fs,seq_file: huge allocations for seq file reads
Hi all,
While fuzzing with trinity inside a KVM tools guest running the latest -next kernel, I've stumbled on the following:
[ 2052.444910] WARNING: CPU: 3 PID: 26525 at mm/page_alloc.c:2513 __alloc_pages_slowpat h+0x6a/0x801() [ 2052.447575] Modules linked in: [ 2052.448438] CPU: 3 PID: 26525 Comm: trinity-c3 Tainted: G W 3.14.0-next-2 0140407-sasha-00023-gd35b0d6 #382 [ 2052.452147] 0000000000000009 ffff88010f485af8 ffffffff9d52ee51 0000000000005d20 [ 2052.454425] 0000000000000000 ffff88010f485b38 ffffffff9a15a2dc 000000174802a016 [ 2052.456676] 00000000001040d0 0000000000000000 0000000000000002 0000000000000000 [ 2052.458851] Call Trace: [ 2052.459587] dump_stack (lib/dump_stack.c:52) [ 2052.461211] warn_slowpath_common (kernel/panic.c:419) [ 2052.462914] warn_slowpath_null (kernel/panic.c:454) [ 2052.464512] __alloc_pages_slowpath (mm/page_alloc.c:2513 (discriminator 3)) [ 2052.466160] ? get_page_from_freelist (mm/page_alloc.c:1939) [ 2052.468020] ? sched_clock_local (kernel/sched/clock.c:213) [ 2052.469633] ? get_parent_ip (kernel/sched/core.c:2471) [ 2052.471329] __alloc_pages_nodemask (mm/page_alloc.c:2766) [ 2052.473084] alloc_pages_current (mm/mempolicy.c:2131) [ 2052.474688] ? __get_free_pages (mm/page_alloc.c:2803) [ 2052.476273] ? __free_pages_ok (arch/x86/include/asm/paravirt.h:809 (discriminator 2) mm/page_alloc.c:766 (discriminator 2)) [ 2052.477980] __get_free_pages (mm/page_alloc.c:2803) [ 2052.481082] kmalloc_order_trace (include/linux/slab.h:379 mm/slab_common.c:525) [ 2052.483193] __kmalloc (include/linux/slab.h:396 mm/slub.c:3303) [ 2052.485417] ? kfree (mm/slub.c:3395) [ 2052.486337] traverse (fs/seq_file.c:141) [ 2052.487289] seq_read (fs/seq_file.c:179 (discriminator 1)) [ 2052.488161] vfs_read (fs/read_write.c:408) [ 2052.489001] SyS_pread64 (include/linux/file.h:38 fs/read_write.c:557 fs/read_write.c:544) [ 2052.489959] tracesys (arch/x86/kernel/entry_64.S:749)
It seems that when we attempt to read huge chunks of data from a seq file there would be no check for the size being read, leading to the kernel attempting to allocate huge chunks of data internally.
As far as I remember, there was a PAGE_SIZE limitation on those, but I'm not certain about that. Could someone please confirm it?
Thanks,
Sasha
Quoted 6 lines · Sasha Levin
It seems that when we attempt to read huge chunks of data from a seq file there would be no check for the size being read, leading to the kernel attempting to allocate huge chunks of data internally.
As far as I remember, there was a
PAGE_SIZElimitation on those, but I'm not certain about that. Could someone please confirm it?
I've had this diff sitting around for a while to figure out which seq_file we're talking about. I think Al wrote it, I forget..
fs/seq_file.c +27 -3
@@ -82,7 +82,7 @@ int seq_open(struct file *file, const struct seq_operations *op) }EXPORT_SYMBOL (seq_open); -static int traverse (struct seq_file *m,loff_t offset) +static int traverse (struct file *file,struct seq_file *m,loff_t offset) {loff_t pos =0 , index;int error =0 ;
@@ -137,6 +137,13 @@ Eoverflow: m->op->stop (m, p);kfree (m->buf); m->count =0 ; +if ((m->size <<=1 ) >= (PAGE_SIZE <<4 )) { +printk (KERN_ERR "traverse on %s. Size: %ld\n" , + file->f_path.dentry->d_name.name, + (unsigned long ) (m->size)); + m->buf =NULL ; +return -ENOMEM ; + } m->buf =kmalloc (m->size <<=1 ,GFP_KERNEL );return !m->buf ? -ENOMEM : -EAGAIN ; }
@@ -176,7 +183,7 @@ ssize_t seq_read(struct file *file, char __user *buf, size_t size, loff_t *ppos)/* Don't assume *ppos is where we left it */ if (unlikely (*ppos != m->read_pos)) { -while ((err =traverse (m, *ppos)) == -EAGAIN ) +while ((err =traverse (file, m, *ppos)) == -EAGAIN ) ;if (err) {/* With prejudice... */
@@ -192,6 +199,14 @@ ssize_t seq_read(struct file *file, char __user *buf, size_t size, loff_t *ppos)/* grab buffer if we didn't have one */ if (!m->buf) { +if ((m->size <<=1 ) >= (PAGE_SIZE <<4 )) { +printk (KERN_ERR "seq_read on %s. Size: %ld\n" , + file->f_path.dentry->d_name.name, + (unsigned long ) (m->size)); + m->buf =NULL ; + goto Enomem;; + } + m->buf =kmalloc (m->size =PAGE_SIZE ,GFP_KERNEL );if (!m->buf) goto Enomem;
@@ -234,6 +249,15 @@ ssize_t seq_read(struct file *file, char __user *buf, size_t size, loff_t *ppos) m->op->stop (m, p);kfree (m->buf); m->count =0 ; + +if ((m->size <<=1 ) >= (PAGE_SIZE <<4 )) { +printk (KERN_ERR "seq_read on %s. Size: %ld\n" , + file->f_path.dentry->d_name.name, + (unsigned long ) (m->size)); + m->buf =NULL ; + goto Enomem;; + } + m->buf =kmalloc (m->size <<=1 ,GFP_KERNEL );if (!m->buf) goto Enomem;
@@ -316,7 +340,7 @@ loff_t seq_lseek(struct file *file, loff_t offset, int whence)break ; retval = offset;if (offset != m->read_pos) { -while ((retval =traverse (m, offset)) == -EAGAIN ) +while ((retval =traverse (file, m, offset)) == -EAGAIN ) ;if (retval) {/* with extreme prejudice... */
Quoted 11 lines · Dave Jones
Quoted 6 lines · Sasha Levin
It seems that when we attempt to read huge chunks of data from a seq file there would be no check for the size being read, leading to the kernel attempting to allocate huge chunks of data internally.
As far as I remember, there was a
PAGE_SIZElimitation on those, but I'm not certain about that. Could someone please confirm it?I've had this diff sitting around for a while to figure out which
seq_filewe're talking about. I think Al wrote it, I forget..
I'll add it in.
I didn't realize it was a problem with a specific file, I thought it was an issue with seq files in general.
Thanks,
Sasha