From: Michael Jeanson Date: Wed, 29 May 2024 19:02:15 +0000 (-0400) Subject: Warn and return on fd overflow fdt X-Git-Url: https://git.lttng.org./?a=commitdiff_plain;h=116e2b72141ca18cdf8c875aac4631f48821b016;p=lttng-modules.git Warn and return on fd overflow fdt The fdt should only grow and iterate_fd() holds file_lock, which should ensure the fdt does not change while the lock is taken but be cautious and check anyway. Change-Id: Icd6a3263026734cbe3f296f6087f79add4148a8f Signed-off-by: Michael Jeanson Signed-off-by: Mathieu Desnoyers --- diff --git a/src/lttng-statedump-impl.c b/src/lttng-statedump-impl.c index e06649e7..5ea790f5 100644 --- a/src/lttng-statedump-impl.c +++ b/src/lttng-statedump-impl.c @@ -435,17 +435,22 @@ int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd) */ flags &= ~FMODE_NONOTIFY; fdt = files_fdtable(ctx->files); + /* - * We need to check here again whether fd is within the fdt - * max_fds range, because we might be seeing a different - * files_fdtable() than iterate_fd(), assuming only RCU is - * protecting the read. In reality, iterate_fd() holds - * file_lock, which should ensure the fdt does not change while - * the lock is taken, but we are not aware whether this is - * guaranteed or not, so play safe. + * The fdt should only grow and iterate_fd() holds file_lock, which + * should ensure the fdt does not change while the lock is taken but be + * cautious and check anyway. */ - if (fd < fdt->max_fds && lttng_close_on_exec(fd, ctx->files)) + if (WARN_ON_ONCE(fd >= fdt->max_fds)) + return 0; + + if (lttng_close_on_exec(fd, ctx->files)) flags |= O_CLOEXEC; + + /* + * If d_path() failed to get a full path for the file, use the dentry + * name instead to at least get a filename. + */ if (IS_ERR(s)) { struct dentry *dentry = file->f_path.dentry;