Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 4 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
b3b94faa DT |
7 | #include <linux/slab.h> |
8 | #include <linux/spinlock.h> | |
8d098070 | 9 | #include <linux/compat.h> |
b3b94faa DT |
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> | |
f58ba889 | 16 | #include <linux/mount.h> |
18ec7d5c | 17 | #include <linux/fs.h> |
5c676f6d | 18 | #include <linux/gfs2_ondisk.h> |
2fe17c10 CH |
19 | #include <linux/falloc.h> |
20 | #include <linux/swap.h> | |
71b86f56 | 21 | #include <linux/crc32.h> |
33c3de32 | 22 | #include <linux/writeback.h> |
7c0f6ba6 | 23 | #include <linux/uaccess.h> |
f057f6cd SW |
24 | #include <linux/dlm.h> |
25 | #include <linux/dlm_plock.h> | |
2ddfbdd6 | 26 | #include <linux/delay.h> |
64bc06bb | 27 | #include <linux/backing-dev.h> |
b3b94faa DT |
28 | |
29 | #include "gfs2.h" | |
5c676f6d | 30 | #include "incore.h" |
b3b94faa | 31 | #include "bmap.h" |
64bc06bb | 32 | #include "aops.h" |
b3b94faa DT |
33 | #include "dir.h" |
34 | #include "glock.h" | |
35 | #include "glops.h" | |
36 | #include "inode.h" | |
b3b94faa DT |
37 | #include "log.h" |
38 | #include "meta_io.h" | |
b3b94faa DT |
39 | #include "quota.h" |
40 | #include "rgrp.h" | |
41 | #include "trans.h" | |
5c676f6d | 42 | #include "util.h" |
b3b94faa | 43 | |
b3b94faa DT |
44 | /** |
45 | * gfs2_llseek - seek to a location in a file | |
46 | * @file: the file | |
47 | * @offset: the offset | |
965c8e59 | 48 | * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) |
b3b94faa DT |
49 | * |
50 | * SEEK_END requires the glock for the file because it references the | |
51 | * file's size. | |
52 | * | |
53 | * Returns: The new offset, or errno | |
54 | */ | |
55 | ||
965c8e59 | 56 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence) |
b3b94faa | 57 | { |
feaa7bba | 58 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa DT |
59 | struct gfs2_holder i_gh; |
60 | loff_t error; | |
61 | ||
965c8e59 | 62 | switch (whence) { |
3a27411c | 63 | case SEEK_END: |
b3b94faa DT |
64 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
65 | &i_gh); | |
66 | if (!error) { | |
965c8e59 | 67 | error = generic_file_llseek(file, offset, whence); |
b3b94faa DT |
68 | gfs2_glock_dq_uninit(&i_gh); |
69 | } | |
9453615a | 70 | break; |
3a27411c AG |
71 | |
72 | case SEEK_DATA: | |
73 | error = gfs2_seek_data(file, offset); | |
74 | break; | |
75 | ||
76 | case SEEK_HOLE: | |
77 | error = gfs2_seek_hole(file, offset); | |
78 | break; | |
79 | ||
9453615a SW |
80 | case SEEK_CUR: |
81 | case SEEK_SET: | |
3a27411c AG |
82 | /* |
83 | * These don't reference inode->i_size and don't depend on the | |
84 | * block mapping, so we don't need the glock. | |
85 | */ | |
965c8e59 | 86 | error = generic_file_llseek(file, offset, whence); |
9453615a SW |
87 | break; |
88 | default: | |
89 | error = -EINVAL; | |
90 | } | |
b3b94faa DT |
91 | |
92 | return error; | |
93 | } | |
94 | ||
b3b94faa | 95 | /** |
d81a8ef5 | 96 | * gfs2_readdir - Iterator for a directory |
b3b94faa | 97 | * @file: The directory to read from |
d81a8ef5 | 98 | * @ctx: What to feed directory entries to |
b3b94faa DT |
99 | * |
100 | * Returns: errno | |
101 | */ | |
102 | ||
d81a8ef5 | 103 | static int gfs2_readdir(struct file *file, struct dir_context *ctx) |
b3b94faa | 104 | { |
71b86f56 | 105 | struct inode *dir = file->f_mapping->host; |
feaa7bba | 106 | struct gfs2_inode *dip = GFS2_I(dir); |
b3b94faa | 107 | struct gfs2_holder d_gh; |
b3b94faa DT |
108 | int error; |
109 | ||
d81a8ef5 AV |
110 | error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); |
111 | if (error) | |
b3b94faa | 112 | return error; |
b3b94faa | 113 | |
d81a8ef5 | 114 | error = gfs2_dir_read(dir, ctx, &file->f_ra); |
b3b94faa DT |
115 | |
116 | gfs2_glock_dq_uninit(&d_gh); | |
117 | ||
b3b94faa DT |
118 | return error; |
119 | } | |
120 | ||
128e5eba | 121 | /** |
b16f7e57 | 122 | * fsflag_gfs2flag |
128e5eba | 123 | * |
b16f7e57 AG |
124 | * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories, |
125 | * and to GFS2_DIF_JDATA for non-directories. | |
128e5eba | 126 | */ |
b16f7e57 AG |
127 | static struct { |
128 | u32 fsflag; | |
129 | u32 gfsflag; | |
130 | } fsflag_gfs2flag[] = { | |
131 | {FS_SYNC_FL, GFS2_DIF_SYNC}, | |
132 | {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE}, | |
133 | {FS_APPEND_FL, GFS2_DIF_APPENDONLY}, | |
134 | {FS_NOATIME_FL, GFS2_DIF_NOATIME}, | |
135 | {FS_INDEX_FL, GFS2_DIF_EXHASH}, | |
136 | {FS_TOPDIR_FL, GFS2_DIF_TOPDIR}, | |
137 | {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA}, | |
7ea9ea83 | 138 | }; |
71b86f56 | 139 | |
5aca2842 DW |
140 | static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags) |
141 | { | |
142 | int i; | |
143 | u32 fsflags = 0; | |
144 | ||
145 | if (S_ISDIR(inode->i_mode)) | |
146 | gfsflags &= ~GFS2_DIF_JDATA; | |
147 | else | |
148 | gfsflags &= ~GFS2_DIF_INHERIT_JDATA; | |
149 | ||
150 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) | |
151 | if (gfsflags & fsflag_gfs2flag[i].gfsflag) | |
152 | fsflags |= fsflag_gfs2flag[i].fsflag; | |
153 | return fsflags; | |
154 | } | |
155 | ||
b09e593d | 156 | static int gfs2_get_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 157 | { |
496ad9aa | 158 | struct inode *inode = file_inode(filp); |
feaa7bba | 159 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 | 160 | struct gfs2_holder gh; |
5aca2842 DW |
161 | int error; |
162 | u32 fsflags; | |
71b86f56 | 163 | |
719ee344 SW |
164 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
165 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 166 | if (error) |
9c7fe835 | 167 | goto out_uninit; |
907b9bce | 168 | |
5aca2842 | 169 | fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); |
b16f7e57 | 170 | |
128e5eba | 171 | if (put_user(fsflags, ptr)) |
71b86f56 SW |
172 | error = -EFAULT; |
173 | ||
3cc3f710 | 174 | gfs2_glock_dq(&gh); |
9c7fe835 | 175 | out_uninit: |
71b86f56 SW |
176 | gfs2_holder_uninit(&gh); |
177 | return error; | |
178 | } | |
179 | ||
6b124d8d SW |
180 | void gfs2_set_inode_flags(struct inode *inode) |
181 | { | |
182 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
183 | unsigned int flags = inode->i_flags; |
184 | ||
9964afbb SW |
185 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
186 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 187 | flags |= S_NOSEC; |
383f01fb | 188 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 189 | flags |= S_IMMUTABLE; |
383f01fb | 190 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 191 | flags |= S_APPEND; |
383f01fb | 192 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 193 | flags |= S_NOATIME; |
383f01fb | 194 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
195 | flags |= S_SYNC; |
196 | inode->i_flags = flags; | |
197 | } | |
198 | ||
71b86f56 SW |
199 | /* Flags that can be set by user space */ |
200 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
201 | GFS2_DIF_IMMUTABLE| \ |
202 | GFS2_DIF_APPENDONLY| \ | |
203 | GFS2_DIF_NOATIME| \ | |
204 | GFS2_DIF_SYNC| \ | |
23d0bb83 | 205 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
206 | GFS2_DIF_INHERIT_JDATA) |
207 | ||
208 | /** | |
9dd868e1 FF |
209 | * do_gfs2_set_flags - set flags on an inode |
210 | * @filp: file pointer | |
211 | * @reqflags: The flags to set | |
71b86f56 | 212 | * @mask: Indicates which flags are valid |
5aca2842 | 213 | * @fsflags: The FS_* inode flags passed in |
71b86f56 SW |
214 | * |
215 | */ | |
5aca2842 DW |
216 | static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask, |
217 | const u32 fsflags) | |
71b86f56 | 218 | { |
496ad9aa | 219 | struct inode *inode = file_inode(filp); |
feaa7bba SW |
220 | struct gfs2_inode *ip = GFS2_I(inode); |
221 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
222 | struct buffer_head *bh; |
223 | struct gfs2_holder gh; | |
224 | int error; | |
5aca2842 | 225 | u32 new_flags, flags, oldflags; |
71b86f56 | 226 | |
a561be71 | 227 | error = mnt_want_write_file(filp); |
52f341cf | 228 | if (error) |
71b86f56 SW |
229 | return error; |
230 | ||
f58ba889 MS |
231 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
232 | if (error) | |
233 | goto out_drop_write; | |
234 | ||
5aca2842 DW |
235 | oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); |
236 | error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags); | |
237 | if (error) | |
238 | goto out; | |
239 | ||
7df0e039 | 240 | error = -EACCES; |
2e149670 | 241 | if (!inode_owner_or_capable(inode)) |
7df0e039 SW |
242 | goto out; |
243 | ||
244 | error = 0; | |
383f01fb | 245 | flags = ip->i_diskflags; |
55eccc6d | 246 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
247 | if ((new_flags ^ flags) == 0) |
248 | goto out; | |
249 | ||
71b86f56 SW |
250 | error = -EPERM; |
251 | if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) | |
252 | goto out; | |
253 | if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) | |
254 | goto out; | |
907b9bce | 255 | if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && |
b9cb9813 | 256 | !capable(CAP_LINUX_IMMUTABLE)) |
71b86f56 | 257 | goto out; |
b9cb9813 | 258 | if (!IS_IMMUTABLE(inode)) { |
10556cb2 | 259 | error = gfs2_permission(inode, MAY_WRITE); |
b9cb9813 SW |
260 | if (error) |
261 | goto out; | |
262 | } | |
5561093e | 263 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
cc555b09 | 264 | if (new_flags & GFS2_DIF_JDATA) |
c1696fb8 | 265 | gfs2_log_flush(sdp, ip->i_gl, |
805c0907 BP |
266 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
267 | GFS2_LFC_SET_FLAGS); | |
5561093e SW |
268 | error = filemap_fdatawrite(inode->i_mapping); |
269 | if (error) | |
270 | goto out; | |
271 | error = filemap_fdatawait(inode->i_mapping); | |
272 | if (error) | |
273 | goto out; | |
cc555b09 BP |
274 | if (new_flags & GFS2_DIF_JDATA) |
275 | gfs2_ordered_del_inode(ip); | |
5561093e | 276 | } |
55eccc6d | 277 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
278 | if (error) |
279 | goto out; | |
55eccc6d SW |
280 | error = gfs2_meta_inode_buffer(ip, &bh); |
281 | if (error) | |
282 | goto out_trans_end; | |
9b7c2ddb | 283 | inode->i_ctime = current_time(inode); |
350a9b0a | 284 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 285 | ip->i_diskflags = new_flags; |
539e5d6b | 286 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 287 | brelse(bh); |
6b124d8d | 288 | gfs2_set_inode_flags(inode); |
5561093e | 289 | gfs2_set_aops(inode); |
55eccc6d SW |
290 | out_trans_end: |
291 | gfs2_trans_end(sdp); | |
71b86f56 SW |
292 | out: |
293 | gfs2_glock_dq_uninit(&gh); | |
f58ba889 | 294 | out_drop_write: |
2a79f17e | 295 | mnt_drop_write_file(filp); |
71b86f56 SW |
296 | return error; |
297 | } | |
298 | ||
b09e593d | 299 | static int gfs2_set_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 300 | { |
496ad9aa | 301 | struct inode *inode = file_inode(filp); |
b16f7e57 AG |
302 | u32 fsflags, gfsflags = 0; |
303 | u32 mask; | |
304 | int i; | |
7df0e039 | 305 | |
128e5eba | 306 | if (get_user(fsflags, ptr)) |
71b86f56 | 307 | return -EFAULT; |
7df0e039 | 308 | |
b16f7e57 AG |
309 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { |
310 | if (fsflags & fsflag_gfs2flag[i].fsflag) { | |
311 | fsflags &= ~fsflag_gfs2flag[i].fsflag; | |
312 | gfsflags |= fsflag_gfs2flag[i].gfsflag; | |
313 | } | |
314 | } | |
315 | if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) | |
316 | return -EINVAL; | |
317 | ||
318 | mask = GFS2_FLAGS_USER_SET; | |
319 | if (S_ISDIR(inode->i_mode)) { | |
320 | mask &= ~GFS2_DIF_JDATA; | |
321 | } else { | |
322 | /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ | |
323 | if (gfsflags & GFS2_DIF_TOPDIR) | |
324 | return -EINVAL; | |
325 | mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); | |
b9af7ca6 | 326 | } |
b16f7e57 | 327 | |
5aca2842 | 328 | return do_gfs2_set_flags(filp, gfsflags, mask, fsflags); |
71b86f56 SW |
329 | } |
330 | ||
6ddc5c3d SW |
331 | static int gfs2_getlabel(struct file *filp, char __user *label) |
332 | { | |
333 | struct inode *inode = file_inode(filp); | |
334 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
335 | ||
336 | if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN)) | |
337 | return -EFAULT; | |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
b09e593d | 342 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
343 | { |
344 | switch(cmd) { | |
128e5eba | 345 | case FS_IOC_GETFLAGS: |
b09e593d | 346 | return gfs2_get_flags(filp, (u32 __user *)arg); |
128e5eba | 347 | case FS_IOC_SETFLAGS: |
b09e593d | 348 | return gfs2_set_flags(filp, (u32 __user *)arg); |
66fc061b SW |
349 | case FITRIM: |
350 | return gfs2_fitrim(filp, (void __user *)arg); | |
6ddc5c3d SW |
351 | case FS_IOC_GETFSLABEL: |
352 | return gfs2_getlabel(filp, (char __user *)arg); | |
71b86f56 | 353 | } |
6ddc5c3d | 354 | |
71b86f56 SW |
355 | return -ENOTTY; |
356 | } | |
357 | ||
8d098070 AB |
358 | #ifdef CONFIG_COMPAT |
359 | static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) | |
360 | { | |
361 | switch(cmd) { | |
362 | /* These are just misnamed, they actually get/put from/to user an int */ | |
363 | case FS_IOC32_GETFLAGS: | |
364 | cmd = FS_IOC_GETFLAGS; | |
365 | break; | |
366 | case FS_IOC32_SETFLAGS: | |
367 | cmd = FS_IOC_SETFLAGS; | |
368 | break; | |
369 | /* Keep this list in sync with gfs2_ioctl */ | |
370 | case FITRIM: | |
371 | case FS_IOC_GETFSLABEL: | |
372 | break; | |
373 | default: | |
374 | return -ENOIOCTLCMD; | |
375 | } | |
376 | ||
377 | return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)); | |
378 | } | |
379 | #else | |
380 | #define gfs2_compat_ioctl NULL | |
381 | #endif | |
382 | ||
da1dfb6a SW |
383 | /** |
384 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 385 | * @filep: The struct file |
da1dfb6a SW |
386 | * @offset: The file offset of the write |
387 | * @size: The length of the write | |
388 | * | |
389 | * When we are about to do a write, this function records the total | |
390 | * write size in order to provide a suitable hint to the lower layers | |
391 | * about how many blocks will be required. | |
392 | * | |
393 | */ | |
394 | ||
395 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
396 | { | |
496ad9aa | 397 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
398 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
399 | struct gfs2_inode *ip = GFS2_I(inode); | |
400 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
401 | int hint = min_t(size_t, INT_MAX, blks); | |
402 | ||
21f09c43 AG |
403 | if (hint > atomic_read(&ip->i_sizehint)) |
404 | atomic_set(&ip->i_sizehint, hint); | |
da1dfb6a SW |
405 | } |
406 | ||
3cc3f710 | 407 | /** |
35af80ae | 408 | * gfs2_allocate_page_backing - Allocate blocks for a write fault |
3cc3f710 | 409 | * @page: The (locked) page to allocate backing for |
f53056c4 | 410 | * @length: Size of the allocation |
3cc3f710 | 411 | * |
35af80ae CH |
412 | * We try to allocate all the blocks required for the page in one go. This |
413 | * might fail for various reasons, so we keep trying until all the blocks to | |
414 | * back this page are allocated. If some of the blocks are already allocated, | |
415 | * that is ok too. | |
3cc3f710 | 416 | */ |
f53056c4 | 417 | static int gfs2_allocate_page_backing(struct page *page, unsigned int length) |
3cc3f710 | 418 | { |
35af80ae | 419 | u64 pos = page_offset(page); |
3cc3f710 SW |
420 | |
421 | do { | |
35af80ae CH |
422 | struct iomap iomap = { }; |
423 | ||
f53056c4 | 424 | if (gfs2_iomap_get_alloc(page->mapping->host, pos, length, &iomap)) |
3cc3f710 | 425 | return -EIO; |
35af80ae | 426 | |
f53056c4 AG |
427 | if (length < iomap.length) |
428 | iomap.length = length; | |
429 | length -= iomap.length; | |
35af80ae | 430 | pos += iomap.length; |
f53056c4 | 431 | } while (length > 0); |
35af80ae | 432 | |
3cc3f710 SW |
433 | return 0; |
434 | } | |
435 | ||
436 | /** | |
437 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
438 | * @vma: The virtual memory area | |
9dd868e1 | 439 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
440 | * |
441 | * When the page becomes writable, we need to ensure that we have | |
442 | * blocks allocated on disk to back that page. | |
443 | */ | |
444 | ||
109dbb1e | 445 | static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) |
3cc3f710 | 446 | { |
c2ec175c | 447 | struct page *page = vmf->page; |
11bac800 | 448 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3cc3f710 SW |
449 | struct gfs2_inode *ip = GFS2_I(inode); |
450 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 451 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
184b4e60 | 452 | u64 offset = page_offset(page); |
3cc3f710 | 453 | unsigned int data_blocks, ind_blocks, rblocks; |
3cc3f710 | 454 | struct gfs2_holder gh; |
184b4e60 | 455 | unsigned int length; |
13d921e3 | 456 | loff_t size; |
3cc3f710 SW |
457 | int ret; |
458 | ||
39263d5e | 459 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 460 | |
719ee344 SW |
461 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
462 | ret = gfs2_glock_nq(&gh); | |
3cc3f710 | 463 | if (ret) |
2b3dcf35 | 464 | goto out_uninit; |
3cc3f710 | 465 | |
184b4e60 AG |
466 | /* Check page index against inode size */ |
467 | size = i_size_read(inode); | |
468 | if (offset >= size) { | |
469 | ret = -EINVAL; | |
470 | goto out_unlock; | |
471 | } | |
472 | ||
d7c436cd | 473 | /* Update file times before taking page lock */ |
11bac800 | 474 | file_update_time(vmf->vma->vm_file); |
d7c436cd | 475 | |
184b4e60 AG |
476 | /* page is wholly or partially inside EOF */ |
477 | if (offset > size - PAGE_SIZE) | |
478 | length = offset_in_page(size); | |
479 | else | |
480 | length = PAGE_SIZE; | |
481 | ||
482 | gfs2_size_hint(vmf->vma->vm_file, offset, length); | |
483 | ||
9c538837 SW |
484 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
485 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
486 | ||
184b4e60 AG |
487 | /* |
488 | * iomap_writepage / iomap_writepages currently don't support inline | |
489 | * files, so always unstuff here. | |
490 | */ | |
491 | ||
492 | if (!gfs2_is_stuffed(ip) && | |
493 | !gfs2_write_alloc_required(ip, offset, length)) { | |
13d921e3 SW |
494 | lock_page(page); |
495 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
496 | ret = -EAGAIN; | |
497 | unlock_page(page); | |
498 | } | |
3cc3f710 | 499 | goto out_unlock; |
13d921e3 SW |
500 | } |
501 | ||
5407e242 BP |
502 | ret = gfs2_rindex_update(sdp); |
503 | if (ret) | |
6dbd8224 SW |
504 | goto out_unlock; |
505 | ||
184b4e60 | 506 | gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks); |
7b9cff46 | 507 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
508 | ret = gfs2_quota_lock_check(ip, &ap); |
509 | if (ret) | |
510 | goto out_unlock; | |
7b9cff46 | 511 | ret = gfs2_inplace_reserve(ip, &ap); |
3cc3f710 SW |
512 | if (ret) |
513 | goto out_quota_unlock; | |
514 | ||
515 | rblocks = RES_DINODE + ind_blocks; | |
516 | if (gfs2_is_jdata(ip)) | |
517 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 518 | if (ind_blocks || data_blocks) { |
3cc3f710 | 519 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 520 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 521 | } |
3cc3f710 SW |
522 | ret = gfs2_trans_begin(sdp, rblocks, 0); |
523 | if (ret) | |
524 | goto out_trans_fail; | |
525 | ||
526 | lock_page(page); | |
13d921e3 SW |
527 | ret = -EAGAIN; |
528 | /* If truncated, we must retry the operation, we may have raced | |
529 | * with the glock demotion code. | |
530 | */ | |
531 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) | |
532 | goto out_trans_end; | |
533 | ||
534 | /* Unstuff, if required, and allocate backing blocks for page */ | |
b7fe2e39 | 535 | ret = 0; |
13d921e3 | 536 | if (gfs2_is_stuffed(ip)) |
3cc3f710 | 537 | ret = gfs2_unstuff_dinode(ip, page); |
13d921e3 | 538 | if (ret == 0) |
184b4e60 | 539 | ret = gfs2_allocate_page_backing(page, length); |
3cc3f710 | 540 | |
13d921e3 SW |
541 | out_trans_end: |
542 | if (ret) | |
543 | unlock_page(page); | |
3cc3f710 SW |
544 | gfs2_trans_end(sdp); |
545 | out_trans_fail: | |
546 | gfs2_inplace_release(ip); | |
547 | out_quota_unlock: | |
548 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
549 | out_unlock: |
550 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 551 | out_uninit: |
3cc3f710 | 552 | gfs2_holder_uninit(&gh); |
13d921e3 SW |
553 | if (ret == 0) { |
554 | set_page_dirty(page); | |
1d1d1a76 | 555 | wait_for_stable_page(page); |
13d921e3 | 556 | } |
39263d5e | 557 | sb_end_pagefault(inode->i_sb); |
13d921e3 | 558 | return block_page_mkwrite_return(ret); |
3cc3f710 SW |
559 | } |
560 | ||
f0f37e2f | 561 | static const struct vm_operations_struct gfs2_vm_ops = { |
3cc3f710 | 562 | .fault = filemap_fault, |
f1820361 | 563 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
564 | .page_mkwrite = gfs2_page_mkwrite, |
565 | }; | |
566 | ||
b3b94faa DT |
567 | /** |
568 | * gfs2_mmap - | |
569 | * @file: The file to map | |
570 | * @vma: The VMA which described the mapping | |
571 | * | |
48bf2b17 SW |
572 | * There is no need to get a lock here unless we should be updating |
573 | * atime. We ignore any locking errors since the only consequence is | |
574 | * a missed atime update (which will just be deferred until later). | |
575 | * | |
576 | * Returns: 0 | |
b3b94faa DT |
577 | */ |
578 | ||
579 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
580 | { | |
feaa7bba | 581 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 582 | |
b9c93bb7 SW |
583 | if (!(file->f_flags & O_NOATIME) && |
584 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
585 | struct gfs2_holder i_gh; |
586 | int error; | |
b3b94faa | 587 | |
3d162688 BM |
588 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
589 | &i_gh); | |
b9c93bb7 SW |
590 | if (error) |
591 | return error; | |
3d162688 BM |
592 | /* grab lock to update inode */ |
593 | gfs2_glock_dq_uninit(&i_gh); | |
594 | file_accessed(file); | |
48bf2b17 | 595 | } |
3cc3f710 | 596 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 597 | |
48bf2b17 | 598 | return 0; |
b3b94faa DT |
599 | } |
600 | ||
601 | /** | |
6d4ade98 SW |
602 | * gfs2_open_common - This is common to open and atomic_open |
603 | * @inode: The inode being opened | |
604 | * @file: The file being opened | |
b3b94faa | 605 | * |
6d4ade98 SW |
606 | * This maybe called under a glock or not depending upon how it has |
607 | * been called. We must always be called under a glock for regular | |
608 | * files, however. For other file types, it does not matter whether | |
609 | * we hold the glock or not. | |
610 | * | |
611 | * Returns: Error code or 0 for success | |
b3b94faa DT |
612 | */ |
613 | ||
6d4ade98 | 614 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 615 | { |
b3b94faa | 616 | struct gfs2_file *fp; |
6d4ade98 SW |
617 | int ret; |
618 | ||
619 | if (S_ISREG(inode->i_mode)) { | |
620 | ret = generic_file_open(inode, file); | |
621 | if (ret) | |
622 | return ret; | |
623 | } | |
b3b94faa | 624 | |
6d4ade98 | 625 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
626 | if (!fp) |
627 | return -ENOMEM; | |
628 | ||
f55ab26a | 629 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 630 | |
feaa7bba | 631 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 632 | file->private_data = fp; |
2fba46a0 BP |
633 | if (file->f_mode & FMODE_WRITE) { |
634 | ret = gfs2_qa_get(GFS2_I(inode)); | |
635 | if (ret) | |
636 | goto fail; | |
637 | } | |
6d4ade98 | 638 | return 0; |
2fba46a0 BP |
639 | |
640 | fail: | |
641 | kfree(file->private_data); | |
642 | file->private_data = NULL; | |
643 | return ret; | |
6d4ade98 SW |
644 | } |
645 | ||
646 | /** | |
647 | * gfs2_open - open a file | |
648 | * @inode: the inode to open | |
649 | * @file: the struct file for this opening | |
650 | * | |
651 | * After atomic_open, this function is only used for opening files | |
652 | * which are already cached. We must still get the glock for regular | |
653 | * files to ensure that we have the file size uptodate for the large | |
654 | * file check which is in the common code. That is only an issue for | |
655 | * regular files though. | |
656 | * | |
657 | * Returns: errno | |
658 | */ | |
659 | ||
660 | static int gfs2_open(struct inode *inode, struct file *file) | |
661 | { | |
662 | struct gfs2_inode *ip = GFS2_I(inode); | |
663 | struct gfs2_holder i_gh; | |
664 | int error; | |
665 | bool need_unlock = false; | |
b3b94faa | 666 | |
b60623c2 | 667 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
668 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
669 | &i_gh); | |
670 | if (error) | |
6d4ade98 SW |
671 | return error; |
672 | need_unlock = true; | |
673 | } | |
b3b94faa | 674 | |
6d4ade98 | 675 | error = gfs2_open_common(inode, file); |
b3b94faa | 676 | |
6d4ade98 | 677 | if (need_unlock) |
b3b94faa | 678 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 679 | |
b3b94faa DT |
680 | return error; |
681 | } | |
682 | ||
683 | /** | |
df3fd117 | 684 | * gfs2_release - called to close a struct file |
b3b94faa DT |
685 | * @inode: the inode the struct file belongs to |
686 | * @file: the struct file being closed | |
687 | * | |
688 | * Returns: errno | |
689 | */ | |
690 | ||
df3fd117 | 691 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 692 | { |
0a305e49 | 693 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 694 | |
8e2e0047 | 695 | kfree(file->private_data); |
5c676f6d | 696 | file->private_data = NULL; |
b3b94faa | 697 | |
1595548f AG |
698 | if (file->f_mode & FMODE_WRITE) { |
699 | gfs2_rs_delete(ip, &inode->i_writecount); | |
700 | gfs2_qa_put(ip); | |
701 | } | |
b3b94faa DT |
702 | return 0; |
703 | } | |
704 | ||
705 | /** | |
706 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
707 | * @file: the file that points to the dentry |
708 | * @start: the start position in the file to sync | |
709 | * @end: the end position in the file to sync | |
dba898b0 | 710 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 711 | * |
2f0264d5 SW |
712 | * We split the data flushing here so that we don't wait for the data |
713 | * until after we've also sent the metadata to disk. Note that for | |
714 | * data=ordered, we will write & wait for the data at the log flush | |
715 | * stage anyway, so this is unlikely to make much of a difference | |
716 | * except in the data=writeback case. | |
717 | * | |
718 | * If the fdatawrite fails due to any reason except -EIO, we will | |
719 | * continue the remainder of the fsync, although we'll still report | |
720 | * the error at the end. This is to match filemap_write_and_wait_range() | |
721 | * behaviour. | |
34126f9f | 722 | * |
b3b94faa DT |
723 | * Returns: errno |
724 | */ | |
725 | ||
02c24a82 JB |
726 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
727 | int datasync) | |
b3b94faa | 728 | { |
2f0264d5 SW |
729 | struct address_space *mapping = file->f_mapping; |
730 | struct inode *inode = mapping->host; | |
0ae45f63 | 731 | int sync_state = inode->i_state & I_DIRTY_ALL; |
dba898b0 | 732 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 733 | int ret = 0, ret1 = 0; |
b3b94faa | 734 | |
2f0264d5 SW |
735 | if (mapping->nrpages) { |
736 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
737 | if (ret1 == -EIO) | |
738 | return ret1; | |
739 | } | |
02c24a82 | 740 | |
0c901809 BM |
741 | if (!gfs2_is_jdata(ip)) |
742 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 743 | if (datasync) |
0ae45f63 | 744 | sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); |
b3b94faa | 745 | |
dba898b0 SW |
746 | if (sync_state) { |
747 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 748 | if (ret) |
dba898b0 | 749 | return ret; |
f1818529 | 750 | if (gfs2_is_jdata(ip)) |
d07a6ac7 JL |
751 | ret = file_write_and_wait(file); |
752 | if (ret) | |
753 | return ret; | |
b5b24d7a | 754 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
755 | } |
756 | ||
2f0264d5 | 757 | if (mapping->nrpages) |
d07a6ac7 | 758 | ret = file_fdatawait_range(file, start, end); |
2f0264d5 SW |
759 | |
760 | return ret ? ret : ret1; | |
b3b94faa DT |
761 | } |
762 | ||
967bcc91 AG |
763 | static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to) |
764 | { | |
765 | struct file *file = iocb->ki_filp; | |
766 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); | |
767 | size_t count = iov_iter_count(to); | |
768 | struct gfs2_holder gh; | |
769 | ssize_t ret; | |
770 | ||
771 | if (!count) | |
772 | return 0; /* skip atime */ | |
773 | ||
774 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); | |
775 | ret = gfs2_glock_nq(&gh); | |
776 | if (ret) | |
777 | goto out_uninit; | |
778 | ||
13ef9544 JK |
779 | ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, |
780 | is_sync_kiocb(iocb)); | |
967bcc91 | 781 | |
967bcc91 AG |
782 | gfs2_glock_dq(&gh); |
783 | out_uninit: | |
784 | gfs2_holder_uninit(&gh); | |
785 | return ret; | |
786 | } | |
787 | ||
788 | static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
789 | { | |
790 | struct file *file = iocb->ki_filp; | |
791 | struct inode *inode = file->f_mapping->host; | |
792 | struct gfs2_inode *ip = GFS2_I(inode); | |
793 | size_t len = iov_iter_count(from); | |
794 | loff_t offset = iocb->ki_pos; | |
795 | struct gfs2_holder gh; | |
796 | ssize_t ret; | |
797 | ||
798 | /* | |
799 | * Deferred lock, even if its a write, since we do no allocation on | |
800 | * this path. All we need to change is the atime, and this lock mode | |
801 | * ensures that other nodes have flushed their buffered read caches | |
802 | * (i.e. their page cache entries for this inode). We do not, | |
803 | * unfortunately, have the option of only flushing a range like the | |
804 | * VFS does. | |
805 | */ | |
806 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); | |
807 | ret = gfs2_glock_nq(&gh); | |
808 | if (ret) | |
809 | goto out_uninit; | |
810 | ||
811 | /* Silently fall back to buffered I/O when writing beyond EOF */ | |
812 | if (offset + len > i_size_read(&ip->i_inode)) | |
813 | goto out; | |
814 | ||
13ef9544 JK |
815 | ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, |
816 | is_sync_kiocb(iocb)); | |
967bcc91 AG |
817 | |
818 | out: | |
819 | gfs2_glock_dq(&gh); | |
820 | out_uninit: | |
821 | gfs2_holder_uninit(&gh); | |
822 | return ret; | |
823 | } | |
824 | ||
825 | static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
826 | { | |
827 | ssize_t ret; | |
828 | ||
829 | if (iocb->ki_flags & IOCB_DIRECT) { | |
830 | ret = gfs2_file_direct_read(iocb, to); | |
831 | if (likely(ret != -ENOTBLK)) | |
832 | return ret; | |
833 | iocb->ki_flags &= ~IOCB_DIRECT; | |
834 | } | |
835 | return generic_file_read_iter(iocb, to); | |
836 | } | |
837 | ||
56aa616a | 838 | /** |
da56e45b | 839 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a | 840 | * @iocb: The io context |
64bc06bb | 841 | * @from: The data to write |
56aa616a SW |
842 | * |
843 | * We have to do a lock/unlock here to refresh the inode size for | |
844 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
845 | * offset. There is still a race, but provided the app is using its | |
846 | * own file locking, this will make O_APPEND work as expected. | |
847 | * | |
848 | */ | |
849 | ||
da56e45b | 850 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
851 | { |
852 | struct file *file = iocb->ki_filp; | |
64bc06bb AG |
853 | struct inode *inode = file_inode(file); |
854 | struct gfs2_inode *ip = GFS2_I(inode); | |
6e5e41e2 | 855 | ssize_t ret; |
0a305e49 | 856 | |
da56e45b | 857 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 858 | |
2ba48ce5 | 859 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a | 860 | struct gfs2_holder gh; |
56aa616a SW |
861 | |
862 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
863 | if (ret) | |
4bd684bc | 864 | return ret; |
56aa616a SW |
865 | gfs2_glock_dq_uninit(&gh); |
866 | } | |
867 | ||
64bc06bb AG |
868 | inode_lock(inode); |
869 | ret = generic_write_checks(iocb, from); | |
870 | if (ret <= 0) | |
4c0e8dda | 871 | goto out_unlock; |
64bc06bb AG |
872 | |
873 | ret = file_remove_privs(file); | |
874 | if (ret) | |
4c0e8dda | 875 | goto out_unlock; |
64bc06bb AG |
876 | |
877 | ret = file_update_time(file); | |
878 | if (ret) | |
4c0e8dda | 879 | goto out_unlock; |
64bc06bb | 880 | |
967bcc91 AG |
881 | if (iocb->ki_flags & IOCB_DIRECT) { |
882 | struct address_space *mapping = file->f_mapping; | |
6e5e41e2 | 883 | ssize_t buffered, ret2; |
967bcc91 | 884 | |
6e5e41e2 AG |
885 | ret = gfs2_file_direct_write(iocb, from); |
886 | if (ret < 0 || !iov_iter_count(from)) | |
4c0e8dda | 887 | goto out_unlock; |
967bcc91 | 888 | |
6e5e41e2 | 889 | iocb->ki_flags |= IOCB_DSYNC; |
4c0e8dda | 890 | current->backing_dev_info = inode_to_bdi(inode); |
6e5e41e2 | 891 | buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); |
4c0e8dda | 892 | current->backing_dev_info = NULL; |
6e5e41e2 | 893 | if (unlikely(buffered <= 0)) |
4c0e8dda | 894 | goto out_unlock; |
967bcc91 AG |
895 | |
896 | /* | |
897 | * We need to ensure that the page cache pages are written to | |
898 | * disk and invalidated to preserve the expected O_DIRECT | |
6e5e41e2 AG |
899 | * semantics. If the writeback or invalidate fails, only report |
900 | * the direct I/O range as we don't know if the buffered pages | |
901 | * made it to disk. | |
967bcc91 | 902 | */ |
6e5e41e2 AG |
903 | iocb->ki_pos += buffered; |
904 | ret2 = generic_write_sync(iocb, buffered); | |
905 | invalidate_mapping_pages(mapping, | |
906 | (iocb->ki_pos - buffered) >> PAGE_SHIFT, | |
907 | (iocb->ki_pos - 1) >> PAGE_SHIFT); | |
908 | if (!ret || ret2 > 0) | |
909 | ret += ret2; | |
967bcc91 | 910 | } else { |
4c0e8dda | 911 | current->backing_dev_info = inode_to_bdi(inode); |
967bcc91 | 912 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); |
4c0e8dda | 913 | current->backing_dev_info = NULL; |
6e5e41e2 | 914 | if (likely(ret > 0)) { |
967bcc91 | 915 | iocb->ki_pos += ret; |
6e5e41e2 AG |
916 | ret = generic_write_sync(iocb, ret); |
917 | } | |
967bcc91 | 918 | } |
64bc06bb | 919 | |
4c0e8dda | 920 | out_unlock: |
64bc06bb | 921 | inode_unlock(inode); |
6e5e41e2 | 922 | return ret; |
56aa616a SW |
923 | } |
924 | ||
2fe17c10 CH |
925 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
926 | int mode) | |
927 | { | |
fffb6412 | 928 | struct super_block *sb = inode->i_sb; |
2fe17c10 | 929 | struct gfs2_inode *ip = GFS2_I(inode); |
fffb6412 | 930 | loff_t end = offset + len; |
2fe17c10 CH |
931 | struct buffer_head *dibh; |
932 | int error; | |
2fe17c10 CH |
933 | |
934 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
935 | if (unlikely(error)) | |
64dd153c | 936 | return error; |
2fe17c10 | 937 | |
350a9b0a | 938 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
939 | |
940 | if (gfs2_is_stuffed(ip)) { | |
941 | error = gfs2_unstuff_dinode(ip, NULL); | |
942 | if (unlikely(error)) | |
943 | goto out; | |
944 | } | |
945 | ||
fffb6412 | 946 | while (offset < end) { |
c2589282 AG |
947 | struct iomap iomap = { }; |
948 | ||
628e366d AG |
949 | error = gfs2_iomap_get_alloc(inode, offset, end - offset, |
950 | &iomap); | |
fffb6412 | 951 | if (error) |
64dd153c | 952 | goto out; |
fffb6412 | 953 | offset = iomap.offset + iomap.length; |
d505a96a | 954 | if (!(iomap.flags & IOMAP_F_NEW)) |
64dd153c | 955 | continue; |
fffb6412 AG |
956 | error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, |
957 | iomap.length >> inode->i_blkbits, | |
958 | GFP_NOFS); | |
959 | if (error) { | |
960 | fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); | |
2fe17c10 | 961 | goto out; |
64dd153c | 962 | } |
2fe17c10 | 963 | } |
2fe17c10 | 964 | out: |
64dd153c | 965 | brelse(dibh); |
2fe17c10 CH |
966 | return error; |
967 | } | |
f3b64b57 | 968 | |
d9be0cda AD |
969 | /** |
970 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
971 | * blocks, determine how many bytes can be written. | |
972 | * @ip: The inode in question. | |
973 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
974 | * @data_blocks: Compute and return the number of data blocks needed | |
975 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
976 | * @max_blocks: The total blocks available to work with. | |
977 | * | |
978 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
979 | */ | |
980 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
981 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
982 | unsigned int max_blocks) | |
2fe17c10 | 983 | { |
d9be0cda | 984 | loff_t max = *len; |
2fe17c10 | 985 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
986 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
987 | ||
988 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
989 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
990 | max_data -= tmp; | |
991 | } | |
d9be0cda | 992 | |
2fe17c10 CH |
993 | *data_blocks = max_data; |
994 | *ind_blocks = max_blocks - max_data; | |
995 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
996 | if (*len > max) { | |
997 | *len = max; | |
998 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
999 | } | |
1000 | } | |
1001 | ||
9c9f1159 | 1002 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 1003 | { |
496ad9aa | 1004 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
1005 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1006 | struct gfs2_inode *ip = GFS2_I(inode); | |
7b9cff46 | 1007 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
2fe17c10 | 1008 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
174d1232 | 1009 | loff_t bytes, max_bytes, max_blks; |
2fe17c10 | 1010 | int error; |
4442f2e0 SW |
1011 | const loff_t pos = offset; |
1012 | const loff_t count = len; | |
6905d9e4 | 1013 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 1014 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 1015 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 1016 | |
2fe17c10 CH |
1017 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
1018 | ||
6905d9e4 | 1019 | offset &= bsize_mask; |
2fe17c10 CH |
1020 | |
1021 | len = next - offset; | |
1022 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
1023 | if (!bytes) | |
1024 | bytes = UINT_MAX; | |
6905d9e4 BM |
1025 | bytes &= bsize_mask; |
1026 | if (bytes == 0) | |
1027 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 1028 | |
da1dfb6a | 1029 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 1030 | |
d9be0cda AD |
1031 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
1032 | ap.min_target = data_blocks + ind_blocks; | |
1033 | ||
2fe17c10 CH |
1034 | while (len > 0) { |
1035 | if (len < bytes) | |
1036 | bytes = len; | |
58a7d5fb BM |
1037 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
1038 | len -= bytes; | |
1039 | offset += bytes; | |
1040 | continue; | |
1041 | } | |
d9be0cda AD |
1042 | |
1043 | /* We need to determine how many bytes we can actually | |
1044 | * fallocate without exceeding quota or going over the | |
1045 | * end of the fs. We start off optimistically by assuming | |
1046 | * we can write max_bytes */ | |
1047 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
1048 | ||
1049 | /* Since max_bytes is most likely a theoretical max, we | |
1050 | * calculate a more realistic 'bytes' to serve as a good | |
1051 | * starting point for the number of bytes we may be able | |
1052 | * to write */ | |
2fe17c10 | 1053 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 1054 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
1055 | |
1056 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 1057 | if (error) |
9c9f1159 | 1058 | return error; |
d9be0cda AD |
1059 | /* ap.allowed tells us how many blocks quota will allow |
1060 | * us to write. Check if this reduces max_blks */ | |
174d1232 AG |
1061 | max_blks = UINT_MAX; |
1062 | if (ap.allowed) | |
d9be0cda | 1063 | max_blks = ap.allowed; |
2fe17c10 | 1064 | |
7b9cff46 | 1065 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 1066 | if (error) |
2fe17c10 | 1067 | goto out_qunlock; |
d9be0cda AD |
1068 | |
1069 | /* check if the selected rgrp limits our max_blks further */ | |
1070 | if (ap.allowed && ap.allowed < max_blks) | |
1071 | max_blks = ap.allowed; | |
1072 | ||
1073 | /* Almost done. Calculate bytes that can be written using | |
1074 | * max_blks. We also recompute max_bytes, data_blocks and | |
1075 | * ind_blocks */ | |
1076 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
1077 | &ind_blocks, max_blks); | |
2fe17c10 CH |
1078 | |
1079 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 1080 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
1081 | if (gfs2_is_jdata(ip)) |
1082 | rblocks += data_blocks ? data_blocks : 1; | |
1083 | ||
1084 | error = gfs2_trans_begin(sdp, rblocks, | |
45eb0504 | 1085 | PAGE_SIZE >> inode->i_blkbits); |
2fe17c10 CH |
1086 | if (error) |
1087 | goto out_trans_fail; | |
1088 | ||
1089 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
1090 | gfs2_trans_end(sdp); | |
1091 | ||
1092 | if (error) | |
1093 | goto out_trans_fail; | |
1094 | ||
1095 | len -= max_bytes; | |
1096 | offset += max_bytes; | |
1097 | gfs2_inplace_release(ip); | |
1098 | gfs2_quota_unlock(ip); | |
2fe17c10 | 1099 | } |
4442f2e0 | 1100 | |
0a6a4abc | 1101 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) |
1885867b | 1102 | i_size_write(inode, pos + count); |
0a6a4abc AG |
1103 | file_update_time(file); |
1104 | mark_inode_dirty(inode); | |
1885867b | 1105 | |
dde0c2e7 CH |
1106 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
1107 | return vfs_fsync_range(file, pos, pos + count - 1, | |
1108 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
1109 | return 0; | |
2fe17c10 CH |
1110 | |
1111 | out_trans_fail: | |
1112 | gfs2_inplace_release(ip); | |
1113 | out_qunlock: | |
1114 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
1115 | return error; |
1116 | } | |
1117 | ||
1118 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
1119 | { | |
1120 | struct inode *inode = file_inode(file); | |
d4d7fc12 | 1121 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9c9f1159 AP |
1122 | struct gfs2_inode *ip = GFS2_I(inode); |
1123 | struct gfs2_holder gh; | |
1124 | int ret; | |
1125 | ||
4e56a641 | 1126 | if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) |
d4d7fc12 AP |
1127 | return -EOPNOTSUPP; |
1128 | /* fallocate is needed by gfs2_grow to reserve space in the rindex */ | |
1129 | if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) | |
9c9f1159 AP |
1130 | return -EOPNOTSUPP; |
1131 | ||
5955102c | 1132 | inode_lock(inode); |
9c9f1159 AP |
1133 | |
1134 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1135 | ret = gfs2_glock_nq(&gh); | |
1136 | if (ret) | |
1137 | goto out_uninit; | |
1138 | ||
1139 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
1140 | (offset + len) > inode->i_size) { | |
1141 | ret = inode_newsize_ok(inode, offset + len); | |
1142 | if (ret) | |
1143 | goto out_unlock; | |
1144 | } | |
1145 | ||
1146 | ret = get_write_access(inode); | |
1147 | if (ret) | |
1148 | goto out_unlock; | |
1149 | ||
4e56a641 AG |
1150 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
1151 | ret = __gfs2_punch_hole(file, offset, len); | |
1152 | } else { | |
4e56a641 | 1153 | ret = __gfs2_fallocate(file, mode, offset, len); |
4e56a641 AG |
1154 | if (ret) |
1155 | gfs2_rs_deltree(&ip->i_res); | |
1156 | } | |
a097dc7e | 1157 | |
9c9f1159 | 1158 | put_write_access(inode); |
2fe17c10 | 1159 | out_unlock: |
a0846a53 | 1160 | gfs2_glock_dq(&gh); |
2fe17c10 | 1161 | out_uninit: |
a0846a53 | 1162 | gfs2_holder_uninit(&gh); |
5955102c | 1163 | inode_unlock(inode); |
9c9f1159 | 1164 | return ret; |
2fe17c10 CH |
1165 | } |
1166 | ||
f1ea6f4e BP |
1167 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
1168 | struct file *out, loff_t *ppos, | |
1169 | size_t len, unsigned int flags) | |
1170 | { | |
2fba46a0 | 1171 | ssize_t ret; |
f1ea6f4e | 1172 | |
f1ea6f4e BP |
1173 | gfs2_size_hint(out, *ppos, len); |
1174 | ||
2fba46a0 | 1175 | ret = iter_file_splice_write(pipe, out, ppos, len, flags); |
2fba46a0 | 1176 | return ret; |
f1ea6f4e BP |
1177 | } |
1178 | ||
f057f6cd SW |
1179 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
1180 | ||
b3b94faa DT |
1181 | /** |
1182 | * gfs2_lock - acquire/release a posix lock on a file | |
1183 | * @file: the file pointer | |
1184 | * @cmd: either modify or retrieve lock state, possibly wait | |
1185 | * @fl: type and range of lock | |
1186 | * | |
1187 | * Returns: errno | |
1188 | */ | |
1189 | ||
1190 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1191 | { | |
feaa7bba SW |
1192 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1193 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1194 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1195 | |
b3b94faa DT |
1196 | if (!(fl->fl_flags & FL_POSIX)) |
1197 | return -ENOLCK; | |
720e7749 | 1198 | if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) |
b3b94faa DT |
1199 | return -ENOLCK; |
1200 | ||
586759f0 ME |
1201 | if (cmd == F_CANCELLK) { |
1202 | /* Hack: */ | |
1203 | cmd = F_SETLK; | |
1204 | fl->fl_type = F_UNLCK; | |
1205 | } | |
eb43e660 | 1206 | if (unlikely(gfs2_withdrawn(sdp))) { |
c2952d20 | 1207 | if (fl->fl_type == F_UNLCK) |
4f656367 | 1208 | locks_lock_file_wait(file, fl); |
f057f6cd | 1209 | return -EIO; |
c2952d20 | 1210 | } |
b3b94faa | 1211 | if (IS_GETLK(cmd)) |
f057f6cd | 1212 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1213 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1214 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1215 | else |
f057f6cd | 1216 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1217 | } |
1218 | ||
b3b94faa DT |
1219 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1220 | { | |
5c676f6d | 1221 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1222 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1223 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1224 | struct gfs2_glock *gl; |
1225 | unsigned int state; | |
b58bf407 | 1226 | u16 flags; |
b3b94faa | 1227 | int error = 0; |
2ddfbdd6 | 1228 | int sleeptime; |
b3b94faa DT |
1229 | |
1230 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
2ddfbdd6 | 1231 | flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; |
b3b94faa | 1232 | |
f55ab26a | 1233 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa | 1234 | |
283c9a97 | 1235 | if (gfs2_holder_initialized(fl_gh)) { |
4d62d3f7 | 1236 | struct file_lock request; |
b3b94faa DT |
1237 | if (fl_gh->gh_state == state) |
1238 | goto out; | |
4d62d3f7 N |
1239 | locks_init_lock(&request); |
1240 | request.fl_type = F_UNLCK; | |
1241 | request.fl_flags = FL_FLOCK; | |
1242 | locks_lock_file_wait(file, &request); | |
5bef3e7c | 1243 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1244 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1245 | } else { |
6802e340 SW |
1246 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1247 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1248 | if (error) |
1249 | goto out; | |
b4c20166 AD |
1250 | gfs2_holder_init(gl, state, flags, fl_gh); |
1251 | gfs2_glock_put(gl); | |
b3b94faa | 1252 | } |
2ddfbdd6 BP |
1253 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1254 | error = gfs2_glock_nq(fl_gh); | |
1255 | if (error != GLR_TRYFAILED) | |
1256 | break; | |
1257 | fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; | |
1258 | fl_gh->gh_error = 0; | |
1259 | msleep(sleeptime); | |
1260 | } | |
b3b94faa DT |
1261 | if (error) { |
1262 | gfs2_holder_uninit(fl_gh); | |
1263 | if (error == GLR_TRYFAILED) | |
1264 | error = -EAGAIN; | |
1265 | } else { | |
4f656367 | 1266 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1267 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1268 | } |
1269 | ||
420b9e5e | 1270 | out: |
f55ab26a | 1271 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1272 | return error; |
1273 | } | |
1274 | ||
1275 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1276 | { | |
5c676f6d | 1277 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1278 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1279 | ||
f55ab26a | 1280 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1281 | locks_lock_file_wait(file, fl); |
6df9f9a2 | 1282 | if (gfs2_holder_initialized(fl_gh)) { |
2ddfbdd6 | 1283 | gfs2_glock_dq(fl_gh); |
0a33443b SW |
1284 | gfs2_holder_uninit(fl_gh); |
1285 | } | |
f55ab26a | 1286 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1287 | } |
1288 | ||
1289 | /** | |
1290 | * gfs2_flock - acquire/release a flock lock on a file | |
1291 | * @file: the file pointer | |
1292 | * @cmd: either modify or retrieve lock state, possibly wait | |
1293 | * @fl: type and range of lock | |
1294 | * | |
1295 | * Returns: errno | |
1296 | */ | |
1297 | ||
1298 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1299 | { | |
b3b94faa DT |
1300 | if (!(fl->fl_flags & FL_FLOCK)) |
1301 | return -ENOLCK; | |
a12af1eb AD |
1302 | if (fl->fl_type & LOCK_MAND) |
1303 | return -EOPNOTSUPP; | |
b3b94faa | 1304 | |
b3b94faa DT |
1305 | if (fl->fl_type == F_UNLCK) { |
1306 | do_unflock(file, fl); | |
1307 | return 0; | |
d00223f1 | 1308 | } else { |
b3b94faa | 1309 | return do_flock(file, cmd, fl); |
d00223f1 | 1310 | } |
b3b94faa DT |
1311 | } |
1312 | ||
10d21988 | 1313 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1314 | .llseek = gfs2_llseek, |
967bcc91 | 1315 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1316 | .write_iter = gfs2_file_write_iter, |
81214bab | 1317 | .iopoll = iomap_dio_iopoll, |
26c1a574 | 1318 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1319 | .compat_ioctl = gfs2_compat_ioctl, |
26c1a574 SW |
1320 | .mmap = gfs2_mmap, |
1321 | .open = gfs2_open, | |
df3fd117 | 1322 | .release = gfs2_release, |
26c1a574 SW |
1323 | .fsync = gfs2_fsync, |
1324 | .lock = gfs2_lock, | |
26c1a574 | 1325 | .flock = gfs2_flock, |
82c156f8 | 1326 | .splice_read = generic_file_splice_read, |
f42a69fa | 1327 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1328 | .setlease = simple_nosetlease, |
2fe17c10 | 1329 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1330 | }; |
1331 | ||
10d21988 | 1332 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1333 | .iterate_shared = gfs2_readdir, |
26c1a574 | 1334 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1335 | .compat_ioctl = gfs2_compat_ioctl, |
26c1a574 | 1336 | .open = gfs2_open, |
df3fd117 | 1337 | .release = gfs2_release, |
26c1a574 SW |
1338 | .fsync = gfs2_fsync, |
1339 | .lock = gfs2_lock, | |
1340 | .flock = gfs2_flock, | |
6038f373 | 1341 | .llseek = default_llseek, |
b3b94faa DT |
1342 | }; |
1343 | ||
f057f6cd SW |
1344 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1345 | ||
10d21988 | 1346 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1347 | .llseek = gfs2_llseek, |
967bcc91 | 1348 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1349 | .write_iter = gfs2_file_write_iter, |
81214bab | 1350 | .iopoll = iomap_dio_iopoll, |
c97bfe43 | 1351 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1352 | .compat_ioctl = gfs2_compat_ioctl, |
c97bfe43 WC |
1353 | .mmap = gfs2_mmap, |
1354 | .open = gfs2_open, | |
df3fd117 | 1355 | .release = gfs2_release, |
c97bfe43 | 1356 | .fsync = gfs2_fsync, |
82c156f8 | 1357 | .splice_read = generic_file_splice_read, |
f42a69fa | 1358 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1359 | .setlease = generic_setlease, |
2fe17c10 | 1360 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1361 | }; |
1362 | ||
10d21988 | 1363 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1364 | .iterate_shared = gfs2_readdir, |
c97bfe43 | 1365 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1366 | .compat_ioctl = gfs2_compat_ioctl, |
c97bfe43 | 1367 | .open = gfs2_open, |
df3fd117 | 1368 | .release = gfs2_release, |
c97bfe43 | 1369 | .fsync = gfs2_fsync, |
6038f373 | 1370 | .llseek = default_llseek, |
c97bfe43 WC |
1371 | }; |
1372 |