From 5ff0e810f04de4b31f605ba3179dec3b3777978a Mon Sep 17 00:00:00 2001
From: "fu.lin" <fulin10@huawei.com>
Date: Mon, 8 Nov 2021 15:08:12 +0800
Subject: [PATCH 28/72] infiniband: fix the infiniband fd conflict

Phenomenon:
  Operating uverbs device will generate anonymous fd named
`anon_inode:[infinibandevent]`. When `anon_inode:[infinibandevent]` fd
is the last opened fd, and some kind of unix socket fd exist, which is
generated by syscalls like `socketpair()` at the same tim,
`anon_inode:[infinibandevent]` will restore fail probabilistically.

log as the following:

```
(00.254523)  63959: open file flags:1
(00.254526)  63959: unix: Opening standalone (stage 0 id 0x1ff ino 1019605 peer 0)
(00.254571)  63959: *******flags: 0
(00.254575)  63959: 		Create fd for 1408 # the fake fd
(00.254578)  63959: *******flags: 1
(00.254580)  63959: 		Create fd for 445 # the restoration fd
```

Reason:
  During the restoration of unix socket, `socketpair()` will generate
two fds, one is used to the current restoration, another is called fake
fd which fd nr is owned by `find_unused_fd()`. When
`anon_inode:[infinibandevent]` fd is the last one, criu don't dump the
fd information for `anon_inode:[infinibandevent]` in original
implementation, and criu think the fd nr which should belong to
`anon_inode:[infinibandevent]` isn't used. Therefore, it cause the
`anon_inode:[infinibandevent]` restoration fail.

This patch fix the above problem. Core: dump
`anon_inode:[infinibandevent]` fd information, make the criu is aware
that that fd nr is used.

Conflict:NA
Reference:https://gitee.com/src-openeuler/criu/pulls/21
Signed-off-by: fu.lin <fulin10@huawei.com>
---
 criu/files-chr.c             | 57 ++++++++++++++++++++++++++++++++++++
 criu/files.c                 | 10 +++----
 criu/include/files-chr.h     |  8 +++++
 criu/include/image-desc.h    |  1 +
 criu/include/protobuf-desc.h |  1 +
 images/chr.proto             |  3 ++
 images/fdinfo.proto          |  2 ++
 7 files changed, 76 insertions(+), 6 deletions(-)

diff --git a/criu/files-chr.c b/criu/files-chr.c
index 95d93e1..6d87c33 100644
--- a/criu/files-chr.c
+++ b/criu/files-chr.c
@@ -215,3 +215,60 @@ static int handle_hisi_vma(struct list_head *fds, struct vma_area *vma)
 
 	return 0;
 }
+
+static void pr_info_infiniband(char *action, InfinibandEntry *infiniband)
+{
+	pr_info("%sinfiniband: id %#08x\n", action, infiniband->id);
+}
+
+static int dump_one_infiniband(int lfd, u32 id, const struct fd_parms *p)
+{
+	FileEntry fe = FILE_ENTRY__INIT;
+	InfinibandEntry infiniband = INFINIBAND_ENTRY__INIT;
+
+	infiniband.id = id;
+
+	fe.type  = FD_TYPES__INFINIBAND;
+	fe.id = infiniband.id;
+	fe.infiniband = &infiniband;
+
+	pr_info_infiniband("Dumping ", &infiniband);
+
+	return pb_write_one(img_from_set(glob_imgset, CR_FD_FILES), &fe, PB_FILE);
+}
+
+const struct fdtype_ops infiniband_dump_ops = {
+	.type	= FD_TYPES__INFINIBAND,
+	.dump	= dump_one_infiniband,
+};
+
+static int infiniband_open(struct file_desc *d, int *new_fd) {
+	/*
+	 * `*new_fd == -1` at this time, it means this open operation shouldn't
+	 * be served out, which is why this function does nothing here.
+	 */
+	return 0;
+};
+
+static struct file_desc_ops infiniband_desc_ops = {
+	.type	= FD_TYPES__INFINIBAND,
+	.open	= infiniband_open,
+};
+
+static int collect_one_infiniband(void *o, ProtobufCMessage *base, struct cr_img *i)
+{
+	struct infiniband_file_info *info = o;
+
+	info->infiniband = pb_msg(base, InfinibandEntry);
+	pr_info_infiniband("Collected ", info->infiniband);
+
+	/* add the fd to `file_desc_hash` list to prevent from NULL pointer */
+	return file_desc_add(&info->d, info->infiniband->id, &infiniband_desc_ops);
+}
+
+struct collect_image_info infiniband_cinfo = {
+	.fd_type = CR_FD_INFINIBAND,
+	.pb_type = PB_INFINIBAND,
+	.priv_size = sizeof(struct infiniband_file_info),
+	.collect = collect_one_infiniband,
+};
diff --git a/criu/files.c b/criu/files.c
index 7b688f5..1ec5281 100644
--- a/criu/files.c
+++ b/criu/files.c
@@ -639,7 +639,7 @@ static int dump_one_file(struct pid *pid, int fd, int lfd, struct fd_opts *opts,
 		else if (is_timerfd_link(link))
 			ops = &timerfd_dump_ops;
 		else if (is_infiniband_link(link))
-			return 1;
+			ops = &infiniband_dump_ops;
 #ifdef CONFIG_HAS_LIBBPF
 		else if (is_bpfmap_link(link))
 			ops = &bpfmap_dump_ops;
@@ -744,11 +744,6 @@ int dump_task_files_seized(struct parasite_ctl *ctl, struct pstree_item *item, s
 			ret = dump_one_file(item->pid, dfds->fds[i + off], lfds[i], opts + i, ctl, &e, dfds);
 			if (ret)
 				break;
-			/* infiniband link file */
-			if (ret > 0) {
-				ret = 0;
-				continue;
-			}
 
 			ret = pb_write_one(img, &e, PB_FDINFO);
 			if (ret)
@@ -1864,6 +1859,9 @@ static int collect_one_file(void *o, ProtobufCMessage *base, struct cr_img *i)
 	case FD_TYPES__CHR:
 		ret = collect_one_file_entry(fe, fe->chr->id, &fe->chr->base, &chrfile_cinfo);
 		break;
+	case FD_TYPES__INFINIBAND:
+		ret = collect_one_file_entry(fe, fe->infiniband->id, &fe->infiniband->base, &infiniband_cinfo);
+		break;
 	}
 
 	return ret;
diff --git a/criu/include/files-chr.h b/criu/include/files-chr.h
index 26b8fb2..261c4b2 100644
--- a/criu/include/files-chr.h
+++ b/criu/include/files-chr.h
@@ -38,4 +38,12 @@ enum hisi_sec_dev {
 	HISI_SEC_DUS	= 0x2000,
 };
 
+struct infiniband_file_info {
+	InfinibandEntry *infiniband;
+	struct file_desc d;
+};
+
+extern const struct fdtype_ops infiniband_dump_ops;
+extern struct collect_image_info infiniband_cinfo;
+
 #endif /* __CRIU_FILES_CHR_H__ */
diff --git a/criu/include/image-desc.h b/criu/include/image-desc.h
index e35f8b2..9ad5fa0 100644
--- a/criu/include/image-desc.h
+++ b/criu/include/image-desc.h
@@ -116,6 +116,7 @@ enum {
 
 	CR_FD_AUTOFS,
 	CR_FD_CHRFILE,
+	CR_FD_INFINIBAND,
 
 	CR_FD_MAX
 };
diff --git a/criu/include/protobuf-desc.h b/criu/include/protobuf-desc.h
index 2468e8f..72a9e1d 100644
--- a/criu/include/protobuf-desc.h
+++ b/criu/include/protobuf-desc.h
@@ -71,6 +71,7 @@ enum {
 	PB_BPFMAP_DATA,
 	PB_APPARMOR,
 	PB_CHRFILE,
+	PB_INFINIBAND,
 
 	/* PB_AUTOGEN_STOP */
 
diff --git a/images/chr.proto b/images/chr.proto
index 67929db..ed65005 100644
--- a/images/chr.proto
+++ b/images/chr.proto
@@ -10,3 +10,6 @@ message chrfile_entry {
 	required bool		repair	= 5;
 };
 
+message infiniband_entry {
+	required uint32		id	= 1;
+};
diff --git a/images/fdinfo.proto b/images/fdinfo.proto
index 6549472..eb52f35 100644
--- a/images/fdinfo.proto
+++ b/images/fdinfo.proto
@@ -44,6 +44,7 @@ enum fd_types {
 	MEMFD		= 18;
 	BPFMAP		= 19;
 	CHR		= 21;
+	INFINIBAND	= 22;
 
 	/* Any number above the real used. Not stored to image */
 	CTL_TTY		= 65534;
@@ -81,4 +82,5 @@ message file_entry {
 	optional memfd_file_entry	memfd	= 20;
 	optional bpfmap_file_entry	bpf	= 21;
 	optional chrfile_entry          chr     = 23;
+	optional infiniband_entry	infiniband	= 25;
 }
-- 
2.34.1