commit 7632f0c2860700b522babb7a6bc0b2ac51390c99
author David Anderson <dvander@google.com> Tue Oct 30 14:58:44 2018 -0700
committer David Anderson <dvander@google.com> Mon Nov 05 19:57:47 2018 +0000
tree c73bc0eb0a1e14f747ed4cfc6584442b85e4458c
parent e0c3a8d97b0a0493fdc3ebd0a648a2cdcd7986ab

liblp: Fix flashing metadata to multiple block devices.

Although metadata is only stored on the first block device, we ideally
want to validate all the block devices before flashing. This patch does
that and in the process converts ValidateAndSerializeMetadata to use
IPartitionOpener.

Bug: 116802789
Test: manual test
      liblp_test gtest

Change-Id: I3f24cfc6f5607309dc3cded0423326c5ba293d26

fs_mgr/liblp/writer.cpp

diff --git a/fs_mgr/liblp/writer.cpp b/fs_mgr/liblp/writer.cpp
index c740bd4..f4c9b99 100644
--- a/fs_mgr/liblp/writer.cpp
+++ b/fs_mgr/liblp/writer.cpp
@@ -82,12 +82,8 @@
 
 // Perform sanity checks so we don't accidentally overwrite valid metadata
 // with potentially invalid metadata, or random partition data with metadata.
-static bool ValidateAndSerializeMetadata(int fd, const LpMetadata& metadata, std::string* blob) {
-    uint64_t blockdevice_size;
-    if (!GetDescriptorSize(fd, &blockdevice_size)) {
-        return false;
-    }
-
+static bool ValidateAndSerializeMetadata(const IPartitionOpener& opener, const LpMetadata& metadata,
+                                         std::string* blob) {
     const LpMetadataHeader& header = metadata.header;
     const LpMetadataGeometry& geometry = metadata.geometry;
 
@@ -104,7 +100,7 @@
     // metadata.
     uint64_t reserved_size = LP_METADATA_GEOMETRY_SIZE +
                              uint64_t(geometry.metadata_max_size) * geometry.metadata_slot_count;
-    uint64_t total_reserved = reserved_size * 2;
+    uint64_t total_reserved = LP_PARTITION_RESERVED_BYTES + reserved_size * 2;
 
     const LpMetadataBlockDevice* super_device = GetMetadataSuperBlockDevice(metadata);
     if (!super_device) {
@@ -112,15 +108,27 @@
         return false;
     }
 
-    if (total_reserved > blockdevice_size ||
-        total_reserved > super_device->first_logical_sector * LP_SECTOR_SIZE) {
+    if (total_reserved > super_device->first_logical_sector * LP_SECTOR_SIZE) {
         LERROR << "Not enough space to store all logical partition metadata slots.";
         return false;
     }
-    if (blockdevice_size != super_device->size) {
-        LERROR << "Block device size " << blockdevice_size
-               << " does not match metadata requested size " << super_device->size;
-        return false;
+    for (const auto& block_device : metadata.block_devices) {
+        std::string partition_name = GetBlockDevicePartitionName(block_device);
+        if ((block_device.first_logical_sector + 1) * LP_SECTOR_SIZE > block_device.size) {
+            LERROR << "Block device " << partition_name << " has invalid first sector "
+                   << block_device.first_logical_sector << " for size " << block_device.size;
+            return false;
+        }
+        BlockDeviceInfo info;
+        if (!opener.GetInfo(partition_name, &info)) {
+            PERROR << partition_name << ": ioctl";
+            return false;
+        }
+        if (info.size != block_device.size) {
+            LERROR << "Block device " << partition_name << " size mismatch (expected"
+                   << block_device.size << ", got " << info.size << ")";
+            return false;
+        }
     }
 
     // Make sure all partition entries reference valid extents.
@@ -230,7 +238,7 @@
     // basic checks that the geometry and tables are coherent, and will fit
     // on the given block device.
     std::string metadata_blob;
-    if (!ValidateAndSerializeMetadata(fd, metadata, &metadata_blob)) {
+    if (!ValidateAndSerializeMetadata(opener, metadata, &metadata_blob)) {
         return false;
     }
 
@@ -295,7 +303,7 @@
     // basic checks that the geometry and tables are coherent, and will fit
     // on the given block device.
     std::string blob;
-    if (!ValidateAndSerializeMetadata(fd, metadata, &blob)) {
+    if (!ValidateAndSerializeMetadata(opener, metadata, &blob)) {
         return false;
     }
 
@@ -327,7 +335,7 @@
         // synchronize the backup copy. This guarantees that a partial write
         // still leaves one copy intact.
         std::string old_blob;
-        if (!ValidateAndSerializeMetadata(fd, *primary.get(), &old_blob)) {
+        if (!ValidateAndSerializeMetadata(opener, *primary.get(), &old_blob)) {
             LERROR << "Error serializing primary metadata to repair corrupted backup";
             return false;
         }
@@ -339,7 +347,7 @@
         // The backup copy is coherent, and the primary is not. Sync it for
         // safety.
         std::string old_blob;
-        if (!ValidateAndSerializeMetadata(fd, *backup.get(), &old_blob)) {
+        if (!ValidateAndSerializeMetadata(opener, *backup.get(), &old_blob)) {
             LERROR << "Error serializing primary metadata to repair corrupted backup";
             return false;
         }