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> | |
9 | #include <linux/completion.h> | |
10 | #include <linux/buffer_head.h> | |
11 | #include <linux/pagemap.h> | |
12 | #include <linux/uio.h> | |
13 | #include <linux/blkdev.h> | |
14 | #include <linux/mm.h> | |
f58ba889 | 15 | #include <linux/mount.h> |
18ec7d5c | 16 | #include <linux/fs.h> |
5c676f6d | 17 | #include <linux/gfs2_ondisk.h> |
2fe17c10 CH |
18 | #include <linux/falloc.h> |
19 | #include <linux/swap.h> | |
71b86f56 | 20 | #include <linux/crc32.h> |
33c3de32 | 21 | #include <linux/writeback.h> |
7c0f6ba6 | 22 | #include <linux/uaccess.h> |
f057f6cd SW |
23 | #include <linux/dlm.h> |
24 | #include <linux/dlm_plock.h> | |
2ddfbdd6 | 25 | #include <linux/delay.h> |
64bc06bb | 26 | #include <linux/backing-dev.h> |
b3b94faa DT |
27 | |
28 | #include "gfs2.h" | |
5c676f6d | 29 | #include "incore.h" |
b3b94faa | 30 | #include "bmap.h" |
64bc06bb | 31 | #include "aops.h" |
b3b94faa DT |
32 | #include "dir.h" |
33 | #include "glock.h" | |
34 | #include "glops.h" | |
35 | #include "inode.h" | |
b3b94faa DT |
36 | #include "log.h" |
37 | #include "meta_io.h" | |
b3b94faa DT |
38 | #include "quota.h" |
39 | #include "rgrp.h" | |
40 | #include "trans.h" | |
5c676f6d | 41 | #include "util.h" |
b3b94faa | 42 | |
b3b94faa DT |
43 | /** |
44 | * gfs2_llseek - seek to a location in a file | |
45 | * @file: the file | |
46 | * @offset: the offset | |
965c8e59 | 47 | * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) |
b3b94faa DT |
48 | * |
49 | * SEEK_END requires the glock for the file because it references the | |
50 | * file's size. | |
51 | * | |
52 | * Returns: The new offset, or errno | |
53 | */ | |
54 | ||
965c8e59 | 55 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence) |
b3b94faa | 56 | { |
feaa7bba | 57 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa DT |
58 | struct gfs2_holder i_gh; |
59 | loff_t error; | |
60 | ||
965c8e59 | 61 | switch (whence) { |
3a27411c | 62 | case SEEK_END: |
b3b94faa DT |
63 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
64 | &i_gh); | |
65 | if (!error) { | |
965c8e59 | 66 | error = generic_file_llseek(file, offset, whence); |
b3b94faa DT |
67 | gfs2_glock_dq_uninit(&i_gh); |
68 | } | |
9453615a | 69 | break; |
3a27411c AG |
70 | |
71 | case SEEK_DATA: | |
72 | error = gfs2_seek_data(file, offset); | |
73 | break; | |
74 | ||
75 | case SEEK_HOLE: | |
76 | error = gfs2_seek_hole(file, offset); | |
77 | break; | |
78 | ||
9453615a SW |
79 | case SEEK_CUR: |
80 | case SEEK_SET: | |
3a27411c AG |
81 | /* |
82 | * These don't reference inode->i_size and don't depend on the | |
83 | * block mapping, so we don't need the glock. | |
84 | */ | |
965c8e59 | 85 | error = generic_file_llseek(file, offset, whence); |
9453615a SW |
86 | break; |
87 | default: | |
88 | error = -EINVAL; | |
89 | } | |
b3b94faa DT |
90 | |
91 | return error; | |
92 | } | |
93 | ||
b3b94faa | 94 | /** |
d81a8ef5 | 95 | * gfs2_readdir - Iterator for a directory |
b3b94faa | 96 | * @file: The directory to read from |
d81a8ef5 | 97 | * @ctx: What to feed directory entries to |
b3b94faa DT |
98 | * |
99 | * Returns: errno | |
100 | */ | |
101 | ||
d81a8ef5 | 102 | static int gfs2_readdir(struct file *file, struct dir_context *ctx) |
b3b94faa | 103 | { |
71b86f56 | 104 | struct inode *dir = file->f_mapping->host; |
feaa7bba | 105 | struct gfs2_inode *dip = GFS2_I(dir); |
b3b94faa | 106 | struct gfs2_holder d_gh; |
b3b94faa DT |
107 | int error; |
108 | ||
d81a8ef5 AV |
109 | error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); |
110 | if (error) | |
b3b94faa | 111 | return error; |
b3b94faa | 112 | |
d81a8ef5 | 113 | error = gfs2_dir_read(dir, ctx, &file->f_ra); |
b3b94faa DT |
114 | |
115 | gfs2_glock_dq_uninit(&d_gh); | |
116 | ||
b3b94faa DT |
117 | return error; |
118 | } | |
119 | ||
128e5eba | 120 | /** |
b16f7e57 | 121 | * fsflag_gfs2flag |
128e5eba | 122 | * |
b16f7e57 AG |
123 | * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories, |
124 | * and to GFS2_DIF_JDATA for non-directories. | |
128e5eba | 125 | */ |
b16f7e57 AG |
126 | static struct { |
127 | u32 fsflag; | |
128 | u32 gfsflag; | |
129 | } fsflag_gfs2flag[] = { | |
130 | {FS_SYNC_FL, GFS2_DIF_SYNC}, | |
131 | {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE}, | |
132 | {FS_APPEND_FL, GFS2_DIF_APPENDONLY}, | |
133 | {FS_NOATIME_FL, GFS2_DIF_NOATIME}, | |
134 | {FS_INDEX_FL, GFS2_DIF_EXHASH}, | |
135 | {FS_TOPDIR_FL, GFS2_DIF_TOPDIR}, | |
136 | {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA}, | |
7ea9ea83 | 137 | }; |
71b86f56 | 138 | |
5aca2842 DW |
139 | static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags) |
140 | { | |
141 | int i; | |
142 | u32 fsflags = 0; | |
143 | ||
144 | if (S_ISDIR(inode->i_mode)) | |
145 | gfsflags &= ~GFS2_DIF_JDATA; | |
146 | else | |
147 | gfsflags &= ~GFS2_DIF_INHERIT_JDATA; | |
148 | ||
149 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) | |
150 | if (gfsflags & fsflag_gfs2flag[i].gfsflag) | |
151 | fsflags |= fsflag_gfs2flag[i].fsflag; | |
152 | return fsflags; | |
153 | } | |
154 | ||
b09e593d | 155 | static int gfs2_get_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 156 | { |
496ad9aa | 157 | struct inode *inode = file_inode(filp); |
feaa7bba | 158 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 | 159 | struct gfs2_holder gh; |
5aca2842 DW |
160 | int error; |
161 | u32 fsflags; | |
71b86f56 | 162 | |
719ee344 SW |
163 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
164 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 165 | if (error) |
9c7fe835 | 166 | goto out_uninit; |
907b9bce | 167 | |
5aca2842 | 168 | fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); |
b16f7e57 | 169 | |
128e5eba | 170 | if (put_user(fsflags, ptr)) |
71b86f56 SW |
171 | error = -EFAULT; |
172 | ||
3cc3f710 | 173 | gfs2_glock_dq(&gh); |
9c7fe835 | 174 | out_uninit: |
71b86f56 SW |
175 | gfs2_holder_uninit(&gh); |
176 | return error; | |
177 | } | |
178 | ||
6b124d8d SW |
179 | void gfs2_set_inode_flags(struct inode *inode) |
180 | { | |
181 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
182 | unsigned int flags = inode->i_flags; |
183 | ||
9964afbb SW |
184 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
185 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 186 | flags |= S_NOSEC; |
383f01fb | 187 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 188 | flags |= S_IMMUTABLE; |
383f01fb | 189 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 190 | flags |= S_APPEND; |
383f01fb | 191 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 192 | flags |= S_NOATIME; |
383f01fb | 193 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
194 | flags |= S_SYNC; |
195 | inode->i_flags = flags; | |
196 | } | |
197 | ||
71b86f56 SW |
198 | /* Flags that can be set by user space */ |
199 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
200 | GFS2_DIF_IMMUTABLE| \ |
201 | GFS2_DIF_APPENDONLY| \ | |
202 | GFS2_DIF_NOATIME| \ | |
203 | GFS2_DIF_SYNC| \ | |
23d0bb83 | 204 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
205 | GFS2_DIF_INHERIT_JDATA) |
206 | ||
207 | /** | |
9dd868e1 FF |
208 | * do_gfs2_set_flags - set flags on an inode |
209 | * @filp: file pointer | |
210 | * @reqflags: The flags to set | |
71b86f56 | 211 | * @mask: Indicates which flags are valid |
5aca2842 | 212 | * @fsflags: The FS_* inode flags passed in |
71b86f56 SW |
213 | * |
214 | */ | |
5aca2842 DW |
215 | static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask, |
216 | const u32 fsflags) | |
71b86f56 | 217 | { |
496ad9aa | 218 | struct inode *inode = file_inode(filp); |
feaa7bba SW |
219 | struct gfs2_inode *ip = GFS2_I(inode); |
220 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
221 | struct buffer_head *bh; |
222 | struct gfs2_holder gh; | |
223 | int error; | |
5aca2842 | 224 | u32 new_flags, flags, oldflags; |
71b86f56 | 225 | |
a561be71 | 226 | error = mnt_want_write_file(filp); |
52f341cf | 227 | if (error) |
71b86f56 SW |
228 | return error; |
229 | ||
f58ba889 MS |
230 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
231 | if (error) | |
232 | goto out_drop_write; | |
233 | ||
5aca2842 DW |
234 | oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); |
235 | error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags); | |
236 | if (error) | |
237 | goto out; | |
238 | ||
7df0e039 | 239 | error = -EACCES; |
2e149670 | 240 | if (!inode_owner_or_capable(inode)) |
7df0e039 SW |
241 | goto out; |
242 | ||
243 | error = 0; | |
383f01fb | 244 | flags = ip->i_diskflags; |
55eccc6d | 245 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
246 | if ((new_flags ^ flags) == 0) |
247 | goto out; | |
248 | ||
71b86f56 SW |
249 | error = -EPERM; |
250 | if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) | |
251 | goto out; | |
252 | if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) | |
253 | goto out; | |
907b9bce | 254 | if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && |
b9cb9813 | 255 | !capable(CAP_LINUX_IMMUTABLE)) |
71b86f56 | 256 | goto out; |
b9cb9813 | 257 | if (!IS_IMMUTABLE(inode)) { |
10556cb2 | 258 | error = gfs2_permission(inode, MAY_WRITE); |
b9cb9813 SW |
259 | if (error) |
260 | goto out; | |
261 | } | |
5561093e | 262 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
cc555b09 | 263 | if (new_flags & GFS2_DIF_JDATA) |
c1696fb8 | 264 | gfs2_log_flush(sdp, ip->i_gl, |
805c0907 BP |
265 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
266 | GFS2_LFC_SET_FLAGS); | |
5561093e SW |
267 | error = filemap_fdatawrite(inode->i_mapping); |
268 | if (error) | |
269 | goto out; | |
270 | error = filemap_fdatawait(inode->i_mapping); | |
271 | if (error) | |
272 | goto out; | |
cc555b09 BP |
273 | if (new_flags & GFS2_DIF_JDATA) |
274 | gfs2_ordered_del_inode(ip); | |
5561093e | 275 | } |
55eccc6d | 276 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
277 | if (error) |
278 | goto out; | |
55eccc6d SW |
279 | error = gfs2_meta_inode_buffer(ip, &bh); |
280 | if (error) | |
281 | goto out_trans_end; | |
9b7c2ddb | 282 | inode->i_ctime = current_time(inode); |
350a9b0a | 283 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 284 | ip->i_diskflags = new_flags; |
539e5d6b | 285 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 286 | brelse(bh); |
6b124d8d | 287 | gfs2_set_inode_flags(inode); |
5561093e | 288 | gfs2_set_aops(inode); |
55eccc6d SW |
289 | out_trans_end: |
290 | gfs2_trans_end(sdp); | |
71b86f56 SW |
291 | out: |
292 | gfs2_glock_dq_uninit(&gh); | |
f58ba889 | 293 | out_drop_write: |
2a79f17e | 294 | mnt_drop_write_file(filp); |
71b86f56 SW |
295 | return error; |
296 | } | |
297 | ||
b09e593d | 298 | static int gfs2_set_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 299 | { |
496ad9aa | 300 | struct inode *inode = file_inode(filp); |
b16f7e57 AG |
301 | u32 fsflags, gfsflags = 0; |
302 | u32 mask; | |
303 | int i; | |
7df0e039 | 304 | |
128e5eba | 305 | if (get_user(fsflags, ptr)) |
71b86f56 | 306 | return -EFAULT; |
7df0e039 | 307 | |
b16f7e57 AG |
308 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { |
309 | if (fsflags & fsflag_gfs2flag[i].fsflag) { | |
310 | fsflags &= ~fsflag_gfs2flag[i].fsflag; | |
311 | gfsflags |= fsflag_gfs2flag[i].gfsflag; | |
312 | } | |
313 | } | |
314 | if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) | |
315 | return -EINVAL; | |
316 | ||
317 | mask = GFS2_FLAGS_USER_SET; | |
318 | if (S_ISDIR(inode->i_mode)) { | |
319 | mask &= ~GFS2_DIF_JDATA; | |
320 | } else { | |
321 | /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ | |
322 | if (gfsflags & GFS2_DIF_TOPDIR) | |
323 | return -EINVAL; | |
324 | mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); | |
b9af7ca6 | 325 | } |
b16f7e57 | 326 | |
5aca2842 | 327 | return do_gfs2_set_flags(filp, gfsflags, mask, fsflags); |
71b86f56 SW |
328 | } |
329 | ||
6ddc5c3d SW |
330 | static int gfs2_getlabel(struct file *filp, char __user *label) |
331 | { | |
332 | struct inode *inode = file_inode(filp); | |
333 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
334 | ||
335 | if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN)) | |
336 | return -EFAULT; | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
b09e593d | 341 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
342 | { |
343 | switch(cmd) { | |
128e5eba | 344 | case FS_IOC_GETFLAGS: |
b09e593d | 345 | return gfs2_get_flags(filp, (u32 __user *)arg); |
128e5eba | 346 | case FS_IOC_SETFLAGS: |
b09e593d | 347 | return gfs2_set_flags(filp, (u32 __user *)arg); |
66fc061b SW |
348 | case FITRIM: |
349 | return gfs2_fitrim(filp, (void __user *)arg); | |
6ddc5c3d SW |
350 | case FS_IOC_GETFSLABEL: |
351 | return gfs2_getlabel(filp, (char __user *)arg); | |
71b86f56 | 352 | } |
6ddc5c3d | 353 | |
71b86f56 SW |
354 | return -ENOTTY; |
355 | } | |
356 | ||
da1dfb6a SW |
357 | /** |
358 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 359 | * @filep: The struct file |
da1dfb6a SW |
360 | * @offset: The file offset of the write |
361 | * @size: The length of the write | |
362 | * | |
363 | * When we are about to do a write, this function records the total | |
364 | * write size in order to provide a suitable hint to the lower layers | |
365 | * about how many blocks will be required. | |
366 | * | |
367 | */ | |
368 | ||
369 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
370 | { | |
496ad9aa | 371 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
372 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
373 | struct gfs2_inode *ip = GFS2_I(inode); | |
374 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
375 | int hint = min_t(size_t, INT_MAX, blks); | |
376 | ||
21f09c43 AG |
377 | if (hint > atomic_read(&ip->i_sizehint)) |
378 | atomic_set(&ip->i_sizehint, hint); | |
da1dfb6a SW |
379 | } |
380 | ||
3cc3f710 | 381 | /** |
35af80ae | 382 | * gfs2_allocate_page_backing - Allocate blocks for a write fault |
3cc3f710 SW |
383 | * @page: The (locked) page to allocate backing for |
384 | * | |
35af80ae CH |
385 | * We try to allocate all the blocks required for the page in one go. This |
386 | * might fail for various reasons, so we keep trying until all the blocks to | |
387 | * back this page are allocated. If some of the blocks are already allocated, | |
388 | * that is ok too. | |
3cc3f710 | 389 | */ |
3cc3f710 SW |
390 | static int gfs2_allocate_page_backing(struct page *page) |
391 | { | |
35af80ae CH |
392 | u64 pos = page_offset(page); |
393 | u64 size = PAGE_SIZE; | |
3cc3f710 SW |
394 | |
395 | do { | |
35af80ae CH |
396 | struct iomap iomap = { }; |
397 | ||
398 | if (gfs2_iomap_get_alloc(page->mapping->host, pos, 1, &iomap)) | |
3cc3f710 | 399 | return -EIO; |
35af80ae CH |
400 | |
401 | iomap.length = min(iomap.length, size); | |
402 | size -= iomap.length; | |
403 | pos += iomap.length; | |
404 | } while (size > 0); | |
405 | ||
3cc3f710 SW |
406 | return 0; |
407 | } | |
408 | ||
409 | /** | |
410 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
411 | * @vma: The virtual memory area | |
9dd868e1 | 412 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
413 | * |
414 | * When the page becomes writable, we need to ensure that we have | |
415 | * blocks allocated on disk to back that page. | |
416 | */ | |
417 | ||
109dbb1e | 418 | static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) |
3cc3f710 | 419 | { |
c2ec175c | 420 | struct page *page = vmf->page; |
11bac800 | 421 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3cc3f710 SW |
422 | struct gfs2_inode *ip = GFS2_I(inode); |
423 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 424 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
3cc3f710 | 425 | unsigned long last_index; |
f3915f83 | 426 | u64 pos = page_offset(page); |
3cc3f710 | 427 | unsigned int data_blocks, ind_blocks, rblocks; |
3cc3f710 | 428 | struct gfs2_holder gh; |
13d921e3 | 429 | loff_t size; |
3cc3f710 SW |
430 | int ret; |
431 | ||
39263d5e | 432 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 433 | |
b54e9a0b | 434 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 | 435 | if (ret) |
a097dc7e | 436 | goto out; |
0a305e49 | 437 | |
11bac800 | 438 | gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE); |
8e2e0047 | 439 | |
719ee344 SW |
440 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
441 | ret = gfs2_glock_nq(&gh); | |
3cc3f710 | 442 | if (ret) |
2b3dcf35 | 443 | goto out_uninit; |
3cc3f710 | 444 | |
d7c436cd | 445 | /* Update file times before taking page lock */ |
11bac800 | 446 | file_update_time(vmf->vma->vm_file); |
d7c436cd | 447 | |
9c538837 SW |
448 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
449 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
450 | ||
09cbfeaf | 451 | if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { |
13d921e3 SW |
452 | lock_page(page); |
453 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
454 | ret = -EAGAIN; | |
455 | unlock_page(page); | |
456 | } | |
3cc3f710 | 457 | goto out_unlock; |
13d921e3 SW |
458 | } |
459 | ||
5407e242 BP |
460 | ret = gfs2_rindex_update(sdp); |
461 | if (ret) | |
6dbd8224 SW |
462 | goto out_unlock; |
463 | ||
09cbfeaf | 464 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
7b9cff46 | 465 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
466 | ret = gfs2_quota_lock_check(ip, &ap); |
467 | if (ret) | |
468 | goto out_unlock; | |
7b9cff46 | 469 | ret = gfs2_inplace_reserve(ip, &ap); |
3cc3f710 SW |
470 | if (ret) |
471 | goto out_quota_unlock; | |
472 | ||
473 | rblocks = RES_DINODE + ind_blocks; | |
474 | if (gfs2_is_jdata(ip)) | |
475 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 476 | if (ind_blocks || data_blocks) { |
3cc3f710 | 477 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 478 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 479 | } |
3cc3f710 SW |
480 | ret = gfs2_trans_begin(sdp, rblocks, 0); |
481 | if (ret) | |
482 | goto out_trans_fail; | |
483 | ||
484 | lock_page(page); | |
485 | ret = -EINVAL; | |
13d921e3 | 486 | size = i_size_read(inode); |
09cbfeaf | 487 | last_index = (size - 1) >> PAGE_SHIFT; |
13d921e3 SW |
488 | /* Check page index against inode size */ |
489 | if (size == 0 || (page->index > last_index)) | |
490 | goto out_trans_end; | |
491 | ||
492 | ret = -EAGAIN; | |
493 | /* If truncated, we must retry the operation, we may have raced | |
494 | * with the glock demotion code. | |
495 | */ | |
496 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) | |
497 | goto out_trans_end; | |
498 | ||
499 | /* Unstuff, if required, and allocate backing blocks for page */ | |
b7fe2e39 | 500 | ret = 0; |
13d921e3 | 501 | if (gfs2_is_stuffed(ip)) |
3cc3f710 | 502 | ret = gfs2_unstuff_dinode(ip, page); |
13d921e3 SW |
503 | if (ret == 0) |
504 | ret = gfs2_allocate_page_backing(page); | |
3cc3f710 | 505 | |
13d921e3 SW |
506 | out_trans_end: |
507 | if (ret) | |
508 | unlock_page(page); | |
3cc3f710 SW |
509 | gfs2_trans_end(sdp); |
510 | out_trans_fail: | |
511 | gfs2_inplace_release(ip); | |
512 | out_quota_unlock: | |
513 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
514 | out_unlock: |
515 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 516 | out_uninit: |
3cc3f710 | 517 | gfs2_holder_uninit(&gh); |
13d921e3 SW |
518 | if (ret == 0) { |
519 | set_page_dirty(page); | |
1d1d1a76 | 520 | wait_for_stable_page(page); |
13d921e3 | 521 | } |
2b3dcf35 | 522 | out: |
39263d5e | 523 | sb_end_pagefault(inode->i_sb); |
13d921e3 | 524 | return block_page_mkwrite_return(ret); |
3cc3f710 SW |
525 | } |
526 | ||
f0f37e2f | 527 | static const struct vm_operations_struct gfs2_vm_ops = { |
3cc3f710 | 528 | .fault = filemap_fault, |
f1820361 | 529 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
530 | .page_mkwrite = gfs2_page_mkwrite, |
531 | }; | |
532 | ||
b3b94faa DT |
533 | /** |
534 | * gfs2_mmap - | |
535 | * @file: The file to map | |
536 | * @vma: The VMA which described the mapping | |
537 | * | |
48bf2b17 SW |
538 | * There is no need to get a lock here unless we should be updating |
539 | * atime. We ignore any locking errors since the only consequence is | |
540 | * a missed atime update (which will just be deferred until later). | |
541 | * | |
542 | * Returns: 0 | |
b3b94faa DT |
543 | */ |
544 | ||
545 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
546 | { | |
feaa7bba | 547 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 548 | |
b9c93bb7 SW |
549 | if (!(file->f_flags & O_NOATIME) && |
550 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
551 | struct gfs2_holder i_gh; |
552 | int error; | |
b3b94faa | 553 | |
3d162688 BM |
554 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
555 | &i_gh); | |
b9c93bb7 SW |
556 | if (error) |
557 | return error; | |
3d162688 BM |
558 | /* grab lock to update inode */ |
559 | gfs2_glock_dq_uninit(&i_gh); | |
560 | file_accessed(file); | |
48bf2b17 | 561 | } |
3cc3f710 | 562 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 563 | |
48bf2b17 | 564 | return 0; |
b3b94faa DT |
565 | } |
566 | ||
567 | /** | |
6d4ade98 SW |
568 | * gfs2_open_common - This is common to open and atomic_open |
569 | * @inode: The inode being opened | |
570 | * @file: The file being opened | |
b3b94faa | 571 | * |
6d4ade98 SW |
572 | * This maybe called under a glock or not depending upon how it has |
573 | * been called. We must always be called under a glock for regular | |
574 | * files, however. For other file types, it does not matter whether | |
575 | * we hold the glock or not. | |
576 | * | |
577 | * Returns: Error code or 0 for success | |
b3b94faa DT |
578 | */ |
579 | ||
6d4ade98 | 580 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 581 | { |
b3b94faa | 582 | struct gfs2_file *fp; |
6d4ade98 SW |
583 | int ret; |
584 | ||
585 | if (S_ISREG(inode->i_mode)) { | |
586 | ret = generic_file_open(inode, file); | |
587 | if (ret) | |
588 | return ret; | |
589 | } | |
b3b94faa | 590 | |
6d4ade98 | 591 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
592 | if (!fp) |
593 | return -ENOMEM; | |
594 | ||
f55ab26a | 595 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 596 | |
feaa7bba | 597 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 598 | file->private_data = fp; |
6d4ade98 SW |
599 | return 0; |
600 | } | |
601 | ||
602 | /** | |
603 | * gfs2_open - open a file | |
604 | * @inode: the inode to open | |
605 | * @file: the struct file for this opening | |
606 | * | |
607 | * After atomic_open, this function is only used for opening files | |
608 | * which are already cached. We must still get the glock for regular | |
609 | * files to ensure that we have the file size uptodate for the large | |
610 | * file check which is in the common code. That is only an issue for | |
611 | * regular files though. | |
612 | * | |
613 | * Returns: errno | |
614 | */ | |
615 | ||
616 | static int gfs2_open(struct inode *inode, struct file *file) | |
617 | { | |
618 | struct gfs2_inode *ip = GFS2_I(inode); | |
619 | struct gfs2_holder i_gh; | |
620 | int error; | |
621 | bool need_unlock = false; | |
b3b94faa | 622 | |
b60623c2 | 623 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
624 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
625 | &i_gh); | |
626 | if (error) | |
6d4ade98 SW |
627 | return error; |
628 | need_unlock = true; | |
629 | } | |
b3b94faa | 630 | |
6d4ade98 | 631 | error = gfs2_open_common(inode, file); |
b3b94faa | 632 | |
6d4ade98 | 633 | if (need_unlock) |
b3b94faa | 634 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 635 | |
b3b94faa DT |
636 | return error; |
637 | } | |
638 | ||
639 | /** | |
df3fd117 | 640 | * gfs2_release - called to close a struct file |
b3b94faa DT |
641 | * @inode: the inode the struct file belongs to |
642 | * @file: the struct file being closed | |
643 | * | |
644 | * Returns: errno | |
645 | */ | |
646 | ||
df3fd117 | 647 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 648 | { |
0a305e49 | 649 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 650 | |
8e2e0047 | 651 | kfree(file->private_data); |
5c676f6d | 652 | file->private_data = NULL; |
b3b94faa | 653 | |
2b3dcf35 BP |
654 | if (!(file->f_mode & FMODE_WRITE)) |
655 | return 0; | |
0a305e49 | 656 | |
b54e9a0b | 657 | gfs2_rsqa_delete(ip, &inode->i_writecount); |
b3b94faa DT |
658 | return 0; |
659 | } | |
660 | ||
661 | /** | |
662 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
663 | * @file: the file that points to the dentry |
664 | * @start: the start position in the file to sync | |
665 | * @end: the end position in the file to sync | |
dba898b0 | 666 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 667 | * |
2f0264d5 SW |
668 | * We split the data flushing here so that we don't wait for the data |
669 | * until after we've also sent the metadata to disk. Note that for | |
670 | * data=ordered, we will write & wait for the data at the log flush | |
671 | * stage anyway, so this is unlikely to make much of a difference | |
672 | * except in the data=writeback case. | |
673 | * | |
674 | * If the fdatawrite fails due to any reason except -EIO, we will | |
675 | * continue the remainder of the fsync, although we'll still report | |
676 | * the error at the end. This is to match filemap_write_and_wait_range() | |
677 | * behaviour. | |
34126f9f | 678 | * |
b3b94faa DT |
679 | * Returns: errno |
680 | */ | |
681 | ||
02c24a82 JB |
682 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
683 | int datasync) | |
b3b94faa | 684 | { |
2f0264d5 SW |
685 | struct address_space *mapping = file->f_mapping; |
686 | struct inode *inode = mapping->host; | |
0ae45f63 | 687 | int sync_state = inode->i_state & I_DIRTY_ALL; |
dba898b0 | 688 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 689 | int ret = 0, ret1 = 0; |
b3b94faa | 690 | |
2f0264d5 SW |
691 | if (mapping->nrpages) { |
692 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
693 | if (ret1 == -EIO) | |
694 | return ret1; | |
695 | } | |
02c24a82 | 696 | |
0c901809 BM |
697 | if (!gfs2_is_jdata(ip)) |
698 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 699 | if (datasync) |
0ae45f63 | 700 | sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); |
b3b94faa | 701 | |
dba898b0 SW |
702 | if (sync_state) { |
703 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 704 | if (ret) |
dba898b0 | 705 | return ret; |
f1818529 | 706 | if (gfs2_is_jdata(ip)) |
d07a6ac7 JL |
707 | ret = file_write_and_wait(file); |
708 | if (ret) | |
709 | return ret; | |
b5b24d7a | 710 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
711 | } |
712 | ||
2f0264d5 | 713 | if (mapping->nrpages) |
d07a6ac7 | 714 | ret = file_fdatawait_range(file, start, end); |
2f0264d5 SW |
715 | |
716 | return ret ? ret : ret1; | |
b3b94faa DT |
717 | } |
718 | ||
967bcc91 AG |
719 | static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to) |
720 | { | |
721 | struct file *file = iocb->ki_filp; | |
722 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); | |
723 | size_t count = iov_iter_count(to); | |
724 | struct gfs2_holder gh; | |
725 | ssize_t ret; | |
726 | ||
727 | if (!count) | |
728 | return 0; /* skip atime */ | |
729 | ||
730 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); | |
731 | ret = gfs2_glock_nq(&gh); | |
732 | if (ret) | |
733 | goto out_uninit; | |
734 | ||
967bcc91 AG |
735 | ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL); |
736 | ||
967bcc91 AG |
737 | gfs2_glock_dq(&gh); |
738 | out_uninit: | |
739 | gfs2_holder_uninit(&gh); | |
740 | return ret; | |
741 | } | |
742 | ||
743 | static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
744 | { | |
745 | struct file *file = iocb->ki_filp; | |
746 | struct inode *inode = file->f_mapping->host; | |
747 | struct gfs2_inode *ip = GFS2_I(inode); | |
748 | size_t len = iov_iter_count(from); | |
749 | loff_t offset = iocb->ki_pos; | |
750 | struct gfs2_holder gh; | |
751 | ssize_t ret; | |
752 | ||
753 | /* | |
754 | * Deferred lock, even if its a write, since we do no allocation on | |
755 | * this path. All we need to change is the atime, and this lock mode | |
756 | * ensures that other nodes have flushed their buffered read caches | |
757 | * (i.e. their page cache entries for this inode). We do not, | |
758 | * unfortunately, have the option of only flushing a range like the | |
759 | * VFS does. | |
760 | */ | |
761 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); | |
762 | ret = gfs2_glock_nq(&gh); | |
763 | if (ret) | |
764 | goto out_uninit; | |
765 | ||
766 | /* Silently fall back to buffered I/O when writing beyond EOF */ | |
767 | if (offset + len > i_size_read(&ip->i_inode)) | |
768 | goto out; | |
769 | ||
770 | ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL); | |
771 | ||
772 | out: | |
773 | gfs2_glock_dq(&gh); | |
774 | out_uninit: | |
775 | gfs2_holder_uninit(&gh); | |
776 | return ret; | |
777 | } | |
778 | ||
779 | static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
780 | { | |
781 | ssize_t ret; | |
782 | ||
783 | if (iocb->ki_flags & IOCB_DIRECT) { | |
784 | ret = gfs2_file_direct_read(iocb, to); | |
785 | if (likely(ret != -ENOTBLK)) | |
786 | return ret; | |
787 | iocb->ki_flags &= ~IOCB_DIRECT; | |
788 | } | |
789 | return generic_file_read_iter(iocb, to); | |
790 | } | |
791 | ||
56aa616a | 792 | /** |
da56e45b | 793 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a | 794 | * @iocb: The io context |
64bc06bb | 795 | * @from: The data to write |
56aa616a SW |
796 | * |
797 | * We have to do a lock/unlock here to refresh the inode size for | |
798 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
799 | * offset. There is still a race, but provided the app is using its | |
800 | * own file locking, this will make O_APPEND work as expected. | |
801 | * | |
802 | */ | |
803 | ||
da56e45b | 804 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
805 | { |
806 | struct file *file = iocb->ki_filp; | |
64bc06bb AG |
807 | struct inode *inode = file_inode(file); |
808 | struct gfs2_inode *ip = GFS2_I(inode); | |
967bcc91 | 809 | ssize_t written = 0, ret; |
0a305e49 | 810 | |
b54e9a0b | 811 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 BP |
812 | if (ret) |
813 | return ret; | |
56aa616a | 814 | |
da56e45b | 815 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 816 | |
2ba48ce5 | 817 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a | 818 | struct gfs2_holder gh; |
56aa616a SW |
819 | |
820 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
821 | if (ret) | |
822 | return ret; | |
823 | gfs2_glock_dq_uninit(&gh); | |
824 | } | |
825 | ||
64bc06bb AG |
826 | inode_lock(inode); |
827 | ret = generic_write_checks(iocb, from); | |
828 | if (ret <= 0) | |
829 | goto out; | |
830 | ||
831 | /* We can write back this queue in page reclaim */ | |
832 | current->backing_dev_info = inode_to_bdi(inode); | |
833 | ||
834 | ret = file_remove_privs(file); | |
835 | if (ret) | |
836 | goto out2; | |
837 | ||
838 | ret = file_update_time(file); | |
839 | if (ret) | |
840 | goto out2; | |
841 | ||
967bcc91 AG |
842 | if (iocb->ki_flags & IOCB_DIRECT) { |
843 | struct address_space *mapping = file->f_mapping; | |
844 | loff_t pos, endbyte; | |
845 | ssize_t buffered; | |
846 | ||
847 | written = gfs2_file_direct_write(iocb, from); | |
848 | if (written < 0 || !iov_iter_count(from)) | |
849 | goto out2; | |
850 | ||
851 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); | |
852 | if (unlikely(ret < 0)) | |
853 | goto out2; | |
854 | buffered = ret; | |
855 | ||
856 | /* | |
857 | * We need to ensure that the page cache pages are written to | |
858 | * disk and invalidated to preserve the expected O_DIRECT | |
859 | * semantics. | |
860 | */ | |
861 | pos = iocb->ki_pos; | |
862 | endbyte = pos + buffered - 1; | |
863 | ret = filemap_write_and_wait_range(mapping, pos, endbyte); | |
864 | if (!ret) { | |
865 | iocb->ki_pos += buffered; | |
866 | written += buffered; | |
867 | invalidate_mapping_pages(mapping, | |
868 | pos >> PAGE_SHIFT, | |
869 | endbyte >> PAGE_SHIFT); | |
870 | } else { | |
871 | /* | |
872 | * We don't know how much we wrote, so just return | |
873 | * the number of bytes which were direct-written | |
874 | */ | |
875 | } | |
876 | } else { | |
877 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); | |
878 | if (likely(ret > 0)) | |
879 | iocb->ki_pos += ret; | |
880 | } | |
64bc06bb AG |
881 | |
882 | out2: | |
883 | current->backing_dev_info = NULL; | |
884 | out: | |
885 | inode_unlock(inode); | |
886 | if (likely(ret > 0)) { | |
64bc06bb AG |
887 | /* Handle various SYNC-type writes */ |
888 | ret = generic_write_sync(iocb, ret); | |
889 | } | |
967bcc91 | 890 | return written ? written : ret; |
56aa616a SW |
891 | } |
892 | ||
2fe17c10 CH |
893 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
894 | int mode) | |
895 | { | |
fffb6412 | 896 | struct super_block *sb = inode->i_sb; |
2fe17c10 | 897 | struct gfs2_inode *ip = GFS2_I(inode); |
fffb6412 | 898 | loff_t end = offset + len; |
2fe17c10 CH |
899 | struct buffer_head *dibh; |
900 | int error; | |
2fe17c10 CH |
901 | |
902 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
903 | if (unlikely(error)) | |
64dd153c | 904 | return error; |
2fe17c10 | 905 | |
350a9b0a | 906 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
907 | |
908 | if (gfs2_is_stuffed(ip)) { | |
909 | error = gfs2_unstuff_dinode(ip, NULL); | |
910 | if (unlikely(error)) | |
911 | goto out; | |
912 | } | |
913 | ||
fffb6412 | 914 | while (offset < end) { |
c2589282 AG |
915 | struct iomap iomap = { }; |
916 | ||
628e366d AG |
917 | error = gfs2_iomap_get_alloc(inode, offset, end - offset, |
918 | &iomap); | |
fffb6412 | 919 | if (error) |
64dd153c | 920 | goto out; |
fffb6412 | 921 | offset = iomap.offset + iomap.length; |
d505a96a | 922 | if (!(iomap.flags & IOMAP_F_NEW)) |
64dd153c | 923 | continue; |
fffb6412 AG |
924 | error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, |
925 | iomap.length >> inode->i_blkbits, | |
926 | GFP_NOFS); | |
927 | if (error) { | |
928 | fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); | |
2fe17c10 | 929 | goto out; |
64dd153c | 930 | } |
2fe17c10 | 931 | } |
2fe17c10 | 932 | out: |
64dd153c | 933 | brelse(dibh); |
2fe17c10 CH |
934 | return error; |
935 | } | |
d9be0cda AD |
936 | /** |
937 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
938 | * blocks, determine how many bytes can be written. | |
939 | * @ip: The inode in question. | |
940 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
941 | * @data_blocks: Compute and return the number of data blocks needed | |
942 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
943 | * @max_blocks: The total blocks available to work with. | |
944 | * | |
945 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
946 | */ | |
947 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
948 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
949 | unsigned int max_blocks) | |
2fe17c10 | 950 | { |
d9be0cda | 951 | loff_t max = *len; |
2fe17c10 | 952 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
953 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
954 | ||
955 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
956 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
957 | max_data -= tmp; | |
958 | } | |
d9be0cda | 959 | |
2fe17c10 CH |
960 | *data_blocks = max_data; |
961 | *ind_blocks = max_blocks - max_data; | |
962 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
963 | if (*len > max) { | |
964 | *len = max; | |
965 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
966 | } | |
967 | } | |
968 | ||
9c9f1159 | 969 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 970 | { |
496ad9aa | 971 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
972 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
973 | struct gfs2_inode *ip = GFS2_I(inode); | |
7b9cff46 | 974 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
2fe17c10 | 975 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
174d1232 | 976 | loff_t bytes, max_bytes, max_blks; |
2fe17c10 | 977 | int error; |
4442f2e0 SW |
978 | const loff_t pos = offset; |
979 | const loff_t count = len; | |
6905d9e4 | 980 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 981 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 982 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 983 | |
2fe17c10 CH |
984 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
985 | ||
6905d9e4 | 986 | offset &= bsize_mask; |
2fe17c10 CH |
987 | |
988 | len = next - offset; | |
989 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
990 | if (!bytes) | |
991 | bytes = UINT_MAX; | |
6905d9e4 BM |
992 | bytes &= bsize_mask; |
993 | if (bytes == 0) | |
994 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 995 | |
da1dfb6a | 996 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 997 | |
d9be0cda AD |
998 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
999 | ap.min_target = data_blocks + ind_blocks; | |
1000 | ||
2fe17c10 CH |
1001 | while (len > 0) { |
1002 | if (len < bytes) | |
1003 | bytes = len; | |
58a7d5fb BM |
1004 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
1005 | len -= bytes; | |
1006 | offset += bytes; | |
1007 | continue; | |
1008 | } | |
d9be0cda AD |
1009 | |
1010 | /* We need to determine how many bytes we can actually | |
1011 | * fallocate without exceeding quota or going over the | |
1012 | * end of the fs. We start off optimistically by assuming | |
1013 | * we can write max_bytes */ | |
1014 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
1015 | ||
1016 | /* Since max_bytes is most likely a theoretical max, we | |
1017 | * calculate a more realistic 'bytes' to serve as a good | |
1018 | * starting point for the number of bytes we may be able | |
1019 | * to write */ | |
2fe17c10 | 1020 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 1021 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
1022 | |
1023 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 1024 | if (error) |
9c9f1159 | 1025 | return error; |
d9be0cda AD |
1026 | /* ap.allowed tells us how many blocks quota will allow |
1027 | * us to write. Check if this reduces max_blks */ | |
174d1232 AG |
1028 | max_blks = UINT_MAX; |
1029 | if (ap.allowed) | |
d9be0cda | 1030 | max_blks = ap.allowed; |
2fe17c10 | 1031 | |
7b9cff46 | 1032 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 1033 | if (error) |
2fe17c10 | 1034 | goto out_qunlock; |
d9be0cda AD |
1035 | |
1036 | /* check if the selected rgrp limits our max_blks further */ | |
1037 | if (ap.allowed && ap.allowed < max_blks) | |
1038 | max_blks = ap.allowed; | |
1039 | ||
1040 | /* Almost done. Calculate bytes that can be written using | |
1041 | * max_blks. We also recompute max_bytes, data_blocks and | |
1042 | * ind_blocks */ | |
1043 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
1044 | &ind_blocks, max_blks); | |
2fe17c10 CH |
1045 | |
1046 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 1047 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
1048 | if (gfs2_is_jdata(ip)) |
1049 | rblocks += data_blocks ? data_blocks : 1; | |
1050 | ||
1051 | error = gfs2_trans_begin(sdp, rblocks, | |
45eb0504 | 1052 | PAGE_SIZE >> inode->i_blkbits); |
2fe17c10 CH |
1053 | if (error) |
1054 | goto out_trans_fail; | |
1055 | ||
1056 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
1057 | gfs2_trans_end(sdp); | |
1058 | ||
1059 | if (error) | |
1060 | goto out_trans_fail; | |
1061 | ||
1062 | len -= max_bytes; | |
1063 | offset += max_bytes; | |
1064 | gfs2_inplace_release(ip); | |
1065 | gfs2_quota_unlock(ip); | |
2fe17c10 | 1066 | } |
4442f2e0 | 1067 | |
0a6a4abc | 1068 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) |
1885867b | 1069 | i_size_write(inode, pos + count); |
0a6a4abc AG |
1070 | file_update_time(file); |
1071 | mark_inode_dirty(inode); | |
1885867b | 1072 | |
dde0c2e7 CH |
1073 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
1074 | return vfs_fsync_range(file, pos, pos + count - 1, | |
1075 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
1076 | return 0; | |
2fe17c10 CH |
1077 | |
1078 | out_trans_fail: | |
1079 | gfs2_inplace_release(ip); | |
1080 | out_qunlock: | |
1081 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
1082 | return error; |
1083 | } | |
1084 | ||
1085 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
1086 | { | |
1087 | struct inode *inode = file_inode(file); | |
d4d7fc12 | 1088 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9c9f1159 AP |
1089 | struct gfs2_inode *ip = GFS2_I(inode); |
1090 | struct gfs2_holder gh; | |
1091 | int ret; | |
1092 | ||
4e56a641 | 1093 | if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) |
d4d7fc12 AP |
1094 | return -EOPNOTSUPP; |
1095 | /* fallocate is needed by gfs2_grow to reserve space in the rindex */ | |
1096 | if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) | |
9c9f1159 AP |
1097 | return -EOPNOTSUPP; |
1098 | ||
5955102c | 1099 | inode_lock(inode); |
9c9f1159 AP |
1100 | |
1101 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1102 | ret = gfs2_glock_nq(&gh); | |
1103 | if (ret) | |
1104 | goto out_uninit; | |
1105 | ||
1106 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
1107 | (offset + len) > inode->i_size) { | |
1108 | ret = inode_newsize_ok(inode, offset + len); | |
1109 | if (ret) | |
1110 | goto out_unlock; | |
1111 | } | |
1112 | ||
1113 | ret = get_write_access(inode); | |
1114 | if (ret) | |
1115 | goto out_unlock; | |
1116 | ||
4e56a641 AG |
1117 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
1118 | ret = __gfs2_punch_hole(file, offset, len); | |
1119 | } else { | |
1120 | ret = gfs2_rsqa_alloc(ip); | |
1121 | if (ret) | |
1122 | goto out_putw; | |
9c9f1159 | 1123 | |
4e56a641 AG |
1124 | ret = __gfs2_fallocate(file, mode, offset, len); |
1125 | ||
1126 | if (ret) | |
1127 | gfs2_rs_deltree(&ip->i_res); | |
1128 | } | |
a097dc7e | 1129 | |
9c9f1159 AP |
1130 | out_putw: |
1131 | put_write_access(inode); | |
2fe17c10 | 1132 | out_unlock: |
a0846a53 | 1133 | gfs2_glock_dq(&gh); |
2fe17c10 | 1134 | out_uninit: |
a0846a53 | 1135 | gfs2_holder_uninit(&gh); |
5955102c | 1136 | inode_unlock(inode); |
9c9f1159 | 1137 | return ret; |
2fe17c10 CH |
1138 | } |
1139 | ||
f1ea6f4e BP |
1140 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
1141 | struct file *out, loff_t *ppos, | |
1142 | size_t len, unsigned int flags) | |
1143 | { | |
1144 | int error; | |
1145 | struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); | |
1146 | ||
b54e9a0b | 1147 | error = gfs2_rsqa_alloc(ip); |
f1ea6f4e BP |
1148 | if (error) |
1149 | return (ssize_t)error; | |
1150 | ||
1151 | gfs2_size_hint(out, *ppos, len); | |
1152 | ||
1153 | return iter_file_splice_write(pipe, out, ppos, len, flags); | |
1154 | } | |
1155 | ||
f057f6cd SW |
1156 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
1157 | ||
b3b94faa DT |
1158 | /** |
1159 | * gfs2_lock - acquire/release a posix lock on a file | |
1160 | * @file: the file pointer | |
1161 | * @cmd: either modify or retrieve lock state, possibly wait | |
1162 | * @fl: type and range of lock | |
1163 | * | |
1164 | * Returns: errno | |
1165 | */ | |
1166 | ||
1167 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1168 | { | |
feaa7bba SW |
1169 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1170 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1171 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1172 | |
b3b94faa DT |
1173 | if (!(fl->fl_flags & FL_POSIX)) |
1174 | return -ENOLCK; | |
720e7749 | 1175 | if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) |
b3b94faa DT |
1176 | return -ENOLCK; |
1177 | ||
586759f0 ME |
1178 | if (cmd == F_CANCELLK) { |
1179 | /* Hack: */ | |
1180 | cmd = F_SETLK; | |
1181 | fl->fl_type = F_UNLCK; | |
1182 | } | |
04aea0ca | 1183 | if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags))) { |
c2952d20 | 1184 | if (fl->fl_type == F_UNLCK) |
4f656367 | 1185 | locks_lock_file_wait(file, fl); |
f057f6cd | 1186 | return -EIO; |
c2952d20 | 1187 | } |
b3b94faa | 1188 | if (IS_GETLK(cmd)) |
f057f6cd | 1189 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1190 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1191 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1192 | else |
f057f6cd | 1193 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1194 | } |
1195 | ||
b3b94faa DT |
1196 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1197 | { | |
5c676f6d | 1198 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1199 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1200 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1201 | struct gfs2_glock *gl; |
1202 | unsigned int state; | |
b58bf407 | 1203 | u16 flags; |
b3b94faa | 1204 | int error = 0; |
2ddfbdd6 | 1205 | int sleeptime; |
b3b94faa DT |
1206 | |
1207 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
2ddfbdd6 | 1208 | flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; |
b3b94faa | 1209 | |
f55ab26a | 1210 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa | 1211 | |
283c9a97 | 1212 | if (gfs2_holder_initialized(fl_gh)) { |
4d62d3f7 | 1213 | struct file_lock request; |
b3b94faa DT |
1214 | if (fl_gh->gh_state == state) |
1215 | goto out; | |
4d62d3f7 N |
1216 | locks_init_lock(&request); |
1217 | request.fl_type = F_UNLCK; | |
1218 | request.fl_flags = FL_FLOCK; | |
1219 | locks_lock_file_wait(file, &request); | |
5bef3e7c | 1220 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1221 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1222 | } else { |
6802e340 SW |
1223 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1224 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1225 | if (error) |
1226 | goto out; | |
b4c20166 AD |
1227 | gfs2_holder_init(gl, state, flags, fl_gh); |
1228 | gfs2_glock_put(gl); | |
b3b94faa | 1229 | } |
2ddfbdd6 BP |
1230 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1231 | error = gfs2_glock_nq(fl_gh); | |
1232 | if (error != GLR_TRYFAILED) | |
1233 | break; | |
1234 | fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; | |
1235 | fl_gh->gh_error = 0; | |
1236 | msleep(sleeptime); | |
1237 | } | |
b3b94faa DT |
1238 | if (error) { |
1239 | gfs2_holder_uninit(fl_gh); | |
1240 | if (error == GLR_TRYFAILED) | |
1241 | error = -EAGAIN; | |
1242 | } else { | |
4f656367 | 1243 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1244 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1245 | } |
1246 | ||
420b9e5e | 1247 | out: |
f55ab26a | 1248 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1249 | return error; |
1250 | } | |
1251 | ||
1252 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1253 | { | |
5c676f6d | 1254 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1255 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1256 | ||
f55ab26a | 1257 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1258 | locks_lock_file_wait(file, fl); |
6df9f9a2 | 1259 | if (gfs2_holder_initialized(fl_gh)) { |
2ddfbdd6 | 1260 | gfs2_glock_dq(fl_gh); |
0a33443b SW |
1261 | gfs2_holder_uninit(fl_gh); |
1262 | } | |
f55ab26a | 1263 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1264 | } |
1265 | ||
1266 | /** | |
1267 | * gfs2_flock - acquire/release a flock lock on a file | |
1268 | * @file: the file pointer | |
1269 | * @cmd: either modify or retrieve lock state, possibly wait | |
1270 | * @fl: type and range of lock | |
1271 | * | |
1272 | * Returns: errno | |
1273 | */ | |
1274 | ||
1275 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1276 | { | |
b3b94faa DT |
1277 | if (!(fl->fl_flags & FL_FLOCK)) |
1278 | return -ENOLCK; | |
a12af1eb AD |
1279 | if (fl->fl_type & LOCK_MAND) |
1280 | return -EOPNOTSUPP; | |
b3b94faa | 1281 | |
b3b94faa DT |
1282 | if (fl->fl_type == F_UNLCK) { |
1283 | do_unflock(file, fl); | |
1284 | return 0; | |
d00223f1 | 1285 | } else { |
b3b94faa | 1286 | return do_flock(file, cmd, fl); |
d00223f1 | 1287 | } |
b3b94faa DT |
1288 | } |
1289 | ||
10d21988 | 1290 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1291 | .llseek = gfs2_llseek, |
967bcc91 | 1292 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1293 | .write_iter = gfs2_file_write_iter, |
81214bab | 1294 | .iopoll = iomap_dio_iopoll, |
26c1a574 SW |
1295 | .unlocked_ioctl = gfs2_ioctl, |
1296 | .mmap = gfs2_mmap, | |
1297 | .open = gfs2_open, | |
df3fd117 | 1298 | .release = gfs2_release, |
26c1a574 SW |
1299 | .fsync = gfs2_fsync, |
1300 | .lock = gfs2_lock, | |
26c1a574 | 1301 | .flock = gfs2_flock, |
82c156f8 | 1302 | .splice_read = generic_file_splice_read, |
f42a69fa | 1303 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1304 | .setlease = simple_nosetlease, |
2fe17c10 | 1305 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1306 | }; |
1307 | ||
10d21988 | 1308 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1309 | .iterate_shared = gfs2_readdir, |
26c1a574 SW |
1310 | .unlocked_ioctl = gfs2_ioctl, |
1311 | .open = gfs2_open, | |
df3fd117 | 1312 | .release = gfs2_release, |
26c1a574 SW |
1313 | .fsync = gfs2_fsync, |
1314 | .lock = gfs2_lock, | |
1315 | .flock = gfs2_flock, | |
6038f373 | 1316 | .llseek = default_llseek, |
b3b94faa DT |
1317 | }; |
1318 | ||
f057f6cd SW |
1319 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1320 | ||
10d21988 | 1321 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1322 | .llseek = gfs2_llseek, |
967bcc91 | 1323 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1324 | .write_iter = gfs2_file_write_iter, |
81214bab | 1325 | .iopoll = iomap_dio_iopoll, |
c97bfe43 WC |
1326 | .unlocked_ioctl = gfs2_ioctl, |
1327 | .mmap = gfs2_mmap, | |
1328 | .open = gfs2_open, | |
df3fd117 | 1329 | .release = gfs2_release, |
c97bfe43 | 1330 | .fsync = gfs2_fsync, |
82c156f8 | 1331 | .splice_read = generic_file_splice_read, |
f42a69fa | 1332 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1333 | .setlease = generic_setlease, |
2fe17c10 | 1334 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1335 | }; |
1336 | ||
10d21988 | 1337 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1338 | .iterate_shared = gfs2_readdir, |
c97bfe43 WC |
1339 | .unlocked_ioctl = gfs2_ioctl, |
1340 | .open = gfs2_open, | |
df3fd117 | 1341 | .release = gfs2_release, |
c97bfe43 | 1342 | .fsync = gfs2_fsync, |
6038f373 | 1343 | .llseek = default_llseek, |
c97bfe43 WC |
1344 | }; |
1345 |