1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/pagemap.h>
15 #include <linux/uio.h>
16 #include <linux/blkdev.h>
17 #include <linux/mm.h>
18 #include <linux/mount.h>
19 #include <linux/fs.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/falloc.h>
22 #include <linux/swap.h>
23 #include <linux/crc32.h>
24 #include <linux/writeback.h>
25 #include <linux/uaccess.h>
26 #include <linux/dlm.h>
27 #include <linux/dlm_plock.h>
28 #include <linux/delay.h>
29 #include <linux/backing-dev.h>
30 
31 #include "gfs2.h"
32 #include "incore.h"
33 #include "bmap.h"
34 #include "aops.h"
35 #include "dir.h"
36 #include "glock.h"
37 #include "glops.h"
38 #include "inode.h"
39 #include "log.h"
40 #include "meta_io.h"
41 #include "quota.h"
42 #include "rgrp.h"
43 #include "trans.h"
44 #include "util.h"
45 
46 /**
47  * gfs2_llseek - seek to a location in a file
48  * @file: the file
49  * @offset: the offset
50  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
51  *
52  * SEEK_END requires the glock for the file because it references the
53  * file's size.
54  *
55  * Returns: The new offset, or errno
56  */
57 
gfs2_llseek(struct file * file,loff_t offset,int whence)58 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
59 {
60 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
61 	struct gfs2_holder i_gh;
62 	loff_t error;
63 
64 	switch (whence) {
65 	case SEEK_END:
66 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
67 					   &i_gh);
68 		if (!error) {
69 			error = generic_file_llseek(file, offset, whence);
70 			gfs2_glock_dq_uninit(&i_gh);
71 		}
72 		break;
73 
74 	case SEEK_DATA:
75 		error = gfs2_seek_data(file, offset);
76 		break;
77 
78 	case SEEK_HOLE:
79 		error = gfs2_seek_hole(file, offset);
80 		break;
81 
82 	case SEEK_CUR:
83 	case SEEK_SET:
84 		/*
85 		 * These don't reference inode->i_size and don't depend on the
86 		 * block mapping, so we don't need the glock.
87 		 */
88 		error = generic_file_llseek(file, offset, whence);
89 		break;
90 	default:
91 		error = -EINVAL;
92 	}
93 
94 	return error;
95 }
96 
97 /**
98  * gfs2_readdir - Iterator for a directory
99  * @file: The directory to read from
100  * @ctx: What to feed directory entries to
101  *
102  * Returns: errno
103  */
104 
gfs2_readdir(struct file * file,struct dir_context * ctx)105 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
106 {
107 	struct inode *dir = file->f_mapping->host;
108 	struct gfs2_inode *dip = GFS2_I(dir);
109 	struct gfs2_holder d_gh;
110 	int error;
111 
112 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
113 	if (error)
114 		return error;
115 
116 	error = gfs2_dir_read(dir, ctx, &file->f_ra);
117 
118 	gfs2_glock_dq_uninit(&d_gh);
119 
120 	return error;
121 }
122 
123 /**
124  * fsflag_gfs2flag
125  *
126  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
127  * and to GFS2_DIF_JDATA for non-directories.
128  */
129 static struct {
130 	u32 fsflag;
131 	u32 gfsflag;
132 } fsflag_gfs2flag[] = {
133 	{FS_SYNC_FL, GFS2_DIF_SYNC},
134 	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
135 	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
136 	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
137 	{FS_INDEX_FL, GFS2_DIF_EXHASH},
138 	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
139 	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
140 };
141 
gfs2_get_flags(struct file * filp,u32 __user * ptr)142 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
143 {
144 	struct inode *inode = file_inode(filp);
145 	struct gfs2_inode *ip = GFS2_I(inode);
146 	struct gfs2_holder gh;
147 	int i, error;
148 	u32 gfsflags, fsflags = 0;
149 
150 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
151 	error = gfs2_glock_nq(&gh);
152 	if (error)
153 		goto out_uninit;
154 
155 	gfsflags = ip->i_diskflags;
156 	if (S_ISDIR(inode->i_mode))
157 		gfsflags &= ~GFS2_DIF_JDATA;
158 	else
159 		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
160 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
161 		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
162 			fsflags |= fsflag_gfs2flag[i].fsflag;
163 
164 	if (put_user(fsflags, ptr))
165 		error = -EFAULT;
166 
167 	gfs2_glock_dq(&gh);
168 out_uninit:
169 	gfs2_holder_uninit(&gh);
170 	return error;
171 }
172 
gfs2_set_inode_flags(struct inode * inode)173 void gfs2_set_inode_flags(struct inode *inode)
174 {
175 	struct gfs2_inode *ip = GFS2_I(inode);
176 	unsigned int flags = inode->i_flags;
177 
178 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
179 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
180 		flags |= S_NOSEC;
181 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
182 		flags |= S_IMMUTABLE;
183 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
184 		flags |= S_APPEND;
185 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
186 		flags |= S_NOATIME;
187 	if (ip->i_diskflags & GFS2_DIF_SYNC)
188 		flags |= S_SYNC;
189 	inode->i_flags = flags;
190 }
191 
192 /* Flags that can be set by user space */
193 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
194 			     GFS2_DIF_IMMUTABLE|		\
195 			     GFS2_DIF_APPENDONLY|		\
196 			     GFS2_DIF_NOATIME|			\
197 			     GFS2_DIF_SYNC|			\
198 			     GFS2_DIF_TOPDIR|			\
199 			     GFS2_DIF_INHERIT_JDATA)
200 
201 /**
202  * do_gfs2_set_flags - set flags on an inode
203  * @filp: file pointer
204  * @reqflags: The flags to set
205  * @mask: Indicates which flags are valid
206  *
207  */
do_gfs2_set_flags(struct file * filp,u32 reqflags,u32 mask)208 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
209 {
210 	struct inode *inode = file_inode(filp);
211 	struct gfs2_inode *ip = GFS2_I(inode);
212 	struct gfs2_sbd *sdp = GFS2_SB(inode);
213 	struct buffer_head *bh;
214 	struct gfs2_holder gh;
215 	int error;
216 	u32 new_flags, flags;
217 
218 	error = mnt_want_write_file(filp);
219 	if (error)
220 		return error;
221 
222 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
223 	if (error)
224 		goto out_drop_write;
225 
226 	error = -EACCES;
227 	if (!inode_owner_or_capable(inode))
228 		goto out;
229 
230 	error = 0;
231 	flags = ip->i_diskflags;
232 	new_flags = (flags & ~mask) | (reqflags & mask);
233 	if ((new_flags ^ flags) == 0)
234 		goto out;
235 
236 	error = -EPERM;
237 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
238 		goto out;
239 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
240 		goto out;
241 	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
242 	    !capable(CAP_LINUX_IMMUTABLE))
243 		goto out;
244 	if (!IS_IMMUTABLE(inode)) {
245 		error = gfs2_permission(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 out_drop_write:
281 	mnt_drop_write_file(filp);
282 	return error;
283 }
284 
gfs2_set_flags(struct file * filp,u32 __user * ptr)285 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
286 {
287 	struct inode *inode = file_inode(filp);
288 	u32 fsflags, gfsflags = 0;
289 	u32 mask;
290 	int i;
291 
292 	if (get_user(fsflags, ptr))
293 		return -EFAULT;
294 
295 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
296 		if (fsflags & fsflag_gfs2flag[i].fsflag) {
297 			fsflags &= ~fsflag_gfs2flag[i].fsflag;
298 			gfsflags |= fsflag_gfs2flag[i].gfsflag;
299 		}
300 	}
301 	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
302 		return -EINVAL;
303 
304 	mask = GFS2_FLAGS_USER_SET;
305 	if (S_ISDIR(inode->i_mode)) {
306 		mask &= ~GFS2_DIF_JDATA;
307 	} else {
308 		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
309 		if (gfsflags & GFS2_DIF_TOPDIR)
310 			return -EINVAL;
311 		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
312 	}
313 
314 	return do_gfs2_set_flags(filp, gfsflags, mask);
315 }
316 
gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)317 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
318 {
319 	switch(cmd) {
320 	case FS_IOC_GETFLAGS:
321 		return gfs2_get_flags(filp, (u32 __user *)arg);
322 	case FS_IOC_SETFLAGS:
323 		return gfs2_set_flags(filp, (u32 __user *)arg);
324 	case FITRIM:
325 		return gfs2_fitrim(filp, (void __user *)arg);
326 	}
327 	return -ENOTTY;
328 }
329 
330 /**
331  * gfs2_size_hint - Give a hint to the size of a write request
332  * @filep: The struct file
333  * @offset: The file offset of the write
334  * @size: The length of the write
335  *
336  * When we are about to do a write, this function records the total
337  * write size in order to provide a suitable hint to the lower layers
338  * about how many blocks will be required.
339  *
340  */
341 
gfs2_size_hint(struct file * filep,loff_t offset,size_t size)342 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
343 {
344 	struct inode *inode = file_inode(filep);
345 	struct gfs2_sbd *sdp = GFS2_SB(inode);
346 	struct gfs2_inode *ip = GFS2_I(inode);
347 	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
348 	int hint = min_t(size_t, INT_MAX, blks);
349 
350 	if (hint > atomic_read(&ip->i_res.rs_sizehint))
351 		atomic_set(&ip->i_res.rs_sizehint, hint);
352 }
353 
354 /**
355  * gfs2_allocate_page_backing - Use bmap to allocate blocks
356  * @page: The (locked) page to allocate backing for
357  *
358  * We try to allocate all the blocks required for the page in
359  * one go. This might fail for various reasons, so we keep
360  * trying until all the blocks to back this page are allocated.
361  * If some of the blocks are already allocated, thats ok too.
362  */
363 
gfs2_allocate_page_backing(struct page * page)364 static int gfs2_allocate_page_backing(struct page *page)
365 {
366 	struct inode *inode = page->mapping->host;
367 	struct buffer_head bh;
368 	unsigned long size = PAGE_SIZE;
369 	u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits);
370 
371 	do {
372 		bh.b_state = 0;
373 		bh.b_size = size;
374 		gfs2_block_map(inode, lblock, &bh, 1);
375 		if (!buffer_mapped(&bh))
376 			return -EIO;
377 		size -= bh.b_size;
378 		lblock += (bh.b_size >> inode->i_blkbits);
379 	} while(size > 0);
380 	return 0;
381 }
382 
383 /**
384  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
385  * @vma: The virtual memory area
386  * @vmf: The virtual memory fault containing the page to become writable
387  *
388  * When the page becomes writable, we need to ensure that we have
389  * blocks allocated on disk to back that page.
390  */
391 
gfs2_page_mkwrite(struct vm_fault * vmf)392 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
393 {
394 	struct page *page = vmf->page;
395 	struct inode *inode = file_inode(vmf->vma->vm_file);
396 	struct gfs2_inode *ip = GFS2_I(inode);
397 	struct gfs2_sbd *sdp = GFS2_SB(inode);
398 	struct gfs2_alloc_parms ap = { .aflags = 0, };
399 	unsigned long last_index;
400 	u64 pos = page->index << PAGE_SHIFT;
401 	unsigned int data_blocks, ind_blocks, rblocks;
402 	struct gfs2_holder gh;
403 	loff_t size;
404 	int ret;
405 
406 	sb_start_pagefault(inode->i_sb);
407 
408 	ret = gfs2_rsqa_alloc(ip);
409 	if (ret)
410 		goto out;
411 
412 	gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE);
413 
414 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
415 	ret = gfs2_glock_nq(&gh);
416 	if (ret)
417 		goto out_uninit;
418 
419 	/* Update file times before taking page lock */
420 	file_update_time(vmf->vma->vm_file);
421 
422 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
423 	set_bit(GIF_SW_PAGED, &ip->i_flags);
424 
425 	if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) {
426 		lock_page(page);
427 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
428 			ret = -EAGAIN;
429 			unlock_page(page);
430 		}
431 		goto out_unlock;
432 	}
433 
434 	ret = gfs2_rindex_update(sdp);
435 	if (ret)
436 		goto out_unlock;
437 
438 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
439 	ap.target = data_blocks + ind_blocks;
440 	ret = gfs2_quota_lock_check(ip, &ap);
441 	if (ret)
442 		goto out_unlock;
443 	ret = gfs2_inplace_reserve(ip, &ap);
444 	if (ret)
445 		goto out_quota_unlock;
446 
447 	rblocks = RES_DINODE + ind_blocks;
448 	if (gfs2_is_jdata(ip))
449 		rblocks += data_blocks ? data_blocks : 1;
450 	if (ind_blocks || data_blocks) {
451 		rblocks += RES_STATFS + RES_QUOTA;
452 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
453 	}
454 	ret = gfs2_trans_begin(sdp, rblocks, 0);
455 	if (ret)
456 		goto out_trans_fail;
457 
458 	lock_page(page);
459 	ret = -EINVAL;
460 	size = i_size_read(inode);
461 	last_index = (size - 1) >> PAGE_SHIFT;
462 	/* Check page index against inode size */
463 	if (size == 0 || (page->index > last_index))
464 		goto out_trans_end;
465 
466 	ret = -EAGAIN;
467 	/* If truncated, we must retry the operation, we may have raced
468 	 * with the glock demotion code.
469 	 */
470 	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
471 		goto out_trans_end;
472 
473 	/* Unstuff, if required, and allocate backing blocks for page */
474 	ret = 0;
475 	if (gfs2_is_stuffed(ip))
476 		ret = gfs2_unstuff_dinode(ip, page);
477 	if (ret == 0)
478 		ret = gfs2_allocate_page_backing(page);
479 
480 out_trans_end:
481 	if (ret)
482 		unlock_page(page);
483 	gfs2_trans_end(sdp);
484 out_trans_fail:
485 	gfs2_inplace_release(ip);
486 out_quota_unlock:
487 	gfs2_quota_unlock(ip);
488 out_unlock:
489 	gfs2_glock_dq(&gh);
490 out_uninit:
491 	gfs2_holder_uninit(&gh);
492 	if (ret == 0) {
493 		set_page_dirty(page);
494 		wait_for_stable_page(page);
495 	}
496 out:
497 	sb_end_pagefault(inode->i_sb);
498 	return block_page_mkwrite_return(ret);
499 }
500 
501 static const struct vm_operations_struct gfs2_vm_ops = {
502 	.fault = filemap_fault,
503 	.map_pages = filemap_map_pages,
504 	.page_mkwrite = gfs2_page_mkwrite,
505 };
506 
507 /**
508  * gfs2_mmap -
509  * @file: The file to map
510  * @vma: The VMA which described the mapping
511  *
512  * There is no need to get a lock here unless we should be updating
513  * atime. We ignore any locking errors since the only consequence is
514  * a missed atime update (which will just be deferred until later).
515  *
516  * Returns: 0
517  */
518 
gfs2_mmap(struct file * file,struct vm_area_struct * vma)519 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
520 {
521 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
522 
523 	if (!(file->f_flags & O_NOATIME) &&
524 	    !IS_NOATIME(&ip->i_inode)) {
525 		struct gfs2_holder i_gh;
526 		int error;
527 
528 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
529 					   &i_gh);
530 		if (error)
531 			return error;
532 		/* grab lock to update inode */
533 		gfs2_glock_dq_uninit(&i_gh);
534 		file_accessed(file);
535 	}
536 	vma->vm_ops = &gfs2_vm_ops;
537 
538 	return 0;
539 }
540 
541 /**
542  * gfs2_open_common - This is common to open and atomic_open
543  * @inode: The inode being opened
544  * @file: The file being opened
545  *
546  * This maybe called under a glock or not depending upon how it has
547  * been called. We must always be called under a glock for regular
548  * files, however. For other file types, it does not matter whether
549  * we hold the glock or not.
550  *
551  * Returns: Error code or 0 for success
552  */
553 
gfs2_open_common(struct inode * inode,struct file * file)554 int gfs2_open_common(struct inode *inode, struct file *file)
555 {
556 	struct gfs2_file *fp;
557 	int ret;
558 
559 	if (S_ISREG(inode->i_mode)) {
560 		ret = generic_file_open(inode, file);
561 		if (ret)
562 			return ret;
563 	}
564 
565 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
566 	if (!fp)
567 		return -ENOMEM;
568 
569 	mutex_init(&fp->f_fl_mutex);
570 
571 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
572 	file->private_data = fp;
573 	return 0;
574 }
575 
576 /**
577  * gfs2_open - open a file
578  * @inode: the inode to open
579  * @file: the struct file for this opening
580  *
581  * After atomic_open, this function is only used for opening files
582  * which are already cached. We must still get the glock for regular
583  * files to ensure that we have the file size uptodate for the large
584  * file check which is in the common code. That is only an issue for
585  * regular files though.
586  *
587  * Returns: errno
588  */
589 
gfs2_open(struct inode * inode,struct file * file)590 static int gfs2_open(struct inode *inode, struct file *file)
591 {
592 	struct gfs2_inode *ip = GFS2_I(inode);
593 	struct gfs2_holder i_gh;
594 	int error;
595 	bool need_unlock = false;
596 
597 	if (S_ISREG(ip->i_inode.i_mode)) {
598 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
599 					   &i_gh);
600 		if (error)
601 			return error;
602 		need_unlock = true;
603 	}
604 
605 	error = gfs2_open_common(inode, file);
606 
607 	if (need_unlock)
608 		gfs2_glock_dq_uninit(&i_gh);
609 
610 	return error;
611 }
612 
613 /**
614  * gfs2_release - called to close a struct file
615  * @inode: the inode the struct file belongs to
616  * @file: the struct file being closed
617  *
618  * Returns: errno
619  */
620 
gfs2_release(struct inode * inode,struct file * file)621 static int gfs2_release(struct inode *inode, struct file *file)
622 {
623 	struct gfs2_inode *ip = GFS2_I(inode);
624 
625 	kfree(file->private_data);
626 	file->private_data = NULL;
627 
628 	if (!(file->f_mode & FMODE_WRITE))
629 		return 0;
630 
631 	gfs2_rsqa_delete(ip, &inode->i_writecount);
632 	return 0;
633 }
634 
635 /**
636  * gfs2_fsync - sync the dirty data for a file (across the cluster)
637  * @file: the file that points to the dentry
638  * @start: the start position in the file to sync
639  * @end: the end position in the file to sync
640  * @datasync: set if we can ignore timestamp changes
641  *
642  * We split the data flushing here so that we don't wait for the data
643  * until after we've also sent the metadata to disk. Note that for
644  * data=ordered, we will write & wait for the data at the log flush
645  * stage anyway, so this is unlikely to make much of a difference
646  * except in the data=writeback case.
647  *
648  * If the fdatawrite fails due to any reason except -EIO, we will
649  * continue the remainder of the fsync, although we'll still report
650  * the error at the end. This is to match filemap_write_and_wait_range()
651  * behaviour.
652  *
653  * Returns: errno
654  */
655 
gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)656 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
657 		      int datasync)
658 {
659 	struct address_space *mapping = file->f_mapping;
660 	struct inode *inode = mapping->host;
661 	int sync_state = inode->i_state & I_DIRTY_ALL;
662 	struct gfs2_inode *ip = GFS2_I(inode);
663 	int ret = 0, ret1 = 0;
664 
665 	if (mapping->nrpages) {
666 		ret1 = filemap_fdatawrite_range(mapping, start, end);
667 		if (ret1 == -EIO)
668 			return ret1;
669 	}
670 
671 	if (!gfs2_is_jdata(ip))
672 		sync_state &= ~I_DIRTY_PAGES;
673 	if (datasync)
674 		sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
675 
676 	if (sync_state) {
677 		ret = sync_inode_metadata(inode, 1);
678 		if (ret)
679 			return ret;
680 		if (gfs2_is_jdata(ip))
681 			ret = file_write_and_wait(file);
682 		if (ret)
683 			return ret;
684 		gfs2_ail_flush(ip->i_gl, 1);
685 	}
686 
687 	if (mapping->nrpages)
688 		ret = file_fdatawait_range(file, start, end);
689 
690 	return ret ? ret : ret1;
691 }
692 
gfs2_file_direct_read(struct kiocb * iocb,struct iov_iter * to)693 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to)
694 {
695 	struct file *file = iocb->ki_filp;
696 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
697 	size_t count = iov_iter_count(to);
698 	struct gfs2_holder gh;
699 	ssize_t ret;
700 
701 	if (!count)
702 		return 0; /* skip atime */
703 
704 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
705 	ret = gfs2_glock_nq(&gh);
706 	if (ret)
707 		goto out_uninit;
708 
709 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL);
710 
711 	gfs2_glock_dq(&gh);
712 out_uninit:
713 	gfs2_holder_uninit(&gh);
714 	return ret;
715 }
716 
gfs2_file_direct_write(struct kiocb * iocb,struct iov_iter * from)717 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
718 {
719 	struct file *file = iocb->ki_filp;
720 	struct inode *inode = file->f_mapping->host;
721 	struct gfs2_inode *ip = GFS2_I(inode);
722 	size_t len = iov_iter_count(from);
723 	loff_t offset = iocb->ki_pos;
724 	struct gfs2_holder gh;
725 	ssize_t ret;
726 
727 	/*
728 	 * Deferred lock, even if its a write, since we do no allocation on
729 	 * this path. All we need to change is the atime, and this lock mode
730 	 * ensures that other nodes have flushed their buffered read caches
731 	 * (i.e. their page cache entries for this inode). We do not,
732 	 * unfortunately, have the option of only flushing a range like the
733 	 * VFS does.
734 	 */
735 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
736 	ret = gfs2_glock_nq(&gh);
737 	if (ret)
738 		goto out_uninit;
739 
740 	/* Silently fall back to buffered I/O when writing beyond EOF */
741 	if (offset + len > i_size_read(&ip->i_inode))
742 		goto out;
743 
744 	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL);
745 
746 out:
747 	gfs2_glock_dq(&gh);
748 out_uninit:
749 	gfs2_holder_uninit(&gh);
750 	return ret;
751 }
752 
gfs2_file_read_iter(struct kiocb * iocb,struct iov_iter * to)753 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
754 {
755 	ssize_t ret;
756 
757 	if (iocb->ki_flags & IOCB_DIRECT) {
758 		ret = gfs2_file_direct_read(iocb, to);
759 		if (likely(ret != -ENOTBLK))
760 			return ret;
761 		iocb->ki_flags &= ~IOCB_DIRECT;
762 	}
763 	return generic_file_read_iter(iocb, to);
764 }
765 
766 /**
767  * gfs2_file_write_iter - Perform a write to a file
768  * @iocb: The io context
769  * @from: The data to write
770  *
771  * We have to do a lock/unlock here to refresh the inode size for
772  * O_APPEND writes, otherwise we can land up writing at the wrong
773  * offset. There is still a race, but provided the app is using its
774  * own file locking, this will make O_APPEND work as expected.
775  *
776  */
777 
gfs2_file_write_iter(struct kiocb * iocb,struct iov_iter * from)778 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
779 {
780 	struct file *file = iocb->ki_filp;
781 	struct inode *inode = file_inode(file);
782 	struct gfs2_inode *ip = GFS2_I(inode);
783 	ssize_t ret;
784 
785 	ret = gfs2_rsqa_alloc(ip);
786 	if (ret)
787 		return ret;
788 
789 	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
790 
791 	if (iocb->ki_flags & IOCB_APPEND) {
792 		struct gfs2_holder gh;
793 
794 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
795 		if (ret)
796 			return ret;
797 		gfs2_glock_dq_uninit(&gh);
798 	}
799 
800 	inode_lock(inode);
801 	ret = generic_write_checks(iocb, from);
802 	if (ret <= 0)
803 		goto out_unlock;
804 
805 	ret = file_remove_privs(file);
806 	if (ret)
807 		goto out_unlock;
808 
809 	ret = file_update_time(file);
810 	if (ret)
811 		goto out_unlock;
812 
813 	if (iocb->ki_flags & IOCB_DIRECT) {
814 		struct address_space *mapping = file->f_mapping;
815 		ssize_t buffered, ret2;
816 
817 		ret = gfs2_file_direct_write(iocb, from);
818 		if (ret < 0 || !iov_iter_count(from))
819 			goto out_unlock;
820 
821 		iocb->ki_flags |= IOCB_DSYNC;
822 		current->backing_dev_info = inode_to_bdi(inode);
823 		buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
824 		current->backing_dev_info = NULL;
825 		if (unlikely(buffered <= 0)) {
826 			if (!ret)
827 				ret = buffered;
828 			goto out_unlock;
829 		}
830 
831 		/*
832 		 * We need to ensure that the page cache pages are written to
833 		 * disk and invalidated to preserve the expected O_DIRECT
834 		 * semantics.  If the writeback or invalidate fails, only report
835 		 * the direct I/O range as we don't know if the buffered pages
836 		 * made it to disk.
837 		 */
838 		iocb->ki_pos += buffered;
839 		ret2 = generic_write_sync(iocb, buffered);
840 		invalidate_mapping_pages(mapping,
841 				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
842 				(iocb->ki_pos - 1) >> PAGE_SHIFT);
843 		if (!ret || ret2 > 0)
844 			ret += ret2;
845 	} else {
846 		current->backing_dev_info = inode_to_bdi(inode);
847 		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
848 		current->backing_dev_info = NULL;
849 		if (likely(ret > 0)) {
850 			iocb->ki_pos += ret;
851 			ret = generic_write_sync(iocb, ret);
852 		}
853 	}
854 
855 out_unlock:
856 	inode_unlock(inode);
857 	return ret;
858 }
859 
fallocate_chunk(struct inode * inode,loff_t offset,loff_t len,int mode)860 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
861 			   int mode)
862 {
863 	struct super_block *sb = inode->i_sb;
864 	struct gfs2_inode *ip = GFS2_I(inode);
865 	loff_t end = offset + len;
866 	struct buffer_head *dibh;
867 	int error;
868 
869 	error = gfs2_meta_inode_buffer(ip, &dibh);
870 	if (unlikely(error))
871 		return error;
872 
873 	gfs2_trans_add_meta(ip->i_gl, dibh);
874 
875 	if (gfs2_is_stuffed(ip)) {
876 		error = gfs2_unstuff_dinode(ip, NULL);
877 		if (unlikely(error))
878 			goto out;
879 	}
880 
881 	while (offset < end) {
882 		struct iomap iomap = { };
883 
884 		error = gfs2_iomap_get_alloc(inode, offset, end - offset,
885 					     &iomap);
886 		if (error)
887 			goto out;
888 		offset = iomap.offset + iomap.length;
889 		if (!(iomap.flags & IOMAP_F_NEW))
890 			continue;
891 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
892 					 iomap.length >> inode->i_blkbits,
893 					 GFP_NOFS);
894 		if (error) {
895 			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
896 			goto out;
897 		}
898 	}
899 out:
900 	brelse(dibh);
901 	return error;
902 }
903 /**
904  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
905  *                     blocks, determine how many bytes can be written.
906  * @ip:          The inode in question.
907  * @len:         Max cap of bytes. What we return in *len must be <= this.
908  * @data_blocks: Compute and return the number of data blocks needed
909  * @ind_blocks:  Compute and return the number of indirect blocks needed
910  * @max_blocks:  The total blocks available to work with.
911  *
912  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
913  */
calc_max_reserv(struct gfs2_inode * ip,loff_t * len,unsigned int * data_blocks,unsigned int * ind_blocks,unsigned int max_blocks)914 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
915 			    unsigned int *data_blocks, unsigned int *ind_blocks,
916 			    unsigned int max_blocks)
917 {
918 	loff_t max = *len;
919 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
920 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
921 
922 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
923 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
924 		max_data -= tmp;
925 	}
926 
927 	*data_blocks = max_data;
928 	*ind_blocks = max_blocks - max_data;
929 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
930 	if (*len > max) {
931 		*len = max;
932 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
933 	}
934 }
935 
__gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)936 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
937 {
938 	struct inode *inode = file_inode(file);
939 	struct gfs2_sbd *sdp = GFS2_SB(inode);
940 	struct gfs2_inode *ip = GFS2_I(inode);
941 	struct gfs2_alloc_parms ap = { .aflags = 0, };
942 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
943 	loff_t bytes, max_bytes, max_blks;
944 	int error;
945 	const loff_t pos = offset;
946 	const loff_t count = len;
947 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
948 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
949 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
950 
951 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
952 
953 	offset &= bsize_mask;
954 
955 	len = next - offset;
956 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
957 	if (!bytes)
958 		bytes = UINT_MAX;
959 	bytes &= bsize_mask;
960 	if (bytes == 0)
961 		bytes = sdp->sd_sb.sb_bsize;
962 
963 	gfs2_size_hint(file, offset, len);
964 
965 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
966 	ap.min_target = data_blocks + ind_blocks;
967 
968 	while (len > 0) {
969 		if (len < bytes)
970 			bytes = len;
971 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
972 			len -= bytes;
973 			offset += bytes;
974 			continue;
975 		}
976 
977 		/* We need to determine how many bytes we can actually
978 		 * fallocate without exceeding quota or going over the
979 		 * end of the fs. We start off optimistically by assuming
980 		 * we can write max_bytes */
981 		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
982 
983 		/* Since max_bytes is most likely a theoretical max, we
984 		 * calculate a more realistic 'bytes' to serve as a good
985 		 * starting point for the number of bytes we may be able
986 		 * to write */
987 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
988 		ap.target = data_blocks + ind_blocks;
989 
990 		error = gfs2_quota_lock_check(ip, &ap);
991 		if (error)
992 			return error;
993 		/* ap.allowed tells us how many blocks quota will allow
994 		 * us to write. Check if this reduces max_blks */
995 		max_blks = UINT_MAX;
996 		if (ap.allowed)
997 			max_blks = ap.allowed;
998 
999 		error = gfs2_inplace_reserve(ip, &ap);
1000 		if (error)
1001 			goto out_qunlock;
1002 
1003 		/* check if the selected rgrp limits our max_blks further */
1004 		if (ap.allowed && ap.allowed < max_blks)
1005 			max_blks = ap.allowed;
1006 
1007 		/* Almost done. Calculate bytes that can be written using
1008 		 * max_blks. We also recompute max_bytes, data_blocks and
1009 		 * ind_blocks */
1010 		calc_max_reserv(ip, &max_bytes, &data_blocks,
1011 				&ind_blocks, max_blks);
1012 
1013 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1014 			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1015 		if (gfs2_is_jdata(ip))
1016 			rblocks += data_blocks ? data_blocks : 1;
1017 
1018 		error = gfs2_trans_begin(sdp, rblocks,
1019 					 PAGE_SIZE/sdp->sd_sb.sb_bsize);
1020 		if (error)
1021 			goto out_trans_fail;
1022 
1023 		error = fallocate_chunk(inode, offset, max_bytes, mode);
1024 		gfs2_trans_end(sdp);
1025 
1026 		if (error)
1027 			goto out_trans_fail;
1028 
1029 		len -= max_bytes;
1030 		offset += max_bytes;
1031 		gfs2_inplace_release(ip);
1032 		gfs2_quota_unlock(ip);
1033 	}
1034 
1035 	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) {
1036 		i_size_write(inode, pos + count);
1037 		file_update_time(file);
1038 		mark_inode_dirty(inode);
1039 	}
1040 
1041 	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1042 		return vfs_fsync_range(file, pos, pos + count - 1,
1043 			       (file->f_flags & __O_SYNC) ? 0 : 1);
1044 	return 0;
1045 
1046 out_trans_fail:
1047 	gfs2_inplace_release(ip);
1048 out_qunlock:
1049 	gfs2_quota_unlock(ip);
1050 	return error;
1051 }
1052 
gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1053 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1054 {
1055 	struct inode *inode = file_inode(file);
1056 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1057 	struct gfs2_inode *ip = GFS2_I(inode);
1058 	struct gfs2_holder gh;
1059 	int ret;
1060 
1061 	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1062 		return -EOPNOTSUPP;
1063 	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1064 	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1065 		return -EOPNOTSUPP;
1066 
1067 	inode_lock(inode);
1068 
1069 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1070 	ret = gfs2_glock_nq(&gh);
1071 	if (ret)
1072 		goto out_uninit;
1073 
1074 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1075 	    (offset + len) > inode->i_size) {
1076 		ret = inode_newsize_ok(inode, offset + len);
1077 		if (ret)
1078 			goto out_unlock;
1079 	}
1080 
1081 	ret = get_write_access(inode);
1082 	if (ret)
1083 		goto out_unlock;
1084 
1085 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1086 		ret = __gfs2_punch_hole(file, offset, len);
1087 	} else {
1088 		ret = gfs2_rsqa_alloc(ip);
1089 		if (ret)
1090 			goto out_putw;
1091 
1092 		ret = __gfs2_fallocate(file, mode, offset, len);
1093 
1094 		if (ret)
1095 			gfs2_rs_deltree(&ip->i_res);
1096 	}
1097 
1098 out_putw:
1099 	put_write_access(inode);
1100 out_unlock:
1101 	gfs2_glock_dq(&gh);
1102 out_uninit:
1103 	gfs2_holder_uninit(&gh);
1104 	inode_unlock(inode);
1105 	return ret;
1106 }
1107 
gfs2_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)1108 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1109 				      struct file *out, loff_t *ppos,
1110 				      size_t len, unsigned int flags)
1111 {
1112 	int error;
1113 	struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
1114 
1115 	error = gfs2_rsqa_alloc(ip);
1116 	if (error)
1117 		return (ssize_t)error;
1118 
1119 	gfs2_size_hint(out, *ppos, len);
1120 
1121 	return iter_file_splice_write(pipe, out, ppos, len, flags);
1122 }
1123 
1124 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1125 
1126 /**
1127  * gfs2_lock - acquire/release a posix lock on a file
1128  * @file: the file pointer
1129  * @cmd: either modify or retrieve lock state, possibly wait
1130  * @fl: type and range of lock
1131  *
1132  * Returns: errno
1133  */
1134 
gfs2_lock(struct file * file,int cmd,struct file_lock * fl)1135 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1136 {
1137 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1138 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1139 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1140 
1141 	if (!(fl->fl_flags & FL_POSIX))
1142 		return -ENOLCK;
1143 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1144 		return -ENOLCK;
1145 
1146 	if (cmd == F_CANCELLK) {
1147 		/* Hack: */
1148 		cmd = F_SETLK;
1149 		fl->fl_type = F_UNLCK;
1150 	}
1151 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1152 		if (fl->fl_type == F_UNLCK)
1153 			locks_lock_file_wait(file, fl);
1154 		return -EIO;
1155 	}
1156 	if (IS_GETLK(cmd))
1157 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1158 	else if (fl->fl_type == F_UNLCK)
1159 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1160 	else
1161 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1162 }
1163 
do_flock(struct file * file,int cmd,struct file_lock * fl)1164 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1165 {
1166 	struct gfs2_file *fp = file->private_data;
1167 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1168 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1169 	struct gfs2_glock *gl;
1170 	unsigned int state;
1171 	u16 flags;
1172 	int error = 0;
1173 	int sleeptime;
1174 
1175 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1176 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1177 
1178 	mutex_lock(&fp->f_fl_mutex);
1179 
1180 	if (gfs2_holder_initialized(fl_gh)) {
1181 		if (fl_gh->gh_state == state)
1182 			goto out;
1183 		locks_lock_file_wait(file,
1184 				     &(struct file_lock) {
1185 					     .fl_type = F_UNLCK,
1186 					     .fl_flags = FL_FLOCK
1187 				     });
1188 		gfs2_glock_dq(fl_gh);
1189 		gfs2_holder_reinit(state, flags, fl_gh);
1190 	} else {
1191 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1192 				       &gfs2_flock_glops, CREATE, &gl);
1193 		if (error)
1194 			goto out;
1195 		gfs2_holder_init(gl, state, flags, fl_gh);
1196 		gfs2_glock_put(gl);
1197 	}
1198 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1199 		error = gfs2_glock_nq(fl_gh);
1200 		if (error != GLR_TRYFAILED)
1201 			break;
1202 		fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1203 		fl_gh->gh_error = 0;
1204 		msleep(sleeptime);
1205 	}
1206 	if (error) {
1207 		gfs2_holder_uninit(fl_gh);
1208 		if (error == GLR_TRYFAILED)
1209 			error = -EAGAIN;
1210 	} else {
1211 		error = locks_lock_file_wait(file, fl);
1212 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1213 	}
1214 
1215 out:
1216 	mutex_unlock(&fp->f_fl_mutex);
1217 	return error;
1218 }
1219 
do_unflock(struct file * file,struct file_lock * fl)1220 static void do_unflock(struct file *file, struct file_lock *fl)
1221 {
1222 	struct gfs2_file *fp = file->private_data;
1223 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1224 
1225 	mutex_lock(&fp->f_fl_mutex);
1226 	locks_lock_file_wait(file, fl);
1227 	if (gfs2_holder_initialized(fl_gh)) {
1228 		gfs2_glock_dq(fl_gh);
1229 		gfs2_holder_uninit(fl_gh);
1230 	}
1231 	mutex_unlock(&fp->f_fl_mutex);
1232 }
1233 
1234 /**
1235  * gfs2_flock - acquire/release a flock lock on a file
1236  * @file: the file pointer
1237  * @cmd: either modify or retrieve lock state, possibly wait
1238  * @fl: type and range of lock
1239  *
1240  * Returns: errno
1241  */
1242 
gfs2_flock(struct file * file,int cmd,struct file_lock * fl)1243 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1244 {
1245 	if (!(fl->fl_flags & FL_FLOCK))
1246 		return -ENOLCK;
1247 	if (fl->fl_type & LOCK_MAND)
1248 		return -EOPNOTSUPP;
1249 
1250 	if (fl->fl_type == F_UNLCK) {
1251 		do_unflock(file, fl);
1252 		return 0;
1253 	} else {
1254 		return do_flock(file, cmd, fl);
1255 	}
1256 }
1257 
1258 const struct file_operations gfs2_file_fops = {
1259 	.llseek		= gfs2_llseek,
1260 	.read_iter	= gfs2_file_read_iter,
1261 	.write_iter	= gfs2_file_write_iter,
1262 	.unlocked_ioctl	= gfs2_ioctl,
1263 	.mmap		= gfs2_mmap,
1264 	.open		= gfs2_open,
1265 	.release	= gfs2_release,
1266 	.fsync		= gfs2_fsync,
1267 	.lock		= gfs2_lock,
1268 	.flock		= gfs2_flock,
1269 	.splice_read	= generic_file_splice_read,
1270 	.splice_write	= gfs2_file_splice_write,
1271 	.setlease	= simple_nosetlease,
1272 	.fallocate	= gfs2_fallocate,
1273 };
1274 
1275 const struct file_operations gfs2_dir_fops = {
1276 	.iterate_shared	= gfs2_readdir,
1277 	.unlocked_ioctl	= gfs2_ioctl,
1278 	.open		= gfs2_open,
1279 	.release	= gfs2_release,
1280 	.fsync		= gfs2_fsync,
1281 	.lock		= gfs2_lock,
1282 	.flock		= gfs2_flock,
1283 	.llseek		= default_llseek,
1284 };
1285 
1286 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1287 
1288 const struct file_operations gfs2_file_fops_nolock = {
1289 	.llseek		= gfs2_llseek,
1290 	.read_iter	= gfs2_file_read_iter,
1291 	.write_iter	= gfs2_file_write_iter,
1292 	.unlocked_ioctl	= gfs2_ioctl,
1293 	.mmap		= gfs2_mmap,
1294 	.open		= gfs2_open,
1295 	.release	= gfs2_release,
1296 	.fsync		= gfs2_fsync,
1297 	.splice_read	= generic_file_splice_read,
1298 	.splice_write	= gfs2_file_splice_write,
1299 	.setlease	= generic_setlease,
1300 	.fallocate	= gfs2_fallocate,
1301 };
1302 
1303 const struct file_operations gfs2_dir_fops_nolock = {
1304 	.iterate_shared	= gfs2_readdir,
1305 	.unlocked_ioctl	= gfs2_ioctl,
1306 	.open		= gfs2_open,
1307 	.release	= gfs2_release,
1308 	.fsync		= gfs2_fsync,
1309 	.llseek		= default_llseek,
1310 };
1311 
1312