New: random.{4,7} [man-pages 5.13] do not reflect changes to /dev/random semantics since kernel 5.6

Hi Alex,

I am still struggling with getting my mail client to send through a proxy. The attached patch is the same as the bugzilla one, except that the commit message has been expanded.

From 343b2ad76dc6d78daa529405f2ec7f9a3e2e65bf Mon Sep 17 00:00:00 2001
From: Mingye Wang <[email protected]>
Date: Thu, 25 May 2023 20:21:08 +0800
Subject: [PATCH] random.{4,7}, getrandom.2: Adapt to Linux 5.6 changes
In-Reply-To: <[email protected]/>
X-Unsent: 1

Linux kernel 5.6 no longer has a blocking random pool. This commit updates all relevant references to reflect this change.

* random.7: Remove references to the blocking pool in the table. Add a note about the blocking pool.
* random.7: Revise "choice of random source" to remove any recommen- dationof the blocking pool. Stop suggesting that the blocking pool is safer (it's not after initialization).
* random.7: Add table entry for GRND_INSECURE.
* random.7: Weaken performance recommendation following the Linux 4.8 CSPRNG speedup.

* getrandom.2: Add flag entry for GRND_INSECURE.
* getrandom.2: Add removal note to GRND_RANDOM.
* random.4: Split DESCRIPTION paragraph on /dev/random into two, one for the new behavior and the other for the old.
* random.4: Adjust size limits and /proc list for 5.6.
* random.4: Mention blocking resolution by high-precision timer entropy.

Changes not made:
* random.4: USAGE section largely unchanged. Old kernels will stick around for a while.

Closes https://bugzilla.kernel.org/show_bug.cgi?id=214885

Signed-off-by Mingye Wang <[email protected]>

---
 man2/getrandom.2 |  9 +++++
 man4/random.4    | 30 +++++++++-------
 man7/random.7    | 89 ++++++++++++++++++++++++++----------------------
 3 files changed, 75 insertions(+), 53 deletions(-)

man2/getrandom.2 +9 -0

index dbd23a0a6..fb2e948be 100644
@@ -64,6 +64,8 @@ argument is a bit mask that can contain zero or more of the following values
 ORed together:
 .TP
 .B GRND_RANDOM
+.\" commit 48446f198f9adcb499b30332488dfd5bc3f176f6
+Ignored since Linux 5.6.
 If this bit is set, then random bytes are drawn from the
 .I random
 source

@@ -105,6 +107,13 @@ does not block in these cases, but instead immediately returns \-1 with
 .I errno
 set to
 .BR EAGAIN .
+.TP GRND_INSECURE
+.\" commit 75551dbf112c992bc6c99a972990b3f272247e23
+Added in Linux 5.6.
+Request best-effort, non-cryptographic-quality random bytes.
+If this bit is set, then
+.BR getrandom ()
+will never block or fail due to a lack of entropy.
 .SH RETURN VALUE
 On success,
 .BR getrandom ()

man4/random.4 +0 -0

index edd047b77..6ce9992fe 100644
@@ -58,16 +58,20 @@ If this is of concern in your application, use
 .BR getrandom (2)
 or \fI/dev/random\fP instead.

man4/random.4 +3 -2

+only within the estimated number of bits of fresh noise in the entropy pool,
+blocking until additional environmental noise is gathered.
+This old \fI/dev/random\fP is suitable for applications that need
 high quality randomness, and can afford indeterminate delays.
 .PP
-When the entropy pool is empty, reads from \fI/dev/random\fP will block
-until additional environmental noise is gathered.
 Since Linux 5.6, the
 .B O_NONBLOCK

flag is ignored as

man4/random.4 +3 -2

 .PP
 Writing to \fI/dev/random\fP or \fI/dev/urandom\fP will update the
 entropy pool with the data written, but this will not result in a
 higher entropy count.
 This means that it will impact the contents
-read from both files, but it will not make reads from
+read from both files, but it will not make reads from a pre-5.6
 \fI/dev/random\fP faster.
 .SS Usage
 The

@@ -145,7 +150,7 @@ soon as it is reloaded in the boot sequence, and perfectly adequate for
 network encryption session keys.
 (All major Linux distributions have saved the seed file across reboots
 since 2000 at least.)
-Since reads from
+Since reads from a pre-5.6
 .I /dev/random
 may block, users will usually want to open it in nonblocking mode
 (or perform a read with timeout),

@@ -246,6 +251,7 @@ It contains the value 4096.
 .RE
 .TP
 .I read_wakeup_threshold
+Removed in Linux 5.6.
 This file
 contains the number of bits of entropy required for waking up processes
 that sleep waiting for entropy from

man7/random.7 +48 -41

index 69e6c2403..e80a1328e 100644
@@ -59,17 +59,16 @@ The kernel collects bits of entropy from the environment.
 When a sufficient number of random bits has been collected, the
 entropy pool is considered to be initialized.
 .SS Choice of random source
-Unless you are doing long-term key generation (and most likely not even
-then), you probably shouldn't be reading from the
+Unless your program may run at early-boot, before the entropy pool
+is initialized, there is no longer any palpable difference between
 .I /dev/random
-device or employing
-.BR getrandom (2)
-with the
-.B GRND_RANDOM
-flag.
-Instead, either read from the
+and
 .I /dev/urandom
-device or employ
+since Linux 5.6. (See the table below.)
+.PP.
+On older kernels, either read from the
+.I /dev/urandom
+device or (especially if you are concerned with early boot) employ
 .BR getrandom (2)
 without the
 .B GRND_RANDOM

@@ -77,18 +76,6 @@ flag.
 The cryptographic algorithms used for the
 .I urandom
 source are quite conservative, and so should be sufficient for all purposes.
-.PP
-The disadvantage of
-.B GRND_RANDOM
-and reads from
-.I /dev/random
-is that the operation can block for an indefinite period of time.
-Furthermore, dealing with the partially fulfilled
-requests that can occur when using
-.B GRND_RANDOM
-or when reading from
-.I /dev/random
-increases code complexity.
 .\"
 .SS Monte Carlo and other probabilistic sampling applications
 Using these interfaces to provide large quantities of data for

@@ -99,6 +86,13 @@ need cryptographically secure random numbers.
 Instead, use the interfaces described in this page to obtain
 a small amount of data to seed a user-space pseudorandom
 number generator for use by such applications.
+.PP
+.\" commit e192be9d9a30555aae2ca1dc3aad37cba484cd4a
+.\" commit 48446f198f9adcb499b30332488dfd5bc3f176f6
+Since Linux 4.8, the random interfaces are backed by a much faster
+NUMA-aware ChaCha20 CSPRNG.
+It is still discouraged to use the random interfaces for sampling,
+but if you do, the performance will be much better than before.
 .\"
 .SS Comparison between getrandom, /dev/urandom, and /dev/random
 The following table summarizes the behavior of the various

@@ -122,9 +116,9 @@ T}
 T{
 .I /dev/random
 T}	T{
-Blocking pool
+CSPRNG output
 T}	T{
-If entropy too low, blocks until there is enough entropy again
+Never blocks
 T}	T{
 Blocks until enough entropy gathered
 T}

@@ -149,17 +143,6 @@ Blocks until pool ready
 T}
 T{
 .BR getrandom ()
-.B GRND_RANDOM
-T}	T{
-Same as
-.I /dev/random
-T}	T{
-If entropy too low, blocks until there is enough entropy again
-T}	T{
-Blocks until pool ready
-T}
-T{
-.BR getrandom ()
 .B GRND_NONBLOCK
 T}	T{
 Same as

@@ -171,21 +154,45 @@ T}	T{
 T}
 T{
 .BR getrandom ()
-.B GRND_RANDOM
-+
-.B GRND_NONBLOCK
+.B GRND_INSECURE
 T}	T{
 Same as
-.I /dev/random
+.I /dev/urandom
 T}	T{
-.B EAGAIN
-if not enough entropy available
+Never blocks
 T}	T{
-.B EAGAIN
+Returns output from uninitialized CSPRNG (may be low entropy and unsuitable for cryptography)
 T}
 .TE
 .ad
 .\"
+.SS The old blocking pool
+The above table describes the behavior of the interfaces since
+Linux 5.6.  In older kernels, the
+.I /dev/random
+used a separate blocking pool, and
+.BR getrandom () 
+had a
+.B GRND_RANDOM
+flag for reading from the blocking pool.
+.\"
+.PP
+The older blocking pool was a vestige of a time when the CSPRNG
+was not trusted.
+It assumed that entropy can run out by reading the CSPRNG.
+This has never been the case.
+Instead, programs using 
+.B GRND_RANDOM
+and
+.I /dev/random
+had to deal with operations blocking indefinitely.
+Furthermore, dealing with the partially fulfilled
+requests that can occur when using
+.B GRND_RANDOM
+or when reading from
+.I /dev/random
+increases code complexity.
+.\"
 .SS Generating cryptographic keys
 The amount of seed material required to generate a cryptographic key
 equals the effective key size of the key.

Signature · 1 line

2.40.0

If some changes come from Linux 4.8, I prefer having two separate patches.

Just wondering: how is this description of /dev/urandom different from GRND_INSECURE?

Blank lines are wrong.

Hi Alex,

attached are two patches split and revised as requested.

Regards,
Mingye

Author: Mingye Wang <[email protected]>
Date:   Mon Jun 5 11:10:12 2023 +0800

random.{4,7}, getrandom.2: Adapt to Linux 5.6 changes In-Reply-To: <[email protected]/>

Linux kernel 5.6 no longer has a blocking random pool. This commit updates all relevant references to reflect this change.

* random.7: Remove references to the blocking pool in the table. Add a note about the blocking pool.
* random.7: Revise "choice of random source" to remove any recommen-

dationof the blocking pool. Stop suggesting that the blocking pool is safer (it's not after initialization).
* random.7: Add table entry for GRND_INSECURE.
* getrandom.2: Add flag entry for GRND_INSECURE.
* getrandom.2: Add removal note to GRND_RANDOM.
* random.4: Split DESCRIPTION paragraph on /dev/random into two, one for the new behavior and the other for the old.
* random.4: Adjust size limits and /proc list for 5.6.
* random.4: Mention blocking resolution by high-precision timer entropy.

Closes https://bugzilla.kernel.org/show_bug.cgi?id=214885

Signed-off-by Mingye Wang <[email protected]>

man2/getrandom.2 +9 -0

index dbd23a0a6..fb2e948be 100644
@@ -64,6 +64,8 @@ .SH DESCRIPTION
 ORed together:
 .TP
 .B GRND_RANDOM
+.\" commit 48446f198f9adcb499b30332488dfd5bc3f176f6
+Ignored since Linux 5.6.
 If this bit is set, then random bytes are drawn from the
 .I random
 source

@@ -105,6 +107,13 @@ .SH DESCRIPTION
 .I errno
 set to
 .BR EAGAIN .
+.TP GRND_INSECURE
+.\" commit 75551dbf112c992bc6c99a972990b3f272247e23
+Added in Linux 5.6.
+Request best-effort, non-cryptographic-quality random bytes.
+If this bit is set, then
+.BR getrandom ()
+will never block or fail due to a lack of entropy.
 .SH RETURN VALUE
 On success,
 .BR getrandom ()

man4/random.4 +0 -1

index edd047b77..8e5b7152b 100644
@@ -58,16 +58,21 @@ .SH DESCRIPTION
 .BR getrandom (2)
 or \fI/dev/random\fP instead.
 .PP
-The \fI/dev/random\fP device is a legacy interface which dates back to

man4/random.4 +9 -3

+.\" commit 50ee7529ec4500c88f8664560770a7a1b65db72b
+.PP
+The pre-5.6 \fI/dev/random\fP device is a legacy interface which dates
+back to a time where the cryptographic primitives used in the implementation
+of \fI/dev/urandom\fP were not widely trusted. It would return random bytes
+only within the estimated number of bits of fresh noise in the entropy pool,
+blocking until additional environmental noise is gathered.
+This old \fI/dev/random\fP is suitable for applications that need
 high quality randomness, and can afford indeterminate delays.
 .PP
-When the entropy pool is empty, reads from \fI/dev/random\fP will block
-until additional environmental noise is gathered.
 Since Linux 5.6, the
 .B O_NONBLOCK
 flag is ignored as

@@ -116,15 +121,15 @@ .SH DESCRIPTION
 .BR read (2)
 from
 .I /dev/random
-will return at most 512 bytes
+has the same maximum size since Linux 5.6. Between Linux 3.16 and 5.5,

man4/random.4 +2 -1

 \fI/dev/random\fP faster.
 .SS Usage
 The

@@ -145,7 +150,7 @@ .SS Usage
 network encryption session keys.
 (All major Linux distributions have saved the seed file across reboots
 since 2000 at least.)
-Since reads from
+Since reads from a pre-5.6
 .I /dev/random
 may block, users will usually want to open it in nonblocking mode
 (or perform a read with timeout),

@@ -246,6 +251,7 @@ .SS /proc interfaces
 .RE
 .TP
 .I read_wakeup_threshold
+Removed in Linux 5.6.
 This file
 contains the number of bits of entropy required for waking up processes
 that sleep waiting for entropy from

man7/random.7 +41 -41

index 69e6c2403..23aecf2a2 100644
@@ -59,17 +59,16 @@ .SS Initialization of the entropy pool
 When a sufficient number of random bits has been collected, the
 entropy pool is considered to be initialized.
 .SS Choice of random source
-Unless you are doing long-term key generation (and most likely not even
-then), you probably shouldn't be reading from the
+Unless your program may run at early-boot, before the entropy pool
+is initialized, there is no longer any palpable difference between
 .I /dev/random
-device or employing
-.BR getrandom (2)
-with the
-.B GRND_RANDOM
-flag.
-Instead, either read from the
+and
 .I /dev/urandom
-device or employ
+since Linux 5.6. (See the table below.)
+.PP.
+On older kernels, either read from the
+.I /dev/urandom
+device or (especially if you are concerned with early boot) employ
 .BR getrandom (2)
 without the
 .B GRND_RANDOM

@@ -77,18 +76,6 @@ .SS Choice of random source
 The cryptographic algorithms used for the
 .I urandom
 source are quite conservative, and so should be sufficient for all purposes.
-.PP
-The disadvantage of
-.B GRND_RANDOM
-and reads from
-.I /dev/random
-is that the operation can block for an indefinite period of time.
-Furthermore, dealing with the partially fulfilled
-requests that can occur when using
-.B GRND_RANDOM
-or when reading from
-.I /dev/random
-increases code complexity.
 .\"
 .SS Monte Carlo and other probabilistic sampling applications
 Using these interfaces to provide large quantities of data for

@@ -122,9 +109,9 @@ .SS Comparison between getrandom, /dev/urandom, and /dev/random
 T{
 .I /dev/random
 T}     T{
-Blocking pool
+CSPRNG output
 T}     T{
-If entropy too low, blocks until there is enough entropy again
+Never blocks
 T}     T{
 Blocks until enough entropy gathered
 T}

@@ -149,17 +136,6 @@ .SS Comparison between getrandom, /dev/urandom, and /dev/random
 T}
 T{
 .BR getrandom ()
-.B GRND_RANDOM
-T}     T{
-Same as
-.I /dev/random
-T}     T{
-If entropy too low, blocks until there is enough entropy again
-T}     T{
-Blocks until pool ready
-T}
-T{
-.BR getrandom ()
 .B GRND_NONBLOCK
 T}     T{
 Same as

@@ -171,21 +147,45 @@ .SS Comparison between getrandom, /dev/urandom, and /dev/random
 T}
 T{
 .BR getrandom ()
-.B GRND_RANDOM
-+
-.B GRND_NONBLOCK
+.B GRND_INSECURE
 T}     T{
 Same as
-.I /dev/random
+.I /dev/urandom
 T}     T{
-.B EAGAIN
-if not enough entropy available
+Never blocks
 T}     T{
-.B EAGAIN
+Returns output from uninitialized CSPRNG (may be low entropy and unsuitable for cryptography)
 T}
 .TE
 .ad
 .\"
+.SS The old blocking pool
+The above table describes the behavior of the interfaces since
+Linux 5.6.  In older kernels, the
+.I /dev/random
+used a separate blocking pool, and
+.BR getrandom ()
+had a
+.B GRND_RANDOM
+flag for reading from the blocking pool.
+.\"
+.PP
+The older blocking pool was a vestige of a time when the CSPRNG
+was not trusted.
+It assumed that entropy can run out by reading the CSPRNG.
+This has never been the case.
+Instead, programs using
+.B GRND_RANDOM
+and
+.I /dev/random
+had to deal with operations blocking indefinitely.
+Furthermore, dealing with the partially fulfilled
+requests that can occur when using
+.B GRND_RANDOM
+or when reading from
+.I /dev/random
+increases code complexity.
+.\"
 .SS Generating cryptographic keys
 The amount of seed material required to generate a cryptographic key
 equals the effective key size of the key.

Quoted 6 lines · Mingye Wang

Hi Alex,

attached are two patches split and revised as requested.

Regards,
Mingye

Hi! Here go some comments for one of the patches:

Quoted 1 line

* random.7: Revise "choice of random source" to remove any recommen-

Don't hyphenate words here.

Review

Quoted 5 lines

+.\" commit 30c08efec8884fb106b8e57094baa51bb4c44e32

(quoted diff) +4 -0

+As of Linux 5.6, \fI/dev/random\fP is identical to \fI/dev/urandom\fP,
+except that it blocks during early boot.
+A jitter-based seeding technique added in Linux 5.4 should help reduce
+block time.

Please use semantic newlines. See man-pages(7):

Use semantic newlines
In the source of a manual page, new sentences should be started on new lines, long sentences should be split into lines at clause breaks (commas, semicolons, colons, and so on), and long clauses should be split at phrase bound‐
aries. This convention, sometimes known as "semantic
newlines", makes it easier to see the effect of patches, which often operate at the level of individual sentences, clauses, or phrases.

Review

Quoted 7 lines

+.\" commit 50ee7529ec4500c88f8664560770a7a1b65db72b +.PP

(quoted diff) +4 -1

+The pre-5.6 \fI/dev/random\fP device is a legacy interface which dates
+back to a time where the cryptographic primitives used in the implementation
+of \fI/dev/urandom\fP were not widely trusted. It would return random bytes
-read from both files, but it will not make reads from
+read from both files, but it will not make reads from a pre-5.6

To make it easier to grep, prefix verions with the software name (this is done consistently in the pages). So something like 'pre-Linux 5.6' would work.

Thanks,
Alex

Hi Alex,

Apologies for being so slow (1.75 years! wow) at fixing my patch.

I've corrected
your mentioned deficits (linebreak, version number) and rebased my patch to the current HEAD at 46b7bca (Feb 27). Please check my work and tell me what you think.

Thanks,
Mingye Wang

Quoted 47 lines · Alejandro Colomar

Quoted 6 lines · Mingye Wang

Hi Alex,

attached are two patches split and revised as requested.

Regards,
Mingye

Hi! Here go some comments for one of the patches:

Quoted 1 line

* random.7: Revise "choice of random source" to remove any recommen-

Don't hyphenate words here.

man/man4/random.4 If this is of concern in your application, use · line 65

Quoted 5 lines

+.\" commit 30c08efec8884fb106b8e57094baa51bb4c44e32

man/man4/random.4 +4 -0

+As of Linux 5.6, \fI/dev/random\fP is identical to \fI/dev/urandom\fP,
+except that it blocks during early boot.
+A jitter-based seeding technique added in Linux 5.4 should help reduce
+block time.

Please use semantic newlines. See man-pages(7):

Use semantic newlines
In the source of a manual page, new sentences should be started on new lines, long sentences should be split into lines at clause breaks (commas, semicolons, colons, and so on), and long clauses should be split at phrase bound‐
aries. This convention, sometimes known as "semantic
newlines", makes it easier to see the effect of patches, which often operate at the level of individual sentences, clauses, or phrases.

Review

Quoted 7 lines

+.\" commit 50ee7529ec4500c88f8664560770a7a1b65db72b +.PP

(quoted diff) +4 -1

+The pre-5.6 \fI/dev/random\fP device is a legacy interface which dates
+back to a time where the cryptographic primitives used in the implementation
+of \fI/dev/urandom\fP were not widely trusted. It would return random bytes
-read from both files, but it will not make reads from
+read from both files, but it will not make reads from a pre-5.6

To make it easier to grep, prefix verions with the software name (this is done consistently in the pages). So something like 'pre-Linux 5.6' would work.

Thanks,
Alex

Linux kernel 5.6 no longer has a blocking random pool. This commit updates all relevant references to reflect this change.

* random.7: Remove references to the blocking pool in the table. Add a note about the blocking pool.
* random.7: Revise "choice of random source" to remove any recommendation of the blocking pool. Stop suggesting that the blocking pool is safer (it's not after initialization).
* random.7: Add table entry for GRND_INSECURE.
* getrandom.2: Add flag entry for GRND_INSECURE.
* getrandom.2: Add removal note to GRND_RANDOM.
* random.4: Split DESCRIPTION paragraph on /dev/random into two, one for the new behavior and the other for the old.
* random.4: Adjust size limits and /proc list for 5.6.
* random.4: Mention blocking resolution by high-precision timer entropy.

Closes https://bugzilla.kernel.org/show_bug.cgi?id=214885

Signed-off-by Mingye Wang <[email protected]>

---
 man/man2/getrandom.2 |  9 ++++++
 man/man4/random.4    | 30 ++++++++++++++-----
 man/man7/random.7    | 70 ++++++++++++++++++++++++++------------------
 3 files changed, 72 insertions(+), 37 deletions(-)

man/man2/getrandom.2 +2 -0

index 9e782e6..5f0a2da 100644
@@ -62,6 +62,8 @@ argument is a bit mask that can contain zero or more of the following values
 ORed together:
 .TP
 .B GRND_RANDOM
+.\" commit 48446f198f9adcb499b30332488dfd5bc3f176f6
+Ignored since Linux 5.6.
 If this bit is set, then random bytes are drawn from the
 .I random
 source

man/man2/getrandom.2 +4 -0

+Request best-effort, non-cryptographic-quality random bytes.
+If this bit is set, then
+.BR getrandom ()
+will never block or fail due to a lack of entropy.
 .SH RETURN VALUE
 On success,
 .BR getrandom ()

man/man4/random.4 +0 -0

index 0a651b0..95200bc 100644

man/man4/random.4 +17 -7

+except that it blocks during early boot.
+A jitter-based seeding technique added in Linux 5.4 should help reduce
+block time.
+.\" commit 50ee7529ec4500c88f8664560770a7a1b65db72b
+.P
+The pre-Linux 5.6
 .I /dev/random
-device is a legacy interface which dates back to
+device was a legacy interface which dates back to
 a time where the cryptographic primitives used in the implementation
 of
 .I /dev/urandom
 were not widely trusted.
-It will return random bytes only within the estimated number of
-bits of fresh noise in the entropy pool, blocking if necessary.
+It would return random bytes only within the estimated number of bits of fresh
+noise in the entropy pool, blocking until additional environmental noise is
+gathered.
+This old
 .I /dev/random
-is suitable for applications that need
+was suitable for applications that need
 high quality randomness, and can afford indeterminate delays.
 .P
 When the entropy pool is empty, reads from

@@ -121,7 +133,8 @@ A
 .BR read (2)
 from
 .I /dev/random
-will return at most 512 bytes
+has the same maximum size since Linux 5.6. Between Linux 3.16 and 5.5,
+the maximum size was 512 bytes. (340 bytes before Linux 2.6.12)
 .\" SEC_XFER_SIZE in drivers/char/random.c
 (340 bytes before Linux 2.6.12).
 .P

@@ -133,7 +146,7 @@ will update the
 entropy pool with the data written, but this will not result in a
 higher entropy count.
 This means that it will impact the contents
-read from both files, but it will not make reads from
+read from both files, but it will not make reads from a pre-Linux 5.6
 .I /dev/random
 faster.
 .SS Usage

@@ -158,7 +171,7 @@ soon as it is reloaded in the boot sequence, and perfectly adequate for
 network encryption session keys.
 (All major Linux distributions have saved the seed file across reboots
 since 2000 at least.)
-Since reads from
+Since reads from a pre-Linux 5.6
 .I /dev/random
 may block, users will usually want to open it in nonblocking mode
 (or perform a read with timeout),

@@ -262,6 +275,7 @@ It contains the value 4096.
 .RE
 .TP
 .I read_wakeup_threshold
+Removed in Linux 5.6.
 This file
 contains the number of bits of entropy required for waking up processes
 that sleep waiting for entropy from

man/man7/random.7 +10 -24

index fda408d..c5e959f 100644
@@ -54,17 +54,16 @@ The kernel collects bits of entropy from the environment.
 When a sufficient number of random bits has been collected, the
 entropy pool is considered to be initialized.
 .SS Choice of random source
-Unless you are doing long-term key generation (and most likely not even
-then), you probably shouldn't be reading from the
+Unless your program may run at early-boot, before the entropy pool
+is initialized, there is no longer any palpable difference between
 .I /dev/random
-device or employing
-.BR getrandom (2)
-with the
-.B GRND_RANDOM
-flag.
-Instead, either read from the
+and
+.I /dev/urandom
+since Linux 5.6. (See the table below.)
+.PP.
+On older kernels, either read from the
 .I /dev/urandom
-device or employ
+device or (especially if you are concerned with early boot) employ
 .BR getrandom (2)
 without the
 .B GRND_RANDOM

@@ -117,9 +116,9 @@ T}
 T{
 .I /dev/random
 T}	T{
-Blocking pool
+CSPRNG output
 T}	T{
-If entropy too low, blocks until there is enough entropy again
+Never blocks
 T}	T{
 Blocks until enough entropy gathered
 T}

@@ -144,17 +143,6 @@ Blocks until pool ready
 T}
 T{
 .BR getrandom ()
-.B GRND_RANDOM
-T}	T{
-Same as
-.I /dev/random
-T}	T{
-If entropy too low, blocks until there is enough entropy again
-T}	T{
-Blocks until pool ready
-T}
-T{
-.BR getrandom ()
 .B GRND_NONBLOCK
 T}	T{
 Same as

@@ -166,21 +154,45 @@ T}	T{
 T}
 T{
 .BR getrandom ()
-.B GRND_RANDOM
-+

man/man7/random.7 +0 -0

 .SS Generating cryptographic keys
 The amount of seed material required to generate a cryptographic key
 equals the effective key size of the key.

Signature · 1 line

2.51.0

Quoted 2 lines

Does /dev/urandom block when reading with read(2) before the pool is

ready? I assume it blocks.

From what I've heard it does not. Working on other comments.

Quoted 4 lines · Mingye Wang

Quoted 2 lines

Does /dev/urandom block when reading with read(2) before the pool is

ready? I assume it blocks.

Quoted 1 line

From what I've heard it does not. Working on other comments.

Done working on the review comments, but git send-email is not working for me again (it accepts no proxy settings and tunnels are flakey; worse, it doesn't save my multi-sentence edited heading, or my in-reply to string which I need to copy again, or the recipient list...). I apologize but I am sending it as an attachment.

I am sending only the revised version of the patch you commented on (the one that touches three files, random.{4,7}, getrandom.2). You might notice that this patch is much larger. This is because when editing the table I noticed that there's really no reason to keep the "Pool" column around for 5.6+, but removing it also felt off. In the end I just made two separate tables, but with identical cell-width settings.

Speaking of cell-width settings, the existing version was not filling the entire screen, which contravenes what man-pages(7) says about using lbx. Adding an "x" does make it look nicer too.

(I have no idea whether this is the fourth version, but it's definitely after the third and five seems too many. In any case, I will be incrementing this number in future patch submissions to keep things navigable.)

Regards,
Mingye Wang (Artoria2e5)

Quoted 5 lines · Mingye Wang

Quoted 2 lines

Does /dev/urandom block when reading with read(2) before the pool is

ready? I assume it blocks.

From what I've heard it does not. Working on other comments.

I am sending only the revised version of the patch you commented on (the one that touches three files, random.{4,7}, getrandom.2). You might notice that this patch is much larger. This is because when editing the table I noticed that there's really no reason to keep the "Pool" column around for 5.6+, but removing it also felt off. In the end I just made two separate tables, but with identical cell-width settings.

Speaking of cell-width settings, the existing version was not filling the entire screen, which contravenes what man-pages(7) says about using lbx. Adding an "x" does make it look nicer too.

(I have no idea whether this is the fourth version, but it's definitely after the third and five seems too many. In any case, I will be incrementing this number in future patch submissions to keep things navigable.)

Regards,
Mingye Wang (Artoria2e5)

Hi Mingye,

Quoted 7 lines · Mingye Wang

Quoted 5 lines · Mingye Wang

Quoted 2 lines

Does /dev/urandom block when reading with read(2) before the pool is

ready? I assume it blocks.

From what I've heard it does not. Working on other comments.

[...]

Quoted 8 lines

I am sending only the revised version of the patch you commented on (the one that touches three files, random.{4,7}, getrandom.2). You might notice that this patch is much larger. This is because when editing the table I noticed that there's really no reason to keep the "Pool" column around for 5.6+, but removing it also felt off. In the end I just made two separate tables, but with identical cell-width settings.

Please split into more patches. I think this patch could be broken into a set of many small patches, each of which does one thing. 5 or 10 patches would be okay.

Quoted 11 lines

Speaking of cell-width settings, the existing version was not filling the entire screen, which contravenes what man-pages(7) says about using lbx. Adding an "x" does make it look nicer too.

(I have no idea whether this is the fourth version, but it's definitely after the third and five seems too many. In any case, I will be incrementing this number in future patch submissions to keep things navigable.)

Regards,
Mingye Wang (Artoria2e5)

Some review of v4:

	diff --git a/man/man4/random.4 b/man/man4/random.4
	index 0a651b03f..071fc99ef 100644
	--- a/man/man4/random.4
	+++ b/man/man4/random.4
	@@ -56,17 +56,29 @@ .SH DESCRIPTION
	 .I /dev/random
	 instead.
	 .P
	-The
	+.\" commit 30c08efec8884fb106b8e57094baa51bb4c44e32
	+Since Linux 5.6,
	 .I /dev/random
	-device is a legacy interface which dates back to
	+is identical to
	+.IR /dev/urandom ,
	+except that it blocks during early boot.
	+A jitter-based seeding technique added in Linux 5.4 should help reduce
	+block time.
	+.\" commit 50ee7529ec4500c88f8664560770a7a1b65db72b
	+.P
	+The pre-Linux 5.6
	+.I /dev/random
	+device was a legacy interface which dates back to

I'd prefer 'Before Linux 5.6,' instead of 'The pre-Linux 5.6'.

a time where the cryptographic primitives used in the implementation of
.I /dev/urandom
were not widely trusted.
-It will return random bytes only within the estimated number of -bits of fresh noise in the entropy pool, blocking if necessary. +It would return random bytes only within the estimated number of bits of fresh +noise in the entropy pool, blocking until additional environmental noise is +gathered.
+This old
.I /dev/random

Please use semantic newlines. See man-pages(7):

$ MANWIDTH=72 man man-pages | sed -n '/Use semantic newlines/,/^$/p'
   Use semantic newlines
     In the source of a manual page, new sentences should be started on
     new lines, long sentences should be split  into  lines  at  clause
     breaks  (commas,  semicolons, colons, and so on), and long clauses
     should be split at phrase boundaries.  This convention,  sometimes
     known as "semantic newlines", makes it easier to see the effect of
     patches, which often operate at the level of individual sentences,
     clauses, or phrases.

I'd write it as:

```
It would return random bytes
only within the estimated number of bits of fresh noise
in the entropy pool,
blocking until additional environmental noise is gathered.
```

Same here:

	@@ -121,7 +133,8 @@ .SH DESCRIPTION
	 .BR read (2)
	 from
	 .I /dev/random
	-will return at most 512 bytes
	+has the same maximum size since Linux 5.6. Between Linux 3.16 and 5.5,
	+the maximum size was 512 bytes
	 .\" SEC_XFER_SIZE in drivers/char/random.c
	 (340 bytes before Linux 2.6.12).
	 .P

Use .P instead of .PP here:

diff --git a/man/man7/random.7 b/man/man7/random.7
index fda408d38..65e21a07e 100644
--- a/man/man7/random.7
+++ b/man/man7/random.7
@@ -54,17 +54,16 @@ .SS Initialization of the entropy pool
When a sufficient number of random bits has been collected, the entropy pool is considered to be initialized.
.SS Choice of random source
-Unless you are doing long-term key generation (and most likely not even -then), you probably shouldn't be reading from the
+Unless your program may run at early-boot, before the entropy pool +is initialized, there is no longer any palpable difference between
.I /dev/random
-device or employing
-.BR getrandom (2)
-with the
-.B GRND_RANDOM
-flag.
-Instead, either read from the
+and
.I /dev/urandom
-device or employ
+since Linux 5.6 (see the table below).
+.PP
+On older kernels, either read from the
+.I /dev/urandom
+device or (especially if you are concerned with early boot) employ
.BR getrandom (2)
without the
.B GRND_RANDOM

Have a lovely day!
Alex

Signature · 2 lines

<https://www.alejandro-colomar.es>
Use port 80 (that is, <...:80/>).

Ping. :)

Quoted 150 lines · Alejandro Colomar

Hi Mingye,

Quoted 7 lines · Mingye Wang

Quoted 5 lines · Mingye Wang

Quoted 2 lines

Does /dev/urandom block when reading with read(2) before the pool is

ready? I assume it blocks.

From what I've heard it does not. Working on other comments.

[...]

Quoted 8 lines

I am sending only the revised version of the patch you commented on (the one that touches three files, random.{4,7}, getrandom.2). You might notice that this patch is much larger. This is because when editing the table I noticed that there's really no reason to keep the "Pool" column around for 5.6+, but removing it also felt off. In the end I just made two separate tables, but with identical cell-width settings.

Please split into more patches. I think this patch could be broken into a set of many small patches, each of which does one thing. 5 or 10 patches would be okay.

Quoted 11 lines

Speaking of cell-width settings, the existing version was not filling the entire screen, which contravenes what man-pages(7) says about using lbx. Adding an "x" does make it look nicer too.

(I have no idea whether this is the fourth version, but it's definitely after the third and five seems too many. In any case, I will be incrementing this number in future patch submissions to keep things navigable.)

Regards,
Mingye Wang (Artoria2e5)

Some review of v4:

	diff --git a/man/man4/random.4 b/man/man4/random.4
	index 0a651b03f..071fc99ef 100644
	--- a/man/man4/random.4
	+++ b/man/man4/random.4
	@@ -56,17 +56,29 @@ .SH DESCRIPTION
	 .I /dev/random
	 instead.
	 .P
	-The
	+.\" commit 30c08efec8884fb106b8e57094baa51bb4c44e32
	+Since Linux 5.6,
	 .I /dev/random
	-device is a legacy interface which dates back to
	+is identical to
	+.IR /dev/urandom ,
	+except that it blocks during early boot.
	+A jitter-based seeding technique added in Linux 5.4 should help reduce
	+block time.
	+.\" commit 50ee7529ec4500c88f8664560770a7a1b65db72b
	+.P
	+The pre-Linux 5.6
	+.I /dev/random
	+device was a legacy interface which dates back to

I'd prefer 'Before Linux 5.6,' instead of 'The pre-Linux 5.6'.

a time where the cryptographic primitives used in the implementation of
.I /dev/urandom
were not widely trusted.
-It will return random bytes only within the estimated number of -bits of fresh noise in the entropy pool, blocking if necessary. +It would return random bytes only within the estimated number of bits of fresh +noise in the entropy pool, blocking until additional environmental noise is +gathered.
+This old
.I /dev/random

Please use semantic newlines. See man-pages(7):

$ MANWIDTH=72 man man-pages | sed -n '/Use semantic newlines/,/^$/p'
   Use semantic newlines
     In the source of a manual page, new sentences should be started on
     new lines, long sentences should be split  into  lines  at  clause
     breaks  (commas,  semicolons, colons, and so on), and long clauses
     should be split at phrase boundaries.  This convention,  sometimes
     known as "semantic newlines", makes it easier to see the effect of
     patches, which often operate at the level of individual sentences,
     clauses, or phrases.

I'd write it as:

```
It would return random bytes
only within the estimated number of bits of fresh noise
in the entropy pool,
blocking until additional environmental noise is gathered.
```

Same here:

	@@ -121,7 +133,8 @@ .SH DESCRIPTION
	 .BR read (2)
	 from
	 .I /dev/random
	-will return at most 512 bytes
	+has the same maximum size since Linux 5.6. Between Linux 3.16 and 5.5,
	+the maximum size was 512 bytes
	 .\" SEC_XFER_SIZE in drivers/char/random.c
	 (340 bytes before Linux 2.6.12).
	 .P

Use .P instead of .PP here:

diff --git a/man/man7/random.7 b/man/man7/random.7
index fda408d38..65e21a07e 100644
--- a/man/man7/random.7
+++ b/man/man7/random.7
@@ -54,17 +54,16 @@ .SS Initialization of the entropy pool
When a sufficient number of random bits has been collected, the entropy pool is considered to be initialized.
.SS Choice of random source
-Unless you are doing long-term key generation (and most likely not even -then), you probably shouldn't be reading from the
+Unless your program may run at early-boot, before the entropy pool +is initialized, there is no longer any palpable difference between
.I /dev/random
-device or employing
-.BR getrandom (2)
-with the
-.B GRND_RANDOM
-flag.
-Instead, either read from the
+and
.I /dev/urandom
-device or employ
+since Linux 5.6 (see the table below).
+.PP
+On older kernels, either read from the
+.I /dev/urandom
+device or (especially if you are concerned with early boot) employ
.BR getrandom (2)
without the
.B GRND_RANDOM

Have a lovely day!
Alex

Signature · 2 lines

<https://www.alejandro-colomar.es>
Use port 80 (that is, <...:80/>).

Signature · 1 line

<https://www.alejandro-colomar.es>

https://bugzilla.kernel.org/show_bug.cgi?id=214885

Mingye Wang ([email protected]) changed:

What |Removed |Added
----------------------------------------------------------------------------

 Attachment #304322|0                           |1

is obsolete| |

--- Comment #3 from Mingye Wang ([email protected]) --- Created attachment 306038
--> https://bugzilla.kernel.org/attachment.cgi?id=306038&action=edit
Patch set as of 2024-03-25

I am not functioning well enough to do man pages right now. The attached tarball contains the current progress of my tree. It contains three patches suitable for the current HEAD at c6a68aef7334f32ffffb74050702b8b53b064f37:

* The first is basically #304322, which changes a bunch of things. It's been e-mailed, but I honestly don't recall whether I've done the requested changes.
* The second is a small change. It has also been emailed, and I also don't know
if the formatting issues have been fixed.
* The third is new. It addresses a pool size change and a write_wakeup_threshold obsoletion. It could have addressed the obsolete "urandom_min_reseed_secs" too, but I don't want to hunt down when it appeared and when it stopped working.