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