Patchsets related to hibernation resume: - enhancement to make the use of an existing resume file more general - enhance name_to_dev_t to ignore trailing newlines coming from userspace.
Both patches are based on the 3.12-rc3 tag. This was tested on a Pandaboard with partial hibernation support, and compiled for x86.
[PATCH 1/2] init/do_mounts.c: ignore final \n in name_to_dev_t init/do_mounts.c | 23 ++++++++++++++++++----- 1 file changed, 18 insertions(+), 5 deletions(-)
Changes name_to_dev_t to handle a trailing newline in the input buffer, which will allow name_to_dev_t to be used directly with user buffers without requiring a copy. Also adds a const to the name parameter which reflects how name_to_dev_t is treating the input buffer currently. This also allows direct use of user buffers (from resume_store for example).
[PATCH 2/2] PM / Hibernate: use name_to_dev_t to parse resume kernel/power/hibernate.c | 15 ++++----------- 1 file changed, 4 insertions(+), 11 deletions(-)
Use name_to_dev_t to parse the /sys/power/resume file making the syntax more flexible. It supports the previous use syntax and additionally can support other formats such as /dev/devicenode and UUID= formats.
By changing /sys/debug/resume to accept the same syntax as the resume=device parameter, we can parse the resume=device in the initrd init script and use the resume device directly from the kernel command line.
Changes in v3: -------------- * Dropped documentation patch as it went in through trivial * Added patch for name_to_dev_t to support directly parsing userspace buffer
Changes in v2: -------------- * Added check for null return of kstrndup in hibernate.c
Thanks,
Sebastian
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
Signed-off-by: Sebastian Capella sebastian.capella@linaro.org Reviewed-by: Pavel Machek pavel@ucw.cz Cc: "Eric W. Biederman" ebiederm@xmission.com Cc: Serge Hallyn serge.hallyn@canonical.com Cc: Andrew Morton akpm@linux-foundation.org Cc: Stephen Warren swarren@nvidia.com Cc: Jens Axboe axboe@kernel.dk Cc: Greg Kroah-Hartman gregkh@linuxfoundation.org Cc: Al Viro viro@zeniv.linux.org.uk --- include/linux/mount.h | 2 +- init/do_mounts.c | 25 +++++++++++++++++++------ 2 files changed, 20 insertions(+), 7 deletions(-)
diff --git a/include/linux/mount.h b/include/linux/mount.h index 38cd98f..fdbb3e6 100644 --- a/include/linux/mount.h +++ b/include/linux/mount.h @@ -77,6 +77,6 @@ extern struct vfsmount *vfs_kern_mount(struct file_system_type *type, extern void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list); extern void mark_mounts_for_expiry(struct list_head *mounts);
-extern dev_t name_to_dev_t(char *name); +extern dev_t name_to_dev_t(const char *name);
#endif /* _LINUX_MOUNT_H */ diff --git a/init/do_mounts.c b/init/do_mounts.c index a51cddc..69d74ff 100644 --- a/init/do_mounts.c +++ b/init/do_mounts.c @@ -145,6 +145,13 @@ static dev_t devt_from_partuuid(const char *uuid_str) clear_root_wait = true; goto done; } + if (uuid_str[cmp.len - 1] == '\n') { + cmp.len--; + if (!cmp.len) { + clear_root_wait = true; + goto done; + } + }
dev = class_find_device(&block_class, NULL, &cmp, &match_dev_by_uuid); @@ -204,12 +211,13 @@ done: * bangs. */
-dev_t name_to_dev_t(char *name) +dev_t name_to_dev_t(const char *name) { char s[32]; char *p; dev_t res = 0; int part; + int n;
#ifdef CONFIG_BLOCK if (strncmp(name, "PARTUUID=", 9) == 0) { @@ -230,7 +238,7 @@ dev_t name_to_dev_t(char *name) goto fail; } else { res = new_decode_dev(simple_strtoul(name, &p, 16)); - if (*p) + if (*p && *p != '\n') goto fail; } goto done; @@ -238,15 +246,20 @@ dev_t name_to_dev_t(char *name)
name += 5; res = Root_NFS; - if (strcmp(name, "nfs") == 0) + if (strncmp(name, "nfs", 3) == 0) goto done; res = Root_RAM0; - if (strcmp(name, "ram") == 0) + if (strncmp(name, "ram", 3) == 0) goto done;
- if (strlen(name) > 31) + n = strlen(name); + if (n != 0 && name[n - 1] == '\n') + n--; + if (n > 31) goto fail; - strcpy(s, name); + strncpy(s, name, n); + s[n] = '\0'; + for (p = s; *p; p++) if (*p == '/') *p = '!';
Use the name_to_dev_t call to parse the device name echo'd to to /sys/power/resume. This imitates the method used in hibernate.c in software_resume, and allows the resume partition to be specified using other equivalent device formats as well. By allowing /sys/debug/resume to accept the same syntax as the resume=device parameter, we can parse the resume=device in the init script and use the resume device directly from the kernel command line.
Signed-off-by: Sebastian Capella sebastian.capella@linaro.org Acked-by: Pavel Machek pavel@ucw.cz Cc: Len Brown len.brown@intel.com Cc: "Rafael J. Wysocki" rjw@sisk.pl --- kernel/power/hibernate.c | 15 ++++----------- 1 file changed, 4 insertions(+), 11 deletions(-)
diff --git a/kernel/power/hibernate.c b/kernel/power/hibernate.c index c9c759d..a29d2a7 100644 --- a/kernel/power/hibernate.c +++ b/kernel/power/hibernate.c @@ -972,16 +972,11 @@ static ssize_t resume_show(struct kobject *kobj, struct kobj_attribute *attr, static ssize_t resume_store(struct kobject *kobj, struct kobj_attribute *attr, const char *buf, size_t n) { - unsigned int maj, min; dev_t res; - int ret = -EINVAL;
- if (sscanf(buf, "%u:%u", &maj, &min) != 2) - goto out; - - res = MKDEV(maj,min); - if (maj != MAJOR(res) || min != MINOR(res)) - goto out; + res = name_to_dev_t(buf); + if (res == 0) + return -EINVAL;
lock_system_sleep(); swsusp_resume_device = res; @@ -989,9 +984,7 @@ static ssize_t resume_store(struct kobject *kobj, struct kobj_attribute *attr, printk(KERN_INFO "PM: Starting manual resume from disk\n"); noresume = 0; software_resume(); - ret = n; - out: - return ret; + return n; }
power_attr(resume);
On Thu, 3 Oct 2013 14:10:37 -0700 Sebastian Capella sebastian.capella@linaro.org wrote:
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
We have lib/string.c:strim() - perhaps this patch would be neater if it were to use it?
Quoting Andrew Morton (2013-10-03 14:15:23)
On Thu, 3 Oct 2013 14:10:37 -0700 Sebastian Capella sebastian.capella@linaro.org wrote:
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
We have lib/string.c:strim() - perhaps this patch would be neater if it were to use it?
Hi Morton,
I was intending to respect the const handling of the input buffer.
The actual buffer in this case is not really const as it comes from the file buffering, but removing the const requires changing the store function defined in the kobj_attribute, and would propagate to many areas in the kernel.
Modifying the buffer and removing the const was also suggested by Pavel. After some discussion I posted this version which did not change the buffer or the prototype.
Please let me know if the preference is to modify the store function definition.
I'll prepare a patchset that removes the consts to see how much is changed.
Thanks,
Sebastian
Quoting Sebastian Capella (2013-10-03 14:42:46)
Quoting Andrew Morton (2013-10-03 14:15:23)
On Thu, 3 Oct 2013 14:10:37 -0700 Sebastian Capella sebastian.capella@linaro.org wrote:
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
We have lib/string.c:strim() - perhaps this patch would be neater if it were to use it?
Hi Morton,
I was intending to respect the const handling of the input buffer.
The actual buffer in this case is not really const as it comes from the file buffering, but removing the const requires changing the store function defined in the kobj_attribute, and would propagate to many areas in the kernel.
Modifying the buffer and removing the const was also suggested by Pavel. After some discussion I posted this version which did not change the buffer or the prototype.
Please let me know if the preference is to modify the store function definition.
I'll prepare a patchset that removes the consts to see how much is changed.
Thanks,
Sebastian
Hi Andrew,
Sorry for calling you Morton earlier.
I looked into removing the const from the store function, but I'm not sure this is the right idea, so I'm going to shelf that for now.
Please let me know your thoughts.
Thanks,
Sebastian
Quoting Sebastian Capella (2013-10-03 16:47:35)
Quoting Sebastian Capella (2013-10-03 14:42:46)
Quoting Andrew Morton (2013-10-03 14:15:23)
On Thu, 3 Oct 2013 14:10:37 -0700 Sebastian Capella sebastian.capella@linaro.org wrote:
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
We have lib/string.c:strim() - perhaps this patch would be neater if it were to use it?
Hi Morton,
I was intending to respect the const handling of the input buffer.
The actual buffer in this case is not really const as it comes from the file buffering, but removing the const requires changing the store function defined in the kobj_attribute, and would propagate to many areas in the kernel.
Modifying the buffer and removing the const was also suggested by Pavel. After some discussion I posted this version which did not change the buffer or the prototype.
Please let me know if the preference is to modify the store function definition.
I'll prepare a patchset that removes the consts to see how much is changed.
Thanks,
Sebastian
Hi Andrew,
Sorry for calling you Morton earlier.
I looked into removing the const from the store function, but I'm not sure this is the right idea, so I'm going to shelf that for now.
Please let me know your thoughts.
Thanks,
Sebastian
Ping...
Hi Andrew,
Do you have any feedback on this?
Below are the three options considered thus far. Do you have any additional suggestions or preferences?
Thanks,
Sebastian
1) copy buffer, remove \n. - v1 patch did this - alternatively could use an array on the stack or a preallocated global Pros: . cleanest change Cons: . adds memcpy 2) make name_to_dev_t work with readonly buffer to ignore \n - v2 and v3 patches do this Pros: . no memcpy, no big modifications to unrelated code Cons: . seems more appropriate to harden store functions to user input than name_to_dev_t. . a little complicated 3) remove const from buffer and modify contents in place to remove \n - remove const from sysfs_ops.store, modify dependent definitions - remove const from kobj_attribute.store, modify dependent definitions Pros: . no memcpy Cons: . a lot of modifications . const contract to not modify the input buffer seems the right way.
Sebastian Capella sebastian.capella@linaro.org writes:
Quoting Sebastian Capella (2013-10-03 16:47:35)
Quoting Sebastian Capella (2013-10-03 14:42:46)
Quoting Andrew Morton (2013-10-03 14:15:23)
On Thu, 3 Oct 2013 14:10:37 -0700 Sebastian Capella sebastian.capella@linaro.org wrote:
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
We have lib/string.c:strim() - perhaps this patch would be neater if it were to use it?
Hi Morton,
I was intending to respect the const handling of the input buffer.
The actual buffer in this case is not really const as it comes from the file buffering, but removing the const requires changing the store function defined in the kobj_attribute, and would propagate to many areas in the kernel.
Modifying the buffer and removing the const was also suggested by Pavel. After some discussion I posted this version which did not change the buffer or the prototype.
Please let me know if the preference is to modify the store function definition.
I'll prepare a patchset that removes the consts to see how much is changed.
Thanks,
Sebastian
Hi Andrew,
Sorry for calling you Morton earlier.
I looked into removing the const from the store function, but I'm not sure this is the right idea, so I'm going to shelf that for now.
Please let me know your thoughts.
Thanks,
Sebastian
Ping...
Hi Andrew,
Do you have any feedback on this?
Below are the three options considered thus far. Do you have any additional suggestions or preferences?
What is wrong with requiring userspace to use echo -n ?
That by far seems the simplest and least error prone solution.
Eric
Quoting Eric W. Biederman (2013-10-10 15:47:45)
Sebastian Capella sebastian.capella@linaro.org writes:
Quoting Sebastian Capella (2013-10-03 16:47:35)
Quoting Sebastian Capella (2013-10-03 14:42:46)
Quoting Andrew Morton (2013-10-03 14:15:23)
On Thu, 3 Oct 2013 14:10:37 -0700 Sebastian Capella sebastian.capella@linaro.org wrote:
Enhance name_to_dev_t to handle trailing newline characters on device paths. Some inputs to name_to_dev_t may come from userspace where oftentimes a '\n' is appended to the path. Added const to the name buffer in both the function declaration and the prototype to reflect input buffer handling.
By handling trailing newlines in name_to_dev_t, userspace buffers may be directly passed to name_to_dev_t without modification.
We have lib/string.c:strim() - perhaps this patch would be neater if it were to use it?
I was intending to respect the const handling of the input buffer.
I looked into removing the const from the store function, but I'm not sure this is the right idea, so I'm going to shelf that for now.
Please let me know your thoughts.
What is wrong with requiring userspace to use echo -n ?
That by far seems the simplest and least error prone solution.
Eric
Trimmed back some of the history above.
Hi Eric,
Thanks for your comment. While this does solve the problem for that case, I haven't generally seen sysfs files requiring 'echo -n' before. If you look in kernel/power/main.c, you can see that in other cases where you're echoing a string onto a sysfs file, the code will handle the \n.
I expect it would likely be confusing for users. The net effect would be that if you omit the -n, it report something like: write error: Invalid argument as if you'd mistyped the resume file.
Thanks,
Sebastian
Quoting Sebastian Capella (2013-10-10 10:50:10)
Quoting Sebastian Capella (2013-10-03 16:47:35)
Quoting Sebastian Capella (2013-10-03 14:42:46) I looked into removing the const from the store function, but I'm not sure this is the right idea, so I'm going to shelf that for now.
Below are the three options considered thus far. Do you have any additional suggestions or preferences?
- copy buffer, remove \n.
- v1 patch did this
- alternatively could use an array on the stack or a preallocated global Pros: . cleanest change . could use strim() proposed by Andrew Cons: . adds memcpy
- make name_to_dev_t work with readonly buffer to ignore \n
- v2 and v3 patches do this Pros: . no memcpy, no big modifications to unrelated code Cons: . seems more appropriate to harden store functions to user input than name_to_dev_t. . a little complicated
- remove const from buffer and modify contents in place to remove \n
- remove const from sysfs_ops.store, modify dependent definitions
- remove const from kobj_attribute.store, modify dependent definitions Pros: . no memcpy Cons: . a lot of modifications . const contract to not modify the input buffer seems the right way.
Hi Andrew,
Do you have any feedback for me on this?
I'm happy do make any changes you think are correct, but I'm unsure if you're asking me for option #3 above. It's quite an intrusive change, and changes old, established code and I'd like confirmation that's what you'd like before proceeding down that path.
I've submitted patches with both options #1 and #2 above.
Thanks,
Sebastian
Quoting Sebastian Capella (2013-10-22 10:54:14)
Quoting Sebastian Capella (2013-10-10 10:50:10)
Quoting Sebastian Capella (2013-10-03 16:47:35)
Quoting Sebastian Capella (2013-10-03 14:42:46) I looked into removing the const from the store function, but I'm not sure this is the right idea, so I'm going to shelf that for now.
Below are the three options considered thus far. Do you have any additional suggestions or preferences?
- copy buffer, remove \n.
- v1 patch did this
- alternatively could use an array on the stack or a preallocated global Pros: . cleanest change . could use strim() proposed by Andrew Cons: . adds memcpy
- make name_to_dev_t work with readonly buffer to ignore \n
- v2 and v3 patches do this Pros: . no memcpy, no big modifications to unrelated code Cons: . seems more appropriate to harden store functions to user input than name_to_dev_t. . a little complicated
- remove const from buffer and modify contents in place to remove \n
- remove const from sysfs_ops.store, modify dependent definitions
- remove const from kobj_attribute.store, modify dependent definitions Pros: . no memcpy Cons: . a lot of modifications . const contract to not modify the input buffer seems the right way.
Hi Andrew,
Do you have any feedback for me on this?
I'm happy do make any changes you think are correct, but I'm unsure if you're asking me for option #3 above. It's quite an intrusive change, and changes old, established code and I'd like confirmation that's what you'd like before proceeding down that path.
I've submitted patches with both options #1 and #2 above.
Thanks,
Sebastian
Ping.
Sorry for the lapse in attention to this.
Could you please clarify what is needed for this to be acceptable? I'm a little confused about what is being asked of me.
Thanks!
Sebastian
On Tue, 28 Jan 2014 10:59:26 -0800 Sebastian Capella sebastian.capella@linaro.org wrote:
Do you have any feedback for me on this?
I'm happy do make any changes you think are correct, but I'm unsure if you're asking me for option #3 above. It's quite an intrusive change, and changes old, established code and I'd like confirmation that's what you'd like before proceeding down that path.
I've submitted patches with both options #1 and #2 above.
Thanks,
Sebastian
Ping.
Sorry for the lapse in attention to this.
Could you please clarify what is needed for this to be acceptable? I'm a little confused about what is being asked of me.
The problem is that kernel/power/hibernate.c:resume_store() is handed a newline-terminated string, yes? And if it blindly hands that string over to name_to_dev_t(), name_to_dev_t() fails because the string is wrong.
This is an oddity of the sysfs->kernel interface and altering name_to_dev_t doesn't really seem appropriate for this problem - it would be better to fix the caller to pass in the correct string.
Something like...
/* * Clean up a string which may have leading and/or trailing whitespace (as * defined by isspace()) by trimming off that whitespace. Returns an address * which the caller must kfree(), or NULL on error. */ char *strim_copy(const char *s, gfp_t gfp) { char *ret = kstrdup(skip_spaces(s), gfp);
if (ret) strim(ret); return ret; } EXPORT_SYMBOL(strim_copy);
Quoting Andrew Morton (2014-01-28 12:54:42)
The problem is that kernel/power/hibernate.c:resume_store() is handed a newline-terminated string, yes? And if it blindly hands that string over to name_to_dev_t(), name_to_dev_t() fails because the string is wrong.
This is an oddity of the sysfs->kernel interface and altering name_to_dev_t doesn't really seem appropriate for this problem - it would be better to fix the caller to pass in the correct string.
Something like...
/*
- Clean up a string which may have leading and/or trailing whitespace (as
- defined by isspace()) by trimming off that whitespace. Returns an address
- which the caller must kfree(), or NULL on error.
*/ char *strim_copy(const char *s, gfp_t gfp) { char *ret = kstrdup(skip_spaces(s), gfp);
if (ret) strim(ret); return ret;} EXPORT_SYMBOL(strim_copy);
Hi Andrew,
Thanks, this is similar, but tidier than my original patch. I'll fix up and repost the patch with a fix like what you're proposing.
Thanks!
Sebastian
Hi Andrew,
By the way, I do see a call (sysfs_streq) in use for this purpose other places. Sorry, I didn't find it while looking at the original problem. I'm not sure if this is preferable, but it appears to have been added specifically for the strings coming through sysfs.
My preference is copying the string and cleaning it up before passing it to internal functions, even though we incur an allocation.
Thanks,
Sebastian
Hi Andrew,
By the way, I do see a call (sysfs_streq) in use for this purpose other places. Sorry, I didn't find it while looking at the original problem. I'm not sure if this is preferable, but it appears to have been added specifically for the strings coming through sysfs.
My preference is copying the string and cleaning it up before passing it to internal functions, even though we incur an allocation.
Thanks,
Sebastian
On Wed, 29 Jan 2014 10:29:56 -0800 Sebastian Capella sebastian.capella@linaro.org wrote:
Hi Andrew,
By the way, I do see a call (sysfs_streq) in use for this purpose other places. Sorry, I didn't find it while looking at the original problem. I'm not sure if this is preferable, but it appears to have been added specifically for the strings coming through sysfs.
Yes, I wrote it ;)
I didn't think sysfs_streq() is well suited to this problem. And the issue of possibly-null-terminated-strings coming in from userspace is a common one, so it is desirable that we build up the suite of utilities to handle this.
There are probably quite a lot of open-coded \n trimming loops which can be cleaned up using such tools.
grep -r "if .* == '\\n'" .
My preference is copying the string and cleaning it up before passing it to internal functions, even though we incur an allocation.
Yes. Here on the kernel/userspace boundary we are typically running in GFP_KERNEL context and the code is not performance critical - it is a good fit.
linaro-kernel@lists.linaro.org
-
Andrew Morton -
ebiederm＠xmission.com -
Sebastian Capella