xref: /linux/fs/gfs2/file.c (revision ec2212088c42ff7d1362629ec26dda4f3e8bdad3)
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/ext2_fs.h>
22 #include <linux/falloc.h>
23 #include <linux/swap.h>
24 #include <linux/crc32.h>
25 #include <linux/writeback.h>
26 #include <asm/uaccess.h>
27 #include <linux/dlm.h>
28 #include <linux/dlm_plock.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  * @origin: 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 origin)
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 (origin) {
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, origin);
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, origin);
76 		break;
77 	default:
78 		error = -EINVAL;
79 	}
80 
81 	return error;
82 }
83 
84 /**
85  * gfs2_readdir - Read directory entries from a directory
86  * @file: The directory to read from
87  * @dirent: Buffer for dirents
88  * @filldir: Function used to do the copying
89  *
90  * Returns: errno
91  */
92 
93 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
94 {
95 	struct inode *dir = file->f_mapping->host;
96 	struct gfs2_inode *dip = GFS2_I(dir);
97 	struct gfs2_holder d_gh;
98 	u64 offset = file->f_pos;
99 	int error;
100 
101 	gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
102 	error = gfs2_glock_nq(&d_gh);
103 	if (error) {
104 		gfs2_holder_uninit(&d_gh);
105 		return error;
106 	}
107 
108 	error = gfs2_dir_read(dir, &offset, dirent, filldir, &file->f_ra);
109 
110 	gfs2_glock_dq_uninit(&d_gh);
111 
112 	file->f_pos = offset;
113 
114 	return error;
115 }
116 
117 /**
118  * fsflags_cvt
119  * @table: A table of 32 u32 flags
120  * @val: a 32 bit value to convert
121  *
122  * This function can be used to convert between fsflags values and
123  * GFS2's own flags values.
124  *
125  * Returns: the converted flags
126  */
127 static u32 fsflags_cvt(const u32 *table, u32 val)
128 {
129 	u32 res = 0;
130 	while(val) {
131 		if (val & 1)
132 			res |= *table;
133 		table++;
134 		val >>= 1;
135 	}
136 	return res;
137 }
138 
139 static const u32 fsflags_to_gfs2[32] = {
140 	[3] = GFS2_DIF_SYNC,
141 	[4] = GFS2_DIF_IMMUTABLE,
142 	[5] = GFS2_DIF_APPENDONLY,
143 	[7] = GFS2_DIF_NOATIME,
144 	[12] = GFS2_DIF_EXHASH,
145 	[14] = GFS2_DIF_INHERIT_JDATA,
146 };
147 
148 static const u32 gfs2_to_fsflags[32] = {
149 	[gfs2fl_Sync] = FS_SYNC_FL,
150 	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
151 	[gfs2fl_AppendOnly] = FS_APPEND_FL,
152 	[gfs2fl_NoAtime] = FS_NOATIME_FL,
153 	[gfs2fl_ExHash] = FS_INDEX_FL,
154 	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
155 };
156 
157 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
158 {
159 	struct inode *inode = filp->f_path.dentry->d_inode;
160 	struct gfs2_inode *ip = GFS2_I(inode);
161 	struct gfs2_holder gh;
162 	int error;
163 	u32 fsflags;
164 
165 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
166 	error = gfs2_glock_nq(&gh);
167 	if (error)
168 		return error;
169 
170 	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
171 	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
172 		fsflags |= FS_JOURNAL_DATA_FL;
173 	if (put_user(fsflags, ptr))
174 		error = -EFAULT;
175 
176 	gfs2_glock_dq(&gh);
177 	gfs2_holder_uninit(&gh);
178 	return error;
179 }
180 
181 void gfs2_set_inode_flags(struct inode *inode)
182 {
183 	struct gfs2_inode *ip = GFS2_I(inode);
184 	unsigned int flags = inode->i_flags;
185 
186 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
187 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
188 		inode->i_flags |= S_NOSEC;
189 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
190 		flags |= S_IMMUTABLE;
191 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
192 		flags |= S_APPEND;
193 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
194 		flags |= S_NOATIME;
195 	if (ip->i_diskflags & GFS2_DIF_SYNC)
196 		flags |= S_SYNC;
197 	inode->i_flags = flags;
198 }
199 
200 /* Flags that can be set by user space */
201 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
202 			     GFS2_DIF_IMMUTABLE|		\
203 			     GFS2_DIF_APPENDONLY|		\
204 			     GFS2_DIF_NOATIME|			\
205 			     GFS2_DIF_SYNC|			\
206 			     GFS2_DIF_SYSTEM|			\
207 			     GFS2_DIF_INHERIT_JDATA)
208 
209 /**
210  * gfs2_set_flags - set flags on an inode
211  * @inode: The inode
212  * @flags: The flags to set
213  * @mask: Indicates which flags are valid
214  *
215  */
216 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
217 {
218 	struct inode *inode = filp->f_path.dentry->d_inode;
219 	struct gfs2_inode *ip = GFS2_I(inode);
220 	struct gfs2_sbd *sdp = GFS2_SB(inode);
221 	struct buffer_head *bh;
222 	struct gfs2_holder gh;
223 	int error;
224 	u32 new_flags, flags;
225 
226 	error = mnt_want_write_file(filp);
227 	if (error)
228 		return error;
229 
230 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
231 	if (error)
232 		goto out_drop_write;
233 
234 	error = -EACCES;
235 	if (!inode_owner_or_capable(inode))
236 		goto out;
237 
238 	error = 0;
239 	flags = ip->i_diskflags;
240 	new_flags = (flags & ~mask) | (reqflags & mask);
241 	if ((new_flags ^ flags) == 0)
242 		goto out;
243 
244 	error = -EINVAL;
245 	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
246 		goto out;
247 
248 	error = -EPERM;
249 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
250 		goto out;
251 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
252 		goto out;
253 	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
254 	    !capable(CAP_LINUX_IMMUTABLE))
255 		goto out;
256 	if (!IS_IMMUTABLE(inode)) {
257 		error = gfs2_permission(inode, MAY_WRITE);
258 		if (error)
259 			goto out;
260 	}
261 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
262 		if (flags & GFS2_DIF_JDATA)
263 			gfs2_log_flush(sdp, ip->i_gl);
264 		error = filemap_fdatawrite(inode->i_mapping);
265 		if (error)
266 			goto out;
267 		error = filemap_fdatawait(inode->i_mapping);
268 		if (error)
269 			goto out;
270 	}
271 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
272 	if (error)
273 		goto out;
274 	error = gfs2_meta_inode_buffer(ip, &bh);
275 	if (error)
276 		goto out_trans_end;
277 	gfs2_trans_add_bh(ip->i_gl, bh, 1);
278 	ip->i_diskflags = new_flags;
279 	gfs2_dinode_out(ip, bh->b_data);
280 	brelse(bh);
281 	gfs2_set_inode_flags(inode);
282 	gfs2_set_aops(inode);
283 out_trans_end:
284 	gfs2_trans_end(sdp);
285 out:
286 	gfs2_glock_dq_uninit(&gh);
287 out_drop_write:
288 	mnt_drop_write_file(filp);
289 	return error;
290 }
291 
292 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
293 {
294 	struct inode *inode = filp->f_path.dentry->d_inode;
295 	u32 fsflags, gfsflags;
296 
297 	if (get_user(fsflags, ptr))
298 		return -EFAULT;
299 
300 	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
301 	if (!S_ISDIR(inode->i_mode)) {
302 		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
303 			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
304 		return do_gfs2_set_flags(filp, gfsflags, ~0);
305 	}
306 	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
307 }
308 
309 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
310 {
311 	switch(cmd) {
312 	case FS_IOC_GETFLAGS:
313 		return gfs2_get_flags(filp, (u32 __user *)arg);
314 	case FS_IOC_SETFLAGS:
315 		return gfs2_set_flags(filp, (u32 __user *)arg);
316 	case FITRIM:
317 		return gfs2_fitrim(filp, (void __user *)arg);
318 	}
319 	return -ENOTTY;
320 }
321 
322 /**
323  * gfs2_allocate_page_backing - Use bmap to allocate blocks
324  * @page: The (locked) page to allocate backing for
325  *
326  * We try to allocate all the blocks required for the page in
327  * one go. This might fail for various reasons, so we keep
328  * trying until all the blocks to back this page are allocated.
329  * If some of the blocks are already allocated, thats ok too.
330  */
331 
332 static int gfs2_allocate_page_backing(struct page *page)
333 {
334 	struct inode *inode = page->mapping->host;
335 	struct buffer_head bh;
336 	unsigned long size = PAGE_CACHE_SIZE;
337 	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
338 
339 	do {
340 		bh.b_state = 0;
341 		bh.b_size = size;
342 		gfs2_block_map(inode, lblock, &bh, 1);
343 		if (!buffer_mapped(&bh))
344 			return -EIO;
345 		size -= bh.b_size;
346 		lblock += (bh.b_size >> inode->i_blkbits);
347 	} while(size > 0);
348 	return 0;
349 }
350 
351 /**
352  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
353  * @vma: The virtual memory area
354  * @page: The page which is about to become writable
355  *
356  * When the page becomes writable, we need to ensure that we have
357  * blocks allocated on disk to back that page.
358  */
359 
360 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
361 {
362 	struct page *page = vmf->page;
363 	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
364 	struct gfs2_inode *ip = GFS2_I(inode);
365 	struct gfs2_sbd *sdp = GFS2_SB(inode);
366 	unsigned long last_index;
367 	u64 pos = page->index << PAGE_CACHE_SHIFT;
368 	unsigned int data_blocks, ind_blocks, rblocks;
369 	struct gfs2_holder gh;
370 	struct gfs2_qadata *qa;
371 	loff_t size;
372 	int ret;
373 
374 	/* Wait if fs is frozen. This is racy so we check again later on
375 	 * and retry if the fs has been frozen after the page lock has
376 	 * been acquired
377 	 */
378 	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
379 
380 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
381 	ret = gfs2_glock_nq(&gh);
382 	if (ret)
383 		goto out;
384 
385 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
386 	set_bit(GIF_SW_PAGED, &ip->i_flags);
387 
388 	if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) {
389 		lock_page(page);
390 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
391 			ret = -EAGAIN;
392 			unlock_page(page);
393 		}
394 		goto out_unlock;
395 	}
396 
397 	ret = -ENOMEM;
398 	qa = gfs2_qadata_get(ip);
399 	if (qa == NULL)
400 		goto out_unlock;
401 
402 	ret = gfs2_quota_lock_check(ip);
403 	if (ret)
404 		goto out_alloc_put;
405 	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
406 	ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
407 	if (ret)
408 		goto out_quota_unlock;
409 
410 	rblocks = RES_DINODE + ind_blocks;
411 	if (gfs2_is_jdata(ip))
412 		rblocks += data_blocks ? data_blocks : 1;
413 	if (ind_blocks || data_blocks) {
414 		rblocks += RES_STATFS + RES_QUOTA;
415 		rblocks += gfs2_rg_blocks(ip);
416 	}
417 	ret = gfs2_trans_begin(sdp, rblocks, 0);
418 	if (ret)
419 		goto out_trans_fail;
420 
421 	lock_page(page);
422 	ret = -EINVAL;
423 	size = i_size_read(inode);
424 	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
425 	/* Check page index against inode size */
426 	if (size == 0 || (page->index > last_index))
427 		goto out_trans_end;
428 
429 	ret = -EAGAIN;
430 	/* If truncated, we must retry the operation, we may have raced
431 	 * with the glock demotion code.
432 	 */
433 	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
434 		goto out_trans_end;
435 
436 	/* Unstuff, if required, and allocate backing blocks for page */
437 	ret = 0;
438 	if (gfs2_is_stuffed(ip))
439 		ret = gfs2_unstuff_dinode(ip, page);
440 	if (ret == 0)
441 		ret = gfs2_allocate_page_backing(page);
442 
443 out_trans_end:
444 	if (ret)
445 		unlock_page(page);
446 	gfs2_trans_end(sdp);
447 out_trans_fail:
448 	gfs2_inplace_release(ip);
449 out_quota_unlock:
450 	gfs2_quota_unlock(ip);
451 out_alloc_put:
452 	gfs2_qadata_put(ip);
453 out_unlock:
454 	gfs2_glock_dq(&gh);
455 out:
456 	gfs2_holder_uninit(&gh);
457 	if (ret == 0) {
458 		set_page_dirty(page);
459 		/* This check must be post dropping of transaction lock */
460 		if (inode->i_sb->s_frozen == SB_UNFROZEN) {
461 			wait_on_page_writeback(page);
462 		} else {
463 			ret = -EAGAIN;
464 			unlock_page(page);
465 		}
466 	}
467 	return block_page_mkwrite_return(ret);
468 }
469 
470 static const struct vm_operations_struct gfs2_vm_ops = {
471 	.fault = filemap_fault,
472 	.page_mkwrite = gfs2_page_mkwrite,
473 };
474 
475 /**
476  * gfs2_mmap -
477  * @file: The file to map
478  * @vma: The VMA which described the mapping
479  *
480  * There is no need to get a lock here unless we should be updating
481  * atime. We ignore any locking errors since the only consequence is
482  * a missed atime update (which will just be deferred until later).
483  *
484  * Returns: 0
485  */
486 
487 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
488 {
489 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
490 
491 	if (!(file->f_flags & O_NOATIME) &&
492 	    !IS_NOATIME(&ip->i_inode)) {
493 		struct gfs2_holder i_gh;
494 		int error;
495 
496 		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
497 		error = gfs2_glock_nq(&i_gh);
498 		if (error == 0) {
499 			file_accessed(file);
500 			gfs2_glock_dq(&i_gh);
501 		}
502 		gfs2_holder_uninit(&i_gh);
503 		if (error)
504 			return error;
505 	}
506 	vma->vm_ops = &gfs2_vm_ops;
507 	vma->vm_flags |= VM_CAN_NONLINEAR;
508 
509 	return 0;
510 }
511 
512 /**
513  * gfs2_open - open a file
514  * @inode: the inode to open
515  * @file: the struct file for this opening
516  *
517  * Returns: errno
518  */
519 
520 static int gfs2_open(struct inode *inode, struct file *file)
521 {
522 	struct gfs2_inode *ip = GFS2_I(inode);
523 	struct gfs2_holder i_gh;
524 	struct gfs2_file *fp;
525 	int error;
526 
527 	fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
528 	if (!fp)
529 		return -ENOMEM;
530 
531 	mutex_init(&fp->f_fl_mutex);
532 
533 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
534 	file->private_data = fp;
535 
536 	if (S_ISREG(ip->i_inode.i_mode)) {
537 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
538 					   &i_gh);
539 		if (error)
540 			goto fail;
541 
542 		if (!(file->f_flags & O_LARGEFILE) &&
543 		    i_size_read(inode) > MAX_NON_LFS) {
544 			error = -EOVERFLOW;
545 			goto fail_gunlock;
546 		}
547 
548 		gfs2_glock_dq_uninit(&i_gh);
549 	}
550 
551 	return 0;
552 
553 fail_gunlock:
554 	gfs2_glock_dq_uninit(&i_gh);
555 fail:
556 	file->private_data = NULL;
557 	kfree(fp);
558 	return error;
559 }
560 
561 /**
562  * gfs2_close - called to close a struct file
563  * @inode: the inode the struct file belongs to
564  * @file: the struct file being closed
565  *
566  * Returns: errno
567  */
568 
569 static int gfs2_close(struct inode *inode, struct file *file)
570 {
571 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
572 	struct gfs2_file *fp;
573 
574 	fp = file->private_data;
575 	file->private_data = NULL;
576 
577 	if (gfs2_assert_warn(sdp, fp))
578 		return -EIO;
579 
580 	kfree(fp);
581 
582 	return 0;
583 }
584 
585 /**
586  * gfs2_fsync - sync the dirty data for a file (across the cluster)
587  * @file: the file that points to the dentry
588  * @start: the start position in the file to sync
589  * @end: the end position in the file to sync
590  * @datasync: set if we can ignore timestamp changes
591  *
592  * We split the data flushing here so that we don't wait for the data
593  * until after we've also sent the metadata to disk. Note that for
594  * data=ordered, we will write & wait for the data at the log flush
595  * stage anyway, so this is unlikely to make much of a difference
596  * except in the data=writeback case.
597  *
598  * If the fdatawrite fails due to any reason except -EIO, we will
599  * continue the remainder of the fsync, although we'll still report
600  * the error at the end. This is to match filemap_write_and_wait_range()
601  * behaviour.
602  *
603  * Returns: errno
604  */
605 
606 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
607 		      int datasync)
608 {
609 	struct address_space *mapping = file->f_mapping;
610 	struct inode *inode = mapping->host;
611 	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
612 	struct gfs2_inode *ip = GFS2_I(inode);
613 	int ret = 0, ret1 = 0;
614 
615 	if (mapping->nrpages) {
616 		ret1 = filemap_fdatawrite_range(mapping, start, end);
617 		if (ret1 == -EIO)
618 			return ret1;
619 	}
620 
621 	if (datasync)
622 		sync_state &= ~I_DIRTY_SYNC;
623 
624 	if (sync_state) {
625 		ret = sync_inode_metadata(inode, 1);
626 		if (ret)
627 			return ret;
628 		if (gfs2_is_jdata(ip))
629 			filemap_write_and_wait(mapping);
630 		gfs2_ail_flush(ip->i_gl, 1);
631 	}
632 
633 	if (mapping->nrpages)
634 		ret = filemap_fdatawait_range(mapping, start, end);
635 
636 	return ret ? ret : ret1;
637 }
638 
639 /**
640  * gfs2_file_aio_write - Perform a write to a file
641  * @iocb: The io context
642  * @iov: The data to write
643  * @nr_segs: Number of @iov segments
644  * @pos: The file position
645  *
646  * We have to do a lock/unlock here to refresh the inode size for
647  * O_APPEND writes, otherwise we can land up writing at the wrong
648  * offset. There is still a race, but provided the app is using its
649  * own file locking, this will make O_APPEND work as expected.
650  *
651  */
652 
653 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
654 				   unsigned long nr_segs, loff_t pos)
655 {
656 	struct file *file = iocb->ki_filp;
657 
658 	if (file->f_flags & O_APPEND) {
659 		struct dentry *dentry = file->f_dentry;
660 		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
661 		struct gfs2_holder gh;
662 		int ret;
663 
664 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
665 		if (ret)
666 			return ret;
667 		gfs2_glock_dq_uninit(&gh);
668 	}
669 
670 	return generic_file_aio_write(iocb, iov, nr_segs, pos);
671 }
672 
673 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
674 			   int mode)
675 {
676 	struct gfs2_inode *ip = GFS2_I(inode);
677 	struct buffer_head *dibh;
678 	int error;
679 	loff_t size = len;
680 	unsigned int nr_blks;
681 	sector_t lblock = offset >> inode->i_blkbits;
682 
683 	error = gfs2_meta_inode_buffer(ip, &dibh);
684 	if (unlikely(error))
685 		return error;
686 
687 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
688 
689 	if (gfs2_is_stuffed(ip)) {
690 		error = gfs2_unstuff_dinode(ip, NULL);
691 		if (unlikely(error))
692 			goto out;
693 	}
694 
695 	while (len) {
696 		struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
697 		bh_map.b_size = len;
698 		set_buffer_zeronew(&bh_map);
699 
700 		error = gfs2_block_map(inode, lblock, &bh_map, 1);
701 		if (unlikely(error))
702 			goto out;
703 		len -= bh_map.b_size;
704 		nr_blks = bh_map.b_size >> inode->i_blkbits;
705 		lblock += nr_blks;
706 		if (!buffer_new(&bh_map))
707 			continue;
708 		if (unlikely(!buffer_zeronew(&bh_map))) {
709 			error = -EIO;
710 			goto out;
711 		}
712 	}
713 	if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE))
714 		i_size_write(inode, offset + size);
715 
716 	mark_inode_dirty(inode);
717 
718 out:
719 	brelse(dibh);
720 	return error;
721 }
722 
723 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
724 			    unsigned int *data_blocks, unsigned int *ind_blocks)
725 {
726 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
727 	unsigned int max_blocks = ip->i_rgd->rd_free_clone;
728 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
729 
730 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
731 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
732 		max_data -= tmp;
733 	}
734 	/* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
735 	   so it might end up with fewer data blocks */
736 	if (max_data <= *data_blocks)
737 		return;
738 	*data_blocks = max_data;
739 	*ind_blocks = max_blocks - max_data;
740 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
741 	if (*len > max) {
742 		*len = max;
743 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
744 	}
745 }
746 
747 static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
748 			   loff_t len)
749 {
750 	struct inode *inode = file->f_path.dentry->d_inode;
751 	struct gfs2_sbd *sdp = GFS2_SB(inode);
752 	struct gfs2_inode *ip = GFS2_I(inode);
753 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
754 	loff_t bytes, max_bytes;
755 	struct gfs2_qadata *qa;
756 	int error;
757 	const loff_t pos = offset;
758 	const loff_t count = len;
759 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
760 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
761 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
762 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
763 
764 	/* We only support the FALLOC_FL_KEEP_SIZE mode */
765 	if (mode & ~FALLOC_FL_KEEP_SIZE)
766 		return -EOPNOTSUPP;
767 
768 	offset &= bsize_mask;
769 
770 	len = next - offset;
771 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
772 	if (!bytes)
773 		bytes = UINT_MAX;
774 	bytes &= bsize_mask;
775 	if (bytes == 0)
776 		bytes = sdp->sd_sb.sb_bsize;
777 
778 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
779 	error = gfs2_glock_nq(&ip->i_gh);
780 	if (unlikely(error))
781 		goto out_uninit;
782 
783 	while (len > 0) {
784 		if (len < bytes)
785 			bytes = len;
786 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
787 			len -= bytes;
788 			offset += bytes;
789 			continue;
790 		}
791 		qa = gfs2_qadata_get(ip);
792 		if (!qa) {
793 			error = -ENOMEM;
794 			goto out_unlock;
795 		}
796 
797 		error = gfs2_quota_lock_check(ip);
798 		if (error)
799 			goto out_alloc_put;
800 
801 retry:
802 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
803 
804 		error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
805 		if (error) {
806 			if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
807 				bytes >>= 1;
808 				bytes &= bsize_mask;
809 				if (bytes == 0)
810 					bytes = sdp->sd_sb.sb_bsize;
811 				goto retry;
812 			}
813 			goto out_qunlock;
814 		}
815 		max_bytes = bytes;
816 		calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len,
817 				&max_bytes, &data_blocks, &ind_blocks);
818 
819 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
820 			  RES_RG_HDR + gfs2_rg_blocks(ip);
821 		if (gfs2_is_jdata(ip))
822 			rblocks += data_blocks ? data_blocks : 1;
823 
824 		error = gfs2_trans_begin(sdp, rblocks,
825 					 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
826 		if (error)
827 			goto out_trans_fail;
828 
829 		error = fallocate_chunk(inode, offset, max_bytes, mode);
830 		gfs2_trans_end(sdp);
831 
832 		if (error)
833 			goto out_trans_fail;
834 
835 		len -= max_bytes;
836 		offset += max_bytes;
837 		gfs2_inplace_release(ip);
838 		gfs2_quota_unlock(ip);
839 		gfs2_qadata_put(ip);
840 	}
841 
842 	if (error == 0)
843 		error = generic_write_sync(file, pos, count);
844 	goto out_unlock;
845 
846 out_trans_fail:
847 	gfs2_inplace_release(ip);
848 out_qunlock:
849 	gfs2_quota_unlock(ip);
850 out_alloc_put:
851 	gfs2_qadata_put(ip);
852 out_unlock:
853 	gfs2_glock_dq(&ip->i_gh);
854 out_uninit:
855 	gfs2_holder_uninit(&ip->i_gh);
856 	return error;
857 }
858 
859 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
860 
861 /**
862  * gfs2_setlease - acquire/release a file lease
863  * @file: the file pointer
864  * @arg: lease type
865  * @fl: file lock
866  *
867  * We don't currently have a way to enforce a lease across the whole
868  * cluster; until we do, disable leases (by just returning -EINVAL),
869  * unless the administrator has requested purely local locking.
870  *
871  * Locking: called under lock_flocks
872  *
873  * Returns: errno
874  */
875 
876 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
877 {
878 	return -EINVAL;
879 }
880 
881 /**
882  * gfs2_lock - acquire/release a posix lock on a file
883  * @file: the file pointer
884  * @cmd: either modify or retrieve lock state, possibly wait
885  * @fl: type and range of lock
886  *
887  * Returns: errno
888  */
889 
890 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
891 {
892 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
893 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
894 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
895 
896 	if (!(fl->fl_flags & FL_POSIX))
897 		return -ENOLCK;
898 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
899 		return -ENOLCK;
900 
901 	if (cmd == F_CANCELLK) {
902 		/* Hack: */
903 		cmd = F_SETLK;
904 		fl->fl_type = F_UNLCK;
905 	}
906 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
907 		return -EIO;
908 	if (IS_GETLK(cmd))
909 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
910 	else if (fl->fl_type == F_UNLCK)
911 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
912 	else
913 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
914 }
915 
916 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
917 {
918 	struct gfs2_file *fp = file->private_data;
919 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
920 	struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
921 	struct gfs2_glock *gl;
922 	unsigned int state;
923 	int flags;
924 	int error = 0;
925 
926 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
927 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
928 
929 	mutex_lock(&fp->f_fl_mutex);
930 
931 	gl = fl_gh->gh_gl;
932 	if (gl) {
933 		if (fl_gh->gh_state == state)
934 			goto out;
935 		flock_lock_file_wait(file,
936 				     &(struct file_lock){.fl_type = F_UNLCK});
937 		gfs2_glock_dq_wait(fl_gh);
938 		gfs2_holder_reinit(state, flags, fl_gh);
939 	} else {
940 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
941 				       &gfs2_flock_glops, CREATE, &gl);
942 		if (error)
943 			goto out;
944 		gfs2_holder_init(gl, state, flags, fl_gh);
945 		gfs2_glock_put(gl);
946 	}
947 	error = gfs2_glock_nq(fl_gh);
948 	if (error) {
949 		gfs2_holder_uninit(fl_gh);
950 		if (error == GLR_TRYFAILED)
951 			error = -EAGAIN;
952 	} else {
953 		error = flock_lock_file_wait(file, fl);
954 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
955 	}
956 
957 out:
958 	mutex_unlock(&fp->f_fl_mutex);
959 	return error;
960 }
961 
962 static void do_unflock(struct file *file, struct file_lock *fl)
963 {
964 	struct gfs2_file *fp = file->private_data;
965 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
966 
967 	mutex_lock(&fp->f_fl_mutex);
968 	flock_lock_file_wait(file, fl);
969 	if (fl_gh->gh_gl) {
970 		gfs2_glock_dq_wait(fl_gh);
971 		gfs2_holder_uninit(fl_gh);
972 	}
973 	mutex_unlock(&fp->f_fl_mutex);
974 }
975 
976 /**
977  * gfs2_flock - acquire/release a flock 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_flock(struct file *file, int cmd, struct file_lock *fl)
986 {
987 	if (!(fl->fl_flags & FL_FLOCK))
988 		return -ENOLCK;
989 	if (fl->fl_type & LOCK_MAND)
990 		return -EOPNOTSUPP;
991 
992 	if (fl->fl_type == F_UNLCK) {
993 		do_unflock(file, fl);
994 		return 0;
995 	} else {
996 		return do_flock(file, cmd, fl);
997 	}
998 }
999 
1000 const struct file_operations gfs2_file_fops = {
1001 	.llseek		= gfs2_llseek,
1002 	.read		= do_sync_read,
1003 	.aio_read	= generic_file_aio_read,
1004 	.write		= do_sync_write,
1005 	.aio_write	= gfs2_file_aio_write,
1006 	.unlocked_ioctl	= gfs2_ioctl,
1007 	.mmap		= gfs2_mmap,
1008 	.open		= gfs2_open,
1009 	.release	= gfs2_close,
1010 	.fsync		= gfs2_fsync,
1011 	.lock		= gfs2_lock,
1012 	.flock		= gfs2_flock,
1013 	.splice_read	= generic_file_splice_read,
1014 	.splice_write	= generic_file_splice_write,
1015 	.setlease	= gfs2_setlease,
1016 	.fallocate	= gfs2_fallocate,
1017 };
1018 
1019 const struct file_operations gfs2_dir_fops = {
1020 	.readdir	= gfs2_readdir,
1021 	.unlocked_ioctl	= gfs2_ioctl,
1022 	.open		= gfs2_open,
1023 	.release	= gfs2_close,
1024 	.fsync		= gfs2_fsync,
1025 	.lock		= gfs2_lock,
1026 	.flock		= gfs2_flock,
1027 	.llseek		= default_llseek,
1028 };
1029 
1030 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1031 
1032 const struct file_operations gfs2_file_fops_nolock = {
1033 	.llseek		= gfs2_llseek,
1034 	.read		= do_sync_read,
1035 	.aio_read	= generic_file_aio_read,
1036 	.write		= do_sync_write,
1037 	.aio_write	= gfs2_file_aio_write,
1038 	.unlocked_ioctl	= gfs2_ioctl,
1039 	.mmap		= gfs2_mmap,
1040 	.open		= gfs2_open,
1041 	.release	= gfs2_close,
1042 	.fsync		= gfs2_fsync,
1043 	.splice_read	= generic_file_splice_read,
1044 	.splice_write	= generic_file_splice_write,
1045 	.setlease	= generic_setlease,
1046 	.fallocate	= gfs2_fallocate,
1047 };
1048 
1049 const struct file_operations gfs2_dir_fops_nolock = {
1050 	.readdir	= gfs2_readdir,
1051 	.unlocked_ioctl	= gfs2_ioctl,
1052 	.open		= gfs2_open,
1053 	.release	= gfs2_close,
1054 	.fsync		= gfs2_fsync,
1055 	.llseek		= default_llseek,
1056 };
1057 
1058