Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

From the manpage:

If the nonblocking read() found no events to read, then it returns -1
with errno set to EAGAIN. In that case, we exit the loop.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

This can be done because of there are only 3 cases:
  - *at syscall: Either dfd is valid or path is absolute and DFD is
  ignored
  - !*at syscall & absolute path: dfd is ignored
  - !*at syscall & !absolute path: dfd defaults to AT_CWD and it behaves
  like a relative path on a !*at syscall.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

AT syscalls continue to work only with absolute paths.  This will be
fixed by next commits.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Move everything into a wrapper in preparation to work on rename(2).

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Should search_ci_path return -ENOENT, we can't assume the syscall would
fail, since we might be doing a file creation.  We could be smart and
catch it from the syscall parameters, but let's not do that now.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

After the first element.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

We can do it, as long as we don't mess up POSIX.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

event->len is not very accurate, so calculate with strlen.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Now that relocation logic is simpler and correct, we no longer a large
initial sg index vector.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

This is done with libunistring.  Since normalization is expensive, we
try to do it only once and cache the result in a pointer stored inside
unused fields of struct record.  We identify if we already did the
normalization for a specific record by flags in the unused bits of
record->d_type.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

This allows us to verify two paths with one test.  If the file was
pre-created, the dcache record buffer will be filled.  Otherwise, the
dynamically created list will be tested.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

This will be reused by the dynamic delete code.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Dynamic unlinking, is what we are calling a file/directory unlink that
occurs after the directory cache was built by libcasefold.  This test
tries to verify the cache was invalidated correctly.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Only directory dentries have valid ownership.  The plan is to handle
directory creation when element is removed from cache.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

We might benefit from reducing the pathwalk in kernelspace

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

New syscalls supported:
 * futimesat

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

New syscalls supported:
 * setxattr
 * lsetxattr
 * getxattr
 * lgetxattr
 * listxattr
 * llistxattr
 * removexattr
 * lremovexattr

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

We don't actually care about fchown(2) because it handles fd instead of
files, the one we actually care is chown(2).

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

New syscalls supported:
 * acct
 * chroot
 * readlink
 * swapon
 * swapoff
 * truncate
 * statfs
 * readlinkat

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>

New syscalls supported:
 * mkdir
 * mkdirat
 * rmdir

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.co.uk>