#include <linux/mm_inline.h>
#endif
#include <linux/pagemap.h>
-#include <linux/smp_lock.h>
#include <linux/writeback.h>
#include <linux/pagevec.h>
-#if defined(AFS_CACHE_BYPASS)
#include "afs/lock.h"
#include "afs/afs_bypasscache.h"
-#endif
#include "osi_compat.h"
#include "osi_pagecopy.h"
return afs_convert_code(code);
}
+#ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
+static ssize_t
+afs_linux_aio_read(struct kiocb *iocb, const struct iovec *iov, unsigned long segs, loff_t pos)
+{
+ struct file *fp = iocb->ki_filp;
+ ssize_t code = 0;
+ struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
+
+ AFS_GLOCK();
+ afs_Trace4(afs_iclSetp, CM_TRACE_AIOREADOP, ICL_TYPE_POINTER, vcp,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32, segs, ICL_TYPE_INT32,
+ 99999);
+ code = afs_linux_VerifyVCache(vcp, NULL);
+
+ if (code == 0) {
+ /* Linux's FlushPages implementation doesn't ever use credp,
+ * so we optimise by not using it */
+ osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
+ AFS_GUNLOCK();
+ code = generic_file_aio_read(iocb, iov, segs, pos);
+ AFS_GLOCK();
+ }
+
+ afs_Trace4(afs_iclSetp, CM_TRACE_AIOREADOP, ICL_TYPE_POINTER, vcp,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32, segs, ICL_TYPE_INT32,
+ code);
+ AFS_GUNLOCK();
+ return code;
+}
+#else
static ssize_t
afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
{
AFS_GUNLOCK();
return code;
}
+#endif
-/* Now we have integrated VM for writes as well as reads. generic_file_write
- * also takes care of re-positioning the pointer if file is open in append
+/* Now we have integrated VM for writes as well as reads. the generic write operations
+ * also take care of re-positioning the pointer if file is open in append
* mode. Call fake open/close to ensure we do writes of core dumps.
*/
+#ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
+static ssize_t
+afs_linux_aio_write(struct kiocb *iocb, const struct iovec *iov, unsigned long segs, loff_t pos)
+{
+ ssize_t code = 0;
+ struct vcache *vcp = VTOAFS(iocb->ki_filp->f_dentry->d_inode);
+ cred_t *credp;
+
+ AFS_GLOCK();
+
+ afs_Trace4(afs_iclSetp, CM_TRACE_AIOWRITEOP, ICL_TYPE_POINTER, vcp,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32, segs, ICL_TYPE_INT32,
+ (iocb->ki_filp->f_flags & O_APPEND) ? 99998 : 99999);
+
+ code = afs_linux_VerifyVCache(vcp, &credp);
+
+ ObtainWriteLock(&vcp->lock, 529);
+ afs_FakeOpen(vcp);
+ ReleaseWriteLock(&vcp->lock);
+ if (code == 0) {
+ AFS_GUNLOCK();
+ code = generic_file_aio_write(iocb, iov, segs, pos);
+ AFS_GLOCK();
+ }
+
+ ObtainWriteLock(&vcp->lock, 530);
+
+ if (vcp->execsOrWriters == 1 && !credp)
+ credp = crref();
+
+ afs_FakeClose(vcp, credp);
+ ReleaseWriteLock(&vcp->lock);
+
+ afs_Trace4(afs_iclSetp, CM_TRACE_AIOWRITEOP, ICL_TYPE_POINTER, vcp,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32, segs, ICL_TYPE_INT32,
+ code);
+
+ if (credp)
+ crfree(credp);
+ AFS_GUNLOCK();
+ return code;
+}
+#else
static ssize_t
afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
{
AFS_GUNLOCK();
return code;
}
+#endif
extern int BlobScan(struct dcache * afile, afs_int32 ablob);
int offset;
int dirpos;
struct DirEntry *de;
+ struct DirBuffer entry;
ino_t ino;
int len;
afs_size_t origOffset, tlen;
code = -ENOENT;
goto out;
}
- ObtainSharedLock(&avc->lock, 810);
- UpgradeSToWLock(&avc->lock, 811);
+ ObtainWriteLock(&avc->lock, 811);
ObtainReadLock(&tdc->lock);
/*
* Make sure that the data in the cache is current. There are two
&& (tdc->dflags & DFFetching)
&& hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
- ReleaseSharedLock(&avc->lock);
+ ReleaseWriteLock(&avc->lock);
afs_osi_Sleep(&tdc->validPos);
- ObtainSharedLock(&avc->lock, 812);
+ ObtainWriteLock(&avc->lock, 812);
ObtainReadLock(&tdc->lock);
}
if (!(avc->f.states & CStatd)
|| !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
- ReleaseSharedLock(&avc->lock);
+ ReleaseWriteLock(&avc->lock);
afs_PutDCache(tdc);
goto tagain;
}
if (!dirpos)
break;
- de = afs_dir_GetBlob(tdc, dirpos);
- if (!de)
- break;
-
- ino = afs_calc_inum (avc->f.fid.Fid.Volume, ntohl(de->fid.vnode));
-
- if (de->name)
- len = strlen(de->name);
- else {
- printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
- (unsigned long)&tdc->f.inode, dirpos);
- DRelease(de, 0);
+ code = afs_dir_GetVerifiedBlob(tdc, dirpos, &entry);
+ if (code) {
+ afs_warn("Corrupt directory (inode %lx, dirpos %d)",
+ (unsigned long)&tdc->f.inode, dirpos);
ReleaseSharedLock(&avc->lock);
afs_PutDCache(tdc);
code = -ENOENT;
goto out;
- }
+ }
+
+ de = (struct DirEntry *)entry.data;
+ ino = afs_calc_inum (avc->f.fid.Cell, avc->f.fid.Fid.Volume,
+ ntohl(de->fid.vnode));
+ len = strlen(de->name);
/* filldir returns -EINVAL when the buffer is full. */
{
code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
AFS_GLOCK();
}
- DRelease(de, 0);
+ DRelease(&entry, 0);
if (code)
break;
offset = dirpos + 1 + ((len + 16) >> 5);
struct vcache *vcp;
cred_t *credp;
int code;
-#if defined(AFS_CACHE_BYPASS)
int bypasscache = 0;
-#endif
AFS_GLOCK();
code = afs_InitReq(&treq, credp);
if (code)
goto out;
-#if defined(AFS_CACHE_BYPASS)
/* If caching is bypassed for this file, or globally, just return 0 */
if (cache_bypass_strategy == ALWAYS_BYPASS_CACHE)
bypasscache = 1;
code = 0;
goto out;
}
-#endif
ObtainSharedLock(&vcp->lock, 535);
if ((vcp->execsOrWriters > 0) && (file_count(fp) == 1)) {
#endif
.open = afs_linux_open,
.release = afs_linux_release,
+ .llseek = default_llseek,
};
struct file_operations afs_file_fops = {
+#ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
+ .aio_read = afs_linux_aio_read,
+ .aio_write = afs_linux_aio_write,
+#else
.read = afs_linux_read,
.write = afs_linux_write,
-#ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
- .aio_read = generic_file_aio_read,
- .aio_write = generic_file_aio_write,
#endif
#ifdef HAVE_UNLOCKED_IOCTL
.unlocked_ioctl = afs_unlocked_xioctl,
#ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
.flock = afs_linux_flock,
#endif
+ .llseek = default_llseek,
};
* AFS Linux dentry operations
**********************************************************************/
-/* check_bad_parent() : Checks if this dentry's vcache is a root vcache
+/* fix_bad_parent() : called if this dentry's vcache is a root vcache
* that has its mvid (parent dir's fid) pointer set to the wrong directory
- * due to being mounted in multiple points at once. If so, check_bad_parent()
+ * due to being mounted in multiple points at once. fix_bad_parent()
* calls afs_lookup() to correct the vcache's mvid, as well as the volume's
* dotdotfid and mtpoint fid members.
* Parameters:
* dp - dentry to be checked.
+ * credp - credentials
+ * vcp, pvc - item's and parent's vcache pointer
* Return Values:
* None.
* Sideeffects:
*/
static inline void
-check_bad_parent(struct dentry *dp)
+fix_bad_parent(struct dentry *dp, cred_t *credp, struct vcache *vcp, struct vcache *pvc)
{
- cred_t *credp;
- struct dentry *parent;
- struct vcache *vcp, *pvc, *avc = NULL;
-
- vcp = VTOAFS(dp->d_inode);
- parent = dget_parent(dp);
- pvc = VTOAFS(parent->d_inode);
-
- if (vcp->mvid->Fid.Volume != pvc->f.fid.Fid.Volume) { /* bad parent */
- credp = crref();
-
- /* force a lookup, so vcp->mvid is fixed up */
- afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
- if (!avc || vcp != avc) { /* bad, very bad.. */
- afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
- "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
- ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
- ICL_TYPE_POINTER, dp);
- }
- if (avc)
- AFS_RELE(AFSTOV(avc));
- crfree(credp);
+ struct vcache *avc = NULL;
+
+ /* force a lookup, so vcp->mvid is fixed up */
+ afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
+ if (!avc || vcp != avc) { /* bad, very bad.. */
+ afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
+ "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
+ ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
+ ICL_TYPE_POINTER, dp);
}
-
- dput(parent);
+ if (avc)
+ AFS_RELE(AFSTOV(avc));
return;
}
#ifdef notyet
/* Make this a fast path (no crref), since it's called so often. */
- if (vcp->f.states & CStatd) {
+ if (vcp->states & CStatd) {
+ struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
- if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
- check_bad_parent(dp); /* check and correct mvid */
-
- AFS_GUNLOCK();
+ if (*dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
+ if (vcp->mvid->Fid.Volume != pvc->fid.Fid.Volume) { /* bad parent */
+ credp = crref();
+ AFS_GLOCK();
+ fix_bad_parent(dp); /* check and correct mvid */
+ AFS_GUNLOCK();
+ crfree(credp);
+ }
+ }
return 0;
}
#endif
code = afs_getattr(vcp, &vattr, credp);
crfree(credp);
}
+
if (!code)
afs_fill_inode(AFSTOV(vcp), &vattr);
* we are advised to follow the entry if it is a link or to make sure that
* it is a directory. But since the kernel itself checks these possibilities
* later on, we shouldn't have to do it until later. Perhaps in the future..
+ *
+ * The code here assumes that on entry the global lock is not held
*/
static int
#ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
struct dentry *parent;
int valid;
struct afs_fakestat_state fakestate;
+ int locked = 0;
+
+#ifdef LOOKUP_RCU
+ /* We don't support RCU path walking */
+ if (nd->flags & LOOKUP_RCU)
+ return -ECHILD;
+#endif
- AFS_GLOCK();
afs_InitFakeStat(&fakestate);
if (dp->d_inode) {
if (vcp == afs_globalVp)
goto good_dentry;
- if (vcp->mvstat == 1) { /* mount point */
+ parent = dget_parent(dp);
+ pvcp = VTOAFS(parent->d_inode);
+
+ if ((vcp->mvstat == 1) || (vcp->mvstat == 2)) { /* need to lock */
+ credp = crref();
+ AFS_GLOCK();
+ locked = 1;
+ }
+
+ if (locked && vcp->mvstat == 1) { /* mount point */
if (vcp->mvid && (vcp->f.states & CMValid)) {
int tryEvalOnly = 0;
int code = 0;
struct vrequest treq;
- credp = crref();
code = afs_InitReq(&treq, credp);
if (
(strcmp(dp->d_name.name, ".directory") == 0)) {
}
}
} else
- if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
- check_bad_parent(dp); /* check and correct mvid */
+ if (locked && *dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
+ if (vcp->mvid->Fid.Volume != pvcp->f.fid.Fid.Volume) { /* bad parent */
+ fix_bad_parent(dp, credp, vcp, pvcp); /* check and correct mvid */
+ }
+ }
#ifdef notdef
/* If the last looker changes, we should make sure the current
}
#endif
- parent = dget_parent(dp);
- pvcp = VTOAFS(parent->d_inode);
/* If the parent's DataVersion has changed or the vnode
* is longer valid, we need to do a full lookup. VerifyVCache
* isn't enough since the vnode may have been renamed.
*/
- if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd)) {
-
+ if ((!locked) && (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd)) ) {
credp = crref();
+ AFS_GLOCK();
+ locked = 1;
+ }
+
+ if (locked && (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd))) {
afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
if (!tvc || tvc != vcp) {
dput(parent);
/* Clean up */
if (tvc)
afs_PutVCache(tvc);
- afs_PutFakeStat(&fakestate);
- AFS_GUNLOCK();
+ afs_PutFakeStat(&fakestate); /* from here on vcp may be no longer valid */
+ if (locked) {
+ /* we hold the global lock if we evaluated a mount point */
+ AFS_GUNLOCK();
+ }
if (credp)
crfree(credp);
afs_getattr(vcp, &vattr, credp);
afs_fill_inode(ip, &vattr);
insert_inode_hash(ip);
+#if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
dp->d_op = &afs_dentry_operations;
+#endif
dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
d_instantiate(dp, ip);
}
if (hlist_unhashed(&ip->i_hash))
insert_inode_hash(ip);
}
+#if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
dp->d_op = &afs_dentry_operations;
+#endif
dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
AFS_GUNLOCK();
afs_getattr(tvcp, &vattr, credp);
afs_fill_inode(ip, &vattr);
+#if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
dp->d_op = &afs_dentry_operations;
+#endif
dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
d_instantiate(dp, ip);
}
rehash = newdp;
}
+#if defined(D_COUNT_INT)
+ spin_lock(&olddp->d_lock);
+ if (olddp->d_count > 1) {
+ spin_unlock(&olddp->d_lock);
+ shrink_dcache_parent(olddp);
+ } else
+ spin_unlock(&olddp->d_lock);
+#else
if (atomic_read(&olddp->d_count) > 1)
shrink_dcache_parent(olddp);
+#endif
AFS_GLOCK();
code = afs_rename(VTOAFS(oldip), (char *)oldname, VTOAFS(newip), (char *)newname, credp);
{
int code;
cred_t *credp = crref();
- uio_t tuio;
+ struct uio tuio;
struct iovec iov;
setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
int code;
char *name;
- name = osi_Alloc(PATH_MAX);
+ name = kmalloc(PATH_MAX, GFP_NOFS);
if (!name) {
return -EIO;
}
AFS_GUNLOCK();
if (code < 0) {
- goto out;
+ return code;
}
name[code] = '\0';
- code = vfs_follow_link(nd, name);
+ nd_set_link(nd, name);
+ return 0;
+}
-out:
- osi_Free(name, PATH_MAX);
+static void
+afs_linux_put_link(struct dentry *dentry, struct nameidata *nd)
+{
+ char *name = nd_get_link(nd);
- return code;
+ if (name && !IS_ERR(name))
+ kfree(name);
}
#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
-#if defined(AFS_CACHE_BYPASS)
-#endif /* defined(AFS_CACHE_BYPASS */
-
/* Populate a page by filling it from the cache file pointed at by cachefp
* (which contains indicated chunk)
* If task is NULL, the page copy occurs syncronously, and the routine
int chunk, struct pagevec *lrupv,
struct afs_pagecopy_task *task) {
loff_t offset = page_offset(page);
+ struct inode *cacheinode = cachefp->f_dentry->d_inode;
struct page *newpage, *cachepage;
struct address_space *cachemapping;
int pageindex;
int code = 0;
- cachemapping = cachefp->f_dentry->d_inode->i_mapping;
+ cachemapping = cacheinode->i_mapping;
newpage = NULL;
cachepage = NULL;
+ /* If we're trying to read a page that's past the end of the disk
+ * cache file, then just return a zeroed page */
+ if (AFS_CHUNKOFFSET(offset) >= i_size_read(cacheinode)) {
+ zero_user_segment(page, 0, PAGE_CACHE_SIZE);
+ SetPageUptodate(page);
+ if (task)
+ unlock_page(page);
+ return 0;
+ }
+
/* From our offset, we now need to work out which page in the disk
* file it corresponds to. This will be fun ... */
pageindex = (offset - AFS_CHUNKTOBASE(chunk)) >> PAGE_CACHE_SHIFT;
{
afs_int32 code;
char *address;
- uio_t *auio;
+ struct uio *auio;
struct iovec *iovecp;
struct inode *ip = FILE_INODE(fp);
afs_int32 cnt = page_count(pp);
address = kmap(pp);
ClearPageError(pp);
- auio = osi_Alloc(sizeof(uio_t));
- iovecp = osi_Alloc(sizeof(struct iovec));
+ auio = kmalloc(sizeof(struct uio), GFP_NOFS);
+ iovecp = kmalloc(sizeof(struct iovec), GFP_NOFS);
setup_uio(auio, iovecp, (char *)address, offset, PAGE_SIZE, UIO_READ,
AFS_UIOSYS);
kunmap(pp);
- osi_Free(auio, sizeof(uio_t));
- osi_Free(iovecp, sizeof(struct iovec));
+ kfree(auio);
+ kfree(iovecp);
crfree(credp);
return afs_convert_code(code);
}
-#if defined(AFS_CACHE_BYPASS)
-
static int
afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
struct list_head *page_list, unsigned num_pages)
{
afs_int32 page_ix;
- uio_t *auio;
+ struct uio *auio;
afs_offs_t offset;
struct iovec* iovecp;
struct nocache_read_request *ancr;
/* background thread must free: iovecp, auio, ancr */
iovecp = osi_Alloc(num_pages * sizeof(struct iovec));
- auio = osi_Alloc(sizeof(uio_t));
+ auio = osi_Alloc(sizeof(struct uio));
auio->uio_iov = iovecp;
auio->uio_iovcnt = num_pages;
auio->uio_flag = UIO_READ;
page_cache_release(pp);
iovecp[page_ix].iov_base = (void *) 0;
base_index++;
- continue;
+ ancr->length -= PAGE_SIZE;
+ continue;
}
base_index++;
if(code) {
/* If there is nothing for the background thread to handle,
* it won't be freeing the things that we never gave it */
osi_Free(iovecp, num_pages * sizeof(struct iovec));
- osi_Free(auio, sizeof(uio_t));
+ osi_Free(auio, sizeof(struct uio));
osi_Free(ancr, sizeof(struct nocache_read_request));
}
/* we do not flush, release, or unmap pages--that will be
afs_linux_bypass_readpage(struct file *fp, struct page *pp)
{
cred_t *credp = NULL;
- uio_t *auio;
+ struct uio *auio;
struct iovec *iovecp;
struct nocache_read_request *ancr;
int code;
+ /*
+ * Special case: if page is at or past end of file, just zero it and set
+ * it as up to date.
+ */
+ if (page_offset(pp) >= i_size_read(fp->f_mapping->host)) {
+ zero_user_segment(pp, 0, PAGE_CACHE_SIZE);
+ SetPageUptodate(pp);
+ unlock_page(pp);
+ return 0;
+ }
+
ClearPageError(pp);
/* receiver frees */
- auio = osi_Alloc(sizeof(uio_t));
+ auio = osi_Alloc(sizeof(struct uio));
iovecp = osi_Alloc(sizeof(struct iovec));
/* address can be NULL, because we overwrite it with 'pp', below */
return bypass;
}
-#else
-static inline int
-afs_linux_bypass_check(struct inode *ip) {
- return 0;
-}
-static inline int
-afs_linux_bypass_readpage(struct file *fp, struct page *pp) {
- return 0;
-}
-static inline int
-afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
- struct list_head *page_list, unsigned int num_pages) {
- return 0;
-}
-#endif
static int
afs_linux_readpage(struct file *fp, struct page *pp)
char *buffer;
afs_offs_t base;
int code = 0;
- uio_t tuio;
+ struct uio tuio;
struct iovec iovec;
int f_flags = 0;
* Check access rights - returns error if can't check or permission denied.
*/
static int
-#ifdef IOP_PERMISSION_TAKES_NAMEIDATA
+#if defined(IOP_PERMISSION_TAKES_FLAGS)
+afs_linux_permission(struct inode *ip, int mode, unsigned int flags)
+#elif defined(IOP_PERMISSION_TAKES_NAMEIDATA)
afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
#else
afs_linux_permission(struct inode *ip, int mode)
#endif
{
int code;
- cred_t *credp = crref();
+ cred_t *credp;
int tmp = 0;
+#if defined(IOP_PERMISSION_TAKES_FLAGS)
+ /* We don't support RCU path walking */
+ if (flags & IPERM_FLAG_RCU)
+ return -ECHILD;
+#endif
+
+ credp = crref();
AFS_GLOCK();
if (mode & MAY_EXEC)
tmp |= VEXEC;
#else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
.readlink = afs_linux_readlink,
.follow_link = afs_linux_follow_link,
+ .put_link = afs_linux_put_link,
#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
.setattr = afs_notify_change,
};