-
Notifications
You must be signed in to change notification settings - Fork 2.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Don't fail in gz_load() if read would block #993
base: develop
Are you sure you want to change the base?
Conversation
I.e. when [errno] is EAGAIN or EWOULDBLOCK) E.g. if file descriptor is a non-blocking TCP socket, and more data are expected. This might cause a problem if compressed bytes are read but do not generate an uncompressed byte, because the return value will be 0, which usually indicates EOF.
N.B. do not merge this PR yet; I still have some work to do to see if what I am trying to do is even feasible. |
@@ -24,7 +24,7 @@ local int gz_load(gz_statep state, unsigned char *buf, unsigned len, | |||
break; | |||
*have += (unsigned)ret; | |||
} while (*have < len); | |||
if (ret < 0) { | |||
if (ret < 0 && errno != EAGAIN && errno != EWOULDBLOCK) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is the core of this PR, because all of the read-related wrappers here in gzread.c
end up calling this gz_load routine: when
- BOTH state-fd (file descriptor) is non-blocking (fcntl(fd, SET_FL, fcntl(fd, GET_FL) | O_NONBLOCK))
- AND there are no data available to read,
then the read(2) return value (ret above, assigned on line 22) will be -1, and errno will be either EWOULDBLOCK or EAGAIN (Resource not available).
The change here prevents that case from returning -1 from gz_load() here, which in turn
allows the calling routine* to continue working with data from the file descriptor.
It gets a little tricky, because no bytes are returned, and 0 as the return value from read(2)
indicates EOF (End-Of-File).
N.B. tt may make sense to clear errno to a value of 0 here in gz_load(), just before line 22 above.
* gz_avail() or gz_fetch() or gz_read(), with the latter being the ultimate caller of the first two.
…atus of client and server suitable for automated testing (make test...)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This Pull Request is ready to merge, pending review and approval of the repo owner (Dr. Adler).
@@ -307,25 +309,31 @@ local z_size_t gz_read(gz_statep state, voidp buf, z_size_t len) { | |||
buffer */ | |||
else if (state->how == LOOK || n < (state->size << 1)) { | |||
/* get more output, looking for header if required */ | |||
errno = 0; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Prepare for errno to become EAGAIN/EWOULDBLOCK, indicating that the read(2) on the
non-blocking file descriptor would block, so the read(2) returns a value of -1.
@@ -268,6 +268,7 @@ local int gz_skip(gz_statep state, z_off64_t len) { | |||
local z_size_t gz_read(gz_statep state, voidp buf, z_size_t len) { | |||
z_size_t got; | |||
unsigned n; | |||
int load_errno; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Variable to keep track of errno from the read(2) in gz_load(), whether gz_load() is called from here (gz_read())
- EITHER directly,
- OR indirectly via gz_decomp() or gz_fetch().
if (gz_fetch(state) == -1) | ||
return 0; | ||
if (!state->x.have) { load_errno = errno; } |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The call to gz_fetch() (line 313may have resulted in a call to gz_load()/read(2) and updated the internal output buffer, so errno may have a value of EAGAIN/EWOULDBLOCK, indicating that read(2) would have blocked on a non-blocking file.
The next line, 316, will continue to the while at the end of this do-while loop.
The state->x internal output buffer is encapsulated in state->x.next and state->x.have (see lines 293-300 above).
So if there is any data in the internal output buffer (state->x.have is non-zero), then don't copy the value, possibly EAGAIN/EWOULDBLOCK, of errno from gz_fetch() here, because that would cause the while to exit the loop and the data in the internal input buffer would be lost.
If the the state->x internal output buffer is empty here (state->x.have is 0), then there was no data read in the gz_fetch() call (line 313), and it is save to copy the errno value to load_errno and continue, which, if errno is EAGAIN/EWOULDBLOCK here, will exit this do-while loop .
if (gz_load(state, (unsigned char *)buf, n, &n) == -1) | ||
return 0; | ||
load_errno = errno; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
gz_load() (line 324) read the data directly into the external output buffer (buf), and state->x was empty and not used here, so put the value of errno into load_errno, which if that value is EAGAIN/EWOULDBLOCK, will cause the do-while loop to exit (line 346).
if (gz_decomp(state) == -1) | ||
return 0; | ||
load_errno = errno; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
gz_decomp() (line 334) read the data directly into the external output buffer (buf; lines 331-332), and that buffer will be declared empty when state->x.have is cleared to 0 (line 338), so put the value of errno into load_errno, which if the value is ...
check: test | ||
|
||
test: all teststatic testshared | ||
|
||
teststatic: static | ||
teststatic: static testgznonblk |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Addubg the gznoblk test target as a dependency to this static test target, so the non-blocking functionality added to gzread.c in this PR can be tested.
else \ | ||
echo ' *** gznonblk test FAILED ***'; false; \ | ||
fi | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Here are the details of the gznonblk test target, which tests the non-blocking functionality added to gzread.c in this PR. The command is
./gznonblk 44444 --client-fork 127.0.0.1 first second --delay third --delay -stopserver-
- ./gznonblk - run the executable built by staticgznb
- N.B. both server and client sides are contained in this one executable.
- 44444 - the port number for the server socket
- --client-fork - the server parent spawns the client child
N.B. the rest of the arguments are strings or directives for the client only
- 127.0.0.1 - the server host string to which the client will connect
- i.e. localhost, but localhost may point to :: (IPV6) on some systems
- first, second - the first two strings that the client will gzwrite to, and the server will gzread from, the socket
- N.B. these two arguments are typically concatenated into a single gzread on the server side
- --delay - a directive to the client to delay before processing the next argument
- third - the third string gz-transferred from client to server
- --delay - another delay to ensure the next and final string is not concatenated to the previous "third"
- -stopserver - the final string gz-transferred from client to server
- which will trigger the server to exit (see variable stopserver in test/gznonblk.c)
Once the server (parent) exits, it will pause, then check for the exit status of the client (child), and set its own return status to success (zero) IFF both server and client exited successfully.
@@ -145,6 +154,9 @@ example.o: $(SRCDIR)test/example.c $(SRCDIR)zlib.h zconf.h | |||
minigzip.o: $(SRCDIR)test/minigzip.c $(SRCDIR)zlib.h zconf.h | |||
$(CC) $(CFLAGS) $(ZINCOUT) -c -o $@ $(SRCDIR)test/minigzip.c | |||
|
|||
gznonblk.o: $(SRCDIR)test/gznonblk.c $(SRCDIR)zlib.h zconf.h | |||
$(CC) $(CFLAGS) $(ZINCOUT) -c -o $@ $(SRCDIR)test/gznonblk.c |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Compiling test/gznonblk.c, patterned after the other test programs
@@ -287,6 +299,9 @@ example$(EXE): example.o $(STATICLIB) | |||
minigzip$(EXE): minigzip.o $(STATICLIB) | |||
$(CC) $(CFLAGS) $(LDFLAGS) -o $@ minigzip.o $(TEST_LIBS) | |||
|
|||
gznonblk$(EXE): gznonblk.o $(STATICLIB) | |||
$(CC) $(CFLAGS) $(LDFLAGS) -o $@ gznonblk.o $(TEST_LIBS) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Builds the gznonblk test program using the static library (libz.a)
@@ -0,0 +1,563 @@ | |||
#include <zlib.h> |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
New file, the source code for testing the new gz non-blocking functionality added to gzread.c in this PR. Please feel free to let me know if the comments could be better.
I added a few more bells and whistles. But note that the primary changes in this PR are the ten new lines of code in gzread.c, and each of those changes has github comments associated with it. Also, those changes should not affect applications using zlib, because errno=EWOULDBLOCK/EAGAIN will not occur for blocking reads, and anyone using non-blocking reads would not be using zlib v1.3.1's gzread in the first place because it doesn't work in that environment. Let me know if you have any questions. |
P.S. My style is wrong in test/gznonblk.c; I can fix that. |
@madler: Have you seen this PR? |
I.e. when [errno] is EAGAIN or EWOULDBLOCK)
E.g. if file descriptor is a non-blocking TCP socket,
and more data are expected. This might cause a
problem if compressed bytes are read but do not
generate an uncompressed byte, because the return
value will be 0, which usually indicates EOF.
See madler/zlib Issue #992