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