X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=c11c561e84f502c54d822ec50601f8052be08dc8;hb=b1a0bfe834d45694851787317da2cd55add2d7bb;hp=3e004fdfc29aee5af00282504fc94dcdc38f04aa;hpb=3cb5ea93e29ba5baf6a5d1f170bbc003d7e2fe35;p=cumulus.git

diff --git a/scandir.cc b/scandir.cc
index 3e004fd..c11c561 100644
--- a/scandir.cc
+++ b/scandir.cc
@@ -12,16 +12,27 @@
 
 #include <algorithm>
 #include <string>
+#include <list>
 #include <vector>
+#include <iostream>
+#include <sstream>
 
+#include "format.h"
 #include "store.h"
+#include "sha1.h"
 
+using std::list;
 using std::string;
 using std::vector;
+using std::ostream;
 
-static OutputStream *info_dump = NULL;
+static TarSegmentStore *tss = NULL;
 
-void scandir(const string& path);
+/* Buffer for holding a single block of data read from a file. */
+static const int LBS_BLOCK_SIZE = 1024 * 1024;
+static char *block_buf;
+
+void scandir(const string& path, std::ostream& metadata);
 
 /* Converts time to microseconds since the epoch. */
 int64_t encode_time(time_t time)
@@ -29,43 +40,109 @@ int64_t encode_time(time_t time)
     return (int64_t)time * 1000000;
 }
 
-void dumpfile(int fd)
+/* Read data from a file descriptor and return the amount of data read.  A
+ * short read (less than the requested size) will only occur if end-of-file is
+ * hit. */
+size_t file_read(int fd, char *buf, size_t maxlen)
+{
+    size_t bytes_read = 0;
+
+    while (true) {
+        ssize_t res = read(fd, buf, maxlen);
+        if (res < 0) {
+            if (errno == EINTR)
+                continue;
+            throw IOException("file_read: error reading");
+        } else if (res == 0) {
+            break;
+        } else {
+            bytes_read += res;
+            buf += res;
+            maxlen -= res;
+        }
+    }
+
+    return bytes_read;
+}
+
+/* Read the contents of a file (specified by an open file descriptor) and copy
+ * the data to the store. */
+void dumpfile(int fd, dictionary &file_info, ostream &metadata)
 {
     struct stat stat_buf;
     fstat(fd, &stat_buf);
     int64_t size = 0;
-
-    char buf[4096];
+    list<string> segment_list;
 
     if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
         printf("file is no longer a regular file!\n");
         return;
     }
 
+    /* The index data consists of a sequence of pointers to the data blocks
+     * that actually comprise the file data.  This level of indirection is used
+     * so that the same data block can be used in multiple files, or multiple
+     * versions of the same file. */
+    SHA1Checksum hash;
     while (true) {
-        ssize_t res = read(fd, buf, sizeof(buf));
-        if (res < 0) {
-            if (errno == EINTR)
-                continue;
-            printf("Error while reading: %m\n");
-            return;
-        } else if (res == 0) {
+        size_t bytes = file_read(fd, block_buf, LBS_BLOCK_SIZE);
+        if (bytes == 0)
             break;
-        } else {
-            size += res;
-        }
+
+        hash.process(block_buf, bytes);
+
+        // tarstore processing
+        LbsObject *o = new LbsObject;
+        o->set_group("data");
+        o->set_data(block_buf, bytes);
+        o->write(tss);
+        segment_list.push_back(o->get_name());
+        delete o;
+
+        size += bytes;
     }
 
-    printf("    bytes=%Ld\n", size);
+    file_info["checksum"] = hash.checksum_str();
+
+    /* For files that only need to be broken apart into a few objects, store
+     * the list of objects directly.  For larger files, store the data
+     * out-of-line and provide a pointer to the indrect object. */
+    if (segment_list.size() < 8) {
+        string blocklist = "";
+        for (list<string>::iterator i = segment_list.begin();
+             i != segment_list.end(); ++i) {
+            if (i != segment_list.begin())
+                blocklist += " ";
+            blocklist += *i;
+        }
+        file_info["data"] = blocklist;
+    } else {
+        string blocklist = "";
+        for (list<string>::iterator i = segment_list.begin();
+             i != segment_list.end(); ++i) {
+            blocklist += *i + "\n";
+        }
+
+        LbsObject *i = new LbsObject;
+        i->set_group("indirect");
+        i->set_data(blocklist.data(), blocklist.size());
+        i->write(tss);
+        file_info["data"] = "@" + i->get_name();
+        delete i;
+    }
 }
 
-void scanfile(const string& path)
+void scanfile(const string& path, ostream &metadata)
 {
     int fd;
     long flags;
     struct stat stat_buf;
     char *buf;
     ssize_t len;
+    list<string> refs;
+
+    // Set to true if the item is a directory and we should recursively scan
+    bool recurse = false;
 
     dictionary file_info;
 
@@ -73,12 +150,14 @@ void scanfile(const string& path)
 
     printf("%s\n", path.c_str());
 
-    file_info["mode"] = encode_u16(stat_buf.st_mode & 07777);
-    file_info["atime"] = encode_u64(encode_time(stat_buf.st_atime));
-    file_info["ctime"] = encode_u64(encode_time(stat_buf.st_ctime));
-    file_info["mtime"] = encode_u64(encode_time(stat_buf.st_mtime));
-    file_info["user"] = encode_u32(stat_buf.st_uid);
-    file_info["group"] = encode_u32(stat_buf.st_gid);
+    metadata << "name: " << uri_encode(path) << "\n";
+
+    file_info["mode"] = encode_int(stat_buf.st_mode & 07777);
+    file_info["atime"] = encode_int(stat_buf.st_atime);
+    file_info["ctime"] = encode_int(stat_buf.st_ctime);
+    file_info["mtime"] = encode_int(stat_buf.st_mtime);
+    file_info["user"] = encode_int(stat_buf.st_uid);
+    file_info["group"] = encode_int(stat_buf.st_gid);
 
     char inode_type;
 
@@ -112,7 +191,7 @@ void scanfile(const string& path)
             printf("error reading symlink: name truncated\n");
         }
 
-        file_info["contents"] = buf;
+        file_info["contents"] = uri_encode(buf);
 
         delete[] buf;
         break;
@@ -137,14 +216,14 @@ void scanfile(const string& path)
         flags = fcntl(fd, F_GETFL);
         fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
 
-        //dumpfile(fd);
-        file_info["size"] = encode_u64(stat_buf.st_size);
+        file_info["size"] = encode_int(stat_buf.st_size);
+        dumpfile(fd, file_info, metadata);
         close(fd);
 
         break;
     case S_IFDIR:
         inode_type = 'd';
-        scandir(path);
+        recurse = true;
         break;
 
     default:
@@ -153,12 +232,18 @@ void scanfile(const string& path)
     }
 
     file_info["type"] = string(1, inode_type);
+    metadata << "type: " << inode_type << "\n";
+
+    dict_output(metadata, file_info);
+    metadata << "\n";
 
-    info_dump->write_string(path);
-    info_dump->write_dictionary(file_info);
+    // If we hit a directory, now that we've written the directory itself,
+    // recursively scan the directory.
+    if (recurse)
+        scandir(path, metadata);
 }
 
-void scandir(const string& path)
+void scandir(const string& path, ostream &metadata)
 {
     DIR *dir = opendir(path.c_str());
 
@@ -181,7 +266,7 @@ void scandir(const string& path)
     for (vector<string>::iterator i = contents.begin();
          i != contents.end(); ++i) {
         const string& filename = *i;
-        scanfile(path + "/" + filename);
+        scanfile(path + "/" + filename, metadata);
     }
 
     closedir(dir);
@@ -189,20 +274,32 @@ void scandir(const string& path)
 
 int main(int argc, char *argv[])
 {
-    FILE *dump = fopen("fileinfo", "w");
-    if (dump == NULL) {
-        fprintf(stderr, "Cannot open fileinfo: %m\n");
-        return 1;
+    block_buf = new char[LBS_BLOCK_SIZE];
+
+    if (argc > 1) {
+        tss = new TarSegmentStore(argv[1]);
+    } else {
+        tss = new TarSegmentStore(".");
     }
 
-    FileOutputStream os(dump);
-    info_dump = &os;
+    std::ostringstream metadata;
 
     try {
-        scanfile(".");
+        scanfile(".", metadata);
     } catch (IOException e) {
         fprintf(stderr, "IOException: %s\n", e.getError().c_str());
     }
 
+    const string md = metadata.str();
+
+    LbsObject *r = new LbsObject;
+    r->set_group("root");
+    r->set_data(md.data(), md.size());
+    r->write(tss);
+    delete r;
+
+    tss->sync();
+    delete tss;
+
     return 0;
 }