/**
* ntfs_pagein - read a range of pages into memory
* @ni: ntfs inode whose data to read into the page range
* @attr_ofs: byte offset in the inode at which to start
* @size: number of bytes to read from the inode
* @upl: page list describing destination page range
* @upl_ofs: byte offset into page list at which to start
* @flags: flags further describing the pagein request
*
* Read @size bytes from the ntfs inode @ni, starting at byte offset @attr_ofs
* into the inode, into the range of pages specified by the page list @upl,
* starting at byte offset @upl_ofs into the page list.
*
* The @flags further describe the pagein request. The following pagein flags
* are currently defined in OSX kernel:
* UPL_IOSYNC - Perform synchronous i/o.
* UPL_NOCOMMIT - Do not commit/abort the page range.
* UPL_NORDAHEAD - Do not perform any speculative read-ahead.
* IO_PASSIVE - This is background i/o so do not throttle other i/o.
*
* Inside the ntfs driver we have the need to perform pageins whilst the inode
* is locked for writing (@ni->lock) thus we cheat and set UPL_NESTED_PAGEOUT
* in @flags when this is the case. We make sure to clear it in @flags before
* calling into the cluster layer so we do not accidentally cause confusion.
*
* For encrypted attributes we abort for now as we do not support them yet.
*
* For non-resident, non-compressed attributes we use cluster_pagein_ext()
* which deals with both normal and multi sector transfer protected attributes.
*
* For resident attributes and non-resident, compressed attributes we read the
* data ourselves by mapping the page list, and in the resident case, mapping
* the mft record, looking up the attribute in it, and copying the requested
* data from the mapped attribute into the page list, then unmapping the mft
* record, whilst for non-resident, compressed attributes, we get the raw inode
* and use it with ntfs_read_compressed() to read and decompress the data into
* our mapped page list. We then unmap the page list and finally, if
* UPL_NOCOMMIT is not specified, we commit (success) or abort (error) the page
* range.
*
* Return 0 on success and errno on error.
*
* Note the pages in the page list are marked busy on entry and the busy bit is
* cleared when we commit the page range. Thus it is perfectly safe for us to
* fill the pages with encrypted or mst protected data and to decrypt or mst
* deprotect in place before committing the page range.
*
* Adapted from cluster_pagein_ext().
*
* Locking: - Caller must hold an iocount reference on the vnode of @ni.
* - Caller must not hold @ni->lock or if it is held it must be for
* reading unless UPL_NESTED_PAGEOUT is set in @flags in which case
* the caller must hold @ni->lock for reading or writing.
*/
int ntfs_pagein(ntfs_inode *ni, s64 attr_ofs, unsigned size, upl_t upl,
upl_offset_t upl_ofs, int flags)
{
s64 attr_size;
u8 *kaddr;
kern_return_t kerr;
unsigned to_read;
int err;
BOOL locked = FALSE;
ntfs_debug("Entering for mft_no 0x%llx, offset 0x%llx, size 0x%x, "
"pagein flags 0x%x, page list offset 0x%llx.",
(unsigned long long)ni->mft_no,
(unsigned long long)attr_ofs, size, flags,
(unsigned long long)upl_ofs);
/*
* If the caller did not specify any i/o, then we are done. We cannot
* issue an abort because we do not have a upl or we do not know its
* size.
*/
if (!upl) {
ntfs_error(ni->vol->mp, "NULL page list passed in (error "
"EINVAL).");
return EINVAL;
}
if (S_ISDIR(ni->mode)) {
ntfs_error(ni->vol->mp, "Called for directory vnode.");
err = EISDIR;
goto err;
}
/*
* Protect against changes in initialized_size and thus against
* truncation also unless UPL_NESTED_PAGEOUT is set in which case the
* caller has already taken @ni->lock for exclusive access. We simply
* leave @locked to be FALSE in this case so we do not try to drop the
* lock later on.
*
* If UPL_NESTED_PAGEOUT is set we clear it in @flags to ensure we do
* not cause confusion in the cluster layer or the VM.
*/
if (flags & UPL_NESTED_PAGEOUT)
flags &= ~UPL_NESTED_PAGEOUT;
else {
locked = TRUE;
lck_rw_lock_shared(&ni->lock);
}
/* Do not allow messing with the inode once it has been deleted. */
if (NInoDeleted(ni)) {
/* Remove the inode from the name cache. */
cache_purge(ni->vn);
err = ENOENT;
goto err;
}
retry_pagein:
/*
* We guarantee that the size in the ubc will be smaller or equal to
* the size in the ntfs inode thus no need to check @ni->data_size.
*/
attr_size = ubc_getsize(ni->vn);
/*
* Only $DATA attributes can be encrypted/compressed. Index root can
* have the flags set but this means to create compressed/encrypted
* files, not that the attribute is compressed/encrypted. Note we need
* to check for AT_INDEX_ALLOCATION since this is the type of directory
* index inodes.
*/
if (ni->type != AT_INDEX_ALLOCATION) {
/* TODO: Deny access to encrypted attributes, just like NT4. */
if (NInoEncrypted(ni)) {
if (ni->type != AT_DATA)
panic("%s(): Encrypted non-data attribute.\n",
__FUNCTION__);
ntfs_warning(ni->vol->mp, "Denying access to "
"encrypted attribute (EACCES).");
err = EACCES;
goto err;
}
/* Compressed data streams need special handling. */
if (NInoNonResident(ni) && NInoCompressed(ni) && !NInoRaw(ni)) {
if (ni->type != AT_DATA)
panic("%s(): Compressed non-data attribute.\n",
__FUNCTION__);
goto compressed;
}
}
/* NInoNonResident() == NInoIndexAllocPresent() */
if (NInoNonResident(ni)) {
int (*callback)(buf_t, void *);
callback = NULL;
if (NInoMstProtected(ni) || NInoEncrypted(ni))
callback = ntfs_cluster_iodone;
/* Non-resident, possibly mst protected, attribute. */
err = cluster_pagein_ext(ni->vn, upl, upl_ofs, attr_ofs, size,
attr_size, flags, callback, NULL);
if (!err)
ntfs_debug("Done (cluster_pagein_ext()).");
else
ntfs_error(ni->vol->mp, "Failed (cluster_pagein_ext(), "
"error %d).", err);
if (locked)
lck_rw_unlock_shared(&ni->lock);
return err;
}
compressed:
/*
* The attribute is resident and/or compressed.
*
* Cannot pagein from a negative offset or if we are starting beyond
* the end of the attribute or if the attribute offset is not page
* aligned or the size requested is not a multiple of PAGE_SIZE.
*/
if (attr_ofs < 0 || attr_ofs >= attr_size || attr_ofs & PAGE_MASK_64 ||
size & PAGE_MASK || upl_ofs & PAGE_MASK) {
err = EINVAL;
goto err;
}
to_read = size;
attr_size -= attr_ofs;
if (to_read > attr_size)
to_read = attr_size;
/*
* We do not need @attr_size any more so reuse it to hold the number of
* bytes available in the attribute starting at offset @attr_ofs up to
* a maximum of the requested number of bytes rounded up to a multiple
* of the system page size.
*/
attr_size = (to_read + PAGE_MASK) & ~PAGE_MASK;
/* Abort any pages outside the end of the attribute. */
if (size > attr_size && !(flags & UPL_NOCOMMIT)) {
ubc_upl_abort_range(upl, upl_ofs + attr_size, size - attr_size,
UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
/* Update @size. */
size = attr_size;
}
/* To access the page list contents, we need to map the page list. */
kerr = ubc_upl_map(upl, (vm_offset_t*)&kaddr);
if (kerr != KERN_SUCCESS) {
ntfs_error(ni->vol->mp, "ubc_upl_map() failed (error %d).",
(int)kerr);
err = EIO;
goto err;
}
if (!NInoNonResident(ni)) {
/*
* Read the data from the resident attribute into the page
* list.
*/
err = ntfs_resident_attr_read(ni, attr_ofs, size,
kaddr + upl_ofs);
if (err && err != EAGAIN)
ntfs_error(ni->vol->mp, "ntfs_resident_attr_read() "
"failed (error %d).", err);
} else {
ntfs_inode *raw_ni;
int ioflags;
/*
* Get the raw inode. We take the inode lock shared to protect
* against concurrent writers as the compressed data is invalid
* whilst a write is in progress.
*/
err = ntfs_raw_inode_get(ni, LCK_RW_TYPE_SHARED, &raw_ni);
if (err)
ntfs_error(ni->vol->mp, "Failed to get raw inode "
"(error %d).", err);
else {
if (!NInoRaw(raw_ni))
panic("%s(): Requested raw inode but got "
"non-raw one.\n", __FUNCTION__);
ioflags = 0;
if (vnode_isnocache(ni->vn) ||
vnode_isnocache(raw_ni->vn))
ioflags |= IO_NOCACHE;
if (vnode_isnoreadahead(ni->vn) ||
vnode_isnoreadahead(raw_ni->vn))
ioflags |= IO_RAOFF;
err = ntfs_read_compressed(ni, raw_ni, attr_ofs, size,
kaddr + upl_ofs, NULL, ioflags);
if (err)
ntfs_error(ni->vol->mp,
"ntfs_read_compressed() "
"failed (error %d).", err);
lck_rw_unlock_shared(&raw_ni->lock);
(void)vnode_put(raw_ni->vn);
}
}
kerr = ubc_upl_unmap(upl);
if (kerr != KERN_SUCCESS) {
ntfs_error(ni->vol->mp, "ubc_upl_unmap() failed (error %d).",
(int)kerr);
if (!err)
err = EIO;
}
if (!err) {
if (!(flags & UPL_NOCOMMIT)) {
/* Commit the page range we brought up to date. */
ubc_upl_commit_range(upl, upl_ofs, size,
UPL_COMMIT_FREE_ON_EMPTY);
}
ntfs_debug("Done (%s).", !NInoNonResident(ni) ?
"ntfs_resident_attr_read()" :
"ntfs_read_compressed()");
} else /* if (err) */ {
/*
* If the attribute was converted to non-resident under our
* nose, retry the pagein.
*
* TODO: This may no longer be possible to happen now that we
* lock against changes in initialized size and thus
* truncation... Revisit this issue when the write code has
* been written and remove the check + goto if appropriate.
*/
if (err == EAGAIN)
goto retry_pagein;
err:
if (!(flags & UPL_NOCOMMIT)) {
int upl_flags = UPL_ABORT_FREE_ON_EMPTY;
if (err != ENOMEM)
upl_flags |= UPL_ABORT_ERROR;
ubc_upl_abort_range(upl, upl_ofs, size, upl_flags);
}
ntfs_error(ni->vol->mp, "Failed (error %d).", err);
}
if (locked)
lck_rw_unlock_shared(&ni->lock);
return err;
}
/**
* ntfs_page_map_ext - map a page of a vnode into memory
* @ni: ntfs inode of which to map a page
* @ofs: byte offset into @ni of which to map a page
* @upl: destination page list for the page
* @pl: destination array of pages containing the page itself
* @kaddr: destination pointer for the address of the mapped page contents
* @uptodate: if true return an uptodate page and if false return it as is
* @rw: if true we intend to modify the page and if false we do not
*
* Map the page corresponding to byte offset @ofs into the ntfs inode @ni into
* memory and return the page list in @upl, the array of pages containing the
* page in @pl and the address of the mapped page contents in @kaddr.
*
* If @uptodate is true the page is returned uptodate, i.e. if the page is
* currently not valid, it will be brought uptodate via a call to ntfs_pagein()
* before it is returned. And if @uptodate is false, the page is just returned
* ignoring its state. This means the page may or may not be uptodate.
*
* The caller must set @rw to true if the page is going to be modified and to
* false otherwise.
*
* Note: @ofs must be page aligned.
*
* Locking: - Caller must hold an iocount reference on the vnode of @ni.
* - Caller must hold @ni->lock for reading or writing.
*
* Return 0 on success and errno on error in which case *@upl is set to NULL.
*/
errno_t ntfs_page_map_ext(ntfs_inode *ni, s64 ofs, upl_t *upl,
upl_page_info_array_t *pl, u8 **kaddr, const BOOL uptodate,
const BOOL rw)
{
s64 size;
kern_return_t kerr;
int abort_flags;
errno_t err;
ntfs_debug("Entering for inode 0x%llx, offset 0x%llx, rw is %s.",
(unsigned long long)ni->mft_no,
(unsigned long long)ofs,
rw ? "true" : "false");
if (ofs & PAGE_MASK)
panic("%s() called with non page aligned offset (0x%llx).",
__FUNCTION__, (unsigned long long)ofs);
lck_spin_lock(&ni->size_lock);
size = ubc_getsize(ni->vn);
if (size > ni->data_size)
size = ni->data_size;
lck_spin_unlock(&ni->size_lock);
if (ofs > size) {
ntfs_error(ni->vol->mp, "Offset 0x%llx is outside the end of "
"the attribute (0x%llx).",
(unsigned long long)ofs,
(unsigned long long)size);
err = EINVAL;
goto err;
}
/* Create a page list for the wanted page. */
kerr = ubc_create_upl(ni->vn, ofs, PAGE_SIZE, upl, pl, UPL_SET_LITE |
(rw ? UPL_WILL_MODIFY : 0));
if (kerr != KERN_SUCCESS)
panic("%s(): Failed to get page (error %d).\n", __FUNCTION__,
(int)kerr);
/*
* If the page is not valid, need to read it in from the vnode now thus
* making it valid.
*
* We set UPL_NESTED_PAGEOUT to let ntfs_pagein() know that we already
* have the inode locked (@ni->lock is held by the caller).
*/
if (uptodate && !upl_valid_page(*pl, 0)) {
ntfs_debug("Reading page as it was not valid.");
err = ntfs_pagein(ni, ofs, PAGE_SIZE, *upl, 0, UPL_IOSYNC |
UPL_NOCOMMIT | UPL_NESTED_PAGEOUT);
if (err) {
ntfs_error(ni->vol->mp, "Failed to read page (error "
"%d).", err);
goto pagein_err;
}
}
/* Map the page into the kernel's address space. */
kerr = ubc_upl_map(*upl, (vm_offset_t*)kaddr);
if (kerr == KERN_SUCCESS) {
ntfs_debug("Done.");
return 0;
}
ntfs_error(ni->vol->mp, "Failed to map page (error %d).",
(int)kerr);
err = EIO;
pagein_err:
abort_flags = UPL_ABORT_FREE_ON_EMPTY;
if (!upl_valid_page(*pl, 0) ||
(vnode_isnocache(ni->vn) && !upl_dirty_page(*pl, 0)))
abort_flags |= UPL_ABORT_DUMP_PAGES;
ubc_upl_abort_range(*upl, 0, PAGE_SIZE, abort_flags);
err:
*upl = NULL;
return err;
}
/* relies on v1 paging */
static int
nullfs_pagein(struct vnop_pagein_args * ap)
{
int error = EIO;
struct vnode *vp, *lvp;
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
if (vnode_vtype(vp) != VREG) {
return ENOTSUP;
}
/*
* Ask VM/UBC/VFS to do our bidding
*/
if (vnode_getwithvid(lvp, NULLVPTOLOWERVID(vp)) == 0) {
vm_offset_t ioaddr;
uio_t auio;
kern_return_t kret;
off_t bytes_to_commit;
off_t lowersize;
upl_t upl = ap->a_pl;
user_ssize_t bytes_remaining = 0;
auio = uio_create(1, ap->a_f_offset, UIO_SYSSPACE, UIO_READ);
if (auio == NULL) {
error = EIO;
goto exit_no_unmap;
}
kret = ubc_upl_map(upl, &ioaddr);
if (KERN_SUCCESS != kret) {
panic("nullfs_pagein: ubc_upl_map() failed with (%d)", kret);
}
ioaddr += ap->a_pl_offset;
error = uio_addiov(auio, (user_addr_t)ioaddr, ap->a_size);
if (error) {
goto exit;
}
lowersize = ubc_getsize(lvp);
if (lowersize != ubc_getsize(vp)) {
(void)ubc_setsize(vp, lowersize); /* ignore failures, nothing can be done */
}
error = VNOP_READ(lvp, auio, ((ap->a_flags & UPL_IOSYNC) ? IO_SYNC : 0), ap->a_context);
bytes_remaining = uio_resid(auio);
if (bytes_remaining > 0 && bytes_remaining <= (user_ssize_t)ap->a_size)
{
/* zero bytes that weren't read in to the upl */
bzero((void*)((uintptr_t)(ioaddr + ap->a_size - bytes_remaining)), (size_t) bytes_remaining);
}
exit:
kret = ubc_upl_unmap(upl);
if (KERN_SUCCESS != kret) {
panic("nullfs_pagein: ubc_upl_unmap() failed with (%d)", kret);
}
if (auio != NULL) {
uio_free(auio);
}
exit_no_unmap:
if ((ap->a_flags & UPL_NOCOMMIT) == 0) {
if (!error && (bytes_remaining >= 0) && (bytes_remaining <= (user_ssize_t)ap->a_size)) {
/* only commit what was read in (page aligned)*/
bytes_to_commit = ap->a_size - bytes_remaining;
if (bytes_to_commit)
{
/* need to make sure bytes_to_commit and byte_remaining are page aligned before calling ubc_upl_commit_range*/
if (bytes_to_commit & PAGE_MASK)
{
bytes_to_commit = (bytes_to_commit & (~PAGE_MASK)) + (PAGE_MASK + 1);
assert(bytes_to_commit <= (off_t)ap->a_size);
bytes_remaining = ap->a_size - bytes_to_commit;
}
ubc_upl_commit_range(upl, ap->a_pl_offset, (upl_size_t)bytes_to_commit, UPL_COMMIT_FREE_ON_EMPTY);
}
/* abort anything thats left */
if (bytes_remaining) {
ubc_upl_abort_range(upl, ap->a_pl_offset + bytes_to_commit, (upl_size_t)bytes_remaining, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
}
} else {
ubc_upl_abort_range(upl, ap->a_pl_offset, (upl_size_t)ap->a_size, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
}
}
vnode_put(lvp);
} else if((ap->a_flags & UPL_NOCOMMIT) == 0) {
ubc_upl_abort_range(ap->a_pl, ap->a_pl_offset, (upl_size_t)ap->a_size, UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
}
return error;
}
