DEBIAN-CVE-2026-63888

In the Linux kernel, the following vulnerability has been resolved:  scsi: target: iscsi: Fix CRC overread and double-free in iscsithandletextcmd()  Two latent bugs in the Text-phase handler, both present since the original LIO integration in commit e48354ce078c ("iscsi-target: Add iSCSI fabric support for target v4.1"):  1) DataDigest CRC buffer overread (4 bytes past textin).     textin is kzalloc()'d at ALIGN(payloadlength, 4).  rxsize is then    incremented by ISCSICRCLEN to make room for the received DataDigest    in the iovec, but the same (now-bumped) rxsize is passed as the    buffer length to iscsitcrcbuf():         if (conn->connops->DataDigest) {                ...                rxsize += ISCSICRCLEN;        }        ...        if (conn->connops->DataDigest) {                datacrc = iscsitcrcbuf(textin, rxsize, 0, NULL);     iscsitcrcbuf() walks rxsize bytes of textin with crc32c(), so    when DataDigest is negotiated it reads 4 bytes past the end of the    textin allocation.  KASAN reproduces this directly on the unpatched    mainline tree as slab-out-of-bounds in crc32c() called from the Text    PDU path.  The OOB bytes feed crc32c() and are then compared against    the initiator-supplied checksum, so the value does not flow back to    the attacker, but the kernel does read past the buffer on every Text    PDU with DataDigest=CRC32C.     Fix by passing the actual padded payload length    (ALIGN(payloadlength, 4)) that was used for the kzalloc().  2) Stale cmd->textinptr re-free (double-free) on ERL>0 bad DataDigest    drop.     On DataDigest mismatch with ErrorRecoveryLevel > 0 the handler    silently drops the PDU and lets the initiator plug the CmdSN gap:                 kfree(textin);                return 0;     cmd->textinptr still points at the freed buffer.  The next Text    Request on the same ITT re-enters iscsitsetuptextcmd(), which    unconditionally does         kfree(cmd->textinptr);        cmd->textinptr = NULL;     freeing the same pointer a second time.  Session teardown via    iscsitreleasecmd() has the same shape and hits the same double-free    if the connection is dropped before a second Text Request arrives.     On an unmodified mainline tree the bug-1 CRC overread fires first on    the initial valid Text Request and perturbs the subsequent state, so    #4 was isolated by building a kernel with only the bug-1 hunk of this    patch applied plus temporary printk() observability around the three    relevant kfree() sites.  The observability prints are not part of    this patch.  On that build, a three-PDU Text Request sequence after    login produces two back-to-back splats:         BUG: KASAN: double-free in iscsitsetuptextcmd+0x??        BUG: KASAN: double-free in iscsitreleasecmd+0x??     showing the same pointer freed in the ERL>0 drop path and again in    iscsitsetuptextcmd() (next Text Request on the same ITT) and once    more in iscsitreleasecmd() (session teardown).  On distro kernels    with CONFIGSLABFREELISTHARDENED=y (default) the double-free    becomes a remote kernel BUG(); on non-hardened kernels it corrupts    the slab freelist.     Fix by clearing cmd->textin_ptr after the kfree() in the ERL>0 drop    path.  With both hunks applied #4 is directly observable on the stock    tree without observability printks; fixing bug-1 alone would mask #4    less, not more, so the hunks are submitted together.  Both fixes are one-liners.  The Text PDU state machine is unchanged and the wire protocol is unaffected.