On Wed, Jun 24, 2020 at 11:07:03AM +0800, Peng Fan wrote: > kmemleak report: > unreferenced object 0x98000002bb591d00 (size 256): > comm "ftest03", pid 24778, jiffies 4301603810 (age 490.665s) > hex dump (first 32 bytes): > 00 01 04 20 01 00 00 00 80 00 00 00 00 00 00 00 ... ............ > f0 02 04 20 01 00 00 00 80 00 00 00 00 00 00 00 ... ............ > backtrace: > [<0000000050b162cb>] __kmalloc+0x234/0x438 > [<00000000491da9c7>] rw_copy_check_uvector+0x1ac/0x1f0 > [<00000000b0dddb43>] import_iovec+0x50/0xe8 > [<00000000ae843d73>] vfs_readv+0x50/0xb0 > [<00000000c7216b06>] do_readv+0x80/0x160 > [<00000000cad79c3f>] syscall_common+0x34/0x58 > > This is because "iov" allocated by kmalloc() is not destroyed. Under normal > circumstances, "ret_pointer" should be equal to "iov". But if the previous > statements fails to execute, and the allocation is successful, then the > block of memory will not be released, because it is necessary to > determine whether they are equal. So we need to change the order. This patch doesn't make sense. It will _introduce_ a memory leak, not fix one.
