From: Suren Baghdasaryan <surenb@google.com>
To: akpm@linux-foundation.org
Cc: viro@zeniv.linux.org.uk, brauner@kernel.org, jack@suse.cz,
dchinner@redhat.com, casey@schaufler-ca.com,
ben.wolsieffer@hefring.com, paulmck@kernel.org,
david@redhat.com, avagin@google.com, usama.anjum@collabora.com,
peterx@redhat.com, hughd@google.com, ryan.roberts@arm.com,
wangkefeng.wang@huawei.com, Liam.Howlett@Oracle.com,
yuzhao@google.com, axelrasmussen@google.com, lstoakes@gmail.com,
talumbau@google.com, willy@infradead.org, vbabka@suse.cz,
mgorman@techsingularity.net, jhubbard@nvidia.com,
vishal.moola@gmail.com, mathieu.desnoyers@efficios.com,
dhowells@redhat.com, jgg@ziepe.ca, sidhartha.kumar@oracle.com,
andriy.shevchenko@linux.intel.com, yangxingui@huawei.com,
keescook@chromium.org, linux-kernel@vger.kernel.org,
linux-fsdevel@vger.kernel.org, linux-mm@kvack.org,
kernel-team@android.com, surenb@google.com
Subject: [PATCH 3/3] mm/maps: read proc/pid/maps under RCU
Date: Sun, 21 Jan 2024 23:13:24 -0800 [thread overview]
Message-ID: <20240122071324.2099712-3-surenb@google.com> (raw)
In-Reply-To: <20240122071324.2099712-1-surenb@google.com>
With maple_tree supporting vma tree traversal under RCU and per-vma locks
making vma access RCU-safe, /proc/pid/maps can be read under RCU and
without the need to read-lock mmap_lock. However vma content can change
from under us, therefore we make a copy of the vma and we pin pointer
fields used when generating the output (currently only vm_file and
anon_name). Afterwards we check for concurrent address space
modifications, wait for them to end and retry. That last check is needed
to avoid possibility of missing a vma during concurrent maple_tree
node replacement, which might report a NULL when a vma is replaced
with another one. While we take the mmap_lock for reading during such
contention, we do that momentarily only to record new mm_wr_seq counter.
This change is designed to reduce mmap_lock contention and prevent a
process reading /proc/pid/maps files (often a low priority task, such as
monitoring/data collection services) from blocking address space updates.
Note that this change has a userspace visible disadvantage: it allows for
sub-page data tearing as opposed to the previous mechanism where data
tearing could happen only between pages of generated output data.
Since current userspace considers data tearing between pages to be
acceptable, we assume is will be able to handle sub-page data tearing
as well.
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
---
fs/proc/internal.h | 2 +
fs/proc/task_mmu.c | 114 ++++++++++++++++++++++++++++++++++++++++++---
2 files changed, 109 insertions(+), 7 deletions(-)
diff --git a/fs/proc/internal.h b/fs/proc/internal.h
index a71ac5379584..e0247225bb68 100644
--- a/fs/proc/internal.h
+++ b/fs/proc/internal.h
@@ -290,6 +290,8 @@ struct proc_maps_private {
struct task_struct *task;
struct mm_struct *mm;
struct vma_iterator iter;
+ unsigned long mm_wr_seq;
+ struct vm_area_struct vma_copy;
#ifdef CONFIG_NUMA
struct mempolicy *task_mempolicy;
#endif
diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 3f78ebbb795f..3886d04afc01 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -126,11 +126,96 @@ static void release_task_mempolicy(struct proc_maps_private *priv)
}
#endif
-static struct vm_area_struct *proc_get_vma(struct proc_maps_private *priv,
- loff_t *ppos)
+#ifdef CONFIG_PER_VMA_LOCK
+
+static const struct seq_operations proc_pid_maps_op;
+/*
+ * Take VMA snapshot and pin vm_file and anon_name as they are used by
+ * show_map_vma.
+ */
+static int get_vma_snapshow(struct proc_maps_private *priv, struct vm_area_struct *vma)
{
+ struct vm_area_struct *copy = &priv->vma_copy;
+ int ret = -EAGAIN;
+
+ memcpy(copy, vma, sizeof(*vma));
+ if (copy->vm_file && !get_file_rcu(©->vm_file))
+ goto out;
+
+ if (copy->anon_name && !anon_vma_name_get_rcu(copy))
+ goto put_file;
+
+ if (priv->mm_wr_seq == mmap_write_seq_read(priv->mm))
+ return 0;
+
+ /* Address space got modified, vma might be stale. Wait and retry. */
+ rcu_read_unlock();
+ ret = mmap_read_lock_killable(priv->mm);
+ mmap_write_seq_record(priv->mm, &priv->mm_wr_seq);
+ mmap_read_unlock(priv->mm);
+ rcu_read_lock();
+
+ if (!ret)
+ ret = -EAGAIN; /* no other errors, ok to retry */
+
+ if (copy->anon_name)
+ anon_vma_name_put(copy->anon_name);
+put_file:
+ if (copy->vm_file)
+ fput(copy->vm_file);
+out:
+ return ret;
+}
+
+static void put_vma_snapshot(struct proc_maps_private *priv)
+{
+ struct vm_area_struct *vma = &priv->vma_copy;
+
+ if (vma->anon_name)
+ anon_vma_name_put(vma->anon_name);
+ if (vma->vm_file)
+ fput(vma->vm_file);
+}
+
+static inline bool needs_mmap_lock(struct seq_file *m)
+{
+ /*
+ * smaps and numa_maps perform page table walk, therefore require
+ * mmap_lock but maps can be read under RCU.
+ */
+ return m->op != &proc_pid_maps_op;
+}
+
+#else /* CONFIG_PER_VMA_LOCK */
+
+/* Without per-vma locks VMA access is not RCU-safe */
+static inline bool needs_mmap_lock(struct seq_file *m) { return true; }
+
+#endif /* CONFIG_PER_VMA_LOCK */
+
+static struct vm_area_struct *proc_get_vma(struct seq_file *m, loff_t *ppos)
+{
+ struct proc_maps_private *priv = m->private;
struct vm_area_struct *vma = vma_next(&priv->iter);
+#ifdef CONFIG_PER_VMA_LOCK
+ if (vma && !needs_mmap_lock(m)) {
+ int ret;
+
+ put_vma_snapshot(priv);
+ while ((ret = get_vma_snapshow(priv, vma)) == -EAGAIN) {
+ /* lookup the vma at the last position again */
+ vma_iter_init(&priv->iter, priv->mm, *ppos);
+ vma = vma_next(&priv->iter);
+ }
+
+ if (ret) {
+ put_vma_snapshot(priv);
+ return NULL;
+ }
+ vma = &priv->vma_copy;
+ }
+#endif
if (vma) {
*ppos = vma->vm_start;
} else {
@@ -169,12 +254,20 @@ static void *m_start(struct seq_file *m, loff_t *ppos)
return ERR_PTR(-EINTR);
}
+ /* Drop mmap_lock if possible */
+ if (!needs_mmap_lock(m)) {
+ mmap_write_seq_record(priv->mm, &priv->mm_wr_seq);
+ mmap_read_unlock(priv->mm);
+ rcu_read_lock();
+ memset(&priv->vma_copy, 0, sizeof(priv->vma_copy));
+ }
+
vma_iter_init(&priv->iter, mm, last_addr);
hold_task_mempolicy(priv);
if (last_addr == -2UL)
return get_gate_vma(mm);
- return proc_get_vma(priv, ppos);
+ return proc_get_vma(m, ppos);
}
static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
@@ -183,7 +276,7 @@ static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
*ppos = -1UL;
return NULL;
}
- return proc_get_vma(m->private, ppos);
+ return proc_get_vma(m, ppos);
}
static void m_stop(struct seq_file *m, void *v)
@@ -195,7 +288,10 @@ static void m_stop(struct seq_file *m, void *v)
return;
release_task_mempolicy(priv);
- mmap_read_unlock(mm);
+ if (needs_mmap_lock(m))
+ mmap_read_unlock(mm);
+ else
+ rcu_read_unlock();
mmput(mm);
put_task_struct(priv->task);
priv->task = NULL;
@@ -283,8 +379,10 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
start = vma->vm_start;
end = vma->vm_end;
show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
- if (mm)
- anon_name = anon_vma_name(vma);
+ if (mm) {
+ anon_name = needs_mmap_lock(m) ? anon_vma_name(vma) :
+ anon_vma_name_get_rcu(vma);
+ }
/*
* Print the dentry name for named mappings, and a
@@ -338,6 +436,8 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
seq_puts(m, name);
}
seq_putc(m, '\n');
+ if (anon_name && !needs_mmap_lock(m))
+ anon_vma_name_put(anon_name);
}
static int show_map(struct seq_file *m, void *v)
--
2.43.0.429.g432eaa2c6b-goog
next prev parent reply other threads:[~2024-01-22 7:13 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-01-22 7:13 [PATCH 1/3] mm: make vm_area_struct anon_name field RCU-safe Suren Baghdasaryan
2024-01-22 7:13 ` [PATCH 2/3] mm: add mm_struct sequence number to detect write locks Suren Baghdasaryan
2024-01-22 7:13 ` Suren Baghdasaryan [this message]
2024-01-23 5:36 ` [PATCH 3/3] mm/maps: read proc/pid/maps under RCU SeongJae Park
2024-01-23 6:07 ` Suren Baghdasaryan
2024-01-23 23:12 ` Suren Baghdasaryan
2024-01-23 23:48 ` SeongJae Park
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240122071324.2099712-3-surenb@google.com \
--to=surenb@google.com \
--cc=Liam.Howlett@Oracle.com \
--cc=akpm@linux-foundation.org \
--cc=andriy.shevchenko@linux.intel.com \
--cc=avagin@google.com \
--cc=axelrasmussen@google.com \
--cc=ben.wolsieffer@hefring.com \
--cc=brauner@kernel.org \
--cc=casey@schaufler-ca.com \
--cc=david@redhat.com \
--cc=dchinner@redhat.com \
--cc=dhowells@redhat.com \
--cc=hughd@google.com \
--cc=jack@suse.cz \
--cc=jgg@ziepe.ca \
--cc=jhubbard@nvidia.com \
--cc=keescook@chromium.org \
--cc=kernel-team@android.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=lstoakes@gmail.com \
--cc=mathieu.desnoyers@efficios.com \
--cc=mgorman@techsingularity.net \
--cc=paulmck@kernel.org \
--cc=peterx@redhat.com \
--cc=ryan.roberts@arm.com \
--cc=sidhartha.kumar@oracle.com \
--cc=talumbau@google.com \
--cc=usama.anjum@collabora.com \
--cc=vbabka@suse.cz \
--cc=viro@zeniv.linux.org.uk \
--cc=vishal.moola@gmail.com \
--cc=wangkefeng.wang@huawei.com \
--cc=willy@infradead.org \
--cc=yangxingui@huawei.com \
--cc=yuzhao@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox