Checkpoint/restart of ptys, pgids, and controlling tty
Subsystems (5)
This post packs together the patches (6 for kernel, 3 for user) to provide checkpoint/restart support for ptys, pgids and controlling terminal. They apply on top of current ckpt-dev-v17 (kernel, user).
I'm not an expert on PTYs, TTYs, line disciplines and the like, comments are mostly welcome:
* What (additional) values we need to sanitize on restart ?
* What (additional) locking should be in place ? (especially checkpoint)
* Need to save/restore echo buffer (and position) ?
* /dev/ptmx and /dev/pts/... paths are currently hardcoded
* Other security concerns ? ...
Oren.
A message that is not in the archive (sent off-list or not yet received): 1252074054-22241-1-git-send-email-orenl-RdfvBDnrOixBDgjK7y7TUQ@public.gmane.org
PATCH 1/6 c/r: [objhash] add ckpt_obj_reserve()
Use ckpt_obj_reserve() during checkpoint to reserve an objref that will not be otherwise used for any other object in checkpoint. Such an objref can be useful for restart to add objects that were not there at checkpoint.
One use case for this is ptys (subsequent patches): when restoring a master pty, we create a file pointer, which must be kept somewhere and later cleaned up. The hash is a good place to store it.
--- checkpoint/objhash.c | 20 +++++++++++++++++++- include/linux/checkpoint.h | 1 + 2 files changed, 20 insertions(+), 1 deletions(-)
checkpoint/objhash.c +19 -1
@@ -465,6 +465,11 @@ static struct ckpt_obj *obj_find_by_objref(struct ckpt_ctx *ctx, int objref)return NULL ; } +static inline int obj_alloc_objref (struct ckpt_ctx *ctx) +{ +return ctx->obj_hash->next_free_objref++; +} +/** * ckpt_obj_new - add an object to the obj_hash * @ctx: checkpoint context
@@ -498,7 +503,7 @@ static struct ckpt_obj *obj_new(struct ckpt_ctx *ctx, void *ptr,if (!objref) {/* use @obj->ptr to index, assign objref (checkpoint) */ - obj->objref = ctx->obj_hash->next_free_objref++;; + obj->objref =obj_alloc_objref (ctx); i =hash_long ((unsigned long ) ptr,CKPT_OBJ_HASH_NBITS ); }else {/* use @obj->objref to index (restart) */
@@ -637,6 +642,19 @@ int ckpt_obj_lookup_add(struct ckpt_ctx *ctx, void *ptr, }/** +* ckpt_obj_reserve - reserve an objref +* @ctx: checkpoint context +* +* The reserved objref will not be used for subsequent objects. This +* gives an objref that can be safely used during restart without a +* matching object in checkpoint. [used during checkpoint]. +*/ +int ckpt_obj_reserve (struct ckpt_ctx *ctx) +{ +return obj_alloc_objref (ctx); +} + +/** * checkpoint_obj - if not already in hash, add object and checkpoint * @ctx: checkpoint context * @ptr: pointer to object
include/linux/checkpoint.h +1 -0
@@ -121,6 +121,7 @@ extern void *ckpt_obj_fetch(struct ckpt_ctx *ctx, int objref, enum obj_type type);extern int ckpt_obj_insert (struct ckpt_ctx *ctx,void *ptr,int objref,enum obj_type type); +extern int ckpt_obj_reserve (struct ckpt_ctx *ctx);extern void ckpt_ctx_get (struct ckpt_ctx *ctx);extern void ckpt_ctx_put (struct ckpt_ctx *ctx);
Signature · 1 line
1.6.0.4
PATCH 2/6 c/r: [pty 1/2] allow allocation of desired pty slave
In reply to a message not in the archiveDuring restart, we need to allocate pty slaves with the same identifiers as recorded during checkpoint. Modify the allocation code to allow an in-kernel caller to request a specific slave identifier.
For this, add a new field to task_struct - 'required_id'. It will hold the desired identifier when restoring a (master) pty.
The code in ptmx_open() will use this value only for tasks that try to open /dev/ptmx that are restarting (PF_RESTARTING), and if the value isn't CKPT_REQUIRED_NONE (-1).
--- checkpoint/sys.c | 7 +++++++ drivers/char/pty.c | 9 +++++++-- fs/devpts/inode.c | 10 ++++++++-- include/linux/checkpoint.h | 2 ++ include/linux/devpts_fs.h | 2 +- include/linux/sched.h | 1 + 6 files changed, 26 insertions(+), 5 deletions(-)
checkpoint/sys.c +7 -0
@@ -339,6 +339,13 @@ SYSCALL_DEFINE3(restart, pid_t, pid, int, fd, unsigned long, flags)return ret; } +static int __init checkpoint_init (void ) +{ + init_task.required_id =CKPT_REQUIRED_NONE ; +return 0 ; +} +__initcall (checkpoint_init); +/* 'ckpt_debug_level' controls the verbosity level of c/r code */ #ifdef CONFIG_CHECKPOINT_DEBUG
drivers/char/pty.c +7 -2
@@ -633,12 +633,17 @@ static int __ptmx_open(struct inode *inode, struct file *filp) {struct tty_struct *tty;int retval; -int index; +int index =0 ;nonseekable_open (inode, filp); +#ifdef CONFIG_CHECKPOINT +/* when restarting, request specific index */ +if (current->flags &PF_RESTARTING ) + index = (int ) current->required_id; +#endif /* find a device that is not in use. */ - index =devpts_new_index (inode); + index =devpts_new_index (inode, index);if (index <0 )return index;
fs/devpts/inode.c +8 -2
@@ -433,7 +433,7 @@ static struct file_system_type devpts_fs_type = {* to the System V naming convention */ -int devpts_new_index (struct inode *ptmx_inode) +int devpts_new_index (struct inode *ptmx_inode,int req_idx) {struct super_block *sb =pts_sb_from_inode (ptmx_inode);struct pts_fs_info *fsi =DEVPTS_SB (sb);
@@ -445,7 +445,8 @@ retry:return -ENOMEM ;mutex_lock (&allocated_ptys_lock); - ida_ret =ida_get_new (&fsi->allocated_ptys, &index); + index = (req_idx >=0 ? req_idx :0 ); + ida_ret =ida_get_new_above (&fsi->allocated_ptys, index, &index);if (ida_ret <0 ) {mutex_unlock (&allocated_ptys_lock);if (ida_ret == -EAGAIN )
@@ -453,6 +454,11 @@ retry:return -EIO ; } +if (req_idx >=0 && index != req_idx) { +ida_remove (&fsi->allocated_ptys, index); +mutex_unlock (&allocated_ptys_lock); +return -EBUSY ; + }if (index >= pty_limit) {ida_remove (&fsi->allocated_ptys, index);mutex_unlock (&allocated_ptys_lock);
include/linux/checkpoint.h +2 -0
@@ -46,6 +46,8 @@#define CHECKPOINT_USER_FLAGS CHECKPOINT_SUBTREE#define RESTART_USER_FLAGS (RESTART_TASKSELF | RESTART_FROZEN) +#define CKPT_REQUIRED_NONE ((unsigned long) -1L) +extern void exit_checkpoint (struct task_struct *tsk);extern int ckpt_kwrite (struct ckpt_ctx *ctx,void *buf,int count);
include/linux/devpts_fs.h +1 -1
@@ -17,7 +17,7 @@#ifdef CONFIG_UNIX98_PTYS -int devpts_new_index (struct inode *ptmx_inode); +int devpts_new_index (struct inode *ptmx_inode,int req_idx);void devpts_kill_index (struct inode *ptmx_inode,int idx);/* mknod in devpts */ int devpts_pty_new (struct inode *ptmx_inode,struct tty_struct *tty);
include/linux/sched.h +1 -0
@@ -1481,6 +1481,7 @@ struct task_struct {#endif /* CONFIG_TRACING */ #ifdef CONFIG_CHECKPOINT struct ckpt_ctx *checkpoint_ctx; +unsigned long required_id;#endif };
Signature · 1 line
1.6.0.4
A message that is not in the archive (sent off-list or not yet received): 1252074054-22241-3-git-send-email-orenl-RdfvBDnrOixBDgjK7y7TUQ@public.gmane.org
Re: [PATCH 2/6] c/r: [pty 1/2] allow allocation of desired pty slave
Quoted 10 lines · Oren Laadan
During restart, we need to allocate pty slaves with the same identifiers as recorded during checkpoint. Modify the allocation code to allow an in-kernel caller to request a specific slave identifier.
For this, add a new field to
task_struct- 'required_id'. It will hold the desired identifier when restoring a (master) pty.The code in
ptmx_open()will use this value only for tasks that try to open/dev/ptmxthat are restarting (PF_RESTARTING), and if the value isn'tCKPT_REQUIRED_NONE(-1).
So noone has indicated any preference for this versus the ptmx_create() approach...
I'm satisfied knowing we have a working fallback in case task->required_id is deemed unacceptable.
However I'd like to not have linux-kernel folks think us morons for not having considered that. Can you add a message to the changelog saying why we're going with this approach (namely, that it lets us re-use filp_open() instead of having to do a custom alloc_file in a new code-path, which introduces maintenance duplication for file permission checking paths)?
thanks,
-serge
A message that is not in the archive (sent off-list or not yet received): 20090904152644.GA15253-r/[email protected]
Re: [PATCH 2/6] c/r: [pty 1/2] allow allocation of desired pty slave
Quoted 18 lines · Serge E. Hallyn
Quoted 10 lines · Oren Laadan
During restart, we need to allocate pty slaves with the same identifiers as recorded during checkpoint. Modify the allocation code to allow an in-kernel caller to request a specific slave identifier.
For this, add a new field to
task_struct- 'required_id'. It will hold the desired identifier when restoring a (master) pty.The code in
ptmx_open()will use this value only for tasks that try to open/dev/ptmxthat are restarting (PF_RESTARTING), and if the value isn'tCKPT_REQUIRED_NONE(-1).So noone has indicated any preference for this versus the
ptmx_create()approach...I'm satisfied knowing we have a working fallback in case
task->required_idis deemed unacceptable.
Quoted 6 lines
However I'd like to not have linux-kernel folks think us morons for not having considered that. Can you add a message to the changelog saying why we're going with this approach (namely, that it lets us re-use
filp_open()instead of having to do a customalloc_filein a new code-path, which introduces maintenance duplication for file permission checking paths)?
Good point, will add.
Oren.
Re: [PATCH 2/6] c/r: [pty 1/2] allow allocation of desired pty slave
In reply to a message not in the archiveQuoted 18 lines · Serge E. Hallyn
Quoted 10 lines · Oren Laadan
During restart, we need to allocate pty slaves with the same identifiers as recorded during checkpoint. Modify the allocation code to allow an in-kernel caller to request a specific slave identifier.
For this, add a new field to
task_struct- 'required_id'. It will hold the desired identifier when restoring a (master) pty.The code in
ptmx_open()will use this value only for tasks that try to open/dev/ptmxthat are restarting (PF_RESTARTING), and if the value isn'tCKPT_REQUIRED_NONE(-1).So noone has indicated any preference for this versus the
ptmx_create()approach...I'm satisfied knowing we have a working fallback in case
task->required_idis deemed unacceptable.
Quoted 6 lines
However I'd like to not have linux-kernel folks think us morons for not having considered that. Can you add a message to the changelog saying why we're going with this approach (namely, that it lets us re-use
filp_open()instead of having to do a customalloc_filein a new code-path, which introduces maintenance duplication for file permission checking paths)?
As far as I am concerned, I do have a preference for the ptmx_create() approach. This task->required_id field reminds me the former approach taken for restarting pids and (and SYSV IPC ids IIRC) from userspace, that was proposed last year and actually deemed unacceptable [ IIRC, this was an argument in favor of a restart() syscall ]. I know that it's not the same since ->required_id is not set from userspace and used in a later syscall, but still ...
Moreover I see no reason whey there would be no way to refactorize ptmx code and have less duplicated code with the ptmx_create() approach.
Thanks,
Louis
Signature · 4 lines
Dr Louis Rilling Kerlabs Skype: louis.rilling Batiment Germanium Phone: (+33|0) 6 80 89 08 23 80 avenue des Buttes de Coesmes http://www.kerlabs.com/ 35700 Rennes
_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linux-foundation.org/mailman/listinfo/containers
A message that is not in the archive (sent off-list or not yet received): 20090908080944.GP12824-Hu8+6S1rdjywhHL9vcZdMVaTQe2KTcn/@public.gmane.org
Re: [PATCH 2/6] c/r: [pty 1/2] allow allocation of desired pty slave
Quoted 32 lines · Louis Rilling
Quoted 17 lines · Serge E. Hallyn
Quoted 10 lines · Oren Laadan
During restart, we need to allocate pty slaves with the same identifiers as recorded during checkpoint. Modify the allocation code to allow an in-kernel caller to request a specific slave identifier.
For this, add a new field to
task_struct- 'required_id'. It will hold the desired identifier when restoring a (master) pty.The code in
ptmx_open()will use this value only for tasks that try to open/dev/ptmxthat are restarting (PF_RESTARTING), and if the value isn'tCKPT_REQUIRED_NONE(-1).So noone has indicated any preference for this versus the
ptmx_create()approach...I'm satisfied knowing we have a working fallback in case
task->required_idis deemed unacceptable.Quoted 6 lines
However I'd like to not have linux-kernel folks think us morons for not having considered that. Can you add a message to the changelog saying why we're going with this approach (namely, that it lets us re-use
filp_open()instead of having to do a customalloc_filein a new code-path, which introduces maintenance duplication for file permission checking paths)?As far as I am concerned, I do have a preference for the
ptmx_create()approach. Thistask->required_idfield reminds me the former approach taken for restarting pids and (and SYSV IPC ids IIRC) from userspace, that was proposed last year and actually deemed unacceptable [ IIRC, this was an argument in favor of arestart()syscall ]. I know that it's not the same since ->required_id is not set from userspace and used in a later syscall, but still ...
Quoted 2 lines
Moreover I see no reason whey there would be no way to refactorize ptmx code and have less duplicated code with the
ptmx_create()approach.
I basically agree - I simply took the easiest/fastest path; if the ptmx code is properly refactored, we should use that instead.
Did you have a chance to look at Serge's attempt to do exactly that ?
https://lists.linux-foundation.org/pipermail/containers/2009-September/020363.html
Thanks,
Oren.
A message that is not in the archive (sent off-list or not yet received): [email protected]
Re: [PATCH 2/6] c/r: [pty 1/2] allow allocation of desired pty slave
Quoted 35 lines · Oren Laadan
Quoted 32 lines · Louis Rilling
Quoted 17 lines · Serge E. Hallyn
Quoted 10 lines · Oren Laadan
During restart, we need to allocate pty slaves with the same identifiers as recorded during checkpoint. Modify the allocation code to allow an in-kernel caller to request a specific slave identifier.
For this, add a new field to
task_struct- 'required_id'. It will hold the desired identifier when restoring a (master) pty.The code in
ptmx_open()will use this value only for tasks that try to open/dev/ptmxthat are restarting (PF_RESTARTING), and if the value isn'tCKPT_REQUIRED_NONE(-1).So noone has indicated any preference for this versus the
ptmx_create()approach...I'm satisfied knowing we have a working fallback in case
task->required_idis deemed unacceptable.Quoted 6 lines
However I'd like to not have linux-kernel folks think us morons for not having considered that. Can you add a message to the changelog saying why we're going with this approach (namely, that it lets us re-use
filp_open()instead of having to do a customalloc_filein a new code-path, which introduces maintenance duplication for file permission checking paths)?As far as I am concerned, I do have a preference for the
ptmx_create()approach. Thistask->required_idfield reminds me the former approach taken for restarting pids and (and SYSV IPC ids IIRC) from userspace, that was proposed last year and actually deemed unacceptable [ IIRC, this was an argument in favor of arestart()syscall ]. I know that it's not the same since ->required_id is not set from userspace and used in a later syscall, but still ...
Quoted 8 lines
Quoted 2 lines
Moreover I see no reason whey there would be no way to refactorize ptmx code and have less duplicated code with the
ptmx_create()approach.I basically agree - I simply took the easiest/fastest path; if the ptmx code is properly refactored, we should use that instead.
Did you have a chance to look at Serge's attempt to do exactly that ?
https://lists.linux-foundation.org/pipermail/containers/2009-September/020363.html
I only had a quick look. Will try to look closer, but don't rely on me being quick at this.
Thanks,
Louis
Signature · 4 lines
Dr Louis Rilling Kerlabs Skype: louis.rilling Batiment Germanium Phone: (+33|0) 6 80 89 08 23 80 avenue des Buttes de Coesmes http://www.kerlabs.com/ 35700 Rennes
_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linux-foundation.org/mailman/listinfo/containers
PATCH 3/6 c/r: [pty 2/2] support for pseudo terminals
In reply to a message not in the archiveThis patch adds support for checkpoint and restart of pseudo terminals (PTYs). Since PTYs are shared (pointed to by file, and signal), they are managed via objhash.
PTYs are master/slave pairs; The code arranges for the master to always be checkpointed first, followed by the slave. This is important since during restart both ends are created when restoring the master.
In this patch only UNIX98 style PTYs are supported.
Currently only PTYs that are referenced by open files are handled. Thus PTYs checkpoint starts with a file in tty_file_checkpoint(). It will first checkpoint the master and slave PTYs via tty_checkpoint(), and then complete the saving of the file descriptor. This means that in the image file, the order of objects is: master-tty, slave-tty, file-desc.
During restart, to restore the master side, we open the /dev/ptmx device and get a file handle. But at this point we don't know the designated objref for this file, because the file is due later on in the image stream. On the other hand, we can't just fput() the file because it will close the PTY too.
Instead, when we checkpoint the master PTY, we _reserve_ an objref for the file (which won't be further used in checkpoint). Then at restart, use it to insert the file to objhash.
TODO:
* Better sanitize input from checkpoint image on restore
* Check the locking when saving/restoring tty_struct state
* Echo position/buffer isn't saved (is it needed ?)
* Handle multiple devpts mounts (namespaces)
* Paths of ptmx and slaves are hard coded (/dev/ptmx, /dev/pts/...)
Changelog[v1]:
- Adjust include/asm/checkpoint_hdr.h for s390 architecture
- Add NCC to kernel constants header (ckpt_hdr_const)
- [Serge Hallyn] fix calculation of canon_datalen
--- arch/s390/include/asm/checkpoint_hdr.h | 11 + arch/x86/include/asm/checkpoint_hdr.h | 11 + checkpoint/checkpoint.c | 2 + checkpoint/files.c | 6 + checkpoint/objhash.c | 26 ++ checkpoint/restart.c | 4 + drivers/char/pty.c | 2 +- drivers/char/tty_io.c | 515 +++++++++++++++++++++++++++++++- include/linux/checkpoint.h | 5 + include/linux/checkpoint_hdr.h | 78 +++++ include/linux/tty.h | 6 + 11 files changed, 664 insertions(+), 2 deletions(-)
arch/s390/include/asm/checkpoint_hdr.h +11 -0
@@ -83,13 +83,24 @@ struct ckpt_hdr_mm_context { };#define CKPT_ARCH_NSIG 64 +#define CKPT_TTY_NCC 8 + +/* arch dependent constants */ #ifdef __KERNEL__ +#include <asm/signal.h> #if CKPT_ARCH_NSIG != _SIGCONTEXT_NSIG#error CKPT_ARCH_NSIG size is wrong (asm/sigcontext.h and asm/checkpoint_hdr.h)#endif + +#include <linux/tty.h> +#if CKPT_TTY_NCC !=NCC +#error CKPT_TTY_NCC size is wrong per asm-generic/termios.h#endif +#endif /* __KERNEL__ */ + +struct ckpt_hdr_header_arch {struct ckpt_hdr h; };
arch/x86/include/asm/checkpoint_hdr.h +11 -0
@@ -48,14 +48,25 @@ enum { CKPT_HDR_MM_CONTEXT_LDT, }; +/* arch dependent constants */ #define CKPT_ARCH_NSIG 64 +#define CKPT_TTY_NCC 8 +#ifdef __KERNEL__ +#include <asm/signal.h> #if CKPT_ARCH_NSIG != _NSIG#error CKPT_ARCH_NSIG size is wrong per asm/signal.h and asm/checkpoint_hdr.h#endif + +#include <linux/tty.h> +#if CKPT_TTY_NCC !=NCC +#error CKPT_TTY_NCC size is wrong per asm-generic/termios.h#endif +#endif /* __KERNEL__ */ + +struct ckpt_hdr_header_arch {struct ckpt_hdr h;/* FIXME: add HAVE_HWFP */
checkpoint/checkpoint.c +2 -0
@@ -201,6 +201,8 @@ static void fill_kernel_const(struct ckpt_constants *h) h->uts_domainname_len =sizeof (uts->domainname);/* rlimit */ h->rlimit_nlimits =RLIM_NLIMITS ; +/* tty */ + h->tty_termios_ncc =NCC ; }/* write the checkpoint header */
checkpoint/files.c +6 -0
@@ -572,6 +572,12 @@ static struct restore_file_ops restore_file_ops[] = { .file_type = CKPT_FILE_SOCKET, .restore = sock_file_restore, }, +/* tty */ + { + .file_name = "TTY", + .file_type = CKPT_FILE_TTY, + .restore = tty_file_restore, + }, };static struct file *do_restore_file (struct ckpt_ctx *ctx)
checkpoint/objhash.c +26 -0
@@ -246,6 +246,22 @@ static void obj_sock_drop(void *ptr)sock_put ((struct sock *) ptr); } +static int obj_tty_grab (void *ptr) +{ +tty_kref_get ((struct tty_struct *) ptr); +return 0 ; +} + +static void obj_tty_drop (void *ptr) +{ +tty_kref_put ((struct tty_struct *) ptr); +} + +static int obj_tty_users (void *ptr) +{ +return atomic_read (&((struct tty_struct *) ptr)->kref.refcount); +} +static struct ckpt_obj_ops ckpt_obj_ops[] = {/* ignored object */ {
@@ -382,6 +398,16 @@ static struct ckpt_obj_ops ckpt_obj_ops[] = { .ref_drop = obj_sock_drop, .ref_grab = obj_sock_grab, }, +/* struct tty_struct */ + { + .obj_name = "TTY", + .obj_type = CKPT_OBJ_TTY, + .ref_drop = obj_tty_drop, + .ref_grab = obj_tty_grab, + .ref_users = obj_tty_users, + .checkpoint = checkpoint_tty, + .restore = restore_tty, + }, };
checkpoint/restart.c +4 -0
@@ -19,6 +19,7 @@#include <linux/freezer.h> #include <linux/magic.h> #include <linux/utsname.h> +#include <linux/termios.h> #include <asm/syscall.h> #include <linux/elf.h> #include <linux/checkpoint.h>
@@ -374,6 +375,9 @@ static int check_kernel_const(struct ckpt_constants *h)/* rlimit */ if (h->rlimit_nlimits !=RLIM_NLIMITS )return -EINVAL ; +/* tty */ +if (h->tty_termios_ncc !=NCC ) +return -EINVAL ;return 0 ; }
drivers/char/pty.c +1 -1
@@ -633,7 +633,7 @@ static int __ptmx_open(struct inode *inode, struct file *filp) {struct tty_struct *tty;int retval; -int index =0 ; +int index =-1 ;nonseekable_open (inode, filp);
drivers/char/tty_io.c +514 -1
551 lines not on this page · raw message
include/linux/checkpoint.h +5 -0
@@ -265,6 +265,11 @@ extern int restore_obj_signal(struct ckpt_ctx *ctx, int signal_objref);extern int checkpoint_task_signal (struct ckpt_ctx *ctx,struct task_struct *t);extern int restore_task_signal (struct ckpt_ctx *ctx); +/* ttys */ +extern int checkpoint_tty (struct ckpt_ctx *ctx,void *ptr); +extern void *restore_tty (struct ckpt_ctx *ctx); + +/* useful macros to copy fields and buffers to/from ckpt_hdr_xxx structures */ #define CKPT_CPT 1#define CKPT_RST 2
include/linux/checkpoint_hdr.h +78 -0
108 lines not on this page · raw message
include/linux/tty.h +6 -0
@@ -467,6 +467,12 @@ extern void tty_ldisc_begin(void);/* This last one is just for the tty layer internals and shouldn't be used elsewhere */ extern void tty_ldisc_enable (struct tty_struct *tty); +#ifdef CONFIG_CHECKPOINT +struct ckpt_ctx ; +struct ckpt_hdr_file ; +extern struct file *tty_file_restore (struct ckpt_ctx *ctx, +struct ckpt_hdr_file *ptr); +#endif /* n_tty.c */ extern struct tty_ldisc_ops tty_ldisc_N_TTY;
Signature · 1 line
1.6.0.4
A message that is not in the archive (sent off-list or not yet received): 1252074054-22241-4-git-send-email-orenl-RdfvBDnrOixBDgjK7y7TUQ@public.gmane.org
Re: [PATCH 3/6] c/r: [pty 2/2] support for pseudo terminals
Quoted 35 lines · Oren Laadan
This patch adds support for checkpoint and restart of pseudo terminals (PTYs). Since PTYs are shared (pointed to by file, and signal), they are managed via objhash.
PTYs are master/slave pairs; The code arranges for the master to always be checkpointed first, followed by the slave. This is important since during restart both ends are created when restoring the master.
In this patch only UNIX98 style PTYs are supported.
Currently only PTYs that are referenced by open files are handled. Thus PTYs checkpoint starts with a file in
tty_file_checkpoint(). It will first checkpoint the master and slave PTYs viatty_checkpoint(), and then complete the saving of the file descriptor. This means that in the image file, the order of objects is: master-tty, slave-tty, file-desc.During restart, to restore the master side, we open the
/dev/ptmxdevice and get a file handle. But at this point we don't know the designated objref for this file, because the file is due later on in the image stream. On the other hand, we can't justfput()the file because it will close the PTY too.Instead, when we checkpoint the master PTY, we _reserve_ an objref for the file (which won't be further used in checkpoint). Then at restart, use it to insert the file to objhash.
TODO:
* Better sanitize input from checkpoint image on restore
* Check the locking when saving/restoringtty_structstate
* Echo position/buffer isn't saved (is it needed ?)
* Handle multiple devpts mounts (namespaces)
* Paths of ptmx and slaves are hard coded (/dev/ptmx,/dev/pts/...)
Quoted 6 lines
Changelog[v1]:
- Adjustinclude/asm/checkpoint_hdr.hfor s390 architecture
- Add NCC to kernel constants header (ckpt_hdr_const)
- [Serge Hallyn] fix calculation ofcanon_datalenSigned-off-by Oren Laadan <[email protected]>
First three patches can get
Re: [PATCH 3/6] c/r: [pty 2/2] support for pseudo terminals
In reply to a message not in the archiveQuoted 41 lines · Oren Laadan
This patch adds support for checkpoint and restart of pseudo terminals (PTYs). Since PTYs are shared (pointed to by file, and signal), they are managed via objhash.
PTYs are master/slave pairs; The code arranges for the master to always be checkpointed first, followed by the slave. This is important since during restart both ends are created when restoring the master.
In this patch only UNIX98 style PTYs are supported.
Currently only PTYs that are referenced by open files are handled. Thus PTYs checkpoint starts with a file in
tty_file_checkpoint(). It will first checkpoint the master and slave PTYs viatty_checkpoint(), and then complete the saving of the file descriptor. This means that in the image file, the order of objects is: master-tty, slave-tty, file-desc.During restart, to restore the master side, we open the
/dev/ptmxdevice and get a file handle. But at this point we don't know the designated objref for this file, because the file is due later on in the image stream. On the other hand, we can't justfput()the file because it will close the PTY too.Instead, when we checkpoint the master PTY, we _reserve_ an objref for the file (which won't be further used in checkpoint). Then at restart, use it to insert the file to objhash.
TODO:
* Better sanitize input from checkpoint image on restore
* Check the locking when saving/restoringtty_structstate
* Echo position/buffer isn't saved (is it needed ?)
* Handle multiple devpts mounts (namespaces)
* Paths of ptmx and slaves are hard coded (/dev/ptmx,/dev/pts/...)Changelog[v1]:
- Adjustinclude/asm/checkpoint_hdr.hfor s390 architecture
- Add NCC to kernel constants header (ckpt_hdr_const)
- [Serge Hallyn] fix calculation ofcanon_datalenSigned-off-by Oren Laadan <[email protected]>
[...]
Quoted 5 lines
drivers/char/tty_io.c +0
-0
drivers/char/tty_io.c static void do_tty_hangup(struct work_struct *work) · line 590
Quoted 10 lines
drivers/char/tty_io.c +3
-1@@ -570,7 +582,9 @@ static void do_tty_hangup(struct work_struct *work)set_bit (TTY_HUPPED , &tty->flags);spin_unlock_irqrestore (&tty->ctrl_lock, flags); -/* Account for the p->signal references we killed */ +/* Account + +for the p->signal references we killed */ while (refs--)tty_kref_put (tty);
Nit: this hunk is certainly not needed ;)
[...]
Thanks,
Louis
Signature · 4 lines
Dr Louis Rilling Kerlabs Skype: louis.rilling Batiment Germanium Phone: (+33|0) 6 80 89 08 23 80 avenue des Buttes de Coesmes http://www.kerlabs.com/ 35700 Rennes
_______________________________________________
Containers mailing list Containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org
https://lists.linux-foundation.org/mailman/listinfo/containers
A message that is not in the archive (sent off-list or not yet received): 20090908085013.GQ12824-Hu8+6S1rdjywhHL9vcZdMVaTQe2KTcn/@public.gmane.org
Re: [PATCH 3/6] c/r: [pty 2/2] support for pseudo terminals
Quoted 53 lines · Louis Rilling
Quoted 41 lines · Oren Laadan
This patch adds support for checkpoint and restart of pseudo terminals (PTYs). Since PTYs are shared (pointed to by file, and signal), they are managed via objhash.
PTYs are master/slave pairs; The code arranges for the master to always be checkpointed first, followed by the slave. This is important since during restart both ends are created when restoring the master.
In this patch only UNIX98 style PTYs are supported.
Currently only PTYs that are referenced by open files are handled. Thus PTYs checkpoint starts with a file in
tty_file_checkpoint(). It will first checkpoint the master and slave PTYs viatty_checkpoint(), and then complete the saving of the file descriptor. This means that in the image file, the order of objects is: master-tty, slave-tty, file-desc.During restart, to restore the master side, we open the
/dev/ptmxdevice and get a file handle. But at this point we don't know the designated objref for this file, because the file is due later on in the image stream. On the other hand, we can't justfput()the file because it will close the PTY too.Instead, when we checkpoint the master PTY, we _reserve_ an objref for the file (which won't be further used in checkpoint). Then at restart, use it to insert the file to objhash.
TODO:
* Better sanitize input from checkpoint image on restore
* Check the locking when saving/restoringtty_structstate
* Echo position/buffer isn't saved (is it needed ?)
* Handle multiple devpts mounts (namespaces)
* Paths of ptmx and slaves are hard coded (/dev/ptmx,/dev/pts/...)Changelog[v1]:
- Adjustinclude/asm/checkpoint_hdr.hfor s390 architecture
- Add NCC to kernel constants header (ckpt_hdr_const)
- [Serge Hallyn] fix calculation ofcanon_datalenSigned-off-by Oren Laadan <[email protected]>[...]
Quoted 8 lines
drivers/char/tty_io.c +0
-0@@ -570,7 +582,9 @@ static void do_tty_hangup(struct work_struct *work)set_bit (TTY_HUPPED , &tty->flags);spin_unlock_irqrestore (&tty->ctrl_lock, flags);
Quoted 9 lines
drivers/char/tty_io.c static void do_tty_hangup(struct work_struct *work) · line 590
Quoted 7 lines
drivers/char/tty_io.c +3
-1-/* Account for the p->signal references we killed */ +/* Account + +for the p->signal references we killed */ while (refs--)tty_kref_put (tty);Nit: this hunk is certainly not needed ;)
Hmm... lemme think about it ... :p
Oren.
PATCH 4/6 c/r: tighten logic to protect against bogus pids in input
In reply to a message not in the archiveDuring restart, each process that completes its restore will wake up the next process in line, using the pid provided in the image file. A malicious user could provide bogus pids and cause arbitrary tasks to be woken up.
This patch tightens the checks and requires that to wake-up a task by pid during restart, the target task must belong to (have) the same restart context (task->checkpoint_ctx) as the current process.
Also, update in-code comments about restart and zombie logic.
--- checkpoint/restart.c | 35 ++++++++++++++++++++--------------- 1 files changed, 20 insertions(+), 15 deletions(-)
checkpoint/restart.c +20 -15
@@ -462,12 +462,12 @@ static int restore_read_tree(struct ckpt_ctx *ctx)return PTR_ERR (h); ret = -EINVAL ; -if (h->nr_tasks <0 ) +if (h->nr_tasks <=0 ) goto out; ctx->nr_pids = h->nr_tasks; size =sizeof (*ctx->pids_arr) * ctx->nr_pids; -if (size <0 )/* overflow ? */ +if (size <=0 )/* overflow ? */ goto out; ctx->pids_arr =kmalloc (size,GFP_KERNEL );
@@ -520,8 +520,11 @@ int ckpt_activate_next(struct ckpt_ctx *ctx)rcu_read_lock (); task =find_task_by_pid_ns (pid, ctx->root_nsproxy->pid_ns); -if (task) +/* target task must have same restart context */ +if (task && task->checkpoint_ctx == ctx)wake_up_process (task); +else + task =NULL ;rcu_read_unlock ();if (!task) {
@@ -568,9 +571,8 @@ static int do_restore_task(void)int ret;/* -* Wait for coordinator to make become visible, then grab a -* reference to its restart context. If we're the last task to -* do it, notify the coordinator. +* Wait for coordinator to become visible, then grab a +* reference to its restart context. */ ret =wait_event_interruptible (waitq, current->checkpoint_ctx);if (ret <0 )
@@ -610,8 +612,9 @@ static int do_restore_task(void) goto out;/* -* zombie: we're done here; Save @ctx on task_struct, to be -* used to ckpt_activate_next(), and released, from do_exit(). +* zombie: we're done here; do_exit() will notice the @ctx on +* our current->checkpoint_ctx (and our PF_RESTARTING) - it +* will call ckpt_activate_next() and release the @ctx. */ if (ret)do_exit (current->exit_code);
@@ -638,11 +641,11 @@ static int do_restore_task(void) }/** -* prepare_descendants - set ->restart_tsk of all descendants +* prepare_descendants - set ->checkpoint_ctx of all descendants * @ctx: checkpoint context * @root: root process for restart * -* Called by the coodinator to set the ->restart_tsk pointer of the +* Called by the coodinator to set the ->checkpoint_ctx pointer of the * root task and all its descendants. */ static int prepare_descendants (struct ckpt_ctx *ctx,struct task_struct *root)
@@ -662,7 +665,7 @@ static int prepare_descendants(struct ckpt_ctx *ctx, struct task_struct *root)break ; }/* -* Set task->restart_tsk of all non-zombie descendants. +* Set task->checkpoint_ctx of all non-zombie descendants. * If a descendant already has a ->checkpoint_ctx, it * must be a coordinator (for a different restart ?) so * we fail.
@@ -727,6 +730,7 @@ static int wait_all_tasks_finish(struct ckpt_ctx *ctx)init_completion (&ctx->complete); +BUG_ON (ctx->active_pid !=-1 ); ret =ckpt_activate_next (ctx);if (ret <0 )return ret;
@@ -839,7 +843,7 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)if (ret <0 ) goto out; }else { -/* prepare descendants' t->restart_tsk point to coord */ +/* prepare descendants' t->checkpoint_ctx point to coord */ ret =prepare_descendants (ctx, ctx->root_task);if (ret <0 ) goto out;
@@ -970,11 +974,12 @@ void exit_checkpoint(struct task_struct *tsk) {struct ckpt_ctx *ctx; - ctx = tsk->checkpoint_ctx; - tsk->checkpoint_ctx =NULL ; +/* no one else will touch this, because @tsk is dead already */ + ctx =xchg (&tsk->checkpoint_ctx,NULL ); -/* restarting zombies will acrivate next task in restart */ +/* restarting zombies will activate next task in restart */ if (tsk->flags &PF_RESTARTING ) { +BUG_ON (ctx->active_pid ==-1 );if (ckpt_activate_next (ctx) <0 )pr_warning ("c/r: [%d] failed zombie exit\n" , tsk->pid); }
Signature · 1 line
1.6.0.4
A message that is not in the archive (sent off-list or not yet received): 1252074054-22241-5-git-send-email-orenl-RdfvBDnrOixBDgjK7y7TUQ@public.gmane.org
Serge E. Hallyn
Acked-byRe: [PATCH 4/6] c/r: tighten logic to protect against bogus pids in input
Quoted 5 lines · Oren Laadan
During restart, each process that completes its restore will wake up the next process in line, using the pid provided in the image file. A malicious user could provide bogus pids and cause arbitrary tasks to be woken up.
Quoted 7 lines
This patch tightens the checks and requires that to wake-up a task by pid during restart, the target task must belong to (have) the same restart context (
task->checkpoint_ctx) as the current process.Also, update in-code comments about restart and zombie logic.
Signed-off-by Oren Laadan <[email protected]>
Quoted 118 lines
--- checkpoint/restart.c | 35 ++++++++++++++++++++--------------- 1 files changed, 20 insertions(+), 15 deletions(-)checkpoint/restart.c +20
-15@@ -462,12 +462,12 @@ static int restore_read_tree(struct ckpt_ctx *ctx)return PTR_ERR (h); ret = -EINVAL ; -if (h->nr_tasks <0 ) +if (h->nr_tasks <=0 ) goto out; ctx->nr_pids = h->nr_tasks; size =sizeof (*ctx->pids_arr) * ctx->nr_pids; -if (size <0 )/* overflow ? */ +if (size <=0 )/* overflow ? */ goto out; ctx->pids_arr =kmalloc (size,GFP_KERNEL );@@ -520,8 +520,11 @@ int ckpt_activate_next(struct ckpt_ctx *ctx)rcu_read_lock (); task =find_task_by_pid_ns (pid, ctx->root_nsproxy->pid_ns); -if (task) +/* target task must have same restart context */ +if (task && task->checkpoint_ctx == ctx)wake_up_process (task); +else + task =NULL ;rcu_read_unlock ();if (!task) {@@ -568,9 +571,8 @@ static int do_restore_task(void)int ret;/* -* Wait for coordinator to make become visible, then grab a -* reference to its restart context. If we're the last task to -* do it, notify the coordinator. +* Wait for coordinator to become visible, then grab a +* reference to its restart context. */ ret =wait_event_interruptible (waitq, current->checkpoint_ctx);if (ret <0 )@@ -610,8 +612,9 @@ static int do_restore_task(void) goto out;/* -* zombie: we're done here; Save @ctx on task_struct, to be -* used to ckpt_activate_next(), and released, from do_exit(). +* zombie: we're done here; do_exit() will notice the @ctx on +* our current->checkpoint_ctx (and our PF_RESTARTING) - it +* will call ckpt_activate_next() and release the @ctx. */ if (ret)do_exit (current->exit_code);@@ -638,11 +641,11 @@ static int do_restore_task(void) }/** -* prepare_descendants - set ->restart_tsk of all descendants +* prepare_descendants - set ->checkpoint_ctx of all descendants * @ctx: checkpoint context * @root: root process for restart * -* Called by the coodinator to set the ->restart_tsk pointer of the +* Called by the coodinator to set the ->checkpoint_ctx pointer of the * root task and all its descendants. */ static int prepare_descendants (struct ckpt_ctx *ctx,struct task_struct *root)@@ -662,7 +665,7 @@ static int prepare_descendants(struct ckpt_ctx *ctx, struct task_struct *root)break ; }/* -* Set task->restart_tsk of all non-zombie descendants. +* Set task->checkpoint_ctx of all non-zombie descendants. * If a descendant already has a ->checkpoint_ctx, it * must be a coordinator (for a different restart ?) so * we fail. @@ -727,6 +730,7 @@ static int wait_all_tasks_finish(struct ckpt_ctx *ctx)init_completion (&ctx->complete); +BUG_ON (ctx->active_pid !=-1 ); ret =ckpt_activate_next (ctx);if (ret <0 )return ret;@@ -839,7 +843,7 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)if (ret <0 ) goto out; }else { -/* prepare descendants' t->restart_tsk point to coord */ +/* prepare descendants' t->checkpoint_ctx point to coord */ ret =prepare_descendants (ctx, ctx->root_task);if (ret <0 ) goto out;@@ -970,11 +974,12 @@ void exit_checkpoint(struct task_struct *tsk) {struct ckpt_ctx *ctx; - ctx = tsk->checkpoint_ctx; - tsk->checkpoint_ctx =NULL ; +/* no one else will touch this, because @tsk is dead already */ + ctx =xchg (&tsk->checkpoint_ctx,NULL ); -/* restarting zombies will acrivate next task in restart */ +/* restarting zombies will activate next task in restart */ if (tsk->flags &PF_RESTARTING ) { +BUG_ON (ctx->active_pid ==-1 );if (ckpt_activate_next (ctx) <0 )pr_warning ("c/r: [%d] failed zombie exit\n" , tsk->pid); }Signature · 1 line
1.6.0.4
PATCH 1/3 mktree: add support to ghost tasks (TASK_GHOST)
In reply to a message not in the archive--- restart.c | 123 +++++++++++++++++++++++++++++++++++++++--------------------- 1 files changed, 80 insertions(+), 43 deletions(-)
restart.c +80 -43
@@ -144,12 +144,13 @@ struct task {pid_t real_parent;/* pid of task's real parent */ }; -#define TASK_ROOT 0x1/* */ -#define TASK_DEAD 0x2/* */ -#define TASK_THREAD 0x4/* */ -#define TASK_SIBLING 0x8/* */ -#define TASK_SESSION 0x10/* */ -#define TASK_NEWPID 0x20/* */ +#define TASK_ROOT 0x1/* root task */ +#define TASK_GHOST 0x2/* dead task (pid used as sid/pgid) */ +#define TASK_THREAD 0x4/* thread (non leader) */ +#define TASK_SIBLING 0x8/* creator's sibling (use CLONE_PARENT) */ +#define TASK_SESSION 0x10/* inherits creator's original sid */ +#define TASK_NEWPID 0x20/* starts a new pid namespace */ +#define TASK_DEAD 0x40/* dead task (dummy) */ struct ckpt_ctx {pid_t init_pid;
@@ -843,10 +844,12 @@ static int ckpt_build_tree(struct ckpt_ctx *ctx)ckpt_dbg_cont (" prev %d" , task->prev_sib->pid);if (task->phantom)ckpt_dbg_cont (" placeholder %d" , task->phantom->pid); -ckpt_dbg_cont (" %c%c%c%c" , +ckpt_dbg_cont (" %c%c%c%c%c%c" , (task->flags &TASK_THREAD ) ?'T' :' ' , (task->flags &TASK_SIBLING ) ?'P' :' ' , (task->flags &TASK_SESSION ) ?'S' :' ' , + (task->flags &TASK_NEWPID ) ?'N' :' ' , + (task->flags &TASK_GHOST ) ?'G' :' ' , (task->flags &TASK_DEAD ) ?'D' :' ' );ckpt_dbg_cont ("\n" ); }
@@ -855,7 +858,7 @@ static int ckpt_build_tree(struct ckpt_ctx *ctx)return 0 ; } -static int ckpt_setup_task (struct ckpt_ctx *ctx,pid_t pid) +static int ckpt_setup_task (struct ckpt_ctx *ctx,pid_t pid,pid_t ppid) {struct task *task;
@@ -864,12 +867,13 @@ static int ckpt_setup_task(struct ckpt_ctx *ctx, pid_t pid) task = &ctx->tasks_arr[ctx->tasks_nr++]; - task->flags =TASK_DEAD ; + task->flags =TASK_GHOST ; +/* */ task->pid = pid; - task->ppid =ckpt_init_task (ctx)->pid; + task->ppid = ppid; task->tgid = pid; - task->sid = pid; + task->sid = ppid; task->children =NULL ; task->next_sib =NULL ;
@@ -892,36 +896,38 @@ static int ckpt_setup_task(struct ckpt_ctx *ctx, pid_t pid)static int ckpt_init_tree (struct ckpt_ctx *ctx) { +struct ckpt_hdr_pids *pids_arr = ctx->pids_arr; +int pids_nr = ctx->pids_nr;struct task *task; -pid_t init_sid; -pid_t init_pid; -pid_t init_pgid; +pid_t root_sid; +pid_t root_pid; +pid_t root_pgid;int i; - init_pid = ctx->pids_arr[0 ].vpid; - init_sid = ctx->pids_arr[0 ].vsid; - init_pgid = ctx->pids_arr[0 ].vpgid; + root_pid = pids_arr[0 ].vpid; + root_sid = pids_arr[0 ].vsid; + root_pgid = pids_arr[0 ].vpgid;/* XXX for out-of-container subtrees */ -for (i =0 ; i < ctx->pids_nr; i++) { -if (ctx->pids_arr[i].vsid == init_sid) - ctx->pids_arr[i].vsid = init_pid; -if (ctx->pids_arr[i].vpgid == init_sid) - ctx->pids_arr[i].vpgid = init_pid; -if (ctx->pids_arr[i].vpgid == init_pgid) - ctx->pids_arr[i].vpgid = init_pid; +for (i =0 ; i < pids_nr; i++) { +if (pids_arr[i].vsid == root_sid) + pids_arr[i].vsid = root_pid; +if (pids_arr[i].vpgid == root_sid) + pids_arr[i].vpgid = root_pid; +if (pids_arr[i].vpgid == root_pgid) + pids_arr[i].vpgid = root_pid; }/* populate with known tasks */ -for (i =0 ; i < ctx->pids_nr; i++) { +for (i =0 ; i < pids_nr; i++) { task = &ctx->tasks_arr[i]; task->flags =0 ; - task->pid = ctx->pids_arr[i].vpid; - task->ppid = ctx->pids_arr[i].vppid; - task->tgid = ctx->pids_arr[i].vtgid; - task->sid = ctx->pids_arr[i].vsid; + task->pid = pids_arr[i].vpid; + task->ppid = pids_arr[i].vppid; + task->tgid = pids_arr[i].vtgid; + task->sid = pids_arr[i].vsid; task->children =NULL ; task->next_sib =NULL ;
@@ -936,19 +942,32 @@ static int ckpt_init_tree(struct ckpt_ctx *ctx)return -1 ; } - ctx->tasks_nr = ctx->pids_nr; + ctx->tasks_nr = pids_nr;/* add pids unaccounted for (no tasks) */ -for (i =0 ; i < ctx->pids_nr; i++) { -if (ckpt_setup_task (ctx, ctx->pids_arr[i].vsid) <0 ) +for (i =0 ; i < pids_nr; i++) { +/* session leader's parent is root task */ +if (ckpt_setup_task (ctx, pids_arr->vsid, root_pid) <0 )return -1 ; -if (ckpt_setup_task (ctx, ctx->pids_arr[i].vpgid) <0 ) + +/* +* If pgrp != sid, pgrp owner's parent is sid. Other +* tasks with same pgrp will need to have threir sid +* matching, too, when the kernel restores their pgrp. +* If pgrp == sid, then the call above would have +* ensured that the pid is hashed: ckpt_setup_task() +* will return promptly. +*/ +if (ckpt_setup_task (ctx, pids_arr->vpgid, pids_arr->vsid) <0 )return -1 ; + + pids_arr++; }/* mark root task(s) */ ctx->tasks_arr[0 ].flags |=TASK_ROOT ; +ckpt_dbg ("total tasks (including ghosts): %d\n" , ctx->tasks_nr);return 0 ; }
@@ -970,10 +989,11 @@ static int ckpt_init_tree(struct ckpt_ctx *ctx)* flags, pid, tgid, sid, pgid, and pointers to the a creator, next * and previous sibling, and first child task. Note that the creator * may not necessarily correspond to the parent. The possible flags -* are TASK_DEAD, TASK_THREAD, TASK_SESSION (that asks inherit a -* session id), and TASK_SIBLING (that asks to inherit the parent via -* CLONE_PARENT). The algorithm loops through all the entries in the -* table: +* are TASK_ROOT, TASK_GHOST, TASK_THREAD, TASK_SIBLING (that asks to +* inherit the parent via CLONE_PARENT), TASK_SESSION (that asks to +* inherit a session id), TASK_NEWPID (that asks to start a new pid +* namespace), and TASK_DEAD. The algorithm loops through all the +* entries in the table: * * If the entry is a thread and not the thread group leader, we set * the creator to be the thread group leader and set TASK_THREAD.
@@ -1023,7 +1043,7 @@ static int ckpt_set_creator(struct ckpt_ctx *ctx, struct task *task)struct task *creator;if (task ==ckpt_init_task (ctx)) { -ckpt_err ("pid %d: init - no creator\n" ,ckpt_init_task (ctx)->pid); +ckpt_err ("pid %d: no init creator\n" ,ckpt_init_task (ctx)->pid);return -1 ; }
@@ -1230,6 +1250,7 @@ static int ckpt_make_tree(struct ckpt_ctx *ctx, struct task *task) {struct task *child;struct pid_swap swap; +unsigned long flags =0 ;pid_t newpid;int ret;
@@ -1307,9 +1328,20 @@ static int ckpt_make_tree(struct ckpt_ctx *ctx, struct task *task) }close (ctx->pipe_out); +/* +* Ghost tasks are not restarted and end up dead, but their +* pids are referred to by other tasks' pgids (also sids, that +* are already properly set by now). Therefore, they stick +* around until those tasks actually restore their pgrp, and +* then exit (more precisely, killed). The RESTART_GHOST flag +* tells the kernel that they are not to be restored. +*/ +if (task->flags &TASK_GHOST ) + flags |=RESTART_GHOST ; +/* on success this doesn't return */ -ckpt_dbg ("about to call sys_restart()\n" ); - ret =restart (0 ,STDIN_FILENO ,0 ); +ckpt_dbg ("about to call sys_restart(), flags %#lx\n" , flags); + ret =restart (0 ,STDIN_FILENO , flags);if (ret <0 )perror ("task restore failed" );return ret;
@@ -1562,7 +1594,10 @@ static int ckpt_adjust_pids(struct ckpt_ctx *ctx)memcpy (ctx->copy_arr, ctx->pids_arr, len);/* read in 'pid_swap' data and adjust ctx->pids_arr */ -for (n =0 ; n < ctx->pids_nr; n++) { +for (n =0 ; n < ctx->tasks_nr; n++) { +/* don't expect data from dead tasks */ +if (ctx->tasks_arr[n].flags &TASK_DEAD ) +continue ; ret =read (ctx->pipe_in, &swap,sizeof (swap));if (ret <0 )ckpt_abort (ctx,"read pipe" );
@@ -1577,8 +1612,10 @@ static int ckpt_adjust_pids(struct ckpt_ctx *ctx) ctx->copy_arr[m].vpid = swap.new;if (ctx->pids_arr[m].vtgid == swap.old) ctx->copy_arr[m].vtgid = swap.new; -if (ctx->pids_arr[m].vppid == swap.old) - ctx->copy_arr[m].vppid = swap.new; +if (ctx->pids_arr[m].vpgid == swap.old) + ctx->copy_arr[m].vpgid = swap.new; +else if (ctx->pids_arr[m].vpgid == -swap.old) + ctx->copy_arr[m].vpgid = -swap.new; } }
Signature · 1 line
1.6.0.4
PATCH 2/3 test/pgrp.c: add test case for process-groups
In reply to a message not in the archive--- test/Makefile | 2 +- test/pgrp.c | 79 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 80 insertions(+), 1 deletions(-) create mode 100644 test/pgrp.c
test/Makefile +1 -1
@@ -7,7 +7,7 @@ CFLAGS += -g $(WARNS) TESTS= onetask multitask bigmem pipes pipes2 fifo shmem \ ipcshm ipcmsq ipcsem ipcshm_multi zombie sigpending \ - itimer pty + itimer pty pgrp LDLIBS = -lm
test/pgrp.c new file +79 -0
79 lines not on this page · raw message
Signature · 1 line
1.6.0.4
PATCH 3/3 restart: correctly handle pgid/ppid/sid = 0
In reply to a message not in the archiveIf pgid, or sid, or ppid is zero, it means that they point to a task from the ancestor pid-ns. Make sure 'mktree' handles these correctly.
If a 0 pid (pgid/sid/ppid) exists, then force the --pidns option to be on - to ensure that we have a new pid-ns (and thus a parent pid-ns).
Ignore pid = 0 when setting up new tasks in ckpt_setup_task().
Improve sanity checks for pid 0 and tgid 0 (both are impossible).
Allow ppid = 0 only for root_task. For others, allow sid = 0.
Introduce dummy zero_task that represents the parent of root_task, likely from ancestor pid-ns. Setting a creator to root_task simplifies the logic.
Also, ensure that a the tgid of threads is valid (has an entry in the pids array) even if the thread group leader died.
--- restart.c | 126 +++++++++++++++++++++++++++++++++++++++++++------------------ 1 files changed, 89 insertions(+), 37 deletions(-)
restart.c +89 -37
@@ -63,8 +63,8 @@ static char usage_str[] =* the time of the checkpoint may be already in use then. Therefore, * by default, 'mktree' creates an equivalen tree without restoring * the original pids, assuming that the application can tolerate this. -* For this, the 'ckpt_hdr_pids' array is transformed on-the-fly -* before it is fed to the kernel. +* For this, the 'ckpt_pids' array is transformed on-the-fly before it +* is fed to the kernel. * * By default, "--pids" implied "--pidns" and vice-versa. The user can * use "--pids --no-pidns" for a restart in the currnet namespace -
@@ -144,6 +144,9 @@ struct task {pid_t real_parent;/* pid of task's real parent */ }; +/* zero_task represents creator of root_task (all pids 0) */ +struct task zero_task; +#define TASK_ROOT 0x1/* root task */ #define TASK_GHOST 0x2/* dead task (pid used as sid/pgid) */ #define TASK_THREAD 0x4/* thread (non leader) */
@@ -153,7 +156,7 @@ struct task {#define TASK_DEAD 0x40/* dead task (dummy) */ struct ckpt_ctx { -pid_t init_pid; +pid_t root_pid;int pipe_in;int pipe_out;int pids_nr;
@@ -196,6 +199,7 @@ int global_sent_sigint;static int ckpt_build_tree (struct ckpt_ctx *ctx);static int ckpt_init_tree (struct ckpt_ctx *ctx); +static int ckpt_need_pidns (struct ckpt_ctx *ctx);static int ckpt_set_creator (struct ckpt_ctx *ctx,struct task *task);static int ckpt_placeholder_task (struct ckpt_ctx *ctx,struct task *task);static int ckpt_propagate_session (struct ckpt_ctx *ctx,struct task *session);
@@ -545,6 +549,16 @@ int main(int argc, char *argv[])if (ret <0 )exit (1 ); +/* +* For a pgid/sid == 0, the corresponding restarting task will +* expect to reference the parent pid-ns (of entire restart). +* We ensure that one does exist by setting ctx.args->pidns. +*/ +if (!ctx.args->pidns &&ckpt_need_pidns (&ctx)) { +ckpt_dbg ("found pgid/sid 0, need pidns\n" ); + ctx.args->pidns =1 ; + } +if (ctx.args->pidns && ctx.tasks_arr[0 ].pid !=1 ) {ckpt_dbg ("new pidns without init\n" );if (global_send_sigint ==-1 )
@@ -821,8 +835,6 @@ static int ckpt_build_tree(struct ckpt_ctx *ctx)/* assign a creator to each task */ for (i =0 ; i < ctx->tasks_nr; i++) { task = &ctx->tasks_arr[i]; -if (task ==ckpt_init_task (ctx)) -continue ;if (task->creator)continue ;if (ckpt_set_creator (ctx, task) <0 ) {
@@ -837,7 +849,7 @@ static int ckpt_build_tree(struct ckpt_ctx *ctx) task = &ctx->tasks_arr[i];ckpt_dbg ("[%d] pid %d ppid %d sid %d creator %d" , i, task->pid, task->ppid, task->sid, - task->creator ? task->creator->pid :0 ); + task->creator->pid);if (task->next_sib)ckpt_dbg_cont (" next %d" , task->next_sib->pid);if (task->prev_sib)
@@ -862,7 +874,10 @@ static int ckpt_setup_task(struct ckpt_ctx *ctx, pid_t pid, pid_t ppid) {struct task *task; -if (hash_lookup (ctx, pid)) +if (pid ==0 )/* ignore if outside namespace */ +return 0 ; + +if (hash_lookup (ctx, pid))/* already handled */ return 0 ; task = &ctx->tasks_arr[ctx->tasks_nr++];
@@ -896,7 +911,7 @@ static int ckpt_setup_task(struct ckpt_ctx *ctx, pid_t pid, pid_t ppid)static int ckpt_init_tree (struct ckpt_ctx *ctx) { -struct ckpt_hdr_pids *pids_arr = ctx->pids_arr; +struct ckpt_pids *pids_arr = ctx->pids_arr;int pids_nr = ctx->pids_nr;struct task *task;pid_t root_sid;
@@ -908,16 +923,6 @@ static int ckpt_init_tree(struct ckpt_ctx *ctx) root_sid = pids_arr[0 ].vsid; root_pgid = pids_arr[0 ].vpgid; -/* XXX for out-of-container subtrees */ -for (i =0 ; i < pids_nr; i++) { -if (pids_arr[i].vsid == root_sid) - pids_arr[i].vsid = root_pid; -if (pids_arr[i].vpgid == root_sid) - pids_arr[i].vpgid = root_pid; -if (pids_arr[i].vpgid == root_pgid) - pids_arr[i].vpgid = root_pid; - } -/* populate with known tasks */ for (i =0 ; i < pids_nr; i++) { task = &ctx->tasks_arr[i];
@@ -938,6 +943,14 @@ static int ckpt_init_tree(struct ckpt_ctx *ctx) task->rpid =-1 ; task->ctx = ctx; +if (task->pid ==0 ) { +ckpt_err ("Invalid pid 0 for task#%d\n" , i); +return -1 ; + }else if (task->tgid ==0 ) { +ckpt_err ("Invalid tgid 0 for task#%d\n" , i); +return -1 ; + } +if (hash_insert (ctx, task->pid, task) <0 )return -1 ; }
@@ -946,31 +959,60 @@ static int ckpt_init_tree(struct ckpt_ctx *ctx)/* add pids unaccounted for (no tasks) */ for (i =0 ; i < pids_nr; i++) { -/* session leader's parent is root task */ -if (ckpt_setup_task (ctx, pids_arr->vsid, root_pid) <0 ) +pid_t sid; + + sid = pids_arr[i].vsid; + +/* +* An unaccounted-for sid belongs to a task that was a +* session leader and died. We can safe set its parent +* (and creator) to be the root task. +*/ +if (ckpt_setup_task (ctx, sid, root_pid) <0 )return -1 ;/* -* If pgrp != sid, pgrp owner's parent is sid. Other -* tasks with same pgrp will need to have threir sid -* matching, too, when the kernel restores their pgrp. -* If pgrp == sid, then the call above would have -* ensured that the pid is hashed: ckpt_setup_task() -* will return promptly. +* If a pid belongs to a dead thread group leader, we +* need to add it with the same sid as current (and +* other) threads. */ -if (ckpt_setup_task (ctx, pids_arr->vpgid, pids_arr->vsid) <0 ) +if (ckpt_setup_task (ctx, pids_arr[i].vtgid, sid) <0 )return -1 ; - pids_arr++; +/* +* If pgrp == sid, then the pgrp/sid will already have +* been hashed by now (e.g. by the call above) and the +* ckpt_setup_task() will return promptly. +* If pgrp != sid, then the pgrp 'owner' must have the +* same sid as us: all tasks with same pgrp must have +* their sid matching. +*/ +if (ckpt_setup_task (ctx, pids_arr[i].vpgid, sid) <0 ) +return -1 ; } -/* mark root task(s) */ - ctx->tasks_arr[0 ].flags |=TASK_ROOT ; +/* mark root task(s), and set its "creator" to be zero_task */ +ckpt_init_task (ctx)->flags |=TASK_ROOT ; +ckpt_init_task (ctx)->creator = &zero_task;ckpt_dbg ("total tasks (including ghosts): %d\n" , ctx->tasks_nr);return 0 ; } +static int ckpt_need_pidns (struct ckpt_ctx *ctx) +{ +int i; + +for (i =0 ; i < ctx->pids_nr; i++) { +if (ctx->pids_arr[i].vpid ==0 || + ctx->pids_arr[i].vpgid ==0 || + ctx->pids_arr[i].vsid ==0 ) +return 1 ; + } + +return 0 ; +} +/* * Algorithm DumpForest * "Transparent Checkpoint/Restart of Multiple Processes on Commodity
@@ -1043,10 +1085,20 @@ static int ckpt_set_creator(struct ckpt_ctx *ctx, struct task *task)struct task *creator;if (task ==ckpt_init_task (ctx)) { -ckpt_err ("pid %d: no init creator\n" ,ckpt_init_task (ctx)->pid); +ckpt_err ("pid %d: logical error\n" ,ckpt_init_task (ctx)->pid);return -1 ; } +/* only root_task can have ppid == 0, parent must always exist */ +if (task->ppid ==0 || !parent) { +ckpt_err ("pid %d: invalid ppid %d\n" , task->pid, task->ppid); +return -1 ; + } + +/* sid == 0 must have been inherited from outside the container */ +if (task->sid ==0 ) + session =ckpt_init_task (ctx); +if (task->tgid != task->pid) {/* thread: creator is thread-group-leader */ ckpt_dbg ("pid %d: thread tgid %d\n" , task->pid, task->tgid);
@@ -1073,7 +1125,7 @@ static int ckpt_set_creator(struct ckpt_ctx *ctx, struct task *task) creator = session->phantom; }else {/* first make sure we know the session's creator */ -if (!session->creator && session !=ckpt_init_task (ctx)) { +if (!session->creator) {/* (non-session-leader) recursive: session's creator */ ckpt_dbg ("pid %d: recursive session creator %d\n" , task->pid, task->sid);
@@ -1206,7 +1258,7 @@ static int ckpt_propagate_session(struct ckpt_ctx *ctx, struct task *task)ckpt_dbg ("pid %d: moving up to %d\n" , task->pid, creator->pid); task = creator; -if (!task->creator && task !=ckpt_init_task (ctx)) { +if (!task->creator) {if (ckpt_set_creator (ctx, task) <0 )return -1 ; }
@@ -1389,7 +1441,7 @@ int ckpt_fork_stub(void *data)/* root has some extra work */ if (task->flags &TASK_ROOT ) { - ctx->init_pid =_getpid (); + ctx->root_pid =_getpid ();ckpt_dbg ("root task pid %d\n" ,_getpid ()); }
@@ -1513,13 +1565,13 @@ static int ckpt_fork_feeder(struct ckpt_ctx *ctx)static void ckpt_abort (struct ckpt_ctx *ctx,char *str) {perror (str); -kill (-(ctx->init_pid),SIGKILL ); +kill (-(ctx->root_pid),SIGKILL );exit (1 ); }/* * feeder process: delegates checkpoint image stream to the kernel. -* In '--no-pids' mode, transform the pids array (struct ckpt_hdr_pids) +* In '--no-pids' mode, transform the pids array (struct ckpt_pids) * on the fly and feed the result to the "init" task of the restart */ static int ckpt_do_feeder (void *data)
@@ -1567,7 +1619,7 @@ static int ckpt_do_feeder(void *data) }/* -* ckpt_adjust_pids: transform the pids array (struct ckpt_hdr_pids) by +* ckpt_adjust_pids: transform the pids array (struct ckpt_pids) by * substituing actual pid values for original pid values. * * Collect pids reported by the newly created tasks; each task sends
Signature · 1 line
1.6.0.4
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): 1252074054-22241-6-git-send-email-orenl-RdfvBDnrOixBDgjK7y7TUQ@public.gmane.org
Serge E. Hallyn
Acked-byRe: [PATCH 5/6] c/r: correctly restore pgid
Quoted 94 lines · Oren Laadan
The main challenge with restoring the pgid of tasks is that the original "owner" (the process with that pid) might have exited already. I call these "ghost" pgids. 'mktree' does create these processes, but they then exit without participating in the restart.
To solve this, this patch introduces a
RESTART_GHOSTflag, used for "ghost" owners that are created only to pass their pgid to other tasks. ('mktree' now makes them call restart(2) instead of exiting).When a "ghost" task calls restart(2), it will be placed on a wait queue until the restart completes and then exit. This guarantees that the pgid that it owns remains available for all (regular) restarting tasks for when they need it.
Regular tasks perform the restart as before, except that they also now restore their old pgrp, which is guaranteed to exist.
Changelog [v1]:
- Verify that pgid owner is a thread-group-leader.
- Handle the case of pgid/sid == 0 using root's parent pid-nsSigned-off-by Oren Laadan <[email protected]>--- checkpoint/process.c | 106 ++++++++++++++++++++++++- checkpoint/restart.c | 158 ++++++++++++++++++++++++++------------ checkpoint/sys.c | 3 +- include/linux/checkpoint.h | 11 ++- include/linux/checkpoint_hdr.h | 3 + include/linux/checkpoint_types.h | 6 +- 6 files changed, 230 insertions(+), 57 deletions(-)checkpoint/process.c +51
-0@@ -23,6 +23,57 @@#include <linux/syscalls.h> +pid_t ckpt_pid_nr (struct ckpt_ctx *ctx,struct pid *pid) +{ +return pid ?pid_nr_ns (pid, ctx->root_nsproxy->pid_ns) :CKPT_PID_NULL ; +} + +/* must be called with tasklist_lock or rcu_read_lock() held */ +struct pid *_ckpt_find_pgrp (struct ckpt_ctx *ctx,pid_t pgid) +{ +struct task_struct *p; +struct pid *pgrp; + +if (pgid ==0 ) { +/* +* At checkpoint the pgid owner lived in an ancestor +* pid-ns. The best we can do (sanely and safely) is +* to examine the parent of this restart's root: if in +* a distinct pid-ns, use its pgrp; otherwise fail. +*/ + p = ctx->root_task->real_parent; +if (p->nsproxy->pid_ns == current->nsproxy->pid_ns) +return NULL ; + pgrp =task_pgrp (p); + }else { +/* +* Find the owner process of this pgid (it must exist +* if pgrp exists). It must be a thread group leader. +*/ + pgrp =find_vpid (pgid); + p =pid_task (pgrp,PIDTYPE_PID ); +if (!p || !thread_group_leader (p)) +return NULL ; +/* +* The pgrp must "belong" to our restart tree (compare +* p->checkpoint_ctx to ours). This prevents malicious +* input from (guessing and) using unrelated pgrps. If +* the owner is dead, then it doesn't have a context, +* so instead compare against its (real) parent's. +*/ +if (p->exit_state ==EXIT_ZOMBIE ) + p = p->real_parent; +if (p->checkpoint_ctx != ctx) +return NULL ; + } + +if (task_session (current) !=task_session (p)) +return NULL ; + +return pgrp; +} + +#ifdef CONFIG_FUTEX static void save_task_robust_futex_list (struct ckpt_hdr_task *h,struct task_struct *t)
checkpoint/process.c static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) · line 148
Quoted 6 lines
checkpoint/process.c +1
-2@@ -94,8 +145,8 @@ static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) h->exit_signal = t->exit_signal; h->pdeath_signal = t->pdeath_signal; - h->set_child_tid = t->set_child_tid; - h->clear_child_tid = t->clear_child_tid; + h->set_child_tid = (unsigned long ) t->set_child_tid;
note that set_child_tid is an int (signed), not a long. Same on x86, but not on other arches. Shouldn't lose info so could be worse.
On the whole,
-serge
A message that is not in the archive (sent off-list or not yet received): 20090908030445.GB16460-r/[email protected]
Re: [PATCH 5/6] c/r: correctly restore pgid
Quoted 98 lines · Serge E. Hallyn
Quoted 97 lines · Oren Laadan
The main challenge with restoring the pgid of tasks is that the original "owner" (the process with that pid) might have exited already. I call these "ghost" pgids. 'mktree' does create these processes, but they then exit without participating in the restart.
To solve this, this patch introduces a
RESTART_GHOSTflag, used for "ghost" owners that are created only to pass their pgid to other tasks. ('mktree' now makes them call restart(2) instead of exiting).When a "ghost" task calls restart(2), it will be placed on a wait queue until the restart completes and then exit. This guarantees that the pgid that it owns remains available for all (regular) restarting tasks for when they need it.
Regular tasks perform the restart as before, except that they also now restore their old pgrp, which is guaranteed to exist.
Changelog [v1]:
- Verify that pgid owner is a thread-group-leader.
- Handle the case of pgid/sid == 0 using root's parent pid-nsSigned-off-by Oren Laadan <[email protected]>--- checkpoint/process.c | 106 ++++++++++++++++++++++++- checkpoint/restart.c | 158 ++++++++++++++++++++++++++------------ checkpoint/sys.c | 3 +- include/linux/checkpoint.h | 11 ++- include/linux/checkpoint_hdr.h | 3 + include/linux/checkpoint_types.h | 6 +- 6 files changed, 230 insertions(+), 57 deletions(-)checkpoint/process.c +51
-0@@ -23,6 +23,57 @@#include <linux/syscalls.h> +pid_t ckpt_pid_nr (struct ckpt_ctx *ctx,struct pid *pid) +{ +return pid ?pid_nr_ns (pid, ctx->root_nsproxy->pid_ns) :CKPT_PID_NULL ; +} + +/* must be called with tasklist_lock or rcu_read_lock() held */ +struct pid *_ckpt_find_pgrp (struct ckpt_ctx *ctx,pid_t pgid) +{ +struct task_struct *p; +struct pid *pgrp; + +if (pgid ==0 ) { +/* +* At checkpoint the pgid owner lived in an ancestor +* pid-ns. The best we can do (sanely and safely) is +* to examine the parent of this restart's root: if in +* a distinct pid-ns, use its pgrp; otherwise fail. +*/ + p = ctx->root_task->real_parent; +if (p->nsproxy->pid_ns == current->nsproxy->pid_ns) +return NULL ; + pgrp =task_pgrp (p); + }else { +/* +* Find the owner process of this pgid (it must exist +* if pgrp exists). It must be a thread group leader. +*/ + pgrp =find_vpid (pgid); + p =pid_task (pgrp,PIDTYPE_PID ); +if (!p || !thread_group_leader (p)) +return NULL ; +/* +* The pgrp must "belong" to our restart tree (compare +* p->checkpoint_ctx to ours). This prevents malicious +* input from (guessing and) using unrelated pgrps. If +* the owner is dead, then it doesn't have a context, +* so instead compare against its (real) parent's. +*/ +if (p->exit_state ==EXIT_ZOMBIE ) + p = p->real_parent; +if (p->checkpoint_ctx != ctx) +return NULL ; + } + +if (task_session (current) !=task_session (p)) +return NULL ; + +return pgrp; +} + +#ifdef CONFIG_FUTEX static void save_task_robust_futex_list (struct ckpt_hdr_task *h,struct task_struct *t)@@ -94,8 +145,8 @@ static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) h->exit_signal = t->exit_signal; h->pdeath_signal = t->pdeath_signal;
Quoted 6 lines
checkpoint/process.c static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) · line 148
Quoted 3 lines
checkpoint/process.c +1
-2- h->set_child_tid = t->set_child_tid; - h->clear_child_tid = t->clear_child_tid; + h->set_child_tid = (unsigned long ) t->set_child_tid;note that
set_child_tidis an int (signed), not a long. Same on x86, but not on other arches. Shouldn't lose info so could be worse.
{set,clear}_child_tid are both pointers to user space: it's an address in userspace, so we save it as 'unsigned long'.
{clear,set}_child_tid is defined in include/linux/sched.h ... how can it differ for different archs ?
Quoted 4 lines
On the whole,
Thanks. I got a few fixes for the code piles up and now c/r of 'screen' with a couple of shells is working :)
Oren.
A message that is not in the archive (sent off-list or not yet received): [email protected]
Re: [PATCH 5/6] c/r: correctly restore pgid
Quoted 108 lines · Oren Laadan
Quoted 98 lines · Serge E. Hallyn
Quoted 97 lines · Oren Laadan
The main challenge with restoring the pgid of tasks is that the original "owner" (the process with that pid) might have exited already. I call these "ghost" pgids. 'mktree' does create these processes, but they then exit without participating in the restart.
To solve this, this patch introduces a
RESTART_GHOSTflag, used for "ghost" owners that are created only to pass their pgid to other tasks. ('mktree' now makes them call restart(2) instead of exiting).When a "ghost" task calls restart(2), it will be placed on a wait queue until the restart completes and then exit. This guarantees that the pgid that it owns remains available for all (regular) restarting tasks for when they need it.
Regular tasks perform the restart as before, except that they also now restore their old pgrp, which is guaranteed to exist.
Changelog [v1]:
- Verify that pgid owner is a thread-group-leader.
- Handle the case of pgid/sid == 0 using root's parent pid-nsSigned-off-by Oren Laadan <[email protected]>--- checkpoint/process.c | 106 ++++++++++++++++++++++++- checkpoint/restart.c | 158 ++++++++++++++++++++++++++------------ checkpoint/sys.c | 3 +- include/linux/checkpoint.h | 11 ++- include/linux/checkpoint_hdr.h | 3 + include/linux/checkpoint_types.h | 6 +- 6 files changed, 230 insertions(+), 57 deletions(-)checkpoint/process.c +51
-0@@ -23,6 +23,57 @@#include <linux/syscalls.h> +pid_t ckpt_pid_nr (struct ckpt_ctx *ctx,struct pid *pid) +{ +return pid ?pid_nr_ns (pid, ctx->root_nsproxy->pid_ns) :CKPT_PID_NULL ; +} + +/* must be called with tasklist_lock or rcu_read_lock() held */ +struct pid *_ckpt_find_pgrp (struct ckpt_ctx *ctx,pid_t pgid) +{ +struct task_struct *p; +struct pid *pgrp; + +if (pgid ==0 ) { +/* +* At checkpoint the pgid owner lived in an ancestor +* pid-ns. The best we can do (sanely and safely) is +* to examine the parent of this restart's root: if in +* a distinct pid-ns, use its pgrp; otherwise fail. +*/ + p = ctx->root_task->real_parent; +if (p->nsproxy->pid_ns == current->nsproxy->pid_ns) +return NULL ; + pgrp =task_pgrp (p); + }else { +/* +* Find the owner process of this pgid (it must exist +* if pgrp exists). It must be a thread group leader. +*/ + pgrp =find_vpid (pgid); + p =pid_task (pgrp,PIDTYPE_PID ); +if (!p || !thread_group_leader (p)) +return NULL ; +/* +* The pgrp must "belong" to our restart tree (compare +* p->checkpoint_ctx to ours). This prevents malicious +* input from (guessing and) using unrelated pgrps. If +* the owner is dead, then it doesn't have a context, +* so instead compare against its (real) parent's. +*/ +if (p->exit_state ==EXIT_ZOMBIE ) + p = p->real_parent; +if (p->checkpoint_ctx != ctx) +return NULL ; + } + +if (task_session (current) !=task_session (p)) +return NULL ; + +return pgrp; +} + +#ifdef CONFIG_FUTEX static void save_task_robust_futex_list (struct ckpt_hdr_task *h,struct task_struct *t)@@ -94,8 +145,8 @@ static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) h->exit_signal = t->exit_signal; h->pdeath_signal = t->pdeath_signal;Quoted 6 lines
checkpoint/process.c static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) · line 148
Quoted 3 lines
checkpoint/process.c +1
-2- h->set_child_tid = t->set_child_tid; - h->clear_child_tid = t->clear_child_tid; + h->set_child_tid = (unsigned long ) t->set_child_tid;note that
set_child_tidis an int (signed), not a long. Same on x86, but not on other arches. Shouldn't lose info so could be worse.
Quoted 5 lines
{set,clear}_child_tid are both pointers to user space: it's an address in userspace, so we save it as 'unsigned long'.
{clear,set}_child_tid is defined in
include/linux/sched.h... how can it differ for different archs ?
sizeof long differs for different archs. Not the type of x_child_tid.
Quoted 7 lines
Quoted 4 lines
On the whole,
Thanks. I got a few fixes for the code piles up and now c/r of 'screen' with a couple of shells is working :)
Cool!
A message that is not in the archive (sent off-list or not yet received): 20090908140056.GA873-r/[email protected]
Re: [PATCH 5/6] c/r: correctly restore pgid
Quoted 106 lines · Serge E. Hallyn
Quoted 105 lines · Oren Laadan
Quoted 103 lines · Serge E. Hallyn
Quoted 94 lines · Oren Laadan
The main challenge with restoring the pgid of tasks is that the original "owner" (the process with that pid) might have exited already. I call these "ghost" pgids. 'mktree' does create these processes, but they then exit without participating in the restart.
To solve this, this patch introduces a
RESTART_GHOSTflag, used for "ghost" owners that are created only to pass their pgid to other tasks. ('mktree' now makes them call restart(2) instead of exiting).When a "ghost" task calls restart(2), it will be placed on a wait queue until the restart completes and then exit. This guarantees that the pgid that it owns remains available for all (regular) restarting tasks for when they need it.
Regular tasks perform the restart as before, except that they also now restore their old pgrp, which is guaranteed to exist.
Changelog [v1]:
- Verify that pgid owner is a thread-group-leader.
- Handle the case of pgid/sid == 0 using root's parent pid-nsSigned-off-by Oren Laadan <[email protected]>--- checkpoint/process.c | 106 ++++++++++++++++++++++++- checkpoint/restart.c | 158 ++++++++++++++++++++++++++------------ checkpoint/sys.c | 3 +- include/linux/checkpoint.h | 11 ++- include/linux/checkpoint_hdr.h | 3 + include/linux/checkpoint_types.h | 6 +- 6 files changed, 230 insertions(+), 57 deletions(-)checkpoint/process.c +51
-0@@ -23,6 +23,57 @@#include <linux/syscalls.h> +pid_t ckpt_pid_nr (struct ckpt_ctx *ctx,struct pid *pid) +{ +return pid ?pid_nr_ns (pid, ctx->root_nsproxy->pid_ns) :CKPT_PID_NULL ; +} + +/* must be called with tasklist_lock or rcu_read_lock() held */ +struct pid *_ckpt_find_pgrp (struct ckpt_ctx *ctx,pid_t pgid) +{ +struct task_struct *p; +struct pid *pgrp; + +if (pgid ==0 ) { +/* +* At checkpoint the pgid owner lived in an ancestor +* pid-ns. The best we can do (sanely and safely) is +* to examine the parent of this restart's root: if in +* a distinct pid-ns, use its pgrp; otherwise fail. +*/ + p = ctx->root_task->real_parent; +if (p->nsproxy->pid_ns == current->nsproxy->pid_ns) +return NULL ; + pgrp =task_pgrp (p); + }else { +/* +* Find the owner process of this pgid (it must exist +* if pgrp exists). It must be a thread group leader. +*/ + pgrp =find_vpid (pgid); + p =pid_task (pgrp,PIDTYPE_PID ); +if (!p || !thread_group_leader (p)) +return NULL ; +/* +* The pgrp must "belong" to our restart tree (compare +* p->checkpoint_ctx to ours). This prevents malicious +* input from (guessing and) using unrelated pgrps. If +* the owner is dead, then it doesn't have a context, +* so instead compare against its (real) parent's. +*/ +if (p->exit_state ==EXIT_ZOMBIE ) + p = p->real_parent; +if (p->checkpoint_ctx != ctx) +return NULL ; + } + +if (task_session (current) !=task_session (p)) +return NULL ; + +return pgrp; +} + +#ifdef CONFIG_FUTEX static void save_task_robust_futex_list (struct ckpt_hdr_task *h,struct task_struct *t)checkpoint/process.c static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) · line 148
Quoted 6 lines
checkpoint/process.c +1
-2@@ -94,8 +145,8 @@ static int checkpoint_task_struct(struct ckpt_ctx *ctx, struct task_struct *t) h->exit_signal = t->exit_signal; h->pdeath_signal = t->pdeath_signal; - h->set_child_tid = t->set_child_tid; - h->clear_child_tid = t->clear_child_tid; + h->set_child_tid = (unsigned long ) t->set_child_tid;note that
set_child_tidis an int (signed), not a long. Same on x86, but not on other arches. Shouldn't lose info so could be worse.
Quoted 7 lines
Quoted 5 lines
{set,clear}_child_tid are both pointers to user space: it's an address in userspace, so we save it as 'unsigned long'.
{clear,set}_child_tid is defined in
include/linux/sched.h... how can it differ for different archs ?sizeof long differs for different archs. Not the type of
x_child_tid.
Sure. In all ckpt headers, all pointers allway get a __u64 regardless of arch, to cover both 32- and 64-bit.
Oren.
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
On the whole,
Thanks. I got a few fixes for the code piles up and now c/r of 'screen' with a couple of shells is working :)
Cool!
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): 1252074054-22241-7-git-send-email-orenl-RdfvBDnrOixBDgjK7y7TUQ@public.gmane.org
Serge E. Hallyn
Acked-byRe: [PATCH 6/6] c/r: support for controlling terminal and job control
Quoted 8 lines · Oren Laadan
Add checkpoint/restart of controlling terminal:
current->signal->tty. This is only done for session leaders.If the session leader belongs to the ancestor pid-ns, then checkpoint skips this tty; On restart, it will not be restored, and whatever tty is in place from parent pid-ns (at restart) will be inherited.
Signed-off-by Oren Laadan <[email protected]>
Not exactly my area, but looks good
Quoted 264 lines
--- checkpoint/signal.c | 78 ++++++++++++++++++++++++++++++++++++++- drivers/char/tty_io.c | 33 +++++++++++++---- include/linux/checkpoint.h | 1 + include/linux/checkpoint_hdr.h | 6 +++ include/linux/tty.h | 5 +++ 5 files changed, 114 insertions(+), 9 deletions(-)checkpoint/signal.c +76
-2@@ -315,11 +315,12 @@ static int checkpoint_signal(struct ckpt_ctx *ctx, struct task_struct *t)struct ckpt_hdr_signal *h;struct signal_struct *signal;struct sigpending shared_pending; +struct tty_struct *tty =NULL ;struct rlimit *rlim;struct timeval tval;cputime_t cputime;unsigned long flags; -int i, ret; +int i, ret =0 ; h =ckpt_hdr_get_type (ctx,sizeof (*h),CKPT_HDR_SIGNAL );if (!h)@@ -398,9 +399,34 @@ static int checkpoint_signal(struct ckpt_ctx *ctx, struct task_struct *t)cputime_to_timeval (signal->it_prof_incr, &tval); h->it_prof_incr =timeval_to_ns (&tval); +/* tty */ +if (signal->leader) { + h->tty_old_pgrp =ckpt_pid_nr (ctx, signal->tty_old_pgrp); + tty =tty_kref_get (signal->tty); +if (tty) { +/* irq is already disabled */ +spin_lock (&tty->ctrl_lock); + h->tty_pgrp =ckpt_pid_nr (ctx, tty->pgrp); +spin_unlock (&tty->ctrl_lock); +tty_kref_put (tty); + } + } +unlock_task_sighand (t, &flags); - ret =ckpt_write_obj (ctx, &h->h); +/* +* If the session is in an ancestor namespace, skip this tty +* and set tty_objref = 0. It will not be explicitly restored, +* but rather inherited from parent pid-ns at restart time. +*/ +if (tty &&ckpt_pid_nr (ctx, tty->session) >0 ) { + h->tty_objref =checkpoint_obj (ctx, tty,CKPT_OBJ_TTY ); +if (h->tty_objref <0 ) + ret = h->tty_objref; + } + +if (!ret) + ret =ckpt_write_obj (ctx, &h->h);if (!ret) ret =checkpoint_sigpending (ctx, &shared_pending);@@ -471,8 +497,10 @@ static int restore_signal(struct ckpt_ctx *ctx)struct ckpt_hdr_signal *h;struct sigpending new_pending;struct sigpending *pending; +struct tty_struct *tty =NULL ;struct itimerval itimer;struct rlimit rlim; +struct pid *pgrp;int i, ret; h =ckpt_read_obj_type (ctx,sizeof (*h),CKPT_HDR_SIGNAL );@@ -492,6 +520,36 @@ static int restore_signal(struct ckpt_ctx *ctx)if (ret <0 ) goto out; +/* tty - session */ +if (h->tty_objref) { + tty =ckpt_obj_fetch (ctx, h->tty_objref,CKPT_OBJ_TTY ); +if (IS_ERR (tty)) { + ret =PTR_ERR (tty); + goto out; + } +/* this will fail unless we're the session leader */ + ret =tiocsctty (tty,0 ); +if (ret <0 ) + goto out; +/* now restore the foreground group (job control) */ +if (h->tty_pgrp) { + ret =do_tiocspgrp (tty,tty_pair_get_tty (tty), + h->tty_pgrp); +if (ret <0 ) + goto out; + } + }else { +/* +* If tty_objref isn't set, we _keep_ whatever tty we +* already have as a ctty. Why does this make sense ? +* - If our session is "within" the restart context, +* then that session has no controlling terminal. +* - If out session is "outside" the restart context, +* then we're like to keep whatever we inherit from +* the parent pid-ns. +*/ + } +/* * Reset real/virt/prof itimer (in case they were set), to * prevent unwanted signals after flushing current signals @@ -503,7 +561,23 @@ static int restore_signal(struct ckpt_ctx *ctx)do_setitimer (ITIMER_VIRTUAL , &itimer,NULL );do_setitimer (ITIMER_PROF , &itimer,NULL ); +/* tty - tty_old_pgrp */ +if (h->tty_old_pgrp) { + ret = -EINVAL ; +if (!current->signal->leader) + goto out; +rcu_read_lock (); + pgrp =get_pid (_ckpt_find_pgrp (ctx, h->tty_old_pgrp)); +rcu_read_unlock (); +if (!pgrp) + goto out; + } +spin_lock_irq (¤t->sighand->siglock); +/* tty - tty_old_pgrp */ +put_pid (current->signal->tty_old_pgrp); + current->signal->tty_old_pgrp = pgrp; +/* pending signals */ pending = ¤t->signal->shared_pending;flush_sigqueue (pending); pending->signal = new_pending.signal;drivers/char/tty_io.c +26
-7@@ -2132,7 +2132,7 @@ static int fionbio(struct file *file, int __user *p)* Takes ->siglock() when updating signal->tty */ -static int tiocsctty (struct tty_struct *tty,int arg) +int tiocsctty (struct tty_struct *tty,int arg) {int ret =0 ;if (current->signal->leader && (task_session (current) == tty->session))@@ -2221,10 +2221,10 @@ static int tiocgpgrp(struct tty_struct *tty, struct tty_struct *real_tty, pid_t }/** -* tiocspgrp - attempt to set process group +* do_tiocspgrp - attempt to set process group * @tty: tty passed by user * @real_tty: tty side device matching tty passed by user -* @p: pid pointer +* @pid: pgrp_nr * * Set the process group of the tty to the session passed. Only * permitted where the tty session is our session. @@ -2232,10 +2232,10 @@ static int tiocgpgrp(struct tty_struct *tty, struct tty_struct *real_tty, pid_t* Locking: RCU, ctrl lock */ -static int tiocspgrp (struct tty_struct *tty,struct tty_struct *real_tty,pid_t __user *p) +int do_tiocspgrp (struct tty_struct *tty, +struct tty_struct *real_tty,pid_t pgrp_nr) {struct pid *pgrp; -pid_t pgrp_nr;int retval =tty_check_change (real_tty);unsigned long flags;@@ -2247,8 +2247,6 @@ static int tiocspgrp(struct tty_struct *tty, struct tty_struct *real_tty, pid_t (current->signal->tty != real_tty) || (real_tty->session != task_session(current))) return -ENOTTY; -if (get_user (pgrp_nr, p)) -return -EFAULT ;if (pgrp_nr <0 )return -EINVAL ;rcu_read_lock ();@@ -2270,6 +2268,27 @@ out_unlock: }/** +* tiocspgrp - attempt to set process group +* @tty: tty passed by user +* @real_tty: tty side device matching tty passed by user +* @p: pid pointer +* +* Set the process group of the tty to the session passed. Only +* permitted where the tty session is our session. +* +* Locking: RCU, ctrl lock +*/ + +static int tiocspgrp (struct tty_struct *tty,struct tty_struct *real_tty,pid_t __user *p) +{ +pid_t pgrp_nr; + +if (get_user (pgrp_nr, p)) +return -EFAULT ; +return do_tiocspgrp (tty, real_tty, pgrp_nr); +} + +/** * tiocgsid - get session id * @tty: tty passed by user * @real_tty: tty side of the tty pased by the user if a pty else the tty include/linux/checkpoint.h +1
-0@@ -87,6 +87,7 @@ extern char *ckpt_fill_fname(struct path *path, struct path *root,/* pids */ extern pid_t ckpt_pid_nr (struct ckpt_ctx *ctx,struct pid *pid); +extern struct pid *_ckpt_find_pgrp (struct ckpt_ctx *ctx,pid_t pgid);/* socket functions */ extern int ckpt_sock_getnames(struct ckpt_ctx *ctx,include/linux/checkpoint_hdr.h +6
-0@@ -555,13 +555,19 @@ struct ckpt_rlimit {struct ckpt_hdr_signal {struct ckpt_hdr h; +/* rlimit */ struct ckpt_rlimit rlim[CKPT_RLIM_NLIMITS ]; +/* itimer */ __u64 it_real_value;__u64 it_real_incr;__u64 it_virt_value;__u64 it_virt_incr;__u64 it_prof_value;__u64 it_prof_incr; +/* tty */ +__s32 tty_objref; +__s32 tty_pgrp; +__s32 tty_old_pgrp; } __attribute__((aligned (8 )));struct ckpt_hdr_signal_task {include/linux/tty.h +5
-0@@ -467,6 +467,11 @@ extern void tty_ldisc_begin(void);/* This last one is just for the tty layer internals and shouldn't be used elsewhere */ extern void tty_ldisc_enable (struct tty_struct *tty); +/* These are for checkpoint/restart */ +extern int tiocsctty (struct tty_struct *tty,int arg); +extern int do_tiocspgrp (struct tty_struct *tty, +struct tty_struct *real_tty,pid_t pgrp_nr); +#ifdef CONFIG_CHECKPOINT struct ckpt_ctx ;struct ckpt_hdr_file ;Signature · 1 line
1.6.0.4