[Devel] [PATCH RHEL7 COMMIT] mm/memflags/procfs: Allow userspace to set kernel memory allocation flags

[Konstantin Khorenko]

The commit is pushed to "branch-rh7-3.10.0-1160.88.1.vz7.195.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh7-3.10.0-1160.88.1.vz7.195.1
------>
commit 137f99214d2ca0b4b81e92c1e8c722a87cd20f32
Author: Alexander Atanasov <alexander.atanasov at virtuozzo.com>
Date:   Mon Mar 20 18:47:30 2023 +0200

    mm/memflags/procfs: Allow userspace to set kernel memory allocation flags
    
    Currently there is no way to control if page reclaim can be done.
    This can be useful in networked file systems, which can deadlock
    in the synchronous reclaim path and streaming which can get
    unexpected jitter when a synchronouse reclaim is done.
    
    To improve this add interface to set PF_MEMALLOC and
    PF_MEMALLOC_NOIO flags on current process.
    
    Reading from /proc/self/memflags returns current flags.
    Writing to /proc/self/memflags sets the flags.
    Flag values used are defined in the kernel header include/linux/sched.h.
    
    https://jira.sw.ru/browse/PSBM-141577
    Signed-off-by: Alexander Atanasov <alexander.atanasov at virtuozzo.com>
    
    khorenko@: in our particular case FUSE-based vStorage processes will
    tweak those flags to themselves in order to avoid possible deadlocks.
    
    Feature: vStorage
---
 fs/proc/base.c | 76 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 76 insertions(+)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index eafb5574c543..f8e06a26e73a 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -1425,6 +1425,80 @@ static const struct file_operations proc_oom_score_adj_operations = {
 	.llseek		= default_llseek,
 };
 
+#define MEMALLOC_FLAGS_MASK (PF_MEMALLOC | PF_MEMALLOC_NOIO)
+
+static ssize_t memalloc_flags_read(struct file *file, char __user *buf,
+					size_t count, loff_t *ppos)
+{
+	struct task_struct *task = get_proc_task(file_inode(file));
+	char buffer[PROC_NUMBUF];
+	unsigned int pflags;
+	size_t len;
+
+	if (!task)
+		return -ESRCH;
+
+	pflags = READ_ONCE(task->flags) & MEMALLOC_FLAGS_MASK;
+	put_task_struct(task);
+	len = snprintf(buffer, sizeof(buffer), "%u\n", pflags);
+	return simple_read_from_buffer(buf, count, ppos, buffer, len);
+}
+
+static ssize_t memalloc_flags_write(struct file *file, const char __user *buf,
+					size_t count, loff_t *ppos)
+{
+	struct task_struct *task = get_proc_task(file_inode(file));
+	char buffer[PROC_NUMBUF] = {0};
+	int memalloc_flags;
+	int err = -ESRCH;
+	unsigned int pflags;
+
+	if (!task)
+		goto out;
+
+	if (!ve_is_super(get_exec_env())) {
+		err = -EPERM;
+		goto out;
+	}
+	/*
+	 * Potential issue here if task != current
+	 * concurrent setting of flags need synchronization
+	 * but currently flags are expected to change only from
+	 * current process so there is none.
+	 */
+	if (task != current) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	count = min(count, sizeof(buffer) -1);
+	if (copy_from_user(buffer, buf, count)) {
+		err = -EFAULT;
+		goto out;
+	}
+
+	err = kstrtoint(strstrip(buffer), 0, &memalloc_flags);
+	if (err)
+		goto out;
+	if (memalloc_flags & ~MEMALLOC_FLAGS_MASK) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	pflags = READ_ONCE(task->flags) & ~MEMALLOC_FLAGS_MASK;
+	WRITE_ONCE(task->flags, pflags | memalloc_flags);
+out:
+	if (task)
+		put_task_struct(task);
+	return err < 0 ? err : count;
+}
+
+static const struct file_operations proc_memalloc_flags_operations = {
+	.read		= memalloc_flags_read,
+	.write		= memalloc_flags_write,
+	.llseek		= default_llseek,
+};
+
 #ifdef CONFIG_AUDITSYSCALL
 #define TMPBUFLEN 21
 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
@@ -3132,6 +3206,7 @@ static const struct pid_entry tgid_base_stuff[] = {
 	INF("oom_score",  S_IRUGO, proc_oom_score),
 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
+	REG("memalloc_flags", S_IRUGO|S_IWUSR, proc_memalloc_flags_operations),
 #ifdef CONFIG_AUDITSYSCALL
 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
@@ -3499,6 +3574,7 @@ static const struct pid_entry tid_base_stuff[] = {
 	INF("oom_score", S_IRUGO, proc_oom_score),
 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
+	REG("memalloc_flags", S_IRUGO|S_IWUSR, proc_memalloc_flags_operations),
 #ifdef CONFIG_AUDITSYSCALL
 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),