libata: sysfs naming
v2 cover libata: sysfs naming
Hi all,
here's my second attempt the align the libata object naming with sysfs. Key point is to introduce an 'ata' bus, which serves to collect all libata object (ata_port, ata_link, and ata_device).
To facilitate that the name of the 'ata_port' object changes from 'ata' to 'port'. To provide backwards compability I've added an additional class object for the ata_port with the name of 'ata'.
This additional object can be disabled with the config option ATA_SYSFS_COMPAT.
As usual, comments and reviews are welcome.
Hannes Reinecke (3):
libata: sysfs naming option
libata: CONFIG_ATA_SYSFS_COMPAT
libata: sanitize PMP link naming
drivers/ata/Kconfig | 10 +++++++ drivers/ata/libata-transport.c | 55 ++++++++++++++++++++++++++++++---- include/linux/libata.h | 54 ++++++++++----------------------- 3 files changed, 76 insertions(+), 43 deletions(-)
Signature · 1 line
2.29.2
PATCH v2 1/3 libata: sysfs naming option
This patch adds a new dummy bus 'ata', which collects all ata device objects like ata_port, ata_link, and ata_dev, and adds an 'ata' prefix to the message log.
For consistency the 'ata_port' name has been changed from 'ata' to 'port'; to cover for sysfs incompabilities I've added a link from the original 'ata' device to the new 'port' device.
--- drivers/ata/libata-transport.c | 38 ++++++++++++++++++++++-- include/linux/libata.h | 54 ++++++++++------------------------ 2 files changed, 51 insertions(+), 41 deletions(-)
drivers/ata/libata-transport.c +35 -3
@@ -81,10 +81,14 @@ struct ata_internal {tdev_to_port ((dev)->parent) -/* Device objects are always created whit link objects */ +/* Device objects are always created with link objects */static int ata_tdev_add (struct ata_device *dev);static void ata_tdev_delete (struct ata_device *dev); +struct bus_type ata_bus_type = { + .name ="ata" , +}; +/* * Hack to allow attributes of the same name in different objects.
@@ -257,12 +261,15 @@ static int ata_tport_match(struct attribute_container *cont,void ata_tport_delete (struct ata_port *ap) {struct device *dev = &ap->tdev; +struct device *cdev = &ap->cdev;ata_tlink_delete (&ap->link); +device_del (cdev);transport_remove_device (dev);device_del (dev);transport_destroy_device (dev); +put_device (cdev);put_device (dev); }
@@ -281,6 +288,7 @@ int ata_tport_add(struct device *parent, {int error;struct device *dev = &ap->tdev; +struct device *cdev = &ap->cdev;device_initialize (dev); dev->type = &ata_port_type;
@@ -288,7 +296,14 @@ int ata_tport_add(struct device *parent, dev->parent = parent;ata_host_get (ap->host); dev->release = ata_tport_release; -dev_set_name (dev,"ata%d" , ap->print_id); + dev->bus = &ata_bus_type; +dev_set_name (dev,"port%d" , ap->print_id); + +device_initialize (cdev); + cdev->parent =get_device (dev); + cdev->class = &ata_port_class.class; +dev_set_name (cdev,"ata%d" , ap->print_id); +transport_setup_device (dev);ata_acpi_bind_port (ap); error =device_add (dev);
@@ -296,6 +311,11 @@ int ata_tport_add(struct device *parent, goto tport_err; } + error =device_add (cdev); +if (error) { + goto cdev_err; + } +device_enable_async_suspend (dev);pm_runtime_set_active (dev);pm_runtime_enable (dev);
@@ -311,11 +331,14 @@ int ata_tport_add(struct device *parent,return 0 ; tport_link_err: +device_del (cdev); + cdev_err:transport_remove_device (dev);device_del (dev); tport_err:transport_destroy_device (dev); +put_device (cdev);put_device (dev);ata_host_put (ap->host);return error;
@@ -444,6 +467,8 @@ int ata_tlink_add(struct ata_link *link)device_initialize (dev); dev->parent = &ap->tdev; dev->release = ata_tlink_release; + dev->bus = &ata_bus_type; +if (ata_is_host_link (link))dev_set_name (dev,"link%d" , ap->print_id);else
@@ -695,8 +720,10 @@ static int ata_tdev_add(struct ata_device *ata_dev)device_initialize (dev); dev->parent = &link->tdev; dev->release = ata_tdev_release; + dev->bus = &ata_bus_type; +if (ata_is_host_link (link)) -dev_set_name (dev,"dev%d.%d" , ap->print_id,ata_dev->devno); +dev_set_name (dev,"dev%d.%d" , ap->print_id, ata_dev->devno);else dev_set_name (dev,"dev%d.%d.0" , ap->print_id, link->pmp);
@@ -822,8 +849,13 @@ __init int libata_transport_init(void) error =transport_class_register (&ata_dev_class);if (error) goto out_unregister_port; + error =bus_register (&ata_bus_type); +if (error) + goto out_unregister_bus;return 0 ; + out_unregister_bus: +bus_unregister (&ata_bus_type); out_unregister_port:transport_class_unregister (&ata_port_class); out_unregister_link:
include/linux/libata.h +16 -38
83 lines not on this page · raw message
Signature · 1 line
2.29.2
PATCH v2 2/3 libata: CONFIG_ATA_SYSFS_COMPAT
In reply to Hannes ReineckeAdd a config option 'ATA_SYSFS_COMPAT' to allow the user to select whether the compability 'ata_port' object with the name of 'ata' should be registered with sysfs.
--- drivers/ata/Kconfig | 10 ++++++++++ drivers/ata/libata-transport.c | 19 ++++++++++++++++--- 2 files changed, 26 insertions(+), 3 deletions(-)
drivers/ata/Kconfig +10 -0
@@ -51,6 +51,16 @@ config ATA_VERBOSE_ERROR If unsure, say Y. +config ATA_SYSFS_COMPAT + bool "Keep original sysfs layout" + default y + help + This option retains the original sysfs layout by adding an + additional ata_port object with the name of 'ataX' in + to the ATA objects like 'ata_port', 'ata_link', and 'ata_device'. + + If unsure, say Y. + config ATA_FORCE bool "\"libata.force=\" kernel parameter support" if EXPERT default y
drivers/ata/libata-transport.c +16 -3
@@ -261,15 +261,21 @@ static int ata_tport_match(struct attribute_container *cont,void ata_tport_delete (struct ata_port *ap) {struct device *dev = &ap->tdev; +#ifdef CONFIG_ATA_SYSFS_COMPAT struct device *cdev = &ap->cdev; +#endif ata_tlink_delete (&ap->link); +#ifdef CONFIG_ATA_SYSFS_COMPAT device_del (cdev); +#endif transport_remove_device (dev);device_del (dev);transport_destroy_device (dev); +#ifdef CONFIG_ATA_SYSFS_COMPAT put_device (cdev); +#endif put_device (dev); }
@@ -288,7 +294,9 @@ int ata_tport_add(struct device *parent, {int error;struct device *dev = &ap->tdev; +#ifdef CONFIG_ATA_SYSFS_COMPAT struct device *cdev = &ap->cdev; +#endif device_initialize (dev); dev->type = &ata_port_type;
@@ -299,23 +307,24 @@ int ata_tport_add(struct device *parent, dev->bus = &ata_bus_type;dev_set_name (dev,"port%d" , ap->print_id); +#ifdef CONFIG_ATA_SYSFS_COMPAT device_initialize (cdev); cdev->parent =get_device (dev); cdev->class = &ata_port_class.class;dev_set_name (cdev,"ata%d" , ap->print_id); - +#endif transport_setup_device (dev);ata_acpi_bind_port (ap); error =device_add (dev);if (error) { goto tport_err; } - +#ifdef CONFIG_ATA_SYSFS_COMPAT error =device_add (cdev);if (error) { goto cdev_err; } - +#endif device_enable_async_suspend (dev);pm_runtime_set_active (dev);pm_runtime_enable (dev);
@@ -331,14 +340,18 @@ int ata_tport_add(struct device *parent,return 0 ; tport_link_err: +#ifdef CONFIG_ATA_SYSFS_COMPATdevice_del (cdev ); cdev_err: +#endiftransport_remove_device (dev);device_del (dev); tport_err:transport_destroy_device (dev); +#ifdef CONFIG_ATA_SYSFS_COMPAT put_device (cdev); +#endif put_device (dev);ata_host_put (ap->host);return error;
Signature · 1 line
2.29.2
Re: [PATCH 2/3] libata: CONFIG_ATA_SYSFS_COMPAT
Quoted 81 lines · Hannes Reinecke
Add a config option '
ATA_SYSFS_COMPAT' to allow the user to select whether the compability 'ata_port' object with the name of 'ata' should be registered with sysfs.Signed-off-by Hannes Reinecke <[email protected]>--- drivers/ata/Kconfig | 10 ++++++++++ drivers/ata/libata-transport.c | 19 ++++++++++++++++--- 2 files changed, 26 insertions(+), 3 deletions(-)drivers/ata/Kconfig +10
-0@@ -51,6 +51,16 @@ config ATA_VERBOSE_ERROR If unsure, say Y. +config ATA_SYSFS_COMPAT + bool "Keep original sysfs layout" + default y + help + This option retains the original sysfs layout by adding an + additional ata_port object with the name of 'ataX' in + to the ATA objects like 'ata_port', 'ata_link', and 'ata_device'. + + If unsure, say Y. + config ATA_FORCE bool "\"libata.force=\" kernel parameter support" if EXPERT default ydrivers/ata/libata-transport.c +10
-1@@ -261,15 +261,21 @@ static int ata_tport_match(struct attribute_container *cont,void ata_tport_delete (struct ata_port *ap) {struct device *dev = &ap->tdev; +#ifdef CONFIG_ATA_SYSFS_COMPAT struct device *cdev = &ap->cdev; +#endif ata_tlink_delete (&ap->link); +#ifdef CONFIG_ATA_SYSFS_COMPAT device_del (cdev); +#endif transport_remove_device (dev);device_del (dev);transport_destroy_device (dev); +#ifdef CONFIG_ATA_SYSFS_COMPAT put_device (cdev); +#endif put_device (dev); }@@ -288,7 +294,9 @@ int ata_tport_add(struct device *parent, {int error;struct device *dev = &ap->tdev; +#ifdef CONFIG_ATA_SYSFS_COMPAT struct device *cdev = &ap->cdev; +#endif device_initialize (dev); dev->type = &ata_port_type;@@ -299,23 +307,24 @@ int ata_tport_add(struct device *parent, dev->bus = &ata_bus_type;dev_set_name (dev,"port%d" , ap->print_id); +#ifdef CONFIG_ATA_SYSFS_COMPAT device_initialize (cdev); cdev->parent =get_device (dev); cdev->class = &ata_port_class.class;dev_set_name (cdev,"ata%d" , ap->print_id); - +#endif transport_setup_device (dev);ata_acpi_bind_port (ap); error =device_add (dev);
drivers/ata/libata-transport.c int ata_tport_add(struct device *parent, · line 318
Quoted 10 lines
drivers/ata/libata-transport.c +1
-2if (error) { goto tport_err; } - +#ifdef CONFIG_ATA_SYSFS_COMPAT error =device_add (cdev);if (error) { goto cdev_err; } -
Instead of adding a device, can't we simply create a symlink ? E.g.:
sprintf(symlink_name, "ata%d", ap->print_id); sysfs_create_link(&parent->kobj, &dev->kobj, symlink_name);
? This seems to work, but I have not checked all possible paths to see if a symlink or directory name expected to be "ataX" ends up being "portX", which could break userspace relying on the old name.
Quoted 23 lines
drivers/ata/libata-transport.c +5
-0+#endif device_enable_async_suspend (dev);pm_runtime_set_active (dev);pm_runtime_enable (dev);@@ -331,14 +340,18 @@ int ata_tport_add(struct device *parent,return 0 ; tport_link_err: +#ifdef CONFIG_ATA_SYSFS_COMPATdevice_del (cdev ); cdev_err: +#endiftransport_remove_device (dev);device_del (dev); tport_err:transport_destroy_device (dev); +#ifdef CONFIG_ATA_SYSFS_COMPAT put_device (cdev); +#endif put_device (dev);ata_host_put (ap->host);return error;
Signature · 2 lines
Damien Le Moal Western Digital Research
Re: [PATCH 2/3] libata: CONFIG_ATA_SYSFS_COMPAT
Quoted 96 lines · Damien Le Moal
Quoted 81 lines · Hannes Reinecke
Add a config option '
ATA_SYSFS_COMPAT' to allow the user to select whether the compability 'ata_port' object with the name of 'ata' should be registered with sysfs.Signed-off-by Hannes Reinecke <[email protected]>--- drivers/ata/Kconfig | 10 ++++++++++ drivers/ata/libata-transport.c | 19 ++++++++++++++++--- 2 files changed, 26 insertions(+), 3 deletions(-)drivers/ata/Kconfig +10
-0@@ -51,6 +51,16 @@ config ATA_VERBOSE_ERROR If unsure, say Y. +config ATA_SYSFS_COMPAT + bool "Keep original sysfs layout" + default y + help + This option retains the original sysfs layout by adding an + additional ata_port object with the name of 'ataX' in + to the ATA objects like 'ata_port', 'ata_link', and 'ata_device'. + + If unsure, say Y. + config ATA_FORCE bool "\"libata.force=\" kernel parameter support" if EXPERT default ydrivers/ata/libata-transport.c +10
-1@@ -261,15 +261,21 @@ static int ata_tport_match(struct attribute_container *cont,void ata_tport_delete (struct ata_port *ap) {struct device *dev = &ap->tdev; +#ifdef CONFIG_ATA_SYSFS_COMPAT struct device *cdev = &ap->cdev; +#endif ata_tlink_delete (&ap->link); +#ifdef CONFIG_ATA_SYSFS_COMPAT device_del (cdev); +#endif transport_remove_device (dev);device_del (dev);transport_destroy_device (dev); +#ifdef CONFIG_ATA_SYSFS_COMPAT put_device (cdev); +#endif put_device (dev); }@@ -288,7 +294,9 @@ int ata_tport_add(struct device *parent, {int error;struct device *dev = &ap->tdev; +#ifdef CONFIG_ATA_SYSFS_COMPAT struct device *cdev = &ap->cdev; +#endif device_initialize (dev); dev->type = &ata_port_type;@@ -299,23 +307,24 @@ int ata_tport_add(struct device *parent, dev->bus = &ata_bus_type;dev_set_name (dev,"port%d" , ap->print_id); +#ifdef CONFIG_ATA_SYSFS_COMPAT device_initialize (cdev); cdev->parent =get_device (dev); cdev->class = &ata_port_class.class;dev_set_name (cdev,"ata%d" , ap->print_id); - +#endif transport_setup_device (dev);ata_acpi_bind_port (ap); error =device_add (dev);drivers/ata/libata-transport.c int ata_tport_add(struct device *parent, · line 318
Quoted 10 lines
drivers/ata/libata-transport.c +1
-2if (error) { goto tport_err; } - +#ifdef CONFIG_ATA_SYSFS_COMPAT error =device_add (cdev);if (error) { goto cdev_err; } -Instead of adding a device, can't we simply create a symlink ? E.g.:
Quoted 7 lines
sprintf(symlink_name, "ata%d", ap->print_id); sysfs_create_link(&parent->kobj, &dev->kobj, symlink_name);? This seems to work, but I have not checked all possible paths to see if a symlink or directory name expected to be "ataX" ends up being "portX", which could break userspace relying on the old name.
Yeah; tried a similar thing which turned out not to be working. Let me test your patch.
Cheers,
Hannes
Signature · 4 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
PATCH v2 3/3 libata: sanitize PMP link naming
In reply to Hannes ReineckeDo not set the ata_link name to 'link<port>.<pmp>.0' as we only have one device attached, rather use the simpler 'link<port>.<pmp>' to avoid having the only three-level naming in libata.
--- drivers/ata/libata-transport.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/ata/libata-transport.c +1 -1
@@ -738,7 +738,7 @@ static int ata_tdev_add(struct ata_device *ata_dev)if (ata_is_host_link (link))dev_set_name (dev,"dev%d.%d" , ap->print_id, ata_dev->devno);else -dev_set_name (dev,"dev%d.%d.0" , ap->print_id, link->pmp); +dev_set_name (dev,"dev%d.%d" , ap->print_id, link->pmp);transport_setup_device (dev);ata_acpi_bind_dev (ata_dev);
Signature · 1 line
2.29.2
Quoted 19 lines · Hannes Reinecke
Hi all,
here's my second attempt the align the libata object naming with sysfs. Key point is to introduce an 'ata' bus, which serves to collect all libata object (
ata_port,ata_link, andata_device).To facilitate that the name of the '
ata_port' object changes from 'ata' to 'port'. To provide backwards compability I've added an additional class object for theata_portwith the name of 'ata'.
This additional object can be disabled with the config optionATA_SYSFS_COMPAT.As usual, comments and reviews are welcome.
Hannes Reinecke (3):
libata: sysfs naming option
libata:CONFIG_ATA_SYSFS_COMPAT
libata: sanitize PMP link naming
Quoted 5 lines
drivers/ata/Kconfig | 10 +++++++ drivers/ata/libata-transport.c | 55 ++++++++++++++++++++++++++++++---- include/linux/libata.h | 54 ++++++++++----------------------- 3 files changed, 76 insertions(+), 43 deletions(-)
Kasan is not happy at all when I do "rmmod ahci"...
[ 1657.438508] BUG: KASAN: double-free or invalid-free in attribute_container_release+0x37/0x50 [ 1657.447070] [ 1657.448597] CPU: 8 PID: 1597 Comm: rmmod Not tainted 5.17.0-libata+ #25 [ 1657.455314] Hardware name: Supermicro Super Server/X11DPL-i, BIOS 3.3 02/21/2020 [ 1657.462809] Call Trace: [ 1657.465304] <TASK> [ 1657.467447] dump_stack_lvl+0x45/0x59 [ 1657.471180] print_address_description.constprop.0+0x1f/0x120 [ 1657.477021] ? attribute_container_release+0x37/0x50 [ 1657.482060] ? attribute_container_release+0x37/0x50 [ 1657.487108] kasan_report_invalid_free+0x51/0x80 [ 1657.491802] __kasan_slab_free+0xf4/0x110 [ 1657.495885] ? attribute_container_release+0x37/0x50 [ 1657.500930] kfree+0xca/0x210 [ 1657.503956] attribute_container_release+0x37/0x50 [ 1657.508826] device_release+0x98/0x200 [ 1657.512642] kobject_put+0x139/0x410 [ 1657.516283] ata_tport_delete+0x4a/0x60 [libata] [ 1657.521015] ata_host_detach+0x336/0x660 [libata] [ 1657.525820] ? kernfs_remove_by_name_ns+0x9a/0xe0 [ 1657.530615] pci_device_remove+0x65/0x110 [ 1657.534696] __device_release_driver+0x316/0x680 [ 1657.539398] driver_detach+0x1ec/0x2d0 [ 1657.543217] bus_remove_driver+0xe7/0x2d0 [ 1657.547293] ? lock_is_held_type+0x98/0x110 [ 1657.551546] pci_unregister_driver+0x26/0x250 [ 1657.555982] __x64_sys_delete_module+0x2fd/0x510 [ 1657.560673] ? free_module+0xaa0/0xaa0 [ 1657.564487] ? __cond_resched+0x1c/0x90 [ 1657.568392] ? lockdep_hardirqs_on_prepare+0x273/0x3e0 [ 1657.573611] ? syscall_enter_from_user_mode+0x21/0x70 [ 1657.578740] ? trace_hardirqs_on+0x1c/0x110 [ 1657.583000] do_syscall_64+0x35/0x80 [ 1657.586635] entry_SYSCALL_64_after_hwframe+0x44/0xae [ 1657.591765] RIP: 0033:0x7f4eb289d84b [ 1657.595403] Code: 73 01 c3 48 8b 0d dd 75 0e 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d ad 75 0e 00 f7 d8 64 89 01 48 [ 1657.618085] RSP: 002b:00007fffa28f9158 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0 [ 1657.627664] RAX: ffffffffffffffda RBX: 000055e1a7f54760 RCX: 00007f4eb289d84b [ 1657.636822] RDX: 000000000000000a RSI: 0000000000000800 RDI: 000055e1a7f547c8 [ 1657.645965] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000 [ 1657.655065] R10: 00007f4eb292eac0 R11: 0000000000000206 R12: 00007fffa28f93d0 [ 1657.664115] R13: 00007fffa28fa780 R14: 000055e1a7f542a0 R15: 000055e1a7f54760 [ 1657.673133] </TASK> [ 1657.677130] [ 1657.680403] Allocated by task 183: [ 1657.685577] kasan_save_stack+0x1e/0x40 [ 1657.691152] __kasan_kmalloc+0x7f/0xa0 [ 1657.696603] kmem_cache_alloc_trace+0x1f9/0x470 [ 1657.702809] ata_port_alloc+0x40/0x5a0 [libata] [ 1657.709031] ata_host_alloc+0x1ca/0x260 [libata] [ 1657.715301] ata_host_alloc_pinfo+0x1d/0x540 [libata] [ 1657.721978] ahci_init_one+0xc5b/0x1d40 [ahci] [ 1657.728037] local_pci_probe+0xc6/0x150 [ 1657.733487] work_for_cpu_fn+0x4e/0xa0 [ 1657.738833] process_one_work+0x7f0/0x1310 [ 1657.744509] worker_thread+0x6e0/0xf70 [ 1657.749855] kthread+0x28f/0x330 [ 1657.754665] ret_from_fork+0x1f/0x30 [ 1657.759818] [ 1657.762850] The buggy address belongs to the object at ffff8885e2ce0000 [ 1657.762850] which belongs to the cache kmalloc-32k of size 32768 [ 1657.778757] The buggy address is located 15088 bytes inside of [ 1657.778757] 32768-byte region [ffff8885e2ce0000, ffff8885e2ce8000) [ 1657.794105] The buggy address belongs to the page: [ 1657.800540] page:00000000ddc44957 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x5e2ce0 [ 1657.811675] head:00000000ddc44957 order:4 compound_mapcount:0 compound_pincount:0 [ 1657.820907] flags: 0x20000000010200(slab|head|node=0|zone=2) [ 1657.828316] raw: 0020000000010200 ffffea0018392808 ffffea00178b3c08 ffff888100040200 [ 1657.837857] raw: 0000000000000000 ffff8885e2ce0000 0000000100000001 0000000000000000 [ 1657.847408] page dumped because: kasan: bad access detected [ 1657.854796] [ 1657.858067] Memory state around the buggy address: [ 1657.864684] ffff8885e2ce3980: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 1657.873803] ffff8885e2ce3a00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 1657.882884] >ffff8885e2ce3a80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 1657.891942] ^ [ 1657.900670] ffff8885e2ce3b00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 [ 1657.909774] ffff8885e2ce3b80: 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00
And after that, doing "rmmod libata; modprobe libata", Things get really bad...
[ 1820.226215] sysfs: cannot create duplicate filename '/bus/ata'
And then NULL pointer dereference oops.
My PMP setup also does not show the devices named devX.Y. I suspect this is because my eSATA box supports FIS based switching. Even with that, the eSATA box has 4 drives all connected behind one port/one link, but I still see each device with its own port and link. Weird. Since I am not 100% up to speed on how the PMP code works, I need to dig into it.
Signature · 2 lines
Damien Le Moal Western Digital Research