diff options
Diffstat (limited to 'meta-fsl-ppc/recipes-kernel/linux/files/0002-shmem-CVE-2014-4171.patch')
| -rw-r--r-- | meta-fsl-ppc/recipes-kernel/linux/files/0002-shmem-CVE-2014-4171.patch | 200 |
1 files changed, 0 insertions, 200 deletions
diff --git a/meta-fsl-ppc/recipes-kernel/linux/files/0002-shmem-CVE-2014-4171.patch b/meta-fsl-ppc/recipes-kernel/linux/files/0002-shmem-CVE-2014-4171.patch deleted file mode 100644 index a43b89563..000000000 --- a/meta-fsl-ppc/recipes-kernel/linux/files/0002-shmem-CVE-2014-4171.patch +++ /dev/null | |||
| @@ -1,200 +0,0 @@ | |||
| 1 | From 38d05809df1ea5272a658e7f4d5f2a3027ad2fd2 Mon Sep 17 00:00:00 2001 | ||
| 2 | From: Hugh Dickins <hughd@google.com> | ||
| 3 | Date: Wed, 23 Jul 2014 14:00:10 -0700 | ||
| 4 | Subject: [PATCH 2/3] shmem: fix faulting into a hole, not taking i_mutex | ||
| 5 | |||
| 6 | commit 8e205f779d1443a94b5ae81aa359cb535dd3021e upstream. | ||
| 7 | |||
| 8 | Commit f00cdc6df7d7 ("shmem: fix faulting into a hole while it's | ||
| 9 | punched") was buggy: Sasha sent a lockdep report to remind us that | ||
| 10 | grabbing i_mutex in the fault path is a no-no (write syscall may already | ||
| 11 | hold i_mutex while faulting user buffer). | ||
| 12 | |||
| 13 | We tried a completely different approach (see following patch) but that | ||
| 14 | proved inadequate: good enough for a rational workload, but not good | ||
| 15 | enough against trinity - which forks off so many mappings of the object | ||
| 16 | that contention on i_mmap_mutex while hole-puncher holds i_mutex builds | ||
| 17 | into serious starvation when concurrent faults force the puncher to fall | ||
| 18 | back to single-page unmap_mapping_range() searches of the i_mmap tree. | ||
| 19 | |||
| 20 | So return to the original umbrella approach, but keep away from i_mutex | ||
| 21 | this time. We really don't want to bloat every shmem inode with a new | ||
| 22 | mutex or completion, just to protect this unlikely case from trinity. | ||
| 23 | So extend the original with wait_queue_head on stack at the hole-punch | ||
| 24 | end, and wait_queue item on the stack at the fault end. | ||
| 25 | |||
| 26 | This involves further use of i_lock to guard against the races: lockdep | ||
| 27 | has been happy so far, and I see fs/inode.c:unlock_new_inode() holds | ||
| 28 | i_lock around wake_up_bit(), which is comparable to what we do here. | ||
| 29 | i_lock is more convenient, but we could switch to shmem's info->lock. | ||
| 30 | |||
| 31 | This issue has been tagged with CVE-2014-4171, which will require commit | ||
| 32 | f00cdc6df7d7 and this and the following patch to be backported: we | ||
| 33 | suggest to 3.1+, though in fact the trinity forkbomb effect might go | ||
| 34 | back as far as 2.6.16, when madvise(,,MADV_REMOVE) came in - or might | ||
| 35 | not, since much has changed, with i_mmap_mutex a spinlock before 3.0. | ||
| 36 | Anyone running trinity on 3.0 and earlier? I don't think we need care. | ||
| 37 | |||
| 38 | Upstream-Status: Backport | ||
| 39 | |||
| 40 | Signed-off-by: Hugh Dickins <hughd@google.com> | ||
| 41 | Reported-by: Sasha Levin <sasha.levin@oracle.com> | ||
| 42 | Tested-by: Sasha Levin <sasha.levin@oracle.com> | ||
| 43 | Cc: Vlastimil Babka <vbabka@suse.cz> | ||
| 44 | Cc: Konstantin Khlebnikov <koct9i@gmail.com> | ||
| 45 | Cc: Johannes Weiner <hannes@cmpxchg.org> | ||
| 46 | Cc: Lukas Czerner <lczerner@redhat.com> | ||
| 47 | Cc: Dave Jones <davej@redhat.com> | ||
| 48 | Cc: <stable@vger.kernel.org> [3.1+] | ||
| 49 | Signed-off-by: Andrew Morton <akpm@linux-foundation.org> | ||
| 50 | Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> | ||
| 51 | Signed-off-by: Jiri Slaby <jslaby@suse.cz> | ||
| 52 | Signed-off-by: Sona Sarmadi <sona.sarmadi@enea.com> | ||
| 53 | --- | ||
| 54 | mm/shmem.c | 78 +++++++++++++++++++++++++++++++++++++++++--------------------- | ||
| 55 | 1 file changed, 52 insertions(+), 26 deletions(-) | ||
| 56 | |||
| 57 | diff --git a/mm/shmem.c b/mm/shmem.c | ||
| 58 | index 00d412f..6f5626f 100644 | ||
| 59 | --- a/mm/shmem.c | ||
| 60 | +++ b/mm/shmem.c | ||
| 61 | @@ -85,7 +85,7 @@ static struct vfsmount *shm_mnt; | ||
| 62 | * a time): we would prefer not to enlarge the shmem inode just for that. | ||
| 63 | */ | ||
| 64 | struct shmem_falloc { | ||
| 65 | - int mode; /* FALLOC_FL mode currently operating */ | ||
| 66 | + wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ | ||
| 67 | pgoff_t start; /* start of range currently being fallocated */ | ||
| 68 | pgoff_t next; /* the next page offset to be fallocated */ | ||
| 69 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | ||
| 70 | @@ -827,7 +827,7 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc) | ||
| 71 | spin_lock(&inode->i_lock); | ||
| 72 | shmem_falloc = inode->i_private; | ||
| 73 | if (shmem_falloc && | ||
| 74 | - !shmem_falloc->mode && | ||
| 75 | + !shmem_falloc->waitq && | ||
| 76 | index >= shmem_falloc->start && | ||
| 77 | index < shmem_falloc->next) | ||
| 78 | shmem_falloc->nr_unswapped++; | ||
| 79 | @@ -1306,38 +1306,58 @@ static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | ||
| 80 | * Trinity finds that probing a hole which tmpfs is punching can | ||
| 81 | * prevent the hole-punch from ever completing: which in turn | ||
| 82 | * locks writers out with its hold on i_mutex. So refrain from | ||
| 83 | - * faulting pages into the hole while it's being punched, and | ||
| 84 | - * wait on i_mutex to be released if vmf->flags permits. | ||
| 85 | + * faulting pages into the hole while it's being punched. Although | ||
| 86 | + * shmem_undo_range() does remove the additions, it may be unable to | ||
| 87 | + * keep up, as each new page needs its own unmap_mapping_range() call, | ||
| 88 | + * and the i_mmap tree grows ever slower to scan if new vmas are added. | ||
| 89 | + * | ||
| 90 | + * It does not matter if we sometimes reach this check just before the | ||
| 91 | + * hole-punch begins, so that one fault then races with the punch: | ||
| 92 | + * we just need to make racing faults a rare case. | ||
| 93 | + * | ||
| 94 | + * The implementation below would be much simpler if we just used a | ||
| 95 | + * standard mutex or completion: but we cannot take i_mutex in fault, | ||
| 96 | + * and bloating every shmem inode for this unlikely case would be sad. | ||
| 97 | */ | ||
| 98 | if (unlikely(inode->i_private)) { | ||
| 99 | struct shmem_falloc *shmem_falloc; | ||
| 100 | |||
| 101 | spin_lock(&inode->i_lock); | ||
| 102 | shmem_falloc = inode->i_private; | ||
| 103 | - if (!shmem_falloc || | ||
| 104 | - shmem_falloc->mode != FALLOC_FL_PUNCH_HOLE || | ||
| 105 | - vmf->pgoff < shmem_falloc->start || | ||
| 106 | - vmf->pgoff >= shmem_falloc->next) | ||
| 107 | - shmem_falloc = NULL; | ||
| 108 | - spin_unlock(&inode->i_lock); | ||
| 109 | - /* | ||
| 110 | - * i_lock has protected us from taking shmem_falloc seriously | ||
| 111 | - * once return from shmem_fallocate() went back up that stack. | ||
| 112 | - * i_lock does not serialize with i_mutex at all, but it does | ||
| 113 | - * not matter if sometimes we wait unnecessarily, or sometimes | ||
| 114 | - * miss out on waiting: we just need to make those cases rare. | ||
| 115 | - */ | ||
| 116 | - if (shmem_falloc) { | ||
| 117 | + if (shmem_falloc && | ||
| 118 | + shmem_falloc->waitq && | ||
| 119 | + vmf->pgoff >= shmem_falloc->start && | ||
| 120 | + vmf->pgoff < shmem_falloc->next) { | ||
| 121 | + wait_queue_head_t *shmem_falloc_waitq; | ||
| 122 | + DEFINE_WAIT(shmem_fault_wait); | ||
| 123 | + | ||
| 124 | + ret = VM_FAULT_NOPAGE; | ||
| 125 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && | ||
| 126 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | ||
| 127 | + /* It's polite to up mmap_sem if we can */ | ||
| 128 | up_read(&vma->vm_mm->mmap_sem); | ||
| 129 | - mutex_lock(&inode->i_mutex); | ||
| 130 | - mutex_unlock(&inode->i_mutex); | ||
| 131 | - return VM_FAULT_RETRY; | ||
| 132 | + ret = VM_FAULT_RETRY; | ||
| 133 | } | ||
| 134 | - /* cond_resched? Leave that to GUP or return to user */ | ||
| 135 | - return VM_FAULT_NOPAGE; | ||
| 136 | + | ||
| 137 | + shmem_falloc_waitq = shmem_falloc->waitq; | ||
| 138 | + prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | ||
| 139 | + TASK_UNINTERRUPTIBLE); | ||
| 140 | + spin_unlock(&inode->i_lock); | ||
| 141 | + schedule(); | ||
| 142 | + | ||
| 143 | + /* | ||
| 144 | + * shmem_falloc_waitq points into the shmem_fallocate() | ||
| 145 | + * stack of the hole-punching task: shmem_falloc_waitq | ||
| 146 | + * is usually invalid by the time we reach here, but | ||
| 147 | + * finish_wait() does not dereference it in that case; | ||
| 148 | + * though i_lock needed lest racing with wake_up_all(). | ||
| 149 | + */ | ||
| 150 | + spin_lock(&inode->i_lock); | ||
| 151 | + finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | ||
| 152 | + spin_unlock(&inode->i_lock); | ||
| 153 | + return ret; | ||
| 154 | } | ||
| 155 | + spin_unlock(&inode->i_lock); | ||
| 156 | } | ||
| 157 | |||
| 158 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); | ||
| 159 | @@ -1855,13 +1875,13 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset, | ||
| 160 | |||
| 161 | mutex_lock(&inode->i_mutex); | ||
| 162 | |||
| 163 | - shmem_falloc.mode = mode & ~FALLOC_FL_KEEP_SIZE; | ||
| 164 | - | ||
| 165 | if (mode & FALLOC_FL_PUNCH_HOLE) { | ||
| 166 | struct address_space *mapping = file->f_mapping; | ||
| 167 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | ||
| 168 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | ||
| 169 | + DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); | ||
| 170 | |||
| 171 | + shmem_falloc.waitq = &shmem_falloc_waitq; | ||
| 172 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; | ||
| 173 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | ||
| 174 | spin_lock(&inode->i_lock); | ||
| 175 | @@ -1873,8 +1893,13 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset, | ||
| 176 | 1 + unmap_end - unmap_start, 0); | ||
| 177 | shmem_truncate_range(inode, offset, offset + len - 1); | ||
| 178 | /* No need to unmap again: hole-punching leaves COWed pages */ | ||
| 179 | + | ||
| 180 | + spin_lock(&inode->i_lock); | ||
| 181 | + inode->i_private = NULL; | ||
| 182 | + wake_up_all(&shmem_falloc_waitq); | ||
| 183 | + spin_unlock(&inode->i_lock); | ||
| 184 | error = 0; | ||
| 185 | - goto undone; | ||
| 186 | + goto out; | ||
| 187 | } | ||
| 188 | |||
| 189 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | ||
| 190 | @@ -1890,6 +1915,7 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset, | ||
| 191 | goto out; | ||
| 192 | } | ||
| 193 | |||
| 194 | + shmem_falloc.waitq = NULL; | ||
| 195 | shmem_falloc.start = start; | ||
| 196 | shmem_falloc.next = start; | ||
| 197 | shmem_falloc.nr_falloced = 0; | ||
| 198 | -- | ||
| 199 | 1.9.1 | ||
| 200 | |||
