/*
* Sanity check volume block size.
*/
int
zvol_check_volblocksize(const char *name, uint64_t volblocksize)
{
/* Record sizes above 128k need the feature to be enabled */
if (volblocksize > SPA_OLD_MAXBLOCKSIZE) {
spa_t *spa;
int error;
if ((error = spa_open(name, &spa, FTAG)) != 0)
return (error);
if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
spa_close(spa, FTAG);
return (SET_ERROR(ENOTSUP));
}
/*
* We don't allow setting the property above 1MB,
* unless the tunable has been changed.
*/
if (volblocksize > zfs_max_recordsize)
return (SET_ERROR(EDOM));
spa_close(spa, FTAG);
}
if (volblocksize < SPA_MINBLOCKSIZE ||
volblocksize > SPA_MAXBLOCKSIZE ||
!ISP2(volblocksize))
return (SET_ERROR(EDOM));
return (0);
}
/*
* Process the buffer of nvlists, unpacking and storing each nvlist record
* into 'records'. 'leftover' is set to the number of bytes that weren't
* processed as there wasn't a complete record.
*/
static int
zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
nvlist_t ***records, uint_t *numrecords)
{
uint64_t reclen;
nvlist_t *nv;
int i;
while (bytes_read > sizeof (reclen)) {
/* get length of packed record (stored as little endian) */
for (i = 0, reclen = 0; i < sizeof (reclen); i++)
reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
if (bytes_read < sizeof (reclen) + reclen)
break;
/* unpack record */
if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
return (ENOMEM);
bytes_read -= sizeof (reclen) + reclen;
buf += sizeof (reclen) + reclen;
/* add record to nvlist array */
(*numrecords)++;
if (ISP2(*numrecords + 1)) {
*records = realloc(*records,
*numrecords * 2 * sizeof (nvlist_t *));
}
(*records)[*numrecords - 1] = nv;
}
*leftover = bytes_read;
return (0);
}
/*
* Create a pool of mblks from which future vio_allocb() requests
* will be serviced.
*
* NOTE: num_mblks has to non-zero and a power-of-2
*
* Returns 0 on success or EINVAL if num_mblks is zero or not
* a power of 2.
*/
int
vio_create_mblks(uint64_t num_mblks, size_t mblk_size, vio_mblk_pool_t **poolp)
{
vio_mblk_pool_t *vmplp;
vio_mblk_t *vmp;
uint8_t *datap;
int i;
if (!(num_mblks) || (!ISP2(num_mblks))) {
*poolp = 0;
return (EINVAL);
}
vmplp = kmem_zalloc(sizeof (*vmplp), KM_SLEEP);
vmplp->quelen = num_mblks;
vmplp->quemask = num_mblks - 1; /* expects quelen is power-of-2 */
vmplp->mblk_size = mblk_size;
mutex_init(&vmplp->hlock, NULL, MUTEX_DRIVER,
DDI_INTR_PRI(DDI_INTR_SOFTPRI_DEFAULT));
mutex_init(&vmplp->tlock, NULL, MUTEX_DRIVER,
DDI_INTR_PRI(DDI_INTR_SOFTPRI_DEFAULT));
vmplp->basep = kmem_zalloc(num_mblks * sizeof (vio_mblk_t), KM_SLEEP);
vmplp->datap = kmem_zalloc(num_mblks * mblk_size, KM_SLEEP);
vmplp->nextp = NULL;
/* create a queue of pointers to free vio_mblk_t's */
vmplp->quep = kmem_zalloc(vmplp->quelen * sizeof (vio_mblk_t *),
KM_SLEEP);
vmplp->head = 0;
vmplp->tail = 0;
for (i = 0, datap = vmplp->datap; i < num_mblks; i++) {
vmp = &(vmplp->basep[i]);
vmp->vmplp = vmplp;
vmp->datap = datap;
vmp->reclaim.free_func = vio_freeb;
vmp->reclaim.free_arg = (caddr_t)vmp;
vmp->mp = desballoc(vmp->datap, mblk_size, BPRI_MED,
&vmp->reclaim);
if (vmp->mp == NULL)
continue;
/* put this vmp on the free stack */
vmplp->quep[vmplp->tail] = vmp;
vmplp->tail = (vmplp->tail + 1) & vmplp->quemask;
datap += mblk_size;
}
*poolp = vmplp;
return (0);
}
int
zvol_check_volblocksize(uint64_t volblocksize)
{
if (volblocksize < SPA_MINBLOCKSIZE ||
volblocksize > SPA_MAXBLOCKSIZE ||
!ISP2(volblocksize))
return (EDOM);
return (0);
}
static void
blksz_changed_cb(void *arg, uint64_t newval)
{
zfs_sb_t *zsb = arg;
ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zsb->z_os)));
ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
ASSERT(ISP2(newval));
zsb->z_max_blksz = newval;
}
int
zvol_check_volblocksize(zfs_cmd_t *zc)
{
if (zc->zc_volblocksize < SPA_MINBLOCKSIZE ||
zc->zc_volblocksize > SPA_MAXBLOCKSIZE ||
!ISP2(zc->zc_volblocksize))
return (EDOM);
return (0);
}
static void
blksz_changed_cb(void *arg, uint64_t newval)
{
zfs_sb_t *zsb = arg;
if (newval < SPA_MINBLOCKSIZE ||
newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
newval = SPA_MAXBLOCKSIZE;
zsb->z_max_blksz = newval;
}
static void
blksz_changed_cb(void *arg, uint64_t newval)
{
zfsvfs_t *zfsvfs = arg;
if (newval < SPA_MINBLOCKSIZE ||
newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
newval = SPA_MAXBLOCKSIZE;
zfsvfs->z_max_blksz = newval;
zfsvfs->z_vfs->vfs_bsize = newval;
}
static boolean_t
pci_cfgacc_valid(pci_cfgacc_req_t *req)
{
int sz = req->size;
if (IS_P2ALIGNED(req->offset, sz) &&
(req->offset + sz - 1 < PCIE_CFG_SPACE_SIZE) &&
((sz & 0xf) && ISP2(sz)))
return (B_TRUE);
cmn_err(CE_WARN, "illegal PCI request: offset = %x, size = %d",
req->offset, sz);
return (B_FALSE);
}
/*
* hermon_cfg_wqe_sizes()
* Context: Only called from attach() path context
*/
static void
hermon_cfg_wqe_sizes(hermon_state_t *state, hermon_cfg_profile_t *cp)
{
uint_t max_size, log2;
uint_t max_sgl, real_max_sgl;
/*
* Get the requested maximum number SGL per WQE from the Hermon
* patchable variable
*/
max_sgl = hermon_wqe_max_sgl;
/*
* Use requested maximum number of SGL to calculate the max descriptor
* size (while guaranteeing that the descriptor size is a power-of-2
* cachelines). We have to use the calculation for QP1 MLX transport
* because the possibility that we might need to inline a GRH, along
* with all the other headers and alignment restrictions, sets the
* maximum for the number of SGLs that we can advertise support for.
*/
max_size = (HERMON_QP_WQE_MLX_QP1_HDRS + (max_sgl << 4));
log2 = highbit(max_size);
if (ISP2(max_size)) {
log2 = log2 - 1;
}
max_size = (1 << log2);
max_size = min(max_size, state->hs_devlim.max_desc_sz_sq);
/*
* Then use the calculated max descriptor size to determine the "real"
* maximum SGL (the number beyond which we would roll over to the next
* power-of-2).
*/
real_max_sgl = (max_size - HERMON_QP_WQE_MLX_QP1_HDRS) >> 4;
/* Then save away this configuration information */
cp->cp_wqe_max_sgl = max_sgl;
cp->cp_wqe_real_max_sgl = real_max_sgl;
/* SRQ SGL gets set to it's own patchable variable value */
cp->cp_srq_max_sgl = hermon_srq_max_sgl;
}
void
rds_hash_init()
{
int i;
if (!ISP2(rds_bind_fanout_size)) {
/* Not a power of two. Round up to nearest power of two */
for (i = 0; i < UINT_32_BITS; i++) {
if (rds_bind_fanout_size < (1 << i))
break;
}
rds_bind_fanout_size = 1 << i;
}
rds_bind_fanout = kmem_zalloc(rds_bind_fanout_size *
sizeof (rds_bf_t), KM_SLEEP);
for (i = 0; i < rds_bind_fanout_size; i++) {
mutex_init(&rds_bind_fanout[i].rds_bf_lock, NULL, MUTEX_DEFAULT,
NULL);
}
}
static uint64_t
zpios_dmu_object_create(run_args_t *run_args, objset_t *os)
{
struct dmu_tx *tx;
uint64_t obj = 0ULL;
uint64_t blksize = run_args->block_size;
int rc;
if (blksize < SPA_MINBLOCKSIZE ||
blksize > spa_maxblocksize(dmu_objset_spa(os)) ||
!ISP2(blksize)) {
zpios_print(run_args->file,
"invalid block size for pool: %d\n", (int)blksize);
return (obj);
}
tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, OBJ_SIZE);
rc = dmu_tx_assign(tx, TXG_WAIT);
if (rc) {
zpios_print(run_args->file,
"dmu_tx_assign() failed: %d\n", rc);
dmu_tx_abort(tx);
return (obj);
}
obj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, DMU_OT_NONE, 0, tx);
rc = dmu_object_set_blocksize(os, obj, blksize, 0, tx);
if (rc) {
zpios_print(run_args->file,
"dmu_object_set_blocksize to %d failed: %d\n",
(int)blksize, rc);
dmu_tx_abort(tx);
return (obj);
}
dmu_tx_commit(tx);
return (obj);
}
/* Check pitch constriants for all chips & tiling formats */
static int
i915_tiling_ok(struct drm_device *dev, int stride, int size, int tiling_mode)
{
int tile_width;
/* Linear is always fine */
if (tiling_mode == I915_TILING_NONE)
return 1;
if (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev))
tile_width = 128;
else
tile_width = 512;
if (stride == 0)
return 0;
/* 965+ just needs multiples of tile width */
if (IS_I965G(dev)) {
if (stride & (tile_width - 1))
return 0;
return 1;
}
/* Pre-965 needs power of two tile widths */
if (stride < tile_width)
return 0;
if (!ISP2(stride))
return 0;
/* We don't handle the aperture area covered by the fence being bigger
* than the object size.
*/
if (i915_get_fence_size(dev, size) != size)
return 0;
return 1;
}
/*ARGSUSED*/
void *
segkmem_alloc_lp(vmem_t *vmp, size_t *sizep, size_t align, int vmflag)
{
size_t size;
kthread_t *t = curthread;
segkmem_lpcb_t *lpcb = &segkmem_lpcb;
ASSERT(sizep != NULL);
size = *sizep;
if (lpcb->lp_uselp && !(t->t_flag & T_PANIC) &&
!(vmflag & SEGKMEM_SHARELOCKED)) {
size_t kmemlp_qnt = segkmem_kmemlp_quantum;
size_t asize = P2ROUNDUP(size, kmemlp_qnt);
void *addr = NULL;
ulong_t *lpthrtp = &lpcb->lp_throttle;
ulong_t lpthrt = *lpthrtp;
int dowakeup = 0;
int doalloc = 1;
ASSERT(kmem_lp_arena != NULL);
ASSERT(asize >= size);
if (lpthrt != 0) {
/* try to update the throttle value */
lpthrt = atomic_inc_ulong_nv(lpthrtp);
if (lpthrt >= segkmem_lpthrottle_max) {
lpthrt = atomic_cas_ulong(lpthrtp, lpthrt,
segkmem_lpthrottle_max / 4);
}
/*
* when we get above throttle start do an exponential
* backoff at trying large pages and reaping
*/
if (lpthrt > segkmem_lpthrottle_start &&
!ISP2(lpthrt)) {
lpcb->allocs_throttled++;
lpthrt--;
if (ISP2(lpthrt))
kmem_reap();
return (segkmem_alloc(vmp, size, vmflag));
}
}
if (!(vmflag & VM_NOSLEEP) &&
segkmem_heaplp_quantum >= (8 * kmemlp_qnt) &&
vmem_size(kmem_lp_arena, VMEM_FREE) <= kmemlp_qnt &&
asize < (segkmem_heaplp_quantum - kmemlp_qnt)) {
/*
* we are low on free memory in kmem_lp_arena
* we let only one guy to allocate heap_lp
* quantum size chunk that everybody is going to
* share
*/
mutex_enter(&lpcb->lp_lock);
if (lpcb->lp_wait) {
/* we are not the first one - wait */
cv_wait(&lpcb->lp_cv, &lpcb->lp_lock);
if (vmem_size(kmem_lp_arena, VMEM_FREE) <
kmemlp_qnt) {
doalloc = 0;
}
} else if (vmem_size(kmem_lp_arena, VMEM_FREE) <=
kmemlp_qnt) {
/*
* we are the first one, make sure we import
* a large page
*/
if (asize == kmemlp_qnt)
asize += kmemlp_qnt;
dowakeup = 1;
lpcb->lp_wait = 1;
}
mutex_exit(&lpcb->lp_lock);
}
/*
* VM_ABORT flag prevents sleeps in vmem_xalloc when
* large pages are not available. In that case this allocation
* attempt will fail and we will retry allocation with small
* pages. We also do not want to panic if this allocation fails
* because we are going to retry.
*/
if (doalloc) {
addr = vmem_alloc(kmem_lp_arena, asize,
(vmflag | VM_ABORT) & ~VM_PANIC);
if (dowakeup) {
mutex_enter(&lpcb->lp_lock);
ASSERT(lpcb->lp_wait != 0);
lpcb->lp_wait = 0;
cv_broadcast(&lpcb->lp_cv);
mutex_exit(&lpcb->lp_lock);
}
}
if (addr != NULL) {
*sizep = asize;
*lpthrtp = 0;
return (addr);
}
if (vmflag & VM_NOSLEEP)
lpcb->nosleep_allocs_failed++;
else
lpcb->sleep_allocs_failed++;
lpcb->alloc_bytes_failed += size;
/* if large page throttling is not started yet do it */
if (segkmem_use_lpthrottle && lpthrt == 0) {
lpthrt = atomic_cas_ulong(lpthrtp, lpthrt, 1);
}
}
return (segkmem_alloc(vmp, size, vmflag));
}
void
cpu_intrq_cleanup(struct cpu *cpu)
{
struct machcpu *mcpup = &cpu->cpu_m;
int cpu_list_size;
uint64_t cpu_q_size;
uint64_t dev_q_size;
uint64_t cpu_rq_size;
uint64_t cpu_nrq_size;
/*
* Free mondo data for xcalls.
*/
if (mcpup->mondo_data) {
contig_mem_free(mcpup->mondo_data, INTR_REPORT_SIZE);
mcpup->mondo_data = NULL;
mcpup->mondo_data_ra = NULL;
}
/*
* Free per-cpu list of ncpu_guest_max for xcalls
*/
cpu_list_size = ncpu_guest_max * sizeof (uint16_t);
if (cpu_list_size < INTR_REPORT_SIZE)
cpu_list_size = INTR_REPORT_SIZE;
/*
* contig_mem_alloc() requires size to be a power of 2.
* Increase size to a power of 2 if necessary.
*/
if (!ISP2(cpu_list_size)) {
cpu_list_size = 1 << highbit(cpu_list_size);
}
if (mcpup->cpu_list) {
contig_mem_free(mcpup->cpu_list, cpu_list_size);
mcpup->cpu_list = NULL;
mcpup->cpu_list_ra = NULL;
}
/*
* Free sun4v interrupt and error queues.
*/
if (mcpup->cpu_q_va) {
cpu_q_size = cpu_q_entries * INTR_REPORT_SIZE;
contig_mem_free(mcpup->cpu_q_va, cpu_q_size);
mcpup->cpu_q_va = NULL;
mcpup->cpu_q_base_pa = NULL;
mcpup->cpu_q_size = 0;
}
if (mcpup->dev_q_va) {
dev_q_size = dev_q_entries * INTR_REPORT_SIZE;
contig_mem_free(mcpup->dev_q_va, dev_q_size);
mcpup->dev_q_va = NULL;
mcpup->dev_q_base_pa = NULL;
mcpup->dev_q_size = 0;
}
if (mcpup->cpu_rq_va) {
cpu_rq_size = cpu_rq_entries * Q_ENTRY_SIZE;
contig_mem_free(mcpup->cpu_rq_va, 2 * cpu_rq_size);
mcpup->cpu_rq_va = NULL;
mcpup->cpu_rq_base_pa = NULL;
mcpup->cpu_rq_size = 0;
}
if (mcpup->cpu_nrq_va) {
cpu_nrq_size = cpu_nrq_entries * Q_ENTRY_SIZE;
contig_mem_free(mcpup->cpu_nrq_va, 2 * cpu_nrq_size);
mcpup->cpu_nrq_va = NULL;
mcpup->cpu_nrq_base_pa = NULL;
mcpup->cpu_nrq_size = 0;
}
}
int
cpu_intrq_setup(struct cpu *cpu)
{
struct machcpu *mcpup = &cpu->cpu_m;
size_t size;
/*
* This routine will return with an error return if any
* contig_mem_alloc() fails. It is expected that the caller will
* call cpu_intrq_cleanup() (or cleanup_cpu_common() which will).
* That will cleanly free only those blocks that were alloc'd.
*/
/*
* Allocate mondo data for xcalls.
*/
mcpup->mondo_data = contig_mem_alloc(INTR_REPORT_SIZE);
if (mcpup->mondo_data == NULL) {
cmn_err(CE_NOTE, "cpu%d: cpu mondo_data allocation failed",
cpu->cpu_id);
return (ENOMEM);
}
/*
* va_to_pa() is too expensive to call for every crosscall
* so we do it here at init time and save it in machcpu.
*/
mcpup->mondo_data_ra = va_to_pa(mcpup->mondo_data);
/*
* Allocate a per-cpu list of ncpu_guest_max for xcalls
*/
size = ncpu_guest_max * sizeof (uint16_t);
if (size < INTR_REPORT_SIZE)
size = INTR_REPORT_SIZE;
/*
* contig_mem_alloc() requires size to be a power of 2.
* Increase size to a power of 2 if necessary.
*/
if (!ISP2(size)) {
size = 1 << highbit(size);
}
mcpup->cpu_list = contig_mem_alloc(size);
if (mcpup->cpu_list == NULL) {
cmn_err(CE_NOTE, "cpu%d: cpu cpu_list allocation failed",
cpu->cpu_id);
return (ENOMEM);
}
mcpup->cpu_list_ra = va_to_pa(mcpup->cpu_list);
/*
* Allocate sun4v interrupt and error queues.
*/
size = cpu_q_entries * INTR_REPORT_SIZE;
mcpup->cpu_q_va = contig_mem_alloc(size);
if (mcpup->cpu_q_va == NULL) {
cmn_err(CE_NOTE, "cpu%d: cpu intrq allocation failed",
cpu->cpu_id);
return (ENOMEM);
}
mcpup->cpu_q_base_pa = va_to_pa(mcpup->cpu_q_va);
mcpup->cpu_q_size = size;
/*
* Allocate device queues
*/
size = dev_q_entries * INTR_REPORT_SIZE;
mcpup->dev_q_va = contig_mem_alloc(size);
if (mcpup->dev_q_va == NULL) {
cmn_err(CE_NOTE, "cpu%d: dev intrq allocation failed",
cpu->cpu_id);
return (ENOMEM);
}
mcpup->dev_q_base_pa = va_to_pa(mcpup->dev_q_va);
mcpup->dev_q_size = size;
/*
* Allocate resumable queue and its kernel buffer
*/
size = cpu_rq_entries * Q_ENTRY_SIZE;
mcpup->cpu_rq_va = contig_mem_alloc(2 * size);
if (mcpup->cpu_rq_va == NULL) {
cmn_err(CE_NOTE, "cpu%d: resumable queue allocation failed",
cpu->cpu_id);
return (ENOMEM);
}
mcpup->cpu_rq_base_pa = va_to_pa(mcpup->cpu_rq_va);
mcpup->cpu_rq_size = size;
/* zero out the memory */
bzero(mcpup->cpu_rq_va, 2 * size);
/*
* Allocate non-resumable queues
*/
size = cpu_nrq_entries * Q_ENTRY_SIZE;
mcpup->cpu_nrq_va = contig_mem_alloc(2 * size);
if (mcpup->cpu_nrq_va == NULL) {
cmn_err(CE_NOTE, "cpu%d: nonresumable queue allocation failed",
cpu->cpu_id);
return (ENOMEM);
}
mcpup->cpu_nrq_base_pa = va_to_pa(mcpup->cpu_nrq_va);
mcpup->cpu_nrq_size = size;
/* zero out the memory */
bzero(mcpup->cpu_nrq_va, 2 * size);
return (0);
}
/*
* Interrupt allocate/free functions
*/
int
ddi_intr_alloc(dev_info_t *dip, ddi_intr_handle_t *h_array, int type, int inum,
int count, int *actualp, int behavior)
{
ddi_intr_handle_impl_t *hdlp, tmp_hdl;
int i, ret, cap = 0, curr_type, nintrs;
uint_t pri, navail, curr_nintrs = 0;
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: name %s dip 0x%p "
"type %x inum %x count %x behavior %x\n", ddi_driver_name(dip),
(void *)dip, type, inum, count, behavior));
/* Validate parameters */
if ((dip == NULL) || (h_array == NULL) || (inum < 0) || (count < 1) ||
(actualp == NULL) || !DDI_INTR_BEHAVIOR_FLAG_VALID(behavior)) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: "
"Invalid input args\n"));
return (DDI_EINVAL);
}
/* Validate interrupt type */
if (!DDI_INTR_TYPE_FLAG_VALID(type) ||
!(i_ddi_intr_get_supported_types(dip) & type)) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: type %x not "
"supported\n", type));
return (DDI_EINVAL);
}
/* Validate inum not previously allocated */
if ((type == DDI_INTR_TYPE_FIXED) &&
(i_ddi_get_intr_handle(dip, inum) != NULL)) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: inum %d is already "
"in use, cannot allocate again!!\n", inum));
return (DDI_EINVAL);
}
/* Get how many interrupts the device supports */
if ((nintrs = i_ddi_intr_get_supported_nintrs(dip, type)) == 0) {
if (ddi_intr_get_nintrs(dip, type, &nintrs) != DDI_SUCCESS) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: no "
"interrupts found of type %d\n", type));
return (DDI_INTR_NOTFOUND);
}
}
/* Get how many interrupts the device is already using */
if ((curr_type = i_ddi_intr_get_current_type(dip)) != 0) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: type %x "
"is already being used\n", curr_type));
curr_nintrs = i_ddi_intr_get_current_nintrs(dip);
}
/* Validate interrupt type consistency */
if ((curr_type != 0) && (type != curr_type)) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: Requested "
"interrupt type %x is different from interrupt type %x"
"already in use\n", type, curr_type));
return (DDI_EINVAL);
}
/* Validate count does not exceed what device supports */
if (count > nintrs) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: no of interrupts "
"requested %d is more than supported %d\n", count, nintrs));
return (DDI_EINVAL);
} else if ((count + curr_nintrs) > nintrs) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: count %d "
"+ intrs in use %d exceeds supported %d intrs\n",
count, curr_nintrs, nintrs));
return (DDI_EINVAL);
}
/* Validate power of 2 requirements for MSI */
if ((type == DDI_INTR_TYPE_MSI) && !ISP2(curr_nintrs + count)) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: "
"MSI count %d is not a power of two\n", count));
return (DDI_EINVAL);
}
/*
* Initialize the device's interrupt information structure,
* and establish an association with IRM if it is supported.
*
* NOTE: IRM checks minimum support, and can return DDI_EAGAIN.
*/
if (curr_nintrs == 0) {
i_ddi_intr_devi_init(dip);
if (i_ddi_irm_insert(dip, type, count) == DDI_EAGAIN) {
cmn_err(CE_WARN, "ddi_intr_alloc: "
"cannot fit into interrupt pool\n");
return (DDI_EAGAIN);
}
}
/* Synchronously adjust IRM associations for non-IRM aware drivers */
if (curr_nintrs && (i_ddi_irm_supported(dip, type) != DDI_SUCCESS))
(void) i_ddi_irm_modify(dip, count + curr_nintrs);
/* Get how many interrupts are currently available */
navail = i_ddi_intr_get_current_navail(dip, type);
/* Validate that requested number of interrupts are available */
if (curr_nintrs == navail) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: max # of intrs %d "
"already allocated\n", navail));
return (DDI_EAGAIN);
}
if ((count + curr_nintrs) > navail) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: requested # of "
"intrs %d exceeds # of available intrs %d\n", count,
navail - curr_nintrs));
if (behavior == DDI_INTR_ALLOC_STRICT) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: "
"DDI_INTR_ALLOC_STRICT flag is passed, "
"return failure\n"));
if (curr_nintrs == 0)
i_ddi_intr_devi_fini(dip);
else if (i_ddi_irm_supported(dip, type) != DDI_SUCCESS)
(void) i_ddi_irm_modify(dip, curr_nintrs);
return (DDI_EAGAIN);
}
count = navail - curr_nintrs;
}
/* Now allocate required number of interrupts */
bzero(&tmp_hdl, sizeof (ddi_intr_handle_impl_t));
tmp_hdl.ih_type = type;
tmp_hdl.ih_inum = inum;
tmp_hdl.ih_scratch1 = count;
tmp_hdl.ih_scratch2 = (void *)(uintptr_t)behavior;
tmp_hdl.ih_dip = dip;
if (i_ddi_intr_ops(dip, dip, DDI_INTROP_ALLOC,
&tmp_hdl, (void *)actualp) != DDI_SUCCESS) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: allocation "
"failed\n"));
i_ddi_intr_devi_fini(dip);
return (*actualp ? DDI_EAGAIN : DDI_INTR_NOTFOUND);
}
if ((ret = i_ddi_intr_ops(dip, dip, DDI_INTROP_GETPRI,
&tmp_hdl, (void *)&pri)) != DDI_SUCCESS) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: get priority "
"failed\n"));
goto fail;
}
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: getting capability\n"));
if ((ret = i_ddi_intr_ops(dip, dip, DDI_INTROP_GETCAP,
&tmp_hdl, (void *)&cap)) != DDI_SUCCESS) {
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: get capability "
"failed\n"));
goto fail;
}
/*
* Save current interrupt type, supported and current intr count.
*/
i_ddi_intr_set_current_type(dip, type);
i_ddi_intr_set_supported_nintrs(dip, nintrs);
i_ddi_intr_set_current_nintrs(dip,
i_ddi_intr_get_current_nintrs(dip) + *actualp);
/* Now, go and handle each "handle" */
for (i = inum; i < (inum + *actualp); i++) {
hdlp = (ddi_intr_handle_impl_t *)kmem_zalloc(
(sizeof (ddi_intr_handle_impl_t)), KM_SLEEP);
rw_init(&hdlp->ih_rwlock, NULL, RW_DRIVER, NULL);
h_array[i] = (struct __ddi_intr_handle *)hdlp;
hdlp->ih_type = type;
hdlp->ih_pri = pri;
hdlp->ih_cap = cap;
hdlp->ih_ver = DDI_INTR_VERSION;
hdlp->ih_state = DDI_IHDL_STATE_ALLOC;
hdlp->ih_dip = dip;
hdlp->ih_inum = i;
i_ddi_alloc_intr_phdl(hdlp);
if (type & DDI_INTR_TYPE_FIXED)
i_ddi_set_intr_handle(dip, hdlp->ih_inum,
(ddi_intr_handle_t)hdlp);
DDI_INTR_APIDBG((CE_CONT, "ddi_intr_alloc: hdlp = 0x%p\n",
(void *)h_array[i]));
}
return (DDI_SUCCESS);
fail:
(void) i_ddi_intr_ops(tmp_hdl.ih_dip, tmp_hdl.ih_dip,
DDI_INTROP_FREE, &tmp_hdl, NULL);
i_ddi_intr_devi_fini(dip);
return (ret);
}
/*
* XXX - right now we aren't attempting to fix anything that looks bad,
* instead we just give up.
*/
void
readBPB(int fd)
{
boot_sector_t ubpb;
/*
* The BPB is the first sector of the file system
*/
if (lseek64(fd, PartitionOffset, SEEK_SET) < 0) {
mountSanityCheckFails();
perror(gettext("Cannot seek to start of disk partition"));
(void) close(fd);
exit(7);
}
if (Verbose)
(void) fprintf(stderr,
gettext("Reading BIOS parameter block\n"));
if (read(fd, ubpb.buf, BPSEC) < BPSEC) {
mountSanityCheckFails();
perror(gettext("Read BIOS parameter block"));
(void) close(fd);
exit(2);
}
if (ltohs(ubpb.mb.signature) != BOOTSECSIG) {
mountSanityCheckFails();
(void) fprintf(stderr,
gettext("Bad signature on BPB. Giving up.\n"));
exit(2);
}
#ifdef _BIG_ENDIAN
swap_pack_grabbpb(&TheBIOSParameterBlock, &(ubpb.bs));
#else
(void) memcpy(&(TheBIOSParameterBlock.bpb), &(ubpb.bs.bs_front.bs_bpb),
sizeof (TheBIOSParameterBlock.bpb));
(void) memcpy(&(TheBIOSParameterBlock.ebpb), &(ubpb.bs.bs_ebpb),
sizeof (TheBIOSParameterBlock.ebpb));
#endif
/*
* In general, we would expect the bytes per sector to
* equal 256 (BPSEC). I personally have yet to see a file
* system where this isn't true but apparently some weird media
* have different sector sizes. So we'll accept a couple of
* other small multiples of 256 as valid sizes.
*/
if (TheBIOSParameterBlock.bpb.bytes_per_sector != BPSEC &&
TheBIOSParameterBlock.bpb.bytes_per_sector != 2 * BPSEC &&
TheBIOSParameterBlock.bpb.bytes_per_sector != 4 * BPSEC) {
mountSanityCheckFails();
(void) fprintf(stderr,
gettext("Bogus bytes per sector value. Giving up.\n"));
exit(2);
}
if (!(ISP2(TheBIOSParameterBlock.bpb.sectors_per_cluster) &&
IN_RANGE(TheBIOSParameterBlock.bpb.sectors_per_cluster,
1, 128))) {
mountSanityCheckFails();
(void) fprintf(stderr,
gettext("Bogus sectors per cluster value. Giving up.\n"));
(void) close(fd);
exit(6);
}
if (TheBIOSParameterBlock.bpb.sectors_per_fat == 0) {
#ifdef _BIG_ENDIAN
swap_pack_grab32bpb(&TheBIOSParameterBlock, &(ubpb.bs));
#else
(void) memcpy(&(TheBIOSParameterBlock.bpb32),
&(ubpb.bs32.bs_bpb32),
sizeof (TheBIOSParameterBlock.bpb32));
#endif
IsFAT32 = 1;
}
if (!IsFAT32) {
if ((TheBIOSParameterBlock.bpb.num_root_entries == 0) ||
((TheBIOSParameterBlock.bpb.num_root_entries *
sizeof (struct pcdir)) %
TheBIOSParameterBlock.bpb.bytes_per_sector) != 0) {
mountSanityCheckFails();
(void) fprintf(stderr,
gettext("Bogus number of root entries. "
"Giving up.\n"));
exit(2);
}
} else {
if (TheBIOSParameterBlock.bpb.num_root_entries != 0) {
mountSanityCheckFails();
(void) fprintf(stderr,
gettext("Bogus number of root entries. "
"Giving up.\n"));
exit(2);
}
}
/*
* In general, we would expect the number of FATs field to
* equal 2. Our mkfs and Windows have this as a default
* value. I suppose someone could override the default,
* though, so we'll sort of arbitrarily accept any number
* between 1 and 4 inclusive as reasonable values.
*
* XXX: Warn, but continue, if value is suspicious? (>2?)
*/
if (TheBIOSParameterBlock.bpb.num_fats > 4 ||
TheBIOSParameterBlock.bpb.num_fats < 1) {
mountSanityCheckFails();
(void) fprintf(stderr,
gettext("Bogus number of FATs. Giving up.\n"));
exit(2);
}
computeFileAreaSize();
}
/*
* Increase the file length
*
* IN: zp - znode of file to free data in.
* end - new end-of-file
*
* RETURN: 0 on success, error code on failure
*/
static int
zfs_extend(znode_t *zp, uint64_t end)
{
zfs_sb_t *zsb = ZTOZSB(zp);
dmu_tx_t *tx;
rl_t *rl;
uint64_t newblksz;
int error;
/*
* We will change zp_size, lock the whole file.
*/
rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
/*
* Nothing to do if file already at desired length.
*/
if (end <= zp->z_size) {
zfs_range_unlock(rl);
return (0);
}
tx = dmu_tx_create(zsb->z_os);
dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
zfs_sa_upgrade_txholds(tx, zp);
if (end > zp->z_blksz &&
(!ISP2(zp->z_blksz) || zp->z_blksz < zsb->z_max_blksz)) {
/*
* We are growing the file past the current block size.
*/
if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
ASSERT(!ISP2(zp->z_blksz));
newblksz = MIN(end, SPA_MAXBLOCKSIZE);
} else {
newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
}
dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
} else {
newblksz = 0;
}
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
zfs_range_unlock(rl);
return (error);
}
if (newblksz)
zfs_grow_blocksize(zp, newblksz, tx);
zp->z_size = end;
VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
&zp->z_size, sizeof (zp->z_size), tx));
zfs_range_unlock(rl);
dmu_tx_commit(tx);
return (0);
}
int i;
ASSERT3U(dn->dn_indblkshift, >=, 0);
ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT);
if (dn->dn_datablkshift) {
ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT);
ASSERT3U(dn->dn_datablkshift, <=, SPA_MAXBLOCKSHIFT);
ASSERT3U(1<<dn->dn_datablkshift, ==, dn->dn_datablksz);
}
ASSERT3U(dn->dn_nlevels, <=, 30);
ASSERT3U(dn->dn_type, <=, DMU_OT_NUMTYPES);
ASSERT3U(dn->dn_nblkptr, >=, 1);
ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR);
ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN);
ASSERT3U(dn->dn_datablksz, ==,
dn->dn_datablkszsec << SPA_MINBLOCKSHIFT);
ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0);
ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) +
dn->dn_bonuslen, <=, DN_MAX_BONUSLEN);
for (i = 0; i < TXG_SIZE; i++) {
ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels);
}
}
if (dn->dn_phys->dn_type != DMU_OT_NONE)
ASSERT3U(dn->dn_phys->dn_nlevels, <=, dn->dn_nlevels);
ASSERT(DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_dbuf != NULL);
if (dn->dn_dbuf != NULL) {
ASSERT3P(dn->dn_phys, ==,
(dnode_phys_t *)dn->dn_dbuf->db.db_data +
(dn->dn_object % (dn->dn_dbuf->db.db_size >> DNODE_SHIFT)));
}
if (drop_struct_lock)
static int
smmap_common(caddr_t *addrp, size_t len,
int prot, int flags, struct file *fp, offset_t pos)
{
struct vnode *vp;
struct as *as = curproc->p_as;
uint_t uprot, maxprot, type;
int error;
int in_crit = 0;
if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | _MAP_NEW |
_MAP_LOW32 | MAP_NORESERVE | MAP_ANON | MAP_ALIGN |
MAP_TEXT | MAP_INITDATA)) != 0) {
/* | MAP_RENAME */ /* not implemented, let user know */
return (EINVAL);
}
if ((flags & MAP_TEXT) && !(prot & PROT_EXEC)) {
return (EINVAL);
}
if ((flags & (MAP_TEXT | MAP_INITDATA)) == (MAP_TEXT | MAP_INITDATA)) {
return (EINVAL);
}
#if defined(__sparc)
/*
* See if this is an "old mmap call". If so, remember this
* fact and convert the flags value given to mmap to indicate
* the specified address in the system call must be used.
* _MAP_NEW is turned set by all new uses of mmap.
*/
if ((flags & _MAP_NEW) == 0)
flags |= MAP_FIXED;
#endif
flags &= ~_MAP_NEW;
type = flags & MAP_TYPE;
if (type != MAP_PRIVATE && type != MAP_SHARED)
return (EINVAL);
if (flags & MAP_ALIGN) {
if (flags & MAP_FIXED)
return (EINVAL);
/* alignment needs to be a power of 2 >= page size */
if (((uintptr_t)*addrp < PAGESIZE && (uintptr_t)*addrp != 0) ||
!ISP2((uintptr_t)*addrp))
return (EINVAL);
}
/*
* Check for bad lengths and file position.
* We let the VOP_MAP routine check for negative lengths
* since on some vnode types this might be appropriate.
*/
if (len == 0 || (pos & (u_offset_t)PAGEOFFSET) != 0)
return (EINVAL);
maxprot = PROT_ALL; /* start out allowing all accesses */
uprot = prot | PROT_USER;
if (fp == NULL) {
ASSERT(flags & MAP_ANON);
/* discard lwpchan mappings, like munmap() */
if ((flags & MAP_FIXED) && curproc->p_lcp != NULL)
lwpchan_delete_mapping(curproc, *addrp, *addrp + len);
as_rangelock(as);
error = zmap(as, addrp, len, uprot, flags, pos);
as_rangeunlock(as);
/*
* Tell machine specific code that lwp has mapped shared memory
*/
if (error == 0 && (flags & MAP_SHARED)) {
/* EMPTY */
LWP_MMODEL_SHARED_AS(*addrp, len);
}
return (error);
} else if ((flags & MAP_ANON) != 0)
return (EINVAL);
vp = fp->f_vnode;
/* Can't execute code from "noexec" mounted filesystem. */
if ((vp->v_vfsp->vfs_flag & VFS_NOEXEC) != 0)
maxprot &= ~PROT_EXEC;
/*
* These checks were added as part of large files.
*
* Return ENXIO if the initial position is negative; return EOVERFLOW
* if (offset + len) would overflow the maximum allowed offset for the
* type of file descriptor being used.
*/
if (vp->v_type == VREG) {
if (pos < 0)
return (ENXIO);
if ((offset_t)len > (OFFSET_MAX(fp) - pos))
return (EOVERFLOW);
}
if (type == MAP_SHARED && (fp->f_flag & FWRITE) == 0) {
/* no write access allowed */
maxprot &= ~PROT_WRITE;
}
/*
* XXX - Do we also adjust maxprot based on protections
* of the vnode? E.g. if no execute permission is given
* on the vnode for the current user, maxprot probably
* should disallow PROT_EXEC also? This is different
* from the write access as this would be a per vnode
* test as opposed to a per fd test for writability.
*/
/*
* Verify that the specified protections are not greater than
* the maximum allowable protections. Also test to make sure
* that the file descriptor does allows for read access since
* "write only" mappings are hard to do since normally we do
* the read from the file before the page can be written.
*/
if (((maxprot & uprot) != uprot) || (fp->f_flag & FREAD) == 0)
return (EACCES);
/*
* If the user specified an address, do some simple checks here
*/
if ((flags & MAP_FIXED) != 0) {
caddr_t userlimit;
/*
* Use the user address. First verify that
* the address to be used is page aligned.
* Then make some simple bounds checks.
*/
if (((uintptr_t)*addrp & PAGEOFFSET) != 0)
return (EINVAL);
userlimit = flags & _MAP_LOW32 ?
(caddr_t)USERLIMIT32 : as->a_userlimit;
switch (valid_usr_range(*addrp, len, uprot, as, userlimit)) {
case RANGE_OKAY:
break;
case RANGE_BADPROT:
return (ENOTSUP);
case RANGE_BADADDR:
default:
return (ENOMEM);
}
}
if ((prot & (PROT_READ | PROT_WRITE | PROT_EXEC)) &&
nbl_need_check(vp)) {
int svmand;
nbl_op_t nop;
nbl_start_crit(vp, RW_READER);
in_crit = 1;
error = nbl_svmand(vp, fp->f_cred, &svmand);
if (error != 0)
goto done;
if ((prot & PROT_WRITE) && (type == MAP_SHARED)) {
if (prot & (PROT_READ | PROT_EXEC)) {
nop = NBL_READWRITE;
} else {
nop = NBL_WRITE;
}
} else {
nop = NBL_READ;
}
if (nbl_conflict(vp, nop, 0, LONG_MAX, svmand, NULL)) {
error = EACCES;
goto done;
}
}
/* discard lwpchan mappings, like munmap() */
if ((flags & MAP_FIXED) && curproc->p_lcp != NULL)
lwpchan_delete_mapping(curproc, *addrp, *addrp + len);
/*
* Ok, now let the vnode map routine do its thing to set things up.
*/
error = VOP_MAP(vp, pos, as,
addrp, len, uprot, maxprot, flags, fp->f_cred, NULL);
if (error == 0) {
/*
* Tell machine specific code that lwp has mapped shared memory
*/
if (flags & MAP_SHARED) {
/* EMPTY */
LWP_MMODEL_SHARED_AS(*addrp, len);
}
if (vp->v_type == VREG &&
(flags & (MAP_TEXT | MAP_INITDATA)) != 0) {
/*
* Mark this as an executable vnode
*/
mutex_enter(&vp->v_lock);
vp->v_flag |= VVMEXEC;
mutex_exit(&vp->v_lock);
}
}
done:
if (in_crit)
nbl_end_crit(vp);
return (error);
}
