| Message ID | 20220315134122.22912-2-rpalethorpe@suse.com |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | [1/2] API/cgroup: Add io controller | expand |
Hi! > 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 | 162 ++++++++++++++++++ > 4 files changed, 172 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 > > 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..706e67c89 > --- /dev/null > +++ b/testcases/kernel/controllers/io/io_control01.c > @@ -0,0 +1,162 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* [Description] This is not the correct docparse hearder. > + * 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 <mntent.h> > + > +#include "tst_test.h" > + > +static unsigned int dev_major, dev_minor; > + > +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)); > + unsigned long st_rbytes = 0, st_wbytes = 0, st_rios = 0, st_wios = 0; > + > + SAFE_CG_READ(tst_cg, "io.stat", buf, BUFSIZ - 1); > + line = strtok_r(buf, "\n", &buf_ptr); > + while (line) { > + unsigned int mjr, mnr; > + unsigned long dbytes, dios; > + > + const int convs = > + sscanf(line, > + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", > + &mjr, &mnr, &st_rbytes, &st_wbytes, &st_rios, &st_wios, > + &dbytes, &dios); > + > + if (convs < 2) > + continue; > + > + tst_res(TINFO, "Found %u:%u in io.stat", dev_major, dev_minor); > + > + if (mjr == dev_major || 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) { > + unsigned int mjr, mnr; > + unsigned long rbytes, wbytes, rios, wios, dbytes, dios; > + > + const int convs = > + sscanf(line, > + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", > + &mjr, &mnr, &rbytes, &wbytes, &rios, &wios, > + &dbytes, &dios); > + > + if (convs < 8) > + break; > + > + if (mjr != dev_major || 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(rbytes > st_rbytes, "(rbytes=%lu) > (st_rbytes=%lu)", rbytes, st_rbytes); > + TST_EXP_EXPR(wbytes > st_wbytes, "(wbytes=%lu) > (st_wbytes=%lu)", wbytes, st_wbytes); > + TST_EXP_EXPR(rios > st_rios, "(rios=%lu) > (st_rios=%lu)", rios, st_rios); > + TST_EXP_EXPR(wios > st_wios, "(wios=%lu) > (st_wios=%lu)", wios, st_wios); So we only test here that the counters are updated, that sounds fine for a simple test. Do you plan to try anything for io.max? Maybe something as basic as running two concurent processes with very different limits and checking that the more limited process transfer less bytes per unit of time? > + goto out; > + } We do have two very similar copies of this parsing code, maybe we should put that into a function, and pack the parameters into a structure o avoid copy&paste like this. e.g. struct iostats { unsigned long st_rbytes; unsigned long st_wbytes; unsigned long st_rios; unsigned long st_wios; }; static int read_iostats(const char *stats, unsigned int dev_min, unsigned int dev_maj, struct iostats *iostats); > + 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 mntent *mnt; > + FILE *mntf = setmntent("/proc/self/mounts", "r"); > + struct stat st; > + > + strcpy(path + strlen(path), "/mnt"); > + > + if (!mntf) { > + tst_brk(TBROK | TERRNO, "Can't open /proc/self/mounts"); > + return; > + } > + > + mnt = getmntent(mntf); > + if (!mnt) { > + tst_brk(TBROK | TERRNO, "Can't read mounts or no mounts?"); > + return; > + } > + > + do { > + if (strcmp(mnt->mnt_dir, path)) > + continue; > + > + SAFE_STAT(mnt->mnt_fsname, &st); > + dev_major = major(st.st_rdev); > + dev_minor = minor(st.st_rdev); > + > + return; > + > + } while ((mnt = getmntent(mntf))); I guess that this should probably go to the test library. We already have tst_find_backding_dev() in there which is doing something a bit similar. Looking at the code what we do here is to translate a mountpoint into a device so it may be something as: int tst_find_dev_by_mntpoint() > + tst_brk(TBROK, "Could not find mount device"); > +} > + > +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 }, > +}; > -- > 2.35.1 > > > -- > Mailing list info: https://lists.linux.it/listinfo/ltp
Hello, >> + >> + tst_res(TPASS, "Found %u:%u in io.stat", dev_major, dev_minor); >> + TST_EXP_EXPR(rbytes > st_rbytes, "(rbytes=%lu) > (st_rbytes=%lu)", rbytes, st_rbytes); >> + TST_EXP_EXPR(wbytes > st_wbytes, "(wbytes=%lu) > (st_wbytes=%lu)", wbytes, st_wbytes); >> + TST_EXP_EXPR(rios > st_rios, "(rios=%lu) > (st_rios=%lu)", rios, st_rios); >> + TST_EXP_EXPR(wios > st_wios, "(wios=%lu) > (st_wios=%lu)", wios, st_wios); > > So we only test here that the counters are updated, that sounds fine for > a simple test. > > Do you plan to try anything for io.max? Maybe something as basic as > running two concurent processes with very different limits and checking > that the more limited process transfer less bytes per unit of time? Yes, although maybe not immediately. In the case of io.max this should not even require comparing concurrent processes as the limits are simply the absolute number of bytes or iops performed per second by a particular group. So we can just set them to a very low value and see that reading or writing takes longer than when unlimited. io.cost.qos is more complicated because limits are a proportion of the overall sibling activity. > >> + goto out; >> + } > > We do have two very similar copies of this parsing code, maybe we should > put that into a function, and pack the parameters into a structure o > avoid copy&paste like this. e.g. > > struct iostats { > unsigned long st_rbytes; > unsigned long st_wbytes; > unsigned long st_rios; > unsigned long st_wios; > }; > > static int read_iostats(const char *stats, > unsigned int dev_min, unsigned int dev_maj, > struct iostats *iostats); > +1 > >> + 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 mntent *mnt; >> + FILE *mntf = setmntent("/proc/self/mounts", "r"); >> + struct stat st; >> + >> + strcpy(path + strlen(path), "/mnt"); >> + >> + if (!mntf) { >> + tst_brk(TBROK | TERRNO, "Can't open /proc/self/mounts"); >> + return; >> + } >> + >> + mnt = getmntent(mntf); >> + if (!mnt) { >> + tst_brk(TBROK | TERRNO, "Can't read mounts or no mounts?"); >> + return; >> + } >> + >> + do { >> + if (strcmp(mnt->mnt_dir, path)) >> + continue; >> + >> + SAFE_STAT(mnt->mnt_fsname, &st); >> + dev_major = major(st.st_rdev); >> + dev_minor = minor(st.st_rdev); >> + >> + return; >> + >> + } while ((mnt = getmntent(mntf))); > > I guess that this should probably go to the test library. We already > have tst_find_backding_dev() in there which is doing something a bit > similar. Looking at the code what we do here is to translate a > mountpoint into a device so it may be something as: > > int tst_find_dev_by_mntpoint() OK, I wasn't sure if to replace tst_find_backing_dev() which doesn't do what the name suggests to me. The actual device it returns will be some anonymous virtual block device in many cases. If it is btrfs, then the "backing device" is probably a real device not an anonymous device created for the subpartion. OTOH maybe this is what was intended. > >> + tst_brk(TBROK, "Could not find mount device"); >> +} >> + >> +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 }, >> +}; >> -- >> 2.35.1 >> >> >> -- >> Mailing list info: https://lists.linux.it/listinfo/ltp
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..706e67c89 --- /dev/null +++ b/testcases/kernel/controllers/io/io_control01.c @@ -0,0 +1,162 @@ +// 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 <mntent.h> + +#include "tst_test.h" + +static unsigned int dev_major, dev_minor; + +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)); + unsigned long st_rbytes = 0, st_wbytes = 0, st_rios = 0, st_wios = 0; + + SAFE_CG_READ(tst_cg, "io.stat", buf, BUFSIZ - 1); + line = strtok_r(buf, "\n", &buf_ptr); + while (line) { + unsigned int mjr, mnr; + unsigned long dbytes, dios; + + const int convs = + sscanf(line, + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", + &mjr, &mnr, &st_rbytes, &st_wbytes, &st_rios, &st_wios, + &dbytes, &dios); + + if (convs < 2) + continue; + + tst_res(TINFO, "Found %u:%u in io.stat", dev_major, dev_minor); + + if (mjr == dev_major || 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) { + unsigned int mjr, mnr; + unsigned long rbytes, wbytes, rios, wios, dbytes, dios; + + const int convs = + sscanf(line, + "%u:%u rbytes=%lu wbytes=%lu rios=%lu wios=%lu dbytes=%lu dios=%lu", + &mjr, &mnr, &rbytes, &wbytes, &rios, &wios, + &dbytes, &dios); + + if (convs < 8) + break; + + if (mjr != dev_major || 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(rbytes > st_rbytes, "(rbytes=%lu) > (st_rbytes=%lu)", rbytes, st_rbytes); + TST_EXP_EXPR(wbytes > st_wbytes, "(wbytes=%lu) > (st_wbytes=%lu)", wbytes, st_wbytes); + TST_EXP_EXPR(rios > st_rios, "(rios=%lu) > (st_rios=%lu)", rios, st_rios); + TST_EXP_EXPR(wios > st_wios, "(wios=%lu) > (st_wios=%lu)", wios, st_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 mntent *mnt; + FILE *mntf = setmntent("/proc/self/mounts", "r"); + struct stat st; + + strcpy(path + strlen(path), "/mnt"); + + if (!mntf) { + tst_brk(TBROK | TERRNO, "Can't open /proc/self/mounts"); + return; + } + + mnt = getmntent(mntf); + if (!mnt) { + tst_brk(TBROK | TERRNO, "Can't read mounts or no mounts?"); + return; + } + + do { + if (strcmp(mnt->mnt_dir, path)) + continue; + + SAFE_STAT(mnt->mnt_fsname, &st); + dev_major = major(st.st_rdev); + dev_minor = minor(st.st_rdev); + + return; + + } while ((mnt = getmntent(mntf))); + + tst_brk(TBROK, "Could not find mount device"); +} + +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 | 162 ++++++++++++++++++ 4 files changed, 172 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