mm: Add reliable memory use limit for user tasks

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I4SK3S
CVE: NA

----------------------------------------------

there is a upper limit for special user tasks's memory allocation.
special user task means user task with reliable flag.

Init tasks will alloc memory from non-mirrored region if their allocation
trigger limit.

The limit can be set or access via /proc/sys/vm/task_reliable_limit

This limit's default value is ULONG_MAX.
Signed-off-by: NPeng Wu <wupeng58@huawei.com>
Signed-off-by: NMa Wupeng <mawupeng1@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

include/linux/mem_reliable.h

@@ -15,6 +15,7 @@ extern struct static_key_false mem_reliable;
 
 extern bool reliable_enabled;
 extern atomic_long_t reliable_user_used_nr_page;
+extern unsigned long task_reliable_limit __read_mostly;
 
 extern void add_reliable_mem_size(long sz);
 extern void mem_reliable_init(bool has_unmirrored_mem,
@@ -22,6 +23,9 @@ extern void mem_reliable_init(bool has_unmirrored_mem,
 extern void reliable_report_meminfo(struct seq_file *m);
 extern bool page_reliable(struct page *page);
 extern void reliable_report_usage(struct seq_file *m, struct mm_struct *mm);
+extern void reliable_show_mem_info(void);
+extern void mem_reliable_out_of_memory(gfp_t gfp_mask, unsigned int order,
+				       int preferred_nid, nodemask_t *nodemask);
 
 static inline bool mem_reliable_is_enabled(void)
 {
@@ -59,6 +63,12 @@ static inline void reliable_page_counter(struct page *page,
 		atomic_long_add(val, &reliable_user_used_nr_page);
 	}
 }
+
+static inline bool reliable_mem_limit_check(unsigned long nr_page)
+{
+	return atomic_long_read(&reliable_user_used_nr_page) + nr_page <=
+	       task_reliable_limit / PAGE_SIZE;
+}
 #else
 #define reliable_enabled 0
 
@@ -78,6 +88,16 @@ static inline void reliable_page_counter(struct page *page,
 static inline void reliable_report_usage(struct seq_file *m,
 					 struct mm_struct *mm) {}
 
+static inline bool reliable_mem_limit_check(unsigned long nr_page)
+{
+	return false;
+}
+static inline void reliable_show_mem_info(void) {}
+static inline void mem_reliable_out_of_memory(gfp_t gfp_mask,
+					      unsigned int order,
+					      int preferred_nid,
+					      nodemask_t *nodemask) {}
+
 #endif
 
 #endif

lib/show_mem.c

@@ -49,4 +49,5 @@ void show_mem(unsigned int filter, nodemask_t *nodemask)
 #ifdef CONFIG_MEMORY_FAILURE
 	printk("%lu pages hwpoisoned\n", atomic_long_read(&num_poisoned_pages));
 #endif
+	reliable_show_mem_info();
 }

mm/khugepaged.c

@@ -1235,6 +1235,12 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
 out_unmap:
 	pte_unmap_unlock(pte, ptl);
 	if (ret) {
+		if (reliable &&
+		    !reliable_mem_limit_check(1 << HPAGE_PMD_ORDER)) {
+			ret = SCAN_ALLOC_HUGE_PAGE_FAIL;
+			goto out;
+		}
+
 		node = khugepaged_find_target_node();
 		/* collapse_huge_page will return with the mmap_sem released */
 		collapse_huge_page(mm, address, hpage, node,
@@ -1692,6 +1698,12 @@ static void khugepaged_scan_shmem(struct mm_struct *mm,
 		if (present < HPAGE_PMD_NR - khugepaged_max_ptes_none) {
 			result = SCAN_EXCEED_NONE_PTE;
 		} else {
+			if (reliable &&
+			    !reliable_mem_limit_check(1 << HPAGE_PMD_ORDER)) {
+				result = SCAN_ALLOC_HUGE_PAGE_FAIL;
+				goto out;
+			}
+
 			node = khugepaged_find_target_node();
 			collapse_shmem(mm, mapping, start, hpage, node,
 				       reliable);
@@ -1699,6 +1711,8 @@ static void khugepaged_scan_shmem(struct mm_struct *mm,
 	}
 
 	/* TODO: tracepoints */
+out:
+	return;
 }
 #else
 static void khugepaged_scan_shmem(struct mm_struct *mm,

mm/mem_reliable.c

@@ -8,12 +8,15 @@
 #include <linux/memory_hotplug.h>
 #include <linux/seq_file.h>
 #include <linux/mmzone.h>
+#include <linux/oom.h>
 
 DEFINE_STATIC_KEY_FALSE(mem_reliable);
 
 bool reliable_enabled;
 static atomic_long_t total_reliable_mem;
 atomic_long_t reliable_user_used_nr_page;
+/* reliable user limit for user tasks with reliable flag */
+unsigned long task_reliable_limit = ULONG_MAX;
 
 void add_reliable_mem_size(long sz)
 {
@@ -118,3 +121,86 @@ void reliable_report_usage(struct seq_file *m, struct mm_struct *mm)
 			   atomic_long_read(&mm->reliable_nr_page));
 	}
 }
+
+#ifdef CONFIG_SYSCTL
+int reliable_limit_handler(struct ctl_table *table, int write,
+	void __user *buffer, size_t *length, loff_t *ppos)
+{
+	unsigned long old = task_reliable_limit;
+	int ret;
+
+	ret = proc_doulongvec_minmax(table, write, buffer, length, ppos);
+	if (ret == 0 && write) {
+		if (task_reliable_limit > total_reliable_mem_sz()) {
+			task_reliable_limit = old;
+			return -EINVAL;
+		}
+	}
+
+	return ret;
+}
+
+static struct ctl_table reliable_ctl_table[] = {
+	{
+		.procname = "task_reliable_limit",
+		.data = &task_reliable_limit,
+		.maxlen = sizeof(task_reliable_limit),
+		.mode = 0644,
+		.proc_handler = reliable_limit_handler,
+	},
+	{}
+};
+
+static struct ctl_table reliable_dir_table[] = {
+	{
+		.procname = "vm",
+		.maxlen = 0,
+		.mode = 0555,
+		.child = reliable_ctl_table,
+	},
+	{}
+};
+
+static int __init reliable_sysctl_init(void)
+{
+	if (!mem_reliable_is_enabled())
+		return 0;
+
+	if (!register_sysctl_table(reliable_dir_table)) {
+		pr_err("register sysctl failed.");
+		return -1;
+	}
+
+	return 0;
+}
+late_initcall(reliable_sysctl_init);
+#endif
+
+void reliable_show_mem_info(void)
+{
+	if (mem_reliable_is_enabled()) {
+		pr_info("ReliableTotal: %lu kB", total_reliable_mem_sz() >> 10);
+		pr_info("ReliableUsed: %lu kB", used_reliable_mem_sz() >> 10);
+		pr_info("task_reliable_limit: %lu kB",
+			task_reliable_limit >> 10);
+		pr_info("reliable_user_used: %ld kB",
+			atomic_long_read(&reliable_user_used_nr_page) * 4);
+	}
+}
+
+void mem_reliable_out_of_memory(gfp_t gfp_mask, unsigned int order,
+				int preferred_nid, nodemask_t *nodemask)
+{
+	struct oom_control oc = {
+		.zonelist = node_zonelist(preferred_nid, gfp_mask),
+		.nodemask = nodemask,
+		.memcg = NULL,
+		.gfp_mask = gfp_mask,
+		.order = order,
+	};
+
+	if (!mutex_trylock(&oom_lock))
+		return;
+	out_of_memory(&oc);
+	mutex_unlock(&oom_lock);
+}

mm/page_alloc.c

@@ -4561,19 +4561,51 @@ static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac)
 					ac->high_zoneidx, ac->nodemask);
 }
 
-static inline void prepare_before_alloc(gfp_t *gfp_mask)
+/*
+ * return false means this allocation is limit by reliable user limit and
+ * this will lead to pagefault_out_of_memory()
+ */
+static inline bool prepare_before_alloc(gfp_t *gfp_mask, unsigned int order)
 {
 	gfp_t gfp_ori = *gfp_mask;
 	*gfp_mask &= gfp_allowed_mask;
 
 	if (!mem_reliable_is_enabled())
-		return;
+		return true;
 
-	if (gfp_ori & ___GFP_RELIABILITY)
+	if (gfp_ori & ___GFP_RELIABILITY) {
 		*gfp_mask |= ___GFP_RELIABILITY;
+		return true;
+	}
 
-	if (current->flags & PF_RELIABLE || is_global_init(current))
-		*gfp_mask |= ___GFP_RELIABILITY;
+	/*
+	 * Init tasks will alloc memory from non-mirrored region if their
+	 * allocation trigger task_reliable_limit
+	 */
+	if (is_global_init(current)) {
+		if (reliable_mem_limit_check(1 << order))
+			*gfp_mask |= ___GFP_RELIABILITY;
+		return true;
+	}
+
+	/*
+	 * This only check task_reliable_limit without ___GFP_RELIABILITY
+	 * or this process is global init.
+	 * For kernel internal mechanism(hugepaged collapse and others)
+	 * If they alloc memory for user and obey task_reliable_limit, they
+	 * need to check this limit before allocing pages.
+	 */
+	if ((current->flags & PF_RELIABLE) && (gfp_ori & __GFP_HIGHMEM) &&
+	    (gfp_ori & __GFP_MOVABLE)) {
+		if (reliable_mem_limit_check(1 << order)) {
+			*gfp_mask |= ___GFP_RELIABILITY;
+			return true;
+		}
+
+		return false;
+	}
+
+	return true;
 }
 
 /*
@@ -4583,7 +4615,7 @@ struct page *
 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
 							nodemask_t *nodemask)
 {
-	struct page *page;
+	struct page *page = NULL;
 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
 	gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
 	struct alloc_context ac = { };
@@ -4597,7 +4629,11 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
 		return NULL;
 	}
 
-	prepare_before_alloc(&gfp_mask);
+	if (!prepare_before_alloc(&gfp_mask, order)) {
+		mem_reliable_out_of_memory(gfp_mask, order, preferred_nid,
+					   nodemask);
+		goto out;
+	}
 
 	alloc_mask = gfp_mask;
 	if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))