#include <linux/pagemap.h>
#include <linux/writeback.h>
#include <linux/pagevec.h>
+#include <linux/aio.h>
#include "afs/lock.h"
#include "afs/afs_bypasscache.h"
#define MAX_ERRNO 1000L
#endif
+int cachefs_noreadpage = 0;
+
extern struct backing_dev_info *afs_backing_dev_info;
extern struct vcache *afs_globalVp;
}
#ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
+# ifdef LINUX_HAS_NONVECTOR_AIO
+static ssize_t
+afs_linux_aio_read(struct kiocb *iocb, char __user *buf, size_t bufsize,
+ loff_t pos)
+# else
static ssize_t
-afs_linux_aio_read(struct kiocb *iocb, const struct iovec *iov, unsigned long segs, loff_t pos)
+afs_linux_aio_read(struct kiocb *iocb, const struct iovec *buf,
+ unsigned long bufsize, loff_t pos)
+# endif
{
struct file *fp = iocb->ki_filp;
ssize_t code = 0;
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);
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
+ (afs_int32)bufsize, ICL_TYPE_INT32, 99999);
code = afs_linux_VerifyVCache(vcp, NULL);
if (code == 0) {
* 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);
+ code = generic_file_aio_read(iocb, buf, bufsize, 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);
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
+ (afs_int32)bufsize, ICL_TYPE_INT32, code);
AFS_GUNLOCK();
return code;
}
* mode. Call fake open/close to ensure we do writes of core dumps.
*/
#ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
+# ifdef LINUX_HAS_NONVECTOR_AIO
+static ssize_t
+afs_linux_aio_write(struct kiocb *iocb, const char __user *buf, size_t bufsize,
+ loff_t pos)
+# else
static ssize_t
-afs_linux_aio_write(struct kiocb *iocb, const struct iovec *iov, unsigned long segs, loff_t pos)
+afs_linux_aio_write(struct kiocb *iocb, const struct iovec *buf,
+ unsigned long bufsize, loff_t pos)
+# endif
{
ssize_t code = 0;
struct vcache *vcp = VTOAFS(iocb->ki_filp->f_dentry->d_inode);
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,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
+ (afs_int32)bufsize, ICL_TYPE_INT32,
(iocb->ki_filp->f_flags & O_APPEND) ? 99998 : 99999);
code = afs_linux_VerifyVCache(vcp, &credp);
ReleaseWriteLock(&vcp->lock);
if (code == 0) {
AFS_GUNLOCK();
- code = generic_file_aio_write(iocb, iov, segs, pos);
+ code = generic_file_aio_write(iocb, buf, bufsize, pos);
AFS_GLOCK();
}
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);
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
+ (afs_int32)bufsize, ICL_TYPE_INT32, code);
if (credp)
crfree(credp);
* handling and use of bulkstats will need to be reflected here as well.
*/
static int
+#if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
+afs_linux_readdir(struct file *fp, struct dir_context *ctx)
+#else
afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
+#endif
{
struct vcache *avc = VTOAFS(FILE_INODE(fp));
struct vrequest treq;
* takes an offset in units of blobs, rather than bytes.
*/
code = 0;
+#if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
+ offset = ctx->pos;
+#else
offset = (int) fp->f_pos;
+#endif
while (1) {
dirpos = BlobScan(tdc, offset);
if (!dirpos)
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);
+ if (!(avc->f.states & CCorrupt)) {
+ struct cell *tc = afs_GetCellStale(avc->f.fid.Cell, READ_LOCK);
+ afs_warn("Corrupt directory (%d.%d.%d.%d [%s] @%lx, pos %d)",
+ avc->f.fid.Cell, avc->f.fid.Fid.Volume,
+ avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique,
+ tc ? tc->cellName : "",
+ (unsigned long)&tdc->f.inode, dirpos);
+ if (tc)
+ afs_PutCell(tc, READ_LOCK);
+ UpgradeSToWLock(&avc->lock, 814);
+ avc->f.states |= CCorrupt;
+ }
code = -ENOENT;
- goto out;
+ goto unlock_out;
}
de = (struct DirEntry *)entry.data;
* holding the GLOCK.
*/
AFS_GUNLOCK();
+#if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
+ /* dir_emit returns a bool - true when it succeeds.
+ * Inverse the result to fit with how we check "code" */
+ code = !dir_emit(ctx, de->name, len, ino, type);
+#else
code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
+#endif
AFS_GLOCK();
}
DRelease(&entry, 0);
/* If filldir didn't fill in the last one this is still pointing to that
* last attempt.
*/
+#if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
+ ctx->pos = (loff_t) offset;
+#else
fp->f_pos = (loff_t) offset;
+#endif
+ code = 0;
+unlock_out:
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
UpgradeSToWLock(&avc->lock, 813);
avc->dcreaddir = 0;
avc->readdir_pid = 0;
ReleaseSharedLock(&avc->lock);
- code = 0;
out:
afs_PutFakeStat(&fakestat);
#endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
AFS_GLOCK();
+ if ((vcp->f.states & CRO)) {
+ if (flp->fl_type == F_WRLCK) {
+ code = EBADF;
+ } else {
+ code = 0;
+ }
+ AFS_GUNLOCK();
+ crfree(credp);
+ return code;
+ }
code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
AFS_GUNLOCK();
struct file_operations afs_dir_fops = {
.read = generic_read_dir,
+#if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
+ .iterate = afs_linux_readdir,
+#else
.readdir = afs_linux_readdir,
+#endif
#ifdef HAVE_UNLOCKED_IOCTL
.unlocked_ioctl = afs_unlocked_xioctl,
#else
.llseek = default_llseek,
};
+static struct dentry *
+canonical_dentry(struct inode *ip)
+{
+ struct vcache *vcp = VTOAFS(ip);
+ struct dentry *first = NULL, *ret = NULL, *cur;
+#if defined(D_ALIAS_IS_HLIST) && !defined(HLIST_ITERATOR_NO_NODE)
+ struct hlist_node *p;
+#endif
+
+ /* general strategy:
+ * if vcp->target_link is set, and can be found in ip->i_dentry, use that.
+ * otherwise, use the first dentry in ip->i_dentry.
+ * if ip->i_dentry is empty, use the 'dentry' argument we were given.
+ */
+ /* note that vcp->target_link specifies which dentry to use, but we have
+ * no reference held on that dentry. so, we cannot use or dereference
+ * vcp->target_link itself, since it may have been freed. instead, we only
+ * use it to compare to pointers in the ip->i_dentry list. */
+
+ d_prune_aliases(ip);
+
+# ifdef HAVE_DCACHE_LOCK
+ spin_lock(&dcache_lock);
+# else
+ spin_lock(&ip->i_lock);
+# endif
+
+#if defined(D_ALIAS_IS_HLIST)
+# if defined(HLIST_ITERATOR_NO_NODE)
+ hlist_for_each_entry(cur, &ip->i_dentry, d_alias) {
+# else
+ hlist_for_each_entry(cur, p, &ip->i_dentry, d_alias) {
+# endif
+#else
+ list_for_each_entry_reverse(cur, &ip->i_dentry, d_alias) {
+#endif
+
+ if (!vcp->target_link || cur == vcp->target_link) {
+ ret = cur;
+ break;
+ }
+
+ if (!first) {
+ first = cur;
+ }
+ }
+ if (!ret && first) {
+ ret = first;
+ }
+
+ vcp->target_link = ret;
+
+# ifdef HAVE_DCACHE_LOCK
+ if (ret) {
+ dget_locked(ret);
+ }
+ spin_unlock(&dcache_lock);
+# else
+ if (ret) {
+ dget(ret);
+ }
+ spin_unlock(&ip->i_lock);
+# endif
+
+ return ret;
+}
/**********************************************************************
* AFS Linux dentry operations
}
AFS_GUNLOCK();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
static int
* The code here assumes that on entry the global lock is not held
*/
static int
-#ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
+#if defined(DOP_REVALIDATE_TAKES_UNSIGNED)
+afs_linux_dentry_revalidate(struct dentry *dp, unsigned int flags)
+#elif defined(DOP_REVALIDATE_TAKES_NAMEIDATA)
afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
#else
afs_linux_dentry_revalidate(struct dentry *dp, int flags)
#ifdef LOOKUP_RCU
/* We don't support RCU path walking */
+# if defined(DOP_REVALIDATE_TAKES_UNSIGNED)
+ if (flags & LOOKUP_RCU)
+# else
if (nd->flags & LOOKUP_RCU)
+# endif
return -ECHILD;
#endif
code = afs_EvalFakeStat(&vcp, &fakestate, &treq);
if ((tryEvalOnly && vcp->mvstat == 1) || code) {
/* a mount point, not yet replaced by its directory */
+ dput(parent);
goto bad_dentry;
}
}
* always require a crref() which would be "slow".
*/
if (vcp->last_looker != treq.uid) {
- if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS))
+ if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
+ dput(parent);
goto bad_dentry;
+ }
vcp->last_looker = treq.uid;
}
return 0;
}
+#ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
+static struct vfsmount *
+afs_dentry_automount(afs_linux_path_t *path)
+{
+ struct dentry *target;
+
+ /* avoid symlink resolution limits when resolving; we cannot contribute to
+ * an infinite symlink loop */
+ current->total_link_count--;
+
+ target = canonical_dentry(path->dentry->d_inode);
+
+ if (target == path->dentry) {
+ dput(target);
+ target = NULL;
+ }
+
+ if (target) {
+ dput(path->dentry);
+ path->dentry = target;
+
+ } else {
+ spin_lock(&path->dentry->d_lock);
+ path->dentry->d_flags &= ~DCACHE_NEED_AUTOMOUNT;
+ spin_unlock(&path->dentry->d_lock);
+ }
+
+ return NULL;
+}
+#endif /* STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
+
struct dentry_operations afs_dentry_operations = {
.d_revalidate = afs_linux_dentry_revalidate,
.d_delete = afs_dentry_delete,
.d_iput = afs_dentry_iput,
+#ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
+ .d_automount = afs_dentry_automount,
+#endif /* STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
};
/**********************************************************************
* name is in kernel space at this point.
*/
static int
-#if defined(IOP_MKDIR_TAKES_UMODE_T)
+#if defined(IOP_CREATE_TAKES_BOOL)
+afs_linux_create(struct inode *dip, struct dentry *dp, umode_t mode,
+ bool excl)
+#elif defined(IOP_CREATE_TAKES_UMODE_T)
afs_linux_create(struct inode *dip, struct dentry *dp, umode_t mode,
struct nameidata *nd)
-#else
-#ifdef IOP_CREATE_TAKES_NAMEIDATA
+#elif defined(IOP_CREATE_TAKES_NAMEIDATA)
afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
struct nameidata *nd)
#else
afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
#endif
-#endif
{
struct vattr vattr;
cred_t *credp = crref();
/* afs_linux_lookup */
static struct dentry *
-#ifdef IOP_LOOKUP_TAKES_NAMEIDATA
+#if defined(IOP_LOOKUP_TAKES_UNSIGNED)
+afs_linux_lookup(struct inode *dip, struct dentry *dp,
+ unsigned flags)
+#elif defined(IOP_LOOKUP_TAKES_NAMEIDATA)
afs_linux_lookup(struct inode *dip, struct dentry *dp,
struct nameidata *nd)
#else
if (vcp) {
struct vattr vattr;
+ struct vcache *parent_vc = VTOAFS(dip);
+
+ if (parent_vc == vcp) {
+ /* This is possible if the parent dir is a mountpoint to a volume,
+ * and the dir entry we looked up is a mountpoint to the same
+ * volume. Linux cannot cope with this, so return an error instead
+ * of risking a deadlock or panic. */
+ afs_PutVCache(vcp);
+ code = EDEADLK;
+ AFS_GUNLOCK();
+ goto done;
+ }
ip = AFSTOV(vcp);
afs_getattr(vcp, &vattr, credp);
AFS_GUNLOCK();
if (ip && S_ISDIR(ip->i_mode)) {
- struct dentry *alias;
-
- /* Try to invalidate an existing alias in favor of our new one */
- alias = d_find_alias(ip);
- /* But not if it's disconnected; then we want d_splice_alias below */
- if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
- if (d_invalidate(alias) == 0) {
- dput(alias);
- } else {
- iput(ip);
- crfree(credp);
- return alias;
- }
- }
+ d_prune_aliases(ip);
+
+#ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
+ ip->i_flags |= S_AUTOMOUNT;
+#endif
}
newdp = d_splice_alias(ip, dp);
+ done:
crfree(credp);
/* It's ok for the file to not be found. That's noted by the caller by
if (cacheDiskType != AFS_FCACHE_TYPE_UFS)
return 0;
+ /* No readpage (ex: tmpfs) , skip */
+ if (cachefs_noreadpage)
+ return 0;
+
/* Can't do anything if the vcache isn't statd , or if the read
* crosses a chunk boundary.
*/
/* Is the dcache we've been given currently up to date */
if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
- (tdc->dflags & DFFetching)) {
- ReleaseWriteLock(&avc->lock);
- ReleaseReadLock(&tdc->lock);
- afs_PutDCache(tdc);
- return 0;
- }
+ (tdc->dflags & DFFetching))
+ goto out;
/* Update our hint for future abuse */
avc->dchint = tdc;
/* XXX - I suspect we should be locking the inodes before we use them! */
AFS_GUNLOCK();
cacheFp = afs_linux_raw_open(&tdc->f.inode);
+ if (!cacheFp->f_dentry->d_inode->i_mapping->a_ops->readpage) {
+ cachefs_noreadpage = 1;
+ AFS_GLOCK();
+ goto out;
+ }
pagevec_init(&lrupv, 0);
code = afs_linux_read_cache(cacheFp, pp, tdc->f.chunk, &lrupv, NULL);
*codep = code;
return 1;
+
+out:
+ ReleaseWriteLock(&avc->lock);
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ return 0;
}
/* afs_linux_readpage
if(page_ix == 0) {
offset = page_offset(pp);
- auio->uio_offset = offset;
+ ancr->offset = auio->uio_offset = offset;
base_index = pp->index;
}
iovecp[page_ix].iov_len = PAGE_SIZE;
static inline int
afs_linux_can_bypass(struct inode *ip) {
+
switch(cache_bypass_strategy) {
case NEVER_BYPASS_CACHE:
return 0;
case ALWAYS_BYPASS_CACHE:
return 1;
case LARGE_FILES_BYPASS_CACHE:
- if(i_size_read(ip) > cache_bypass_threshold)
+ if (i_size_read(ip) > cache_bypass_threshold)
return 1;
default:
return 0;
if (cacheDiskType == AFS_FCACHE_TYPE_MEM)
return 0;
+ /* No readpage (ex: tmpfs) , skip */
+ if (cachefs_noreadpage)
+ return 0;
+
AFS_GLOCK();
if ((code = afs_linux_VerifyVCache(avc, NULL))) {
AFS_GUNLOCK();
ObtainReadLock(&tdc->lock);
if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
(tdc->dflags & DFFetching)) {
+ goto out;
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
AFS_GUNLOCK();
- if (tdc)
+ if (tdc) {
cacheFp = afs_linux_raw_open(&tdc->f.inode);
+ if (!cacheFp->f_dentry->d_inode->i_mapping->a_ops->readpage) {
+ cachefs_noreadpage = 1;
+ goto out;
+ }
+ }
}
if (tdc && !add_to_page_cache(page, mapping, page->index,
if (pagevec_count(&lrupv))
__pagevec_lru_add_file(&lrupv);
+out:
if (tdc)
filp_close(cacheFp, NULL);
}
#endif
+#ifndef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
+static void *
+afs_linux_dir_follow_link(struct dentry *dentry, struct nameidata *nd)
+{
+ struct dentry **dpp;
+ struct dentry *target;
+
+ if (current->total_link_count > 0) {
+ /* avoid symlink resolution limits when resolving; we cannot contribute to
+ * an infinite symlink loop */
+ /* only do this for follow_link when total_link_count is positive to be
+ * on the safe side; there is at least one code path in the Linux
+ * kernel where it seems like it may be possible to get here without
+ * total_link_count getting incremented. it is not clear on how that
+ * path is actually reached, but guard against it just to be safe */
+ current->total_link_count--;
+ }
+
+ target = canonical_dentry(dentry->d_inode);
+
+# ifdef STRUCT_NAMEIDATA_HAS_PATH
+ dpp = &nd->path.dentry;
+# else
+ dpp = &nd->dentry;
+# endif
+
+ dput(*dpp);
+
+ if (target) {
+ *dpp = target;
+ } else {
+ *dpp = dget(dentry);
+ }
+
+ nd->last_type = LAST_BIND;
+
+ return NULL;
+}
+#endif /* !STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
+
static struct inode_operations afs_file_iops = {
.permission = afs_linux_permission,
.rename = afs_linux_rename,
.getattr = afs_linux_getattr,
.permission = afs_linux_permission,
+#ifndef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
+ .follow_link = afs_linux_dir_follow_link,
+#endif
};
/* We really need a separate symlink set of ops, since do_follow_link()
git://git.openafs.org / openafs.git / blobdiff