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> |
5970e15d | 18 | #include <linux/filelock.h> |
5c676f6d | 19 | #include <linux/gfs2_ondisk.h> |
2fe17c10 CH |
20 | #include <linux/falloc.h> |
21 | #include <linux/swap.h> | |
71b86f56 | 22 | #include <linux/crc32.h> |
33c3de32 | 23 | #include <linux/writeback.h> |
7c0f6ba6 | 24 | #include <linux/uaccess.h> |
f057f6cd SW |
25 | #include <linux/dlm.h> |
26 | #include <linux/dlm_plock.h> | |
2ddfbdd6 | 27 | #include <linux/delay.h> |
64bc06bb | 28 | #include <linux/backing-dev.h> |
88b631cb | 29 | #include <linux/fileattr.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 | ||
c551f66c LJ |
123 | /* |
124 | * struct 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 | |
5aca2842 DW |
142 | static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags) |
143 | { | |
144 | int i; | |
145 | u32 fsflags = 0; | |
146 | ||
147 | if (S_ISDIR(inode->i_mode)) | |
148 | gfsflags &= ~GFS2_DIF_JDATA; | |
149 | else | |
150 | gfsflags &= ~GFS2_DIF_INHERIT_JDATA; | |
151 | ||
152 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) | |
153 | if (gfsflags & fsflag_gfs2flag[i].gfsflag) | |
154 | fsflags |= fsflag_gfs2flag[i].fsflag; | |
155 | return fsflags; | |
156 | } | |
157 | ||
88b631cb | 158 | int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa) |
71b86f56 | 159 | { |
88b631cb | 160 | struct inode *inode = d_inode(dentry); |
feaa7bba | 161 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 | 162 | struct gfs2_holder gh; |
5aca2842 DW |
163 | int error; |
164 | u32 fsflags; | |
71b86f56 | 165 | |
88b631cb MS |
166 | if (d_is_special(dentry)) |
167 | return -ENOTTY; | |
168 | ||
719ee344 SW |
169 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
170 | error = gfs2_glock_nq(&gh); | |
71b86f56 | 171 | if (error) |
9c7fe835 | 172 | goto out_uninit; |
907b9bce | 173 | |
5aca2842 | 174 | fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags); |
b16f7e57 | 175 | |
88b631cb | 176 | fileattr_fill_flags(fa, fsflags); |
71b86f56 | 177 | |
3cc3f710 | 178 | gfs2_glock_dq(&gh); |
9c7fe835 | 179 | out_uninit: |
71b86f56 SW |
180 | gfs2_holder_uninit(&gh); |
181 | return error; | |
182 | } | |
183 | ||
6b124d8d SW |
184 | void gfs2_set_inode_flags(struct inode *inode) |
185 | { | |
186 | struct gfs2_inode *ip = GFS2_I(inode); | |
6b124d8d SW |
187 | unsigned int flags = inode->i_flags; |
188 | ||
9964afbb SW |
189 | flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC); |
190 | if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode)) | |
01e64ee4 | 191 | flags |= S_NOSEC; |
383f01fb | 192 | if (ip->i_diskflags & GFS2_DIF_IMMUTABLE) |
6b124d8d | 193 | flags |= S_IMMUTABLE; |
383f01fb | 194 | if (ip->i_diskflags & GFS2_DIF_APPENDONLY) |
6b124d8d | 195 | flags |= S_APPEND; |
383f01fb | 196 | if (ip->i_diskflags & GFS2_DIF_NOATIME) |
6b124d8d | 197 | flags |= S_NOATIME; |
383f01fb | 198 | if (ip->i_diskflags & GFS2_DIF_SYNC) |
6b124d8d SW |
199 | flags |= S_SYNC; |
200 | inode->i_flags = flags; | |
201 | } | |
202 | ||
71b86f56 SW |
203 | /* Flags that can be set by user space */ |
204 | #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \ | |
71b86f56 SW |
205 | GFS2_DIF_IMMUTABLE| \ |
206 | GFS2_DIF_APPENDONLY| \ | |
207 | GFS2_DIF_NOATIME| \ | |
208 | GFS2_DIF_SYNC| \ | |
23d0bb83 | 209 | GFS2_DIF_TOPDIR| \ |
71b86f56 SW |
210 | GFS2_DIF_INHERIT_JDATA) |
211 | ||
212 | /** | |
9dd868e1 | 213 | * do_gfs2_set_flags - set flags on an inode |
0f1616f6 | 214 | * @inode: The inode |
9dd868e1 | 215 | * @reqflags: The flags to set |
71b86f56 SW |
216 | * @mask: Indicates which flags are valid |
217 | * | |
218 | */ | |
a500bd31 | 219 | static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask) |
71b86f56 | 220 | { |
feaa7bba SW |
221 | struct gfs2_inode *ip = GFS2_I(inode); |
222 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
71b86f56 SW |
223 | struct buffer_head *bh; |
224 | struct gfs2_holder gh; | |
225 | int error; | |
88b631cb | 226 | u32 new_flags, flags; |
71b86f56 | 227 | |
f58ba889 MS |
228 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
229 | if (error) | |
88b631cb | 230 | return error; |
7df0e039 SW |
231 | |
232 | error = 0; | |
383f01fb | 233 | flags = ip->i_diskflags; |
55eccc6d | 234 | new_flags = (flags & ~mask) | (reqflags & mask); |
71b86f56 SW |
235 | if ((new_flags ^ flags) == 0) |
236 | goto out; | |
237 | ||
b9cb9813 | 238 | if (!IS_IMMUTABLE(inode)) { |
4609e1f1 | 239 | error = gfs2_permission(&nop_mnt_idmap, inode, MAY_WRITE); |
b9cb9813 SW |
240 | if (error) |
241 | goto out; | |
242 | } | |
5561093e | 243 | if ((flags ^ new_flags) & GFS2_DIF_JDATA) { |
cc555b09 | 244 | if (new_flags & GFS2_DIF_JDATA) |
c1696fb8 | 245 | gfs2_log_flush(sdp, ip->i_gl, |
805c0907 BP |
246 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
247 | GFS2_LFC_SET_FLAGS); | |
5561093e SW |
248 | error = filemap_fdatawrite(inode->i_mapping); |
249 | if (error) | |
250 | goto out; | |
251 | error = filemap_fdatawait(inode->i_mapping); | |
252 | if (error) | |
253 | goto out; | |
cc555b09 BP |
254 | if (new_flags & GFS2_DIF_JDATA) |
255 | gfs2_ordered_del_inode(ip); | |
5561093e | 256 | } |
55eccc6d | 257 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); |
71b86f56 SW |
258 | if (error) |
259 | goto out; | |
55eccc6d SW |
260 | error = gfs2_meta_inode_buffer(ip, &bh); |
261 | if (error) | |
262 | goto out_trans_end; | |
8a8b8d91 | 263 | inode_set_ctime_current(inode); |
350a9b0a | 264 | gfs2_trans_add_meta(ip->i_gl, bh); |
383f01fb | 265 | ip->i_diskflags = new_flags; |
539e5d6b | 266 | gfs2_dinode_out(ip, bh->b_data); |
71b86f56 | 267 | brelse(bh); |
6b124d8d | 268 | gfs2_set_inode_flags(inode); |
5561093e | 269 | gfs2_set_aops(inode); |
55eccc6d SW |
270 | out_trans_end: |
271 | gfs2_trans_end(sdp); | |
71b86f56 SW |
272 | out: |
273 | gfs2_glock_dq_uninit(&gh); | |
274 | return error; | |
275 | } | |
276 | ||
8782a9ae | 277 | int gfs2_fileattr_set(struct mnt_idmap *idmap, |
88b631cb | 278 | struct dentry *dentry, struct fileattr *fa) |
71b86f56 | 279 | { |
88b631cb MS |
280 | struct inode *inode = d_inode(dentry); |
281 | u32 fsflags = fa->flags, gfsflags = 0; | |
b16f7e57 AG |
282 | u32 mask; |
283 | int i; | |
7df0e039 | 284 | |
88b631cb MS |
285 | if (d_is_special(dentry)) |
286 | return -ENOTTY; | |
287 | ||
288 | if (fileattr_has_fsx(fa)) | |
289 | return -EOPNOTSUPP; | |
7df0e039 | 290 | |
b16f7e57 AG |
291 | for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) { |
292 | if (fsflags & fsflag_gfs2flag[i].fsflag) { | |
293 | fsflags &= ~fsflag_gfs2flag[i].fsflag; | |
294 | gfsflags |= fsflag_gfs2flag[i].gfsflag; | |
295 | } | |
296 | } | |
297 | if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET) | |
298 | return -EINVAL; | |
299 | ||
300 | mask = GFS2_FLAGS_USER_SET; | |
301 | if (S_ISDIR(inode->i_mode)) { | |
302 | mask &= ~GFS2_DIF_JDATA; | |
303 | } else { | |
304 | /* The GFS2_DIF_TOPDIR flag is only valid for directories. */ | |
305 | if (gfsflags & GFS2_DIF_TOPDIR) | |
306 | return -EINVAL; | |
307 | mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA); | |
b9af7ca6 | 308 | } |
b16f7e57 | 309 | |
a500bd31 | 310 | return do_gfs2_set_flags(inode, gfsflags, mask); |
71b86f56 SW |
311 | } |
312 | ||
6ddc5c3d SW |
313 | static int gfs2_getlabel(struct file *filp, char __user *label) |
314 | { | |
315 | struct inode *inode = file_inode(filp); | |
316 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
317 | ||
318 | if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN)) | |
319 | return -EFAULT; | |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
b09e593d | 324 | static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
71b86f56 SW |
325 | { |
326 | switch(cmd) { | |
66fc061b SW |
327 | case FITRIM: |
328 | return gfs2_fitrim(filp, (void __user *)arg); | |
6ddc5c3d SW |
329 | case FS_IOC_GETFSLABEL: |
330 | return gfs2_getlabel(filp, (char __user *)arg); | |
71b86f56 | 331 | } |
6ddc5c3d | 332 | |
71b86f56 SW |
333 | return -ENOTTY; |
334 | } | |
335 | ||
8d098070 AB |
336 | #ifdef CONFIG_COMPAT |
337 | static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) | |
338 | { | |
339 | switch(cmd) { | |
8d098070 AB |
340 | /* Keep this list in sync with gfs2_ioctl */ |
341 | case FITRIM: | |
342 | case FS_IOC_GETFSLABEL: | |
343 | break; | |
344 | default: | |
345 | return -ENOIOCTLCMD; | |
346 | } | |
347 | ||
348 | return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg)); | |
349 | } | |
350 | #else | |
351 | #define gfs2_compat_ioctl NULL | |
352 | #endif | |
353 | ||
da1dfb6a SW |
354 | /** |
355 | * gfs2_size_hint - Give a hint to the size of a write request | |
9dd868e1 | 356 | * @filep: The struct file |
da1dfb6a SW |
357 | * @offset: The file offset of the write |
358 | * @size: The length of the write | |
359 | * | |
360 | * When we are about to do a write, this function records the total | |
361 | * write size in order to provide a suitable hint to the lower layers | |
362 | * about how many blocks will be required. | |
363 | * | |
364 | */ | |
365 | ||
366 | static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size) | |
367 | { | |
496ad9aa | 368 | struct inode *inode = file_inode(filep); |
da1dfb6a SW |
369 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
370 | struct gfs2_inode *ip = GFS2_I(inode); | |
371 | size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift; | |
372 | int hint = min_t(size_t, INT_MAX, blks); | |
373 | ||
21f09c43 AG |
374 | if (hint > atomic_read(&ip->i_sizehint)) |
375 | atomic_set(&ip->i_sizehint, hint); | |
da1dfb6a SW |
376 | } |
377 | ||
3cc3f710 | 378 | /** |
35af80ae | 379 | * gfs2_allocate_page_backing - Allocate blocks for a write fault |
3cc3f710 | 380 | * @page: The (locked) page to allocate backing for |
f53056c4 | 381 | * @length: Size of the allocation |
3cc3f710 | 382 | * |
35af80ae CH |
383 | * We try to allocate all the blocks required for the page in one go. This |
384 | * might fail for various reasons, so we keep trying until all the blocks to | |
385 | * back this page are allocated. If some of the blocks are already allocated, | |
386 | * that is ok too. | |
3cc3f710 | 387 | */ |
f53056c4 | 388 | static int gfs2_allocate_page_backing(struct page *page, unsigned int length) |
3cc3f710 | 389 | { |
35af80ae | 390 | u64 pos = page_offset(page); |
3cc3f710 SW |
391 | |
392 | do { | |
35af80ae CH |
393 | struct iomap iomap = { }; |
394 | ||
54992257 | 395 | if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap)) |
3cc3f710 | 396 | return -EIO; |
35af80ae | 397 | |
f53056c4 AG |
398 | if (length < iomap.length) |
399 | iomap.length = length; | |
400 | length -= iomap.length; | |
35af80ae | 401 | pos += iomap.length; |
f53056c4 | 402 | } while (length > 0); |
35af80ae | 403 | |
3cc3f710 SW |
404 | return 0; |
405 | } | |
406 | ||
407 | /** | |
408 | * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable | |
9dd868e1 | 409 | * @vmf: The virtual memory fault containing the page to become writable |
3cc3f710 SW |
410 | * |
411 | * When the page becomes writable, we need to ensure that we have | |
412 | * blocks allocated on disk to back that page. | |
413 | */ | |
414 | ||
109dbb1e | 415 | static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf) |
3cc3f710 | 416 | { |
c2ec175c | 417 | struct page *page = vmf->page; |
11bac800 | 418 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3cc3f710 SW |
419 | struct gfs2_inode *ip = GFS2_I(inode); |
420 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
f7e4c610 | 421 | struct gfs2_alloc_parms ap = {}; |
184b4e60 | 422 | u64 offset = page_offset(page); |
3cc3f710 | 423 | unsigned int data_blocks, ind_blocks, rblocks; |
0fc3bcd6 | 424 | vm_fault_t ret = VM_FAULT_LOCKED; |
3cc3f710 | 425 | struct gfs2_holder gh; |
184b4e60 | 426 | unsigned int length; |
13d921e3 | 427 | loff_t size; |
0fc3bcd6 | 428 | int err; |
3cc3f710 | 429 | |
39263d5e | 430 | sb_start_pagefault(inode->i_sb); |
13d921e3 | 431 | |
719ee344 | 432 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
0fc3bcd6 AG |
433 | err = gfs2_glock_nq(&gh); |
434 | if (err) { | |
2ba39cc4 | 435 | ret = vmf_fs_error(err); |
2b3dcf35 | 436 | goto out_uninit; |
0fc3bcd6 | 437 | } |
3cc3f710 | 438 | |
184b4e60 AG |
439 | /* Check page index against inode size */ |
440 | size = i_size_read(inode); | |
441 | if (offset >= size) { | |
0fc3bcd6 | 442 | ret = VM_FAULT_SIGBUS; |
184b4e60 AG |
443 | goto out_unlock; |
444 | } | |
445 | ||
d7c436cd | 446 | /* Update file times before taking page lock */ |
11bac800 | 447 | file_update_time(vmf->vma->vm_file); |
d7c436cd | 448 | |
184b4e60 | 449 | /* page is wholly or partially inside EOF */ |
d3c51c55 AG |
450 | if (size - offset < PAGE_SIZE) |
451 | length = size - offset; | |
184b4e60 AG |
452 | else |
453 | length = PAGE_SIZE; | |
454 | ||
455 | gfs2_size_hint(vmf->vma->vm_file, offset, length); | |
456 | ||
9c538837 SW |
457 | set_bit(GLF_DIRTY, &ip->i_gl->gl_flags); |
458 | set_bit(GIF_SW_PAGED, &ip->i_flags); | |
459 | ||
184b4e60 AG |
460 | /* |
461 | * iomap_writepage / iomap_writepages currently don't support inline | |
462 | * files, so always unstuff here. | |
463 | */ | |
464 | ||
465 | if (!gfs2_is_stuffed(ip) && | |
466 | !gfs2_write_alloc_required(ip, offset, length)) { | |
13d921e3 SW |
467 | lock_page(page); |
468 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | |
0fc3bcd6 | 469 | ret = VM_FAULT_NOPAGE; |
13d921e3 SW |
470 | unlock_page(page); |
471 | } | |
3cc3f710 | 472 | goto out_unlock; |
13d921e3 SW |
473 | } |
474 | ||
0fc3bcd6 AG |
475 | err = gfs2_rindex_update(sdp); |
476 | if (err) { | |
2ba39cc4 | 477 | ret = vmf_fs_error(err); |
6dbd8224 | 478 | goto out_unlock; |
0fc3bcd6 | 479 | } |
6dbd8224 | 480 | |
184b4e60 | 481 | gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks); |
7b9cff46 | 482 | ap.target = data_blocks + ind_blocks; |
0fc3bcd6 AG |
483 | err = gfs2_quota_lock_check(ip, &ap); |
484 | if (err) { | |
2ba39cc4 | 485 | ret = vmf_fs_error(err); |
b8fbf471 | 486 | goto out_unlock; |
0fc3bcd6 AG |
487 | } |
488 | err = gfs2_inplace_reserve(ip, &ap); | |
489 | if (err) { | |
2ba39cc4 | 490 | ret = vmf_fs_error(err); |
3cc3f710 | 491 | goto out_quota_unlock; |
0fc3bcd6 | 492 | } |
3cc3f710 SW |
493 | |
494 | rblocks = RES_DINODE + ind_blocks; | |
495 | if (gfs2_is_jdata(ip)) | |
496 | rblocks += data_blocks ? data_blocks : 1; | |
bf97b673 | 497 | if (ind_blocks || data_blocks) { |
3cc3f710 | 498 | rblocks += RES_STATFS + RES_QUOTA; |
71f890f7 | 499 | rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
bf97b673 | 500 | } |
0fc3bcd6 AG |
501 | err = gfs2_trans_begin(sdp, rblocks, 0); |
502 | if (err) { | |
2ba39cc4 | 503 | ret = vmf_fs_error(err); |
3cc3f710 | 504 | goto out_trans_fail; |
0fc3bcd6 | 505 | } |
3cc3f710 | 506 | |
64090cbe AG |
507 | /* Unstuff, if required, and allocate backing blocks for page */ |
508 | if (gfs2_is_stuffed(ip)) { | |
7a607a41 | 509 | err = gfs2_unstuff_dinode(ip); |
64090cbe | 510 | if (err) { |
2ba39cc4 | 511 | ret = vmf_fs_error(err); |
64090cbe AG |
512 | goto out_trans_end; |
513 | } | |
514 | } | |
515 | ||
3cc3f710 | 516 | lock_page(page); |
13d921e3 SW |
517 | /* If truncated, we must retry the operation, we may have raced |
518 | * with the glock demotion code. | |
519 | */ | |
0fc3bcd6 AG |
520 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { |
521 | ret = VM_FAULT_NOPAGE; | |
64090cbe | 522 | goto out_page_locked; |
0fc3bcd6 | 523 | } |
13d921e3 | 524 | |
0fc3bcd6 AG |
525 | err = gfs2_allocate_page_backing(page, length); |
526 | if (err) | |
2ba39cc4 | 527 | ret = vmf_fs_error(err); |
3cc3f710 | 528 | |
64090cbe | 529 | out_page_locked: |
0fc3bcd6 | 530 | if (ret != VM_FAULT_LOCKED) |
13d921e3 | 531 | unlock_page(page); |
64090cbe | 532 | out_trans_end: |
3cc3f710 SW |
533 | gfs2_trans_end(sdp); |
534 | out_trans_fail: | |
535 | gfs2_inplace_release(ip); | |
536 | out_quota_unlock: | |
537 | gfs2_quota_unlock(ip); | |
3cc3f710 SW |
538 | out_unlock: |
539 | gfs2_glock_dq(&gh); | |
2b3dcf35 | 540 | out_uninit: |
3cc3f710 | 541 | gfs2_holder_uninit(&gh); |
0fc3bcd6 | 542 | if (ret == VM_FAULT_LOCKED) { |
13d921e3 | 543 | set_page_dirty(page); |
1d1d1a76 | 544 | wait_for_stable_page(page); |
13d921e3 | 545 | } |
39263d5e | 546 | sb_end_pagefault(inode->i_sb); |
0fc3bcd6 | 547 | return ret; |
3cc3f710 SW |
548 | } |
549 | ||
20f82999 AG |
550 | static vm_fault_t gfs2_fault(struct vm_fault *vmf) |
551 | { | |
552 | struct inode *inode = file_inode(vmf->vma->vm_file); | |
553 | struct gfs2_inode *ip = GFS2_I(inode); | |
554 | struct gfs2_holder gh; | |
555 | vm_fault_t ret; | |
556 | int err; | |
557 | ||
d5b81454 | 558 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
20f82999 AG |
559 | err = gfs2_glock_nq(&gh); |
560 | if (err) { | |
2ba39cc4 | 561 | ret = vmf_fs_error(err); |
20f82999 AG |
562 | goto out_uninit; |
563 | } | |
564 | ret = filemap_fault(vmf); | |
565 | gfs2_glock_dq(&gh); | |
566 | out_uninit: | |
567 | gfs2_holder_uninit(&gh); | |
568 | return ret; | |
569 | } | |
570 | ||
f0f37e2f | 571 | static const struct vm_operations_struct gfs2_vm_ops = { |
20f82999 | 572 | .fault = gfs2_fault, |
f1820361 | 573 | .map_pages = filemap_map_pages, |
3cc3f710 SW |
574 | .page_mkwrite = gfs2_page_mkwrite, |
575 | }; | |
576 | ||
b3b94faa | 577 | /** |
c551f66c | 578 | * gfs2_mmap |
b3b94faa DT |
579 | * @file: The file to map |
580 | * @vma: The VMA which described the mapping | |
581 | * | |
48bf2b17 SW |
582 | * There is no need to get a lock here unless we should be updating |
583 | * atime. We ignore any locking errors since the only consequence is | |
584 | * a missed atime update (which will just be deferred until later). | |
585 | * | |
586 | * Returns: 0 | |
b3b94faa DT |
587 | */ |
588 | ||
589 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
590 | { | |
feaa7bba | 591 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
b3b94faa | 592 | |
b9c93bb7 SW |
593 | if (!(file->f_flags & O_NOATIME) && |
594 | !IS_NOATIME(&ip->i_inode)) { | |
48bf2b17 SW |
595 | struct gfs2_holder i_gh; |
596 | int error; | |
b3b94faa | 597 | |
3d162688 BM |
598 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
599 | &i_gh); | |
b9c93bb7 SW |
600 | if (error) |
601 | return error; | |
3d162688 BM |
602 | /* grab lock to update inode */ |
603 | gfs2_glock_dq_uninit(&i_gh); | |
604 | file_accessed(file); | |
48bf2b17 | 605 | } |
3cc3f710 | 606 | vma->vm_ops = &gfs2_vm_ops; |
b3b94faa | 607 | |
48bf2b17 | 608 | return 0; |
b3b94faa DT |
609 | } |
610 | ||
611 | /** | |
6d4ade98 SW |
612 | * gfs2_open_common - This is common to open and atomic_open |
613 | * @inode: The inode being opened | |
614 | * @file: The file being opened | |
b3b94faa | 615 | * |
6d4ade98 SW |
616 | * This maybe called under a glock or not depending upon how it has |
617 | * been called. We must always be called under a glock for regular | |
618 | * files, however. For other file types, it does not matter whether | |
619 | * we hold the glock or not. | |
620 | * | |
621 | * Returns: Error code or 0 for success | |
b3b94faa DT |
622 | */ |
623 | ||
6d4ade98 | 624 | int gfs2_open_common(struct inode *inode, struct file *file) |
b3b94faa | 625 | { |
b3b94faa | 626 | struct gfs2_file *fp; |
6d4ade98 SW |
627 | int ret; |
628 | ||
629 | if (S_ISREG(inode->i_mode)) { | |
630 | ret = generic_file_open(inode, file); | |
631 | if (ret) | |
632 | return ret; | |
7b7b06d5 CH |
633 | |
634 | if (!gfs2_is_jdata(GFS2_I(inode))) | |
635 | file->f_mode |= FMODE_CAN_ODIRECT; | |
6d4ade98 | 636 | } |
b3b94faa | 637 | |
6d4ade98 | 638 | fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS); |
b3b94faa DT |
639 | if (!fp) |
640 | return -ENOMEM; | |
641 | ||
f55ab26a | 642 | mutex_init(&fp->f_fl_mutex); |
b3b94faa | 643 | |
feaa7bba | 644 | gfs2_assert_warn(GFS2_SB(inode), !file->private_data); |
5c676f6d | 645 | file->private_data = fp; |
2fba46a0 BP |
646 | if (file->f_mode & FMODE_WRITE) { |
647 | ret = gfs2_qa_get(GFS2_I(inode)); | |
648 | if (ret) | |
649 | goto fail; | |
650 | } | |
6d4ade98 | 651 | return 0; |
2fba46a0 BP |
652 | |
653 | fail: | |
654 | kfree(file->private_data); | |
655 | file->private_data = NULL; | |
656 | return ret; | |
6d4ade98 SW |
657 | } |
658 | ||
659 | /** | |
660 | * gfs2_open - open a file | |
661 | * @inode: the inode to open | |
662 | * @file: the struct file for this opening | |
663 | * | |
664 | * After atomic_open, this function is only used for opening files | |
665 | * which are already cached. We must still get the glock for regular | |
666 | * files to ensure that we have the file size uptodate for the large | |
667 | * file check which is in the common code. That is only an issue for | |
668 | * regular files though. | |
669 | * | |
670 | * Returns: errno | |
671 | */ | |
672 | ||
673 | static int gfs2_open(struct inode *inode, struct file *file) | |
674 | { | |
675 | struct gfs2_inode *ip = GFS2_I(inode); | |
676 | struct gfs2_holder i_gh; | |
677 | int error; | |
678 | bool need_unlock = false; | |
b3b94faa | 679 | |
b60623c2 | 680 | if (S_ISREG(ip->i_inode.i_mode)) { |
b3b94faa DT |
681 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, |
682 | &i_gh); | |
683 | if (error) | |
6d4ade98 SW |
684 | return error; |
685 | need_unlock = true; | |
686 | } | |
b3b94faa | 687 | |
6d4ade98 | 688 | error = gfs2_open_common(inode, file); |
b3b94faa | 689 | |
6d4ade98 | 690 | if (need_unlock) |
b3b94faa | 691 | gfs2_glock_dq_uninit(&i_gh); |
b3b94faa | 692 | |
b3b94faa DT |
693 | return error; |
694 | } | |
695 | ||
696 | /** | |
df3fd117 | 697 | * gfs2_release - called to close a struct file |
b3b94faa DT |
698 | * @inode: the inode the struct file belongs to |
699 | * @file: the struct file being closed | |
700 | * | |
701 | * Returns: errno | |
702 | */ | |
703 | ||
df3fd117 | 704 | static int gfs2_release(struct inode *inode, struct file *file) |
b3b94faa | 705 | { |
0a305e49 | 706 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa | 707 | |
8e2e0047 | 708 | kfree(file->private_data); |
5c676f6d | 709 | file->private_data = NULL; |
b3b94faa | 710 | |
d3add1a9 BP |
711 | if (file->f_mode & FMODE_WRITE) { |
712 | if (gfs2_rs_active(&ip->i_res)) | |
7336905a | 713 | gfs2_rs_delete(ip); |
1595548f | 714 | gfs2_qa_put(ip); |
d3add1a9 | 715 | } |
b3b94faa DT |
716 | return 0; |
717 | } | |
718 | ||
719 | /** | |
720 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
02c24a82 JB |
721 | * @file: the file that points to the dentry |
722 | * @start: the start position in the file to sync | |
723 | * @end: the end position in the file to sync | |
dba898b0 | 724 | * @datasync: set if we can ignore timestamp changes |
b3b94faa | 725 | * |
2f0264d5 SW |
726 | * We split the data flushing here so that we don't wait for the data |
727 | * until after we've also sent the metadata to disk. Note that for | |
728 | * data=ordered, we will write & wait for the data at the log flush | |
729 | * stage anyway, so this is unlikely to make much of a difference | |
730 | * except in the data=writeback case. | |
731 | * | |
732 | * If the fdatawrite fails due to any reason except -EIO, we will | |
733 | * continue the remainder of the fsync, although we'll still report | |
734 | * the error at the end. This is to match filemap_write_and_wait_range() | |
735 | * behaviour. | |
34126f9f | 736 | * |
b3b94faa DT |
737 | * Returns: errno |
738 | */ | |
739 | ||
02c24a82 JB |
740 | static int gfs2_fsync(struct file *file, loff_t start, loff_t end, |
741 | int datasync) | |
b3b94faa | 742 | { |
2f0264d5 SW |
743 | struct address_space *mapping = file->f_mapping; |
744 | struct inode *inode = mapping->host; | |
3aac630b | 745 | int sync_state = inode->i_state & I_DIRTY; |
dba898b0 | 746 | struct gfs2_inode *ip = GFS2_I(inode); |
87654896 | 747 | int ret = 0, ret1 = 0; |
b3b94faa | 748 | |
2f0264d5 SW |
749 | if (mapping->nrpages) { |
750 | ret1 = filemap_fdatawrite_range(mapping, start, end); | |
751 | if (ret1 == -EIO) | |
752 | return ret1; | |
753 | } | |
02c24a82 | 754 | |
0c901809 BM |
755 | if (!gfs2_is_jdata(ip)) |
756 | sync_state &= ~I_DIRTY_PAGES; | |
dba898b0 | 757 | if (datasync) |
3aac630b | 758 | sync_state &= ~I_DIRTY_SYNC; |
b3b94faa | 759 | |
dba898b0 SW |
760 | if (sync_state) { |
761 | ret = sync_inode_metadata(inode, 1); | |
b5b24d7a | 762 | if (ret) |
dba898b0 | 763 | return ret; |
f1818529 | 764 | if (gfs2_is_jdata(ip)) |
d07a6ac7 JL |
765 | ret = file_write_and_wait(file); |
766 | if (ret) | |
767 | return ret; | |
b5b24d7a | 768 | gfs2_ail_flush(ip->i_gl, 1); |
33c3de32 SW |
769 | } |
770 | ||
2f0264d5 | 771 | if (mapping->nrpages) |
d07a6ac7 | 772 | ret = file_fdatawait_range(file, start, end); |
2f0264d5 SW |
773 | |
774 | return ret ? ret : ret1; | |
b3b94faa DT |
775 | } |
776 | ||
72382264 | 777 | static inline bool should_fault_in_pages(struct iov_iter *i, |
324d116c | 778 | struct kiocb *iocb, |
00bfe02f AG |
779 | size_t *prev_count, |
780 | size_t *window_size) | |
781 | { | |
00bfe02f | 782 | size_t count = iov_iter_count(i); |
bb7f5d96 | 783 | size_t size, offs; |
00bfe02f | 784 | |
fa5dfa64 | 785 | if (!count) |
00bfe02f | 786 | return false; |
fcb14cb1 | 787 | if (!user_backed_iter(i)) |
00bfe02f AG |
788 | return false; |
789 | ||
fa58cc88 AG |
790 | /* |
791 | * Try to fault in multiple pages initially. When that doesn't result | |
792 | * in any progress, fall back to a single page. | |
793 | */ | |
bb7f5d96 | 794 | size = PAGE_SIZE; |
324d116c | 795 | offs = offset_in_page(iocb->ki_pos); |
fa58cc88 | 796 | if (*prev_count != count) { |
bb7f5d96 | 797 | size_t nr_dirtied; |
00bfe02f | 798 | |
00bfe02f | 799 | nr_dirtied = max(current->nr_dirtied_pause - |
bb7f5d96 | 800 | current->nr_dirtied, 8); |
324d116c | 801 | size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT); |
00bfe02f AG |
802 | } |
803 | ||
804 | *prev_count = count; | |
bb7f5d96 | 805 | *window_size = size - offs; |
00bfe02f AG |
806 | return true; |
807 | } | |
808 | ||
4c5c3010 AG |
809 | static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to, |
810 | struct gfs2_holder *gh) | |
967bcc91 AG |
811 | { |
812 | struct file *file = iocb->ki_filp; | |
813 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); | |
b01b2d72 | 814 | size_t prev_count = 0, window_size = 0; |
42e4c3bd | 815 | size_t read = 0; |
967bcc91 AG |
816 | ssize_t ret; |
817 | ||
b01b2d72 AG |
818 | /* |
819 | * In this function, we disable page faults when we're holding the | |
820 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
821 | * that the inode glock may be dropped, fault in the pages manually, | |
822 | * and retry. | |
823 | * | |
824 | * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger | |
825 | * physical as well as manual page faults, and we need to disable both | |
826 | * kinds. | |
827 | * | |
828 | * For direct I/O, gfs2 takes the inode glock in deferred mode. This | |
829 | * locking mode is compatible with other deferred holders, so multiple | |
830 | * processes and nodes can do direct I/O to a file at the same time. | |
831 | * There's no guarantee that reads or writes will be atomic. Any | |
832 | * coordination among readers and writers needs to happen externally. | |
833 | */ | |
834 | ||
835 | if (!iov_iter_count(to)) | |
967bcc91 AG |
836 | return 0; /* skip atime */ |
837 | ||
4c5c3010 | 838 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh); |
b01b2d72 | 839 | retry: |
4c5c3010 | 840 | ret = gfs2_glock_nq(gh); |
967bcc91 AG |
841 | if (ret) |
842 | goto out_uninit; | |
b01b2d72 AG |
843 | pagefault_disable(); |
844 | to->nofault = true; | |
845 | ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, | |
786f847f | 846 | IOMAP_DIO_PARTIAL, NULL, read); |
b01b2d72 AG |
847 | to->nofault = false; |
848 | pagefault_enable(); | |
72382264 AG |
849 | if (ret <= 0 && ret != -EFAULT) |
850 | goto out_unlock; | |
53bb540f | 851 | /* No increment (+=) because iomap_dio_rw returns a cumulative value. */ |
b01b2d72 | 852 | if (ret > 0) |
42e4c3bd | 853 | read = ret; |
b01b2d72 | 854 | |
324d116c | 855 | if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) { |
e1fa9ea8 | 856 | gfs2_glock_dq(gh); |
6d22ff47 | 857 | window_size -= fault_in_iov_iter_writeable(to, window_size); |
e1fa9ea8 | 858 | if (window_size) |
124c458a | 859 | goto retry; |
b01b2d72 | 860 | } |
72382264 | 861 | out_unlock: |
b01b2d72 AG |
862 | if (gfs2_holder_queued(gh)) |
863 | gfs2_glock_dq(gh); | |
967bcc91 | 864 | out_uninit: |
4c5c3010 | 865 | gfs2_holder_uninit(gh); |
53bb540f | 866 | /* User space doesn't expect partial success. */ |
b01b2d72 AG |
867 | if (ret < 0) |
868 | return ret; | |
42e4c3bd | 869 | return read; |
967bcc91 AG |
870 | } |
871 | ||
4c5c3010 AG |
872 | static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from, |
873 | struct gfs2_holder *gh) | |
967bcc91 AG |
874 | { |
875 | struct file *file = iocb->ki_filp; | |
876 | struct inode *inode = file->f_mapping->host; | |
877 | struct gfs2_inode *ip = GFS2_I(inode); | |
b01b2d72 | 878 | size_t prev_count = 0, window_size = 0; |
42e4c3bd | 879 | size_t written = 0; |
fa58cc88 | 880 | bool enough_retries; |
967bcc91 AG |
881 | ssize_t ret; |
882 | ||
b01b2d72 AG |
883 | /* |
884 | * In this function, we disable page faults when we're holding the | |
885 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
886 | * that the inode glock may be dropped, fault in the pages manually, | |
887 | * and retry. | |
888 | * | |
889 | * For writes, iomap_dio_rw only triggers manual page faults, so we | |
890 | * don't need to disable physical ones. | |
891 | */ | |
892 | ||
967bcc91 AG |
893 | /* |
894 | * Deferred lock, even if its a write, since we do no allocation on | |
895 | * this path. All we need to change is the atime, and this lock mode | |
896 | * ensures that other nodes have flushed their buffered read caches | |
897 | * (i.e. their page cache entries for this inode). We do not, | |
898 | * unfortunately, have the option of only flushing a range like the | |
899 | * VFS does. | |
900 | */ | |
4c5c3010 | 901 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh); |
b01b2d72 | 902 | retry: |
4c5c3010 | 903 | ret = gfs2_glock_nq(gh); |
967bcc91 AG |
904 | if (ret) |
905 | goto out_uninit; | |
967bcc91 | 906 | /* Silently fall back to buffered I/O when writing beyond EOF */ |
b01b2d72 | 907 | if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode)) |
72382264 | 908 | goto out_unlock; |
967bcc91 | 909 | |
b01b2d72 AG |
910 | from->nofault = true; |
911 | ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, | |
786f847f | 912 | IOMAP_DIO_PARTIAL, NULL, written); |
b01b2d72 | 913 | from->nofault = false; |
72382264 AG |
914 | if (ret <= 0) { |
915 | if (ret == -ENOTBLK) | |
916 | ret = 0; | |
917 | if (ret != -EFAULT) | |
918 | goto out_unlock; | |
919 | } | |
53bb540f | 920 | /* No increment (+=) because iomap_dio_rw returns a cumulative value. */ |
b01b2d72 | 921 | if (ret > 0) |
42e4c3bd | 922 | written = ret; |
b01b2d72 | 923 | |
fa58cc88 AG |
924 | enough_retries = prev_count == iov_iter_count(from) && |
925 | window_size <= PAGE_SIZE; | |
324d116c | 926 | if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) { |
e1fa9ea8 | 927 | gfs2_glock_dq(gh); |
6d22ff47 | 928 | window_size -= fault_in_iov_iter_readable(from, window_size); |
fa58cc88 AG |
929 | if (window_size) { |
930 | if (!enough_retries) | |
931 | goto retry; | |
932 | /* fall back to buffered I/O */ | |
933 | ret = 0; | |
934 | } | |
b01b2d72 | 935 | } |
72382264 | 936 | out_unlock: |
b01b2d72 AG |
937 | if (gfs2_holder_queued(gh)) |
938 | gfs2_glock_dq(gh); | |
967bcc91 | 939 | out_uninit: |
4c5c3010 | 940 | gfs2_holder_uninit(gh); |
53bb540f | 941 | /* User space doesn't expect partial success. */ |
b01b2d72 AG |
942 | if (ret < 0) |
943 | return ret; | |
42e4c3bd | 944 | return written; |
967bcc91 AG |
945 | } |
946 | ||
947 | static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
948 | { | |
20f82999 AG |
949 | struct gfs2_inode *ip; |
950 | struct gfs2_holder gh; | |
00bfe02f | 951 | size_t prev_count = 0, window_size = 0; |
42e4c3bd | 952 | size_t read = 0; |
967bcc91 AG |
953 | ssize_t ret; |
954 | ||
00bfe02f AG |
955 | /* |
956 | * In this function, we disable page faults when we're holding the | |
957 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
958 | * that the inode glock may be dropped, fault in the pages manually, | |
959 | * and retry. | |
960 | */ | |
961 | ||
11661835 AG |
962 | if (iocb->ki_flags & IOCB_DIRECT) |
963 | return gfs2_file_direct_read(iocb, to, &gh); | |
964 | ||
52f3f033 | 965 | pagefault_disable(); |
20f82999 AG |
966 | iocb->ki_flags |= IOCB_NOIO; |
967 | ret = generic_file_read_iter(iocb, to); | |
968 | iocb->ki_flags &= ~IOCB_NOIO; | |
52f3f033 | 969 | pagefault_enable(); |
20f82999 AG |
970 | if (ret >= 0) { |
971 | if (!iov_iter_count(to)) | |
972 | return ret; | |
42e4c3bd | 973 | read = ret; |
52f3f033 | 974 | } else if (ret != -EFAULT) { |
20f82999 AG |
975 | if (ret != -EAGAIN) |
976 | return ret; | |
977 | if (iocb->ki_flags & IOCB_NOWAIT) | |
978 | return ret; | |
979 | } | |
980 | ip = GFS2_I(iocb->ki_filp->f_mapping->host); | |
981 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); | |
00bfe02f | 982 | retry: |
20f82999 AG |
983 | ret = gfs2_glock_nq(&gh); |
984 | if (ret) | |
985 | goto out_uninit; | |
00bfe02f | 986 | pagefault_disable(); |
20f82999 | 987 | ret = generic_file_read_iter(iocb, to); |
00bfe02f | 988 | pagefault_enable(); |
72382264 AG |
989 | if (ret <= 0 && ret != -EFAULT) |
990 | goto out_unlock; | |
20f82999 | 991 | if (ret > 0) |
42e4c3bd | 992 | read += ret; |
00bfe02f | 993 | |
324d116c | 994 | if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) { |
e1fa9ea8 | 995 | gfs2_glock_dq(&gh); |
6d22ff47 | 996 | window_size -= fault_in_iov_iter_writeable(to, window_size); |
e1fa9ea8 | 997 | if (window_size) |
124c458a | 998 | goto retry; |
00bfe02f | 999 | } |
72382264 | 1000 | out_unlock: |
00bfe02f AG |
1001 | if (gfs2_holder_queued(&gh)) |
1002 | gfs2_glock_dq(&gh); | |
20f82999 AG |
1003 | out_uninit: |
1004 | gfs2_holder_uninit(&gh); | |
42e4c3bd | 1005 | return read ? read : ret; |
967bcc91 AG |
1006 | } |
1007 | ||
1b223f70 AG |
1008 | static ssize_t gfs2_file_buffered_write(struct kiocb *iocb, |
1009 | struct iov_iter *from, | |
1010 | struct gfs2_holder *gh) | |
2eb7509a AG |
1011 | { |
1012 | struct file *file = iocb->ki_filp; | |
1013 | struct inode *inode = file_inode(file); | |
b924bdab AG |
1014 | struct gfs2_inode *ip = GFS2_I(inode); |
1015 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1b223f70 | 1016 | struct gfs2_holder *statfs_gh = NULL; |
00bfe02f | 1017 | size_t prev_count = 0, window_size = 0; |
554c577c | 1018 | size_t orig_count = iov_iter_count(from); |
42e4c3bd | 1019 | size_t written = 0; |
2eb7509a AG |
1020 | ssize_t ret; |
1021 | ||
00bfe02f AG |
1022 | /* |
1023 | * In this function, we disable page faults when we're holding the | |
1024 | * inode glock while doing I/O. If a page fault occurs, we indicate | |
1025 | * that the inode glock may be dropped, fault in the pages manually, | |
1026 | * and retry. | |
1027 | */ | |
1028 | ||
1b223f70 AG |
1029 | if (inode == sdp->sd_rindex) { |
1030 | statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS); | |
1031 | if (!statfs_gh) | |
1032 | return -ENOMEM; | |
1033 | } | |
1034 | ||
1035 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh); | |
fa5dfa64 | 1036 | if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) { |
c8ed1b35 | 1037 | retry: |
fa5dfa64 | 1038 | window_size -= fault_in_iov_iter_readable(from, window_size); |
fa5dfa64 AG |
1039 | if (!window_size) { |
1040 | ret = -EFAULT; | |
e1fa9ea8 | 1041 | goto out_uninit; |
fa5dfa64 | 1042 | } |
fa5dfa64 AG |
1043 | from->count = min(from->count, window_size); |
1044 | } | |
e1fa9ea8 AG |
1045 | ret = gfs2_glock_nq(gh); |
1046 | if (ret) | |
1047 | goto out_uninit; | |
fa5dfa64 | 1048 | |
b924bdab AG |
1049 | if (inode == sdp->sd_rindex) { |
1050 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
1051 | ||
1052 | ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
1b223f70 | 1053 | GL_NOCACHE, statfs_gh); |
b924bdab AG |
1054 | if (ret) |
1055 | goto out_unlock; | |
1056 | } | |
1057 | ||
00bfe02f | 1058 | pagefault_disable(); |
2eb7509a | 1059 | ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops); |
00bfe02f | 1060 | pagefault_enable(); |
219580ee | 1061 | if (ret > 0) |
42e4c3bd | 1062 | written += ret; |
b924bdab | 1063 | |
1b223f70 AG |
1064 | if (inode == sdp->sd_rindex) |
1065 | gfs2_glock_dq_uninit(statfs_gh); | |
b924bdab | 1066 | |
72382264 AG |
1067 | if (ret <= 0 && ret != -EFAULT) |
1068 | goto out_unlock; | |
1069 | ||
42e4c3bd | 1070 | from->count = orig_count - written; |
e1fa9ea8 AG |
1071 | if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) { |
1072 | gfs2_glock_dq(gh); | |
1073 | goto retry; | |
1074 | } | |
b924bdab | 1075 | out_unlock: |
00bfe02f AG |
1076 | if (gfs2_holder_queued(gh)) |
1077 | gfs2_glock_dq(gh); | |
b924bdab | 1078 | out_uninit: |
1b223f70 | 1079 | gfs2_holder_uninit(gh); |
ab37c305 | 1080 | kfree(statfs_gh); |
42e4c3bd AG |
1081 | from->count = orig_count - written; |
1082 | return written ? written : ret; | |
2eb7509a AG |
1083 | } |
1084 | ||
56aa616a | 1085 | /** |
da56e45b | 1086 | * gfs2_file_write_iter - Perform a write to a file |
56aa616a | 1087 | * @iocb: The io context |
64bc06bb | 1088 | * @from: The data to write |
56aa616a SW |
1089 | * |
1090 | * We have to do a lock/unlock here to refresh the inode size for | |
1091 | * O_APPEND writes, otherwise we can land up writing at the wrong | |
1092 | * offset. There is still a race, but provided the app is using its | |
1093 | * own file locking, this will make O_APPEND work as expected. | |
1094 | * | |
1095 | */ | |
1096 | ||
da56e45b | 1097 | static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
56aa616a SW |
1098 | { |
1099 | struct file *file = iocb->ki_filp; | |
64bc06bb AG |
1100 | struct inode *inode = file_inode(file); |
1101 | struct gfs2_inode *ip = GFS2_I(inode); | |
4c5c3010 | 1102 | struct gfs2_holder gh; |
6e5e41e2 | 1103 | ssize_t ret; |
0a305e49 | 1104 | |
da56e45b | 1105 | gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from)); |
da1dfb6a | 1106 | |
2ba48ce5 | 1107 | if (iocb->ki_flags & IOCB_APPEND) { |
56aa616a SW |
1108 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
1109 | if (ret) | |
4bd684bc | 1110 | return ret; |
56aa616a SW |
1111 | gfs2_glock_dq_uninit(&gh); |
1112 | } | |
1113 | ||
64bc06bb AG |
1114 | inode_lock(inode); |
1115 | ret = generic_write_checks(iocb, from); | |
1116 | if (ret <= 0) | |
4c0e8dda | 1117 | goto out_unlock; |
64bc06bb AG |
1118 | |
1119 | ret = file_remove_privs(file); | |
1120 | if (ret) | |
4c0e8dda | 1121 | goto out_unlock; |
64bc06bb | 1122 | |
967bcc91 AG |
1123 | if (iocb->ki_flags & IOCB_DIRECT) { |
1124 | struct address_space *mapping = file->f_mapping; | |
6e5e41e2 | 1125 | ssize_t buffered, ret2; |
967bcc91 | 1126 | |
089f4eb0 AG |
1127 | /* |
1128 | * Note that under direct I/O, we don't allow and inode | |
1129 | * timestamp updates, so we're not calling file_update_time() | |
1130 | * here. | |
1131 | */ | |
1132 | ||
4c5c3010 | 1133 | ret = gfs2_file_direct_write(iocb, from, &gh); |
6e5e41e2 | 1134 | if (ret < 0 || !iov_iter_count(from)) |
4c0e8dda | 1135 | goto out_unlock; |
967bcc91 | 1136 | |
6e5e41e2 | 1137 | iocb->ki_flags |= IOCB_DSYNC; |
1b223f70 | 1138 | buffered = gfs2_file_buffered_write(iocb, from, &gh); |
43a511c4 AG |
1139 | if (unlikely(buffered <= 0)) { |
1140 | if (!ret) | |
1141 | ret = buffered; | |
4c0e8dda | 1142 | goto out_unlock; |
43a511c4 | 1143 | } |
967bcc91 AG |
1144 | |
1145 | /* | |
1146 | * We need to ensure that the page cache pages are written to | |
1147 | * disk and invalidated to preserve the expected O_DIRECT | |
6e5e41e2 AG |
1148 | * semantics. If the writeback or invalidate fails, only report |
1149 | * the direct I/O range as we don't know if the buffered pages | |
1150 | * made it to disk. | |
967bcc91 | 1151 | */ |
6e5e41e2 AG |
1152 | ret2 = generic_write_sync(iocb, buffered); |
1153 | invalidate_mapping_pages(mapping, | |
1154 | (iocb->ki_pos - buffered) >> PAGE_SHIFT, | |
1155 | (iocb->ki_pos - 1) >> PAGE_SHIFT); | |
1156 | if (!ret || ret2 > 0) | |
1157 | ret += ret2; | |
967bcc91 | 1158 | } else { |
089f4eb0 AG |
1159 | ret = file_update_time(file); |
1160 | if (ret) | |
1161 | goto out_unlock; | |
1162 | ||
1b223f70 | 1163 | ret = gfs2_file_buffered_write(iocb, from, &gh); |
2eb7509a | 1164 | if (likely(ret > 0)) |
6e5e41e2 | 1165 | ret = generic_write_sync(iocb, ret); |
967bcc91 | 1166 | } |
64bc06bb | 1167 | |
4c0e8dda | 1168 | out_unlock: |
64bc06bb | 1169 | inode_unlock(inode); |
6e5e41e2 | 1170 | return ret; |
56aa616a SW |
1171 | } |
1172 | ||
2fe17c10 CH |
1173 | static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len, |
1174 | int mode) | |
1175 | { | |
fffb6412 | 1176 | struct super_block *sb = inode->i_sb; |
2fe17c10 | 1177 | struct gfs2_inode *ip = GFS2_I(inode); |
fffb6412 | 1178 | loff_t end = offset + len; |
2fe17c10 CH |
1179 | struct buffer_head *dibh; |
1180 | int error; | |
2fe17c10 CH |
1181 | |
1182 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
1183 | if (unlikely(error)) | |
64dd153c | 1184 | return error; |
2fe17c10 | 1185 | |
350a9b0a | 1186 | gfs2_trans_add_meta(ip->i_gl, dibh); |
2fe17c10 CH |
1187 | |
1188 | if (gfs2_is_stuffed(ip)) { | |
7a607a41 | 1189 | error = gfs2_unstuff_dinode(ip); |
2fe17c10 CH |
1190 | if (unlikely(error)) |
1191 | goto out; | |
1192 | } | |
1193 | ||
fffb6412 | 1194 | while (offset < end) { |
c2589282 AG |
1195 | struct iomap iomap = { }; |
1196 | ||
54992257 | 1197 | error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap); |
fffb6412 | 1198 | if (error) |
64dd153c | 1199 | goto out; |
fffb6412 | 1200 | offset = iomap.offset + iomap.length; |
d505a96a | 1201 | if (!(iomap.flags & IOMAP_F_NEW)) |
64dd153c | 1202 | continue; |
fffb6412 AG |
1203 | error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits, |
1204 | iomap.length >> inode->i_blkbits, | |
1205 | GFP_NOFS); | |
1206 | if (error) { | |
1207 | fs_err(GFS2_SB(inode), "Failed to zero data buffers\n"); | |
2fe17c10 | 1208 | goto out; |
64dd153c | 1209 | } |
2fe17c10 | 1210 | } |
2fe17c10 | 1211 | out: |
64dd153c | 1212 | brelse(dibh); |
2fe17c10 CH |
1213 | return error; |
1214 | } | |
f3b64b57 | 1215 | |
d9be0cda AD |
1216 | /** |
1217 | * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of | |
1218 | * blocks, determine how many bytes can be written. | |
1219 | * @ip: The inode in question. | |
1220 | * @len: Max cap of bytes. What we return in *len must be <= this. | |
1221 | * @data_blocks: Compute and return the number of data blocks needed | |
1222 | * @ind_blocks: Compute and return the number of indirect blocks needed | |
1223 | * @max_blocks: The total blocks available to work with. | |
1224 | * | |
1225 | * Returns: void, but @len, @data_blocks and @ind_blocks are filled in. | |
1226 | */ | |
1227 | static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len, | |
1228 | unsigned int *data_blocks, unsigned int *ind_blocks, | |
1229 | unsigned int max_blocks) | |
2fe17c10 | 1230 | { |
d9be0cda | 1231 | loff_t max = *len; |
2fe17c10 | 1232 | const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
2fe17c10 CH |
1233 | unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1); |
1234 | ||
1235 | for (tmp = max_data; tmp > sdp->sd_diptrs;) { | |
1236 | tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs); | |
1237 | max_data -= tmp; | |
1238 | } | |
d9be0cda | 1239 | |
2fe17c10 CH |
1240 | *data_blocks = max_data; |
1241 | *ind_blocks = max_blocks - max_data; | |
1242 | *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift; | |
1243 | if (*len > max) { | |
1244 | *len = max; | |
1245 | gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks); | |
1246 | } | |
1247 | } | |
1248 | ||
9c9f1159 | 1249 | static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
2fe17c10 | 1250 | { |
496ad9aa | 1251 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
1252 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1253 | struct gfs2_inode *ip = GFS2_I(inode); | |
f7e4c610 | 1254 | struct gfs2_alloc_parms ap = {}; |
2fe17c10 | 1255 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
174d1232 | 1256 | loff_t bytes, max_bytes, max_blks; |
2fe17c10 | 1257 | int error; |
4442f2e0 SW |
1258 | const loff_t pos = offset; |
1259 | const loff_t count = len; | |
6905d9e4 | 1260 | loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1); |
2fe17c10 | 1261 | loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift; |
64dd153c | 1262 | loff_t max_chunk_size = UINT_MAX & bsize_mask; |
a0846a53 | 1263 | |
2fe17c10 CH |
1264 | next = (next + 1) << sdp->sd_sb.sb_bsize_shift; |
1265 | ||
6905d9e4 | 1266 | offset &= bsize_mask; |
2fe17c10 CH |
1267 | |
1268 | len = next - offset; | |
1269 | bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2; | |
1270 | if (!bytes) | |
1271 | bytes = UINT_MAX; | |
6905d9e4 BM |
1272 | bytes &= bsize_mask; |
1273 | if (bytes == 0) | |
1274 | bytes = sdp->sd_sb.sb_bsize; | |
2fe17c10 | 1275 | |
da1dfb6a | 1276 | gfs2_size_hint(file, offset, len); |
8e2e0047 | 1277 | |
d9be0cda AD |
1278 | gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks); |
1279 | ap.min_target = data_blocks + ind_blocks; | |
1280 | ||
2fe17c10 CH |
1281 | while (len > 0) { |
1282 | if (len < bytes) | |
1283 | bytes = len; | |
58a7d5fb BM |
1284 | if (!gfs2_write_alloc_required(ip, offset, bytes)) { |
1285 | len -= bytes; | |
1286 | offset += bytes; | |
1287 | continue; | |
1288 | } | |
d9be0cda AD |
1289 | |
1290 | /* We need to determine how many bytes we can actually | |
1291 | * fallocate without exceeding quota or going over the | |
1292 | * end of the fs. We start off optimistically by assuming | |
1293 | * we can write max_bytes */ | |
1294 | max_bytes = (len > max_chunk_size) ? max_chunk_size : len; | |
1295 | ||
1296 | /* Since max_bytes is most likely a theoretical max, we | |
1297 | * calculate a more realistic 'bytes' to serve as a good | |
1298 | * starting point for the number of bytes we may be able | |
1299 | * to write */ | |
2fe17c10 | 1300 | gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks); |
7b9cff46 | 1301 | ap.target = data_blocks + ind_blocks; |
b8fbf471 AD |
1302 | |
1303 | error = gfs2_quota_lock_check(ip, &ap); | |
2fe17c10 | 1304 | if (error) |
9c9f1159 | 1305 | return error; |
d9be0cda AD |
1306 | /* ap.allowed tells us how many blocks quota will allow |
1307 | * us to write. Check if this reduces max_blks */ | |
174d1232 AG |
1308 | max_blks = UINT_MAX; |
1309 | if (ap.allowed) | |
d9be0cda | 1310 | max_blks = ap.allowed; |
2fe17c10 | 1311 | |
7b9cff46 | 1312 | error = gfs2_inplace_reserve(ip, &ap); |
d9be0cda | 1313 | if (error) |
2fe17c10 | 1314 | goto out_qunlock; |
d9be0cda AD |
1315 | |
1316 | /* check if the selected rgrp limits our max_blks further */ | |
725d0e9d AG |
1317 | if (ip->i_res.rs_reserved < max_blks) |
1318 | max_blks = ip->i_res.rs_reserved; | |
d9be0cda AD |
1319 | |
1320 | /* Almost done. Calculate bytes that can be written using | |
1321 | * max_blks. We also recompute max_bytes, data_blocks and | |
1322 | * ind_blocks */ | |
1323 | calc_max_reserv(ip, &max_bytes, &data_blocks, | |
1324 | &ind_blocks, max_blks); | |
2fe17c10 CH |
1325 | |
1326 | rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA + | |
71f890f7 | 1327 | RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks); |
2fe17c10 CH |
1328 | if (gfs2_is_jdata(ip)) |
1329 | rblocks += data_blocks ? data_blocks : 1; | |
1330 | ||
1331 | error = gfs2_trans_begin(sdp, rblocks, | |
45eb0504 | 1332 | PAGE_SIZE >> inode->i_blkbits); |
2fe17c10 CH |
1333 | if (error) |
1334 | goto out_trans_fail; | |
1335 | ||
1336 | error = fallocate_chunk(inode, offset, max_bytes, mode); | |
1337 | gfs2_trans_end(sdp); | |
1338 | ||
1339 | if (error) | |
1340 | goto out_trans_fail; | |
1341 | ||
1342 | len -= max_bytes; | |
1343 | offset += max_bytes; | |
1344 | gfs2_inplace_release(ip); | |
1345 | gfs2_quota_unlock(ip); | |
2fe17c10 | 1346 | } |
4442f2e0 | 1347 | |
0a6a4abc | 1348 | if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size) |
1885867b | 1349 | i_size_write(inode, pos + count); |
0a6a4abc AG |
1350 | file_update_time(file); |
1351 | mark_inode_dirty(inode); | |
1885867b | 1352 | |
dde0c2e7 CH |
1353 | if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host)) |
1354 | return vfs_fsync_range(file, pos, pos + count - 1, | |
1355 | (file->f_flags & __O_SYNC) ? 0 : 1); | |
1356 | return 0; | |
2fe17c10 CH |
1357 | |
1358 | out_trans_fail: | |
1359 | gfs2_inplace_release(ip); | |
1360 | out_qunlock: | |
1361 | gfs2_quota_unlock(ip); | |
9c9f1159 AP |
1362 | return error; |
1363 | } | |
1364 | ||
1365 | static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len) | |
1366 | { | |
1367 | struct inode *inode = file_inode(file); | |
d4d7fc12 | 1368 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9c9f1159 AP |
1369 | struct gfs2_inode *ip = GFS2_I(inode); |
1370 | struct gfs2_holder gh; | |
1371 | int ret; | |
1372 | ||
4e56a641 | 1373 | if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) |
d4d7fc12 AP |
1374 | return -EOPNOTSUPP; |
1375 | /* fallocate is needed by gfs2_grow to reserve space in the rindex */ | |
1376 | if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex) | |
9c9f1159 AP |
1377 | return -EOPNOTSUPP; |
1378 | ||
5955102c | 1379 | inode_lock(inode); |
9c9f1159 AP |
1380 | |
1381 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
1382 | ret = gfs2_glock_nq(&gh); | |
1383 | if (ret) | |
1384 | goto out_uninit; | |
1385 | ||
1386 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
1387 | (offset + len) > inode->i_size) { | |
1388 | ret = inode_newsize_ok(inode, offset + len); | |
1389 | if (ret) | |
1390 | goto out_unlock; | |
1391 | } | |
1392 | ||
1393 | ret = get_write_access(inode); | |
1394 | if (ret) | |
1395 | goto out_unlock; | |
1396 | ||
4e56a641 AG |
1397 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
1398 | ret = __gfs2_punch_hole(file, offset, len); | |
1399 | } else { | |
4e56a641 | 1400 | ret = __gfs2_fallocate(file, mode, offset, len); |
4e56a641 AG |
1401 | if (ret) |
1402 | gfs2_rs_deltree(&ip->i_res); | |
1403 | } | |
a097dc7e | 1404 | |
9c9f1159 | 1405 | put_write_access(inode); |
2fe17c10 | 1406 | out_unlock: |
a0846a53 | 1407 | gfs2_glock_dq(&gh); |
2fe17c10 | 1408 | out_uninit: |
a0846a53 | 1409 | gfs2_holder_uninit(&gh); |
5955102c | 1410 | inode_unlock(inode); |
9c9f1159 | 1411 | return ret; |
2fe17c10 CH |
1412 | } |
1413 | ||
f1ea6f4e BP |
1414 | static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe, |
1415 | struct file *out, loff_t *ppos, | |
1416 | size_t len, unsigned int flags) | |
1417 | { | |
2fba46a0 | 1418 | ssize_t ret; |
f1ea6f4e | 1419 | |
f1ea6f4e BP |
1420 | gfs2_size_hint(out, *ppos, len); |
1421 | ||
2fba46a0 | 1422 | ret = iter_file_splice_write(pipe, out, ppos, len, flags); |
2fba46a0 | 1423 | return ret; |
f1ea6f4e BP |
1424 | } |
1425 | ||
f057f6cd SW |
1426 | #ifdef CONFIG_GFS2_FS_LOCKING_DLM |
1427 | ||
b3b94faa DT |
1428 | /** |
1429 | * gfs2_lock - acquire/release a posix lock on a file | |
1430 | * @file: the file pointer | |
1431 | * @cmd: either modify or retrieve lock state, possibly wait | |
1432 | * @fl: type and range of lock | |
1433 | * | |
1434 | * Returns: errno | |
1435 | */ | |
1436 | ||
1437 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
1438 | { | |
feaa7bba SW |
1439 | struct gfs2_inode *ip = GFS2_I(file->f_mapping->host); |
1440 | struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host); | |
f057f6cd | 1441 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
b3b94faa | 1442 | |
b3b94faa DT |
1443 | if (!(fl->fl_flags & FL_POSIX)) |
1444 | return -ENOLCK; | |
4d927b03 | 1445 | if (gfs2_withdrawing_or_withdrawn(sdp)) { |
c2952d20 | 1446 | if (fl->fl_type == F_UNLCK) |
4f656367 | 1447 | locks_lock_file_wait(file, fl); |
f057f6cd | 1448 | return -EIO; |
c2952d20 | 1449 | } |
dc52cd2e AA |
1450 | if (cmd == F_CANCELLK) |
1451 | return dlm_posix_cancel(ls->ls_dlm, ip->i_no_addr, file, fl); | |
1452 | else if (IS_GETLK(cmd)) | |
f057f6cd | 1453 | return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1454 | else if (fl->fl_type == F_UNLCK) |
f057f6cd | 1455 | return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl); |
b3b94faa | 1456 | else |
f057f6cd | 1457 | return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl); |
b3b94faa DT |
1458 | } |
1459 | ||
56535dc6 AG |
1460 | static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh) |
1461 | { | |
4ad02083 | 1462 | struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl); |
56535dc6 AG |
1463 | |
1464 | /* | |
1465 | * Make sure gfs2_glock_put() won't sleep under the file->f_lock | |
1466 | * spinlock. | |
1467 | */ | |
1468 | ||
56535dc6 AG |
1469 | spin_lock(&file->f_lock); |
1470 | gfs2_holder_uninit(fl_gh); | |
1471 | spin_unlock(&file->f_lock); | |
1472 | gfs2_glock_put(gl); | |
1473 | } | |
1474 | ||
b3b94faa DT |
1475 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) |
1476 | { | |
5c676f6d | 1477 | struct gfs2_file *fp = file->private_data; |
b3b94faa | 1478 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
496ad9aa | 1479 | struct gfs2_inode *ip = GFS2_I(file_inode(file)); |
b3b94faa DT |
1480 | struct gfs2_glock *gl; |
1481 | unsigned int state; | |
b58bf407 | 1482 | u16 flags; |
b3b94faa | 1483 | int error = 0; |
2ddfbdd6 | 1484 | int sleeptime; |
b3b94faa DT |
1485 | |
1486 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
b582d5f0 AG |
1487 | flags = GL_EXACT | GL_NOPID; |
1488 | if (!IS_SETLKW(cmd)) | |
1489 | flags |= LM_FLAG_TRY_1CB; | |
b3b94faa | 1490 | |
f55ab26a | 1491 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa | 1492 | |
283c9a97 | 1493 | if (gfs2_holder_initialized(fl_gh)) { |
4d62d3f7 | 1494 | struct file_lock request; |
b3b94faa DT |
1495 | if (fl_gh->gh_state == state) |
1496 | goto out; | |
4d62d3f7 N |
1497 | locks_init_lock(&request); |
1498 | request.fl_type = F_UNLCK; | |
1499 | request.fl_flags = FL_FLOCK; | |
1500 | locks_lock_file_wait(file, &request); | |
5bef3e7c | 1501 | gfs2_glock_dq(fl_gh); |
b4c20166 | 1502 | gfs2_holder_reinit(state, flags, fl_gh); |
b3b94faa | 1503 | } else { |
6802e340 SW |
1504 | error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr, |
1505 | &gfs2_flock_glops, CREATE, &gl); | |
b3b94faa DT |
1506 | if (error) |
1507 | goto out; | |
56535dc6 | 1508 | spin_lock(&file->f_lock); |
b4c20166 | 1509 | gfs2_holder_init(gl, state, flags, fl_gh); |
56535dc6 | 1510 | spin_unlock(&file->f_lock); |
b4c20166 | 1511 | gfs2_glock_put(gl); |
b3b94faa | 1512 | } |
2ddfbdd6 BP |
1513 | for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) { |
1514 | error = gfs2_glock_nq(fl_gh); | |
1515 | if (error != GLR_TRYFAILED) | |
1516 | break; | |
b582d5f0 AG |
1517 | fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB; |
1518 | fl_gh->gh_flags |= LM_FLAG_TRY; | |
2ddfbdd6 BP |
1519 | msleep(sleeptime); |
1520 | } | |
b3b94faa | 1521 | if (error) { |
56535dc6 | 1522 | __flock_holder_uninit(file, fl_gh); |
b3b94faa DT |
1523 | if (error == GLR_TRYFAILED) |
1524 | error = -EAGAIN; | |
1525 | } else { | |
4f656367 | 1526 | error = locks_lock_file_wait(file, fl); |
feaa7bba | 1527 | gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); |
b3b94faa DT |
1528 | } |
1529 | ||
420b9e5e | 1530 | out: |
f55ab26a | 1531 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1532 | return error; |
1533 | } | |
1534 | ||
1535 | static void do_unflock(struct file *file, struct file_lock *fl) | |
1536 | { | |
5c676f6d | 1537 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
1538 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
1539 | ||
f55ab26a | 1540 | mutex_lock(&fp->f_fl_mutex); |
4f656367 | 1541 | locks_lock_file_wait(file, fl); |
6df9f9a2 | 1542 | if (gfs2_holder_initialized(fl_gh)) { |
2ddfbdd6 | 1543 | gfs2_glock_dq(fl_gh); |
56535dc6 | 1544 | __flock_holder_uninit(file, fl_gh); |
0a33443b | 1545 | } |
f55ab26a | 1546 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
1547 | } |
1548 | ||
1549 | /** | |
1550 | * gfs2_flock - acquire/release a flock lock on a file | |
1551 | * @file: the file pointer | |
1552 | * @cmd: either modify or retrieve lock state, possibly wait | |
1553 | * @fl: type and range of lock | |
1554 | * | |
1555 | * Returns: errno | |
1556 | */ | |
1557 | ||
1558 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
1559 | { | |
b3b94faa DT |
1560 | if (!(fl->fl_flags & FL_FLOCK)) |
1561 | return -ENOLCK; | |
b3b94faa | 1562 | |
b3b94faa DT |
1563 | if (fl->fl_type == F_UNLCK) { |
1564 | do_unflock(file, fl); | |
1565 | return 0; | |
d00223f1 | 1566 | } else { |
b3b94faa | 1567 | return do_flock(file, cmd, fl); |
d00223f1 | 1568 | } |
b3b94faa DT |
1569 | } |
1570 | ||
10d21988 | 1571 | const struct file_operations gfs2_file_fops = { |
26c1a574 | 1572 | .llseek = gfs2_llseek, |
967bcc91 | 1573 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1574 | .write_iter = gfs2_file_write_iter, |
3e08773c | 1575 | .iopoll = iocb_bio_iopoll, |
26c1a574 | 1576 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1577 | .compat_ioctl = gfs2_compat_ioctl, |
26c1a574 SW |
1578 | .mmap = gfs2_mmap, |
1579 | .open = gfs2_open, | |
df3fd117 | 1580 | .release = gfs2_release, |
26c1a574 SW |
1581 | .fsync = gfs2_fsync, |
1582 | .lock = gfs2_lock, | |
26c1a574 | 1583 | .flock = gfs2_flock, |
0be84321 | 1584 | .splice_read = copy_splice_read, |
f42a69fa | 1585 | .splice_write = gfs2_file_splice_write, |
1c994a09 | 1586 | .setlease = simple_nosetlease, |
2fe17c10 | 1587 | .fallocate = gfs2_fallocate, |
b3b94faa DT |
1588 | }; |
1589 | ||
10d21988 | 1590 | const struct file_operations gfs2_dir_fops = { |
1d1bb236 | 1591 | .iterate_shared = gfs2_readdir, |
26c1a574 | 1592 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1593 | .compat_ioctl = gfs2_compat_ioctl, |
26c1a574 | 1594 | .open = gfs2_open, |
df3fd117 | 1595 | .release = gfs2_release, |
26c1a574 SW |
1596 | .fsync = gfs2_fsync, |
1597 | .lock = gfs2_lock, | |
1598 | .flock = gfs2_flock, | |
6038f373 | 1599 | .llseek = default_llseek, |
b3b94faa DT |
1600 | }; |
1601 | ||
f057f6cd SW |
1602 | #endif /* CONFIG_GFS2_FS_LOCKING_DLM */ |
1603 | ||
10d21988 | 1604 | const struct file_operations gfs2_file_fops_nolock = { |
c97bfe43 | 1605 | .llseek = gfs2_llseek, |
967bcc91 | 1606 | .read_iter = gfs2_file_read_iter, |
da56e45b | 1607 | .write_iter = gfs2_file_write_iter, |
3e08773c | 1608 | .iopoll = iocb_bio_iopoll, |
c97bfe43 | 1609 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1610 | .compat_ioctl = gfs2_compat_ioctl, |
c97bfe43 WC |
1611 | .mmap = gfs2_mmap, |
1612 | .open = gfs2_open, | |
df3fd117 | 1613 | .release = gfs2_release, |
c97bfe43 | 1614 | .fsync = gfs2_fsync, |
0be84321 | 1615 | .splice_read = copy_splice_read, |
f42a69fa | 1616 | .splice_write = gfs2_file_splice_write, |
f057f6cd | 1617 | .setlease = generic_setlease, |
2fe17c10 | 1618 | .fallocate = gfs2_fallocate, |
c97bfe43 WC |
1619 | }; |
1620 | ||
10d21988 | 1621 | const struct file_operations gfs2_dir_fops_nolock = { |
1d1bb236 | 1622 | .iterate_shared = gfs2_readdir, |
c97bfe43 | 1623 | .unlocked_ioctl = gfs2_ioctl, |
8d098070 | 1624 | .compat_ioctl = gfs2_compat_ioctl, |
c97bfe43 | 1625 | .open = gfs2_open, |
df3fd117 | 1626 | .release = gfs2_release, |
c97bfe43 | 1627 | .fsync = gfs2_fsync, |
6038f373 | 1628 | .llseek = default_llseek, |
c97bfe43 WC |
1629 | }; |
1630 |