forked from iovisor/bcc
-
Notifications
You must be signed in to change notification settings - Fork 4
/
inject_example.txt
154 lines (115 loc) · 6.67 KB
/
inject_example.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
Some examples for inject
inject guarantees the appropriate erroneous return of the specified injection
mode (kmalloc,bio,etc) given a call chain and an optional set of predicates. You
can also optionally print out the generated BPF program for
modification/debugging purposes.
As a simple example, let's say you wanted to fail all mounts. As of 4.17 we can
fail syscalls directly, so let's do that:
# ./inject.py kmalloc -v 'SyS_mount()'
The first argument indicates the mode (or what to fail). Appropriate headers are
specified, if necessary. The verbosity flag prints the generated program. Note
that some syscalls will be available as 'SyS_xyz' and some will be available as
'sys_xyz'. This is largely dependent on the number of arguments each syscall
takes.
Trying to mount various filesystems will fail and report an inability to
allocate memory, as expected.
Whenever a predicate is missing, an implicit "(true)" is inserted. The example
above can be explicitly written as:
# ./inject.py kmalloc -v '(true) => SyS_mount()(true)'
The "(true)" without an associated function is a predicate for the error
injection mechanism of the current mode. In the case of kmalloc, the predicate
would have access to the arguments of:
should_failslab(struct kmem_cache *s, gfp_t gfpflags)
Other modes work similarly.
"bio" has access to the arguments of:
should_fail_bio(struct bio *bio)
"alloc_page" has access to the arguments of:
should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
We also note that it's unnecessary to state the arguments of the function if you
have no intention to reference them in the associated predicate.
Now let's say we want to be a bit more specific; suppose you want to fail
kmalloc() from mount_subtree() when called from btrfs_mount(). This will fail
only btrfs mounts:
# ./inject.py kmalloc -v 'mount_subtree() => btrfs_mount()'
Attempting to mount btrfs filesystem during the execution of this command will
yield an error, but other filesystems will be fine.
Next, lets say we want to hit one of the BUG_ONs in fs/btrfs. As of 4.16-rc3,
there is a BUG_ON in btrfs_prepare_close_one_device() at fs/btrfs/volumes.c:1002
To hit this, we can use the following:
# ./inject.py kmalloc -v 'btrfs_alloc_device() => btrfs_close_devices()'
While the script was executing, I mounted and unmounted btrfs, causing a
segfault on umount(since that satisfied the call path indicated). A look at
dmesg will confirm that the erroneous return value injected by the script
tripped the BUG_ON, causing a segfault down the line.
In general, it's worth noting that the required specificity of the call chain is
dependent on how much granularity you need. The example above might have
performed as expected without the intermediate btrfs_alloc_device, but might
have also done something unexpected(an earlier kmalloc could have failed before
the one we were targeting).
For hot paths, the approach outlined above isn't enough. If a path is traversed
very often, we can distinguish distinct calls with function arguments. Let's say
we want to fail the dentry allocation of a file creatively named 'bananas'. We
can do the following:
# ./inject.py kmalloc -v 'd_alloc_parallel(struct dentry *parent, const struct
qstr *name)(STRCMP(name->name, 'bananas'))'
While this script is executing, any operation that would cause a dentry
allocation where the name is 'bananas' fails, as expected.
Here, since we're referencing a function argument in our predicate, we need to
provide the function signature up to the argument we're using.
To note, STRCMP is a workaround for some rewriter issues. It will take input of
the form (x->...->z, 'literal'), and generate some equivalent code that the
verifier is more friendly about. It's not horribly robust, but works for the
purposes of making string comparisons a bit easier.
Finally, we briefly demonstrate how to inject bio failures. The mechanism is
identical, so any information from above will apply.
Let's say we want to fail bio requests when the request is to some specific
sector. An example use case would be to fail superblock writes in btrfs. For
btrfs, we know that there must be a superblock at 65536 bytes, or sector 128.
This allows us to run the following:
# ./inject.py bio -v -I 'linux/blkdev.h' '(({struct gendisk *d = bio->bi_disk;
struct disk_part_tbl *tbl = d->part_tbl; struct hd_struct **parts = (void *)tbl +
sizeof(struct disk_part_tbl); struct hd_struct **partp = parts + bio->bi_partno;
struct hd_struct *p = *partp; dev_t disk = p->__dev.devt; disk ==
MKDEV(254,16);}) && bio->bi_iter.bi_sector == 128)'
The predicate in the command above has two parts. The first is a compound
statement which shortens to "only if the system is btrfs", but is long due
to rewriter/verifier shenanigans. The major/minor information can be found
however; I used Python. The second part simply checks the starting
address of bi_iter. While executing, this script effectively fails superblock
writes to the superblock at sector 128 without affecting other filesystems.
As an extension to the above, one could easily fail all btrfs superblock writes
(we only fail the primary) by calculating the sector number of the mirrors and
amending the predicate accordingly.
Inject also provides a probability option; this allows you to fail the
path+predicates some percentage of the time. For example, let's say we want to
fail our mounts half the time:
# ./inject.py kmalloc -v -P 0.01 'SyS_mount()'
USAGE message:
usage: inject.py [-h] [-I header] [-P probability] [-v] [-c COUNT]
{kmalloc,bio,alloc_page} spec
Fail specified kernel functionality when call chain and predicates are met
positional arguments:
{kmalloc,bio,alloc_page}
indicate which base kernel function to fail
spec specify call chain
optional arguments:
-h, --help show this help message and exit
-I header, --include header
additional header files to include in the BPF program
-P probability, --probability probability
probability that this call chain will fail
-v, --verbose print BPF program
-c COUNT, --count COUNT
Number of fails before bypassing the override
EXAMPLES:
# ./inject.py kmalloc -v 'SyS_mount()'
Fails all calls to syscall mount
# ./inject.py kmalloc -v '(true) => SyS_mount()(true)'
Explicit rewriting of above
# ./inject.py kmalloc -v 'mount_subtree() => btrfs_mount()'
Fails btrfs mounts only
# ./inject.py kmalloc -v 'd_alloc_parallel(struct dentry *parent, const struct \
qstr *name)(STRCMP(name->name, 'bananas'))'
Fails dentry allocations of files named 'bananas'
# ./inject.py kmalloc -v -P 0.01 'SyS_mount()'
Fails calls to syscall mount with 1% probability