| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Shell scripts moved from Unix may contain hard-coded paths to
binaries such as /bin/sh. A recent commit made it possible to
execute such binaries reliably, but that does require them to be
installed. As an alternative solution: if a binary with a
standard Unix path prefix can't be found but is available as a
built-in applet, run the applet.
Add the function unix_path() to detect paths starting with /bin,
/usr/bin, /sbin or /usr/sbin.
Use this function in:
- the 'which' applet
- shellexec(), describe_command() and find_command() in ash
- mingw_spawn_1()
See GitHub issue #195.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As noted in commit 548ec7045 (win32: interpret absolute paths as
relative to %SYSTEMDRIVE%) a path starting with a '/' in the Unix
world is treated as relative to the current drive by Windows.
To avoid ambiguity that commit considered certain such paths to
be relative to %SYSTEMDRIVE%. Extend this to paths representing
executables.
Add the functions need_system_drive() and auto_add_system_drive()
to detect the need for a system drive prefix and to add it if
necessary. Use these functions in:
- the 'which' applet
- the find_executable() function
- tab-completion code
- PATH look-up, shellexec(), describe_command() and find_command() in ash
- parse_interpreter() and mingw_spawn_1()
With these changes executable paths starting with a slash are
handled consistently, whatever the current drive.
|
|
|
|
|
|
|
| |
Allocate static storage for the system drive string instead of
making a new allocation on every call. This is easier to manage.
Adds 16 bytes.
|
|
|
|
|
|
|
|
|
|
| |
Add a new function, has_path(), to detect that an executable name
doesn't require a path look-up.
Also, since is_absolute_path() is now only used in shell/ash.c move
its definition there from include/mingw.h.
Saves 128 bytes.
|
|
|
|
|
|
| |
Until now the emulated stat(2) system calls have only returned a
synthesised Unix-style mode value. Also return the raw Windows
file attributes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Be more strict about identifying UNC paths in unc_root_len().
In updatepwd() in ash:
- Skip duplicate leading slashes unless the directory is a UNC path.
- Rewrite detection and handling of the five possible types of path.
This improves cases like 'cd ///' and 'cd /xyz' when the current
directory is a UNC path.
See GitHub issue #192.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Change how busybox.exe expands wildcards on the command line.
When globbing is enabled at compile time provide an implementation
of _setargv(), which is run early during startup of C programs. This:
- enables globbing by setting _dowildcard to -1
- checks for the presence of the environment BB_GLOBBING
- if it exists and is set to 0 disables globbing
- if it doesn't exist sets BB_GLOBBING=0 but continues to apply
Windows' globbing in the current process
The consequences of this are:
- When busybox.exe is initially run from a Command Prompt Windows'
globbing is applied;
- Windows' globbing is turned off for future child processes, thus
allowing the shell re-execute busybox.exe without it interfering
with wildcards;
- this behaviour can be overridden by setting BB_GLOBBING explicitly.
Globbing can still be disabled at compile time if required. In that
case BB_GLOBBING has no effect.
With these changes globbing can be enabled by default and BusyBox
will do the right thing in most circumstances.
(See GitHub issues #172 and #189.)
|
|
|
|
|
|
|
|
|
| |
Only a handful of functions are used from shell32.dll, userenv.dll
and psapi.dll. Mostly these functions are in out of the way places.
By loading the functions only when required we can avoid the startup
cost of linking the three DLLs in the common case that they aren't
needed.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Commit e6680912a (dd: create a sparse file when seek=N is used)
broke the use of 'conv=notrunc seek=N' to modify existing files.
Rename seek_sparse() to make_sparse() and:
- add an argument to specify the start of the sparse region;
- call make_sparse() before ftruncate();
- call make_sparse() only if:
* we can determine the size of the file;
* the file is not open in append mode;
* the file is being extended.
This should fix GitHub issue #186.
|
|
|
|
|
|
|
| |
When the seek=N argument is used mark the file as sparse and set
the range that is sparse.
See https://stackoverflow.com/questions/4011508/how-to-create-a-sparse-file-on-ntfs
|
|
|
|
|
|
| |
Use GetCompressedFileSize to obtain the actual number of blocks
for compressed or sparse files. Use this to return a more accurate
value for st_blocks.
|
|
|
|
|
|
|
| |
Define NOT_DEVICE as -1 so the return value of index_in_strings()
when no device name is matched is NOT_DEVICE.
Make the logic in mingw_open() clearer.
|
|
|
|
|
|
| |
Improve get_dev_type() and use it to dectect /dev/null in mingw_xopen().
Saves 64 bytes.
|
|
|
|
|
| |
Have bs_to_slash() return a pointer to its argument. This allows
some calls to be chained, saving 32 bytes.
|
|
|
|
|
|
|
| |
Since we don't have a working clock_settime(2) there's no point
in claiming to support the '-s' option.
Saves 96 bytes
|
|
|
|
|
|
|
|
|
|
|
|
| |
When a format specification is replaced the loop variable 't'
should point to the last character of the replacement string in
the new format buffer.
In an extreme case if the original format string is "%z" and
tm->tm_isdst is negative to indicate that no DST information is
available the replacement string will be empty and 't' will point
to the location before the start of the new format buffer. This
is OK.
|
|
|
|
|
| |
The security descriptor was being freed before its contents were
accessed.
|
|
|
|
|
|
| |
- Update configuration files
- Omit unnecessary libraries
- Replace fake stime(2) with fake clock_settime(2)
|
|
|
|
|
|
| |
Modern Linux kernels use struct timespec to represent file times,
thus allowing nanosecond precision. Update the WIN32 emulation of
struct stat and stat(2) to do the same.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Standardise the path names used for the current working directory by:
- resolving with realpath(3);
- making the drive name or host name uppercase.
The first only really works for physical drives; results for mapped
drives are patchy.
The standardisation is applied in two places:
- at the end of updatepwd() in ash;
- when a symbolic link is resolved in mingw_chdir().
|
|
|
|
|
|
|
|
|
|
|
| |
Drives mapped to a network share or path didn't have their current
directory tracked when using path names of the form 'c:path'.
This was because commit 585d17d26 used realpath(3) to canonicalise
paths in chdir(2). Use readlink(2) instead so that mapped drives
aren't canonicalised but symlinks are resolved.
See GitHub issue #147.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Microsoft Windows permits path names of the form 'c:path', without a
path separator after the colon. The system records a current directory
for each drive and the path is interpreted relative to that.
Since Windows API calls understand 'c:path' path names many commands
in busybox-w32 already work with them. This commit adds the following:
- The 'cd' shell built-in interprets 'c:path' path names correctly.
Previously it treated them as relative to the shell's concept of
the current working directory, not the current directory of the
specified drive.
- The 'pwd' shell built-in takes the '-a' option to list the current
directory for all drives.
- 'c:path' path names are subject to tab-completion.
Paths of the form 'c:path' don't work for mapped network drives or
paths that have been associated with a drive using SUBST.
See GitHub issue #147.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
BusyBox contains hardcoded references to absolute paths which
are unique in the *nix world but on Microsoft Windows are
interpreted as being on the current drive. To make these unique
again consider them to be relative to %SYSTEMDRIVE%.
Support this by adding functions to:
- determine the system drive (not using the environment variable);
- change a process's current directory to the root of the system drive;
- make relative paths absolute before changing directory (if needed).
The following applications have been modified:
- ash references /etc/profile from the system drive;
- dpkg places its data store on and installs files to the system drive;
- rpm installs files to the system drive;
- man looks for configuration files and man pages on the system drive.
See GitHub issue #158.
|
|
|
|
|
|
|
| |
Move unc_root_len() from ash to mingw32.c and use it in the new
function root_len(), which can be used in make_directory().
This reduces changes to upstream code and saves a few bytes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Commit 325fee1f9 (win32: change handling of trailing slashes in
stat(2)) caused stat(2) to return EINVAL in some cases when ENOENT
might have been more appropriate.
This caused the command:
rm -f dir/*
to report an error when the directory was empty, contrary to its
expected behaviour of saying nothing.
Fixes GitHub issue #155.
|
|
|
|
|
|
|
|
|
| |
There are now two places where slashes are converted to backslashes
throughout a string so it makes sense to create a function to do
this.
To avoid confusion rename convert_slashes() to bs_to_slash() and
call the new function slash_to_bs().
|
|
|
|
|
|
|
|
|
| |
If the path is that of a drive mapped to a network share _fullpath()
leaves the trailing slash on the drive name and it remains present
after the call to resolve_symlinks().
Remove a trailing slash from the resolved path unless it's preceded
by a colon.
|
|
|
|
| |
Merge FEATURE_IDENTIFY_OWNER into FEATURE_EXTRA_FILE_DATA.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Formalise the use of 0 as the uid of a process running with elevated
privileges:
- Rewrite getuid(2) to return DEFAULT_UID by default and 0 if the
process has elevated privileges.
- geteuid(2) and the corresponding functions for groups are aliases
for getuid(2).
- Change root's home directory to be whatever GetSystemDirectory()
returns, probably C:/Windows/System32 in most cases.
- Remove the special handling of geteuid(2) in the line editing code.
With these changes the shell started by 'su' is a lot more like a
*nix root shell.
|
|
|
|
|
| |
Add is_admin() and use it to alter the command prompt in the line
editor when running with admin privileges.
|
|
|
|
|
|
|
|
|
|
|
| |
The code to normalise paths in resolve_symlinks(), which is used
by realpath(3), was incomplete and unable to handle UNC paths.
Make an ASCII version of normalize_ntpath() to extend the cases
covered.
This fixes a regression introduced by commit 585d17d26 (win32:
canonicalize path in chdir(2)): it wasn't possible to change
to a directory with a UNC path.
|
|
|
|
|
|
|
|
|
|
|
|
| |
Provide an implementation of chdir(2) which canonicalizes the
path to resolve symlinks. Otherwise changing to a symlinked
directory confuses 'ls -l' because it thinks '.' is a link
rather than a directory.
OTOH, using 'cd' in the shell to change to a symlinked directory
now results in a mismatch between the shell's idea of where we are
and what's displayed in the prompt. But upstream BusyBox does
that too so it must be OK.
|
|
|
|
|
|
|
|
|
|
| |
Provide an implementation of readlink(2) based on code from Git
for Windows. This version only supports symbolic links, not
mount points, as the latter seem to work well enough as-is.
With this change the ls and stat applets can display the targets
of symbolic links. The readlink applet has been enabled in the
default configuration.
|
|
|
|
|
|
| |
Allow realpath(3) to return successfully on platforms that don't
support GetFinalPathNameByHandleA(). It may still have done some
useful work.
|
| |
|
|
|
|
|
|
|
|
| |
The file size and times reported when Windows follows a symlink
are incorrect. To get the correct values canonicalize the path
and try again.
Also fetch the correct device id and inode for symlinks.
|
|
|
|
|
|
|
|
|
|
| |
The code to implement stat(2) works around the fact that getting
the attributes of a file fails if its name ends with a path
separator. Directory names with a trailing path separator work
fine.
Why bother with this workaround? Linux doesn't. Instead try to
return a meaningful error code.
|
|
|
|
|
|
|
|
|
|
|
| |
SetFileTime only needs FILE_WRITE_ATTRIBUTES access, not full
read/write access. Therefore it isn't necessary to change the
permissions of read-only files.
The flag FILE_FLAG_BACKUP_SEMANTICS is required to access directories
but does no harm if used on a file.
As a result there's no need to get file attributes.
|
|
|
|
|
|
|
|
|
|
|
| |
Further extend file identification so stat(2) returns the relative
identifier as a numeric uid for files with owner SIDs that look like
a local or domain user.
See:
https://blogs.technet.microsoft.com/markrussinovich/2009/11/03/the-machine-sid-duplication-myth-and-why-sysprep-matters/
https://cygwin.com/cygwin-ug-net/ntsec.html
|
|
|
|
|
| |
Extend stat(2) so it tries to determine whether a file belongs to
the current user or not. If not it's said to belong to root.
|
| |
|
|
|
|
|
| |
If stat(2) knows a file exists but can't obtain additional metadata
for it give it root ownership and no permissions for other.
|
|
|
|
|
|
|
|
|
| |
Some files can't be opened to fetch additional metadata. When
that happens allow stat(2) to successfully return what data it
has.
In a few cases where the inode number is used to determine if
files are identical ignore invalid inode numbers.
|
|
|
|
|
|
|
| |
Since st_nlink now depends on st_mode it should be set after
st_mode is initialised.
Rearrange the code to fetch extra metadata.
|
|
|
|
|
|
| |
When additional metadata was being fetched the code for non-disk
files used uninitialised data from the BY_HANDLE_FILE_INFORMATION
structure.
|
|
|
|
|
|
|
|
| |
Move the code to hide the console to a separate function in
win32/mingw.c. Use lazy loading to avoid problems on platforms
where the require APIs aren't supported (PR #70).
Enable console hiding in the default 64-bit configuration.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Modify the WIN32 implementation of stat(2) to fetch inode number,
device id and number of hardlinks. This requires opening a handle
to the target file so it will be slower.
A number of features can be enabled or start to work:
- tar can detect if an archive is being stored in itself;
- find can support the -inum and -links options;
- ls can display inode numbers;
- diff can detect attempts to compare a file with itself;
- du has better support for hardlinked files;
- cp can detect attempts to copy a file over itself.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
umask() in the Microsoft C runtime takes different arguments to
umask(2). Implement a fake umask(2) that remembers the mask and
calls the Microsoft umask() with an appropriate value.
Since the mask won't be inherited by children use an environment
variable to pass any value set by the shell built-in umask.
|