X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=scandir.cc;h=c8a41800f82254479a430c62b8cde0cdcb0d48f3;hb=5c10647d667fc4408b73402db9247181a54a57ad;hp=d0f76d1a0eeba81ef286a4fb5669ae1234b86366;hpb=38c66f088ed65d2f42264c92add6e0b33eac2bfc;p=cumulus.git

diff --git a/scandir.cc b/scandir.cc
index d0f76d1..c8a4180 100644
--- a/scandir.cc
+++ b/scandir.cc
@@ -12,51 +12,155 @@
 
 #include <algorithm>
 #include <string>
+#include <list>
 #include <vector>
+#include <iostream>
+#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 int LBS_BLOCK_SIZE = 1024 * 1024;
+static char *block_buf;
+
+/* 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;
 }
 
-void dumpfile(int fd)
+/* 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");
         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);
+        object_list.push_back(o->get_name());
+        segment_list.insert(o->get_ref().get_segment());
+        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 (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)
@@ -66,45 +170,42 @@ 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;
 
     dictionary file_info;
 
     lstat(path.c_str(), &stat_buf);
 
-    printf("%s:\n", path.c_str());
-    printf("  ino=%Ld, perm=%04o, uid=%d, gid=%d, nlink=%d, blksize=%d, size=%Ld\n",
-           (int64_t)stat_buf.st_ino, stat_buf.st_mode & 07777,
-           stat_buf.st_uid, stat_buf.st_gid, stat_buf.st_nlink,
-           (int)stat_buf.st_blksize, (int64_t)stat_buf.st_size);
+    printf("%s\n", path.c_str());
+
+    metadata << "name: " << uri_encode(path) << "\n";
 
-    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);
+    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;
 
     switch (stat_buf.st_mode & S_IFMT) {
     case S_IFIFO:
-        printf("    FIFO\n");
         inode_type = 'p';
         break;
     case S_IFSOCK:
-        printf("    socket\n");
         inode_type = 's';
         break;
     case S_IFCHR:
-        printf("    character device\n");
         inode_type = 'c';
         break;
     case S_IFBLK:
-        printf("    block device\n");
         inode_type = 'b';
         break;
     case S_IFLNK:
-        printf("    symlink\n");
         inode_type = 'l';
 
         /* Use the reported file size to allocate a buffer large enough to read
@@ -121,12 +222,11 @@ 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;
     case S_IFREG:
-        printf("    regular file\n");
         inode_type = '-';
 
         /* Be paranoid when opening the file.  We have no guarantee that the
@@ -147,15 +247,14 @@ 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);
         close(fd);
 
         break;
     case S_IFDIR:
-        printf("    directory\n");
         inode_type = 'd';
-        scandir(path);
+        recurse = true;
         break;
 
     default:
@@ -164,9 +263,19 @@ void scanfile(const string& path)
     }
 
     file_info["type"] = string(1, inode_type);
+    metadata << "type: " << inode_type << "\n";
+
+    dict_output(metadata, file_info);
+    metadata << "\n";
+
+    // Break apart metadata listing if it becomes too large.
+    if (metadata.str().size() > 4096)
+        metadata_flush();
 
-    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);
 }
 
 void scandir(const string& path)
@@ -200,20 +309,42 @@ 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];
 
-    FileOutputStream os(dump);
-    info_dump = &os;
+    if (argc > 1) {
+        tss = new TarSegmentStore(argv[1]);
+    } else {
+        tss = new TarSegmentStore(".");
+    }
 
     try {
-        scandir(".");
+        scanfile(".");
     } catch (IOException e) {
         fprintf(stderr, "IOException: %s\n", e.getError().c_str());
     }
 
+    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());
+    string r = root->get_ref().to_string();
+    printf("root: %s\n\n", r.c_str());
+    delete root;
+
+    printf("segments:\n");
+    for (std::set<string>::iterator i = segment_list.begin();
+         i != segment_list.end(); ++i) {
+        printf("    %s\n", i->c_str());
+    }
+
+    tss->sync();
+    delete tss;
+
     return 0;
 }