MMC: Fix race condition in resume/card detect code
PATCH 1/1 MMC: Fix race condition in resume/card detect code
When device wakes up by card change interrupt and MMC_UNSAFE_RESUME is enabled then race condition between mmc_rescan() and mmc_resume()/mmc_sd_resume() appeared.
Resume functions can sleep into mmc_remove_card() and at this time mmc_rescan() can be called by delayed work handler. Double-free of kobject or double-remove of host->card can be result of this.
This patch adds an mutex which deny simultaneous executing of mmc_sd_resume()/mmc_resume() and mmc_rescan() functions. Probably, it is not right way.
--- drivers/mmc/core/core.c | 7 +++++++ drivers/mmc/core/host.c | 3 +++ drivers/mmc/core/mmc.c | 3 +++ drivers/mmc/core/sd.c | 3 +++ include/linux/mmc/host.h | 3 +++ 5 files changed, 19 insertions(+), 0 deletions(-)
drivers/mmc/core/core.c +7 -0
@@ -657,6 +657,9 @@ void mmc_rescan(struct work_struct *work)u32 ocr;int err; +#ifdef CONFIG_MMC_UNSAFE_RESUME +mutex_lock (&host->carddetect_lock); +#endif mmc_bus_get (host);if (host->bus_ops ==NULL ) {
@@ -717,6 +720,10 @@ void mmc_rescan(struct work_struct *work) out:if (host->caps &MMC_CAP_NEEDS_POLL )mmc_schedule_delayed_work (&host->detect,HZ ); + +#ifdef CONFIG_MMC_UNSAFE_RESUME +mutex_unlock (&host->carddetect_lock); +#endif }void mmc_start_host (struct mmc_host *host)
drivers/mmc/core/host.c +3 -0
@@ -82,6 +82,9 @@ struct mmc_host *mmc_alloc_host(int extra, struct device *dev)device_initialize (&host->class_dev);spin_lock_init (&host->lock); +#ifdef CONFIG_MMC_UNSAFE_RESUME +mutex_init (&host->carddetect_lock); +#endif init_waitqueue_head (&host->wq);INIT_DELAYED_WORK (&host->detect, mmc_rescan);
drivers/mmc/core/mmc.c +3 -0
@@ -523,6 +523,8 @@ static void mmc_resume(struct mmc_host *host) {int err; +mutex_lock (&host->carddetect_lock); +BUG_ON (!host);BUG_ON (!host->card);
@@ -538,6 +540,7 @@ static void mmc_resume(struct mmc_host *host) mmc_release_host(host); } +mutex_unlock (&host->carddetect_lock); }#else
drivers/mmc/core/sd.c +3 -0
@@ -585,6 +585,8 @@ static void mmc_sd_resume(struct mmc_host *host) {int err; +mutex_lock (&host->carddetect_lock); +BUG_ON (!host);BUG_ON (!host->card);
@@ -600,6 +602,7 @@ static void mmc_sd_resume(struct mmc_host *host) mmc_release_host(host); } +mutex_unlock (&host->carddetect_lock); }#else
include/linux/mmc/host.h +3 -0
@@ -129,6 +129,9 @@ struct mmc_host {/* private data */ spinlock_t lock;/* lock for claim and bus ops */ +#ifdef CONFIG_MMC_UNSAFE_RESUME +struct mutex carddetect_lock; +#endif struct mmc_ios ios;/* current io bus settings */ u32 ocr;/* the current OCR setting */
Signature · 1 line
1.5.6.5
Quoted 8 lines · On Thu, 16 Oct 2008 19:09:36 +0300 Yauhen Kharuzhy <[email protected]> wrote:
When device wakes up by card change interrupt and
MMC_UNSAFE_RESUMEis enabled then race condition betweenmmc_rescan()and mmc_resume()/mmc_sd_resume() appeared.Resume functions can sleep into
mmc_remove_card()and at this timemmc_rescan()can be called by delayed work handler. Double-free of kobject or double-remove ofhost->cardcan be result of this.
Quoted 6 lines
This patch adds an mutex which deny simultaneous executing of mmc_sd_resume()/mmc_resume() and
mmc_rescan()functions. Probably, it is not right way.Signed-off-by Yauhen Kharuzhy <[email protected]>---
Can't we just ask the PM layer if this device is currently resuming, and if so ignore card notifications from the driver?
Signature · 9 lines
-- Pierre Ossman Linux kernel, MMC maintainer http://www.kernel.org rdesktop, core developer http://www.rdesktop.org WARNING: This correspondence is being monitored by the Swedish government. Make sure your server uses encryption for SMTP traffic and consider using PGP for end-to-end encryption.
Quoted 15 lines · Pierre Ossman
Quoted 12 lines · On Thu, 16 Oct 2008 19:09:36 +0300 Yauhen Kharuzhy <[email protected]> wrote:
When device wakes up by card change interrupt and
MMC_UNSAFE_RESUMEis enabled then race condition betweenmmc_rescan()and mmc_resume()/mmc_sd_resume() appeared.Resume functions can sleep into
mmc_remove_card()and at this timemmc_rescan()can be called by delayed work handler. Double-free of kobject or double-remove ofhost->cardcan be result of this.This patch adds an mutex which deny simultaneous executing of mmc_sd_resume()/mmc_resume() and
mmc_rescan()functions. Probably, it is not right way.
Quoted 5 lines
Quoted 2 lines
Signed-off-by Yauhen Kharuzhy <[email protected]>---
Can't we just ask the PM layer if this device is currently resuming, and if so ignore card notifications from the driver?
What about another idea: mmc_sd_resume() checks if host->detect is
scheduled and if true then it don't try to reinitialize card.host->detect can be scheduled at this moment only in one case: if device has been waked up by card change interrupt.
Signature · 5 lines
Yauhen Kharuzhy jekhor _at_ gmail.com JID: [email protected] A: No Q: Should I quote below my post?
2008/10/19 Yauhen Kharuzhy <[email protected]>:
Quoted 19 lines · Pierre Ossman
Quoted 18 lines
Quoted 14 lines · On Thu, 16 Oct 2008 19:09:36 +0300 Yauhen Kharuzhy <[email protected]> wrote:
When device wakes up by card change interrupt and
MMC_UNSAFE_RESUMEis enabled then race condition betweenmmc_rescan()and mmc_resume()/mmc_sd_resume() appeared.Resume functions can sleep into
mmc_remove_card()and at this timemmc_rescan()can be called by delayed work handler. Double-free of kobject or double-remove ofhost->cardcan be result of this.This patch adds an mutex which deny simultaneous executing of mmc_sd_resume()/mmc_resume() and
mmc_rescan()functions. Probably, it is not right way.Signed-off-by Yauhen Kharuzhy <[email protected]>---
Quoted 7 lines
Quoted 2 lines
Can't we just ask the PM layer if this device is currently resuming, and if so ignore card notifications from the driver?
What about another idea:
mmc_sd_resume()checks ifhost->detectis
scheduled and if true then it don't try to reinitialize card.host->detectcan be scheduled at this moment only in one case: if device has been waked up by card change interrupt.
Hmm... But it will cause a race condition if SD card change interrupt will be raised during resume process.
Signature · 5 lines
Best regards,
Yauhen Kharuzhy jekhor_(at)_gmail.com
A: No
Q: Should I quote below my post?PATCH 1/1 MMC: Fix race condition in resume/card detect code
In reply to Pierre OssmanWhen device wakes up by card change interrupt and MMC_UNSAFE_RESUME is enabled then race condition between mmc_rescan() and mmc_resume()/mmc_sd_resume() appeared.
Resume functions can sleep into mmc_remove_card() and at this time mmc_rescan() can be called by delayed work handler. Double-free of kobject or double-remove of host->card can be result of this.
This patch adds an host->suspended flag which indicated that host is in suspend state. mmc_rescan() checks it and returned when host->suspended == 1. It's safe because resume code calls mmc_detect_change() at end of resume process.
--- drivers/mmc/core/core.c | 7 +++++++ include/linux/mmc/host.h | 3 +++ 2 files changed, 10 insertions(+), 0 deletions(-)
drivers/mmc/core/core.c +7 -0
@@ -657,6 +657,9 @@ void mmc_rescan(struct work_struct *work)u32 ocr;int err; +if (host->suspended) +return ; +mmc_bus_get (host);if (host->bus_ops ==NULL ) {
@@ -780,6 +783,8 @@ int mmc_suspend_host(struct mmc_host *host, pm_message_t state)mmc_power_off (host); + host->suspended =1 ; +return 0 ; }
@@ -805,6 +810,8 @@ int mmc_resume_host(struct mmc_host *host)*/ mmc_detect_change (host,1 ); + host->suspended =0 ; +return 0 ; }
include/linux/mmc/host.h +3 -0
@@ -139,6 +139,9 @@ struct mmc_host {#ifdef CONFIG_MMC_DEBUG unsigned int removed:1; /* host is being removed */#endif +#ifdef CONFIG_MMC_UNSAFE_RESUME + unsigned int suspended:1; +#endif struct mmc_card *card;/* device attached to this host */
Signature · 1 line
1.5.6.5
Quoted 56 lines · Yauhen Kharuzhy
When device wakes up by card change interrupt and
MMC_UNSAFE_RESUMEis enabled then race condition betweenmmc_rescan()and mmc_resume()/mmc_sd_resume() appeared.Resume functions can sleep into
mmc_remove_card()and at this timemmc_rescan()can be called by delayed work handler. Double-free of kobject or double-remove ofhost->cardcan be result of this.This patch adds an
host->suspendedflag which indicated that host is in suspend state.mmc_rescan()checks it and returned whenhost->suspended== 1. It's safe because resume code callsmmc_detect_change()at end of resume process.Signed-off-by Yauhen Kharuzhy <[email protected]>--- drivers/mmc/core/core.c | 7 +++++++ include/linux/mmc/host.h | 3 +++ 2 files changed, 10 insertions(+), 0 deletions(-)drivers/mmc/core/core.c +7
-0@@ -657,6 +657,9 @@ void mmc_rescan(struct work_struct *work)u32 ocr;int err; +if (host->suspended) +return ; +mmc_bus_get (host);if (host->bus_ops ==NULL ) {@@ -780,6 +783,8 @@ int mmc_suspend_host(struct mmc_host *host, pm_message_t state)mmc_power_off (host); + host->suspended =1 ; +return 0 ; }@@ -805,6 +810,8 @@ int mmc_resume_host(struct mmc_host *host)*/ mmc_detect_change (host,1 ); + host->suspended =0 ; +return 0 ; }include/linux/mmc/host.h +0
-0
include/linux/mmc/host.h struct mmc_host { · line 147
Quoted 9 lines
include/linux/mmc/host.h +3
-0@@ -139,6 +139,9 @@ struct mmc_host {#ifdef CONFIG_MMC_DEBUG unsigned int removed:1; /* host is being removed */#endif +#ifdef CONFIG_MMC_UNSAFE_RESUME + unsigned int suspended:1; +#endif struct mmc_card *card;/* device attached to this host */
Shouldn't you also bracket any acces to ->suspended with the ifdefs ? Compilation failed on my box...and I had to enable MMC_UNSAFE_RESUME.
Signature · 4 lines
J.A. Magallon <jamagallon()ono!com> \ Software is like sex:
\ It's better when it's free
Mandriva Linux release 2009.0 (Cooker) for i586
Linux 2.6.25-jam18 (gcc 4.3.1 20080626 (GCC) #1 SMPQuoted 4 lines · On Mon, 20 Oct 2008 22:41:48 +0300 Yauhen Kharuzhy <[email protected]> wrote:
When device wakes up by card change interrupt and
MMC_UNSAFE_RESUMEis enabled then race condition betweenmmc_rescan()and mmc_resume()/mmc_sd_resume() appeared.
Having thought a bit more about this, I'm not sure where the race is. mmc_sd_resume() will be called before mmc_detect_change() is.
There is a race if the drivers call mmc_detect_change() before mmc_resume_host() has returned, but that is a driver bug. I won't object to adding a safe guard against that, but the commit message should reflect that scenario and not something else. There should also be some printk() to indicate that the driver is up to no good.
Quoted 5 lines
drivers/mmc/core/core.c +0
-0
drivers/mmc/core/core.c void mmc_rescan(struct work_struct *work) · line 665
Quoted 9 lines
drivers/mmc/core/core.c +3
-0@@ -657,6 +657,9 @@ void mmc_rescan(struct work_struct *work)u32 ocr;int err; +if (host->suspended) +return ; +mmc_bus_get (host);if (host->bus_ops ==NULL ) {
Was there no way to query the PM layer for this information?
drivers/mmc/core/core.c int mmc_resume_host(struct mmc_host *host) · line 817
Quoted 9 lines
drivers/mmc/core/core.c +2
-0@@ -805,6 +810,8 @@ int mmc_resume_host(struct mmc_host *host)*/ mmc_detect_change (host,1 ); + host->suspended =0 ; +return 0 ; }
You've added a new race here. ;)
You should set suspended to 0 before calling mmc_detect_change(), not after. The point was to protect bus_ops->resume(), nothing else.
Quoted 5 lines
include/linux/mmc/host.h +0
-0
include/linux/mmc/host.h struct mmc_host { · line 147
Quoted 9 lines
include/linux/mmc/host.h +3
-0@@ -139,6 +139,9 @@ struct mmc_host {#ifdef CONFIG_MMC_DEBUG unsigned int removed:1; /* host is being removed */#endif +#ifdef CONFIG_MMC_UNSAFE_RESUME + unsigned int suspended:1; +#endif struct mmc_card *card;/* device attached to this host */
No #ifdef for this as there are no where this variable is referenced.
Rgds
Signature · 9 lines
-- Pierre Ossman Linux kernel, MMC maintainer http://www.kernel.org rdesktop, core developer http://www.rdesktop.org WARNING: This correspondence is being monitored by the Swedish government. Make sure your server uses encryption for SMTP traffic and consider using PGP for end-to-end encryption.