From: Martin Doucha Date: Tue, 27 Apr 2021 16:14:01 +0200 Subject: [LTP] [PATCH 3/4] RFC: Add rtnetlink helper library In-Reply-To: References: <20210426111918.4304-1-mdoucha@suse.cz> <20210426111918.4304-4-mdoucha@suse.cz> Message-ID: List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ltp@lists.linux.it On 27. 04. 21 15:41, Cyril Hrubis wrote: > Hi! >> +static int tst_rtnl_grow_buffer(const char *file, const int lineno, >> + struct tst_rtnl_context *ctx, size_t size) >> +{ >> + size_t needed, offset, curlen = NLMSG_ALIGN(ctx->datalen); >> + char *buf; >> + >> + if (ctx->bufsize - curlen >= size) >> + return 1; >> + >> + needed = size - (ctx->bufsize - curlen); >> + size = ctx->bufsize + (ctx->bufsize > needed ? ctx->bufsize : needed); >> + size = NLMSG_ALIGN(size); >> + buf = safe_realloc(file, lineno, ctx->buffer, size); >> + >> + if (!buf) >> + return 0; > > You are calling safe_realloc() here yet you check the return value. And > it's the same for safe_malloc(), safe_bind(), safe_socket() and a few > more in the code. > > So either we get rid of the error checks and of the error > propagation or we avoid using safe_ variants. Both rtnetlink and netdevice management functions will be called in cleanup() so I have to assume that the safe functions will only print a warning instead of terminating the program. But I still want to use the error reporting code in the safe functions. > Other than that the code looks sane but it's hard to review the API > without an example that would excersize it. What about adding something > simple in newlib_tests? There are short examples for both rtnetlink and netdevice management in the cover letter. The netdevice library itself is also a detailed example for the rtnetlink API. The final patchset will include the SADDNS CVE test which will use most of the netdevice management functions in setup(). >> +void tst_rtnl_free_context(const char *file, const int lineno, >> + struct tst_rtnl_context *ctx) > > This should be probably called destroy_context() but that is very minor. OK, why not. >> +int tst_rtnl_wait(struct tst_rtnl_context *ctx) >> +{ >> + fd_set fdlist; >> + struct timeval timeout = {0}; >> + >> + FD_ZERO(&fdlist); >> + FD_SET(ctx->socket, &fdlist); >> + timeout.tv_sec = 1; >> + >> + return select(ctx->socket + 1, &fdlist, NULL, NULL, &timeout); > > I find the poll() syscall to have a bit saner API than this. I'll have a look at it. -- Martin Doucha mdoucha@suse.cz QA Engineer for Software Maintenance SUSE LINUX, s.r.o. CORSO IIa Krizikova 148/34 186 00 Prague 8 Czech Republic