]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blobdiff - mm/filemap.c
vfs: enable remap callers that can handle short operations
[mirror_ubuntu-jammy-kernel.git] / mm / filemap.c
index 52517f28e6f4a69020cfc60867f5a57461c89fb8..1775d4ad3317c988a97682c971f48900a2fbe9f6 100644 (file)
@@ -2915,6 +2915,42 @@ struct page *read_cache_page_gfp(struct address_space *mapping,
 }
 EXPORT_SYMBOL(read_cache_page_gfp);
 
+/*
+ * Don't operate on ranges the page cache doesn't support, and don't exceed the
+ * LFS limits.  If pos is under the limit it becomes a short access.  If it
+ * exceeds the limit we return -EFBIG.
+ */
+static int generic_access_check_limits(struct file *file, loff_t pos,
+                                      loff_t *count)
+{
+       struct inode *inode = file->f_mapping->host;
+       loff_t max_size = inode->i_sb->s_maxbytes;
+
+       if (!(file->f_flags & O_LARGEFILE))
+               max_size = MAX_NON_LFS;
+
+       if (unlikely(pos >= max_size))
+               return -EFBIG;
+       *count = min(*count, max_size - pos);
+       return 0;
+}
+
+static int generic_write_check_limits(struct file *file, loff_t pos,
+                                     loff_t *count)
+{
+       loff_t limit = rlimit(RLIMIT_FSIZE);
+
+       if (limit != RLIM_INFINITY) {
+               if (pos >= limit) {
+                       send_sig(SIGXFSZ, current, 0);
+                       return -EFBIG;
+               }
+               *count = min(*count, limit - pos);
+       }
+
+       return generic_access_check_limits(file, pos, count);
+}
+
 /*
  * Performs necessary checks before doing a write
  *
@@ -2926,8 +2962,8 @@ inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
 {
        struct file *file = iocb->ki_filp;
        struct inode *inode = file->f_mapping->host;
-       unsigned long limit = rlimit(RLIMIT_FSIZE);
-       loff_t pos;
+       loff_t count;
+       int ret;
 
        if (!iov_iter_count(from))
                return 0;
@@ -2936,43 +2972,99 @@ inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
        if (iocb->ki_flags & IOCB_APPEND)
                iocb->ki_pos = i_size_read(inode);
 
-       pos = iocb->ki_pos;
-
        if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
                return -EINVAL;
 
-       if (limit != RLIM_INFINITY) {
-               if (iocb->ki_pos >= limit) {
-                       send_sig(SIGXFSZ, current, 0);
-                       return -EFBIG;
-               }
-               iov_iter_truncate(from, limit - (unsigned long)pos);
-       }
+       count = iov_iter_count(from);
+       ret = generic_write_check_limits(file, iocb->ki_pos, &count);
+       if (ret)
+               return ret;
+
+       iov_iter_truncate(from, count);
+       return iov_iter_count(from);
+}
+EXPORT_SYMBOL(generic_write_checks);
+
+/*
+ * Performs necessary checks before doing a clone.
+ *
+ * Can adjust amount of bytes to clone.
+ * Returns appropriate error code that caller should return or
+ * zero in case the clone should be allowed.
+ */
+int generic_remap_checks(struct file *file_in, loff_t pos_in,
+                        struct file *file_out, loff_t pos_out,
+                        loff_t *req_count, unsigned int remap_flags)
+{
+       struct inode *inode_in = file_in->f_mapping->host;
+       struct inode *inode_out = file_out->f_mapping->host;
+       uint64_t count = *req_count;
+       uint64_t bcount;
+       loff_t size_in, size_out;
+       loff_t bs = inode_out->i_sb->s_blocksize;
+       int ret;
+
+       /* The start of both ranges must be aligned to an fs block. */
+       if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
+               return -EINVAL;
+
+       /* Ensure offsets don't wrap. */
+       if (pos_in + count < pos_in || pos_out + count < pos_out)
+               return -EINVAL;
+
+       size_in = i_size_read(inode_in);
+       size_out = i_size_read(inode_out);
+
+       /* Dedupe requires both ranges to be within EOF. */
+       if ((remap_flags & REMAP_FILE_DEDUP) &&
+           (pos_in >= size_in || pos_in + count > size_in ||
+            pos_out >= size_out || pos_out + count > size_out))
+               return -EINVAL;
+
+       /* Ensure the infile range is within the infile. */
+       if (pos_in >= size_in)
+               return -EINVAL;
+       count = min(count, size_in - (uint64_t)pos_in);
+
+       ret = generic_access_check_limits(file_in, pos_in, &count);
+       if (ret)
+               return ret;
+
+       ret = generic_write_check_limits(file_out, pos_out, &count);
+       if (ret)
+               return ret;
 
        /*
-        * LFS rule
+        * If the user wanted us to link to the infile's EOF, round up to the
+        * next block boundary for this check.
+        *
+        * Otherwise, make sure the count is also block-aligned, having
+        * already confirmed the starting offsets' block alignment.
         */
-       if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
-                               !(file->f_flags & O_LARGEFILE))) {
-               if (pos >= MAX_NON_LFS)
-                       return -EFBIG;
-               iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
+       if (pos_in + count == size_in) {
+               bcount = ALIGN(size_in, bs) - pos_in;
+       } else {
+               if (!IS_ALIGNED(count, bs))
+                       count = ALIGN_DOWN(count, bs);
+               bcount = count;
        }
 
+       /* Don't allow overlapped cloning within the same file. */
+       if (inode_in == inode_out &&
+           pos_out + bcount > pos_in &&
+           pos_out < pos_in + bcount)
+               return -EINVAL;
+
        /*
-        * Are we about to exceed the fs block limit ?
-        *
-        * If we have written data it becomes a short write.  If we have
-        * exceeded without writing data we send a signal and return EFBIG.
-        * Linus frestrict idea will clean these up nicely..
+        * We shortened the request but the caller can't deal with that, so
+        * bounce the request back to userspace.
         */
-       if (unlikely(pos >= inode->i_sb->s_maxbytes))
-               return -EFBIG;
+       if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
+               return -EINVAL;
 
-       iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
-       return iov_iter_count(from);
+       *req_count = count;
+       return 0;
 }
-EXPORT_SYMBOL(generic_write_checks);
 
 int pagecache_write_begin(struct file *file, struct address_space *mapping,
                                loff_t pos, unsigned len, unsigned flags,