))
1102 			rblocks += data_blocks ? data_blocks : 1;
1103 
1104 		error = gfs2_trans_begin(sdp, rblocks,
1105 					 PAGE_SIZE >> inode->i_blkbits);
1106 		if (error)
1107 			goto out_trans_fail;
1108 
1109 		error = fallocate_chunk(inode, offset, max_bytes, mode);
1110 		gfs2_trans_end(sdp);
1111 
1112 		if (error)
1113 			goto out_trans_fail;
1114 
1115 		len -= max_bytes;
1116 		offset += max_bytes;
1117 		gfs2_inplace_release(ip);
1118 		gfs2_quota_unlock(ip);
1119 	}
1120 
1121 	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1122 		i_size_write(inode, pos + count);
1123 	file_update_time(file);
1124 	mark_inode_dirty(inode);
1125 
1126 	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1127 		return vfs_fsync_range(file, pos, pos + count - 1,
1128 			       (file->f_flags & __O_SYNC) ? 0 : 1);
1129 	return 0;
1130 
1131 out_trans_fail:
1132 	gfs2_inplace_release(ip);
1133 out_qunlock:
1134 	gfs2_quota_unlock(ip);
1135 	return error;
1136 }
1137 
1138 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1139 {
1140 	struct inode *inode = file_inode(file);
1141 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1142 	struct gfs2_inode *ip = GFS2_I(inode);
1143 	struct gfs2_holder gh;
1144 	int ret;
1145 
1146 	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1147 		return -EOPNOTSUPP;
1148 	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1149 	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1150 		return -EOPNOTSUPP;
1151 
1152 	inode_lock(inode);
1153 
1154 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1155 	ret = gfs2_glock_nq(&gh);
1156 	if (ret)
1157 		goto out_uninit;
1158 
1159 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1160 	    (offset + len) > inode->i_size) {
1161 		ret = inode_newsize_ok(inode, offset + len);
1162 		if (ret)
1163 			goto out_unlock;
1164 	}
1165 
1166 	ret = get_write_access(inode);
1167 	if (ret)
1168 		goto out_unlock;
1169 
1170 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1171 		ret = __gfs2_punch_hole(file, offset, len);
1172 	} else {
1173 		ret = __gfs2_fallocate(file, mode, offset, len);
1174 		if (ret)
1175 			gfs2_rs_deltree(&ip->i_res);
1176 	}
1177 
1178 	put_write_access(inode);
1179 out_unlock:
1180 	gfs2_glock_dq(&gh);
1181 out_uninit:
1182 	gfs2_holder_uninit(&gh);
1183 	inode_unlock(inode);
1184 	return ret;
1185 }
1186 
1187 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1188 				      struct file *out, loff_t *ppos,
1189 				      size_t len, unsigned int flags)
1190 {
1191 	ssize_t ret;
1192 
1193 	gfs2_size_hint(out, *ppos, len);
1194 
1195 	ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1196 	return ret;
1197 }
1198 
1199 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1200 
1201 /**
1202  * gfs2_lock - acquire/release a posix lock on a file
1203  * @file: the file pointer
1204  * @cmd: either modify or retrieve lock state, possibly wait
1205  * @fl: type and range of lock
1206  *
1207  * Returns: errno
1208  */
1209 
1210 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1211 {
1212 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1213 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1214 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1215 
1216 	if (!(fl->fl_flags & FL_POSIX))
1217 		return -ENOLCK;
1218 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1219 		return -ENOLCK;
1220 
1221 	if (cmd == F_CANCELLK) {
1222 		/* Hack: */
1223 		cmd = F_SETLK;
1224 		fl->fl_type = F_UNLCK;
1225 	}
1226 	if (unlikely(gfs2_withdrawn(sdp))) {
1227 		if (fl->fl_type == F_UNLCK)
1228 			locks_lock_file_wait(file, fl);
1229 		return -EIO;
1230 	}
1231 	if (IS_GETLK(cmd))
1232 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1233 	else if (fl->fl_type == F_UNLCK)
1234 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1235 	else
1236 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1237 }
1238 
1239 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1240 {
1241 	struct gfs2_file *fp = file->private_data;
1242 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1243 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1244 	struct gfs2_glock *gl;
1245 	unsigned int state;
1246 	u16 flags;
1247 	int error = 0;
1248 	int sleeptime;
1249 
1250 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1251 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1252 
1253 	mutex_lock(&fp->f_fl_mutex);
1254 
1255 	if (gfs2_holder_initialized(fl_gh)) {
1256 		struct file_lock request;
1257 		if (fl_gh->gh_state == state)
1258 			goto out;
1259 		locks_init_lock(&request);
1260 		request.fl_type = F_UNLCK;
1261 		request.fl_flags = FL_FLOCK;
1262 		locks_lock_file_wait(file, &request);
1263 		gfs2_glock_dq(fl_gh);
1264 		gfs2_holder_reinit(state, flags, fl_gh);
1265 	} else {
1266 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1267 				       &gfs2_flock_glops, CREATE, &gl);
1268 		if (error)
1269 			goto out;
1270 		gfs2_holder_init(gl, state, flags, fl_gh);
1271 		gfs2_glock_put(gl);
1272 	}
1273 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1274 		error = gfs2_glock_nq(fl_gh);
1275 		if (error != GLR_TRYFAILED)
1276 			break;
1277 		fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1278 		fl_gh->gh_error = 0;
1279 		msleep(sleeptime);
1280 	}
1281 	if (error) {
1282 		gfs2_holder_uninit(fl_gh);
1283 		if (error == GLR_TRYFAILED)
1284 			error = -EAGAIN;
1285 	} else {
1286 		error = locks_lock_file_wait(file, fl);
1287 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1288 	}
1289 
1290 out:
1291 	mutex_unlock(&fp->f_fl_mutex);
1292 	return error;
1293 }
1294 
1295 static void do_unflock(struct file *file, struct file_lock *fl)
1296 {
1297 	struct gfs2_file *fp = file->private_data;
1298 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1299 
1300 	mutex_lock(&fp->f_fl_mutex);
1301 	locks_lock_file_wait(file, fl);
1302 	if (gfs2_holder_initialized(fl_gh)) {
1303 		gfs2_glock_dq(fl_gh);
1304 		gfs2_holder_uninit(fl_gh);
1305 	}
1306 	mutex_unlock(&fp->f_fl_mutex);
1307 }
1308 
1309 /**
1310  * gfs2_flock - acquire/release a flock lock on a file
1311  * @file: the file pointer
1312  * @cmd: either modify or retrieve lock state, possibly wait
1313  * @fl: type and range of lock
1314  *
1315  * Returns: errno
1316  */
1317 
1318 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1319 {
1320 	if (!(fl->fl_flags & FL_FLOCK))
1321 		return -ENOLCK;
1322 	if (fl->fl_type & LOCK_MAND)
1323 		return -EOPNOTSUPP;
1324 
1325 	if (fl->fl_type == F_UNLCK) {
1326 		do_unflock(file, fl);
1327 		return 0;
1328 	} else {
1329 		return do_flock(file, cmd, fl);
1330 	}
1331 }
1332 
1333 const struct file_operations gfs2_file_fops = {
1334 	.llseek		= gfs2_llseek,
1335 	.read_iter	= gfs2_file_read_iter,
1336 	.write_iter	= gfs2_file_write_iter,
1337 	.iopoll		= iomap_dio_iopoll,
1338 	.unlocked_ioctl	= gfs2_ioctl,
1339 	.compat_ioctl	= gfs2_compat_ioctl,
1340 	.mmap		= gfs2_mmap,
1341 	.open		= gfs2_open,
1342 	.release	= gfs2_release,
1343 	.fsync		= gfs2_fsync,
1344 	.lock		= gfs2_lock,
1345 	.flock		= gfs2_flock,
1346 	.splice_read	= generic_file_splice_read,
1347 	.splice_write	= gfs2_file_splice_write,
1348 	.setlease	= simple_nosetlease,
1349 	.fallocate	= gfs2_fallocate,
1350 };
1351 
1352 const struct file_operations gfs2_dir_fops = {
1353 	.iterate_shared	= gfs2_readdir,
1354 	.unlocked_ioctl	= gfs2_ioctl,
1355 	.compat_ioctl	= gfs2_compat_ioctl,
1356 	.open		= gfs2_open,
1357 	.release	= gfs2_release,
1358 	.fsync		= gfs2_fsync,
1359 	.lock		= gfs2_lock,
1360 	.flock		= gfs2_flock,
1361 	.llseek		= default_llseek,
1362 };
1363 
1364 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1365 
1366 const struct file_operations gfs2_file_fops_nolock = {
1367 	.llseek		= gfs2_llseek,
1368 	.read_iter	= gfs2_file_read_iter,
1369 	.write_iter	= gfs2_file_write_iter,
1370 	.iopoll		= iomap_dio_iopoll,
1371 	.unlocked_ioctl	= gfs2_ioctl,
1372 	.compat_ioctl	= gfs2_compat_ioctl,
1373 	.mmap		= gfs2_mmap,
1374 	.open		= gfs2_open,
1375 	.release	= gfs2_release,
1376 	.fsync		= gfs2_fsync,
1377 	.splice_read	= generic_file_splice_read,
1378 	.splice_write	= gfs2_file_splice_write,
1379 	.setlease	= generic_setlease,
1380 	.fallocate	= gfs2_fallocate,
1381 };
1382 
1383 const struct file_operations gfs2_dir_fops_nolock = {
1384 	.iterate_shared	= gfs2_readdir,
1385 	.unlocked_ioctl	= gfs2_ioctl,
1386 	.compat_ioctl	= gfs2_compat_ioctl,
1387 	.open		= gfs2_open,
1388 	.release	= gfs2_release,
1389 	.fsync		= gfs2_fsync,
1390 	.llseek		= default_llseek,
1391 };
1392 
1393