user/sven/linux.git/lib, branch v3.2.45

kobject: fix kset_find_obj() race with concurrent last kobject_put()

2013-04-25T19:25:39Z

commit a49b7e82cab0f9b41f483359be83f44fbb6b4979 upstream. Anatol Pomozov identified a race condition that hits module unloading and re-loading. To quote Anatol: "This is a race codition that exists between kset_find_obj() and kobject_put(). kset_find_obj() might return kobject that has refcount equal to 0 if this kobject is freeing by kobject_put() in other thread. Here is timeline for the crash in case if kset_find_obj() searches for an object tht nobody holds and other thread is doing kobject_put() on the same kobject: THREAD A (calls kset_find_obj()) THREAD B (calls kobject_put()) splin_lock() atomic_dec_return(kobj->kref), counter gets zero here ... starts kobject cleanup .... spin_lock() // WAIT thread A in kobj_kset_leave() iterate over kset->list atomic_inc(kobj->kref) (counter becomes 1) spin_unlock() spin_lock() // taken // it does not know that thread A increased counter so it remove obj from list spin_unlock() vfree(module) // frees module object with containing kobj // kobj points to freed memory area!! kobject_put(kobj) // OOPS!!!! The race above happens because module.c tries to use kset_find_obj() when somebody unloads module. The module.c code was introduced in commit 6494a93d55fa" Anatol supplied a patch specific for module.c that worked around the problem by simply not using kset_find_obj() at all, but rather than make a local band-aid, this just fixes kset_find_obj() to be thread-safe using the proper model of refusing the get a new reference if the refcount has already dropped to zero. See examples of this proper refcount handling not only in the kref documentation, but in various other equivalent uses of this pattern by grepping for atomic_inc_not_zero(). [ Side note: the module race does indicate that module loading and unloading is not properly serialized wrt sysfs information using the module mutex. That may require further thought, but this is the correct fix at the kobject layer regardless. ] Reported-analyzed-and-tested-by: Anatol Pomozov Cc: Greg Kroah-Hartman Cc: Al Viro Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

idr: fix top layer handling

2013-03-06T03:24:17Z

commit 326cf0f0f308933c10236280a322031f0097205d upstream. Most functions in idr fail to deal with the high bits when the idr tree grows to the maximum height. * idr_get_empty_slot() stops growing idr tree once the depth reaches MAX_IDR_LEVEL - 1, which is one depth shallower than necessary to cover the whole range. The function doesn't even notice that it didn't grow the tree enough and ends up allocating the wrong ID given sufficiently high @starting_id. For example, on 64 bit, if the starting id is 0x7fffff01, idr_get_empty_slot() will grow the tree 5 layer deep, which only covers the 30 bits and then proceed to allocate as if the bit 30 wasn't specified. It ends up allocating 0x3fffff01 without the bit 30 but still returns 0x7fffff01. * __idr_remove_all() will not remove anything if the tree is fully grown. * idr_find() can't find anything if the tree is fully grown. * idr_for_each() and idr_get_next() can't iterate anything if the tree is fully grown. Fix it by introducing idr_max() which returns the maximum possible ID given the depth of tree and replacing the id limit checks in all affected places. As the idr_layer pointer array pa[] needs to be 1 larger than the maximum depth, enlarge pa[] arrays by one. While this plugs the discovered issues, the whole code base is horrible and in desparate need of rewrite. It's fragile like hell, Signed-off-by: Tejun Heo Cc: Rusty Russell Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds [bwh: Backported to 3.2: - Adjust context - s/MAX_IDR_LEVEL/MAX_LEVEL/; s/MAX_IDR_SHIFT/MAX_ID_SHIFT/ - Drop change to idr_alloc()] Signed-off-by: Ben Hutchings

idr: make idr_get_next() good for rcu_read_lock()

2013-03-06T03:24:16Z

commit 9f7de8275b46d9d11b1505adbfe6c2bb48df4741 upstream. Make one small adjustment to idr_get_next(): take the height from the top layer (stable under RCU) instead of from the root (unprotected by RCU), as idr_find() does: so that it can be used with RCU locking. Copied comment on RCU locking from idr_find(). Signed-off-by: Hugh Dickins Acked-by: KAMEZAWA Hiroyuki Acked-by: Li Zefan Cc: Eric Dumazet Acked-by: Tejun Heo Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

idr: fix a subtle bug in idr_get_next()

2013-03-06T03:24:16Z

commit 6cdae7416a1c45c2ce105a78187d9b7e8feb9e24 upstream. The iteration logic of idr_get_next() is borrowed mostly verbatim from idr_for_each(). It walks down the tree looking for the slot matching the current ID. If the matching slot is not found, the ID is incremented by the distance of single slot at the given level and repeats. The implementation assumes that during the whole iteration id is aligned to the layer boundaries of the level closest to the leaf, which is true for all iterations starting from zero or an existing element and thus is fine for idr_for_each(). However, idr_get_next() may be given any point and if the starting id hits in the middle of a non-existent layer, increment to the next layer will end up skipping the same offset into it. For example, an IDR with IDs filled between [64, 127] would look like the following. [ 0 64 ... ] /----/ | | | NULL [ 64 ... 127 ] If idr_get_next() is called with 63 as the starting point, it will try to follow down the pointer from 0. As it is NULL, it will then try to proceed to the next slot in the same level by adding the slot distance at that level which is 64 - making the next try 127. It goes around the loop and finds and returns 127 skipping [64, 126]. Note that this bug also triggers in idr_for_each_entry() loop which deletes during iteration as deletions can make layers go away leaving the iteration with unaligned ID into missing layers. Fix it by ensuring proceeding to the next slot doesn't carry over the unaligned offset - ie. use round_up(id + 1, slot_distance) instead of id += slot_distance. Signed-off-by: Tejun Heo Reported-by: David Teigland Cc: KAMEZAWA Hiroyuki Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

genalloc: stop crashing the system when destroying a pool

2012-10-30T23:27:06Z

commit eedce141cd2dad8d0cefc5468ef41898949a7031 upstream. The genalloc code uses the bitmap API from include/linux/bitmap.h and lib/bitmap.c, which is based on long values. Both bitmap_set from lib/bitmap.c and bitmap_set_ll, which is the lockless version from genalloc.c, use BITMAP_LAST_WORD_MASK to set the first bits in a long in the bitmap. That one uses (1 << bits) - 1, 0b111, if you are setting the first three bits. This means that the API counts from the least significant bits (LSB from now on) to the MSB. The LSB in the first long is bit 0, then. The same works for the lookup functions. The genalloc code uses longs for the bitmap, as it should. In include/linux/genalloc.h, struct gen_pool_chunk has unsigned long bits[0] as its last member. When allocating the struct, genalloc should reserve enough space for the bitmap. This should be a proper number of longs that can fit the amount of bits in the bitmap. However, genalloc allocates an integer number of bytes that fit the amount of bits, but may not be an integer amount of longs. 9 bytes, for example, could be allocated for 70 bits. This is a problem in itself if the Least Significat Bit in a long is in the byte with the largest address, which happens in Big Endian machines. This means genalloc is not allocating the byte in which it will try to set or check for a bit. This may end up in memory corruption, where genalloc will try to set the bits it has not allocated. In fact, genalloc may not set these bits because it may find them already set, because they were not zeroed since they were not allocated. And that's what causes a BUG when gen_pool_destroy is called and check for any set bits. What really happens is that genalloc uses kmalloc_node with __GFP_ZERO on gen_pool_add_virt. With SLAB and SLUB, this means the whole slab will be cleared, not only the requested bytes. Since struct gen_pool_chunk has a size that is a multiple of 8, and slab sizes are multiples of 8, we get lucky and allocate and clear the right amount of bytes. Hower, this is not the case with SLOB or with older code that did memset after allocating instead of using __GFP_ZERO. So, a simple module as this (running 3.6.0), will cause a crash when rmmod'ed. [root@phantom-lp2 foo]# cat foo.c #include #include #include #include MODULE_LICENSE("GPL"); MODULE_VERSION("0.1"); static struct gen_pool *foo_pool; static __init int foo_init(void) { int ret; foo_pool = gen_pool_create(10, -1); if (!foo_pool) return -ENOMEM; ret = gen_pool_add(foo_pool, 0xa0000000, 32 << 10, -1); if (ret) { gen_pool_destroy(foo_pool); return ret; } return 0; } static __exit void foo_exit(void) { gen_pool_destroy(foo_pool); } module_init(foo_init); module_exit(foo_exit); [root@phantom-lp2 foo]# zcat /proc/config.gz | grep SLOB CONFIG_SLOB=y [root@phantom-lp2 foo]# insmod ./foo.ko [root@phantom-lp2 foo]# rmmod foo ------------[ cut here ]------------ kernel BUG at lib/genalloc.c:243! cpu 0x4: Vector: 700 (Program Check) at [c0000000bb0e7960] pc: c0000000003cb50c: .gen_pool_destroy+0xac/0x110 lr: c0000000003cb4fc: .gen_pool_destroy+0x9c/0x110 sp: c0000000bb0e7be0 msr: 8000000000029032 current = 0xc0000000bb0e0000 paca = 0xc000000006d30e00 softe: 0 irq_happened: 0x01 pid = 13044, comm = rmmod kernel BUG at lib/genalloc.c:243! [c0000000bb0e7ca0] d000000004b00020 .foo_exit+0x20/0x38 [foo] [c0000000bb0e7d20] c0000000000dff98 .SyS_delete_module+0x1a8/0x290 [c0000000bb0e7e30] c0000000000097d4 syscall_exit+0x0/0x94 --- Exception: c00 (System Call) at 000000800753d1a0 SP (fffd0b0e640) is in userspace Signed-off-by: Thadeu Lima de Souza Cascardo Cc: Paul Gortmaker Cc: Benjamin Gaignard Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

lib/gcd.c: prevent possible div by 0

2012-10-17T02:49:37Z

commit e96875677fb2b7cb739c5d7769824dff7260d31d upstream. Account for all properties when a and/or b are 0: gcd(0, 0) = 0 gcd(a, 0) = a gcd(0, b) = b Fixes no known problems in current kernels. Signed-off-by: Davidlohr Bueso Cc: Eric Dumazet Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

lib/vsprintf.c: kptr_restrict: fix pK-error in SysRq show-all-timers(Q)

2012-08-09T23:25:01Z

commit 3715c5309f6d175c3053672b73fd4f73be16fd07 upstream. When using ALT+SysRq+Q all the pointers are replaced with "pK-error" like this: [23153.208033] .base: pK-error with echo h > /proc/sysrq-trigger it works: [23107.776363] .base: ffff88023e60d540 The intent behind this behavior was to return "pK-error" in cases where the %pK format specifier was used in interrupt context, because the CAP_SYSLOG check wouldn't be meaningful. Clearly this should only apply when kptr_restrict is actually enabled though. Reported-by: Stevie Trujillo Signed-off-by: Dan Rosenberg Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

btree: fix tree corruption in btree_get_prev()

2012-06-10T13:42:10Z

commit cbf8ae32f66a9ceb8907ad9e16663c2a29e48990 upstream. The memory the parameter __key points to is used as an iterator in btree_get_prev(), so if we save off a bkey() pointer in retry_key and then assign that to __key, we'll end up corrupting the btree internals when we do eg longcpy(__key, bkey(geo, node, i), geo->keylen); to return the key value. What we should do instead is use longcpy() to copy the key value that retry_key points to __key. This can cause a btree to get corrupted by seemingly read-only operations such as btree_for_each_safe. [akpm@linux-foundation.org: avoid the double longcpy()] Signed-off-by: Roland Dreier Acked-by: Joern Engel Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Ben Hutchings

uevent: send events in correct order according to seqnum (v3)

2012-04-02T16:52:35Z

commit 7b60a18da393ed70db043a777fd9e6d5363077c4 upstream. The queue handling in the udev daemon assumes that the events are ordered. Before this patch uevent_seqnum is incremented under sequence_lock, than an event is send uner uevent_sock_mutex. I want to say that code contained a window between incrementing seqnum and sending an event. This patch locks uevent_sock_mutex before incrementing uevent_seqnum. v2: delete sequence_lock, uevent_seqnum is protected by uevent_sock_mutex v3: unlock the mutex before the goto exit Thanks for Kay for the comments. Signed-off-by: Andrew Vagin Tested-By: Kay Sievers Signed-off-by: Greg Kroah-Hartman

Fix comparison using wrong pointer variable in dma debug code

2011-11-21T10:35:37Z

cppcheck reported: [lib/dma-debug.c:248] -> [lib/dma-debug.c:248]: (style) Same expression on both sides of '=='. Signed-off-by: Thomas Jarosch Signed-off-by: Joerg Roedel