fuse: add kernel-enforced request timeout option
Subsystems (3)
v9 cover fuse: add kernel-enforced request timeout option
There are situations where fuse servers can become unresponsive or stuck, for example if the server is in a deadlock. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This patchset adds a timeout option where if the server does not reply to a request by the time the timeout elapses, the connection will be aborted. This patchset also adds two dynamically configurable fuse sysctls "default_request_timeout" and "max_request_timeout" for controlling/enforcing timeout behavior system-wide.
Existing systems running fuse servers will not be affected unless they explicitly opt into the timeout.
v8:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v8 -> v9:
* Fix comment for u16 fs_parse_result, ULONG_MAX instead of U32_MAX, fix spacing (Bernd)
v7:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v7 -> v8:
* Use existing lists for checking expirations (Miklos)
v6:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v6 -> v7:
- Make timer per-connection instead of per-request (Miklos)
- Make default granularity of time minutes instead of seconds
- Removed the reviewed-bys since the interface of this has changed (now minutes, instead of seconds)
v5:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v5 -> v6:
- Gate sysctl.o behind CONFIG_SYSCTL in makefile (kernel test robot)
- Reword/clarify last sentence in cover letter (Miklos)
v4:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v4 -> v5:
- Change timeout behavior from aborting request to aborting connection (Miklos)
- Clarify wording for sysctl documentation (Jingbo)
v3:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v3 -> v4:
- Fix wording on some comments to make it more clear
- Use simpler logic for timer (eg remove extra if checks, use mod timer API) (Josef)
- Sanity-check should be on FR_FINISHING not FR_FINISHED (Jingbo)
- Fix comment for "processing queue", add req->fpq = NULL safeguard (Bernd)
v2:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v2 -> v3:
- Disarm / rearm timer in dev_do_read to handle race conditions (Bernrd)
- Disarm timer in error handling for fatal interrupt (Yafang)
- Clean up do_fuse_request_end (Jingbo)
- Add timer for notify retrieve requests
- Fix kernel test robot errors for #define no-op functions
v1:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v1 -> v2:
- Add timeout for background requests
- Handle resend race condition
- Add sysctls
Joanne Koong (3):
fs_parser: add fsparam_u16 helper
fuse: add optional kernel-enforced timeout for requests
fuse: add default_request_timeout and max_request_timeout sysctls
Documentation/admin-guide/sysctl/fs.rst | 27 +++++++++ fs/fs_parser.c | 14 +++++ fs/fuse/dev.c | 80 +++++++++++++++++++++++++ fs/fuse/fuse_i.h | 31 ++++++++++ fs/fuse/inode.c | 33 ++++++++++ fs/fuse/sysctl.c | 20 +++++++ include/linux/fs_parser.h | 9 ++- 7 files changed, 211 insertions(+), 3 deletions(-)
Signature · 1 line
2.43.5
PATCH RESEND v9 1/3 fs_parser: add fsparam_u16 helper
Add a fsparam helper for unsigned 16 bit values.
--- fs/fs_parser.c | 14 ++++++++++++++ include/linux/fs_parser.h | 9 ++++++--- 2 files changed, 20 insertions(+), 3 deletions(-)
fs/fs_parser.c +14 -0
@@ -210,6 +210,20 @@ int fs_param_is_bool(struct p_log *log, const struct fs_parameter_spec *p, }EXPORT_SYMBOL (fs_param_is_bool); +int fs_param_is_u16 (struct p_log *log,const struct fs_parameter_spec *p, +struct fs_parameter *param,struct fs_parse_result *result) +{ +int base = (unsigned long )p->data; +if (param->type != fs_value_is_string) +return fs_param_bad_value (log, param); +if (!*param->string && (p->flags & fs_param_can_be_empty)) +return 0 ; +if (kstrtou16 (param->string, base, &result->uint_16) <0 ) +return fs_param_bad_value (log, param); +return 0 ; +} +EXPORT_SYMBOL (fs_param_is_u16); +int fs_param_is_u32 (struct p_log *log,const struct fs_parameter_spec *p,struct fs_parameter *param,struct fs_parse_result *result) {
include/linux/fs_parser.h +6 -3
@@ -26,9 +26,10 @@ typedef int fs_param_type(struct p_log *,/* * The type of parameter expected. */ -fs_param_type fs_param_is_bool, fs_param_is_u32, fs_param_is_s32, fs_param_is_u64, - fs_param_is_enum, fs_param_is_string, fs_param_is_blob, fs_param_is_blockdev, - fs_param_is_path, fs_param_is_fd, fs_param_is_uid, fs_param_is_gid; +fs_param_type fs_param_is_bool, fs_param_is_u16, fs_param_is_u32, fs_param_is_s32, + fs_param_is_u64, fs_param_is_enum, fs_param_is_string, fs_param_is_blob, + fs_param_is_blockdev, fs_param_is_path, fs_param_is_fd, fs_param_is_uid, + fs_param_is_gid;/* * Specification of the type of value a parameter wants.
@@ -55,6 +56,7 @@ struct fs_parse_result {union {bool boolean;/* For spec_bool */ int int_32;/* For spec_s32/spec_enum */ +u16 uint_16;/* For spec_u16{,_octal,_hex}/spec_enum */ unsigned int uint_32;/* For spec_u32{,_octal,_hex}/spec_enum */ u64 uint_64;/* For spec_u64 */ kuid_t uid;
@@ -119,6 +121,7 @@ static inline bool fs_validate_description(const char *name,#define fsparam_flag_no (NAME ,OPT ) \ __fsparam(NULL, NAME, OPT, fs_param_neg_with_no, NULL)#define fsparam_bool (NAME ,OPT ) __fsparam(fs_param_is_bool, NAME, OPT, 0, NULL) +#define fsparam_u16 (NAME ,OPT ) __fsparam(fs_param_is_u16, NAME, OPT, 0, NULL)#define fsparam_u32 (NAME ,OPT ) __fsparam(fs_param_is_u32, NAME, OPT, 0, NULL)#define fsparam_u32oct (NAME ,OPT ) \ __fsparam(fs_param_is_u32, NAME, OPT, 0, (void *)8)
Signature · 1 line
2.43.5
PATCH RESEND v9 2/3 fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongThere are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Please note that these timeouts are not 100% precise. The request may take an extra FUSE_TIMEOUT_TIMER_FREQ seconds beyond the requested max timeout due to how it's internally implemented.
--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)
fs/fuse/dev.c +80 -0
@@ -45,6 +45,82 @@ static struct fuse_dev *fuse_get_dev(struct file *file)return READ_ONCE (file->private_data); } +static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +} + +/* +* Check if any requests aren't being completed by the specified request +* timeout. To do so, we: +* - check the fiq pending list +* - check the bg queue +* - check the fpq io and processing lists +* +* To make this fast, we only check against the head request on each list since +* these are generally queued in order of creation time (eg newer requests get +* queued to the tail). We might miss a few edge cases (eg requests transitioning +* between lists, re-sent requests at the head of the pending list having a +* later creation time than other requests on that list, etc.) but that is fine +* since if the request never gets fulfilled, it will eventually be caught. +*/ +void fuse_check_timeout (struct timer_list *timer) +{ + struct fuse_conn *fc = container_of(timer, struct fuse_conn, timeout.timer); +struct fuse_iqueue *fiq = &fc->iq; +struct fuse_req *req; +struct fuse_dev *fud; +struct fuse_pqueue *fpq; +bool expired = false; +int i; + +spin_lock (&fiq->lock); + req = list_first_entry_or_null(&fiq->pending, struct fuse_req, list); +if (req) + expired =request_expired (fc, req); +spin_unlock (&fiq->lock); +if (expired) + goto abort_conn; + +spin_lock (&fc->bg_lock); + req = list_first_entry_or_null(&fc->bg_queue, struct fuse_req, list); +if (req) + expired =request_expired (fc, req); +spin_unlock (&fc->bg_lock); +if (expired) + goto abort_conn; + +spin_lock (&fc->lock); +if (!fc->connected) { +spin_unlock (&fc->lock); +return ; + } +list_for_each_entry (fud, &fc->devices, entry) { + fpq = &fud->pq; +spin_lock (&fpq->lock); + req = list_first_entry_or_null(&fpq->io, struct fuse_req, list); +if (req &&request_expired (fc, req)) + goto fpq_abort; + +for (i =0 ; i <FUSE_PQ_HASH_SIZE ; i++) { + req = list_first_entry_or_null(&fpq->processing[i], struct fuse_req, list); +if (req &&request_expired (fc, req)) + goto fpq_abort; + } +spin_unlock (&fpq->lock); + } +spin_unlock (&fc->lock); + +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); +return ; + +fpq_abort: +spin_unlock (&fpq->lock); +spin_unlock (&fc->lock); +abort_conn: +fuse_abort_conn (fc); +} +static void fuse_request_init (struct fuse_mount *fm,struct fuse_req *req) {INIT_LIST_HEAD (&req->list);
@@ -53,6 +129,7 @@ static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)refcount_set (&req->count,1 );__set_bit (FR_PENDING , &req->flags); req->fm = fm; + req->create_time = jiffies; }static struct fuse_req *fuse_request_alloc (struct fuse_mount *fm,gfp_t flags)
@@ -2308,6 +2385,9 @@ void fuse_abort_conn(struct fuse_conn *fc)spin_unlock (&fc->lock);end_requests (&to_end); + +if (fc->timeout.req_timeout) +timer_delete (&fc->timeout.timer); }else {spin_unlock (&fc->lock); }
fs/fuse/fuse_i.h +21 -0
@@ -438,6 +438,9 @@ struct fuse_req {/** fuse_mount this request belongs to */ struct fuse_mount *fm; + +/** When (in jiffies) the request was created */ +unsigned long create_time; };struct fuse_iqueue ;
@@ -528,6 +531,16 @@ struct fuse_pqueue {struct list_head io; }; +/* Frequency (in seconds) of request timeout checks, if opted into */ +#define FUSE_TIMEOUT_TIMER_FREQ 60 * HZ + +struct fuse_timeout { +struct timer_list timer; + +/* Request timeout (in jiffies). 0 = no timeout */ +unsigned long req_timeout; +}; +/** * Fuse device instance */
@@ -574,6 +587,8 @@ struct fuse_fs_context {enum fuse_dax_mode dax_mode;unsigned int max_read;unsigned int blksize; +/* Request timeout (in minutes). 0 = no timeout (infinite wait) */ +unsigned int req_timeout;const char *subtype;/* DAX device, may be NULL */
@@ -920,6 +935,9 @@ struct fuse_conn {/** IDR for backing files ids */ struct idr backing_files_map;#endif + +/** Only used if the connection enforces request timeouts */ +struct fuse_timeout timeout; };/*
@@ -1181,6 +1199,9 @@ void fuse_request_end(struct fuse_req *req);void fuse_abort_conn (struct fuse_conn *fc);void fuse_wait_aborted (struct fuse_conn *fc); +/* Check if any requests timed out */ +void fuse_check_timeout (struct timer_list *timer); +/** * Invalidate inode attributes */
fs/fuse/inode.c +21 -0
@@ -735,6 +735,7 @@ enum {OPT_ALLOW_OTHER ,OPT_MAX_READ ,OPT_BLKSIZE , +OPT_REQUEST_TIMEOUT , OPT_ERR };
@@ -749,6 +750,7 @@ static const struct fs_parameter_spec fuse_fs_parameters[] = {fsparam_u32 ("max_read" ,OPT_MAX_READ ),fsparam_u32 ("blksize" ,OPT_BLKSIZE ),fsparam_string ("subtype" ,OPT_SUBTYPE ), + fsparam_u16 ("request_timeout", OPT_REQUEST_TIMEOUT), {} };
@@ -844,6 +846,10 @@ static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) ctx->blksize = result.uint_32;break ; +case OPT_REQUEST_TIMEOUT : + ctx->req_timeout = result.uint_16; +break ; +default :return -EINVAL ; }
@@ -973,6 +979,8 @@ void fuse_conn_put(struct fuse_conn *fc)if (IS_ENABLED (CONFIG_FUSE_DAX ))fuse_dax_conn_free (fc); +if (fc->timeout.req_timeout) +timer_shutdown_sync (&fc->timeout.timer);if (fiq->ops->release) fiq->ops->release (fiq);put_pid_ns (fc->pid_ns);
@@ -1691,6 +1699,18 @@ int fuse_init_fs_context_submount(struct fs_context *fsc) }EXPORT_SYMBOL_GPL (fuse_init_fs_context_submount); +static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{ +if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); + }else { + fc->timeout.req_timeout =0 ; + } +} +int fuse_fill_super_common (struct super_block *sb,struct fuse_fs_context *ctx) {struct fuse_dev *fud =NULL ;
@@ -1753,6 +1773,7 @@ int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) fc->destroy = ctx->destroy; fc->no_control = ctx->no_control; fc->no_force_umount = ctx->no_force_umount; +fuse_init_fc_timeout (fc, ctx); err = -ENOMEM ; root =fuse_get_root_inode (sb, ctx->rootmode);
Signature · 1 line
2.43.5
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Hi Joanne,
Quoted 7 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Does it make sense to configure timeout in seconds? hung-task watchdog operates in seconds and can be set to anything, e.g. 45 seconds, so it panic the system before fuse timeout has a chance to trigger.
Another question is: this will terminate the connection. Does it make sense to run timeout per request and just "abort" individual requests? What I'm currently playing with here on our side is something like this:
----
fs/fuse/dev.c +18 -4
@@ -21,6 +21,7 @@#include <linux/swap.h> #include <linux/splice.h> #include <linux/sched.h> +#include <linux/sched/sysctl.h> MODULE_ALIAS_MISCDEV (FUSE_MINOR );MODULE_ALIAS ("devname:fuse" );
@@ -368,11 +369,24 @@ static void request_wait_answer(struct fuse_req *req)int err;if (!fc->no_interrupt) { -/* Any signal may interrupt this */ - err =wait_event_interruptible (req->waitq, +/* We can use CONFIG_DEFAULT_HUNG_TASK_TIMEOUT here */ +unsigned long hang_check = sysctl_hung_task_timeout_secs; + +if (hang_check) { +/* Any signal or timeout may interrupt this */ + err =wait_event_interruptible_timeout (req->waitq, +test_bit (FR_FINISHED , &req->flags), + hang_check * (HZ /2 )); +if (err >0 ) +return ; + }else { +/* Any signal may interrupt this */ + err =wait_event_interruptible (req->waitq,test_bit (FR_FINISHED , &req->flags)); -if (!err) -return ; + +if (!err) +return ; + }set_bit (FR_INTERRUPTED , &req->flags);/* matches barrier in fuse_dev_do_read() */
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 11 lines · Sergey Senozhatsky
Hi Joanne,
Quoted 7 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Quoted 8 lines
Does it make sense to configure timeout in seconds? hung-task watchdog operates in seconds and can be set to anything, e.g. 45 seconds, so it panic the system before fuse timeout has a chance to trigger.
Another question is: this will terminate the connection. Does it make sense to run timeout per request and just "abort" individual requests? What I'm currently playing with here on our side is something like this:
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/raw
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 12 lines · Bernd Schubert
Quoted 11 lines · Sergey Senozhatsky
Hi Joanne,
Quoted 7 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Quoted 8 lines
Quoted 8 lines
Does it make sense to configure timeout in seconds? hung-task watchdog operates in seconds and can be set to anything, e.g. 45 seconds, so it panic the system before fuse timeout has a chance to trigger.
Another question is: this will terminate the connection. Does it make sense to run timeout per request and just "abort" individual requests? What I'm currently playing with here on our side is something like this:
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/raw
OK, sounds reasonable. I'll try to give the series some testing in the coming days.
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 26 lines · Sergey Senozhatsky
[You don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
Quoted 12 lines · Bernd Schubert
Quoted 11 lines · Sergey Senozhatsky
Hi Joanne,
Quoted 7 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Quoted 8 lines
Quoted 8 lines
Does it make sense to configure timeout in seconds? hung-task watchdog operates in seconds and can be set to anything, e.g. 45 seconds, so it panic the system before fuse timeout has a chance to trigger.
Another question is: this will terminate the connection. Does it make sense to run timeout per request and just "abort" individual requests? What I'm currently playing with here on our side is something like this:
Thanks for the pointers again, Bernd.
Quoted 8 lines
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/raw
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/raw
That's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our DEFAULT_HUNG_TASK_TIMEOUT is larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:
arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
In those cases 1 minute fuse timeout will overshot HUNG_TASK_TIMEOUT and then the question is whether HUNG_TASK_PANIC is set.
On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Hi everyone,
Quoted 47 lines · Sergey Senozhatsky
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/rawThat's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our
DEFAULT_HUNG_TASK_TIMEOUTis larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
Quoted 8 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
First of all, thanks everyone for looking into this.
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.
(In fact, we could optimize it even further by opportunistically scheduling a timer slightly later and opportunistically handling timed out requests when other requests are being completed, but this would be optimizing for the slow path, so probably an overkill.)
As for the length of the request timeout vs the hung task watchdog timeout, my opinion is that we should make sure that the hung task watchdog doesn't hit in any case, simply because a misbehaving userspace process must not be able to panic the kernel. In the blk-core, the blk_io_schedule() function [1] uses sysctl_hung_task_timeout_secs to determine the maximum length of a single uninterruptible sleep. I suppose we could use the same calculation to obtain our timeout number. What does everyone think?
[1] https://elixir.bootlin.com/linux/v6.12.1/source/block/blk-core.c#L1232
Best regards,
Tomasz
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 60 lines · Tomasz Figa
Hi everyone,
Quoted 47 lines · Sergey Senozhatsky
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/rawThat's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our
DEFAULT_HUNG_TASK_TIMEOUTis larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
Quoted 8 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
Quoted 8 lines
First of all, thanks everyone for looking into this.
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.
Requests are in FIFO order on the list and only head is checked. There are 256 hash lists per fuse device for requests currently in user space, though.
Quoted 15 lines
(In fact, we could optimize it even further by opportunistically scheduling a timer slightly later and opportunistically handling timed out requests when other requests are being completed, but this would be optimizing for the slow path, so probably an overkill.)
As for the length of the request timeout vs the hung task watchdog timeout, my opinion is that we should make sure that the hung task watchdog doesn't hit in any case, simply because a misbehaving userspace process must not be able to panic the kernel. In the blk-core, the
blk_io_schedule()function [1] usessysctl_hung_task_timeout_secsto determine the maximum length of a single uninterruptible sleep. I suppose we could use the same calculation to obtain our timeout number. What does everyone think?
I think that is a good idea.
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 56 lines · Bernd Schubert
Quoted 54 lines · Tomasz Figa
Hi everyone,
Quoted 50 lines · Sergey Senozhatsky
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/rawThat's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our
DEFAULT_HUNG_TASK_TIMEOUTis larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
First of all, thanks everyone for looking into this.
Hi Sergey and Tomasz,
Sorry for the late reply - I was out the last couple of days. Thanks Bernd for weighing in and answering the questions!
Quoted 7 lines
Quoted 7 lines
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.
I think the issue with this is that we likely would end up wasting more cpu cycles. For a busy FUSE server, there could be hundreds (thousands?) of requests that happen within the span of FUSE_TIMEOUT_TIMER_FREQ seconds.
While working on the patch, one thing I considered was disarming the timer in the timeout handler fuse_check_timeout() if no requests are on the list, in order to accomodate for "quiet periods" (eg if the FUSE server is inactive for a few minutes or hours) but ultimately decided against it because of the overhead it'd incur per request (eg check if the timer is disarmed, would most likely need to grab the fc->lock as well since timer rearming would need to be synchronized between background and non-background requests, etc.).
All in all, imo I don't think having the timer trigger every 60 seconds (what FUSE_TIMEOUT_TIMER_FREQ is set to) is too costly.
Quoted 20 lines
Requests are in FIFO order on the list and only head is checked. There are 256 hash lists per fuse device for requests currently in user space, though.
Quoted 15 lines
(In fact, we could optimize it even further by opportunistically scheduling a timer slightly later and opportunistically handling timed out requests when other requests are being completed, but this would be optimizing for the slow path, so probably an overkill.)
As for the length of the request timeout vs the hung task watchdog timeout, my opinion is that we should make sure that the hung task watchdog doesn't hit in any case, simply because a misbehaving userspace process must not be able to panic the kernel. In the blk-core, the
blk_io_schedule()function [1] usessysctl_hung_task_timeout_secsto determine the maximum length of a single uninterruptible sleep. I suppose we could use the same calculation to obtain our timeout number. What does everyone think?
Quoted 3 lines
I think that is a good idea.
Btw, just something to note, the fuse request timeout has an upper margin of error associated with it.
Copying over from the commit message -
"Please note that these timeouts are not 100% precise. The request may take an extra FUSE_TIMEOUT_TIMER_FREQ seconds beyond the requested max timeout due to how it's internally implemented."
For example, if a server sets the request timeout config to 10 minutes, the server could be aborted after 11 minutes (FUSE_TIMEOUT_TIMER_FREQ is set to 60 seconds internally) instead of 10 minutes.
Thanks,
Joanne
Quoted 4 lines
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 24 lines · Joanne Koong
Quoted 10 lines
Quoted 10 lines
Quoted 8 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
First of all, thanks everyone for looking into this.
Hi Sergey and Tomasz,
Sorry for the late reply - I was out the last couple of days. Thanks Bernd for weighing in and answering the questions!
Quoted 7 lines
Quoted 7 lines
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.
Quoted 4 lines
I think the issue with this is that we likely would end up wasting more cpu cycles. For a busy FUSE server, there could be hundreds (thousands?) of requests that happen within the span of
FUSE_TIMEOUT_TIMER_FREQseconds.
So, a silly question - can we not do that maybe?
What I'm thinking about is what if instead of implementing fuse-watchdog and tracking jiffies per request we'd switch to timeout aware operations and use what's already in the kernel? E.g. instead of wait_event() we'd use wait_event_timeout() and would configure timeout per connection (also bringing in current hung-task-watchdog timeout value into the equation), using MAX_SCHEDULE_TIMEOUT as a default (similarly to what core kernel does). The first req that timeouts kills its siblings and the connection.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 33 lines · Sergey Senozhatsky
Quoted 24 lines · Joanne Koong
Quoted 10 lines
Quoted 10 lines
Quoted 8 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
First of all, thanks everyone for looking into this.
Hi Sergey and Tomasz,
Sorry for the late reply - I was out the last couple of days. Thanks Bernd for weighing in and answering the questions!
Quoted 7 lines
Quoted 7 lines
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.Quoted 4 lines
I think the issue with this is that we likely would end up wasting more cpu cycles. For a busy FUSE server, there could be hundreds (thousands?) of requests that happen within the span of
FUSE_TIMEOUT_TIMER_FREQseconds.So, a silly question - can we not do that maybe?
Quoted 8 lines
What I'm thinking about is what if instead of implementing fuse-watchdog and tracking jiffies per request we'd switch to timeout aware operations and use what's already in the kernel? E.g. instead of
wait_event()we'd usewait_event_timeout()and would configure timeout per connection (also bringing in current hung-task-watchdog timeout value into the equation), usingMAX_SCHEDULE_TIMEOUTas a default (similarly to what core kernel does). The first req that timeouts kills its siblings and the connection.
Using timeout aware operations like wait_event_timeout() associates a timer per request (see schedule_timeout()) and this approach was tried in v6 [1] but the overhead of having a timer per request showed about a 1.5% drop in throughput [1], which is why we ended up pivoting to a periodic watchdog timer that triggers at set intervals.
Thanks,
Joanne
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongQuoted 126 lines · Joanne Koong
Quoted 56 lines · Bernd Schubert
Quoted 54 lines · Tomasz Figa
Hi everyone,
Quoted 50 lines · Sergey Senozhatsky
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/rawThat's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our
DEFAULT_HUNG_TASK_TIMEOUTis larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
First of all, thanks everyone for looking into this.
Hi Sergey and Tomasz,
Sorry for the late reply - I was out the last couple of days. Thanks Bernd for weighing in and answering the questions!
Quoted 7 lines
Quoted 7 lines
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.I think the issue with this is that we likely would end up wasting more cpu cycles. For a busy FUSE server, there could be hundreds (thousands?) of requests that happen within the span of
FUSE_TIMEOUT_TIMER_FREQseconds.While working on the patch, one thing I considered was disarming the timer in the timeout handler
fuse_check_timeout()if no requests are on the list, in order to accomodate for "quiet periods" (eg if the FUSE server is inactive for a few minutes or hours) but ultimately decided against it because of the overhead it'd incur per request (eg check if the timer is disarmed, would most likely need to grab thefc->lockas well since timer rearming would need to be synchronized between background and non-background requests, etc.).All in all, imo I don't think having the timer trigger every 60 seconds (what
FUSE_TIMEOUT_TIMER_FREQis set to) is too costly.Quoted 20 lines
Requests are in FIFO order on the list and only head is checked. There are 256 hash lists per fuse device for requests currently in user space, though.
Quoted 15 lines
(In fact, we could optimize it even further by opportunistically scheduling a timer slightly later and opportunistically handling timed out requests when other requests are being completed, but this would be optimizing for the slow path, so probably an overkill.)
As for the length of the request timeout vs the hung task watchdog timeout, my opinion is that we should make sure that the hung task watchdog doesn't hit in any case, simply because a misbehaving userspace process must not be able to panic the kernel. In the blk-core, the
blk_io_schedule()function [1] usessysctl_hung_task_timeout_secsto determine the maximum length of a single uninterruptible sleep. I suppose we could use the same calculation to obtain our timeout number. What does everyone think?Quoted 3 lines
I think that is a good idea.
Btw, just something to note, the fuse request timeout has an upper margin of error associated with it.
Copying over from the commit message -
Quoted 9 lines
"Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented."For example, if a server sets the request timeout config to 10 minutes, the server could be aborted after 11 minutes (
FUSE_TIMEOUT_TIMER_FREQis set to 60 seconds internally) instead of 10 minutes.
Let me add +Brian Geffon who also was thinking about the right timeout value.
Quoted 7 lines
Thanks,
JoanneQuoted 4 lines
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 58 lines · Tomasz Figa
Quoted 56 lines · Joanne Koong
Quoted 53 lines · Bernd Schubert
Quoted 51 lines · Tomasz Figa
Hi everyone,
Quoted 47 lines · Sergey Senozhatsky
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/rawThat's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our
DEFAULT_HUNG_TASK_TIMEOUTis larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.
In my opinion this is a good argument for having the hung task timeout and a fuse timeout independent. The hung task timeout is for hung kernel threads, in this situation we're potentially taking too long in userspace but that doesn't necessarily mean the system is hung. I think a loop which does an interruptible wait with a timeout of 1/2 the hung task timeout would make sense to ensure the hung task timeout doesn't hit. There might be situations where we want a fuse timeout which is larger than the hung task timeout, perhaps a file system being read over a satellite internet connection?
Quoted 87 lines
Quoted 68 lines
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
On the other hand, setups that set much lower timeout than DEFAULT_HUNG_TASK_TIMEOUT=120 will have extra CPU activities regardless, just because watchdogs will run more often.
Tomasz, any opinions?
First of all, thanks everyone for looking into this.
Hi Sergey and Tomasz,
Sorry for the late reply - I was out the last couple of days. Thanks Bernd for weighing in and answering the questions!
Quoted 7 lines
Quoted 7 lines
How about keeping a list of requests in the FIFO order (in other
words: first entry is the first to timeout) and whenever the first entry is being removed from the list (aka the request actually completes), re-arming the timer to the timeout of the next request in the list? This way we don't really have any timer firing unless there is really a request that timed out.I think the issue with this is that we likely would end up wasting more cpu cycles. For a busy FUSE server, there could be hundreds (thousands?) of requests that happen within the span of
FUSE_TIMEOUT_TIMER_FREQseconds.While working on the patch, one thing I considered was disarming the timer in the timeout handler
fuse_check_timeout()if no requests are on the list, in order to accomodate for "quiet periods" (eg if the FUSE server is inactive for a few minutes or hours) but ultimately decided against it because of the overhead it'd incur per request (eg check if the timer is disarmed, would most likely need to grab thefc->lockas well since timer rearming would need to be synchronized between background and non-background requests, etc.).All in all, imo I don't think having the timer trigger every 60 seconds (what
FUSE_TIMEOUT_TIMER_FREQis set to) is too costly.Quoted 20 lines
Requests are in FIFO order on the list and only head is checked. There are 256 hash lists per fuse device for requests currently in user space, though.
Quoted 15 lines
(In fact, we could optimize it even further by opportunistically scheduling a timer slightly later and opportunistically handling timed out requests when other requests are being completed, but this would be optimizing for the slow path, so probably an overkill.)
As for the length of the request timeout vs the hung task watchdog timeout, my opinion is that we should make sure that the hung task watchdog doesn't hit in any case, simply because a misbehaving userspace process must not be able to panic the kernel. In the blk-core, the
blk_io_schedule()function [1] usessysctl_hung_task_timeout_secsto determine the maximum length of a single uninterruptible sleep. I suppose we could use the same calculation to obtain our timeout number. What does everyone think?Quoted 3 lines
I think that is a good idea.
Btw, just something to note, the fuse request timeout has an upper margin of error associated with it.
Copying over from the commit message -
Quoted 9 lines
"Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented."For example, if a server sets the request timeout config to 10 minutes, the server could be aborted after 11 minutes (
FUSE_TIMEOUT_TIMER_FREQis set to 60 seconds internally) instead of 10 minutes.Let me add +Brian Geffon who also was thinking about the right timeout value.
Quoted 7 lines
Thanks,
JoanneQuoted 4 lines
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 62 lines · Brian Geffon
Quoted 58 lines · Tomasz Figa
Quoted 56 lines · Joanne Koong
Quoted 53 lines · Bernd Schubert
Quoted 51 lines · Tomasz Figa
Hi everyone,
Quoted 47 lines · Sergey Senozhatsky
Cc-ing Tomasz
Quoted 8 lines · Bernd Schubert
Quoted 8 lines
Thanks for the pointers again, Bernd.
Quoted 2 lines
Miklos had asked for to abort the connection in v4
https://lore.kernel.org/all/CAJfpegsiRNnJx7OAoH58XRq3zujrcXx94S2JACFdgJJ_b8FdHw@mail.gmail.com/rawOK, sounds reasonable. I'll try to give the series some testing in the coming days.
Quoted 8 lines
Quoted 2 lines
// I still would probably prefer "seconds" timeout granularity. // Unless this also has been discussed already and Bernd has a link ;)
The issue is that is currently iterating through 256 hash lists + pending + bg.
https://lore.kernel.org/all/CAJnrk1b7bfAWWq_pFP=4XH3ddc_9GtAM2mE7EgWnx2Od+UUUjQ@mail.gmail.com/raw
Oh, I see.
Quoted 2 lines
Personally I would prefer a second list to avoid the check spike and latency
https://lore.kernel.org/linux-fsdevel/[email protected]/rawThat's good to know. I like the idea of less CPU usage in general, our devices a battery powered so everything counts, to some extent.
Quoted 2 lines
What is your opinion about that? I guess android and chromium have an interest low latencies and avoiding cpu spikes?
Good question.
Can't speak for android, in chromeos we probably will keep it at 1 minute, but this is because our
DEFAULT_HUNG_TASK_TIMEOUTis larger than that (we use default value of 120 sec). There are setups that might use lower values, or even re-define default value, e.g.:arch/arc/configs/axs101_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/axs103_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/hsdk_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/arc/configs/vdk_hs38_smp_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=10 arch/powerpc/configs/mvme5100_defconfig:CONFIG_DEFAULT_HUNG_TASK_TIMEOUT=20
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.
Quoted 9 lines
In my opinion this is a good argument for having the hung task timeout and a fuse timeout independent. The hung task timeout is for hung kernel threads, in this situation we're potentially taking too long in userspace but that doesn't necessarily mean the system is hung. I think a loop which does an interruptible wait with a timeout of 1/2 the hung task timeout would make sense to ensure the hung task timeout doesn't hit. There might be situations where we want a fuse timeout which is larger than the hung task timeout, perhaps a file system being read over a satellite internet connection?
For a network file system the remote server also might just hang and one might want to wait much longer than 1/2 hung task timeout for recovery.
Thanks,
Bernd
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Brian GeffonQuoted 7 lines · Brian Geffon
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.In my opinion this is a good argument for having the hung task timeout and a fuse timeout independent. The hung task timeout is for hung kernel threads
Sorry, but no, it's not only for kernel threads.
Quoted 2 lines
in this situation we're potentially taking too long in userspace but that doesn't necessarily mean the system is hung.
And it's not for hung system. It's for tasks that stuck unreasonably long waiting for a particular event or resource. And those tasks very well can be user-space processes, being stuck in syscall waiting for something.
Quoted 3 lines
I think a loop which does an interruptible wait with a timeout of 1/2 the hung task timeout would make sense to ensure the hung task timeout doesn't hit.
The point here is not to silent watchdog, we might as well just disable it and call it a day. The point here is that fuse can be used (and IS in our particular case) as a network filesystem, and the problem can be way outside of your system, so spinning in a wait loop doesn't fix any problem on that remote system no matter how long we spin, and that's what watchdog signals us.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 24 lines · Sergey Senozhatsky
Quoted 7 lines · Brian Geffon
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
In those cases 1 minute fuse timeout will overshot
HUNG_TASK_TIMEOUTand then the question is whetherHUNG_TASK_PANICis set.In my opinion this is a good argument for having the hung task timeout and a fuse timeout independent. The hung task timeout is for hung kernel threads
Sorry, but no, it's not only for kernel threads.
Quoted 2 lines
in this situation we're potentially taking too long in userspace but that doesn't necessarily mean the system is hung.
And it's not for hung system. It's for tasks that stuck unreasonably long waiting for a particular event or resource. And those tasks very well can be user-space processes, being stuck in syscall waiting for something.
Quoted 3 lines
I think a loop which does an interruptible wait with a timeout of 1/2 the hung task timeout would make sense to ensure the hung task timeout doesn't hit.
Quoted 6 lines
The point here is not to silent watchdog, we might as well just disable it and call it a day. The point here is that fuse can be used (and IS in our particular case) as a network filesystem, and the problem can be way outside of your system, so spinning in a wait loop doesn't fix any problem on that remote system no matter how long we spin, and that's what watchdog signals us.
Sergey, to clarify again what I was suggesting, I was not suggesting that no timeout is appropriate. I suggested that the hung task timeout is not appropriate and instead we should be using a different timeout. My suggestion about looping was simply to say: **once we've transitioned to userspace to service the request the hung task timeout should not apply and instead a different fuse timeout should.**
The loop is nothing more than to allow the task to break out of the wait so it's not triggering the hung task timeout and frankly an implementation detail that I should have probably omitted. There are situations where a timeout less than the hung task timeout might be appropriate and there are situations where a timeout longer than the hung task timeout might be appropriate.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongOn (24/11/14 11:13), Joanne Koong wrote:
[..]
fs/fuse/inode.c int fuse_init_fs_context_submount(struct fs_context *fsc) · line 1708
Quoted 7 lines
fs/fuse/inode.c +7
-0+static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{ +if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ );
So I think this will require much bigger changes in the code. fuse_check_timeout() is executed from IRQ context and it takes the same locks that are acquired by preemptive task contexts. So all of those now need to disable local IRQs before they
acquire: fc->bg_lock, fig->lock, fc->lock. Otherwise we are in a deadlock scenario (in many places, unfortunately).
A simple example:
[ 91.466408] _raw_spin_lock+0x39/0x70
[ 91.466420] fuse_simple_background+0x902/0x1130 [fuse]
[ 91.466453] fuse_send_init+0x337/0x680 [fuse]
[ 91.466485] fuse_fill_super+0x1c8/0x200 [fuse]
[ 91.466516] get_tree_nodev+0xaf/0x140
[ 91.466527] fuse_get_tree+0x27e/0x450 [fuse]
[ 91.466559] vfs_get_tree+0x88/0x240
[ 91.466569] path_mount+0xf26/0x1ed0
[ 91.466579] __se_sys_mount+0x1c9/0x240
[ 91.466588] do_syscall_64+0x6f/0xa0
[ 91.466598] entry_SYSCALL_64_after_hwframe+0x73/0xdd
[ 91.466666]
other info that might help us debug this:
[ 91.466672] Possible unsafe locking scenario:[ 91.466679] CPU0
[ 91.466684] ----
[ 91.466689] lock(&fiq->lock);
[ 91.466701] <Interrupt>
[ 91.466707] lock(&fiq->lock);
[ 91.466718]
*** DEADLOCK ***[ 91.466724] 4 locks held by runtime_probe/5043:
[ 91.466732] #0: ffff888005812448 (sb_writers#3){.+.+}-{0:0}, at: filename_create+0xb2/0x320
[ 91.466762] #1: ffff8881499ea3f0 (&type->i_mutex_dir_key#5/1){+.+.}-{3:3}, at: filename_create+0x14c/0x320
[ 91.466791] #2: ffffffff9d864ce0 (rcu_read_lock){....}-{1:2}, at: security_sid_to_context_core+0xa4/0x4f0
[ 91.466817] #3: ffff88815c009ec0 ((&fc->timeout.timer)){+.-.}-{0:0}, at: run_timer_softirq+0x702/0x1700
[ 91.466841]
stack backtrace:
[ 91.466850] CPU: 2 PID: 5043 Comm: runtime_probe Tainted: G U 6.6.63-lockdep #1 f2b045305e587e03c4766ca818bb77742f953c87
[ 91.466864] Hardware name: HP Lantis/Lantis, BIOS Google_Lantis.13606.437.0 01/21/2022
[ 91.466872] Call Trace:
[ 91.466881] <IRQ>
[ 91.466889] dump_stack_lvl+0x6d/0xb0
[ 91.466904] print_usage_bug+0x8a4/0xbb0
[ 91.466917] mark_lock+0x13ca/0x1940
[ 91.466930] __lock_acquire+0xc10/0x2670
[ 91.466944] lock_acquire+0x119/0x3a0
[ 91.466955] ? fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5]
[ 91.466992] ? fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5]
[ 91.467025] _raw_spin_lock+0x39/0x70
[ 91.467036] ? fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5]
[ 91.467070] fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5]
[ 91.467104] ? run_timer_softirq+0x702/0x1700
[ 91.467114] ? run_timer_softirq+0x702/0x1700
[ 91.467123] ? __pfx_fuse_check_timeout+0x10/0x10 [fuse c290dfa1c114772028056af34b9662cba2d155b5]
[ 91.467156] run_timer_softirq+0x763/0x1700
[ 91.467172] irq_exit_rcu+0x300/0x7d0
[ 91.467183] ? sysvec_apic_timer_interrupt+0x56/0x90
[ 91.467195] sysvec_apic_timer_interrupt+0x56/0x90
[ 91.467205] </IRQ>Do we want to run fuse-watchdog as a preemptive task instead of IRQ context? Simlar to the way the hung-task watchdog runs, for example. Yes, it now may starve and not get scheduled in corner cases (under some crazy pressure), but at least we don't need to patch the entire fuse code to use irq_safe/irq_restore locking variants.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 73 lines · Sergey Senozhatsky
On (24/11/14 11:13), Joanne Koong wrote:
[..]fs/fuse/inode.c int fuse_init_fs_context_submount(struct fs_context *fsc) · line 1708
Quoted 7 lines
fs/fuse/inode.c +7
-0+static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{ +if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ );So I think this will require much bigger changes in the code.
fuse_check_timeout()is executed from IRQ context and it takes the same locks that are acquired by preemptive task contexts. So all of those now need to disable local IRQs before they
acquire:fc->bg_lock,fig->lock,fc->lock. Otherwise we are in a deadlock scenario (in many places, unfortunately).A simple example:
[ 91.466408] _raw_spin_lock+0x39/0x70 [ 91.466420] fuse_simple_background+0x902/0x1130 [fuse] [ 91.466453] fuse_send_init+0x337/0x680 [fuse] [ 91.466485] fuse_fill_super+0x1c8/0x200 [fuse] [ 91.466516] get_tree_nodev+0xaf/0x140 [ 91.466527] fuse_get_tree+0x27e/0x450 [fuse] [ 91.466559] vfs_get_tree+0x88/0x240 [ 91.466569] path_mount+0xf26/0x1ed0 [ 91.466579] __se_sys_mount+0x1c9/0x240 [ 91.466588] do_syscall_64+0x6f/0xa0 [ 91.466598] entry_SYSCALL_64_after_hwframe+0x73/0xdd [ 91.466666] other info that might help us debug this: [ 91.466672] Possible unsafe locking scenario:[ 91.466679] CPU0 [ 91.466684] ---- [ 91.466689] lock(&fiq->lock); [ 91.466701] <Interrupt> [ 91.466707] lock(&fiq->lock); [ 91.466718] *** DEADLOCK ***[ 91.466724] 4 locks held by runtime_probe/5043: [ 91.466732] #0: ffff888005812448 (sb_writers#3){.+.+}-{0:0}, at: filename_create+0xb2/0x320 [ 91.466762] #1: ffff8881499ea3f0 (&type->i_mutex_dir_key#5/1){+.+.}-{3:3}, at: filename_create+0x14c/0x320 [ 91.466791] #2: ffffffff9d864ce0 (rcu_read_lock){....}-{1:2}, at: security_sid_to_context_core+0xa4/0x4f0 [ 91.466817] #3: ffff88815c009ec0 ((&fc->timeout.timer)){+.-.}-{0:0}, at: run_timer_softirq+0x702/0x1700 [ 91.466841] stack backtrace: [ 91.466850] CPU: 2 PID: 5043 Comm: runtime_probe Tainted: G U 6.6.63-lockdep #1 f2b045305e587e03c4766ca818bb77742f953c87 [ 91.466864] Hardware name: HP Lantis/Lantis, BIOS Google_Lantis.13606.437.0 01/21/2022 [ 91.466872] Call Trace: [ 91.466881] <IRQ> [ 91.466889] dump_stack_lvl+0x6d/0xb0 [ 91.466904] print_usage_bug+0x8a4/0xbb0 [ 91.466917] mark_lock+0x13ca/0x1940 [ 91.466930] __lock_acquire+0xc10/0x2670 [ 91.466944] lock_acquire+0x119/0x3a0 [ 91.466955] ? fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5] [ 91.466992] ? fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5] [ 91.467025] _raw_spin_lock+0x39/0x70 [ 91.467036] ? fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5] [ 91.467070] fuse_check_timeout+0x32/0x630 [fuse c290dfa1c114772028056af34b9662cba2d155b5] [ 91.467104] ? run_timer_softirq+0x702/0x1700 [ 91.467114] ? run_timer_softirq+0x702/0x1700 [ 91.467123] ? __pfx_fuse_check_timeout+0x10/0x10 [fuse c290dfa1c114772028056af34b9662cba2d155b5] [ 91.467156] run_timer_softirq+0x763/0x1700 [ 91.467172] irq_exit_rcu+0x300/0x7d0 [ 91.467183] ? sysvec_apic_timer_interrupt+0x56/0x90 [ 91.467195] sysvec_apic_timer_interrupt+0x56/0x90 [ 91.467205] </IRQ>
Quoted 6 lines
Do we want to run fuse-watchdog as a preemptive task instead of IRQ context? Simlar to the way the hung-task watchdog runs, for example. Yes, it now may starve and not get scheduled in corner cases (under some crazy pressure), but at least we don't need to patch the entire fuse code to use irq_safe/irq_restore locking variants.
Interesting! Thanks for noting this.
It looks like the choices we have here then are to either:
* have this run in a kthread like hung-task watchdog, as you mentioned above
* have all fuse code use irq_safe/irq_restore for locks
* use a separate spinlock/list for request timeouts as per the implementation in v7 [1], though will need to benchmark this to make sure performance doesn't degrade from lock contention if lots of requests are submitted in parallel
Having 1 extra kthread per server as overhead doesn't seem too bad to me. There's already an upper margin of imprecision with the timer implementation, so I don't see the kthread scheduling imprecision as a big deal.
I'll restructure this to use a kthread in v10. If anyone has thoughts on a better or more preferred approach though, would love to hear them.
Thanks,
Joanne
[1] https://lore.kernel.org/linux-fsdevel/[email protected]/
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 5 lines · Joanne Koong
Interesting! Thanks for noting this.
It looks like the choices we have here then are to either:
* have this run in a kthread like hung-task watchdog, as you mentioned above
Works for me.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongOn (24/11/14 11:13), Joanne Koong wrote:
[..]
Quoted 8 lines
fs/fuse/fuse_i.h +3
-0@@ -920,6 +935,9 @@ struct fuse_conn {/** IDR for backing files ids */ struct idr backing_files_map;#endif + +/** Only used if the connection enforces request timeouts */ +struct fuse_timeout timeout; };
[..]
Quoted 9 lines
fs/fuse/inode.c +1
-0@@ -749,6 +750,7 @@ static const struct fs_parameter_spec fuse_fs_parameters[] = {fsparam_u32 ("max_read" ,OPT_MAX_READ ),fsparam_u32 ("blksize" ,OPT_BLKSIZE ),fsparam_string ("subtype" ,OPT_SUBTYPE ), + fsparam_u16 ("request_timeout", OPT_REQUEST_TIMEOUT), {} };
fs/fuse/inode.c static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) · line 852
Quoted 7 lines
fs/fuse/inode.c +4
-0@@ -844,6 +846,10 @@ static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) ctx->blksize = result.uint_32;break ; +case OPT_REQUEST_TIMEOUT : + ctx->req_timeout = result.uint_16; +break ; +
A quick question: so for this user-space should be updated to request fuse-watchdog on particular connection? Would it make sense to have a way to simply enforce watchdog on all connections?
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 24 lines · Sergey Senozhatsky
On (24/11/14 11:13), Joanne Koong wrote:
[..]Quoted 8 lines
fs/fuse/fuse_i.h +3
-0@@ -920,6 +935,9 @@ struct fuse_conn {/** IDR for backing files ids */ struct idr backing_files_map;#endif + +/** Only used if the connection enforces request timeouts */ +struct fuse_timeout timeout; };[..]
Quoted 12 lines
fs/fuse/inode.c +1
-0@@ -749,6 +750,7 @@ static const struct fs_parameter_spec fuse_fs_parameters[] = {fsparam_u32 ("max_read" ,OPT_MAX_READ ),fsparam_u32 ("blksize" ,OPT_BLKSIZE ),fsparam_string ("subtype" ,OPT_SUBTYPE ), + fsparam_u16 ("request_timeout", OPT_REQUEST_TIMEOUT), {} };@@ -844,6 +846,10 @@ static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) ctx->blksize = result.uint_32;break ;
Quoted 9 lines
fs/fuse/inode.c static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) · line 852
Quoted 4 lines
fs/fuse/inode.c +4
-0+case OPT_REQUEST_TIMEOUT : + ctx->req_timeout = result.uint_16; +break ; +A quick question: so for this user-space should be updated to request fuse-watchdog on particular connection? Would it make sense to have a way to simply enforce watchdog on all connections?
Hi Sergey,
The third patch
(https://lore.kernel.org/linux-fsdevel/[email protected]/)
in this patchset adds this through the sysctl interface. The sys admin can set /proc/sys/fs/fuse/max_request_timeout and this will ensure requests don't take longer than this value.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne Koongfs/fuse/dev.c static struct fuse_dev *fuse_get_dev(struct file *file) · line 51
Quoted 5 lines · Joanne Koong
fs/fuse/dev.c +4
-0+static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +}
With jiffies we need to use time_after() and such, so we'd deal with jiffies wrap-around.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 10 lines · Sergey Senozhatsky
fs/fuse/dev.c static struct fuse_dev *fuse_get_dev(struct file *file) · line 51
Quoted 5 lines · Joanne Koong
fs/fuse/dev.c +4
-0+static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +}With jiffies we need to use
time_after()and such, so we'd deal with jiffies wrap-around.
Ohh I see, I guess this is because on 32-bit systems unsigned longs are 4 bytes? Thanks for noting - I'll push out a new version with this change.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 13 lines · Joanne Koong
Quoted 10 lines · Sergey Senozhatsky
fs/fuse/dev.c static struct fuse_dev *fuse_get_dev(struct file *file) · line 51
Quoted 5 lines · Joanne Koong
fs/fuse/dev.c +4
-0+static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +}With jiffies we need to use
time_after()and such, so we'd deal with jiffies wrap-around.
Quoted 2 lines
Ohh I see, I guess this is because on 32-bit systems unsigned longs are 4 bytes?
Correct, IIRC on 32bit system jiffies wraparound every 47 or 49 days. But I also recall that jiffies (at least in the past) were initialized to -5 minutes so that they would wraparound during first 5 minutes of uptime to reveal bugs. I don't know if it's still the case though.
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongQuoted 8 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
I haven't been keeping up with the earlier series, but I think I agree with Sergey that this timeout would be better expressed in seconds.
Most filesystems that deal with timeouts (NFS, CIFS, etc.) specify them as a number of seconds, and expressing this in minutes goes against that convention. It also seems rather coarse-grained. I could easily see a situation where 5 minutes is too short, but 6 minutes is too long.
With that too, you probably wouldn't need patch #1. You could treat it as a 32-bit integer and just clamp the value as necessary.
Quoted 244 lines
Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented.Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)fs/fuse/dev.c +80
-0@@ -45,6 +45,82 @@ static struct fuse_dev *fuse_get_dev(struct file *file)return READ_ONCE (file->private_data); } +static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +} + +/* +* Check if any requests aren't being completed by the specified request +* timeout. To do so, we: +* - check the fiq pending list +* - check the bg queue +* - check the fpq io and processing lists +* +* To make this fast, we only check against the head request on each list since +* these are generally queued in order of creation time (eg newer requests get +* queued to the tail). We might miss a few edge cases (eg requests transitioning +* between lists, re-sent requests at the head of the pending list having a +* later creation time than other requests on that list, etc.) but that is fine +* since if the request never gets fulfilled, it will eventually be caught. +*/ +void fuse_check_timeout (struct timer_list *timer) +{ + struct fuse_conn *fc = container_of(timer, struct fuse_conn, timeout.timer); +struct fuse_iqueue *fiq = &fc->iq; +struct fuse_req *req; +struct fuse_dev *fud; +struct fuse_pqueue *fpq; +bool expired = false; +int i; + +spin_lock (&fiq->lock); + req = list_first_entry_or_null(&fiq->pending, struct fuse_req, list); +if (req) + expired =request_expired (fc, req); +spin_unlock (&fiq->lock); +if (expired) + goto abort_conn; + +spin_lock (&fc->bg_lock); + req = list_first_entry_or_null(&fc->bg_queue, struct fuse_req, list); +if (req) + expired =request_expired (fc, req); +spin_unlock (&fc->bg_lock); +if (expired) + goto abort_conn; + +spin_lock (&fc->lock); +if (!fc->connected) { +spin_unlock (&fc->lock); +return ; + } +list_for_each_entry (fud, &fc->devices, entry) { + fpq = &fud->pq; +spin_lock (&fpq->lock); + req = list_first_entry_or_null(&fpq->io, struct fuse_req, list); +if (req &&request_expired (fc, req)) + goto fpq_abort; + +for (i =0 ; i <FUSE_PQ_HASH_SIZE ; i++) { + req = list_first_entry_or_null(&fpq->processing[i], struct fuse_req, list); +if (req &&request_expired (fc, req)) + goto fpq_abort; + } +spin_unlock (&fpq->lock); + } +spin_unlock (&fc->lock); + +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); +return ; + +fpq_abort: +spin_unlock (&fpq->lock); +spin_unlock (&fc->lock); +abort_conn: +fuse_abort_conn (fc); +} +static void fuse_request_init (struct fuse_mount *fm,struct fuse_req *req) {INIT_LIST_HEAD (&req->list);@@ -53,6 +129,7 @@ static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)refcount_set (&req->count,1 );__set_bit (FR_PENDING , &req->flags); req->fm = fm; + req->create_time = jiffies; }static struct fuse_req *fuse_request_alloc (struct fuse_mount *fm,gfp_t flags)@@ -2308,6 +2385,9 @@ void fuse_abort_conn(struct fuse_conn *fc)spin_unlock (&fc->lock);end_requests (&to_end); + +if (fc->timeout.req_timeout) +timer_delete (&fc->timeout.timer); }else {spin_unlock (&fc->lock); }fs/fuse/fuse_i.h +21
-0@@ -438,6 +438,9 @@ struct fuse_req {/** fuse_mount this request belongs to */ struct fuse_mount *fm; + +/** When (in jiffies) the request was created */ +unsigned long create_time; };struct fuse_iqueue ;@@ -528,6 +531,16 @@ struct fuse_pqueue {struct list_head io; }; +/* Frequency (in seconds) of request timeout checks, if opted into */ +#define FUSE_TIMEOUT_TIMER_FREQ 60 * HZ + +struct fuse_timeout { +struct timer_list timer; + +/* Request timeout (in jiffies). 0 = no timeout */ +unsigned long req_timeout; +}; +/** * Fuse device instance */ @@ -574,6 +587,8 @@ struct fuse_fs_context {enum fuse_dax_mode dax_mode;unsigned int max_read;unsigned int blksize; +/* Request timeout (in minutes). 0 = no timeout (infinite wait) */ +unsigned int req_timeout;const char *subtype;/* DAX device, may be NULL */ @@ -920,6 +935,9 @@ struct fuse_conn {/** IDR for backing files ids */ struct idr backing_files_map;#endif + +/** Only used if the connection enforces request timeouts */ +struct fuse_timeout timeout; };/* @@ -1181,6 +1199,9 @@ void fuse_request_end(struct fuse_req *req);void fuse_abort_conn (struct fuse_conn *fc);void fuse_wait_aborted (struct fuse_conn *fc); +/* Check if any requests timed out */ +void fuse_check_timeout (struct timer_list *timer); +/** * Invalidate inode attributes */ fs/fuse/inode.c +21
-0@@ -735,6 +735,7 @@ enum {OPT_ALLOW_OTHER ,OPT_MAX_READ ,OPT_BLKSIZE , +OPT_REQUEST_TIMEOUT , OPT_ERR };@@ -749,6 +750,7 @@ static const struct fs_parameter_spec fuse_fs_parameters[] = {fsparam_u32 ("max_read" ,OPT_MAX_READ ),fsparam_u32 ("blksize" ,OPT_BLKSIZE ),fsparam_string ("subtype" ,OPT_SUBTYPE ), + fsparam_u16 ("request_timeout", OPT_REQUEST_TIMEOUT), {} };@@ -844,6 +846,10 @@ static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) ctx->blksize = result.uint_32;break ; +case OPT_REQUEST_TIMEOUT : + ctx->req_timeout = result.uint_16; +break ; +default :return -EINVAL ; }@@ -973,6 +979,8 @@ void fuse_conn_put(struct fuse_conn *fc)if (IS_ENABLED (CONFIG_FUSE_DAX ))fuse_dax_conn_free (fc); +if (fc->timeout.req_timeout) +timer_shutdown_sync (&fc->timeout.timer);if (fiq->ops->release) fiq->ops->release (fiq);put_pid_ns (fc->pid_ns);@@ -1691,6 +1699,18 @@ int fuse_init_fs_context_submount(struct fs_context *fsc) }EXPORT_SYMBOL_GPL (fuse_init_fs_context_submount); +static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{ +if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); + }else { + fc->timeout.req_timeout =0 ; + } +} +int fuse_fill_super_common (struct super_block *sb,struct fuse_fs_context *ctx) {struct fuse_dev *fud =NULL ;@@ -1753,6 +1773,7 @@ int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) fc->destroy = ctx->destroy; fc->no_control = ctx->no_control; fc->no_force_umount = ctx->no_force_umount; +fuse_init_fc_timeout (fc, ctx); err = -ENOMEM ; root =fuse_get_root_inode (sb, ctx->rootmode);
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 11 lines · Jeff Layton
Quoted 8 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Quoted 8 lines
I haven't been keeping up with the earlier series, but I think I agree with Sergey that this timeout would be better expressed in seconds.
Most filesystems that deal with timeouts (NFS, CIFS, etc.) specify them as a number of seconds, and expressing this in minutes goes against that convention. It also seems rather coarse-grained. I could easily see a situation where 5 minutes is too short, but 6 minutes is too long.
Sounds good, I'll change the timeout to seconds. The reason it was set in minutes is because the timeouts have an upper margin of error (right now, up to 1 minute) and I didn't want to give a misleading illusion of precision.
It sounds like it'd be useful if the timer is run more frequently (eg instead of every 1 minute, maybe running this every 15 or 30 secs) as well. I'll make this change in the next version.
Quoted 16 lines
With that too, you probably wouldn't need patch #1. You could treat it as a 32-bit integer and just clamp the value as necessary.
Quoted 12 lines
Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented.Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)
...
Quoted 3 lines
--
Jeff Layton <[email protected]>
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 21 lines · Joanne Koong
Quoted 11 lines · Jeff Layton
Quoted 8 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Quoted 8 lines
I haven't been keeping up with the earlier series, but I think I agree with Sergey that this timeout would be better expressed in seconds.
Most filesystems that deal with timeouts (NFS, CIFS, etc.) specify them as a number of seconds, and expressing this in minutes goes against that convention. It also seems rather coarse-grained. I could easily see a situation where 5 minutes is too short, but 6 minutes is too long.
Quoted 5 lines
Sounds good, I'll change the timeout to seconds. The reason it was set in minutes is because the timeouts have an upper margin of error (right now, up to 1 minute) and I didn't want to give a misleading illusion of precision.
Understood. That is a concern, but I think you can mitigate it by documenting that the timeout just makes it eligible to be reaped, rather than a hard timeout guarantee. That would also allow you to change the implementation in the future to reap more (or less) aggressively, without breaking the interface.
Quoted 4 lines
It sounds like it'd be useful if the timer is run more frequently (eg instead of every 1 minute, maybe running this every 15 or 30 secs) as well. I'll make this change in the next version.
Yeah, 1 minute seems like quite a long time. I don't know FUSE well enough to get a feel for how heavyweight fuse_check_timeout() is, but it might be good to check more often, especially if you don't expect it to take long in the common case.
Quoted 20 lines
Quoted 16 lines
With that too, you probably wouldn't need patch #1. You could treat it as a 32-bit integer and just clamp the value as necessary.
Quoted 12 lines
Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented.Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)...
Quoted 3 lines
--
Jeff Layton <[email protected]>
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongQuoted 8 lines · Etienne Martineau
From: Joanne Koong <[email protected]>
Quoted 3 lines
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
I tried this V9 patchset and realized that upon
max_request_timeoutthe connection will be dropped irrespectively if the task is in D state or not. I guess this is expected behavior.
Yes, the connection will be dropped regardless of what state the task is in.
Quoted 1 line
To me the concerning aspect is when tasks are going in D state because of the consequence when running with
hung_task_timeout_secsand hung_task_panic=1.
Could you elaborate on this a bit more? When running with hung_task_timeout_secs and hung_task_panic=1, how does it cause the task to go into D state?
Quoted 42 lines
Quoted 42 lines
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented.Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)fs/fuse/dev.c +16
-0@@ -45,6 +45,82 @@ static struct fuse_dev *fuse_get_dev(struct file *file)return READ_ONCE (file->private_data); } +static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +} + +/* +* Check if any requests aren't being completed by the specified request +* timeout. To do so, we: +* - check the fiq pending list +* - check the bg queue +* - check the fpq io and processing lists +* +* To make this fast, we only check against the head request on each list since +* these are generally queued in order of creation time (eg newer requests get +* queued to the tail). We might miss a few edge cases (eg requests transitioning +* between lists, re-sent requests at the head of the pending list having a
Quoted 8 lines
fs/fuse/dev.c static struct fuse_dev *fuse_get_dev(struct file *file) · line 69
Quoted 6 lines
fs/fuse/dev.c +6
-0+* later creation time than other requests on that list, etc.) but that is fine +* since if the request never gets fulfilled, it will eventually be caught. +*/ +void fuse_check_timeout (struct timer_list *timer) +{ + struct fuse_conn *fc = container_of(timer, struct fuse_conn, timeout.timer);Here this timer may get queued and if echo 1 > /sys/fs/fuse/connections/'nn'/abort is done at more or less the same time over the same connection I'm wondering what will happen? At least I think we may need
timer_delete_sync()instead oftimer_delete()infuse_abort_conn()and potentially call it from the top offuse_abort_conn()instead.
I don't think this is an issue because there's still a reference on the "struct fuse_conn" when fuse_abort_conn() is called. The fuse_conn struct is freed in fuse_conn_put() when the last refcount is dropped, and in fuse_conn_put() there's this line
if (fc->timeout.req_timeout)
timer_shutdown_sync(&fc->timeout.timer);
that guarantees the timer is not queued / callback of timer is not running / cannot be rearmed.
Quoted 10 lines
fs/fuse/dev.c static struct fuse_dev *fuse_get_dev(struct file *file) · line 77
Quoted 9 lines
fs/fuse/dev.c +8
-0+struct fuse_iqueue *fiq = &fc->iq; +struct fuse_req *req; +struct fuse_dev *fud; +struct fuse_pqueue *fpq; +bool expired = false; +int i; + +spin_lock (&fiq->lock);Do we need
spin_lock_irqinstead bcos this is irq context no?
Yeah, I missed that spin_lock doesn't disable interrupts. Sergey noted this as well in [1] and for the next iteration of this patchset, I'm planning to use use a kthread per server connection, similar to what the hung task watchdog does.
Quoted 24 lines
Quoted 18 lines
fs/fuse/dev.c +12
-0+static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{ +if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); + }else { + fc->timeout.req_timeout =0 ; + } +} +int fuse_fill_super_common (struct super_block *sb,struct fuse_fs_context *ctx) {struct fuse_dev *fud =NULL ;fs/fuse/dev.c int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) · line 1776
Quoted 4 lines
fs/fuse/dev.c +1
-0@@ -1753,6 +1773,7 @@ int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) fc->destroy = ctx->destroy; fc->no_control = ctx->no_control; fc->no_force_umount = ctx->no_force_umount; +fuse_init_fc_timeout (fc, ctx);The
max_request_timeoutis latched infc->timeout.req_timeoutduringfuse_fill_super_common()so any further modification are not taking into effect
Quoted 9 lines
Say:
sudo bash -c 'echo 100 >/proc/sys/fs/fuse/max_request_timeout' sshfs -o allow_other,default_permissions you@localhost:/home/you/test ./mnt kill -STOPpidof /usr/lib/openssh/sftp-server
ls ./mnt/
^C
# Trying to change back the timeout doesn't have any effect
sudo bash -c 'echo 10 >/proc/sys/fs/fuse/max_request_timeout'
This is expected behavior. The timeout that's set at mount time will be the timeout for the duration of the connection.
Thanks,
Joanne
Quoted 2 lines
Thanks
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 3 lines
Quoted 1 line
I tried this V9 patchset and realized that upon
max_request_timeoutthe connection will be dropped irrespectively if the task is in D state or not. I guess this is expected behavior.Yes, the connection will be dropped regardless of what state the task is in.
Thanks for confirmation
Quoted 6 lines
Quoted 2 lines
To me the concerning aspect is when tasks are going in D state because of the consequence when running with
hung_task_timeout_secsand hung_task_panic=1.Could you elaborate on this a bit more? When running with
hung_task_timeout_secsand hung_task_panic=1, how does it cause the task to go into D state?
Sorry for the confusion. It doesn't cause tasks to go in D state. What I meant is that I've been looking for a way to terminate tasks stuck in D state because we have hung_task_panic=1 and this is causing bad consequences when they trigger the hung task timer.
Quoted 8 lines
Quoted 2 lines
Here this timer may get queued and if echo 1 > /sys/fs/fuse/connections/'nn'/abort is done at more or less the same time over the same connection I'm wondering what will happen? At least I think we may need
timer_delete_sync()instead oftimer_delete()infuse_abort_conn()and potentially call it from the top offuse_abort_conn()instead.I don't think this is an issue because there's still a reference on the "struct
fuse_conn" whenfuse_abort_conn()is called. Thefuse_connstruct is freed infuse_conn_put()when the last refcount is dropped, and infuse_conn_put()there's this line
Quoted 5 lines
if (
fc->timeout.req_timeout)
timer_shutdown_sync(&fc->timeout.timer);that guarantees the timer is not queued / callback of timer is not running / cannot be rearmed.
Got it. I missed that last part in fuse_conn_put() Thanks
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 9 lines · Etienne
Quoted 4 lines
Quoted 2 lines
I tried this V9 patchset and realized that upon
max_request_timeoutthe connection will be dropped irrespectively if the task is in D state or not. I guess this is expected behavior.Yes, the connection will be dropped regardless of what state the task is in.
Thanks for confirmation
Quoted 3 lines
Quoted 2 lines
To me the concerning aspect is when tasks are going in D state because of the consequence when running with
hung_task_timeout_secsand hung_task_panic=1.
Quoted 8 lines
Quoted 3 lines
Could you elaborate on this a bit more? When running with
hung_task_timeout_secsand hung_task_panic=1, how does it cause the task to go into D state?Sorry for the confusion. It doesn't cause tasks to go in D state. What I meant is that I've been looking for a way to terminate tasks stuck in D state because we have hung_task_panic=1 and this is causing bad consequences when they trigger the hung task timer.
Gotcha, thanks for clarifying!
Quoted 17 lines
Quoted 9 lines
Quoted 3 lines
Here this timer may get queued and if echo 1 > /sys/fs/fuse/connections/'nn'/abort is done at more or less the same time over the same connection I'm wondering what will happen? At least I think we may need
timer_delete_sync()instead oftimer_delete()infuse_abort_conn()and potentially call it from the top offuse_abort_conn()instead.I don't think this is an issue because there's still a reference on the "struct
fuse_conn" whenfuse_abort_conn()is called. Thefuse_connstruct is freed infuse_conn_put()when the last refcount is dropped, and infuse_conn_put()there's this lineQuoted 5 lines
if (
fc->timeout.req_timeout)
timer_shutdown_sync(&fc->timeout.timer);that guarantees the timer is not queued / callback of timer is not running / cannot be rearmed.
Got it. I missed that last part in
fuse_conn_put()Thanks
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
In reply to Joanne KoongQuoted 231 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented.Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)fs/fuse/dev.c +80
-0@@ -45,6 +45,82 @@ static struct fuse_dev *fuse_get_dev(struct file *file)return READ_ONCE (file->private_data); } +static bool request_expired (struct fuse_conn *fc,struct fuse_req *req) +{ +return jiffies > req->create_time + fc->timeout.req_timeout; +} + +/* +* Check if any requests aren't being completed by the specified request +* timeout. To do so, we: +* - check the fiq pending list +* - check the bg queue +* - check the fpq io and processing lists +* +* To make this fast, we only check against the head request on each list since +* these are generally queued in order of creation time (eg newer requests get +* queued to the tail). We might miss a few edge cases (eg requests transitioning +* between lists, re-sent requests at the head of the pending list having a +* later creation time than other requests on that list, etc.) but that is fine +* since if the request never gets fulfilled, it will eventually be caught. +*/ +void fuse_check_timeout (struct timer_list *timer) +{ + struct fuse_conn *fc = container_of(timer, struct fuse_conn, timeout.timer); +struct fuse_iqueue *fiq = &fc->iq; +struct fuse_req *req; +struct fuse_dev *fud; +struct fuse_pqueue *fpq; +bool expired = false; +int i; + +spin_lock (&fiq->lock); + req = list_first_entry_or_null(&fiq->pending, struct fuse_req, list); +if (req) + expired =request_expired (fc, req); +spin_unlock (&fiq->lock); +if (expired) + goto abort_conn; + +spin_lock (&fc->bg_lock); + req = list_first_entry_or_null(&fc->bg_queue, struct fuse_req, list); +if (req) + expired =request_expired (fc, req); +spin_unlock (&fc->bg_lock); +if (expired) + goto abort_conn; + +spin_lock (&fc->lock); +if (!fc->connected) { +spin_unlock (&fc->lock); +return ; + } +list_for_each_entry (fud, &fc->devices, entry) { + fpq = &fud->pq; +spin_lock (&fpq->lock); + req = list_first_entry_or_null(&fpq->io, struct fuse_req, list); +if (req &&request_expired (fc, req)) + goto fpq_abort; + +for (i =0 ; i <FUSE_PQ_HASH_SIZE ; i++) { + req = list_first_entry_or_null(&fpq->processing[i], struct fuse_req, list); +if (req &&request_expired (fc, req)) + goto fpq_abort; + } +spin_unlock (&fpq->lock); + } +spin_unlock (&fc->lock); + +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); +return ; + +fpq_abort: +spin_unlock (&fpq->lock); +spin_unlock (&fc->lock); +abort_conn: +fuse_abort_conn (fc); +} +static void fuse_request_init (struct fuse_mount *fm,struct fuse_req *req) {INIT_LIST_HEAD (&req->list);@@ -53,6 +129,7 @@ static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)refcount_set (&req->count,1 );__set_bit (FR_PENDING , &req->flags); req->fm = fm; + req->create_time = jiffies; }static struct fuse_req *fuse_request_alloc (struct fuse_mount *fm,gfp_t flags)@@ -2308,6 +2385,9 @@ void fuse_abort_conn(struct fuse_conn *fc)spin_unlock (&fc->lock);end_requests (&to_end); + +if (fc->timeout.req_timeout) +timer_delete (&fc->timeout.timer); }else {spin_unlock (&fc->lock); }fs/fuse/fuse_i.h +21
-0@@ -438,6 +438,9 @@ struct fuse_req {/** fuse_mount this request belongs to */ struct fuse_mount *fm; + +/** When (in jiffies) the request was created */ +unsigned long create_time; };struct fuse_iqueue ;@@ -528,6 +531,16 @@ struct fuse_pqueue {struct list_head io; }; +/* Frequency (in seconds) of request timeout checks, if opted into */ +#define FUSE_TIMEOUT_TIMER_FREQ 60 * HZ + +struct fuse_timeout { +struct timer_list timer; + +/* Request timeout (in jiffies). 0 = no timeout */ +unsigned long req_timeout; +}; +/** * Fuse device instance */ @@ -574,6 +587,8 @@ struct fuse_fs_context {enum fuse_dax_mode dax_mode;unsigned int max_read;unsigned int blksize; +/* Request timeout (in minutes). 0 = no timeout (infinite wait) */ +unsigned int req_timeout;const char *subtype;/* DAX device, may be NULL */ @@ -920,6 +935,9 @@ struct fuse_conn {/** IDR for backing files ids */ struct idr backing_files_map;#endif + +/** Only used if the connection enforces request timeouts */ +struct fuse_timeout timeout; };/* @@ -1181,6 +1199,9 @@ void fuse_request_end(struct fuse_req *req);void fuse_abort_conn (struct fuse_conn *fc);void fuse_wait_aborted (struct fuse_conn *fc); +/* Check if any requests timed out */ +void fuse_check_timeout (struct timer_list *timer); +/** * Invalidate inode attributes */ fs/fuse/inode.c +10
-0@@ -735,6 +735,7 @@ enum {OPT_ALLOW_OTHER ,OPT_MAX_READ ,OPT_BLKSIZE , +OPT_REQUEST_TIMEOUT , OPT_ERR };@@ -749,6 +750,7 @@ static const struct fs_parameter_spec fuse_fs_parameters[] = {fsparam_u32 ("max_read" ,OPT_MAX_READ ),fsparam_u32 ("blksize" ,OPT_BLKSIZE ),fsparam_string ("subtype" ,OPT_SUBTYPE ), + fsparam_u16 ("request_timeout", OPT_REQUEST_TIMEOUT), {} };@@ -844,6 +846,10 @@ static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) ctx->blksize = result.uint_32;break ; +case OPT_REQUEST_TIMEOUT : + ctx->req_timeout = result.uint_16; +break ; +default :return -EINVAL ; }@@ -973,6 +979,8 @@ void fuse_conn_put(struct fuse_conn *fc)if (IS_ENABLED (CONFIG_FUSE_DAX ))fuse_dax_conn_free (fc); +if (fc->timeout.req_timeout) +timer_shutdown_sync (&fc->timeout.timer);if (fiq->ops->release) fiq->ops->release (fiq);put_pid_ns (fc->pid_ns);@@ -1691,6 +1699,18 @@ int fuse_init_fs_context_submount(struct fs_context *fsc) }EXPORT_SYMBOL_GPL (fuse_init_fs_context_submount); +static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{
fs/fuse/inode.c int fuse_init_fs_context_submount(struct fs_context *fsc) · line 1713
Quoted 10 lines
fs/fuse/inode.c +10
-0+if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); + }else { + fc->timeout.req_timeout =0 ; + } +} +
Does fuse_check_timeout need to run in IRQ context? It doesn't seem like it does. Have you considered setting up a recurring delayed workqueue job instead? That would run in process context, which might make the locking in that function less hairy.
Quoted 11 lines
fs/fuse/inode.c +1
-0int fuse_fill_super_common (struct super_block *sb,struct fuse_fs_context *ctx) {struct fuse_dev *fud =NULL ;@@ -1753,6 +1773,7 @@ int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) fc->destroy = ctx->destroy; fc->no_control = ctx->no_control; fc->no_force_umount = ctx->no_force_umount; +fuse_init_fc_timeout (fc, ctx); err = -ENOMEM ; root =fuse_get_root_inode (sb, ctx->rootmode);
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH RESEND v9 2/3] fuse: add optional kernel-enforced timeout for requests
Quoted 51 lines · Jeff Layton
Quoted 49 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is deadlocked. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This commit adds an option for enforcing a timeout (in minutes) for requests where if the timeout elapses without the server responding to the request, the connection will be automatically aborted.
Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the requested max timeout due to how it's internally implemented.Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- fs/fuse/dev.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++ fs/fuse/fuse_i.h | 21 +++++++++++++ fs/fuse/inode.c | 21 +++++++++++++ 3 files changed, 122 insertions(+)fs/fuse/dev.c +12
-0return -EINVAL ; }@@ -973,6 +979,8 @@ void fuse_conn_put(struct fuse_conn *fc)if (IS_ENABLED (CONFIG_FUSE_DAX ))fuse_dax_conn_free (fc); +if (fc->timeout.req_timeout) +timer_shutdown_sync (&fc->timeout.timer);if (fiq->ops->release) fiq->ops->release (fiq);put_pid_ns (fc->pid_ns);@@ -1691,6 +1699,18 @@ int fuse_init_fs_context_submount(struct fs_context *fsc) }EXPORT_SYMBOL_GPL (fuse_init_fs_context_submount); +static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) +{ +if (ctx->req_timeout) { +if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) + fc->timeout.req_timeout =ULONG_MAX ; +timer_setup (&fc->timeout.timer, fuse_check_timeout,0 ); +mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ ); + }else { + fc->timeout.req_timeout =0 ; + }
Quoted 9 lines
fs/fuse/dev.c int fuse_init_fs_context_submount(struct fs_context *fsc) · line 1713
Quoted 2 lines
fs/fuse/dev.c +2
-0+} +Does
fuse_check_timeoutneed to run in IRQ context? It doesn't seem like it does. Have you considered setting up a recurring delayed workqueue job instead? That would run in process context, which might make the locking in that function less hairy.
Great idea, I'll use a recurring delayed workqueue job instead of a kthread, since it's more lightweight.
Thanks,
Joanne
Quoted 15 lines
fs/fuse/dev.c int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) · line 1779
Quoted 12 lines
fs/fuse/dev.c +1
-0int fuse_fill_super_common (struct super_block *sb,struct fuse_fs_context *ctx) {struct fuse_dev *fud =NULL ;@@ -1753,6 +1773,7 @@ int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) fc->destroy = ctx->destroy; fc->no_control = ctx->no_control; fc->no_force_umount = ctx->no_force_umount; +fuse_init_fc_timeout (fc, ctx); err = -ENOMEM ; root =fuse_get_root_inode (sb, ctx->rootmode);--
Jeff Layton <[email protected]>
PATCH RESEND v9 3/3 fuse: add default_request_timeout and max_request_timeout sysctls
In reply to Joanne KoongIntroduce two new sysctls, "default_request_timeout" and "max_request_timeout". These control how long (in minutes) a server can take to reply to a request. If the server does not reply by the timeout, then the connection will be aborted.
"default_request_timeout" sets the default timeout if no timeout is specified by the fuse server on mount. 0 (default) indicates no default timeout should be enforced. If the server did specify a timeout, then default_request_timeout will be ignored.
"max_request_timeout" sets the max amount of time the server may take to reply to a request. 0 (default) indicates no maximum timeout. If max_request_timeout is set and the fuse server attempts to set a timeout greater than max_request_timeout, the system will use max_request_timeout as the timeout. Similarly, if default_request_timeout is greater than max_request_timeout, the system will use max_request_timeout as the timeout. If the server does not request a timeout and default_request_timeout is set to 0 but max_request_timeout is set, then the timeout will be max_request_timeout.
Please note that these timeouts are not 100% precise. The request may take an extra FUSE_TIMEOUT_TIMER_FREQ seconds beyond the set max timeout due to how it's internally implemented.
$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 0
$ echo 65536 | sudo tee /proc/sys/fs/fuse/default_request_timeout tee: /proc/sys/fs/fuse/default_request_timeout: Invalid argument
$ echo 65535 | sudo tee /proc/sys/fs/fuse/default_request_timeout 65535
$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 65535
$ echo 0 | sudo tee /proc/sys/fs/fuse/default_request_timeout 0
$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 0
--- Documentation/admin-guide/sysctl/fs.rst | 27 +++++++++++++++++++++++++ fs/fuse/fuse_i.h | 10 +++++++++ fs/fuse/inode.c | 16 +++++++++++++-- fs/fuse/sysctl.c | 20 ++++++++++++++++++ 4 files changed, 71 insertions(+), 2 deletions(-)
Documentation/admin-guide/sysctl/fs.rst +27 -0
@@ -342,3 +342,30 @@ filesystems: ``/proc/sys/fs/fuse/max_pages_limit`` is a read/write file for setting/getting the maximum number of pages that can be used for servicing requests in FUSE. + +``/proc/sys/fs/fuse/default_request_timeout`` is a read/write file for +setting/getting the default timeout (in minutes) for a fuse server to +reply to a kernel-issued request in the event where the server did not +specify a timeout at mount. If the server set a timeout, +then default_request_timeout will be ignored. The default +"default_request_timeout" is set to 0. 0 indicates a no-op (eg +requests will not have a default request timeout set if no timeout was +specified by the server). + +``/proc/sys/fs/fuse/max_request_timeout`` is a read/write file for +setting/getting the maximum timeout (in minutes) for a fuse server to +reply to a kernel-issued request. A value greater than 0 automatically opts +the server into a timeout that will be at most "max_request_timeout", even if +the server did not specify a timeout and default_request_timeout is set to 0. +If max_request_timeout is greater than 0 and the server set a timeout greater +than max_request_timeout or default_request_timeout is set to a value greater +than max_request_timeout, the system will use max_request_timeout as the +timeout. 0 indicates a no-op (eg requests will not have an upper bound on the +timeout and if the server did not request a timeout and default_request_timeout +was not set, there will be no timeout). + +Please note that for the timeout options, if the server does not respond to +the request by the time the timeout elapses, then the connection to the fuse +server will be aborted. Please also note that the timeouts are not 100% +precise (eg you may set 10 minutes but the timeout may kick in after 11 +minutes).
fs/fuse/fuse_i.h +10 -0
@@ -46,6 +46,16 @@/** Maximum of max_pages received in init_out */ extern unsigned int fuse_max_pages_limit; +/* +* Default timeout (in minutes) for the server to reply to a request +* before the connection is aborted, if no timeout was specified on mount. +*/ +extern unsigned int fuse_default_req_timeout; +/* +* Max timeout (in minutes) for the server to reply to a request before +* the connection is aborted. +*/ +extern unsigned int fuse_max_req_timeout;/** List of active connections */ extern struct list_head fuse_conn_list;
fs/fuse/inode.c +14 -2
@@ -36,6 +36,9 @@ DEFINE_MUTEX(fuse_mutex);static int set_global_limit (const char *val,const struct kernel_param *kp);unsigned int fuse_max_pages_limit =256 ; +/* default is no timeout */ +unsigned int fuse_default_req_timeout =0 ; +unsigned int fuse_max_req_timeout =0 ;unsigned max_user_bgreq; module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
@@ -1701,8 +1704,17 @@ EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) { -if (ctx->req_timeout) { -if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) +unsigned int timeout = ctx->req_timeout ?: fuse_default_req_timeout; + +if (fuse_max_req_timeout) { +if (!timeout) + timeout = fuse_max_req_timeout; +else + timeout =min (timeout, fuse_max_req_timeout); + } + +if (timeout) { +if (check_mul_overflow (timeout *60 ,HZ , &fc->timeout.req_timeout)) fc->timeout.req_timeout =ULONG_MAX ;timer_setup (&fc->timeout.timer, fuse_check_timeout,0 );mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ );
fs/fuse/sysctl.c +20 -0
@@ -13,6 +13,8 @@ static struct ctl_table_header *fuse_table_header;/* Bound by fuse_init_out max_pages, which is a u16 */ static unsigned int sysctl_fuse_max_pages_limit =65535 ; +static unsigned int sysctl_fuse_max_req_timeout_limit =U16_MAX ; +static struct ctl_table fuse_sysctl_table[] = { { .procname ="max_pages_limit" ,
@@ -23,6 +25,24 @@ static struct ctl_table fuse_sysctl_table[] = { .extra1 = SYSCTL_ONE, .extra2 = &sysctl_fuse_max_pages_limit, }, + { + .procname = "default_request_timeout", + .data = &fuse_default_req_timeout, + .maxlen = sizeof(fuse_default_req_timeout), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sysctl_fuse_max_req_timeout_limit, + }, + { + .procname = "max_request_timeout", + .data = &fuse_max_req_timeout, + .maxlen = sizeof(fuse_max_req_timeout), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sysctl_fuse_max_req_timeout_limit, + }, };int fuse_sysctl_register (void )
Signature · 1 line
2.43.5
Re: [PATCH RESEND v9 3/3] fuse: add default_request_timeout and max_request_timeout sysctls
Quoted 5 lines · Joanne Koong
Introduce two new sysctls, "
default_request_timeout" and "max_request_timeout". These control how long (in minutes) a server can take to reply to a request. If the server does not reply by the timeout, then the connection will be aborted.
Is this patch really needed? You have a per-mount limit already, do we need a global default and max?
Also, ditto here wrt seconds vs. minutes. If these *are* needed, then having them expressed in seconds would be more intuitive for most admins.
Quoted 174 lines
"
default_request_timeout" sets the default timeout if no timeout is specified by the fuse server on mount. 0 (default) indicates no default timeout should be enforced. If the server did specify a timeout, thendefault_request_timeoutwill be ignored."
max_request_timeout" sets the max amount of time the server may take to reply to a request. 0 (default) indicates no maximum timeout. Ifmax_request_timeoutis set and the fuse server attempts to set a timeout greater thanmax_request_timeout, the system will usemax_request_timeoutas the timeout. Similarly, ifdefault_request_timeoutis greater thanmax_request_timeout, the system will usemax_request_timeoutas the timeout. If the server does not request a timeout anddefault_request_timeoutis set to 0 butmax_request_timeoutis set, then the timeout will bemax_request_timeout.Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the set max timeout due to how it's internally implemented.$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 0$ echo 65536 | sudo tee /proc/sys/fs/fuse/default_request_timeout tee: /proc/sys/fs/fuse/default_request_timeout: Invalid argument$ echo 65535 | sudo tee /proc/sys/fs/fuse/default_request_timeout 65535$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 65535$ echo 0 | sudo tee /proc/sys/fs/fuse/default_request_timeout 0$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 0Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- Documentation/admin-guide/sysctl/fs.rst | 27 +++++++++++++++++++++++++ fs/fuse/fuse_i.h | 10 +++++++++ fs/fuse/inode.c | 16 +++++++++++++-- fs/fuse/sysctl.c | 20 ++++++++++++++++++ 4 files changed, 71 insertions(+), 2 deletions(-)Documentation/admin-guide/sysctl/fs.rst +27
-0@@ -342,3 +342,30 @@ filesystems: ``/proc/sys/fs/fuse/max_pages_limit`` is a read/write file for setting/getting the maximum number of pages that can be used for servicing requests in FUSE. + +``/proc/sys/fs/fuse/default_request_timeout`` is a read/write file for +setting/getting the default timeout (in minutes) for a fuse server to +reply to a kernel-issued request in the event where the server did not +specify a timeout at mount. If the server set a timeout, +then default_request_timeout will be ignored. The default +"default_request_timeout" is set to 0. 0 indicates a no-op (eg +requests will not have a default request timeout set if no timeout was +specified by the server). + +``/proc/sys/fs/fuse/max_request_timeout`` is a read/write file for +setting/getting the maximum timeout (in minutes) for a fuse server to +reply to a kernel-issued request. A value greater than 0 automatically opts +the server into a timeout that will be at most "max_request_timeout", even if +the server did not specify a timeout and default_request_timeout is set to 0. +If max_request_timeout is greater than 0 and the server set a timeout greater +than max_request_timeout or default_request_timeout is set to a value greater +than max_request_timeout, the system will use max_request_timeout as the +timeout. 0 indicates a no-op (eg requests will not have an upper bound on the +timeout and if the server did not request a timeout and default_request_timeout +was not set, there will be no timeout). + +Please note that for the timeout options, if the server does not respond to +the request by the time the timeout elapses, then the connection to the fuse +server will be aborted. Please also note that the timeouts are not 100% +precise (eg you may set 10 minutes but the timeout may kick in after 11 +minutes).fs/fuse/fuse_i.h +10
-0@@ -46,6 +46,16 @@/** Maximum of max_pages received in init_out */ extern unsigned int fuse_max_pages_limit; +/* +* Default timeout (in minutes) for the server to reply to a request +* before the connection is aborted, if no timeout was specified on mount. +*/ +extern unsigned int fuse_default_req_timeout; +/* +* Max timeout (in minutes) for the server to reply to a request before +* the connection is aborted. +*/ +extern unsigned int fuse_max_req_timeout;/** List of active connections */ extern struct list_head fuse_conn_list;fs/fuse/inode.c +14
-2@@ -36,6 +36,9 @@ DEFINE_MUTEX(fuse_mutex);static int set_global_limit (const char *val,const struct kernel_param *kp);unsigned int fuse_max_pages_limit =256 ; +/* default is no timeout */ +unsigned int fuse_default_req_timeout =0 ; +unsigned int fuse_max_req_timeout =0 ;unsigned max_user_bgreq; module_param_call(max_user_bgreq, set_global_limit, param_get_uint,@@ -1701,8 +1704,17 @@ EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) { -if (ctx->req_timeout) { -if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) +unsigned int timeout = ctx->req_timeout ?: fuse_default_req_timeout; + +if (fuse_max_req_timeout) { +if (!timeout) + timeout = fuse_max_req_timeout; +else + timeout =min (timeout, fuse_max_req_timeout); + } + +if (timeout) { +if (check_mul_overflow (timeout *60 ,HZ , &fc->timeout.req_timeout)) fc->timeout.req_timeout =ULONG_MAX ;timer_setup (&fc->timeout.timer, fuse_check_timeout,0 );mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ );fs/fuse/sysctl.c +20
-0@@ -13,6 +13,8 @@ static struct ctl_table_header *fuse_table_header;/* Bound by fuse_init_out max_pages, which is a u16 */ static unsigned int sysctl_fuse_max_pages_limit =65535 ; +static unsigned int sysctl_fuse_max_req_timeout_limit =U16_MAX ; +static struct ctl_table fuse_sysctl_table[] = { { .procname ="max_pages_limit" ,@@ -23,6 +25,24 @@ static struct ctl_table fuse_sysctl_table[] = { .extra1 = SYSCTL_ONE, .extra2 = &sysctl_fuse_max_pages_limit, }, + { + .procname = "default_request_timeout", + .data = &fuse_default_req_timeout, + .maxlen = sizeof(fuse_default_req_timeout), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sysctl_fuse_max_req_timeout_limit, + }, + { + .procname = "max_request_timeout", + .data = &fuse_max_req_timeout, + .maxlen = sizeof(fuse_max_req_timeout), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sysctl_fuse_max_req_timeout_limit, + }, };int fuse_sysctl_register (void )
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH RESEND v9 3/3] fuse: add default_request_timeout and max_request_timeout sysctls
Quoted 10 lines · Jeff Layton
Quoted 5 lines · Joanne Koong
Introduce two new sysctls, "
default_request_timeout" and "max_request_timeout". These control how long (in minutes) a server can take to reply to a request. If the server does not reply by the timeout, then the connection will be aborted.Is this patch really needed? You have a per-mount limit already, do we need a global default and max?
Yes, I do think this patch would be useful. Since fuse servers can run unprivileged, I think giving system admins some way to enforce a default and max limit that applies automatically to all fuse servers would help protect against malicious or buggy servers on the system.
Quoted 4 lines
Also, ditto here wrt seconds vs. minutes. If these *are* needed, then having them expressed in seconds would be more intuitive for most admins.
Noted. I'll change this to seconds in the next iteration of this patchset.
Quoted 179 lines
Quoted 166 lines
"
default_request_timeout" sets the default timeout if no timeout is specified by the fuse server on mount. 0 (default) indicates no default timeout should be enforced. If the server did specify a timeout, thendefault_request_timeoutwill be ignored."
max_request_timeout" sets the max amount of time the server may take to reply to a request. 0 (default) indicates no maximum timeout. Ifmax_request_timeoutis set and the fuse server attempts to set a timeout greater thanmax_request_timeout, the system will usemax_request_timeoutas the timeout. Similarly, ifdefault_request_timeoutis greater thanmax_request_timeout, the system will usemax_request_timeoutas the timeout. If the server does not request a timeout anddefault_request_timeoutis set to 0 butmax_request_timeoutis set, then the timeout will bemax_request_timeout.Please note that these timeouts are not 100% precise. The request may take an extra
FUSE_TIMEOUT_TIMER_FREQseconds beyond the set max timeout due to how it's internally implemented.$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 0$ echo 65536 | sudo tee /proc/sys/fs/fuse/default_request_timeout tee: /proc/sys/fs/fuse/default_request_timeout: Invalid argument$ echo 65535 | sudo tee /proc/sys/fs/fuse/default_request_timeout 65535$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 65535$ echo 0 | sudo tee /proc/sys/fs/fuse/default_request_timeout 0$ sysctl -a | grep fuse.default_request_timeout fs.fuse.default_request_timeout = 0Signed-off-by Joanne Koong <[email protected]>Reviewed-by Bernd Schubert <[email protected]>--- Documentation/admin-guide/sysctl/fs.rst | 27 +++++++++++++++++++++++++ fs/fuse/fuse_i.h | 10 +++++++++ fs/fuse/inode.c | 16 +++++++++++++-- fs/fuse/sysctl.c | 20 ++++++++++++++++++ 4 files changed, 71 insertions(+), 2 deletions(-)Documentation/admin-guide/sysctl/fs.rst +27
-0@@ -342,3 +342,30 @@ filesystems: ``/proc/sys/fs/fuse/max_pages_limit`` is a read/write file for setting/getting the maximum number of pages that can be used for servicing requests in FUSE. + +``/proc/sys/fs/fuse/default_request_timeout`` is a read/write file for +setting/getting the default timeout (in minutes) for a fuse server to +reply to a kernel-issued request in the event where the server did not +specify a timeout at mount. If the server set a timeout, +then default_request_timeout will be ignored. The default +"default_request_timeout" is set to 0. 0 indicates a no-op (eg +requests will not have a default request timeout set if no timeout was +specified by the server). + +``/proc/sys/fs/fuse/max_request_timeout`` is a read/write file for +setting/getting the maximum timeout (in minutes) for a fuse server to +reply to a kernel-issued request. A value greater than 0 automatically opts +the server into a timeout that will be at most "max_request_timeout", even if +the server did not specify a timeout and default_request_timeout is set to 0. +If max_request_timeout is greater than 0 and the server set a timeout greater +than max_request_timeout or default_request_timeout is set to a value greater +than max_request_timeout, the system will use max_request_timeout as the +timeout. 0 indicates a no-op (eg requests will not have an upper bound on the +timeout and if the server did not request a timeout and default_request_timeout +was not set, there will be no timeout). + +Please note that for the timeout options, if the server does not respond to +the request by the time the timeout elapses, then the connection to the fuse +server will be aborted. Please also note that the timeouts are not 100% +precise (eg you may set 10 minutes but the timeout may kick in after 11 +minutes).fs/fuse/fuse_i.h +10
-0@@ -46,6 +46,16 @@/** Maximum of max_pages received in init_out */ extern unsigned int fuse_max_pages_limit; +/* +* Default timeout (in minutes) for the server to reply to a request +* before the connection is aborted, if no timeout was specified on mount. +*/ +extern unsigned int fuse_default_req_timeout; +/* +* Max timeout (in minutes) for the server to reply to a request before +* the connection is aborted. +*/ +extern unsigned int fuse_max_req_timeout;/** List of active connections */ extern struct list_head fuse_conn_list;fs/fuse/inode.c +14
-2@@ -36,6 +36,9 @@ DEFINE_MUTEX(fuse_mutex);static int set_global_limit (const char *val,const struct kernel_param *kp);unsigned int fuse_max_pages_limit =256 ; +/* default is no timeout */ +unsigned int fuse_default_req_timeout =0 ; +unsigned int fuse_max_req_timeout =0 ;unsigned max_user_bgreq; module_param_call(max_user_bgreq, set_global_limit, param_get_uint,@@ -1701,8 +1704,17 @@ EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);static void fuse_init_fc_timeout (struct fuse_conn *fc,struct fuse_fs_context *ctx) { -if (ctx->req_timeout) { -if (check_mul_overflow (ctx->req_timeout *60 ,HZ , &fc->timeout.req_timeout)) +unsigned int timeout = ctx->req_timeout ?: fuse_default_req_timeout; + +if (fuse_max_req_timeout) { +if (!timeout) + timeout = fuse_max_req_timeout; +else + timeout =min (timeout, fuse_max_req_timeout); + } + +if (timeout) { +if (check_mul_overflow (timeout *60 ,HZ , &fc->timeout.req_timeout)) fc->timeout.req_timeout =ULONG_MAX ;timer_setup (&fc->timeout.timer, fuse_check_timeout,0 );mod_timer (&fc->timeout.timer, jiffies +FUSE_TIMEOUT_TIMER_FREQ );fs/fuse/sysctl.c +13
-0@@ -13,6 +13,8 @@ static struct ctl_table_header *fuse_table_header;/* Bound by fuse_init_out max_pages, which is a u16 */ static unsigned int sysctl_fuse_max_pages_limit =65535 ; +static unsigned int sysctl_fuse_max_req_timeout_limit =U16_MAX ; +static struct ctl_table fuse_sysctl_table[] = { { .procname ="max_pages_limit" ,@@ -23,6 +25,24 @@ static struct ctl_table fuse_sysctl_table[] = { .extra1 = SYSCTL_ONE, .extra2 = &sysctl_fuse_max_pages_limit, }, + { + .procname = "default_request_timeout", + .data = &fuse_default_req_timeout, + .maxlen = sizeof(fuse_default_req_timeout), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sysctl_fuse_max_req_timeout_limit, + }, + { + .procname = "max_request_timeout",fs/fuse/sysctl.c static struct ctl_table fuse_sysctl_table[] = { · line 48
Quoted 10 lines
fs/fuse/sysctl.c +7
-0+ .data = &fuse_max_req_timeout, + .maxlen = sizeof(fuse_max_req_timeout), + .mode = 0644, + .proc_handler = proc_douintvec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = &sysctl_fuse_max_req_timeout_limit, + }, };int fuse_sysctl_register (void )--
Jeff Layton <[email protected]>
Quoted 71 lines · Joanne Koong
There are situations where fuse servers can become unresponsive or stuck, for example if the server is in a deadlock. Currently, there's no good way to detect if a server is stuck and needs to be killed manually.
This patchset adds a timeout option where if the server does not reply to a request by the time the timeout elapses, the connection will be aborted. This patchset also adds two dynamically configurable fuse sysctls "
default_request_timeout" and "max_request_timeout" for controlling/enforcing timeout behavior system-wide.Existing systems running fuse servers will not be affected unless they explicitly opt into the timeout.
v8:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v8 -> v9:
* Fix comment for u16fs_parse_result,ULONG_MAXinstead ofU32_MAX, fix spacing (Bernd)v7:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v7 -> v8:
* Use existing lists for checking expirations (Miklos)v6:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v6 -> v7:
- Make timer per-connection instead of per-request (Miklos)
- Make default granularity of time minutes instead of seconds
- Removed the reviewed-bys since the interface of this has changed (now minutes, instead of seconds)v5:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v5 -> v6:
- Gate sysctl.o behindCONFIG_SYSCTLin makefile (kernel test robot)
- Reword/clarify last sentence in cover letter (Miklos)v4:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v4 -> v5:
- Change timeout behavior from aborting request to aborting connection (Miklos)
- Clarify wording for sysctl documentation (Jingbo)v3:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v3 -> v4:
- Fix wording on some comments to make it more clear
- Use simpler logic for timer (eg remove extra if checks, use mod timer API) (Josef)
- Sanity-check should be onFR_FINISHINGnotFR_FINISHED(Jingbo)
- Fix comment for "processing queue", addreq->fpq= NULL safeguard (Bernd)v2:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v2 -> v3:
- Disarm / rearm timer indev_do_readto handle race conditions (Bernrd)
- Disarm timer in error handling for fatal interrupt (Yafang)
- Clean updo_fuse_request_end(Jingbo)
- Add timer for notify retrieve requests
- Fix kernel test robot errors for #define no-op functionsv1:
https://lore.kernel.org/linux-fsdevel/[email protected]/
Changes from v1 -> v2:
- Add timeout for background requests
- Handle resend race condition
- Add sysctls
Quoted 5 lines
Joanne Koong (3):
fs_parser: addfsparam_u16helper
fuse: add optional kernel-enforced timeout for requests
fuse: adddefault_request_timeoutandmax_request_timeoutsysctls
These look great Joanne, you can add
to the series. Thanks,
Josef