Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
b3b94faa DT |
10 | #include <linux/slab.h> |
11 | #include <linux/spinlock.h> | |
12 | #include <linux/completion.h> | |
13 | #include <linux/buffer_head.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/uio.h> | |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/mm.h> | |
f58ba889 | 18 | #include <linux/mount.h> |
18ec7d5c | 19 | #include <linux/fs.h> |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
2fe17c10 CH |
21 | #include <linux/falloc.h> |
22 | #include <linux/swap.h> | |
71b86f56 | 23 | #include <linux/crc32.h> |
33c3de32 | 24 | #include <linux/writeback.h> |
7c0f6ba6 | 25 | #include <linux/uaccess.h> |
f057f6cd SW |
26 | #include <linux/dlm.h> |
27 | #include <linux/dlm_plock.h> | |
2ddfbdd6 | 28 | #include <linux/delay.h> |
b3b94faa DT |
29 | |
30 | #include "gfs2.h" | |
5c676f6d | 31 | #include "incore.h" |
b3b94faa DT |
32 | #include "bmap.h" |
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 | |
b09e593d | 140 | static int gfs2_get_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 141 | { |
496ad9aa | 142 | struct inode *inode = file_inode(filp); |
feaa7bba | 143 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 | 144 | struct gfs2_holder gh; |
b16f7e57 AG |
145 | int i, error; |
146 | u32 gfsflags, fsflags = 0; | |
71b86f56 | 147 | |
719ee344 SW |
148 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
149 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 150 | if (error) |
9c7fe835 | 151 | goto out_uninit; |
907b9bce | 152 | |
b16f7e57 AG |
153 | gfsflags = ip->i_diskflags; |
154 | if (S_ISDIR(inode->i_mode)) | |
155 | gfsflags &= ~GFS2_DIF_JDATA; | |
156 | else | |
157 | gfsflags &= ~GFS2_DIF_INHERIT_JDATA; | |
158 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) | |
159 | if (gfsflags & fsflag_gfs2flag[i].gfsflag) | |
160 | fsflags |= fsflag_gfs2flag[i].fsflag; | |
161 | ||
128e5eba | 162 | if (put_user(fsflags, ptr)) |
71b86f56 SW |
163 | error = -EFAULT; |
164 | ||
3cc3f710 | 165 | gfs2_glock_dq(&gh); |
9c7fe835 | 166 | out_uninit: |
71b86f56 SW |
167 | gfs2_holder_uninit(&gh); |
168 | return error; | |
169 | } | |
170 | ||
6b124d8d SW |
171 | void gfs2_set_inode_flags(struct inode *inode) |
172 | { | |
173 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
174 | unsigned int flags = inode->i_flags; |
175 | ||
9964afbb SW |
176 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
177 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 178 | flags |= S_NOSEC; |
383f01fb | 179 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 180 | flags |= S_IMMUTABLE; |
383f01fb | 181 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 182 | flags |= S_APPEND; |
383f01fb | 183 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 184 | flags |= S_NOATIME; |
383f01fb | 185 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
186 | flags |= S_SYNC; |
187 | inode->i_flags = flags; | |
188 | } | |
189 | ||
71b86f56 SW |
190 | /* Flags that can be set by user space */ |
191 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
192 | GFS2_DIF_IMMUTABLE| \ |
193 | GFS2_DIF_APPENDONLY| \ | |
194 | GFS2_DIF_NOATIME| \ | |
195 | GFS2_DIF_SYNC| \ | |
23d0bb83 | 196 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
197 | GFS2_DIF_INHERIT_JDATA) |
198 | ||
199 | /** | |
9dd868e1 FF |
200 | * do_gfs2_set_flags - set flags on an inode |
201 | * @filp: file pointer | |
202 | * @reqflags: The flags to set | |
71b86f56 SW |
203 | * @mask: Indicates which flags are valid |
204 | * | |
205 | */ | |
b09e593d | 206 | static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) |
71b86f56 | 207 | { |
496ad9aa | 208 | struct inode *inode = file_inode(filp); |
feaa7bba SW |
209 | struct gfs2_inode *ip = GFS2_I(inode); |
210 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
211 | struct buffer_head *bh; |
212 | struct gfs2_holder gh; | |
213 | int error; | |
55eccc6d | 214 | u32 new_flags, flags; |
71b86f56 | 215 | |
a561be71 | 216 | error = mnt_want_write_file(filp); |
52f341cf | 217 | if (error) |
71b86f56 SW |
218 | return error; |
219 | ||
f58ba889 MS |
220 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
221 | if (error) | |
222 | goto out_drop_write; | |
223 | ||
7df0e039 | 224 | error = -EACCES; |
2e149670 | 225 | if (!inode_owner_or_capable(inode)) |
7df0e039 SW |
226 | goto out; |
227 | ||
228 | error = 0; | |
383f01fb | 229 | flags = ip->i_diskflags; |
55eccc6d | 230 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
231 | if ((new_flags ^ flags) == 0) |
232 | goto out; | |
233 | ||
71b86f56 SW |
234 | error = -EPERM; |
235 | if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) | |
236 | goto out; | |
237 | if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) | |
238 | goto out; | |
907b9bce | 239 | if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && |
b9cb9813 | 240 | !capable(CAP_LINUX_IMMUTABLE)) |
71b86f56 | 241 | goto out; |
b9cb9813 | 242 | if (!IS_IMMUTABLE(inode)) { |
10556cb2 | 243 | error = gfs2_permission(inode, MAY_WRITE); |
b9cb9813 SW |
244 | if (error) |
245 | goto out; | |
246 | } | |
5561093e | 247 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
cc555b09 | 248 | if (new_flags & GFS2_DIF_JDATA) |
c1696fb8 | 249 | gfs2_log_flush(sdp, ip->i_gl, |
805c0907 BP |
250 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
251 | GFS2_LFC_SET_FLAGS); | |
5561093e SW |
252 | error = filemap_fdatawrite(inode->i_mapping); |
253 | if (error) | |
254 | goto out; | |
255 | error = filemap_fdatawait(inode->i_mapping); | |
256 | if (error) | |
257 | goto out; | |
cc555b09 BP |
258 | if (new_flags & GFS2_DIF_JDATA) |
259 | gfs2_ordered_del_inode(ip); | |
5561093e | 260 | } |
55eccc6d | 261 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
262 | if (error) |
263 | goto out; | |
55eccc6d SW |
264 | error = gfs2_meta_inode_buffer(ip, &bh); |
265 | if (error) | |
266 | goto out_trans_end; | |
9b7c2ddb | 267 | inode->i_ctime = current_time(inode); |
350a9b0a | 268 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 269 | ip->i_diskflags = new_flags; |
539e5d6b | 270 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 271 | brelse(bh); |
6b124d8d | 272 | gfs2_set_inode_flags(inode); |
5561093e | 273 | gfs2_set_aops(inode); |
55eccc6d SW |
274 | out_trans_end: |
275 | gfs2_trans_end(sdp); | |
71b86f56 SW |
276 | out: |
277 | gfs2_glock_dq_uninit(&gh); | |
f58ba889 | 278 | out_drop_write: |
2a79f17e | 279 | mnt_drop_write_file(filp); |
71b86f56 SW |
280 | return error; |
281 | } | |
282 | ||
b09e593d | 283 | static int gfs2_set_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 284 | { |
496ad9aa | 285 | struct inode *inode = file_inode(filp); |
b16f7e57 AG |
286 | u32 fsflags, gfsflags = 0; |
287 | u32 mask; | |
288 | int i; | |
7df0e039 | 289 | |
128e5eba | 290 | if (get_user(fsflags, ptr)) |
71b86f56 | 291 | return -EFAULT; |
7df0e039 | 292 | |
b16f7e57 AG |
293 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { |
294 | if (fsflags & fsflag_gfs2flag[i].fsflag) { | |
295 | fsflags &= ~fsflag_gfs2flag[i].fsflag; | |
296 | gfsflags |= fsflag_gfs2flag[i].gfsflag; | |
297 | } | |
298 | } | |
299 | if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) | |
300 | return -EINVAL; | |
301 | ||
302 | mask = GFS2_FLAGS_USER_SET; | |
303 | if (S_ISDIR(inode->i_mode)) { | |
304 | mask &= ~GFS2_DIF_JDATA; | |
305 | } else { | |
306 | /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ | |
307 | if (gfsflags & GFS2_DIF_TOPDIR) | |
308 | return -EINVAL; | |
309 | mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); | |
b9af7ca6 | 310 | } |
b16f7e57 AG |
311 | |
312 | return do_gfs2_set_flags(filp, gfsflags, mask); | |
71b86f56 SW |
313 | } |
314 | ||
b09e593d | 315 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
316 | { |
317 | switch(cmd) { | |
128e5eba | 318 | case FS_IOC_GETFLAGS: |
b09e593d | 319 | return gfs2_get_flags(filp, (u32 __user *)arg); |
128e5eba | 320 | case FS_IOC_SETFLAGS: |
b09e593d | 321 | return gfs2_set_flags(filp, (u32 __user *)arg); |
66fc061b SW |
322 | case FITRIM: |
323 | return gfs2_fitrim(filp, (void __user *)arg); | |
71b86f56 SW |
324 | } |
325 | return -ENOTTY; | |
326 | } | |
327 | ||
da1dfb6a SW |
328 | /** |
329 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 330 | * @filep: The struct file |
da1dfb6a SW |
331 | * @offset: The file offset of the write |
332 | * @size: The length of the write | |
333 | * | |
334 | * When we are about to do a write, this function records the total | |
335 | * write size in order to provide a suitable hint to the lower layers | |
336 | * about how many blocks will be required. | |
337 | * | |
338 | */ | |
339 | ||
340 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
341 | { | |
496ad9aa | 342 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
343 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
344 | struct gfs2_inode *ip = GFS2_I(inode); | |
345 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
346 | int hint = min_t(size_t, INT_MAX, blks); | |
347 | ||
a097dc7e BP |
348 | if (hint > atomic_read(&ip->i_res.rs_sizehint)) |
349 | atomic_set(&ip->i_res.rs_sizehint, hint); | |
da1dfb6a SW |
350 | } |
351 | ||
3cc3f710 SW |
352 | /** |
353 | * gfs2_allocate_page_backing - Use bmap to allocate blocks | |
354 | * @page: The (locked) page to allocate backing for | |
355 | * | |
356 | * We try to allocate all the blocks required for the page in | |
357 | * one go. This might fail for various reasons, so we keep | |
358 | * trying until all the blocks to back this page are allocated. | |
359 | * If some of the blocks are already allocated, thats ok too. | |
360 | */ | |
361 | ||
362 | static int gfs2_allocate_page_backing(struct page *page) | |
363 | { | |
364 | struct inode *inode = page->mapping->host; | |
365 | struct buffer_head bh; | |
09cbfeaf KS |
366 | unsigned long size = PAGE_SIZE; |
367 | u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits); | |
3cc3f710 SW |
368 | |
369 | do { | |
370 | bh.b_state = 0; | |
371 | bh.b_size = size; | |
e9e1ef2b | 372 | gfs2_block_map(inode, lblock, &bh, 1); |
3cc3f710 SW |
373 | if (!buffer_mapped(&bh)) |
374 | return -EIO; | |
375 | size -= bh.b_size; | |
376 | lblock += (bh.b_size >> inode->i_blkbits); | |
377 | } while(size > 0); | |
378 | return 0; | |
379 | } | |
380 | ||
381 | /** | |
382 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
383 | * @vma: The virtual memory area | |
9dd868e1 | 384 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
385 | * |
386 | * When the page becomes writable, we need to ensure that we have | |
387 | * blocks allocated on disk to back that page. | |
388 | */ | |
389 | ||
11bac800 | 390 | static int gfs2_page_mkwrite(struct vm_fault *vmf) |
3cc3f710 | 391 | { |
c2ec175c | 392 | struct page *page = vmf->page; |
11bac800 | 393 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3cc3f710 SW |
394 | struct gfs2_inode *ip = GFS2_I(inode); |
395 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 396 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
3cc3f710 | 397 | unsigned long last_index; |
09cbfeaf | 398 | u64 pos = page->index << PAGE_SHIFT; |
3cc3f710 | 399 | unsigned int data_blocks, ind_blocks, rblocks; |
3cc3f710 | 400 | struct gfs2_holder gh; |
13d921e3 | 401 | loff_t size; |
3cc3f710 SW |
402 | int ret; |
403 | ||
39263d5e | 404 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 405 | |
b54e9a0b | 406 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 | 407 | if (ret) |
a097dc7e | 408 | goto out; |
0a305e49 | 409 | |
11bac800 | 410 | gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE); |
8e2e0047 | 411 | |
719ee344 SW |
412 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
413 | ret = gfs2_glock_nq(&gh); | |
3cc3f710 | 414 | if (ret) |
2b3dcf35 | 415 | goto out_uninit; |
3cc3f710 | 416 | |
d7c436cd | 417 | /* Update file times before taking page lock */ |
11bac800 | 418 | file_update_time(vmf->vma->vm_file); |
d7c436cd | 419 | |
9c538837 SW |
420 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
421 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
422 | ||
09cbfeaf | 423 | if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { |
13d921e3 SW |
424 | lock_page(page); |
425 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
426 | ret = -EAGAIN; | |
427 | unlock_page(page); | |
428 | } | |
3cc3f710 | 429 | goto out_unlock; |
13d921e3 SW |
430 | } |
431 | ||
5407e242 BP |
432 | ret = gfs2_rindex_update(sdp); |
433 | if (ret) | |
6dbd8224 SW |
434 | goto out_unlock; |
435 | ||
09cbfeaf | 436 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
7b9cff46 | 437 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
438 | ret = gfs2_quota_lock_check(ip, &ap); |
439 | if (ret) | |
440 | goto out_unlock; | |
7b9cff46 | 441 | ret = gfs2_inplace_reserve(ip, &ap); |
3cc3f710 SW |
442 | if (ret) |
443 | goto out_quota_unlock; | |
444 | ||
445 | rblocks = RES_DINODE + ind_blocks; | |
446 | if (gfs2_is_jdata(ip)) | |
447 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 448 | if (ind_blocks || data_blocks) { |
3cc3f710 | 449 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 450 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 451 | } |
3cc3f710 SW |
452 | ret = gfs2_trans_begin(sdp, rblocks, 0); |
453 | if (ret) | |
454 | goto out_trans_fail; | |
455 | ||
456 | lock_page(page); | |
457 | ret = -EINVAL; | |
13d921e3 | 458 | size = i_size_read(inode); |
09cbfeaf | 459 | last_index = (size - 1) >> PAGE_SHIFT; |
13d921e3 SW |
460 | /* Check page index against inode size */ |
461 | if (size == 0 || (page->index > last_index)) | |
462 | goto out_trans_end; | |
463 | ||
464 | ret = -EAGAIN; | |
465 | /* If truncated, we must retry the operation, we may have raced | |
466 | * with the glock demotion code. | |
467 | */ | |
468 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) | |
469 | goto out_trans_end; | |
470 | ||
471 | /* Unstuff, if required, and allocate backing blocks for page */ | |
b7fe2e39 | 472 | ret = 0; |
13d921e3 | 473 | if (gfs2_is_stuffed(ip)) |
3cc3f710 | 474 | ret = gfs2_unstuff_dinode(ip, page); |
13d921e3 SW |
475 | if (ret == 0) |
476 | ret = gfs2_allocate_page_backing(page); | |
3cc3f710 | 477 | |
13d921e3 SW |
478 | out_trans_end: |
479 | if (ret) | |
480 | unlock_page(page); | |
3cc3f710 SW |
481 | gfs2_trans_end(sdp); |
482 | out_trans_fail: | |
483 | gfs2_inplace_release(ip); | |
484 | out_quota_unlock: | |
485 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
486 | out_unlock: |
487 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 488 | out_uninit: |
3cc3f710 | 489 | gfs2_holder_uninit(&gh); |
13d921e3 SW |
490 | if (ret == 0) { |
491 | set_page_dirty(page); | |
1d1d1a76 | 492 | wait_for_stable_page(page); |
13d921e3 | 493 | } |
2b3dcf35 | 494 | out: |
39263d5e | 495 | sb_end_pagefault(inode->i_sb); |
13d921e3 | 496 | return block_page_mkwrite_return(ret); |
3cc3f710 SW |
497 | } |
498 | ||
f0f37e2f | 499 | static const struct vm_operations_struct gfs2_vm_ops = { |
3cc3f710 | 500 | .fault = filemap_fault, |
f1820361 | 501 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
502 | .page_mkwrite = gfs2_page_mkwrite, |
503 | }; | |
504 | ||
b3b94faa DT |
505 | /** |
506 | * gfs2_mmap - | |
507 | * @file: The file to map | |
508 | * @vma: The VMA which described the mapping | |
509 | * | |
48bf2b17 SW |
510 | * There is no need to get a lock here unless we should be updating |
511 | * atime. We ignore any locking errors since the only consequence is | |
512 | * a missed atime update (which will just be deferred until later). | |
513 | * | |
514 | * Returns: 0 | |
b3b94faa DT |
515 | */ |
516 | ||
517 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
518 | { | |
feaa7bba | 519 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 520 | |
b9c93bb7 SW |
521 | if (!(file->f_flags & O_NOATIME) && |
522 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
523 | struct gfs2_holder i_gh; |
524 | int error; | |
b3b94faa | 525 | |
3d162688 BM |
526 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
527 | &i_gh); | |
b9c93bb7 SW |
528 | if (error) |
529 | return error; | |
3d162688 BM |
530 | /* grab lock to update inode */ |
531 | gfs2_glock_dq_uninit(&i_gh); | |
532 | file_accessed(file); | |
48bf2b17 | 533 | } |
3cc3f710 | 534 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 535 | |
48bf2b17 | 536 | return 0; |
b3b94faa DT |
537 | } |
538 | ||
539 | /** | |
6d4ade98 SW |
540 | * gfs2_open_common - This is common to open and atomic_open |
541 | * @inode: The inode being opened | |
542 | * @file: The file being opened | |
b3b94faa | 543 | * |
6d4ade98 SW |
544 | * This maybe called under a glock or not depending upon how it has |
545 | * been called. We must always be called under a glock for regular | |
546 | * files, however. For other file types, it does not matter whether | |
547 | * we hold the glock or not. | |
548 | * | |
549 | * Returns: Error code or 0 for success | |
b3b94faa DT |
550 | */ |
551 | ||
6d4ade98 | 552 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 553 | { |
b3b94faa | 554 | struct gfs2_file *fp; |
6d4ade98 SW |
555 | int ret; |
556 | ||
557 | if (S_ISREG(inode->i_mode)) { | |
558 | ret = generic_file_open(inode, file); | |
559 | if (ret) | |
560 | return ret; | |
561 | } | |
b3b94faa | 562 | |
6d4ade98 | 563 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
564 | if (!fp) |
565 | return -ENOMEM; | |
566 | ||
f55ab26a | 567 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 568 | |
feaa7bba | 569 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 570 | file->private_data = fp; |
6d4ade98 SW |
571 | return 0; |
572 | } | |
573 | ||
574 | /** | |
575 | * gfs2_open - open a file | |
576 | * @inode: the inode to open | |
577 | * @file: the struct file for this opening | |
578 | * | |
579 | * After atomic_open, this function is only used for opening files | |
580 | * which are already cached. We must still get the glock for regular | |
581 | * files to ensure that we have the file size uptodate for the large | |
582 | * file check which is in the common code. That is only an issue for | |
583 | * regular files though. | |
584 | * | |
585 | * Returns: errno | |
586 | */ | |
587 | ||
588 | static int gfs2_open(struct inode *inode, struct file *file) | |
589 | { | |
590 | struct gfs2_inode *ip = GFS2_I(inode); | |
591 | struct gfs2_holder i_gh; | |
592 | int error; | |
593 | bool need_unlock = false; | |
b3b94faa | 594 | |
b60623c2 | 595 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
596 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
597 | &i_gh); | |
598 | if (error) | |
6d4ade98 SW |
599 | return error; |
600 | need_unlock = true; | |
601 | } | |
b3b94faa | 602 | |
6d4ade98 | 603 | error = gfs2_open_common(inode, file); |
b3b94faa | 604 | |
6d4ade98 | 605 | if (need_unlock) |
b3b94faa | 606 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 607 | |
b3b94faa DT |
608 | return error; |
609 | } | |
610 | ||
611 | /** | |
df3fd117 | 612 | * gfs2_release - called to close a struct file |
b3b94faa DT |
613 | * @inode: the inode the struct file belongs to |
614 | * @file: the struct file being closed | |
615 | * | |
616 | * Returns: errno | |
617 | */ | |
618 | ||
df3fd117 | 619 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 620 | { |
0a305e49 | 621 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 622 | |
8e2e0047 | 623 | kfree(file->private_data); |
5c676f6d | 624 | file->private_data = NULL; |
b3b94faa | 625 | |
2b3dcf35 BP |
626 | if (!(file->f_mode & FMODE_WRITE)) |
627 | return 0; | |
0a305e49 | 628 | |
b54e9a0b | 629 | gfs2_rsqa_delete(ip, &inode->i_writecount); |
b3b94faa DT |
630 | return 0; |
631 | } | |
632 | ||
633 | /** | |
634 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
635 | * @file: the file that points to the dentry |
636 | * @start: the start position in the file to sync | |
637 | * @end: the end position in the file to sync | |
dba898b0 | 638 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 639 | * |
2f0264d5 SW |
640 | * We split the data flushing here so that we don't wait for the data |
641 | * until after we've also sent the metadata to disk. Note that for | |
642 | * data=ordered, we will write & wait for the data at the log flush | |
643 | * stage anyway, so this is unlikely to make much of a difference | |
644 | * except in the data=writeback case. | |
645 | * | |
646 | * If the fdatawrite fails due to any reason except -EIO, we will | |
647 | * continue the remainder of the fsync, although we'll still report | |
648 | * the error at the end. This is to match filemap_write_and_wait_range() | |
649 | * behaviour. | |
34126f9f | 650 | * |
b3b94faa DT |
651 | * Returns: errno |
652 | */ | |
653 | ||
02c24a82 JB |
654 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
655 | int datasync) | |
b3b94faa | 656 | { |
2f0264d5 SW |
657 | struct address_space *mapping = file->f_mapping; |
658 | struct inode *inode = mapping->host; | |
0ae45f63 | 659 | int sync_state = inode->i_state & I_DIRTY_ALL; |
dba898b0 | 660 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 661 | int ret = 0, ret1 = 0; |
b3b94faa | 662 | |
2f0264d5 SW |
663 | if (mapping->nrpages) { |
664 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
665 | if (ret1 == -EIO) | |
666 | return ret1; | |
667 | } | |
02c24a82 | 668 | |
0c901809 BM |
669 | if (!gfs2_is_jdata(ip)) |
670 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 671 | if (datasync) |
0ae45f63 | 672 | sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); |
b3b94faa | 673 | |
dba898b0 SW |
674 | if (sync_state) { |
675 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 676 | if (ret) |
dba898b0 | 677 | return ret; |
f1818529 | 678 | if (gfs2_is_jdata(ip)) |
d07a6ac7 JL |
679 | ret = file_write_and_wait(file); |
680 | if (ret) | |
681 | return ret; | |
b5b24d7a | 682 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
683 | } |
684 | ||
2f0264d5 | 685 | if (mapping->nrpages) |
d07a6ac7 | 686 | ret = file_fdatawait_range(file, start, end); |
2f0264d5 SW |
687 | |
688 | return ret ? ret : ret1; | |
b3b94faa DT |
689 | } |
690 | ||
56aa616a | 691 | /** |
da56e45b | 692 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a SW |
693 | * @iocb: The io context |
694 | * @iov: The data to write | |
695 | * @nr_segs: Number of @iov segments | |
696 | * @pos: The file position | |
697 | * | |
698 | * We have to do a lock/unlock here to refresh the inode size for | |
699 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
700 | * offset. There is still a race, but provided the app is using its | |
701 | * own file locking, this will make O_APPEND work as expected. | |
702 | * | |
703 | */ | |
704 | ||
da56e45b | 705 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
706 | { |
707 | struct file *file = iocb->ki_filp; | |
496ad9aa | 708 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
0a305e49 BP |
709 | int ret; |
710 | ||
b54e9a0b | 711 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 BP |
712 | if (ret) |
713 | return ret; | |
56aa616a | 714 | |
da56e45b | 715 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 716 | |
2ba48ce5 | 717 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a | 718 | struct gfs2_holder gh; |
56aa616a SW |
719 | |
720 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
721 | if (ret) | |
722 | return ret; | |
723 | gfs2_glock_dq_uninit(&gh); | |
724 | } | |
725 | ||
da56e45b | 726 | return generic_file_write_iter(iocb, from); |
56aa616a SW |
727 | } |
728 | ||
2fe17c10 CH |
729 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
730 | int mode) | |
731 | { | |
fffb6412 | 732 | struct super_block *sb = inode->i_sb; |
2fe17c10 | 733 | struct gfs2_inode *ip = GFS2_I(inode); |
fffb6412 | 734 | loff_t end = offset + len; |
2fe17c10 | 735 | struct buffer_head *dibh; |
628e366d | 736 | struct iomap iomap = { }; |
2fe17c10 | 737 | int error; |
2fe17c10 CH |
738 | |
739 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
740 | if (unlikely(error)) | |
64dd153c | 741 | return error; |
2fe17c10 | 742 | |
350a9b0a | 743 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
744 | |
745 | if (gfs2_is_stuffed(ip)) { | |
746 | error = gfs2_unstuff_dinode(ip, NULL); | |
747 | if (unlikely(error)) | |
748 | goto out; | |
749 | } | |
750 | ||
fffb6412 | 751 | while (offset < end) { |
628e366d AG |
752 | error = gfs2_iomap_get_alloc(inode, offset, end - offset, |
753 | &iomap); | |
fffb6412 | 754 | if (error) |
64dd153c | 755 | goto out; |
fffb6412 AG |
756 | offset = iomap.offset + iomap.length; |
757 | if (iomap.type != IOMAP_HOLE) | |
64dd153c | 758 | continue; |
fffb6412 AG |
759 | error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, |
760 | iomap.length >> inode->i_blkbits, | |
761 | GFP_NOFS); | |
762 | if (error) { | |
763 | fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); | |
2fe17c10 | 764 | goto out; |
64dd153c | 765 | } |
2fe17c10 | 766 | } |
2fe17c10 | 767 | out: |
64dd153c | 768 | brelse(dibh); |
2fe17c10 CH |
769 | return error; |
770 | } | |
d9be0cda AD |
771 | /** |
772 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
773 | * blocks, determine how many bytes can be written. | |
774 | * @ip: The inode in question. | |
775 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
776 | * @data_blocks: Compute and return the number of data blocks needed | |
777 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
778 | * @max_blocks: The total blocks available to work with. | |
779 | * | |
780 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
781 | */ | |
782 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
783 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
784 | unsigned int max_blocks) | |
2fe17c10 | 785 | { |
d9be0cda | 786 | loff_t max = *len; |
2fe17c10 | 787 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
788 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
789 | ||
790 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
791 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
792 | max_data -= tmp; | |
793 | } | |
d9be0cda | 794 | |
2fe17c10 CH |
795 | *data_blocks = max_data; |
796 | *ind_blocks = max_blocks - max_data; | |
797 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
798 | if (*len > max) { | |
799 | *len = max; | |
800 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
801 | } | |
802 | } | |
803 | ||
9c9f1159 | 804 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 805 | { |
496ad9aa | 806 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
807 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
808 | struct gfs2_inode *ip = GFS2_I(inode); | |
7b9cff46 | 809 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
2fe17c10 | 810 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
174d1232 | 811 | loff_t bytes, max_bytes, max_blks; |
2fe17c10 | 812 | int error; |
4442f2e0 SW |
813 | const loff_t pos = offset; |
814 | const loff_t count = len; | |
6905d9e4 | 815 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 816 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 817 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 818 | |
2fe17c10 CH |
819 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
820 | ||
6905d9e4 | 821 | offset &= bsize_mask; |
2fe17c10 CH |
822 | |
823 | len = next - offset; | |
824 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
825 | if (!bytes) | |
826 | bytes = UINT_MAX; | |
6905d9e4 BM |
827 | bytes &= bsize_mask; |
828 | if (bytes == 0) | |
829 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 830 | |
da1dfb6a | 831 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 832 | |
d9be0cda AD |
833 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
834 | ap.min_target = data_blocks + ind_blocks; | |
835 | ||
2fe17c10 CH |
836 | while (len > 0) { |
837 | if (len < bytes) | |
838 | bytes = len; | |
58a7d5fb BM |
839 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
840 | len -= bytes; | |
841 | offset += bytes; | |
842 | continue; | |
843 | } | |
d9be0cda AD |
844 | |
845 | /* We need to determine how many bytes we can actually | |
846 | * fallocate without exceeding quota or going over the | |
847 | * end of the fs. We start off optimistically by assuming | |
848 | * we can write max_bytes */ | |
849 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
850 | ||
851 | /* Since max_bytes is most likely a theoretical max, we | |
852 | * calculate a more realistic 'bytes' to serve as a good | |
853 | * starting point for the number of bytes we may be able | |
854 | * to write */ | |
2fe17c10 | 855 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 856 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
857 | |
858 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 859 | if (error) |
9c9f1159 | 860 | return error; |
d9be0cda AD |
861 | /* ap.allowed tells us how many blocks quota will allow |
862 | * us to write. Check if this reduces max_blks */ | |
174d1232 AG |
863 | max_blks = UINT_MAX; |
864 | if (ap.allowed) | |
d9be0cda | 865 | max_blks = ap.allowed; |
2fe17c10 | 866 | |
7b9cff46 | 867 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 868 | if (error) |
2fe17c10 | 869 | goto out_qunlock; |
d9be0cda AD |
870 | |
871 | /* check if the selected rgrp limits our max_blks further */ | |
872 | if (ap.allowed && ap.allowed < max_blks) | |
873 | max_blks = ap.allowed; | |
874 | ||
875 | /* Almost done. Calculate bytes that can be written using | |
876 | * max_blks. We also recompute max_bytes, data_blocks and | |
877 | * ind_blocks */ | |
878 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
879 | &ind_blocks, max_blks); | |
2fe17c10 CH |
880 | |
881 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 882 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
883 | if (gfs2_is_jdata(ip)) |
884 | rblocks += data_blocks ? data_blocks : 1; | |
885 | ||
886 | error = gfs2_trans_begin(sdp, rblocks, | |
09cbfeaf | 887 | PAGE_SIZE/sdp->sd_sb.sb_bsize); |
2fe17c10 CH |
888 | if (error) |
889 | goto out_trans_fail; | |
890 | ||
891 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
892 | gfs2_trans_end(sdp); | |
893 | ||
894 | if (error) | |
895 | goto out_trans_fail; | |
896 | ||
897 | len -= max_bytes; | |
898 | offset += max_bytes; | |
899 | gfs2_inplace_release(ip); | |
900 | gfs2_quota_unlock(ip); | |
2fe17c10 | 901 | } |
4442f2e0 | 902 | |
1885867b AP |
903 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) { |
904 | i_size_write(inode, pos + count); | |
98f1a696 | 905 | file_update_time(file); |
4b813f09 | 906 | mark_inode_dirty(inode); |
1885867b AP |
907 | } |
908 | ||
dde0c2e7 CH |
909 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
910 | return vfs_fsync_range(file, pos, pos + count - 1, | |
911 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
912 | return 0; | |
2fe17c10 CH |
913 | |
914 | out_trans_fail: | |
915 | gfs2_inplace_release(ip); | |
916 | out_qunlock: | |
917 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
918 | return error; |
919 | } | |
920 | ||
921 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
922 | { | |
923 | struct inode *inode = file_inode(file); | |
d4d7fc12 | 924 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9c9f1159 AP |
925 | struct gfs2_inode *ip = GFS2_I(inode); |
926 | struct gfs2_holder gh; | |
927 | int ret; | |
928 | ||
4e56a641 | 929 | if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) |
d4d7fc12 AP |
930 | return -EOPNOTSUPP; |
931 | /* fallocate is needed by gfs2_grow to reserve space in the rindex */ | |
932 | if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) | |
9c9f1159 AP |
933 | return -EOPNOTSUPP; |
934 | ||
5955102c | 935 | inode_lock(inode); |
9c9f1159 AP |
936 | |
937 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
938 | ret = gfs2_glock_nq(&gh); | |
939 | if (ret) | |
940 | goto out_uninit; | |
941 | ||
942 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
943 | (offset + len) > inode->i_size) { | |
944 | ret = inode_newsize_ok(inode, offset + len); | |
945 | if (ret) | |
946 | goto out_unlock; | |
947 | } | |
948 | ||
949 | ret = get_write_access(inode); | |
950 | if (ret) | |
951 | goto out_unlock; | |
952 | ||
4e56a641 AG |
953 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
954 | ret = __gfs2_punch_hole(file, offset, len); | |
955 | } else { | |
956 | ret = gfs2_rsqa_alloc(ip); | |
957 | if (ret) | |
958 | goto out_putw; | |
9c9f1159 | 959 | |
4e56a641 AG |
960 | ret = __gfs2_fallocate(file, mode, offset, len); |
961 | ||
962 | if (ret) | |
963 | gfs2_rs_deltree(&ip->i_res); | |
964 | } | |
a097dc7e | 965 | |
9c9f1159 AP |
966 | out_putw: |
967 | put_write_access(inode); | |
2fe17c10 | 968 | out_unlock: |
a0846a53 | 969 | gfs2_glock_dq(&gh); |
2fe17c10 | 970 | out_uninit: |
a0846a53 | 971 | gfs2_holder_uninit(&gh); |
5955102c | 972 | inode_unlock(inode); |
9c9f1159 | 973 | return ret; |
2fe17c10 CH |
974 | } |
975 | ||
f1ea6f4e BP |
976 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
977 | struct file *out, loff_t *ppos, | |
978 | size_t len, unsigned int flags) | |
979 | { | |
980 | int error; | |
981 | struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); | |
982 | ||
b54e9a0b | 983 | error = gfs2_rsqa_alloc(ip); |
f1ea6f4e BP |
984 | if (error) |
985 | return (ssize_t)error; | |
986 | ||
987 | gfs2_size_hint(out, *ppos, len); | |
988 | ||
989 | return iter_file_splice_write(pipe, out, ppos, len, flags); | |
990 | } | |
991 | ||
f057f6cd SW |
992 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
993 | ||
b3b94faa DT |
994 | /** |
995 | * gfs2_lock - acquire/release a posix lock on a file | |
996 | * @file: the file pointer | |
997 | * @cmd: either modify or retrieve lock state, possibly wait | |
998 | * @fl: type and range of lock | |
999 | * | |
1000 | * Returns: errno | |
1001 | */ | |
1002 | ||
1003 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1004 | { | |
feaa7bba SW |
1005 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1006 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1007 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1008 | |
b3b94faa DT |
1009 | if (!(fl->fl_flags & FL_POSIX)) |
1010 | return -ENOLCK; | |
720e7749 | 1011 | if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) |
b3b94faa DT |
1012 | return -ENOLCK; |
1013 | ||
586759f0 ME |
1014 | if (cmd == F_CANCELLK) { |
1015 | /* Hack: */ | |
1016 | cmd = F_SETLK; | |
1017 | fl->fl_type = F_UNLCK; | |
1018 | } | |
c2952d20 SW |
1019 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { |
1020 | if (fl->fl_type == F_UNLCK) | |
4f656367 | 1021 | locks_lock_file_wait(file, fl); |
f057f6cd | 1022 | return -EIO; |
c2952d20 | 1023 | } |
b3b94faa | 1024 | if (IS_GETLK(cmd)) |
f057f6cd | 1025 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1026 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1027 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1028 | else |
f057f6cd | 1029 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1030 | } |
1031 | ||
b3b94faa DT |
1032 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1033 | { | |
5c676f6d | 1034 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1035 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1036 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1037 | struct gfs2_glock *gl; |
1038 | unsigned int state; | |
b58bf407 | 1039 | u16 flags; |
b3b94faa | 1040 | int error = 0; |
2ddfbdd6 | 1041 | int sleeptime; |
b3b94faa DT |
1042 | |
1043 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
2ddfbdd6 | 1044 | flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; |
b3b94faa | 1045 | |
f55ab26a | 1046 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa | 1047 | |
283c9a97 | 1048 | if (gfs2_holder_initialized(fl_gh)) { |
b3b94faa DT |
1049 | if (fl_gh->gh_state == state) |
1050 | goto out; | |
4f656367 | 1051 | locks_lock_file_wait(file, |
a93a9983 JB |
1052 | &(struct file_lock) { |
1053 | .fl_type = F_UNLCK, | |
1054 | .fl_flags = FL_FLOCK | |
1055 | }); | |
5bef3e7c | 1056 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1057 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1058 | } else { |
6802e340 SW |
1059 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1060 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1061 | if (error) |
1062 | goto out; | |
b4c20166 AD |
1063 | gfs2_holder_init(gl, state, flags, fl_gh); |
1064 | gfs2_glock_put(gl); | |
b3b94faa | 1065 | } |
2ddfbdd6 BP |
1066 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1067 | error = gfs2_glock_nq(fl_gh); | |
1068 | if (error != GLR_TRYFAILED) | |
1069 | break; | |
1070 | fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; | |
1071 | fl_gh->gh_error = 0; | |
1072 | msleep(sleeptime); | |
1073 | } | |
b3b94faa DT |
1074 | if (error) { |
1075 | gfs2_holder_uninit(fl_gh); | |
1076 | if (error == GLR_TRYFAILED) | |
1077 | error = -EAGAIN; | |
1078 | } else { | |
4f656367 | 1079 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1080 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1081 | } |
1082 | ||
420b9e5e | 1083 | out: |
f55ab26a | 1084 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1085 | return error; |
1086 | } | |
1087 | ||
1088 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1089 | { | |
5c676f6d | 1090 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1091 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1092 | ||
f55ab26a | 1093 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1094 | locks_lock_file_wait(file, fl); |
6df9f9a2 | 1095 | if (gfs2_holder_initialized(fl_gh)) { |
2ddfbdd6 | 1096 | gfs2_glock_dq(fl_gh); |
0a33443b SW |
1097 | gfs2_holder_uninit(fl_gh); |
1098 | } | |
f55ab26a | 1099 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1100 | } |
1101 | ||
1102 | /** | |
1103 | * gfs2_flock - acquire/release a flock lock on a file | |
1104 | * @file: the file pointer | |
1105 | * @cmd: either modify or retrieve lock state, possibly wait | |
1106 | * @fl: type and range of lock | |
1107 | * | |
1108 | * Returns: errno | |
1109 | */ | |
1110 | ||
1111 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1112 | { | |
b3b94faa DT |
1113 | if (!(fl->fl_flags & FL_FLOCK)) |
1114 | return -ENOLCK; | |
a12af1eb AD |
1115 | if (fl->fl_type & LOCK_MAND) |
1116 | return -EOPNOTSUPP; | |
b3b94faa | 1117 | |
b3b94faa DT |
1118 | if (fl->fl_type == F_UNLCK) { |
1119 | do_unflock(file, fl); | |
1120 | return 0; | |
d00223f1 | 1121 | } else { |
b3b94faa | 1122 | return do_flock(file, cmd, fl); |
d00223f1 | 1123 | } |
b3b94faa DT |
1124 | } |
1125 | ||
10d21988 | 1126 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1127 | .llseek = gfs2_llseek, |
aad4f8bb | 1128 | .read_iter = generic_file_read_iter, |
da56e45b | 1129 | .write_iter = gfs2_file_write_iter, |
26c1a574 SW |
1130 | .unlocked_ioctl = gfs2_ioctl, |
1131 | .mmap = gfs2_mmap, | |
1132 | .open = gfs2_open, | |
df3fd117 | 1133 | .release = gfs2_release, |
26c1a574 SW |
1134 | .fsync = gfs2_fsync, |
1135 | .lock = gfs2_lock, | |
26c1a574 | 1136 | .flock = gfs2_flock, |
82c156f8 | 1137 | .splice_read = generic_file_splice_read, |
f42a69fa | 1138 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1139 | .setlease = simple_nosetlease, |
2fe17c10 | 1140 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1141 | }; |
1142 | ||
10d21988 | 1143 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1144 | .iterate_shared = gfs2_readdir, |
26c1a574 SW |
1145 | .unlocked_ioctl = gfs2_ioctl, |
1146 | .open = gfs2_open, | |
df3fd117 | 1147 | .release = gfs2_release, |
26c1a574 SW |
1148 | .fsync = gfs2_fsync, |
1149 | .lock = gfs2_lock, | |
1150 | .flock = gfs2_flock, | |
6038f373 | 1151 | .llseek = default_llseek, |
b3b94faa DT |
1152 | }; |
1153 | ||
f057f6cd SW |
1154 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1155 | ||
10d21988 | 1156 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1157 | .llseek = gfs2_llseek, |
aad4f8bb | 1158 | .read_iter = generic_file_read_iter, |
da56e45b | 1159 | .write_iter = gfs2_file_write_iter, |
c97bfe43 WC |
1160 | .unlocked_ioctl = gfs2_ioctl, |
1161 | .mmap = gfs2_mmap, | |
1162 | .open = gfs2_open, | |
df3fd117 | 1163 | .release = gfs2_release, |
c97bfe43 | 1164 | .fsync = gfs2_fsync, |
82c156f8 | 1165 | .splice_read = generic_file_splice_read, |
f42a69fa | 1166 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1167 | .setlease = generic_setlease, |
2fe17c10 | 1168 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1169 | }; |
1170 | ||
10d21988 | 1171 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1172 | .iterate_shared = gfs2_readdir, |
c97bfe43 WC |
1173 | .unlocked_ioctl = gfs2_ioctl, |
1174 | .open = gfs2_open, | |
df3fd117 | 1175 | .release = gfs2_release, |
c97bfe43 | 1176 | .fsync = gfs2_fsync, |
6038f373 | 1177 | .llseek = default_llseek, |
c97bfe43 WC |
1178 | }; |
1179 |