Explorar el Código

Finally adope the patch from Morten Hustveit <morten@debian.org> to be
able to optional use mmaps and iovec to increase patch speed -
but as this increase memory usage we can always fall back to the "old"
method which doesn't relay on mmaps.

David Kalnischkies hace 16 años
padre
commit
1a37c61961
Se han modificado 1 ficheros con 243 adiciones y 18 borrados
  1. 243 18
      methods/rred.cc

+ 243 - 18
methods/rred.cc

@@ -1,11 +1,13 @@
 // Includes									/*{{{*/
 #include <apt-pkg/fileutl.h>
+#include <apt-pkg/mmap.h>
 #include <apt-pkg/error.h>
 #include <apt-pkg/acquire-method.h>
 #include <apt-pkg/strutl.h>
 #include <apt-pkg/hashes.h>
 
 #include <sys/stat.h>
+#include <sys/uio.h>
 #include <unistd.h>
 #include <utime.h>
 #include <stdio.h>
@@ -29,7 +31,7 @@ class RredMethod : public pkgAcqMethod {
 	// the supported ed commands
 	enum Mode {MODE_CHANGED='c', MODE_DELETED='d', MODE_ADDED='a'};
 	// return values
-	enum State {ED_OK=0, ED_ORDERING=1, ED_PARSER=2, ED_FAILURE=3};
+	enum State {ED_OK, ED_ORDERING, ED_PARSER, ED_FAILURE, MMAP_FAILED};
 
 	State applyFile(FILE *ed_cmds, FILE *in_file, FILE *out_file,
 	             unsigned long &line, char *buffer, Hashes *hash) const;
@@ -37,7 +39,8 @@ class RredMethod : public pkgAcqMethod {
 	void copyLinesFromFileToFile(FILE *fin, FILE *fout, unsigned int lines,
 	                            Hashes *hash, char *buffer) const;
 
-	State patchFile(FILE *ed_cmds, FILE *in_file, FILE *out_file, Hashes *hash) const;
+	State patchFile(FileFd &Patch, FileFd &From, FileFd &out_file, Hashes *hash) const;
+	State patchMMap(FileFd &Patch, FileFd &From, FileFd &out_file, Hashes *hash) const;
 
 protected:
 	// the methods main method
@@ -67,7 +70,7 @@ RredMethod::State RredMethod::applyFile(FILE *ed_cmds, FILE *in_file, FILE *out_
 	// get the current command and parse it
 	if (fgets(buffer, BUF_SIZE, ed_cmds) == NULL) {
 		if (Debug == true)
-			std::clog << "rred: encounter end of file - we can start patching now.";
+			std::clog << "rred: encounter end of file - we can start patching now." << std::endl;
 		line = 0;
 		return ED_OK;
 	}
@@ -189,22 +192,240 @@ void RredMethod::ignoreLineInFile(FILE *fin, char *buffer) const {		/*{{{*/
 	}
 }
 										/*}}}*/
-RredMethod::State RredMethod::patchFile(FILE *ed_cmds, FILE *in_file, FILE *out_file,		/*{{{*/
-      Hashes *hash) const {
+RredMethod::State RredMethod::patchFile(FileFd &Patch, FileFd &From,		/*{{{*/
+					FileFd &out_file, Hashes *hash) const {
    char buffer[BUF_SIZE];
-   
+   FILE* fFrom = fdopen(From.Fd(), "r");
+   FILE* fPatch = fdopen(Patch.Fd(), "r");
+   FILE* fTo = fdopen(out_file.Fd(), "w");
+
    /* we do a tail recursion to read the commands in the right order */
    unsigned long line = -1; // assign highest possible value
-   State result = applyFile(ed_cmds, in_file, out_file, line, buffer, hash);
+   State const result = applyFile(fPatch, fFrom, fTo, line, buffer, hash);
    
    /* read the rest from infile */
    if (result == ED_OK) {
-      while (fgets(buffer, BUF_SIZE, in_file) != NULL) {
-         size_t const written = fwrite(buffer, 1, strlen(buffer), out_file);
+      while (fgets(buffer, BUF_SIZE, fFrom) != NULL) {
+         size_t const written = fwrite(buffer, 1, strlen(buffer), fTo);
          hash->Add((unsigned char*)buffer, written);
       }
+      fflush(fTo);
    }
    return result;
+}
+										/*}}}*/
+struct EdCommand {								/*{{{*/
+  size_t data_start;
+  size_t data_end;
+  size_t data_lines;
+  size_t first_line;
+  size_t last_line;
+  char type;
+};
+#define IOV_COUNT 1024 /* Don't really want IOV_MAX since it can be arbitrarily large */
+										/*}}}*/
+RredMethod::State RredMethod::patchMMap(FileFd &Patch, FileFd &From,		/*{{{*/
+					FileFd &out_file, Hashes *hash) const {
+#ifdef _POSIX_MAPPED_FILES
+	MMap ed_cmds(Patch, MMap::ReadOnly);
+	MMap in_file(From, MMap::ReadOnly);
+	FILE* fTo = fdopen(out_file.Fd(), "w");
+
+	if (ed_cmds.Size() == 0 || in_file.Size() == 0)
+		return MMAP_FAILED;
+
+	EdCommand* commands = 0;
+	size_t command_count = 0;
+	size_t command_alloc = 0;
+
+	const char* begin = (char*) ed_cmds.Data();
+	const char* end = begin;
+	const char* ed_end = (char*) ed_cmds.Data() + ed_cmds.Size();
+
+	const char* input = (char*) in_file.Data();
+	const char* input_end = (char*) in_file.Data() + in_file.Size();
+
+	size_t i;
+
+	/* 1. Parse entire script.  It is executed in reverse order, so we cather it
+	 *    in the `commands' buffer first
+	 */
+
+	for(;;) {
+		EdCommand cmd;
+		cmd.data_start = 0;
+		cmd.data_end = 0;
+
+		while(begin != ed_end && *begin == '\n')
+			++begin;
+		while(end != ed_end && *end != '\n')
+			++end;
+		if(end == ed_end && begin == end)
+			break;
+
+		/* Determine command range */
+		const char* tmp = begin;
+
+		for(;;) {
+			/* atoll is safe despite lacking NUL-termination; we know there's an
+			 * alphabetic character at end[-1]
+			 */
+			if(tmp == end) {
+				cmd.first_line = atol(begin);
+				cmd.last_line = cmd.first_line;
+				break;
+			}
+			if(*tmp == ',') {
+				cmd.first_line = atol(begin);
+				cmd.last_line = atol(tmp + 1);
+				break;
+			}
+			++tmp;
+		}
+
+		// which command to execute on this line(s)?
+		switch (end[-1]) {
+			case MODE_CHANGED:
+				if (Debug == true)
+					std::clog << "Change from line " << cmd.first_line << " to " << cmd.last_line << std::endl;
+				break;
+			case MODE_ADDED:
+				if (Debug == true)
+					std::clog << "Insert after line " << cmd.first_line << std::endl;
+				break;
+			case MODE_DELETED:
+				if (Debug == true)
+					std::clog << "Delete from line " << cmd.first_line << " to " << cmd.last_line << std::endl;
+				break;
+			default:
+				_error->Error("rred: Unknown ed command '%c'. Abort.", end[-1]);
+				free(commands);
+				return ED_PARSER;
+		}
+		cmd.type = end[-1];
+
+		/* Determine the size of the inserted text, so we don't have to scan this
+		 * text again later.
+		 */
+		begin = end + 1;
+		end = begin;
+		cmd.data_lines = 0;
+
+		if(cmd.type == MODE_ADDED || cmd.type == MODE_CHANGED) {
+			cmd.data_start = begin - (char*) ed_cmds.Data();
+			while(end != ed_end) {
+				if(*end == '\n') {
+					if(end[-1] == '.' && end[-2] == '\n')
+						break;
+					++cmd.data_lines;
+				}
+				++end;
+			}
+			cmd.data_end = end - (char*) ed_cmds.Data() - 1;
+			begin = end + 1;
+			end = begin;
+		}
+		if(command_count == command_alloc) {
+			command_alloc = (command_alloc + 64) * 3 / 2;
+			commands = (EdCommand*) realloc(commands, command_alloc * sizeof(EdCommand));
+		}
+		commands[command_count++] = cmd;
+	}
+
+	struct iovec* iov = new struct iovec[IOV_COUNT];
+	size_t iov_size = 0;
+
+	size_t amount, remaining;
+	size_t line = 1;
+	EdCommand* cmd;
+
+	/* 2. Execute script.  We gather writes in a `struct iov' array, and flush
+	 *    using writev to minimize the number of system calls.  Data is read
+	 *    directly from the memory mappings of the input file and the script.
+	 */
+
+	for(i = command_count; i-- > 0; ) {
+		cmd = &commands[i];
+		if(cmd->type == MODE_ADDED)
+			amount = cmd->first_line + 1;
+		else
+			amount = cmd->first_line;
+
+		if(line < amount) {
+			begin = input;
+			while(line != amount) {
+				input = (const char*) memchr(input, '\n', input_end - input);
+				if(!input)
+					break;
+				++line;
+				++input;
+			}
+
+			iov[iov_size].iov_base = (void*) begin;
+			iov[iov_size].iov_len = input - begin;
+			hash->Add((const unsigned char*) begin, input - begin);
+
+			if(++iov_size == IOV_COUNT) {
+				writev(out_file.Fd(), iov, IOV_COUNT);
+				iov_size = 0;
+			}
+		}
+
+		if(cmd->type == MODE_DELETED || cmd->type == MODE_CHANGED) {
+			remaining = (cmd->last_line - cmd->first_line) + 1;
+			line += remaining;
+			while(remaining) {
+				input = (const char*) memchr(input, '\n', input_end - input);
+				if(!input)
+					break;
+				--remaining;
+				++input;
+			}
+		}
+
+		if(cmd->type == MODE_CHANGED || cmd->type == MODE_ADDED) {
+			if(cmd->data_end != cmd->data_start) {
+				iov[iov_size].iov_base = (void*) ((char*)ed_cmds.Data() + cmd->data_start);
+				iov[iov_size].iov_len = cmd->data_end - cmd->data_start;
+				hash->Add((const unsigned char*) ((char*)ed_cmds.Data() + cmd->data_start),
+				iov[iov_size].iov_len);
+
+				if(++iov_size == IOV_COUNT) {
+					writev(out_file.Fd(), iov, IOV_COUNT);
+					iov_size = 0;
+				}
+			}
+		}
+	}
+
+	if(input != input_end) {
+		iov[iov_size].iov_base = (void*) input;
+		iov[iov_size].iov_len = input_end - input;
+		hash->Add((const unsigned char*) input, input_end - input);
+		++iov_size;
+	}
+
+	if(iov_size) {
+		writev(out_file.Fd(), iov, iov_size);
+		iov_size = 0;
+	}
+
+	for(i = 0; i < iov_size; i += IOV_COUNT) {
+		if(iov_size - i < IOV_COUNT)
+			writev(out_file.Fd(), iov + i, iov_size - i);
+		else
+			writev(out_file.Fd(), iov + i, IOV_COUNT);
+	}
+
+	delete [] iov;
+	free(commands);
+
+	fflush(fTo);
+
+	return ED_OK;
+#else
+	return MMAP_FAILED;
+#endif
 }
 										/*}}}*/
 bool RredMethod::Fetch(FetchItem *Itm)						/*{{{*/
@@ -234,19 +455,23 @@ bool RredMethod::Fetch(FetchItem *Itm)						/*{{{*/
       return false;
    
    Hashes Hash;
-   FILE* fFrom = fdopen(From.Fd(), "r");
-   FILE* fPatch = fdopen(Patch.Fd(), "r");
-   FILE* fTo = fdopen(To.Fd(), "w");
    // now do the actual patching
-   if (patchFile(fPatch, fFrom, fTo, &Hash) != ED_OK) {
-     _error->Errno("rred", _("Could not patch file"));  
-      return false;
+   State const result = patchMMap(Patch, From, To, &Hash);
+   if (result == MMAP_FAILED) {
+      // retry with patchFile
+      lseek(Patch.Fd(), 0, SEEK_SET);
+      lseek(From.Fd(), 0, SEEK_SET);
+      To.Open(Itm->DestFile,FileFd::WriteEmpty);
+      if (_error->PendingError() == true)
+         return false;
+      if (patchFile(Patch, From, To, &Hash) != ED_OK) {
+	 return _error->Errno("rred", _("Could not patch file %s"), Path.append(" (1)").c_str());
+      }
+   } else if (result != ED_OK) {
+      return _error->Errno("rred", _("Could not patch file %s"), Path.append(" (2)").c_str());
    }
 
    // write out the result
-   fflush(fFrom);
-   fflush(fPatch);
-   fflush(fTo);
    From.Close();
    Patch.Close();
    To.Close();