xref: /linux/fs/gfs2/file.c (revision bdd1a21b52557ea8f61d0a5dc2f77151b576eb70)
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  * @inode: The inode
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 	vm_fault_t ret = VM_FAULT_LOCKED;
431 	struct gfs2_holder gh;
432 	unsigned int length;
433 	loff_t size;
434 	int err;
435 
436 	sb_start_pagefault(inode->i_sb);
437 
438 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
439 	err = gfs2_glock_nq(&gh);
440 	if (err) {
441 		ret = block_page_mkwrite_return(err);
442 		goto out_uninit;
443 	}
444 
445 	/* Check page index against inode size */
446 	size = i_size_read(inode);
447 	if (offset >= size) {
448 		ret = VM_FAULT_SIGBUS;
449 		goto out_unlock;
450 	}
451 
452 	/* Update file times before taking page lock */
453 	file_update_time(vmf->vma->vm_file);
454 
455 	/* page is wholly or partially inside EOF */
456 	if (size - offset < PAGE_SIZE)
457 		length = size - offset;
458 	else
459 		length = PAGE_SIZE;
460 
461 	gfs2_size_hint(vmf->vma->vm_file, offset, length);
462 
463 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
464 	set_bit(GIF_SW_PAGED, &ip->i_flags);
465 
466 	/*
467 	 * iomap_writepage / iomap_writepages currently don't support inline
468 	 * files, so always unstuff here.
469 	 */
470 
471 	if (!gfs2_is_stuffed(ip) &&
472 	    !gfs2_write_alloc_required(ip, offset, length)) {
473 		lock_page(page);
474 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
475 			ret = VM_FAULT_NOPAGE;
476 			unlock_page(page);
477 		}
478 		goto out_unlock;
479 	}
480 
481 	err = gfs2_rindex_update(sdp);
482 	if (err) {
483 		ret = block_page_mkwrite_return(err);
484 		goto out_unlock;
485 	}
486 
487 	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
488 	ap.target = data_blocks + ind_blocks;
489 	err = gfs2_quota_lock_check(ip, &ap);
490 	if (err) {
491 		ret = block_page_mkwrite_return(err);
492 		goto out_unlock;
493 	}
494 	err = gfs2_inplace_reserve(ip, &ap);
495 	if (err) {
496 		ret = block_page_mkwrite_return(err);
497 		goto out_quota_unlock;
498 	}
499 
500 	rblocks = RES_DINODE + ind_blocks;
501 	if (gfs2_is_jdata(ip))
502 		rblocks += data_blocks ? data_blocks : 1;
503 	if (ind_blocks || data_blocks) {
504 		rblocks += RES_STATFS + RES_QUOTA;
505 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
506 	}
507 	err = gfs2_trans_begin(sdp, rblocks, 0);
508 	if (err) {
509 		ret = block_page_mkwrite_return(err);
510 		goto out_trans_fail;
511 	}
512 
513 	/* Unstuff, if required, and allocate backing blocks for page */
514 	if (gfs2_is_stuffed(ip)) {
515 		err = gfs2_unstuff_dinode(ip);
516 		if (err) {
517 			ret = block_page_mkwrite_return(err);
518 			goto out_trans_end;
519 		}
520 	}
521 
522 	lock_page(page);
523 	/* If truncated, we must retry the operation, we may have raced
524 	 * with the glock demotion code.
525 	 */
526 	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
527 		ret = VM_FAULT_NOPAGE;
528 		goto out_page_locked;
529 	}
530 
531 	err = gfs2_allocate_page_backing(page, length);
532 	if (err)
533 		ret = block_page_mkwrite_return(err);
534 
535 out_page_locked:
536 	if (ret != VM_FAULT_LOCKED)
537 		unlock_page(page);
538 out_trans_end:
539 	gfs2_trans_end(sdp);
540 out_trans_fail:
541 	gfs2_inplace_release(ip);
542 out_quota_unlock:
543 	gfs2_quota_unlock(ip);
544 out_unlock:
545 	gfs2_glock_dq(&gh);
546 out_uninit:
547 	gfs2_holder_uninit(&gh);
548 	if (ret == VM_FAULT_LOCKED) {
549 		set_page_dirty(page);
550 		wait_for_stable_page(page);
551 	}
552 	sb_end_pagefault(inode->i_sb);
553 	return ret;
554 }
555 
556 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
557 {
558 	struct inode *inode = file_inode(vmf->vma->vm_file);
559 	struct gfs2_inode *ip = GFS2_I(inode);
560 	struct gfs2_holder gh;
561 	vm_fault_t ret;
562 	int err;
563 
564 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
565 	err = gfs2_glock_nq(&gh);
566 	if (err) {
567 		ret = block_page_mkwrite_return(err);
568 		goto out_uninit;
569 	}
570 	ret = filemap_fault(vmf);
571 	gfs2_glock_dq(&gh);
572 out_uninit:
573 	gfs2_holder_uninit(&gh);
574 	return ret;
575 }
576 
577 static const struct vm_operations_struct gfs2_vm_ops = {
578 	.fault = gfs2_fault,
579 	.map_pages = filemap_map_pages,
580 	.page_mkwrite = gfs2_page_mkwrite,
581 };
582 
583 /**
584  * gfs2_mmap
585  * @file: The file to map
586  * @vma: The VMA which described the mapping
587  *
588  * There is no need to get a lock here unless we should be updating
589  * atime. We ignore any locking errors since the only consequence is
590  * a missed atime update (which will just be deferred until later).
591  *
592  * Returns: 0
593  */
594 
595 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
596 {
597 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
598 
599 	if (!(file->f_flags & O_NOATIME) &&
600 	    !IS_NOATIME(&ip->i_inode)) {
601 		struct gfs2_holder i_gh;
602 		int error;
603 
604 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
605 					   &i_gh);
606 		if (error)
607 			return error;
608 		/* grab lock to update inode */
609 		gfs2_glock_dq_uninit(&i_gh);
610 		file_accessed(file);
611 	}
612 	vma->vm_ops = &gfs2_vm_ops;
613 
614 	return 0;
615 }
616 
617 /**
618  * gfs2_open_common - This is common to open and atomic_open
619  * @inode: The inode being opened
620  * @file: The file being opened
621  *
622  * This maybe called under a glock or not depending upon how it has
623  * been called. We must always be called under a glock for regular
624  * files, however. For other file types, it does not matter whether
625  * we hold the glock or not.
626  *
627  * Returns: Error code or 0 for success
628  */
629 
630 int gfs2_open_common(struct inode *inode, struct file *file)
631 {
632 	struct gfs2_file *fp;
633 	int ret;
634 
635 	if (S_ISREG(inode->i_mode)) {
636 		ret = generic_file_open(inode, file);
637 		if (ret)
638 			return ret;
639 	}
640 
641 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
642 	if (!fp)
643 		return -ENOMEM;
644 
645 	mutex_init(&fp->f_fl_mutex);
646 
647 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
648 	file->private_data = fp;
649 	if (file->f_mode & FMODE_WRITE) {
650 		ret = gfs2_qa_get(GFS2_I(inode));
651 		if (ret)
652 			goto fail;
653 	}
654 	return 0;
655 
656 fail:
657 	kfree(file->private_data);
658 	file->private_data = NULL;
659 	return ret;
660 }
661 
662 /**
663  * gfs2_open - open a file
664  * @inode: the inode to open
665  * @file: the struct file for this opening
666  *
667  * After atomic_open, this function is only used for opening files
668  * which are already cached. We must still get the glock for regular
669  * files to ensure that we have the file size uptodate for the large
670  * file check which is in the common code. That is only an issue for
671  * regular files though.
672  *
673  * Returns: errno
674  */
675 
676 static int gfs2_open(struct inode *inode, struct file *file)
677 {
678 	struct gfs2_inode *ip = GFS2_I(inode);
679 	struct gfs2_holder i_gh;
680 	int error;
681 	bool need_unlock = false;
682 
683 	if (S_ISREG(ip->i_inode.i_mode)) {
684 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
685 					   &i_gh);
686 		if (error)
687 			return error;
688 		need_unlock = true;
689 	}
690 
691 	error = gfs2_open_common(inode, file);
692 
693 	if (need_unlock)
694 		gfs2_glock_dq_uninit(&i_gh);
695 
696 	return error;
697 }
698 
699 /**
700  * gfs2_release - called to close a struct file
701  * @inode: the inode the struct file belongs to
702  * @file: the struct file being closed
703  *
704  * Returns: errno
705  */
706 
707 static int gfs2_release(struct inode *inode, struct file *file)
708 {
709 	struct gfs2_inode *ip = GFS2_I(inode);
710 
711 	kfree(file->private_data);
712 	file->private_data = NULL;
713 
714 	if (gfs2_rs_active(&ip->i_res))
715 		gfs2_rs_delete(ip, &inode->i_writecount);
716 	if (file->f_mode & FMODE_WRITE)
717 		gfs2_qa_put(ip);
718 	return 0;
719 }
720 
721 /**
722  * gfs2_fsync - sync the dirty data for a file (across the cluster)
723  * @file: the file that points to the dentry
724  * @start: the start position in the file to sync
725  * @end: the end position in the file to sync
726  * @datasync: set if we can ignore timestamp changes
727  *
728  * We split the data flushing here so that we don't wait for the data
729  * until after we've also sent the metadata to disk. Note that for
730  * data=ordered, we will write & wait for the data at the log flush
731  * stage anyway, so this is unlikely to make much of a difference
732  * except in the data=writeback case.
733  *
734  * If the fdatawrite fails due to any reason except -EIO, we will
735  * continue the remainder of the fsync, although we'll still report
736  * the error at the end. This is to match filemap_write_and_wait_range()
737  * behaviour.
738  *
739  * Returns: errno
740  */
741 
742 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
743 		      int datasync)
744 {
745 	struct address_space *mapping = file->f_mapping;
746 	struct inode *inode = mapping->host;
747 	int sync_state = inode->i_state & I_DIRTY;
748 	struct gfs2_inode *ip = GFS2_I(inode);
749 	int ret = 0, ret1 = 0;
750 
751 	if (mapping->nrpages) {
752 		ret1 = filemap_fdatawrite_range(mapping, start, end);
753 		if (ret1 == -EIO)
754 			return ret1;
755 	}
756 
757 	if (!gfs2_is_jdata(ip))
758 		sync_state &= ~I_DIRTY_PAGES;
759 	if (datasync)
760 		sync_state &= ~I_DIRTY_SYNC;
761 
762 	if (sync_state) {
763 		ret = sync_inode_metadata(inode, 1);
764 		if (ret)
765 			return ret;
766 		if (gfs2_is_jdata(ip))
767 			ret = file_write_and_wait(file);
768 		if (ret)
769 			return ret;
770 		gfs2_ail_flush(ip->i_gl, 1);
771 	}
772 
773 	if (mapping->nrpages)
774 		ret = file_fdatawait_range(file, start, end);
775 
776 	return ret ? ret : ret1;
777 }
778 
779 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
780 				     struct gfs2_holder *gh)
781 {
782 	struct file *file = iocb->ki_filp;
783 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
784 	size_t count = iov_iter_count(to);
785 	ssize_t ret;
786 
787 	if (!count)
788 		return 0; /* skip atime */
789 
790 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
791 	ret = gfs2_glock_nq(gh);
792 	if (ret)
793 		goto out_uninit;
794 
795 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, 0);
796 	gfs2_glock_dq(gh);
797 out_uninit:
798 	gfs2_holder_uninit(gh);
799 	return ret;
800 }
801 
802 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
803 				      struct gfs2_holder *gh)
804 {
805 	struct file *file = iocb->ki_filp;
806 	struct inode *inode = file->f_mapping->host;
807 	struct gfs2_inode *ip = GFS2_I(inode);
808 	size_t len = iov_iter_count(from);
809 	loff_t offset = iocb->ki_pos;
810 	ssize_t ret;
811 
812 	/*
813 	 * Deferred lock, even if its a write, since we do no allocation on
814 	 * this path. All we need to change is the atime, and this lock mode
815 	 * ensures that other nodes have flushed their buffered read caches
816 	 * (i.e. their page cache entries for this inode). We do not,
817 	 * unfortunately, have the option of only flushing a range like the
818 	 * VFS does.
819 	 */
820 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
821 	ret = gfs2_glock_nq(gh);
822 	if (ret)
823 		goto out_uninit;
824 
825 	/* Silently fall back to buffered I/O when writing beyond EOF */
826 	if (offset + len > i_size_read(&ip->i_inode))
827 		goto out;
828 
829 	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, 0);
830 	if (ret == -ENOTBLK)
831 		ret = 0;
832 out:
833 	gfs2_glock_dq(gh);
834 out_uninit:
835 	gfs2_holder_uninit(gh);
836 	return ret;
837 }
838 
839 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
840 {
841 	struct gfs2_inode *ip;
842 	struct gfs2_holder gh;
843 	size_t written = 0;
844 	ssize_t ret;
845 
846 	if (iocb->ki_flags & IOCB_DIRECT) {
847 		ret = gfs2_file_direct_read(iocb, to, &gh);
848 		if (likely(ret != -ENOTBLK))
849 			return ret;
850 		iocb->ki_flags &= ~IOCB_DIRECT;
851 	}
852 	iocb->ki_flags |= IOCB_NOIO;
853 	ret = generic_file_read_iter(iocb, to);
854 	iocb->ki_flags &= ~IOCB_NOIO;
855 	if (ret >= 0) {
856 		if (!iov_iter_count(to))
857 			return ret;
858 		written = ret;
859 	} else {
860 		if (ret != -EAGAIN)
861 			return ret;
862 		if (iocb->ki_flags & IOCB_NOWAIT)
863 			return ret;
864 	}
865 	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
866 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
867 	ret = gfs2_glock_nq(&gh);
868 	if (ret)
869 		goto out_uninit;
870 	ret = generic_file_read_iter(iocb, to);
871 	if (ret > 0)
872 		written += ret;
873 	gfs2_glock_dq(&gh);
874 out_uninit:
875 	gfs2_holder_uninit(&gh);
876 	return written ? written : ret;
877 }
878 
879 /**
880  * gfs2_file_write_iter - Perform a write to a file
881  * @iocb: The io context
882  * @from: The data to write
883  *
884  * We have to do a lock/unlock here to refresh the inode size for
885  * O_APPEND writes, otherwise we can land up writing at the wrong
886  * offset. There is still a race, but provided the app is using its
887  * own file locking, this will make O_APPEND work as expected.
888  *
889  */
890 
891 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
892 {
893 	struct file *file = iocb->ki_filp;
894 	struct inode *inode = file_inode(file);
895 	struct gfs2_inode *ip = GFS2_I(inode);
896 	struct gfs2_holder gh;
897 	ssize_t ret;
898 
899 	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
900 
901 	if (iocb->ki_flags & IOCB_APPEND) {
902 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
903 		if (ret)
904 			return ret;
905 		gfs2_glock_dq_uninit(&gh);
906 	}
907 
908 	inode_lock(inode);
909 	ret = generic_write_checks(iocb, from);
910 	if (ret <= 0)
911 		goto out_unlock;
912 
913 	ret = file_remove_privs(file);
914 	if (ret)
915 		goto out_unlock;
916 
917 	ret = file_update_time(file);
918 	if (ret)
919 		goto out_unlock;
920 
921 	if (iocb->ki_flags & IOCB_DIRECT) {
922 		struct address_space *mapping = file->f_mapping;
923 		ssize_t buffered, ret2;
924 
925 		ret = gfs2_file_direct_write(iocb, from, &gh);
926 		if (ret < 0 || !iov_iter_count(from))
927 			goto out_unlock;
928 
929 		iocb->ki_flags |= IOCB_DSYNC;
930 		current->backing_dev_info = inode_to_bdi(inode);
931 		buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
932 		current->backing_dev_info = NULL;
933 		if (unlikely(buffered <= 0)) {
934 			if (!ret)
935 				ret = buffered;
936 			goto out_unlock;
937 		}
938 
939 		/*
940 		 * We need to ensure that the page cache pages are written to
941 		 * disk and invalidated to preserve the expected O_DIRECT
942 		 * semantics.  If the writeback or invalidate fails, only report
943 		 * the direct I/O range as we don't know if the buffered pages
944 		 * made it to disk.
945 		 */
946 		iocb->ki_pos += buffered;
947 		ret2 = generic_write_sync(iocb, buffered);
948 		invalidate_mapping_pages(mapping,
949 				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
950 				(iocb->ki_pos - 1) >> PAGE_SHIFT);
951 		if (!ret || ret2 > 0)
952 			ret += ret2;
953 	} else {
954 		current->backing_dev_info = inode_to_bdi(inode);
955 		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
956 		current->backing_dev_info = NULL;
957 		if (likely(ret > 0)) {
958 			iocb->ki_pos += ret;
959 			ret = generic_write_sync(iocb, ret);
960 		}
961 	}
962 
963 out_unlock:
964 	inode_unlock(inode);
965 	return ret;
966 }
967 
968 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
969 			   int mode)
970 {
971 	struct super_block *sb = inode->i_sb;
972 	struct gfs2_inode *ip = GFS2_I(inode);
973 	loff_t end = offset + len;
974 	struct buffer_head *dibh;
975 	int error;
976 
977 	error = gfs2_meta_inode_buffer(ip, &dibh);
978 	if (unlikely(error))
979 		return error;
980 
981 	gfs2_trans_add_meta(ip->i_gl, dibh);
982 
983 	if (gfs2_is_stuffed(ip)) {
984 		error = gfs2_unstuff_dinode(ip);
985 		if (unlikely(error))
986 			goto out;
987 	}
988 
989 	while (offset < end) {
990 		struct iomap iomap = { };
991 
992 		error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
993 		if (error)
994 			goto out;
995 		offset = iomap.offset + iomap.length;
996 		if (!(iomap.flags & IOMAP_F_NEW))
997 			continue;
998 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
999 					 iomap.length >> inode->i_blkbits,
1000 					 GFP_NOFS);
1001 		if (error) {
1002 			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1003 			goto out;
1004 		}
1005 	}
1006 out:
1007 	brelse(dibh);
1008 	return error;
1009 }
1010 
1011 /**
1012  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1013  *                     blocks, determine how many bytes can be written.
1014  * @ip:          The inode in question.
1015  * @len:         Max cap of bytes. What we return in *len must be <= this.
1016  * @data_blocks: Compute and return the number of data blocks needed
1017  * @ind_blocks:  Compute and return the number of indirect blocks needed
1018  * @max_blocks:  The total blocks available to work with.
1019  *
1020  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1021  */
1022 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1023 			    unsigned int *data_blocks, unsigned int *ind_blocks,
1024 			    unsigned int max_blocks)
1025 {
1026 	loff_t max = *len;
1027 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1028 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1029 
1030 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1031 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1032 		max_data -= tmp;
1033 	}
1034 
1035 	*data_blocks = max_data;
1036 	*ind_blocks = max_blocks - max_data;
1037 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1038 	if (*len > max) {
1039 		*len = max;
1040 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1041 	}
1042 }
1043 
1044 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1045 {
1046 	struct inode *inode = file_inode(file);
1047 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1048 	struct gfs2_inode *ip = GFS2_I(inode);
1049 	struct gfs2_alloc_parms ap = { .aflags = 0, };
1050 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1051 	loff_t bytes, max_bytes, max_blks;
1052 	int error;
1053 	const loff_t pos = offset;
1054 	const loff_t count = len;
1055 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1056 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1057 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1058 
1059 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1060 
1061 	offset &= bsize_mask;
1062 
1063 	len = next - offset;
1064 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1065 	if (!bytes)
1066 		bytes = UINT_MAX;
1067 	bytes &= bsize_mask;
1068 	if (bytes == 0)
1069 		bytes = sdp->sd_sb.sb_bsize;
1070 
1071 	gfs2_size_hint(file, offset, len);
1072 
1073 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1074 	ap.min_target = data_blocks + ind_blocks;
1075 
1076 	while (len > 0) {
1077 		if (len < bytes)
1078 			bytes = len;
1079 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1080 			len -= bytes;
1081 			offset += bytes;
1082 			continue;
1083 		}
1084 
1085 		/* We need to determine how many bytes we can actually
1086 		 * fallocate without exceeding quota or going over the
1087 		 * end of the fs. We start off optimistically by assuming
1088 		 * we can write max_bytes */
1089 		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1090 
1091 		/* Since max_bytes is most likely a theoretical max, we
1092 		 * calculate a more realistic 'bytes' to serve as a good
1093 		 * starting point for the number of bytes we may be able
1094 		 * to write */
1095 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1096 		ap.target = data_blocks + ind_blocks;
1097 
1098 		error = gfs2_quota_lock_check(ip, &ap);
1099 		if (error)
1100 			return error;
1101 		/* ap.allowed tells us how many blocks quota will allow
1102 		 * us to write. Check if this reduces max_blks */
1103 		max_blks = UINT_MAX;
1104 		if (ap.allowed)
1105 			max_blks = ap.allowed;
1106 
1107 		error = gfs2_inplace_reserve(ip, &ap);
1108 		if (error)
1109 			goto out_qunlock;
1110 
1111 		/* check if the selected rgrp limits our max_blks further */
1112 		if (ip->i_res.rs_reserved < max_blks)
1113 			max_blks = ip->i_res.rs_reserved;
1114 
1115 		/* Almost done. Calculate bytes that can be written using
1116 		 * max_blks. We also recompute max_bytes, data_blocks and
1117 		 * ind_blocks */
1118 		calc_max_reserv(ip, &max_bytes, &data_blocks,
1119 				&ind_blocks, max_blks);
1120 
1121 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1122 			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1123 		if (gfs2_is_jdata(ip))
1124 			rblocks += data_blocks ? data_blocks : 1;
1125 
1126 		error = gfs2_trans_begin(sdp, rblocks,
1127 					 PAGE_SIZE >> inode->i_blkbits);
1128 		if (error)
1129 			goto out_trans_fail;
1130 
1131 		error = fallocate_chunk(inode, offset, max_bytes, mode);
1132 		gfs2_trans_end(sdp);
1133 
1134 		if (error)
1135 			goto out_trans_fail;
1136 
1137 		len -= max_bytes;
1138 		offset += max_bytes;
1139 		gfs2_inplace_release(ip);
1140 		gfs2_quota_unlock(ip);
1141 	}
1142 
1143 	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1144 		i_size_write(inode, pos + count);
1145 	file_update_time(file);
1146 	mark_inode_dirty(inode);
1147 
1148 	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1149 		return vfs_fsync_range(file, pos, pos + count - 1,
1150 			       (file->f_flags & __O_SYNC) ? 0 : 1);
1151 	return 0;
1152 
1153 out_trans_fail:
1154 	gfs2_inplace_release(ip);
1155 out_qunlock:
1156 	gfs2_quota_unlock(ip);
1157 	return error;
1158 }
1159 
1160 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1161 {
1162 	struct inode *inode = file_inode(file);
1163 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1164 	struct gfs2_inode *ip = GFS2_I(inode);
1165 	struct gfs2_holder gh;
1166 	int ret;
1167 
1168 	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1169 		return -EOPNOTSUPP;
1170 	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1171 	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1172 		return -EOPNOTSUPP;
1173 
1174 	inode_lock(inode);
1175 
1176 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1177 	ret = gfs2_glock_nq(&gh);
1178 	if (ret)
1179 		goto out_uninit;
1180 
1181 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1182 	    (offset + len) > inode->i_size) {
1183 		ret = inode_newsize_ok(inode, offset + len);
1184 		if (ret)
1185 			goto out_unlock;
1186 	}
1187 
1188 	ret = get_write_access(inode);
1189 	if (ret)
1190 		goto out_unlock;
1191 
1192 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1193 		ret = __gfs2_punch_hole(file, offset, len);
1194 	} else {
1195 		ret = __gfs2_fallocate(file, mode, offset, len);
1196 		if (ret)
1197 			gfs2_rs_deltree(&ip->i_res);
1198 	}
1199 
1200 	put_write_access(inode);
1201 out_unlock:
1202 	gfs2_glock_dq(&gh);
1203 out_uninit:
1204 	gfs2_holder_uninit(&gh);
1205 	inode_unlock(inode);
1206 	return ret;
1207 }
1208 
1209 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1210 				      struct file *out, loff_t *ppos,
1211 				      size_t len, unsigned int flags)
1212 {
1213 	ssize_t ret;
1214 
1215 	gfs2_size_hint(out, *ppos, len);
1216 
1217 	ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1218 	return ret;
1219 }
1220 
1221 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1222 
1223 /**
1224  * gfs2_lock - acquire/release a posix lock on a file
1225  * @file: the file pointer
1226  * @cmd: either modify or retrieve lock state, possibly wait
1227  * @fl: type and range of lock
1228  *
1229  * Returns: errno
1230  */
1231 
1232 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1233 {
1234 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1235 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1236 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1237 
1238 	if (!(fl->fl_flags & FL_POSIX))
1239 		return -ENOLCK;
1240 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1241 		return -ENOLCK;
1242 
1243 	if (cmd == F_CANCELLK) {
1244 		/* Hack: */
1245 		cmd = F_SETLK;
1246 		fl->fl_type = F_UNLCK;
1247 	}
1248 	if (unlikely(gfs2_withdrawn(sdp))) {
1249 		if (fl->fl_type == F_UNLCK)
1250 			locks_lock_file_wait(file, fl);
1251 		return -EIO;
1252 	}
1253 	if (IS_GETLK(cmd))
1254 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1255 	else if (fl->fl_type == F_UNLCK)
1256 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1257 	else
1258 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1259 }
1260 
1261 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1262 {
1263 	struct gfs2_file *fp = file->private_data;
1264 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1265 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1266 	struct gfs2_glock *gl;
1267 	unsigned int state;
1268 	u16 flags;
1269 	int error = 0;
1270 	int sleeptime;
1271 
1272 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1273 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1274 
1275 	mutex_lock(&fp->f_fl_mutex);
1276 
1277 	if (gfs2_holder_initialized(fl_gh)) {
1278 		struct file_lock request;
1279 		if (fl_gh->gh_state == state)
1280 			goto out;
1281 		locks_init_lock(&request);
1282 		request.fl_type = F_UNLCK;
1283 		request.fl_flags = FL_FLOCK;
1284 		locks_lock_file_wait(file, &request);
1285 		gfs2_glock_dq(fl_gh);
1286 		gfs2_holder_reinit(state, flags, fl_gh);
1287 	} else {
1288 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1289 				       &gfs2_flock_glops, CREATE, &gl);
1290 		if (error)
1291 			goto out;
1292 		gfs2_holder_init(gl, state, flags, fl_gh);
1293 		gfs2_glock_put(gl);
1294 	}
1295 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1296 		error = gfs2_glock_nq(fl_gh);
1297 		if (error != GLR_TRYFAILED)
1298 			break;
1299 		fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1300 		fl_gh->gh_error = 0;
1301 		msleep(sleeptime);
1302 	}
1303 	if (error) {
1304 		gfs2_holder_uninit(fl_gh);
1305 		if (error == GLR_TRYFAILED)
1306 			error = -EAGAIN;
1307 	} else {
1308 		error = locks_lock_file_wait(file, fl);
1309 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1310 	}
1311 
1312 out:
1313 	mutex_unlock(&fp->f_fl_mutex);
1314 	return error;
1315 }
1316 
1317 static void do_unflock(struct file *file, struct file_lock *fl)
1318 {
1319 	struct gfs2_file *fp = file->private_data;
1320 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1321 
1322 	mutex_lock(&fp->f_fl_mutex);
1323 	locks_lock_file_wait(file, fl);
1324 	if (gfs2_holder_initialized(fl_gh)) {
1325 		gfs2_glock_dq(fl_gh);
1326 		gfs2_holder_uninit(fl_gh);
1327 	}
1328 	mutex_unlock(&fp->f_fl_mutex);
1329 }
1330 
1331 /**
1332  * gfs2_flock - acquire/release a flock lock on a file
1333  * @file: the file pointer
1334  * @cmd: either modify or retrieve lock state, possibly wait
1335  * @fl: type and range of lock
1336  *
1337  * Returns: errno
1338  */
1339 
1340 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1341 {
1342 	if (!(fl->fl_flags & FL_FLOCK))
1343 		return -ENOLCK;
1344 	if (fl->fl_type & LOCK_MAND)
1345 		return -EOPNOTSUPP;
1346 
1347 	if (fl->fl_type == F_UNLCK) {
1348 		do_unflock(file, fl);
1349 		return 0;
1350 	} else {
1351 		return do_flock(file, cmd, fl);
1352 	}
1353 }
1354 
1355 const struct file_operations gfs2_file_fops = {
1356 	.llseek		= gfs2_llseek,
1357 	.read_iter	= gfs2_file_read_iter,
1358 	.write_iter	= gfs2_file_write_iter,
1359 	.iopoll		= iomap_dio_iopoll,
1360 	.unlocked_ioctl	= gfs2_ioctl,
1361 	.compat_ioctl	= gfs2_compat_ioctl,
1362 	.mmap		= gfs2_mmap,
1363 	.open		= gfs2_open,
1364 	.release	= gfs2_release,
1365 	.fsync		= gfs2_fsync,
1366 	.lock		= gfs2_lock,
1367 	.flock		= gfs2_flock,
1368 	.splice_read	= generic_file_splice_read,
1369 	.splice_write	= gfs2_file_splice_write,
1370 	.setlease	= simple_nosetlease,
1371 	.fallocate	= gfs2_fallocate,
1372 };
1373 
1374 const struct file_operations gfs2_dir_fops = {
1375 	.iterate_shared	= gfs2_readdir,
1376 	.unlocked_ioctl	= gfs2_ioctl,
1377 	.compat_ioctl	= gfs2_compat_ioctl,
1378 	.open		= gfs2_open,
1379 	.release	= gfs2_release,
1380 	.fsync		= gfs2_fsync,
1381 	.lock		= gfs2_lock,
1382 	.flock		= gfs2_flock,
1383 	.llseek		= default_llseek,
1384 };
1385 
1386 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1387 
1388 const struct file_operations gfs2_file_fops_nolock = {
1389 	.llseek		= gfs2_llseek,
1390 	.read_iter	= gfs2_file_read_iter,
1391 	.write_iter	= gfs2_file_write_iter,
1392 	.iopoll		= iomap_dio_iopoll,
1393 	.unlocked_ioctl	= gfs2_ioctl,
1394 	.compat_ioctl	= gfs2_compat_ioctl,
1395 	.mmap		= gfs2_mmap,
1396 	.open		= gfs2_open,
1397 	.release	= gfs2_release,
1398 	.fsync		= gfs2_fsync,
1399 	.splice_read	= generic_file_splice_read,
1400 	.splice_write	= gfs2_file_splice_write,
1401 	.setlease	= generic_setlease,
1402 	.fallocate	= gfs2_fallocate,
1403 };
1404 
1405 const struct file_operations gfs2_dir_fops_nolock = {
1406 	.iterate_shared	= gfs2_readdir,
1407 	.unlocked_ioctl	= gfs2_ioctl,
1408 	.compat_ioctl	= gfs2_compat_ioctl,
1409 	.open		= gfs2_open,
1410 	.release	= gfs2_release,
1411 	.fsync		= gfs2_fsync,
1412 	.llseek		= default_llseek,
1413 };
1414 
1415