X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fsymlink.c;h=f005cd715f025ef2730472001669f1e68007afab;hp=3e89a88eead6cfed31980c86f9f4fd51a2e5ca4e;hb=ca8727cf22a445adc20eb1f8b7a12b8eededc32d;hpb=a2c56fd332b6c37026b24c8a216e0ff0bf140d80

diff --git a/src/symlink.c b/src/symlink.c
index 3e89a88e..f005cd71 100644
--- a/src/symlink.c
+++ b/src/symlink.c
@@ -5,7 +5,7 @@
  */
 
 /*
- * Copyright (C) 2012 Eric Biggers
+ * Copyright (C) 2012, 2013 Eric Biggers
  *
  * This file is part of wimlib, a library for working with WIM files.
  *
@@ -24,35 +24,46 @@
  */
 
 #include "dentry.h"
-#include "io.h"
+#include "buffer_io.h"
 #include "lookup_table.h"
 #include "sha1.h"
 #include <errno.h>
 
+/* UNIX version of getting and setting the data in reparse points */
+#if !defined(__WIN32__)
+
+#include <sys/stat.h>
+
+#ifdef HAVE_ALLOCA_H
+#  include <alloca.h>
+#endif
+
 /*
  * Find the symlink target of a symbolic link or junction point in the WIM.
  *
- * See http://msdn.microsoft.com/en-us/library/cc232006(v=prot.10).aspx
- * Except the first 8 bytes aren't included in the resource (presumably because
- * we already know the reparse tag from the dentry, and we already know the
- * reparse tag len from the lookup table entry resource length).
+ * See http://msdn.microsoft.com/en-us/library/cc232006(v=prot.10).aspx for a
+ * description of the format of the so-called "reparse point data buffers".
+ *
+ * But, in the WIM format, the first 8 bytes of the reparse point data buffer
+ * are omitted, presumably because we already know the reparse tag from the
+ * dentry, and we already know the reparse tag length from the lookup table
+ * entry resource length.
  */
-static ssize_t get_symlink_name(const u8 *resource, size_t resource_len,
-			        char *buf, size_t buf_len,
-			        u32 reparse_tag)
+static ssize_t
+get_symlink_name(const void *resource, size_t resource_len, char *buf,
+		 size_t buf_len, u32 reparse_tag)
 {
-	const u8 *p = resource;
+	const void *p = resource;
 	u16 substitute_name_offset;
 	u16 substitute_name_len;
 	u16 print_name_offset;
 	u16 print_name_len;
 	char *link_target;
+	char *translated_target;
 	size_t link_target_len;
 	ssize_t ret;
 	unsigned header_size;
-	char *translated_target;
-	bool is_absolute;
-	u32 flags;
+	bool translate_slashes;
 
 	if (resource_len < 12)
 		return -EIO;
@@ -60,113 +71,230 @@ static ssize_t get_symlink_name(const u8 *resource, size_t resource_len,
 	p = get_u16(p, &substitute_name_len);
 	p = get_u16(p, &print_name_offset);
 	p = get_u16(p, &print_name_len);
-	get_u32(p, &flags);
 
 	wimlib_assert(reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK ||
 		      reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT);
 
-	/* I think that some junction points incorrectly get marked as symbolic
-	 * links.  So, parse the link buffer as a symlink if the flags seem
-	 * plausible. */
-	if (flags <= 1)
-		reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
-
-	if (reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT) {
+	if (reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT)
 		header_size = 8;
-	} else {
-		is_absolute = (flags & 1) ? false : true;
+	else {
 		header_size = 12;
 		p += 4;
 	}
-	if (header_size + substitute_name_offset + substitute_name_len > resource_len)
+	if (header_size +
+	    substitute_name_offset + substitute_name_len > resource_len)
 		return -EIO;
-	link_target = utf16_to_utf8((const char *)p + substitute_name_offset,
-				    substitute_name_len,
-				    &link_target_len);
 
-	if (!link_target)
-		return -EIO;
+	ret = utf16le_to_tstr((const utf16lechar*)(p + substitute_name_offset),
+			      substitute_name_len,
+			      &link_target, &link_target_len);
+	if (ret)
+		return -errno;
 
-	if (link_target_len + 1 > buf_len) {
-		ret = -ENAMETOOLONG;
+	DEBUG("Interpeting substitute name \"%s\" (ReparseTag=0x%x)",
+	      link_target, reparse_tag);
+	translate_slashes = true;
+	translated_target = link_target;
+	if (link_target_len >= 7 &&
+	    translated_target[0] == '\\' &&
+	    translated_target[1] == '?' &&
+	    translated_target[2] == '?' &&
+	    translated_target[3] == '\\' &&
+	    translated_target[4] != '\0' &&
+	    translated_target[5] == ':' &&
+	    translated_target[6] == '\\')
+	{
+		/* "Full" symlink or junction (\??\x:\ prefixed path) */
+		translated_target += 6;
+		link_target_len -= 6;
+	} else if (reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT &&
+		   link_target_len >= 12 &&
+		   memcmp(translated_target, "\\\\?\\Volume{", 11) == 0 &&
+		   translated_target[link_target_len - 1] == '\\')
+	{
+		/* Volume junction.  Can't really do anything with it. */
+		translate_slashes = false;
+	} else if (reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK &&
+		   link_target_len >= 3 &&
+		   translated_target[0] != '\0' &&
+		   translated_target[1] == ':' &&
+		   translated_target[2] == '\\')
+	{
+		/* "Absolute" symlink, with drive letter */
+		translated_target += 2;
+		link_target_len -= 2;
+	} else if (reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK &&
+		   link_target_len >= 1)
+	{
+		if (translated_target[0] == '\\')
+			/* "Absolute" symlink, without drive letter */
+			;
+		else
+			/* "Relative" symlink, without drive letter */
+			;
+	} else {
+		ERROR("Invalid reparse point substitute name: \"%s\"", translated_target);
+		ret = -EIO;
 		goto out;
 	}
 
-	translated_target = link_target;
-	if (reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT || is_absolute) {
-		if (link_target_len < 7
-		      || memcmp(translated_target, "\\??\\", 4) != 0
-		      || translated_target[4] == '\0'
-		      || translated_target[5] != ':'
-		      || translated_target[6] != '\\') {
-			ret = -EIO;
-			goto out;
-		}
-		translated_target += 4;
-		link_target_len -= 4;
-		/* There's a drive letter, so just leave the backslashes since
-		 * it won't go anyhwere on UNIX anyway...
-		 *
-		 * XXX
-		 * NTFS-3g tries to re-map these links to actually point to
-		 * something, so maybe we could do something like that here
-		 * XXX*/
-	} else {
+	if (translate_slashes)
 		for (size_t i = 0; i < link_target_len; i++)
 			if (translated_target[i] == '\\')
 				translated_target[i] = '/';
-	}
 
-	memcpy(buf, translated_target, link_target_len + 1);
-	ret = link_target_len;
+	if (link_target_len > buf_len) {
+		link_target_len = buf_len;
+		ret = -ENAMETOOLONG;
+	} else {
+		ret = link_target_len;
+	}
+	memcpy(buf, translated_target, link_target_len);
 out:
 	FREE(link_target);
 	return ret;
 }
 
-void *make_symlink_reparse_data_buf(const char *symlink_target, size_t *len_ret)
+#define SYMBOLIC_LINK_RELATIVE 0x00000001
+
+/* Given a UNIX symlink target, prepare the corresponding symbolic link reparse
+ * data buffer. */
+static int
+make_symlink_reparse_data_buf(const char *symlink_target, void *rpdata,
+			      size_t *rplen_ret)
 {
-	size_t utf8_len = strlen(symlink_target);
-	size_t utf16_len;
-	char *name_utf16 = utf8_to_utf16(symlink_target, utf8_len, &utf16_len);
-	if (!name_utf16)
-		return NULL;
-	/*DEBUG("utf16_len = %zu", utf16_len);*/
-	for (size_t i = 0; i < utf16_len / 2; i++)
-		if (((u16*)name_utf16)[i] == cpu_to_le16('/'))
-			((u16*)name_utf16)[i] = cpu_to_le16('\\');
-	size_t len = 12 + utf16_len * 2 + 4;
-	void *buf = MALLOC(len);
-	if (!buf)
-		goto out;
-	/* XXX Fix absolute paths */
-
-	u8 *p = buf;
-	p = put_u16(p, utf16_len + 2); /* Substitute name offset */
-	p = put_u16(p, utf16_len); /* Substitute name length */
-	p = put_u16(p, 0); /* Print name offset */
-	p = put_u16(p, utf16_len); /* Print name length */
-	p = put_u32(p, 1);
-	p = put_bytes(p, utf16_len, (const u8*)name_utf16);
+	int ret;
+	utf16lechar *name_utf16le;
+	size_t name_utf16le_nbytes;
+	size_t substitute_name_nbytes;
+	size_t print_name_nbytes;
+	static const char abs_subst_name_prefix[12] = "\\\0?\0?\0\\\0C\0:\0";
+	static const char abs_print_name_prefix[4] = "C\0:\0";
+	u32 flags;
+	size_t rplen;
+	void *p;
+
+	ret = tstr_to_utf16le(symlink_target, strlen(symlink_target),
+			      &name_utf16le, &name_utf16le_nbytes);
+	if (ret)
+		return ret;
+
+	for (size_t i = 0; i < name_utf16le_nbytes / 2; i++)
+		if (name_utf16le[i] == cpu_to_le16('/'))
+			name_utf16le[i] = cpu_to_le16('\\');
+
+	/* Compatability notes:
+	 *
+	 * On UNIX, an absolute symbolic link begins with '/'; everything else
+	 * is a relative symbolic link.  (Quite simple compared to the various
+	 * ways to provide Windows paths.)
+	 *
+	 * To change a UNIX relative symbolic link to Windows format, we only
+	 * need to translate it to UTF-16LE and replace backslashes with forward
+	 * slashes.  We do not make any attempt to handle filename character
+	 * problems, such as a link target that itself contains backslashes on
+	 * UNIX.  Then, for these relative links, we set the reparse header
+	 * @flags field to SYMBOLIC_LINK_RELATIVE.
+	 *
+	 * For UNIX absolute symbolic links, we must set the @flags field to 0.
+	 * Then, there are multiple options as to actually represent the
+	 * absolute link targets:
+	 *
+	 * (1) An absolute path beginning with one backslash character. similar
+	 * to UNIX-style, just with a different path separator.  Print name same
+	 * as substitute name.
+	 *
+	 * (2) Absolute path beginning with drive letter followed by a
+	 * backslash.  Print name same as substitute name.
+	 *
+	 * (3) Absolute path beginning with drive letter followed by a
+	 * backslash; substitute name prefixed with \??\, otherwise same as
+	 * print name.
+	 *
+	 * We choose option (3) here, and we just assume C: for the drive
+	 * letter.  The reasoning for this is:
+	 *
+	 * (1) Microsoft imagex.exe has a bug where it does not attempt to do
+	 * reparse point fixups for these links, even though they are valid
+	 * absolute links.  (Note: in this case prefixing the substitute name
+	 * with \??\ does not work; it just makes the data unable to be restored
+	 * at all.)
+	 * (2) Microsoft imagex.exe will fail when doing reparse point fixups
+	 * for these.  It apparently contains a bug that causes it to create an
+	 * invalid reparse point, which then cannot be restored.
+	 * (3) This is the only option I tested for which reparse point fixups
+	 * worked properly in Microsoft imagex.exe.
+	 *
+	 * So option (3) it is.
+	 */
+
+	substitute_name_nbytes = name_utf16le_nbytes;
+	print_name_nbytes = name_utf16le_nbytes;
+	if (symlink_target[0] == '/') {
+		substitute_name_nbytes += sizeof(abs_subst_name_prefix);
+		print_name_nbytes += sizeof(abs_print_name_prefix);
+	}
+
+	rplen = 12 + substitute_name_nbytes + print_name_nbytes +
+			2 * sizeof(utf16lechar);
+
+	if (rplen > REPARSE_POINT_MAX_SIZE) {
+		ERROR("Symlink \"%s\" is too long!", symlink_target);
+		return WIMLIB_ERR_LINK;
+	}
+
+	p = rpdata;
+
+	/* Substitute name offset */
 	p = put_u16(p, 0);
-	p = put_bytes(p, utf16_len, (const u8*)name_utf16);
+
+	/* Substitute name length */
+	p = put_u16(p, substitute_name_nbytes);
+
+	/* Print name offset */
+	p = put_u16(p, substitute_name_nbytes + sizeof(utf16lechar));
+
+	/* Print name length */
+	p = put_u16(p, print_name_nbytes);
+
+	/* Flags */
+	flags = 0;
+	if (symlink_target[0] != '/')
+		flags |= SYMBOLIC_LINK_RELATIVE;
+	p = put_u32(p, flags);
+
+	/* Substitute name */
+	if (symlink_target[0] == '/')
+		p = put_bytes(p, sizeof(abs_subst_name_prefix), abs_subst_name_prefix);
+	p = put_bytes(p, name_utf16le_nbytes, name_utf16le);
 	p = put_u16(p, 0);
-	/*DEBUG("utf16_len = %zu, len = %zu", utf16_len, len);*/
-	*len_ret = len;
-out:
-	FREE(name_utf16);
-	return buf;
+
+	/* Print name */
+	if (symlink_target[0] == '/')
+		p = put_bytes(p, sizeof(abs_print_name_prefix), abs_print_name_prefix);
+	p = put_bytes(p, name_utf16le_nbytes, name_utf16le);
+	p = put_u16(p, 0);
+
+	*rplen_ret = rplen;
+	ret = 0;
+out_free_name_utf16le:
+	FREE(name_utf16le);
+	return ret;
 }
 
-/* Get the symlink target from a dentry.
+/* Get the symlink target from a WIM inode.
  *
- * The dentry may be either "real" symlink or a junction point.
+ * The inode may be either a "real" symlink (reparse tag
+ * WIM_IO_REPARSE_TAG_SYMLINK), or it may be a junction point (reparse tag
+ * WIM_IO_REPARSE_TAG_MOUNT_POINT).
  */
-ssize_t inode_readlink(const struct inode *inode, char *buf, size_t buf_len,
-		       const WIMStruct *w)
+ssize_t
+inode_readlink(const struct wim_inode *inode, char *buf, size_t buf_len,
+	       const WIMStruct *w, bool threadsafe)
 {
-	const struct lookup_table_entry *lte;
+	const struct wim_lookup_table_entry *lte;
 	int ret;
+	u8 *res_buf;
 
 	wimlib_assert(inode_is_symlink(inode));
 
@@ -174,15 +302,15 @@ ssize_t inode_readlink(const struct inode *inode, char *buf, size_t buf_len,
 	if (!lte)
 		return -EIO;
 
-	if (wim_resource_size(lte) > 10000)
+	if (wim_resource_size(lte) > REPARSE_POINT_MAX_SIZE)
 		return -EIO;
 
-	u8 res_buf[wim_resource_size(lte)];
-	ret = read_full_wim_resource(lte, res_buf);
-	if (ret != 0)
+	res_buf = alloca(wim_resource_size(lte));
+	ret = read_full_resource_into_buf(lte, res_buf, threadsafe);
+	if (ret)
 		return -EIO;
-	return get_symlink_name(res_buf, wim_resource_size(lte), buf,
-				buf_len, inode->reparse_tag);
+	return get_symlink_name(res_buf, wim_resource_size(lte),
+				buf, buf_len, inode->i_reparse_tag);
 }
 
 /*
@@ -196,59 +324,110 @@ ssize_t inode_readlink(const struct inode *inode, char *buf, size_t buf_len,
  *
  * On failure @dentry and @lookup_table are not modified.
  */
-int inode_set_symlink(struct inode *inode, const char *target,
-		      struct lookup_table *lookup_table,
-		      struct lookup_table_entry **lte_ret)
+int
+inode_set_symlink(struct wim_inode *inode,
+		  const char *target,
+		  struct wim_lookup_table *lookup_table,
+		  struct wim_lookup_table_entry **lte_ret)
 
 {
 	int ret;
-	size_t symlink_buf_len;
-	struct lookup_table_entry *lte = NULL, *existing_lte;
-	u8 symlink_buf_hash[SHA1_HASH_SIZE];
-	void *symlink_buf;
 
-	symlink_buf = make_symlink_reparse_data_buf(target, &symlink_buf_len);
-	if (!symlink_buf)
-		return WIMLIB_ERR_NOMEM;
+	/* Buffer for reparse point data */
+	u8 rpdata[REPARSE_POINT_MAX_SIZE];
+
+	/* Actual length of the reparse point data (to be calculated by
+	 * make_symlink_reparse_data_buf()) */
+	size_t rplen;
+
+	DEBUG("Creating reparse point data buffer "
+	      "for UNIX symlink target \"%s\"", target);
 
-	DEBUG("Made symlink reparse data buf (len = %zu, name len = %zu)",
-			symlink_buf_len, symlink_buf_len);
+	ret = make_symlink_reparse_data_buf(target, rpdata, &rplen);
+	if (ret)
+		return ret;
 
-	sha1_buffer(symlink_buf, symlink_buf_len, symlink_buf_hash);
+	ret = inode_set_unnamed_stream(inode, rpdata, rplen, lookup_table);
+	if (ret)
+		return ret;
 
-	existing_lte = __lookup_resource(lookup_table, symlink_buf_hash);
+	if (lte_ret)
+		*lte_ret = inode->i_lte;
+	return 0;
+}
 
-	if (existing_lte) {
-		lte = existing_lte;
-		FREE(symlink_buf);
-		symlink_buf = NULL;
+static int
+unix_get_ino_and_dev(const char *path, u64 *ino_ret, u64 *dev_ret)
+{
+	struct stat stbuf;
+	if (stat(path, &stbuf)) {
+		WARNING_WITH_ERRNO("Failed to stat \"%s\"", path);
+		/* Treat as a link pointing outside the capture root (it
+		 * most likely is). */
+		return WIMLIB_ERR_STAT;
 	} else {
-		DEBUG("Creating new lookup table entry for symlink buf");
-		lte = new_lookup_table_entry();
-		if (!lte) {
-			ret = WIMLIB_ERR_NOMEM;
-			goto out_free_symlink_buf;
-		}
-		lte->resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
-		lte->attached_buffer              = symlink_buf;
-		lte->resource_entry.original_size = symlink_buf_len;
-		lte->resource_entry.size          = symlink_buf_len;
-		copy_hash(lte->hash, symlink_buf_hash);
+		*ino_ret = stbuf.st_ino;
+		*dev_ret = stbuf.st_dev;
+		return 0;
 	}
+}
 
-	inode->lte = lte;
-	inode->resolved = true;
+#endif /* !defined(__WIN32__) */
 
-	DEBUG("Loaded symlink buf");
+#ifdef __WIN32__
+#  include "win32.h"
+#  define RP_PATH_SEPARATOR L'\\'
+#  define is_rp_path_separator(c) ((c) == L'\\' || (c) == L'/')
+#  define os_get_ino_and_dev win32_get_file_and_vol_ids
+#else
+#  define RP_PATH_SEPARATOR '/'
+#  define is_rp_path_separator(c) ((c) == '/')
+#  define os_get_ino_and_dev unix_get_ino_and_dev
+#endif
 
-	if (existing_lte)
-		lte->refcnt++;
-	else
-		lookup_table_insert(lookup_table, lte);
-	if (lte_ret)
-		*lte_ret = lte;
-	return 0;
-out_free_symlink_buf:
-	FREE(symlink_buf);
-	return ret;
+/* Fix up absolute symbolic link targets--- mostly shared between UNIX and
+ * Windows */
+tchar *
+fixup_symlink(tchar *dest, u64 capture_root_ino, u64 capture_root_dev)
+{
+	tchar *p = dest;
+
+	DEBUG("Fixing symlink or junction \"%"TS"\"", dest);
+	for (;;) {
+		tchar save;
+		int ret;
+		u64 ino;
+		u64 dev;
+
+		while (is_rp_path_separator(*p))
+			p++;
+
+		save = *p;
+		*p = T('\0');
+		ret = os_get_ino_and_dev(dest, &ino, &dev);
+		*p = save;
+
+		if (ret) /* stat() failed before we got to the capture root---
+			    assume the link points outside it. */
+			return NULL;
+
+		if (ino == capture_root_ino && dev == capture_root_dev) {
+			/* Link points inside capture root.  Return abbreviated
+			 * path. */
+			if (*p == T('\0'))
+				*(p - 1) = RP_PATH_SEPARATOR;
+			while (p - 1 >= dest && is_rp_path_separator(*(p - 1)))
+				p--;
+			return p;
+		}
+
+		if (*p == T('\0')) {
+			/* Link points outside capture root. */
+			return NULL;
+		}
+
+		do {
+			p++;
+		} while (!is_rp_path_separator(*p) && *p != T('\0'));
+	}
 }