linux/fs/notify/dnotify/dnotify.c

399 lines
11 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* Directory notifications for Linux.
*
* Copyright (C) 2000,2001,2002 Stephen Rothwell
*
* Copyright (C) 2009 Eric Paris <Red Hat Inc>
* dnotify was largly rewritten to use the new fsnotify infrastructure
*/
#include <linux/fs.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/sched/signal.h>
#include <linux/dnotify.h>
#include <linux/init.h>
fanotify, inotify, dnotify, security: add security hook for fs notifications As of now, setting watches on filesystem objects has, at most, applied a check for read access to the inode, and in the case of fanotify, requires CAP_SYS_ADMIN. No specific security hook or permission check has been provided to control the setting of watches. Using any of inotify, dnotify, or fanotify, it is possible to observe, not only write-like operations, but even read access to a file. Modeling the watch as being merely a read from the file is insufficient for the needs of SELinux. This is due to the fact that read access should not necessarily imply access to information about when another process reads from a file. Furthermore, fanotify watches grant more power to an application in the form of permission events. While notification events are solely, unidirectional (i.e. they only pass information to the receiving application), permission events are blocking. Permission events make a request to the receiving application which will then reply with a decision as to whether or not that action may be completed. This causes the issue of the watching application having the ability to exercise control over the triggering process. Without drawing a distinction within the permission check, the ability to read would imply the greater ability to control an application. Additionally, mount and superblock watches apply to all files within the same mount or superblock. Read access to one file should not necessarily imply the ability to watch all files accessed within a given mount or superblock. In order to solve these issues, a new LSM hook is implemented and has been placed within the system calls for marking filesystem objects with inotify, fanotify, and dnotify watches. These calls to the hook are placed at the point at which the target path has been resolved and are provided with the path struct, the mask of requested notification events, and the type of object on which the mark is being set (inode, superblock, or mount). The mask and obj_type have already been translated into common FS_* values shared by the entirety of the fs notification infrastructure. The path struct is passed rather than just the inode so that the mount is available, particularly for mount watches. This also allows for use of the hook by pathname-based security modules. However, since the hook is intended for use even by inode based security modules, it is not placed under the CONFIG_SECURITY_PATH conditional. Otherwise, the inode-based security modules would need to enable all of the path hooks, even though they do not use any of them. This only provides a hook at the point of setting a watch, and presumes that permission to set a particular watch implies the ability to receive all notification about that object which match the mask. This is all that is required for SELinux. If other security modules require additional hooks or infrastructure to control delivery of notification, these can be added by them. It does not make sense for us to propose hooks for which we have no implementation. The understanding that all notifications received by the requesting application are all strictly of a type for which the application has been granted permission shows that this implementation is sufficient in its coverage. Security modules wishing to provide complete control over fanotify must also implement a security_file_open hook that validates that the access requested by the watching application is authorized. Fanotify has the issue that it returns a file descriptor with the file mode specified during fanotify_init() to the watching process on event. This is already covered by the LSM security_file_open hook if the security module implements checking of the requested file mode there. Otherwise, a watching process can obtain escalated access to a file for which it has not been authorized. The selinux_path_notify hook implementation works by adding five new file permissions: watch, watch_mount, watch_sb, watch_reads, and watch_with_perm (descriptions about which will follow), and one new filesystem permission: watch (which is applied to superblock checks). The hook then decides which subset of these permissions must be held by the requesting application based on the contents of the provided mask and the obj_type. The selinux_file_open hook already checks the requested file mode and therefore ensures that a watching process cannot escalate its access through fanotify. The watch, watch_mount, and watch_sb permissions are the baseline permissions for setting a watch on an object and each are a requirement for any watch to be set on a file, mount, or superblock respectively. It should be noted that having either of the other two permissions (watch_reads and watch_with_perm) does not imply the watch, watch_mount, or watch_sb permission. Superblock watches further require the filesystem watch permission to the superblock. As there is no labeled object in view for mounts, there is no specific check for mount watches beyond watch_mount to the inode. Such a check could be added in the future, if a suitable labeled object existed representing the mount. The watch_reads permission is required to receive notifications from read-exclusive events on filesystem objects. These events include accessing a file for the purpose of reading and closing a file which has been opened read-only. This distinction has been drawn in order to provide a direct indication in the policy for this otherwise not obvious capability. Read access to a file should not necessarily imply the ability to observe read events on a file. Finally, watch_with_perm only applies to fanotify masks since it is the only way to set a mask which allows for the blocking, permission event. This permission is needed for any watch which is of this type. Though fanotify requires CAP_SYS_ADMIN, this is insufficient as it gives implicit trust to root, which we do not do, and does not support least privilege. Signed-off-by: Aaron Goidel <acgoide@tycho.nsa.gov> Acked-by: Casey Schaufler <casey@schaufler-ca.com> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-08-12 23:20:00 +08:00
#include <linux/security.h>
#include <linux/spinlock.h>
#include <linux/slab.h>
#include <linux/fdtable.h>
#include <linux/fsnotify_backend.h>
int dir_notify_enable __read_mostly = 1;
static struct kmem_cache *dnotify_struct_cache __read_mostly;
static struct kmem_cache *dnotify_mark_cache __read_mostly;
static struct fsnotify_group *dnotify_group __read_mostly;
/*
* dnotify will attach one of these to each inode (i_fsnotify_marks) which
* is being watched by dnotify. If multiple userspace applications are watching
* the same directory with dnotify their information is chained in dn
*/
struct dnotify_mark {
struct fsnotify_mark fsn_mark;
struct dnotify_struct *dn;
};
/*
* When a process starts or stops watching an inode the set of events which
* dnotify cares about for that inode may change. This function runs the
* list of everything receiving dnotify events about this directory and calculates
* the set of all those events. After it updates what dnotify is interested in
* it calls the fsnotify function so it can update the set of all events relevant
* to this inode.
*/
static void dnotify_recalc_inode_mask(struct fsnotify_mark *fsn_mark)
{
__u32 new_mask = 0;
struct dnotify_struct *dn;
struct dnotify_mark *dn_mark = container_of(fsn_mark,
struct dnotify_mark,
fsn_mark);
assert_spin_locked(&fsn_mark->lock);
for (dn = dn_mark->dn; dn != NULL; dn = dn->dn_next)
new_mask |= (dn->dn_mask & ~FS_DN_MULTISHOT);
if (fsn_mark->mask == new_mask)
return;
fsn_mark->mask = new_mask;
fsnotify_recalc_mask(fsn_mark->connector);
}
/*
* Mains fsnotify call where events are delivered to dnotify.
* Find the dnotify mark on the relevant inode, run the list of dnotify structs
* on that mark and determine which of them has expressed interest in receiving
* events of this type. When found send the correct process and signal and
* destroy the dnotify struct if it was not registered to receive multiple
* events.
*/
static int dnotify_handle_event(struct fsnotify_group *group,
fsnotify: do not share events between notification groups Currently fsnotify framework creates one event structure for each notification event and links this event into all interested notification groups. This is done so that we save memory when several notification groups are interested in the event. However the need for event structure shared between inotify & fanotify bloats the event structure so the result is often higher memory consumption. Another problem is that fsnotify framework keeps path references with outstanding events so that fanotify can return open file descriptors with its events. This has the undesirable effect that filesystem cannot be unmounted while there are outstanding events - a regression for inotify compared to a situation before it was converted to fsnotify framework. For fanotify this problem is hard to avoid and users of fanotify should kind of expect this behavior when they ask for file descriptors from notified files. This patch changes fsnotify and its users to create separate event structure for each group. This allows for much simpler code (~400 lines removed by this patch) and also smaller event structures. For example on 64-bit system original struct fsnotify_event consumes 120 bytes, plus additional space for file name, additional 24 bytes for second and each subsequent group linking the event, and additional 32 bytes for each inotify group for private data. After the conversion inotify event consumes 48 bytes plus space for file name which is considerably less memory unless file names are long and there are several groups interested in the events (both of which are uncommon). Fanotify event fits in 56 bytes after the conversion (fanotify doesn't care about file names so its events don't have to have it allocated). A win unless there are four or more fanotify groups interested in the event. The conversion also solves the problem with unmount when only inotify is used as we don't have to grab path references for inotify events. [hughd@google.com: fanotify: fix corruption preventing startup] Signed-off-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Cc: Eric Paris <eparis@parisplace.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-22 07:48:14 +08:00
struct inode *inode,
u32 mask, const void *data, int data_type,
const struct qstr *file_name, u32 cookie,
struct fsnotify_iter_info *iter_info)
{
struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info);
struct dnotify_mark *dn_mark;
struct dnotify_struct *dn;
struct dnotify_struct **prev;
struct fown_struct *fown;
fsnotify: do not share events between notification groups Currently fsnotify framework creates one event structure for each notification event and links this event into all interested notification groups. This is done so that we save memory when several notification groups are interested in the event. However the need for event structure shared between inotify & fanotify bloats the event structure so the result is often higher memory consumption. Another problem is that fsnotify framework keeps path references with outstanding events so that fanotify can return open file descriptors with its events. This has the undesirable effect that filesystem cannot be unmounted while there are outstanding events - a regression for inotify compared to a situation before it was converted to fsnotify framework. For fanotify this problem is hard to avoid and users of fanotify should kind of expect this behavior when they ask for file descriptors from notified files. This patch changes fsnotify and its users to create separate event structure for each group. This allows for much simpler code (~400 lines removed by this patch) and also smaller event structures. For example on 64-bit system original struct fsnotify_event consumes 120 bytes, plus additional space for file name, additional 24 bytes for second and each subsequent group linking the event, and additional 32 bytes for each inotify group for private data. After the conversion inotify event consumes 48 bytes plus space for file name which is considerably less memory unless file names are long and there are several groups interested in the events (both of which are uncommon). Fanotify event fits in 56 bytes after the conversion (fanotify doesn't care about file names so its events don't have to have it allocated). A win unless there are four or more fanotify groups interested in the event. The conversion also solves the problem with unmount when only inotify is used as we don't have to grab path references for inotify events. [hughd@google.com: fanotify: fix corruption preventing startup] Signed-off-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Cc: Eric Paris <eparis@parisplace.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-22 07:48:14 +08:00
__u32 test_mask = mask & ~FS_EVENT_ON_CHILD;
/* not a dir, dnotify doesn't care */
if (!S_ISDIR(inode->i_mode))
return 0;
if (WARN_ON(fsnotify_iter_vfsmount_mark(iter_info)))
return 0;
dn_mark = container_of(inode_mark, struct dnotify_mark, fsn_mark);
spin_lock(&inode_mark->lock);
prev = &dn_mark->dn;
while ((dn = *prev) != NULL) {
dnotify: ignore FS_EVENT_ON_CHILD Mask off FS_EVENT_ON_CHILD in dnotify_handle_event(). Otherwise, when there is more than one watch on a directory and dnotify_should_send_event() succeeds, events with FS_EVENT_ON_CHILD set will trigger all watches and cause spurious events. This case was overlooked in commit e42e2773. #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <unistd.h> #include <signal.h> #include <sys/types.h> #include <sys/stat.h> #include <fcntl.h> #include <string.h> static void create_event(int s, siginfo_t* si, void* p) { printf("create\n"); } static void delete_event(int s, siginfo_t* si, void* p) { printf("delete\n"); } int main (void) { struct sigaction action; char *tmpdir, *file; int fd1, fd2; sigemptyset (&action.sa_mask); action.sa_flags = SA_SIGINFO; action.sa_sigaction = create_event; sigaction (SIGRTMIN + 0, &action, NULL); action.sa_sigaction = delete_event; sigaction (SIGRTMIN + 1, &action, NULL); # define TMPDIR "/tmp/test.XXXXXX" tmpdir = malloc(strlen(TMPDIR) + 1); strcpy(tmpdir, TMPDIR); mkdtemp(tmpdir); # define TMPFILE "/file" file = malloc(strlen(tmpdir) + strlen(TMPFILE) + 1); sprintf(file, "%s/%s", tmpdir, TMPFILE); fd1 = open (tmpdir, O_RDONLY); fcntl(fd1, F_SETSIG, SIGRTMIN); fcntl(fd1, F_NOTIFY, DN_MULTISHOT | DN_CREATE); fd2 = open (tmpdir, O_RDONLY); fcntl(fd2, F_SETSIG, SIGRTMIN + 1); fcntl(fd2, F_NOTIFY, DN_MULTISHOT | DN_DELETE); if (fork()) { /* This triggers a create event */ creat(file, 0600); /* This triggers a create and delete event (!) */ unlink(file); } else { sleep(1); rmdir(tmpdir); } return 0; } Signed-off-by: Andreas Gruenbacher <agruen@suse.de> Signed-off-by: Eric Paris <eparis@redhat.com>
2009-10-15 06:13:23 +08:00
if ((dn->dn_mask & test_mask) == 0) {
prev = &dn->dn_next;
continue;
}
fown = &dn->dn_filp->f_owner;
send_sigio(fown, dn->dn_fd, POLL_MSG);
if (dn->dn_mask & FS_DN_MULTISHOT)
prev = &dn->dn_next;
else {
*prev = dn->dn_next;
kmem_cache_free(dnotify_struct_cache, dn);
dnotify_recalc_inode_mask(inode_mark);
}
}
spin_unlock(&inode_mark->lock);
return 0;
}
static void dnotify_free_mark(struct fsnotify_mark *fsn_mark)
{
struct dnotify_mark *dn_mark = container_of(fsn_mark,
struct dnotify_mark,
fsn_mark);
BUG_ON(dn_mark->dn);
kmem_cache_free(dnotify_mark_cache, dn_mark);
}
static const struct fsnotify_ops dnotify_fsnotify_ops = {
.handle_event = dnotify_handle_event,
.free_mark = dnotify_free_mark,
};
/*
* Called every time a file is closed. Looks first for a dnotify mark on the
* inode. If one is found run all of the ->dn structures attached to that
* mark for one relevant to this process closing the file and remove that
* dnotify_struct. If that was the last dnotify_struct also remove the
* fsnotify_mark.
*/
void dnotify_flush(struct file *filp, fl_owner_t id)
{
struct fsnotify_mark *fsn_mark;
struct dnotify_mark *dn_mark;
struct dnotify_struct *dn;
struct dnotify_struct **prev;
struct inode *inode;
bool free = false;
inode = file_inode(filp);
if (!S_ISDIR(inode->i_mode))
return;
fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, dnotify_group);
if (!fsn_mark)
return;
dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
mutex_lock(&dnotify_group->mark_mutex);
spin_lock(&fsn_mark->lock);
prev = &dn_mark->dn;
while ((dn = *prev) != NULL) {
if ((dn->dn_owner == id) && (dn->dn_filp == filp)) {
*prev = dn->dn_next;
kmem_cache_free(dnotify_struct_cache, dn);
dnotify_recalc_inode_mask(fsn_mark);
break;
}
prev = &dn->dn_next;
}
spin_unlock(&fsn_mark->lock);
/* nothing else could have found us thanks to the dnotify_groups
mark_mutex */
if (dn_mark->dn == NULL) {
fsnotify_detach_mark(fsn_mark);
free = true;
}
mutex_unlock(&dnotify_group->mark_mutex);
if (free)
fsnotify_free_mark(fsn_mark);
fsnotify_put_mark(fsn_mark);
}
/* this conversion is done only at watch creation */
static __u32 convert_arg(unsigned long arg)
{
__u32 new_mask = FS_EVENT_ON_CHILD;
if (arg & DN_MULTISHOT)
new_mask |= FS_DN_MULTISHOT;
if (arg & DN_DELETE)
new_mask |= (FS_DELETE | FS_MOVED_FROM);
if (arg & DN_MODIFY)
new_mask |= FS_MODIFY;
if (arg & DN_ACCESS)
new_mask |= FS_ACCESS;
if (arg & DN_ATTRIB)
new_mask |= FS_ATTRIB;
if (arg & DN_RENAME)
new_mask |= FS_DN_RENAME;
if (arg & DN_CREATE)
new_mask |= (FS_CREATE | FS_MOVED_TO);
return new_mask;
}
/*
* If multiple processes watch the same inode with dnotify there is only one
* dnotify mark in inode->i_fsnotify_marks but we chain a dnotify_struct
* onto that mark. This function either attaches the new dnotify_struct onto
* that list, or it |= the mask onto an existing dnofiy_struct.
*/
static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark *dn_mark,
fl_owner_t id, int fd, struct file *filp, __u32 mask)
{
struct dnotify_struct *odn;
odn = dn_mark->dn;
while (odn != NULL) {
/* adding more events to existing dnofiy_struct? */
if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
odn->dn_fd = fd;
odn->dn_mask |= mask;
return -EEXIST;
}
odn = odn->dn_next;
}
dn->dn_mask = mask;
dn->dn_fd = fd;
dn->dn_filp = filp;
dn->dn_owner = id;
dn->dn_next = dn_mark->dn;
dn_mark->dn = dn;
return 0;
}
/*
* When a process calls fcntl to attach a dnotify watch to a directory it ends
* up here. Allocate both a mark for fsnotify to add and a dnotify_struct to be
* attached to the fsnotify_mark.
*/
int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
{
struct dnotify_mark *new_dn_mark, *dn_mark;
struct fsnotify_mark *new_fsn_mark, *fsn_mark;
struct dnotify_struct *dn;
struct inode *inode;
fl_owner_t id = current->files;
struct file *f;
int destroy = 0, error = 0;
__u32 mask;
/* we use these to tell if we need to kfree */
new_fsn_mark = NULL;
dn = NULL;
if (!dir_notify_enable) {
error = -EINVAL;
goto out_err;
}
/* a 0 mask means we are explicitly removing the watch */
if ((arg & ~DN_MULTISHOT) == 0) {
dnotify_flush(filp, id);
error = 0;
goto out_err;
}
/* dnotify only works on directories */
inode = file_inode(filp);
if (!S_ISDIR(inode->i_mode)) {
error = -ENOTDIR;
goto out_err;
}
fanotify, inotify, dnotify, security: add security hook for fs notifications As of now, setting watches on filesystem objects has, at most, applied a check for read access to the inode, and in the case of fanotify, requires CAP_SYS_ADMIN. No specific security hook or permission check has been provided to control the setting of watches. Using any of inotify, dnotify, or fanotify, it is possible to observe, not only write-like operations, but even read access to a file. Modeling the watch as being merely a read from the file is insufficient for the needs of SELinux. This is due to the fact that read access should not necessarily imply access to information about when another process reads from a file. Furthermore, fanotify watches grant more power to an application in the form of permission events. While notification events are solely, unidirectional (i.e. they only pass information to the receiving application), permission events are blocking. Permission events make a request to the receiving application which will then reply with a decision as to whether or not that action may be completed. This causes the issue of the watching application having the ability to exercise control over the triggering process. Without drawing a distinction within the permission check, the ability to read would imply the greater ability to control an application. Additionally, mount and superblock watches apply to all files within the same mount or superblock. Read access to one file should not necessarily imply the ability to watch all files accessed within a given mount or superblock. In order to solve these issues, a new LSM hook is implemented and has been placed within the system calls for marking filesystem objects with inotify, fanotify, and dnotify watches. These calls to the hook are placed at the point at which the target path has been resolved and are provided with the path struct, the mask of requested notification events, and the type of object on which the mark is being set (inode, superblock, or mount). The mask and obj_type have already been translated into common FS_* values shared by the entirety of the fs notification infrastructure. The path struct is passed rather than just the inode so that the mount is available, particularly for mount watches. This also allows for use of the hook by pathname-based security modules. However, since the hook is intended for use even by inode based security modules, it is not placed under the CONFIG_SECURITY_PATH conditional. Otherwise, the inode-based security modules would need to enable all of the path hooks, even though they do not use any of them. This only provides a hook at the point of setting a watch, and presumes that permission to set a particular watch implies the ability to receive all notification about that object which match the mask. This is all that is required for SELinux. If other security modules require additional hooks or infrastructure to control delivery of notification, these can be added by them. It does not make sense for us to propose hooks for which we have no implementation. The understanding that all notifications received by the requesting application are all strictly of a type for which the application has been granted permission shows that this implementation is sufficient in its coverage. Security modules wishing to provide complete control over fanotify must also implement a security_file_open hook that validates that the access requested by the watching application is authorized. Fanotify has the issue that it returns a file descriptor with the file mode specified during fanotify_init() to the watching process on event. This is already covered by the LSM security_file_open hook if the security module implements checking of the requested file mode there. Otherwise, a watching process can obtain escalated access to a file for which it has not been authorized. The selinux_path_notify hook implementation works by adding five new file permissions: watch, watch_mount, watch_sb, watch_reads, and watch_with_perm (descriptions about which will follow), and one new filesystem permission: watch (which is applied to superblock checks). The hook then decides which subset of these permissions must be held by the requesting application based on the contents of the provided mask and the obj_type. The selinux_file_open hook already checks the requested file mode and therefore ensures that a watching process cannot escalate its access through fanotify. The watch, watch_mount, and watch_sb permissions are the baseline permissions for setting a watch on an object and each are a requirement for any watch to be set on a file, mount, or superblock respectively. It should be noted that having either of the other two permissions (watch_reads and watch_with_perm) does not imply the watch, watch_mount, or watch_sb permission. Superblock watches further require the filesystem watch permission to the superblock. As there is no labeled object in view for mounts, there is no specific check for mount watches beyond watch_mount to the inode. Such a check could be added in the future, if a suitable labeled object existed representing the mount. The watch_reads permission is required to receive notifications from read-exclusive events on filesystem objects. These events include accessing a file for the purpose of reading and closing a file which has been opened read-only. This distinction has been drawn in order to provide a direct indication in the policy for this otherwise not obvious capability. Read access to a file should not necessarily imply the ability to observe read events on a file. Finally, watch_with_perm only applies to fanotify masks since it is the only way to set a mask which allows for the blocking, permission event. This permission is needed for any watch which is of this type. Though fanotify requires CAP_SYS_ADMIN, this is insufficient as it gives implicit trust to root, which we do not do, and does not support least privilege. Signed-off-by: Aaron Goidel <acgoide@tycho.nsa.gov> Acked-by: Casey Schaufler <casey@schaufler-ca.com> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: Paul Moore <paul@paul-moore.com>
2019-08-12 23:20:00 +08:00
/*
* convert the userspace DN_* "arg" to the internal FS_*
* defined in fsnotify
*/
mask = convert_arg(arg);
error = security_path_notify(&filp->f_path, mask,
FSNOTIFY_OBJ_TYPE_INODE);
if (error)
goto out_err;
/* expect most fcntl to add new rather than augment old */
dn = kmem_cache_alloc(dnotify_struct_cache, GFP_KERNEL);
if (!dn) {
error = -ENOMEM;
goto out_err;
}
/* new fsnotify mark, we expect most fcntl calls to add a new mark */
new_dn_mark = kmem_cache_alloc(dnotify_mark_cache, GFP_KERNEL);
if (!new_dn_mark) {
error = -ENOMEM;
goto out_err;
}
/* set up the new_fsn_mark and new_dn_mark */
new_fsn_mark = &new_dn_mark->fsn_mark;
fsnotify_init_mark(new_fsn_mark, dnotify_group);
new_fsn_mark->mask = mask;
new_dn_mark->dn = NULL;
/* this is needed to prevent the fcntl/close race described below */
mutex_lock(&dnotify_group->mark_mutex);
/* add the new_fsn_mark or find an old one. */
fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, dnotify_group);
if (fsn_mark) {
dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
spin_lock(&fsn_mark->lock);
} else {
error = fsnotify_add_inode_mark_locked(new_fsn_mark, inode, 0);
if (error) {
mutex_unlock(&dnotify_group->mark_mutex);
goto out_err;
}
spin_lock(&new_fsn_mark->lock);
fsn_mark = new_fsn_mark;
dn_mark = new_dn_mark;
/* we used new_fsn_mark, so don't free it */
new_fsn_mark = NULL;
}
rcu_read_lock();
f = fcheck(fd);
rcu_read_unlock();
/* if (f != filp) means that we lost a race and another task/thread
* actually closed the fd we are still playing with before we grabbed
* the dnotify_groups mark_mutex and fsn_mark->lock. Since closing the
* fd is the only time we clean up the marks we need to get our mark
* off the list. */
if (f != filp) {
/* if we added ourselves, shoot ourselves, it's possible that
* the flush actually did shoot this fsn_mark. That's fine too
* since multiple calls to destroy_mark is perfectly safe, if
* we found a dn_mark already attached to the inode, just sod
* off silently as the flush at close time dealt with it.
*/
if (dn_mark == new_dn_mark)
destroy = 1;
error = 0;
goto out;
}
__f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
error = attach_dn(dn, dn_mark, id, fd, filp, mask);
/* !error means that we attached the dn to the dn_mark, so don't free it */
if (!error)
dn = NULL;
/* -EEXIST means that we didn't add this new dn and used an old one.
* that isn't an error (and the unused dn should be freed) */
else if (error == -EEXIST)
error = 0;
dnotify_recalc_inode_mask(fsn_mark);
out:
spin_unlock(&fsn_mark->lock);
if (destroy)
fsnotify_detach_mark(fsn_mark);
mutex_unlock(&dnotify_group->mark_mutex);
if (destroy)
fsnotify_free_mark(fsn_mark);
fsnotify_put_mark(fsn_mark);
out_err:
if (new_fsn_mark)
fsnotify_put_mark(new_fsn_mark);
if (dn)
kmem_cache_free(dnotify_struct_cache, dn);
return error;
}
static int __init dnotify_init(void)
{
fs: fsnotify: account fsnotify metadata to kmemcg Patch series "Directed kmem charging", v8. The Linux kernel's memory cgroup allows limiting the memory usage of the jobs running on the system to provide isolation between the jobs. All the kernel memory allocated in the context of the job and marked with __GFP_ACCOUNT will also be included in the memory usage and be limited by the job's limit. The kernel memory can only be charged to the memcg of the process in whose context kernel memory was allocated. However there are cases where the allocated kernel memory should be charged to the memcg different from the current processes's memcg. This patch series contains two such concrete use-cases i.e. fsnotify and buffer_head. The fsnotify event objects can consume a lot of system memory for large or unlimited queues if there is either no or slow listener. The events are allocated in the context of the event producer. However they should be charged to the event consumer. Similarly the buffer_head objects can be allocated in a memcg different from the memcg of the page for which buffer_head objects are being allocated. To solve this issue, this patch series introduces mechanism to charge kernel memory to a given memcg. In case of fsnotify events, the memcg of the consumer can be used for charging and for buffer_head, the memcg of the page can be charged. For directed charging, the caller can use the scope API memalloc_[un]use_memcg() to specify the memcg to charge for all the __GFP_ACCOUNT allocations within the scope. This patch (of 2): A lot of memory can be consumed by the events generated for the huge or unlimited queues if there is either no or slow listener. This can cause system level memory pressure or OOMs. So, it's better to account the fsnotify kmem caches to the memcg of the listener. However the listener can be in a different memcg than the memcg of the producer and these allocations happen in the context of the event producer. This patch introduces remote memcg charging API which the producer can use to charge the allocations to the memcg of the listener. There are seven fsnotify kmem caches and among them allocations from dnotify_struct_cache, dnotify_mark_cache, fanotify_mark_cache and inotify_inode_mark_cachep happens in the context of syscall from the listener. So, SLAB_ACCOUNT is enough for these caches. The objects from fsnotify_mark_connector_cachep are not accounted as they are small compared to the notification mark or events and it is unclear whom to account connector to since it is shared by all events attached to the inode. The allocations from the event caches happen in the context of the event producer. For such caches we will need to remote charge the allocations to the listener's memcg. Thus we save the memcg reference in the fsnotify_group structure of the listener. This patch has also moved the members of fsnotify_group to keep the size same, at least for 64 bit build, even with additional member by filling the holes. [shakeelb@google.com: use GFP_KERNEL_ACCOUNT rather than open-coding it] Link: http://lkml.kernel.org/r/20180702215439.211597-1-shakeelb@google.com Link: http://lkml.kernel.org/r/20180627191250.209150-2-shakeelb@google.com Signed-off-by: Shakeel Butt <shakeelb@google.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Jan Kara <jack@suse.cz> Cc: Amir Goldstein <amir73il@gmail.com> Cc: Greg Thelen <gthelen@google.com> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Cc: Roman Gushchin <guro@fb.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-08-18 06:46:39 +08:00
dnotify_struct_cache = KMEM_CACHE(dnotify_struct,
SLAB_PANIC|SLAB_ACCOUNT);
dnotify_mark_cache = KMEM_CACHE(dnotify_mark, SLAB_PANIC|SLAB_ACCOUNT);
dnotify_group = fsnotify_alloc_group(&dnotify_fsnotify_ops);
if (IS_ERR(dnotify_group))
panic("unable to allocate fsnotify group for dnotify\n");
return 0;
}
module_init(dnotify_init)