xref: /openbmc/linux/fs/gfs2/file.c (revision d6b412c5)
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/filelock.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/falloc.h>
21 #include <linux/swap.h>
22 #include <linux/crc32.h>
23 #include <linux/writeback.h>
24 #include <linux/uaccess.h>
25 #include <linux/dlm.h>
26 #include <linux/dlm_plock.h>
27 #include <linux/delay.h>
28 #include <linux/backing-dev.h>
29 #include <linux/fileattr.h>
30 
31 #include "gfs2.h"
32 #include "incore.h"
33 #include "bmap.h"
34 #include "aops.h"
35 #include "dir.h"
36 #include "glock.h"
37 #include "glops.h"
38 #include "inode.h"
39 #include "log.h"
40 #include "meta_io.h"
41 #include "quota.h"
42 #include "rgrp.h"
43 #include "trans.h"
44 #include "util.h"
45 
46 /**
47  * gfs2_llseek - seek to a location in a file
48  * @file: the file
49  * @offset: the offset
50  * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
51  *
52  * SEEK_END requires the glock for the file because it references the
53  * file's size.
54  *
55  * Returns: The new offset, or errno
56  */
57 
gfs2_llseek(struct file * file,loff_t offset,int whence)58 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
59 {
60 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
61 	struct gfs2_holder i_gh;
62 	loff_t error;
63 
64 	switch (whence) {
65 	case SEEK_END:
66 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
67 					   &i_gh);
68 		if (!error) {
69 			error = generic_file_llseek(file, offset, whence);
70 			gfs2_glock_dq_uninit(&i_gh);
71 		}
72 		break;
73 
74 	case SEEK_DATA:
75 		error = gfs2_seek_data(file, offset);
76 		break;
77 
78 	case SEEK_HOLE:
79 		error = gfs2_seek_hole(file, offset);
80 		break;
81 
82 	case SEEK_CUR:
83 	case SEEK_SET:
84 		/*
85 		 * These don't reference inode->i_size and don't depend on the
86 		 * block mapping, so we don't need the glock.
87 		 */
88 		error = generic_file_llseek(file, offset, whence);
89 		break;
90 	default:
91 		error = -EINVAL;
92 	}
93 
94 	return error;
95 }
96 
97 /**
98  * gfs2_readdir - Iterator for a directory
99  * @file: The directory to read from
100  * @ctx: What to feed directory entries to
101  *
102  * Returns: errno
103  */
104 
gfs2_readdir(struct file * file,struct dir_context * ctx)105 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
106 {
107 	struct inode *dir = file->f_mapping->host;
108 	struct gfs2_inode *dip = GFS2_I(dir);
109 	struct gfs2_holder d_gh;
110 	int error;
111 
112 	error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
113 	if (error)
114 		return error;
115 
116 	error = gfs2_dir_read(dir, ctx, &file->f_ra);
117 
118 	gfs2_glock_dq_uninit(&d_gh);
119 
120 	return error;
121 }
122 
123 /*
124  * struct fsflag_gfs2flag
125  *
126  * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
127  * and to GFS2_DIF_JDATA for non-directories.
128  */
129 static struct {
130 	u32 fsflag;
131 	u32 gfsflag;
132 } fsflag_gfs2flag[] = {
133 	{FS_SYNC_FL, GFS2_DIF_SYNC},
134 	{FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
135 	{FS_APPEND_FL, GFS2_DIF_APPENDONLY},
136 	{FS_NOATIME_FL, GFS2_DIF_NOATIME},
137 	{FS_INDEX_FL, GFS2_DIF_EXHASH},
138 	{FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
139 	{FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
140 };
141 
gfs2_gfsflags_to_fsflags(struct inode * inode,u32 gfsflags)142 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
143 {
144 	int i;
145 	u32 fsflags = 0;
146 
147 	if (S_ISDIR(inode->i_mode))
148 		gfsflags &= ~GFS2_DIF_JDATA;
149 	else
150 		gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
151 
152 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
153 		if (gfsflags & fsflag_gfs2flag[i].gfsflag)
154 			fsflags |= fsflag_gfs2flag[i].fsflag;
155 	return fsflags;
156 }
157 
gfs2_fileattr_get(struct dentry * dentry,struct fileattr * fa)158 int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
159 {
160 	struct inode *inode = d_inode(dentry);
161 	struct gfs2_inode *ip = GFS2_I(inode);
162 	struct gfs2_holder gh;
163 	int error;
164 	u32 fsflags;
165 
166 	if (d_is_special(dentry))
167 		return -ENOTTY;
168 
169 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
170 	error = gfs2_glock_nq(&gh);
171 	if (error)
172 		goto out_uninit;
173 
174 	fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
175 
176 	fileattr_fill_flags(fa, fsflags);
177 
178 	gfs2_glock_dq(&gh);
179 out_uninit:
180 	gfs2_holder_uninit(&gh);
181 	return error;
182 }
183 
gfs2_set_inode_flags(struct inode * inode)184 void gfs2_set_inode_flags(struct inode *inode)
185 {
186 	struct gfs2_inode *ip = GFS2_I(inode);
187 	unsigned int flags = inode->i_flags;
188 
189 	flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
190 	if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
191 		flags |= S_NOSEC;
192 	if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
193 		flags |= S_IMMUTABLE;
194 	if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
195 		flags |= S_APPEND;
196 	if (ip->i_diskflags & GFS2_DIF_NOATIME)
197 		flags |= S_NOATIME;
198 	if (ip->i_diskflags & GFS2_DIF_SYNC)
199 		flags |= S_SYNC;
200 	inode->i_flags = flags;
201 }
202 
203 /* Flags that can be set by user space */
204 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA|			\
205 			     GFS2_DIF_IMMUTABLE|		\
206 			     GFS2_DIF_APPENDONLY|		\
207 			     GFS2_DIF_NOATIME|			\
208 			     GFS2_DIF_SYNC|			\
209 			     GFS2_DIF_TOPDIR|			\
210 			     GFS2_DIF_INHERIT_JDATA)
211 
212 /**
213  * do_gfs2_set_flags - set flags on an inode
214  * @inode: The inode
215  * @reqflags: The flags to set
216  * @mask: Indicates which flags are valid
217  *
218  */
do_gfs2_set_flags(struct inode * inode,u32 reqflags,u32 mask)219 static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
220 {
221 	struct gfs2_inode *ip = GFS2_I(inode);
222 	struct gfs2_sbd *sdp = GFS2_SB(inode);
223 	struct buffer_head *bh;
224 	struct gfs2_holder gh;
225 	int error;
226 	u32 new_flags, flags;
227 
228 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
229 	if (error)
230 		return error;
231 
232 	error = 0;
233 	flags = ip->i_diskflags;
234 	new_flags = (flags & ~mask) | (reqflags & mask);
235 	if ((new_flags ^ flags) == 0)
236 		goto out;
237 
238 	if (!IS_IMMUTABLE(inode)) {
239 		error = gfs2_permission(&nop_mnt_idmap, inode, MAY_WRITE);
240 		if (error)
241 			goto out;
242 	}
243 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
244 		if (new_flags & GFS2_DIF_JDATA)
245 			gfs2_log_flush(sdp, ip->i_gl,
246 				       GFS2_LOG_HEAD_FLUSH_NORMAL |
247 				       GFS2_LFC_SET_FLAGS);
248 		error = filemap_fdatawrite(inode->i_mapping);
249 		if (error)
250 			goto out;
251 		error = filemap_fdatawait(inode->i_mapping);
252 		if (error)
253 			goto out;
254 		if (new_flags & GFS2_DIF_JDATA)
255 			gfs2_ordered_del_inode(ip);
256 	}
257 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
258 	if (error)
259 		goto out;
260 	error = gfs2_meta_inode_buffer(ip, &bh);
261 	if (error)
262 		goto out_trans_end;
263 	inode_set_ctime_current(inode);
264 	gfs2_trans_add_meta(ip->i_gl, bh);
265 	ip->i_diskflags = new_flags;
266 	gfs2_dinode_out(ip, bh->b_data);
267 	brelse(bh);
268 	gfs2_set_inode_flags(inode);
269 	gfs2_set_aops(inode);
270 out_trans_end:
271 	gfs2_trans_end(sdp);
272 out:
273 	gfs2_glock_dq_uninit(&gh);
274 	return error;
275 }
276 
gfs2_fileattr_set(struct mnt_idmap * idmap,struct dentry * dentry,struct fileattr * fa)277 int gfs2_fileattr_set(struct mnt_idmap *idmap,
278 		      struct dentry *dentry, struct fileattr *fa)
279 {
280 	struct inode *inode = d_inode(dentry);
281 	u32 fsflags = fa->flags, gfsflags = 0;
282 	u32 mask;
283 	int i;
284 
285 	if (d_is_special(dentry))
286 		return -ENOTTY;
287 
288 	if (fileattr_has_fsx(fa))
289 		return -EOPNOTSUPP;
290 
291 	for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
292 		if (fsflags & fsflag_gfs2flag[i].fsflag) {
293 			fsflags &= ~fsflag_gfs2flag[i].fsflag;
294 			gfsflags |= fsflag_gfs2flag[i].gfsflag;
295 		}
296 	}
297 	if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
298 		return -EINVAL;
299 
300 	mask = GFS2_FLAGS_USER_SET;
301 	if (S_ISDIR(inode->i_mode)) {
302 		mask &= ~GFS2_DIF_JDATA;
303 	} else {
304 		/* The GFS2_DIF_TOPDIR flag is only valid for directories. */
305 		if (gfsflags & GFS2_DIF_TOPDIR)
306 			return -EINVAL;
307 		mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
308 	}
309 
310 	return do_gfs2_set_flags(inode, gfsflags, mask);
311 }
312 
gfs2_getlabel(struct file * filp,char __user * label)313 static int gfs2_getlabel(struct file *filp, char __user *label)
314 {
315 	struct inode *inode = file_inode(filp);
316 	struct gfs2_sbd *sdp = GFS2_SB(inode);
317 
318 	if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
319 		return -EFAULT;
320 
321 	return 0;
322 }
323 
gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)324 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
325 {
326 	switch(cmd) {
327 	case FITRIM:
328 		return gfs2_fitrim(filp, (void __user *)arg);
329 	case FS_IOC_GETFSLABEL:
330 		return gfs2_getlabel(filp, (char __user *)arg);
331 	}
332 
333 	return -ENOTTY;
334 }
335 
336 #ifdef CONFIG_COMPAT
gfs2_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)337 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
338 {
339 	switch(cmd) {
340 	/* Keep this list in sync with gfs2_ioctl */
341 	case FITRIM:
342 	case FS_IOC_GETFSLABEL:
343 		break;
344 	default:
345 		return -ENOIOCTLCMD;
346 	}
347 
348 	return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
349 }
350 #else
351 #define gfs2_compat_ioctl NULL
352 #endif
353 
354 /**
355  * gfs2_size_hint - Give a hint to the size of a write request
356  * @filep: The struct file
357  * @offset: The file offset of the write
358  * @size: The length of the write
359  *
360  * When we are about to do a write, this function records the total
361  * write size in order to provide a suitable hint to the lower layers
362  * about how many blocks will be required.
363  *
364  */
365 
gfs2_size_hint(struct file * filep,loff_t offset,size_t size)366 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
367 {
368 	struct inode *inode = file_inode(filep);
369 	struct gfs2_sbd *sdp = GFS2_SB(inode);
370 	struct gfs2_inode *ip = GFS2_I(inode);
371 	size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
372 	int hint = min_t(size_t, INT_MAX, blks);
373 
374 	if (hint > atomic_read(&ip->i_sizehint))
375 		atomic_set(&ip->i_sizehint, hint);
376 }
377 
378 /**
379  * gfs2_allocate_page_backing - Allocate blocks for a write fault
380  * @page: The (locked) page to allocate backing for
381  * @length: Size of the allocation
382  *
383  * We try to allocate all the blocks required for the page in one go.  This
384  * might fail for various reasons, so we keep trying until all the blocks to
385  * back this page are allocated.  If some of the blocks are already allocated,
386  * that is ok too.
387  */
gfs2_allocate_page_backing(struct page * page,unsigned int length)388 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
389 {
390 	u64 pos = page_offset(page);
391 
392 	do {
393 		struct iomap iomap = { };
394 
395 		if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
396 			return -EIO;
397 
398 		if (length < iomap.length)
399 			iomap.length = length;
400 		length -= iomap.length;
401 		pos += iomap.length;
402 	} while (length > 0);
403 
404 	return 0;
405 }
406 
407 /**
408  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
409  * @vmf: The virtual memory fault containing the page to become writable
410  *
411  * When the page becomes writable, we need to ensure that we have
412  * blocks allocated on disk to back that page.
413  */
414 
gfs2_page_mkwrite(struct vm_fault * vmf)415 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
416 {
417 	struct page *page = vmf->page;
418 	struct inode *inode = file_inode(vmf->vma->vm_file);
419 	struct gfs2_inode *ip = GFS2_I(inode);
420 	struct gfs2_sbd *sdp = GFS2_SB(inode);
421 	struct gfs2_alloc_parms ap = { .aflags = 0, };
422 	u64 offset = page_offset(page);
423 	unsigned int data_blocks, ind_blocks, rblocks;
424 	vm_fault_t ret = VM_FAULT_LOCKED;
425 	struct gfs2_holder gh;
426 	unsigned int length;
427 	loff_t size;
428 	int err;
429 
430 	sb_start_pagefault(inode->i_sb);
431 
432 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
433 	err = gfs2_glock_nq(&gh);
434 	if (err) {
435 		ret = vmf_fs_error(err);
436 		goto out_uninit;
437 	}
438 
439 	/* Check page index against inode size */
440 	size = i_size_read(inode);
441 	if (offset >= size) {
442 		ret = VM_FAULT_SIGBUS;
443 		goto out_unlock;
444 	}
445 
446 	/* Update file times before taking page lock */
447 	file_update_time(vmf->vma->vm_file);
448 
449 	/* page is wholly or partially inside EOF */
450 	if (size - offset < PAGE_SIZE)
451 		length = size - offset;
452 	else
453 		length = PAGE_SIZE;
454 
455 	gfs2_size_hint(vmf->vma->vm_file, offset, length);
456 
457 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
458 	set_bit(GIF_SW_PAGED, &ip->i_flags);
459 
460 	/*
461 	 * iomap_writepage / iomap_writepages currently don't support inline
462 	 * files, so always unstuff here.
463 	 */
464 
465 	if (!gfs2_is_stuffed(ip) &&
466 	    !gfs2_write_alloc_required(ip, offset, length)) {
467 		lock_page(page);
468 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
469 			ret = VM_FAULT_NOPAGE;
470 			unlock_page(page);
471 		}
472 		goto out_unlock;
473 	}
474 
475 	err = gfs2_rindex_update(sdp);
476 	if (err) {
477 		ret = vmf_fs_error(err);
478 		goto out_unlock;
479 	}
480 
481 	gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
482 	ap.target = data_blocks + ind_blocks;
483 	err = gfs2_quota_lock_check(ip, &ap);
484 	if (err) {
485 		ret = vmf_fs_error(err);
486 		goto out_unlock;
487 	}
488 	err = gfs2_inplace_reserve(ip, &ap);
489 	if (err) {
490 		ret = vmf_fs_error(err);
491 		goto out_quota_unlock;
492 	}
493 
494 	rblocks = RES_DINODE + ind_blocks;
495 	if (gfs2_is_jdata(ip))
496 		rblocks += data_blocks ? data_blocks : 1;
497 	if (ind_blocks || data_blocks) {
498 		rblocks += RES_STATFS + RES_QUOTA;
499 		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
500 	}
501 	err = gfs2_trans_begin(sdp, rblocks, 0);
502 	if (err) {
503 		ret = vmf_fs_error(err);
504 		goto out_trans_fail;
505 	}
506 
507 	/* Unstuff, if required, and allocate backing blocks for page */
508 	if (gfs2_is_stuffed(ip)) {
509 		err = gfs2_unstuff_dinode(ip);
510 		if (err) {
511 			ret = vmf_fs_error(err);
512 			goto out_trans_end;
513 		}
514 	}
515 
516 	lock_page(page);
517 	/* If truncated, we must retry the operation, we may have raced
518 	 * with the glock demotion code.
519 	 */
520 	if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
521 		ret = VM_FAULT_NOPAGE;
522 		goto out_page_locked;
523 	}
524 
525 	err = gfs2_allocate_page_backing(page, length);
526 	if (err)
527 		ret = vmf_fs_error(err);
528 
529 out_page_locked:
530 	if (ret != VM_FAULT_LOCKED)
531 		unlock_page(page);
532 out_trans_end:
533 	gfs2_trans_end(sdp);
534 out_trans_fail:
535 	gfs2_inplace_release(ip);
536 out_quota_unlock:
537 	gfs2_quota_unlock(ip);
538 out_unlock:
539 	gfs2_glock_dq(&gh);
540 out_uninit:
541 	gfs2_holder_uninit(&gh);
542 	if (ret == VM_FAULT_LOCKED) {
543 		set_page_dirty(page);
544 		wait_for_stable_page(page);
545 	}
546 	sb_end_pagefault(inode->i_sb);
547 	return ret;
548 }
549 
gfs2_fault(struct vm_fault * vmf)550 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
551 {
552 	struct inode *inode = file_inode(vmf->vma->vm_file);
553 	struct gfs2_inode *ip = GFS2_I(inode);
554 	struct gfs2_holder gh;
555 	vm_fault_t ret;
556 	int err;
557 
558 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
559 	err = gfs2_glock_nq(&gh);
560 	if (err) {
561 		ret = vmf_fs_error(err);
562 		goto out_uninit;
563 	}
564 	ret = filemap_fault(vmf);
565 	gfs2_glock_dq(&gh);
566 out_uninit:
567 	gfs2_holder_uninit(&gh);
568 	return ret;
569 }
570 
571 static const struct vm_operations_struct gfs2_vm_ops = {
572 	.fault = gfs2_fault,
573 	.map_pages = filemap_map_pages,
574 	.page_mkwrite = gfs2_page_mkwrite,
575 };
576 
577 /**
578  * gfs2_mmap
579  * @file: The file to map
580  * @vma: The VMA which described the mapping
581  *
582  * There is no need to get a lock here unless we should be updating
583  * atime. We ignore any locking errors since the only consequence is
584  * a missed atime update (which will just be deferred until later).
585  *
586  * Returns: 0
587  */
588 
gfs2_mmap(struct file * file,struct vm_area_struct * vma)589 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
590 {
591 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
592 
593 	if (!(file->f_flags & O_NOATIME) &&
594 	    !IS_NOATIME(&ip->i_inode)) {
595 		struct gfs2_holder i_gh;
596 		int error;
597 
598 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
599 					   &i_gh);
600 		if (error)
601 			return error;
602 		/* grab lock to update inode */
603 		gfs2_glock_dq_uninit(&i_gh);
604 		file_accessed(file);
605 	}
606 	vma->vm_ops = &gfs2_vm_ops;
607 
608 	return 0;
609 }
610 
611 /**
612  * gfs2_open_common - This is common to open and atomic_open
613  * @inode: The inode being opened
614  * @file: The file being opened
615  *
616  * This maybe called under a glock or not depending upon how it has
617  * been called. We must always be called under a glock for regular
618  * files, however. For other file types, it does not matter whether
619  * we hold the glock or not.
620  *
621  * Returns: Error code or 0 for success
622  */
623 
gfs2_open_common(struct inode * inode,struct file * file)624 int gfs2_open_common(struct inode *inode, struct file *file)
625 {
626 	struct gfs2_file *fp;
627 	int ret;
628 
629 	if (S_ISREG(inode->i_mode)) {
630 		ret = generic_file_open(inode, file);
631 		if (ret)
632 			return ret;
633 
634 		if (!gfs2_is_jdata(GFS2_I(inode)))
635 			file->f_mode |= FMODE_CAN_ODIRECT;
636 	}
637 
638 	fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
639 	if (!fp)
640 		return -ENOMEM;
641 
642 	mutex_init(&fp->f_fl_mutex);
643 
644 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
645 	file->private_data = fp;
646 	if (file->f_mode & FMODE_WRITE) {
647 		ret = gfs2_qa_get(GFS2_I(inode));
648 		if (ret)
649 			goto fail;
650 	}
651 	return 0;
652 
653 fail:
654 	kfree(file->private_data);
655 	file->private_data = NULL;
656 	return ret;
657 }
658 
659 /**
660  * gfs2_open - open a file
661  * @inode: the inode to open
662  * @file: the struct file for this opening
663  *
664  * After atomic_open, this function is only used for opening files
665  * which are already cached. We must still get the glock for regular
666  * files to ensure that we have the file size uptodate for the large
667  * file check which is in the common code. That is only an issue for
668  * regular files though.
669  *
670  * Returns: errno
671  */
672 
gfs2_open(struct inode * inode,struct file * file)673 static int gfs2_open(struct inode *inode, struct file *file)
674 {
675 	struct gfs2_inode *ip = GFS2_I(inode);
676 	struct gfs2_holder i_gh;
677 	int error;
678 	bool need_unlock = false;
679 
680 	if (S_ISREG(ip->i_inode.i_mode)) {
681 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
682 					   &i_gh);
683 		if (error)
684 			return error;
685 		need_unlock = true;
686 	}
687 
688 	error = gfs2_open_common(inode, file);
689 
690 	if (need_unlock)
691 		gfs2_glock_dq_uninit(&i_gh);
692 
693 	return error;
694 }
695 
696 /**
697  * gfs2_release - called to close a struct file
698  * @inode: the inode the struct file belongs to
699  * @file: the struct file being closed
700  *
701  * Returns: errno
702  */
703 
gfs2_release(struct inode * inode,struct file * file)704 static int gfs2_release(struct inode *inode, struct file *file)
705 {
706 	struct gfs2_inode *ip = GFS2_I(inode);
707 
708 	kfree(file->private_data);
709 	file->private_data = NULL;
710 
711 	if (file->f_mode & FMODE_WRITE) {
712 		if (gfs2_rs_active(&ip->i_res))
713 			gfs2_rs_delete(ip);
714 		gfs2_qa_put(ip);
715 	}
716 	return 0;
717 }
718 
719 /**
720  * gfs2_fsync - sync the dirty data for a file (across the cluster)
721  * @file: the file that points to the dentry
722  * @start: the start position in the file to sync
723  * @end: the end position in the file to sync
724  * @datasync: set if we can ignore timestamp changes
725  *
726  * We split the data flushing here so that we don't wait for the data
727  * until after we've also sent the metadata to disk. Note that for
728  * data=ordered, we will write & wait for the data at the log flush
729  * stage anyway, so this is unlikely to make much of a difference
730  * except in the data=writeback case.
731  *
732  * If the fdatawrite fails due to any reason except -EIO, we will
733  * continue the remainder of the fsync, although we'll still report
734  * the error at the end. This is to match filemap_write_and_wait_range()
735  * behaviour.
736  *
737  * Returns: errno
738  */
739 
gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)740 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
741 		      int datasync)
742 {
743 	struct address_space *mapping = file->f_mapping;
744 	struct inode *inode = mapping->host;
745 	int sync_state = inode->i_state & I_DIRTY;
746 	struct gfs2_inode *ip = GFS2_I(inode);
747 	int ret = 0, ret1 = 0;
748 
749 	if (mapping->nrpages) {
750 		ret1 = filemap_fdatawrite_range(mapping, start, end);
751 		if (ret1 == -EIO)
752 			return ret1;
753 	}
754 
755 	if (!gfs2_is_jdata(ip))
756 		sync_state &= ~I_DIRTY_PAGES;
757 	if (datasync)
758 		sync_state &= ~I_DIRTY_SYNC;
759 
760 	if (sync_state) {
761 		ret = sync_inode_metadata(inode, 1);
762 		if (ret)
763 			return ret;
764 		if (gfs2_is_jdata(ip))
765 			ret = file_write_and_wait(file);
766 		if (ret)
767 			return ret;
768 		gfs2_ail_flush(ip->i_gl, 1);
769 	}
770 
771 	if (mapping->nrpages)
772 		ret = file_fdatawait_range(file, start, end);
773 
774 	return ret ? ret : ret1;
775 }
776 
should_fault_in_pages(struct iov_iter * i,struct kiocb * iocb,size_t * prev_count,size_t * window_size)777 static inline bool should_fault_in_pages(struct iov_iter *i,
778 					 struct kiocb *iocb,
779 					 size_t *prev_count,
780 					 size_t *window_size)
781 {
782 	size_t count = iov_iter_count(i);
783 	size_t size, offs;
784 
785 	if (!count)
786 		return false;
787 	if (!user_backed_iter(i))
788 		return false;
789 
790 	/*
791 	 * Try to fault in multiple pages initially.  When that doesn't result
792 	 * in any progress, fall back to a single page.
793 	 */
794 	size = PAGE_SIZE;
795 	offs = offset_in_page(iocb->ki_pos);
796 	if (*prev_count != count) {
797 		size_t nr_dirtied;
798 
799 		nr_dirtied = max(current->nr_dirtied_pause -
800 				 current->nr_dirtied, 8);
801 		size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
802 	}
803 
804 	*prev_count = count;
805 	*window_size = size - offs;
806 	return true;
807 }
808 
gfs2_file_direct_read(struct kiocb * iocb,struct iov_iter * to,struct gfs2_holder * gh)809 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
810 				     struct gfs2_holder *gh)
811 {
812 	struct file *file = iocb->ki_filp;
813 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
814 	size_t prev_count = 0, window_size = 0;
815 	size_t read = 0;
816 	ssize_t ret;
817 
818 	/*
819 	 * In this function, we disable page faults when we're holding the
820 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
821 	 * that the inode glock may be dropped, fault in the pages manually,
822 	 * and retry.
823 	 *
824 	 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
825 	 * physical as well as manual page faults, and we need to disable both
826 	 * kinds.
827 	 *
828 	 * For direct I/O, gfs2 takes the inode glock in deferred mode.  This
829 	 * locking mode is compatible with other deferred holders, so multiple
830 	 * processes and nodes can do direct I/O to a file at the same time.
831 	 * There's no guarantee that reads or writes will be atomic.  Any
832 	 * coordination among readers and writers needs to happen externally.
833 	 */
834 
835 	if (!iov_iter_count(to))
836 		return 0; /* skip atime */
837 
838 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
839 retry:
840 	ret = gfs2_glock_nq(gh);
841 	if (ret)
842 		goto out_uninit;
843 	pagefault_disable();
844 	to->nofault = true;
845 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
846 			   IOMAP_DIO_PARTIAL, NULL, read);
847 	to->nofault = false;
848 	pagefault_enable();
849 	if (ret <= 0 && ret != -EFAULT)
850 		goto out_unlock;
851 	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
852 	if (ret > 0)
853 		read = ret;
854 
855 	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
856 		gfs2_glock_dq(gh);
857 		window_size -= fault_in_iov_iter_writeable(to, window_size);
858 		if (window_size)
859 			goto retry;
860 	}
861 out_unlock:
862 	if (gfs2_holder_queued(gh))
863 		gfs2_glock_dq(gh);
864 out_uninit:
865 	gfs2_holder_uninit(gh);
866 	/* User space doesn't expect partial success. */
867 	if (ret < 0)
868 		return ret;
869 	return read;
870 }
871 
gfs2_file_direct_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)872 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
873 				      struct gfs2_holder *gh)
874 {
875 	struct file *file = iocb->ki_filp;
876 	struct inode *inode = file->f_mapping->host;
877 	struct gfs2_inode *ip = GFS2_I(inode);
878 	size_t prev_count = 0, window_size = 0;
879 	size_t written = 0;
880 	bool enough_retries;
881 	ssize_t ret;
882 
883 	/*
884 	 * In this function, we disable page faults when we're holding the
885 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
886 	 * that the inode glock may be dropped, fault in the pages manually,
887 	 * and retry.
888 	 *
889 	 * For writes, iomap_dio_rw only triggers manual page faults, so we
890 	 * don't need to disable physical ones.
891 	 */
892 
893 	/*
894 	 * Deferred lock, even if its a write, since we do no allocation on
895 	 * this path. All we need to change is the atime, and this lock mode
896 	 * ensures that other nodes have flushed their buffered read caches
897 	 * (i.e. their page cache entries for this inode). We do not,
898 	 * unfortunately, have the option of only flushing a range like the
899 	 * VFS does.
900 	 */
901 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
902 retry:
903 	ret = gfs2_glock_nq(gh);
904 	if (ret)
905 		goto out_uninit;
906 	/* Silently fall back to buffered I/O when writing beyond EOF */
907 	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
908 		goto out_unlock;
909 
910 	from->nofault = true;
911 	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
912 			   IOMAP_DIO_PARTIAL, NULL, written);
913 	from->nofault = false;
914 	if (ret <= 0) {
915 		if (ret == -ENOTBLK)
916 			ret = 0;
917 		if (ret != -EFAULT)
918 			goto out_unlock;
919 	}
920 	/* No increment (+=) because iomap_dio_rw returns a cumulative value. */
921 	if (ret > 0)
922 		written = ret;
923 
924 	enough_retries = prev_count == iov_iter_count(from) &&
925 			 window_size <= PAGE_SIZE;
926 	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
927 		gfs2_glock_dq(gh);
928 		window_size -= fault_in_iov_iter_readable(from, window_size);
929 		if (window_size) {
930 			if (!enough_retries)
931 				goto retry;
932 			/* fall back to buffered I/O */
933 			ret = 0;
934 		}
935 	}
936 out_unlock:
937 	if (gfs2_holder_queued(gh))
938 		gfs2_glock_dq(gh);
939 out_uninit:
940 	gfs2_holder_uninit(gh);
941 	/* User space doesn't expect partial success. */
942 	if (ret < 0)
943 		return ret;
944 	return written;
945 }
946 
gfs2_file_read_iter(struct kiocb * iocb,struct iov_iter * to)947 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
948 {
949 	struct gfs2_inode *ip;
950 	struct gfs2_holder gh;
951 	size_t prev_count = 0, window_size = 0;
952 	size_t read = 0;
953 	ssize_t ret;
954 
955 	/*
956 	 * In this function, we disable page faults when we're holding the
957 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
958 	 * that the inode glock may be dropped, fault in the pages manually,
959 	 * and retry.
960 	 */
961 
962 	if (iocb->ki_flags & IOCB_DIRECT)
963 		return gfs2_file_direct_read(iocb, to, &gh);
964 
965 	pagefault_disable();
966 	iocb->ki_flags |= IOCB_NOIO;
967 	ret = generic_file_read_iter(iocb, to);
968 	iocb->ki_flags &= ~IOCB_NOIO;
969 	pagefault_enable();
970 	if (ret >= 0) {
971 		if (!iov_iter_count(to))
972 			return ret;
973 		read = ret;
974 	} else if (ret != -EFAULT) {
975 		if (ret != -EAGAIN)
976 			return ret;
977 		if (iocb->ki_flags & IOCB_NOWAIT)
978 			return ret;
979 	}
980 	ip = GFS2_I(iocb->ki_filp->f_mapping->host);
981 	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
982 retry:
983 	ret = gfs2_glock_nq(&gh);
984 	if (ret)
985 		goto out_uninit;
986 	pagefault_disable();
987 	ret = generic_file_read_iter(iocb, to);
988 	pagefault_enable();
989 	if (ret <= 0 && ret != -EFAULT)
990 		goto out_unlock;
991 	if (ret > 0)
992 		read += ret;
993 
994 	if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
995 		gfs2_glock_dq(&gh);
996 		window_size -= fault_in_iov_iter_writeable(to, window_size);
997 		if (window_size)
998 			goto retry;
999 	}
1000 out_unlock:
1001 	if (gfs2_holder_queued(&gh))
1002 		gfs2_glock_dq(&gh);
1003 out_uninit:
1004 	gfs2_holder_uninit(&gh);
1005 	return read ? read : ret;
1006 }
1007 
gfs2_file_buffered_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)1008 static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1009 					struct iov_iter *from,
1010 					struct gfs2_holder *gh)
1011 {
1012 	struct file *file = iocb->ki_filp;
1013 	struct inode *inode = file_inode(file);
1014 	struct gfs2_inode *ip = GFS2_I(inode);
1015 	struct gfs2_sbd *sdp = GFS2_SB(inode);
1016 	struct gfs2_holder *statfs_gh = NULL;
1017 	size_t prev_count = 0, window_size = 0;
1018 	size_t orig_count = iov_iter_count(from);
1019 	size_t written = 0;
1020 	ssize_t ret;
1021 
1022 	/*
1023 	 * In this function, we disable page faults when we're holding the
1024 	 * inode glock while doing I/O.  If a page fault occurs, we indicate
1025 	 * that the inode glock may be dropped, fault in the pages manually,
1026 	 * and retry.
1027 	 */
1028 
1029 	if (inode == sdp->sd_rindex) {
1030 		statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1031 		if (!statfs_gh)
1032 			return -ENOMEM;
1033 	}
1034 
1035 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1036 	if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1037 retry:
1038 		window_size -= fault_in_iov_iter_readable(from, window_size);
1039 		if (!window_size) {
1040 			ret = -EFAULT;
1041 			goto out_uninit;
1042 		}
1043 		from->count = min(from->count, window_size);
1044 	}
1045 	ret = gfs2_glock_nq(gh);
1046 	if (ret)
1047 		goto out_uninit;
1048 
1049 	if (inode == sdp->sd_rindex) {
1050 		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1051 
1052 		ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1053 					 GL_NOCACHE, statfs_gh);
1054 		if (ret)
1055 			goto out_unlock;
1056 	}
1057 
1058 	pagefault_disable();
1059 	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1060 	pagefault_enable();
1061 	if (ret > 0)
1062 		written += ret;
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 (gfs2_withdrawing_or_withdrawn(sdp)) {
1440 		if (fl->fl_type == F_UNLCK)
1441 			locks_lock_file_wait(file, fl);
1442 		return -EIO;
1443 	}
1444 	if (cmd == F_CANCELLK)
1445 		return dlm_posix_cancel(ls->ls_dlm, ip->i_no_addr, file, fl);
1446 	else if (IS_GETLK(cmd))
1447 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1448 	else if (fl->fl_type == F_UNLCK)
1449 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1450 	else
1451 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1452 }
1453 
__flock_holder_uninit(struct file * file,struct gfs2_holder * fl_gh)1454 static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1455 {
1456 	struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl);
1457 
1458 	/*
1459 	 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1460 	 * spinlock.
1461 	 */
1462 
1463 	spin_lock(&file->f_lock);
1464 	gfs2_holder_uninit(fl_gh);
1465 	spin_unlock(&file->f_lock);
1466 	gfs2_glock_put(gl);
1467 }
1468 
do_flock(struct file * file,int cmd,struct file_lock * fl)1469 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1470 {
1471 	struct gfs2_file *fp = file->private_data;
1472 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1473 	struct gfs2_inode *ip = GFS2_I(file_inode(file));
1474 	struct gfs2_glock *gl;
1475 	unsigned int state;
1476 	u16 flags;
1477 	int error = 0;
1478 	int sleeptime;
1479 
1480 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1481 	flags = GL_EXACT | GL_NOPID;
1482 	if (!IS_SETLKW(cmd))
1483 		flags |= LM_FLAG_TRY_1CB;
1484 
1485 	mutex_lock(&fp->f_fl_mutex);
1486 
1487 	if (gfs2_holder_initialized(fl_gh)) {
1488 		struct file_lock request;
1489 		if (fl_gh->gh_state == state)
1490 			goto out;
1491 		locks_init_lock(&request);
1492 		request.fl_type = F_UNLCK;
1493 		request.fl_flags = FL_FLOCK;
1494 		locks_lock_file_wait(file, &request);
1495 		gfs2_glock_dq(fl_gh);
1496 		gfs2_holder_reinit(state, flags, fl_gh);
1497 	} else {
1498 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1499 				       &gfs2_flock_glops, CREATE, &gl);
1500 		if (error)
1501 			goto out;
1502 		spin_lock(&file->f_lock);
1503 		gfs2_holder_init(gl, state, flags, fl_gh);
1504 		spin_unlock(&file->f_lock);
1505 		gfs2_glock_put(gl);
1506 	}
1507 	for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1508 		error = gfs2_glock_nq(fl_gh);
1509 		if (error != GLR_TRYFAILED)
1510 			break;
1511 		fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1512 		fl_gh->gh_flags |= LM_FLAG_TRY;
1513 		msleep(sleeptime);
1514 	}
1515 	if (error) {
1516 		__flock_holder_uninit(file, fl_gh);
1517 		if (error == GLR_TRYFAILED)
1518 			error = -EAGAIN;
1519 	} else {
1520 		error = locks_lock_file_wait(file, fl);
1521 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1522 	}
1523 
1524 out:
1525 	mutex_unlock(&fp->f_fl_mutex);
1526 	return error;
1527 }
1528 
do_unflock(struct file * file,struct file_lock * fl)1529 static void do_unflock(struct file *file, struct file_lock *fl)
1530 {
1531 	struct gfs2_file *fp = file->private_data;
1532 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1533 
1534 	mutex_lock(&fp->f_fl_mutex);
1535 	locks_lock_file_wait(file, fl);
1536 	if (gfs2_holder_initialized(fl_gh)) {
1537 		gfs2_glock_dq(fl_gh);
1538 		__flock_holder_uninit(file, fl_gh);
1539 	}
1540 	mutex_unlock(&fp->f_fl_mutex);
1541 }
1542 
1543 /**
1544  * gfs2_flock - acquire/release a flock lock on a file
1545  * @file: the file pointer
1546  * @cmd: either modify or retrieve lock state, possibly wait
1547  * @fl: type and range of lock
1548  *
1549  * Returns: errno
1550  */
1551 
gfs2_flock(struct file * file,int cmd,struct file_lock * fl)1552 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1553 {
1554 	if (!(fl->fl_flags & FL_FLOCK))
1555 		return -ENOLCK;
1556 
1557 	if (fl->fl_type == F_UNLCK) {
1558 		do_unflock(file, fl);
1559 		return 0;
1560 	} else {
1561 		return do_flock(file, cmd, fl);
1562 	}
1563 }
1564 
1565 const struct file_operations gfs2_file_fops = {
1566 	.llseek		= gfs2_llseek,
1567 	.read_iter	= gfs2_file_read_iter,
1568 	.write_iter	= gfs2_file_write_iter,
1569 	.iopoll		= iocb_bio_iopoll,
1570 	.unlocked_ioctl	= gfs2_ioctl,
1571 	.compat_ioctl	= gfs2_compat_ioctl,
1572 	.mmap		= gfs2_mmap,
1573 	.open		= gfs2_open,
1574 	.release	= gfs2_release,
1575 	.fsync		= gfs2_fsync,
1576 	.lock		= gfs2_lock,
1577 	.flock		= gfs2_flock,
1578 	.splice_read	= copy_splice_read,
1579 	.splice_write	= gfs2_file_splice_write,
1580 	.setlease	= simple_nosetlease,
1581 	.fallocate	= gfs2_fallocate,
1582 };
1583 
1584 const struct file_operations gfs2_dir_fops = {
1585 	.iterate_shared	= gfs2_readdir,
1586 	.unlocked_ioctl	= gfs2_ioctl,
1587 	.compat_ioctl	= gfs2_compat_ioctl,
1588 	.open		= gfs2_open,
1589 	.release	= gfs2_release,
1590 	.fsync		= gfs2_fsync,
1591 	.lock		= gfs2_lock,
1592 	.flock		= gfs2_flock,
1593 	.llseek		= default_llseek,
1594 };
1595 
1596 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1597 
1598 const struct file_operations gfs2_file_fops_nolock = {
1599 	.llseek		= gfs2_llseek,
1600 	.read_iter	= gfs2_file_read_iter,
1601 	.write_iter	= gfs2_file_write_iter,
1602 	.iopoll		= iocb_bio_iopoll,
1603 	.unlocked_ioctl	= gfs2_ioctl,
1604 	.compat_ioctl	= gfs2_compat_ioctl,
1605 	.mmap		= gfs2_mmap,
1606 	.open		= gfs2_open,
1607 	.release	= gfs2_release,
1608 	.fsync		= gfs2_fsync,
1609 	.splice_read	= copy_splice_read,
1610 	.splice_write	= gfs2_file_splice_write,
1611 	.setlease	= generic_setlease,
1612 	.fallocate	= gfs2_fallocate,
1613 };
1614 
1615 const struct file_operations gfs2_dir_fops_nolock = {
1616 	.iterate_shared	= gfs2_readdir,
1617 	.unlocked_ioctl	= gfs2_ioctl,
1618 	.compat_ioctl	= gfs2_compat_ioctl,
1619 	.open		= gfs2_open,
1620 	.release	= gfs2_release,
1621 	.fsync		= gfs2_fsync,
1622 	.llseek		= default_llseek,
1623 };
1624 
1625