1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22 /*
23 * Copyright (c) 2023, Klara Inc.
24 */
25
26 #ifdef CONFIG_COMPAT
27 #include <linux/compat.h>
28 #endif
29 #include <linux/fs.h>
30 #ifdef HAVE_VFS_SPLICE_COPY_FILE_RANGE
31 #include <linux/splice.h>
32 #endif
33 #include <sys/file.h>
34 #include <sys/zfs_znode.h>
35 #include <sys/zfs_vnops.h>
36 #include <sys/zfeature.h>
37
38 /*
39 * Clone part of a file via block cloning.
40 *
41 * Note that we are not required to update file offsets; the kernel will take
42 * care of that depending on how it was called.
43 */
44 static ssize_t
zpl_clone_file_range_impl(struct file * src_file,loff_t src_off,struct file * dst_file,loff_t dst_off,size_t len)45 zpl_clone_file_range_impl(struct file *src_file, loff_t src_off,
46 struct file *dst_file, loff_t dst_off, size_t len)
47 {
48 struct inode *src_i = file_inode(src_file);
49 struct inode *dst_i = file_inode(dst_file);
50 uint64_t src_off_o = (uint64_t)src_off;
51 uint64_t dst_off_o = (uint64_t)dst_off;
52 uint64_t len_o = (uint64_t)len;
53 cred_t *cr = CRED();
54 fstrans_cookie_t cookie;
55 int err;
56
57 if (!zfs_bclone_enabled)
58 return (-EOPNOTSUPP);
59
60 if (!spa_feature_is_enabled(
61 dmu_objset_spa(ITOZSB(dst_i)->z_os), SPA_FEATURE_BLOCK_CLONING))
62 return (-EOPNOTSUPP);
63
64 if (src_i != dst_i)
65 spl_inode_lock_shared(src_i);
66 spl_inode_lock(dst_i);
67
68 crhold(cr);
69 cookie = spl_fstrans_mark();
70
71 err = -zfs_clone_range(ITOZ(src_i), &src_off_o, ITOZ(dst_i),
72 &dst_off_o, &len_o, cr);
73
74 spl_fstrans_unmark(cookie);
75 crfree(cr);
76
77 spl_inode_unlock(dst_i);
78 if (src_i != dst_i)
79 spl_inode_unlock_shared(src_i);
80
81 if (err < 0)
82 return (err);
83
84 return ((ssize_t)len_o);
85 }
86
87 /*
88 * Entry point for copy_file_range(). Copy len bytes from src_off in src_file
89 * to dst_off in dst_file. We are permitted to do this however we like, so we
90 * try to just clone the blocks, and if we can't support it, fall back to the
91 * kernel's generic byte copy function.
92 */
93 ssize_t
zpl_copy_file_range(struct file * src_file,loff_t src_off,struct file * dst_file,loff_t dst_off,size_t len,unsigned int flags)94 zpl_copy_file_range(struct file *src_file, loff_t src_off,
95 struct file *dst_file, loff_t dst_off, size_t len, unsigned int flags)
96 {
97 ssize_t ret;
98
99 /* Flags is reserved for future extensions and must be zero. */
100 if (flags != 0)
101 return (-EINVAL);
102
103 /* Try to do it via zfs_clone_range() and allow shortening. */
104 ret = zpl_clone_file_range_impl(src_file, src_off,
105 dst_file, dst_off, len);
106
107 #if defined(HAVE_VFS_GENERIC_COPY_FILE_RANGE)
108 /*
109 * Since Linux 5.3 the filesystem driver is responsible for executing
110 * an appropriate fallback, and a generic fallback function is provided.
111 */
112 if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -EXDEV ||
113 ret == -EAGAIN)
114 ret = generic_copy_file_range(src_file, src_off, dst_file,
115 dst_off, len, flags);
116 #elif defined(HAVE_VFS_SPLICE_COPY_FILE_RANGE)
117 /*
118 * Since 6.8 the fallback function is called splice_copy_file_range
119 * and has a slightly different signature.
120 */
121 if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -EXDEV ||
122 ret == -EAGAIN)
123 ret = splice_copy_file_range(src_file, src_off, dst_file,
124 dst_off, len);
125 #else
126 /*
127 * Before Linux 5.3 the filesystem has to return -EOPNOTSUPP to signal
128 * to the kernel that it should fallback to a content copy.
129 */
130 if (ret == -EINVAL || ret == -EXDEV || ret == -EAGAIN)
131 ret = -EOPNOTSUPP;
132 #endif /* HAVE_VFS_GENERIC_COPY_FILE_RANGE || HAVE_VFS_SPLICE_COPY_FILE_RANGE */
133
134 return (ret);
135 }
136
137 #ifdef HAVE_VFS_REMAP_FILE_RANGE
138 /*
139 * Entry point for FICLONE/FICLONERANGE/FIDEDUPERANGE.
140 *
141 * FICLONE and FICLONERANGE are basically the same as copy_file_range(), except
142 * that they must clone - they cannot fall back to copying. FICLONE is exactly
143 * FICLONERANGE, for the entire file. We don't need to try to tell them apart;
144 * the kernel will sort that out for us.
145 *
146 * FIDEDUPERANGE is for turning a non-clone into a clone, that is, compare the
147 * range in both files and if they're the same, arrange for them to be backed
148 * by the same storage.
149 *
150 * REMAP_FILE_CAN_SHORTEN lets us know we can clone less than the given range
151 * if we want. It's designed for filesystems that may need to shorten the
152 * length for alignment, EOF, or any other requirement. ZFS may shorten the
153 * request when there is outstanding dirty data which hasn't been written.
154 */
155 loff_t
zpl_remap_file_range(struct file * src_file,loff_t src_off,struct file * dst_file,loff_t dst_off,loff_t len,unsigned int flags)156 zpl_remap_file_range(struct file *src_file, loff_t src_off,
157 struct file *dst_file, loff_t dst_off, loff_t len, unsigned int flags)
158 {
159 if (flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_CAN_SHORTEN))
160 return (-EINVAL);
161
162 /* No support for dedup yet */
163 if (flags & REMAP_FILE_DEDUP)
164 return (-EOPNOTSUPP);
165
166 /* Zero length means to clone everything to the end of the file */
167 if (len == 0)
168 len = i_size_read(file_inode(src_file)) - src_off;
169
170 ssize_t ret = zpl_clone_file_range_impl(src_file, src_off,
171 dst_file, dst_off, len);
172
173 if (!(flags & REMAP_FILE_CAN_SHORTEN) && ret >= 0 && ret != len)
174 ret = -EINVAL;
175
176 return (ret);
177 }
178 #endif /* HAVE_VFS_REMAP_FILE_RANGE */
179
180 #if defined(HAVE_VFS_CLONE_FILE_RANGE)
181 /*
182 * Entry point for FICLONE and FICLONERANGE, before Linux 4.20.
183 */
184 int
zpl_clone_file_range(struct file * src_file,loff_t src_off,struct file * dst_file,loff_t dst_off,uint64_t len)185 zpl_clone_file_range(struct file *src_file, loff_t src_off,
186 struct file *dst_file, loff_t dst_off, uint64_t len)
187 {
188 /* Zero length means to clone everything to the end of the file */
189 if (len == 0)
190 len = i_size_read(file_inode(src_file)) - src_off;
191
192 /* The entire length must be cloned or this is an error. */
193 ssize_t ret = zpl_clone_file_range_impl(src_file, src_off,
194 dst_file, dst_off, len);
195
196 if (ret >= 0 && ret != len)
197 ret = -EINVAL;
198
199 return (ret);
200 }
201 #endif /* HAVE_VFS_CLONE_FILE_RANGE */
202
203 #ifdef HAVE_VFS_DEDUPE_FILE_RANGE
204 /*
205 * Entry point for FIDEDUPERANGE, before Linux 4.20.
206 */
207 int
zpl_dedupe_file_range(struct file * src_file,loff_t src_off,struct file * dst_file,loff_t dst_off,uint64_t len)208 zpl_dedupe_file_range(struct file *src_file, loff_t src_off,
209 struct file *dst_file, loff_t dst_off, uint64_t len)
210 {
211 /* No support for dedup yet */
212 return (-EOPNOTSUPP);
213 }
214 #endif /* HAVE_VFS_DEDUPE_FILE_RANGE */
215
216 /* Entry point for FICLONE, before Linux 4.5. */
217 long
zpl_ioctl_ficlone(struct file * dst_file,void * arg)218 zpl_ioctl_ficlone(struct file *dst_file, void *arg)
219 {
220 unsigned long sfd = (unsigned long)arg;
221
222 struct file *src_file = fget(sfd);
223 if (src_file == NULL)
224 return (-EBADF);
225
226 if (dst_file->f_op != src_file->f_op) {
227 fput(src_file);
228 return (-EXDEV);
229 }
230
231 size_t len = i_size_read(file_inode(src_file));
232
233 ssize_t ret = zpl_clone_file_range_impl(src_file, 0, dst_file, 0, len);
234
235 fput(src_file);
236
237 if (ret < 0) {
238 if (ret == -EOPNOTSUPP)
239 return (-ENOTTY);
240 return (ret);
241 }
242
243 if (ret != len)
244 return (-EINVAL);
245
246 return (0);
247 }
248
249 /* Entry point for FICLONERANGE, before Linux 4.5. */
250 long
zpl_ioctl_ficlonerange(struct file * dst_file,void __user * arg)251 zpl_ioctl_ficlonerange(struct file *dst_file, void __user *arg)
252 {
253 zfs_ioc_compat_file_clone_range_t fcr;
254
255 if (copy_from_user(&fcr, arg, sizeof (fcr)))
256 return (-EFAULT);
257
258 struct file *src_file = fget(fcr.fcr_src_fd);
259 if (src_file == NULL)
260 return (-EBADF);
261
262 if (dst_file->f_op != src_file->f_op) {
263 fput(src_file);
264 return (-EXDEV);
265 }
266
267 size_t len = fcr.fcr_src_length;
268 if (len == 0)
269 len = i_size_read(file_inode(src_file)) - fcr.fcr_src_offset;
270
271 ssize_t ret = zpl_clone_file_range_impl(src_file, fcr.fcr_src_offset,
272 dst_file, fcr.fcr_dest_offset, len);
273
274 fput(src_file);
275
276 if (ret < 0) {
277 if (ret == -EOPNOTSUPP)
278 return (-ENOTTY);
279 return (ret);
280 }
281
282 if (ret != len)
283 return (-EINVAL);
284
285 return (0);
286 }
287
288 /* Entry point for FIDEDUPERANGE, before Linux 4.5. */
289 long
zpl_ioctl_fideduperange(struct file * filp,void * arg)290 zpl_ioctl_fideduperange(struct file *filp, void *arg)
291 {
292 (void) arg;
293
294 /* No support for dedup yet */
295 return (-ENOTTY);
296 }
297