/*
* Report critical errors. ip may be NULL.
*/
void
hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
int error, const char *msg)
{
hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
krateprintf(&hmp->krate,
"HAMMER(%s): Critical error inode=%jd error=%d %s\n",
hmp->mp->mnt_stat.f_mntfromname,
(intmax_t)(ip ? ip->obj_id : -1),
error, msg);
if (hmp->ronly == 0) {
hmp->ronly = 2; /* special errored read-only mode */
hmp->mp->mnt_flag |= MNT_RDONLY;
RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
hammer_adjust_volume_mode, NULL);
kprintf("HAMMER(%s): Forcing read-only mode\n",
hmp->mp->mnt_stat.f_mntfromname);
}
hmp->error = error;
if (hammer_debug_critical)
Debugger("Entering debugger");
}
/*
* Exception, fault, and trap interface to the kernel.
* This common code is called from assembly language IDT gate entry
* routines that prepare a suitable stack frame, and restore this
* frame after the exception has been processed.
*
* This function is also called from doreti in an interlock to handle ASTs.
* For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
*
* NOTE! We have to retrieve the fault address prior to potentially
* blocking, including blocking on any token.
*
* NOTE! NMI and kernel DBG traps remain on their respective pcpu IST
* stacks if taken from a kernel RPL. trap() cannot block in this
* situation. DDB entry or a direct report-and-return is ok.
*
* XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
* if an attempt is made to switch from a fast interrupt or IPI.
*/
void
trap(struct trapframe *frame)
{
static struct krate sscpubugrate = { 1 };
struct globaldata *gd = mycpu;
struct thread *td = gd->gd_curthread;
struct lwp *lp = td->td_lwp;
struct proc *p;
int sticks = 0;
int i = 0, ucode = 0, type, code;
#ifdef INVARIANTS
int crit_count = td->td_critcount;
lwkt_tokref_t curstop = td->td_toks_stop;
#endif
vm_offset_t eva;
p = td->td_proc;
clear_quickret();
#ifdef DDB
/*
* We need to allow T_DNA faults when the debugger is active since
* some dumping paths do large bcopy() which use the floating
* point registers for faster copying.
*/
if (db_active && frame->tf_trapno != T_DNA) {
eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
++gd->gd_trap_nesting_level;
trap_fatal(frame, eva);
--gd->gd_trap_nesting_level;
goto out2;
}
#endif
eva = 0;
if ((frame->tf_rflags & PSL_I) == 0) {
/*
* Buggy application or kernel code has disabled interrupts
* and then trapped. Enabling interrupts now is wrong, but
* it is better than running with interrupts disabled until
* they are accidentally enabled later.
*/
type = frame->tf_trapno;
if (ISPL(frame->tf_cs) == SEL_UPL) {
/* JG curproc can be NULL */
kprintf(
"pid %ld (%s): trap %d with interrupts disabled\n",
(long)curproc->p_pid, curproc->p_comm, type);
} else if ((type == T_STKFLT || type == T_PROTFLT ||
type == T_SEGNPFLT) &&
frame->tf_rip == (long)doreti_iret) {
/*
* iretq fault from kernel mode during return to
* userland.
*
* This situation is expected, don't complain.
*/
} else if (type != T_NMI && type != T_BPTFLT &&
type != T_TRCTRAP) {
/*
* XXX not quite right, since this may be for a
* multiple fault in user mode.
*/
kprintf("kernel trap %d (%s @ 0x%016jx) with "
"interrupts disabled\n",
type,
td->td_comm,
frame->tf_rip);
}
cpu_enable_intr();
}
type = frame->tf_trapno;
code = frame->tf_err;
if (ISPL(frame->tf_cs) == SEL_UPL) {
/* user trap */
KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
frame->tf_trapno, eva);
userenter(td, p);
sticks = (int)td->td_sticks;
KASSERT(lp->lwp_md.md_regs == frame,
("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
switch (type) {
case T_PRIVINFLT: /* privileged instruction fault */
i = SIGILL;
ucode = ILL_PRVOPC;
break;
case T_BPTFLT: /* bpt instruction fault */
case T_TRCTRAP: /* trace trap */
frame->tf_rflags &= ~PSL_T;
i = SIGTRAP;
ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
break;
case T_ARITHTRAP: /* arithmetic trap */
ucode = code;
i = SIGFPE;
break;
case T_ASTFLT: /* Allow process switch */
mycpu->gd_cnt.v_soft++;
if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
atomic_clear_int(&mycpu->gd_reqflags,
RQF_AST_OWEUPC);
addupc_task(p, p->p_prof.pr_addr,
p->p_prof.pr_ticks);
}
goto out;
case T_PROTFLT: /* general protection fault */
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_STKFLT: /* stack fault */
case T_SEGNPFLT: /* segment not present fault */
i = SIGBUS;
ucode = BUS_ADRERR;
break;
case T_TSSFLT: /* invalid TSS fault */
case T_DOUBLEFLT: /* double fault */
default:
i = SIGBUS;
ucode = BUS_OBJERR;
break;
case T_PAGEFLT: /* page fault */
i = trap_pfault(frame, TRUE);
#ifdef DDB
if (frame->tf_rip == 0) {
/* used for kernel debugging only */
while (freeze_on_seg_fault)
tsleep(p, 0, "freeze", hz * 20);
}
#endif
if (i == -1 || i == 0)
goto out;
if (i == SIGSEGV) {
ucode = SEGV_MAPERR;
} else {
i = SIGSEGV;
ucode = SEGV_ACCERR;
}
break;
case T_DIVIDE: /* integer divide fault */
ucode = FPE_INTDIV;
i = SIGFPE;
break;
#if NISA > 0
case T_NMI:
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef DDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (ddb_on_nmi) {
kprintf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* DDB */
goto out2;
} else if (panic_on_nmi)
panic("NMI indicates hardware failure");
break;
#endif /* NISA > 0 */
case T_OFLOW: /* integer overflow fault */
ucode = FPE_INTOVF;
i = SIGFPE;
break;
case T_BOUND: /* bounds check fault */
ucode = FPE_FLTSUB;
i = SIGFPE;
break;
case T_DNA:
/*
* Virtual kernel intercept - pass the DNA exception
* to the virtual kernel if it asked to handle it.
* This occurs when the virtual kernel is holding
* onto the FP context for a different emulated
* process then the one currently running.
*
* We must still call npxdna() since we may have
* saved FP state that the virtual kernel needs
* to hand over to a different emulated process.
*/
if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
(td->td_pcb->pcb_flags & FP_VIRTFP)
) {
npxdna();
break;
}
/*
* The kernel may have switched out the FP unit's
* state, causing the user process to take a fault
* when it tries to use the FP unit. Restore the
* state here
*/
if (npxdna()) {
gd->gd_cnt.v_trap++;
goto out;
}
i = SIGFPE;
ucode = FPE_FPU_NP_TRAP;
break;
case T_FPOPFLT: /* FPU operand fetch fault */
ucode = ILL_COPROC;
i = SIGILL;
break;
case T_XMMFLT: /* SIMD floating-point exception */
ucode = 0; /* XXX */
i = SIGFPE;
break;
}
} else {
/* kernel trap */
switch (type) {
case T_PAGEFLT: /* page fault */
trap_pfault(frame, FALSE);
goto out2;
case T_DNA:
/*
* The kernel is apparently using fpu for copying.
* XXX this should be fatal unless the kernel has
* registered such use.
*/
if (npxdna()) {
gd->gd_cnt.v_trap++;
goto out2;
}
break;
case T_STKFLT: /* stack fault */
case T_PROTFLT: /* general protection fault */
case T_SEGNPFLT: /* segment not present fault */
/*
* Invalid segment selectors and out of bounds
* %rip's and %rsp's can be set up in user mode.
* This causes a fault in kernel mode when the
* kernel tries to return to user mode. We want
* to get this fault so that we can fix the
* problem here and not have to check all the
* selectors and pointers when the user changes
* them.
*/
if (mycpu->gd_intr_nesting_level == 0) {
/*
* NOTE: in 64-bit mode traps push rsp/ss
* even if no ring change occurs.
*/
if (td->td_pcb->pcb_onfault &&
td->td_pcb->pcb_onfault_sp ==
frame->tf_rsp) {
frame->tf_rip = (register_t)
td->td_pcb->pcb_onfault;
goto out2;
}
/*
* If the iretq in doreti faults during
* return to user, it will be special-cased
* in IDTVEC(prot) to get here. We want
* to 'return' to doreti_iret_fault in
* ipl.s in approximately the same state we
* were in at the iretq.
*/
if (frame->tf_rip == (long)doreti_iret) {
frame->tf_rip = (long)doreti_iret_fault;
goto out2;
}
}
break;
case T_TSSFLT:
/*
* PSL_NT can be set in user mode and isn't cleared
* automatically when the kernel is entered. This
* causes a TSS fault when the kernel attempts to
* `iret' because the TSS link is uninitialized. We
* want to get this fault so that we can fix the
* problem here and not every time the kernel is
* entered.
*/
if (frame->tf_rflags & PSL_NT) {
frame->tf_rflags &= ~PSL_NT;
#if 0
/* do we need this? */
if (frame->tf_rip == (long)doreti_iret)
frame->tf_rip = (long)doreti_iret_fault;
#endif
goto out2;
}
break;
case T_TRCTRAP: /* trace trap */
/*
* Detect historical CPU artifact on syscall or int $3
* entry (if not shortcutted in exception.s via
* DIRECT_DISALLOW_SS_CPUBUG).
*/
gd->gd_cnt.v_trap++;
if (frame->tf_rip == (register_t)IDTVEC(fast_syscall)) {
krateprintf(&sscpubugrate,
"Caught #DB at syscall cpu artifact\n");
goto out2;
}
if (frame->tf_rip == (register_t)IDTVEC(bpt)) {
krateprintf(&sscpubugrate,
"Caught #DB at int $N cpu artifact\n");
goto out2;
}
/*
* Ignore debug register trace traps due to
* accesses in the user's address space, which
* can happen under several conditions such as
* if a user sets a watchpoint on a buffer and
* then passes that buffer to a system call.
* We still want to get TRCTRAPS for addresses
* in kernel space because that is useful when
* debugging the kernel.
*/
if (user_dbreg_trap()) {
/*
* Reset breakpoint bits because the
* processor doesn't
*/
load_dr6(rdr6() & ~0xf);
goto out2;
}
/*
* FALLTHROUGH (TRCTRAP kernel mode, kernel address)
*/
case T_BPTFLT:
/*
* If DDB is enabled, let it handle the debugger trap.
* Otherwise, debugger traps "can't happen".
*/
ucode = TRAP_BRKPT;
#ifdef DDB
if (kdb_trap(type, 0, frame))
goto out2;
#endif
break;
#if NISA > 0
case T_NMI:
/* machine/parity/power fail/"kitchen sink" faults */
if (isa_nmi(code) == 0) {
#ifdef DDB
/*
* NMI can be hooked up to a pushbutton
* for debugging.
*/
if (ddb_on_nmi) {
kprintf ("NMI ... going to debugger\n");
kdb_trap(type, 0, frame);
}
#endif /* DDB */
goto out2;
} else if (panic_on_nmi == 0)
goto out2;
/* FALL THROUGH */
#endif /* NISA > 0 */
}
trap_fatal(frame, 0);
goto out2;
}
/*
* Fault from user mode, virtual kernel interecept.
*
* If the fault is directly related to a VM context managed by a
* virtual kernel then let the virtual kernel handle it.
*/
if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
vkernel_trap(lp, frame);
goto out;
}
/* Translate fault for emulators (e.g. Linux) */
if (*p->p_sysent->sv_transtrap)
i = (*p->p_sysent->sv_transtrap)(i, type);
gd->gd_cnt.v_trap++;
trapsignal(lp, i, ucode);
#ifdef DEBUG
if (type <= MAX_TRAP_MSG) {
uprintf("fatal process exception: %s",
trap_msg[type]);
if ((type == T_PAGEFLT) || (type == T_PROTFLT))
uprintf(", fault VA = 0x%lx", frame->tf_addr);
uprintf("\n");
}
#endif
out:
userret(lp, frame, sticks);
userexit(lp);
out2: ;
if (p != NULL && lp != NULL)
KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
#ifdef INVARIANTS
KASSERT(crit_count == td->td_critcount,
("trap: critical section count mismatch! %d/%d",
crit_count, td->td_pri));
KASSERT(curstop == td->td_toks_stop,
("trap: extra tokens held after trap! %ld/%ld",
curstop - &td->td_toks_base,
td->td_toks_stop - &td->td_toks_base));
#endif
}
/*
* This is now called from local media FS's to operate against their
* own vnodes if they fail to implement VOP_PUTPAGES.
*
* This is typically called indirectly via the pageout daemon and
* clustering has already typically occured, so in general we ask the
* underlying filesystem to write the data out asynchronously rather
* then delayed.
*/
int
vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *m, int bytecount,
int flags, int *rtvals)
{
int i;
int maxsize, ncount, count;
vm_ooffset_t poffset;
struct uio auio;
struct iovec aiov;
int error;
int ioflags;
count = bytecount / PAGE_SIZE;
for (i = 0; i < count; i++)
rtvals[i] = VM_PAGER_AGAIN;
if ((int) m[0]->pindex < 0) {
kprintf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n",
(long)m[0]->pindex, m[0]->dirty);
rtvals[0] = VM_PAGER_BAD;
return VM_PAGER_BAD;
}
maxsize = count * PAGE_SIZE;
ncount = count;
poffset = IDX_TO_OFF(m[0]->pindex);
/*
* If the page-aligned write is larger then the actual file we
* have to invalidate pages occuring beyond the file EOF.
*
* If the file EOF resides in the middle of a page we still clear
* all of that page's dirty bits later on. If we didn't it would
* endlessly re-write.
*
* We do not under any circumstances truncate the valid bits, as
* this will screw up bogus page replacement.
*
* The caller has already read-protected the pages. The VFS must
* use the buffer cache to wrap the pages. The pages might not
* be immediately flushed by the buffer cache but once under its
* control the pages themselves can wind up being marked clean
* and their covering buffer cache buffer can be marked dirty.
*/
if (poffset + maxsize > vp->v_filesize) {
if (poffset < vp->v_filesize) {
maxsize = vp->v_filesize - poffset;
ncount = btoc(maxsize);
} else {
maxsize = 0;
ncount = 0;
}
if (ncount < count) {
for (i = ncount; i < count; i++) {
rtvals[i] = VM_PAGER_BAD;
}
}
}
/*
* pageouts are already clustered, use IO_ASYNC to force a bawrite()
* rather then a bdwrite() to prevent paging I/O from saturating
* the buffer cache. Dummy-up the sequential heuristic to cause
* large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
* the system decides how to cluster.
*/
ioflags = IO_VMIO;
if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
ioflags |= IO_SYNC;
else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
ioflags |= IO_ASYNC;
ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
ioflags |= IO_SEQMAX << IO_SEQSHIFT;
aiov.iov_base = (caddr_t) 0;
aiov.iov_len = maxsize;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = poffset;
auio.uio_segflg = UIO_NOCOPY;
auio.uio_rw = UIO_WRITE;
auio.uio_resid = maxsize;
auio.uio_td = NULL;
error = VOP_WRITE(vp, &auio, ioflags, proc0.p_ucred);
mycpu->gd_cnt.v_vnodeout++;
mycpu->gd_cnt.v_vnodepgsout += ncount;
if (error) {
krateprintf(&vbadrate,
"vnode_pager_putpages: I/O error %d\n", error);
}
if (auio.uio_resid) {
krateprintf(&vresrate,
"vnode_pager_putpages: residual I/O %zd at %lu\n",
auio.uio_resid, (u_long)m[0]->pindex);
}
if (error == 0) {
for (i = 0; i < ncount; i++) {
rtvals[i] = VM_PAGER_OK;
vm_page_undirty(m[i]);
}
}
return rtvals[0];
}