module/os/linux/zfs/zpl_file_range.c

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