#endif
#if defined(AFS_LINUX26_ENV)
#include "h/writeback.h"
+#include "h/pagevec.h"
+#endif
+#if defined(AFS_CACHE_BYPASS)
+#include "afs/lock.h"
+#include "afs/afs_bypasscache.h"
#endif
#ifdef pgoff2loff
#endif
#if defined(AFS_LINUX26_ENV)
+#define LockPage(pp) lock_page(pp)
#define UnlockPage(pp) unlock_page(pp)
#endif
extern struct vcache *afs_globalVp;
+#if defined(AFS_LINUX26_ENV)
+/* Some uses of BKL are perhaps not needed for bypass or memcache--
+ * why don't we try it out? */
+extern struct afs_cacheOps afs_UfsCacheOps;
+#define maybe_lock_kernel() \
+ do { \
+ if(afs_cacheType == &afs_UfsCacheOps) \
+ lock_kernel(); \
+ } while(0);
+
+
+#define maybe_unlock_kernel() \
+ do { \
+ if(afs_cacheType == &afs_UfsCacheOps) \
+ unlock_kernel(); \
+ } while(0);
+#endif /* AFS_CACHE_BYPASS */
+
static ssize_t
afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
{
struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
cred_t *credp = crref();
struct vrequest treq;
-
+ afs_size_t isize, offindex;
AFS_GLOCK();
afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
99999);
-
/* get a validated vcache entry */
code = afs_InitReq(&treq, credp);
if (!code)
if (code)
code = -code;
else {
- osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
- AFS_GUNLOCK();
- code = generic_file_read(fp, buf, count, offp);
- AFS_GLOCK();
+#if defined(AFS_CACHE_BYPASS)
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+ isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
+ offindex = *offp >> PAGE_CACHE_SHIFT;
+ if(offindex > isize) {
+ code=0;
+ goto done;
+ }
+#endif
+#endif
+ osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
+ AFS_GUNLOCK();
+#ifdef DO_SYNC_READ
+ code = do_sync_read(fp, buf, count, offp);
+#else
+ code = generic_file_read(fp, buf, count, offp);
+#endif
+ AFS_GLOCK();
}
afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
code);
-
+done:
AFS_GUNLOCK();
crfree(credp);
return code;
afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
{
ssize_t code = 0;
- int code2 = 0;
struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
struct vrequest treq;
cred_t *credp = crref();
code = -code;
else {
AFS_GUNLOCK();
+#ifdef DO_SYNC_READ
+ code = do_sync_write(fp, buf, count, offp);
+#else
code = generic_file_write(fp, buf, count, offp);
+#endif
AFS_GLOCK();
}
struct afs_fakestat_state fakestat;
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
AFS_STATCNT(afs_readdir);
code = -ENOENT;
goto out;
}
- ObtainReadLock(&avc->lock);
+ ObtainSharedLock(&avc->lock, 810);
+ UpgradeSToWLock(&avc->lock, 811);
ObtainReadLock(&tdc->lock);
/*
* Make sure that the data in the cache is current. There are two
&& (tdc->dflags & DFFetching)
&& hsame(avc->m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
- ReleaseReadLock(&avc->lock);
+ ReleaseSharedLock(&avc->lock);
afs_osi_Sleep(&tdc->validPos);
- ObtainReadLock(&avc->lock);
+ ObtainSharedLock(&avc->lock, 812);
ObtainReadLock(&tdc->lock);
}
if (!(avc->states & CStatd)
|| !hsame(avc->m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
- ReleaseReadLock(&avc->lock);
+ ReleaseSharedLock(&avc->lock);
afs_PutDCache(tdc);
goto tagain;
}
+ /* Set the readdir-in-progress flag, and downgrade the lock
+ * to shared so others will be able to acquire a read lock.
+ */
+ avc->states |= CReadDir;
+ avc->dcreaddir = tdc;
+ avc->readdir_pid = MyPidxx;
+ ConvertWToSLock(&avc->lock);
+
/* Fill in until we get an error or we're done. This implementation
* takes an offset in units of blobs, rather than bytes.
*/
printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
(unsigned long)&tdc->f.inode, dirpos);
DRelease((struct buffer *) de, 0);
+ ReleaseSharedLock(&avc->lock);
afs_PutDCache(tdc);
- ReleaseReadLock(&avc->lock);
code = -ENOENT;
goto out;
}
/* clean up from afs_FindVCache */
afs_PutVCache(tvc);
}
+ /*
+ * If this is NFS readdirplus, then the filler is going to
+ * call getattr on this inode, which will deadlock if we're
+ * holding the GLOCK.
+ */
+ AFS_GUNLOCK();
code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
+ AFS_GLOCK();
}
#else
code = (*filldir) (dirbuf, de->name, len, offset, ino);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
- ReleaseReadLock(&avc->lock);
+ UpgradeSToWLock(&avc->lock, 813);
+ avc->states &= ~CReadDir;
+ avc->dcreaddir = 0;
+ avc->readdir_pid = 0;
+ ReleaseSharedLock(&avc->lock);
code = 0;
out:
out1:
AFS_GUNLOCK();
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
return code;
}
int code;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
code = afs_open(&vcp, fp->f_flags, credp);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
int code = 0;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
code = afs_close(vcp, fp->f_flags, credp);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
cred_t *credp = crref();
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
code = afs_fsync(VTOAFS(ip), credp);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- if (code == 0 && (cmd == F_SETLK || cmd == F_SETLKW)) {
- struct file_lock flp2;
- flp2 = *flp;
-#ifdef AFS_LINUX26_ENV
- flp2.fl_flags &=~ FL_SLEEP;
-#endif
- code = posix_lock_file(fp, &flp2);
- osi_Assert(code != -EAGAIN); /* there should be no conflicts */
- if (code) {
- struct AFS_FLOCK flock2;
- flock2 = flock;
- flock2.l_type = F_UNLCK;
- AFS_GLOCK();
- afs_lockctl(vcp, &flock2, F_SETLK, credp);
- AFS_GUNLOCK();
- }
+ if ((code == 0 || flp->fl_type == F_UNLCK) &&
+ (cmd == F_SETLK || cmd == F_SETLKW)) {
+#ifdef POSIX_LOCK_FILE_WAIT_ARG
+ code = posix_lock_file(fp, flp, 0);
+#else
+ flp->fl_flags &=~ FL_SLEEP;
+ code = posix_lock_file(fp, flp);
+#endif
+ if (code && flp->fl_type != F_UNLCK) {
+ struct AFS_FLOCK flock2;
+ flock2 = flock;
+ flock2.l_type = F_UNLCK;
+ AFS_GLOCK();
+ afs_lockctl(vcp, &flock2, F_SETLK, credp);
+ AFS_GUNLOCK();
+ }
}
#endif
/* Convert flock back to Linux's file_lock */
}
+#ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
+static int
+afs_linux_flock(struct file *fp, int cmd, struct file_lock *flp) {
+ int code = 0;
+ struct vcache *vcp = VTOAFS(FILE_INODE(fp));
+ cred_t *credp = crref();
+ struct AFS_FLOCK flock;
+ /* Convert to a lock format afs_lockctl understands. */
+ memset((char *)&flock, 0, sizeof(flock));
+ flock.l_type = flp->fl_type;
+ flock.l_pid = flp->fl_pid;
+ flock.l_whence = 0;
+ flock.l_start = 0;
+ flock.l_len = OFFSET_MAX;
+
+ /* Safe because there are no large files, yet */
+#if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
+ if (cmd == F_GETLK64)
+ cmd = F_GETLK;
+ else if (cmd == F_SETLK64)
+ cmd = F_SETLK;
+ else if (cmd == F_SETLKW64)
+ cmd = F_SETLKW;
+#endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
+
+ AFS_GLOCK();
+ code = afs_lockctl(vcp, &flock, cmd, credp);
+ AFS_GUNLOCK();
+
+ if ((code == 0 || flp->fl_type == F_UNLCK) &&
+ (cmd == F_SETLK || cmd == F_SETLKW)) {
+ flp->fl_flags &=~ FL_SLEEP;
+ code = flock_lock_file_wait(fp, flp);
+ if (code && flp->fl_type != F_UNLCK) {
+ struct AFS_FLOCK flock2;
+ flock2 = flock;
+ flock2.l_type = F_UNLCK;
+ AFS_GLOCK();
+ afs_lockctl(vcp, &flock2, F_SETLK, credp);
+ AFS_GUNLOCK();
+ }
+ }
+ /* Convert flock back to Linux's file_lock */
+ flp->fl_type = flock.l_type;
+ flp->fl_pid = flock.l_pid;
+
+ crfree(credp);
+ return -code;
+}
+#endif
+
/* afs_linux_flush
* essentially the same as afs_fsync() but we need to get the return
* code for the sys_close() here, not afs_linux_release(), so call
* afs_StoreAllSegments() with AFS_LASTSTORE
*/
static int
+#if defined(FOP_FLUSH_TAKES_FL_OWNER_T)
+afs_linux_flush(struct file *fp, fl_owner_t id)
+#else
afs_linux_flush(struct file *fp)
+#endif
{
struct vrequest treq;
- struct vcache *vcp = VTOAFS(FILE_INODE(fp));
- cred_t *credp = crref();
+ struct vcache *vcp;
+ cred_t *credp;
int code;
+#if defined(AFS_CACHE_BYPASS)
+ int bypasscache;
+#endif
AFS_GLOCK();
+ if ((fp->f_flags & O_ACCMODE) == O_RDONLY) { /* readers dont flush */
+ AFS_GUNLOCK();
+ return 0;
+ }
+
+ AFS_DISCON_LOCK();
+
+ credp = crref();
+ vcp = VTOAFS(FILE_INODE(fp));
+
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;
+ else {
+ ObtainReadLock(&vcp->lock);
+ if(vcp->cachingStates & FCSBypass)
+ bypasscache = 1;
+ ReleaseReadLock(&vcp->lock);
+ }
+ if(bypasscache) {
+ /* future proof: don't rely on 0 return from afs_InitReq */
+ code = 0; goto out;
+ }
+#endif
ObtainSharedLock(&vcp->lock, 535);
- if (vcp->execsOrWriters > 0) {
+ if ((vcp->execsOrWriters > 0) && (file_count(fp) == 1)) {
UpgradeSToWLock(&vcp->lock, 536);
- code = afs_StoreAllSegments(vcp, &treq, AFS_SYNC | AFS_LASTSTORE);
+ if (!AFS_IS_DISCONNECTED) {
+ code = afs_StoreAllSegments(vcp,
+ &treq,
+ AFS_SYNC | AFS_LASTSTORE);
+ } else {
+#if defined(AFS_DISCON_ENV)
+ if (!vcp->ddirty_flags ||
+ (vcp->ddirty_flags == VDisconShadowed)) {
+
+ ObtainWriteLock(&afs_DDirtyVCListLock, 710);
+ AFS_DISCON_ADD_DIRTY(vcp);
+ ReleaseWriteLock(&afs_DDirtyVCListLock);
+ }
+
+ /* Set disconnected write flag. */
+ vcp->ddirty_flags |= VDisconWriteOsiFlush;
+#endif
+ }
+
ConvertWToSLock(&vcp->lock);
}
code = afs_CheckCode(code, &treq, 54);
ReleaseSharedLock(&vcp->lock);
out:
+ AFS_DISCON_UNLOCK();
AFS_GUNLOCK();
crfree(credp);
struct file_operations afs_file_fops = {
.read = afs_linux_read,
.write = afs_linux_write,
+#ifdef 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,
#else
.mmap = afs_linux_mmap,
.open = afs_linux_open,
.flush = afs_linux_flush,
-#ifdef AFS_LINUX26_ENV
+#if defined(AFS_LINUX26_ENV) && defined(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
.sendfile = generic_file_sendfile,
#endif
+#if defined(AFS_LINUX26_ENV) && defined(STRUCT_FILE_OPERATIONS_HAS_SPLICE)
+ .splice_write = generic_file_splice_write,
+ .splice_read = generic_file_splice_read,
+#endif
.release = afs_linux_release,
.fsync = afs_linux_fsync,
.lock = afs_linux_lock,
+#ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
+ .flock = afs_linux_flock,
+#endif
};
int code;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
credp = crref();
code = afs_getattr(vcp, &vattr, credp);
if (!code)
- vattr2inode(AFSTOV(vcp), &vattr);
+ afs_fill_inode(AFSTOV(vcp), &vattr);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
cred_t *credp = NULL;
struct vcache *vcp, *pvcp, *tvc = NULL;
int valid;
+ struct afs_fakestat_state fakestate;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
+ afs_InitFakeStat(&fakestate);
if (dp->d_inode) {
if (vcp == afs_globalVp)
goto good_dentry;
- if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
- check_bad_parent(dp); /* check and correct mvid */
+ if (vcp->mvstat == 1) { /* mount point */
+ if (vcp->mvid && (vcp->states & CMValid)) {
+ int tryEvalOnly = 0;
+ int code = 0;
+ struct vrequest treq;
+
+ credp = crref();
+ code = afs_InitReq(&treq, credp);
+ if (
+#ifdef AFS_DARWIN_ENV
+ (strcmp(dp->d_name.name, ".DS_Store") == 0) ||
+ (strcmp(dp->d_name.name, "Contents") == 0) ||
+#endif
+ (strcmp(dp->d_name.name, ".directory") == 0)) {
+ tryEvalOnly = 1;
+ }
+ if (tryEvalOnly)
+ code = afs_TryEvalFakeStat(&vcp, &fakestate, &treq);
+ else
+ code = afs_EvalFakeStat(&vcp, &fakestate, &treq);
+ if ((tryEvalOnly && vcp->mvstat == 1) || code) {
+ /* a mount point, not yet replaced by its directory */
+ goto bad_dentry;
+ }
+ }
+ } else
+ if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
+ check_bad_parent(dp); /* check and correct mvid */
#ifdef notdef
/* If the last looker changes, we should make sure the current
/* Clean up */
if (tvc)
afs_PutVCache(tvc);
+ afs_PutFakeStat(&fakestate);
AFS_GUNLOCK();
if (credp)
crfree(credp);
d_drop(dp);
}
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
return valid;
bad_dentry:
- valid = 0;
+ if (have_submounts(dp))
+ valid = 1;
+ else
+ valid = 0;
goto done;
}
struct vcache *vcp = VTOAFS(ip);
AFS_GLOCK();
- if (vcp->states & CUnlinked)
- (void) afs_InactiveVCache(vcp, NULL);
+ (void) afs_InactiveVCache(vcp, NULL);
AFS_GUNLOCK();
+#ifdef DCACHE_NFSFS_RENAMED
+#ifdef AFS_LINUX26_ENV
+ spin_lock(&dp->d_lock);
+#endif
+ dp->d_flags &= ~DCACHE_NFSFS_RENAMED;
+#ifdef AFS_LINUX26_ENV
+ spin_unlock(&dp->d_lock);
+#endif
+#endif
iput(ip);
}
vattr.va_type = mode & S_IFMT;
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
afs_getattr(vcp, &vattr, credp);
afs_fill_inode(ip, &vattr);
+ insert_inode_hash(ip);
dp->d_op = &afs_dentry_operations;
dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
d_instantiate(dp, ip);
AFS_GUNLOCK();
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
struct vcache *vcp = NULL;
const char *comp = dp->d_name.name;
struct inode *ip = NULL;
+#if defined(AFS_LINUX26_ENV)
+ struct dentry *newdp = NULL;
+#endif
int code;
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
code = afs_lookup(VTOAFS(dip), comp, &vcp, credp);
ip = AFSTOV(vcp);
afs_getattr(vcp, &vattr, credp);
afs_fill_inode(ip, &vattr);
+ if (
+#ifdef HAVE_KERNEL_HLIST_UNHASHED
+ hlist_unhashed(&ip->i_hash)
+#elif defined(AFS_LINUX26_ENV)
+ ip->i_hash.pprev == NULL
+#else
+ ip->i_hash.prev == NULL
+#endif
+ )
+ insert_inode_hash(ip);
}
dp->d_op = &afs_dentry_operations;
dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
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);
+#if defined(AFS_LINUX26_ENV)
+ /* But not if it's disconnected; then we want d_splice_alias below */
+ if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
+#else
if (alias) {
+#endif
if (d_invalidate(alias) == 0) {
dput(alias);
} else {
#if defined(AFS_LINUX26_ENV)
unlock_kernel();
#endif
+ crfree(credp);
return alias;
}
}
}
#endif
+#if defined(AFS_LINUX26_ENV)
+ newdp = d_splice_alias(ip, dp);
+#else
d_add(dp, ip);
+#endif
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
* seeing that the dp->d_inode field is NULL.
*/
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
+#if defined(AFS_LINUX26_ENV)
+ if (!code || code == ENOENT)
+ return newdp;
+#else
if (code == ENOENT)
return ERR_PTR(0);
+#endif
else
return ERR_PTR(-code);
#else
struct vcache *tvc = VTOAFS(dp->d_inode);
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
&& !(tvc->states & CUnlinked)) {
}
tvc->uncred = credp;
tvc->states |= CUnlinked;
+#ifdef DCACHE_NFSFS_RENAMED
+#ifdef AFS_LINUX26_ENV
+ spin_lock(&dp->d_lock);
+#endif
+ dp->d_flags |= DCACHE_NFSFS_RENAMED;
+#ifdef AFS_LINUX26_ENV
+ spin_unlock(&dp->d_lock);
+#endif
+#endif
+ } else {
+ osi_FreeSmallSpace(__name);
}
AFS_GUNLOCK();
d_drop(dp);
out:
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
const char *name = dp->d_name.name;
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
VATTR_NULL(&vattr);
vattr.va_mask = ATTR_MODE;
AFS_GUNLOCK();
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
#if defined(AFS_LINUX26_ENV)
/* Prevent any new references during rename operation. */
- lock_kernel();
-#endif
- /* Remove old and new entries from name hash. New one will change below.
- * While it's optimal to catch failures and re-insert newdp into hash,
- * it's also error prone and in that case we're already dealing with error
- * cases. Let another lookup put things right, if need be.
- */
-#if defined(AFS_LINUX26_ENV)
+ maybe_lock_kernel();
+
if (!d_unhashed(newdp)) {
d_drop(newdp);
rehash = newdp;
code = afs_rename(VTOAFS(oldip), oldname, VTOAFS(newip), newname, credp);
AFS_GUNLOCK();
+ if (!code)
+ olddp->d_time = 0; /* force to revalidate */
+
if (rehash)
d_rehash(rehash);
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
}
-#if !defined(AFS_LINUX24_ENV)
+#if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
/* afs_linux_readlink
* Fill target (which is in user space) with contents of symlink.
*/
/* afs_linux_follow_link
* a file system dependent link following routine.
*/
+#if defined(AFS_LINUX24_ENV)
+static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
+{
+ int code;
+ char *name;
+
+ name = osi_Alloc(PATH_MAX);
+ if (!name) {
+ return -EIO;
+ }
+
+ AFS_GLOCK();
+ code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
+ AFS_GUNLOCK();
+
+ if (code < 0) {
+ goto out;
+ }
+
+ name[code] = '\0';
+ code = vfs_follow_link(nd, name);
+
+out:
+ osi_Free(name, PATH_MAX);
+
+ return code;
+}
+
+#else /* !defined(AFS_LINUX24_ENV) */
+
static struct dentry *
afs_linux_follow_link(struct dentry *dp, struct dentry *basep,
unsigned int follow)
AFS_GUNLOCK();
return res;
}
-#endif
+#endif /* AFS_LINUX24_ENV */
+#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
+
+#if defined(AFS_CACHE_BYPASS)
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+
+/* The kernel calls readpages before trying readpage, with a list of
+ * pages. The readahead algorithm expands num_pages when it thinks
+ * the application will benefit. Unlike readpage, the pages are not
+ * necessarily allocated. If we do not a) allocate required pages and
+ * b) remove them from page_list, linux will re-enter at afs_linux_readpage
+ * for each required page (and the page will be pre-allocated) */
+
+static int
+afs_linux_readpages(struct file *fp, struct address_space *mapping,
+ struct list_head *page_list, unsigned num_pages)
+{
+ afs_int32 page_ix;
+ uio_t *auio;
+ afs_offs_t offset;
+ struct iovec* iovecp;
+ struct nocache_read_request *ancr;
+ struct page *pp, *ppt;
+ struct pagevec lrupv;
+ afs_int32 code = 0;
+
+ cred_t *credp;
+ struct inode *ip = FILE_INODE(fp);
+ struct vcache *avc = VTOAFS(ip);
+ afs_int32 bypasscache = 0; /* bypass for this read */
+ afs_int32 base_index = 0;
+ afs_int32 page_count = 0;
+ afs_int32 isize;
+
+ credp = crref();
+
+ switch(cache_bypass_strategy) {
+ case NEVER_BYPASS_CACHE:
+ break;
+ case ALWAYS_BYPASS_CACHE:
+ bypasscache = 1;
+ break;
+ case LARGE_FILES_BYPASS_CACHE:
+ if(i_size_read(ip) > cache_bypass_threshold) {
+ bypasscache = 1;
+ }
+ break;
+ default:
+ break;
+ }
+
+ /* In the new incarnation of selective caching, a file's caching policy
+ * can change, eg because file size exceeds threshold, etc. */
+ trydo_cache_transition(avc, credp, bypasscache);
+
+ if(!bypasscache) {
+ while(!list_empty(page_list)) {
+ pp = list_entry(page_list->prev, struct page, lru);
+ list_del(&pp->lru);
+ }
+ goto out;
+ }
+ /* background thread must free: iovecp, auio, ancr */
+ iovecp = osi_Alloc(num_pages * sizeof(struct iovec));
+
+ auio = osi_Alloc(sizeof(uio_t));
+ auio->uio_iov = iovecp;
+ auio->uio_iovcnt = num_pages;
+ auio->uio_flag = UIO_READ;
+ auio->uio_seg = AFS_UIOSYS;
+ auio->uio_resid = num_pages * PAGE_SIZE;
+
+ ancr = osi_Alloc(sizeof(struct nocache_read_request));
+ ancr->auio = auio;
+ ancr->offset = auio->uio_offset;
+ ancr->length = auio->uio_resid;
+
+ pagevec_init(&lrupv, 0);
+
+ for(page_ix = 0; page_ix < num_pages; ++page_ix) {
+
+ if(list_empty(page_list))
+ break;
+
+ pp = list_entry(page_list->prev, struct page, lru);
+ /* If we allocate a page and don't remove it from page_list,
+ * the page cache gets upset. */
+ list_del(&pp->lru);
+ isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
+ if(pp->index > isize) {
+ if(PageLocked(pp))
+ UnlockPage(pp);
+ continue;
+ }
+
+ if(page_ix == 0) {
+ offset = ((loff_t) pp->index) << PAGE_CACHE_SHIFT;
+ auio->uio_offset = offset;
+ base_index = pp->index;
+ }
+ iovecp[page_ix].iov_len = PAGE_SIZE;
+ code = add_to_page_cache(pp, mapping, pp->index, GFP_KERNEL);
+ if(base_index != pp->index) {
+ if(PageLocked(pp))
+ UnlockPage(pp);
+ page_cache_release(pp);
+ iovecp[page_ix].iov_base = (void *) 0;
+ base_index++;
+ continue;
+ }
+ base_index++;
+ if(code) {
+ if(PageLocked(pp))
+ UnlockPage(pp);
+ page_cache_release(pp);
+ iovecp[page_ix].iov_base = (void *) 0;
+ } else {
+ page_count++;
+ if(!PageLocked(pp)) {
+ LockPage(pp);
+ }
+
+ /* save the page for background map */
+ iovecp[page_ix].iov_base = (void*) pp;
+
+ /* and put it on the LRU cache */
+ if (!pagevec_add(&lrupv, pp))
+ __pagevec_lru_add(&lrupv);
+ }
+ }
+
+ /* If there were useful pages in the page list, make sure all pages
+ * are in the LRU cache, then schedule the read */
+ if(page_count) {
+ pagevec_lru_add(&lrupv);
+ code = afs_ReadNoCache(avc, ancr, credp);
+ } else {
+ /* 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(ancr, sizeof(struct nocache_read_request));
+ }
+ /* we do not flush, release, or unmap pages--that will be
+ * done for us by the background thread as each page comes in
+ * from the fileserver */
+ crfree(credp);
+
+out:
+ return -code;
+}
+
+#endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0) */
+#endif /* defined(AFS_CACHE_BYPASS */
+
/* afs_linux_readpage
* all reads come through here. A strategy-like read call.
static int
afs_linux_readpage(struct file *fp, struct page *pp)
{
- int code;
- cred_t *credp = crref();
+ afs_int32 code;
+ cred_t *credp = crref();
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
- char *address;
- afs_offs_t offset = ((loff_t) pp->index) << PAGE_CACHE_SHIFT;
+ char *address;
+ afs_offs_t offset = ((loff_t) pp->index) << PAGE_CACHE_SHIFT;
#else
- ulong address = afs_linux_page_address(pp);
- afs_offs_t offset = pageoff(pp);
+ ulong address = afs_linux_page_address(pp);
+ afs_offs_t offset = pageoff(pp);
#endif
- uio_t tuio;
- struct iovec iovec;
- struct inode *ip = FILE_INODE(fp);
- int cnt = page_count(pp);
- struct vcache *avc = VTOAFS(ip);
-
+#if defined(AFS_CACHE_BYPASS)
+ afs_int32 bypasscache = 0; /* bypass for this read */
+ struct nocache_read_request *ancr;
+#endif
+ afs_int32 isize;
+ uio_t *auio;
+ struct iovec *iovecp;
+ struct inode *ip = FILE_INODE(fp);
+ afs_int32 cnt = page_count(pp);
+ struct vcache *avc = VTOAFS(ip);
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
- address = kmap(pp);
- ClearPageError(pp);
+ address = kmap(pp);
+ ClearPageError(pp);
#else
- atomic_add(1, &pp->count);
- set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
- clear_bit(PG_error, &pp->flags);
-#endif
-
- setup_uio(&tuio, &iovec, (char *)address, offset, PAGESIZE, UIO_READ,
- AFS_UIOSYS);
+ atomic_add(1, &pp->count);
+ set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
+ clear_bit(PG_error, &pp->flags);
+#endif
+#if defined(AFS_CACHE_BYPASS)
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+ /* If the page is past the end of the file, skip it */
+ isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
+ if(pp->index > isize) {
+ if(PageLocked(pp))
+ UnlockPage(pp);
+ goto done;
+ }
+#endif
+#endif
+ /* if bypasscache, receiver frees, else we do */
+ auio = osi_Alloc(sizeof(uio_t));
+ iovecp = osi_Alloc(sizeof(struct iovec));
+
+ setup_uio(auio, iovecp, (char *)address, offset, PAGE_SIZE, UIO_READ,
+ AFS_UIOSYS);
+
+#if defined(AFS_CACHE_BYPASS)
+
+ switch(cache_bypass_strategy) {
+ case NEVER_BYPASS_CACHE:
+ break;
+ case ALWAYS_BYPASS_CACHE:
+ bypasscache = 1;
+ break;
+ case LARGE_FILES_BYPASS_CACHE:
+ if(i_size_read(ip) > cache_bypass_threshold) {
+ bypasscache = 1;
+ }
+ break;
+ default:
+ break;
+ }
+
+ /* In the new incarnation of selective caching, a file's caching policy
+ * can change, eg because file size exceeds threshold, etc. */
+ trydo_cache_transition(avc, credp, bypasscache);
+
+ if(bypasscache) {
+ if(address)
+ kunmap(pp);
+ /* save the page for background map */
+ auio->uio_iov->iov_base = (void*) pp;
+ /* the background thread will free this */
+ ancr = osi_Alloc(sizeof(struct nocache_read_request));
+ ancr->auio = auio;
+ ancr->offset = offset;
+ ancr->length = PAGE_SIZE;
+
+ maybe_lock_kernel();
+ code = afs_ReadNoCache(avc, ancr, credp);
+ maybe_unlock_kernel();
+
+ goto done; /* skips release page, doing it in bg thread */
+ }
+#endif
+
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
- AFS_GLOCK();
- afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip, ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32, 99999); /* not a possible code value */
- code = afs_rdwr(avc, &tuio, UIO_READ, 0, credp);
- afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
- ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
- code);
- AFS_GUNLOCK();
+ AFS_GLOCK();
+ afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip, ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32, 99999); /* not a possible code value */
+
+ code = afs_rdwr(avc, auio, UIO_READ, 0, credp);
+
+ afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
+ ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
+ code);
+ AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
+ if (!code) {
+ /* XXX valid for no-cache also? Check last bits of files... :)
+ * Cognate code goes in afs_NoCacheFetchProc. */
+ if (auio->uio_resid) /* zero remainder of page */
+ memset((void *)(address + (PAGE_SIZE - auio->uio_resid)), 0,
+ auio->uio_resid);
- if (!code) {
- if (tuio.uio_resid) /* zero remainder of page */
- memset((void *)(address + (PAGESIZE - tuio.uio_resid)), 0,
- tuio.uio_resid);
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
- flush_dcache_page(pp);
- SetPageUptodate(pp);
+ flush_dcache_page(pp);
+ SetPageUptodate(pp);
#else
- set_bit(PG_uptodate, &pp->flags);
+ set_bit(PG_uptodate, &pp->flags);
#endif
- }
+ } /* !code */
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
- kunmap(pp);
- UnlockPage(pp);
+ kunmap(pp);
+ UnlockPage(pp);
#else
- clear_bit(PG_locked, &pp->flags);
- wake_up(&pp->wait);
- free_page(address);
+ clear_bit(PG_locked, &pp->flags);
+ wake_up(&pp->wait);
+ free_page(address);
#endif
- if (!code && AFS_CHUNKOFFSET(offset) == 0) {
- struct dcache *tdc;
- struct vrequest treq;
+#if defined(AFS_CACHE_BYPASS)
- AFS_GLOCK();
- code = afs_InitReq(&treq, credp);
- if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
- tdc = afs_FindDCache(avc, offset);
- if (tdc) {
- if (!(tdc->mflags & DFNextStarted))
- afs_PrefetchChunk(avc, tdc, credp, &treq);
- afs_PutDCache(tdc);
- }
- ReleaseWriteLock(&avc->lock);
- }
- AFS_GUNLOCK();
- }
+/* do not call afs_GetDCache if cache is bypassed */
+ if(bypasscache)
+ goto done;
+
+#endif
- crfree(credp);
- return -code;
+ /* free if not bypassing cache */
+ osi_Free(auio, sizeof(uio_t));
+ osi_Free(iovecp, sizeof(struct iovec));
+
+ if (!code && AFS_CHUNKOFFSET(offset) == 0) {
+ struct dcache *tdc;
+ struct vrequest treq;
+
+ AFS_GLOCK();
+ code = afs_InitReq(&treq, credp);
+ if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
+ tdc = afs_FindDCache(avc, offset);
+ if (tdc) {
+ if (!(tdc->mflags & DFNextStarted))
+ afs_PrefetchChunk(avc, tdc, credp, &treq);
+ afs_PutDCache(tdc);
+ }
+ ReleaseWriteLock(&avc->lock);
+ }
+ AFS_GUNLOCK();
+ }
+
+done:
+ crfree(credp);
+ return -code;
}
base = (((loff_t) pp->index) << PAGE_CACHE_SHIFT) + offset;
credp = crref();
- lock_kernel();
+ maybe_lock_kernel();
AFS_GLOCK();
afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
ICL_TYPE_INT32, 99999);
+ ObtainReadLock(&vcp->lock);
+ if (vcp->states & CPageWrite) {
+ ReleaseReadLock(&vcp->lock);
+ AFS_GUNLOCK();
+ maybe_unlock_kernel();
+ crfree(credp);
+ kunmap(pp);
+#ifdef AFS_LINUX26_ENV
+#if defined(WRITEPAGE_ACTIVATE)
+ return WRITEPAGE_ACTIVATE;
+#else
+ return AOP_WRITEPAGE_ACTIVATE;
+#endif
+#else
+ /* should mark it dirty? */
+ return(0);
+#endif
+ }
+ ReleaseReadLock(&vcp->lock);
+
setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
code = afs_write(vcp, &tuio, f_flags, credp, 0);
ICL_TYPE_INT32, code);
AFS_GUNLOCK();
- unlock_kernel();
+ maybe_unlock_kernel();
crfree(credp);
kunmap(pp);
.setattr = afs_notify_change,
#else
.default_file_ops = &afs_file_fops,
- .readpage = afs_linux_readpage,
+ .readpage = afs_linux_readpage,
.revalidate = afs_linux_revalidate,
.updatepage = afs_linux_updatepage,
#endif
#if defined(AFS_LINUX24_ENV)
static struct address_space_operations afs_file_aops = {
.readpage = afs_linux_readpage,
+#if defined(AFS_CACHE_BYPASS) && LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+ .readpages = afs_linux_readpages,
+#endif
.writepage = afs_linux_writepage,
.commit_write = afs_linux_commit_write,
.prepare_write = afs_linux_prepare_write,
/* We really need a separate symlink set of ops, since do_follow_link()
* determines if it _is_ a link by checking if the follow_link op is set.
*/
-#if defined(AFS_LINUX24_ENV)
+#if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
static int
afs_symlink_filler(struct file *file, struct page *page)
{
char *p = (char *)kmap(page);
int code;
- lock_kernel();
+ maybe_lock_kernel();
AFS_GLOCK();
code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
AFS_GUNLOCK();
if (code < 0)
goto fail;
p[code] = '\0'; /* null terminate? */
- unlock_kernel();
+ maybe_unlock_kernel();
SetPageUptodate(page);
kunmap(page);
return 0;
fail:
- unlock_kernel();
+ maybe_unlock_kernel();
SetPageError(page);
kunmap(page);
static struct address_space_operations afs_symlink_aops = {
.readpage = afs_symlink_filler
};
-#endif
+#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
static struct inode_operations afs_symlink_iops = {
-#if defined(AFS_LINUX24_ENV)
+#if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
.readlink = page_readlink,
#if defined(HAVE_KERNEL_PAGE_FOLLOW_LINK)
.follow_link = page_follow_link,
.follow_link = page_follow_link_light,
.put_link = page_put_link,
#endif
- .setattr = afs_notify_change,
-#else
+#else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
.readlink = afs_linux_readlink,
.follow_link = afs_linux_follow_link,
+#if !defined(AFS_LINUX24_ENV)
.permission = afs_linux_permission,
.revalidate = afs_linux_revalidate,
#endif
+#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
+#if defined(AFS_LINUX24_ENV)
+ .setattr = afs_notify_change,
+#endif
};
void
} else if (S_ISLNK(ip->i_mode)) {
ip->i_op = &afs_symlink_iops;
-#if defined(AFS_LINUX24_ENV)
+#if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
ip->i_data.a_ops = &afs_symlink_aops;
ip->i_mapping = &ip->i_data;
#endif
}
- /* insert_inode_hash(ip); -- this would make iget() work (if we used it) */
}