Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 4 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
b3b94faa DT |
7 | #include <linux/slab.h> |
8 | #include <linux/spinlock.h> | |
8d098070 | 9 | #include <linux/compat.h> |
b3b94faa DT |
10 | #include <linux/completion.h> |
11 | #include <linux/buffer_head.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/uio.h> | |
14 | #include <linux/blkdev.h> | |
15 | #include <linux/mm.h> | |
f58ba889 | 16 | #include <linux/mount.h> |
18ec7d5c | 17 | #include <linux/fs.h> |
5c676f6d | 18 | #include <linux/gfs2_ondisk.h> |
2fe17c10 CH |
19 | #include <linux/falloc.h> |
20 | #include <linux/swap.h> | |
71b86f56 | 21 | #include <linux/crc32.h> |
33c3de32 | 22 | #include <linux/writeback.h> |
7c0f6ba6 | 23 | #include <linux/uaccess.h> |
f057f6cd SW |
24 | #include <linux/dlm.h> |
25 | #include <linux/dlm_plock.h> | |
2ddfbdd6 | 26 | #include <linux/delay.h> |
64bc06bb | 27 | #include <linux/backing-dev.h> |
88b631cb | 28 | #include <linux/fileattr.h> |
b3b94faa DT |
29 | |
30 | #include "gfs2.h" | |
5c676f6d | 31 | #include "incore.h" |
b3b94faa | 32 | #include "bmap.h" |
64bc06bb | 33 | #include "aops.h" |
b3b94faa DT |
34 | #include "dir.h" |
35 | #include "glock.h" | |
36 | #include "glops.h" | |
37 | #include "inode.h" | |
b3b94faa DT |
38 | #include "log.h" |
39 | #include "meta_io.h" | |
b3b94faa DT |
40 | #include "quota.h" |
41 | #include "rgrp.h" | |
42 | #include "trans.h" | |
5c676f6d | 43 | #include "util.h" |
b3b94faa | 44 | |
b3b94faa DT |
45 | /** |
46 | * gfs2_llseek - seek to a location in a file | |
47 | * @file: the file | |
48 | * @offset: the offset | |
965c8e59 | 49 | * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) |
b3b94faa DT |
50 | * |
51 | * SEEK_END requires the glock for the file because it references the | |
52 | * file's size. | |
53 | * | |
54 | * Returns: The new offset, or errno | |
55 | */ | |
56 | ||
965c8e59 | 57 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence) |
b3b94faa | 58 | { |
feaa7bba | 59 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa DT |
60 | struct gfs2_holder i_gh; |
61 | loff_t error; | |
62 | ||
965c8e59 | 63 | switch (whence) { |
3a27411c | 64 | case SEEK_END: |
b3b94faa DT |
65 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
66 | &i_gh); | |
67 | if (!error) { | |
965c8e59 | 68 | error = generic_file_llseek(file, offset, whence); |
b3b94faa DT |
69 | gfs2_glock_dq_uninit(&i_gh); |
70 | } | |
9453615a | 71 | break; |
3a27411c AG |
72 | |
73 | case SEEK_DATA: | |
74 | error = gfs2_seek_data(file, offset); | |
75 | break; | |
76 | ||
77 | case SEEK_HOLE: | |
78 | error = gfs2_seek_hole(file, offset); | |
79 | break; | |
80 | ||
9453615a SW |
81 | case SEEK_CUR: |
82 | case SEEK_SET: | |
3a27411c AG |
83 | /* |
84 | * These don't reference inode->i_size and don't depend on the | |
85 | * block mapping, so we don't need the glock. | |
86 | */ | |
965c8e59 | 87 | error = generic_file_llseek(file, offset, whence); |
9453615a SW |
88 | break; |
89 | default: | |
90 | error = -EINVAL; | |
91 | } | |
b3b94faa DT |
92 | |
93 | return error; | |
94 | } | |
95 | ||
b3b94faa | 96 | /** |
d81a8ef5 | 97 | * gfs2_readdir - Iterator for a directory |
b3b94faa | 98 | * @file: The directory to read from |
d81a8ef5 | 99 | * @ctx: What to feed directory entries to |
b3b94faa DT |
100 | * |
101 | * Returns: errno | |
102 | */ | |
103 | ||
d81a8ef5 | 104 | static int gfs2_readdir(struct file *file, struct dir_context *ctx) |
b3b94faa | 105 | { |
71b86f56 | 106 | struct inode *dir = file->f_mapping->host; |
feaa7bba | 107 | struct gfs2_inode *dip = GFS2_I(dir); |
b3b94faa | 108 | struct gfs2_holder d_gh; |
b3b94faa DT |
109 | int error; |
110 | ||
d81a8ef5 AV |
111 | error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); |
112 | if (error) | |
b3b94faa | 113 | return error; |
b3b94faa | 114 | |
d81a8ef5 | 115 | error = gfs2_dir_read(dir, ctx, &file->f_ra); |
b3b94faa DT |
116 | |
117 | gfs2_glock_dq_uninit(&d_gh); | |
118 | ||
b3b94faa DT |
119 | return error; |
120 | } | |
121 | ||
c551f66c LJ |
122 | /* |
123 | * struct fsflag_gfs2flag | |
128e5eba | 124 | * |
b16f7e57 AG |
125 | * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories, |
126 | * and to GFS2_DIF_JDATA for non-directories. | |
128e5eba | 127 | */ |
b16f7e57 AG |
128 | static struct { |
129 | u32 fsflag; | |
130 | u32 gfsflag; | |
131 | } fsflag_gfs2flag[] = { | |
132 | {FS_SYNC_FL, GFS2_DIF_SYNC}, | |
133 | {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE}, | |
134 | {FS_APPEND_FL, GFS2_DIF_APPENDONLY}, | |
135 | {FS_NOATIME_FL, GFS2_DIF_NOATIME}, | |
136 | {FS_INDEX_FL, GFS2_DIF_EXHASH}, | |
137 | {FS_TOPDIR_FL, GFS2_DIF_TOPDIR}, | |
138 | {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA}, | |
7ea9ea83 | 139 | }; |
71b86f56 | 140 | |
5aca2842 DW |
141 | static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags) |
142 | { | |
143 | int i; | |
144 | u32 fsflags = 0; | |
145 | ||
146 | if (S_ISDIR(inode->i_mode)) | |
147 | gfsflags &= ~GFS2_DIF_JDATA; | |
148 | else | |
149 | gfsflags &= ~GFS2_DIF_INHERIT_JDATA; | |
150 | ||
151 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) | |
152 | if (gfsflags & fsflag_gfs2flag[i].gfsflag) | |
153 | fsflags |= fsflag_gfs2flag[i].fsflag; | |
154 | return fsflags; | |
155 | } | |
156 | ||
88b631cb | 157 | int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa) |
71b86f56 | 158 | { |
88b631cb | 159 | struct inode *inode = d_inode(dentry); |
feaa7bba | 160 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 | 161 | struct gfs2_holder gh; |
5aca2842 DW |
162 | int error; |
163 | u32 fsflags; | |
71b86f56 | 164 | |
88b631cb MS |
165 | if (d_is_special(dentry)) |
166 | return -ENOTTY; | |
167 | ||
719ee344 SW |
168 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
169 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 170 | if (error) |
9c7fe835 | 171 | goto out_uninit; |
907b9bce | 172 | |
5aca2842 | 173 | fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); |
b16f7e57 | 174 | |
88b631cb | 175 | fileattr_fill_flags(fa, fsflags); |
71b86f56 | 176 | |
3cc3f710 | 177 | gfs2_glock_dq(&gh); |
9c7fe835 | 178 | out_uninit: |
71b86f56 SW |
179 | gfs2_holder_uninit(&gh); |
180 | return error; | |
181 | } | |
182 | ||
6b124d8d SW |
183 | void gfs2_set_inode_flags(struct inode *inode) |
184 | { | |
185 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
186 | unsigned int flags = inode->i_flags; |
187 | ||
9964afbb SW |
188 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
189 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 190 | flags |= S_NOSEC; |
383f01fb | 191 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 192 | flags |= S_IMMUTABLE; |
383f01fb | 193 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 194 | flags |= S_APPEND; |
383f01fb | 195 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 196 | flags |= S_NOATIME; |
383f01fb | 197 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
198 | flags |= S_SYNC; |
199 | inode->i_flags = flags; | |
200 | } | |
201 | ||
71b86f56 SW |
202 | /* Flags that can be set by user space */ |
203 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
204 | GFS2_DIF_IMMUTABLE| \ |
205 | GFS2_DIF_APPENDONLY| \ | |
206 | GFS2_DIF_NOATIME| \ | |
207 | GFS2_DIF_SYNC| \ | |
23d0bb83 | 208 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
209 | GFS2_DIF_INHERIT_JDATA) |
210 | ||
211 | /** | |
9dd868e1 | 212 | * do_gfs2_set_flags - set flags on an inode |
0f1616f6 | 213 | * @inode: The inode |
9dd868e1 | 214 | * @reqflags: The flags to set |
71b86f56 SW |
215 | * @mask: Indicates which flags are valid |
216 | * | |
217 | */ | |
a500bd31 | 218 | static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask) |
71b86f56 | 219 | { |
feaa7bba SW |
220 | struct gfs2_inode *ip = GFS2_I(inode); |
221 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
222 | struct buffer_head *bh; |
223 | struct gfs2_holder gh; | |
224 | int error; | |
88b631cb | 225 | u32 new_flags, flags; |
71b86f56 | 226 | |
f58ba889 MS |
227 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
228 | if (error) | |
88b631cb | 229 | return error; |
7df0e039 SW |
230 | |
231 | error = 0; | |
383f01fb | 232 | flags = ip->i_diskflags; |
55eccc6d | 233 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
234 | if ((new_flags ^ flags) == 0) |
235 | goto out; | |
236 | ||
b9cb9813 | 237 | if (!IS_IMMUTABLE(inode)) { |
549c7297 | 238 | error = gfs2_permission(&init_user_ns, inode, MAY_WRITE); |
b9cb9813 SW |
239 | if (error) |
240 | goto out; | |
241 | } | |
5561093e | 242 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
cc555b09 | 243 | if (new_flags & GFS2_DIF_JDATA) |
c1696fb8 | 244 | gfs2_log_flush(sdp, ip->i_gl, |
805c0907 BP |
245 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
246 | GFS2_LFC_SET_FLAGS); | |
5561093e SW |
247 | error = filemap_fdatawrite(inode->i_mapping); |
248 | if (error) | |
249 | goto out; | |
250 | error = filemap_fdatawait(inode->i_mapping); | |
251 | if (error) | |
252 | goto out; | |
cc555b09 BP |
253 | if (new_flags & GFS2_DIF_JDATA) |
254 | gfs2_ordered_del_inode(ip); | |
5561093e | 255 | } |
55eccc6d | 256 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
257 | if (error) |
258 | goto out; | |
55eccc6d SW |
259 | error = gfs2_meta_inode_buffer(ip, &bh); |
260 | if (error) | |
261 | goto out_trans_end; | |
9b7c2ddb | 262 | inode->i_ctime = current_time(inode); |
350a9b0a | 263 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 264 | ip->i_diskflags = new_flags; |
539e5d6b | 265 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 266 | brelse(bh); |
6b124d8d | 267 | gfs2_set_inode_flags(inode); |
5561093e | 268 | gfs2_set_aops(inode); |
55eccc6d SW |
269 | out_trans_end: |
270 | gfs2_trans_end(sdp); | |
71b86f56 SW |
271 | out: |
272 | gfs2_glock_dq_uninit(&gh); | |
273 | return error; | |
274 | } | |
275 | ||
88b631cb MS |
276 | int gfs2_fileattr_set(struct user_namespace *mnt_userns, |
277 | struct dentry *dentry, struct fileattr *fa) | |
71b86f56 | 278 | { |
88b631cb MS |
279 | struct inode *inode = d_inode(dentry); |
280 | u32 fsflags = fa->flags, gfsflags = 0; | |
b16f7e57 AG |
281 | u32 mask; |
282 | int i; | |
7df0e039 | 283 | |
88b631cb MS |
284 | if (d_is_special(dentry)) |
285 | return -ENOTTY; | |
286 | ||
287 | if (fileattr_has_fsx(fa)) | |
288 | return -EOPNOTSUPP; | |
7df0e039 | 289 | |
b16f7e57 AG |
290 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { |
291 | if (fsflags & fsflag_gfs2flag[i].fsflag) { | |
292 | fsflags &= ~fsflag_gfs2flag[i].fsflag; | |
293 | gfsflags |= fsflag_gfs2flag[i].gfsflag; | |
294 | } | |
295 | } | |
296 | if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) | |
297 | return -EINVAL; | |
298 | ||
299 | mask = GFS2_FLAGS_USER_SET; | |
300 | if (S_ISDIR(inode->i_mode)) { | |
301 | mask &= ~GFS2_DIF_JDATA; | |
302 | } else { | |
303 | /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ | |
304 | if (gfsflags & GFS2_DIF_TOPDIR) | |
305 | return -EINVAL; | |
306 | mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); | |
b9af7ca6 | 307 | } |
b16f7e57 | 308 | |
a500bd31 | 309 | return do_gfs2_set_flags(inode, gfsflags, mask); |
71b86f56 SW |
310 | } |
311 | ||
6ddc5c3d SW |
312 | static int gfs2_getlabel(struct file *filp, char __user *label) |
313 | { | |
314 | struct inode *inode = file_inode(filp); | |
315 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
316 | ||
317 | if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN)) | |
318 | return -EFAULT; | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
b09e593d | 323 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
324 | { |
325 | switch(cmd) { | |
66fc061b SW |
326 | case FITRIM: |
327 | return gfs2_fitrim(filp, (void __user *)arg); | |
6ddc5c3d SW |
328 | case FS_IOC_GETFSLABEL: |
329 | return gfs2_getlabel(filp, (char __user *)arg); | |
71b86f56 | 330 | } |
6ddc5c3d | 331 | |
71b86f56 SW |
332 | return -ENOTTY; |
333 | } | |
334 | ||
8d098070 AB |
335 | #ifdef CONFIG_COMPAT |
336 | static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) | |
337 | { | |
338 | switch(cmd) { | |
8d098070 AB |
339 | /* Keep this list in sync with gfs2_ioctl */ |
340 | case FITRIM: | |
341 | case FS_IOC_GETFSLABEL: | |
342 | break; | |
343 | default: | |
344 | return -ENOIOCTLCMD; | |
345 | } | |
346 | ||
347 | return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)); | |
348 | } | |
349 | #else | |
350 | #define gfs2_compat_ioctl NULL | |
351 | #endif | |
352 | ||
da1dfb6a SW |
353 | /** |
354 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 355 | * @filep: The struct file |
da1dfb6a SW |
356 | * @offset: The file offset of the write |
357 | * @size: The length of the write | |
358 | * | |
359 | * When we are about to do a write, this function records the total | |
360 | * write size in order to provide a suitable hint to the lower layers | |
361 | * about how many blocks will be required. | |
362 | * | |
363 | */ | |
364 | ||
365 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
366 | { | |
496ad9aa | 367 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
368 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
369 | struct gfs2_inode *ip = GFS2_I(inode); | |
370 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
371 | int hint = min_t(size_t, INT_MAX, blks); | |
372 | ||
21f09c43 AG |
373 | if (hint > atomic_read(&ip->i_sizehint)) |
374 | atomic_set(&ip->i_sizehint, hint); | |
da1dfb6a SW |
375 | } |
376 | ||
3cc3f710 | 377 | /** |
35af80ae | 378 | * gfs2_allocate_page_backing - Allocate blocks for a write fault |
3cc3f710 | 379 | * @page: The (locked) page to allocate backing for |
f53056c4 | 380 | * @length: Size of the allocation |
3cc3f710 | 381 | * |
35af80ae CH |
382 | * We try to allocate all the blocks required for the page in one go. This |
383 | * might fail for various reasons, so we keep trying until all the blocks to | |
384 | * back this page are allocated. If some of the blocks are already allocated, | |
385 | * that is ok too. | |
3cc3f710 | 386 | */ |
f53056c4 | 387 | static int gfs2_allocate_page_backing(struct page *page, unsigned int length) |
3cc3f710 | 388 | { |
35af80ae | 389 | u64 pos = page_offset(page); |
3cc3f710 SW |
390 | |
391 | do { | |
35af80ae CH |
392 | struct iomap iomap = { }; |
393 | ||
54992257 | 394 | if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap)) |
3cc3f710 | 395 | return -EIO; |
35af80ae | 396 | |
f53056c4 AG |
397 | if (length < iomap.length) |
398 | iomap.length = length; | |
399 | length -= iomap.length; | |
35af80ae | 400 | pos += iomap.length; |
f53056c4 | 401 | } while (length > 0); |
35af80ae | 402 | |
3cc3f710 SW |
403 | return 0; |
404 | } | |
405 | ||
406 | /** | |
407 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
9dd868e1 | 408 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
409 | * |
410 | * When the page becomes writable, we need to ensure that we have | |
411 | * blocks allocated on disk to back that page. | |
412 | */ | |
413 | ||
109dbb1e | 414 | static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) |
3cc3f710 | 415 | { |
c2ec175c | 416 | struct page *page = vmf->page; |
11bac800 | 417 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3cc3f710 SW |
418 | struct gfs2_inode *ip = GFS2_I(inode); |
419 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
7b9cff46 | 420 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
184b4e60 | 421 | u64 offset = page_offset(page); |
3cc3f710 | 422 | unsigned int data_blocks, ind_blocks, rblocks; |
0fc3bcd6 | 423 | vm_fault_t ret = VM_FAULT_LOCKED; |
3cc3f710 | 424 | struct gfs2_holder gh; |
184b4e60 | 425 | unsigned int length; |
13d921e3 | 426 | loff_t size; |
0fc3bcd6 | 427 | int err; |
3cc3f710 | 428 | |
39263d5e | 429 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 430 | |
719ee344 | 431 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
0fc3bcd6 AG |
432 | err = gfs2_glock_nq(&gh); |
433 | if (err) { | |
434 | ret = block_page_mkwrite_return(err); | |
2b3dcf35 | 435 | goto out_uninit; |
0fc3bcd6 | 436 | } |
3cc3f710 | 437 | |
184b4e60 AG |
438 | /* Check page index against inode size */ |
439 | size = i_size_read(inode); | |
440 | if (offset >= size) { | |
0fc3bcd6 | 441 | ret = VM_FAULT_SIGBUS; |
184b4e60 AG |
442 | goto out_unlock; |
443 | } | |
444 | ||
d7c436cd | 445 | /* Update file times before taking page lock */ |
11bac800 | 446 | file_update_time(vmf->vma->vm_file); |
d7c436cd | 447 | |
184b4e60 | 448 | /* page is wholly or partially inside EOF */ |
d3c51c55 AG |
449 | if (size - offset < PAGE_SIZE) |
450 | length = size - offset; | |
184b4e60 AG |
451 | else |
452 | length = PAGE_SIZE; | |
453 | ||
454 | gfs2_size_hint(vmf->vma->vm_file, offset, length); | |
455 | ||
9c538837 SW |
456 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
457 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
458 | ||
184b4e60 AG |
459 | /* |
460 | * iomap_writepage / iomap_writepages currently don't support inline | |
461 | * files, so always unstuff here. | |
462 | */ | |
463 | ||
464 | if (!gfs2_is_stuffed(ip) && | |
465 | !gfs2_write_alloc_required(ip, offset, length)) { | |
13d921e3 SW |
466 | lock_page(page); |
467 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
0fc3bcd6 | 468 | ret = VM_FAULT_NOPAGE; |
13d921e3 SW |
469 | unlock_page(page); |
470 | } | |
3cc3f710 | 471 | goto out_unlock; |
13d921e3 SW |
472 | } |
473 | ||
0fc3bcd6 AG |
474 | err = gfs2_rindex_update(sdp); |
475 | if (err) { | |
476 | ret = block_page_mkwrite_return(err); | |
6dbd8224 | 477 | goto out_unlock; |
0fc3bcd6 | 478 | } |
6dbd8224 | 479 | |
184b4e60 | 480 | gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks); |
7b9cff46 | 481 | ap.target = data_blocks + ind_blocks; |
0fc3bcd6 AG |
482 | err = gfs2_quota_lock_check(ip, &ap); |
483 | if (err) { | |
484 | ret = block_page_mkwrite_return(err); | |
b8fbf471 | 485 | goto out_unlock; |
0fc3bcd6 AG |
486 | } |
487 | err = gfs2_inplace_reserve(ip, &ap); | |
488 | if (err) { | |
489 | ret = block_page_mkwrite_return(err); | |
3cc3f710 | 490 | goto out_quota_unlock; |
0fc3bcd6 | 491 | } |
3cc3f710 SW |
492 | |
493 | rblocks = RES_DINODE + ind_blocks; | |
494 | if (gfs2_is_jdata(ip)) | |
495 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 496 | if (ind_blocks || data_blocks) { |
3cc3f710 | 497 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 498 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 499 | } |
0fc3bcd6 AG |
500 | err = gfs2_trans_begin(sdp, rblocks, 0); |
501 | if (err) { | |
502 | ret = block_page_mkwrite_return(err); | |
3cc3f710 | 503 | goto out_trans_fail; |
0fc3bcd6 | 504 | } |
3cc3f710 | 505 | |
64090cbe AG |
506 | /* Unstuff, if required, and allocate backing blocks for page */ |
507 | if (gfs2_is_stuffed(ip)) { | |
7a607a41 | 508 | err = gfs2_unstuff_dinode(ip); |
64090cbe AG |
509 | if (err) { |
510 | ret = block_page_mkwrite_return(err); | |
511 | goto out_trans_end; | |
512 | } | |
513 | } | |
514 | ||
3cc3f710 | 515 | lock_page(page); |
13d921e3 SW |
516 | /* If truncated, we must retry the operation, we may have raced |
517 | * with the glock demotion code. | |
518 | */ | |
0fc3bcd6 AG |
519 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { |
520 | ret = VM_FAULT_NOPAGE; | |
64090cbe | 521 | goto out_page_locked; |
0fc3bcd6 | 522 | } |
13d921e3 | 523 | |
0fc3bcd6 AG |
524 | err = gfs2_allocate_page_backing(page, length); |
525 | if (err) | |
526 | ret = block_page_mkwrite_return(err); | |
3cc3f710 | 527 | |
64090cbe | 528 | out_page_locked: |
0fc3bcd6 | 529 | if (ret != VM_FAULT_LOCKED) |
13d921e3 | 530 | unlock_page(page); |
64090cbe | 531 | out_trans_end: |
3cc3f710 SW |
532 | gfs2_trans_end(sdp); |
533 | out_trans_fail: | |
534 | gfs2_inplace_release(ip); | |
535 | out_quota_unlock: | |
536 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
537 | out_unlock: |
538 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 539 | out_uninit: |
3cc3f710 | 540 | gfs2_holder_uninit(&gh); |
0fc3bcd6 | 541 | if (ret == VM_FAULT_LOCKED) { |
13d921e3 | 542 | set_page_dirty(page); |
1d1d1a76 | 543 | wait_for_stable_page(page); |
13d921e3 | 544 | } |
39263d5e | 545 | sb_end_pagefault(inode->i_sb); |
0fc3bcd6 | 546 | return ret; |
3cc3f710 SW |
547 | } |
548 | ||
20f82999 AG |
549 | static vm_fault_t gfs2_fault(struct vm_fault *vmf) |
550 | { | |
551 | struct inode *inode = file_inode(vmf->vma->vm_file); | |
552 | struct gfs2_inode *ip = GFS2_I(inode); | |
553 | struct gfs2_holder gh; | |
554 | vm_fault_t ret; | |
555 | int err; | |
556 | ||
d5b81454 | 557 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
20f82999 AG |
558 | err = gfs2_glock_nq(&gh); |
559 | if (err) { | |
560 | ret = block_page_mkwrite_return(err); | |
561 | goto out_uninit; | |
562 | } | |
563 | ret = filemap_fault(vmf); | |
564 | gfs2_glock_dq(&gh); | |
565 | out_uninit: | |
566 | gfs2_holder_uninit(&gh); | |
567 | return ret; | |
568 | } | |
569 | ||
f0f37e2f | 570 | static const struct vm_operations_struct gfs2_vm_ops = { |
20f82999 | 571 | .fault = gfs2_fault, |
f1820361 | 572 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
573 | .page_mkwrite = gfs2_page_mkwrite, |
574 | }; | |
575 | ||
b3b94faa | 576 | /** |
c551f66c | 577 | * gfs2_mmap |
b3b94faa DT |
578 | * @file: The file to map |
579 | * @vma: The VMA which described the mapping | |
580 | * | |
48bf2b17 SW |
581 | * There is no need to get a lock here unless we should be updating |
582 | * atime. We ignore any locking errors since the only consequence is | |
583 | * a missed atime update (which will just be deferred until later). | |
584 | * | |
585 | * Returns: 0 | |
b3b94faa DT |
586 | */ |
587 | ||
588 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
589 | { | |
feaa7bba | 590 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 591 | |
b9c93bb7 SW |
592 | if (!(file->f_flags & O_NOATIME) && |
593 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
594 | struct gfs2_holder i_gh; |
595 | int error; | |
b3b94faa | 596 | |
3d162688 BM |
597 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
598 | &i_gh); | |
b9c93bb7 SW |
599 | if (error) |
600 | return error; | |
3d162688 BM |
601 | /* grab lock to update inode */ |
602 | gfs2_glock_dq_uninit(&i_gh); | |
603 | file_accessed(file); | |
48bf2b17 | 604 | } |
3cc3f710 | 605 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 606 | |
48bf2b17 | 607 | return 0; |
b3b94faa DT |
608 | } |
609 | ||
610 | /** | |
6d4ade98 SW |
611 | * gfs2_open_common - This is common to open and atomic_open |
612 | * @inode: The inode being opened | |
613 | * @file: The file being opened | |
b3b94faa | 614 | * |
6d4ade98 SW |
615 | * This maybe called under a glock or not depending upon how it has |
616 | * been called. We must always be called under a glock for regular | |
617 | * files, however. For other file types, it does not matter whether | |
618 | * we hold the glock or not. | |
619 | * | |
620 | * Returns: Error code or 0 for success | |
b3b94faa DT |
621 | */ |
622 | ||
6d4ade98 | 623 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 624 | { |
b3b94faa | 625 | struct gfs2_file *fp; |
6d4ade98 SW |
626 | int ret; |
627 | ||
628 | if (S_ISREG(inode->i_mode)) { | |
629 | ret = generic_file_open(inode, file); | |
630 | if (ret) | |
631 | return ret; | |
632 | } | |
b3b94faa | 633 | |
6d4ade98 | 634 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
635 | if (!fp) |
636 | return -ENOMEM; | |
637 | ||
f55ab26a | 638 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 639 | |
feaa7bba | 640 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 641 | file->private_data = fp; |
2fba46a0 BP |
642 | if (file->f_mode & FMODE_WRITE) { |
643 | ret = gfs2_qa_get(GFS2_I(inode)); | |
644 | if (ret) | |
645 | goto fail; | |
646 | } | |
6d4ade98 | 647 | return 0; |
2fba46a0 BP |
648 | |
649 | fail: | |
650 | kfree(file->private_data); | |
651 | file->private_data = NULL; | |
652 | return ret; | |
6d4ade98 SW |
653 | } |
654 | ||
655 | /** | |
656 | * gfs2_open - open a file | |
657 | * @inode: the inode to open | |
658 | * @file: the struct file for this opening | |
659 | * | |
660 | * After atomic_open, this function is only used for opening files | |
661 | * which are already cached. We must still get the glock for regular | |
662 | * files to ensure that we have the file size uptodate for the large | |
663 | * file check which is in the common code. That is only an issue for | |
664 | * regular files though. | |
665 | * | |
666 | * Returns: errno | |
667 | */ | |
668 | ||
669 | static int gfs2_open(struct inode *inode, struct file *file) | |
670 | { | |
671 | struct gfs2_inode *ip = GFS2_I(inode); | |
672 | struct gfs2_holder i_gh; | |
673 | int error; | |
674 | bool need_unlock = false; | |
b3b94faa | 675 | |
b60623c2 | 676 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
677 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
678 | &i_gh); | |
679 | if (error) | |
6d4ade98 SW |
680 | return error; |
681 | need_unlock = true; | |
682 | } | |
b3b94faa | 683 | |
6d4ade98 | 684 | error = gfs2_open_common(inode, file); |
b3b94faa | 685 | |
6d4ade98 | 686 | if (need_unlock) |
b3b94faa | 687 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 688 | |
b3b94faa DT |
689 | return error; |
690 | } | |
691 | ||
692 | /** | |
df3fd117 | 693 | * gfs2_release - called to close a struct file |
b3b94faa DT |
694 | * @inode: the inode the struct file belongs to |
695 | * @file: the struct file being closed | |
696 | * | |
697 | * Returns: errno | |
698 | */ | |
699 | ||
df3fd117 | 700 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 701 | { |
0a305e49 | 702 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 703 | |
8e2e0047 | 704 | kfree(file->private_data); |
5c676f6d | 705 | file->private_data = NULL; |
b3b94faa | 706 | |
d3add1a9 BP |
707 | if (file->f_mode & FMODE_WRITE) { |
708 | if (gfs2_rs_active(&ip->i_res)) | |
7336905a | 709 | gfs2_rs_delete(ip); |
1595548f | 710 | gfs2_qa_put(ip); |
d3add1a9 | 711 | } |
b3b94faa DT |
712 | return 0; |
713 | } | |
714 | ||
715 | /** | |
716 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
717 | * @file: the file that points to the dentry |
718 | * @start: the start position in the file to sync | |
719 | * @end: the end position in the file to sync | |
dba898b0 | 720 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 721 | * |
2f0264d5 SW |
722 | * We split the data flushing here so that we don't wait for the data |
723 | * until after we've also sent the metadata to disk. Note that for | |
724 | * data=ordered, we will write & wait for the data at the log flush | |
725 | * stage anyway, so this is unlikely to make much of a difference | |
726 | * except in the data=writeback case. | |
727 | * | |
728 | * If the fdatawrite fails due to any reason except -EIO, we will | |
729 | * continue the remainder of the fsync, although we'll still report | |
730 | * the error at the end. This is to match filemap_write_and_wait_range() | |
731 | * behaviour. | |
34126f9f | 732 | * |
b3b94faa DT |
733 | * Returns: errno |
734 | */ | |
735 | ||
02c24a82 JB |
736 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
737 | int datasync) | |
b3b94faa | 738 | { |
2f0264d5 SW |
739 | struct address_space *mapping = file->f_mapping; |
740 | struct inode *inode = mapping->host; | |
3aac630b | 741 | int sync_state = inode->i_state & I_DIRTY; |
dba898b0 | 742 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 743 | int ret = 0, ret1 = 0; |
b3b94faa | 744 | |
2f0264d5 SW |
745 | if (mapping->nrpages) { |
746 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
747 | if (ret1 == -EIO) | |
748 | return ret1; | |
749 | } | |
02c24a82 | 750 | |
0c901809 BM |
751 | if (!gfs2_is_jdata(ip)) |
752 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 753 | if (datasync) |
3aac630b | 754 | sync_state &= ~I_DIRTY_SYNC; |
b3b94faa | 755 | |
dba898b0 SW |
756 | if (sync_state) { |
757 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 758 | if (ret) |
dba898b0 | 759 | return ret; |
f1818529 | 760 | if (gfs2_is_jdata(ip)) |
d07a6ac7 JL |
761 | ret = file_write_and_wait(file); |
762 | if (ret) | |
763 | return ret; | |
b5b24d7a | 764 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
765 | } |
766 | ||
2f0264d5 | 767 | if (mapping->nrpages) |
d07a6ac7 | 768 | ret = file_fdatawait_range(file, start, end); |
2f0264d5 SW |
769 | |
770 | return ret ? ret : ret1; | |
b3b94faa DT |
771 | } |
772 | ||
72382264 | 773 | static inline bool should_fault_in_pages(struct iov_iter *i, |
324d116c | 774 | struct kiocb *iocb, |
00bfe02f AG |
775 | size_t *prev_count, |
776 | size_t *window_size) | |
777 | { | |
00bfe02f | 778 | size_t count = iov_iter_count(i); |
bb7f5d96 | 779 | size_t size, offs; |
00bfe02f | 780 | |
fa5dfa64 | 781 | if (!count) |
00bfe02f | 782 | return false; |
fcb14cb1 | 783 | if (!user_backed_iter(i)) |
00bfe02f AG |
784 | return false; |
785 | ||
bb7f5d96 | 786 | size = PAGE_SIZE; |
324d116c | 787 | offs = offset_in_page(iocb->ki_pos); |
00bfe02f | 788 | if (*prev_count != count || !*window_size) { |
bb7f5d96 | 789 | size_t nr_dirtied; |
00bfe02f | 790 | |
00bfe02f | 791 | nr_dirtied = max(current->nr_dirtied_pause - |
bb7f5d96 | 792 | current->nr_dirtied, 8); |
324d116c | 793 | size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT); |
00bfe02f AG |
794 | } |
795 | ||
796 | *prev_count = count; | |
bb7f5d96 | 797 | *window_size = size - offs; |
00bfe02f AG |
798 | return true; |
799 | } | |
800 | ||
4c5c3010 AG |
801 | static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to, |
802 | struct gfs2_holder *gh) | |
967bcc91 AG |
803 | { |
804 | struct file *file = iocb->ki_filp; | |
805 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); | |
b01b2d72 | 806 | size_t prev_count = 0, window_size = 0; |
42e4c3bd | 807 | size_t read = 0; |
967bcc91 AG |
808 | ssize_t ret; |
809 | ||
b01b2d72 AG |
810 | /* |
811 | * In this function, we disable page faults when we're holding the | |
812 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
813 | * that the inode glock may be dropped, fault in the pages manually, | |
814 | * and retry. | |
815 | * | |
816 | * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger | |
817 | * physical as well as manual page faults, and we need to disable both | |
818 | * kinds. | |
819 | * | |
820 | * For direct I/O, gfs2 takes the inode glock in deferred mode. This | |
821 | * locking mode is compatible with other deferred holders, so multiple | |
822 | * processes and nodes can do direct I/O to a file at the same time. | |
823 | * There's no guarantee that reads or writes will be atomic. Any | |
824 | * coordination among readers and writers needs to happen externally. | |
825 | */ | |
826 | ||
827 | if (!iov_iter_count(to)) | |
967bcc91 AG |
828 | return 0; /* skip atime */ |
829 | ||
4c5c3010 | 830 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh); |
b01b2d72 | 831 | retry: |
4c5c3010 | 832 | ret = gfs2_glock_nq(gh); |
967bcc91 AG |
833 | if (ret) |
834 | goto out_uninit; | |
b01b2d72 AG |
835 | pagefault_disable(); |
836 | to->nofault = true; | |
837 | ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, | |
786f847f | 838 | IOMAP_DIO_PARTIAL, NULL, read); |
b01b2d72 AG |
839 | to->nofault = false; |
840 | pagefault_enable(); | |
72382264 AG |
841 | if (ret <= 0 && ret != -EFAULT) |
842 | goto out_unlock; | |
53bb540f | 843 | /* No increment (+=) because iomap_dio_rw returns a cumulative value. */ |
b01b2d72 | 844 | if (ret > 0) |
42e4c3bd | 845 | read = ret; |
b01b2d72 | 846 | |
324d116c | 847 | if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) { |
e1fa9ea8 | 848 | gfs2_glock_dq(gh); |
6d22ff47 | 849 | window_size -= fault_in_iov_iter_writeable(to, window_size); |
e1fa9ea8 | 850 | if (window_size) |
124c458a | 851 | goto retry; |
b01b2d72 | 852 | } |
72382264 | 853 | out_unlock: |
b01b2d72 AG |
854 | if (gfs2_holder_queued(gh)) |
855 | gfs2_glock_dq(gh); | |
967bcc91 | 856 | out_uninit: |
4c5c3010 | 857 | gfs2_holder_uninit(gh); |
53bb540f | 858 | /* User space doesn't expect partial success. */ |
b01b2d72 AG |
859 | if (ret < 0) |
860 | return ret; | |
42e4c3bd | 861 | return read; |
967bcc91 AG |
862 | } |
863 | ||
4c5c3010 AG |
864 | static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from, |
865 | struct gfs2_holder *gh) | |
967bcc91 AG |
866 | { |
867 | struct file *file = iocb->ki_filp; | |
868 | struct inode *inode = file->f_mapping->host; | |
869 | struct gfs2_inode *ip = GFS2_I(inode); | |
b01b2d72 | 870 | size_t prev_count = 0, window_size = 0; |
42e4c3bd | 871 | size_t written = 0; |
967bcc91 AG |
872 | ssize_t ret; |
873 | ||
b01b2d72 AG |
874 | /* |
875 | * In this function, we disable page faults when we're holding the | |
876 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
877 | * that the inode glock may be dropped, fault in the pages manually, | |
878 | * and retry. | |
879 | * | |
880 | * For writes, iomap_dio_rw only triggers manual page faults, so we | |
881 | * don't need to disable physical ones. | |
882 | */ | |
883 | ||
967bcc91 AG |
884 | /* |
885 | * Deferred lock, even if its a write, since we do no allocation on | |
886 | * this path. All we need to change is the atime, and this lock mode | |
887 | * ensures that other nodes have flushed their buffered read caches | |
888 | * (i.e. their page cache entries for this inode). We do not, | |
889 | * unfortunately, have the option of only flushing a range like the | |
890 | * VFS does. | |
891 | */ | |
4c5c3010 | 892 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh); |
b01b2d72 | 893 | retry: |
4c5c3010 | 894 | ret = gfs2_glock_nq(gh); |
967bcc91 AG |
895 | if (ret) |
896 | goto out_uninit; | |
967bcc91 | 897 | /* Silently fall back to buffered I/O when writing beyond EOF */ |
b01b2d72 | 898 | if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode)) |
72382264 | 899 | goto out_unlock; |
967bcc91 | 900 | |
b01b2d72 AG |
901 | from->nofault = true; |
902 | ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, | |
786f847f | 903 | IOMAP_DIO_PARTIAL, NULL, written); |
b01b2d72 | 904 | from->nofault = false; |
72382264 AG |
905 | if (ret <= 0) { |
906 | if (ret == -ENOTBLK) | |
907 | ret = 0; | |
908 | if (ret != -EFAULT) | |
909 | goto out_unlock; | |
910 | } | |
53bb540f | 911 | /* No increment (+=) because iomap_dio_rw returns a cumulative value. */ |
b01b2d72 | 912 | if (ret > 0) |
42e4c3bd | 913 | written = ret; |
b01b2d72 | 914 | |
324d116c | 915 | if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) { |
e1fa9ea8 | 916 | gfs2_glock_dq(gh); |
6d22ff47 | 917 | window_size -= fault_in_iov_iter_readable(from, window_size); |
e1fa9ea8 | 918 | if (window_size) |
124c458a | 919 | goto retry; |
b01b2d72 | 920 | } |
72382264 | 921 | out_unlock: |
b01b2d72 AG |
922 | if (gfs2_holder_queued(gh)) |
923 | gfs2_glock_dq(gh); | |
967bcc91 | 924 | out_uninit: |
4c5c3010 | 925 | gfs2_holder_uninit(gh); |
53bb540f | 926 | /* User space doesn't expect partial success. */ |
b01b2d72 AG |
927 | if (ret < 0) |
928 | return ret; | |
42e4c3bd | 929 | return written; |
967bcc91 AG |
930 | } |
931 | ||
932 | static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
933 | { | |
20f82999 AG |
934 | struct gfs2_inode *ip; |
935 | struct gfs2_holder gh; | |
00bfe02f | 936 | size_t prev_count = 0, window_size = 0; |
42e4c3bd | 937 | size_t read = 0; |
967bcc91 AG |
938 | ssize_t ret; |
939 | ||
00bfe02f AG |
940 | /* |
941 | * In this function, we disable page faults when we're holding the | |
942 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
943 | * that the inode glock may be dropped, fault in the pages manually, | |
944 | * and retry. | |
945 | */ | |
946 | ||
11661835 AG |
947 | if (iocb->ki_flags & IOCB_DIRECT) |
948 | return gfs2_file_direct_read(iocb, to, &gh); | |
949 | ||
52f3f033 | 950 | pagefault_disable(); |
20f82999 AG |
951 | iocb->ki_flags |= IOCB_NOIO; |
952 | ret = generic_file_read_iter(iocb, to); | |
953 | iocb->ki_flags &= ~IOCB_NOIO; | |
52f3f033 | 954 | pagefault_enable(); |
20f82999 AG |
955 | if (ret >= 0) { |
956 | if (!iov_iter_count(to)) | |
957 | return ret; | |
42e4c3bd | 958 | read = ret; |
52f3f033 | 959 | } else if (ret != -EFAULT) { |
20f82999 AG |
960 | if (ret != -EAGAIN) |
961 | return ret; | |
962 | if (iocb->ki_flags & IOCB_NOWAIT) | |
963 | return ret; | |
964 | } | |
965 | ip = GFS2_I(iocb->ki_filp->f_mapping->host); | |
966 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
00bfe02f | 967 | retry: |
20f82999 AG |
968 | ret = gfs2_glock_nq(&gh); |
969 | if (ret) | |
970 | goto out_uninit; | |
00bfe02f | 971 | pagefault_disable(); |
20f82999 | 972 | ret = generic_file_read_iter(iocb, to); |
00bfe02f | 973 | pagefault_enable(); |
72382264 AG |
974 | if (ret <= 0 && ret != -EFAULT) |
975 | goto out_unlock; | |
20f82999 | 976 | if (ret > 0) |
42e4c3bd | 977 | read += ret; |
00bfe02f | 978 | |
324d116c | 979 | if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) { |
e1fa9ea8 | 980 | gfs2_glock_dq(&gh); |
6d22ff47 | 981 | window_size -= fault_in_iov_iter_writeable(to, window_size); |
e1fa9ea8 | 982 | if (window_size) |
124c458a | 983 | goto retry; |
00bfe02f | 984 | } |
72382264 | 985 | out_unlock: |
00bfe02f AG |
986 | if (gfs2_holder_queued(&gh)) |
987 | gfs2_glock_dq(&gh); | |
20f82999 AG |
988 | out_uninit: |
989 | gfs2_holder_uninit(&gh); | |
42e4c3bd | 990 | return read ? read : ret; |
967bcc91 AG |
991 | } |
992 | ||
1b223f70 AG |
993 | static ssize_t gfs2_file_buffered_write(struct kiocb *iocb, |
994 | struct iov_iter *from, | |
995 | struct gfs2_holder *gh) | |
2eb7509a AG |
996 | { |
997 | struct file *file = iocb->ki_filp; | |
998 | struct inode *inode = file_inode(file); | |
b924bdab AG |
999 | struct gfs2_inode *ip = GFS2_I(inode); |
1000 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1b223f70 | 1001 | struct gfs2_holder *statfs_gh = NULL; |
00bfe02f | 1002 | size_t prev_count = 0, window_size = 0; |
554c577c | 1003 | size_t orig_count = iov_iter_count(from); |
42e4c3bd | 1004 | size_t written = 0; |
2eb7509a AG |
1005 | ssize_t ret; |
1006 | ||
00bfe02f AG |
1007 | /* |
1008 | * In this function, we disable page faults when we're holding the | |
1009 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
1010 | * that the inode glock may be dropped, fault in the pages manually, | |
1011 | * and retry. | |
1012 | */ | |
1013 | ||
1b223f70 AG |
1014 | if (inode == sdp->sd_rindex) { |
1015 | statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS); | |
1016 | if (!statfs_gh) | |
1017 | return -ENOMEM; | |
1018 | } | |
1019 | ||
1020 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh); | |
00bfe02f | 1021 | retry: |
fa5dfa64 | 1022 | if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) { |
fa5dfa64 | 1023 | window_size -= fault_in_iov_iter_readable(from, window_size); |
fa5dfa64 AG |
1024 | if (!window_size) { |
1025 | ret = -EFAULT; | |
e1fa9ea8 | 1026 | goto out_uninit; |
fa5dfa64 | 1027 | } |
fa5dfa64 AG |
1028 | from->count = min(from->count, window_size); |
1029 | } | |
e1fa9ea8 AG |
1030 | ret = gfs2_glock_nq(gh); |
1031 | if (ret) | |
1032 | goto out_uninit; | |
fa5dfa64 | 1033 | |
b924bdab AG |
1034 | if (inode == sdp->sd_rindex) { |
1035 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
1036 | ||
1037 | ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
1b223f70 | 1038 | GL_NOCACHE, statfs_gh); |
b924bdab AG |
1039 | if (ret) |
1040 | goto out_unlock; | |
1041 | } | |
1042 | ||
2eb7509a | 1043 | current->backing_dev_info = inode_to_bdi(inode); |
00bfe02f | 1044 | pagefault_disable(); |
2eb7509a | 1045 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); |
00bfe02f | 1046 | pagefault_enable(); |
2eb7509a | 1047 | current->backing_dev_info = NULL; |
00bfe02f | 1048 | if (ret > 0) { |
2eb7509a | 1049 | iocb->ki_pos += ret; |
42e4c3bd | 1050 | written += ret; |
00bfe02f | 1051 | } |
b924bdab | 1052 | |
1b223f70 AG |
1053 | if (inode == sdp->sd_rindex) |
1054 | gfs2_glock_dq_uninit(statfs_gh); | |
b924bdab | 1055 | |
72382264 AG |
1056 | if (ret <= 0 && ret != -EFAULT) |
1057 | goto out_unlock; | |
1058 | ||
42e4c3bd | 1059 | from->count = orig_count - written; |
e1fa9ea8 AG |
1060 | if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) { |
1061 | gfs2_glock_dq(gh); | |
1062 | goto retry; | |
1063 | } | |
b924bdab | 1064 | out_unlock: |
00bfe02f AG |
1065 | if (gfs2_holder_queued(gh)) |
1066 | gfs2_glock_dq(gh); | |
b924bdab | 1067 | out_uninit: |
1b223f70 | 1068 | gfs2_holder_uninit(gh); |
ab37c305 | 1069 | kfree(statfs_gh); |
42e4c3bd AG |
1070 | from->count = orig_count - written; |
1071 | return written ? written : ret; | |
2eb7509a AG |
1072 | } |
1073 | ||
56aa616a | 1074 | /** |
da56e45b | 1075 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a | 1076 | * @iocb: The io context |
64bc06bb | 1077 | * @from: The data to write |
56aa616a SW |
1078 | * |
1079 | * We have to do a lock/unlock here to refresh the inode size for | |
1080 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
1081 | * offset. There is still a race, but provided the app is using its | |
1082 | * own file locking, this will make O_APPEND work as expected. | |
1083 | * | |
1084 | */ | |
1085 | ||
da56e45b | 1086 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
1087 | { |
1088 | struct file *file = iocb->ki_filp; | |
64bc06bb AG |
1089 | struct inode *inode = file_inode(file); |
1090 | struct gfs2_inode *ip = GFS2_I(inode); | |
4c5c3010 | 1091 | struct gfs2_holder gh; |
6e5e41e2 | 1092 | ssize_t ret; |
0a305e49 | 1093 | |
da56e45b | 1094 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 1095 | |
2ba48ce5 | 1096 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a SW |
1097 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
1098 | if (ret) | |
4bd684bc | 1099 | return ret; |
56aa616a SW |
1100 | gfs2_glock_dq_uninit(&gh); |
1101 | } | |
1102 | ||
64bc06bb AG |
1103 | inode_lock(inode); |
1104 | ret = generic_write_checks(iocb, from); | |
1105 | if (ret <= 0) | |
4c0e8dda | 1106 | goto out_unlock; |
64bc06bb AG |
1107 | |
1108 | ret = file_remove_privs(file); | |
1109 | if (ret) | |
4c0e8dda | 1110 | goto out_unlock; |
64bc06bb AG |
1111 | |
1112 | ret = file_update_time(file); | |
1113 | if (ret) | |
4c0e8dda | 1114 | goto out_unlock; |
64bc06bb | 1115 | |
967bcc91 AG |
1116 | if (iocb->ki_flags & IOCB_DIRECT) { |
1117 | struct address_space *mapping = file->f_mapping; | |
6e5e41e2 | 1118 | ssize_t buffered, ret2; |
967bcc91 | 1119 | |
4c5c3010 | 1120 | ret = gfs2_file_direct_write(iocb, from, &gh); |
6e5e41e2 | 1121 | if (ret < 0 || !iov_iter_count(from)) |
4c0e8dda | 1122 | goto out_unlock; |
967bcc91 | 1123 | |
6e5e41e2 | 1124 | iocb->ki_flags |= IOCB_DSYNC; |
1b223f70 | 1125 | buffered = gfs2_file_buffered_write(iocb, from, &gh); |
43a511c4 AG |
1126 | if (unlikely(buffered <= 0)) { |
1127 | if (!ret) | |
1128 | ret = buffered; | |
4c0e8dda | 1129 | goto out_unlock; |
43a511c4 | 1130 | } |
967bcc91 AG |
1131 | |
1132 | /* | |
1133 | * We need to ensure that the page cache pages are written to | |
1134 | * disk and invalidated to preserve the expected O_DIRECT | |
6e5e41e2 AG |
1135 | * semantics. If the writeback or invalidate fails, only report |
1136 | * the direct I/O range as we don't know if the buffered pages | |
1137 | * made it to disk. | |
967bcc91 | 1138 | */ |
6e5e41e2 AG |
1139 | ret2 = generic_write_sync(iocb, buffered); |
1140 | invalidate_mapping_pages(mapping, | |
1141 | (iocb->ki_pos - buffered) >> PAGE_SHIFT, | |
1142 | (iocb->ki_pos - 1) >> PAGE_SHIFT); | |
1143 | if (!ret || ret2 > 0) | |
1144 | ret += ret2; | |
967bcc91 | 1145 | } else { |
1b223f70 | 1146 | ret = gfs2_file_buffered_write(iocb, from, &gh); |
2eb7509a | 1147 | if (likely(ret > 0)) |
6e5e41e2 | 1148 | ret = generic_write_sync(iocb, ret); |
967bcc91 | 1149 | } |
64bc06bb | 1150 | |
4c0e8dda | 1151 | out_unlock: |
64bc06bb | 1152 | inode_unlock(inode); |
6e5e41e2 | 1153 | return ret; |
56aa616a SW |
1154 | } |
1155 | ||
2fe17c10 CH |
1156 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
1157 | int mode) | |
1158 | { | |
fffb6412 | 1159 | struct super_block *sb = inode->i_sb; |
2fe17c10 | 1160 | struct gfs2_inode *ip = GFS2_I(inode); |
fffb6412 | 1161 | loff_t end = offset + len; |
2fe17c10 CH |
1162 | struct buffer_head *dibh; |
1163 | int error; | |
2fe17c10 CH |
1164 | |
1165 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1166 | if (unlikely(error)) | |
64dd153c | 1167 | return error; |
2fe17c10 | 1168 | |
350a9b0a | 1169 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
1170 | |
1171 | if (gfs2_is_stuffed(ip)) { | |
7a607a41 | 1172 | error = gfs2_unstuff_dinode(ip); |
2fe17c10 CH |
1173 | if (unlikely(error)) |
1174 | goto out; | |
1175 | } | |
1176 | ||
fffb6412 | 1177 | while (offset < end) { |
c2589282 AG |
1178 | struct iomap iomap = { }; |
1179 | ||
54992257 | 1180 | error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap); |
fffb6412 | 1181 | if (error) |
64dd153c | 1182 | goto out; |
fffb6412 | 1183 | offset = iomap.offset + iomap.length; |
d505a96a | 1184 | if (!(iomap.flags & IOMAP_F_NEW)) |
64dd153c | 1185 | continue; |
fffb6412 AG |
1186 | error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, |
1187 | iomap.length >> inode->i_blkbits, | |
1188 | GFP_NOFS); | |
1189 | if (error) { | |
1190 | fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); | |
2fe17c10 | 1191 | goto out; |
64dd153c | 1192 | } |
2fe17c10 | 1193 | } |
2fe17c10 | 1194 | out: |
64dd153c | 1195 | brelse(dibh); |
2fe17c10 CH |
1196 | return error; |
1197 | } | |
f3b64b57 | 1198 | |
d9be0cda AD |
1199 | /** |
1200 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
1201 | * blocks, determine how many bytes can be written. | |
1202 | * @ip: The inode in question. | |
1203 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
1204 | * @data_blocks: Compute and return the number of data blocks needed | |
1205 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
1206 | * @max_blocks: The total blocks available to work with. | |
1207 | * | |
1208 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
1209 | */ | |
1210 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
1211 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
1212 | unsigned int max_blocks) | |
2fe17c10 | 1213 | { |
d9be0cda | 1214 | loff_t max = *len; |
2fe17c10 | 1215 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
1216 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
1217 | ||
1218 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
1219 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
1220 | max_data -= tmp; | |
1221 | } | |
d9be0cda | 1222 | |
2fe17c10 CH |
1223 | *data_blocks = max_data; |
1224 | *ind_blocks = max_blocks - max_data; | |
1225 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
1226 | if (*len > max) { | |
1227 | *len = max; | |
1228 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
1229 | } | |
1230 | } | |
1231 | ||
9c9f1159 | 1232 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 1233 | { |
496ad9aa | 1234 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
1235 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1236 | struct gfs2_inode *ip = GFS2_I(inode); | |
7b9cff46 | 1237 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
2fe17c10 | 1238 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
174d1232 | 1239 | loff_t bytes, max_bytes, max_blks; |
2fe17c10 | 1240 | int error; |
4442f2e0 SW |
1241 | const loff_t pos = offset; |
1242 | const loff_t count = len; | |
6905d9e4 | 1243 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 1244 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 1245 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 1246 | |
2fe17c10 CH |
1247 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
1248 | ||
6905d9e4 | 1249 | offset &= bsize_mask; |
2fe17c10 CH |
1250 | |
1251 | len = next - offset; | |
1252 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
1253 | if (!bytes) | |
1254 | bytes = UINT_MAX; | |
6905d9e4 BM |
1255 | bytes &= bsize_mask; |
1256 | if (bytes == 0) | |
1257 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 1258 | |
da1dfb6a | 1259 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 1260 | |
d9be0cda AD |
1261 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
1262 | ap.min_target = data_blocks + ind_blocks; | |
1263 | ||
2fe17c10 CH |
1264 | while (len > 0) { |
1265 | if (len < bytes) | |
1266 | bytes = len; | |
58a7d5fb BM |
1267 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
1268 | len -= bytes; | |
1269 | offset += bytes; | |
1270 | continue; | |
1271 | } | |
d9be0cda AD |
1272 | |
1273 | /* We need to determine how many bytes we can actually | |
1274 | * fallocate without exceeding quota or going over the | |
1275 | * end of the fs. We start off optimistically by assuming | |
1276 | * we can write max_bytes */ | |
1277 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
1278 | ||
1279 | /* Since max_bytes is most likely a theoretical max, we | |
1280 | * calculate a more realistic 'bytes' to serve as a good | |
1281 | * starting point for the number of bytes we may be able | |
1282 | * to write */ | |
2fe17c10 | 1283 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 1284 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
1285 | |
1286 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 1287 | if (error) |
9c9f1159 | 1288 | return error; |
d9be0cda AD |
1289 | /* ap.allowed tells us how many blocks quota will allow |
1290 | * us to write. Check if this reduces max_blks */ | |
174d1232 AG |
1291 | max_blks = UINT_MAX; |
1292 | if (ap.allowed) | |
d9be0cda | 1293 | max_blks = ap.allowed; |
2fe17c10 | 1294 | |
7b9cff46 | 1295 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 1296 | if (error) |
2fe17c10 | 1297 | goto out_qunlock; |
d9be0cda AD |
1298 | |
1299 | /* check if the selected rgrp limits our max_blks further */ | |
725d0e9d AG |
1300 | if (ip->i_res.rs_reserved < max_blks) |
1301 | max_blks = ip->i_res.rs_reserved; | |
d9be0cda AD |
1302 | |
1303 | /* Almost done. Calculate bytes that can be written using | |
1304 | * max_blks. We also recompute max_bytes, data_blocks and | |
1305 | * ind_blocks */ | |
1306 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
1307 | &ind_blocks, max_blks); | |
2fe17c10 CH |
1308 | |
1309 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 1310 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
1311 | if (gfs2_is_jdata(ip)) |
1312 | rblocks += data_blocks ? data_blocks : 1; | |
1313 | ||
1314 | error = gfs2_trans_begin(sdp, rblocks, | |
45eb0504 | 1315 | PAGE_SIZE >> inode->i_blkbits); |
2fe17c10 CH |
1316 | if (error) |
1317 | goto out_trans_fail; | |
1318 | ||
1319 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
1320 | gfs2_trans_end(sdp); | |
1321 | ||
1322 | if (error) | |
1323 | goto out_trans_fail; | |
1324 | ||
1325 | len -= max_bytes; | |
1326 | offset += max_bytes; | |
1327 | gfs2_inplace_release(ip); | |
1328 | gfs2_quota_unlock(ip); | |
2fe17c10 | 1329 | } |
4442f2e0 | 1330 | |
0a6a4abc | 1331 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) |
1885867b | 1332 | i_size_write(inode, pos + count); |
0a6a4abc AG |
1333 | file_update_time(file); |
1334 | mark_inode_dirty(inode); | |
1885867b | 1335 | |
dde0c2e7 CH |
1336 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
1337 | return vfs_fsync_range(file, pos, pos + count - 1, | |
1338 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
1339 | return 0; | |
2fe17c10 CH |
1340 | |
1341 | out_trans_fail: | |
1342 | gfs2_inplace_release(ip); | |
1343 | out_qunlock: | |
1344 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
1345 | return error; |
1346 | } | |
1347 | ||
1348 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
1349 | { | |
1350 | struct inode *inode = file_inode(file); | |
d4d7fc12 | 1351 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9c9f1159 AP |
1352 | struct gfs2_inode *ip = GFS2_I(inode); |
1353 | struct gfs2_holder gh; | |
1354 | int ret; | |
1355 | ||
4e56a641 | 1356 | if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) |
d4d7fc12 AP |
1357 | return -EOPNOTSUPP; |
1358 | /* fallocate is needed by gfs2_grow to reserve space in the rindex */ | |
1359 | if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) | |
9c9f1159 AP |
1360 | return -EOPNOTSUPP; |
1361 | ||
5955102c | 1362 | inode_lock(inode); |
9c9f1159 AP |
1363 | |
1364 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1365 | ret = gfs2_glock_nq(&gh); | |
1366 | if (ret) | |
1367 | goto out_uninit; | |
1368 | ||
1369 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
1370 | (offset + len) > inode->i_size) { | |
1371 | ret = inode_newsize_ok(inode, offset + len); | |
1372 | if (ret) | |
1373 | goto out_unlock; | |
1374 | } | |
1375 | ||
1376 | ret = get_write_access(inode); | |
1377 | if (ret) | |
1378 | goto out_unlock; | |
1379 | ||
4e56a641 AG |
1380 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
1381 | ret = __gfs2_punch_hole(file, offset, len); | |
1382 | } else { | |
4e56a641 | 1383 | ret = __gfs2_fallocate(file, mode, offset, len); |
4e56a641 AG |
1384 | if (ret) |
1385 | gfs2_rs_deltree(&ip->i_res); | |
1386 | } | |
a097dc7e | 1387 | |
9c9f1159 | 1388 | put_write_access(inode); |
2fe17c10 | 1389 | out_unlock: |
a0846a53 | 1390 | gfs2_glock_dq(&gh); |
2fe17c10 | 1391 | out_uninit: |
a0846a53 | 1392 | gfs2_holder_uninit(&gh); |
5955102c | 1393 | inode_unlock(inode); |
9c9f1159 | 1394 | return ret; |
2fe17c10 CH |
1395 | } |
1396 | ||
f1ea6f4e BP |
1397 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
1398 | struct file *out, loff_t *ppos, | |
1399 | size_t len, unsigned int flags) | |
1400 | { | |
2fba46a0 | 1401 | ssize_t ret; |
f1ea6f4e | 1402 | |
f1ea6f4e BP |
1403 | gfs2_size_hint(out, *ppos, len); |
1404 | ||
2fba46a0 | 1405 | ret = iter_file_splice_write(pipe, out, ppos, len, flags); |
2fba46a0 | 1406 | return ret; |
f1ea6f4e BP |
1407 | } |
1408 | ||
f057f6cd SW |
1409 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
1410 | ||
b3b94faa DT |
1411 | /** |
1412 | * gfs2_lock - acquire/release a posix lock on a file | |
1413 | * @file: the file pointer | |
1414 | * @cmd: either modify or retrieve lock state, possibly wait | |
1415 | * @fl: type and range of lock | |
1416 | * | |
1417 | * Returns: errno | |
1418 | */ | |
1419 | ||
1420 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1421 | { | |
feaa7bba SW |
1422 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1423 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1424 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1425 | |
b3b94faa DT |
1426 | if (!(fl->fl_flags & FL_POSIX)) |
1427 | return -ENOLCK; | |
586759f0 ME |
1428 | if (cmd == F_CANCELLK) { |
1429 | /* Hack: */ | |
1430 | cmd = F_SETLK; | |
1431 | fl->fl_type = F_UNLCK; | |
1432 | } | |
eb43e660 | 1433 | if (unlikely(gfs2_withdrawn(sdp))) { |
c2952d20 | 1434 | if (fl->fl_type == F_UNLCK) |
4f656367 | 1435 | locks_lock_file_wait(file, fl); |
f057f6cd | 1436 | return -EIO; |
c2952d20 | 1437 | } |
b3b94faa | 1438 | if (IS_GETLK(cmd)) |
f057f6cd | 1439 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1440 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1441 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1442 | else |
f057f6cd | 1443 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1444 | } |
1445 | ||
56535dc6 AG |
1446 | static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh) |
1447 | { | |
4ad02083 | 1448 | struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl); |
56535dc6 AG |
1449 | |
1450 | /* | |
1451 | * Make sure gfs2_glock_put() won't sleep under the file->f_lock | |
1452 | * spinlock. | |
1453 | */ | |
1454 | ||
56535dc6 AG |
1455 | spin_lock(&file->f_lock); |
1456 | gfs2_holder_uninit(fl_gh); | |
1457 | spin_unlock(&file->f_lock); | |
1458 | gfs2_glock_put(gl); | |
1459 | } | |
1460 | ||
b3b94faa DT |
1461 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1462 | { | |
5c676f6d | 1463 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1464 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1465 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1466 | struct gfs2_glock *gl; |
1467 | unsigned int state; | |
b58bf407 | 1468 | u16 flags; |
b3b94faa | 1469 | int error = 0; |
2ddfbdd6 | 1470 | int sleeptime; |
b3b94faa DT |
1471 | |
1472 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
b582d5f0 AG |
1473 | flags = GL_EXACT | GL_NOPID; |
1474 | if (!IS_SETLKW(cmd)) | |
1475 | flags |= LM_FLAG_TRY_1CB; | |
b3b94faa | 1476 | |
f55ab26a | 1477 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa | 1478 | |
283c9a97 | 1479 | if (gfs2_holder_initialized(fl_gh)) { |
4d62d3f7 | 1480 | struct file_lock request; |
b3b94faa DT |
1481 | if (fl_gh->gh_state == state) |
1482 | goto out; | |
4d62d3f7 N |
1483 | locks_init_lock(&request); |
1484 | request.fl_type = F_UNLCK; | |
1485 | request.fl_flags = FL_FLOCK; | |
1486 | locks_lock_file_wait(file, &request); | |
5bef3e7c | 1487 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1488 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1489 | } else { |
6802e340 SW |
1490 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1491 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1492 | if (error) |
1493 | goto out; | |
56535dc6 | 1494 | spin_lock(&file->f_lock); |
b4c20166 | 1495 | gfs2_holder_init(gl, state, flags, fl_gh); |
56535dc6 | 1496 | spin_unlock(&file->f_lock); |
b4c20166 | 1497 | gfs2_glock_put(gl); |
b3b94faa | 1498 | } |
2ddfbdd6 BP |
1499 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1500 | error = gfs2_glock_nq(fl_gh); | |
1501 | if (error != GLR_TRYFAILED) | |
1502 | break; | |
b582d5f0 AG |
1503 | fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB; |
1504 | fl_gh->gh_flags |= LM_FLAG_TRY; | |
2ddfbdd6 BP |
1505 | msleep(sleeptime); |
1506 | } | |
b3b94faa | 1507 | if (error) { |
56535dc6 | 1508 | __flock_holder_uninit(file, fl_gh); |
b3b94faa DT |
1509 | if (error == GLR_TRYFAILED) |
1510 | error = -EAGAIN; | |
1511 | } else { | |
4f656367 | 1512 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1513 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1514 | } |
1515 | ||
420b9e5e | 1516 | out: |
f55ab26a | 1517 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1518 | return error; |
1519 | } | |
1520 | ||
1521 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1522 | { | |
5c676f6d | 1523 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1524 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1525 | ||
f55ab26a | 1526 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1527 | locks_lock_file_wait(file, fl); |
6df9f9a2 | 1528 | if (gfs2_holder_initialized(fl_gh)) { |
2ddfbdd6 | 1529 | gfs2_glock_dq(fl_gh); |
56535dc6 | 1530 | __flock_holder_uninit(file, fl_gh); |
0a33443b | 1531 | } |
f55ab26a | 1532 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1533 | } |
1534 | ||
1535 | /** | |
1536 | * gfs2_flock - acquire/release a flock lock on a file | |
1537 | * @file: the file pointer | |
1538 | * @cmd: either modify or retrieve lock state, possibly wait | |
1539 | * @fl: type and range of lock | |
1540 | * | |
1541 | * Returns: errno | |
1542 | */ | |
1543 | ||
1544 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1545 | { | |
b3b94faa DT |
1546 | if (!(fl->fl_flags & FL_FLOCK)) |
1547 | return -ENOLCK; | |
b3b94faa | 1548 | |
b3b94faa DT |
1549 | if (fl->fl_type == F_UNLCK) { |
1550 | do_unflock(file, fl); | |
1551 | return 0; | |
d00223f1 | 1552 | } else { |
b3b94faa | 1553 | return do_flock(file, cmd, fl); |
d00223f1 | 1554 | } |
b3b94faa DT |
1555 | } |
1556 | ||
10d21988 | 1557 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1558 | .llseek = gfs2_llseek, |
967bcc91 | 1559 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1560 | .write_iter = gfs2_file_write_iter, |
3e08773c | 1561 | .iopoll = iocb_bio_iopoll, |
26c1a574 | 1562 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1563 | .compat_ioctl = gfs2_compat_ioctl, |
26c1a574 SW |
1564 | .mmap = gfs2_mmap, |
1565 | .open = gfs2_open, | |
df3fd117 | 1566 | .release = gfs2_release, |
26c1a574 SW |
1567 | .fsync = gfs2_fsync, |
1568 | .lock = gfs2_lock, | |
26c1a574 | 1569 | .flock = gfs2_flock, |
82c156f8 | 1570 | .splice_read = generic_file_splice_read, |
f42a69fa | 1571 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1572 | .setlease = simple_nosetlease, |
2fe17c10 | 1573 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1574 | }; |
1575 | ||
10d21988 | 1576 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1577 | .iterate_shared = gfs2_readdir, |
26c1a574 | 1578 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1579 | .compat_ioctl = gfs2_compat_ioctl, |
26c1a574 | 1580 | .open = gfs2_open, |
df3fd117 | 1581 | .release = gfs2_release, |
26c1a574 SW |
1582 | .fsync = gfs2_fsync, |
1583 | .lock = gfs2_lock, | |
1584 | .flock = gfs2_flock, | |
6038f373 | 1585 | .llseek = default_llseek, |
b3b94faa DT |
1586 | }; |
1587 | ||
f057f6cd SW |
1588 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1589 | ||
10d21988 | 1590 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1591 | .llseek = gfs2_llseek, |
967bcc91 | 1592 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1593 | .write_iter = gfs2_file_write_iter, |
3e08773c | 1594 | .iopoll = iocb_bio_iopoll, |
c97bfe43 | 1595 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1596 | .compat_ioctl = gfs2_compat_ioctl, |
c97bfe43 WC |
1597 | .mmap = gfs2_mmap, |
1598 | .open = gfs2_open, | |
df3fd117 | 1599 | .release = gfs2_release, |
c97bfe43 | 1600 | .fsync = gfs2_fsync, |
82c156f8 | 1601 | .splice_read = generic_file_splice_read, |
f42a69fa | 1602 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1603 | .setlease = generic_setlease, |
2fe17c10 | 1604 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1605 | }; |
1606 | ||
10d21988 | 1607 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1608 | .iterate_shared = gfs2_readdir, |
c97bfe43 | 1609 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1610 | .compat_ioctl = gfs2_compat_ioctl, |
c97bfe43 | 1611 | .open = gfs2_open, |
df3fd117 | 1612 | .release = gfs2_release, |
c97bfe43 | 1613 | .fsync = gfs2_fsync, |
6038f373 | 1614 | .llseek = default_llseek, |
c97bfe43 WC |
1615 | }; |
1616 |