Re: [PATCH v3] Implement "info proc mappings" for NetBSD

[Kamil Rytarowski <n54@gmx.com>]

On 11.04.2020 23:05, Simon Marchi wrote:
> On 2020-04-06 5:37 a.m., Kamil Rytarowski wrote:
>> @@ -39,3 +42,148 @@ nbsd_nat_target::pid_to_exec_file (int pid)
>>      return NULL;
>>    return buf;
>>  }
>> +
>> +/* Retrieve all the memory regions in the specified process.  */
>> +
>> +static gdb::unique_xmalloc_ptr<struct kinfo_vmentry[]>
>> +nbsd_kinfo_get_vmmap (pid_t pid, size_t *size)
>> +{
>> +  int mib[5] = {CTL_VM, VM_PROC, VM_PROC_MAP, pid,
>> +		sizeof (struct kinfo_vmentry)};
>> +
>> +  size_t length = 0;
>> +  if (sysctl (mib, ARRAY_SIZE (mib), NULL, &length, NULL, 0))
>> +    {
>> +      *size = 0;
>> +      return NULL;
>> +    }
>> +
>> +  /* Prereserve more space. */
>> +  length = length * 5 / 3;
>
> Why is this needed?
>

This length is volatile and can be larger on the 2nd sysctl(3) call. We
can call this function against a running process that can enlarge its
usage of address space in this time window.

>> +
>> +  gdb::unique_xmalloc_ptr<struct kinfo_vmentry[]> kiv
>> +    ((struct kinfo_vmentry *) xcalloc (length, 1));
>
> Let's use XCNEWVAR here.  It doesn't matter in this case, but for consistency.  It does
> check that we are not trying to allocate space for a non-POD type.  You can use like:
>
>   XCNEWVAR (kinfo_vmentry, length)
>
> You can also use XNEWVAR if you don't care about the buffer being initialized to zeros.
>

OK.

>> diff --git a/gdb/nbsd-tdep.h b/gdb/nbsd-tdep.h
>> index c99a8b537b6..81bdb2510f5 100644
>> --- a/gdb/nbsd-tdep.h
>> +++ b/gdb/nbsd-tdep.h
>> @@ -25,4 +25,22 @@ struct link_map_offsets *nbsd_lp64_solib_svr4_fetch_link_map_offsets (void);
>>
>>  int nbsd_pc_in_sigtramp (CORE_ADDR, const char *);
>>
>> +/* Output the header for "info proc mappings".  ADDR_BIT is the size
>> +   of a virtual address in bits.  */
>> +
>> +extern void nbsd_info_proc_mappings_header (int addr_bit);
>> +
>> +/* Output description of a single memory range for "info proc
>> +   mappings".  ADDR_BIT is the size of a virtual address in bits.  The
>> +   KVE_START, KVE_END, KVE_OFFSET, KVE_FLAGS, and KVE_PROTECTION
>> +   parameters should contain the value of the corresponding fields in
>> +   a 'struct kinfo_vmentry'.  The KVE_PATH parameter should contain a
>> +   pointer to the 'kve_path' field in a 'struct kinfo_vmentry'. */
>> +
>> +extern void nbsd_info_proc_mappings_entry (int addr_bit, ULONGEST kve_start,
>> +					   ULONGEST kve_end,
>> +					   ULONGEST kve_offset,
>> +					   int kve_flags, int kve_protection,
>> +					   const void *kve_path);
>
> I'd make the last parameter a `const char *` directly, since this is what the function
> expects to receive.  And remove the reinterpret_casts.
>

OK.

> Simon
>