X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=30b8d4572aa6eab67a9796018f8ba8c59310e6a4;hb=6ac0cb32a08c81ddb87cd3a1373e8e64e8757534;hp=d0ed67d5167bc59e620d8442acc101afa64f11a5;hpb=4a6945983fa171fa843c6f8955ba601f733f3ca5;p=cumulus.git

diff --git a/scandir.cc b/scandir.cc
index d0ed67d..30b8d45 100644
--- a/scandir.cc
+++ b/scandir.cc
@@ -12,55 +12,158 @@
 
 #include <algorithm>
 #include <string>
+#include <list>
 #include <vector>
+#include <iostream>
+#include <fstream>
+#include <sstream>
+#include <set>
 
+#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;
+
+/* Buffer for holding a single block of data read from a file. */
+static const size_t LBS_BLOCK_SIZE = 1024 * 1024;
+static char *block_buf;
+
+static const size_t LBS_METADATA_BLOCK_SIZE = 65536;
+
+/* Contents of the root object.  This will contain a set of indirect links to
+ * the metadata objects. */
+std::ostringstream metadata_root;
+
+/* Buffer for building up metadata. */
+std::ostringstream metadata;
+
+/* Keep track of all segments which are needed to reconstruct the snapshot. */
+std::set<string> segment_list;
 
 void scandir(const string& path);
 
-/* Converts time to microseconds since the epoch. */
-int64_t encode_time(time_t time)
+/* Ensure contents of metadata are flushed to an object. */
+void metadata_flush()
 {
-    return (int64_t)time * 1000000;
+    string m = metadata.str();
+    if (m.size() == 0)
+        return;
+
+    /* Write current metadata information to a new object. */
+    LbsObject *meta = new LbsObject;
+    meta->set_group("root");
+    meta->set_data(m.data(), m.size());
+    meta->write(tss);
+    meta->checksum();
+
+    /* Write a reference to this block in the root. */
+    ObjectReference ref = meta->get_ref();
+    metadata_root << "@" << ref.to_string() << "\n";
+    segment_list.insert(ref.get_segment());
+
+    delete meta;
+
+    metadata.str("");
 }
 
+/* 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)
 {
     struct stat stat_buf;
     fstat(fd, &stat_buf);
     int64_t size = 0;
-
-    char buf[4096];
+    list<string> object_list;
 
     if ((stat_buf.st_mode & S_IFMT) != S_IFREG) {
-        printf("file is no longer a regular file!\n");
+        fprintf(stderr, "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 {
-            hash.process(buf, res);
-            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);
+        object_list.push_back(o->get_name());
+        segment_list.insert(o->get_ref().get_segment());
+        delete o;
+
+        size += bytes;
     }
 
-    file_info["sha1"] = string((const char *)hash.checksum(),
-                               hash.checksum_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 (object_list.size() < 8) {
+        string blocklist = "";
+        for (list<string>::iterator i = object_list.begin();
+             i != object_list.end(); ++i) {
+            if (i != object_list.begin())
+                blocklist += " ";
+            blocklist += *i;
+        }
+        file_info["data"] = blocklist;
+    } else {
+        string blocklist = "";
+        for (list<string>::iterator i = object_list.begin();
+             i != object_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();
+        segment_list.insert(i->get_ref().get_segment());
+        delete i;
+    }
 }
 
 void scanfile(const string& path)
@@ -70,6 +173,7 @@ void scanfile(const string& path)
     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;
@@ -80,12 +184,14 @@ 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;
 
@@ -111,16 +217,14 @@ void scanfile(const string& path)
         buf = new char[stat_buf.st_size + 2];
         len = readlink(path.c_str(), buf, stat_buf.st_size + 1);
         if (len < 0) {
-            printf("error reading symlink: %m\n");
+            fprintf(stderr, "error reading symlink: %m\n");
         } else if (len <= stat_buf.st_size) {
             buf[len] = '\0';
-            printf("    contents=%s\n", buf);
+            file_info["contents"] = uri_encode(buf);
         } else if (len > stat_buf.st_size) {
-            printf("error reading symlink: name truncated\n");
+            fprintf(stderr, "error reading symlink: name truncated\n");
         }
 
-        file_info["contents"] = buf;
-
         delete[] buf;
         break;
     case S_IFREG:
@@ -144,7 +248,7 @@ void scanfile(const string& path)
         flags = fcntl(fd, F_GETFL);
         fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
 
-        file_info["size"] = encode_u64(stat_buf.st_size);
+        file_info["size"] = encode_int(stat_buf.st_size);
         dumpfile(fd, file_info);
         close(fd);
 
@@ -161,8 +265,12 @@ void scanfile(const string& path)
 
     file_info["type"] = string(1, inode_type);
 
-    info_dump->write_string(path);
-    info_dump->write_dictionary(file_info);
+    dict_output(metadata, file_info);
+    metadata << "\n";
+
+    // Break apart metadata listing if it becomes too large.
+    if (metadata.str().size() > LBS_METADATA_BLOCK_SIZE)
+        metadata_flush();
 
     // If we hit a directory, now that we've written the directory itself,
     // recursively scan the directory.
@@ -175,7 +283,7 @@ void scandir(const string& path)
     DIR *dir = opendir(path.c_str());
 
     if (dir == NULL) {
-        printf("Error: %m\n");
+        fprintf(stderr, "Error: %m\n");
         return;
     }
 
@@ -201,12 +309,17 @@ void scandir(const string& path)
 
 int main(int argc, char *argv[])
 {
-    SegmentStore ss(".");
-    SegmentWriter *sw = ss.new_segment();
-    info_dump = sw->new_object();
+    block_buf = new char[LBS_BLOCK_SIZE];
+
+    string backup_dest = ".";
+
+    if (argc > 1)
+        backup_dest = argv[1];
+
+    tss = new TarSegmentStore(backup_dest);
 
-    string uuid = SegmentWriter::format_uuid(sw->get_uuid());
-    printf("Backup UUID: %s\n", uuid.c_str());
+    string desc_filename = backup_dest + "/snapshot.lbs";
+    std::ofstream descriptor(desc_filename.c_str());
 
     try {
         scanfile(".");
@@ -214,7 +327,28 @@ int main(int argc, char *argv[])
         fprintf(stderr, "IOException: %s\n", e.getError().c_str());
     }
 
-    delete sw;
+    metadata_flush();
+    const string md = metadata_root.str();
+
+    LbsObject *root = new LbsObject;
+    root->set_group("root");
+    root->set_data(md.data(), md.size());
+    root->write(tss);
+    root->checksum();
+
+    segment_list.insert(root->get_ref().get_segment());
+    descriptor << "root: " << root->get_ref().to_string() << "\n\n";
+
+    delete root;
+
+    descriptor << "segments:\n";
+    for (std::set<string>::iterator i = segment_list.begin();
+         i != segment_list.end(); ++i) {
+        descriptor << "    " << *i << "\n";
+    }
+
+    tss->sync();
+    delete tss;
 
     return 0;
 }