What is the correct way to get a consistent snapshot of /proc/pid/smaps? - linux

Question

I am trying to parse the PSS value from /proc/<pid>/smaps of a process in my C++ binary.
According to this SO answer, naively reading the /proc/<pid>/smaps file for example with ifstream::getLine() will result in an inconsistent dataset. The solution suggested is to use the read() system call to read the whole data in one go, something like:
#include <unistd.h>
#include <fcntl.h>
...
char rawData[102400];
int file = open("/proc/12345/smaps", O_RDONLY, 0);
auto bytesRead = read(file, rawData, 102400); // this returns 3722 instead of expected ~64k
close(file);
std::cout << bytesRead << std::endl;
// do some parsing here after null-terminating the buffer
...
My problem now is that despite me using a 100kB buffer, only 3722 bytes are returned. Looking at what cat does when parsing the file using strace, I see that it is using multiple calls to read() (also getting around 3k bytes on every read) until read() returns 0 - as described in the documentation of read():
...
read(3, "7fa8db3d7000-7fa8db3d8000 r--p 0"..., 131072) = 3588
write(1, "7fa8db3d7000-7fa8db3d8000 r--p 0"..., 3588) = 3588
read(3, "7fa8db3df000-7fa8db3e0000 r--p 0"..., 131072) = 3632
write(1, "7fa8db3df000-7fa8db3e0000 r--p 0"..., 3632) = 3632
read(3, "7fa8db3e8000-7fa8db3ed000 r--s 0"..., 131072) = 3603
write(1, "7fa8db3e8000-7fa8db3ed000 r--s 0"..., 3603) = 3603
read(3, "7fa8db41d000-7fa8db425000 r--p 0"..., 131072) = 3445
write(1, "7fa8db41d000-7fa8db425000 r--p 0"..., 3445) = 3445
read(3, "7fff05467000-7fff05496000 rw-p 0"..., 131072) = 2725
write(1, "7fff05467000-7fff05496000 rw-p 0"..., 2725) = 2725
read(3, "", 131072) = 0
munmap(0x7f8d29ad4000, 139264) = 0
close(3) = 0
close(1) = 0
close(2) = 0
exit_group(0) = ?
+++ exited with 0 +++
But isn't this supposed to produce inconsistent data according to the SO answer linked above?
I have also found some information about proc here, that seem to support the previous SO answer:
To see a precise
snapshot of a moment, you can see /proc/<pid>/smaps file and scan page table.
Then later in the text it says:
Note: reading /proc/PID/maps or /proc/PID/smaps is inherently racy (consistent
output can be achieved only in the single read call).
This typically manifests when doing partial reads of these files while the
memory map is being modified.
Despite the races, we do provide the following
guarantees:
1) The mapped addresses never go backwards, which implies no two regions will ever overlap.
2) If there is something at a given vaddr during the entirety of the
life of the smaps/maps walk, there will be some output for it.
So it seems to me, I can only trust the data I'm getting if I get it in a single read() call.
Which only returns a small chunk of data despite the buffer being big enough.
Which in turn means there is actually no way to get a consistent snapshot of /proc/<pid>/smaps and the data returned by cat/using multiple read() calls may be garbage depending on the sun to moon light ratio?
Or does 2) actually mean I'm too hung up on the previous SO answer listed above?

Answer 1

You are being limited with the internal kernel buffer size in fs/seq_file.c, which is used to generate many /proc files.
Buffer is first set to be the size of a page, then is exponentially grown to fit at least one record, and then is crammed with as many entire records as will fit, but is not grown any more after being able needed to fit the first entry. When the internal buffer cannot fit any more entries, the read is ended.