1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
|
--- src/sys/kern/kern_descrip.c 2006/09/29 12:29:26
+++ src/sys/kern/kern_descrip.c 2006/12/29 16:06:02
@@ -2150,6 +2150,17 @@
FILE_UNLOCK(fp);
return (0);
}
+
+ /*
+ * We might have just dropped the last reference to a file
+ * object that is for a UNIX domain socket whose message
+ * buffers are being examined in unp_gc(). If that is the
+ * case, FWAIT will be set in f_gcflag and we need to wait for
+ * unp_gc() to finish its scan.
+ */
+ while (fp->f_gcflag & FWAIT)
+ msleep(&fp->f_gcflag, fp->f_mtxp, 0, "fpdrop", 0);
+
/* We have the last ref so we can proceed without the file lock. */
FILE_UNLOCK(fp);
if (fp->f_count < 0)
--- src/sys/kern/uipc_usrreq.c 2006/07/13 07:28:19
+++ src/sys/kern/uipc_usrreq.c 2006/12/29 16:16:12
@@ -68,6 +69,17 @@
#include <vm/uma.h>
+/*
+ * We allocate wrapper objects that add the reference count to an existing
+ * unpcb in 6.x to preserve the ABI layout of unpcb.
+ */
+struct unpcb_wrapper {
+ struct unpcb unpw_unpcb;
+ u_int unpw_refcount;
+};
+
+#define UNP_REFCOUNT(unp) (((struct unpcb_wrapper *)(unp))->unpw_refcount)
+
static uma_zone_t unp_zone;
static unp_gen_t unp_gencnt;
static u_int unp_count;
@@ -769,6 +781,7 @@
unp->unp_socket = so;
so->so_pcb = unp;
+ UNP_REFCOUNT(unp) = 1;
UNP_LOCK();
unp->unp_gencnt = ++unp_gencnt;
unp_count++;
@@ -782,8 +795,10 @@
static void
unp_detach(struct unpcb *unp)
{
+ struct sockaddr_un *saved_unp_addr;
struct vnode *vp;
int local_unp_rights;
+ int freeunp;
UNP_LOCK_ASSERT();
@@ -807,10 +822,15 @@
soisdisconnected(unp->unp_socket);
unp->unp_socket->so_pcb = NULL;
local_unp_rights = unp_rights;
+ saved_unp_addr = unp->unp_addr;
+ unp->unp_addr = NULL;
+ UNP_REFCOUNT(unp)--;
+ freeunp = (UNP_REFCOUNT(unp) == 0);
UNP_UNLOCK();
- if (unp->unp_addr != NULL)
- FREE(unp->unp_addr, M_SONAME);
- uma_zfree(unp_zone, unp);
+ if (saved_unp_addr != NULL)
+ FREE(saved_unp_addr, M_SONAME);
+ if (freeunp)
+ uma_zfree(unp_zone, unp);
if (vp) {
int vfslocked;
@@ -1126,6 +1146,7 @@
unp_pcblist(SYSCTL_HANDLER_ARGS)
{
int error, i, n;
+ int freeunp;
struct unpcb *unp, **unp_list;
unp_gen_t gencnt;
struct xunpgen *xug;
@@ -1177,6 +1198,7 @@
unp->unp_socket->so_cred))
continue;
unp_list[i++] = unp;
+ UNP_REFCOUNT(unp)++;
}
}
UNP_UNLOCK();
@@ -1186,7 +1208,9 @@
xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
for (i = 0; i < n; i++) {
unp = unp_list[i];
- if (unp->unp_gencnt <= gencnt) {
+ UNP_LOCK();
+ UNP_REFCOUNT(unp)--;
+ if (UNP_REFCOUNT(unp) != 0 && unp->unp_gencnt <= gencnt) {
xu->xu_len = sizeof *xu;
xu->xu_unpp = unp;
/*
@@ -1203,7 +1227,13 @@
unp->unp_conn->unp_addr->sun_len);
bcopy(unp, &xu->xu_unp, sizeof *unp);
sotoxsocket(unp->unp_socket, &xu->xu_socket);
+ UNP_UNLOCK();
error = SYSCTL_OUT(req, xu, sizeof *xu);
+ } else {
+ freeunp = (UNP_REFCOUNT(unp) == 0);
+ UNP_UNLOCK();
+ if (freeunp)
+ uma_zfree(unp_zone, unp);
}
}
free(xu, M_TEMP);
@@ -1401,8 +1431,8 @@
void
unp_init(void)
{
- unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
- NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
+ unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb_wrapper), NULL,
+ NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
if (unp_zone == NULL)
panic("unp_init");
uma_zone_set_max(unp_zone, maxsockets);
@@ -1636,7 +1666,7 @@
unp_defer = 0;
/*
* before going through all this, set all FDs to
- * be NOT defered and NOT externally accessible
+ * be NOT deferred and NOT externally accessible
*/
sx_slock(&filelist_lock);
LIST_FOREACH(fp, &filehead, f_list)
@@ -1659,16 +1689,16 @@
continue;
}
/*
- * If we already marked it as 'defer' in a
- * previous pass, then try process it this time
- * and un-mark it
+ * If we already marked it as 'defer' in a
+ * previous pass, then try to process it this
+ * time and un-mark it
*/
if (fp->f_gcflag & FDEFER) {
fp->f_gcflag &= ~FDEFER;
unp_defer--;
} else {
/*
- * if it's not defered, then check if it's
+ * if it's not deferred, then check if it's
* already marked.. if so skip it
*/
if (fp->f_gcflag & FMARK) {
@@ -1691,7 +1721,7 @@
fp->f_gcflag |= FMARK;
}
/*
- * either it was defered, or it is externally
+ * either it was deferred, or it is externally
* accessible and not already marked so.
* Now check if it is possibly one of OUR sockets.
*/
@@ -1700,13 +1730,23 @@
FILE_UNLOCK(fp);
continue;
}
- FILE_UNLOCK(fp);
if (so->so_proto->pr_domain != &localdomain ||
- (so->so_proto->pr_flags&PR_RIGHTS) == 0)
+ (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
+ FILE_UNLOCK(fp);
continue;
+ }
+
+ /*
+ * Tell any other threads that do a subsequent
+ * fdrop() that we are scanning the message
+ * buffers.
+ */
+ fp->f_gcflag |= FWAIT;
+ FILE_UNLOCK(fp);
+
/*
* So, Ok, it's one of our sockets and it IS externally
- * accessible (or was defered). Now we look
+ * accessible (or was deferred). Now we look
* to see if we hold any file descriptors in its
* message buffers. Follow those links and mark them
* as accessible too.
@@ -1714,6 +1754,14 @@
SOCKBUF_LOCK(&so->so_rcv);
unp_scan(so->so_rcv.sb_mb, unp_mark);
SOCKBUF_UNLOCK(&so->so_rcv);
+
+ /*
+ * Wake up any threads waiting in fdrop().
+ */
+ FILE_LOCK(fp);
+ fp->f_gcflag &= ~FWAIT;
+ wakeup(&fp->f_gcflag);
+ FILE_UNLOCK(fp);
}
} while (unp_defer);
sx_sunlock(&filelist_lock);
--- src/sys/sys/file.h 2006/05/29 20:28:17
+++ src/sys/sys/file.h 2006/12/29 16:16:12
@@ -127,6 +127,7 @@
short f_gcflag; /* used by thread doing fd garbage collection */
#define FMARK 0x1 /* mark during gc() */
#define FDEFER 0x2 /* defer for next gc pass */
+#define FWAIT 0x4 /* gc is scanning message buffers */
int f_msgcount; /* (f) references from message queue */
/* DTYPE_VNODE specific fields */
|
GitWeb