xref: /linux/fs/gfs2/file.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
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 <asm/uaccess.h>
26 #include <linux/dlm.h>
27 #include <linux/dlm_plock.h>
28 #include <linux/delay.h>
29 
30 #include "gfs2.h"
31 #include "incore.h"
32 #include "bmap.h"
33 #include "dir.h"
34 #include "glock.h"
35 #include "glops.h"
36 #include "inode.h"
37 #include "log.h"
38 #include "meta_io.h"
39 #include "quota.h"
40 #include "rgrp.h"
41 #include "trans.h"
42 #include "util.h"
43 
44 /**
45  * gfs2_llseek - seek to a location in a file
46  * @file: the file
47  * @offset: the offset
48  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
49  *
50  * SEEK_END requires the glock for the file because it references the
51  * file's size.
52  *
53  * Returns: The new offset, or errno
54  */
55 
56 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
57 {
58 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 	struct gfs2_holder i_gh;
60 	loff_t error;
61 
62 	switch (whence) {
63 	case SEEK_END: /* These reference inode->i_size */
64 	case SEEK_DATA:
65 	case SEEK_HOLE:
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 	case SEEK_CUR:
74 	case SEEK_SET:
75 		error = generic_file_llseek(file, offset, whence);
76 		break;
77 	default:
78 		error = -EINVAL;
79 	}
80 
81 	return error;
82 }
83 
84 /**
85  * gfs2_readdir - Iterator for a directory
86  * @file: The directory to read from
87  * @ctx: What to feed directory entries to
88  *
89  * Returns: errno
90  */
91 
92 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
93 {
94 	struct inode *dir = file->f_mapping->host;
95 	struct gfs2_inode *dip = GFS2_I(dir);
96 	struct gfs2_holder d_gh;
97 	int error;
98 
99 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
100 	if (error)
101 		return error;
102 
103 	error = gfs2_dir_read(dir, ctx, &file->f_ra);
104 
105 	gfs2_glock_dq_uninit(&d_gh);
106 
107 	return error;
108 }
109 
110 /**
111  * fsflags_cvt
112  * @table: A table of 32 u32 flags
113  * @val: a 32 bit value to convert
114  *
115  * This function can be used to convert between fsflags values and
116  * GFS2's own flags values.
117  *
118  * Returns: the converted flags
119  */
120 static u32 fsflags_cvt(const u32 *table, u32 val)
121 {
122 	u32 res = 0;
123 	while(val) {
124 		if (val & 1)
125 			res |= *table;
126 		table++;
127 		val >>= 1;
128 	}
129 	return res;
130 }
131 
132 static const u32 fsflags_to_gfs2[32] = {
133 	[3] = GFS2_DIF_SYNC,
134 	[4] = GFS2_DIF_IMMUTABLE,
135 	[5] = GFS2_DIF_APPENDONLY,
136 	[7] = GFS2_DIF_NOATIME,
137 	[12] = GFS2_DIF_EXHASH,
138 	[14] = GFS2_DIF_INHERIT_JDATA,
139 	[17] = GFS2_DIF_TOPDIR,
140 };
141 
142 static const u32 gfs2_to_fsflags[32] = {
143 	[gfs2fl_Sync] = FS_SYNC_FL,
144 	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
145 	[gfs2fl_AppendOnly] = FS_APPEND_FL,
146 	[gfs2fl_NoAtime] = FS_NOATIME_FL,
147 	[gfs2fl_ExHash] = FS_INDEX_FL,
148 	[gfs2fl_TopLevel] = FS_TOPDIR_FL,
149 	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
150 };
151 
152 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
153 {
154 	struct inode *inode = file_inode(filp);
155 	struct gfs2_inode *ip = GFS2_I(inode);
156 	struct gfs2_holder gh;
157 	int error;
158 	u32 fsflags;
159 
160 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
161 	error = gfs2_glock_nq(&gh);
162 	if (error)
163 		return error;
164 
165 	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
166 	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
167 		fsflags |= FS_JOURNAL_DATA_FL;
168 	if (put_user(fsflags, ptr))
169 		error = -EFAULT;
170 
171 	gfs2_glock_dq(&gh);
172 	gfs2_holder_uninit(&gh);
173 	return error;
174 }
175 
176 void gfs2_set_inode_flags(struct inode *inode)
177 {
178 	struct gfs2_inode *ip = GFS2_I(inode);
179 	unsigned int flags = inode->i_flags;
180 
181 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
182 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
183 		flags |= S_NOSEC;
184 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
185 		flags |= S_IMMUTABLE;
186 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
187 		flags |= S_APPEND;
188 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
189 		flags |= S_NOATIME;
190 	if (ip->i_diskflags & GFS2_DIF_SYNC)
191 		flags |= S_SYNC;
192 	inode->i_flags = flags;
193 }
194 
195 /* Flags that can be set by user space */
196 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
197 			     GFS2_DIF_IMMUTABLE|		\
198 			     GFS2_DIF_APPENDONLY|		\
199 			     GFS2_DIF_NOATIME|			\
200 			     GFS2_DIF_SYNC|			\
201 			     GFS2_DIF_SYSTEM|			\
202 			     GFS2_DIF_TOPDIR|			\
203 			     GFS2_DIF_INHERIT_JDATA)
204 
205 /**
206  * do_gfs2_set_flags - set flags on an inode
207  * @filp: file pointer
208  * @reqflags: The flags to set
209  * @mask: Indicates which flags are valid
210  *
211  */
212 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
213 {
214 	struct inode *inode = file_inode(filp);
215 	struct gfs2_inode *ip = GFS2_I(inode);
216 	struct gfs2_sbd *sdp = GFS2_SB(inode);
217 	struct buffer_head *bh;
218 	struct gfs2_holder gh;
219 	int error;
220 	u32 new_flags, flags;
221 
222 	error = mnt_want_write_file(filp);
223 	if (error)
224 		return error;
225 
226 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
227 	if (error)
228 		goto out_drop_write;
229 
230 	error = -EACCES;
231 	if (!inode_owner_or_capable(inode))
232 		goto out;
233 
234 	error = 0;
235 	flags = ip->i_diskflags;
236 	new_flags = (flags & ~mask) | (reqflags & mask);
237 	if ((new_flags ^ flags) == 0)
238 		goto out;
239 
240 	error = -EINVAL;
241 	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
242 		goto out;
243 
244 	error = -EPERM;
245 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
246 		goto out;
247 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
248 		goto out;
249 	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
250 	    !capable(CAP_LINUX_IMMUTABLE))
251 		goto out;
252 	if (!IS_IMMUTABLE(inode)) {
253 		error = gfs2_permission(inode, MAY_WRITE);
254 		if (error)
255 			goto out;
256 	}
257 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
258 		if (flags & GFS2_DIF_JDATA)
259 			gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH);
260 		error = filemap_fdatawrite(inode->i_mapping);
261 		if (error)
262 			goto out;
263 		error = filemap_fdatawait(inode->i_mapping);
264 		if (error)
265 			goto out;
266 	}
267 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
268 	if (error)
269 		goto out;
270 	error = gfs2_meta_inode_buffer(ip, &bh);
271 	if (error)
272 		goto out_trans_end;
273 	gfs2_trans_add_meta(ip->i_gl, bh);
274 	ip->i_diskflags = new_flags;
275 	gfs2_dinode_out(ip, bh->b_data);
276 	brelse(bh);
277 	gfs2_set_inode_flags(inode);
278 	gfs2_set_aops(inode);
279 out_trans_end:
280 	gfs2_trans_end(sdp);
281 out:
282 	gfs2_glock_dq_uninit(&gh);
283 out_drop_write:
284 	mnt_drop_write_file(filp);
285 	return error;
286 }
287 
288 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
289 {
290 	struct inode *inode = file_inode(filp);
291 	u32 fsflags, gfsflags;
292 
293 	if (get_user(fsflags, ptr))
294 		return -EFAULT;
295 
296 	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
297 	if (!S_ISDIR(inode->i_mode)) {
298 		gfsflags &= ~GFS2_DIF_TOPDIR;
299 		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
300 			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
301 		return do_gfs2_set_flags(filp, gfsflags, ~0);
302 	}
303 	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
304 }
305 
306 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
307 {
308 	switch(cmd) {
309 	case FS_IOC_GETFLAGS:
310 		return gfs2_get_flags(filp, (u32 __user *)arg);
311 	case FS_IOC_SETFLAGS:
312 		return gfs2_set_flags(filp, (u32 __user *)arg);
313 	case FITRIM:
314 		return gfs2_fitrim(filp, (void __user *)arg);
315 	}
316 	return -ENOTTY;
317 }
318 
319 /**
320  * gfs2_size_hint - Give a hint to the size of a write request
321  * @filep: The struct file
322  * @offset: The file offset of the write
323  * @size: The length of the write
324  *
325  * When we are about to do a write, this function records the total
326  * write size in order to provide a suitable hint to the lower layers
327  * about how many blocks will be required.
328  *
329  */
330 
331 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
332 {
333 	struct inode *inode = file_inode(filep);
334 	struct gfs2_sbd *sdp = GFS2_SB(inode);
335 	struct gfs2_inode *ip = GFS2_I(inode);
336 	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
337 	int hint = min_t(size_t, INT_MAX, blks);
338 
339 	if (hint > atomic_read(&ip->i_res->rs_sizehint))
340 		atomic_set(&ip->i_res->rs_sizehint, hint);
341 }
342 
343 /**
344  * gfs2_allocate_page_backing - Use bmap to allocate blocks
345  * @page: The (locked) page to allocate backing for
346  *
347  * We try to allocate all the blocks required for the page in
348  * one go. This might fail for various reasons, so we keep
349  * trying until all the blocks to back this page are allocated.
350  * If some of the blocks are already allocated, thats ok too.
351  */
352 
353 static int gfs2_allocate_page_backing(struct page *page)
354 {
355 	struct inode *inode = page->mapping->host;
356 	struct buffer_head bh;
357 	unsigned long size = PAGE_CACHE_SIZE;
358 	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
359 
360 	do {
361 		bh.b_state = 0;
362 		bh.b_size = size;
363 		gfs2_block_map(inode, lblock, &bh, 1);
364 		if (!buffer_mapped(&bh))
365 			return -EIO;
366 		size -= bh.b_size;
367 		lblock += (bh.b_size >> inode->i_blkbits);
368 	} while(size > 0);
369 	return 0;
370 }
371 
372 /**
373  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
374  * @vma: The virtual memory area
375  * @vmf: The virtual memory fault containing the page to become writable
376  *
377  * When the page becomes writable, we need to ensure that we have
378  * blocks allocated on disk to back that page.
379  */
380 
381 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
382 {
383 	struct page *page = vmf->page;
384 	struct inode *inode = file_inode(vma->vm_file);
385 	struct gfs2_inode *ip = GFS2_I(inode);
386 	struct gfs2_sbd *sdp = GFS2_SB(inode);
387 	struct gfs2_alloc_parms ap = { .aflags = 0, };
388 	unsigned long last_index;
389 	u64 pos = page->index << PAGE_CACHE_SHIFT;
390 	unsigned int data_blocks, ind_blocks, rblocks;
391 	struct gfs2_holder gh;
392 	loff_t size;
393 	int ret;
394 
395 	sb_start_pagefault(inode->i_sb);
396 
397 	/* Update file times before taking page lock */
398 	file_update_time(vma->vm_file);
399 
400 	ret = get_write_access(inode);
401 	if (ret)
402 		goto out;
403 
404 	ret = gfs2_rs_alloc(ip);
405 	if (ret)
406 		goto out_write_access;
407 
408 	gfs2_size_hint(vma->vm_file, pos, PAGE_CACHE_SIZE);
409 
410 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
411 	ret = gfs2_glock_nq(&gh);
412 	if (ret)
413 		goto out_uninit;
414 
415 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
416 	set_bit(GIF_SW_PAGED, &ip->i_flags);
417 
418 	if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) {
419 		lock_page(page);
420 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
421 			ret = -EAGAIN;
422 			unlock_page(page);
423 		}
424 		goto out_unlock;
425 	}
426 
427 	ret = gfs2_rindex_update(sdp);
428 	if (ret)
429 		goto out_unlock;
430 
431 	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
432 	ap.target = data_blocks + ind_blocks;
433 	ret = gfs2_quota_lock_check(ip, &ap);
434 	if (ret)
435 		goto out_unlock;
436 	ret = gfs2_inplace_reserve(ip, &ap);
437 	if (ret)
438 		goto out_quota_unlock;
439 
440 	rblocks = RES_DINODE + ind_blocks;
441 	if (gfs2_is_jdata(ip))
442 		rblocks += data_blocks ? data_blocks : 1;
443 	if (ind_blocks || data_blocks) {
444 		rblocks += RES_STATFS + RES_QUOTA;
445 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
446 	}
447 	ret = gfs2_trans_begin(sdp, rblocks, 0);
448 	if (ret)
449 		goto out_trans_fail;
450 
451 	lock_page(page);
452 	ret = -EINVAL;
453 	size = i_size_read(inode);
454 	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
455 	/* Check page index against inode size */
456 	if (size == 0 || (page->index > last_index))
457 		goto out_trans_end;
458 
459 	ret = -EAGAIN;
460 	/* If truncated, we must retry the operation, we may have raced
461 	 * with the glock demotion code.
462 	 */
463 	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
464 		goto out_trans_end;
465 
466 	/* Unstuff, if required, and allocate backing blocks for page */
467 	ret = 0;
468 	if (gfs2_is_stuffed(ip))
469 		ret = gfs2_unstuff_dinode(ip, page);
470 	if (ret == 0)
471 		ret = gfs2_allocate_page_backing(page);
472 
473 out_trans_end:
474 	if (ret)
475 		unlock_page(page);
476 	gfs2_trans_end(sdp);
477 out_trans_fail:
478 	gfs2_inplace_release(ip);
479 out_quota_unlock:
480 	gfs2_quota_unlock(ip);
481 out_unlock:
482 	gfs2_glock_dq(&gh);
483 out_uninit:
484 	gfs2_holder_uninit(&gh);
485 	if (ret == 0) {
486 		set_page_dirty(page);
487 		wait_for_stable_page(page);
488 	}
489 out_write_access:
490 	put_write_access(inode);
491 out:
492 	sb_end_pagefault(inode->i_sb);
493 	return block_page_mkwrite_return(ret);
494 }
495 
496 static const struct vm_operations_struct gfs2_vm_ops = {
497 	.fault = filemap_fault,
498 	.map_pages = filemap_map_pages,
499 	.page_mkwrite = gfs2_page_mkwrite,
500 };
501 
502 /**
503  * gfs2_mmap -
504  * @file: The file to map
505  * @vma: The VMA which described the mapping
506  *
507  * There is no need to get a lock here unless we should be updating
508  * atime. We ignore any locking errors since the only consequence is
509  * a missed atime update (which will just be deferred until later).
510  *
511  * Returns: 0
512  */
513 
514 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
515 {
516 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
517 
518 	if (!(file->f_flags & O_NOATIME) &&
519 	    !IS_NOATIME(&ip->i_inode)) {
520 		struct gfs2_holder i_gh;
521 		int error;
522 
523 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
524 					   &i_gh);
525 		if (error)
526 			return error;
527 		/* grab lock to update inode */
528 		gfs2_glock_dq_uninit(&i_gh);
529 		file_accessed(file);
530 	}
531 	vma->vm_ops = &gfs2_vm_ops;
532 
533 	return 0;
534 }
535 
536 /**
537  * gfs2_open_common - This is common to open and atomic_open
538  * @inode: The inode being opened
539  * @file: The file being opened
540  *
541  * This maybe called under a glock or not depending upon how it has
542  * been called. We must always be called under a glock for regular
543  * files, however. For other file types, it does not matter whether
544  * we hold the glock or not.
545  *
546  * Returns: Error code or 0 for success
547  */
548 
549 int gfs2_open_common(struct inode *inode, struct file *file)
550 {
551 	struct gfs2_file *fp;
552 	int ret;
553 
554 	if (S_ISREG(inode->i_mode)) {
555 		ret = generic_file_open(inode, file);
556 		if (ret)
557 			return ret;
558 	}
559 
560 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
561 	if (!fp)
562 		return -ENOMEM;
563 
564 	mutex_init(&fp->f_fl_mutex);
565 
566 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
567 	file->private_data = fp;
568 	return 0;
569 }
570 
571 /**
572  * gfs2_open - open a file
573  * @inode: the inode to open
574  * @file: the struct file for this opening
575  *
576  * After atomic_open, this function is only used for opening files
577  * which are already cached. We must still get the glock for regular
578  * files to ensure that we have the file size uptodate for the large
579  * file check which is in the common code. That is only an issue for
580  * regular files though.
581  *
582  * Returns: errno
583  */
584 
585 static int gfs2_open(struct inode *inode, struct file *file)
586 {
587 	struct gfs2_inode *ip = GFS2_I(inode);
588 	struct gfs2_holder i_gh;
589 	int error;
590 	bool need_unlock = false;
591 
592 	if (S_ISREG(ip->i_inode.i_mode)) {
593 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
594 					   &i_gh);
595 		if (error)
596 			return error;
597 		need_unlock = true;
598 	}
599 
600 	error = gfs2_open_common(inode, file);
601 
602 	if (need_unlock)
603 		gfs2_glock_dq_uninit(&i_gh);
604 
605 	return error;
606 }
607 
608 /**
609  * gfs2_release - called to close a struct file
610  * @inode: the inode the struct file belongs to
611  * @file: the struct file being closed
612  *
613  * Returns: errno
614  */
615 
616 static int gfs2_release(struct inode *inode, struct file *file)
617 {
618 	struct gfs2_inode *ip = GFS2_I(inode);
619 
620 	kfree(file->private_data);
621 	file->private_data = NULL;
622 
623 	if (!(file->f_mode & FMODE_WRITE))
624 		return 0;
625 
626 	gfs2_rs_delete(ip, &inode->i_writecount);
627 	return 0;
628 }
629 
630 /**
631  * gfs2_fsync - sync the dirty data for a file (across the cluster)
632  * @file: the file that points to the dentry
633  * @start: the start position in the file to sync
634  * @end: the end position in the file to sync
635  * @datasync: set if we can ignore timestamp changes
636  *
637  * We split the data flushing here so that we don't wait for the data
638  * until after we've also sent the metadata to disk. Note that for
639  * data=ordered, we will write & wait for the data at the log flush
640  * stage anyway, so this is unlikely to make much of a difference
641  * except in the data=writeback case.
642  *
643  * If the fdatawrite fails due to any reason except -EIO, we will
644  * continue the remainder of the fsync, although we'll still report
645  * the error at the end. This is to match filemap_write_and_wait_range()
646  * behaviour.
647  *
648  * Returns: errno
649  */
650 
651 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
652 		      int datasync)
653 {
654 	struct address_space *mapping = file->f_mapping;
655 	struct inode *inode = mapping->host;
656 	int sync_state = inode->i_state & I_DIRTY_ALL;
657 	struct gfs2_inode *ip = GFS2_I(inode);
658 	int ret = 0, ret1 = 0;
659 
660 	if (mapping->nrpages) {
661 		ret1 = filemap_fdatawrite_range(mapping, start, end);
662 		if (ret1 == -EIO)
663 			return ret1;
664 	}
665 
666 	if (!gfs2_is_jdata(ip))
667 		sync_state &= ~I_DIRTY_PAGES;
668 	if (datasync)
669 		sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
670 
671 	if (sync_state) {
672 		ret = sync_inode_metadata(inode, 1);
673 		if (ret)
674 			return ret;
675 		if (gfs2_is_jdata(ip))
676 			filemap_write_and_wait(mapping);
677 		gfs2_ail_flush(ip->i_gl, 1);
678 	}
679 
680 	if (mapping->nrpages)
681 		ret = filemap_fdatawait_range(mapping, start, end);
682 
683 	return ret ? ret : ret1;
684 }
685 
686 /**
687  * gfs2_file_write_iter - Perform a write to a file
688  * @iocb: The io context
689  * @iov: The data to write
690  * @nr_segs: Number of @iov segments
691  * @pos: The file position
692  *
693  * We have to do a lock/unlock here to refresh the inode size for
694  * O_APPEND writes, otherwise we can land up writing at the wrong
695  * offset. There is still a race, but provided the app is using its
696  * own file locking, this will make O_APPEND work as expected.
697  *
698  */
699 
700 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
701 {
702 	struct file *file = iocb->ki_filp;
703 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
704 	int ret;
705 
706 	ret = gfs2_rs_alloc(ip);
707 	if (ret)
708 		return ret;
709 
710 	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
711 
712 	if (iocb->ki_flags & IOCB_APPEND) {
713 		struct gfs2_holder gh;
714 
715 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
716 		if (ret)
717 			return ret;
718 		gfs2_glock_dq_uninit(&gh);
719 	}
720 
721 	return generic_file_write_iter(iocb, from);
722 }
723 
724 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
725 			   int mode)
726 {
727 	struct gfs2_inode *ip = GFS2_I(inode);
728 	struct buffer_head *dibh;
729 	int error;
730 	unsigned int nr_blks;
731 	sector_t lblock = offset >> inode->i_blkbits;
732 
733 	error = gfs2_meta_inode_buffer(ip, &dibh);
734 	if (unlikely(error))
735 		return error;
736 
737 	gfs2_trans_add_meta(ip->i_gl, dibh);
738 
739 	if (gfs2_is_stuffed(ip)) {
740 		error = gfs2_unstuff_dinode(ip, NULL);
741 		if (unlikely(error))
742 			goto out;
743 	}
744 
745 	while (len) {
746 		struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
747 		bh_map.b_size = len;
748 		set_buffer_zeronew(&bh_map);
749 
750 		error = gfs2_block_map(inode, lblock, &bh_map, 1);
751 		if (unlikely(error))
752 			goto out;
753 		len -= bh_map.b_size;
754 		nr_blks = bh_map.b_size >> inode->i_blkbits;
755 		lblock += nr_blks;
756 		if (!buffer_new(&bh_map))
757 			continue;
758 		if (unlikely(!buffer_zeronew(&bh_map))) {
759 			error = -EIO;
760 			goto out;
761 		}
762 	}
763 out:
764 	brelse(dibh);
765 	return error;
766 }
767 /**
768  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
769  *                     blocks, determine how many bytes can be written.
770  * @ip:          The inode in question.
771  * @len:         Max cap of bytes. What we return in *len must be <= this.
772  * @data_blocks: Compute and return the number of data blocks needed
773  * @ind_blocks:  Compute and return the number of indirect blocks needed
774  * @max_blocks:  The total blocks available to work with.
775  *
776  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
777  */
778 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
779 			    unsigned int *data_blocks, unsigned int *ind_blocks,
780 			    unsigned int max_blocks)
781 {
782 	loff_t max = *len;
783 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
784 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
785 
786 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
787 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
788 		max_data -= tmp;
789 	}
790 
791 	*data_blocks = max_data;
792 	*ind_blocks = max_blocks - max_data;
793 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
794 	if (*len > max) {
795 		*len = max;
796 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
797 	}
798 }
799 
800 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
801 {
802 	struct inode *inode = file_inode(file);
803 	struct gfs2_sbd *sdp = GFS2_SB(inode);
804 	struct gfs2_inode *ip = GFS2_I(inode);
805 	struct gfs2_alloc_parms ap = { .aflags = 0, };
806 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
807 	loff_t bytes, max_bytes, max_blks = UINT_MAX;
808 	int error;
809 	const loff_t pos = offset;
810 	const loff_t count = len;
811 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
812 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
813 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
814 
815 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
816 
817 	offset &= bsize_mask;
818 
819 	len = next - offset;
820 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
821 	if (!bytes)
822 		bytes = UINT_MAX;
823 	bytes &= bsize_mask;
824 	if (bytes == 0)
825 		bytes = sdp->sd_sb.sb_bsize;
826 
827 	gfs2_size_hint(file, offset, len);
828 
829 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
830 	ap.min_target = data_blocks + ind_blocks;
831 
832 	while (len > 0) {
833 		if (len < bytes)
834 			bytes = len;
835 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
836 			len -= bytes;
837 			offset += bytes;
838 			continue;
839 		}
840 
841 		/* We need to determine how many bytes we can actually
842 		 * fallocate without exceeding quota or going over the
843 		 * end of the fs. We start off optimistically by assuming
844 		 * we can write max_bytes */
845 		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
846 
847 		/* Since max_bytes is most likely a theoretical max, we
848 		 * calculate a more realistic 'bytes' to serve as a good
849 		 * starting point for the number of bytes we may be able
850 		 * to write */
851 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
852 		ap.target = data_blocks + ind_blocks;
853 
854 		error = gfs2_quota_lock_check(ip, &ap);
855 		if (error)
856 			return error;
857 		/* ap.allowed tells us how many blocks quota will allow
858 		 * us to write. Check if this reduces max_blks */
859 		if (ap.allowed && ap.allowed < max_blks)
860 			max_blks = ap.allowed;
861 
862 		error = gfs2_inplace_reserve(ip, &ap);
863 		if (error)
864 			goto out_qunlock;
865 
866 		/* check if the selected rgrp limits our max_blks further */
867 		if (ap.allowed && ap.allowed < max_blks)
868 			max_blks = ap.allowed;
869 
870 		/* Almost done. Calculate bytes that can be written using
871 		 * max_blks. We also recompute max_bytes, data_blocks and
872 		 * ind_blocks */
873 		calc_max_reserv(ip, &max_bytes, &data_blocks,
874 				&ind_blocks, max_blks);
875 
876 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
877 			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
878 		if (gfs2_is_jdata(ip))
879 			rblocks += data_blocks ? data_blocks : 1;
880 
881 		error = gfs2_trans_begin(sdp, rblocks,
882 					 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
883 		if (error)
884 			goto out_trans_fail;
885 
886 		error = fallocate_chunk(inode, offset, max_bytes, mode);
887 		gfs2_trans_end(sdp);
888 
889 		if (error)
890 			goto out_trans_fail;
891 
892 		len -= max_bytes;
893 		offset += max_bytes;
894 		gfs2_inplace_release(ip);
895 		gfs2_quota_unlock(ip);
896 	}
897 
898 	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) {
899 		i_size_write(inode, pos + count);
900 		/* Marks the inode as dirty */
901 		file_update_time(file);
902 	}
903 
904 	return generic_write_sync(file, pos, count);
905 
906 out_trans_fail:
907 	gfs2_inplace_release(ip);
908 out_qunlock:
909 	gfs2_quota_unlock(ip);
910 	return error;
911 }
912 
913 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
914 {
915 	struct inode *inode = file_inode(file);
916 	struct gfs2_inode *ip = GFS2_I(inode);
917 	struct gfs2_holder gh;
918 	int ret;
919 
920 	if ((mode & ~FALLOC_FL_KEEP_SIZE) || gfs2_is_jdata(ip))
921 		return -EOPNOTSUPP;
922 
923 	mutex_lock(&inode->i_mutex);
924 
925 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
926 	ret = gfs2_glock_nq(&gh);
927 	if (ret)
928 		goto out_uninit;
929 
930 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
931 	    (offset + len) > inode->i_size) {
932 		ret = inode_newsize_ok(inode, offset + len);
933 		if (ret)
934 			goto out_unlock;
935 	}
936 
937 	ret = get_write_access(inode);
938 	if (ret)
939 		goto out_unlock;
940 
941 	ret = gfs2_rs_alloc(ip);
942 	if (ret)
943 		goto out_putw;
944 
945 	ret = __gfs2_fallocate(file, mode, offset, len);
946 	if (ret)
947 		gfs2_rs_deltree(ip->i_res);
948 out_putw:
949 	put_write_access(inode);
950 out_unlock:
951 	gfs2_glock_dq(&gh);
952 out_uninit:
953 	gfs2_holder_uninit(&gh);
954 	mutex_unlock(&inode->i_mutex);
955 	return ret;
956 }
957 
958 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
959 				      struct file *out, loff_t *ppos,
960 				      size_t len, unsigned int flags)
961 {
962 	int error;
963 	struct gfs2_inode *ip = GFS2_I(out->f_mapping->host);
964 
965 	error = gfs2_rs_alloc(ip);
966 	if (error)
967 		return (ssize_t)error;
968 
969 	gfs2_size_hint(out, *ppos, len);
970 
971 	return iter_file_splice_write(pipe, out, ppos, len, flags);
972 }
973 
974 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
975 
976 /**
977  * gfs2_lock - acquire/release a posix lock on a file
978  * @file: the file pointer
979  * @cmd: either modify or retrieve lock state, possibly wait
980  * @fl: type and range of lock
981  *
982  * Returns: errno
983  */
984 
985 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
986 {
987 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
988 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
989 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
990 
991 	if (!(fl->fl_flags & FL_POSIX))
992 		return -ENOLCK;
993 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
994 		return -ENOLCK;
995 
996 	if (cmd == F_CANCELLK) {
997 		/* Hack: */
998 		cmd = F_SETLK;
999 		fl->fl_type = F_UNLCK;
1000 	}
1001 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1002 		if (fl->fl_type == F_UNLCK)
1003 			posix_lock_file_wait(file, fl);
1004 		return -EIO;
1005 	}
1006 	if (IS_GETLK(cmd))
1007 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1008 	else if (fl->fl_type == F_UNLCK)
1009 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1010 	else
1011 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1012 }
1013 
1014 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1015 {
1016 	struct gfs2_file *fp = file->private_data;
1017 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1018 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1019 	struct gfs2_glock *gl;
1020 	unsigned int state;
1021 	int flags;
1022 	int error = 0;
1023 	int sleeptime;
1024 
1025 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1026 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1027 
1028 	mutex_lock(&fp->f_fl_mutex);
1029 
1030 	gl = fl_gh->gh_gl;
1031 	if (gl) {
1032 		if (fl_gh->gh_state == state)
1033 			goto out;
1034 		flock_lock_file_wait(file,
1035 				     &(struct file_lock){.fl_type = F_UNLCK});
1036 		gfs2_glock_dq(fl_gh);
1037 		gfs2_holder_reinit(state, flags, fl_gh);
1038 	} else {
1039 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1040 				       &gfs2_flock_glops, CREATE, &gl);
1041 		if (error)
1042 			goto out;
1043 		gfs2_holder_init(gl, state, flags, fl_gh);
1044 		gfs2_glock_put(gl);
1045 	}
1046 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1047 		error = gfs2_glock_nq(fl_gh);
1048 		if (error != GLR_TRYFAILED)
1049 			break;
1050 		fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1051 		fl_gh->gh_error = 0;
1052 		msleep(sleeptime);
1053 	}
1054 	if (error) {
1055 		gfs2_holder_uninit(fl_gh);
1056 		if (error == GLR_TRYFAILED)
1057 			error = -EAGAIN;
1058 	} else {
1059 		error = flock_lock_file_wait(file, fl);
1060 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1061 	}
1062 
1063 out:
1064 	mutex_unlock(&fp->f_fl_mutex);
1065 	return error;
1066 }
1067 
1068 static void do_unflock(struct file *file, struct file_lock *fl)
1069 {
1070 	struct gfs2_file *fp = file->private_data;
1071 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1072 
1073 	mutex_lock(&fp->f_fl_mutex);
1074 	flock_lock_file_wait(file, fl);
1075 	if (fl_gh->gh_gl) {
1076 		gfs2_glock_dq(fl_gh);
1077 		gfs2_holder_uninit(fl_gh);
1078 	}
1079 	mutex_unlock(&fp->f_fl_mutex);
1080 }
1081 
1082 /**
1083  * gfs2_flock - acquire/release a flock lock on a file
1084  * @file: the file pointer
1085  * @cmd: either modify or retrieve lock state, possibly wait
1086  * @fl: type and range of lock
1087  *
1088  * Returns: errno
1089  */
1090 
1091 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1092 {
1093 	if (!(fl->fl_flags & FL_FLOCK))
1094 		return -ENOLCK;
1095 	if (fl->fl_type & LOCK_MAND)
1096 		return -EOPNOTSUPP;
1097 
1098 	if (fl->fl_type == F_UNLCK) {
1099 		do_unflock(file, fl);
1100 		return 0;
1101 	} else {
1102 		return do_flock(file, cmd, fl);
1103 	}
1104 }
1105 
1106 const struct file_operations gfs2_file_fops = {
1107 	.llseek		= gfs2_llseek,
1108 	.read_iter	= generic_file_read_iter,
1109 	.write_iter	= gfs2_file_write_iter,
1110 	.unlocked_ioctl	= gfs2_ioctl,
1111 	.mmap		= gfs2_mmap,
1112 	.open		= gfs2_open,
1113 	.release	= gfs2_release,
1114 	.fsync		= gfs2_fsync,
1115 	.lock		= gfs2_lock,
1116 	.flock		= gfs2_flock,
1117 	.splice_read	= generic_file_splice_read,
1118 	.splice_write	= gfs2_file_splice_write,
1119 	.setlease	= simple_nosetlease,
1120 	.fallocate	= gfs2_fallocate,
1121 };
1122 
1123 const struct file_operations gfs2_dir_fops = {
1124 	.iterate	= gfs2_readdir,
1125 	.unlocked_ioctl	= gfs2_ioctl,
1126 	.open		= gfs2_open,
1127 	.release	= gfs2_release,
1128 	.fsync		= gfs2_fsync,
1129 	.lock		= gfs2_lock,
1130 	.flock		= gfs2_flock,
1131 	.llseek		= default_llseek,
1132 };
1133 
1134 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1135 
1136 const struct file_operations gfs2_file_fops_nolock = {
1137 	.llseek		= gfs2_llseek,
1138 	.read_iter	= generic_file_read_iter,
1139 	.write_iter	= gfs2_file_write_iter,
1140 	.unlocked_ioctl	= gfs2_ioctl,
1141 	.mmap		= gfs2_mmap,
1142 	.open		= gfs2_open,
1143 	.release	= gfs2_release,
1144 	.fsync		= gfs2_fsync,
1145 	.splice_read	= generic_file_splice_read,
1146 	.splice_write	= gfs2_file_splice_write,
1147 	.setlease	= generic_setlease,
1148 	.fallocate	= gfs2_fallocate,
1149 };
1150 
1151 const struct file_operations gfs2_dir_fops_nolock = {
1152 	.iterate	= gfs2_readdir,
1153 	.unlocked_ioctl	= gfs2_ioctl,
1154 	.open		= gfs2_open,
1155 	.release	= gfs2_release,
1156 	.fsync		= gfs2_fsync,
1157 	.llseek		= default_llseek,
1158 };
1159 
1160