linux/sound/core/control.c

2563 lines
67 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* Routines for driver control interface
* Copyright (c) by Jaroslav Kysela <perex@perex.cz>
*/
#include <linux/threads.h>
#include <linux/interrupt.h>
#include <linux/module.h>
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
#include <linux/moduleparam.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/time.h>
#include <linux/mm.h>
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
#include <linux/math64.h>
#include <linux/sched/signal.h>
#include <sound/core.h>
#include <sound/minors.h>
#include <sound/info.h>
#include <sound/control.h>
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// Max allocation size for user controls.
static int max_user_ctl_alloc_size = 8 * 1024 * 1024;
module_param_named(max_user_ctl_alloc_size, max_user_ctl_alloc_size, int, 0444);
MODULE_PARM_DESC(max_user_ctl_alloc_size, "Max allocation size for user controls");
#define MAX_CONTROL_COUNT 1028
struct snd_kctl_ioctl {
struct list_head list; /* list of all ioctls */
snd_kctl_ioctl_func_t fioctl;
};
static DECLARE_RWSEM(snd_ioctl_rwsem);
static DECLARE_RWSEM(snd_ctl_layer_rwsem);
static LIST_HEAD(snd_control_ioctls);
#ifdef CONFIG_COMPAT
static LIST_HEAD(snd_control_compat_ioctls);
#endif
static struct snd_ctl_layer_ops *snd_ctl_layer;
static int snd_ctl_remove_locked(struct snd_card *card,
struct snd_kcontrol *kcontrol);
static int snd_ctl_open(struct inode *inode, struct file *file)
{
unsigned long flags;
struct snd_card *card;
struct snd_ctl_file *ctl;
int i, err;
*: convert stream-like files from nonseekable_open -> stream_open Using scripts/coccinelle/api/stream_open.cocci added in 10dce8af3422 ("fs: stream_open - opener for stream-like files so that read and write can run simultaneously without deadlock"), search and convert to stream_open all in-kernel nonseekable_open users for which read and write actually do not depend on ppos and where there is no other methods in file_operations which assume @offset access. I've verified each generated change manually - that it is correct to convert - and each other nonseekable_open instance left - that it is either not correct to convert there, or that it is not converted due to current stream_open.cocci limitations. The script also does not convert files that should be valid to convert, but that currently have .llseek = noop_llseek or generic_file_llseek for unknown reason despite file being opened with nonseekable_open (e.g. drivers/input/mousedev.c) Among cases converted 14 were potentially vulnerable to read vs write deadlock (see details in 10dce8af3422): drivers/char/pcmcia/cm4000_cs.c:1685:7-23: ERROR: cm4000_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/gnss/core.c:45:1-17: ERROR: gnss_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/hid/uhid.c:635:1-17: ERROR: uhid_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/infiniband/core/user_mad.c:988:1-17: ERROR: umad_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/input/evdev.c:527:1-17: ERROR: evdev_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/input/misc/uinput.c:401:1-17: ERROR: uinput_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/isdn/capi/capi.c:963:8-24: ERROR: capi_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/leds/uleds.c:77:1-17: ERROR: uleds_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/media/rc/lirc_dev.c:198:1-17: ERROR: lirc_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/s390/char/fs3270.c:488:1-17: ERROR: fs3270_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/usb/misc/ldusb.c:310:1-17: ERROR: ld_usb_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. drivers/xen/evtchn.c:667:8-24: ERROR: evtchn_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. net/batman-adv/icmp_socket.c:80:1-17: ERROR: batadv_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. net/rfkill/core.c:1146:8-24: ERROR: rfkill_fops: .read() can deadlock .write(); change nonseekable_open -> stream_open to fix. and the rest were just safe to convert to stream_open because their read and write do not use ppos at all and corresponding file_operations do not have methods that assume @offset file access(*): arch/powerpc/platforms/52xx/mpc52xx_gpt.c:631:8-24: WARNING: mpc52xx_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. arch/powerpc/platforms/cell/spufs/file.c:591:8-24: WARNING: spufs_ibox_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. arch/powerpc/platforms/cell/spufs/file.c:591:8-24: WARNING: spufs_ibox_stat_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. arch/powerpc/platforms/cell/spufs/file.c:591:8-24: WARNING: spufs_mbox_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. arch/powerpc/platforms/cell/spufs/file.c:591:8-24: WARNING: spufs_mbox_stat_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. arch/powerpc/platforms/cell/spufs/file.c:591:8-24: WARNING: spufs_wbox_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. arch/powerpc/platforms/cell/spufs/file.c:591:8-24: WARNING: spufs_wbox_stat_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. arch/um/drivers/harddog_kern.c:88:8-24: WARNING: harddog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. arch/x86/kernel/cpu/microcode/core.c:430:33-49: WARNING: microcode_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/char/ds1620.c:215:8-24: WARNING: ds1620_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/char/dtlk.c:301:1-17: WARNING: dtlk_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/char/ipmi/ipmi_watchdog.c:840:9-25: WARNING: ipmi_wdog_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/char/pcmcia/scr24x_cs.c:95:8-24: WARNING: scr24x_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/char/tb0219.c:246:9-25: WARNING: tb0219_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/firewire/nosy.c:306:8-24: WARNING: nosy_ops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/hwmon/fschmd.c:840:8-24: WARNING: watchdog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/hwmon/w83793.c:1344:8-24: WARNING: watchdog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/infiniband/core/ucma.c:1747:8-24: WARNING: ucma_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/infiniband/core/ucm.c:1178:8-24: WARNING: ucm_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/infiniband/core/uverbs_main.c:1086:8-24: WARNING: uverbs_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/input/joydev.c:282:1-17: WARNING: joydev_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/pci/switch/switchtec.c:393:1-17: WARNING: switchtec_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/platform/chrome/cros_ec_debugfs.c:135:8-24: WARNING: cros_ec_console_log_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/rtc/rtc-ds1374.c:470:9-25: WARNING: ds1374_wdt_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/rtc/rtc-m41t80.c:805:9-25: WARNING: wdt_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/s390/char/tape_char.c:293:2-18: WARNING: tape_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/s390/char/zcore.c:194:8-24: WARNING: zcore_reipl_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/s390/crypto/zcrypt_api.c:528:8-24: WARNING: zcrypt_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/spi/spidev.c:594:1-17: WARNING: spidev_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/staging/pi433/pi433_if.c:974:1-17: WARNING: pi433_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/acquirewdt.c:203:8-24: WARNING: acq_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/advantechwdt.c:202:8-24: WARNING: advwdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/alim1535_wdt.c:252:8-24: WARNING: ali_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/alim7101_wdt.c:217:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/ar7_wdt.c:166:8-24: WARNING: ar7_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/at91rm9200_wdt.c:113:8-24: WARNING: at91wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/ath79_wdt.c:135:8-24: WARNING: ath79_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/bcm63xx_wdt.c:119:8-24: WARNING: bcm63xx_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/cpu5wdt.c:143:8-24: WARNING: cpu5wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/cpwd.c:397:8-24: WARNING: cpwd_fops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/eurotechwdt.c:319:8-24: WARNING: eurwdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/f71808e_wdt.c:528:8-24: WARNING: watchdog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/gef_wdt.c:232:8-24: WARNING: gef_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/geodewdt.c:95:8-24: WARNING: geodewdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/ib700wdt.c:241:8-24: WARNING: ibwdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/ibmasr.c:326:8-24: WARNING: asr_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/indydog.c:80:8-24: WARNING: indydog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/intel_scu_watchdog.c:307:8-24: WARNING: intel_scu_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/iop_wdt.c:104:8-24: WARNING: iop_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/it8712f_wdt.c:330:8-24: WARNING: it8712f_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/ixp4xx_wdt.c:68:8-24: WARNING: ixp4xx_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/ks8695_wdt.c:145:8-24: WARNING: ks8695wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/m54xx_wdt.c:88:8-24: WARNING: m54xx_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/machzwd.c:336:8-24: WARNING: zf_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/mixcomwd.c:153:8-24: WARNING: mixcomwd_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/mtx-1_wdt.c:121:8-24: WARNING: mtx1_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/mv64x60_wdt.c:136:8-24: WARNING: mv64x60_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/nuc900_wdt.c:134:8-24: WARNING: nuc900wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/nv_tco.c:164:8-24: WARNING: nv_tco_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pc87413_wdt.c:289:8-24: WARNING: pc87413_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pcwd.c:698:8-24: WARNING: pcwd_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pcwd.c:737:8-24: WARNING: pcwd_temp_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pcwd_pci.c:581:8-24: WARNING: pcipcwd_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pcwd_pci.c:623:8-24: WARNING: pcipcwd_temp_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pcwd_usb.c:488:8-24: WARNING: usb_pcwd_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pcwd_usb.c:527:8-24: WARNING: usb_pcwd_temperature_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pika_wdt.c:121:8-24: WARNING: pikawdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/pnx833x_wdt.c:119:8-24: WARNING: pnx833x_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/rc32434_wdt.c:153:8-24: WARNING: rc32434_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/rdc321x_wdt.c:145:8-24: WARNING: rdc321x_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/riowd.c:79:1-17: WARNING: riowd_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sa1100_wdt.c:62:8-24: WARNING: sa1100dog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sbc60xxwdt.c:211:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sbc7240_wdt.c:139:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sbc8360.c:274:8-24: WARNING: sbc8360_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sbc_epx_c3.c:81:8-24: WARNING: epx_c3_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sbc_fitpc2_wdt.c:78:8-24: WARNING: fitpc2_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sb_wdog.c:108:1-17: WARNING: sbwdog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sc1200wdt.c:181:8-24: WARNING: sc1200wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sc520_wdt.c:261:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/sch311x_wdt.c:319:8-24: WARNING: sch311x_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/scx200_wdt.c:105:8-24: WARNING: scx200_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/smsc37b787_wdt.c:369:8-24: WARNING: wb_smsc_wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/w83877f_wdt.c:227:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/w83977f_wdt.c:301:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wafer5823wdt.c:200:8-24: WARNING: wafwdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/watchdog_dev.c:828:8-24: WARNING: watchdog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdrtas.c:379:8-24: WARNING: wdrtas_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdrtas.c:445:8-24: WARNING: wdrtas_temp_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdt285.c:104:1-17: WARNING: watchdog_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdt977.c:276:8-24: WARNING: wdt977_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdt.c:424:8-24: WARNING: wdt_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdt.c:484:8-24: WARNING: wdt_temp_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdt_pci.c:464:8-24: WARNING: wdtpci_fops: .write() has stream semantic; safe to change nonseekable_open -> stream_open. drivers/watchdog/wdt_pci.c:527:8-24: WARNING: wdtpci_temp_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. net/batman-adv/log.c:105:1-17: WARNING: batadv_log_fops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. sound/core/control.c:57:7-23: WARNING: snd_ctl_f_ops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. sound/core/rawmidi.c:385:7-23: WARNING: snd_rawmidi_f_ops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. sound/core/seq/seq_clientmgr.c:310:7-23: WARNING: snd_seq_f_ops: .read() and .write() have stream semantic; safe to change nonseekable_open -> stream_open. sound/core/timer.c:1428:7-23: WARNING: snd_timer_f_ops: .read() has stream semantic; safe to change nonseekable_open -> stream_open. One can also recheck/review the patch via generating it with explanation comments included via $ make coccicheck MODE=patch COCCI=scripts/coccinelle/api/stream_open.cocci SPFLAGS="-D explain" (*) This second group also contains cases with read/write deadlocks that stream_open.cocci don't yet detect, but which are still valid to convert to stream_open since ppos is not used. For example drivers/pci/switch/switchtec.c calls wait_for_completion_interruptible() in its .read, but stream_open.cocci currently detects only "wait_event*" as blocking. Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Yongzhi Pan <panyongzhi@gmail.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: David Vrabel <david.vrabel@citrix.com> Cc: Juergen Gross <jgross@suse.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Tejun Heo <tj@kernel.org> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Christoph Hellwig <hch@lst.de> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Julia Lawall <Julia.Lawall@lip6.fr> Cc: Nikolaus Rath <Nikolaus@rath.org> Cc: Han-Wen Nienhuys <hanwen@google.com> Cc: Anatolij Gustschin <agust@denx.de> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Anton Ivanov <anton.ivanov@cambridgegreys.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: "James R. Van Zandt" <jrv@vanzandt.mv.com> Cc: Corey Minyard <minyard@acm.org> Cc: Harald Welte <laforge@gnumonks.org> Acked-by: Lubomir Rintel <lkundrak@v3.sk> [scr24x_cs] Cc: Stefan Richter <stefanr@s5r6.in-berlin.de> Cc: Johan Hovold <johan@kernel.org> Cc: David Herrmann <dh.herrmann@googlemail.com> Cc: Jiri Kosina <jikos@kernel.org> Cc: Benjamin Tissoires <benjamin.tissoires@redhat.com> Cc: Jean Delvare <jdelvare@suse.com> Acked-by: Guenter Roeck <linux@roeck-us.net> [watchdog/* hwmon/*] Cc: Rudolf Marek <r.marek@assembler.cz> Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com> Cc: Karsten Keil <isdn@linux-pingi.de> Cc: Jacek Anaszewski <jacek.anaszewski@gmail.com> Cc: Pavel Machek <pavel@ucw.cz> Cc: Mauro Carvalho Chehab <mchehab@kernel.org> Cc: Kurt Schwemmer <kurt.schwemmer@microsemi.com> Acked-by: Logan Gunthorpe <logang@deltatee.com> [drivers/pci/switch/switchtec] Acked-by: Bjorn Helgaas <bhelgaas@google.com> [drivers/pci/switch/switchtec] Cc: Benson Leung <bleung@chromium.org> Acked-by: Enric Balletbo i Serra <enric.balletbo@collabora.com> [platform/chrome] Cc: Alessandro Zummo <a.zummo@towertech.it> Acked-by: Alexandre Belloni <alexandre.belloni@bootlin.com> [rtc/*] Cc: Mark Brown <broonie@kernel.org> Cc: Wim Van Sebroeck <wim@linux-watchdog.org> Cc: Florian Fainelli <f.fainelli@gmail.com> Cc: bcm-kernel-feedback-list@broadcom.com Cc: Wan ZongShun <mcuos.com@gmail.com> Cc: Zwane Mwaikambo <zwanem@gmail.com> Cc: Marek Lindner <mareklindner@neomailbox.ch> Cc: Simon Wunderlich <sw@simonwunderlich.de> Cc: Antonio Quartulli <a@unstable.cc> Cc: "David S. Miller" <davem@davemloft.net> Cc: Johannes Berg <johannes@sipsolutions.net> Cc: Jaroslav Kysela <perex@perex.cz> Cc: Takashi Iwai <tiwai@suse.com> Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
2019-03-26 20:51:19 +00:00
err = stream_open(inode, file);
if (err < 0)
return err;
card = snd_lookup_minor_data(iminor(inode), SNDRV_DEVICE_TYPE_CONTROL);
if (!card) {
err = -ENODEV;
goto __error1;
}
err = snd_card_file_add(card, file);
if (err < 0) {
err = -ENODEV;
goto __error1;
}
if (!try_module_get(card->module)) {
err = -EFAULT;
goto __error2;
}
ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
if (ctl == NULL) {
err = -ENOMEM;
goto __error;
}
INIT_LIST_HEAD(&ctl->events);
init_waitqueue_head(&ctl->change_sleep);
spin_lock_init(&ctl->read_lock);
ctl->card = card;
for (i = 0; i < SND_CTL_SUBDEV_ITEMS; i++)
ctl->preferred_subdevice[i] = -1;
ctl->pid = get_pid(task_pid(current));
file->private_data = ctl;
write_lock_irqsave(&card->ctl_files_rwlock, flags);
list_add_tail(&ctl->list, &card->ctl_files);
write_unlock_irqrestore(&card->ctl_files_rwlock, flags);
snd_card_unref(card);
return 0;
__error:
module_put(card->module);
__error2:
snd_card_file_remove(card, file);
__error1:
if (card)
snd_card_unref(card);
return err;
}
static void snd_ctl_empty_read_queue(struct snd_ctl_file * ctl)
{
unsigned long flags;
struct snd_kctl_event *cread;
spin_lock_irqsave(&ctl->read_lock, flags);
while (!list_empty(&ctl->events)) {
cread = snd_kctl_event(ctl->events.next);
list_del(&cread->list);
kfree(cread);
}
spin_unlock_irqrestore(&ctl->read_lock, flags);
}
static int snd_ctl_release(struct inode *inode, struct file *file)
{
unsigned long flags;
struct snd_card *card;
struct snd_ctl_file *ctl;
struct snd_kcontrol *control;
unsigned int idx;
ctl = file->private_data;
file->private_data = NULL;
card = ctl->card;
write_lock_irqsave(&card->ctl_files_rwlock, flags);
list_del(&ctl->list);
write_unlock_irqrestore(&card->ctl_files_rwlock, flags);
down_write(&card->controls_rwsem);
list_for_each_entry(control, &card->controls, list)
for (idx = 0; idx < control->count; idx++)
if (control->vd[idx].owner == ctl)
control->vd[idx].owner = NULL;
up_write(&card->controls_rwsem);
snd_fasync_free(ctl->fasync);
snd_ctl_empty_read_queue(ctl);
put_pid(ctl->pid);
kfree(ctl);
module_put(card->module);
snd_card_file_remove(card, file);
return 0;
}
/**
* snd_ctl_notify - Send notification to user-space for a control change
* @card: the card to send notification
* @mask: the event mask, SNDRV_CTL_EVENT_*
* @id: the ctl element id to send notification
*
* This function adds an event record with the given id and mask, appends
* to the list and wakes up the user-space for notification. This can be
* called in the atomic context.
*/
void snd_ctl_notify(struct snd_card *card, unsigned int mask,
struct snd_ctl_elem_id *id)
{
unsigned long flags;
struct snd_ctl_file *ctl;
struct snd_kctl_event *ev;
if (snd_BUG_ON(!card || !id))
return;
if (card->shutdown)
return;
read_lock_irqsave(&card->ctl_files_rwlock, flags);
#if IS_ENABLED(CONFIG_SND_MIXER_OSS)
card->mixer_oss_change_count++;
#endif
list_for_each_entry(ctl, &card->ctl_files, list) {
if (!ctl->subscribed)
continue;
spin_lock(&ctl->read_lock);
list_for_each_entry(ev, &ctl->events, list) {
if (ev->id.numid == id->numid) {
ev->mask |= mask;
goto _found;
}
}
ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
if (ev) {
ev->id = *id;
ev->mask = mask;
list_add_tail(&ev->list, &ctl->events);
} else {
dev_err(card->dev, "No memory available to allocate event\n");
}
_found:
wake_up(&ctl->change_sleep);
spin_unlock(&ctl->read_lock);
snd_kill_fasync(ctl->fasync, SIGIO, POLL_IN);
}
read_unlock_irqrestore(&card->ctl_files_rwlock, flags);
}
EXPORT_SYMBOL(snd_ctl_notify);
/**
* snd_ctl_notify_one - Send notification to user-space for a control change
* @card: the card to send notification
* @mask: the event mask, SNDRV_CTL_EVENT_*
* @kctl: the pointer with the control instance
* @ioff: the additional offset to the control index
*
* This function calls snd_ctl_notify() and does additional jobs
* like LED state changes.
*/
void snd_ctl_notify_one(struct snd_card *card, unsigned int mask,
struct snd_kcontrol *kctl, unsigned int ioff)
{
struct snd_ctl_elem_id id = kctl->id;
struct snd_ctl_layer_ops *lops;
id.index += ioff;
id.numid += ioff;
snd_ctl_notify(card, mask, &id);
down_read(&snd_ctl_layer_rwsem);
for (lops = snd_ctl_layer; lops; lops = lops->next)
lops->lnotify(card, mask, kctl, ioff);
up_read(&snd_ctl_layer_rwsem);
}
EXPORT_SYMBOL(snd_ctl_notify_one);
/**
* snd_ctl_new - create a new control instance with some elements
* @kctl: the pointer to store new control instance
* @count: the number of elements in this control
* @access: the default access flags for elements in this control
* @file: given when locking these elements
*
* Allocates a memory object for a new control instance. The instance has
* elements as many as the given number (@count). Each element has given
* access permissions (@access). Each element is locked when @file is given.
*
* Return: 0 on success, error code on failure
*/
static int snd_ctl_new(struct snd_kcontrol **kctl, unsigned int count,
unsigned int access, struct snd_ctl_file *file)
{
unsigned int idx;
if (count == 0 || count > MAX_CONTROL_COUNT)
return -EINVAL;
*kctl = kzalloc(struct_size(*kctl, vd, count), GFP_KERNEL);
if (!*kctl)
return -ENOMEM;
for (idx = 0; idx < count; idx++) {
(*kctl)->vd[idx].access = access;
(*kctl)->vd[idx].owner = file;
}
(*kctl)->count = count;
return 0;
}
/**
* snd_ctl_new1 - create a control instance from the template
* @ncontrol: the initialization record
* @private_data: the private data to set
*
* Allocates a new struct snd_kcontrol instance and initialize from the given
* template. When the access field of ncontrol is 0, it's assumed as
* READWRITE access. When the count field is 0, it's assumes as one.
*
* Return: The pointer of the newly generated instance, or %NULL on failure.
*/
struct snd_kcontrol *snd_ctl_new1(const struct snd_kcontrol_new *ncontrol,
void *private_data)
{
struct snd_kcontrol *kctl;
unsigned int count;
unsigned int access;
int err;
if (snd_BUG_ON(!ncontrol || !ncontrol->info))
return NULL;
count = ncontrol->count;
if (count == 0)
count = 1;
access = ncontrol->access;
if (access == 0)
access = SNDRV_CTL_ELEM_ACCESS_READWRITE;
access &= (SNDRV_CTL_ELEM_ACCESS_READWRITE |
SNDRV_CTL_ELEM_ACCESS_VOLATILE |
SNDRV_CTL_ELEM_ACCESS_INACTIVE |
SNDRV_CTL_ELEM_ACCESS_TLV_READWRITE |
SNDRV_CTL_ELEM_ACCESS_TLV_COMMAND |
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
SNDRV_CTL_ELEM_ACCESS_TLV_CALLBACK |
SNDRV_CTL_ELEM_ACCESS_LED_MASK |
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK);
err = snd_ctl_new(&kctl, count, access, NULL);
if (err < 0)
return NULL;
/* The 'numid' member is decided when calling snd_ctl_add(). */
kctl->id.iface = ncontrol->iface;
kctl->id.device = ncontrol->device;
kctl->id.subdevice = ncontrol->subdevice;
if (ncontrol->name) {
strscpy(kctl->id.name, ncontrol->name, sizeof(kctl->id.name));
if (strcmp(ncontrol->name, kctl->id.name) != 0)
pr_warn("ALSA: Control name '%s' truncated to '%s'\n",
ncontrol->name, kctl->id.name);
}
kctl->id.index = ncontrol->index;
kctl->info = ncontrol->info;
kctl->get = ncontrol->get;
kctl->put = ncontrol->put;
kctl->tlv.p = ncontrol->tlv.p;
kctl->private_value = ncontrol->private_value;
kctl->private_data = private_data;
return kctl;
}
EXPORT_SYMBOL(snd_ctl_new1);
/**
* snd_ctl_free_one - release the control instance
* @kcontrol: the control instance
*
* Releases the control instance created via snd_ctl_new()
* or snd_ctl_new1().
* Don't call this after the control was added to the card.
*/
void snd_ctl_free_one(struct snd_kcontrol *kcontrol)
{
if (kcontrol) {
if (kcontrol->private_free)
kcontrol->private_free(kcontrol);
kfree(kcontrol);
}
}
EXPORT_SYMBOL(snd_ctl_free_one);
static bool snd_ctl_remove_numid_conflict(struct snd_card *card,
unsigned int count)
{
struct snd_kcontrol *kctl;
/* Make sure that the ids assigned to the control do not wrap around */
if (card->last_numid >= UINT_MAX - count)
card->last_numid = 0;
list_for_each_entry(kctl, &card->controls, list) {
if (kctl->id.numid < card->last_numid + 1 + count &&
kctl->id.numid + kctl->count > card->last_numid + 1) {
card->last_numid = kctl->id.numid + kctl->count - 1;
return true;
}
}
return false;
}
static int snd_ctl_find_hole(struct snd_card *card, unsigned int count)
{
unsigned int iter = 100000;
while (snd_ctl_remove_numid_conflict(card, count)) {
if (--iter == 0) {
/* this situation is very unlikely */
dev_err(card->dev, "unable to allocate new control numid\n");
return -ENOMEM;
}
}
return 0;
}
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
/* check whether the given id is contained in the given kctl */
static bool elem_id_matches(const struct snd_kcontrol *kctl,
const struct snd_ctl_elem_id *id)
{
return kctl->id.iface == id->iface &&
kctl->id.device == id->device &&
kctl->id.subdevice == id->subdevice &&
!strncmp(kctl->id.name, id->name, sizeof(kctl->id.name)) &&
kctl->id.index <= id->index &&
kctl->id.index + kctl->count > id->index;
}
#ifdef CONFIG_SND_CTL_FAST_LOOKUP
/* Compute a hash key for the corresponding ctl id
* It's for the name lookup, hence the numid is excluded.
* The hash key is bound in LONG_MAX to be used for Xarray key.
*/
#define MULTIPLIER 37
static unsigned long get_ctl_id_hash(const struct snd_ctl_elem_id *id)
{
int i;
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
unsigned long h;
h = id->iface;
h = MULTIPLIER * h + id->device;
h = MULTIPLIER * h + id->subdevice;
for (i = 0; i < SNDRV_CTL_ELEM_ID_NAME_MAXLEN && id->name[i]; i++)
h = MULTIPLIER * h + id->name[i];
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
h = MULTIPLIER * h + id->index;
h &= LONG_MAX;
return h;
}
/* add hash entries to numid and ctl xarray tables */
static void add_hash_entries(struct snd_card *card,
struct snd_kcontrol *kcontrol)
{
struct snd_ctl_elem_id id = kcontrol->id;
int i;
xa_store_range(&card->ctl_numids, kcontrol->id.numid,
kcontrol->id.numid + kcontrol->count - 1,
kcontrol, GFP_KERNEL);
for (i = 0; i < kcontrol->count; i++) {
id.index = kcontrol->id.index + i;
if (xa_insert(&card->ctl_hash, get_ctl_id_hash(&id),
kcontrol, GFP_KERNEL)) {
/* skip hash for this entry, noting we had collision */
card->ctl_hash_collision = true;
dev_dbg(card->dev, "ctl_hash collision %d:%s:%d\n",
id.iface, id.name, id.index);
}
}
}
/* remove hash entries that have been added */
static void remove_hash_entries(struct snd_card *card,
struct snd_kcontrol *kcontrol)
{
struct snd_ctl_elem_id id = kcontrol->id;
struct snd_kcontrol *matched;
unsigned long h;
int i;
for (i = 0; i < kcontrol->count; i++) {
xa_erase(&card->ctl_numids, id.numid);
h = get_ctl_id_hash(&id);
matched = xa_load(&card->ctl_hash, h);
if (matched && (matched == kcontrol ||
elem_id_matches(matched, &id)))
xa_erase(&card->ctl_hash, h);
id.index++;
id.numid++;
}
}
#else /* CONFIG_SND_CTL_FAST_LOOKUP */
static inline void add_hash_entries(struct snd_card *card,
struct snd_kcontrol *kcontrol)
{
}
static inline void remove_hash_entries(struct snd_card *card,
struct snd_kcontrol *kcontrol)
{
}
#endif /* CONFIG_SND_CTL_FAST_LOOKUP */
enum snd_ctl_add_mode {
CTL_ADD_EXCLUSIVE, CTL_REPLACE, CTL_ADD_ON_REPLACE,
};
/* add/replace a new kcontrol object; call with card->controls_rwsem locked */
static int __snd_ctl_add_replace(struct snd_card *card,
struct snd_kcontrol *kcontrol,
enum snd_ctl_add_mode mode)
{
struct snd_ctl_elem_id id;
unsigned int idx;
struct snd_kcontrol *old;
int err;
lockdep_assert_held_write(&card->controls_rwsem);
id = kcontrol->id;
if (id.index > UINT_MAX - kcontrol->count)
return -EINVAL;
old = snd_ctl_find_id_locked(card, &id);
if (!old) {
if (mode == CTL_REPLACE)
return -EINVAL;
} else {
if (mode == CTL_ADD_EXCLUSIVE) {
dev_err(card->dev,
"control %i:%i:%i:%s:%i is already present\n",
id.iface, id.device, id.subdevice, id.name,
id.index);
return -EBUSY;
}
err = snd_ctl_remove_locked(card, old);
if (err < 0)
return err;
}
if (snd_ctl_find_hole(card, kcontrol->count) < 0)
return -ENOMEM;
list_add_tail(&kcontrol->list, &card->controls);
card->controls_count += kcontrol->count;
kcontrol->id.numid = card->last_numid + 1;
card->last_numid += kcontrol->count;
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
add_hash_entries(card, kcontrol);
for (idx = 0; idx < kcontrol->count; idx++)
snd_ctl_notify_one(card, SNDRV_CTL_EVENT_MASK_ADD, kcontrol, idx);
return 0;
}
static int snd_ctl_add_replace(struct snd_card *card,
struct snd_kcontrol *kcontrol,
enum snd_ctl_add_mode mode)
{
int err = -EINVAL;
if (! kcontrol)
return err;
if (snd_BUG_ON(!card || !kcontrol->info))
goto error;
down_write(&card->controls_rwsem);
err = __snd_ctl_add_replace(card, kcontrol, mode);
up_write(&card->controls_rwsem);
if (err < 0)
goto error;
return 0;
error:
snd_ctl_free_one(kcontrol);
return err;
}
/**
* snd_ctl_add - add the control instance to the card
* @card: the card instance
* @kcontrol: the control instance to add
*
* Adds the control instance created via snd_ctl_new() or
* snd_ctl_new1() to the given card. Assigns also an unique
* numid used for fast search.
*
* It frees automatically the control which cannot be added.
*
* Return: Zero if successful, or a negative error code on failure.
*
*/
int snd_ctl_add(struct snd_card *card, struct snd_kcontrol *kcontrol)
{
return snd_ctl_add_replace(card, kcontrol, CTL_ADD_EXCLUSIVE);
}
EXPORT_SYMBOL(snd_ctl_add);
/**
* snd_ctl_replace - replace the control instance of the card
* @card: the card instance
* @kcontrol: the control instance to replace
* @add_on_replace: add the control if not already added
*
* Replaces the given control. If the given control does not exist
* and the add_on_replace flag is set, the control is added. If the
* control exists, it is destroyed first.
*
* It frees automatically the control which cannot be added or replaced.
*
* Return: Zero if successful, or a negative error code on failure.
*/
int snd_ctl_replace(struct snd_card *card, struct snd_kcontrol *kcontrol,
bool add_on_replace)
{
return snd_ctl_add_replace(card, kcontrol,
add_on_replace ? CTL_ADD_ON_REPLACE : CTL_REPLACE);
}
EXPORT_SYMBOL(snd_ctl_replace);
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
static int __snd_ctl_remove(struct snd_card *card,
struct snd_kcontrol *kcontrol,
bool remove_hash)
{
unsigned int idx;
lockdep_assert_held_write(&card->controls_rwsem);
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
if (snd_BUG_ON(!card || !kcontrol))
return -EINVAL;
list_del(&kcontrol->list);
if (remove_hash)
remove_hash_entries(card, kcontrol);
card->controls_count -= kcontrol->count;
for (idx = 0; idx < kcontrol->count; idx++)
snd_ctl_notify_one(card, SNDRV_CTL_EVENT_MASK_REMOVE, kcontrol, idx);
snd_ctl_free_one(kcontrol);
return 0;
}
static inline int snd_ctl_remove_locked(struct snd_card *card,
struct snd_kcontrol *kcontrol)
{
return __snd_ctl_remove(card, kcontrol, true);
}
/**
* snd_ctl_remove - remove the control from the card and release it
* @card: the card instance
* @kcontrol: the control instance to remove
*
* Removes the control from the card and then releases the instance.
* You don't need to call snd_ctl_free_one().
*
* Return: 0 if successful, or a negative error code on failure.
*
* Note that this function takes card->controls_rwsem lock internally.
*/
int snd_ctl_remove(struct snd_card *card, struct snd_kcontrol *kcontrol)
{
int ret;
down_write(&card->controls_rwsem);
ret = snd_ctl_remove_locked(card, kcontrol);
up_write(&card->controls_rwsem);
return ret;
}
EXPORT_SYMBOL(snd_ctl_remove);
/**
* snd_ctl_remove_id - remove the control of the given id and release it
* @card: the card instance
* @id: the control id to remove
*
* Finds the control instance with the given id, removes it from the
* card list and releases it.
*
* Return: 0 if successful, or a negative error code on failure.
*/
int snd_ctl_remove_id(struct snd_card *card, struct snd_ctl_elem_id *id)
{
struct snd_kcontrol *kctl;
int ret;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, id);
if (kctl == NULL) {
up_write(&card->controls_rwsem);
return -ENOENT;
}
ret = snd_ctl_remove_locked(card, kctl);
up_write(&card->controls_rwsem);
return ret;
}
EXPORT_SYMBOL(snd_ctl_remove_id);
/**
* snd_ctl_remove_user_ctl - remove and release the unlocked user control
* @file: active control handle
* @id: the control id to remove
*
* Finds the control instance with the given id, removes it from the
* card list and releases it.
*
* Return: 0 if successful, or a negative error code on failure.
*/
static int snd_ctl_remove_user_ctl(struct snd_ctl_file * file,
struct snd_ctl_elem_id *id)
{
struct snd_card *card = file->card;
struct snd_kcontrol *kctl;
int idx, ret;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, id);
if (kctl == NULL) {
ret = -ENOENT;
goto error;
}
if (!(kctl->vd[0].access & SNDRV_CTL_ELEM_ACCESS_USER)) {
ret = -EINVAL;
goto error;
}
for (idx = 0; idx < kctl->count; idx++)
if (kctl->vd[idx].owner != NULL && kctl->vd[idx].owner != file) {
ret = -EBUSY;
goto error;
}
ret = snd_ctl_remove_locked(card, kctl);
error:
up_write(&card->controls_rwsem);
return ret;
}
/**
* snd_ctl_activate_id - activate/inactivate the control of the given id
* @card: the card instance
* @id: the control id to activate/inactivate
* @active: non-zero to activate
*
* Finds the control instance with the given id, and activate or
* inactivate the control together with notification, if changed.
* The given ID data is filled with full information.
*
* Return: 0 if unchanged, 1 if changed, or a negative error code on failure.
*/
int snd_ctl_activate_id(struct snd_card *card, struct snd_ctl_elem_id *id,
int active)
{
struct snd_kcontrol *kctl;
struct snd_kcontrol_volatile *vd;
unsigned int index_offset;
int ret;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, id);
if (kctl == NULL) {
ret = -ENOENT;
goto unlock;
}
index_offset = snd_ctl_get_ioff(kctl, id);
vd = &kctl->vd[index_offset];
ret = 0;
if (active) {
if (!(vd->access & SNDRV_CTL_ELEM_ACCESS_INACTIVE))
goto unlock;
vd->access &= ~SNDRV_CTL_ELEM_ACCESS_INACTIVE;
} else {
if (vd->access & SNDRV_CTL_ELEM_ACCESS_INACTIVE)
goto unlock;
vd->access |= SNDRV_CTL_ELEM_ACCESS_INACTIVE;
}
snd_ctl_build_ioff(id, kctl, index_offset);
downgrade_write(&card->controls_rwsem);
snd_ctl_notify_one(card, SNDRV_CTL_EVENT_MASK_INFO, kctl, index_offset);
up_read(&card->controls_rwsem);
return 1;
unlock:
up_write(&card->controls_rwsem);
return ret;
}
EXPORT_SYMBOL_GPL(snd_ctl_activate_id);
/**
* snd_ctl_rename_id - replace the id of a control on the card
* @card: the card instance
* @src_id: the old id
* @dst_id: the new id
*
* Finds the control with the old id from the card, and replaces the
* id with the new one.
*
* The function tries to keep the already assigned numid while replacing
* the rest.
*
* Note that this function should be used only in the card initialization
* phase. Calling after the card instantiation may cause issues with
* user-space expecting persistent numids.
*
* Return: Zero if successful, or a negative error code on failure.
*/
int snd_ctl_rename_id(struct snd_card *card, struct snd_ctl_elem_id *src_id,
struct snd_ctl_elem_id *dst_id)
{
struct snd_kcontrol *kctl;
int saved_numid;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, src_id);
if (kctl == NULL) {
up_write(&card->controls_rwsem);
return -ENOENT;
}
saved_numid = kctl->id.numid;
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
remove_hash_entries(card, kctl);
kctl->id = *dst_id;
kctl->id.numid = saved_numid;
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
add_hash_entries(card, kctl);
up_write(&card->controls_rwsem);
return 0;
}
EXPORT_SYMBOL(snd_ctl_rename_id);
/**
* snd_ctl_rename - rename the control on the card
* @card: the card instance
* @kctl: the control to rename
* @name: the new name
*
* Renames the specified control on the card to the new name.
*
* Note that this function takes card->controls_rwsem lock internally.
*/
void snd_ctl_rename(struct snd_card *card, struct snd_kcontrol *kctl,
const char *name)
{
down_write(&card->controls_rwsem);
remove_hash_entries(card, kctl);
if (strscpy(kctl->id.name, name, sizeof(kctl->id.name)) < 0)
pr_warn("ALSA: Renamed control new name '%s' truncated to '%s'\n",
name, kctl->id.name);
add_hash_entries(card, kctl);
up_write(&card->controls_rwsem);
}
EXPORT_SYMBOL(snd_ctl_rename);
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
#ifndef CONFIG_SND_CTL_FAST_LOOKUP
static struct snd_kcontrol *
snd_ctl_find_numid_slow(struct snd_card *card, unsigned int numid)
{
struct snd_kcontrol *kctl;
list_for_each_entry(kctl, &card->controls, list) {
if (kctl->id.numid <= numid && kctl->id.numid + kctl->count > numid)
return kctl;
}
return NULL;
}
#endif /* !CONFIG_SND_CTL_FAST_LOOKUP */
/**
* snd_ctl_find_numid_locked - find the control instance with the given number-id
* @card: the card instance
* @numid: the number-id to search
*
* Finds the control instance with the given number-id from the card.
*
* The caller must down card->controls_rwsem before calling this function
* (if the race condition can happen).
*
* Return: The pointer of the instance if found, or %NULL if not.
*/
struct snd_kcontrol *
snd_ctl_find_numid_locked(struct snd_card *card, unsigned int numid)
{
if (snd_BUG_ON(!card || !numid))
return NULL;
lockdep_assert_held(&card->controls_rwsem);
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
#ifdef CONFIG_SND_CTL_FAST_LOOKUP
return xa_load(&card->ctl_numids, numid);
#else
return snd_ctl_find_numid_slow(card, numid);
#endif
}
EXPORT_SYMBOL(snd_ctl_find_numid_locked);
/**
* snd_ctl_find_numid - find the control instance with the given number-id
* @card: the card instance
* @numid: the number-id to search
*
* Finds the control instance with the given number-id from the card.
*
* Return: The pointer of the instance if found, or %NULL if not.
*
* Note that this function takes card->controls_rwsem lock internally.
*/
struct snd_kcontrol *snd_ctl_find_numid(struct snd_card *card,
unsigned int numid)
{
struct snd_kcontrol *kctl;
down_read(&card->controls_rwsem);
kctl = snd_ctl_find_numid_locked(card, numid);
up_read(&card->controls_rwsem);
return kctl;
}
EXPORT_SYMBOL(snd_ctl_find_numid);
/**
* snd_ctl_find_id_locked - find the control instance with the given id
* @card: the card instance
* @id: the id to search
*
* Finds the control instance with the given id from the card.
*
* The caller must down card->controls_rwsem before calling this function
* (if the race condition can happen).
*
* Return: The pointer of the instance if found, or %NULL if not.
*/
struct snd_kcontrol *snd_ctl_find_id_locked(struct snd_card *card,
const struct snd_ctl_elem_id *id)
{
struct snd_kcontrol *kctl;
if (snd_BUG_ON(!card || !id))
return NULL;
lockdep_assert_held(&card->controls_rwsem);
if (id->numid != 0)
return snd_ctl_find_numid_locked(card, id->numid);
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
#ifdef CONFIG_SND_CTL_FAST_LOOKUP
kctl = xa_load(&card->ctl_hash, get_ctl_id_hash(id));
if (kctl && elem_id_matches(kctl, id))
return kctl;
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
if (!card->ctl_hash_collision)
return NULL; /* we can rely on only hash table */
#endif
/* no matching in hash table - try all as the last resort */
list_for_each_entry(kctl, &card->controls, list)
if (elem_id_matches(kctl, id))
return kctl;
return NULL;
}
EXPORT_SYMBOL(snd_ctl_find_id_locked);
/**
* snd_ctl_find_id - find the control instance with the given id
* @card: the card instance
* @id: the id to search
*
* Finds the control instance with the given id from the card.
*
* Return: The pointer of the instance if found, or %NULL if not.
*
* Note that this function takes card->controls_rwsem lock internally.
*/
struct snd_kcontrol *snd_ctl_find_id(struct snd_card *card,
const struct snd_ctl_elem_id *id)
{
struct snd_kcontrol *kctl;
down_read(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, id);
up_read(&card->controls_rwsem);
return kctl;
}
EXPORT_SYMBOL(snd_ctl_find_id);
static int snd_ctl_card_info(struct snd_card *card, struct snd_ctl_file * ctl,
unsigned int cmd, void __user *arg)
{
struct snd_ctl_card_info *info;
info = kzalloc(sizeof(*info), GFP_KERNEL);
if (! info)
return -ENOMEM;
down_read(&snd_ioctl_rwsem);
info->card = card->number;
strscpy(info->id, card->id, sizeof(info->id));
strscpy(info->driver, card->driver, sizeof(info->driver));
strscpy(info->name, card->shortname, sizeof(info->name));
strscpy(info->longname, card->longname, sizeof(info->longname));
strscpy(info->mixername, card->mixername, sizeof(info->mixername));
strscpy(info->components, card->components, sizeof(info->components));
up_read(&snd_ioctl_rwsem);
if (copy_to_user(arg, info, sizeof(struct snd_ctl_card_info))) {
kfree(info);
return -EFAULT;
}
kfree(info);
return 0;
}
static int snd_ctl_elem_list(struct snd_card *card,
struct snd_ctl_elem_list *list)
{
struct snd_kcontrol *kctl;
struct snd_ctl_elem_id id;
unsigned int offset, space, jidx;
int err = 0;
offset = list->offset;
space = list->space;
down_read(&card->controls_rwsem);
list->count = card->controls_count;
list->used = 0;
if (space > 0) {
list_for_each_entry(kctl, &card->controls, list) {
if (offset >= kctl->count) {
offset -= kctl->count;
continue;
}
for (jidx = offset; jidx < kctl->count; jidx++) {
snd_ctl_build_ioff(&id, kctl, jidx);
if (copy_to_user(list->pids + list->used, &id,
sizeof(id))) {
err = -EFAULT;
goto out;
}
list->used++;
if (!--space)
goto out;
}
offset = 0;
}
}
out:
up_read(&card->controls_rwsem);
return err;
}
static int snd_ctl_elem_list_user(struct snd_card *card,
struct snd_ctl_elem_list __user *_list)
{
struct snd_ctl_elem_list list;
int err;
if (copy_from_user(&list, _list, sizeof(list)))
return -EFAULT;
err = snd_ctl_elem_list(card, &list);
if (err)
return err;
if (copy_to_user(_list, &list, sizeof(list)))
return -EFAULT;
return 0;
}
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
/* Check whether the given kctl info is valid */
static int snd_ctl_check_elem_info(struct snd_card *card,
const struct snd_ctl_elem_info *info)
{
static const unsigned int max_value_counts[] = {
[SNDRV_CTL_ELEM_TYPE_BOOLEAN] = 128,
[SNDRV_CTL_ELEM_TYPE_INTEGER] = 128,
[SNDRV_CTL_ELEM_TYPE_ENUMERATED] = 128,
[SNDRV_CTL_ELEM_TYPE_BYTES] = 512,
[SNDRV_CTL_ELEM_TYPE_IEC958] = 1,
[SNDRV_CTL_ELEM_TYPE_INTEGER64] = 64,
};
if (info->type < SNDRV_CTL_ELEM_TYPE_BOOLEAN ||
info->type > SNDRV_CTL_ELEM_TYPE_INTEGER64) {
if (card)
dev_err(card->dev,
"control %i:%i:%i:%s:%i: invalid type %d\n",
info->id.iface, info->id.device,
info->id.subdevice, info->id.name,
info->id.index, info->type);
return -EINVAL;
}
if (info->type == SNDRV_CTL_ELEM_TYPE_ENUMERATED &&
info->value.enumerated.items == 0) {
if (card)
dev_err(card->dev,
"control %i:%i:%i:%s:%i: zero enum items\n",
info->id.iface, info->id.device,
info->id.subdevice, info->id.name,
info->id.index);
return -EINVAL;
}
if (info->count > max_value_counts[info->type]) {
if (card)
dev_err(card->dev,
"control %i:%i:%i:%s:%i: invalid count %d\n",
info->id.iface, info->id.device,
info->id.subdevice, info->id.name,
info->id.index, info->count);
return -EINVAL;
}
return 0;
}
/* The capacity of struct snd_ctl_elem_value.value.*/
static const unsigned int value_sizes[] = {
[SNDRV_CTL_ELEM_TYPE_BOOLEAN] = sizeof(long),
[SNDRV_CTL_ELEM_TYPE_INTEGER] = sizeof(long),
[SNDRV_CTL_ELEM_TYPE_ENUMERATED] = sizeof(unsigned int),
[SNDRV_CTL_ELEM_TYPE_BYTES] = sizeof(unsigned char),
[SNDRV_CTL_ELEM_TYPE_IEC958] = sizeof(struct snd_aes_iec958),
[SNDRV_CTL_ELEM_TYPE_INTEGER64] = sizeof(long long),
};
/* fill the remaining snd_ctl_elem_value data with the given pattern */
static void fill_remaining_elem_value(struct snd_ctl_elem_value *control,
struct snd_ctl_elem_info *info,
u32 pattern)
{
size_t offset = value_sizes[info->type] * info->count;
offset = DIV_ROUND_UP(offset, sizeof(u32));
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
memset32((u32 *)control->value.bytes.data + offset, pattern,
sizeof(control->value) / sizeof(u32) - offset);
}
/* check whether the given integer ctl value is valid */
static int sanity_check_int_value(struct snd_card *card,
const struct snd_ctl_elem_value *control,
const struct snd_ctl_elem_info *info,
int i, bool print_error)
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
{
long long lval, lmin, lmax, lstep;
u64 rem;
switch (info->type) {
default:
case SNDRV_CTL_ELEM_TYPE_BOOLEAN:
lval = control->value.integer.value[i];
lmin = 0;
lmax = 1;
lstep = 0;
break;
case SNDRV_CTL_ELEM_TYPE_INTEGER:
lval = control->value.integer.value[i];
lmin = info->value.integer.min;
lmax = info->value.integer.max;
lstep = info->value.integer.step;
break;
case SNDRV_CTL_ELEM_TYPE_INTEGER64:
lval = control->value.integer64.value[i];
lmin = info->value.integer64.min;
lmax = info->value.integer64.max;
lstep = info->value.integer64.step;
break;
case SNDRV_CTL_ELEM_TYPE_ENUMERATED:
lval = control->value.enumerated.item[i];
lmin = 0;
lmax = info->value.enumerated.items - 1;
lstep = 0;
break;
}
if (lval < lmin || lval > lmax) {
if (print_error)
dev_err(card->dev,
"control %i:%i:%i:%s:%i: value out of range %lld (%lld/%lld) at count %i\n",
control->id.iface, control->id.device,
control->id.subdevice, control->id.name,
control->id.index, lval, lmin, lmax, i);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
return -EINVAL;
}
if (lstep) {
div64_u64_rem(lval, lstep, &rem);
if (rem) {
if (print_error)
dev_err(card->dev,
"control %i:%i:%i:%s:%i: unaligned value %lld (step %lld) at count %i\n",
control->id.iface, control->id.device,
control->id.subdevice, control->id.name,
control->id.index, lval, lstep, i);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
return -EINVAL;
}
}
return 0;
}
/* check whether the all input values are valid for the given elem value */
static int sanity_check_input_values(struct snd_card *card,
const struct snd_ctl_elem_value *control,
const struct snd_ctl_elem_info *info,
bool print_error)
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
{
int i, ret;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
switch (info->type) {
case SNDRV_CTL_ELEM_TYPE_BOOLEAN:
case SNDRV_CTL_ELEM_TYPE_INTEGER:
case SNDRV_CTL_ELEM_TYPE_INTEGER64:
case SNDRV_CTL_ELEM_TYPE_ENUMERATED:
for (i = 0; i < info->count; i++) {
ret = sanity_check_int_value(card, control, info, i,
print_error);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
if (ret < 0)
return ret;
}
break;
default:
break;
}
return 0;
}
/* perform sanity checks to the given snd_ctl_elem_value object */
static int sanity_check_elem_value(struct snd_card *card,
const struct snd_ctl_elem_value *control,
const struct snd_ctl_elem_info *info,
u32 pattern)
{
size_t offset;
int ret;
u32 *p;
ret = sanity_check_input_values(card, control, info, true);
if (ret < 0)
return ret;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
/* check whether the remaining area kept untouched */
offset = value_sizes[info->type] * info->count;
offset = DIV_ROUND_UP(offset, sizeof(u32));
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
p = (u32 *)control->value.bytes.data + offset;
for (; offset < sizeof(control->value) / sizeof(u32); offset++, p++) {
if (*p != pattern) {
ret = -EINVAL;
break;
}
*p = 0; /* clear the checked area */
}
return ret;
}
static int __snd_ctl_elem_info(struct snd_card *card,
struct snd_kcontrol *kctl,
struct snd_ctl_elem_info *info,
struct snd_ctl_file *ctl)
{
struct snd_kcontrol_volatile *vd;
unsigned int index_offset;
int result;
#ifdef CONFIG_SND_DEBUG
info->access = 0;
#endif
ALSA: control: Track in-flight control read/write/tlv accesses Although the power state check is performed in various places (e.g. at the entrance of quite a few ioctls), there can be still some pending tasks that already went into the ioctl handler or other ops, and those may access the hardware even after the power state check. For example, kcontrol access ioctl paths that call info/get/put callbacks may update the hardware registers. If a system wants to assure the free from such hw access (like the case of PCI rescan feature we're going to implement in future), this situation must be avoided, and we have to sync such in-flight tasks finishing beforehand. For that purpose, this patch introduces a few new things in core code: - A refcount, power_ref, and a wait queue, power_ref_sleep, to the card object - A few new helpers, snd_power_ref(), snd_power_unref(), snd_power_ref_and_wait(), and snd_power_sync_ref() In the code paths that call kctl info/read/write/tlv ops, we check the power state with the newly introduced snd_power_ref_and_wait(). This function also takes the card.power_ref refcount for tracking this in-flight task. Once after the access finishes, snd_power_unref() is called to released the refcount in return. So the driver can sync via snd_power_sync_ref() assuring that all in-flight tasks have been finished. As of this patch, snd_power_sync_ref() is called only at snd_card_disconnect(), but it'll be used in other places in future. Note that atomic_t is used for power_ref intentionally instead of refcount_t. It's because of the design of refcount_t type; refcount_t cannot be zero-based, and it cannot do dec_and_test() call for multiple times, hence it's not suitable for our purpose. Also, this patch changes snd_power_wait() to accept only SNDRV_CTL_POWER_D0, which is the only value that makes sense. In later patch, the snd_power_wait() calls will be cleaned up. Reviewed-by: Jaroslav Kysela <perex@perex.cz> Link: https://lore.kernel.org/r/20210523090920.15345-3-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-23 09:09:16 +00:00
result = snd_power_ref_and_wait(card);
if (!result)
result = kctl->info(kctl, info);
snd_power_unref(card);
if (result >= 0) {
snd_BUG_ON(info->access);
index_offset = snd_ctl_get_ioff(kctl, &info->id);
vd = &kctl->vd[index_offset];
snd_ctl_build_ioff(&info->id, kctl, index_offset);
info->access = vd->access;
if (vd->owner) {
info->access |= SNDRV_CTL_ELEM_ACCESS_LOCK;
if (vd->owner == ctl)
info->access |= SNDRV_CTL_ELEM_ACCESS_OWNER;
info->owner = pid_vnr(vd->owner->pid);
} else {
info->owner = -1;
}
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
if (!snd_ctl_skip_validation(info) &&
snd_ctl_check_elem_info(card, info) < 0)
result = -EINVAL;
}
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
return result;
}
static int snd_ctl_elem_info(struct snd_ctl_file *ctl,
struct snd_ctl_elem_info *info)
{
struct snd_card *card = ctl->card;
struct snd_kcontrol *kctl;
int result;
down_read(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, &info->id);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
if (kctl == NULL)
result = -ENOENT;
else
result = __snd_ctl_elem_info(card, kctl, info, ctl);
up_read(&card->controls_rwsem);
return result;
}
static int snd_ctl_elem_info_user(struct snd_ctl_file *ctl,
struct snd_ctl_elem_info __user *_info)
{
struct snd_ctl_elem_info info;
int result;
if (copy_from_user(&info, _info, sizeof(info)))
return -EFAULT;
result = snd_ctl_elem_info(ctl, &info);
if (result < 0)
return result;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
/* drop internal access flags */
info.access &= ~(SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK|
SNDRV_CTL_ELEM_ACCESS_LED_MASK);
if (copy_to_user(_info, &info, sizeof(info)))
return -EFAULT;
return result;
}
static int snd_ctl_elem_read(struct snd_card *card,
struct snd_ctl_elem_value *control)
{
struct snd_kcontrol *kctl;
struct snd_kcontrol_volatile *vd;
unsigned int index_offset;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
struct snd_ctl_elem_info info;
const u32 pattern = 0xdeadbeef;
int ret;
down_read(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, &control->id);
if (kctl == NULL) {
ret = -ENOENT;
goto unlock;
}
index_offset = snd_ctl_get_ioff(kctl, &control->id);
vd = &kctl->vd[index_offset];
if (!(vd->access & SNDRV_CTL_ELEM_ACCESS_READ) || kctl->get == NULL) {
ret = -EPERM;
goto unlock;
}
snd_ctl_build_ioff(&control->id, kctl, index_offset);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
#ifdef CONFIG_SND_CTL_DEBUG
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
/* info is needed only for validation */
memset(&info, 0, sizeof(info));
info.id = control->id;
ret = __snd_ctl_elem_info(card, kctl, &info, NULL);
if (ret < 0)
goto unlock;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
#endif
if (!snd_ctl_skip_validation(&info))
fill_remaining_elem_value(control, &info, pattern);
ALSA: control: Track in-flight control read/write/tlv accesses Although the power state check is performed in various places (e.g. at the entrance of quite a few ioctls), there can be still some pending tasks that already went into the ioctl handler or other ops, and those may access the hardware even after the power state check. For example, kcontrol access ioctl paths that call info/get/put callbacks may update the hardware registers. If a system wants to assure the free from such hw access (like the case of PCI rescan feature we're going to implement in future), this situation must be avoided, and we have to sync such in-flight tasks finishing beforehand. For that purpose, this patch introduces a few new things in core code: - A refcount, power_ref, and a wait queue, power_ref_sleep, to the card object - A few new helpers, snd_power_ref(), snd_power_unref(), snd_power_ref_and_wait(), and snd_power_sync_ref() In the code paths that call kctl info/read/write/tlv ops, we check the power state with the newly introduced snd_power_ref_and_wait(). This function also takes the card.power_ref refcount for tracking this in-flight task. Once after the access finishes, snd_power_unref() is called to released the refcount in return. So the driver can sync via snd_power_sync_ref() assuring that all in-flight tasks have been finished. As of this patch, snd_power_sync_ref() is called only at snd_card_disconnect(), but it'll be used in other places in future. Note that atomic_t is used for power_ref intentionally instead of refcount_t. It's because of the design of refcount_t type; refcount_t cannot be zero-based, and it cannot do dec_and_test() call for multiple times, hence it's not suitable for our purpose. Also, this patch changes snd_power_wait() to accept only SNDRV_CTL_POWER_D0, which is the only value that makes sense. In later patch, the snd_power_wait() calls will be cleaned up. Reviewed-by: Jaroslav Kysela <perex@perex.cz> Link: https://lore.kernel.org/r/20210523090920.15345-3-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-23 09:09:16 +00:00
ret = snd_power_ref_and_wait(card);
if (!ret)
ret = kctl->get(kctl, control);
snd_power_unref(card);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
if (ret < 0)
goto unlock;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
if (!snd_ctl_skip_validation(&info) &&
sanity_check_elem_value(card, control, &info, pattern) < 0) {
dev_err(card->dev,
"control %i:%i:%i:%s:%i: access overflow\n",
control->id.iface, control->id.device,
control->id.subdevice, control->id.name,
control->id.index);
ret = -EINVAL;
goto unlock;
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
}
unlock:
up_read(&card->controls_rwsem);
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
return ret;
}
static int snd_ctl_elem_read_user(struct snd_card *card,
struct snd_ctl_elem_value __user *_control)
{
struct snd_ctl_elem_value *control;
int result;
control = memdup_user(_control, sizeof(*control));
if (IS_ERR(control))
return PTR_ERR(control);
result = snd_ctl_elem_read(card, control);
if (result < 0)
goto error;
if (copy_to_user(_control, control, sizeof(*control)))
result = -EFAULT;
error:
kfree(control);
return result;
}
static int snd_ctl_elem_write(struct snd_card *card, struct snd_ctl_file *file,
struct snd_ctl_elem_value *control)
{
struct snd_kcontrol *kctl;
struct snd_kcontrol_volatile *vd;
unsigned int index_offset;
int result;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, &control->id);
if (kctl == NULL) {
up_write(&card->controls_rwsem);
return -ENOENT;
}
index_offset = snd_ctl_get_ioff(kctl, &control->id);
vd = &kctl->vd[index_offset];
if (!(vd->access & SNDRV_CTL_ELEM_ACCESS_WRITE) || kctl->put == NULL ||
(file && vd->owner && vd->owner != file)) {
up_write(&card->controls_rwsem);
return -EPERM;
}
snd_ctl_build_ioff(&control->id, kctl, index_offset);
ALSA: control: Track in-flight control read/write/tlv accesses Although the power state check is performed in various places (e.g. at the entrance of quite a few ioctls), there can be still some pending tasks that already went into the ioctl handler or other ops, and those may access the hardware even after the power state check. For example, kcontrol access ioctl paths that call info/get/put callbacks may update the hardware registers. If a system wants to assure the free from such hw access (like the case of PCI rescan feature we're going to implement in future), this situation must be avoided, and we have to sync such in-flight tasks finishing beforehand. For that purpose, this patch introduces a few new things in core code: - A refcount, power_ref, and a wait queue, power_ref_sleep, to the card object - A few new helpers, snd_power_ref(), snd_power_unref(), snd_power_ref_and_wait(), and snd_power_sync_ref() In the code paths that call kctl info/read/write/tlv ops, we check the power state with the newly introduced snd_power_ref_and_wait(). This function also takes the card.power_ref refcount for tracking this in-flight task. Once after the access finishes, snd_power_unref() is called to released the refcount in return. So the driver can sync via snd_power_sync_ref() assuring that all in-flight tasks have been finished. As of this patch, snd_power_sync_ref() is called only at snd_card_disconnect(), but it'll be used in other places in future. Note that atomic_t is used for power_ref intentionally instead of refcount_t. It's because of the design of refcount_t type; refcount_t cannot be zero-based, and it cannot do dec_and_test() call for multiple times, hence it's not suitable for our purpose. Also, this patch changes snd_power_wait() to accept only SNDRV_CTL_POWER_D0, which is the only value that makes sense. In later patch, the snd_power_wait() calls will be cleaned up. Reviewed-by: Jaroslav Kysela <perex@perex.cz> Link: https://lore.kernel.org/r/20210523090920.15345-3-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-23 09:09:16 +00:00
result = snd_power_ref_and_wait(card);
/* validate input values */
if (IS_ENABLED(CONFIG_SND_CTL_INPUT_VALIDATION) && !result) {
struct snd_ctl_elem_info info;
memset(&info, 0, sizeof(info));
info.id = control->id;
result = __snd_ctl_elem_info(card, kctl, &info, NULL);
if (!result)
result = sanity_check_input_values(card, control, &info,
false);
}
ALSA: control: Track in-flight control read/write/tlv accesses Although the power state check is performed in various places (e.g. at the entrance of quite a few ioctls), there can be still some pending tasks that already went into the ioctl handler or other ops, and those may access the hardware even after the power state check. For example, kcontrol access ioctl paths that call info/get/put callbacks may update the hardware registers. If a system wants to assure the free from such hw access (like the case of PCI rescan feature we're going to implement in future), this situation must be avoided, and we have to sync such in-flight tasks finishing beforehand. For that purpose, this patch introduces a few new things in core code: - A refcount, power_ref, and a wait queue, power_ref_sleep, to the card object - A few new helpers, snd_power_ref(), snd_power_unref(), snd_power_ref_and_wait(), and snd_power_sync_ref() In the code paths that call kctl info/read/write/tlv ops, we check the power state with the newly introduced snd_power_ref_and_wait(). This function also takes the card.power_ref refcount for tracking this in-flight task. Once after the access finishes, snd_power_unref() is called to released the refcount in return. So the driver can sync via snd_power_sync_ref() assuring that all in-flight tasks have been finished. As of this patch, snd_power_sync_ref() is called only at snd_card_disconnect(), but it'll be used in other places in future. Note that atomic_t is used for power_ref intentionally instead of refcount_t. It's because of the design of refcount_t type; refcount_t cannot be zero-based, and it cannot do dec_and_test() call for multiple times, hence it's not suitable for our purpose. Also, this patch changes snd_power_wait() to accept only SNDRV_CTL_POWER_D0, which is the only value that makes sense. In later patch, the snd_power_wait() calls will be cleaned up. Reviewed-by: Jaroslav Kysela <perex@perex.cz> Link: https://lore.kernel.org/r/20210523090920.15345-3-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-23 09:09:16 +00:00
if (!result)
result = kctl->put(kctl, control);
snd_power_unref(card);
if (result < 0) {
up_write(&card->controls_rwsem);
return result;
}
if (result > 0) {
downgrade_write(&card->controls_rwsem);
snd_ctl_notify_one(card, SNDRV_CTL_EVENT_MASK_VALUE, kctl, index_offset);
up_read(&card->controls_rwsem);
} else {
up_write(&card->controls_rwsem);
}
return 0;
}
static int snd_ctl_elem_write_user(struct snd_ctl_file *file,
struct snd_ctl_elem_value __user *_control)
{
struct snd_ctl_elem_value *control;
struct snd_card *card;
int result;
control = memdup_user(_control, sizeof(*control));
if (IS_ERR(control))
return PTR_ERR(control);
card = file->card;
result = snd_ctl_elem_write(card, file, control);
if (result < 0)
goto error;
if (copy_to_user(_control, control, sizeof(*control)))
result = -EFAULT;
error:
kfree(control);
return result;
}
static int snd_ctl_elem_lock(struct snd_ctl_file *file,
struct snd_ctl_elem_id __user *_id)
{
struct snd_card *card = file->card;
struct snd_ctl_elem_id id;
struct snd_kcontrol *kctl;
struct snd_kcontrol_volatile *vd;
int result;
if (copy_from_user(&id, _id, sizeof(id)))
return -EFAULT;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, &id);
if (kctl == NULL) {
result = -ENOENT;
} else {
vd = &kctl->vd[snd_ctl_get_ioff(kctl, &id)];
if (vd->owner != NULL)
result = -EBUSY;
else {
vd->owner = file;
result = 0;
}
}
up_write(&card->controls_rwsem);
return result;
}
static int snd_ctl_elem_unlock(struct snd_ctl_file *file,
struct snd_ctl_elem_id __user *_id)
{
struct snd_card *card = file->card;
struct snd_ctl_elem_id id;
struct snd_kcontrol *kctl;
struct snd_kcontrol_volatile *vd;
int result;
if (copy_from_user(&id, _id, sizeof(id)))
return -EFAULT;
down_write(&card->controls_rwsem);
kctl = snd_ctl_find_id_locked(card, &id);
if (kctl == NULL) {
result = -ENOENT;
} else {
vd = &kctl->vd[snd_ctl_get_ioff(kctl, &id)];
if (vd->owner == NULL)
result = -EINVAL;
else if (vd->owner != file)
result = -EPERM;
else {
vd->owner = NULL;
result = 0;
}
}
up_write(&card->controls_rwsem);
return result;
}
struct user_element {
struct snd_ctl_elem_info info;
struct snd_card *card;
char *elem_data; /* element data */
unsigned long elem_data_size; /* size of element data in bytes */
void *tlv_data; /* TLV data */
unsigned long tlv_data_size; /* TLV data size */
void *priv_data; /* private data (like strings for enumerated type) */
};
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// check whether the addition (in bytes) of user ctl element may overflow the limit.
static bool check_user_elem_overflow(struct snd_card *card, ssize_t add)
{
return (ssize_t)card->user_ctl_alloc_size + add > max_user_ctl_alloc_size;
}
static int snd_ctl_elem_user_info(struct snd_kcontrol *kcontrol,
struct snd_ctl_elem_info *uinfo)
{
struct user_element *ue = kcontrol->private_data;
unsigned int offset;
offset = snd_ctl_get_ioff(kcontrol, &uinfo->id);
*uinfo = ue->info;
snd_ctl_build_ioff(&uinfo->id, kcontrol, offset);
return 0;
}
static int snd_ctl_elem_user_enum_info(struct snd_kcontrol *kcontrol,
struct snd_ctl_elem_info *uinfo)
{
struct user_element *ue = kcontrol->private_data;
const char *names;
unsigned int item;
unsigned int offset;
item = uinfo->value.enumerated.item;
offset = snd_ctl_get_ioff(kcontrol, &uinfo->id);
*uinfo = ue->info;
snd_ctl_build_ioff(&uinfo->id, kcontrol, offset);
item = min(item, uinfo->value.enumerated.items - 1);
uinfo->value.enumerated.item = item;
names = ue->priv_data;
for (; item > 0; --item)
names += strlen(names) + 1;
strcpy(uinfo->value.enumerated.name, names);
return 0;
}
static int snd_ctl_elem_user_get(struct snd_kcontrol *kcontrol,
struct snd_ctl_elem_value *ucontrol)
{
struct user_element *ue = kcontrol->private_data;
unsigned int size = ue->elem_data_size;
char *src = ue->elem_data +
snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size;
memcpy(&ucontrol->value, src, size);
return 0;
}
static int snd_ctl_elem_user_put(struct snd_kcontrol *kcontrol,
struct snd_ctl_elem_value *ucontrol)
{
int change;
struct user_element *ue = kcontrol->private_data;
unsigned int size = ue->elem_data_size;
char *dst = ue->elem_data +
snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size;
change = memcmp(&ucontrol->value, dst, size) != 0;
if (change)
memcpy(dst, &ucontrol->value, size);
return change;
}
ALSA: control: Fix racy management of user ctl memory size account We've got a report about the possible race in the user control element counts (card->user_ctl_count), and it was confirmed that the race wasn't serious in the old code up to 5.12. There, the value modification itself was exclusive and protected via a write semaphore, hence it's at most concurrent reads and evaluations before the increment. Since it's only about the soft-limit to avoid the exhausting memory usage, one-off isn't a big problem at all. Meanwhile, the relevant code has been largely modified recently, and now card->user_ctl_count was replaced with card->user_ctl_alloc_size, and a few more places were added to access this field. And, in this new code, it turned out to be more serious: the modifications are scattered in various places, and a few of them are without protection. It implies that it may lead to an inconsistent value by racy accesses. For addressing it, this patch extends the range covered by the card->controls_rwsem write lock at snd_ctl_elem_add() so that the all code paths that modify and refer to card->user_ctl_alloc_size are protected by the rwsem properly. The patch adds also comments in a couple of functions to indicate that they are under the rwsem lock. Fixes: 66c6d1ef86ff ("ALSA: control: Add memory consumption limit to user controls") Link: https://lore.kernel.org/r/FEEBF384-44BE-42CF-8FB3-93470933F64F@purdue.edu Link: https://lore.kernel.org/r/20210415131856.13113-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-15 13:18:56 +00:00
/* called in controls_rwsem write lock */
static int replace_user_tlv(struct snd_kcontrol *kctl, unsigned int __user *buf,
unsigned int size)
{
struct user_element *ue = kctl->private_data;
unsigned int *container;
unsigned int mask = 0;
int i;
int change;
lockdep_assert_held_write(&ue->card->controls_rwsem);
if (size > 1024 * 128) /* sane value */
return -EINVAL;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// does the TLV size change cause overflow?
if (check_user_elem_overflow(ue->card, (ssize_t)(size - ue->tlv_data_size)))
return -ENOMEM;
container = vmemdup_user(buf, size);
if (IS_ERR(container))
return PTR_ERR(container);
change = ue->tlv_data_size != size;
if (!change)
change = memcmp(ue->tlv_data, container, size) != 0;
if (!change) {
kvfree(container);
return 0;
}
if (ue->tlv_data == NULL) {
/* Now TLV data is available. */
for (i = 0; i < kctl->count; ++i)
kctl->vd[i].access |= SNDRV_CTL_ELEM_ACCESS_TLV_READ;
mask = SNDRV_CTL_EVENT_MASK_INFO;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
} else {
ue->card->user_ctl_alloc_size -= ue->tlv_data_size;
ue->tlv_data_size = 0;
kvfree(ue->tlv_data);
}
ue->tlv_data = container;
ue->tlv_data_size = size;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// decremented at private_free.
ue->card->user_ctl_alloc_size += size;
mask |= SNDRV_CTL_EVENT_MASK_TLV;
for (i = 0; i < kctl->count; ++i)
snd_ctl_notify_one(ue->card, mask, kctl, i);
ALSA: control: delegate TLV eventing to each driver In a design of ALSA control core, a set of elements is represented by 'struct snd_kcontrol' to share common attributes. The set of elements shares TLV (Type-Length-Value) data, too. On the other hand, in ALSA control interface/protocol for applications, a TLV operation is committed to an element. Totally, the operation can have sub-effect to the other elements in the set. For example, TLV_WRITE operation is expected to change TLV data, which returns to applications. Applications attempt to change the TLV data per element, but in the above design, they can effect to elements in the same set. As a default, ALSA control core has no implementation except for TLV_READ operation. Thus, the above design looks to have no issue. However, in kernel APIs of ALSA control component, developers can program a handler for any request of the TLV operation. Therefore, for elements in a set which has the handler, applications can commit TLV_WRITE and TLV_COMMAND requests. For the above scenario, ALSA control core assist notification. When the handler returns positive value, the core queueing an event for a requested element. However, this includes design defects that the event is not queued for the other element in a set. Actually, developers can program the handlers to keep per-element TLV data, but it depends on each driver. As of v4.13-rc6, there's no driver in tree to utilize the notification, except for user-defined element set. This commit delegates the notification into each driver to prevent developers from the design defects. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2017-08-24 01:46:14 +00:00
return change;
}
static int read_user_tlv(struct snd_kcontrol *kctl, unsigned int __user *buf,
unsigned int size)
{
struct user_element *ue = kctl->private_data;
if (ue->tlv_data_size == 0 || ue->tlv_data == NULL)
return -ENXIO;
if (size < ue->tlv_data_size)
return -ENOSPC;
if (copy_to_user(buf, ue->tlv_data, ue->tlv_data_size))
return -EFAULT;
return 0;
}
static int snd_ctl_elem_user_tlv(struct snd_kcontrol *kctl, int op_flag,
unsigned int size, unsigned int __user *buf)
{
if (op_flag == SNDRV_CTL_TLV_OP_WRITE)
return replace_user_tlv(kctl, buf, size);
else
return read_user_tlv(kctl, buf, size);
}
ALSA: control: Fix racy management of user ctl memory size account We've got a report about the possible race in the user control element counts (card->user_ctl_count), and it was confirmed that the race wasn't serious in the old code up to 5.12. There, the value modification itself was exclusive and protected via a write semaphore, hence it's at most concurrent reads and evaluations before the increment. Since it's only about the soft-limit to avoid the exhausting memory usage, one-off isn't a big problem at all. Meanwhile, the relevant code has been largely modified recently, and now card->user_ctl_count was replaced with card->user_ctl_alloc_size, and a few more places were added to access this field. And, in this new code, it turned out to be more serious: the modifications are scattered in various places, and a few of them are without protection. It implies that it may lead to an inconsistent value by racy accesses. For addressing it, this patch extends the range covered by the card->controls_rwsem write lock at snd_ctl_elem_add() so that the all code paths that modify and refer to card->user_ctl_alloc_size are protected by the rwsem properly. The patch adds also comments in a couple of functions to indicate that they are under the rwsem lock. Fixes: 66c6d1ef86ff ("ALSA: control: Add memory consumption limit to user controls") Link: https://lore.kernel.org/r/FEEBF384-44BE-42CF-8FB3-93470933F64F@purdue.edu Link: https://lore.kernel.org/r/20210415131856.13113-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-15 13:18:56 +00:00
/* called in controls_rwsem write lock */
static int snd_ctl_elem_init_enum_names(struct user_element *ue)
{
char *names, *p;
size_t buf_len, name_len;
unsigned int i;
const uintptr_t user_ptrval = ue->info.value.enumerated.names_ptr;
lockdep_assert_held_write(&ue->card->controls_rwsem);
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
buf_len = ue->info.value.enumerated.names_length;
if (buf_len > 64 * 1024)
return -EINVAL;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
if (check_user_elem_overflow(ue->card, buf_len))
return -ENOMEM;
names = vmemdup_user((const void __user *)user_ptrval, buf_len);
if (IS_ERR(names))
return PTR_ERR(names);
/* check that there are enough valid names */
p = names;
for (i = 0; i < ue->info.value.enumerated.items; ++i) {
name_len = strnlen(p, buf_len);
if (name_len == 0 || name_len >= 64 || name_len == buf_len) {
kvfree(names);
return -EINVAL;
}
p += name_len + 1;
buf_len -= name_len + 1;
}
ue->priv_data = names;
ue->info.value.enumerated.names_ptr = 0;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// increment the allocation size; decremented again at private_free.
ue->card->user_ctl_alloc_size += ue->info.value.enumerated.names_length;
return 0;
}
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
static size_t compute_user_elem_size(size_t size, unsigned int count)
{
return sizeof(struct user_element) + size * count;
}
static void snd_ctl_elem_user_free(struct snd_kcontrol *kcontrol)
{
struct user_element *ue = kcontrol->private_data;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// decrement the allocation size.
ue->card->user_ctl_alloc_size -= compute_user_elem_size(ue->elem_data_size, kcontrol->count);
ue->card->user_ctl_alloc_size -= ue->tlv_data_size;
if (ue->priv_data)
ue->card->user_ctl_alloc_size -= ue->info.value.enumerated.names_length;
kvfree(ue->tlv_data);
kvfree(ue->priv_data);
kfree(ue);
}
static int snd_ctl_elem_add(struct snd_ctl_file *file,
struct snd_ctl_elem_info *info, int replace)
{
struct snd_card *card = file->card;
struct snd_kcontrol *kctl;
unsigned int count;
unsigned int access;
long private_size;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
size_t alloc_size;
struct user_element *ue;
unsigned int offset;
int err;
ALSA: control: Fix replacing user controls There are two issues with the current implementation for replacing user controls. The first is that the code does not check if the control is actually a user control and neither does it check if the control is owned by the process that tries to remove it. That allows userspace applications to remove arbitrary controls, which can cause a user after free if a for example a driver does not expect a control to be removed from under its feed. The second issue is that on one hand when a control is replaced the user_ctl_count limit is not checked and on the other hand the user_ctl_count is increased (even though the number of user controls does not change). This allows userspace, once the user_ctl_count limit as been reached, to repeatedly replace a control until user_ctl_count overflows. Once that happens new controls can be added effectively bypassing the user_ctl_count limit. Both issues can be fixed by instead of open-coding the removal of the control that is to be replaced to use snd_ctl_remove_user_ctl(). This function does proper permission checks as well as decrements user_ctl_count after the control has been removed. Note that by using snd_ctl_remove_user_ctl() the check which returns -EBUSY at beginning of the function if the control already exists is removed. This is not a problem though since the check is quite useless, because the lock that is protecting the control list is released between the check and before adding the new control to the list, which means that it is possible that a different control with the same settings is added to the list after the check. Luckily there is another check that is done while holding the lock in snd_ctl_add(), so we'll rely on that to make sure that the same control is not added twice. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Jaroslav Kysela <perex@perex.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-06-18 11:32:32 +00:00
if (!*info->id.name)
return -EINVAL;
if (strnlen(info->id.name, sizeof(info->id.name)) >= sizeof(info->id.name))
return -EINVAL;
ALSA: control: Fix replacing user controls There are two issues with the current implementation for replacing user controls. The first is that the code does not check if the control is actually a user control and neither does it check if the control is owned by the process that tries to remove it. That allows userspace applications to remove arbitrary controls, which can cause a user after free if a for example a driver does not expect a control to be removed from under its feed. The second issue is that on one hand when a control is replaced the user_ctl_count limit is not checked and on the other hand the user_ctl_count is increased (even though the number of user controls does not change). This allows userspace, once the user_ctl_count limit as been reached, to repeatedly replace a control until user_ctl_count overflows. Once that happens new controls can be added effectively bypassing the user_ctl_count limit. Both issues can be fixed by instead of open-coding the removal of the control that is to be replaced to use snd_ctl_remove_user_ctl(). This function does proper permission checks as well as decrements user_ctl_count after the control has been removed. Note that by using snd_ctl_remove_user_ctl() the check which returns -EBUSY at beginning of the function if the control already exists is removed. This is not a problem though since the check is quite useless, because the lock that is protecting the control list is released between the check and before adding the new control to the list, which means that it is possible that a different control with the same settings is added to the list after the check. Luckily there is another check that is done while holding the lock in snd_ctl_add(), so we'll rely on that to make sure that the same control is not added twice. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Jaroslav Kysela <perex@perex.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-06-18 11:32:32 +00:00
/* Delete a control to replace them if needed. */
ALSA: control: Fix replacing user controls There are two issues with the current implementation for replacing user controls. The first is that the code does not check if the control is actually a user control and neither does it check if the control is owned by the process that tries to remove it. That allows userspace applications to remove arbitrary controls, which can cause a user after free if a for example a driver does not expect a control to be removed from under its feed. The second issue is that on one hand when a control is replaced the user_ctl_count limit is not checked and on the other hand the user_ctl_count is increased (even though the number of user controls does not change). This allows userspace, once the user_ctl_count limit as been reached, to repeatedly replace a control until user_ctl_count overflows. Once that happens new controls can be added effectively bypassing the user_ctl_count limit. Both issues can be fixed by instead of open-coding the removal of the control that is to be replaced to use snd_ctl_remove_user_ctl(). This function does proper permission checks as well as decrements user_ctl_count after the control has been removed. Note that by using snd_ctl_remove_user_ctl() the check which returns -EBUSY at beginning of the function if the control already exists is removed. This is not a problem though since the check is quite useless, because the lock that is protecting the control list is released between the check and before adding the new control to the list, which means that it is possible that a different control with the same settings is added to the list after the check. Luckily there is another check that is done while holding the lock in snd_ctl_add(), so we'll rely on that to make sure that the same control is not added twice. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Jaroslav Kysela <perex@perex.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-06-18 11:32:32 +00:00
if (replace) {
info->id.numid = 0;
ALSA: control: Fix replacing user controls There are two issues with the current implementation for replacing user controls. The first is that the code does not check if the control is actually a user control and neither does it check if the control is owned by the process that tries to remove it. That allows userspace applications to remove arbitrary controls, which can cause a user after free if a for example a driver does not expect a control to be removed from under its feed. The second issue is that on one hand when a control is replaced the user_ctl_count limit is not checked and on the other hand the user_ctl_count is increased (even though the number of user controls does not change). This allows userspace, once the user_ctl_count limit as been reached, to repeatedly replace a control until user_ctl_count overflows. Once that happens new controls can be added effectively bypassing the user_ctl_count limit. Both issues can be fixed by instead of open-coding the removal of the control that is to be replaced to use snd_ctl_remove_user_ctl(). This function does proper permission checks as well as decrements user_ctl_count after the control has been removed. Note that by using snd_ctl_remove_user_ctl() the check which returns -EBUSY at beginning of the function if the control already exists is removed. This is not a problem though since the check is quite useless, because the lock that is protecting the control list is released between the check and before adding the new control to the list, which means that it is possible that a different control with the same settings is added to the list after the check. Luckily there is another check that is done while holding the lock in snd_ctl_add(), so we'll rely on that to make sure that the same control is not added twice. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Jaroslav Kysela <perex@perex.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-06-18 11:32:32 +00:00
err = snd_ctl_remove_user_ctl(file, &info->id);
if (err)
return err;
}
ALSA: control: Fix replacing user controls There are two issues with the current implementation for replacing user controls. The first is that the code does not check if the control is actually a user control and neither does it check if the control is owned by the process that tries to remove it. That allows userspace applications to remove arbitrary controls, which can cause a user after free if a for example a driver does not expect a control to be removed from under its feed. The second issue is that on one hand when a control is replaced the user_ctl_count limit is not checked and on the other hand the user_ctl_count is increased (even though the number of user controls does not change). This allows userspace, once the user_ctl_count limit as been reached, to repeatedly replace a control until user_ctl_count overflows. Once that happens new controls can be added effectively bypassing the user_ctl_count limit. Both issues can be fixed by instead of open-coding the removal of the control that is to be replaced to use snd_ctl_remove_user_ctl(). This function does proper permission checks as well as decrements user_ctl_count after the control has been removed. Note that by using snd_ctl_remove_user_ctl() the check which returns -EBUSY at beginning of the function if the control already exists is removed. This is not a problem though since the check is quite useless, because the lock that is protecting the control list is released between the check and before adding the new control to the list, which means that it is possible that a different control with the same settings is added to the list after the check. Luckily there is another check that is done while holding the lock in snd_ctl_add(), so we'll rely on that to make sure that the same control is not added twice. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Jaroslav Kysela <perex@perex.cz> Cc: <stable@vger.kernel.org> Signed-off-by: Takashi Iwai <tiwai@suse.de>
2014-06-18 11:32:32 +00:00
/* Check the number of elements for this userspace control. */
count = info->owner;
if (count == 0)
count = 1;
/* Arrange access permissions if needed. */
access = info->access;
if (access == 0)
access = SNDRV_CTL_ELEM_ACCESS_READWRITE;
access &= (SNDRV_CTL_ELEM_ACCESS_READWRITE |
SNDRV_CTL_ELEM_ACCESS_INACTIVE |
SNDRV_CTL_ELEM_ACCESS_TLV_WRITE);
/* In initial state, nothing is available as TLV container. */
if (access & SNDRV_CTL_ELEM_ACCESS_TLV_WRITE)
access |= SNDRV_CTL_ELEM_ACCESS_TLV_CALLBACK;
access |= SNDRV_CTL_ELEM_ACCESS_USER;
/*
* Check information and calculate the size of data specific to
* this userspace control.
*/
ALSA: control: Add verification for kctl accesses The current implementation of ALSA control API fully relies on the callbacks of each driver, and there is no verification of the values passed via API. This patch is an attempt to improve the situation slightly by adding the validation code for the values stored via info and get callbacks. The patch adds a new kconfig, CONFIG_SND_CTL_VALIDATION. It depends on CONFIG_SND_DEBUG and off as default since the validation would require a slight overhead including the additional call of info callback at each get callback invocation. When this config is enabled, the values stored by each info callback invocation are verified, namely: - Whether the info type is valid - Whether the number of enum items is non-zero - Whether the given info count is within the allowed boundary Similarly, the values stored at each get callback are verified as well: - Whether the values are within the given range - Whether the values are aligned with the given step - Whether any further changes are seen in the data array over the given info count The last point helps identifying a possibly invalid data type access, typically a case where the info callback declares the type being SNDRV_CTL_ELEM_TYPE_ENUMERATED while the get/put callbacks store the values in value.integer.value[] array. When a validation fails, the ALSA core logs an error message including the device and the control ID, and the API call also returns an error. So, with the new validation turned on, the driver behavior difference may be visible on user-space, too -- it's intentional, though, so that we can catch an error more clearly. The patch also introduces a new ctl access type, SNDRV_CTL_ELEM_ACCESS_SKIP_CHECK. A driver may pass this flag with other access bits to indicate that the ctl element won't be verified. It's useful when a driver code is specially written to access the data greater than info->count size by some reason. For example, this flag is actually set now in HD-audio HDMI codec driver which needs to clear the data array in the case of the disconnected monitor. Also, the PCM channel-map helper code is slightly modified to avoid the false-positive hit by this validation code, too. Link: https://lore.kernel.org/r/20200104083556.27789-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2020-01-04 08:35:56 +00:00
/* pass NULL to card for suppressing error messages */
err = snd_ctl_check_elem_info(NULL, info);
if (err < 0)
return err;
/* user-space control doesn't allow zero-size data */
if (info->count < 1)
return -EINVAL;
private_size = value_sizes[info->type] * info->count;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
alloc_size = compute_user_elem_size(private_size, count);
ALSA: control: Fix racy management of user ctl memory size account We've got a report about the possible race in the user control element counts (card->user_ctl_count), and it was confirmed that the race wasn't serious in the old code up to 5.12. There, the value modification itself was exclusive and protected via a write semaphore, hence it's at most concurrent reads and evaluations before the increment. Since it's only about the soft-limit to avoid the exhausting memory usage, one-off isn't a big problem at all. Meanwhile, the relevant code has been largely modified recently, and now card->user_ctl_count was replaced with card->user_ctl_alloc_size, and a few more places were added to access this field. And, in this new code, it turned out to be more serious: the modifications are scattered in various places, and a few of them are without protection. It implies that it may lead to an inconsistent value by racy accesses. For addressing it, this patch extends the range covered by the card->controls_rwsem write lock at snd_ctl_elem_add() so that the all code paths that modify and refer to card->user_ctl_alloc_size are protected by the rwsem properly. The patch adds also comments in a couple of functions to indicate that they are under the rwsem lock. Fixes: 66c6d1ef86ff ("ALSA: control: Add memory consumption limit to user controls") Link: https://lore.kernel.org/r/FEEBF384-44BE-42CF-8FB3-93470933F64F@purdue.edu Link: https://lore.kernel.org/r/20210415131856.13113-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-15 13:18:56 +00:00
down_write(&card->controls_rwsem);
if (check_user_elem_overflow(card, alloc_size)) {
err = -ENOMEM;
goto unlock;
}
/*
* Keep memory object for this userspace control. After passing this
* code block, the instance should be freed by snd_ctl_free_one().
*
* Note that these elements in this control are locked.
*/
err = snd_ctl_new(&kctl, count, access, file);
if (err < 0)
ALSA: control: Fix racy management of user ctl memory size account We've got a report about the possible race in the user control element counts (card->user_ctl_count), and it was confirmed that the race wasn't serious in the old code up to 5.12. There, the value modification itself was exclusive and protected via a write semaphore, hence it's at most concurrent reads and evaluations before the increment. Since it's only about the soft-limit to avoid the exhausting memory usage, one-off isn't a big problem at all. Meanwhile, the relevant code has been largely modified recently, and now card->user_ctl_count was replaced with card->user_ctl_alloc_size, and a few more places were added to access this field. And, in this new code, it turned out to be more serious: the modifications are scattered in various places, and a few of them are without protection. It implies that it may lead to an inconsistent value by racy accesses. For addressing it, this patch extends the range covered by the card->controls_rwsem write lock at snd_ctl_elem_add() so that the all code paths that modify and refer to card->user_ctl_alloc_size are protected by the rwsem properly. The patch adds also comments in a couple of functions to indicate that they are under the rwsem lock. Fixes: 66c6d1ef86ff ("ALSA: control: Add memory consumption limit to user controls") Link: https://lore.kernel.org/r/FEEBF384-44BE-42CF-8FB3-93470933F64F@purdue.edu Link: https://lore.kernel.org/r/20210415131856.13113-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-15 13:18:56 +00:00
goto unlock;
memcpy(&kctl->id, &info->id, sizeof(kctl->id));
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
ue = kzalloc(alloc_size, GFP_KERNEL);
if (!ue) {
kfree(kctl);
ALSA: control: Fix racy management of user ctl memory size account We've got a report about the possible race in the user control element counts (card->user_ctl_count), and it was confirmed that the race wasn't serious in the old code up to 5.12. There, the value modification itself was exclusive and protected via a write semaphore, hence it's at most concurrent reads and evaluations before the increment. Since it's only about the soft-limit to avoid the exhausting memory usage, one-off isn't a big problem at all. Meanwhile, the relevant code has been largely modified recently, and now card->user_ctl_count was replaced with card->user_ctl_alloc_size, and a few more places were added to access this field. And, in this new code, it turned out to be more serious: the modifications are scattered in various places, and a few of them are without protection. It implies that it may lead to an inconsistent value by racy accesses. For addressing it, this patch extends the range covered by the card->controls_rwsem write lock at snd_ctl_elem_add() so that the all code paths that modify and refer to card->user_ctl_alloc_size are protected by the rwsem properly. The patch adds also comments in a couple of functions to indicate that they are under the rwsem lock. Fixes: 66c6d1ef86ff ("ALSA: control: Add memory consumption limit to user controls") Link: https://lore.kernel.org/r/FEEBF384-44BE-42CF-8FB3-93470933F64F@purdue.edu Link: https://lore.kernel.org/r/20210415131856.13113-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-15 13:18:56 +00:00
err = -ENOMEM;
goto unlock;
}
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
kctl->private_data = ue;
kctl->private_free = snd_ctl_elem_user_free;
ALSA: control: Add memory consumption limit to user controls ALSA control interface allows users to add arbitrary control elements (called "user controls" or "user elements"), and its resource usage is limited just by the max number of control sets (currently 32). This limit, however, is quite loose: each allocation of control set may have 1028 elements, and each element may have up to 512 bytes (ILP32) or 1024 bytes (LP64) of value data. Moreover, each control set may contain the enum strings and TLV data, which can be up to 64kB and 128kB, respectively. Totally, the whole memory consumption may go over 38MB -- it's quite large, and we'd rather like to reduce the size. OTOH, there have been other requests even to increase the max number of user elements; e.g. ALSA firewire stack require the more user controls, hence we want to raise the bar, too. For satisfying both requirements, this patch changes the management of user controls: instead of setting the upper limit of the number of user controls, we check the actual memory allocation size and set the upper limit of the total allocation in bytes. As long as the memory consumption stays below the limit, more user controls are allowed than the current limit 32. At the same time, we set the lower limit (8MB) as default than the current theoretical limit, in order to lower the risk of DoS. As a compromise for lowering the default limit, now the actual memory limit is defined as a module option, 'max_user_ctl_alloc_size', so that user can increase/decrease the limit if really needed, too. Link: https://lore.kernel.org/r/s5htur3zl5e.wl-tiwai@suse.de Co-developed-by: Takashi Iwai <tiwai@suse.de> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Tested-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> Link: https://lore.kernel.org/r/20210408103149.40357-1-o-takashi@sakamocchi.jp Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-08 10:31:49 +00:00
// increment the allocated size; decremented again at private_free.
card->user_ctl_alloc_size += alloc_size;
/* Set private data for this userspace control. */
ue->card = card;
ue->info = *info;
ue->info.access = 0;
ue->elem_data = (char *)ue + sizeof(*ue);
ue->elem_data_size = private_size;
if (ue->info.type == SNDRV_CTL_ELEM_TYPE_ENUMERATED) {
err = snd_ctl_elem_init_enum_names(ue);
if (err < 0) {
snd_ctl_free_one(kctl);
ALSA: control: Fix racy management of user ctl memory size account We've got a report about the possible race in the user control element counts (card->user_ctl_count), and it was confirmed that the race wasn't serious in the old code up to 5.12. There, the value modification itself was exclusive and protected via a write semaphore, hence it's at most concurrent reads and evaluations before the increment. Since it's only about the soft-limit to avoid the exhausting memory usage, one-off isn't a big problem at all. Meanwhile, the relevant code has been largely modified recently, and now card->user_ctl_count was replaced with card->user_ctl_alloc_size, and a few more places were added to access this field. And, in this new code, it turned out to be more serious: the modifications are scattered in various places, and a few of them are without protection. It implies that it may lead to an inconsistent value by racy accesses. For addressing it, this patch extends the range covered by the card->controls_rwsem write lock at snd_ctl_elem_add() so that the all code paths that modify and refer to card->user_ctl_alloc_size are protected by the rwsem properly. The patch adds also comments in a couple of functions to indicate that they are under the rwsem lock. Fixes: 66c6d1ef86ff ("ALSA: control: Add memory consumption limit to user controls") Link: https://lore.kernel.org/r/FEEBF384-44BE-42CF-8FB3-93470933F64F@purdue.edu Link: https://lore.kernel.org/r/20210415131856.13113-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-04-15 13:18:56 +00:00
goto unlock;
}
}
/* Set callback functions. */
if (info->type == SNDRV_CTL_ELEM_TYPE_ENUMERATED)
kctl->info = snd_ctl_elem_user_enum_info;
else
kctl->info = snd_ctl_elem_user_info;
if (access & SNDRV_CTL_ELEM_ACCESS_READ)
kctl->get = snd_ctl_elem_user_get;
if (access & SNDRV_CTL_ELEM_ACCESS_WRITE)
kctl->put = snd_ctl_elem_user_put;
if (access & SNDRV_CTL_ELEM_ACCESS_TLV_WRITE)
kctl->tlv.c = snd_ctl_elem_user_tlv;
/* This function manage to free the instance on failure. */
err = __snd_ctl_add_replace(card, kctl, CTL_ADD_EXCLUSIVE);
if (err < 0) {
snd_ctl_free_one(kctl);
goto unlock;
}
offset = snd_ctl_get_ioff(kctl, &info->id);
snd_ctl_build_ioff(&info->id, kctl, offset);
/*
* Here we cannot fill any field for the number of elements added by
* this operation because there're no specific fields. The usage of
* 'owner' field for this purpose may cause any bugs to userspace
* applications because the field originally means PID of a process
* which locks the element.
*/
unlock:
up_write(&card->controls_rwsem);
return err;
}
static int snd_ctl_elem_add_user(struct snd_ctl_file *file,
struct snd_ctl_elem_info __user *_info, int replace)
{
struct snd_ctl_elem_info info;
int err;
if (copy_from_user(&info, _info, sizeof(info)))
return -EFAULT;
err = snd_ctl_elem_add(file, &info, replace);
if (err < 0)
return err;
if (copy_to_user(_info, &info, sizeof(info))) {
snd_ctl_remove_user_ctl(file, &info.id);
return -EFAULT;
}
return 0;
}
static int snd_ctl_elem_remove(struct snd_ctl_file *file,
struct snd_ctl_elem_id __user *_id)
{
struct snd_ctl_elem_id id;
if (copy_from_user(&id, _id, sizeof(id)))
return -EFAULT;
return snd_ctl_remove_user_ctl(file, &id);
}
static int snd_ctl_subscribe_events(struct snd_ctl_file *file, int __user *ptr)
{
int subscribe;
if (get_user(subscribe, ptr))
return -EFAULT;
if (subscribe < 0) {
subscribe = file->subscribed;
if (put_user(subscribe, ptr))
return -EFAULT;
return 0;
}
if (subscribe) {
file->subscribed = 1;
return 0;
} else if (file->subscribed) {
snd_ctl_empty_read_queue(file);
file->subscribed = 0;
}
return 0;
}
static int call_tlv_handler(struct snd_ctl_file *file, int op_flag,
struct snd_kcontrol *kctl,
struct snd_ctl_elem_id *id,
unsigned int __user *buf, unsigned int size)
{
static const struct {
int op;
int perm;
} pairs[] = {
{SNDRV_CTL_TLV_OP_READ, SNDRV_CTL_ELEM_ACCESS_TLV_READ},
{SNDRV_CTL_TLV_OP_WRITE, SNDRV_CTL_ELEM_ACCESS_TLV_WRITE},
{SNDRV_CTL_TLV_OP_CMD, SNDRV_CTL_ELEM_ACCESS_TLV_COMMAND},
};
struct snd_kcontrol_volatile *vd = &kctl->vd[snd_ctl_get_ioff(kctl, id)];
ALSA: control: Track in-flight control read/write/tlv accesses Although the power state check is performed in various places (e.g. at the entrance of quite a few ioctls), there can be still some pending tasks that already went into the ioctl handler or other ops, and those may access the hardware even after the power state check. For example, kcontrol access ioctl paths that call info/get/put callbacks may update the hardware registers. If a system wants to assure the free from such hw access (like the case of PCI rescan feature we're going to implement in future), this situation must be avoided, and we have to sync such in-flight tasks finishing beforehand. For that purpose, this patch introduces a few new things in core code: - A refcount, power_ref, and a wait queue, power_ref_sleep, to the card object - A few new helpers, snd_power_ref(), snd_power_unref(), snd_power_ref_and_wait(), and snd_power_sync_ref() In the code paths that call kctl info/read/write/tlv ops, we check the power state with the newly introduced snd_power_ref_and_wait(). This function also takes the card.power_ref refcount for tracking this in-flight task. Once after the access finishes, snd_power_unref() is called to released the refcount in return. So the driver can sync via snd_power_sync_ref() assuring that all in-flight tasks have been finished. As of this patch, snd_power_sync_ref() is called only at snd_card_disconnect(), but it'll be used in other places in future. Note that atomic_t is used for power_ref intentionally instead of refcount_t. It's because of the design of refcount_t type; refcount_t cannot be zero-based, and it cannot do dec_and_test() call for multiple times, hence it's not suitable for our purpose. Also, this patch changes snd_power_wait() to accept only SNDRV_CTL_POWER_D0, which is the only value that makes sense. In later patch, the snd_power_wait() calls will be cleaned up. Reviewed-by: Jaroslav Kysela <perex@perex.cz> Link: https://lore.kernel.org/r/20210523090920.15345-3-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-23 09:09:16 +00:00
int i, ret;
/* Check support of the request for this element. */
for (i = 0; i < ARRAY_SIZE(pairs); ++i) {
if (op_flag == pairs[i].op && (vd->access & pairs[i].perm))
break;
}
if (i == ARRAY_SIZE(pairs))
return -ENXIO;
if (kctl->tlv.c == NULL)
return -ENXIO;
2019-12-23 09:33:47 +00:00
/* Write and command operations are not allowed for locked element. */
if (op_flag != SNDRV_CTL_TLV_OP_READ &&
vd->owner != NULL && vd->owner != file)
return -EPERM;
ALSA: control: Track in-flight control read/write/tlv accesses Although the power state check is performed in various places (e.g. at the entrance of quite a few ioctls), there can be still some pending tasks that already went into the ioctl handler or other ops, and those may access the hardware even after the power state check. For example, kcontrol access ioctl paths that call info/get/put callbacks may update the hardware registers. If a system wants to assure the free from such hw access (like the case of PCI rescan feature we're going to implement in future), this situation must be avoided, and we have to sync such in-flight tasks finishing beforehand. For that purpose, this patch introduces a few new things in core code: - A refcount, power_ref, and a wait queue, power_ref_sleep, to the card object - A few new helpers, snd_power_ref(), snd_power_unref(), snd_power_ref_and_wait(), and snd_power_sync_ref() In the code paths that call kctl info/read/write/tlv ops, we check the power state with the newly introduced snd_power_ref_and_wait(). This function also takes the card.power_ref refcount for tracking this in-flight task. Once after the access finishes, snd_power_unref() is called to released the refcount in return. So the driver can sync via snd_power_sync_ref() assuring that all in-flight tasks have been finished. As of this patch, snd_power_sync_ref() is called only at snd_card_disconnect(), but it'll be used in other places in future. Note that atomic_t is used for power_ref intentionally instead of refcount_t. It's because of the design of refcount_t type; refcount_t cannot be zero-based, and it cannot do dec_and_test() call for multiple times, hence it's not suitable for our purpose. Also, this patch changes snd_power_wait() to accept only SNDRV_CTL_POWER_D0, which is the only value that makes sense. In later patch, the snd_power_wait() calls will be cleaned up. Reviewed-by: Jaroslav Kysela <perex@perex.cz> Link: https://lore.kernel.org/r/20210523090920.15345-3-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2021-05-23 09:09:16 +00:00
ret = snd_power_ref_and_wait(file->card);
if (!ret)
ret = kctl->tlv.c(kctl, op_flag, size, buf);
snd_power_unref(file->card);
return ret;
}
static int read_tlv_buf(struct snd_kcontrol *kctl, struct snd_ctl_elem_id *id,
unsigned int __user *buf, unsigned int size)
{
struct snd_kcontrol_volatile *vd = &kctl->vd[snd_ctl_get_ioff(kctl, id)];
unsigned int len;
if (!(vd->access & SNDRV_CTL_ELEM_ACCESS_TLV_READ))
return -ENXIO;
if (kctl->tlv.p == NULL)
return -ENXIO;
len = sizeof(unsigned int) * 2 + kctl->tlv.p[1];
if (size < len)
return -ENOMEM;
if (copy_to_user(buf, kctl->tlv.p, len))
return -EFAULT;
return 0;
}
static int snd_ctl_tlv_ioctl(struct snd_ctl_file *file,
struct snd_ctl_tlv __user *buf,
int op_flag)
{
struct snd_ctl_tlv header;
unsigned int __user *container;
unsigned int container_size;
struct snd_kcontrol *kctl;
struct snd_ctl_elem_id id;
struct snd_kcontrol_volatile *vd;
lockdep_assert_held(&file->card->controls_rwsem);
if (copy_from_user(&header, buf, sizeof(header)))
return -EFAULT;
/* In design of control core, numerical ID starts at 1. */
if (header.numid == 0)
return -EINVAL;
/* At least, container should include type and length fields. */
if (header.length < sizeof(unsigned int) * 2)
return -EINVAL;
container_size = header.length;
container = buf->tlv;
kctl = snd_ctl_find_numid_locked(file->card, header.numid);
if (kctl == NULL)
return -ENOENT;
/* Calculate index of the element in this set. */
id = kctl->id;
snd_ctl_build_ioff(&id, kctl, header.numid - id.numid);
vd = &kctl->vd[snd_ctl_get_ioff(kctl, &id)];
if (vd->access & SNDRV_CTL_ELEM_ACCESS_TLV_CALLBACK) {
return call_tlv_handler(file, op_flag, kctl, &id, container,
container_size);
} else {
if (op_flag == SNDRV_CTL_TLV_OP_READ) {
return read_tlv_buf(kctl, &id, container,
container_size);
}
}
/* Not supported. */
return -ENXIO;
}
static long snd_ctl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
struct snd_ctl_file *ctl;
struct snd_card *card;
struct snd_kctl_ioctl *p;
void __user *argp = (void __user *)arg;
int __user *ip = argp;
int err;
ctl = file->private_data;
card = ctl->card;
if (snd_BUG_ON(!card))
return -ENXIO;
switch (cmd) {
case SNDRV_CTL_IOCTL_PVERSION:
return put_user(SNDRV_CTL_VERSION, ip) ? -EFAULT : 0;
case SNDRV_CTL_IOCTL_CARD_INFO:
return snd_ctl_card_info(card, ctl, cmd, argp);
case SNDRV_CTL_IOCTL_ELEM_LIST:
return snd_ctl_elem_list_user(card, argp);
case SNDRV_CTL_IOCTL_ELEM_INFO:
return snd_ctl_elem_info_user(ctl, argp);
case SNDRV_CTL_IOCTL_ELEM_READ:
return snd_ctl_elem_read_user(card, argp);
case SNDRV_CTL_IOCTL_ELEM_WRITE:
return snd_ctl_elem_write_user(ctl, argp);
case SNDRV_CTL_IOCTL_ELEM_LOCK:
return snd_ctl_elem_lock(ctl, argp);
case SNDRV_CTL_IOCTL_ELEM_UNLOCK:
return snd_ctl_elem_unlock(ctl, argp);
case SNDRV_CTL_IOCTL_ELEM_ADD:
return snd_ctl_elem_add_user(ctl, argp, 0);
case SNDRV_CTL_IOCTL_ELEM_REPLACE:
return snd_ctl_elem_add_user(ctl, argp, 1);
case SNDRV_CTL_IOCTL_ELEM_REMOVE:
return snd_ctl_elem_remove(ctl, argp);
case SNDRV_CTL_IOCTL_SUBSCRIBE_EVENTS:
return snd_ctl_subscribe_events(ctl, ip);
case SNDRV_CTL_IOCTL_TLV_READ:
down_read(&ctl->card->controls_rwsem);
err = snd_ctl_tlv_ioctl(ctl, argp, SNDRV_CTL_TLV_OP_READ);
up_read(&ctl->card->controls_rwsem);
return err;
case SNDRV_CTL_IOCTL_TLV_WRITE:
down_write(&ctl->card->controls_rwsem);
err = snd_ctl_tlv_ioctl(ctl, argp, SNDRV_CTL_TLV_OP_WRITE);
up_write(&ctl->card->controls_rwsem);
return err;
case SNDRV_CTL_IOCTL_TLV_COMMAND:
down_write(&ctl->card->controls_rwsem);
err = snd_ctl_tlv_ioctl(ctl, argp, SNDRV_CTL_TLV_OP_CMD);
up_write(&ctl->card->controls_rwsem);
return err;
case SNDRV_CTL_IOCTL_POWER:
return -ENOPROTOOPT;
case SNDRV_CTL_IOCTL_POWER_STATE:
return put_user(SNDRV_CTL_POWER_D0, ip) ? -EFAULT : 0;
}
down_read(&snd_ioctl_rwsem);
list_for_each_entry(p, &snd_control_ioctls, list) {
err = p->fioctl(card, ctl, cmd, arg);
if (err != -ENOIOCTLCMD) {
up_read(&snd_ioctl_rwsem);
return err;
}
}
up_read(&snd_ioctl_rwsem);
dev_dbg(card->dev, "unknown ioctl = 0x%x\n", cmd);
return -ENOTTY;
}
static ssize_t snd_ctl_read(struct file *file, char __user *buffer,
size_t count, loff_t * offset)
{
struct snd_ctl_file *ctl;
int err = 0;
ssize_t result = 0;
ctl = file->private_data;
if (snd_BUG_ON(!ctl || !ctl->card))
return -ENXIO;
if (!ctl->subscribed)
return -EBADFD;
if (count < sizeof(struct snd_ctl_event))
return -EINVAL;
spin_lock_irq(&ctl->read_lock);
while (count >= sizeof(struct snd_ctl_event)) {
struct snd_ctl_event ev;
struct snd_kctl_event *kev;
while (list_empty(&ctl->events)) {
wait_queue_entry_t wait;
if ((file->f_flags & O_NONBLOCK) != 0 || result > 0) {
err = -EAGAIN;
goto __end_lock;
}
init_waitqueue_entry(&wait, current);
add_wait_queue(&ctl->change_sleep, &wait);
set_current_state(TASK_INTERRUPTIBLE);
spin_unlock_irq(&ctl->read_lock);
schedule();
remove_wait_queue(&ctl->change_sleep, &wait);
if (ctl->card->shutdown)
return -ENODEV;
if (signal_pending(current))
return -ERESTARTSYS;
spin_lock_irq(&ctl->read_lock);
}
kev = snd_kctl_event(ctl->events.next);
ev.type = SNDRV_CTL_EVENT_ELEM;
ev.data.elem.mask = kev->mask;
ev.data.elem.id = kev->id;
list_del(&kev->list);
spin_unlock_irq(&ctl->read_lock);
kfree(kev);
if (copy_to_user(buffer, &ev, sizeof(struct snd_ctl_event))) {
err = -EFAULT;
goto __end;
}
spin_lock_irq(&ctl->read_lock);
buffer += sizeof(struct snd_ctl_event);
count -= sizeof(struct snd_ctl_event);
result += sizeof(struct snd_ctl_event);
}
__end_lock:
spin_unlock_irq(&ctl->read_lock);
__end:
return result > 0 ? result : err;
}
static __poll_t snd_ctl_poll(struct file *file, poll_table * wait)
{
__poll_t mask;
struct snd_ctl_file *ctl;
ctl = file->private_data;
if (!ctl->subscribed)
return 0;
poll_wait(file, &ctl->change_sleep, wait);
mask = 0;
if (!list_empty(&ctl->events))
mask |= EPOLLIN | EPOLLRDNORM;
return mask;
}
/*
* register the device-specific control-ioctls.
* called from each device manager like pcm.c, hwdep.c, etc.
*/
static int _snd_ctl_register_ioctl(snd_kctl_ioctl_func_t fcn, struct list_head *lists)
{
struct snd_kctl_ioctl *pn;
pn = kzalloc(sizeof(struct snd_kctl_ioctl), GFP_KERNEL);
if (pn == NULL)
return -ENOMEM;
pn->fioctl = fcn;
down_write(&snd_ioctl_rwsem);
list_add_tail(&pn->list, lists);
up_write(&snd_ioctl_rwsem);
return 0;
}
/**
* snd_ctl_register_ioctl - register the device-specific control-ioctls
* @fcn: ioctl callback function
*
* called from each device manager like pcm.c, hwdep.c, etc.
*
* Return: zero if successful, or a negative error code
*/
int snd_ctl_register_ioctl(snd_kctl_ioctl_func_t fcn)
{
return _snd_ctl_register_ioctl(fcn, &snd_control_ioctls);
}
EXPORT_SYMBOL(snd_ctl_register_ioctl);
#ifdef CONFIG_COMPAT
/**
* snd_ctl_register_ioctl_compat - register the device-specific 32bit compat
* control-ioctls
* @fcn: ioctl callback function
*
* Return: zero if successful, or a negative error code
*/
int snd_ctl_register_ioctl_compat(snd_kctl_ioctl_func_t fcn)
{
return _snd_ctl_register_ioctl(fcn, &snd_control_compat_ioctls);
}
EXPORT_SYMBOL(snd_ctl_register_ioctl_compat);
#endif
/*
* de-register the device-specific control-ioctls.
*/
static int _snd_ctl_unregister_ioctl(snd_kctl_ioctl_func_t fcn,
struct list_head *lists)
{
struct snd_kctl_ioctl *p;
if (snd_BUG_ON(!fcn))
return -EINVAL;
down_write(&snd_ioctl_rwsem);
list_for_each_entry(p, lists, list) {
if (p->fioctl == fcn) {
list_del(&p->list);
up_write(&snd_ioctl_rwsem);
kfree(p);
return 0;
}
}
up_write(&snd_ioctl_rwsem);
snd_BUG();
return -EINVAL;
}
/**
* snd_ctl_unregister_ioctl - de-register the device-specific control-ioctls
* @fcn: ioctl callback function to unregister
*
* Return: zero if successful, or a negative error code
*/
int snd_ctl_unregister_ioctl(snd_kctl_ioctl_func_t fcn)
{
return _snd_ctl_unregister_ioctl(fcn, &snd_control_ioctls);
}
EXPORT_SYMBOL(snd_ctl_unregister_ioctl);
#ifdef CONFIG_COMPAT
/**
* snd_ctl_unregister_ioctl_compat - de-register the device-specific compat
* 32bit control-ioctls
* @fcn: ioctl callback function to unregister
*
* Return: zero if successful, or a negative error code
*/
int snd_ctl_unregister_ioctl_compat(snd_kctl_ioctl_func_t fcn)
{
return _snd_ctl_unregister_ioctl(fcn, &snd_control_compat_ioctls);
}
EXPORT_SYMBOL(snd_ctl_unregister_ioctl_compat);
#endif
static int snd_ctl_fasync(int fd, struct file * file, int on)
{
struct snd_ctl_file *ctl;
ctl = file->private_data;
return snd_fasync_helper(fd, file, on, &ctl->fasync);
}
/* return the preferred subdevice number if already assigned;
* otherwise return -1
*/
int snd_ctl_get_preferred_subdevice(struct snd_card *card, int type)
{
struct snd_ctl_file *kctl;
int subdevice = -1;
unsigned long flags;
read_lock_irqsave(&card->ctl_files_rwlock, flags);
list_for_each_entry(kctl, &card->ctl_files, list) {
if (kctl->pid == task_pid(current)) {
subdevice = kctl->preferred_subdevice[type];
if (subdevice != -1)
break;
}
}
read_unlock_irqrestore(&card->ctl_files_rwlock, flags);
return subdevice;
}
EXPORT_SYMBOL_GPL(snd_ctl_get_preferred_subdevice);
/*
* ioctl32 compat
*/
#ifdef CONFIG_COMPAT
#include "control_compat.c"
#else
#define snd_ctl_ioctl_compat NULL
#endif
/*
* control layers (audio LED etc.)
*/
/**
* snd_ctl_request_layer - request to use the layer
* @module_name: Name of the kernel module (NULL == build-in)
*
* Return: zero if successful, or an error code when the module cannot be loaded
*/
int snd_ctl_request_layer(const char *module_name)
{
struct snd_ctl_layer_ops *lops;
if (module_name == NULL)
return 0;
down_read(&snd_ctl_layer_rwsem);
for (lops = snd_ctl_layer; lops; lops = lops->next)
if (strcmp(lops->module_name, module_name) == 0)
break;
up_read(&snd_ctl_layer_rwsem);
if (lops)
return 0;
return request_module(module_name);
}
EXPORT_SYMBOL_GPL(snd_ctl_request_layer);
/**
* snd_ctl_register_layer - register new control layer
* @lops: operation structure
*
* The new layer can track all control elements and do additional
* operations on top (like audio LED handling).
*/
void snd_ctl_register_layer(struct snd_ctl_layer_ops *lops)
{
struct snd_card *card;
int card_number;
down_write(&snd_ctl_layer_rwsem);
lops->next = snd_ctl_layer;
snd_ctl_layer = lops;
up_write(&snd_ctl_layer_rwsem);
for (card_number = 0; card_number < SNDRV_CARDS; card_number++) {
card = snd_card_ref(card_number);
if (card) {
down_read(&card->controls_rwsem);
lops->lregister(card);
up_read(&card->controls_rwsem);
snd_card_unref(card);
}
}
}
EXPORT_SYMBOL_GPL(snd_ctl_register_layer);
/**
* snd_ctl_disconnect_layer - disconnect control layer
* @lops: operation structure
*
* It is expected that the information about tracked cards
* is freed before this call (the disconnect callback is
* not called here).
*/
void snd_ctl_disconnect_layer(struct snd_ctl_layer_ops *lops)
{
struct snd_ctl_layer_ops *lops2, *prev_lops2;
down_write(&snd_ctl_layer_rwsem);
for (lops2 = snd_ctl_layer, prev_lops2 = NULL; lops2; lops2 = lops2->next) {
if (lops2 == lops) {
if (!prev_lops2)
snd_ctl_layer = lops->next;
else
prev_lops2->next = lops->next;
break;
}
prev_lops2 = lops2;
}
up_write(&snd_ctl_layer_rwsem);
}
EXPORT_SYMBOL_GPL(snd_ctl_disconnect_layer);
/*
* INIT PART
*/
static const struct file_operations snd_ctl_f_ops =
{
.owner = THIS_MODULE,
.read = snd_ctl_read,
.open = snd_ctl_open,
.release = snd_ctl_release,
.llseek = no_llseek,
.poll = snd_ctl_poll,
.unlocked_ioctl = snd_ctl_ioctl,
.compat_ioctl = snd_ctl_ioctl_compat,
.fasync = snd_ctl_fasync,
};
/*
* registration of the control device
*/
static int snd_ctl_dev_register(struct snd_device *device)
{
struct snd_card *card = device->device_data;
struct snd_ctl_layer_ops *lops;
int err;
err = snd_register_device(SNDRV_DEVICE_TYPE_CONTROL, card, -1,
&snd_ctl_f_ops, card, card->ctl_dev);
if (err < 0)
return err;
down_read(&card->controls_rwsem);
down_read(&snd_ctl_layer_rwsem);
for (lops = snd_ctl_layer; lops; lops = lops->next)
lops->lregister(card);
up_read(&snd_ctl_layer_rwsem);
up_read(&card->controls_rwsem);
return 0;
}
/*
* disconnection of the control device
*/
static int snd_ctl_dev_disconnect(struct snd_device *device)
{
struct snd_card *card = device->device_data;
struct snd_ctl_file *ctl;
struct snd_ctl_layer_ops *lops;
unsigned long flags;
read_lock_irqsave(&card->ctl_files_rwlock, flags);
list_for_each_entry(ctl, &card->ctl_files, list) {
wake_up(&ctl->change_sleep);
snd_kill_fasync(ctl->fasync, SIGIO, POLL_ERR);
}
read_unlock_irqrestore(&card->ctl_files_rwlock, flags);
down_read(&card->controls_rwsem);
down_read(&snd_ctl_layer_rwsem);
for (lops = snd_ctl_layer; lops; lops = lops->next)
lops->ldisconnect(card);
up_read(&snd_ctl_layer_rwsem);
up_read(&card->controls_rwsem);
return snd_unregister_device(card->ctl_dev);
}
/*
* free all controls
*/
static int snd_ctl_dev_free(struct snd_device *device)
{
struct snd_card *card = device->device_data;
struct snd_kcontrol *control;
down_write(&card->controls_rwsem);
while (!list_empty(&card->controls)) {
control = snd_kcontrol(card->controls.next);
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
__snd_ctl_remove(card, control, false);
}
ALSA: control: Use xarray for faster lookups The control elements are managed in a single linked list and we traverse the whole list for matching each numid or ctl id per every inquiry of a control element. This is OK-ish for a small number of elements but obviously it doesn't scale. Especially the matching with the ctl id takes time because it checks each field of the snd_ctl_id element, e.g. the name string is matched with strcmp(). This patch adds the hash tables with Xarray for improving the lookup speed of a control element. There are two xarray tables added to the card; one for numid and another for ctl id. For the numid, we use the numid as the index, while for the ctl id, we calculate a hash key. The lookup is done via a single xa_load() execution. As long as the given control element is found on the Xarray table, that's fine, we can give back a quick lookup result. The problem is when no entry hits on the table, and for this case, we have a slight optimization. Namely, the driver checks whether we had a collision on Xarray table, and do a fallback search (linear lookup of the full entries) only if a hash key collision happened beforehand. So, in theory, the inquiry for a non-existing element might take still time even with this patch in a worst case, but this must be pretty rare. The feature is enabled via CONFIG_SND_CTL_FAST_LOOKUP, which is turned on as default. For simplicity, the option can be turned off only when CONFIG_EXPERT is set ("You are expert? Then you manage 1000 knobs"). Link: https://lore.kernel.org/r/20211028130027.18764-1-tiwai@suse.de Link: https://lore.kernel.org/r/20220609180504.775-1-tiwai@suse.de Link: https://lore.kernel.org/all/cover.1653813866.git.quic_rbankapu@quicinc.com/ Link: https://lore.kernel.org/r/20220610064537.18660-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de>
2022-06-10 06:45:37 +00:00
#ifdef CONFIG_SND_CTL_FAST_LOOKUP
xa_destroy(&card->ctl_numids);
xa_destroy(&card->ctl_hash);
#endif
up_write(&card->controls_rwsem);
put_device(card->ctl_dev);
return 0;
}
/*
* create control core:
* called from init.c
*/
int snd_ctl_create(struct snd_card *card)
{
static const struct snd_device_ops ops = {
.dev_free = snd_ctl_dev_free,
.dev_register = snd_ctl_dev_register,
.dev_disconnect = snd_ctl_dev_disconnect,
};
int err;
if (snd_BUG_ON(!card))
return -ENXIO;
if (snd_BUG_ON(card->number < 0 || card->number >= SNDRV_CARDS))
return -ENXIO;
err = snd_device_alloc(&card->ctl_dev, card);
if (err < 0)
return err;
dev_set_name(card->ctl_dev, "controlC%d", card->number);
err = snd_device_new(card, SNDRV_DEV_CONTROL, card, &ops);
if (err < 0)
put_device(card->ctl_dev);
return err;
}
/*
* Frequently used control callbacks/helpers
*/
/**
* snd_ctl_boolean_mono_info - Helper function for a standard boolean info
* callback with a mono channel
* @kcontrol: the kcontrol instance
* @uinfo: info to store
*
* This is a function that can be used as info callback for a standard
* boolean control with a single mono channel.
*
* Return: Zero (always successful)
*/
int snd_ctl_boolean_mono_info(struct snd_kcontrol *kcontrol,
struct snd_ctl_elem_info *uinfo)
{
uinfo->type = SNDRV_CTL_ELEM_TYPE_BOOLEAN;
uinfo->count = 1;
uinfo->value.integer.min = 0;
uinfo->value.integer.max = 1;
return 0;
}
EXPORT_SYMBOL(snd_ctl_boolean_mono_info);
/**
* snd_ctl_boolean_stereo_info - Helper function for a standard boolean info
* callback with stereo two channels
* @kcontrol: the kcontrol instance
* @uinfo: info to store
*
* This is a function that can be used as info callback for a standard
* boolean control with stereo two channels.
*
* Return: Zero (always successful)
*/
int snd_ctl_boolean_stereo_info(struct snd_kcontrol *kcontrol,
struct snd_ctl_elem_info *uinfo)
{
uinfo->type = SNDRV_CTL_ELEM_TYPE_BOOLEAN;
uinfo->count = 2;
uinfo->value.integer.min = 0;
uinfo->value.integer.max = 1;
return 0;
}
EXPORT_SYMBOL(snd_ctl_boolean_stereo_info);
/**
* snd_ctl_enum_info - fills the info structure for an enumerated control
* @info: the structure to be filled
* @channels: the number of the control's channels; often one
* @items: the number of control values; also the size of @names
* @names: an array containing the names of all control values
*
* Sets all required fields in @info to their appropriate values.
* If the control's accessibility is not the default (readable and writable),
* the caller has to fill @info->access.
*
* Return: Zero (always successful)
*/
int snd_ctl_enum_info(struct snd_ctl_elem_info *info, unsigned int channels,
unsigned int items, const char *const names[])
{
info->type = SNDRV_CTL_ELEM_TYPE_ENUMERATED;
info->count = channels;
info->value.enumerated.items = items;
if (!items)
return 0;
if (info->value.enumerated.item >= items)
info->value.enumerated.item = items - 1;
WARN(strlen(names[info->value.enumerated.item]) >= sizeof(info->value.enumerated.name),
"ALSA: too long item name '%s'\n",
names[info->value.enumerated.item]);
strscpy(info->value.enumerated.name,
names[info->value.enumerated.item],
sizeof(info->value.enumerated.name));
return 0;
}
EXPORT_SYMBOL(snd_ctl_enum_info);