proxmox-mirrors/mirror_lxc
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_lxc synced 2025-08-09 12:43:43 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #3937 from brauner/2021-08-13.fixes

Browse Source 
cgroups: cpumask fixes
This commit is contained in:
Stéphane Graber 2021-08-14 11:12:27 -04:00 committed by GitHub
commit 6f76b9e528
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 71 additions and 104 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

175
src/lxc/cgroups/cgfsng.c
View File

@ -208,49 +208,68 @@ static bool is_set(unsigned bit, uint32_t *bitarr)
* *
* 1 0 1 1 * 1 0 1 1
*/ */
static uint32_t *lxc_cpumask(char *buf, size_t nbits) static int lxc_cpumask(char *buf, uint32_t **bitarr, size_t *last_set_bit)
{ {
__do_free uint32_t *bitarr = NULL; __do_free uint32_t *arr_u32 = NULL;
size_t cur_last_set_bit = 0, nbits = 256;
size_t nr_u32;
char *token; char *token;
size_t arrlen;
arrlen = BITS_TO_LONGS(nbits); nr_u32 = BITS_TO_LONGS(nbits);
bitarr = zalloc(arrlen * sizeof(uint32_t)); arr_u32 = zalloc(nr_u32 * sizeof(uint32_t));
if (!bitarr) if (!arr_u32)
return ret_set_errno(NULL, ENOMEM); return ret_errno(ENOMEM);
lxc_iterate_parts(token, buf, ",") { lxc_iterate_parts(token, buf, ",") {
errno = 0; unsigned last_bit, first_bit;
unsigned end, start;
char *range; char *range;
start = strtoul(token, NULL, 0); errno = 0;
end = start; first_bit = strtoul(token, NULL, 0);
last_bit = first_bit;
range = strchr(token, '-'); range = strchr(token, '-');
if (range) if (range)
end = strtoul(range + 1, NULL, 0); last_bit = strtoul(range + 1, NULL, 0);
if (!(start <= end)) if (!(first_bit <= last_bit))
return ret_set_errno(NULL, EINVAL); return ret_errno(EINVAL);
if (end >= nbits) if (last_bit >= nbits) {
return ret_set_errno(NULL, EINVAL); size_t add_bits = last_bit - nbits + 32;
size_t new_nr_u32;
uint32_t *p;
while (start <= end) new_nr_u32 = BITS_TO_LONGS(nbits + add_bits);
set_bit(start++, bitarr); p = realloc(arr_u32, new_nr_u32 * sizeof(uint32_t));
if (!p)
return ret_errno(ENOMEM);
arr_u32 = move_ptr(p);
memset(arr_u32 + nr_u32, 0,
(new_nr_u32 - nr_u32) * sizeof(uint32_t));
nbits += add_bits;
}
while (first_bit <= last_bit)
set_bit(first_bit++, arr_u32);
if (last_bit > cur_last_set_bit)
cur_last_set_bit = last_bit;
} }
return move_ptr(bitarr); *last_set_bit = cur_last_set_bit;
*bitarr = move_ptr(arr_u32);
return 0;
} }
/* Turn cpumask into simple, comma-separated cpulist. */ /* Turn cpumask into simple, comma-separated cpulist. */
static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t last_set_bit)
{ {
__do_free_string_list char **cpulist = NULL; __do_free_string_list char **cpulist = NULL;
char numstr[INTTYPE_TO_STRLEN(size_t)] = {0}; char numstr[INTTYPE_TO_STRLEN(size_t)] = {0};
int ret; int ret;
for (size_t i = 0; i <= nbits; i++) { for (size_t i = 0; i <= last_set_bit; i++) {
if (!is_set(i, bitarr)) if (!is_set(i, bitarr))
continue; continue;
@ -269,37 +288,6 @@ static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits)
return lxc_string_join(",", (const char **)cpulist, false); return lxc_string_join(",", (const char **)cpulist, false);
} }
static ssize_t get_max_cpus(char *cpulist)
{
char *c1, *c2;
char *maxcpus = cpulist;
size_t cpus = 0;
c1 = strrchr(maxcpus, ',');
if (c1)
c1++;
c2 = strrchr(maxcpus, '-');
if (c2)
c2++;
if (!c1 && !c2)
c1 = maxcpus;
else if (c1 > c2)
c2 = c1;
else if (c1 < c2)
c1 = c2;
else if (!c1 && c2)
c1 = c2;
errno = 0;
cpus = strtoul(c1, NULL, 0);
if (errno != 0)
return -1;
return cpus;
}
static inline bool is_unified_hierarchy(const struct hierarchy *h) static inline bool is_unified_hierarchy(const struct hierarchy *h)
{ {
return h->fs_type == UNIFIED_HIERARCHY; return h->fs_type == UNIFIED_HIERARCHY;
@ -569,33 +557,21 @@ static bool cpuset1_cpus_initialize(int dfd_parent, int dfd_child,
__do_free uint32_t *isolmask = NULL, *offlinemask = NULL, __do_free uint32_t *isolmask = NULL, *offlinemask = NULL,
*possmask = NULL; *possmask = NULL;
int ret; int ret;
ssize_t i; size_t isol_last_set_bit = 0, offline_last_set_bit = 0,
ssize_t maxisol = 0, maxoffline = 0, maxposs = 0; poss_last_set_bit = 0;
bool flipped_bit = false; bool flipped_bit = false;
posscpus = read_file_at(dfd_parent, "cpuset.cpus", PROTECT_OPEN, 0); posscpus = read_file_at(dfd_parent, "cpuset.cpus", PROTECT_OPEN, 0);
if (!posscpus) if (!posscpus)
return log_error_errno(false, errno, "Failed to read file \"%s\"", fpath); return log_error_errno(false, errno, "Failed to read file \"%s\"", fpath);
/* Get maximum number of cpus found in possible cpuset. */
maxposs = get_max_cpus(posscpus);
if (maxposs < 0 || maxposs >= INT_MAX - 1)
return false;
if (file_exists(__ISOL_CPUS)) { if (file_exists(__ISOL_CPUS)) {
isolcpus = read_file_at(-EBADF, __ISOL_CPUS, PROTECT_OPEN, 0); isolcpus = read_file_at(-EBADF, __ISOL_CPUS, PROTECT_OPEN, 0);
if (!isolcpus) if (!isolcpus)
return log_error_errno(false, errno, "Failed to read file \"%s\"", __ISOL_CPUS); return log_error_errno(false, errno, "Failed to read file \"%s\"", __ISOL_CPUS);
if (isdigit(isolcpus[0])) { if (!isdigit(isolcpus[0]))
/* Get maximum number of cpus found in isolated cpuset. */ free_disarm(isolcpus);
maxisol = get_max_cpus(isolcpus);
if (maxisol < 0 || maxisol >= INT_MAX - 1)
return false;
}
if (maxposs < maxisol)
maxposs = maxisol;
} else { } else {
TRACE("The path \""__ISOL_CPUS"\" to read isolated cpus from does not exist"); TRACE("The path \""__ISOL_CPUS"\" to read isolated cpus from does not exist");
} }
@ -605,52 +581,49 @@ static bool cpuset1_cpus_initialize(int dfd_parent, int dfd_child,
if (!offlinecpus) if (!offlinecpus)
return log_error_errno(false, errno, "Failed to read file \"%s\"", __OFFLINE_CPUS); return log_error_errno(false, errno, "Failed to read file \"%s\"", __OFFLINE_CPUS);
if (isdigit(offlinecpus[0])) { if (!isdigit(offlinecpus[0]))
/* Get maximum number of cpus found in offline cpuset. */ free_disarm(offlinecpus);
maxoffline = get_max_cpus(offlinecpus);
if (maxoffline < 0 || maxoffline >= INT_MAX - 1)
return false;
}
if (maxposs < maxoffline)
maxposs = maxoffline;
} else { } else {
TRACE("The path \""__OFFLINE_CPUS"\" to read offline cpus from does not exist"); TRACE("The path \""__OFFLINE_CPUS"\" to read offline cpus from does not exist");
} }
if ((maxisol == 0) && (maxoffline == 0)) { if (!isolcpus && !offlinecpus) {
cpulist = move_ptr(posscpus); cpulist = move_ptr(posscpus);
goto copy_parent; goto copy_parent;
} }
possmask = lxc_cpumask(posscpus, maxposs); ret = lxc_cpumask(posscpus, &possmask, &poss_last_set_bit);
if (!possmask) if (ret)
return log_error_errno(false, errno, "Failed to create cpumask for possible cpus"); return log_error_errno(false, errno, "Failed to create cpumask for possible cpus");
if (maxisol > 0) { if (isolcpus) {
isolmask = lxc_cpumask(isolcpus, maxposs); ret = lxc_cpumask(isolcpus, &isolmask, &isol_last_set_bit);
if (!isolmask) if (ret)
return log_error_errno(false, errno, "Failed to create cpumask for isolated cpus"); return log_error_errno(false, errno, "Failed to create cpumask for isolated cpus");
} }
if (maxoffline > 0) { if (offlinecpus > 0) {
offlinemask = lxc_cpumask(offlinecpus, maxposs); ret = lxc_cpumask(offlinecpus, &offlinemask, &offline_last_set_bit);
if (!offlinemask) if (ret)
return log_error_errno(false, errno, "Failed to create cpumask for offline cpus"); return log_error_errno(false, errno, "Failed to create cpumask for offline cpus");
} }
for (i = 0; i <= maxposs; i++) { for (size_t i = 0; i <= poss_last_set_bit; i++) {
if ((isolmask && !is_set(i, isolmask)) || if (isolmask && (i <= isol_last_set_bit) &&
(offlinemask && !is_set(i, offlinemask)) || is_set(i, isolmask) && is_set(i, possmask)) {
!is_set(i, possmask)) flipped_bit = true;
continue; clear_bit(i, possmask);
}
flipped_bit = true; if (offlinemask && (i <= offline_last_set_bit) &&
clear_bit(i, possmask); is_set(i, offlinemask) && is_set(i, possmask)) {
flipped_bit = true;
clear_bit(i, possmask);
}
} }
if (!flipped_bit) { if (!flipped_bit) {
cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); cpulist = lxc_cpumask_to_cpulist(possmask, poss_last_set_bit);
TRACE("No isolated or offline cpus present in cpuset"); TRACE("No isolated or offline cpus present in cpuset");
} else { } else {
cpulist = move_ptr(posscpus); cpulist = move_ptr(posscpus);
@ -677,18 +650,12 @@ static bool cpuset1_initialize(int dfd_base, int dfd_next)
ssize_t bytes; ssize_t bytes;
char v; char v;
/* /* Determine whether the base cgroup has cpuset inheritance turned on. */
* Determine whether the base cgroup has cpuset
* inheritance turned on.
*/
bytes = lxc_readat(dfd_base, "cgroup.clone_children", &v, 1); bytes = lxc_readat(dfd_base, "cgroup.clone_children", &v, 1);
if (bytes < 0) if (bytes < 0)
return syserror_ret(false, "Failed to read file %d(cgroup.clone_children)", dfd_base); return syserror_ret(false, "Failed to read file %d(cgroup.clone_children)", dfd_base);
/* /* Initialize cpuset.cpus removing any isolated and offline cpus. */
* Initialize cpuset.cpus and make remove any isolated
* and offline cpus.
*/
if (!cpuset1_cpus_initialize(dfd_base, dfd_next, v == '1')) if (!cpuset1_cpus_initialize(dfd_base, dfd_next, v == '1'))
return syserror_ret(false, "Failed to initialize cpuset.cpus"); return syserror_ret(false, "Failed to initialize cpuset.cpus");
@ -697,12 +664,12 @@ static bool cpuset1_initialize(int dfd_base, int dfd_next)
if (bytes < 0) if (bytes < 0)
return syserror_ret(false, "Failed to read file %d(cpuset.mems)", dfd_base); return syserror_ret(false, "Failed to read file %d(cpuset.mems)", dfd_base);
/* ... and copy to first cgroup in the tree... */ /* and copy to first cgroup in the tree... */
bytes = lxc_writeat(dfd_next, "cpuset.mems", mems, bytes); bytes = lxc_writeat(dfd_next, "cpuset.mems", mems, bytes);
if (bytes < 0) if (bytes < 0)
return syserror_ret(false, "Failed to write %d(cpuset.mems)", dfd_next); return syserror_ret(false, "Failed to write %d(cpuset.mems)", dfd_next);
/* ... and finally turn on cpuset inheritance. */ /* and finally turn on cpuset inheritance. */
bytes = lxc_writeat(dfd_next, "cgroup.clone_children", "1", 1); bytes = lxc_writeat(dfd_next, "cgroup.clone_children", "1", 1);
if (bytes < 0) if (bytes < 0)
return syserror_ret(false, "Failed to write %d(cgroup.clone_children)", dfd_next); return syserror_ret(false, "Failed to write %d(cgroup.clone_children)", dfd_next);