• 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 #include <linux/fileattr.h>
29 
30 #include "gfs2.h"
31 #include "incore.h"
32 #include "bmap.h"
33 #include "aops.h"
34 #include "dir.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "log.h"
39 #include "meta_io.h"
40 #include "quota.h"
41 #include "rgrp.h"
42 #include "trans.h"
43 #include "util.h"
44 
45 /**
46  * gfs2_llseek - seek to a location in a file
47  * @file: the file
48  * @offset: the offset
49  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
50  *
51  * SEEK_END requires the glock for the file because it references the
52  * file's size.
53  *
54  * Returns: The new offset, or errno
55  */
56 
gfs2_llseek(struct file * file,loff_t offset,int whence)57 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
58 {
59 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
60 	struct gfs2_holder i_gh;
61 	loff_t error;
62 
63 	switch (whence) {
64 	case SEEK_END:
65 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
66 					   &i_gh);
67 		if (!error) {
68 			error = generic_file_llseek(file, offset, whence);
69 			gfs2_glock_dq_uninit(&i_gh);
70 		}
71 		break;
72 
73 	case SEEK_DATA:
74 		error = gfs2_seek_data(file, offset);
75 		break;
76 
77 	case SEEK_HOLE:
78 		error = gfs2_seek_hole(file, offset);
79 		break;
80 
81 	case SEEK_CUR:
82 	case SEEK_SET:
83 		/*
84 		 * These don't reference inode->i_size and don't depend on the
85 		 * block mapping, so we don't need the glock.
86 		 */
87 		error = generic_file_llseek(file, offset, whence);
88 		break;
89 	default:
90 		error = -EINVAL;
91 	}
92 
93 	return error;
94 }
95 
96 /**
97  * gfs2_readdir - Iterator for a directory
98  * @file: The directory to read from
99  * @ctx: What to feed directory entries to
100  *
101  * Returns: errno
102  */
103 
gfs2_readdir(struct file * file,struct dir_context * ctx)104 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
105 {
106 	struct inode *dir = file->f_mapping->host;
107 	struct gfs2_inode *dip = GFS2_I(dir);
108 	struct gfs2_holder d_gh;
109 	int error;
110 
111 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
112 	if (error)
113 		return error;
114 
115 	error = gfs2_dir_read(dir, ctx, &file->f_ra);
116 
117 	gfs2_glock_dq_uninit(&d_gh);
118 
119 	return error;
120 }
121 
122 /*
123  * struct fsflag_gfs2flag
124  *
125  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126  * and to GFS2_DIF_JDATA for non-directories.
127  */
128 static struct {
129 	u32 fsflag;
130 	u32 gfsflag;
131 } fsflag_gfs2flag[] = {
132 	{FS_SYNC_FL, GFS2_DIF_SYNC},
133 	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
134 	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
135 	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
136 	{FS_INDEX_FL, GFS2_DIF_EXHASH},
137 	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
138 	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
139 };
140 
gfs2_gfsflags_to_fsflags(struct inode * inode,u32 gfsflags)141 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
142 {
143 	int i;
144 	u32 fsflags = 0;
145 
146 	if (S_ISDIR(inode->i_mode))
147 		gfsflags &= ~GFS2_DIF_JDATA;
148 	else
149 		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
150 
151 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
152 		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
153 			fsflags |= fsflag_gfs2flag[i].fsflag;
154 	return fsflags;
155 }
156 
gfs2_fileattr_get(struct dentry * dentry,struct fileattr * fa)157 int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
158 {
159 	struct inode *inode = d_inode(dentry);
160 	struct gfs2_inode *ip = GFS2_I(inode);
161 	struct gfs2_holder gh;
162 	int error;
163 	u32 fsflags;
164 
165 	if (d_is_special(dentry))
166 		return -ENOTTY;
167 
168 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
169 	error = gfs2_glock_nq(&gh);
170 	if (error)
171 		goto out_uninit;
172 
173 	fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
174 
175 	fileattr_fill_flags(fa, fsflags);
176 
177 	gfs2_glock_dq(&gh);
178 out_uninit:
179 	gfs2_holder_uninit(&gh);
180 	return error;
181 }
182 
gfs2_set_inode_flags(struct inode * inode)183 void gfs2_set_inode_flags(struct inode *inode)
184 {
185 	struct gfs2_inode *ip = GFS2_I(inode);
186 	unsigned int flags = inode->i_flags;
187 
188 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
189 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
190 		flags |= S_NOSEC;
191 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
192 		flags |= S_IMMUTABLE;
193 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
194 		flags |= S_APPEND;
195 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
196 		flags |= S_NOATIME;
197 	if (ip->i_diskflags & GFS2_DIF_SYNC)
198 		flags |= S_SYNC;
199 	inode->i_flags = flags;
200 }
201 
202 /* Flags that can be set by user space */
203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
204 			     GFS2_DIF_IMMUTABLE|		\
205 			     GFS2_DIF_APPENDONLY|		\
206 			     GFS2_DIF_NOATIME|			\
207 			     GFS2_DIF_SYNC|			\
208 			     GFS2_DIF_TOPDIR|			\
209 			     GFS2_DIF_INHERIT_JDATA)
210 
211 /**
212  * do_gfs2_set_flags - set flags on an inode
213  * @inode: The inode
214  * @reqflags: The flags to set
215  * @mask: Indicates which flags are valid
216  *
217  */
do_gfs2_set_flags(struct inode * inode,u32 reqflags,u32 mask)218 static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
219 {
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;
226 
227 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
228 	if (error)
229 		return error;
230 
231 	error = 0;
232 	flags = ip->i_diskflags;
233 	new_flags = (flags & ~mask) | (reqflags & mask);
234 	if ((new_flags ^ flags) == 0)
235 		goto out;
236 
237 	if (!IS_IMMUTABLE(inode)) {
238 		error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
239 		if (error)
240 			goto out;
241 	}
242 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
243 		if (new_flags & GFS2_DIF_JDATA)
244 			gfs2_log_flush(sdp, ip->i_gl,
245 				       GFS2_LOG_HEAD_FLUSH_NORMAL |
246 				       GFS2_LFC_SET_FLAGS);
247 		error = filemap_fdatawrite(inode->i_mapping);
248 		if (error)
249 			goto out;
250 		error = filemap_fdatawait(inode->i_mapping);
251 		if (error)
252 			goto out;
253 		if (new_flags & GFS2_DIF_JDATA)
254 			gfs2_ordered_del_inode(ip);
255 	}
256 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
257 	if (error)
258 		goto out;
259 	error = gfs2_meta_inode_buffer(ip, &bh);
260 	if (error)
261 		goto out_trans_end;
262 	inode->i_ctime = current_time(inode);
263 	gfs2_trans_add_meta(ip->i_gl, bh);
264 	ip->i_diskflags = new_flags;
265 	gfs2_dinode_out(ip, bh->b_data);
266 	brelse(bh);
267 	gfs2_set_inode_flags(inode);
268 	gfs2_set_aops(inode);
269 out_trans_end:
270 	gfs2_trans_end(sdp);
271 out:
272 	gfs2_glock_dq_uninit(&gh);
273 	return error;
274 }
275 
gfs2_fileattr_set(struct user_namespace * mnt_userns,struct dentry * dentry,struct fileattr * fa)276 int gfs2_fileattr_set(struct user_namespace *mnt_userns,
277 		      struct dentry *dentry, struct fileattr *fa)
278 {
279 	struct inode *inode = d_inode(dentry);
280 	u32 fsflags = fa->flags, gfsflags = 0;
281 	u32 mask;
282 	int i;
283 
284 	if (d_is_special(dentry))
285 		return -ENOTTY;
286 
287 	if (fileattr_has_fsx(fa))
288 		return -EOPNOTSUPP;
289 
290 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
291 		if (fsflags & fsflag_gfs2flag[i].fsflag) {
292 			fsflags &= ~fsflag_gfs2flag[i].fsflag;
293 			gfsflags |= fsflag_gfs2flag[i].gfsflag;
294 		}
295 	}
296 	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
297 		return -EINVAL;
298 
299 	mask = GFS2_FLAGS_USER_SET;
300 	if (S_ISDIR(inode->i_mode)) {
301 		mask &= ~GFS2_DIF_JDATA;
302 	} else {
303 		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
304 		if (gfsflags & GFS2_DIF_TOPDIR)
305 			return -EINVAL;
306 		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
307 	}
308 
309 	return do_gfs2_set_flags(inode, gfsflags, mask);
310 }
311 
gfs2_getlabel(struct file * filp,char __user * label)312 static int gfs2_getlabel(struct file *filp, char __user *label)
313 {
314 	struct inode *inode = file_inode(filp);
315 	struct gfs2_sbd *sdp = GFS2_SB(inode);
316 
317 	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
318 		return -EFAULT;
319 
320 	return 0;
321 }
322 
gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)323 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
324 {
325 	switch(cmd) {
326 	case FITRIM:
327 		return gfs2_fitrim(filp, (void __user *)arg);
328 	case FS_IOC_GETFSLABEL:
329 		return gfs2_getlabel(filp, (char __user *)arg);
330 	}
331 
332 	return -ENOTTY;
333 }
334 
335 #ifdef CONFIG_COMPAT
gfs2_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)336 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
337 {
338 	switch(cmd) {
339 	/* Keep this list in sync with gfs2_ioctl */
340 	case FITRIM:
341 	case FS_IOC_GETFSLABEL:
342 		break;
343 	default:
344 		return -ENOIOCTLCMD;
345 	}
346 
347 	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
348 }
349 #else
350 #define gfs2_compat_ioctl NULL
351 #endif
352 
353 /**
354  * gfs2_size_hint - Give a hint to the size of a write request
355  * @filep: The struct file
356  * @offset: The file offset of the write
357  * @size: The length of the write
358  *
359  * When we are about to do a write, this function records the total
360  * write size in order to provide a suitable hint to the lower layers
361  * about how many blocks will be required.
362  *
363  */
364 
gfs2_size_hint(struct file * filep,loff_t offset,size_t size)365 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
366 {
367 	struct inode *inode = file_inode(filep);
368 	struct gfs2_sbd *sdp = GFS2_SB(inode);
369 	struct gfs2_inode *ip = GFS2_I(inode);
370 	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
371 	int hint = min_t(size_t, INT_MAX, blks);
372 
373 	if (hint > atomic_read(&ip->i_sizehint))
374 		atomic_set(&ip->i_sizehint, hint);
375 }
376 
377 /**
378  * gfs2_allocate_page_backing - Allocate blocks for a write fault
379  * @page: The (locked) page to allocate backing for
380  * @length: Size of the allocation
381  *
382  * We try to allocate all the blocks required for the page in one go.  This
383  * might fail for various reasons, so we keep trying until all the blocks to
384  * back this page are allocated.  If some of the blocks are already allocated,
385  * that is ok too.
386  */
gfs2_allocate_page_backing(struct page * page,unsigned int length)387 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
388 {
389 	u64 pos = page_offset(page);
390 
391 	do {
392 		struct iomap iomap = { };
393 
394 		if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
395 			return -EIO;
396 
397 		if (length < iomap.length)
398 			iomap.length = length;
399 		length -= iomap.length;
400 		pos += iomap.length;
401 	} while (length > 0);
402 
403 	return 0;
404 }
405 
406 /**
407  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
408  * @vmf: The virtual memory fault containing the page to become writable
409  *
410  * When the page becomes writable, we need to ensure that we have
411  * blocks allocated on disk to back that page.
412  */
413 
gfs2_page_mkwrite(struct vm_fault * vmf)414 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
415 {
416 	struct page *page = vmf->page;
417 	struct inode *inode = file_inode(vmf->vma->vm_file);
418 	struct gfs2_inode *ip = GFS2_I(inode);
419 	struct gfs2_sbd *sdp = GFS2_SB(inode);
420 	struct gfs2_alloc_parms ap = { .aflags = 0, };
421 	u64 offset = page_offset(page);
422 	unsigned int data_blocks, ind_blocks, rblocks;
423 	vm_fault_t ret = VM_FAULT_LOCKED;
424 	struct gfs2_holder gh;
425 	unsigned int length;
426 	loff_t size;
427 	int err;
428 
429 	sb_start_pagefault(inode->i_sb);
430 
431 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
432 	err = gfs2_glock_nq(&gh);
433 	if (err) {
434 		ret = block_page_mkwrite_return(err);
435 		goto out_uninit;
436 	}
437 
438 	/* Check page index against inode size */
439 	size = i_size_read(inode);
440 	if (offset >= size) {
441 		ret = VM_FAULT_SIGBUS;
442 		goto out_unlock;
443 	}
444 
445 	/* Update file times before taking page lock */
446 	file_update_time(vmf->vma->vm_file);
447 
448 	/* page is wholly or partially inside EOF */
449 	if (size - offset < PAGE_SIZE)
450 		length = size - offset;
451 	else
452 		length = PAGE_SIZE;
453 
454 	gfs2_size_hint(vmf->vma->vm_file, offset, length);
455 
456 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
457 	set_bit(GIF_SW_PAGED, &ip->i_flags);
458 
459 	/*
460 	 * iomap_writepage / iomap_writepages currently don't support inline
461 	 * files, so always unstuff here.
462 	 */
463 
464 	if (!gfs2_is_stuffed(ip) &&
465 	    !gfs2_write_alloc_required(ip, offset, length)) {
466 		lock_page(page);
467 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
468 			ret = VM_FAULT_NOPAGE;
469 			unlock_page(page);
470 		}
471 		goto out_unlock;
472 	}
473 
474 	err = gfs2_rindex_update(sdp);
475 	if (err) {
476 		ret = block_page_mkwrite_return(err);
477 		goto out_unlock;
478 	}
479 
480 	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
481 	ap.target = data_blocks + ind_blocks;
482 	err = gfs2_quota_lock_check(ip, &ap);
483 	if (err) {
484 		ret = block_page_mkwrite_return(err);
485 		goto out_unlock;
486 	}
487 	err = gfs2_inplace_reserve(ip, &ap);
488 	if (err) {
489 		ret = block_page_mkwrite_return(err);
490 		goto out_quota_unlock;
491 	}
492 
493 	rblocks = RES_DINODE + ind_blocks;
494 	if (gfs2_is_jdata(ip))
495 		rblocks += data_blocks ? data_blocks : 1;
496 	if (ind_blocks || data_blocks) {
497 		rblocks += RES_STATFS + RES_QUOTA;
498 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
499 	}
500 	err = gfs2_trans_begin(sdp, rblocks, 0);
501 	if (err) {
502 		ret = block_page_mkwrite_return(err);
503 		goto out_trans_fail;
504 	}
505 
506 	/* Unstuff, if required, and allocate backing blocks for page */
507 	if (gfs2_is_stuffed(ip)) {
508 		err = gfs2_unstuff_dinode(ip);
509 		if (err) {
510 			ret = block_page_mkwrite_return(err);
511 			goto out_trans_end;
512 		}
513 	}
514 
515 	lock_page(page);
516 	/* If truncated, we must retry the operation, we may have raced
517 	 * with the glock demotion code.
518 	 */
519 	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
520 		ret = VM_FAULT_NOPAGE;
521 		goto out_page_locked;
522 	}
523 
524 	err = gfs2_allocate_page_backing(page, length);
525 	if (err)
526 		ret = block_page_mkwrite_return(err);
527 
528 out_page_locked:
529 	if (ret != VM_FAULT_LOCKED)
530 		unlock_page(page);
531 out_trans_end:
532 	gfs2_trans_end(sdp);
533 out_trans_fail:
534 	gfs2_inplace_release(ip);
535 out_quota_unlock:
536 	gfs2_quota_unlock(ip);
537 out_unlock:
538 	gfs2_glock_dq(&gh);
539 out_uninit:
540 	gfs2_holder_uninit(&gh);
541 	if (ret == VM_FAULT_LOCKED) {
542 		set_page_dirty(page);
543 		wait_for_stable_page(page);
544 	}
545 	sb_end_pagefault(inode->i_sb);
546 	return ret;
547 }
548 
gfs2_fault(struct vm_fault * vmf)549 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
550 {
551 	struct inode *inode = file_inode(vmf->vma->vm_file);
552 	struct gfs2_inode *ip = GFS2_I(inode);
553 	struct gfs2_holder gh;
554 	vm_fault_t ret;
555 	int err;
556 
557 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
558 	err = gfs2_glock_nq(&gh);
559 	if (err) {
560 		ret = block_page_mkwrite_return(err);
561 		goto out_uninit;
562 	}
563 	ret = filemap_fault(vmf);
564 	gfs2_glock_dq(&gh);
565 out_uninit:
566 	gfs2_holder_uninit(&gh);
567 	return ret;
568 }
569 
570 static const struct vm_operations_struct gfs2_vm_ops = {
571 	.fault = gfs2_fault,
572 	.map_pages = filemap_map_pages,
573 	.page_mkwrite = gfs2_page_mkwrite,
574 };
575 
576 /**
577  * gfs2_mmap
578  * @file: The file to map
579  * @vma: The VMA which described the mapping
580  *
581  * There is no need to get a lock here unless we should be updating
582  * atime. We ignore any locking errors since the only consequence is
583  * a missed atime update (which will just be deferred until later).
584  *
585  * Returns: 0
586  */
587 
gfs2_mmap(struct file * file,struct vm_area_struct * vma)588 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
589 {
590 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
591 
592 	if (!(file->f_flags & O_NOATIME) &&
593 	    !IS_NOATIME(&ip->i_inode)) {
594 		struct gfs2_holder i_gh;
595 		int error;
596 
597 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
598 					   &i_gh);
599 		if (error)
600 			return error;
601 		/* grab lock to update inode */
602 		gfs2_glock_dq_uninit(&i_gh);
603 		file_accessed(file);
604 	}
605 	vma->vm_ops = &gfs2_vm_ops;
606 
607 	return 0;
608 }
609 
610 /**
611  * gfs2_open_common - This is common to open and atomic_open
612  * @inode: The inode being opened
613  * @file: The file being opened
614  *
615  * This maybe called under a glock or not depending upon how it has
616  * been called. We must always be called under a glock for regular
617  * files, however. For other file types, it does not matter whether
618  * we hold the glock or not.
619  *
620  * Returns: Error code or 0 for success
621  */
622 
gfs2_open_common(struct inode * inode,struct file * file)623 int gfs2_open_common(struct inode *inode, struct file *file)
624 {
625 	struct gfs2_file *fp;
626 	int ret;
627 
628 	if (S_ISREG(inode->i_mode)) {
629 		ret = generic_file_open(inode, file);
630 		if (ret)
631 			return ret;
632 	}
633 
634 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
635 	if (!fp)
636 		return -ENOMEM;
637 
638 	mutex_init(&fp->f_fl_mutex);
639 
640 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
641 	file->private_data = fp;
642 	if (file->f_mode & FMODE_WRITE) {
643 		ret = gfs2_qa_get(GFS2_I(inode));
644 		if (ret)
645 			goto fail;
646 	}
647 	return 0;
648 
649 fail:
650 	kfree(file->private_data);
651 	file->private_data = NULL;
652 	return ret;
653 }
654 
655 /**
656  * gfs2_open - open a file
657  * @inode: the inode to open
658  * @file: the struct file for this opening
659  *
660  * After atomic_open, this function is only used for opening files
661  * which are already cached. We must still get the glock for regular
662  * files to ensure that we have the file size uptodate for the large
663  * file check which is in the common code. That is only an issue for
664  * regular files though.
665  *
666  * Returns: errno
667  */
668 
gfs2_open(struct inode * inode,struct file * file)669 static int gfs2_open(struct inode *inode, struct file *file)
670 {
671 	struct gfs2_inode *ip = GFS2_I(inode);
672 	struct gfs2_holder i_gh;
673 	int error;
674 	bool need_unlock = false;
675 
676 	if (S_ISREG(ip->i_inode.i_mode)) {
677 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
678 					   &i_gh);
679 		if (error)
680 			return error;
681 		need_unlock = true;
682 	}
683 
684 	error = gfs2_open_common(inode, file);
685 
686 	if (need_unlock)
687 		gfs2_glock_dq_uninit(&i_gh);
688 
689 	return error;
690 }
691 
692 /**
693  * gfs2_release - called to close a struct file
694  * @inode: the inode the struct file belongs to
695  * @file: the struct file being closed
696  *
697  * Returns: errno
698  */
699 
gfs2_release(struct inode * inode,struct file * file)700 static int gfs2_release(struct inode *inode, struct file *file)
701 {
702 	struct gfs2_inode *ip = GFS2_I(inode);
703 
704 	kfree(file->private_data);
705 	file->private_data = NULL;
706 
707 	if (file->f_mode & FMODE_WRITE) {
708 		if (gfs2_rs_active(&ip->i_res))
709 			gfs2_rs_delete(ip);
710 		gfs2_qa_put(ip);
711 	}
712 	return 0;
713 }
714 
715 /**
716  * gfs2_fsync - sync the dirty data for a file (across the cluster)
717  * @file: the file that points to the dentry
718  * @start: the start position in the file to sync
719  * @end: the end position in the file to sync
720  * @datasync: set if we can ignore timestamp changes
721  *
722  * We split the data flushing here so that we don't wait for the data
723  * until after we've also sent the metadata to disk. Note that for
724  * data=ordered, we will write & wait for the data at the log flush
725  * stage anyway, so this is unlikely to make much of a difference
726  * except in the data=writeback case.
727  *
728  * If the fdatawrite fails due to any reason except -EIO, we will
729  * continue the remainder of the fsync, although we'll still report
730  * the error at the end. This is to match filemap_write_and_wait_range()
731  * behaviour.
732  *
733  * Returns: errno
734  */
735 
gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)736 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
737 		      int datasync)
738 {
739 	struct address_space *mapping = file->f_mapping;
740 	struct inode *inode = mapping->host;
741 	int sync_state = inode->i_state & I_DIRTY;
742 	struct gfs2_inode *ip = GFS2_I(inode);
743 	int ret = 0, ret1 = 0;
744 
745 	if (mapping->nrpages) {
746 		ret1 = filemap_fdatawrite_range(mapping, start, end);
747 		if (ret1 == -EIO)
748 			return ret1;
749 	}
750 
751 	if (!gfs2_is_jdata(ip))
752 		sync_state &= ~I_DIRTY_PAGES;
753 	if (datasync)
754 		sync_state &= ~I_DIRTY_SYNC;
755 
756 	if (sync_state) {
757 		ret = sync_inode_metadata(inode, 1);
758 		if (ret)
759 			return ret;
760 		if (gfs2_is_jdata(ip))
761 			ret = file_write_and_wait(file);
762 		if (ret)
763 			return ret;
764 		gfs2_ail_flush(ip->i_gl, 1);
765 	}
766 
767 	if (mapping->nrpages)
768 		ret = file_fdatawait_range(file, start, end);
769 
770 	return ret ? ret : ret1;
771 }
772 
should_fault_in_pages(struct iov_iter * i,struct kiocb * iocb,size_t * prev_count,size_t * window_size)773 static inline bool should_fault_in_pages(struct iov_iter *i,
774 					 struct kiocb *iocb,
775 					 size_t *prev_count,
776 					 size_t *window_size)
777 {
778 	size_t count = iov_iter_count(i);
779 	size_t size, offs;
780 
781 	if (!count)
782 		return false;
783 	if (!user_backed_iter(i))
784 		return false;
785 
786 	/*
787 	 * Try to fault in multiple pages initially.  When that doesn't result
788 	 * in any progress, fall back to a single page.
789 	 */
790 	size = PAGE_SIZE;
791 	offs = offset_in_page(iocb->ki_pos);
792 	if (*prev_count != count) {
793 		size_t nr_dirtied;
794 
795 		nr_dirtied = max(current->nr_dirtied_pause -
796 				 current->nr_dirtied, 8);
797 		size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
798 	}
799 
800 	*prev_count = count;
801 	*window_size = size - offs;
802 	return true;
803 }
804 
gfs2_file_direct_read(struct kiocb * iocb,struct iov_iter * to,struct gfs2_holder * gh)805 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
806 				     struct gfs2_holder *gh)
807 {
808 	struct file *file = iocb->ki_filp;
809 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
810 	size_t prev_count = 0, window_size = 0;
811 	size_t read = 0;
812 	ssize_t ret;
813 
814 	/*
815 	 * In this function, we disable page faults when we're holding the
816 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
817 	 * that the inode glock may be dropped, fault in the pages manually,
818 	 * and retry.
819 	 *
820 	 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
821 	 * physical as well as manual page faults, and we need to disable both
822 	 * kinds.
823 	 *
824 	 * For direct I/O, gfs2 takes the inode glock in deferred mode.  This
825 	 * locking mode is compatible with other deferred holders, so multiple
826 	 * processes and nodes can do direct I/O to a file at the same time.
827 	 * There's no guarantee that reads or writes will be atomic.  Any
828 	 * coordination among readers and writers needs to happen externally.
829 	 */
830 
831 	if (!iov_iter_count(to))
832 		return 0; /* skip atime */
833 
834 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
835 retry:
836 	ret = gfs2_glock_nq(gh);
837 	if (ret)
838 		goto out_uninit;
839 	pagefault_disable();
840 	to->nofault = true;
841 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
842 			   IOMAP_DIO_PARTIAL, NULL, read);
843 	to->nofault = false;
844 	pagefault_enable();
845 	if (ret <= 0 && ret != -EFAULT)
846 		goto out_unlock;
847 	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
848 	if (ret > 0)
849 		read = ret;
850 
851 	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
852 		gfs2_glock_dq(gh);
853 		window_size -= fault_in_iov_iter_writeable(to, window_size);
854 		if (window_size)
855 			goto retry;
856 	}
857 out_unlock:
858 	if (gfs2_holder_queued(gh))
859 		gfs2_glock_dq(gh);
860 out_uninit:
861 	gfs2_holder_uninit(gh);
862 	/* User space doesn't expect partial success. */
863 	if (ret < 0)
864 		return ret;
865 	return read;
866 }
867 
gfs2_file_direct_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)868 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
869 				      struct gfs2_holder *gh)
870 {
871 	struct file *file = iocb->ki_filp;
872 	struct inode *inode = file->f_mapping->host;
873 	struct gfs2_inode *ip = GFS2_I(inode);
874 	size_t prev_count = 0, window_size = 0;
875 	size_t written = 0;
876 	bool enough_retries;
877 	ssize_t ret;
878 
879 	/*
880 	 * In this function, we disable page faults when we're holding the
881 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
882 	 * that the inode glock may be dropped, fault in the pages manually,
883 	 * and retry.
884 	 *
885 	 * For writes, iomap_dio_rw only triggers manual page faults, so we
886 	 * don't need to disable physical ones.
887 	 */
888 
889 	/*
890 	 * Deferred lock, even if its a write, since we do no allocation on
891 	 * this path. All we need to change is the atime, and this lock mode
892 	 * ensures that other nodes have flushed their buffered read caches
893 	 * (i.e. their page cache entries for this inode). We do not,
894 	 * unfortunately, have the option of only flushing a range like the
895 	 * VFS does.
896 	 */
897 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
898 retry:
899 	ret = gfs2_glock_nq(gh);
900 	if (ret)
901 		goto out_uninit;
902 	/* Silently fall back to buffered I/O when writing beyond EOF */
903 	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
904 		goto out_unlock;
905 
906 	from->nofault = true;
907 	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
908 			   IOMAP_DIO_PARTIAL, NULL, written);
909 	from->nofault = false;
910 	if (ret <= 0) {
911 		if (ret == -ENOTBLK)
912 			ret = 0;
913 		if (ret != -EFAULT)
914 			goto out_unlock;
915 	}
916 	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
917 	if (ret > 0)
918 		written = ret;
919 
920 	enough_retries = prev_count == iov_iter_count(from) &&
921 			 window_size <= PAGE_SIZE;
922 	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
923 		gfs2_glock_dq(gh);
924 		window_size -= fault_in_iov_iter_readable(from, window_size);
925 		if (window_size) {
926 			if (!enough_retries)
927 				goto retry;
928 			/* fall back to buffered I/O */
929 			ret = 0;
930 		}
931 	}
932 out_unlock:
933 	if (gfs2_holder_queued(gh))
934 		gfs2_glock_dq(gh);
935 out_uninit:
936 	gfs2_holder_uninit(gh);
937 	/* User space doesn't expect partial success. */
938 	if (ret < 0)
939 		return ret;
940 	return written;
941 }
942 
gfs2_file_read_iter(struct kiocb * iocb,struct iov_iter * to)943 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
944 {
945 	struct gfs2_inode *ip;
946 	struct gfs2_holder gh;
947 	size_t prev_count = 0, window_size = 0;
948 	size_t read = 0;
949 	ssize_t ret;
950 
951 	/*
952 	 * In this function, we disable page faults when we're holding the
953 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
954 	 * that the inode glock may be dropped, fault in the pages manually,
955 	 * and retry.
956 	 */
957 
958 	if (iocb->ki_flags & IOCB_DIRECT)
959 		return gfs2_file_direct_read(iocb, to, &gh);
960 
961 	pagefault_disable();
962 	iocb->ki_flags |= IOCB_NOIO;
963 	ret = generic_file_read_iter(iocb, to);
964 	iocb->ki_flags &= ~IOCB_NOIO;
965 	pagefault_enable();
966 	if (ret >= 0) {
967 		if (!iov_iter_count(to))
968 			return ret;
969 		read = ret;
970 	} else if (ret != -EFAULT) {
971 		if (ret != -EAGAIN)
972 			return ret;
973 		if (iocb->ki_flags & IOCB_NOWAIT)
974 			return ret;
975 	}
976 	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
977 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
978 retry:
979 	ret = gfs2_glock_nq(&gh);
980 	if (ret)
981 		goto out_uninit;
982 	pagefault_disable();
983 	ret = generic_file_read_iter(iocb, to);
984 	pagefault_enable();
985 	if (ret <= 0 && ret != -EFAULT)
986 		goto out_unlock;
987 	if (ret > 0)
988 		read += ret;
989 
990 	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
991 		gfs2_glock_dq(&gh);
992 		window_size -= fault_in_iov_iter_writeable(to, window_size);
993 		if (window_size)
994 			goto retry;
995 	}
996 out_unlock:
997 	if (gfs2_holder_queued(&gh))
998 		gfs2_glock_dq(&gh);
999 out_uninit:
1000 	gfs2_holder_uninit(&gh);
1001 	return read ? read : ret;
1002 }
1003 
gfs2_file_buffered_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)1004 static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1005 					struct iov_iter *from,
1006 					struct gfs2_holder *gh)
1007 {
1008 	struct file *file = iocb->ki_filp;
1009 	struct inode *inode = file_inode(file);
1010 	struct gfs2_inode *ip = GFS2_I(inode);
1011 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1012 	struct gfs2_holder *statfs_gh = NULL;
1013 	size_t prev_count = 0, window_size = 0;
1014 	size_t orig_count = iov_iter_count(from);
1015 	size_t written = 0;
1016 	ssize_t ret;
1017 
1018 	/*
1019 	 * In this function, we disable page faults when we're holding the
1020 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
1021 	 * that the inode glock may be dropped, fault in the pages manually,
1022 	 * and retry.
1023 	 */
1024 
1025 	if (inode == sdp->sd_rindex) {
1026 		statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1027 		if (!statfs_gh)
1028 			return -ENOMEM;
1029 	}
1030 
1031 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1032 	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1033 retry:
1034 		window_size -= fault_in_iov_iter_readable(from, window_size);
1035 		if (!window_size) {
1036 			ret = -EFAULT;
1037 			goto out_uninit;
1038 		}
1039 		from->count = min(from->count, window_size);
1040 	}
1041 	ret = gfs2_glock_nq(gh);
1042 	if (ret)
1043 		goto out_uninit;
1044 
1045 	if (inode == sdp->sd_rindex) {
1046 		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1047 
1048 		ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1049 					 GL_NOCACHE, statfs_gh);
1050 		if (ret)
1051 			goto out_unlock;
1052 	}
1053 
1054 	current->backing_dev_info = inode_to_bdi(inode);
1055 	pagefault_disable();
1056 	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1057 	pagefault_enable();
1058 	current->backing_dev_info = NULL;
1059 	if (ret > 0) {
1060 		iocb->ki_pos += ret;
1061 		written += ret;
1062 	}
1063 
1064 	if (inode == sdp->sd_rindex)
1065 		gfs2_glock_dq_uninit(statfs_gh);
1066 
1067 	if (ret <= 0 && ret != -EFAULT)
1068 		goto out_unlock;
1069 
1070 	from->count = orig_count - written;
1071 	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1072 		gfs2_glock_dq(gh);
1073 		goto retry;
1074 	}
1075 out_unlock:
1076 	if (gfs2_holder_queued(gh))
1077 		gfs2_glock_dq(gh);
1078 out_uninit:
1079 	gfs2_holder_uninit(gh);
1080 	kfree(statfs_gh);
1081 	from->count = orig_count - written;
1082 	return written ? written : ret;
1083 }
1084 
1085 /**
1086  * gfs2_file_write_iter - Perform a write to a file
1087  * @iocb: The io context
1088  * @from: The data to write
1089  *
1090  * We have to do a lock/unlock here to refresh the inode size for
1091  * O_APPEND writes, otherwise we can land up writing at the wrong
1092  * offset. There is still a race, but provided the app is using its
1093  * own file locking, this will make O_APPEND work as expected.
1094  *
1095  */
1096 
gfs2_file_write_iter(struct kiocb * iocb,struct iov_iter * from)1097 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1098 {
1099 	struct file *file = iocb->ki_filp;
1100 	struct inode *inode = file_inode(file);
1101 	struct gfs2_inode *ip = GFS2_I(inode);
1102 	struct gfs2_holder gh;
1103 	ssize_t ret;
1104 
1105 	gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1106 
1107 	if (iocb->ki_flags & IOCB_APPEND) {
1108 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1109 		if (ret)
1110 			return ret;
1111 		gfs2_glock_dq_uninit(&gh);
1112 	}
1113 
1114 	inode_lock(inode);
1115 	ret = generic_write_checks(iocb, from);
1116 	if (ret <= 0)
1117 		goto out_unlock;
1118 
1119 	ret = file_remove_privs(file);
1120 	if (ret)
1121 		goto out_unlock;
1122 
1123 	ret = file_update_time(file);
1124 	if (ret)
1125 		goto out_unlock;
1126 
1127 	if (iocb->ki_flags & IOCB_DIRECT) {
1128 		struct address_space *mapping = file->f_mapping;
1129 		ssize_t buffered, ret2;
1130 
1131 		ret = gfs2_file_direct_write(iocb, from, &gh);
1132 		if (ret < 0 || !iov_iter_count(from))
1133 			goto out_unlock;
1134 
1135 		iocb->ki_flags |= IOCB_DSYNC;
1136 		buffered = gfs2_file_buffered_write(iocb, from, &gh);
1137 		if (unlikely(buffered <= 0)) {
1138 			if (!ret)
1139 				ret = buffered;
1140 			goto out_unlock;
1141 		}
1142 
1143 		/*
1144 		 * We need to ensure that the page cache pages are written to
1145 		 * disk and invalidated to preserve the expected O_DIRECT
1146 		 * semantics.  If the writeback or invalidate fails, only report
1147 		 * the direct I/O range as we don't know if the buffered pages
1148 		 * made it to disk.
1149 		 */
1150 		ret2 = generic_write_sync(iocb, buffered);
1151 		invalidate_mapping_pages(mapping,
1152 				(iocb->ki_pos - buffered) >> PAGE_SHIFT,
1153 				(iocb->ki_pos - 1) >> PAGE_SHIFT);
1154 		if (!ret || ret2 > 0)
1155 			ret += ret2;
1156 	} else {
1157 		ret = gfs2_file_buffered_write(iocb, from, &gh);
1158 		if (likely(ret > 0))
1159 			ret = generic_write_sync(iocb, ret);
1160 	}
1161 
1162 out_unlock:
1163 	inode_unlock(inode);
1164 	return ret;
1165 }
1166 
fallocate_chunk(struct inode * inode,loff_t offset,loff_t len,int mode)1167 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1168 			   int mode)
1169 {
1170 	struct super_block *sb = inode->i_sb;
1171 	struct gfs2_inode *ip = GFS2_I(inode);
1172 	loff_t end = offset + len;
1173 	struct buffer_head *dibh;
1174 	int error;
1175 
1176 	error = gfs2_meta_inode_buffer(ip, &dibh);
1177 	if (unlikely(error))
1178 		return error;
1179 
1180 	gfs2_trans_add_meta(ip->i_gl, dibh);
1181 
1182 	if (gfs2_is_stuffed(ip)) {
1183 		error = gfs2_unstuff_dinode(ip);
1184 		if (unlikely(error))
1185 			goto out;
1186 	}
1187 
1188 	while (offset < end) {
1189 		struct iomap iomap = { };
1190 
1191 		error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1192 		if (error)
1193 			goto out;
1194 		offset = iomap.offset + iomap.length;
1195 		if (!(iomap.flags & IOMAP_F_NEW))
1196 			continue;
1197 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1198 					 iomap.length >> inode->i_blkbits,
1199 					 GFP_NOFS);
1200 		if (error) {
1201 			fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1202 			goto out;
1203 		}
1204 	}
1205 out:
1206 	brelse(dibh);
1207 	return error;
1208 }
1209 
1210 /**
1211  * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1212  *                     blocks, determine how many bytes can be written.
1213  * @ip:          The inode in question.
1214  * @len:         Max cap of bytes. What we return in *len must be <= this.
1215  * @data_blocks: Compute and return the number of data blocks needed
1216  * @ind_blocks:  Compute and return the number of indirect blocks needed
1217  * @max_blocks:  The total blocks available to work with.
1218  *
1219  * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1220  */
calc_max_reserv(struct gfs2_inode * ip,loff_t * len,unsigned int * data_blocks,unsigned int * ind_blocks,unsigned int max_blocks)1221 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1222 			    unsigned int *data_blocks, unsigned int *ind_blocks,
1223 			    unsigned int max_blocks)
1224 {
1225 	loff_t max = *len;
1226 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1227 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1228 
1229 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1230 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1231 		max_data -= tmp;
1232 	}
1233 
1234 	*data_blocks = max_data;
1235 	*ind_blocks = max_blocks - max_data;
1236 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1237 	if (*len > max) {
1238 		*len = max;
1239 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1240 	}
1241 }
1242 
__gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1243 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1244 {
1245 	struct inode *inode = file_inode(file);
1246 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1247 	struct gfs2_inode *ip = GFS2_I(inode);
1248 	struct gfs2_alloc_parms ap = { .aflags = 0, };
1249 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1250 	loff_t bytes, max_bytes, max_blks;
1251 	int error;
1252 	const loff_t pos = offset;
1253 	const loff_t count = len;
1254 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1255 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1256 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
1257 
1258 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1259 
1260 	offset &= bsize_mask;
1261 
1262 	len = next - offset;
1263 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1264 	if (!bytes)
1265 		bytes = UINT_MAX;
1266 	bytes &= bsize_mask;
1267 	if (bytes == 0)
1268 		bytes = sdp->sd_sb.sb_bsize;
1269 
1270 	gfs2_size_hint(file, offset, len);
1271 
1272 	gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1273 	ap.min_target = data_blocks + ind_blocks;
1274 
1275 	while (len > 0) {
1276 		if (len < bytes)
1277 			bytes = len;
1278 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1279 			len -= bytes;
1280 			offset += bytes;
1281 			continue;
1282 		}
1283 
1284 		/* We need to determine how many bytes we can actually
1285 		 * fallocate without exceeding quota or going over the
1286 		 * end of the fs. We start off optimistically by assuming
1287 		 * we can write max_bytes */
1288 		max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1289 
1290 		/* Since max_bytes is most likely a theoretical max, we
1291 		 * calculate a more realistic 'bytes' to serve as a good
1292 		 * starting point for the number of bytes we may be able
1293 		 * to write */
1294 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1295 		ap.target = data_blocks + ind_blocks;
1296 
1297 		error = gfs2_quota_lock_check(ip, &ap);
1298 		if (error)
1299 			return error;
1300 		/* ap.allowed tells us how many blocks quota will allow
1301 		 * us to write. Check if this reduces max_blks */
1302 		max_blks = UINT_MAX;
1303 		if (ap.allowed)
1304 			max_blks = ap.allowed;
1305 
1306 		error = gfs2_inplace_reserve(ip, &ap);
1307 		if (error)
1308 			goto out_qunlock;
1309 
1310 		/* check if the selected rgrp limits our max_blks further */
1311 		if (ip->i_res.rs_reserved < max_blks)
1312 			max_blks = ip->i_res.rs_reserved;
1313 
1314 		/* Almost done. Calculate bytes that can be written using
1315 		 * max_blks. We also recompute max_bytes, data_blocks and
1316 		 * ind_blocks */
1317 		calc_max_reserv(ip, &max_bytes, &data_blocks,
1318 				&ind_blocks, max_blks);
1319 
1320 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1321 			  RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1322 		if (gfs2_is_jdata(ip))
1323 			rblocks += data_blocks ? data_blocks : 1;
1324 
1325 		error = gfs2_trans_begin(sdp, rblocks,
1326 					 PAGE_SIZE >> inode->i_blkbits);
1327 		if (error)
1328 			goto out_trans_fail;
1329 
1330 		error = fallocate_chunk(inode, offset, max_bytes, mode);
1331 		gfs2_trans_end(sdp);
1332 
1333 		if (error)
1334 			goto out_trans_fail;
1335 
1336 		len -= max_bytes;
1337 		offset += max_bytes;
1338 		gfs2_inplace_release(ip);
1339 		gfs2_quota_unlock(ip);
1340 	}
1341 
1342 	if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1343 		i_size_write(inode, pos + count);
1344 	file_update_time(file);
1345 	mark_inode_dirty(inode);
1346 
1347 	if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1348 		return vfs_fsync_range(file, pos, pos + count - 1,
1349 			       (file->f_flags & __O_SYNC) ? 0 : 1);
1350 	return 0;
1351 
1352 out_trans_fail:
1353 	gfs2_inplace_release(ip);
1354 out_qunlock:
1355 	gfs2_quota_unlock(ip);
1356 	return error;
1357 }
1358 
gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1359 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1360 {
1361 	struct inode *inode = file_inode(file);
1362 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1363 	struct gfs2_inode *ip = GFS2_I(inode);
1364 	struct gfs2_holder gh;
1365 	int ret;
1366 
1367 	if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1368 		return -EOPNOTSUPP;
1369 	/* fallocate is needed by gfs2_grow to reserve space in the rindex */
1370 	if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1371 		return -EOPNOTSUPP;
1372 
1373 	inode_lock(inode);
1374 
1375 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1376 	ret = gfs2_glock_nq(&gh);
1377 	if (ret)
1378 		goto out_uninit;
1379 
1380 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1381 	    (offset + len) > inode->i_size) {
1382 		ret = inode_newsize_ok(inode, offset + len);
1383 		if (ret)
1384 			goto out_unlock;
1385 	}
1386 
1387 	ret = get_write_access(inode);
1388 	if (ret)
1389 		goto out_unlock;
1390 
1391 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1392 		ret = __gfs2_punch_hole(file, offset, len);
1393 	} else {
1394 		ret = __gfs2_fallocate(file, mode, offset, len);
1395 		if (ret)
1396 			gfs2_rs_deltree(&ip->i_res);
1397 	}
1398 
1399 	put_write_access(inode);
1400 out_unlock:
1401 	gfs2_glock_dq(&gh);
1402 out_uninit:
1403 	gfs2_holder_uninit(&gh);
1404 	inode_unlock(inode);
1405 	return ret;
1406 }
1407 
gfs2_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)1408 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1409 				      struct file *out, loff_t *ppos,
1410 				      size_t len, unsigned int flags)
1411 {
1412 	ssize_t ret;
1413 
1414 	gfs2_size_hint(out, *ppos, len);
1415 
1416 	ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1417 	return ret;
1418 }
1419 
1420 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1421 
1422 /**
1423  * gfs2_lock - acquire/release a posix lock on a file
1424  * @file: the file pointer
1425  * @cmd: either modify or retrieve lock state, possibly wait
1426  * @fl: type and range of lock
1427  *
1428  * Returns: errno
1429  */
1430 
gfs2_lock(struct file * file,int cmd,struct file_lock * fl)1431 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1432 {
1433 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1434 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1435 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1436 
1437 	if (!(fl->fl_flags & FL_POSIX))
1438 		return -ENOLCK;
1439 	if (cmd == F_CANCELLK) {
1440 		/* Hack: */
1441 		cmd = F_SETLK;
1442 		fl->fl_type = F_UNLCK;
1443 	}
1444 	if (unlikely(gfs2_withdrawn(sdp))) {
1445 		if (fl->fl_type == F_UNLCK)
1446 			locks_lock_file_wait(file, fl);
1447 		return -EIO;
1448 	}
1449 	if (IS_GETLK(cmd))
1450 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1451 	else if (fl->fl_type == F_UNLCK)
1452 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1453 	else
1454 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1455 }
1456 
__flock_holder_uninit(struct file * file,struct gfs2_holder * fl_gh)1457 static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1458 {
1459 	struct gfs2_glock *gl = fl_gh->gh_gl;
1460 
1461 	/*
1462 	 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1463 	 * spinlock.
1464 	 */
1465 
1466 	gfs2_glock_hold(gl);
1467 	spin_lock(&file->f_lock);
1468 	gfs2_holder_uninit(fl_gh);
1469 	spin_unlock(&file->f_lock);
1470 	gfs2_glock_put(gl);
1471 }
1472 
do_flock(struct file * file,int cmd,struct file_lock * fl)1473 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1474 {
1475 	struct gfs2_file *fp = file->private_data;
1476 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1477 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1478 	struct gfs2_glock *gl;
1479 	unsigned int state;
1480 	u16 flags;
1481 	int error = 0;
1482 	int sleeptime;
1483 
1484 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1485 	flags = GL_EXACT | GL_NOPID;
1486 	if (!IS_SETLKW(cmd))
1487 		flags |= LM_FLAG_TRY_1CB;
1488 
1489 	mutex_lock(&fp->f_fl_mutex);
1490 
1491 	if (gfs2_holder_initialized(fl_gh)) {
1492 		struct file_lock request;
1493 		if (fl_gh->gh_state == state)
1494 			goto out;
1495 		locks_init_lock(&request);
1496 		request.fl_type = F_UNLCK;
1497 		request.fl_flags = FL_FLOCK;
1498 		locks_lock_file_wait(file, &request);
1499 		gfs2_glock_dq(fl_gh);
1500 		gfs2_holder_reinit(state, flags, fl_gh);
1501 	} else {
1502 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1503 				       &gfs2_flock_glops, CREATE, &gl);
1504 		if (error)
1505 			goto out;
1506 		spin_lock(&file->f_lock);
1507 		gfs2_holder_init(gl, state, flags, fl_gh);
1508 		spin_unlock(&file->f_lock);
1509 		gfs2_glock_put(gl);
1510 	}
1511 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1512 		error = gfs2_glock_nq(fl_gh);
1513 		if (error != GLR_TRYFAILED)
1514 			break;
1515 		fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1516 		fl_gh->gh_flags |= LM_FLAG_TRY;
1517 		msleep(sleeptime);
1518 	}
1519 	if (error) {
1520 		__flock_holder_uninit(file, fl_gh);
1521 		if (error == GLR_TRYFAILED)
1522 			error = -EAGAIN;
1523 	} else {
1524 		error = locks_lock_file_wait(file, fl);
1525 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1526 	}
1527 
1528 out:
1529 	mutex_unlock(&fp->f_fl_mutex);
1530 	return error;
1531 }
1532 
do_unflock(struct file * file,struct file_lock * fl)1533 static void do_unflock(struct file *file, struct file_lock *fl)
1534 {
1535 	struct gfs2_file *fp = file->private_data;
1536 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1537 
1538 	mutex_lock(&fp->f_fl_mutex);
1539 	locks_lock_file_wait(file, fl);
1540 	if (gfs2_holder_initialized(fl_gh)) {
1541 		gfs2_glock_dq(fl_gh);
1542 		__flock_holder_uninit(file, fl_gh);
1543 	}
1544 	mutex_unlock(&fp->f_fl_mutex);
1545 }
1546 
1547 /**
1548  * gfs2_flock - acquire/release a flock lock on a file
1549  * @file: the file pointer
1550  * @cmd: either modify or retrieve lock state, possibly wait
1551  * @fl: type and range of lock
1552  *
1553  * Returns: errno
1554  */
1555 
gfs2_flock(struct file * file,int cmd,struct file_lock * fl)1556 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1557 {
1558 	if (!(fl->fl_flags & FL_FLOCK))
1559 		return -ENOLCK;
1560 
1561 	if (fl->fl_type == F_UNLCK) {
1562 		do_unflock(file, fl);
1563 		return 0;
1564 	} else {
1565 		return do_flock(file, cmd, fl);
1566 	}
1567 }
1568 
1569 const struct file_operations gfs2_file_fops = {
1570 	.llseek		= gfs2_llseek,
1571 	.read_iter	= gfs2_file_read_iter,
1572 	.write_iter	= gfs2_file_write_iter,
1573 	.iopoll		= iocb_bio_iopoll,
1574 	.unlocked_ioctl	= gfs2_ioctl,
1575 	.compat_ioctl	= gfs2_compat_ioctl,
1576 	.mmap		= gfs2_mmap,
1577 	.open		= gfs2_open,
1578 	.release	= gfs2_release,
1579 	.fsync		= gfs2_fsync,
1580 	.lock		= gfs2_lock,
1581 	.flock		= gfs2_flock,
1582 	.splice_read	= generic_file_splice_read,
1583 	.splice_write	= gfs2_file_splice_write,
1584 	.setlease	= simple_nosetlease,
1585 	.fallocate	= gfs2_fallocate,
1586 };
1587 
1588 const struct file_operations gfs2_dir_fops = {
1589 	.iterate_shared	= gfs2_readdir,
1590 	.unlocked_ioctl	= gfs2_ioctl,
1591 	.compat_ioctl	= gfs2_compat_ioctl,
1592 	.open		= gfs2_open,
1593 	.release	= gfs2_release,
1594 	.fsync		= gfs2_fsync,
1595 	.lock		= gfs2_lock,
1596 	.flock		= gfs2_flock,
1597 	.llseek		= default_llseek,
1598 };
1599 
1600 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1601 
1602 const struct file_operations gfs2_file_fops_nolock = {
1603 	.llseek		= gfs2_llseek,
1604 	.read_iter	= gfs2_file_read_iter,
1605 	.write_iter	= gfs2_file_write_iter,
1606 	.iopoll		= iocb_bio_iopoll,
1607 	.unlocked_ioctl	= gfs2_ioctl,
1608 	.compat_ioctl	= gfs2_compat_ioctl,
1609 	.mmap		= gfs2_mmap,
1610 	.open		= gfs2_open,
1611 	.release	= gfs2_release,
1612 	.fsync		= gfs2_fsync,
1613 	.splice_read	= generic_file_splice_read,
1614 	.splice_write	= gfs2_file_splice_write,
1615 	.setlease	= generic_setlease,
1616 	.fallocate	= gfs2_fallocate,
1617 };
1618 
1619 const struct file_operations gfs2_dir_fops_nolock = {
1620 	.iterate_shared	= gfs2_readdir,
1621 	.unlocked_ioctl	= gfs2_ioctl,
1622 	.compat_ioctl	= gfs2_compat_ioctl,
1623 	.open		= gfs2_open,
1624 	.release	= gfs2_release,
1625 	.fsync		= gfs2_fsync,
1626 	.llseek		= default_llseek,
1627 };
1628 
1629