posixshm: Fix handling of sendfile() with largepage objects

sendfile(2) can transmit POSIX shared memory objects.  Typically it will
look up and wire each page before sending it to a socket; once
transmission is complete, the page is unwired and typically released
back into the page queues.  sendfile() has an advisory flag, SF_NOCACHE,
which means, "try to free the page once transmission is complete."  This
is implemented in vm_page_release(), which expects to operate on managed
pages.

Pages belonging a largepage object are de-facto wired not explicitly so.
Thus, vm_page_release() will unwire and, having found no additional
references, free the page.  Because mappings of largepage objects are
unmanaged, userspace can still access the now freed page.

Fix the problem by explicitly wiring largepage pages.  Make the VM
object destructor responsible for unwiring and freeing them.

Add a regression test.

Approved by:	so
Security:	FreeBSD-SA-26:44.posixshm
Security:	CVE-2026-49427
Reported by:	Chris Jarrett-Davies <chrisjd@openai.com>
Reviewed by:	kib
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D57832

3 files changed, 121 insertions, 13 deletions

diff --git a/sys/kern/uipc_shm.c b/sys/kern/uipc_shm.c
index b1b6a8657d69..8341364f4d16 100644
--- a/sys/kern/uipc_shm.c
+++ b/sys/kern/uipc_shm.c
@@ -326,6 +326,7 @@ static void
 shm_largepage_phys_ctor(vm_object_t object, vm_prot_t prot,
     vm_ooffset_t foff, struct ucred *cred)
 {
+	object->flags |= OBJ_PG_DTOR;
 }
 
 static void
@@ -333,11 +334,27 @@ shm_largepage_phys_dtor(vm_object_t object)
 {
 	int psind;
 
+	VM_OBJECT_ASSERT_WLOCKED(object);
+
 	psind = object->un_pager.phys.data_val;
 	if (psind != 0) {
+		struct pctrie_iter pages;
+		vm_page_t m;
+		bool removed __diagused;
+
+		vm_page_iter_init(&pages, object);
+restart:
+		VM_RADIX_FOREACH(m, &pages) {
+			if (!vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL)) {
+				pctrie_iter_reset(&pages);
+				goto restart;
+			}
+			removed = vm_page_iter_remove(&pages, m);
+			KASSERT(!removed, ("%s: page %p not wired", __func__, m));
+			vm_page_unwire(m, PQ_NONE);
+		}
 		atomic_subtract_long(&count_largepages[psind],
 		    object->size / (pagesizes[psind] / PAGE_SIZE));
-		vm_wire_sub(object->size);
 	} else {
 		KASSERT(object->size == 0,
 		    ("largepage phys obj %p not initialized bit size %#jx > 0",
@@ -818,7 +835,7 @@ shm_dotruncate_largepage(struct shmfd *shmfd, off_t length, void *rl_cookie)
 	if ((shmfd->shm_seals & F_SEAL_GROW) != 0)
 		return (EPERM);
 
-	aflags = VM_ALLOC_NORMAL | VM_ALLOC_ZERO;
+	aflags = VM_ALLOC_NORMAL | VM_ALLOC_ZERO | VM_ALLOC_WIRED;
 	if (shmfd->shm_lp_alloc_policy == SHM_LARGEPAGE_ALLOC_NOWAIT)
 		aflags |= VM_ALLOC_WAITFAIL;
 	try = 0;
@@ -874,7 +891,6 @@ shm_dotruncate_largepage(struct shmfd *shmfd, off_t length, void *rl_cookie)
 		object->size += OFF_TO_IDX(pagesizes[psind]);
 		shmfd->shm_size += pagesizes[psind];
 		atomic_add_long(&count_largepages[psind], 1);
-		vm_wire_add(atop(pagesizes[psind]));
 	}
 	return (0);
 }
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
index 3753079f2fde..cd95cc620ad4 100644
--- a/sys/vm/vm_page.c
+++ b/sys/vm/vm_page.c
@@ -4272,6 +4272,9 @@ vm_page_unwire_managed(vm_page_t m, uint8_t nqueue, bool noreuse)
 {
 	u_int old;
 
+	KASSERT(nqueue < PQ_COUNT,
+	    ("vm_page_unwire: invalid queue %u request for page %p",
+	    nqueue, m));
 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
 	    ("%s: page %p is unmanaged", __func__, m));
 
@@ -4330,17 +4333,15 @@ vm_page_unwire_managed(vm_page_t m, uint8_t nqueue, bool noreuse)
 void
 vm_page_unwire(vm_page_t m, uint8_t nqueue)
 {
-
-	KASSERT(nqueue < PQ_COUNT,
-	    ("vm_page_unwire: invalid queue %u request for page %p",
-	    nqueue, m));
+        KASSERT(nqueue < PQ_COUNT || nqueue == PQ_NONE,
+            ("%s: invalid queue %u request for page %p", __func__, nqueue, m));
 
 	if ((m->oflags & VPO_UNMANAGED) != 0) {
 		if (vm_page_unwire_noq(m) && m->ref_count == 0)
 			vm_page_free(m);
-		return;
+	} else {
+		vm_page_unwire_managed(m, nqueue, false);
 	}
-	vm_page_unwire_managed(m, nqueue, false);
 }
 
 /*
@@ -4514,13 +4515,15 @@ vm_page_release_toq(vm_page_t m, uint8_t nqueue, const bool noreuse)
 void
 vm_page_release(vm_page_t m, int flags)
 {
-	vm_object_t object;
-
-	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
-	    ("vm_page_release: page %p is unmanaged", m));
+	if ((m->oflags & VPO_UNMANAGED) != 0) {
+		vm_page_unwire(m, PQ_NONE);
+		return;
+	}
 
 	if ((flags & VPR_TRYFREE) != 0) {
 		for (;;) {
+			vm_object_t object;
+
 			object = atomic_load_ptr(&m->object);
 			if (object == NULL)
 				break;
diff --git a/tests/sys/posixshm/posixshm_test.c b/tests/sys/posixshm/posixshm_test.c
index e816e2d1efa7..e4b57dec6311 100644
--- a/tests/sys/posixshm/posixshm_test.c
+++ b/tests/sys/posixshm/posixshm_test.c
@@ -33,6 +33,7 @@
 #include <sys/ioctl.h>
 #include <sys/mman.h>
 #include <sys/resource.h>
+#include <sys/socket.h>
 #include <sys/stat.h>
 #include <sys/syscall.h>
 #include <sys/sysctl.h>
@@ -2148,6 +2149,93 @@ ATF_TC_BODY(largepage_reopen, tc)
 	    "close failed; errno=%d", errno);
 }
 
+static unsigned char
+largepage_sendfile_expected(size_t off)
+{
+
+	return ((unsigned char)(off * 131 + (off >> 8)));
+}
+
+ATF_TC_WITHOUT_HEAD(largepage_sendfile);
+ATF_TC_BODY(largepage_sendfile, tc)
+{
+	static const int flags[] = { 0, SF_NOCACHE };
+	char *addr;
+	off_t sbytes;
+	size_t ps[MAXPAGESIZES];
+	int error, fd, pscnt, sd[2], status;
+	pid_t child;
+
+	pscnt = pagesizes(ps, true);
+
+	for (int i = 1; i < pscnt; i++) {
+		for (int fi = 0; fi < (int)nitems(flags); fi++) {
+			fd = shm_open_large(i, SHM_LARGEPAGE_ALLOC_DEFAULT,
+			    ps[i]);
+			addr = mmap(NULL, ps[i], PROT_READ | PROT_WRITE,
+			    MAP_SHARED, fd, 0);
+			ATF_REQUIRE_MSG(addr != MAP_FAILED,
+			    "mmap(%zu bytes) failed; error=%d", ps[i], errno);
+
+			/* Fill with a verifiable pattern. */
+			for (size_t j = 0; j < ps[i]; j++)
+				addr[j] = largepage_sendfile_expected(j);
+
+			ATF_REQUIRE(socketpair(PF_LOCAL, SOCK_STREAM, 0,
+			    sd) == 0);
+
+			child = fork();
+			ATF_REQUIRE_MSG(child != -1,
+			    "fork() failed; error=%d", errno);
+			if (child == 0) {
+				char buf[BUFSIZ];
+				ssize_t len;
+				size_t off, resid;
+
+				(void)close(sd[0]);
+				off = 0;
+				for (resid = ps[i]; resid > 0; resid -= len) {
+					len = read(sd[1], buf, sizeof(buf));
+					if (len <= 0)
+						_exit(1);
+					for (ssize_t k = 0; k < len; k++) {
+						if ((unsigned char)buf[k] !=
+						    largepage_sendfile_expected(
+						    off + k))
+							_exit(2);
+					}
+					off += len;
+				}
+				_exit(0);
+			}
+			ATF_REQUIRE(close(sd[1]) == 0);
+
+			sbytes = 0;
+			error = sendfile(fd, sd[0], 0, ps[i], NULL, &sbytes,
+			    flags[fi]);
+			ATF_REQUIRE_MSG(error == 0,
+			    "sendfile() failed; error=%d flags=%#x",
+			    errno, flags[fi]);
+			ATF_REQUIRE_MSG(sbytes == (off_t)ps[i],
+			    "sendfile() short; sbytes=%jd expected=%zu flags=%#x",
+			    (intmax_t)sbytes, ps[i], flags[fi]);
+
+			ATF_REQUIRE(close(sd[0]) == 0);
+
+			ATF_REQUIRE_MSG(waitpid(child, &status, 0) == child,
+			    "waitpid() failed; error=%d", errno);
+			ATF_REQUIRE_MSG(WIFEXITED(status),
+			    "child killed by signal %d", WTERMSIG(status));
+			ATF_REQUIRE_MSG(WEXITSTATUS(status) == 0,
+			    "child exited with status %d (flags=%#x)",
+			    WEXITSTATUS(status), flags[fi]);
+
+			ATF_REQUIRE(munmap(addr, ps[i]) == 0);
+			ATF_REQUIRE(close(fd) == 0);
+		}
+	}
+}
+
 ATF_TC_WITHOUT_HEAD(largepage_truncate);
 ATF_TC_BODY(largepage_truncate, tc)
 {
@@ -2236,6 +2324,7 @@ ATF_TP_ADD_TCS(tp)
 	ATF_TP_ADD_TC(tp, largepage_pkru);
 #endif
 	ATF_TP_ADD_TC(tp, largepage_reopen);
+	ATF_TP_ADD_TC(tp, largepage_sendfile);
 	ATF_TP_ADD_TC(tp, largepage_truncate);
 
 	return (atf_no_error());