#include "afs/sysincludes.h"
#include "afsincludes.h"
#include "afs/afs_stats.h"
-#include "afs/afs_osidnlc.h"
#include "h/mm.h"
#ifdef HAVE_MM_INLINE_H
#include "h/mm_inline.h"
#if defined(AFS_LINUX24_ENV)
#include "h/smp_lock.h"
#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
#define pageoff(pp) pgoff2loff((pp)->index)
#endif
#if defined(AFS_LINUX26_ENV)
+#define LockPage(pp) lock_page(pp)
#define UnlockPage(pp) unlock_page(pp)
+extern struct backing_dev_info afs_backing_dev_info;
#endif
extern struct vcache *afs_globalVp;
-extern afs_rwlock_t afs_xvcache;
-
-#if defined(AFS_LINUX24_ENV)
-extern struct inode_operations afs_file_iops;
-extern struct address_space_operations afs_file_aops;
-struct address_space_operations afs_symlink_aops;
-#endif
-extern struct inode_operations afs_dir_iops;
-extern struct inode_operations afs_symlink_iops;
-
+#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)
{
ssize_t code;
- struct vcache *vcp = ITOAFS(fp->f_dentry->d_inode);
+ 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 {
-#ifdef AFS_64BIT_CLIENT
- if (*offp + count > afs_vmMappingEnd) {
- uio_t tuio;
- struct iovec iov;
- afs_size_t oldOffset = *offp;
- afs_int32 xfered = 0;
-
- if (*offp < afs_vmMappingEnd) {
- /* special case of a buffer crossing the VM mapping end */
- afs_int32 tcount = afs_vmMappingEnd - *offp;
- count -= tcount;
- osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
- AFS_GUNLOCK();
- code = generic_file_read(fp, buf, tcount, offp);
- AFS_GLOCK();
- if (code != tcount) {
- goto done;
- }
- xfered = tcount;
- }
- setup_uio(&tuio, &iov, buf + xfered, (afs_offs_t) * offp, count,
- UIO_READ, AFS_UIOSYS);
- code = afs_read(vcp, &tuio, credp, 0, 0, 0);
- xfered += count - tuio.uio_resid;
- if (code != 0) {
- afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER,
- vcp, ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, -1,
- ICL_TYPE_INT32, code);
- code = xfered;
- *offp += count - tuio.uio_resid;
- } else {
- code = xfered;
- *offp += count;
- }
- done:
- ;
- } else {
-#endif /* AFS_64BIT_CLIENT */
- osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
- AFS_GUNLOCK();
- code = generic_file_read(fp, buf, count, offp);
- AFS_GLOCK();
-#ifdef AFS_64BIT_CLIENT
- }
-#endif /* AFS_64BIT_CLIENT */
+#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;
- struct vcache *vcp = ITOAFS(fp->f_dentry->d_inode);
+ struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
struct vrequest treq;
cred_t *credp = crref();
- afs_offs_t toffs;
AFS_GLOCK();
if (code)
code = -code;
else {
-#ifdef AFS_64BIT_CLIENT
- toffs = *offp;
- if (fp->f_flags & O_APPEND)
- toffs += vcp->m.Length;
- if (toffs + count > afs_vmMappingEnd) {
- uio_t tuio;
- struct iovec iov;
- afs_size_t oldOffset = *offp;
- afs_int32 xfered = 0;
-
- if (toffs < afs_vmMappingEnd) {
- /* special case of a buffer crossing the VM mapping end */
- afs_int32 tcount = afs_vmMappingEnd - *offp;
- count -= tcount;
- AFS_GUNLOCK();
- code = generic_file_write(fp, buf, tcount, offp);
- AFS_GLOCK();
- if (code != tcount) {
- goto done;
- }
- xfered = tcount;
- toffs += tcount;
- }
- setup_uio(&tuio, &iov, buf + xfered, (afs_offs_t) toffs, count,
- UIO_WRITE, AFS_UIOSYS);
- code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
- xfered += count - tuio.uio_resid;
- if (code != 0) {
- code = xfered;
- *offp += count - tuio.uio_resid;
- } else {
- /* Purge dirty chunks of file if there are too many dirty chunks.
- * Inside the write loop, we only do this at a chunk boundary.
- * Clean up partial chunk if necessary at end of loop.
- */
- if (AFS_CHUNKBASE(tuio.afsio_offset) !=
- AFS_CHUNKBASE(oldOffset)) {
- ObtainWriteLock(&vcp->lock, 402);
- code = afs_DoPartialWrite(vcp, &treq);
- vcp->states |= CDirty;
- ReleaseWriteLock(&vcp->lock);
- }
- code = xfered;
- *offp += count;
- toffs += count;
- ObtainWriteLock(&vcp->lock, 400);
- vcp->m.Date = osi_Time(); /* Set file date (for ranlib) */
- /* extend file */
- if (!(fp->f_flags & O_APPEND) && toffs > vcp->m.Length) {
- vcp->m.Length = toffs;
- }
- ReleaseWriteLock(&vcp->lock);
- }
- done:
- ;
- } else {
-#endif /* AFS_64BIT_CLIENT */
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();
-#ifdef AFS_64BIT_CLIENT
- }
-#endif /* AFS_64BIT_CLIENT */
}
ObtainWriteLock(&vcp->lock, 530);
- vcp->m.Date = osi_Time(); /* set modification time */
afs_FakeClose(vcp, credp);
- if (code >= 0)
- code2 = afs_DoPartialWrite(vcp, &treq);
- if (code2 && code >= 0)
- code = (ssize_t) - code2;
ReleaseWriteLock(&vcp->lock);
afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
return code;
}
+extern int BlobScan(struct dcache * afile, afs_int32 ablob);
+
/* This is a complete rewrite of afs_readdir, since we can make use of
* filldir instead of afs_readdir_move. Note that changes to vcache/dcache
* handling and use of bulkstats will need to be reflected here as well.
afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
{
extern struct DirEntry *afs_dir_GetBlob();
- struct vcache *avc = ITOAFS(FILE_INODE(fp));
+ struct vcache *avc = VTOAFS(FILE_INODE(fp));
struct vrequest treq;
register struct dcache *tdc;
int code;
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.
*/
if (!de)
break;
- ino = (avc->fid.Fid.Volume << 16) + ntohl(de->fid.vnode);
- ino &= 0x7fffffff; /* Assumes 32 bit ino_t ..... */
+ ino = afs_calc_inum (avc->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 %x, dirpos %d)\n",
- &tdc->f.inode, dirpos);
+ 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;
}
extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
unsigned long arg);
+#if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
+static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
+ unsigned long arg) {
+ return afs_xioctl(FILE_INODE(fp), fp, com, arg);
-/* We need to detect unmap's after close. To do that, we need our own
- * vm_operations_struct's. And we need to set them up for both the
- * private and shared mappings. The fun part is that these are all static
- * so we'll have to initialize on the fly!
- */
-static struct vm_operations_struct afs_private_mmap_ops;
-static int afs_private_mmap_ops_inited = 0;
-static struct vm_operations_struct afs_shared_mmap_ops;
-static int afs_shared_mmap_ops_inited = 0;
-
-void
-afs_linux_vma_close(struct vm_area_struct *vmap)
-{
- struct vcache *vcp;
- cred_t *credp;
- int need_unlock = 0;
-
- if (!vmap->vm_file)
- return;
-
- vcp = ITOAFS(FILE_INODE(vmap->vm_file));
- if (!vcp)
- return;
-
- AFS_GLOCK();
- afs_Trace4(afs_iclSetp, CM_TRACE_VM_CLOSE, ICL_TYPE_POINTER, vcp,
- ICL_TYPE_INT32, vcp->mapcnt, ICL_TYPE_INT32, vcp->opens,
- ICL_TYPE_INT32, vcp->execsOrWriters);
- if ((&vcp->lock)->excl_locked == 0 || (&vcp->lock)->pid_writer == MyPidxx) {
- ObtainWriteLock(&vcp->lock, 532);
- need_unlock = 1;
- } else
- printk("AFS_VMA_CLOSE(%d): Skipping Already locked vcp=%p vmap=%p\n",
- MyPidxx, &vcp, &vmap);
- if (vcp->mapcnt) {
- vcp->mapcnt--;
- if (need_unlock)
- ReleaseWriteLock(&vcp->lock);
- if (!vcp->mapcnt) {
- if (need_unlock && vcp->execsOrWriters < 2) {
- credp = crref();
- (void)afs_close(vcp, vmap->vm_file->f_flags, credp);
- /* only decrement the execsOrWriters flag if this is not a
- * writable file. */
- if (!(vcp->states & CRO) )
- if (! (vmap->vm_file->f_flags & (FWRITE | FTRUNC)))
- vcp->execsOrWriters--;
- vcp->states &= ~CMAPPED;
- crfree(credp);
- } else if ((vmap->vm_file->f_flags & (FWRITE | FTRUNC)))
- vcp->execsOrWriters--;
- /* If we did not have the lock */
- if (!need_unlock) {
- vcp->mapcnt++;
- if (!vcp->execsOrWriters)
- vcp->execsOrWriters = 1;
- }
- }
- } else {
- if (need_unlock)
- ReleaseWriteLock(&vcp->lock);
- }
-
- unlock_exit:
- AFS_GUNLOCK();
}
+#endif
+
static int
afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
{
- struct vcache *vcp = ITOAFS(FILE_INODE(fp));
+ struct vcache *vcp = VTOAFS(FILE_INODE(fp));
cred_t *credp = crref();
struct vrequest treq;
int code;
/* get a validated vcache entry */
code = afs_InitReq(&treq, credp);
- if (!code)
- code = afs_VerifyVCache(vcp, &treq);
-
-
if (code)
- code = -code;
- else {
- osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
-
- AFS_GUNLOCK();
- code = generic_file_mmap(fp, vmap);
- AFS_GLOCK();
- }
+ goto out_err;
- if (code == 0) {
- ObtainWriteLock(&vcp->lock, 531);
- /* Set out vma ops so we catch the close. The following test should be
- * the same as used in generic_file_mmap.
- */
- if ((vmap->vm_flags & VM_SHARED) && (vmap->vm_flags & VM_MAYWRITE)) {
- if (!afs_shared_mmap_ops_inited) {
- afs_shared_mmap_ops_inited = 1;
- afs_shared_mmap_ops = *vmap->vm_ops;
- afs_shared_mmap_ops.close = afs_linux_vma_close;
- }
- vmap->vm_ops = &afs_shared_mmap_ops;
- } else {
- if (!afs_private_mmap_ops_inited) {
- afs_private_mmap_ops_inited = 1;
- afs_private_mmap_ops = *vmap->vm_ops;
- afs_private_mmap_ops.close = afs_linux_vma_close;
- }
- vmap->vm_ops = &afs_private_mmap_ops;
- }
+ code = afs_VerifyVCache(vcp, &treq);
+ if (code)
+ goto out_err;
+ osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
- /* Add an open reference on the first mapping. */
- if (vcp->mapcnt == 0) {
- if (!(vcp->states & CRO))
- vcp->execsOrWriters++;
- vcp->opens++;
- vcp->states |= CMAPPED;
- }
- ReleaseWriteLock(&vcp->lock);
- vcp->mapcnt++;
- }
+ AFS_GUNLOCK();
+ code = generic_file_mmap(fp, vmap);
+ AFS_GLOCK();
+ if (!code)
+ vcp->states |= CMAPPED;
+out:
AFS_GUNLOCK();
crfree(credp);
return code;
+
+out_err:
+ code = -code;
+ goto out;
}
-int
+static int
afs_linux_open(struct inode *ip, struct file *fp)
{
- int code;
+ struct vcache *vcp = VTOAFS(ip);
cred_t *credp = crref();
+ int code;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
- code = afs_open((struct vcache **)&ip, fp->f_flags, credp);
+ code = afs_open(&vcp, fp->f_flags, credp);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
}
-/* afs_Close is called from release, since release is used to handle all
- * file closings. In addition afs_linux_flush is called from sys_close to
- * handle flushing the data back to the server. The kicker is that we could
- * ignore flush completely if only sys_close took it's return value from
- * fput. See afs_linux_flush for notes on interactions between release and
- * flush.
- */
static int
afs_linux_release(struct inode *ip, struct file *fp)
{
- int code = 0;
+ struct vcache *vcp = VTOAFS(ip);
cred_t *credp = crref();
- struct vcache *vcp = ITOAFS(ip);
+ int code = 0;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
- if (vcp->flushcnt) {
- vcp->flushcnt--; /* protected by AFS global lock. */
- } else {
- code = afs_close(vcp, fp->f_flags, credp);
- }
+ code = afs_close(vcp, fp->f_flags, credp);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
}
-#if defined(AFS_LINUX24_ENV)
static int
+#if defined(AFS_LINUX24_ENV)
afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
#else
-static int
afs_linux_fsync(struct file *fp, struct dentry *dp)
#endif
{
cred_t *credp = crref();
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
- code = afs_fsync(ITOAFS(ip), credp);
+ code = afs_fsync(VTOAFS(ip), credp);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
{
int code = 0;
- struct vcache *vcp = ITOAFS(FILE_INODE(fp));
+ struct vcache *vcp = VTOAFS(FILE_INODE(fp));
cred_t *credp = crref();
struct AFS_FLOCK flock;
/* Convert to a lock format afs_lockctl understands. */
code = afs_lockctl(vcp, &flock, cmd, credp);
AFS_GUNLOCK();
+#ifdef AFS_LINUX24_ENV
+ 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 */
flp->fl_type = flock.l_type;
flp->fl_pid = flock.l_pid;
}
+#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
- * flush is called from sys_close. We could ignore it, but sys_close return
- * code comes from flush, not release. We need to use release to keep
- * the vcache open count correct. Note that flush is called before release
- * (via fput) in sys_close. vcp->flushcnt is a bit of ugliness to avoid
- * races and also avoid calling afs_close twice when closing the file.
- * If we merely checked for opens > 0 in afs_linux_release, then if an
- * new open occurred when storing back the file, afs_linux_release would
- * incorrectly close the file and decrement the opens count. Calling afs_close
- * on the just flushed file is wasteful, since the background daemon will
- * execute the code that finally decides there is nothing to do.
+ * 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
*/
-int
+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 vcache *vcp = ITOAFS(FILE_INODE(fp));
- int code = 0;
+ struct vrequest treq;
+ struct vcache *vcp;
cred_t *credp;
-
- /* Only do this on the last close of the file pointer. */
-#if defined(AFS_LINUX24_ENV)
- if (atomic_read(&fp->f_count) > 1)
-#else
- if (fp->f_count > 1)
+ 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));
- AFS_GLOCK();
- code = afs_close(vcp, fp->f_flags, credp);
- vcp->flushcnt++; /* protected by AFS global lock. */
+ 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) && (file_count(fp) == 1)) {
+ UpgradeSToWLock(&vcp->lock, 536);
+ 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);
.read = generic_read_dir,
#endif
.readdir = afs_linux_readdir,
+#ifdef HAVE_UNLOCKED_IOCTL
+ .unlocked_ioctl = afs_unlocked_xioctl,
+#else
.ioctl = afs_xioctl,
+#endif
+#ifdef HAVE_COMPAT_IOCTL
+ .compat_ioctl = afs_unlocked_xioctl,
+#endif
.open = afs_linux_open,
.release = afs_linux_release,
};
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
.ioctl = afs_xioctl,
+#endif
+#ifdef HAVE_COMPAT_IOCTL
+ .compat_ioctl = afs_unlocked_xioctl,
+#endif
.mmap = afs_linux_mmap,
.open = afs_linux_open,
.flush = afs_linux_flush,
+#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
};
* AFS Linux dentry operations
**********************************************************************/
+/* check_bad_parent() : Checks 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()
+ * 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.
+ * Return Values:
+ * None.
+ * Sideeffects:
+ * This dentry's vcache's mvid will be set to the correct parent directory's
+ * fid.
+ * This root vnode's volume will have its dotdotfid and mtpoint fids set
+ * to the correct parent and mountpoint fids.
+ */
+
+static inline void
+check_bad_parent(struct dentry *dp)
+{
+ cred_t *credp;
+ struct vcache *vcp = VTOAFS(dp->d_inode), *avc = NULL;
+ struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
+
+ if (vcp->mvid->Fid.Volume != pvc->fid.Fid.Volume) { /* bad parent */
+ credp = crref();
+
+ /* force a lookup, so vcp->mvid is fixed up */
+ afs_lookup(pvc, 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);
+ }
+
+ return;
+}
+
/* afs_linux_revalidate
* Ensure vcache is stat'd before use. Return 0 if entry is valid.
*/
static int
afs_linux_revalidate(struct dentry *dp)
{
- int code;
+ struct vattr vattr;
+ struct vcache *vcp = VTOAFS(dp->d_inode);
cred_t *credp;
- struct vrequest treq;
- struct vcache *vcp = ITOAFS(dp->d_inode);
- struct vcache *rootvp = NULL;
+ int code;
#ifdef AFS_LINUX24_ENV
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
- if (afs_fakestat_enable && vcp->mvstat == 1 && vcp->mvid
- && (vcp->states & CMValid) && (vcp->states & CStatd)) {
- ObtainSharedLock(&afs_xvcache, 680);
- rootvp = afs_FindVCache(vcp->mvid, 0, 0);
- ReleaseSharedLock(&afs_xvcache);
- }
-
+#ifdef notyet
/* Make this a fast path (no crref), since it's called so often. */
if (vcp->states & CStatd) {
+
if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
check_bad_parent(dp); /* check and correct mvid */
- if (rootvp)
- vcache2fakeinode(rootvp, vcp);
- else
- vcache2inode(vcp);
- if (rootvp)
- afs_PutVCache(rootvp);
+
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
unlock_kernel();
#endif
return 0;
}
+#endif
credp = crref();
- code = afs_InitReq(&treq, credp);
+ code = afs_getattr(vcp, &vattr, credp);
if (!code)
- code = afs_VerifyVCache(vcp, &treq);
+ afs_fill_inode(AFSTOV(vcp), &vattr);
AFS_GUNLOCK();
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
{
int err = afs_linux_revalidate(dentry);
- if (!err)
+ if (!err) {
generic_fillattr(dentry->d_inode, stat);
+}
return err;
}
#endif
* 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..
*/
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
static int
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
+#ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
+afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
+#else
afs_linux_dentry_revalidate(struct dentry *dp, int flags)
+#endif
#else
-static int
afs_linux_dentry_revalidate(struct dentry *dp)
#endif
{
- char *name;
- cred_t *credp = crref();
- struct vrequest treq;
- struct vcache *lookupvcp = NULL;
- int code, bad_dentry = 1;
- struct sysname_info sysState;
- struct vcache *vcp, *parentvcp;
-
- sysState.allocked = 0;
+ struct vattr vattr;
+ 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);
- vcp = ITOAFS(dp->d_inode);
- parentvcp = ITOAFS(dp->d_parent->d_inode);
+ if (dp->d_inode) {
- /* If it's a negative dentry, then there's nothing to do. */
- if (!vcp || !parentvcp)
- goto done;
+ vcp = VTOAFS(dp->d_inode);
+ pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
- /* If it is the AFS root, then there's no chance it needs
- * revalidating */
- if (vcp == afs_globalVp) {
- bad_dentry = 0;
- goto done;
- }
+ if (vcp == afs_globalVp)
+ goto good_dentry;
- if ((code = afs_InitReq(&treq, credp)))
- goto done;
-
- Check_AtSys(parentvcp, dp->d_name.name, &sysState, &treq);
- name = sysState.name;
-
- /* First try looking up the DNLC */
- if ((lookupvcp = osi_dnlc_lookup(parentvcp, name, WRITE_LOCK))) {
- /* Verify that the dentry does not point to an old inode */
- if (vcp != lookupvcp)
- goto done;
- /* Check and correct mvid */
- if (*name != '/' && vcp->mvstat == 2)
- check_bad_parent(dp);
- vcache2inode(vcp);
- bad_dentry = 0;
- goto done;
- }
+ 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
+ * looker still has permission to examine this file. This would
+ * 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))
+ goto bad_dentry;
+
+ vcp->last_looker = treq.uid;
+ }
+#endif
+
+ /* 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->m.DataVersion) > dp->d_time || !(vcp->states & CStatd)) {
+
+ credp = crref();
+ afs_lookup(pvcp, dp->d_name.name, &tvc, credp);
+ if (!tvc || tvc != vcp)
+ goto bad_dentry;
+
+ if (afs_getattr(vcp, &vattr, credp))
+ goto bad_dentry;
+
+ vattr2inode(AFSTOV(vcp), &vattr);
+ dp->d_time = hgetlo(pvcp->m.DataVersion);
+ }
+
+ /* should we always update the attributes at this point? */
+ /* unlikely--the vcache entry hasn't changed */
+
+ } else {
+#ifdef notyet
+ pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
+ if (hgetlo(pvcp->m.DataVersion) > dp->d_time)
+ goto bad_dentry;
+#endif
- /* A DNLC lookup failure cannot be trusted. Try a real lookup.
- Make sure to try the real name and not the @sys expansion;
- afs_lookup will expand @sys itself. */
-
- code = afs_lookup(parentvcp, dp->d_name.name, &lookupvcp, credp);
+ /* No change in parent's DataVersion so this negative
+ * lookup is still valid. BUT, if a server is down a
+ * negative lookup can result so there should be a
+ * liftime as well. For now, always expire.
+ */
- /* Verify that the dentry does not point to an old inode */
- if (vcp != lookupvcp)
- goto done;
+ goto bad_dentry;
+ }
- bad_dentry = 0;
+ good_dentry:
+ valid = 1;
done:
/* Clean up */
- if (lookupvcp)
- afs_PutVCache(lookupvcp);
- if (sysState.allocked)
- osi_FreeLargeSpace(name);
-
+ if (tvc)
+ afs_PutVCache(tvc);
+ afs_PutFakeStat(&fakestate);
AFS_GUNLOCK();
+ if (credp)
+ crfree(credp);
- if (bad_dentry) {
+ if (!valid) {
shrink_dcache_parent(dp);
d_drop(dp);
}
-
#ifdef AFS_LINUX24_ENV
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
- crfree(credp);
+ return valid;
- return !bad_dentry;
+ bad_dentry:
+ if (have_submounts(dp))
+ valid = 1;
+ else
+ valid = 0;
+ goto done;
}
-#if !defined(AFS_LINUX26_ENV)
-/* afs_dentry_iput */
static void
afs_dentry_iput(struct dentry *dp, struct inode *ip)
{
- int isglock;
-
- if (ICL_SETACTIVE(afs_iclSetp)) {
- isglock = ISAFS_GLOCK();
- if (!isglock) AFS_GLOCK();
- afs_Trace3(afs_iclSetp, CM_TRACE_DENTRYIPUT, ICL_TYPE_POINTER, ip,
- ICL_TYPE_STRING, dp->d_parent->d_name.name,
- ICL_TYPE_STRING, dp->d_name.name);
- if (!isglock) AFS_GUNLOCK();
- }
+ struct vcache *vcp = VTOAFS(ip);
- osi_iput(ip);
-}
+ AFS_GLOCK();
+ (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);
+}
static int
afs_dentry_delete(struct dentry *dp)
{
- int isglock;
- if (ICL_SETACTIVE(afs_iclSetp)) {
- isglock = ISAFS_GLOCK();
- if (!isglock) AFS_GLOCK();
- afs_Trace3(afs_iclSetp, CM_TRACE_DENTRYDELETE, ICL_TYPE_POINTER,
- dp->d_inode, ICL_TYPE_STRING, dp->d_parent->d_name.name,
- ICL_TYPE_STRING, dp->d_name.name);
- if (!isglock) AFS_GUNLOCK();
- }
-
- if (dp->d_inode && (ITOAFS(dp->d_inode)->states & CUnlinked))
+ if (dp->d_inode && (VTOAFS(dp->d_inode)->states & CUnlinked))
return 1; /* bad inode? */
return 0;
struct dentry_operations afs_dentry_operations = {
.d_revalidate = afs_linux_dentry_revalidate,
.d_delete = afs_dentry_delete,
-#if !defined(AFS_LINUX26_ENV)
.d_iput = afs_dentry_iput,
-#endif
};
/**********************************************************************
*
* name is in kernel space at this point.
*/
-int
+static int
+#ifdef 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
{
- int code;
- cred_t *credp = crref();
struct vattr vattr;
- enum vcexcl excl;
+ cred_t *credp = crref();
const char *name = dp->d_name.name;
- struct inode *ip;
+ struct vcache *vcp;
+ int code;
VATTR_NULL(&vattr);
vattr.va_mode = mode;
+ vattr.va_type = mode & S_IFMT;
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
AFS_GLOCK();
- code =
- afs_create(ITOAFS(dip), name, &vattr, NONEXCL, mode,
- (struct vcache **)&ip, credp);
+ code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
+ &vcp, credp);
if (!code) {
- vattr2inode(ip, &vattr);
- /* Reset ops if symlink or directory. */
-#if defined(AFS_LINUX24_ENV)
- if (S_ISREG(ip->i_mode)) {
- ip->i_op = &afs_file_iops;
- ip->i_fop = &afs_file_fops;
- ip->i_data.a_ops = &afs_file_aops;
- } else if (S_ISDIR(ip->i_mode)) {
- ip->i_op = &afs_dir_iops;
- ip->i_fop = &afs_dir_fops;
- } else if (S_ISLNK(ip->i_mode)) {
- ip->i_op = &afs_symlink_iops;
- ip->i_data.a_ops = &afs_symlink_aops;
- ip->i_mapping = &ip->i_data;
- } else
- printk("afs_linux_create: FIXME\n");
-#else
- if (S_ISDIR(ip->i_mode))
- ip->i_op = &afs_dir_iops;
- else if (S_ISLNK(ip->i_mode))
- ip->i_op = &afs_symlink_iops;
-#endif
+ struct inode *ip = AFSTOV(vcp);
+ afs_getattr(vcp, &vattr, credp);
+ afs_fill_inode(ip, &vattr);
+ insert_inode_hash(ip);
dp->d_op = &afs_dentry_operations;
- dp->d_time = jiffies;
+ 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;
/* afs_linux_lookup */
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
-struct dentry *
+static struct dentry *
+#ifdef IOP_LOOKUP_TAKES_NAMEIDATA
+afs_linux_lookup(struct inode *dip, struct dentry *dp,
+ struct nameidata *nd)
+#else
afs_linux_lookup(struct inode *dip, struct dentry *dp)
+#endif
#else
-int
+static int
afs_linux_lookup(struct inode *dip, struct dentry *dp)
#endif
{
- int code = 0;
cred_t *credp = crref();
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(ITOAFS(dip), comp, &vcp, credp);
- AFS_GUNLOCK();
+ code = afs_lookup(VTOAFS(dip), comp, &vcp, credp);
if (vcp) {
- struct inode *ip = AFSTOI(vcp);
- /* Reset ops if symlink or directory. */
-#if defined(AFS_LINUX24_ENV)
- if (S_ISREG(ip->i_mode)) {
- ip->i_op = &afs_file_iops;
- ip->i_fop = &afs_file_fops;
- ip->i_data.a_ops = &afs_file_aops;
- } else if (S_ISDIR(ip->i_mode)) {
- ip->i_op = &afs_dir_iops;
- ip->i_fop = &afs_dir_fops;
- } else if (S_ISLNK(ip->i_mode)) {
- ip->i_op = &afs_symlink_iops;
- ip->i_data.a_ops = &afs_symlink_aops;
- ip->i_mapping = &ip->i_data;
- } else
- printk
- ("afs_linux_lookup: ip->i_mode 0x%x dp->d_name.name %s code %d\n",
- ip->i_mode, dp->d_name.name, code);
-#ifdef STRUCT_INODE_HAS_I_SECURITY
- if (ip->i_security == NULL) {
- if (security_inode_alloc(ip))
- panic("afs_linux_lookup: Cannot allocate inode security");
- }
-#endif
+ struct vattr vattr;
+
+ 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
- if (S_ISDIR(ip->i_mode))
- ip->i_op = &afs_dir_iops;
- else if (S_ISLNK(ip->i_mode))
- ip->i_op = &afs_symlink_iops;
+ ip->i_hash.prev == NULL
#endif
+ )
+ insert_inode_hash(ip);
}
- dp->d_time = jiffies;
dp->d_op = &afs_dentry_operations;
- d_add(dp, AFSTOI(vcp));
+ dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
+ AFS_GUNLOCK();
+
+#if defined(AFS_LINUX24_ENV)
+ 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 {
+ iput(ip);
+#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
#endif
}
-int
+static int
afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
{
int code;
d_drop(newdp);
AFS_GLOCK();
- code = afs_link(ITOAFS(oldip), ITOAFS(dip), name, credp);
+ code = afs_link(VTOAFS(oldip), VTOAFS(dip), name, credp);
AFS_GUNLOCK();
crfree(credp);
return -code;
}
-int
+static int
afs_linux_unlink(struct inode *dip, struct dentry *dp)
{
- int code;
+ int code = EBUSY;
cred_t *credp = crref();
const char *name = dp->d_name.name;
+ 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)) {
+ struct dentry *__dp;
+ char *__name;
+ extern char *afs_newname();
+
+ __dp = NULL;
+ __name = NULL;
+ do {
+ dput(__dp);
+
+ AFS_GLOCK();
+ if (__name)
+ osi_FreeSmallSpace(__name);
+ __name = afs_newname();
+ AFS_GUNLOCK();
+
+ __dp = lookup_one_len(__name, dp->d_parent, strlen(__name));
+
+ if (IS_ERR(__dp))
+ goto out;
+ } while (__dp->d_inode != NULL);
+
+ AFS_GLOCK();
+ code = afs_rename(VTOAFS(dip), dp->d_name.name, VTOAFS(dip), __dp->d_name.name, credp);
+ if (!code) {
+ tvc->mvid = (void *) __name;
+ crhold(credp);
+ if (tvc->uncred) {
+ crfree(tvc->uncred);
+ }
+ 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();
+
+ if (!code) {
+ __dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
+ d_move(dp, __dp);
+ }
+ dput(__dp);
+
+ goto out;
+ }
+
AFS_GLOCK();
- code = afs_remove(ITOAFS(dip), name, credp);
+ code = afs_remove(VTOAFS(dip), name, credp);
AFS_GUNLOCK();
if (!code)
d_drop(dp);
+out:
#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
+ maybe_unlock_kernel();
#endif
crfree(credp);
return -code;
}
-int
+static int
afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
{
int code;
*/
d_drop(dp);
- AFS_GLOCK();
VATTR_NULL(&vattr);
- code = afs_symlink(ITOAFS(dip), name, &vattr, target, credp);
+ AFS_GLOCK();
+ code = afs_symlink(VTOAFS(dip), name, &vattr, target, credp);
AFS_GUNLOCK();
crfree(credp);
return -code;
}
-int
+static int
afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
{
int code;
const char *name = dp->d_name.name;
#if defined(AFS_LINUX26_ENV)
- lock_kernel();
+ maybe_lock_kernel();
#endif
- AFS_GLOCK();
VATTR_NULL(&vattr);
vattr.va_mask = ATTR_MODE;
vattr.va_mode = mode;
- code = afs_mkdir(ITOAFS(dip), name, &vattr, &tvcp, credp);
- AFS_GUNLOCK();
+ AFS_GLOCK();
+ code = afs_mkdir(VTOAFS(dip), name, &vattr, &tvcp, credp);
if (tvcp) {
- tvcp->v.v_op = &afs_dir_iops;
-#if defined(AFS_LINUX24_ENV)
- tvcp->v.v_fop = &afs_dir_fops;
-#endif
+ struct inode *ip = AFSTOV(tvcp);
+
+ afs_getattr(tvcp, &vattr, credp);
+ afs_fill_inode(ip, &vattr);
+
dp->d_op = &afs_dentry_operations;
- dp->d_time = jiffies;
- d_instantiate(dp, AFSTOI(tvcp));
+ 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;
}
-int
+static int
afs_linux_rmdir(struct inode *dip, struct dentry *dp)
{
int code;
cred_t *credp = crref();
const char *name = dp->d_name.name;
-#if defined(AFS_LINUX26_ENV)
- lock_kernel();
-#endif
+ /* locking kernel conflicts with glock? */
+
AFS_GLOCK();
- code = afs_rmdir(ITOAFS(dip), name, credp);
+ code = afs_rmdir(VTOAFS(dip), name, credp);
AFS_GUNLOCK();
/* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
d_drop(dp);
}
-#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
-#endif
crfree(credp);
return -code;
}
-
-int
+static int
afs_linux_rename(struct inode *oldip, struct dentry *olddp,
struct inode *newip, struct dentry *newdp)
{
cred_t *credp = crref();
const char *oldname = olddp->d_name.name;
const char *newname = newdp->d_name.name;
+ struct dentry *rehash = NULL;
#if defined(AFS_LINUX26_ENV)
- 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)
- if (!d_unhashed(olddp))
- d_drop(olddp);
- if (!d_unhashed(newdp))
+ /* Prevent any new references during rename operation. */
+ maybe_lock_kernel();
+
+ if (!d_unhashed(newdp)) {
d_drop(newdp);
+ rehash = newdp;
+ }
#else
- if (!list_empty(&olddp->d_hash))
- d_drop(olddp);
- if (!list_empty(&newdp->d_hash))
+ if (!list_empty(&newdp->d_hash)) {
d_drop(newdp);
+ rehash = newdp;
+ }
#endif
+
+#if defined(AFS_LINUX24_ENV)
+ if (atomic_read(&olddp->d_count) > 1)
+ shrink_dcache_parent(olddp);
+#endif
+
AFS_GLOCK();
- code = afs_rename(ITOAFS(oldip), oldname, ITOAFS(newip), newname, credp);
+ code = afs_rename(VTOAFS(oldip), oldname, VTOAFS(newip), newname, credp);
AFS_GUNLOCK();
- if (!code) {
- /* update time so it doesn't expire immediately */
- newdp->d_time = jiffies;
- d_move(olddp, newdp);
- }
+ 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);
struct iovec iov;
setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
- code = afs_readlink(ITOAFS(ip), &tuio, credp);
+ code = afs_readlink(VTOAFS(ip), &tuio, credp);
crfree(credp);
if (!code)
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.
*/
-int
+static int
afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
{
int code;
/* afs_linux_follow_link
* a file system dependent link following routine.
*/
-struct dentry *
+#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.
*/
-int
+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 = 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 = ITOAFS(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);
-#endif
-
- 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();
- }
-
- crfree(credp);
- return -code;
-}
-
-#if defined(AFS_LINUX24_ENV)
-int
-afs_linux_writepage(struct page *pp)
-{
- struct address_space *mapping = pp->mapping;
- struct inode *inode;
- unsigned long end_index;
- unsigned offset = PAGE_CACHE_SIZE;
- long status;
-
-#ifdef PageLaunder
- if (PageLaunder(pp)) {
- return(fail_writepage(pp));
- }
-#endif
-
- inode = (struct inode *)mapping->host;
- end_index = inode->i_size >> PAGE_CACHE_SHIFT;
-
- /* easy case */
- if (pp->index < end_index)
- goto do_it;
- /* things got complicated... */
- offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
- /* OK, are we completely out? */
- if (pp->index >= end_index + 1 || !offset)
- return -EIO;
- do_it:
- AFS_GLOCK();
- status = afs_linux_writepage_sync(inode, pp, 0, offset);
- AFS_GUNLOCK();
- SetPageUptodate(pp);
- UnlockPage(pp);
- if (status == offset)
- return 0;
- else
- return status;
-}
-#endif
-
-/* afs_linux_permission
- * Check access rights - returns error if can't check or permission denied.
- */
-int
-afs_linux_permission(struct inode *ip, int mode)
-{
- int code;
- cred_t *credp = crref();
- int tmp = 0;
-
- AFS_GLOCK();
- if (mode & MAY_EXEC)
- tmp |= VEXEC;
- if (mode & MAY_READ)
- tmp |= VREAD;
- if (mode & MAY_WRITE)
- tmp |= VWRITE;
- code = afs_access(ITOAFS(ip), tmp, credp);
-
- AFS_GUNLOCK();
- crfree(credp);
- return -code;
+ clear_bit(PG_locked, &pp->flags);
+ wake_up(&pp->wait);
+ free_page(address);
+#endif
+
+#if defined(AFS_CACHE_BYPASS)
+
+/* do not call afs_GetDCache if cache is bypassed */
+ if(bypasscache)
+ goto done;
+
+#endif
+
+ /* 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;
}
#if defined(AFS_LINUX24_ENV)
-int
+static int
afs_linux_writepage_sync(struct inode *ip, struct page *pp,
unsigned long offset, unsigned int count)
{
- struct vcache *vcp = ITOAFS(ip);
+ struct vcache *vcp = VTOAFS(ip);
char *buffer;
afs_offs_t base;
int code = 0;
int f_flags = 0;
buffer = kmap(pp) + offset;
- base = (pp->index << PAGE_CACHE_SHIFT) + offset;
+ base = (((loff_t) pp->index) << PAGE_CACHE_SHIFT) + offset;
credp = crref();
+ 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);
- vcache2inode(vcp);
+ ip->i_size = vcp->m.Length;
+ ip->i_blocks = ((vcp->m.Length + 1023) >> 10) << 1;
- if (!code
- && afs_stats_cmperf.cacheCurrDirtyChunks >
- afs_stats_cmperf.cacheMaxDirtyChunks) {
+ if (!code) {
struct vrequest treq;
ObtainWriteLock(&vcp->lock, 533);
ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
ICL_TYPE_INT32, code);
+ AFS_GUNLOCK();
+ maybe_unlock_kernel();
crfree(credp);
kunmap(pp);
return code;
}
+
static int
-afs_linux_updatepage(struct file *file, struct page *page,
- unsigned long offset, unsigned int count)
+#ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
+afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
+#else
+afs_linux_writepage(struct page *pp)
+#endif
{
- struct dentry *dentry = file->f_dentry;
+ struct address_space *mapping = pp->mapping;
+ struct inode *inode;
+ unsigned long end_index;
+ unsigned offset = PAGE_CACHE_SIZE;
+ long status;
- return afs_linux_writepage_sync(dentry->d_inode, page, offset, count);
+#if defined(AFS_LINUX26_ENV)
+ if (PageReclaim(pp)) {
+# if defined(WRITEPAGE_ACTIVATE)
+ return WRITEPAGE_ACTIVATE;
+# else
+ return AOP_WRITEPAGE_ACTIVATE;
+# endif
+ }
+#else
+ if (PageLaunder(pp)) {
+ return(fail_writepage(pp));
+ }
+#endif
+
+ inode = (struct inode *)mapping->host;
+ end_index = inode->i_size >> PAGE_CACHE_SHIFT;
+
+ /* easy case */
+ if (pp->index < end_index)
+ goto do_it;
+ /* things got complicated... */
+ offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
+ /* OK, are we completely out? */
+ if (pp->index >= end_index + 1 || !offset)
+ return -EIO;
+ do_it:
+ status = afs_linux_writepage_sync(inode, pp, 0, offset);
+ SetPageUptodate(pp);
+ UnlockPage(pp);
+ if (status == offset)
+ return 0;
+ else
+ return status;
}
#else
/* afs_linux_updatepage
* Called from generic_file_write. buffer is still in user space. pagep
* has been filled in with old data if we're updating less than a page.
*/
-int
+static int
afs_linux_updatepage(struct file *fp, struct page *pp, unsigned long offset,
unsigned int count, int sync)
{
- struct vcache *vcp = ITOAFS(FILE_INODE(fp));
+ struct vcache *vcp = VTOAFS(FILE_INODE(fp));
u8 *page_addr = (u8 *) afs_linux_page_address(pp);
int code = 0;
cred_t *credp;
code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
- vcache2inode(vcp);
+ ip->i_size = vcp->m.Length;
+ ip->i_blocks = ((vcp->m.Length + 1023) >> 10) << 1;
+
+ if (!code) {
+ struct vrequest treq;
+
+ ObtainWriteLock(&vcp->lock, 533);
+ vcp->m.Date = osi_Time(); /* set modification time */
+ if (!afs_InitReq(&treq, credp))
+ code = afs_DoPartialWrite(vcp, &treq);
+ ReleaseWriteLock(&vcp->lock);
+ }
code = code ? -code : count - tuio.uio_resid;
afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
}
#endif
+/* afs_linux_permission
+ * Check access rights - returns error if can't check or permission denied.
+ */
+static int
+#ifdef 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();
+ int tmp = 0;
+
+ AFS_GLOCK();
+ if (mode & MAY_EXEC)
+ tmp |= VEXEC;
+ if (mode & MAY_READ)
+ tmp |= VREAD;
+ if (mode & MAY_WRITE)
+ tmp |= VWRITE;
+ code = afs_access(VTOAFS(ip), tmp, credp);
+
+ AFS_GUNLOCK();
+ crfree(credp);
+ return -code;
+}
+
#if defined(AFS_LINUX24_ENV)
static int
afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
{
int code;
- lock_kernel();
- AFS_GLOCK();
- code = afs_linux_updatepage(file, page, offset, to - offset);
- AFS_GUNLOCK();
- unlock_kernel();
+ code = afs_linux_writepage_sync(file->f_dentry->d_inode, page,
+ offset, to - offset);
+#if !defined(AFS_LINUX26_ENV)
kunmap(page);
+#endif
return code;
}
afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
unsigned to)
{
+/* sometime between 2.4.0 and 2.4.19, the callers of prepare_write began to
+ call kmap directly instead of relying on us to do it */
+#if !defined(AFS_LINUX26_ENV)
kmap(page);
+#endif
return 0;
}
extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
#endif
-struct inode_operations afs_file_iops = {
+static struct inode_operations afs_file_iops = {
#if defined(AFS_LINUX26_ENV)
.permission = afs_linux_permission,
.getattr = afs_linux_getattr,
.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)
-struct address_space_operations afs_file_aops = {
+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,
* by what sort of operation is allowed.....
*/
-struct inode_operations afs_dir_iops = {
+static struct inode_operations afs_dir_iops = {
#if !defined(AFS_LINUX24_ENV)
.default_file_ops = &afs_dir_fops,
#else
/* 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);
return code;
}
-struct address_space_operations afs_symlink_aops = {
+static struct address_space_operations afs_symlink_aops = {
.readpage = afs_symlink_filler
};
-#endif
+#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
-struct inode_operations afs_symlink_iops = {
-#if defined(AFS_LINUX24_ENV)
+static struct inode_operations afs_symlink_iops = {
+#if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
.readlink = page_readlink,
+#if defined(HAVE_KERNEL_PAGE_FOLLOW_LINK)
.follow_link = page_follow_link,
- .setattr = afs_notify_change,
#else
+ .follow_link = page_follow_link_light,
+ .put_link = page_put_link,
+#endif
+#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
+afs_fill_inode(struct inode *ip, struct vattr *vattr)
+{
+
+ if (vattr)
+ vattr2inode(ip, vattr);
+
+ ip->i_mapping->backing_dev_info = &afs_backing_dev_info;
+/* Reset ops if symlink or directory. */
+ if (S_ISREG(ip->i_mode)) {
+ ip->i_op = &afs_file_iops;
+#if defined(AFS_LINUX24_ENV)
+ ip->i_fop = &afs_file_fops;
+ ip->i_data.a_ops = &afs_file_aops;
+#endif
+
+ } else if (S_ISDIR(ip->i_mode)) {
+ ip->i_op = &afs_dir_iops;
+#if defined(AFS_LINUX24_ENV)
+ ip->i_fop = &afs_dir_fops;
+#endif
+
+ } else if (S_ISLNK(ip->i_mode)) {
+ ip->i_op = &afs_symlink_iops;
+#if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
+ ip->i_data.a_ops = &afs_symlink_aops;
+ ip->i_mapping = &ip->i_data;
+#endif
+ }
+
+}