uprobes: document mmap_lock, don't abuse get_user_pages_remote()
Subsystems (6)
Oleg Nesterov
maintainercover uprobes: document mmap_lock, don't abuse get_user_pages_remote()
Peter, these simple cleanups should not conflict with your changes, but I can resend them later if it causes any inconvenience.
Oleg.
Oleg Nesterov
maintainerPATCH 1/2 uprobes: document the usage of mm->mmap_lock
The comment above uprobe_write_opcode() is wrong, unapply_uprobe() calls it under mmap_read_lock() and this is correct.
And it is completely unclear why register_for_each_vma() takes mmap_lock for writing, add a comment to explain that mmap_write_lock() is needed to avoid the following race:
- A task T hits the bp installed by uprobe and calls find_active_uprobe()
- uprobe_unregister() removes this uprobe/bp
- T calls find_uprobe() which returns NULL
- another uprobe_register() installs the bp at the same address
- T calls is_trap_at_addr() which returns true
- T returns to handle_swbp() and gets SIGTRAP.
--- kernel/events/uprobes.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
kernel/events/uprobes.c +7 -2
@@ -453,7 +453,7 @@ static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,* @vaddr: the virtual address to store the opcode. * @opcode: opcode to be written at @vaddr. * -* Called with mm->mmap_lock held for write. +* Called with mm->mmap_lock held for read or write. * Return 0 (success) or a negative errno. */ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
@@ -1046,7 +1046,12 @@ register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)if (err && is_register) goto free; - +/* +* We take mmap_lock for writing to avoid the race with +* find_active_uprobe(), install_breakpoint() must not +* make is_trap_at_addr() true right after find_uprobe() +* returns NULL. +*/ mmap_write_lock (mm); vma =find_vma (mm, info->vaddr); if (!vma || !valid_vma(vma, is_register) ||
Signature · 1 line
2.25.1.362.g51ebf55
Masami Hiramatsu (Google)
maintainerRe: [PATCH 1/2] uprobes: document the usage of mm->mmap_lock
Quoted 40 lines · On Wed, 10 Jul 2024 16:00:45 +0200 Oleg Nesterov <[email protected]> wrote:
The comment above
uprobe_write_opcode()is wrong,unapply_uprobe()calls it undermmap_read_lock()and this is correct.And it is completely unclear why
register_for_each_vma()takesmmap_lockfor writing, add a comment to explain thatmmap_write_lock()is needed to avoid the following race:- A task T hits the bp installed by uprobe and calls
find_active_uprobe()-
uprobe_unregister()removes this uprobe/bp- T calls
find_uprobe()which returns NULL- another
uprobe_register()installs the bp at the same address- T calls
is_trap_at_addr()which returns true- T returns to
handle_swbp()and gets SIGTRAP.Reported-by Andrii Nakryiko <[email protected]>Signed-off-by Oleg Nesterov <[email protected]>--- kernel/events/uprobes.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)kernel/events/uprobes.c +1
-1@@ -453,7 +453,7 @@ static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,* @vaddr: the virtual address to store the opcode. * @opcode: opcode to be written at @vaddr. * -* Called with mm->mmap_lock held for write. +* Called with mm->mmap_lock held for read or write. * Return 0 (success) or a negative errno. */ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
kernel/events/uprobes.c register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) · line 1053
Quoted 9 lines
kernel/events/uprobes.c +5
-1@@ -1046,7 +1046,12 @@ register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)if (err && is_register) goto free; - +/* +* We take mmap_lock for writing to avoid the race with +* find_active_uprobe(), install_breakpoint() must not +* make is_trap_at_addr() true right after find_uprobe() +* returns NULL.
Sorry, I couldn't catch the latter part. What is the relationship of taking the mmap_lock and install_breakpoint() and is_trap_at_addr() here?
You meant that find_active_uprobe() is using find_uprobe() which searchs uprobe form rbtree? But it seems uprobe is already inserted to the rbtree in alloc_uprobe() so find_uprobe() will not return NULL here, right?
Thank you,
Quoted 8 lines
kernel/events/uprobes.c +1
-0+*/ mmap_write_lock (mm); vma =find_vma (mm, info->vaddr); if (!vma || !valid_vma(vma, is_register) ||Signature · 1 line
2.25.1.362.g51ebf55
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Oleg Nesterov
maintainerRe: [PATCH 1/2] uprobes: document the usage of mm->mmap_lock
Quoted 23 lines · Masami Hiramatsu
Quoted 19 lines · On Wed, 10 Jul 2024 16:00:45 +0200 Oleg Nesterov <[email protected]> wrote:
The comment above
uprobe_write_opcode()is wrong,unapply_uprobe()calls it undermmap_read_lock()and this is correct.And it is completely unclear why
register_for_each_vma()takesmmap_lockfor writing, add a comment to explain thatmmap_write_lock()is needed to avoid the following race:- A task T hits the bp installed by uprobe and calls
find_active_uprobe()-
uprobe_unregister()removes this uprobe/bp- T calls
find_uprobe()which returns NULL- another
uprobe_register()installs the bp at the same address- T calls
is_trap_at_addr()which returns true- T returns to
handle_swbp()and gets SIGTRAP.
...
Quoted 14 lines
kernel/events/uprobes.c register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) · line 1053
Quoted 11 lines
int uprobe_write_opcode(struct
arch_uprobe*auprobe, structmm_struct*mm,kernel/events/uprobes.c +5
-1@@ -1046,7 +1046,12 @@ register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)if (err && is_register) goto free; - +/* +* We take mmap_lock for writing to avoid the race with +* find_active_uprobe(), install_breakpoint() must not +* make is_trap_at_addr() true right after find_uprobe() +* returns NULL. Sorry, I couldn't catch the latter part. What is the relationship of taking the
mmap_lockandinstall_breakpoint()andis_trap_at_addr()here?
Please the the changelog above, it tries to explain this race with more details...
Quoted 2 lines
You meant that
find_active_uprobe()is usingfind_uprobe()which searchs uprobe form rbtree?
Yes,
Quoted 2 lines
But it seems uprobe is already inserted to the rbtree
inalloc_uprobe()sofind_uprobe()will not return NULL here, right?
uprobe_register() -> alloc_uprobe() can come after find_active_uprobe() -> find_uprobe() returns NULL.
Now, if uprobe_register() -> register_for_each_vma() used mmap_read_lock(), it could do install_breakpoint() before find_active_uprobe() calls is_trap_at_addr().
In this case find_active_uprobe() returns with uprobe == NULL and is_swbp == 1, handle_swbp() treat this case as the "normal" int3 without uprobe and do
if (!uprobe) {
if (is_swbp > 0) {
/* No matching uprobe; signal SIGTRAP. */
force_sig(SIGTRAP);Does this answer your question?
Oleg.
Masami Hiramatsu (Google)
maintainerRe: [PATCH 1/2] uprobes: document the usage of mm->mmap_lock
Quoted 42 lines · On Wed, 10 Jul 2024 17:10:07 +0200 Oleg Nesterov <[email protected]> wrote:
Quoted 23 lines · Masami Hiramatsu
Quoted 19 lines · On Wed, 10 Jul 2024 16:00:45 +0200 Oleg Nesterov <[email protected]> wrote:
The comment above
uprobe_write_opcode()is wrong,unapply_uprobe()calls it undermmap_read_lock()and this is correct.And it is completely unclear why
register_for_each_vma()takesmmap_lockfor writing, add a comment to explain thatmmap_write_lock()is needed to avoid the following race:- A task T hits the bp installed by uprobe and calls
find_active_uprobe()-
uprobe_unregister()removes this uprobe/bp- T calls
find_uprobe()which returns NULL- another
uprobe_register()installs the bp at the same address- T calls
is_trap_at_addr()which returns true- T returns to
handle_swbp()and gets SIGTRAP....
Quoted 14 lines
kernel/events/uprobes.c register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new) · line 1053
Quoted 11 lines
int uprobe_write_opcode(struct
arch_uprobe*auprobe, structmm_struct*mm,kernel/events/uprobes.c +5
-1@@ -1046,7 +1046,12 @@ register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)if (err && is_register) goto free; - +/* +* We take mmap_lock for writing to avoid the race with +* find_active_uprobe(), install_breakpoint() must not +* make is_trap_at_addr() true right after find_uprobe() +* returns NULL. Sorry, I couldn't catch the latter part. What is the relationship of taking the
mmap_lockandinstall_breakpoint()andis_trap_at_addr()here?
Quoted 2 lines
Please the the changelog above, it tries to explain this race with more details...
OK, but it seems we should write the above longer explanation here. What about the comment like this?
/*
* We take mmap_lock for writing to avoid the race with
* find_active_uprobe() and is_trap_at_adder() in reader
* side.
* If the reader, which hits a swbp and is handling it,
* does not take mmap_lock for reading, it is possible
* that find_active_uprobe() returns NULL (because
* uprobe_unregister() removes uprobes right before that),
* but is_trap_at_addr() can return true afterwards (because
* another thread calls uprobe_register() on the same address).
* This causes unexpected SIGTRAP on reader thread.
* Taking mmap_lock avoids this race.
*/
Quoted 18 lines
Quoted 3 lines
You meant that
find_active_uprobe()is usingfind_uprobe()which searchs uprobe form rbtree?Yes,
Quoted 2 lines
But it seems uprobe is already inserted to the rbtree
inalloc_uprobe()sofind_uprobe()will not return NULL here, right?
uprobe_register()->alloc_uprobe()can come afterfind_active_uprobe()->find_uprobe()returns NULL.Now, if
uprobe_register()->register_for_each_vma()usedmmap_read_lock(), it could doinstall_breakpoint()beforefind_active_uprobe()callsis_trap_at_addr().In this case
find_active_uprobe()returns with uprobe == NULL andis_swbp== 1,handle_swbp()treat this case as the "normal" int3 without uprobe and do
Quoted 6 lines
if (!uprobe) { if (is_swbp > 0) { /* No matching uprobe; signal SIGTRAP. */ force_sig(SIGTRAP);Does this answer your question?
No, thanks for the explanation.
Thank you!
Quoted 3 lines
Oleg.
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Oleg Nesterov
maintainerRe: [PATCH 1/2] uprobes: document the usage of mm->mmap_lock
Quoted 16 lines · Masami Hiramatsu
Quoted 12 lines · On Wed, 10 Jul 2024 17:10:07 +0200 Oleg Nesterov <[email protected]> wrote:
Quoted 12 lines
Quoted 11 lines
int uprobe_write_opcode(struct
arch_uprobe*auprobe, structmm_struct*mm,kernel/events/uprobes.c +5
-1@@ -1046,7 +1046,12 @@ register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)if (err && is_register) goto free; - +/* +* We take mmap_lock for writing to avoid the race with +* find_active_uprobe(), install_breakpoint() must not +* make is_trap_at_addr() true right after find_uprobe() +* returns NULL.
...
Quoted 2 lines
OK, but it seems we should write the above longer explanation here. What about the comment like this?
Well, I am biased, but your version looks much more confusing to me...
Quoted 6 lines
/*
* We takemmap_lockfor writing to avoid the race with
*find_active_uprobe()andis_trap_at_adder()in reader
* side.
* If the reader, which hits a swbp and is handling it,
* does not takemmap_lockfor reading,
this looks as if the reader which hits a swbp takes mmap_lock for reading because of this race. No, find_active_uprobe() needs mmap_read_lock() for vma_lookup, get_user_pages, etc.
Quoted 5 lines
it is possible
* thatfind_active_uprobe()returns NULL (because
*uprobe_unregister()removes uprobes right before that),
* butis_trap_at_addr()can return true afterwards (because
* another thread callsuprobe_register()on the same address).
^^^^^^^^^^^^^^^
We are the thread which called uprobe_register(), we are going to do install_breakpoint().
And btw, not that I think this makes sense, but register_for_each_vma() could probably do
if (is_register) mmap_write_lock(mm); else mmap_read_lock(mm);
Oleg.
Masami Hiramatsu (Google)
maintainerRe: [PATCH 1/2] uprobes: document the usage of mm->mmap_lock
Quoted 32 lines · On Thu, 11 Jul 2024 11:49:40 +0200 Oleg Nesterov <[email protected]> wrote:
Quoted 16 lines · Masami Hiramatsu
Quoted 12 lines · On Wed, 10 Jul 2024 17:10:07 +0200 Oleg Nesterov <[email protected]> wrote:
Quoted 12 lines
Quoted 11 lines
int uprobe_write_opcode(struct
arch_uprobe*auprobe, structmm_struct*mm,kernel/events/uprobes.c +5
-1@@ -1046,7 +1046,12 @@ register_for_each_vma(struct uprobe *uprobe, struct uprobe_consumer *new)if (err && is_register) goto free; - +/* +* We take mmap_lock for writing to avoid the race with +* find_active_uprobe(), install_breakpoint() must not +* make is_trap_at_addr() true right after find_uprobe() +* returns NULL. ...
Quoted 2 lines
OK, but it seems we should write the above longer explanation here. What about the comment like this?
Well, I am biased, but your version looks much more confusing to me...
Quoted 6 lines
/*
* We takemmap_lockfor writing to avoid the race with
*find_active_uprobe()andis_trap_at_adder()in reader
* side.
* If the reader, which hits a swbp and is handling it,
* does not takemmap_lockfor reading,
Quoted 3 lines
this looks as if the reader which hits a swbp takes
mmap_lockfor reading because of this race. No,find_active_uprobe()needsmmap_read_lock()forvma_lookup,get_user_pages, etc.
OK, so it is for looking up VMA. (But in the end, this rock protects both the VMAs and uprobes, right?)
Quoted 9 lines
Quoted 6 lines
it is possible
* thatfind_active_uprobe()returns NULL (because
*uprobe_unregister()removes uprobes right before that),
* butis_trap_at_addr()can return true afterwards (because
* another thread callsuprobe_register()on the same address).^^^^^^^^^^^^^^^We are the thread which called
uprobe_register(), we are going to doinstall_breakpoint().
Ah, yes :)
What about this?
* We take mmap_lock for writing to avoid the race with
* find_active_uprobe(), which takes mmap_lock for reading.
* Thus this install_breakpoint() must not make
* is_trap_at_addr() true right after find_uprobe()
* returns NULL in find_active_uprobe().
Quoted 8 lines
And btw, not that I think this makes sense, but
register_for_each_vma()could probably doif (is_register) mmap_write_lock(mm); else mmap_read_lock(mm);
Agreed.
Thank you,
Quoted 3 lines
Oleg.
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Oleg Nesterov
maintainerRe: [PATCH 1/2] uprobes: document the usage of mm->mmap_lock
Quoted 8 lines · Masami Hiramatsu
What about this?
* We take
mmap_lockfor writing to avoid the race with
*find_active_uprobe(), which takesmmap_lockfor reading.
* Thus thisinstall_breakpoint()must not make
*is_trap_at_addr()true right afterfind_uprobe()
* returns NULL infind_active_uprobe().
Thanks! will change.
Oleg.
Oleg Nesterov
maintainerPATCH 2/2 uprobes: is_trap_at_addr: don't use get_user_pages_remote()
In reply to Oleg Nesterovget_user_pages_remote() and the comment above it make no sense.
There is no task_struct passed into get_user_pages_remote() anymore, and nowadays mm_account_fault() increments the current->min/maj_flt counters regardless of FAULT_FLAG_REMOTE.
--- kernel/events/uprobes.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
kernel/events/uprobes.c +2 -8
@@ -453,7 +453,7 @@ static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,* @vaddr: the virtual address to store the opcode. * @opcode: opcode to be written at @vaddr. * -* Called with mm->mmap_lock held for read or write. +* Called with mm->mmap_lock held. * Return 0 (success) or a negative errno. */ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
@@ -2024,13 +2024,7 @@ static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)if (likely (result ==0 )) goto out; -/* -* The NULL 'tsk' here ensures that any faults that occur here -* will not be accounted to the task. 'mm' *is* current->mm, -* but we treat this as a 'remote' access since it is -* essentially a kernel access to the memory. -*/ - result =get_user_pages_remote (mm, vaddr,1 ,FOLL_FORCE , &page,NULL ); + result =get_user_pages (vaddr,1 ,FOLL_FORCE , &page);if (result <0 )return result;
Signature · 1 line
2.25.1.362.g51ebf55
Re: [PATCH 2/2] uprobes: is_trap_at_addr: don't use get_user_pages_remote()
Quoted 7 lines · Oleg Nesterov
get_user_pages_remote()and the comment above it make no sense.There is no
task_structpassed intoget_user_pages_remote()anymore, and nowadaysmm_account_fault()increments the current->min/maj_flt counters regardless ofFAULT_FLAG_REMOTE.
Quoted 6 lines
Reported-by Andrii Nakryiko <[email protected]>Signed-off-by Oleg Nesterov <[email protected]>--- kernel/events/uprobes.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
Makes sense
Quoted 32 lines
kernel/events/uprobes.c +2
-9@@ -453,7 +453,7 @@ static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,* @vaddr: the virtual address to store the opcode. * @opcode: opcode to be written at @vaddr. * -* Called with mm->mmap_lock held for read or write. +* Called with mm->mmap_lock held. * Return 0 (success) or a negative errno. */ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,@@ -2024,13 +2024,7 @@ static int is_trap_at_addr(struct mm_struct *mm, unsigned long vaddr)if (likely (result ==0 )) goto out; -/* -* The NULL 'tsk' here ensures that any faults that occur here -* will not be accounted to the task. 'mm' *is* current->mm, -* but we treat this as a 'remote' access since it is -* essentially a kernel access to the memory. -*/ - result =get_user_pages_remote (mm, vaddr,1 ,FOLL_FORCE , &page,NULL ); + result =get_user_pages (vaddr,1 ,FOLL_FORCE , &page);if (result <0 )return result; --2.25.1.362.g51ebf55
Oleg Nesterov
maintainercover uprobes: future cleanups for review
In reply to Oleg NesterovQuoted 3 lines · Oleg Nesterov
Peter, these simple cleanups should not conflict with your changes, but I can resend them later if it causes any inconvenience.
In fact I would like to push 2 more cleanups before the more significant changes, but they certainly conflict with your ongoing work, albeit only textually.
Let me send the patches for review anyway, perhaps you can take at least the 1st one.
3/3 is only compile tested so far. Andrii, can you take a look?
Oleg.
Oleg Nesterov
maintainerPATCH 1/3 uprobes: kill uprobe_register_refctr()
It doesn't make any sense to have 2 versions of _register(). Note that trace_uprobe_enable(), the only user of uprobe_register(), doesn't need to check tu->ref_ctr_offset to decide which one should be used, it could safely pass ref_ctr_offset == 0 to uprobe_register_refctr().
Add this argument to uprobe_register(), update the callers, and kill uprobe_register_refctr().
--- include/linux/uprobes.h | 9 ++------- kernel/events/uprobes.c | 23 +++++------------------ kernel/trace/bpf_trace.c | 2 +- kernel/trace/trace_uprobe.c | 8 ++------ 4 files changed, 10 insertions(+), 32 deletions(-)
include/linux/uprobes.h +2 -7
@@ -110,8 +110,7 @@ extern bool is_trap_insn(uprobe_opcode_t *insn);extern unsigned long uprobe_get_swbp_addr (struct pt_regs *regs);extern unsigned long uprobe_get_trap_addr (struct pt_regs *regs);extern int uprobe_write_opcode (struct arch_uprobe *auprobe,struct mm_struct *mm,unsigned long vaddr,uprobe_opcode_t ); -extern int uprobe_register (struct inode *inode,loff_t offset,struct uprobe_consumer *uc); -extern int uprobe_register_refctr (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc); +extern int uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc);extern int uprobe_apply (struct inode *inode,loff_t offset,struct uprobe_consumer *uc,bool );extern void uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc);extern int uprobe_mmap (struct vm_area_struct *vma);
@@ -149,11 +148,7 @@ static inline void uprobes_init(void)#define uprobe_get_trap_addr (regs) instruction_pointer(regs)static inline int -uprobe_register (struct inode *inode,loff_t offset,struct uprobe_consumer *uc) -{ -return -ENOSYS ; -} -static inline int uprobe_register_refctr (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc) +uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc) {return -ENOSYS ; }
kernel/events/uprobes.c +5 -18
@@ -1120,25 +1120,25 @@ void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumeEXPORT_SYMBOL_GPL (uprobe_unregister);/* -* __uprobe_register - register a probe +* uprobe_register - register a probe * @inode: the file in which the probe has to be placed. * @offset: offset from the start of the file. * @uc: information on howto handle the probe.. * -* Apart from the access refcount, __uprobe_register() takes a creation +* Apart from the access refcount, uprobe_register() takes a creation * refcount (thro alloc_uprobe) if and only if this @uprobe is getting * inserted into the rbtree (i.e first consumer for a @inode:@offset * tuple). Creation refcount stops uprobe_unregister from freeing the * @uprobe even before the register operation is complete. Creation * refcount is released when the last @uc for the @uprobe -* unregisters. Caller of __uprobe_register() is required to keep @inode +* unregisters. Caller of uprobe_register() is required to keep @inode * (and the containing mount) referenced. * * Return errno if it cannot successully install probes * else return 0 (success) */ -static int __uprobe_register (struct inode *inode,loff_t offset, -loff_t ref_ctr_offset,struct uprobe_consumer *uc) +int uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset, +struct uprobe_consumer *uc) {struct uprobe *uprobe;int ret;
@@ -1190,21 +1190,8 @@ static int __uprobe_register(struct inode *inode, loff_t offset, goto retry;return ret; } - -int uprobe_register (struct inode *inode,loff_t offset, -struct uprobe_consumer *uc) -{ -return __uprobe_register (inode, offset,0 , uc); -}EXPORT_SYMBOL_GPL (uprobe_register); -int uprobe_register_refctr (struct inode *inode,loff_t offset, -loff_t ref_ctr_offset,struct uprobe_consumer *uc) -{ -return __uprobe_register (inode, offset, ref_ctr_offset, uc); -} -EXPORT_SYMBOL_GPL (uprobe_register_refctr); -/* * uprobe_apply - unregister an already registered probe. * @inode: the file in which the probe has to be removed.
kernel/trace/bpf_trace.c +1 -1
@@ -3477,7 +3477,7 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register_refctr (d_real_inode (link->path.dentry), + err =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer);
kernel/trace/trace_uprobe.c +2 -6
@@ -1089,12 +1089,8 @@ static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter) tu->consumer.filter = filter; tu->inode =d_real_inode (tu->path.dentry); -if (tu->ref_ctr_offset) - ret =uprobe_register_refctr (tu->inode, tu->offset, - tu->ref_ctr_offset, &tu->consumer); -else - ret =uprobe_register (tu->inode, tu->offset, &tu->consumer); - + ret =uprobe_register (tu->inode, tu->offset, tu->ref_ctr_offset, + &tu->consumer);if (ret) tu->inode =NULL ;
Signature · 1 line
2.25.1.362.g51ebf55
Re: [PATCH 1/3] uprobes: kill uprobe_register_refctr()
Quoted 9 lines · Oleg Nesterov
It doesn't make any sense to have 2 versions of
_register(). Note thattrace_uprobe_enable(), the only user ofuprobe_register(), doesn't need to checktu->ref_ctr_offsetto decide which one should be used, it could safely passref_ctr_offset== 0 touprobe_register_refctr().Add this argument to
uprobe_register(), update the callers, and killuprobe_register_refctr().
Quoted 8 lines
Signed-off-by Oleg Nesterov <[email protected]>--- include/linux/uprobes.h | 9 ++------- kernel/events/uprobes.c | 23 +++++------------------ kernel/trace/bpf_trace.c | 2 +- kernel/trace/trace_uprobe.c | 8 ++------ 4 files changed, 10 insertions(+), 32 deletions(-)
LGTM with few nits below.
Quoted 8 lines
/*
*uprobe_apply- unregister an already registered probe.
* @inode: the file in which the probe has to be removed.kernel/trace/bpf_trace.c +0
-0
kernel/trace/bpf_trace.c int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr · line 3483
Quoted 8 lines
kernel/trace/bpf_trace.c +1
-1@@ -3477,7 +3477,7 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register_refctr (d_real_inode (link->path.dentry), + err =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer);
please adjust indentation here
Quoted 6 lines
kernel/trace/trace_uprobe.c +0
-0@@ -1089,12 +1089,8 @@ static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter) tu->consumer.filter = filter;
kernel/trace/trace_uprobe.c static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter) · line 1093
Quoted 10 lines
kernel/trace/trace_uprobe.c +2
-6tu->inode =d_real_inode (tu->path.dentry); -if (tu->ref_ctr_offset) - ret =uprobe_register_refctr (tu->inode, tu->offset, - tu->ref_ctr_offset, &tu->consumer); -else - ret =uprobe_register (tu->inode, tu->offset, &tu->consumer); - + ret =uprobe_register (tu->inode, tu->offset, tu->ref_ctr_offset, + &tu->consumer);
doesn't fit under 100 characters? If it does, please keep as a single line.
Quoted 7 lines
kernel/trace/trace_uprobe.c +0
-1if (ret) tu->inode =NULL ; --2.25.1.362.g51ebf55
Oleg Nesterov
maintainerRe: [PATCH 1/3] uprobes: kill uprobe_register_refctr()
Quoted 4 lines · Andrii Nakryiko
LGTM with few nits below.
Acked-by Andrii Nakryiko <[email protected]>
Thanks for looking at this.
Quoted 11 lines
kernel/trace/bpf_trace.c int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr · line 3483
Quoted 9 lines
kernel/trace/bpf_trace.c +1
-1@@ -3477,7 +3477,7 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register_refctr (d_real_inode (link->path.dentry), + err =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer);please adjust indentation here
OK,
Quoted 10 lines
kernel/trace/trace_uprobe.c static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter) · line 1093
Quoted 8 lines
kernel/trace/trace_uprobe.c +2
-6-if (tu->ref_ctr_offset) - ret =uprobe_register_refctr (tu->inode, tu->offset, - tu->ref_ctr_offset, &tu->consumer); -else - ret =uprobe_register (tu->inode, tu->offset, &tu->consumer); - + ret =uprobe_register (tu->inode, tu->offset, tu->ref_ctr_offset, + &tu->consumer);doesn't fit under 100 characters? If it does, please keep as a single line.
OK, will do.
Oleg.
Oleg Nesterov
maintainerPATCH 2/3 uprobes: simplify error handling for alloc_uprobe()
In reply to Oleg NesterovFrom: Andrii Nakryiko <[email protected]>
Return -ENOMEM instead of NULL, which makes caller's error handling just a touch simpler.
--- kernel/events/uprobes.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
kernel/events/uprobes.c +1 -3
@@ -725,7 +725,7 @@ static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset, uprobe =kzalloc (sizeof (struct uprobe ),GFP_KERNEL );if (!uprobe) -return NULL ; +return ERR_PTR (-ENOMEM ); uprobe->inode = inode; uprobe->offset = offset;
@@ -1166,8 +1166,6 @@ int uprobe_register(struct inode *inode, loff_t offset, loff_t ref_ctr_offset, retry: uprobe =alloc_uprobe (inode, offset, ref_ctr_offset); -if (!uprobe) -return -ENOMEM ;if (IS_ERR (uprobe))return PTR_ERR (uprobe);
Signature · 1 line
2.25.1.362.g51ebf55
Re: [PATCH 2/3] uprobes: simplify error handling for alloc_uprobe()
Quoted 7 lines · On Wed, 10 Jul 2024 18:31:11 +0200 Oleg Nesterov <[email protected]> wrote:
From: Andrii Nakryiko <[email protected]>
Return
-ENOMEMinstead of NULL, which makes caller's error handling just a touch simpler.Signed-off-by Andrii Nakryiko <[email protected]>Signed-off-by Oleg Nesterov <[email protected]>
Looks good to me.
Quoted 30 lines
--- kernel/events/uprobes.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)kernel/events/uprobes.c +1
-3@@ -725,7 +725,7 @@ static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset, uprobe =kzalloc (sizeof (struct uprobe ),GFP_KERNEL );if (!uprobe) -return NULL ; +return ERR_PTR (-ENOMEM ); uprobe->inode = inode; uprobe->offset = offset;@@ -1166,8 +1166,6 @@ int uprobe_register(struct inode *inode, loff_t offset, loff_t ref_ctr_offset, retry: uprobe =alloc_uprobe (inode, offset, ref_ctr_offset); -if (!uprobe) -return -ENOMEM ;if (IS_ERR (uprobe))return PTR_ERR (uprobe);Signature · 1 line
2.25.1.362.g51ebf55
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Oleg Nesterov
maintainerPATCH 3/3 uprobes: make uprobe_register() return struct uprobe *
In reply to Oleg NesterovThis way uprobe_unregister() and uprobe_apply() do not need find_uprobe() + put_uprobe(). And to me this change simplifies the code a bit.
--- include/linux/uprobes.h | 14 ++++++------ kernel/events/uprobes.c | 45 ++++++++++++------------------------- kernel/trace/bpf_trace.c | 12 +++++----- kernel/trace/trace_uprobe.c | 28 +++++++++++------------ 4 files changed, 41 insertions(+), 58 deletions(-)
include/linux/uprobes.h +7 -7
@@ -110,9 +110,9 @@ extern bool is_trap_insn(uprobe_opcode_t *insn);extern unsigned long uprobe_get_swbp_addr (struct pt_regs *regs);extern unsigned long uprobe_get_trap_addr (struct pt_regs *regs);extern int uprobe_write_opcode (struct arch_uprobe *auprobe,struct mm_struct *mm,unsigned long vaddr,uprobe_opcode_t ); -extern int uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc); -extern int uprobe_apply (struct inode *inode,loff_t offset,struct uprobe_consumer *uc,bool ); -extern void uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc); +extern struct uprobe *uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc); +extern int uprobe_apply (struct uprobe *uprobe,struct uprobe_consumer *uc,bool ); +extern void uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc);extern int uprobe_mmap (struct vm_area_struct *vma);extern void uprobe_munmap (struct vm_area_struct *vma,unsigned long start,unsigned long end);extern void uprobe_start_dup_mmap (void );
@@ -147,18 +147,18 @@ static inline void uprobes_init(void)#define uprobe_get_trap_addr (regs) instruction_pointer(regs) -static inline int +static inline struct uprobe *uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc) { -return -ENOSYS ; +return ERR_PTR (-ENOSYS ); }static inline int -uprobe_apply (struct inode *inode,loff_t offset,struct uprobe_consumer *uc,bool add) +uprobe_apply (struct uprobe * uprobe,struct uprobe_consumer *uc,bool add) {return -ENOSYS ; }static inline void -uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc) +uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc) { }static inline int uprobe_mmap (struct vm_area_struct *vma)
kernel/events/uprobes.c +14 -31
112 lines not on this page · raw message
kernel/trace/bpf_trace.c +6 -6
@@ -3157,6 +3157,7 @@ struct bpf_uprobe {loff_t offset;unsigned long ref_ctr_offset;u64 cookie; +struct uprobe *uprobe;struct uprobe_consumer consumer; };
@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i; -for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - } +for (i =0 ; i < cnt; i++) +uprobe_unregister (uprobes[i].uprobe, &uprobes[i].consumer); }static void bpf_uprobe_multi_link_release (struct bpf_link *link)
@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register (d_real_inode (link->path.dentry), + uprobes[i].uprobe =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer); -if (err) { +if (IS_ERR (uprobes[i].uprobe)) { + err =PTR_ERR (uprobes[i].uprobe);bpf_uprobe_unregister (&path, uprobes, i); goto error_free; }
kernel/trace/trace_uprobe.c +14 -14
@@ -58,8 +58,8 @@ struct trace_uprobe {struct dyn_event devent;struct uprobe_consumer consumer;struct path path; -struct inode *inode;char *filename; +struct uprobe *uprobe;unsigned long offset;unsigned long ref_ctr_offset;unsigned long nhit;
@@ -1084,17 +1084,17 @@ typedef bool (*filter_func_t)(struct uprobe_consumer *self,static int trace_uprobe_enable (struct trace_uprobe *tu,filter_func_t filter) { -int ret; +struct inode *inode =d_real_inode (tu->path.dentry); +struct uprobe *uprobe; tu->consumer.filter = filter; - tu->inode =d_real_inode (tu->path.dentry); - - ret =uprobe_register (tu->inode, tu->offset, tu->ref_ctr_offset, - &tu->consumer); -if (ret) - tu->inode =NULL ; + uprobe =uprobe_register (inode, tu->offset, tu->ref_ctr_offset, + &tu->consumer); +if (IS_ERR (uprobe)) +return PTR_ERR (uprobe); -return ret; + tu->uprobe = uprobe; +return 0 ; }static void __probe_event_disable (struct trace_probe *tp)
@@ -1105,11 +1105,11 @@ static void __probe_event_disable(struct trace_probe *tp)WARN_ON (!uprobe_filter_is_empty (tu->tp.event->filter));list_for_each_entry (tu,trace_probe_probe_list (tp), tp.list) { -if (!tu->inode) +if (!tu->uprobe)continue ; -uprobe_unregister (tu->inode, tu->offset, &tu->consumer); - tu->inode =NULL ; +uprobe_unregister (tu->uprobe, &tu->consumer); + tu->uprobe =NULL ; } }
@@ -1306,7 +1306,7 @@ static int uprobe_perf_close(struct trace_event_call *call,return 0 ;list_for_each_entry (tu,trace_probe_probe_list (tp), tp.list) { - ret =uprobe_apply (tu->inode, tu->offset, &tu->consumer, false); + ret =uprobe_apply (tu->uprobe, &tu->consumer, false);if (ret)break ; }
@@ -1330,7 +1330,7 @@ static int uprobe_perf_open(struct trace_event_call *call,return 0 ;list_for_each_entry (tu,trace_probe_probe_list (tp), tp.list) { - err =uprobe_apply (tu->inode, tu->offset, &tu->consumer, true); + err =uprobe_apply (tu->uprobe, &tu->consumer, true);if (err) {uprobe_perf_close (call, event);break ;
Signature · 1 line
2.25.1.362.g51ebf55
Re: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
On Wed, Jul 10, 2024 at 06:31:33PM +0200, Oleg Nesterov wrote:
SNIP
Quoted 13 lines
kernel/trace/bpf_trace.c +1
-0@@ -3157,6 +3157,7 @@ struct bpf_uprobe {loff_t offset;unsigned long ref_ctr_offset;u64 cookie; +struct uprobe *uprobe;struct uprobe_consumer consumer; };
kernel/trace/bpf_trace.c static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3186
Quoted 7 lines
kernel/trace/bpf_trace.c +0
-4@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i; -for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - }
nice, we could also drop path argument now
jirka
Quoted 20 lines
kernel/trace/bpf_trace.c +5
-2+for (i =0 ; i < cnt; i++) +uprobe_unregister (uprobes[i].uprobe, &uprobes[i].consumer); }static void bpf_uprobe_multi_link_release (struct bpf_link *link)@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register (d_real_inode (link->path.dentry), + uprobes[i].uprobe =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer); -if (err) { +if (IS_ERR (uprobes[i].uprobe)) { + err =PTR_ERR (uprobes[i].uprobe);bpf_uprobe_unregister (&path, uprobes, i); goto error_free; }
Re: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 21 lines · Jiri Olsa
On Wed, Jul 10, 2024 at 06:31:33PM +0200, Oleg Nesterov wrote:
SNIP
Quoted 16 lines
kernel/trace/bpf_trace.c +1
-0@@ -3157,6 +3157,7 @@ struct bpf_uprobe {loff_t offset;unsigned long ref_ctr_offset;u64 cookie; +struct uprobe *uprobe;struct uprobe_consumer consumer; };@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i;
Quoted 6 lines
kernel/trace/bpf_trace.c static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3186
Quoted 4 lines
kernel/trace/bpf_trace.c +0
-4-for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - }nice, we could also drop path argument now
see my comments to Oleg, I think we can/should get rid of link->path altogether if uprobe itself keeps inode alive.
BTW, Jiri, do we have any test for multi-uprobe that simulates partial attachment success/failure (whichever way you want to look at it). It would be super useful to have to check at least some error handling code in the uprobe code base. If we don't, do you mind adding something simple to BPF selftests?
Quoted 23 lines
jirka
Quoted 20 lines
kernel/trace/bpf_trace.c +5
-2+for (i =0 ; i < cnt; i++) +uprobe_unregister (uprobes[i].uprobe, &uprobes[i].consumer); }static void bpf_uprobe_multi_link_release (struct bpf_link *link)@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register (d_real_inode (link->path.dentry), + uprobes[i].uprobe =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer); -if (err) { +if (IS_ERR (uprobes[i].uprobe)) { + err =PTR_ERR (uprobes[i].uprobe);bpf_uprobe_unregister (&path, uprobes, i); goto error_free; }
Re: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 29 lines · Andrii Nakryiko
Quoted 21 lines · Jiri Olsa
On Wed, Jul 10, 2024 at 06:31:33PM +0200, Oleg Nesterov wrote:
SNIP
Quoted 16 lines
kernel/trace/bpf_trace.c +1
-0@@ -3157,6 +3157,7 @@ struct bpf_uprobe {loff_t offset;unsigned long ref_ctr_offset;u64 cookie; +struct uprobe *uprobe;struct uprobe_consumer consumer; };@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i;Quoted 6 lines
kernel/trace/bpf_trace.c static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3186
Quoted 4 lines
kernel/trace/bpf_trace.c +0
-4-for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - }nice, we could also drop path argument now
Quoted 2 lines
see my comments to Oleg, I think we can/should get rid of
link->pathaltogether if uprobe itself keeps inode alive.
yea, I was thinking of that, but then it's kind of useful to have it in bpf_uprobe_multi_link_fill_link_info, otherwise we have to take it from first uprobe in the link, but ok, still probably worth to remove it ;-)
anyway as you wrote it's ok for follow up cleanup, I'll check on that
Quoted 6 lines
BTW, Jiri, do we have any test for multi-uprobe that simulates partial attachment success/failure (whichever way you want to look at it). It would be super useful to have to check at least some error handling code in the uprobe code base. If we don't, do you mind adding something simple to BPF selftests?
there's test_attach_api_fails, but I think all checked fails are before actually calling uprobe_register function
I think there are few ways to fail the uprobe_register, like install it on top of int3.. will check add some test for that
jirka
Quoted 24 lines
Quoted 24 lines
jirka
Quoted 20 lines
kernel/trace/bpf_trace.c +5
-2+for (i =0 ; i < cnt; i++) +uprobe_unregister (uprobes[i].uprobe, &uprobes[i].consumer); }static void bpf_uprobe_multi_link_release (struct bpf_link *link)@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register (d_real_inode (link->path.dentry), + uprobes[i].uprobe =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer); -if (err) { +if (IS_ERR (uprobes[i].uprobe)) { + err =PTR_ERR (uprobes[i].uprobe);bpf_uprobe_unregister (&path, uprobes, i); goto error_free; }
Re: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 31 lines · Jiri Olsa
Quoted 29 lines · Andrii Nakryiko
Quoted 27 lines · Jiri Olsa
On Wed, Jul 10, 2024 at 06:31:33PM +0200, Oleg Nesterov wrote:
SNIP
Quoted 13 lines
kernel/trace/bpf_trace.c +1
-0@@ -3157,6 +3157,7 @@ struct bpf_uprobe {loff_t offset;unsigned long ref_ctr_offset;u64 cookie; +struct uprobe *uprobe;struct uprobe_consumer consumer; };kernel/trace/bpf_trace.c static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3186
Quoted 7 lines
kernel/trace/bpf_trace.c +0
-4@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i; -for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - }nice, we could also drop path argument now
Quoted 6 lines
Quoted 2 lines
see my comments to Oleg, I think we can/should get rid of
link->pathaltogether if uprobe itself keeps inode alive.yea, I was thinking of that, but then it's kind of useful to have it in
bpf_uprobe_multi_link_fill_link_info, otherwise we have to take it from first uprobe in the link, but ok, still probably worth to remove it ;-)
if we need it for link_info, probably cleaner to just keep it, no big deal then
Quoted 10 lines
anyway as you wrote it's ok for follow up cleanup, I'll check on that
Quoted 6 lines
BTW, Jiri, do we have any test for multi-uprobe that simulates partial attachment success/failure (whichever way you want to look at it). It would be super useful to have to check at least some error handling code in the uprobe code base. If we don't, do you mind adding something simple to BPF selftests?
Quoted 6 lines
there's
test_attach_api_fails, but I think all checked fails are before actually callinguprobe_registerfunctionI think there are few ways to fail the
uprobe_register, like install it on top of int3.. will check add some test for that
great, thank you!
Quoted 26 lines
jirka
Quoted 24 lines
Quoted 24 lines
jirka
Quoted 20 lines
kernel/trace/bpf_trace.c +5
-2+for (i =0 ; i < cnt; i++) +uprobe_unregister (uprobes[i].uprobe, &uprobes[i].consumer); }static void bpf_uprobe_multi_link_release (struct bpf_link *link)@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err =uprobe_register (d_real_inode (link->path.dentry), + uprobes[i].uprobe =uprobe_register (d_real_inode (link->path.dentry), uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer); -if (err) { +if (IS_ERR (uprobes[i].uprobe)) { + err =PTR_ERR (uprobes[i].uprobe);bpf_uprobe_unregister (&path, uprobes, i); goto error_free; }
Oleg Nesterov
maintainerRe: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
In reply to Jiri OlsaQuoted 11 lines · Jiri Olsa
kernel/trace/bpf_trace.c static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3186
Quoted 9 lines
kernel/trace/bpf_trace.c +0
-4@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i; -for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - }nice, we could also drop path argument now
Indeed, I missed that. Will send V2 when/if this makes any sense.
Thanks!
Oleg.
Re: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
In reply to Oleg Nesterovinclude/linux/uprobes.h
Quoted 16 lines · Oleg Nesterov
This way
uprobe_unregister()anduprobe_apply()do not needfind_uprobe()+put_uprobe(). And to me this change simplifies the code a bit.Signed-off-by Oleg Nesterov <[email protected]>--- include/linux/uprobes.h | 14 ++++++------ kernel/events/uprobes.c | 45 ++++++++++++------------------------- kernel/trace/bpf_trace.c | 12 +++++----- kernel/trace/trace_uprobe.c | 28 +++++++++++------------ 4 files changed, 41 insertions(+), 58 deletions(-)include/linux/uprobes.h +0
-0
I don't see struct uprobe forward-declared in this header, maybe we should add it?
Quoted 20 lines
include/linux/uprobes.h +4
-4@@ -110,9 +110,9 @@ extern bool is_trap_insn(uprobe_opcode_t *insn);extern unsigned long uprobe_get_swbp_addr (struct pt_regs *regs);extern unsigned long uprobe_get_trap_addr (struct pt_regs *regs);extern int uprobe_write_opcode (struct arch_uprobe *auprobe,struct mm_struct *mm,unsigned long vaddr,uprobe_opcode_t ); -extern int uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc); -extern int uprobe_apply (struct inode *inode,loff_t offset,struct uprobe_consumer *uc,bool ); -extern void uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc); +extern struct uprobe *uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc); +extern int uprobe_apply (struct uprobe *uprobe,struct uprobe_consumer *uc,bool ); +extern void uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc);extern int uprobe_mmap (struct vm_area_struct *vma);extern void uprobe_munmap (struct vm_area_struct *vma,unsigned long start,unsigned long end);extern void uprobe_start_dup_mmap (void );@@ -147,18 +147,18 @@ static inline void uprobes_init(void)#define uprobe_get_trap_addr (regs) instruction_pointer(regs) -static inline int +static inline struct uprobe *uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset,struct uprobe_consumer *uc)
include/linux/uprobes.h static inline void uprobes_init(void) · line 159
Quoted 10 lines
include/linux/uprobes.h +2
-2{ -return -ENOSYS ; +return ERR_PTR (-ENOSYS ); }static inline int -uprobe_apply (struct inode *inode,loff_t offset,struct uprobe_consumer *uc,bool add) +uprobe_apply (struct uprobe * uprobe,struct uprobe_consumer *uc,bool add) {return -ENOSYS ; }
complete aside, when I was looking at this code I was wondering why we even need uprobe_apply, it looks like some hacky variant of uprobe_register and uprobe_unregister. I didn't dig deeper, but think whether we even need this? If this is just to avoid (for some period) some consumer callback calling, then that could be handled at the consumer side by ignoring such calls.
callback call is cheap, it's the int3 handling that's expensive and with uprobe_apply we are already paying it anyways, so what is this for?
Quoted 6 lines
include/linux/uprobes.h +1
-1static inline void -uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc) +uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc) { }static inline int uprobe_mmap (struct vm_area_struct *vma)
[...]
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1138
Quoted 8 lines
include/linux/uprobes.h +0
-3@@ -1133,41 +1126,39 @@ EXPORT_SYMBOL_GPL(uprobe_unregister);* refcount is released when the last @uc for the @uprobe * unregisters. Caller of uprobe_register() is required to keep @inode * (and the containing mount) referenced. -* -* Return errno if it cannot successully install probes -* else return 0 (success)
mention that it never returns NULL, but rather encodes error code inside the pointer on error? It's an important part of the contract.
Quoted 7 lines
include/linux/uprobes.h +3
-2*/ -int uprobe_register (struct inode *inode,loff_t offset,loff_t ref_ctr_offset, -struct uprobe_consumer *uc) +struct uprobe *uprobe_register (struct inode *inode, +loff_t offset,loff_t ref_ctr_offset, +struct uprobe_consumer *uc) {
[...]
include/linux/uprobes.h int uprobe_register(struct inode *inode, loff_t offset, loff_t ref_ctr_offset, · line 1185
Quoted 11 lines
include/linux/uprobes.h +2
-1@@ -1186,35 +1177,27 @@ int uprobe_register(struct inode *inode, loff_t offset, loff_t ref_ctr_offset,if (unlikely (ret == -EAGAIN )) goto retry; -return ret; + +return ret ?ERR_PTR (ret) : uprobe; }EXPORT_SYMBOL_GPL (uprobe_register);/*
this should be /** for doccomment checking (you'd get a warning for missing @uprobe if there was this extra star)
include/linux/uprobes.h int uprobe_register(struct inode *inode, loff_t offset, loff_t ref_ctr_offset, · line 1196
Quoted 3 lines
include/linux/uprobes.h +0
-2* uprobe_apply - unregister an already registered probe. -* @inode: the file in which the probe has to be removed. -* @offset: offset from the start of the file.
add @uprobe description now?
Quoted 11 lines
include/linux/uprobes.h +1
-3* @uc: consumer which wants to add more or remove some breakpoints * @add: add or remove the breakpoints */ -int uprobe_apply (struct inode *inode,loff_t offset, -struct uprobe_consumer *uc,bool add) +int uprobe_apply (struct uprobe *uprobe,struct uprobe_consumer *uc,bool add) { -struct uprobe *uprobe;struct uprobe_consumer *con;int ret = -ENOENT ;
[...]
include/linux/uprobes.h static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3184
Quoted 9 lines
include/linux/uprobes.h +1
-4@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i; -for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - } +for (i =0 ; i < cnt; i++)
you'll now need !IS_ERR_OR_NULL(uprobes[i].uprobe) check (or just NULL check if you null-out it below)
include/linux/uprobes.h int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr · line 3479
Quoted 10 lines
include/linux/uprobes.h +2
-1+uprobe_unregister (uprobes[i].uprobe, &uprobes[i].consumer); }static void bpf_uprobe_multi_link_release (struct bpf_link *link)@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err = uprobe_register(d_real_inode(link->path.dentry), + uprobes[i].uprobe = uprobe_register(d_real_inode(link->path.dentry),
will uprobe keep inode alive as long as uprobe is attached? If that's the case we can get rid of link->path (have it only as a local variable which we put as soon as we are done with registration). We can probably do that clean up separately, I'll defer to Jiri.
include/linux/uprobes.h int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr · line 3484
Quoted 6 lines
include/linux/uprobes.h +2
-1uprobes[i].offset, uprobes[i].ref_ctr_offset, &uprobes[i].consumer); -if (err) { +if (IS_ERR (uprobes[i].uprobe)) { + err =PTR_ERR (uprobes[i].uprobe);
we can NULL-out uprobe on error for bpf_uprobe_unregister() to handle only NULL vs non-NULL case
or maybe better yet let's just have local struct uprobe variable and only assign it if registration succeeded (still need NULL check in bpf_uprobe_unregister above)
Quoted 3 lines
include/linux/uprobes.h +0
-0bpf_uprobe_unregister (&path, uprobes, i); goto error_free; }
[...]
Oleg Nesterov
maintainerRe: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 14 lines · Andrii Nakryiko
Quoted 12 lines · Oleg Nesterov
This way
uprobe_unregister()anduprobe_apply()do not needfind_uprobe()+put_uprobe(). And to me this change simplifies the code a bit.Signed-off-by Oleg Nesterov <[email protected]>--- include/linux/uprobes.h | 14 ++++++------ kernel/events/uprobes.c | 45 ++++++++++++------------------------- kernel/trace/bpf_trace.c | 12 +++++----- kernel/trace/trace_uprobe.c | 28 +++++++++++------------ 4 files changed, 41 insertions(+), 58 deletions(-)
Quoted 7 lines
include/linux/uprobes.h
Quoted 4 lines
include/linux/uprobes.h +0
-0I don't see struct uprobe forward-declared in this header, maybe we should add it?
Probably yes, thanks... Although the current code already uses struct uprobes * without forward-declaration at least if CONFIG_UPROBES=y. Thanks, will add.
Quoted 10 lines
include/linux/uprobes.h static inline void uprobes_init(void) · line 159
Quoted 6 lines
static inline int
include/linux/uprobes.h +1
-1-uprobe_apply(struct inode *inode, loff_t offset, struct uprobe_consumer *uc, bool add) +uprobe_apply(struct uprobe* uprobe, struct uprobe_consumer *uc, bool add) { return -ENOSYS; }complete aside, when I was looking at this code I was wondering why we even need
uprobe_apply, it looks like some hacky variant ofuprobe_registeranduprobe_unregister.
All I can say is that
- I can hardly recall this logic, I'll try to do this tomorrow and write another email
- in any case this logic is ugly and needs more cleanups
- but this patch only tries to simplify this code without any visible changes.
Quoted 10 lines
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1138
Quoted 7 lines
include/linux/uprobes.h +0
-3@@ -1133,41 +1126,39 @@ EXPORT_SYMBOL_GPL(uprobe_unregister);* refcount is released when the last @uc for the @uprobe * unregisters. Caller of uprobe_register() is required to keep @inode * (and the containing mount) referenced. -* -* Return errno if it cannot successully install probes -* else return 0 (success) mention that it never returns NULL, but rather encodes error code inside the pointer on error? It's an important part of the contract.
OK...
Quoted 4 lines
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1129
Quoted 1 line
include/linux/uprobes.h +0
-0/* this should be /** for doccomment checking (you'd get a warning for missing @uprobe if there was this extra star)
Well, this is what we have before this patch, but OK
Quoted 5 lines
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1142
Quoted 3 lines
include/linux/uprobes.h +0
-2* uprobe_apply - unregister an already registered probe. -* @inode: the file in which the probe has to be removed. -* @offset: offset from the start of the file. add @uprobe description now?
If only I knew what this @uprobe description can say ;)
Quoted 4 lines
Quoted 4 lines
include/linux/uprobes.h +0
-0@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i;
Quoted 8 lines
include/linux/uprobes.h static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3184
Quoted 5 lines
include/linux/uprobes.h +1
-4-for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - } +for (i =0 ; i < cnt; i++)you'll now need !IS_ERR_OR_NULL(uprobes[i].uprobe) check (or just NULL check if you null-out it below)
Hmm... are you sure? I'll re-check... See also the end of my email.
Quoted 8 lines
include/linux/uprobes.h int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr · line 3479
Quoted 6 lines
include/linux/uprobes.h +1
-1@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err = uprobe_register(d_real_inode(link->path.dentry), + uprobes[i].uprobe = uprobe_register(d_real_inode(link->path.dentry),will uprobe keep inode alive as long as uprobe is attached?
Why? No, this patch doesn't (shouldn't) change the current behaviour.
The users still need kern_path/path_put to, say, prevent umount.
Quoted 2 lines
we can NULL-out uprobe on error for
bpf_uprobe_unregister()to handle only NULL vs non-NULL case
Not sure I understand...
Quoted 2 lines
or maybe better yet let's just have local struct uprobe variable and only assign it if registration succeeded
We can, but
Quoted 2 lines
(still need NULL check in
bpf_uprobe_unregisterabove)
Why?
If bpf_uprobe_unregister() is called by bpf_uprobe_multi_link_attach() on error, it passes cnt = i where is the 1st failed uprobe index. IOW, uptobes[i].uprobe can't be IS_ERR_OR_NULL() in the 0..cnt-1 range.
If it is called by bpf_uprobe_multi_link_release() then the whole 0..umulti_link->cnt-1 range should be fine?
Oleg.
Andrii Nakryiko
Acked-byRe: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 24 lines · Oleg Nesterov
Quoted 14 lines · Andrii Nakryiko
Quoted 12 lines · Oleg Nesterov
This way
uprobe_unregister()anduprobe_apply()do not needfind_uprobe()+put_uprobe(). And to me this change simplifies the code a bit.Signed-off-by Oleg Nesterov <[email protected]>--- include/linux/uprobes.h | 14 ++++++------ kernel/events/uprobes.c | 45 ++++++++++++------------------------- kernel/trace/bpf_trace.c | 12 +++++----- kernel/trace/trace_uprobe.c | 28 +++++++++++------------ 4 files changed, 41 insertions(+), 58 deletions(-)Quoted 7 lines
include/linux/uprobes.h
Quoted 4 lines
include/linux/uprobes.h +0
-0I don't see struct uprobe forward-declared in this header, maybe we should add it?
Quoted 4 lines
Probably yes, thanks... Although the current code already uses struct uprobes * without forward-declaration at least if
CONFIG_UPROBES=y. Thanks, will add.
Yep, I saw that and was wondering as well.
Quoted 13 lines
Quoted 10 lines
include/linux/uprobes.h static inline void uprobes_init(void) · line 159
Quoted 6 lines
static inline int
include/linux/uprobes.h +1
-1-uprobe_apply(struct inode *inode, loff_t offset, struct uprobe_consumer *uc, bool add) +uprobe_apply(struct uprobe* uprobe, struct uprobe_consumer *uc, bool add) { return -ENOSYS; }complete aside, when I was looking at this code I was wondering why we even need
uprobe_apply, it looks like some hacky variant ofuprobe_registeranduprobe_unregister.All I can say is that
Quoted 7 lines
- I can hardly recall this logic, I'll try to do this tomorrow and write another email
- in any case this logic is ugly and needs more cleanups
- but this patch only tries to simplify this code without any visible changes.
yep, that's why it's an aside, up to you
Quoted 21 lines
Quoted 11 lines
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1138
Quoted 8 lines
include/linux/uprobes.h +0
-3@@ -1133,41 +1126,39 @@ EXPORT_SYMBOL_GPL(uprobe_unregister);* refcount is released when the last @uc for the @uprobe * unregisters. Caller of uprobe_register() is required to keep @inode * (and the containing mount) referenced. -* -* Return errno if it cannot successully install probes -* else return 0 (success) mention that it never returns NULL, but rather encodes error code inside the pointer on error? It's an important part of the contract.
OK...
Quoted 4 lines
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1129
Quoted 1 line
include/linux/uprobes.h +0
-0/* this should be /** for doccomment checking (you'd get a warning for missing @uprobe if there was this extra star)
Well, this is what we have before this patch, but OK
Quoted 7 lines
Quoted 5 lines
include/linux/uprobes.h EXPORT_SYMBOL_GPL(uprobe_unregister); · line 1142
Quoted 3 lines
include/linux/uprobes.h +0
-2* uprobe_apply - unregister an already registered probe. -* @inode: the file in which the probe has to be removed. -* @offset: offset from the start of the file. add @uprobe description now?
If only I knew what this @uprobe description can say ;)
I'm pointing this out because I accidentally used /** for comment for some function, and I got some bot report about missing argument. I think /** makes sense for documenting "public API" function, so which is why all the above.
Quoted 14 lines
Quoted 5 lines
Quoted 5 lines
include/linux/uprobes.h +0
-0@@ -3180,10 +3181,8 @@ static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, {u32 i;Quoted 8 lines
include/linux/uprobes.h static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes, · line 3184
Quoted 5 lines
include/linux/uprobes.h +1
-4-for (i =0 ; i < cnt; i++) { -uprobe_unregister (d_real_inode (path->dentry), uprobes[i].offset, - &uprobes[i].consumer); - } +for (i =0 ; i < cnt; i++)you'll now need !IS_ERR_OR_NULL(uprobes[i].uprobe) check (or just NULL check if you null-out it below)
Quoted 1 line
Hmm... are you sure? I'll re-check... See also the end of my email.
no, you are right, it should be fine
Quoted 10 lines
Quoted 9 lines
include/linux/uprobes.h int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr · line 3479
Quoted 7 lines
include/linux/uprobes.h +1
-1@@ -3477,11 +3476,12 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr &bpf_uprobe_multi_link_lops, prog);for (i =0 ; i < cnt; i++) { - err = uprobe_register(d_real_inode(link->path.dentry), + uprobes[i].uprobe = uprobe_register(d_real_inode(link->path.dentry),will uprobe keep inode alive as long as uprobe is attached?
Quoted 3 lines
Why? No, this patch doesn't (shouldn't) change the current behaviour.
The users still need kern_path/path_put to, say, prevent umount.
ok, then link->path has to stay, I was just wondering if we need to keep it alive
Quoted 14 lines
Quoted 3 lines
we can NULL-out uprobe on error for
bpf_uprobe_unregister()to handle only NULL vs non-NULL caseNot sure I understand...
Quoted 2 lines
or maybe better yet let's just have local struct uprobe variable and only assign it if registration succeeded
We can, but
Quoted 2 lines
(still need NULL check in
bpf_uprobe_unregisterabove)
Quoted 8 lines
Why?
If
bpf_uprobe_unregister()is called bybpf_uprobe_multi_link_attach()on error, it passes cnt = i where is the 1st failed uprobe index. IOW, uptobes[i].uprobe can't beIS_ERR_OR_NULL()in the 0..cnt-1 range.If it is called by
bpf_uprobe_multi_link_release()then the whole 0..umulti_link->cnt-1 range should be fine?
yes, you are absolutely right, I missed that for this partial failure case we pass i as count into bpf_uprobe_unregister(), sorry about the noise!
please feel free to add my ack for the next revision:
Quoted 3 lines
Oleg.
Oleg Nesterov
maintainerRe: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
In reply to Oleg NesterovQuoted 4 lines · Oleg Nesterov
kernel/events/uprobes.c +1
-1-void uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc) +void uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc) {
kernel/events/uprobes.c __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) · line 1118
Quoted 10 lines
kernel/events/uprobes.c +0
-7-struct uprobe *uprobe; - - uprobe =find_uprobe (inode, offset); -if (WARN_ON (!uprobe)) -return ; -down_write (&uprobe->register_rwsem);__uprobe_unregister (uprobe, uc);up_write (&uprobe->register_rwsem); -put_uprobe (uprobe);
OK, this is obviously wrong, needs get_uprobe/put_uprobe. __uprobe_unregister() can free this uprobe, so up_write(&uprobe->register_rwsem) is not safe.
I'll send V2 on top of Peter's new version.
Oleg.
Re: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 19 lines · Oleg Nesterov
Quoted 4 lines · Oleg Nesterov
kernel/events/uprobes.c +1
-1-void uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc) +void uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc) {kernel/events/uprobes.c __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) · line 1118
Quoted 10 lines
kernel/events/uprobes.c +0
-7-struct uprobe *uprobe; - - uprobe =find_uprobe (inode, offset); -if (WARN_ON (!uprobe)) -return ; -down_write (&uprobe->register_rwsem);__uprobe_unregister (uprobe, uc);up_write (&uprobe->register_rwsem); -put_uprobe (uprobe);OK, this is obviously wrong, needs get_uprobe/put_uprobe.
__uprobe_unregister()can free this uprobe, soup_write(&uprobe->register_rwsem)is not safe.
uprobe_register(), given it returns an uprobe instance to the caller should keep refcount on it (it belongs to uprobe_consumer). That's what I did for my patches, are you going to do that as well?
We basically do the same thing, just interfaces look a bit different.
Quoted 5 lines
I'll send V2 on top of Peter's new version.
Oleg.
Oleg Nesterov
maintainerRe: [PATCH 3/3] uprobes: make uprobe_register() return struct uprobe *
Quoted 22 lines · Andrii Nakryiko
Quoted 19 lines · Oleg Nesterov
Quoted 4 lines · Oleg Nesterov
kernel/events/uprobes.c +1
-1-void uprobe_unregister (struct inode *inode,loff_t offset,struct uprobe_consumer *uc) +void uprobe_unregister (struct uprobe *uprobe,struct uprobe_consumer *uc) {kernel/events/uprobes.c __uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc) · line 1118
Quoted 10 lines
kernel/events/uprobes.c +0
-7-struct uprobe *uprobe; - - uprobe =find_uprobe (inode, offset); -if (WARN_ON (!uprobe)) -return ; -down_write (&uprobe->register_rwsem);__uprobe_unregister (uprobe, uc);up_write (&uprobe->register_rwsem); -put_uprobe (uprobe);OK, this is obviously wrong, needs get_uprobe/put_uprobe.
__uprobe_unregister()can free this uprobe, soup_write(&uprobe->register_rwsem)is not safe.
Quoted 2 lines
uprobe_register(), given it returns an uprobe instance to the caller should keep refcount on it (it belongs touprobe_consumer).
Of course. And again, this patch doesn't change the curent behaviour.
Quoted 4 lines
That's
what I did for my patches, are you going to do that as well?We basically do the same thing, just interfaces look a bit different.
Not sure. Well I do not really know, I didn't read your series to the end, sorry ;) The same for V1/V2 from Peter so far.
But let me say this just in case... With or without this change, currently uprobe_consumer doesn't have an "individual" ref to uprobe. The fact that uprobe->consumers != NULL adds a reference.
Lets not discuss if this is good or bad right now, this cleanup is only cleanup.
------------------------------------------------------------------------
Now, let me add another "just in case" note to explain what I am going to do in V2.
So. this patch should turn uprobe_unregister() into something like
void uprobe_unregister(struct uprobe *uprobe, struct uprobe_consumer *uc)
{
// Ugly !!!! please kill me!!!
get_uprobe(uprobe);
down_write(&uprobe->register_rwsem);
__uprobe_unregister(uprobe, uc);
up_write(&uprobe->register_rwsem);
put_uprobe(uprobe);
}to simplify this change. And the next (simple) patch will kill these get_uprobe + put_uprobe, we just need to shift the (possibly) final put_uprobe() from delete_uprobe() to unregister().
But of course, I will recheck before I send V2.
Oleg.
Peter Zijlstra
maintainerRe: [PATCH 0/3] uprobes: future cleanups for review
In reply to Oleg NesterovQuoted 5 lines · Oleg Nesterov
Quoted 3 lines · Oleg Nesterov
Peter, these simple cleanups should not conflict with your changes, but I can resend them later if it causes any inconvenience.
Quoted 8 lines
In fact I would like to push 2 more cleanups before the more significant changes, but they certainly conflict with your ongoing work, albeit only textually.
Let me send the patches for review anyway, perhaps you can take at least the 1st one.
3/3 is only compile tested so far. Andrii, can you take a look?
I was going to post a new version today, but I can wait and rebase on top of / include these 5 patches (or more, these things tend to grow).
Oleg Nesterov
maintainerRe: [PATCH 0/3] uprobes: future cleanups for review
Quoted 7 lines · Peter Zijlstra
Quoted 5 lines · Oleg Nesterov
In fact I would like to push 2 more cleanups before the more significant changes, but they certainly conflict with your ongoing work, albeit only textually.
Quoted 6 lines
Quoted 4 lines
Let me send the patches for review anyway, perhaps you can take at least the 1st one.
3/3 is only compile tested so far. Andrii, can you take a look?
I was going to post a new version today, but I can wait and rebase on
No, I do not want to delay your changes, please send the new version, I'll rebase these cleanups on top of it.
Oleg.