Message ID | 20220329074440.26214-3-rpalethorpe@suse.com |
---|---|
State | Accepted |
Headers | show |
Series | [v2,1/3] API/cgroup: Add io controller | expand |
Hi Richie, > +++ b/testcases/kernel/controllers/io/io_control01.c > @@ -0,0 +1,155 @@ > +// SPDX-License-Identifier: GPL-2.0 nit: not sure if it was deliberate not adding it, but you may want to add your/SUSE copyright. > +/*\ > + * > + * [Description] > + * > + * Perform some I/O on a file and check if at least some of it is > + * recorded by the I/O controller. > + * > + * The exact amount of I/O performed is dependent on the file system, > + * page cache, scheduler and block driver. We call sync and drop the > + * file's page cache to force reading and writing. We also write > + * random data to try to prevent compression. > + * > + * The pagecache is a particular issue for reading. If the call to > + * fadvise is ignored then the data may only be read from the > + * cache. So that no I/O requests are made. > + */ > + ... > +static int read_io_stats(const char *const line, struct io_stats *const stat) > +{ > + return sscanf(line, > + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", > + &stat->mjr, &stat->mnr, > + &stat->rbytes, &stat->wbytes, &stat->rios, &stat->wios, > + &stat->dbytes, &stat->dios); > +} checkpatch.pl false positive: io_control01.c:40: WARNING: unchecked sscanf return value Obviously perl parsing has some limitations as we check read_io_stats() return value. ... > +static void setup(void) > +{ > + char buf[PATH_MAX] = { 0 }; > + char *path = SAFE_GETCWD(buf, PATH_MAX - sizeof("mnt") - 1); > + struct stat st; > + > + strcpy(path + strlen(path), "/mnt"); > + > + tst_stat_mount_dev(path, &st); > + dev_major = major(st.st_rdev); > + dev_minor = minor(st.st_rdev); > +} > + > +static struct tst_test test = { > + .test_all = run, > + .setup = setup, > + .needs_device = 1, nit: testcases/kernel/controllers/io/io_control01.c: useless tag: needs_device Kind regards, Petr > + .mntpoint = "mnt", > + .mount_device = 1, > + .all_filesystems = 1, > + .skip_filesystems = (const char *const[]){ "ntfs", "tmpfs", NULL }, > + .needs_cgroup_ver = TST_CG_V2, > + .needs_cgroup_ctrls = (const char *const[]){ "io", NULL }, > +};
Hi Richie, LGTM, just please remove needs_device as not needed. Reviewed-by: Petr Vorel <pvorel@suse.cz> > +static void setup(void) > +{ > + char buf[PATH_MAX] = { 0 }; > + char *path = SAFE_GETCWD(buf, PATH_MAX - sizeof("mnt") - 1); > + struct stat st; > + > + strcpy(path + strlen(path), "/mnt"); > + > + tst_stat_mount_dev(path, &st); > + dev_major = major(st.st_rdev); > + dev_minor = minor(st.st_rdev); If we ever support test setup run inside do_setup() (i.e. setup function which would be run just once - before for_each_variant(), for_each_filesystem() and fork_testrun()), this test would have use case for it (I suppose more tests which use .all_filesystems). Because tst_stat_mount_dev() could be run just for first filesystem. But that's not related to the testcase. Kind regards, Petr > +} > + > +static struct tst_test test = { > + .test_all = run, > + .setup = setup, > + .needs_device = 1, > + .mntpoint = "mnt", > + .mount_device = 1, > + .all_filesystems = 1, > + .skip_filesystems = (const char *const[]){ "ntfs", "tmpfs", NULL }, > + .needs_cgroup_ver = TST_CG_V2, > + .needs_cgroup_ctrls = (const char *const[]){ "io", NULL }, > +};
Hello Petr, Petr Vorel <pvorel@suse.cz> writes: > Hi Richie, > >> +++ b/testcases/kernel/controllers/io/io_control01.c >> @@ -0,0 +1,155 @@ >> +// SPDX-License-Identifier: GPL-2.0 > nit: not sure if it was deliberate not adding it, but you may want to > add your/SUSE copyright. > >> +/*\ >> + * >> + * [Description] >> + * >> + * Perform some I/O on a file and check if at least some of it is >> + * recorded by the I/O controller. >> + * >> + * The exact amount of I/O performed is dependent on the file system, >> + * page cache, scheduler and block driver. We call sync and drop the >> + * file's page cache to force reading and writing. We also write >> + * random data to try to prevent compression. >> + * >> + * The pagecache is a particular issue for reading. If the call to >> + * fadvise is ignored then the data may only be read from the >> + * cache. So that no I/O requests are made. >> + */ >> + > ... >> +static int read_io_stats(const char *const line, struct io_stats *const stat) >> +{ >> + return sscanf(line, >> + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", >> + &stat->mjr, &stat->mnr, >> + &stat->rbytes, &stat->wbytes, &stat->rios, &stat->wios, >> + &stat->dbytes, &stat->dios); >> +} > checkpatch.pl false positive: > io_control01.c:40: WARNING: unchecked sscanf return value > Obviously perl parsing has some limitations as we check read_io_stats() return > value. I'm not sure what to do about this other than switch to a macro which is a bit silly. IMO sscanf should have the warn_unused_result attribute and this should be inherited by read_io_stats. All of which is better handled by the compiler. > > ... >> +static void setup(void) >> +{ >> + char buf[PATH_MAX] = { 0 }; >> + char *path = SAFE_GETCWD(buf, PATH_MAX - sizeof("mnt") - 1); >> + struct stat st; >> + >> + strcpy(path + strlen(path), "/mnt"); >> + >> + tst_stat_mount_dev(path, &st); >> + dev_major = major(st.st_rdev); >> + dev_minor = minor(st.st_rdev); >> +} >> + >> +static struct tst_test test = { >> + .test_all = run, >> + .setup = setup, >> + .needs_device = 1, > nit: testcases/kernel/controllers/io/io_control01.c: useless tag: needs_device Pushed with fixes (including needs_device).
diff --git a/runtest/controllers b/runtest/controllers index 3108a2561..22d482050 100644 --- a/runtest/controllers +++ b/runtest/controllers @@ -360,6 +360,9 @@ cpuset_regression_test cpuset_regression_test.sh cgroup_xattr cgroup_xattr +# V2 IO controller (was blkio) +io_control01 io_control01 + pids_1_1 pids.sh 1 1 0 pids_1_2 pids.sh 1 2 0 pids_1_10 pids.sh 1 10 0 diff --git a/testcases/kernel/controllers/io/.gitignore b/testcases/kernel/controllers/io/.gitignore new file mode 100644 index 000000000..d626fa80d --- /dev/null +++ b/testcases/kernel/controllers/io/.gitignore @@ -0,0 +1 @@ +io_control01 diff --git a/testcases/kernel/controllers/io/Makefile b/testcases/kernel/controllers/io/Makefile new file mode 100644 index 000000000..5ea7d67db --- /dev/null +++ b/testcases/kernel/controllers/io/Makefile @@ -0,0 +1,6 @@ +# SPDX-License-Identifier: GPL-2.0-or-later + +top_srcdir ?= ../../../.. + +include $(top_srcdir)/include/mk/testcases.mk +include $(top_srcdir)/include/mk/generic_leaf_target.mk diff --git a/testcases/kernel/controllers/io/io_control01.c b/testcases/kernel/controllers/io/io_control01.c new file mode 100644 index 000000000..a6bf01639 --- /dev/null +++ b/testcases/kernel/controllers/io/io_control01.c @@ -0,0 +1,155 @@ +// SPDX-License-Identifier: GPL-2.0 +/*\ + * + * [Description] + * + * Perform some I/O on a file and check if at least some of it is + * recorded by the I/O controller. + * + * The exact amount of I/O performed is dependent on the file system, + * page cache, scheduler and block driver. We call sync and drop the + * file's page cache to force reading and writing. We also write + * random data to try to prevent compression. + * + * The pagecache is a particular issue for reading. If the call to + * fadvise is ignored then the data may only be read from the + * cache. So that no I/O requests are made. + */ + +#include <stdio.h> +#include <stdlib.h> +#include <sys/sysmacros.h> + +#include "tst_test.h" + +struct io_stats { + unsigned int mjr; + unsigned int mnr; + unsigned long rbytes; + unsigned long wbytes; + unsigned long rios; + unsigned long wios; + unsigned long dbytes; + unsigned long dios; +}; + +static unsigned int dev_major, dev_minor; + +static int read_io_stats(const char *const line, struct io_stats *const stat) +{ + return sscanf(line, + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", + &stat->mjr, &stat->mnr, + &stat->rbytes, &stat->wbytes, &stat->rios, &stat->wios, + &stat->dbytes, &stat->dios); +} + +static void run(void) +{ + int i, fd; + char *line, *buf_ptr; + const size_t pgsz = SAFE_SYSCONF(_SC_PAGESIZE); + char *buf = SAFE_MALLOC(MAX((size_t)BUFSIZ, pgsz)); + struct io_stats start; + + SAFE_CG_READ(tst_cg, "io.stat", buf, BUFSIZ - 1); + line = strtok_r(buf, "\n", &buf_ptr); + while (line) { + const int convs = read_io_stats(line, &start); + + if (convs < 2) + continue; + + tst_res(TINFO, "Found %u:%u in io.stat", dev_major, dev_minor); + + if (start.mjr == dev_major || start.mnr == dev_minor) + break; + + line = strtok_r(NULL, "\n", &buf_ptr); + } + + SAFE_CG_PRINTF(tst_cg, "cgroup.procs", "%d", getpid()); + + fd = SAFE_OPEN("/dev/urandom", O_RDONLY, 0600); + SAFE_READ(1, fd, buf, pgsz); + SAFE_CLOSE(fd); + + fd = SAFE_OPEN("mnt/dat", O_WRONLY | O_CREAT, 0600); + + for (i = 0; i < 4; i++) { + SAFE_WRITE(1, fd, buf, pgsz); + SAFE_FSYNC(fd); + TST_EXP_PASS_SILENT(posix_fadvise(fd, pgsz * i, pgsz, POSIX_FADV_DONTNEED)); + } + + SAFE_CLOSE(fd); + fd = SAFE_OPEN("mnt/dat", O_RDONLY, 0600); + + for (i = 0; i < 4; i++) + SAFE_READ(1, fd, buf, pgsz); + + tst_res(TPASS, "Did some IO in the IO controller"); + + SAFE_CG_READ(tst_cg, "io.stat", buf, BUFSIZ - 1); + line = strtok_r(buf, "\n", &buf_ptr); + while (line) { + struct io_stats end; + const int convs = read_io_stats(line, &end); + + if (convs < 8) + break; + + if (end.mjr != dev_major || end.mnr != dev_minor) { + line = strtok_r(NULL, "\n", &buf_ptr); + continue; + } + + tst_res(TPASS, "Found %u:%u in io.stat", dev_major, dev_minor); + TST_EXP_EXPR(end.rbytes > start.rbytes, + "(rbytes=%lu) > (st_rbytes=%lu)", + end.rbytes, start.rbytes); + TST_EXP_EXPR(end.wbytes > start.wbytes, + "(wbytes=%lu) > (st_wbytes=%lu)", + end.wbytes, start.wbytes); + TST_EXP_EXPR(end.rios > start.rios, + "(rios=%lu) > (st_rios=%lu)", + end.rios, start.rios); + TST_EXP_EXPR(end.wios > start.wios, + "(wios=%lu) > (st_wios=%lu)", + end.wios, start.wios); + + goto out; + } + + tst_res(TINFO, "io.stat:\n%s", buf); + tst_res(TFAIL, "Did not find %u:%u in io.stat", dev_major, dev_minor); +out: + free(buf); + SAFE_CLOSE(fd); + SAFE_UNLINK("mnt/dat"); +} + +static void setup(void) +{ + char buf[PATH_MAX] = { 0 }; + char *path = SAFE_GETCWD(buf, PATH_MAX - sizeof("mnt") - 1); + struct stat st; + + strcpy(path + strlen(path), "/mnt"); + + tst_stat_mount_dev(path, &st); + dev_major = major(st.st_rdev); + dev_minor = minor(st.st_rdev); +} + +static struct tst_test test = { + .test_all = run, + .setup = setup, + .needs_device = 1, + .mntpoint = "mnt", + .mount_device = 1, + .all_filesystems = 1, + .skip_filesystems = (const char *const[]){ "ntfs", "tmpfs", NULL }, + .needs_cgroup_ver = TST_CG_V2, + .needs_cgroup_ctrls = (const char *const[]){ "io", NULL }, +};
In V1 there is the blkio controller. This was renamed to just io on V2. The interface and functionality is significantly different. Presently there do not appear to be any tests for the V2 controller. Note that one can not simply stat a file on BTRFS to find the actual block device the filesystem is using. Nor can you read /proc/self/mountinfo. BTRFS seems to generate "anonymous" devices (e.g. 0:27) and this is what is reported by stat. These however are invisible to the IO controller. So instead we have to look in /proc/mounts for the device path then stat the special (/dev/<device>) file to get the actual major and minor device number. Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com> --- runtest/controllers | 3 + testcases/kernel/controllers/io/.gitignore | 1 + testcases/kernel/controllers/io/Makefile | 6 + .../kernel/controllers/io/io_control01.c | 155 ++++++++++++++++++ 4 files changed, 165 insertions(+) create mode 100644 testcases/kernel/controllers/io/.gitignore create mode 100644 testcases/kernel/controllers/io/Makefile create mode 100644 testcases/kernel/controllers/io/io_control01.c