Hello Robin,
On 10 Oct 2022, at 13:27, Robin Roevens robin.roevens@disroot.org wrote:
Hi
Michael Tremer schreef op ma 10-10-2022 om 11:04 [+0100]:
Hello,
On 9 Oct 2022, at 00:09, Robin Roevens robin.roevens@disroot.org wrote:
Hi Michael
Thanks again for the review and comments.
Michael Tremer schreef op vr 07-10-2022 om 16:01 [+0100]:
Hello again,
On 6 Oct 2022, at 18:59, Robin Roevens robin.roevens@disroot.org wrote:
- Addonctrl will now check in addon metadata for the exact
initscript names (Services). If more than one initscript is defined for an addon, the requested action will be performed on all listed initscripts.
- Added posibility to perform action on a specific initscript
of an addon instead of on all initscripts of the addon.
- New action 'list-services' to display a list of services
related to an addon.
- New action 'boot-status' to display wether service(s) are
enabled to start on boot or not.
- More error checking and cleaner error reporting to user
- General cleanup and code restructuring to avoid code
duplication
- Updated and made usage instructions more verbose.
Fixes: Bug#12935 Signed-off-by: Robin Roevens robin.roevens@disroot.org
src/misc-progs/addonctrl.c | 397
+++++++++++++++++++++++++++++++---
1 file changed, 336 insertions(+), 61 deletions(-)
diff --git a/src/misc-progs/addonctrl.c b/src/misc- progs/addonctrl.c index 14b4b1325..1687aac19 100644 --- a/src/misc-progs/addonctrl.c +++ b/src/misc-progs/addonctrl.c @@ -10,71 +10,346 @@ #include <string.h> #include <unistd.h> #include <sys/types.h> +#include <sys/stat.h> #include <fcntl.h> +#include <dirent.h> +#include <fnmatch.h> +#include <errno.h> #include "setuid.h"
#define BUFFER_SIZE 1024
+const char *enabled_path = "/etc/rc.d/rc3.d"; +const char *disabled_path = "/etc/rc.d/rc3.d/off";
+char errormsg[BUFFER_SIZE] = ""; +const char *usage =
- "Usage\n"
- " addonctrl <addon>
(start|stop|restart|reload|enable|disable|status|boot- status|list- services) [<service>]\n"
- "\n"
- "Options:\n"
- " <addon>\t\tName of the addon to control\n"
- " <service>\t\tSpecific service of the addon to control
(optional)\n"
- " \t\t\tBy default the requested action is performed on all
related services. See also 'list-services'.\n"
- " start\t\t\tStart service(s) of the addon\n"
- " stop\t\t\tStop service(s) of the addon\n"
- " restart\t\tRestart service(s) of the addon\n"
- " enable\t\tEnable service(s) of the addon to start at
boot\n"
- " disable\t\tDisable service(s) of the addon to start at
boot\n"
- " status\t\tDisplay current state of addon service(s)\n"
- " boot-status\t\tDisplay wether service(s) is enabled on
boot or not\n"
- " list-services\t\tDisplay a list of services related to
the addon";
+// Find a file <path> using <filepattern> as glob pattern. +// Returns the found filename or NULL if not found +char *find_file_in_dir(const char *path, const char *filepattern) +{
- struct dirent *entry;
- DIR *dp;
- char *found = NULL;
- if ((dp = opendir(path)) != NULL) {
- while(found == NULL && (entry = readdir(dp)) != NULL)
- if (fnmatch(filepattern, entry->d_name, FNM_PATHNAME)
== 0)
- found = strdup(entry->d_name);
- closedir(dp);
- }
- return found;
+}
+// Reads Services metadata for <addon>. +// Returns pointer to array of strings containing the services for
<addon> +// and sets <servicescnt> to the number of found services +char **get_addon_services(const char *addon, int *servicescnt, const char *filter) { + const char *metafile_prefix = "/opt/pakfire/db/installed/meta- "; + const char *metadata_key = "Services"; + const char *keyvalue_delim = ":"; + const char *service_delim = " "; + char *token; + char **services = NULL; + char *service; + char *line = NULL; + size_t line_len = 0; + int i = 0; + char *metafile; + + if (addon == NULL) { + errno = EINVAL; + return NULL; + } + + if (asprintf(&metafile, "%s%s", metafile_prefix, addon) == - 1) { + errno = ENOMEM; + return NULL; + }
You should initialise metafile with NULL before passing it to asprintf().
I'm not sure why? I can't find any information on why it should be initialized to NULL as it is passed to malloc, which doesn't seem to require it to be NULL beforehand. Or at least I didn't find any documentation stating as such.
What I did find is that many implementations of C leave the string in undefined state when it was unable to malloc. While others (like GNU) set it to NULL explicitly if malloc fails.
I suppose my concern is rooted in the different implementations of such functions.
Since this program is exclusively running on IPFire and using glibc, you might get away with this. However, I try to write my code as portable as possible and avoid any uninitialised pointers wherever possible (because they are normally a recipe for disaster), that I initialise everything.
Should this be absolutely unnecessary, the compiler will skip the initialisation anyways.
But since I check the return value of asprintf, I never read or touch the returned string, undefined or not, so I didn't think it was required.
But I'll initialize them. It's probably indeed the safest to do anyway.
Yes, better be safe than sorry.
reallocarray() would require this, and so it would be more uniform to initialise everything.
asprintf() will automatically set errno, actually.
Ok. I was not sure as the manpage for asprintf does not mention that while the manpage for for example strdup does mention that explicitly.
Depending on implementation, asprintf might never set it itself, but the function it calls (i.e. malloc, etc.).
And I have investigated the source of asprintf, and it does not set errno when there is a problem with malloc.. But now I found that malloc itself actually sets errno. So it will indeed be set by asprintf too as that calls malloc. So I'll remove all my setting of ENOMEM.
- FILE *fp = fopen(metafile,"r");
- if (fp != NULL) {
Just FYI, I like writing this check as “if (!fp)” because it is a a lot shorter...
It seems there are 2 camps about this, one just writes what you like to write, while the other argues it is more clear that you are working with pointers by explicitly checking against NULL.
Yes, there are indeed two camps. One of them is being so funny that they set NULL to a value like “5” and check if your code still works.
My code won’t. But I am sure their code won’t work any more if I define 5 as 6, and the number 2 because 9 and free becomes malloc. Changing arbitrary things doesn’t make any sense.
But if you think that is unnecessary I'll change it.
This is more advisory than a demand. When reading code, the brain optimises for certain patterns and if you recognise those patterns again somewhere else, the brain will understand things quicker. So I kind of like it when people code in the same style as I do because it helps me to review things faster and better.
Should I change all my other checks against NULL also? Or only the filepointer ?
If you don’t mind doing that, I would like it :)
No problem. I think you are in the right position to demand a specific code style for use on IPFire :-)
I don’t know what it requires to demand something very specific.
I am just trying to forward my experience since I made lots of mistakes myself and learned from it here and there - and there is tons of stuff that I didn’t quite learn, yet.
- // Get initscript(s) for addon from meta-file
- while (getline(&line, &line_len, fp) != -1 && services ==
NULL) {
- // Strip newline
- char *newline = strchr(line, '\n');
- if (newline) *newline = 0;
Yes, this will cut off the trailing newline.
- // Parse key/value and look for required key.
- token = strtok(line, keyvalue_delim);
- if (token != NULL && strcmp(token, metadata_key) == 0)
{
- token = strtok(NULL, keyvalue_delim);
- if (token != NULL) {
- // Put each service in services array
- service = strtok(token, service_delim);
- while (service != NULL) {
- // if filter is set, only select filtered
service
- if ((filter != NULL && strcmp(filter,
service) == 0) ||
- filter == NULL) {
- services = reallocarray(services ,i+1
,sizeof (char *));
- if (services != NULL)
- services[i++] = strdup(service);
In this block, I would find the !filter way a lot easier to read.
Agreed
Technically you could check whether strdup() was successful, but I would accept this version.
Indeed, I forgot. As you gave enough comments to need a new version of the patchset, I'll include this too anyway. :-)
- else
- break;
- }
- service = strtok(NULL, service_delim);
- }
- }
- }
- }
- free(line);
Are you sure that line will always be non-NULL?
What happens if you read an empty file and getline() exists immediately without ever allocating a buffer?
You are right. Not sure what will happen then. So I'll check if 'line' is not NULL before trying to free it.
I always do that (pretty much) and I am sure the compiler optimises that out in most cases.
- fclose(fp);
- } else {
- snprintf(errormsg, BUFFER_SIZE - 1, "Addon '%s' not
found.\n\n%s", addon, usage);
- }
- free(metafile);
- *servicescnt = i;
- return services;
+}
+// Calls initscript <service> with parameter <action> +int initscript_action(const char *service, const char *action) {
- const char *initd_path = "/etc/rc.d/init.d";
- char *initscript;
Same here. Please initialise it.
- char *argv[] = {
- action,
- NULL
- };
- int r = 0;
- if ((r = asprintf(&initscript, "%s/%s", initd_path,
service)) != -1) {
- r = run(initscript, argv);
- free(initscript);
- } else {
- errno = ENOMEM;
- }
This could look slightly cleaner as:
…
r = asprintf(&initscript, ….); if (r > 0) { r = run(initscript, argv); }
if (initscript) free(initscript);
return r;
Maybe… I don’t know.
Yes, indeed, as I don't have to set ENOMEM
- return r;
+}
+// Move an initscript with filepattern from <src_path> to
<dest_path> +// Returns: +// -1: Error during move or memory allocation. Details in errno +// 0: Success +// 1: file was not moved, but is already in <dest_path> +// 2: file does not exist in either in <src_path> or <dest_path> +int move_initscript_by_pattern(const char *src_path, const char *dest_path, const char *filepattern) { + char *src = NULL; + char *dest = NULL; + int r = 1; + char *filename = NULL; + + if ((filename = find_file_in_dir(src_path, filepattern)) != NULL) { + if ((r = asprintf(&src, "%s/%s", src_path, filename)) != - 1 && + (r = asprintf(&dest, "%s/%s", dest_path, filename) != -1)) { + // move initscript + r = rename(src, dest); + } else { + errno = ENOMEM; + } + + if (src != NULL) + free(src); + if (dest != NULL) + free(dest); + } else { + if ((filename = find_file_in_dir(dest_path, filepattern)) == NULL) + r = 2; + } + + if (filename != NULL) + free(filename); + + return r; +} + +// Enable/Disable addon service(s) by moving initscript symlink from/to disabled_path +int toggle_service(const char *service, const char *action) { + const char *src_path, *dest_path; + char *filepattern;
Init to NULL.
- int r = 0;
- if (asprintf(&filepattern, "S??%s", service) == -1) {
- errno = ENOMEM;
- return -1;
- }
- if (strcmp(action, "enable") == 0) {
- src_path = disabled_path;
- dest_path = enabled_path;
- } else {
- src_path = enabled_path;
- dest_path = disabled_path;
- }
- // Ensure disabled_path exists
- errno = 0;
- if (mkdir(disabled_path, S_IRWXU + S_IRGRP + S_IXGRP +
S_IROTH
- S_IXOTH) == -1 && errno != EEXIST) {
- r = 1;
- snprintf(errormsg, BUFFER_SIZE -1, "Error creating %s.
(Error: %d)", disabled_path, errno);
- } else {
- r = move_initscript_by_pattern(src_path, dest_path,
filepattern);
- if (r == -1 ) {
- r = 1;
- snprintf(errormsg, BUFFER_SIZE - 1, "Could not %s %s.
(Error: %d)", action, service, errno);
- } else if (r == 1) {
- snprintf(errormsg, BUFFER_SIZE - 1, "Service %s is
already %sd. Skipping...", service, action);
- } else if (r == 2) {
- snprintf(errormsg, BUFFER_SIZE - 1, "Unable to %s
service %s. (Service has no valid symlink in %s).", action, service, src_path);
- }
- }
- free(filepattern);
- return r;
+}
+// Print to stdout wether <service> is enabled or disabled on boot +// Prints <service> as Not available when initscript is not found +// in either enabled_path or disabled_path. +void print_boot_status(char *service) {
- char *filepattern;
NULL. How does this even run without it?
- if (asprintf(&filepattern, "S??%s", service) == -1) {
- errno = ENOMEM;
- return;
- }
- if (find_file_in_dir(enabled_path, filepattern) != NULL)
- fprintf(stdout, "%s is enabled on boot.\n", service);
- else if (find_file_in_dir(disabled_path, filepattern) !=
NULL)
I started doubting this piece of code: find_file_in_dir returns a malloc'ed filename. Which in this case will never be free'd, I think?, at least not by my code. I assume it is better to catch the return value in a char *filename and then free it?
I don’t quite get what you are asking for.
I personally always store stuff in a variable first, and then usually evaluate it. One logical step in the code, per line.
In that case, you will have the pointer, and you can free the space later.
- fprintf(stdout, "%s is disabled on boot.\n", service);
- else
- fprintf(stdout, "%s is not available for boot. (Service
has no valid symlink in either %s or %s).\n", service, enabled_path, disabled_path);
- free(filepattern);
+}
int main(int argc, char *argv[]) {
- char command[BUFFER_SIZE];
- if (!(initsetuid()))
- exit(1);
- if (argc < 3) {
- fprintf(stderr, "\nMissing arguments.\n\naddonctrl
addon (start|stop|restart|reload|enable|disable)\n\n");
- exit(1);
- }
- const char* name = argv[1];
- if (strlen(name) > 32) {
- fprintf(stderr, "\nString to large.\n\naddonctrl addon
(start|stop|restart|reload|enable|disable)\n\n");
- exit(1);
- }
- // Check if the input argument is valid
- if (!is_valid_argument_alnum(name)) {
- fprintf(stderr, "Invalid add-on name: %s\n", name);
- exit(2);
- }
- sprintf(command, "/opt/pakfire/db/installed/meta-%s",
name);
- FILE *fp = fopen(command,"r");
- if ( fp ) {
- fclose(fp);
- } else {
- fprintf(stderr, "\nAddon '%s' not found.\n\naddonctrl
addon (start|stop|restart|reload|status|enable|disable)\n\n", name);
- exit(1);
- }
- if (strcmp(argv[2], "start") == 0) {
- snprintf(command, BUFFER_SIZE - 1,
"/etc/rc.d/init.d/%s start", name);
- safe_system(command);
- } else if (strcmp(argv[2], "stop") == 0) {
- snprintf(command, BUFFER_SIZE - 1,
"/etc/rc.d/init.d/%s stop", name);
- safe_system(command);
- } else if (strcmp(argv[2], "restart") == 0) {
- snprintf(command, BUFFER_SIZE - 1,
"/etc/rc.d/init.d/%s restart", name);
- safe_system(command);
- } else if (strcmp(argv[2], "reload") == 0) {
- snprintf(command, BUFFER_SIZE - 1,
"/etc/rc.d/init.d/%s reload", name);
- safe_system(command);
- } else if (strcmp(argv[2], "status") == 0) {
- snprintf(command, BUFFER_SIZE - 1,
"/etc/rc.d/init.d/%s status", name);
- safe_system(command);
- } else if (strcmp(argv[2], "enable") == 0) {
- snprintf(command, BUFFER_SIZE - 1, "mv -f
/etc/rc.d/rc3.d/off/S??%s /etc/rc.d/rc3.d" , name);
- safe_system(command);
- } else if (strcmp(argv[2], "disable") == 0) {
- snprintf(command, BUFFER_SIZE - 1, "mkdir -p
/etc/rc.d/rc3.d/off");
- safe_system(command);
- snprintf(command, BUFFER_SIZE - 1, "mv -f
/etc/rc.d/rc3.d/S??%s /etc/rc.d/rc3.d/off" , name);
- safe_system(command);
- } else {
- fprintf(stderr, "\nBad argument given.\n\naddonctrl
addon (start|stop|restart|reload|enable|disable)\n\n");
- exit(1);
- }
- return 0;
- char **services = NULL;
- int servicescnt = 0;
- char *addon = argv[1];
- char *action = argv[2];
- char *service_filter = NULL;
- int r = 0;
- if (!(initsetuid()))
- exit(1);
- if (argc < 3) {
- fprintf(stderr, "\nMissing arguments.\n\n%s\n\n", usage);
- exit(1);
- }
- if (argc == 4 && strcmp(action, "list-services") != 0)
- service_filter = argv[3];
- if (strlen(addon) > 32) {
- fprintf(stderr, "\nString too large.\n\n%s\n\n", usage);
- exit(1);
- }
- // Check if the input argument is valid
- if (!is_valid_argument_alnum(addon)) {
- fprintf(stderr, "Invalid add-on name: %s.\n", addon);
- exit(2);
- }
- // Get initscript name(s) from addon metadata
- errno = 0;
There is usually no need to reset this.
I noticed before, during testing, that, since I do an mkdir in toggle_service, but ignore the error if it is EEXIST, that later in the code after using other functions that could produce errno, errno keeps the value EEXIST if those functions just succeed.
Yes, you cannot rely on errno alone. The man pages for all those functions that might set errno will always say something like “will return a non-zero value and set errno accordingly”. So you will always have to check the return value first and find some more detail about the error in errno.
And as get_addon_services could return NULL when there are no services but could also return NULL due to an error with errno set.
Yes, this is kind of a flaw in the system of using errno. I don’t know anything better though.
However when no error occurred and the EEXIST error is still in errno, the code would falsely conclude that services is NULL due to an error. So now for safety, I make sure it is set to 0 before I call get_addon_services.
Better reset in get_addon_services then.
I found this cert advisory write-up recommending always settings errno to 0 before a library call: https://wiki.sei.cmu.edu/confluence/display/c/ERR30- .+Take+care+when+reading+errno
Hmm, that will cause new problems then…
If you get an error from a function, you would want to handle that. And if during that, errno gets reset, you would have to copy it first which will make code more complicated.
Maybe this is because people started to rely on errno? That would be a bad idea, because if you call other functions, they might use and set errno, too.
I'll do it in get_addon_services, right before the library calls that could set errno.
- services = get_addon_services(addon, &servicescnt,
service_filter);
- if (services == NULL || *services == 0) {
- if (errno != 0)
This is not reliable error checking. The function should return a defined value if an error has occurred (e.g. NULL) and an error is detected, that error should be handled.
So in this case:
services = get_addon_services(…); if (!services) { printf(“ERROR: …\”);
}
… Other code, on success…
Now, NULL is not a very good indicator because your problem might be that an error has been found and therefore the function aborted, or that there is actually no services available for this package.
You could handle this by setting errno to something like ENOENT (No such file or directory) and handle that separately:
services = get_addon_services(…); if (!services) { switch (errno) { case ENOENT: DO SOMETHING; Break;
default: printf(“ERROR: …\”); DO SOMETING ELSE }
I'm not sure why it is not reliable?
Reliable in that sense, that the function has the same return value for two different cases.
Here, it might not play the biggest role, but if some other software would call this function (because it was part of some library), the caller might get confused.
After reading above mentioned cert advisory, I think I clearly understand why this code is indeed not reliable :-). At least it is not 'compliant' to that advisory.
I am not sure whether that advise is that practical.
I understand the intention, but I believe it makes more complex code even more complex.
I had thought about setting errno before; but I saw it being strongly discouraged in many discussions on the web; unless you where writing an actual system library.
Really? And what do they suggest using instead?
That varies :-) Most common suggestion is the non-helpful "use another mechanism to report errors"
Some suggest to return an error of type int * or size_t * as an output parameter, or returning a negative value assuming the return type is a signed type eventually with: enum myerrors { ERR_NO_MEMORY = -1, ERR_BAD_ARGS = -2, ERR_CPU_EXPLODED = -3, // and so on }; for clarity.
Then you will make the folks angry who hate enums because they might change in size and what not...
Some even suggest implementing a simulation of the C standard errno mechanism: /* your_errno.c */ __thread int g_your_error_code;
Ah, and of course threads :)
/* your_errno.h */ extern __thread int g_your_error_code #define set_your_errno(err) (g_your_error_code = (err)) #define your_errno (g_your_error_code)
Also some refer to this cert advisory as implicit recommendation to not set errno for non system-library functions, but rather make the function return an errno using return type errno_t: https://wiki.sei.cmu.edu/confluence/display/c/DCL09-C.+Declare+functions+tha...
But I think I will go with an output parameter indicating success or error, and when error the value can indicate whether it is a system error or another error. So depending on the value of that returned error, check errno or display a more suitable error message.
(I tried before to have the function return an int indicating success or error, but I then failed to reliably pass the array back in an output parameter. So in the end I went with returning the array.)
I do not know many libraries which follow this. Maybe because libraries are by definition something that is close to the core system?
For many functions, it is very convenient to just return a pointer (or to return NULL). For example strdup(). It would get really complicated if strdup() would return an integer (or enum stuff) and the duplicated string would be handed over by another pointer.
Maybe this is the reason why people hate C? I am not aware that other languages are doing it that much differently.
So I went for my own 'errormsg'; And I check if errno is set for system errors, if not, check my own errormsg for function errors. If both are clear, services is NULL due to no services. (which could then be due to an invalid filter, or just no services for the addon)
Finally, instead of printing the errno number in the error message, you can use %m (without any argument) which will be automatically converted into something human readable.
Are you sure? As I found this:
The %m directive is not portable, use %s mapped to an argument of strerror(errno) (or a version of strerror_r) instead.
here: https://www.gnu.org/software/gnulib/manual/html_node/asprintf.html
We are only using glibc here.
As far as I know, glibc, musl, all the BSDs support it. What else do we need?
This binary won’t be POSIX compatible, indeed.
Ok. I was under the impression that the document spoke about glibc but as I understand it now, it is about gnulib :-)
I should be able to produce a new version of this patchset by Tuesday evening.
Cool!
-Michael
Regards Robin
- fprintf(stderr, "\nSystem error occured. (Error:
%d)\n\n", errno);
- else if (strcmp(errormsg, "") != 0)
- fprintf(stderr, "\n%s\n\n", errormsg);
- else if (service_filter != NULL)
- fprintf(stderr, "\nNo service '%s' found for addon
'%s'. Use 'list-services' to get a list of available services\n\n%s\n\n", service_filter, addon, usage);
- else
- fprintf(stderr, "\nAddon '%s' has no services.\n\n",
addon);
- exit(1);
- }
- // Handle requested action
- if (strcmp(action, "start") == 0 ||
- strcmp(action, "stop") == 0 ||
- strcmp(action, "restart") == 0 ||
- strcmp(action, "reload") == 0 ||
- strcmp(action, "status") == 0) {
- errno = 0;
- for(int i = 0; i < servicescnt; i++) {
- if (initscript_action(services[i], action) != 0) {
- r = 1;
- if (errno != 0)
- fprintf(stderr, "\nSystem error occured.
(Error: %d)\n\n", errno);
- break;
- }
- }
- } else if (strcmp(action, "enable") == 0 ||
- strcmp(action, "disable") == 0) {
- errno = 0;
- for(int i = 0; i < servicescnt; i++) {
- if (toggle_service(services[i], action) == 0) {
- fprintf(stdout, "%sd service %s\n", action,
services[i]);
If you want to just print something, just use printf.
Ok
- } else if (errno != 0) {
- r = 1;
- fprintf(stderr, "\nSystem error occured. (Error:
%d)\n\n", errno);
- break;
- } else {
- r = 1;
- fprintf(stderr, "\n%s\n\n", errormsg);
- }
- }
- } else if (strcmp(action, "boot-status") == 0) {
- errno = 0;
- for(int i = 0; i < servicescnt; i++) {
- print_boot_status(services[i]);
- if (errno != 0) {
- r = 1;
- fprintf(stderr, "\nSystem error occured. (Error:
%d)\n\n", errno);
- break;
- }
- }
- } else if (strcmp(action, "list-services") == 0) {
- fprintf(stdout, "\nServices for addon %s:\n", addon);
- for(int i = 0; i < servicescnt; i++) {
- fprintf(stdout, " %s\n", services[i]);
- }
- fprintf(stdout, "\n");
- } else {
- fprintf(stderr, "\nBad argument given.\n\n%s\n\n", usage);
- r = 1;
- }
- // Cleanup
- for(int i = 0; i < servicescnt; i++)
- free(services[i]);
- free(services);
- return r;
}
Otherwise, this looks a lot better since the last version :)
Thanks. I'm learning a lot.
Happy to help :)
Robin
-Michael
-- Dit bericht is gescanned op virussen en andere gevaarlijke inhoud door MailScanner en lijkt schoon te zijn.