file reference counts can get corrupted

Callers of zfs_file_get and zfs_file_put can corrupt the reference
counts for the file structure resulting in a panic or a soft lockup.
When zfs send/recv runs, it will add a reference count to the
open file, and begin to send or recv the stream. If the file descriptor
is closed, then when dmu_recv_stream() or dmu_send() return we will
call zfs_file_put to remove the reference we placed on the file
structure. Unfortunately, because zfs_file_put() uses the file
descriptor to lookup the file structure, it may end up finding that
the file descriptor table no longer contains the file struct, thus
leaking the file structure. Or it might end up finding a file
descriptor for a different file and blindly updating its reference
counts. Other failure modes probably exists.

This change reworks the zfs_file_[get|put] interface to not rely
on the file descriptor but instead pass the zfs_file_t pointer around.

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Co-authored-by: Allan Jude <allan@klarasystems.com>
Signed-off-by: George Wilson <gwilson@delphix.com>
External-issue: DLPX-76119
Closes #12299

10 files changed, 91 insertions, 107 deletions

diff --git a/include/sys/fm/util.h b/include/sys/fm/util.h
index ea8c61a8b..8fa82b7b6 100644
--- a/include/sys/fm/util.h
+++ b/include/sys/fm/util.h
@@ -31,6 +31,7 @@ extern "C" {
 #endif
 
 #include <sys/nvpair.h>
+#include <sys/zfs_file.h>
 
 /*
  * Shared user/kernel definitions for class length, error channel name,
@@ -96,8 +97,8 @@ extern void fm_nvprint(nvlist_t *);
 extern void zfs_zevent_post_cb(nvlist_t *nvl, nvlist_t *detector);
 extern int zfs_zevent_post(nvlist_t *, nvlist_t *, zevent_cb_t *);
 extern void zfs_zevent_drain_all(int *);
-extern int zfs_zevent_fd_hold(int, minor_t *, zfs_zevent_t **);
-extern void zfs_zevent_fd_rele(int);
+extern zfs_file_t *zfs_zevent_fd_hold(int, minor_t *, zfs_zevent_t **);
+extern void zfs_zevent_fd_rele(zfs_file_t *);
 extern int zfs_zevent_next(zfs_zevent_t *, nvlist_t **, uint64_t *, uint64_t *);
 extern int zfs_zevent_wait(zfs_zevent_t *);
 extern int zfs_zevent_seek(zfs_zevent_t *, uint64_t);
diff --git a/include/sys/zfs_file.h b/include/sys/zfs_file.h
index d117933a6..02cd1a6f0 100644
--- a/include/sys/zfs_file.h
+++ b/include/sys/zfs_file.h
@@ -22,6 +22,8 @@
 #ifndef	_SYS_ZFS_FILE_H
 #define	_SYS_ZFS_FILE_H
 
+#include <sys/zfs_context.h>
+
 #ifndef _KERNEL
 typedef struct zfs_file {
 	int f_fd;
@@ -55,8 +57,8 @@ int zfs_file_fallocate(zfs_file_t *fp, int mode, loff_t offset, loff_t len);
 loff_t zfs_file_off(zfs_file_t *fp);
 int zfs_file_unlink(const char *);
 
-int zfs_file_get(int fd, zfs_file_t **fp);
-void zfs_file_put(int fd);
+zfs_file_t *zfs_file_get(int fd);
+void zfs_file_put(zfs_file_t *fp);
 void *zfs_file_private(zfs_file_t *fp);
 
 #endif /* _SYS_ZFS_FILE_H */
diff --git a/include/sys/zfs_ioctl.h b/include/sys/zfs_ioctl.h
index 8834c5299..1ca3f211b 100644
--- a/include/sys/zfs_ioctl.h
+++ b/include/sys/zfs_ioctl.h
@@ -566,7 +566,7 @@ typedef struct zfsdev_state {
 } zfsdev_state_t;
 
 extern void *zfsdev_get_state(minor_t minor, enum zfsdev_state_type which);
-extern int zfsdev_getminor(int fd, minor_t *minorp);
+extern int zfsdev_getminor(zfs_file_t *fp, minor_t *minorp);
 extern minor_t zfsdev_minor_alloc(void);
 
 extern uint_t zfs_fsyncer_key;
diff --git a/include/sys/zfs_onexit.h b/include/sys/zfs_onexit.h
index 0fab23ff8..fd3030e3a 100644
--- a/include/sys/zfs_onexit.h
+++ b/include/sys/zfs_onexit.h
@@ -51,8 +51,8 @@ extern void zfs_onexit_destroy(zfs_onexit_t *zo);
 
 #endif
 
-extern int zfs_onexit_fd_hold(int fd, minor_t *minorp);
-extern void zfs_onexit_fd_rele(int fd);
+extern zfs_file_t *zfs_onexit_fd_hold(int fd, minor_t *minorp);
+extern void zfs_onexit_fd_rele(zfs_file_t *);
 extern int zfs_onexit_add_cb(minor_t minor, void (*func)(void *), void *data,
     uint64_t *action_handle);
 
diff --git a/lib/libzpool/kernel.c b/lib/libzpool/kernel.c
index ca3578993..5abeb6779 100644
--- a/lib/libzpool/kernel.c
+++ b/lib/libzpool/kernel.c
@@ -966,16 +966,16 @@ kmem_asprintf(const char *fmt, ...)
 }
 
 /* ARGSUSED */
-int
+zfs_file_t *
 zfs_onexit_fd_hold(int fd, minor_t *minorp)
 {
 	*minorp = 0;
-	return (0);
+	return (NULL);
 }
 
 /* ARGSUSED */
 void
-zfs_onexit_fd_rele(int fd)
+zfs_onexit_fd_rele(zfs_file_t *fp)
 {
 }
 
@@ -1385,28 +1385,26 @@ zfs_file_unlink(const char *path)
  * Get reference to file pointer
  *
  * fd - input file descriptor
- * fpp - pointer to file pointer
  *
- * Returns 0 on success EBADF on failure.
+ * Returns pointer to file struct or NULL.
  * Unsupported in user space.
  */
-int
-zfs_file_get(int fd, zfs_file_t **fpp)
+zfs_file_t *
+zfs_file_get(int fd)
 {
 	abort();
 
-	return (EOPNOTSUPP);
+	return (NULL);
 }
-
 /*
  * Drop reference to file pointer
  *
- * fd - input file descriptor
+ * fp - pointer to file struct
  *
  * Unsupported in user space.
  */
 void
-zfs_file_put(int fd)
+zfs_file_put(zfs_file_t *fp)
 {
 	abort();
 }
diff --git a/module/os/freebsd/zfs/zfs_file_os.c b/module/os/freebsd/zfs/zfs_file_os.c
index bfdc4159a..a8c774a4e 100644
--- a/module/os/freebsd/zfs/zfs_file_os.c
+++ b/module/os/freebsd/zfs/zfs_file_os.c
@@ -241,28 +241,21 @@ zfs_file_fsync(zfs_file_t *fp, int flags)
 	return (zfs_vop_fsync(fp->f_vnode));
 }
 
-int
-zfs_file_get(int fd, zfs_file_t **fpp)
+zfs_file_t *
+zfs_file_get(int fd)
 {
 	struct file *fp;
 
 	if (fget(curthread, fd, &cap_no_rights, &fp))
-		return (SET_ERROR(EBADF));
+		return (NULL);
 
-	*fpp = fp;
-	return (0);
+	return (fp);
 }
 
 void
-zfs_file_put(int fd)
+zfs_file_put(zfs_file_t *fp)
 {
-	struct file *fp;
-
-	/* No CAP_ rights required, as we're only releasing. */
-	if (fget(curthread, fd, &cap_no_rights, &fp) == 0) {
-		fdrop(fp, curthread);
-		fdrop(fp, curthread);
-	}
+	fdrop(fp, curthread);
 }
 
 loff_t
diff --git a/module/os/linux/zfs/zfs_file_os.c b/module/os/linux/zfs/zfs_file_os.c
index 99c6ffc95..fe522d257 100644
--- a/module/os/linux/zfs/zfs_file_os.c
+++ b/module/os/linux/zfs/zfs_file_os.c
@@ -405,36 +405,22 @@ zfs_file_unlink(const char *path)
  * Get reference to file pointer
  *
  * fd - input file descriptor
- * fpp - pointer to file pointer
  *
- * Returns 0 on success EBADF on failure.
+ * Returns pointer to file struct or NULL
  */
-int
-zfs_file_get(int fd, zfs_file_t **fpp)
+zfs_file_t *
+zfs_file_get(int fd)
 {
-	zfs_file_t *fp;
-
-	fp = fget(fd);
-	if (fp == NULL)
-		return (EBADF);
-
-	*fpp = fp;
-
-	return (0);
+	return (fget(fd));
 }
 
 /*
  * Drop reference to file pointer
  *
- * fd - input file descriptor
+ * fp - input file struct pointer
  */
 void
-zfs_file_put(int fd)
+zfs_file_put(zfs_file_t *fp)
 {
-	struct file *fp;
-
-	if ((fp = fget(fd)) != NULL) {
-		fput(fp);
-		fput(fp);
-	}
+	fput(fp);
 }
diff --git a/module/zfs/fm.c b/module/zfs/fm.c
index 6ad4f582e..e868798c8 100644
--- a/module/zfs/fm.c
+++ b/module/zfs/fm.c
@@ -586,25 +586,29 @@ zfs_zevent_minor_to_state(minor_t minor, zfs_zevent_t **ze)
 	return (0);
 }
 
-int
+zfs_file_t *
 zfs_zevent_fd_hold(int fd, minor_t *minorp, zfs_zevent_t **ze)
 {
-	int error;
+	zfs_file_t *fp = zfs_file_get(fd);
+	if (fp == NULL)
+		return (NULL);
 
-	error = zfsdev_getminor(fd, minorp);
+	int error = zfsdev_getminor(fp, minorp);
 	if (error == 0)
 		error = zfs_zevent_minor_to_state(*minorp, ze);
 
-	if (error)
-		zfs_zevent_fd_rele(fd);
+	if (error) {
+		zfs_zevent_fd_rele(fp);
+		fp = NULL;
+	}
 
-	return (error);
+	return (fp);
 }
 
 void
-zfs_zevent_fd_rele(int fd)
+zfs_zevent_fd_rele(zfs_file_t *fp)
 {
-	zfs_file_put(fd);
+	zfs_file_put(fp);
 }
 
 /*
diff --git a/module/zfs/zfs_ioctl.c b/module/zfs/zfs_ioctl.c
index 9f670d50a..6f629c984 100644
--- a/module/zfs/zfs_ioctl.c
+++ b/module/zfs/zfs_ioctl.c
@@ -4834,8 +4834,8 @@ zfs_ioc_recv_impl(char *tofs, char *tosnap, char *origin, nvlist_t *recvprops,
 	*errors = fnvlist_alloc();
 	off = 0;
 
-	if ((error = zfs_file_get(input_fd, &input_fp)))
-		return (error);
+	if ((input_fp = zfs_file_get(input_fd)) == NULL)
+		return (SET_ERROR(EBADF));
 
 	noff = off = zfs_file_off(input_fp);
 	error = dmu_recv_begin(tofs, tosnap, begin_record, force,
@@ -5115,7 +5115,7 @@ zfs_ioc_recv_impl(char *tofs, char *tosnap, char *origin, nvlist_t *recvprops,
 		nvlist_free(inheritprops);
 	}
 out:
-	zfs_file_put(input_fd);
+	zfs_file_put(input_fp);
 	nvlist_free(origrecvd);
 	nvlist_free(origprops);
 
@@ -5445,8 +5445,8 @@ zfs_ioc_send(zfs_cmd_t *zc)
 		zfs_file_t *fp;
 		dmu_send_outparams_t out = {0};
 
-		if ((error = zfs_file_get(zc->zc_cookie, &fp)))
-			return (error);
+		if ((fp = zfs_file_get(zc->zc_cookie)) == NULL)
+			return (SET_ERROR(EBADF));
 
 		off = zfs_file_off(fp);
 		out.dso_outfunc = dump_bytes;
@@ -5456,7 +5456,7 @@ zfs_ioc_send(zfs_cmd_t *zc)
 		    zc->zc_fromobj, embedok, large_block_ok, compressok,
 		    rawok, savedok, zc->zc_cookie, &off, &out);
 
-		zfs_file_put(zc->zc_cookie);
+		zfs_file_put(fp);
 	}
 	return (error);
 }
@@ -6020,25 +6020,24 @@ zfs_ioc_tmp_snapshot(zfs_cmd_t *zc)
 {
 	char *snap_name;
 	char *hold_name;
-	int error;
 	minor_t minor;
 
-	error = zfs_onexit_fd_hold(zc->zc_cleanup_fd, &minor);
-	if (error != 0)
-		return (error);
+	zfs_file_t *fp = zfs_onexit_fd_hold(zc->zc_cleanup_fd, &minor);
+	if (fp == NULL)
+		return (SET_ERROR(EBADF));
 
 	snap_name = kmem_asprintf("%s-%016llx", zc->zc_value,
 	    (u_longlong_t)ddi_get_lbolt64());
 	hold_name = kmem_asprintf("%%%s", zc->zc_value);
 
-	error = dsl_dataset_snapshot_tmp(zc->zc_name, snap_name, minor,
+	int error = dsl_dataset_snapshot_tmp(zc->zc_name, snap_name, minor,
 	    hold_name);
 	if (error == 0)
 		(void) strlcpy(zc->zc_value, snap_name,
 		    sizeof (zc->zc_value));
 	kmem_strfree(snap_name);
 	kmem_strfree(hold_name);
-	zfs_onexit_fd_rele(zc->zc_cleanup_fd);
+	zfs_onexit_fd_rele(fp);
 	return (error);
 }
 
@@ -6058,13 +6057,13 @@ zfs_ioc_diff(zfs_cmd_t *zc)
 	offset_t off;
 	int error;
 
-	if ((error = zfs_file_get(zc->zc_cookie, &fp)))
-		return (error);
+	if ((fp = zfs_file_get(zc->zc_cookie)) == NULL)
+		return (SET_ERROR(EBADF));
 
 	off = zfs_file_off(fp);
 	error = dmu_diff(zc->zc_name, zc->zc_value, fp, &off);
 
-	zfs_file_put(zc->zc_cookie);
+	zfs_file_put(fp);
 
 	return (error);
 }
@@ -6100,6 +6099,7 @@ zfs_ioc_hold(const char *pool, nvlist_t *args, nvlist_t *errlist)
 	int cleanup_fd = -1;
 	int error;
 	minor_t minor = 0;
+	zfs_file_t *fp = NULL;
 
 	holds = fnvlist_lookup_nvlist(args, "holds");
 
@@ -6117,14 +6117,16 @@ zfs_ioc_hold(const char *pool, nvlist_t *args, nvlist_t *errlist)
 	}
 
 	if (nvlist_lookup_int32(args, "cleanup_fd", &cleanup_fd) == 0) {
-		error = zfs_onexit_fd_hold(cleanup_fd, &minor);
-		if (error != 0)
-			return (SET_ERROR(error));
+		fp = zfs_onexit_fd_hold(cleanup_fd, &minor);
+		if (fp == NULL)
+			return (SET_ERROR(EBADF));
 	}
 
 	error = dsl_dataset_user_hold(holds, minor, errlist);
-	if (minor != 0)
-		zfs_onexit_fd_rele(cleanup_fd);
+	if (fp != NULL) {
+		ASSERT3U(minor, !=, 0);
+		zfs_onexit_fd_rele(fp);
+	}
 	return (SET_ERROR(error));
 }
 
@@ -6187,9 +6189,9 @@ zfs_ioc_events_next(zfs_cmd_t *zc)
 	uint64_t dropped = 0;
 	int error;
 
-	error = zfs_zevent_fd_hold(zc->zc_cleanup_fd, &minor, &ze);
-	if (error != 0)
-		return (error);
+	zfs_file_t *fp = zfs_zevent_fd_hold(zc->zc_cleanup_fd, &minor, &ze);
+	if (fp == NULL)
+		return (SET_ERROR(EBADF));
 
 	do {
 		error = zfs_zevent_next(ze, &event,
@@ -6211,7 +6213,7 @@ zfs_ioc_events_next(zfs_cmd_t *zc)
 			break;
 	} while (1);
 
-	zfs_zevent_fd_rele(zc->zc_cleanup_fd);
+	zfs_zevent_fd_rele(fp);
 
 	return (error);
 }
@@ -6243,12 +6245,12 @@ zfs_ioc_events_seek(zfs_cmd_t *zc)
 	minor_t minor;
 	int error;
 
-	error = zfs_zevent_fd_hold(zc->zc_cleanup_fd, &minor, &ze);
-	if (error != 0)
-		return (error);
+	zfs_file_t *fp = zfs_zevent_fd_hold(zc->zc_cleanup_fd, &minor, &ze);
+	if (fp == NULL)
+		return (SET_ERROR(EBADF));
 
 	error = zfs_zevent_seek(ze, zc->zc_guid);
-	zfs_zevent_fd_rele(zc->zc_cleanup_fd);
+	zfs_zevent_fd_rele(fp);
 
 	return (error);
 }
@@ -6432,8 +6434,8 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl)
 
 	(void) nvlist_lookup_string(innvl, "redactbook", &redactbook);
 
-	if ((error = zfs_file_get(fd, &fp)))
-		return (error);
+	if ((fp = zfs_file_get(fd)) == NULL)
+		return (SET_ERROR(EBADF));
 
 	off = zfs_file_off(fp);
 
@@ -6445,7 +6447,7 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl)
 	    compressok, rawok, savedok, resumeobj, resumeoff,
 	    redactbook, fd, &off, &out);
 
-	zfs_file_put(fd);
+	zfs_file_put(fp);
 	return (error);
 }
 
@@ -7318,17 +7320,12 @@ pool_status_check(const char *name, zfs_ioc_namecheck_t type,
 }
 
 int
-zfsdev_getminor(int fd, minor_t *minorp)
+zfsdev_getminor(zfs_file_t *fp, minor_t *minorp)
 {
 	zfsdev_state_t *zs, *fpd;
-	zfs_file_t *fp;
-	int rc;
 
 	ASSERT(!MUTEX_HELD(&zfsdev_state_lock));
 
-	if ((rc = zfs_file_get(fd, &fp)))
-		return (rc);
-
 	fpd = zfs_file_private(fp);
 	if (fpd == NULL)
 		return (SET_ERROR(EBADF));
diff --git a/module/zfs/zfs_onexit.c b/module/zfs/zfs_onexit.c
index 2a1332e71..7c56dd9c9 100644
--- a/module/zfs/zfs_onexit.c
+++ b/module/zfs/zfs_onexit.c
@@ -107,30 +107,33 @@ zfs_onexit_destroy(zfs_onexit_t *zo)
  * of this function must call zfs_onexit_fd_rele() when they're finished
  * using the minor number.
  */
-int
+zfs_file_t *
 zfs_onexit_fd_hold(int fd, minor_t *minorp)
 {
 	zfs_onexit_t *zo = NULL;
-	int error;
 
-	error = zfsdev_getminor(fd, minorp);
+	zfs_file_t *fp = zfs_file_get(fd);
+	if (fp == NULL)
+		return (NULL);
+
+	int error = zfsdev_getminor(fp, minorp);
 	if (error) {
-		zfs_onexit_fd_rele(fd);
-		return (error);
+		zfs_onexit_fd_rele(fp);
+		return (NULL);
 	}
 
 	zo = zfsdev_get_state(*minorp, ZST_ONEXIT);
 	if (zo == NULL) {
-		zfs_onexit_fd_rele(fd);
-		return (SET_ERROR(EBADF));
+		zfs_onexit_fd_rele(fp);
+		return (NULL);
 	}
-	return (0);
+	return (fp);
 }
 
 void
-zfs_onexit_fd_rele(int fd)
+zfs_onexit_fd_rele(zfs_file_t *fp)
 {
-	zfs_file_put(fd);
+	zfs_file_put(fp);
 }
 
 static int