pg /c database 通过。似乎是pg使用 open/write 写入了某个文件，通过 fopen/fscanf 绕过了hook路径导致的。出现了新的段错误。

scripts/do_trace.md

@@ -0,0 +1,47 @@
+```shell
+pm=$(pgrep -u postgres -xo postgres)
+sudo strace -ff -f -yy -s 256 \
+    -e trace=%file,read,pread64,readv \
+    -p "$pm" -o /tmp/pg.sys.trace
+
+
+strace: Process 233035 attached
+strace: Process 235357 attached
+^Cstrace: Process 233035 detached
+
+
+fatal_pid=$(grep 'FATAL:  "pg_tblspc' /tmp/pg.log | tail -1 | sed -n 's/.*\[\([0-9]\+\)\].*/\1/p'); echo "fatal_pid=$fatal_pid"
+
+fatal_pid=235357
+
+
+grep -n "PG_VERSION\|pg_tblspc/16384/PG_12_201909212/16385" /tmp/pg.sys.trace.$fatal_pid
+
+18:access("pg_tblspc/16384/PG_12_201909212/16385", F_OK) = 0
+19:openat(AT_FDCWD, "pg_tblspc/16384/PG_12_201909212/16385/PG_VERSION", O_RDONLY) = 4</zvfs/pg_ts_bench/PG_12_201909212/16385/PG_VERSION>
+20:read(4</zvfs/pg_ts_bench/PG_12_201909212/16385/PG_VERSION>, "\0\0\0", 4096) = 3
+21:read(4</zvfs/pg_ts_bench/PG_12_201909212/16385/PG_VERSION>, "", 4096) = 0
+
+
+
+fatal_pid=$(grep 'FATAL:  "pg_tblspc' /tmp/pg.log | tail -1 | sed -n 's/.*\[\([0-9]\+\)\].*/\1/p'); echo "fatal_pid=$fatal_pid"
+
+grep -n "\[zvfs\]\[rw-dbg\].*pid=${fatal_pid}" /tmp/pg.log | tail -n 120
+
+6461:[zvfs][rw-dbg][pid=235357][tid=140193133607488][fd=4] read passthrough reason=fd_not_zvfs count=512
+6462:[zvfs][rw-dbg][pid=235357][tid=140193133607488][fd=4] read passthrough ret=512 errno=0(OK)
+```
+
+
+```shell
+sudo strace -ff -f -k -yy -s 256 -e trace=openat,read,pread64 -p "$pm" -o /tmp/pg.ktrace
+```
+
+调用链：
+
+libc(__open64) -> _IO_file_open -> fopen -> AllocateFile -> ValidatePgVersion
+
+是fopen，内部调用 libc_(__open64)
+
+
+