freebsd-dev/sys/fs/fuse/fuse_device.c
Alan Somers 7e1f5432f4 fusefs: don't leak memory of unsent operations on unmount
Sponsored by:	The FreeBSD Foundation
2019-06-28 18:48:02 +00:00

592 lines
16 KiB
C

/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 2007-2009 Google Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following disclaimer
* in the documentation and/or other materials provided with the
* distribution.
* * Neither the name of Google Inc. nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* Copyright (C) 2005 Csaba Henk.
* All rights reserved.
*
* Copyright (c) 2019 The FreeBSD Foundation
*
* Portions of this software were developed by BFF Storage Systems, LLC under
* sponsorship from the FreeBSD Foundation.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/types.h>
#include <sys/module.h>
#include <sys/systm.h>
#include <sys/errno.h>
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/conf.h>
#include <sys/uio.h>
#include <sys/malloc.h>
#include <sys/queue.h>
#include <sys/lock.h>
#include <sys/sx.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/mount.h>
#include <sys/sdt.h>
#include <sys/stat.h>
#include <sys/fcntl.h>
#include <sys/sysctl.h>
#include <sys/poll.h>
#include <sys/selinfo.h>
#include "fuse.h"
#include "fuse_internal.h"
#include "fuse_ipc.h"
SDT_PROVIDER_DECLARE(fusefs);
/*
* Fuse trace probe:
* arg0: verbosity. Higher numbers give more verbose messages
* arg1: Textual message
*/
SDT_PROBE_DEFINE2(fusefs, , device, trace, "int", "char*");
static struct cdev *fuse_dev;
static d_kqfilter_t fuse_device_filter;
static d_open_t fuse_device_open;
static d_poll_t fuse_device_poll;
static d_read_t fuse_device_read;
static d_write_t fuse_device_write;
static struct cdevsw fuse_device_cdevsw = {
.d_kqfilter = fuse_device_filter,
.d_open = fuse_device_open,
.d_name = "fuse",
.d_poll = fuse_device_poll,
.d_read = fuse_device_read,
.d_write = fuse_device_write,
.d_version = D_VERSION,
};
static int fuse_device_filt_read(struct knote *kn, long hint);
static void fuse_device_filt_detach(struct knote *kn);
struct filterops fuse_device_rfiltops = {
.f_isfd = 1,
.f_detach = fuse_device_filt_detach,
.f_event = fuse_device_filt_read,
};
/****************************
*
* >>> Fuse device op defs
*
****************************/
static void
fdata_dtor(void *arg)
{
struct fuse_data *fdata;
struct fuse_ticket *tick;
fdata = arg;
if (fdata == NULL)
return;
fdata_set_dead(fdata);
FUSE_LOCK();
fuse_lck_mtx_lock(fdata->aw_mtx);
/* wakup poll()ers */
selwakeuppri(&fdata->ks_rsel, PZERO + 1);
/* Don't let syscall handlers wait in vain */
while ((tick = fuse_aw_pop(fdata))) {
fuse_lck_mtx_lock(tick->tk_aw_mtx);
fticket_set_answered(tick);
tick->tk_aw_errno = ENOTCONN;
wakeup(tick);
fuse_lck_mtx_unlock(tick->tk_aw_mtx);
FUSE_ASSERT_AW_DONE(tick);
fuse_ticket_drop(tick);
}
fuse_lck_mtx_unlock(fdata->aw_mtx);
/* Cleanup unsent operations */
fuse_lck_mtx_lock(fdata->ms_mtx);
while ((tick = fuse_ms_pop(fdata))) {
fuse_ticket_drop(tick);
}
fuse_lck_mtx_unlock(fdata->ms_mtx);
FUSE_UNLOCK();
fdata_trydestroy(fdata);
}
static int
fuse_device_filter(struct cdev *dev, struct knote *kn)
{
struct fuse_data *data;
int error;
error = devfs_get_cdevpriv((void **)&data);
/* EVFILT_WRITE is not supported; the device is always ready to write */
if (error == 0 && kn->kn_filter == EVFILT_READ) {
kn->kn_fop = &fuse_device_rfiltops;
kn->kn_hook = data;
knlist_add(&data->ks_rsel.si_note, kn, 0);
error = 0;
} else if (error == 0) {
error = EINVAL;
kn->kn_data = error;
}
return (error);
}
static void
fuse_device_filt_detach(struct knote *kn)
{
struct fuse_data *data;
data = (struct fuse_data*)kn->kn_hook;
MPASS(data != NULL);
knlist_remove(&data->ks_rsel.si_note, kn, 0);
kn->kn_hook = NULL;
}
static int
fuse_device_filt_read(struct knote *kn, long hint)
{
struct fuse_data *data;
int ready;
data = (struct fuse_data*)kn->kn_hook;
MPASS(data != NULL);
mtx_assert(&data->ms_mtx, MA_OWNED);
if (fdata_get_dead(data)) {
kn->kn_flags |= EV_EOF;
kn->kn_fflags = ENODEV;
kn->kn_data = 1;
ready = 1;
} else if (STAILQ_FIRST(&data->ms_head)) {
MPASS(data->ms_count >= 1);
kn->kn_data = data->ms_count;
ready = 1;
} else {
ready = 0;
}
return (ready);
}
/*
* Resources are set up on a per-open basis
*/
static int
fuse_device_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
{
struct fuse_data *fdata;
int error;
SDT_PROBE2(fusefs, , device, trace, 1, "device open");
fdata = fdata_alloc(dev, td->td_ucred);
error = devfs_set_cdevpriv(fdata, fdata_dtor);
if (error != 0)
fdata_trydestroy(fdata);
else
SDT_PROBE2(fusefs, , device, trace, 1, "device open success");
return (error);
}
int
fuse_device_poll(struct cdev *dev, int events, struct thread *td)
{
struct fuse_data *data;
int error, revents = 0;
error = devfs_get_cdevpriv((void **)&data);
if (error != 0)
return (events &
(POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
if (events & (POLLIN | POLLRDNORM)) {
fuse_lck_mtx_lock(data->ms_mtx);
if (fdata_get_dead(data) || STAILQ_FIRST(&data->ms_head))
revents |= events & (POLLIN | POLLRDNORM);
else
selrecord(td, &data->ks_rsel);
fuse_lck_mtx_unlock(data->ms_mtx);
}
if (events & (POLLOUT | POLLWRNORM)) {
revents |= events & (POLLOUT | POLLWRNORM);
}
return (revents);
}
/*
* fuse_device_read hangs on the queue of VFS messages.
* When it's notified that there is a new one, it picks that and
* passes up to the daemon
*/
int
fuse_device_read(struct cdev *dev, struct uio *uio, int ioflag)
{
int err;
struct fuse_data *data;
struct fuse_ticket *tick;
void *buf[] = {NULL, NULL, NULL};
int buflen[3];
int i;
SDT_PROBE2(fusefs, , device, trace, 1, "fuse device read");
err = devfs_get_cdevpriv((void **)&data);
if (err != 0)
return (err);
fuse_lck_mtx_lock(data->ms_mtx);
again:
if (fdata_get_dead(data)) {
SDT_PROBE2(fusefs, , device, trace, 2,
"we know early on that reader should be kicked so we "
"don't wait for news");
fuse_lck_mtx_unlock(data->ms_mtx);
return (ENODEV);
}
if (!(tick = fuse_ms_pop(data))) {
/* check if we may block */
if (ioflag & O_NONBLOCK) {
/* get outa here soon */
fuse_lck_mtx_unlock(data->ms_mtx);
return (EAGAIN);
} else {
err = msleep(data, &data->ms_mtx, PCATCH, "fu_msg", 0);
if (err != 0) {
fuse_lck_mtx_unlock(data->ms_mtx);
return (fdata_get_dead(data) ? ENODEV : err);
}
tick = fuse_ms_pop(data);
}
}
if (!tick) {
/*
* We can get here if fuse daemon suddenly terminates,
* eg, by being hit by a SIGKILL
* -- and some other cases, too, tho not totally clear, when
* (cv_signal/wakeup_one signals the whole process ?)
*/
SDT_PROBE2(fusefs, , device, trace, 1, "no message on thread");
goto again;
}
fuse_lck_mtx_unlock(data->ms_mtx);
if (fdata_get_dead(data)) {
/*
* somebody somewhere -- eg., umount routine --
* wants this liaison finished off
*/
SDT_PROBE2(fusefs, , device, trace, 2,
"reader is to be sacked");
if (tick) {
SDT_PROBE2(fusefs, , device, trace, 2, "weird -- "
"\"kick\" is set tho there is message");
FUSE_ASSERT_MS_DONE(tick);
fuse_ticket_drop(tick);
}
return (ENODEV); /* This should make the daemon get off
* of us */
}
SDT_PROBE2(fusefs, , device, trace, 1,
"fuse device read message successfully");
KASSERT(tick->tk_ms_bufdata || tick->tk_ms_bufsize == 0,
("non-null buf pointer with positive size"));
switch (tick->tk_ms_type) {
case FT_M_FIOV:
buf[0] = tick->tk_ms_fiov.base;
buflen[0] = tick->tk_ms_fiov.len;
break;
case FT_M_BUF:
buf[0] = tick->tk_ms_fiov.base;
buflen[0] = tick->tk_ms_fiov.len;
buf[1] = tick->tk_ms_bufdata;
buflen[1] = tick->tk_ms_bufsize;
break;
default:
panic("unknown message type for fuse_ticket %p", tick);
}
for (i = 0; buf[i]; i++) {
/*
* Why not ban mercilessly stupid daemons who can't keep up
* with us? (There is no much use of a partial read here...)
*/
/*
* XXX note that in such cases Linux FUSE throws EIO at the
* syscall invoker and stands back to the message queue. The
* rationale should be made clear (and possibly adopt that
* behaviour). Keeping the current scheme at least makes
* fallacy as loud as possible...
*/
if (uio->uio_resid < buflen[i]) {
fdata_set_dead(data);
SDT_PROBE2(fusefs, , device, trace, 2,
"daemon is stupid, kick it off...");
err = ENODEV;
break;
}
err = uiomove(buf[i], buflen[i], uio);
if (err)
break;
}
FUSE_ASSERT_MS_DONE(tick);
fuse_ticket_drop(tick);
return (err);
}
static inline int
fuse_ohead_audit(struct fuse_out_header *ohead, struct uio *uio)
{
if (uio->uio_resid + sizeof(struct fuse_out_header) != ohead->len) {
SDT_PROBE2(fusefs, , device, trace, 1,
"Format error: body size "
"differs from size claimed by header");
return (EINVAL);
}
if (uio->uio_resid && ohead->unique != 0 && ohead->error) {
SDT_PROBE2(fusefs, , device, trace, 1,
"Format error: non zero error but message had a body");
return (EINVAL);
}
return (0);
}
SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_notify,
"struct fuse_out_header*");
SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_missing_ticket,
"uint64_t");
SDT_PROBE_DEFINE1(fusefs, , device, fuse_device_write_found,
"struct fuse_ticket*");
/*
* fuse_device_write first reads the header sent by the daemon.
* If that's OK, looks up ticket/callback node by the unique id seen in header.
* If the callback node contains a handler function, the uio is passed over
* that.
*/
static int
fuse_device_write(struct cdev *dev, struct uio *uio, int ioflag)
{
struct fuse_out_header ohead;
int err = 0;
struct fuse_data *data;
struct mount *mp;
struct fuse_ticket *tick, *itick, *x_tick;
int found = 0;
err = devfs_get_cdevpriv((void **)&data);
if (err != 0)
return (err);
mp = data->mp;
if (uio->uio_resid < sizeof(struct fuse_out_header)) {
SDT_PROBE2(fusefs, , device, trace, 1,
"fuse_device_write got less than a header!");
fdata_set_dead(data);
return (EINVAL);
}
if ((err = uiomove(&ohead, sizeof(struct fuse_out_header), uio)) != 0)
return (err);
/*
* We check header information (which is redundant) and compare it
* with what we see. If we see some inconsistency we discard the
* whole answer and proceed on as if it had never existed. In
* particular, no pretender will be woken up, regardless the
* "unique" value in the header.
*/
if ((err = fuse_ohead_audit(&ohead, uio))) {
fdata_set_dead(data);
return (err);
}
/* Pass stuff over to callback if there is one installed */
/* Looking for ticket with the unique id of header */
fuse_lck_mtx_lock(data->aw_mtx);
TAILQ_FOREACH_SAFE(tick, &data->aw_head, tk_aw_link,
x_tick) {
if (tick->tk_unique == ohead.unique) {
SDT_PROBE1(fusefs, , device, fuse_device_write_found,
tick);
found = 1;
fuse_aw_remove(tick);
break;
}
}
if (found && tick->irq_unique > 0) {
/*
* Discard the FUSE_INTERRUPT ticket that tried to interrupt
* this operation
*/
TAILQ_FOREACH_SAFE(itick, &data->aw_head, tk_aw_link,
x_tick) {
if (itick->tk_unique == tick->irq_unique) {
fuse_aw_remove(itick);
fuse_ticket_drop(itick);
break;
}
}
tick->irq_unique = 0;
}
fuse_lck_mtx_unlock(data->aw_mtx);
if (found) {
if (tick->tk_aw_handler) {
/*
* We found a callback with proper handler. In this
* case the out header will be 0wnd by the callback,
* so the fun of freeing that is left for her.
* (Then, by all chance, she'll just get that's done
* via ticket_drop(), so no manual mucking
* around...)
*/
SDT_PROBE2(fusefs, , device, trace, 1,
"pass ticket to a callback");
/* Sanitize the linuxism of negative errnos */
ohead.error *= -1;
memcpy(&tick->tk_aw_ohead, &ohead, sizeof(ohead));
err = tick->tk_aw_handler(tick, uio);
} else {
/* pretender doesn't wanna do anything with answer */
SDT_PROBE2(fusefs, , device, trace, 1,
"stuff devalidated, so we drop it");
}
/*
* As aw_mtx was not held during the callback execution the
* ticket may have been inserted again. However, this is safe
* because fuse_ticket_drop() will deal with refcount anyway.
*/
fuse_ticket_drop(tick);
} else if (ohead.unique == 0){
/* unique == 0 means asynchronous notification */
SDT_PROBE1(fusefs, , device, fuse_device_write_notify, &ohead);
switch (ohead.error) {
case FUSE_NOTIFY_INVAL_ENTRY:
err = fuse_internal_invalidate_entry(mp, uio);
break;
case FUSE_NOTIFY_INVAL_INODE:
err = fuse_internal_invalidate_inode(mp, uio);
break;
case FUSE_NOTIFY_RETRIEVE:
case FUSE_NOTIFY_STORE:
/*
* Unimplemented. I don't know of any file systems
* that use them, and the protocol isn't sound anyway,
* since the notification messages don't include the
* inode's generation number. Without that, it's
* possible to manipulate the cache of the wrong vnode.
* Finally, it's not defined what this message should
* do for a file with dirty cache.
*/
case FUSE_NOTIFY_POLL:
/* Unimplemented. See comments in fuse_vnops */
default:
/* Not implemented */
err = ENOSYS;
}
} else {
/* no callback at all! */
SDT_PROBE1(fusefs, , device, fuse_device_write_missing_ticket,
ohead.unique);
if (ohead.error == -EAGAIN) {
/*
* This was probably a response to a FUSE_INTERRUPT
* operation whose original operation is already
* complete. We can't store FUSE_INTERRUPT tickets
* indefinitely because their responses are optional.
* So we delete them when the original operation
* completes. And sadly the fuse_header_out doesn't
* identify the opcode, so we have to guess.
*/
err = 0;
} else {
err = EINVAL;
}
}
return (err);
}
int
fuse_device_init(void)
{
fuse_dev = make_dev(&fuse_device_cdevsw, 0, UID_ROOT, GID_OPERATOR,
S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH, "fuse");
if (fuse_dev == NULL)
return (ENOMEM);
return (0);
}
void
fuse_device_destroy(void)
{
MPASS(fuse_dev != NULL);
destroy_dev(fuse_dev);
}