/*
* Gets the first page of kernel binary at disk into a buffer
* Uses KPI VFS functions and a ripped uio_createwithbuffer() from XNU
*/
static kern_return_t get_k_mh(void *buffer, vnode_t k_vnode, struct kernel_info *kinfo)
{
uio_t uio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
if(uio == NULL) {
return KERN_FAILURE;
}
/* imitate the kernel and read a single page from the header */
if(uio_addiov(uio, CAST_USER_ADDR_T(buffer), HEADER_SIZE) != 0) {
return KERN_FAILURE;
}
/* read kernel vnode into the buffer */
if(VNOP_READ(k_vnode, uio, 0, NULL) != 0) {
return KERN_FAILURE;
} else if(uio_resid(uio)) {
return KERN_FAILURE;
}
/* process the header */
uint32_t magic = *(uint32_t *)buffer;
if(magic == FAT_CIGAM) {
struct fat_header *fh = (struct fat_header *)buffer;
struct fat_arch *fa = (struct fat_arch *)(buffer + sizeof(struct fat_header));
uint32_t file_off = 0;
for(uint32_t i = 0; i < ntohl(fh->nfat_arch); i++) {
if(sizeof(void *) == 8 && ntohl(fa->cputype) == CPU_TYPE_X86_64) {
file_off = ntohl(fa->offset);
break;
} else if(sizeof(void *) == 4 && ntohl(fa->cputype) == CPU_TYPE_X86) {
file_off = ntohl(fa->offset);
break;
}
fa++;
}
/* read again */
uio = uio_create(1, file_off, UIO_SYSSPACE, UIO_READ);
uio_addiov(uio, CAST_USER_ADDR_T(buffer), HEADER_SIZE);
VNOP_READ(k_vnode, uio, 0, NULL);
kinfo->fat_offset = file_off;
} else {
kinfo->fat_offset = 0;
}
return KERN_SUCCESS;
}
int readFile(char *file, uint8_t *buffer, off_t offset, user_size_t size) {
int res = EIO;
vfs_context_t vfsContext = vfs_context_create(NULL);
if (vfsContext == NULL) {
return EIO;
}
vnode_t fileVnode = NULLVP;
if (vnode_lookup(file, 0, &fileVnode, vfsContext) == 0) {
uio_t uio = uio_create(1, offset, UIO_SYSSPACE, UIO_READ);
if (uio == NULL)
goto exit;
if (uio_addiov(uio, CAST_USER_ADDR_T(buffer), size))
goto exit;
if (VNOP_READ(fileVnode, uio, 0, vfsContext))
goto exit;
if (uio_resid(uio))
goto exit;
res = 0;
} else {
vfs_context_rele(vfsContext);
return ENOENT;
}
exit:
vnode_put(fileVnode);
vfs_context_rele(vfsContext);
return res;
}
static int
vnread_shadow(struct vn_softc * vn, struct uio *uio, int ioflag,
vfs_context_t ctx)
{
u_int32_t blocksize = vn->sc_secsize;
int error = 0;
off_t offset;
user_ssize_t resid;
off_t orig_offset;
user_ssize_t orig_resid;
orig_resid = resid = uio_resid(uio);
orig_offset = offset = uio_offset(uio);
while (resid > 0) {
u_int32_t remainder;
u_int32_t this_block_number;
u_int32_t this_block_count;
off_t this_offset;
user_ssize_t this_resid;
struct vnode * vp;
/* figure out which blocks to read */
remainder = block_remainder(offset, blocksize);
if (shadow_map_read(vn->sc_shadow_map,
block_truncate(offset, blocksize),
block_round(resid + remainder, blocksize),
&this_block_number, &this_block_count)) {
vp = vn->sc_shadow_vp;
}
else {
vp = vn->sc_vp;
}
/* read the blocks (or parts thereof) */
this_offset = (off_t)this_block_number * blocksize + remainder;
uio_setoffset(uio, this_offset);
this_resid = this_block_count * blocksize - remainder;
if (this_resid > resid) {
this_resid = resid;
}
uio_setresid(uio, this_resid);
error = VNOP_READ(vp, uio, ioflag, ctx);
if (error) {
break;
}
/* figure out how much we actually read */
this_resid -= uio_resid(uio);
if (this_resid == 0) {
printf("vn device: vnread_shadow zero length read\n");
break;
}
resid -= this_resid;
offset += this_resid;
}
uio_setresid(uio, resid);
uio_setoffset(uio, offset);
return (error);
}
/*
* retrieve the first page of kernel binary at disk into a buffer
* version that uses KPI VFS functions and a ripped uio_createwithbuffer() from XNU
*/
kern_return_t get_mach_header(void *buffer, vnode_t vnode, vfs_context_t ctxt) {
int error = 0;
uio_t uio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
if (uio == NULL) {
return KERN_FAILURE;
}
// imitate the kernel and read a single page from the header
error = uio_addiov(uio, CAST_USER_ADDR_T(buffer), HEADER_SIZE);
if (error) {
return error;
}
// read kernel vnode into the buffer
error = VNOP_READ(vnode, uio, 0, ctxt);
if (error) {
return error;
}
else if (uio_resid(uio)) return EINVAL;
// process the header
uint32_t magic = *(uint32_t*)buffer;
if (magic != MH_MAGIC_64) {
return KERN_FAILURE;
}
return KERN_SUCCESS;
}
int zfs_vn_rdwr(enum uio_rw rw, struct vnode *vp, caddr_t base, ssize_t len,
offset_t offset, enum uio_seg seg, int ioflag, rlim64_t ulimit,
cred_t *cr, ssize_t *residp)
{
uio_t *auio;
int spacetype;
int error=0;
vfs_context_t vctx;
spacetype = UIO_SEG_IS_USER_SPACE(seg) ? UIO_USERSPACE32 : UIO_SYSSPACE;
vctx = vfs_context_create((vfs_context_t)0);
auio = uio_create(1, 0, spacetype, rw);
uio_reset(auio, offset, spacetype, rw);
uio_addiov(auio, (uint64_t)(uintptr_t)base, len);
if (rw == UIO_READ) {
error = VNOP_READ(vp, auio, ioflag, vctx);
} else {
error = VNOP_WRITE(vp, auio, ioflag, vctx);
}
if (residp) {
*residp = uio_resid(auio);
} else {
if (uio_resid(auio) && error == 0)
error = EIO;
}
uio_free(auio);
vfs_context_rele(vctx);
return (error);
}
/* Generic read interface */
int
afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
afs_int32 asize)
{
afs_ucred_t *oldCred;
afs_size_t resid;
afs_int32 code;
#ifdef AFS_DARWIN80_ENV
uio_t uio;
#endif
AFS_STATCNT(osi_Read);
/**
* If the osi_file passed in is NULL, panic only if AFS is not shutting
* down. No point in crashing when we are already shutting down
*/
if (!afile) {
if (afs_shuttingdown == AFS_RUNNING)
osi_Panic("osi_Read called with null param");
else
return -EIO;
}
if (offset != -1)
afile->offset = offset;
AFS_GUNLOCK();
#ifdef AFS_DARWIN80_ENV
uio=uio_create(1, afile->offset, AFS_UIOSYS, UIO_READ);
uio_addiov(uio, CAST_USER_ADDR_T(aptr), asize);
code = VNOP_READ(afile->vnode, uio, IO_UNIT, afs_osi_ctxtp);
resid = AFS_UIO_RESID(uio);
uio_free(uio);
#else
code =
gop_rdwr(UIO_READ, afile->vnode, (caddr_t) aptr, asize, afile->offset,
AFS_UIOSYS, IO_UNIT, &afs_osi_cred, &resid);
#endif
AFS_GLOCK();
if (code == 0) {
code = asize - resid;
afile->offset += code;
osi_DisableAtimes(afile->vnode);
} else {
afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32, resid,
ICL_TYPE_INT32, code);
if (code > 0) {
code *= -1;
}
}
return code;
}
int
cttyread(__unused dev_t dev, struct uio *uio, int flag)
{
vnode_t ttyvp = cttyvp(current_proc());
struct vfs_context context;
int error;
if (ttyvp == NULL)
return (EIO);
context.vc_thread = current_thread();
context.vc_ucred = NOCRED;
error = VNOP_READ(ttyvp, uio, flag, &context);
vnode_put(ttyvp);
return (error);
}
/*
* retrieve the whole linkedit segment into target buffer from kernel binary at disk
* we keep this buffer until we don't need to solve symbols anymore
*/
static kern_return_t get_kernel_linkedit(vnode_t kernel_vnode, vfs_context_t ctxt, kernel_info *kinfo) {
int error = 0;
uio_t uio = uio_create(1, kinfo->linkedit_fileoff, UIO_SYSSPACE, UIO_READ);
if (uio == NULL) {
return KERN_FAILURE;
}
error = uio_addiov(uio, CAST_USER_ADDR_T(kinfo->linkedit_buf), kinfo->linkedit_size);
if (error) return error;
error = VNOP_READ(kernel_vnode, uio, 0, ctxt);
if (error) {
return error;
} else if (uio_resid(uio)) {
return EINVAL;
}
return KERN_SUCCESS;
}
static int
nullfs_read(struct vnop_read_args * ap)
{
int error = EIO;
struct vnode *vp, *lvp;
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
if (nullfs_checkspecialvp(ap->a_vp)) {
return ENOTSUP; /* the special vnodes can't be read */
}
vp = ap->a_vp;
lvp = NULLVPTOLOWERVP(vp);
/*
* First some house keeping
*/
if (vnode_getwithvid(lvp, NULLVPTOLOWERVID(vp)) == 0) {
if (!vnode_isreg(lvp) && !vnode_islnk(lvp)) {
error = EPERM;
goto end;
}
if (uio_resid(ap->a_uio) == 0) {
error = 0;
goto end;
}
/*
* Now ask VM/UBC/VFS to do our bidding
*/
error = VNOP_READ(lvp, ap->a_uio, ap->a_ioflag, ap->a_context);
if (error) {
NULLFSDEBUG("VNOP_READ failed: %d\n", error);
}
end:
vnode_put(lvp);
}
return error;
}
/*
* Get the whole __LINKEDIT segment into target buffer from kernel binary at disk
* This buffer is keeped until we don't need to solve symbols anymore
*/
static kern_return_t get_k_linkedit(vnode_t k_vnode, struct kernel_info *kinfo)
{
uio_t uio = uio_create(1, kinfo->linkedit_fileoff, UIO_SYSSPACE, UIO_READ);
if(uio == NULL) {
return KERN_FAILURE;
}
if(uio_addiov(uio, CAST_USER_ADDR_T(kinfo->linkedit_buf), kinfo->linkedit_size) != 0) {
return KERN_FAILURE;
}
if(VNOP_READ(k_vnode, uio, 0, NULL) != 0) {
return KERN_FAILURE;
} else if(uio_resid(uio)) {
return KERN_FAILURE;
}
return KERN_SUCCESS;
}
static int
file_io(struct vnode * vp, vfs_context_t ctx,
enum uio_rw op, char * base, off_t offset, user_ssize_t count,
user_ssize_t * resid)
{
uio_t auio;
int error;
char uio_buf[UIO_SIZEOF(1)];
auio = uio_createwithbuffer(1, offset, UIO_SYSSPACE, op,
&uio_buf[0], sizeof(uio_buf));
uio_addiov(auio, CAST_USER_ADDR_T(base), count);
if (op == UIO_READ)
error = VNOP_READ(vp, auio, IO_SYNC, ctx);
else
error = VNOP_WRITE(vp, auio, IO_SYNC, ctx);
if (resid != NULL) {
*resid = uio_resid(auio);
}
return (error);
}
/* kauth file scope listener
* this allows to detect files written to the filesystem
* arg2 contains a flag KAUTH_FILEOP_CLOSE which is set if a modified file is being closed
* this way we don't need to trace every close(), only the ones writing to the filesystem
*/
static int
fileop_scope_listener(kauth_cred_t credential,
void * idata,
kauth_action_t action,
uintptr_t arg0, /* vnode reference */
uintptr_t arg1, /* full path to file being closed */
uintptr_t arg2, /* flags */
uintptr_t arg3)
{
/* ignore all actions except FILE_CLOSE */
if (action != KAUTH_FILEOP_CLOSE)
{
return KAUTH_RESULT_DEFER;
}
/* ignore operations with bad data */
if (credential == NULL || (vnode_t)arg0 == NULL || (char*)arg1 == NULL)
{
ERROR_MSG("Arguments contain null pointers!");
return KAUTH_RESULT_DEFER;
}
/* ignore closes on folders, character and block devices */
switch ( vnode_vtype((vnode_t)arg0) )
{
case VDIR:
case VCHR:
case VBLK:
return KAUTH_RESULT_DEFER;
default:
break;
}
/* we are only interested when a modified file is being closed */
if ((int)arg2 != KAUTH_FILEOP_CLOSE_MODIFIED)
{
return KAUTH_RESULT_DEFER;
}
char *file_path = (char*)arg1;
/* get information from current proc trying to write to the vnode */
proc_t proc = current_proc();
pid_t mypid = proc_pid(proc);
char myprocname[MAXCOMLEN+1] = {0};
proc_name(mypid, myprocname, sizeof(myprocname));
/* retrieve the vnode attributes, we can get a lot of vnode information from here */
struct vnode_attr vap = {0};
vfs_context_t context = vfs_context_create(NULL);
/* initialize the structure fields we are interested in
* reference vn_stat_noauth() xnu/bsd/vfs/vfs_vnops.c
*/
VATTR_INIT(&vap);
VATTR_WANTED(&vap, va_mode);
VATTR_WANTED(&vap, va_type);
VATTR_WANTED(&vap, va_uid);
VATTR_WANTED(&vap, va_gid);
VATTR_WANTED(&vap, va_data_size);
VATTR_WANTED(&vap, va_flags);
int attr_ok = 1;
if ( vnode_getattr((vnode_t)arg0, &vap, context) != 0 )
{
/* in case of error permissions and filesize will be bogus */
ERROR_MSG("failed to vnode_getattr");
attr_ok = 0;
}
/* release the context we created, else kab00m! */
vfs_context_rele(context);
int error = 0;
/* make sure we :
* - were able to read the attributes
* - file size is at least uint32_t
* - path starts with /Users
*/
if ( attr_ok == 1 &&
vap.va_data_size >= sizeof(uint32_t) &&
strprefix(file_path, "/Users/") )
{
uint32_t magic = 0;
/* read target vnode */
uio_t uio = NULL;
/* read from offset 0 */
uio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);
if (uio == NULL)
{
ERROR_MSG("uio_create returned null!");
return KAUTH_RESULT_DEFER;
}
/* we just want to read 4 bytes to match the header */
if ( (error = uio_addiov(uio, CAST_USER_ADDR_T(&magic), sizeof(uint32_t))) )
{
ERROR_MSG("uio_addiov returned error %d!", error);
return KAUTH_RESULT_DEFER;
}
if ( (error = VNOP_READ((vnode_t)arg0, uio, 0, NULL)) )
{
ERROR_MSG("VNOP_READ failed %d!", error);
return KAUTH_RESULT_DEFER;
}
else if (uio_resid(uio))
{
ERROR_MSG("uio_resid!");
return KAUTH_RESULT_DEFER;
}
/* verify if it's a Mach-O file */
if (magic == MH_MAGIC || magic == MH_MAGIC_64 || magic == FAT_CIGAM)
{
char *token = NULL;
char *string = NULL;
char *tofree = NULL;
int library = 0;
int preferences = 0;
tofree = string = STRDUP(file_path, M_TEMP);
while ((token = strsep(&string, "/")) != NULL)
{
if (strcmp(token, "Library") == 0)
{
library = 1;
}
else if (library == 1 && strcmp(token, "Preferences") == 0)
{
preferences = 1;
}
}
_FREE(tofree, M_TEMP);
/* we got a match into /Users/username/Library/Preferences, warn user about it */
if (library == 1 && preferences == 1)
{
DEBUG_MSG("Found Mach-O written to %s by %s.", file_path, myprocname);
char alert_msg[1025] = {0};
snprintf(alert_msg, sizeof(alert_msg), "Process \"%s\" wrote Mach-O binary %s.\n This could be Hacking Team's malware!", myprocname, file_path);
alert_msg[sizeof(alert_msg)-1] = '\0';
/* log to syslog */
printf("[WARNING] Process \"%s\" wrote Mach-O binary %s.\n This could be Hacking Team's malware!", myprocname, file_path);
/* deprecated but still usable to display the alert */
KUNCUserNotificationDisplayNotice(10, // Timeout
0, // Flags - default is Stop alert level
NULL, // iconpath
NULL, // soundpath
NULL, // localization path
"Security Alert", // alert header
alert_msg, // alert message
"OK"); // button title
}
}
}
/* don't deny access, we are just here to observe */
return KAUTH_RESULT_DEFER;
}
/* 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;
}
static int
vnread(dev_t dev, struct uio *uio, int ioflag)
{
struct vfs_context context;
int error = 0;
off_t offset;
proc_t p;
user_ssize_t resid;
struct vn_softc * vn;
int unit;
unit = vnunit(dev);
if (vnunit(dev) >= NVNDEVICE) {
return (ENXIO);
}
p = current_proc();
vn = vn_table + unit;
if ((vn->sc_flags & VNF_INITED) == 0) {
error = ENXIO;
goto done;
}
context.vc_thread = current_thread();
context.vc_ucred = vn->sc_cred;
error = vnode_getwithvid(vn->sc_vp, vn->sc_vid);
if (error != 0) {
/* the vnode is no longer available, abort */
error = ENXIO;
vnclear(vn, &context);
goto done;
}
resid = uio_resid(uio);
offset = uio_offset(uio);
/*
* If out of bounds return an error. If at the EOF point,
* simply read less.
*/
if (offset >= (off_t)vn->sc_fsize) {
if (offset > (off_t)vn->sc_fsize) {
error = EINVAL;
}
goto done;
}
/*
* If the request crosses EOF, truncate the request.
*/
if ((offset + resid) > (off_t)vn->sc_fsize) {
resid = vn->sc_fsize - offset;
uio_setresid(uio, resid);
}
if (vn->sc_shadow_vp != NULL) {
error = vnode_getwithvid(vn->sc_shadow_vp,
vn->sc_shadow_vid);
if (error != 0) {
/* the vnode is no longer available, abort */
error = ENXIO;
vnode_put(vn->sc_vp);
vnclear(vn, &context);
goto done;
}
error = vnread_shadow(vn, uio, ioflag, &context);
vnode_put(vn->sc_shadow_vp);
} else {
error = VNOP_READ(vn->sc_vp, uio, ioflag, &context);
}
vnode_put(vn->sc_vp);
done:
return (error);
}
int
afs_MemRead(struct vcache *avc, struct uio *auio,
afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, tlen;
afs_size_t len = 0;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error, trybusy = 1;
afs_int32 code;
struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop = NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
memset(&tuio, 0, sizeof(tuio));
#endif
AFS_STATCNT(afs_MemRead);
if (avc->vc_error)
return EIO;
/* check that we have the latest status info in the vnode cache */
if ((code = afs_CreateReq(&treq, acred)))
return code;
if (!noLock) {
code = afs_VerifyVCache(avc, treq);
if (code) {
code = afs_CheckCode(code, treq, 8); /* failed to get it */
afs_DestroyReq(treq);
return code;
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
code = afs_CheckCode(EACCES, treq, 9);
afs_DestroyReq(treq);
return code;
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
memset(tvec, 0, sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
/*
* Locks held:
* avc->lock(R)
*/
/* This bit is bogus. We're checking to see if the read goes past the
* end of the file. If so, we should be zeroing out all of the buffers
* that the client has passed into us (there is a danger that we may leak
* kernel memory if we do not). However, this behaviour is disabled by
* not setting len before this segment runs, and by setting len to 0
* immediately we enter it. In addition, we also need to check for a read
* which partially goes off the end of the file in the while loop below.
*/
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->f.chunkBytes - offset;
}
} else {
int versionOk;
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
try_background:
#endif
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw.
*/
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now,
* and we'll need new data */
tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
if (trybusy && !afs_BBusy()) {
#endif
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 665);
if (!(tdc->mflags & DFFetchReq)) {
int dontwait = B_DONTWAIT;
/* start the daemon (may already be running, however) */
UpgradeSToWLock(&tdc->mflock, 666);
tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS)
dontwait = 0;
#endif
bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc, (void *)0, (void *)0);
if (!bp) {
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
/* don't use bp pointer! */
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
len = tdc->validPos - filePos;
versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* no longer fetching, verify data version
* (avoid new GetDCache call) */
if (versionOk && len > 0) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %lld m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
goto try_background;
}
#if 0
printf("afs_read: forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
#endif
}
#endif
/* If we get here, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (!tdc) {
error = EIO;
break;
}
/*
* Locks held:
* avc->lock(R)
* tdc->lock(R)
*/
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the mem cache */
/* mung uio structure to be right for this transfer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
code = afs_MemReadUIO(&tdc->f.inode, tuiop);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
} /* the whole while loop */
/*
* Locks held:
* avc->lock(R)
* tdc->lock(R) if tdc
*/
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
/*
* try to queue prefetch, if needed. If DataVersion is zero there
* should not be any more: files with DV 0 never have been stored
* on the fileserver, symbolic links and directories never require
* more than a single chunk.
*/
if (!noLock && !(hiszero(avc->f.m.DataVersion)) &&
#ifndef AFS_VM_RDWR_ENV
afs_preCache
#else
1
#endif
) {
afs_PrefetchChunk(avc, tdc, acred, treq);
}
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
error = afs_CheckCode(error, treq, 10);
afs_DestroyReq(treq);
return error;
}
/* called with the dcache entry triggering the fetch, the vcache entry involved,
* and a vrequest for the read call. Marks the dcache entry as having already
* triggered a prefetch, starts the prefetch going and sets the DFFetchReq
* flag in the prefetched block, so that the next call to read knows to wait
* for the daemon to start doing things.
*
* This function must be called with the vnode at least read-locked, and
* no locks on the dcache, because it plays around with dcache entries.
*/
void
afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
afs_ucred_t *acred, struct vrequest *areq)
{
struct dcache *tdc;
afs_size_t offset;
afs_size_t j1, j2; /* junk vbls for GetDCache to trash */
offset = adc->f.chunk + 1; /* next chunk we'll need */
offset = AFS_CHUNKTOBASE(offset); /* base of next chunk */
ObtainReadLock(&adc->lock);
ObtainSharedLock(&adc->mflock, 662);
if (offset < avc->f.m.Length && !(adc->mflags & DFNextStarted)
&& !afs_BBusy()) {
struct brequest *bp;
UpgradeSToWLock(&adc->mflock, 663);
adc->mflags |= DFNextStarted; /* we've tried to prefetch for this guy */
ReleaseWriteLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
tdc = afs_GetDCache(avc, offset, areq, &j1, &j2, 2); /* type 2 never returns 0 */
/*
* In disconnected mode, type 2 can return 0 because it doesn't
* make any sense to allocate a dcache we can never fill
*/
if (tdc == NULL)
return;
ObtainSharedLock(&tdc->mflock, 651);
if (!(tdc->mflags & DFFetchReq)) {
/* ask the daemon to do the work */
UpgradeSToWLock(&tdc->mflock, 652);
tdc->mflags |= DFFetchReq; /* guaranteed to be cleared by BKG or GetDCache */
/* last parm (1) tells bkg daemon to do an afs_PutDCache when it is done,
* since we don't want to wait for it to finish before doing so ourselves.
*/
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) offset, (afs_size_t) 1, tdc,
(void *)0, (void *)0);
if (!bp) {
/* Bkg table full; just abort non-important prefetching to avoid deadlocks */
tdc->mflags &= ~DFFetchReq;
ReleaseWriteLock(&tdc->mflock);
afs_PutDCache(tdc);
/*
* DCLOCKXXX: This is a little sketchy, since someone else
* could have already started a prefetch.. In practice,
* this probably doesn't matter; at most it would cause an
* extra slot in the BKG table to be used up when someone
* prefetches this for the second time.
*/
ObtainReadLock(&adc->lock);
ObtainWriteLock(&adc->mflock, 664);
adc->mflags &= ~DFNextStarted;
ReleaseWriteLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
} else {
ReleaseWriteLock(&tdc->mflock);
}
} else {
ReleaseSharedLock(&tdc->mflock);
afs_PutDCache(tdc);
}
} else {
ReleaseSharedLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
}
}
int
afs_UFSRead(struct vcache *avc, struct uio *auio,
afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, tlen;
afs_size_t len = 0;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error;
struct osi_file *tfile;
afs_int32 code;
int trybusy = 1;
struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop=NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
memset(&tuio, 0, sizeof(tuio));
#endif
AFS_STATCNT(afs_UFSRead);
if (avc && avc->vc_error)
return EIO;
AFS_DISCON_LOCK();
/* check that we have the latest status info in the vnode cache */
if ((code = afs_CreateReq(&treq, acred)))
return code;
if (!noLock) {
if (!avc)
osi_Panic("null avc in afs_UFSRead");
else {
code = afs_VerifyVCache(avc, treq);
if (code) {
code = afs_CheckCode(code, treq, 11); /* failed to get it */
afs_DestroyReq(treq);
AFS_DISCON_UNLOCK();
return code;
}
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
AFS_DISCON_UNLOCK();
code = afs_CheckCode(EACCES, treq, 12);
afs_DestroyReq(treq);
return code;
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
memset(tvec, 0, sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
/* This bit is bogus. We're checking to see if the read goes past the
* end of the file. If so, we should be zeroing out all of the buffers
* that the client has passed into us (there is a danger that we may leak
* kernel memory if we do not). However, this behaviour is disabled by
* not setting len before this segment runs, and by setting len to 0
* immediately we enter it. In addition, we also need to check for a read
* which partially goes off the end of the file in the while loop below.
*/
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
}
} else {
int versionOk;
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
try_background:
#endif
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
if (!tdc) {
error = ENETDOWN;
break;
}
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw. */
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now, and we'll need new data */
tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
if (trybusy && !afs_BBusy()) {
#endif
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 667);
if (!(tdc->mflags & DFFetchReq)) {
int dontwait = B_DONTWAIT;
UpgradeSToWLock(&tdc->mflock, 668);
tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS)
dontwait = 0;
#endif
bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc, (void *)0, (void *)0);
if (!bp) {
/* Bkg table full; retry deadlocks */
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
len = tdc->validPos - filePos;
versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* no longer fetching, verify data version (avoid new
* GetDCache call) */
if (versionOk && len > 0) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %d m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
goto try_background;
}
}
#endif
/* If we get here, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
if (!tdc) {
error = EIO;
break;
}
len = tdc->validPos - filePos;
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
tdc->dflags);
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the file */
tfile = (struct osi_file *)osi_UFSOpen(&tdc->f.inode);
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
/* mung uio structure to be right for this transfer */
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
#if defined(AFS_AIX41_ENV)
AFS_GUNLOCK();
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
NULL, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
/* Flush all JFS pages now for big performance gain in big file cases
* If we do something like this, must check to be sure that AFS file
* isn't mmapped... see afs_gn_map() for why.
*/
/*
if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
many different ways to do similar things:
so far, the best performing one is #2, but #1 might match it if we
straighten out the confusion regarding which pages to flush. It
really does matter.
1. vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
2. vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
3. vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
4. vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
tfile->vnode->v_gnode->gn_seg = NULL;
5. deletep
6. ipgrlse
7. ifreeseg
Unfortunately, this seems to cause frequent "cache corruption" episodes.
vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
}
*/
#elif defined(AFS_AIX_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
&tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
VOP_RWLOCK(tfile->vnode, 0, NULL);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, NULL);
VOP_RWUNLOCK(tfile->vnode, 0, NULL);
#else
VOP_RWLOCK(tfile->vnode, 0);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_RWUNLOCK(tfile->vnode, 0);
#endif
AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
AFS_GUNLOCK();
AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
code);
AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
AFS_GUNLOCK();
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
AFS_GUNLOCK();
code = osi_rdwr(tfile, &tuio, UIO_READ);
AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
AFS_GUNLOCK();
code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, current_proc());
AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_FBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curthread);
AFS_GLOCK();
#elif defined(AFS_NBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_XBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curproc);
AFS_GLOCK();
#else
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif
osi_UFSClose(tfile);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
}
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch, obviously.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
/*
* try to queue prefetch, if needed. If DataVersion is zero there
* should not be any more: files with DV 0 never have been stored
* on the fileserver, symbolic links and directories never require
* more than a single chunk.
*/
if (!noLock && !(hiszero(avc->f.m.DataVersion))) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, acred, treq);
}
#endif
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
AFS_DISCON_UNLOCK();
error = afs_CheckCode(error, treq, 13);
afs_DestroyReq(treq);
return error;
}
int
vn_rdwr_64FromKernelCode(
enum uio_rw rw,
struct vnode *vp,
uint64_t base,
int64_t len,
off_t offset,
enum uio_seg segflg,
int ioflg,
int *aresid,
proc_t p)
{
uio_t auio;
//int spacetype;
vfs_context_t context;
int error=0;
//char uio_buf[ UIO_SIZEOF(1) ];
//context.vc_thread = current_thread();
//context.vc_ucred = cred;
// if (UIO_SEG_IS_USER_SPACE(segflg)) {
// spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
// }
// else {
// spacetype = UIO_SYSSPACE;
// }
auio = uio_create(1, offset, UIO_SYSSPACE, rw);
uio_addiov(auio, base, len);
//#if CONFIG_MACF
// /* XXXMAC
// * IO_NOAUTH should be re-examined.
// * Likely that mediation should be performed in caller.
// */
// if ((ioflg & IO_NOAUTH) == 0) {
// /* passed cred is fp->f_cred */
// if (rw == UIO_READ)
// error = mac_vnode_check_read(&context, cred, vp);
// else
// error = mac_vnode_check_write(&context, cred, vp);
// }
//#endif
if (rw == UIO_READ)
{
error = VNOP_READ(vp, auio, ioflg, context);
}
else
{
error = VNOP_WRITE(vp, auio, ioflg, context);
}
if (aresid)
// LP64todo - fix this
*aresid = uio_resid(auio);
else if (uio_resid(auio) && error == 0)
error = EIO;
return (error);
}
int
afs_UFSRead(register struct vcache *avc, struct uio *auio,
struct AFS_UCRED *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, len, tlen;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop=NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
#endif
struct osi_file *tfile;
afs_int32 code;
int trybusy = 1;
struct vrequest treq;
AFS_STATCNT(afs_UFSRead);
if (avc && avc->vc_error)
return EIO;
AFS_DISCON_LOCK();
/* check that we have the latest status info in the vnode cache */
if ((code = afs_InitReq(&treq, acred)))
return code;
if (!noLock) {
if (!avc)
osi_Panic("null avc in afs_UFSRead");
else {
code = afs_VerifyVCache(avc, &treq);
if (code) {
code = afs_CheckCode(code, &treq, 11); /* failed to get it */
AFS_DISCON_UNLOCK();
return code;
}
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, &treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
AFS_DISCON_UNLOCK();
return afs_CheckCode(EACCES, &treq, 12);
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
}
} else {
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
#ifdef AFS_DISCON_ENV
if (!tdc) {
printf("Network down in afs_read");
error = ENETDOWN;
break;
}
#endif /* AFS_DISCON_ENV */
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw. */
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now, and we'll need new data */
tagain:
if (trybusy && !afs_BBusy()) {
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 667);
if (!(tdc->mflags & DFFetchReq)) {
UpgradeSToWLock(&tdc->mflock, 668);
tdc->mflags |= DFFetchReq;
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc);
if (!bp) {
/* Bkg table full; retry deadlocks */
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
} else {
/* no longer fetching, verify data version (avoid new
* GetDCache call) */
if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
&& ((len = tdc->validPos - filePos) > 0)) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
/* If we get, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
if (!tdc) {
error = EIO;
break;
}
len = tdc->validPos - filePos;
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
tdc->dflags);
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the file */
#ifdef IHINT
if (tfile = tdc->ihint) {
if (tdc->f.inode != tfile->inum) {
afs_warn("afs_UFSRead: %x hint mismatch tdc %d inum %d\n",
tdc, tdc->f.inode, tfile->inum);
osi_UFSClose(tfile);
tdc->ihint = tfile = 0;
nihints--;
}
}
if (tfile != 0) {
usedihint++;
} else
#endif /* IHINT */
#if defined(LINUX_USE_FH)
tfile = (struct osi_file *)osi_UFSOpen_fh(&tdc->f.fh, tdc->f.fh_type);
#else
tfile = (struct osi_file *)osi_UFSOpen(tdc->f.inode);
#endif
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
/* mung uio structure to be right for this transfer */
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
#if defined(AFS_AIX41_ENV)
AFS_GUNLOCK();
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
NULL, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
/* Flush all JFS pages now for big performance gain in big file cases
* If we do something like this, must check to be sure that AFS file
* isn't mmapped... see afs_gn_map() for why.
*/
/*
if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
many different ways to do similar things:
so far, the best performing one is #2, but #1 might match it if we
straighten out the confusion regarding which pages to flush. It
really does matter.
1. vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
2. vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
3. vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
4. vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
tfile->vnode->v_gnode->gn_seg = NULL;
5. deletep
6. ipgrlse
7. ifreeseg
Unfortunately, this seems to cause frequent "cache corruption" episodes.
vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
}
*/
#elif defined(AFS_AIX_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
&tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
{
caller_context_t ct;
VOP_RWLOCK(tfile->vnode, 0, &ct);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, &ct);
VOP_RWUNLOCK(tfile->vnode, 0, &ct);
}
#else
VOP_RWLOCK(tfile->vnode, 0);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_RWUNLOCK(tfile->vnode, 0);
#endif
AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
AFS_GUNLOCK();
AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
code);
AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
AFS_GLOCK();
#elif defined(AFS_OSF_ENV)
tuio.uio_rw = UIO_READ;
AFS_GUNLOCK();
VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, code);
AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
AFS_GUNLOCK();
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
AFS_GUNLOCK();
code = osi_rdwr(tfile, &tuio, UIO_READ);
AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
AFS_GUNLOCK();
code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, current_proc());
AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_FBSD50_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curthread);
AFS_GLOCK();
#elif defined(AFS_XBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curproc);
AFS_GLOCK();
#else
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif
#ifdef IHINT
if (!tdc->ihint && nihints < maxIHint) {
tdc->ihint = tfile;
nihints++;
} else
#endif /* IHINT */
osi_UFSClose(tfile);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
}
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch, obviously.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
/* try to queue prefetch, if needed */
if (!noLock) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, acred, &treq);
}
#endif
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
AFS_DISCON_UNLOCK();
error = afs_CheckCode(error, &treq, 13);
return error;
}
