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> |
64bc06bb | 29 | #include <linux/backing-dev.h> |
b3b94faa DT |
30 | |
31 | #include "gfs2.h" | |
5c676f6d | 32 | #include "incore.h" |
b3b94faa | 33 | #include "bmap.h" |
64bc06bb | 34 | #include "aops.h" |
b3b94faa DT |
35 | #include "dir.h" |
36 | #include "glock.h" | |
37 | #include "glops.h" | |
38 | #include "inode.h" | |
b3b94faa DT |
39 | #include "log.h" |
40 | #include "meta_io.h" | |
b3b94faa DT |
41 | #include "quota.h" |
42 | #include "rgrp.h" | |
43 | #include "trans.h" | |
5c676f6d | 44 | #include "util.h" |
b3b94faa | 45 | |
b3b94faa DT |
46 | /** |
47 | * gfs2_llseek - seek to a location in a file | |
48 | * @file: the file | |
49 | * @offset: the offset | |
965c8e59 | 50 | * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) |
b3b94faa DT |
51 | * |
52 | * SEEK_END requires the glock for the file because it references the | |
53 | * file's size. | |
54 | * | |
55 | * Returns: The new offset, or errno | |
56 | */ | |
57 | ||
965c8e59 | 58 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence) |
b3b94faa | 59 | { |
feaa7bba | 60 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa DT |
61 | struct gfs2_holder i_gh; |
62 | loff_t error; | |
63 | ||
965c8e59 | 64 | switch (whence) { |
3a27411c | 65 | case SEEK_END: |
b3b94faa DT |
66 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
67 | &i_gh); | |
68 | if (!error) { | |
965c8e59 | 69 | error = generic_file_llseek(file, offset, whence); |
b3b94faa DT |
70 | gfs2_glock_dq_uninit(&i_gh); |
71 | } | |
9453615a | 72 | break; |
3a27411c AG |
73 | |
74 | case SEEK_DATA: | |
75 | error = gfs2_seek_data(file, offset); | |
76 | break; | |
77 | ||
78 | case SEEK_HOLE: | |
79 | error = gfs2_seek_hole(file, offset); | |
80 | break; | |
81 | ||
9453615a SW |
82 | case SEEK_CUR: |
83 | case SEEK_SET: | |
3a27411c AG |
84 | /* |
85 | * These don't reference inode->i_size and don't depend on the | |
86 | * block mapping, so we don't need the glock. | |
87 | */ | |
965c8e59 | 88 | error = generic_file_llseek(file, offset, whence); |
9453615a SW |
89 | break; |
90 | default: | |
91 | error = -EINVAL; | |
92 | } | |
b3b94faa DT |
93 | |
94 | return error; | |
95 | } | |
96 | ||
b3b94faa | 97 | /** |
d81a8ef5 | 98 | * gfs2_readdir - Iterator for a directory |
b3b94faa | 99 | * @file: The directory to read from |
d81a8ef5 | 100 | * @ctx: What to feed directory entries to |
b3b94faa DT |
101 | * |
102 | * Returns: errno | |
103 | */ | |
104 | ||
d81a8ef5 | 105 | static int gfs2_readdir(struct file *file, struct dir_context *ctx) |
b3b94faa | 106 | { |
71b86f56 | 107 | struct inode *dir = file->f_mapping->host; |
feaa7bba | 108 | struct gfs2_inode *dip = GFS2_I(dir); |
b3b94faa | 109 | struct gfs2_holder d_gh; |
b3b94faa DT |
110 | int error; |
111 | ||
d81a8ef5 AV |
112 | error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); |
113 | if (error) | |
b3b94faa | 114 | return error; |
b3b94faa | 115 | |
d81a8ef5 | 116 | error = gfs2_dir_read(dir, ctx, &file->f_ra); |
b3b94faa DT |
117 | |
118 | gfs2_glock_dq_uninit(&d_gh); | |
119 | ||
b3b94faa DT |
120 | return error; |
121 | } | |
122 | ||
128e5eba | 123 | /** |
b16f7e57 | 124 | * fsflag_gfs2flag |
128e5eba | 125 | * |
b16f7e57 AG |
126 | * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories, |
127 | * and to GFS2_DIF_JDATA for non-directories. | |
128e5eba | 128 | */ |
b16f7e57 AG |
129 | static struct { |
130 | u32 fsflag; | |
131 | u32 gfsflag; | |
132 | } fsflag_gfs2flag[] = { | |
133 | {FS_SYNC_FL, GFS2_DIF_SYNC}, | |
134 | {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE}, | |
135 | {FS_APPEND_FL, GFS2_DIF_APPENDONLY}, | |
136 | {FS_NOATIME_FL, GFS2_DIF_NOATIME}, | |
137 | {FS_INDEX_FL, GFS2_DIF_EXHASH}, | |
138 | {FS_TOPDIR_FL, GFS2_DIF_TOPDIR}, | |
139 | {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA}, | |
7ea9ea83 | 140 | }; |
71b86f56 | 141 | |
b09e593d | 142 | static int gfs2_get_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 143 | { |
496ad9aa | 144 | struct inode *inode = file_inode(filp); |
feaa7bba | 145 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 | 146 | struct gfs2_holder gh; |
b16f7e57 AG |
147 | int i, error; |
148 | u32 gfsflags, fsflags = 0; | |
71b86f56 | 149 | |
719ee344 SW |
150 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
151 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 152 | if (error) |
9c7fe835 | 153 | goto out_uninit; |
907b9bce | 154 | |
b16f7e57 AG |
155 | gfsflags = ip->i_diskflags; |
156 | if (S_ISDIR(inode->i_mode)) | |
157 | gfsflags &= ~GFS2_DIF_JDATA; | |
158 | else | |
159 | gfsflags &= ~GFS2_DIF_INHERIT_JDATA; | |
160 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) | |
161 | if (gfsflags & fsflag_gfs2flag[i].gfsflag) | |
162 | fsflags |= fsflag_gfs2flag[i].fsflag; | |
163 | ||
128e5eba | 164 | if (put_user(fsflags, ptr)) |
71b86f56 SW |
165 | error = -EFAULT; |
166 | ||
3cc3f710 | 167 | gfs2_glock_dq(&gh); |
9c7fe835 | 168 | out_uninit: |
71b86f56 SW |
169 | gfs2_holder_uninit(&gh); |
170 | return error; | |
171 | } | |
172 | ||
6b124d8d SW |
173 | void gfs2_set_inode_flags(struct inode *inode) |
174 | { | |
175 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
176 | unsigned int flags = inode->i_flags; |
177 | ||
9964afbb SW |
178 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
179 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 180 | flags |= S_NOSEC; |
383f01fb | 181 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 182 | flags |= S_IMMUTABLE; |
383f01fb | 183 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 184 | flags |= S_APPEND; |
383f01fb | 185 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 186 | flags |= S_NOATIME; |
383f01fb | 187 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
188 | flags |= S_SYNC; |
189 | inode->i_flags = flags; | |
190 | } | |
191 | ||
71b86f56 SW |
192 | /* Flags that can be set by user space */ |
193 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
194 | GFS2_DIF_IMMUTABLE| \ |
195 | GFS2_DIF_APPENDONLY| \ | |
196 | GFS2_DIF_NOATIME| \ | |
197 | GFS2_DIF_SYNC| \ | |
23d0bb83 | 198 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
199 | GFS2_DIF_INHERIT_JDATA) |
200 | ||
201 | /** | |
9dd868e1 FF |
202 | * do_gfs2_set_flags - set flags on an inode |
203 | * @filp: file pointer | |
204 | * @reqflags: The flags to set | |
71b86f56 SW |
205 | * @mask: Indicates which flags are valid |
206 | * | |
207 | */ | |
b09e593d | 208 | static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask) |
71b86f56 | 209 | { |
496ad9aa | 210 | struct inode *inode = file_inode(filp); |
feaa7bba SW |
211 | struct gfs2_inode *ip = GFS2_I(inode); |
212 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
213 | struct buffer_head *bh; |
214 | struct gfs2_holder gh; | |
215 | int error; | |
55eccc6d | 216 | u32 new_flags, flags; |
71b86f56 | 217 | |
a561be71 | 218 | error = mnt_want_write_file(filp); |
52f341cf | 219 | if (error) |
71b86f56 SW |
220 | return error; |
221 | ||
f58ba889 MS |
222 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
223 | if (error) | |
224 | goto out_drop_write; | |
225 | ||
7df0e039 | 226 | error = -EACCES; |
2e149670 | 227 | if (!inode_owner_or_capable(inode)) |
7df0e039 SW |
228 | goto out; |
229 | ||
230 | error = 0; | |
383f01fb | 231 | flags = ip->i_diskflags; |
55eccc6d | 232 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
233 | if ((new_flags ^ flags) == 0) |
234 | goto out; | |
235 | ||
71b86f56 SW |
236 | error = -EPERM; |
237 | if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE)) | |
238 | goto out; | |
239 | if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY)) | |
240 | goto out; | |
907b9bce | 241 | if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) && |
b9cb9813 | 242 | !capable(CAP_LINUX_IMMUTABLE)) |
71b86f56 | 243 | goto out; |
b9cb9813 | 244 | if (!IS_IMMUTABLE(inode)) { |
10556cb2 | 245 | error = gfs2_permission(inode, MAY_WRITE); |
b9cb9813 SW |
246 | if (error) |
247 | goto out; | |
248 | } | |
5561093e | 249 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
cc555b09 | 250 | if (new_flags & GFS2_DIF_JDATA) |
c1696fb8 | 251 | gfs2_log_flush(sdp, ip->i_gl, |
805c0907 BP |
252 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
253 | GFS2_LFC_SET_FLAGS); | |
5561093e SW |
254 | error = filemap_fdatawrite(inode->i_mapping); |
255 | if (error) | |
256 | goto out; | |
257 | error = filemap_fdatawait(inode->i_mapping); | |
258 | if (error) | |
259 | goto out; | |
cc555b09 BP |
260 | if (new_flags & GFS2_DIF_JDATA) |
261 | gfs2_ordered_del_inode(ip); | |
5561093e | 262 | } |
55eccc6d | 263 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
264 | if (error) |
265 | goto out; | |
55eccc6d SW |
266 | error = gfs2_meta_inode_buffer(ip, &bh); |
267 | if (error) | |
268 | goto out_trans_end; | |
9b7c2ddb | 269 | inode->i_ctime = current_time(inode); |
350a9b0a | 270 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 271 | ip->i_diskflags = new_flags; |
539e5d6b | 272 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 273 | brelse(bh); |
6b124d8d | 274 | gfs2_set_inode_flags(inode); |
5561093e | 275 | gfs2_set_aops(inode); |
55eccc6d SW |
276 | out_trans_end: |
277 | gfs2_trans_end(sdp); | |
71b86f56 SW |
278 | out: |
279 | gfs2_glock_dq_uninit(&gh); | |
f58ba889 | 280 | out_drop_write: |
2a79f17e | 281 | mnt_drop_write_file(filp); |
71b86f56 SW |
282 | return error; |
283 | } | |
284 | ||
b09e593d | 285 | static int gfs2_set_flags(struct file *filp, u32 __user *ptr) |
71b86f56 | 286 | { |
496ad9aa | 287 | struct inode *inode = file_inode(filp); |
b16f7e57 AG |
288 | u32 fsflags, gfsflags = 0; |
289 | u32 mask; | |
290 | int i; | |
7df0e039 | 291 | |
128e5eba | 292 | if (get_user(fsflags, ptr)) |
71b86f56 | 293 | return -EFAULT; |
7df0e039 | 294 | |
b16f7e57 AG |
295 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { |
296 | if (fsflags & fsflag_gfs2flag[i].fsflag) { | |
297 | fsflags &= ~fsflag_gfs2flag[i].fsflag; | |
298 | gfsflags |= fsflag_gfs2flag[i].gfsflag; | |
299 | } | |
300 | } | |
301 | if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) | |
302 | return -EINVAL; | |
303 | ||
304 | mask = GFS2_FLAGS_USER_SET; | |
305 | if (S_ISDIR(inode->i_mode)) { | |
306 | mask &= ~GFS2_DIF_JDATA; | |
307 | } else { | |
308 | /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ | |
309 | if (gfsflags & GFS2_DIF_TOPDIR) | |
310 | return -EINVAL; | |
311 | mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); | |
b9af7ca6 | 312 | } |
b16f7e57 AG |
313 | |
314 | return do_gfs2_set_flags(filp, gfsflags, mask); | |
71b86f56 SW |
315 | } |
316 | ||
b09e593d | 317 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
318 | { |
319 | switch(cmd) { | |
128e5eba | 320 | case FS_IOC_GETFLAGS: |
b09e593d | 321 | return gfs2_get_flags(filp, (u32 __user *)arg); |
128e5eba | 322 | case FS_IOC_SETFLAGS: |
b09e593d | 323 | return gfs2_set_flags(filp, (u32 __user *)arg); |
66fc061b SW |
324 | case FITRIM: |
325 | return gfs2_fitrim(filp, (void __user *)arg); | |
71b86f56 SW |
326 | } |
327 | return -ENOTTY; | |
328 | } | |
329 | ||
da1dfb6a SW |
330 | /** |
331 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 332 | * @filep: The struct file |
da1dfb6a SW |
333 | * @offset: The file offset of the write |
334 | * @size: The length of the write | |
335 | * | |
336 | * When we are about to do a write, this function records the total | |
337 | * write size in order to provide a suitable hint to the lower layers | |
338 | * about how many blocks will be required. | |
339 | * | |
340 | */ | |
341 | ||
342 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
343 | { | |
496ad9aa | 344 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
345 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
346 | struct gfs2_inode *ip = GFS2_I(inode); | |
347 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
348 | int hint = min_t(size_t, INT_MAX, blks); | |
349 | ||
a097dc7e BP |
350 | if (hint > atomic_read(&ip->i_res.rs_sizehint)) |
351 | atomic_set(&ip->i_res.rs_sizehint, hint); | |
da1dfb6a SW |
352 | } |
353 | ||
3cc3f710 SW |
354 | /** |
355 | * gfs2_allocate_page_backing - Use bmap to allocate blocks | |
356 | * @page: The (locked) page to allocate backing for | |
357 | * | |
358 | * We try to allocate all the blocks required for the page in | |
359 | * one go. This might fail for various reasons, so we keep | |
360 | * trying until all the blocks to back this page are allocated. | |
361 | * If some of the blocks are already allocated, thats ok too. | |
362 | */ | |
363 | ||
364 | static int gfs2_allocate_page_backing(struct page *page) | |
365 | { | |
366 | struct inode *inode = page->mapping->host; | |
367 | struct buffer_head bh; | |
09cbfeaf KS |
368 | unsigned long size = PAGE_SIZE; |
369 | u64 lblock = page->index << (PAGE_SHIFT - inode->i_blkbits); | |
3cc3f710 SW |
370 | |
371 | do { | |
372 | bh.b_state = 0; | |
373 | bh.b_size = size; | |
e9e1ef2b | 374 | gfs2_block_map(inode, lblock, &bh, 1); |
3cc3f710 SW |
375 | if (!buffer_mapped(&bh)) |
376 | return -EIO; | |
377 | size -= bh.b_size; | |
378 | lblock += (bh.b_size >> inode->i_blkbits); | |
379 | } while(size > 0); | |
380 | return 0; | |
381 | } | |
382 | ||
383 | /** | |
384 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
385 | * @vma: The virtual memory area | |
9dd868e1 | 386 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
387 | * |
388 | * When the page becomes writable, we need to ensure that we have | |
389 | * blocks allocated on disk to back that page. | |
390 | */ | |
391 | ||
109dbb1e | 392 | static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) |
3cc3f710 | 393 | { |
c2ec175c | 394 | struct page *page = vmf->page; |
11bac800 | 395 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3cc3f710 SW |
396 | struct gfs2_inode *ip = GFS2_I(inode); |
397 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 398 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
3cc3f710 | 399 | unsigned long last_index; |
09cbfeaf | 400 | u64 pos = page->index << PAGE_SHIFT; |
3cc3f710 | 401 | unsigned int data_blocks, ind_blocks, rblocks; |
3cc3f710 | 402 | struct gfs2_holder gh; |
13d921e3 | 403 | loff_t size; |
3cc3f710 SW |
404 | int ret; |
405 | ||
39263d5e | 406 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 407 | |
b54e9a0b | 408 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 | 409 | if (ret) |
a097dc7e | 410 | goto out; |
0a305e49 | 411 | |
11bac800 | 412 | gfs2_size_hint(vmf->vma->vm_file, pos, PAGE_SIZE); |
8e2e0047 | 413 | |
719ee344 SW |
414 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
415 | ret = gfs2_glock_nq(&gh); | |
3cc3f710 | 416 | if (ret) |
2b3dcf35 | 417 | goto out_uninit; |
3cc3f710 | 418 | |
d7c436cd | 419 | /* Update file times before taking page lock */ |
11bac800 | 420 | file_update_time(vmf->vma->vm_file); |
d7c436cd | 421 | |
9c538837 SW |
422 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
423 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
424 | ||
09cbfeaf | 425 | if (!gfs2_write_alloc_required(ip, pos, PAGE_SIZE)) { |
13d921e3 SW |
426 | lock_page(page); |
427 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
428 | ret = -EAGAIN; | |
429 | unlock_page(page); | |
430 | } | |
3cc3f710 | 431 | goto out_unlock; |
13d921e3 SW |
432 | } |
433 | ||
5407e242 BP |
434 | ret = gfs2_rindex_update(sdp); |
435 | if (ret) | |
6dbd8224 SW |
436 | goto out_unlock; |
437 | ||
09cbfeaf | 438 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
7b9cff46 | 439 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
440 | ret = gfs2_quota_lock_check(ip, &ap); |
441 | if (ret) | |
442 | goto out_unlock; | |
7b9cff46 | 443 | ret = gfs2_inplace_reserve(ip, &ap); |
3cc3f710 SW |
444 | if (ret) |
445 | goto out_quota_unlock; | |
446 | ||
447 | rblocks = RES_DINODE + ind_blocks; | |
448 | if (gfs2_is_jdata(ip)) | |
449 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 450 | if (ind_blocks || data_blocks) { |
3cc3f710 | 451 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 452 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 453 | } |
3cc3f710 SW |
454 | ret = gfs2_trans_begin(sdp, rblocks, 0); |
455 | if (ret) | |
456 | goto out_trans_fail; | |
457 | ||
458 | lock_page(page); | |
459 | ret = -EINVAL; | |
13d921e3 | 460 | size = i_size_read(inode); |
09cbfeaf | 461 | last_index = (size - 1) >> PAGE_SHIFT; |
13d921e3 SW |
462 | /* Check page index against inode size */ |
463 | if (size == 0 || (page->index > last_index)) | |
464 | goto out_trans_end; | |
465 | ||
466 | ret = -EAGAIN; | |
467 | /* If truncated, we must retry the operation, we may have raced | |
468 | * with the glock demotion code. | |
469 | */ | |
470 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) | |
471 | goto out_trans_end; | |
472 | ||
473 | /* Unstuff, if required, and allocate backing blocks for page */ | |
b7fe2e39 | 474 | ret = 0; |
13d921e3 | 475 | if (gfs2_is_stuffed(ip)) |
3cc3f710 | 476 | ret = gfs2_unstuff_dinode(ip, page); |
13d921e3 SW |
477 | if (ret == 0) |
478 | ret = gfs2_allocate_page_backing(page); | |
3cc3f710 | 479 | |
13d921e3 SW |
480 | out_trans_end: |
481 | if (ret) | |
482 | unlock_page(page); | |
3cc3f710 SW |
483 | gfs2_trans_end(sdp); |
484 | out_trans_fail: | |
485 | gfs2_inplace_release(ip); | |
486 | out_quota_unlock: | |
487 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
488 | out_unlock: |
489 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 490 | out_uninit: |
3cc3f710 | 491 | gfs2_holder_uninit(&gh); |
13d921e3 SW |
492 | if (ret == 0) { |
493 | set_page_dirty(page); | |
1d1d1a76 | 494 | wait_for_stable_page(page); |
13d921e3 | 495 | } |
2b3dcf35 | 496 | out: |
39263d5e | 497 | sb_end_pagefault(inode->i_sb); |
13d921e3 | 498 | return block_page_mkwrite_return(ret); |
3cc3f710 SW |
499 | } |
500 | ||
f0f37e2f | 501 | static const struct vm_operations_struct gfs2_vm_ops = { |
3cc3f710 | 502 | .fault = filemap_fault, |
f1820361 | 503 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
504 | .page_mkwrite = gfs2_page_mkwrite, |
505 | }; | |
506 | ||
b3b94faa DT |
507 | /** |
508 | * gfs2_mmap - | |
509 | * @file: The file to map | |
510 | * @vma: The VMA which described the mapping | |
511 | * | |
48bf2b17 SW |
512 | * There is no need to get a lock here unless we should be updating |
513 | * atime. We ignore any locking errors since the only consequence is | |
514 | * a missed atime update (which will just be deferred until later). | |
515 | * | |
516 | * Returns: 0 | |
b3b94faa DT |
517 | */ |
518 | ||
519 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
520 | { | |
feaa7bba | 521 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 522 | |
b9c93bb7 SW |
523 | if (!(file->f_flags & O_NOATIME) && |
524 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
525 | struct gfs2_holder i_gh; |
526 | int error; | |
b3b94faa | 527 | |
3d162688 BM |
528 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
529 | &i_gh); | |
b9c93bb7 SW |
530 | if (error) |
531 | return error; | |
3d162688 BM |
532 | /* grab lock to update inode */ |
533 | gfs2_glock_dq_uninit(&i_gh); | |
534 | file_accessed(file); | |
48bf2b17 | 535 | } |
3cc3f710 | 536 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 537 | |
48bf2b17 | 538 | return 0; |
b3b94faa DT |
539 | } |
540 | ||
541 | /** | |
6d4ade98 SW |
542 | * gfs2_open_common - This is common to open and atomic_open |
543 | * @inode: The inode being opened | |
544 | * @file: The file being opened | |
b3b94faa | 545 | * |
6d4ade98 SW |
546 | * This maybe called under a glock or not depending upon how it has |
547 | * been called. We must always be called under a glock for regular | |
548 | * files, however. For other file types, it does not matter whether | |
549 | * we hold the glock or not. | |
550 | * | |
551 | * Returns: Error code or 0 for success | |
b3b94faa DT |
552 | */ |
553 | ||
6d4ade98 | 554 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 555 | { |
b3b94faa | 556 | struct gfs2_file *fp; |
6d4ade98 SW |
557 | int ret; |
558 | ||
559 | if (S_ISREG(inode->i_mode)) { | |
560 | ret = generic_file_open(inode, file); | |
561 | if (ret) | |
562 | return ret; | |
563 | } | |
b3b94faa | 564 | |
6d4ade98 | 565 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
566 | if (!fp) |
567 | return -ENOMEM; | |
568 | ||
f55ab26a | 569 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 570 | |
feaa7bba | 571 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 572 | file->private_data = fp; |
6d4ade98 SW |
573 | return 0; |
574 | } | |
575 | ||
576 | /** | |
577 | * gfs2_open - open a file | |
578 | * @inode: the inode to open | |
579 | * @file: the struct file for this opening | |
580 | * | |
581 | * After atomic_open, this function is only used for opening files | |
582 | * which are already cached. We must still get the glock for regular | |
583 | * files to ensure that we have the file size uptodate for the large | |
584 | * file check which is in the common code. That is only an issue for | |
585 | * regular files though. | |
586 | * | |
587 | * Returns: errno | |
588 | */ | |
589 | ||
590 | static int gfs2_open(struct inode *inode, struct file *file) | |
591 | { | |
592 | struct gfs2_inode *ip = GFS2_I(inode); | |
593 | struct gfs2_holder i_gh; | |
594 | int error; | |
595 | bool need_unlock = false; | |
b3b94faa | 596 | |
b60623c2 | 597 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
598 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
599 | &i_gh); | |
600 | if (error) | |
6d4ade98 SW |
601 | return error; |
602 | need_unlock = true; | |
603 | } | |
b3b94faa | 604 | |
6d4ade98 | 605 | error = gfs2_open_common(inode, file); |
b3b94faa | 606 | |
6d4ade98 | 607 | if (need_unlock) |
b3b94faa | 608 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 609 | |
b3b94faa DT |
610 | return error; |
611 | } | |
612 | ||
613 | /** | |
df3fd117 | 614 | * gfs2_release - called to close a struct file |
b3b94faa DT |
615 | * @inode: the inode the struct file belongs to |
616 | * @file: the struct file being closed | |
617 | * | |
618 | * Returns: errno | |
619 | */ | |
620 | ||
df3fd117 | 621 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 622 | { |
0a305e49 | 623 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 624 | |
8e2e0047 | 625 | kfree(file->private_data); |
5c676f6d | 626 | file->private_data = NULL; |
b3b94faa | 627 | |
2b3dcf35 BP |
628 | if (!(file->f_mode & FMODE_WRITE)) |
629 | return 0; | |
0a305e49 | 630 | |
b54e9a0b | 631 | gfs2_rsqa_delete(ip, &inode->i_writecount); |
b3b94faa DT |
632 | return 0; |
633 | } | |
634 | ||
635 | /** | |
636 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
637 | * @file: the file that points to the dentry |
638 | * @start: the start position in the file to sync | |
639 | * @end: the end position in the file to sync | |
dba898b0 | 640 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 641 | * |
2f0264d5 SW |
642 | * We split the data flushing here so that we don't wait for the data |
643 | * until after we've also sent the metadata to disk. Note that for | |
644 | * data=ordered, we will write & wait for the data at the log flush | |
645 | * stage anyway, so this is unlikely to make much of a difference | |
646 | * except in the data=writeback case. | |
647 | * | |
648 | * If the fdatawrite fails due to any reason except -EIO, we will | |
649 | * continue the remainder of the fsync, although we'll still report | |
650 | * the error at the end. This is to match filemap_write_and_wait_range() | |
651 | * behaviour. | |
34126f9f | 652 | * |
b3b94faa DT |
653 | * Returns: errno |
654 | */ | |
655 | ||
02c24a82 JB |
656 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
657 | int datasync) | |
b3b94faa | 658 | { |
2f0264d5 SW |
659 | struct address_space *mapping = file->f_mapping; |
660 | struct inode *inode = mapping->host; | |
0ae45f63 | 661 | int sync_state = inode->i_state & I_DIRTY_ALL; |
dba898b0 | 662 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 663 | int ret = 0, ret1 = 0; |
b3b94faa | 664 | |
2f0264d5 SW |
665 | if (mapping->nrpages) { |
666 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
667 | if (ret1 == -EIO) | |
668 | return ret1; | |
669 | } | |
02c24a82 | 670 | |
0c901809 BM |
671 | if (!gfs2_is_jdata(ip)) |
672 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 673 | if (datasync) |
0ae45f63 | 674 | sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME); |
b3b94faa | 675 | |
dba898b0 SW |
676 | if (sync_state) { |
677 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 678 | if (ret) |
dba898b0 | 679 | return ret; |
f1818529 | 680 | if (gfs2_is_jdata(ip)) |
d07a6ac7 JL |
681 | ret = file_write_and_wait(file); |
682 | if (ret) | |
683 | return ret; | |
b5b24d7a | 684 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
685 | } |
686 | ||
2f0264d5 | 687 | if (mapping->nrpages) |
d07a6ac7 | 688 | ret = file_fdatawait_range(file, start, end); |
2f0264d5 SW |
689 | |
690 | return ret ? ret : ret1; | |
b3b94faa DT |
691 | } |
692 | ||
967bcc91 AG |
693 | static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to) |
694 | { | |
695 | struct file *file = iocb->ki_filp; | |
696 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); | |
697 | size_t count = iov_iter_count(to); | |
698 | struct gfs2_holder gh; | |
699 | ssize_t ret; | |
700 | ||
701 | if (!count) | |
702 | return 0; /* skip atime */ | |
703 | ||
704 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); | |
705 | ret = gfs2_glock_nq(&gh); | |
706 | if (ret) | |
707 | goto out_uninit; | |
708 | ||
967bcc91 AG |
709 | ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL); |
710 | ||
967bcc91 AG |
711 | gfs2_glock_dq(&gh); |
712 | out_uninit: | |
713 | gfs2_holder_uninit(&gh); | |
714 | return ret; | |
715 | } | |
716 | ||
717 | static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from) | |
718 | { | |
719 | struct file *file = iocb->ki_filp; | |
720 | struct inode *inode = file->f_mapping->host; | |
721 | struct gfs2_inode *ip = GFS2_I(inode); | |
722 | size_t len = iov_iter_count(from); | |
723 | loff_t offset = iocb->ki_pos; | |
724 | struct gfs2_holder gh; | |
725 | ssize_t ret; | |
726 | ||
727 | /* | |
728 | * Deferred lock, even if its a write, since we do no allocation on | |
729 | * this path. All we need to change is the atime, and this lock mode | |
730 | * ensures that other nodes have flushed their buffered read caches | |
731 | * (i.e. their page cache entries for this inode). We do not, | |
732 | * unfortunately, have the option of only flushing a range like the | |
733 | * VFS does. | |
734 | */ | |
735 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); | |
736 | ret = gfs2_glock_nq(&gh); | |
737 | if (ret) | |
738 | goto out_uninit; | |
739 | ||
740 | /* Silently fall back to buffered I/O when writing beyond EOF */ | |
741 | if (offset + len > i_size_read(&ip->i_inode)) | |
742 | goto out; | |
743 | ||
744 | ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL); | |
745 | ||
746 | out: | |
747 | gfs2_glock_dq(&gh); | |
748 | out_uninit: | |
749 | gfs2_holder_uninit(&gh); | |
750 | return ret; | |
751 | } | |
752 | ||
753 | static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
754 | { | |
755 | ssize_t ret; | |
756 | ||
757 | if (iocb->ki_flags & IOCB_DIRECT) { | |
758 | ret = gfs2_file_direct_read(iocb, to); | |
759 | if (likely(ret != -ENOTBLK)) | |
760 | return ret; | |
761 | iocb->ki_flags &= ~IOCB_DIRECT; | |
762 | } | |
763 | return generic_file_read_iter(iocb, to); | |
764 | } | |
765 | ||
56aa616a | 766 | /** |
da56e45b | 767 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a | 768 | * @iocb: The io context |
64bc06bb | 769 | * @from: The data to write |
56aa616a SW |
770 | * |
771 | * We have to do a lock/unlock here to refresh the inode size for | |
772 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
773 | * offset. There is still a race, but provided the app is using its | |
774 | * own file locking, this will make O_APPEND work as expected. | |
775 | * | |
776 | */ | |
777 | ||
da56e45b | 778 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
779 | { |
780 | struct file *file = iocb->ki_filp; | |
64bc06bb AG |
781 | struct inode *inode = file_inode(file); |
782 | struct gfs2_inode *ip = GFS2_I(inode); | |
967bcc91 | 783 | ssize_t written = 0, ret; |
0a305e49 | 784 | |
b54e9a0b | 785 | ret = gfs2_rsqa_alloc(ip); |
0a305e49 BP |
786 | if (ret) |
787 | return ret; | |
56aa616a | 788 | |
da56e45b | 789 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 790 | |
2ba48ce5 | 791 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a | 792 | struct gfs2_holder gh; |
56aa616a SW |
793 | |
794 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
795 | if (ret) | |
796 | return ret; | |
797 | gfs2_glock_dq_uninit(&gh); | |
798 | } | |
799 | ||
64bc06bb AG |
800 | inode_lock(inode); |
801 | ret = generic_write_checks(iocb, from); | |
802 | if (ret <= 0) | |
803 | goto out; | |
804 | ||
805 | /* We can write back this queue in page reclaim */ | |
806 | current->backing_dev_info = inode_to_bdi(inode); | |
807 | ||
808 | ret = file_remove_privs(file); | |
809 | if (ret) | |
810 | goto out2; | |
811 | ||
812 | ret = file_update_time(file); | |
813 | if (ret) | |
814 | goto out2; | |
815 | ||
967bcc91 AG |
816 | if (iocb->ki_flags & IOCB_DIRECT) { |
817 | struct address_space *mapping = file->f_mapping; | |
818 | loff_t pos, endbyte; | |
819 | ssize_t buffered; | |
820 | ||
821 | written = gfs2_file_direct_write(iocb, from); | |
822 | if (written < 0 || !iov_iter_count(from)) | |
823 | goto out2; | |
824 | ||
825 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); | |
826 | if (unlikely(ret < 0)) | |
827 | goto out2; | |
828 | buffered = ret; | |
829 | ||
830 | /* | |
831 | * We need to ensure that the page cache pages are written to | |
832 | * disk and invalidated to preserve the expected O_DIRECT | |
833 | * semantics. | |
834 | */ | |
835 | pos = iocb->ki_pos; | |
836 | endbyte = pos + buffered - 1; | |
837 | ret = filemap_write_and_wait_range(mapping, pos, endbyte); | |
838 | if (!ret) { | |
839 | iocb->ki_pos += buffered; | |
840 | written += buffered; | |
841 | invalidate_mapping_pages(mapping, | |
842 | pos >> PAGE_SHIFT, | |
843 | endbyte >> PAGE_SHIFT); | |
844 | } else { | |
845 | /* | |
846 | * We don't know how much we wrote, so just return | |
847 | * the number of bytes which were direct-written | |
848 | */ | |
849 | } | |
850 | } else { | |
851 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); | |
852 | if (likely(ret > 0)) | |
853 | iocb->ki_pos += ret; | |
854 | } | |
64bc06bb AG |
855 | |
856 | out2: | |
857 | current->backing_dev_info = NULL; | |
858 | out: | |
859 | inode_unlock(inode); | |
860 | if (likely(ret > 0)) { | |
64bc06bb AG |
861 | /* Handle various SYNC-type writes */ |
862 | ret = generic_write_sync(iocb, ret); | |
863 | } | |
967bcc91 | 864 | return written ? written : ret; |
56aa616a SW |
865 | } |
866 | ||
2fe17c10 CH |
867 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
868 | int mode) | |
869 | { | |
fffb6412 | 870 | struct super_block *sb = inode->i_sb; |
2fe17c10 | 871 | struct gfs2_inode *ip = GFS2_I(inode); |
fffb6412 | 872 | loff_t end = offset + len; |
2fe17c10 CH |
873 | struct buffer_head *dibh; |
874 | int error; | |
2fe17c10 CH |
875 | |
876 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
877 | if (unlikely(error)) | |
64dd153c | 878 | return error; |
2fe17c10 | 879 | |
350a9b0a | 880 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
881 | |
882 | if (gfs2_is_stuffed(ip)) { | |
883 | error = gfs2_unstuff_dinode(ip, NULL); | |
884 | if (unlikely(error)) | |
885 | goto out; | |
886 | } | |
887 | ||
fffb6412 | 888 | while (offset < end) { |
c2589282 AG |
889 | struct iomap iomap = { }; |
890 | ||
628e366d AG |
891 | error = gfs2_iomap_get_alloc(inode, offset, end - offset, |
892 | &iomap); | |
fffb6412 | 893 | if (error) |
64dd153c | 894 | goto out; |
fffb6412 | 895 | offset = iomap.offset + iomap.length; |
d505a96a | 896 | if (!(iomap.flags & IOMAP_F_NEW)) |
64dd153c | 897 | continue; |
fffb6412 AG |
898 | error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, |
899 | iomap.length >> inode->i_blkbits, | |
900 | GFP_NOFS); | |
901 | if (error) { | |
902 | fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); | |
2fe17c10 | 903 | goto out; |
64dd153c | 904 | } |
2fe17c10 | 905 | } |
2fe17c10 | 906 | out: |
64dd153c | 907 | brelse(dibh); |
2fe17c10 CH |
908 | return error; |
909 | } | |
d9be0cda AD |
910 | /** |
911 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
912 | * blocks, determine how many bytes can be written. | |
913 | * @ip: The inode in question. | |
914 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
915 | * @data_blocks: Compute and return the number of data blocks needed | |
916 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
917 | * @max_blocks: The total blocks available to work with. | |
918 | * | |
919 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
920 | */ | |
921 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
922 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
923 | unsigned int max_blocks) | |
2fe17c10 | 924 | { |
d9be0cda | 925 | loff_t max = *len; |
2fe17c10 | 926 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
927 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
928 | ||
929 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
930 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
931 | max_data -= tmp; | |
932 | } | |
d9be0cda | 933 | |
2fe17c10 CH |
934 | *data_blocks = max_data; |
935 | *ind_blocks = max_blocks - max_data; | |
936 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
937 | if (*len > max) { | |
938 | *len = max; | |
939 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
940 | } | |
941 | } | |
942 | ||
9c9f1159 | 943 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 944 | { |
496ad9aa | 945 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
946 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
947 | struct gfs2_inode *ip = GFS2_I(inode); | |
7b9cff46 | 948 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
2fe17c10 | 949 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
174d1232 | 950 | loff_t bytes, max_bytes, max_blks; |
2fe17c10 | 951 | int error; |
4442f2e0 SW |
952 | const loff_t pos = offset; |
953 | const loff_t count = len; | |
6905d9e4 | 954 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 955 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 956 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 957 | |
2fe17c10 CH |
958 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
959 | ||
6905d9e4 | 960 | offset &= bsize_mask; |
2fe17c10 CH |
961 | |
962 | len = next - offset; | |
963 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
964 | if (!bytes) | |
965 | bytes = UINT_MAX; | |
6905d9e4 BM |
966 | bytes &= bsize_mask; |
967 | if (bytes == 0) | |
968 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 969 | |
da1dfb6a | 970 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 971 | |
d9be0cda AD |
972 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
973 | ap.min_target = data_blocks + ind_blocks; | |
974 | ||
2fe17c10 CH |
975 | while (len > 0) { |
976 | if (len < bytes) | |
977 | bytes = len; | |
58a7d5fb BM |
978 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
979 | len -= bytes; | |
980 | offset += bytes; | |
981 | continue; | |
982 | } | |
d9be0cda AD |
983 | |
984 | /* We need to determine how many bytes we can actually | |
985 | * fallocate without exceeding quota or going over the | |
986 | * end of the fs. We start off optimistically by assuming | |
987 | * we can write max_bytes */ | |
988 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
989 | ||
990 | /* Since max_bytes is most likely a theoretical max, we | |
991 | * calculate a more realistic 'bytes' to serve as a good | |
992 | * starting point for the number of bytes we may be able | |
993 | * to write */ | |
2fe17c10 | 994 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 995 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
996 | |
997 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 998 | if (error) |
9c9f1159 | 999 | return error; |
d9be0cda AD |
1000 | /* ap.allowed tells us how many blocks quota will allow |
1001 | * us to write. Check if this reduces max_blks */ | |
174d1232 AG |
1002 | max_blks = UINT_MAX; |
1003 | if (ap.allowed) | |
d9be0cda | 1004 | max_blks = ap.allowed; |
2fe17c10 | 1005 | |
7b9cff46 | 1006 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 1007 | if (error) |
2fe17c10 | 1008 | goto out_qunlock; |
d9be0cda AD |
1009 | |
1010 | /* check if the selected rgrp limits our max_blks further */ | |
1011 | if (ap.allowed && ap.allowed < max_blks) | |
1012 | max_blks = ap.allowed; | |
1013 | ||
1014 | /* Almost done. Calculate bytes that can be written using | |
1015 | * max_blks. We also recompute max_bytes, data_blocks and | |
1016 | * ind_blocks */ | |
1017 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
1018 | &ind_blocks, max_blks); | |
2fe17c10 CH |
1019 | |
1020 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 1021 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
1022 | if (gfs2_is_jdata(ip)) |
1023 | rblocks += data_blocks ? data_blocks : 1; | |
1024 | ||
1025 | error = gfs2_trans_begin(sdp, rblocks, | |
09cbfeaf | 1026 | PAGE_SIZE/sdp->sd_sb.sb_bsize); |
2fe17c10 CH |
1027 | if (error) |
1028 | goto out_trans_fail; | |
1029 | ||
1030 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
1031 | gfs2_trans_end(sdp); | |
1032 | ||
1033 | if (error) | |
1034 | goto out_trans_fail; | |
1035 | ||
1036 | len -= max_bytes; | |
1037 | offset += max_bytes; | |
1038 | gfs2_inplace_release(ip); | |
1039 | gfs2_quota_unlock(ip); | |
2fe17c10 | 1040 | } |
4442f2e0 | 1041 | |
1885867b AP |
1042 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) { |
1043 | i_size_write(inode, pos + count); | |
98f1a696 | 1044 | file_update_time(file); |
4b813f09 | 1045 | mark_inode_dirty(inode); |
1885867b AP |
1046 | } |
1047 | ||
dde0c2e7 CH |
1048 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
1049 | return vfs_fsync_range(file, pos, pos + count - 1, | |
1050 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
1051 | return 0; | |
2fe17c10 CH |
1052 | |
1053 | out_trans_fail: | |
1054 | gfs2_inplace_release(ip); | |
1055 | out_qunlock: | |
1056 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
1057 | return error; |
1058 | } | |
1059 | ||
1060 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
1061 | { | |
1062 | struct inode *inode = file_inode(file); | |
d4d7fc12 | 1063 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9c9f1159 AP |
1064 | struct gfs2_inode *ip = GFS2_I(inode); |
1065 | struct gfs2_holder gh; | |
1066 | int ret; | |
1067 | ||
4e56a641 | 1068 | if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) |
d4d7fc12 AP |
1069 | return -EOPNOTSUPP; |
1070 | /* fallocate is needed by gfs2_grow to reserve space in the rindex */ | |
1071 | if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) | |
9c9f1159 AP |
1072 | return -EOPNOTSUPP; |
1073 | ||
5955102c | 1074 | inode_lock(inode); |
9c9f1159 AP |
1075 | |
1076 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1077 | ret = gfs2_glock_nq(&gh); | |
1078 | if (ret) | |
1079 | goto out_uninit; | |
1080 | ||
1081 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
1082 | (offset + len) > inode->i_size) { | |
1083 | ret = inode_newsize_ok(inode, offset + len); | |
1084 | if (ret) | |
1085 | goto out_unlock; | |
1086 | } | |
1087 | ||
1088 | ret = get_write_access(inode); | |
1089 | if (ret) | |
1090 | goto out_unlock; | |
1091 | ||
4e56a641 AG |
1092 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
1093 | ret = __gfs2_punch_hole(file, offset, len); | |
1094 | } else { | |
1095 | ret = gfs2_rsqa_alloc(ip); | |
1096 | if (ret) | |
1097 | goto out_putw; | |
9c9f1159 | 1098 | |
4e56a641 AG |
1099 | ret = __gfs2_fallocate(file, mode, offset, len); |
1100 | ||
1101 | if (ret) | |
1102 | gfs2_rs_deltree(&ip->i_res); | |
1103 | } | |
a097dc7e | 1104 | |
9c9f1159 AP |
1105 | out_putw: |
1106 | put_write_access(inode); | |
2fe17c10 | 1107 | out_unlock: |
a0846a53 | 1108 | gfs2_glock_dq(&gh); |
2fe17c10 | 1109 | out_uninit: |
a0846a53 | 1110 | gfs2_holder_uninit(&gh); |
5955102c | 1111 | inode_unlock(inode); |
9c9f1159 | 1112 | return ret; |
2fe17c10 CH |
1113 | } |
1114 | ||
f1ea6f4e BP |
1115 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
1116 | struct file *out, loff_t *ppos, | |
1117 | size_t len, unsigned int flags) | |
1118 | { | |
1119 | int error; | |
1120 | struct gfs2_inode *ip = GFS2_I(out->f_mapping->host); | |
1121 | ||
b54e9a0b | 1122 | error = gfs2_rsqa_alloc(ip); |
f1ea6f4e BP |
1123 | if (error) |
1124 | return (ssize_t)error; | |
1125 | ||
1126 | gfs2_size_hint(out, *ppos, len); | |
1127 | ||
1128 | return iter_file_splice_write(pipe, out, ppos, len, flags); | |
1129 | } | |
1130 | ||
f057f6cd SW |
1131 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
1132 | ||
b3b94faa DT |
1133 | /** |
1134 | * gfs2_lock - acquire/release a posix lock on a file | |
1135 | * @file: the file pointer | |
1136 | * @cmd: either modify or retrieve lock state, possibly wait | |
1137 | * @fl: type and range of lock | |
1138 | * | |
1139 | * Returns: errno | |
1140 | */ | |
1141 | ||
1142 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1143 | { | |
feaa7bba SW |
1144 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1145 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1146 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1147 | |
b3b94faa DT |
1148 | if (!(fl->fl_flags & FL_POSIX)) |
1149 | return -ENOLCK; | |
720e7749 | 1150 | if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK) |
b3b94faa DT |
1151 | return -ENOLCK; |
1152 | ||
586759f0 ME |
1153 | if (cmd == F_CANCELLK) { |
1154 | /* Hack: */ | |
1155 | cmd = F_SETLK; | |
1156 | fl->fl_type = F_UNLCK; | |
1157 | } | |
c2952d20 SW |
1158 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { |
1159 | if (fl->fl_type == F_UNLCK) | |
4f656367 | 1160 | locks_lock_file_wait(file, fl); |
f057f6cd | 1161 | return -EIO; |
c2952d20 | 1162 | } |
b3b94faa | 1163 | if (IS_GETLK(cmd)) |
f057f6cd | 1164 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1165 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1166 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1167 | else |
f057f6cd | 1168 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1169 | } |
1170 | ||
b3b94faa DT |
1171 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1172 | { | |
5c676f6d | 1173 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1174 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1175 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1176 | struct gfs2_glock *gl; |
1177 | unsigned int state; | |
b58bf407 | 1178 | u16 flags; |
b3b94faa | 1179 | int error = 0; |
2ddfbdd6 | 1180 | int sleeptime; |
b3b94faa DT |
1181 | |
1182 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
2ddfbdd6 | 1183 | flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT; |
b3b94faa | 1184 | |
f55ab26a | 1185 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa | 1186 | |
283c9a97 | 1187 | if (gfs2_holder_initialized(fl_gh)) { |
b3b94faa DT |
1188 | if (fl_gh->gh_state == state) |
1189 | goto out; | |
4f656367 | 1190 | locks_lock_file_wait(file, |
a93a9983 JB |
1191 | &(struct file_lock) { |
1192 | .fl_type = F_UNLCK, | |
1193 | .fl_flags = FL_FLOCK | |
1194 | }); | |
5bef3e7c | 1195 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1196 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1197 | } else { |
6802e340 SW |
1198 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1199 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1200 | if (error) |
1201 | goto out; | |
b4c20166 AD |
1202 | gfs2_holder_init(gl, state, flags, fl_gh); |
1203 | gfs2_glock_put(gl); | |
b3b94faa | 1204 | } |
2ddfbdd6 BP |
1205 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1206 | error = gfs2_glock_nq(fl_gh); | |
1207 | if (error != GLR_TRYFAILED) | |
1208 | break; | |
1209 | fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT; | |
1210 | fl_gh->gh_error = 0; | |
1211 | msleep(sleeptime); | |
1212 | } | |
b3b94faa DT |
1213 | if (error) { |
1214 | gfs2_holder_uninit(fl_gh); | |
1215 | if (error == GLR_TRYFAILED) | |
1216 | error = -EAGAIN; | |
1217 | } else { | |
4f656367 | 1218 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1219 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1220 | } |
1221 | ||
420b9e5e | 1222 | out: |
f55ab26a | 1223 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1224 | return error; |
1225 | } | |
1226 | ||
1227 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1228 | { | |
5c676f6d | 1229 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1230 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1231 | ||
f55ab26a | 1232 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1233 | locks_lock_file_wait(file, fl); |
6df9f9a2 | 1234 | if (gfs2_holder_initialized(fl_gh)) { |
2ddfbdd6 | 1235 | gfs2_glock_dq(fl_gh); |
0a33443b SW |
1236 | gfs2_holder_uninit(fl_gh); |
1237 | } | |
f55ab26a | 1238 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1239 | } |
1240 | ||
1241 | /** | |
1242 | * gfs2_flock - acquire/release a flock lock on a file | |
1243 | * @file: the file pointer | |
1244 | * @cmd: either modify or retrieve lock state, possibly wait | |
1245 | * @fl: type and range of lock | |
1246 | * | |
1247 | * Returns: errno | |
1248 | */ | |
1249 | ||
1250 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1251 | { | |
b3b94faa DT |
1252 | if (!(fl->fl_flags & FL_FLOCK)) |
1253 | return -ENOLCK; | |
a12af1eb AD |
1254 | if (fl->fl_type & LOCK_MAND) |
1255 | return -EOPNOTSUPP; | |
b3b94faa | 1256 | |
b3b94faa DT |
1257 | if (fl->fl_type == F_UNLCK) { |
1258 | do_unflock(file, fl); | |
1259 | return 0; | |
d00223f1 | 1260 | } else { |
b3b94faa | 1261 | return do_flock(file, cmd, fl); |
d00223f1 | 1262 | } |
b3b94faa DT |
1263 | } |
1264 | ||
10d21988 | 1265 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1266 | .llseek = gfs2_llseek, |
967bcc91 | 1267 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1268 | .write_iter = gfs2_file_write_iter, |
26c1a574 SW |
1269 | .unlocked_ioctl = gfs2_ioctl, |
1270 | .mmap = gfs2_mmap, | |
1271 | .open = gfs2_open, | |
df3fd117 | 1272 | .release = gfs2_release, |
26c1a574 SW |
1273 | .fsync = gfs2_fsync, |
1274 | .lock = gfs2_lock, | |
26c1a574 | 1275 | .flock = gfs2_flock, |
82c156f8 | 1276 | .splice_read = generic_file_splice_read, |
f42a69fa | 1277 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1278 | .setlease = simple_nosetlease, |
2fe17c10 | 1279 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1280 | }; |
1281 | ||
10d21988 | 1282 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1283 | .iterate_shared = gfs2_readdir, |
26c1a574 SW |
1284 | .unlocked_ioctl = gfs2_ioctl, |
1285 | .open = gfs2_open, | |
df3fd117 | 1286 | .release = gfs2_release, |
26c1a574 SW |
1287 | .fsync = gfs2_fsync, |
1288 | .lock = gfs2_lock, | |
1289 | .flock = gfs2_flock, | |
6038f373 | 1290 | .llseek = default_llseek, |
b3b94faa DT |
1291 | }; |
1292 | ||
f057f6cd SW |
1293 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1294 | ||
10d21988 | 1295 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1296 | .llseek = gfs2_llseek, |
967bcc91 | 1297 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1298 | .write_iter = gfs2_file_write_iter, |
c97bfe43 WC |
1299 | .unlocked_ioctl = gfs2_ioctl, |
1300 | .mmap = gfs2_mmap, | |
1301 | .open = gfs2_open, | |
df3fd117 | 1302 | .release = gfs2_release, |
c97bfe43 | 1303 | .fsync = gfs2_fsync, |
82c156f8 | 1304 | .splice_read = generic_file_splice_read, |
f42a69fa | 1305 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1306 | .setlease = generic_setlease, |
2fe17c10 | 1307 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1308 | }; |
1309 | ||
10d21988 | 1310 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1311 | .iterate_shared = gfs2_readdir, |
c97bfe43 WC |
1312 | .unlocked_ioctl = gfs2_ioctl, |
1313 | .open = gfs2_open, | |
df3fd117 | 1314 | .release = gfs2_release, |
c97bfe43 | 1315 | .fsync = gfs2_fsync, |
6038f373 | 1316 | .llseek = default_llseek, |
c97bfe43 WC |
1317 | }; |
1318 |