From: Radoslaw Zarzynski <rzarzyns@redhat.com>
Date: Wed, 13 Sep 2023 19:45:47 +0000 (+0200)
Subject: Merge pull request #53178 from aclamk/wip-aclamk-bs-esb-5-enabled
X-Git-Tag: v19.0.0~486
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=f417f38beef37f8a069b82ca082a543f5099f533;p=ceph.git

Merge pull request #53178 from aclamk/wip-aclamk-bs-esb-5-enabled

bluestore: Elastic Shared Blobs 5 - enabled, final

Reviewed-by: Igor Fedotov <ifedotov@suse.com>
Reviewed-by: Radoslaw Zarzynski <rzarzyns@redhat.com
---

f417f38beef37f8a069b82ca082a543f5099f533
diff --cc src/os/bluestore/BlueStore.cc
index 10a9aac59408e,a7ee1a280bb7c..7456f16c6df80
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@@ -3633,55 -4512,95 +4505,143 @@@ BlueStore::BlobRef BlueStore::ExtentMap
    return rb;
  }
  
+ BlueStore::ExtentMap::debug_au_vector_t
+ BlueStore::ExtentMap::debug_list_disk_layout()
+ {
+   BlueStore::ExtentMap::debug_au_vector_t res;
+   uint32_t l_pos = 0;
+   for (auto ep = extent_map.begin(); ep != extent_map.end(); ++ep) {
+     if (l_pos < ep->logical_offset) {
+       // a hole in logical mapping, mark it
+       res.emplace_back(-1ULL, ep->logical_offset - l_pos, 0, 0);
+     }
+     l_pos = ep->logical_offset + ep->length;
+     const bluestore_blob_t& bblob = ep->blob->get_blob();
+     uint32_t chunk_size = bblob.get_chunk_size(onode->c->store->block_size);
+     uint32_t length_left = ep->length;
+ 
+     bluestore_extent_ref_map_t* ref_map = nullptr;
+     if (bblob.is_shared()) {
+       ceph_assert(ep->blob->shared_blob->is_loaded());
+       bluestore_shared_blob_t* bsblob = ep->blob->shared_blob->persistent;
+       ref_map = &bsblob->ref_map;
+     }
+ 
+     unsigned csum_i = 0;
+     size_t csum_cnt;
+     uint32_t length;
+     if (bblob.has_csum()) {
+       csum_cnt = bblob.get_csum_count();
+       uint32_t csum_chunk_size = bblob.get_csum_chunk_size();
+       uint64_t csum_offset_align = p2align(ep->blob_offset, csum_chunk_size);
+       csum_i = csum_offset_align / csum_chunk_size;
+       // size of first chunk
+       length = p2align(ep->blob_offset + csum_chunk_size, csum_chunk_size) - ep->blob_offset;
+       length = std::min<uint32_t>(length_left, length);
+       if (csum_chunk_size < chunk_size) {
+ 	chunk_size = csum_chunk_size;
+       }
+     } else {
+       length = p2align(ep->blob_offset + chunk_size, chunk_size) - ep->blob_offset;
+       length = std::min<uint32_t>(length_left, length);
+     }
+ 
+     uint32_t bo = ep->blob_offset;
+     while (length_left > 0) {
+       uint64_t csum_val = 0;
+       if (bblob.has_csum()) {
+ 	ceph_assert(csum_cnt > csum_i);
+ 	csum_val = bblob.get_csum_item(csum_i);
+ 	++csum_i;
+       }
+       //extract AU from extents
+       uint64_t disk_extent_left; // length till the end of disk extent
+       uint64_t disk_offset = bblob.calc_offset(bo, &disk_extent_left);
+       bluestore_extent_ref_map_t::debug_len_cnt l_c = {0, std::numeric_limits<uint32_t>::max()};
+       if (bblob.is_shared()) {
+ 	l_c = ref_map->debug_peek(disk_offset);
+ 	if (l_c.len < length) {
+ 	  length = l_c.len;
+ 	}
+       }
+       res.emplace_back(disk_offset, length, csum_val, l_c.cnt);
+       bo += length;
+       length_left -= length;
+       length = chunk_size;
+     };
+   }
+   return res;
+ }
+ 
+ std::ostream& operator<<(std::ostream& out, const BlueStore::ExtentMap::debug_au_vector_t& auv)
+ {
+   out << "[";
+   for (size_t i = 0; i < auv.size(); ++i) {
+     if (i != 0) {
+       out << " ";
+     }
+     out << "0x" << std::hex;
+     if (auv[i].disk_offset != -1ULL) {
+       out << auv[i].disk_offset << "~" << auv[i].disk_length
+ 	  << "(" << std::dec << int32_t(auv[i].ref_cnts)
+ 	  << "):" << std::hex << auv[i].chksum;
+     } else {
+       out << "~" << auv[i].disk_length << std::dec;
+     }
+   }
+   out << "]" << std::dec;
+   return out;
+ }
+ 
  // Onode
 +//
 +// Mapping blobs over Onode's logical offsets.
 +//
 +// Blob is always continous. Blobs may overlap.
 +// Non-mapped regions are "0" when read.
 +//                 1               2               3
 +// 0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
 +// <blob.a.blob.a><blob.b.blo>        <blob.c.blob.c.blob.c.blob>
 +//       <blob.d.blob.d.b>                      <blob.e.blob.e>
 +// blob.a starts at 0x0 length 0xe
 +// blob.b starts at 0xf length 0xb
 +// blob.c starts at 0x23 length 0x1b
 +// blob.d starts at 0x06 length 0x12
 +// blob.e starts at 0x2d length 0xf
 +//
 +// Blobs can have non-encoded parts:
 +//                 1               2               3
 +// 0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
 +// aaaaaa......aaabbbbb...bbbb        ccccccccccccccc..........cc
 +//       dddddd........ddd                      .....eeeeeeeeee
 +// "." - non-encoded parts of blob (holes)
 +//
 +// Mapping logical to blob:
 +// extent_map maps {Onode's logical offset, length}=>{Blob, in-blob offset}
 +// {0x0, 0x6}=>{blob.a, 0x0}
 +// {0x6, 0x6}=>{blob.d, 0x0}
 +// {0xc, 0x3}=>{blob.a, 0xc}
 +// {0xf, 0x5}=>{blob.b, 0x0}
 +// {0x14, 0x3}=>{blob.d, 0xe}
 +// {0x17, 0x4}=>{blob.b, 0x8}
 +// a hole here
 +// {0x23, 0xe}=>{blob.c, 0x0}
 +// and so on...
 +//
 +// Compressed blobs do not have non-encoded parts.
 +// Same example as above but all blobs are compressed:
 +//                 1               2               3
 +// 0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
 +// aaaaaaAAAAAAaaabbbbbBBBbbbb        cccccccccccccccCCCCCCCCCCcc
 +//       ddddddDDDDDDDDddd                      EEEEEeeeeeeeeee
 +// A-E: parts of blobs that are never used.
 +// This can happen when a compressed blob is overwritten partially.
 +// The target ranges are no longer used, but are left there because they are necessary
 +// for successful decompression.
 +//
 +// In compressed blobs PExtentVector and csum refer to actually occupied disk space.
 +// Blob's logical length is larger then occupied disk space.
 +// Mapping from extent_map always uses offsets of decompressed data.
  
  #undef dout_prefix
  #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " "