#include <afsconfig.h>
#include "afs/param.h"
-RCSID
- ("$Header$");
#include "afs/sysincludes.h"
#include "afsincludes.h"
#include "afs/afs_stats.h"
-#include "h/mm.h"
+#include <linux/mm.h>
#ifdef HAVE_MM_INLINE_H
-#include "h/mm_inline.h"
+#include <linux/mm_inline.h>
#endif
-#include "h/pagemap.h"
-#if defined(AFS_LINUX24_ENV)
-#include "h/smp_lock.h"
-#endif
-#if defined(AFS_LINUX26_ENV)
-#include "h/writeback.h"
+#include <linux/pagemap.h>
+#include <linux/smp_lock.h>
+#include <linux/writeback.h>
+#include <linux/pagevec.h>
+#if defined(AFS_CACHE_BYPASS)
+#include "afs/lock.h"
+#include "afs/afs_bypasscache.h"
#endif
-#ifdef pgoff2loff
-#define pageoff(pp) pgoff2loff((pp)->index)
-#else
-#define pageoff(pp) pp->offset
+#include "osi_compat.h"
+#include "osi_pagecopy.h"
+
+#ifndef HAVE_PAGEVEC_LRU_ADD_FILE
+#define __pagevec_lru_add_file __pagevec_lru_add
#endif
-#if defined(AFS_LINUX26_ENV)
-#define UnlockPage(pp) unlock_page(pp)
+#ifndef MAX_ERRNO
+#define MAX_ERRNO 1000L
#endif
+extern struct backing_dev_info afs_backing_dev_info;
+
extern struct vcache *afs_globalVp;
+extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
+/* 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;
+
+static inline void
+afs_maybe_lock_kernel(void) {
+ if(afs_cacheType == &afs_UfsCacheOps)
+ lock_kernel();
+}
+
+static inline void
+afs_maybe_unlock_kernel(void) {
+ if(afs_cacheType == &afs_UfsCacheOps)
+ unlock_kernel();
+}
+
+/* This function converts a positive error code from AFS into a negative
+ * code suitable for passing into the Linux VFS layer. It checks that the
+ * error code is within the permissable bounds for the ERR_PTR mechanism.
+ *
+ * _All_ error codes which come from the AFS layer should be passed through
+ * this function before being returned to the kernel.
+ */
+
+static inline int
+afs_convert_code(int code) {
+ if ((code >= 0) && (code <= MAX_ERRNO))
+ return -code;
+ else
+ return -EIO;
+}
+
+/* Linux doesn't require a credp for many functions, and crref is an expensive
+ * operation. This helper function avoids obtaining it for VerifyVCache calls
+ */
+
+static inline int
+afs_linux_VerifyVCache(struct vcache *avc, cred_t **retcred) {
+ cred_t *credp = NULL;
+ struct vrequest treq;
+ int code;
+
+ if (avc->f.states & CStatd) {
+ if (retcred)
+ *retcred = NULL;
+ return 0;
+ }
+
+ credp = crref();
+
+ code = afs_InitReq(&treq, credp);
+ if (code == 0)
+ code = afs_VerifyVCache2(avc, &treq);
+
+ if (retcred != NULL)
+ *retcred = credp;
+ else
+ crfree(credp);
+ return afs_convert_code(code);
+}
static ssize_t
afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
{
- ssize_t code;
+ ssize_t code = 0;
struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
- cred_t *credp = crref();
- struct vrequest treq;
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);
+ code = afs_linux_VerifyVCache(vcp, NULL);
- /* 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_read(fp, buf, count, offp);
- AFS_GLOCK();
+ if (code == 0) {
+ /* Linux's FlushPages implementation doesn't ever use credp,
+ * so we optimise by not using it */
+ osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
+ AFS_GUNLOCK();
+ code = do_sync_read(fp, buf, count, offp);
+ 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);
-
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();
+ cred_t *credp;
AFS_GLOCK();
ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
(fp->f_flags & O_APPEND) ? 99998 : 99999);
-
- /* get a validated vcache entry */
- code = (ssize_t) afs_InitReq(&treq, credp);
- if (!code)
- code = (ssize_t) afs_VerifyVCache(vcp, &treq);
+ code = afs_linux_VerifyVCache(vcp, &credp);
ObtainWriteLock(&vcp->lock, 529);
afs_FakeOpen(vcp);
ReleaseWriteLock(&vcp->lock);
- if (code)
- code = -code;
- else {
+ if (code == 0) {
AFS_GUNLOCK();
- code = generic_file_write(fp, buf, count, offp);
+ code = do_sync_write(fp, buf, count, offp);
AFS_GLOCK();
}
ObtainWriteLock(&vcp->lock, 530);
- vcp->m.Date = osi_Time(); /* set modification time */
+
+ if (vcp->execsOrWriters == 1 && !credp)
+ credp = crref();
+
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,
ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
code);
+ if (credp)
+ crfree(credp);
AFS_GUNLOCK();
- crfree(credp);
return code;
}
static int
afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
{
- extern struct DirEntry *afs_dir_GetBlob();
struct vcache *avc = VTOAFS(FILE_INODE(fp));
struct vrequest treq;
register struct dcache *tdc;
cred_t *credp = crref();
struct afs_fakestat_state fakestat;
-#if defined(AFS_LINUX26_ENV)
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
AFS_GLOCK();
AFS_STATCNT(afs_readdir);
- code = afs_InitReq(&treq, credp);
+ code = afs_convert_code(afs_InitReq(&treq, credp));
crfree(credp);
if (code)
goto out1;
afs_InitFakeStat(&fakestat);
- code = afs_EvalFakeStat(&avc, &fakestat, &treq);
+ code = afs_convert_code(afs_EvalFakeStat(&avc, &fakestat, &treq));
if (code)
goto out;
/* update the cache entry */
tagain:
- code = afs_VerifyVCache(avc, &treq);
+ code = afs_convert_code(afs_VerifyVCache2(avc, &treq));
if (code)
goto out;
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
* 1. The cache data is being fetched by another process.
* 2. The cache data is no longer valid
*/
- while ((avc->states & CStatd)
+ while ((avc->f.states & CStatd)
&& (tdc->dflags & DFFetching)
- && hsame(avc->m.DataVersion, tdc->f.versionNo)) {
+ && hsame(avc->f.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)) {
+ if (!(avc->f.states & CStatd)
+ || !hsame(avc->f.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->f.states |= CReadDir;
+ avc->dcreaddir = tdc;
+ avc->readdir_pid = MyPidxx2Pid(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->f.fid.Fid.Volume, ntohl(de->fid.vnode));
+
if (de->name)
len = strlen(de->name);
else {
printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
(unsigned long)&tdc->f.inode, dirpos);
- DRelease((struct buffer *) de, 0);
+ DRelease(de, 0);
+ ReleaseSharedLock(&avc->lock);
afs_PutDCache(tdc);
- ReleaseReadLock(&avc->lock);
code = -ENOENT;
goto out;
}
/* filldir returns -EINVAL when the buffer is full. */
-#if defined(AFS_LINUX26_ENV) || ((defined(AFS_LINUX24_ENV) || defined(pgoff2loff)) && defined(DECLARE_FSTYPE))
{
unsigned int type = DT_UNKNOWN;
struct VenusFid afid;
struct vcache *tvc;
int vtype;
- afid.Cell = avc->fid.Cell;
- afid.Fid.Volume = avc->fid.Fid.Volume;
+ afid.Cell = avc->f.fid.Cell;
+ afid.Fid.Volume = avc->f.fid.Fid.Volume;
afid.Fid.Vnode = ntohl(de->fid.vnode);
afid.Fid.Unique = ntohl(de->fid.vunique);
- if ((avc->states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
+ if ((avc->f.states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
type = DT_DIR;
} else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
if (tvc->mvstat) {
type = DT_DIR;
- } else if (((tvc->states) & (CStatd | CTruth))) {
+ } else if (((tvc->f.states) & (CStatd | CTruth))) {
/* CTruth will be set if the object has
*ever* been statd */
vtype = vType(tvc);
/* 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);
-#endif
- DRelease((struct buffer *)de, 0);
+ DRelease(de, 0);
if (code)
break;
offset = dirpos + 1 + ((len + 16) >> 5);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
- ReleaseReadLock(&avc->lock);
+ UpgradeSToWLock(&avc->lock, 813);
+ avc->f.states &= ~CReadDir;
+ avc->dcreaddir = 0;
+ avc->readdir_pid = 0;
+ ReleaseSharedLock(&avc->lock);
code = 0;
out:
afs_PutFakeStat(&fakestat);
out1:
AFS_GUNLOCK();
-#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
return code;
}
afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
{
struct vcache *vcp = VTOAFS(FILE_INODE(fp));
- cred_t *credp = crref();
- struct vrequest treq;
int code;
AFS_GLOCK();
-#if defined(AFS_LINUX24_ENV)
afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
vmap->vm_end - vmap->vm_start);
-#else
- afs_Trace4(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
- ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
- vmap->vm_end - vmap->vm_start, ICL_TYPE_INT32,
- vmap->vm_offset);
-#endif
/* get a validated vcache entry */
- code = afs_InitReq(&treq, credp);
- if (code)
- goto out_err;
-
- code = afs_VerifyVCache(vcp, &treq);
- if (code)
- goto out_err;
-
- osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
+ code = afs_linux_VerifyVCache(vcp, NULL);
+ /* Linux's Flushpage implementation doesn't use credp, so optimise
+ * our code to not need to crref() it */
+ osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
AFS_GUNLOCK();
code = generic_file_mmap(fp, vmap);
AFS_GLOCK();
if (!code)
- vcp->states |= CMAPPED;
+ vcp->f.states |= CMAPPED;
-out:
AFS_GUNLOCK();
- crfree(credp);
return code;
-
-out_err:
- code = -code;
- goto out;
}
static int
cred_t *credp = crref();
int code;
-#ifdef AFS_LINUX24_ENV
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
AFS_GLOCK();
code = afs_open(&vcp, fp->f_flags, credp);
AFS_GUNLOCK();
-#ifdef AFS_LINUX24_ENV
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
static int
cred_t *credp = crref();
int code = 0;
-#ifdef AFS_LINUX24_ENV
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
AFS_GLOCK();
code = afs_close(vcp, fp->f_flags, credp);
+ ObtainWriteLock(&vcp->lock, 807);
+ if (vcp->cred) {
+ crfree(vcp->cred);
+ vcp->cred = NULL;
+ }
+ ReleaseWriteLock(&vcp->lock);
AFS_GUNLOCK();
-#ifdef AFS_LINUX24_ENV
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
static int
-#if defined(AFS_LINUX24_ENV)
afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
-#else
-afs_linux_fsync(struct file *fp, struct dentry *dp)
-#endif
{
int code;
struct inode *ip = FILE_INODE(fp);
cred_t *credp = crref();
-#ifdef AFS_LINUX24_ENV
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
AFS_GLOCK();
code = afs_fsync(VTOAFS(ip), credp);
AFS_GUNLOCK();
-#ifdef AFS_LINUX24_ENV
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
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));
+ memset(&flock, 0, sizeof(flock));
flock.l_type = flp->fl_type;
flock.l_pid = flp->fl_pid;
flock.l_whence = 0;
flock.l_start = flp->fl_start;
- flock.l_len = flp->fl_end - flp->fl_start;
+ flock.l_len = flp->fl_end - flp->fl_start + 1;
/* Safe because there are no large files, yet */
#if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
code = afs_lockctl(vcp, &flock, cmd, credp);
AFS_GUNLOCK();
+ if ((code == 0 || flp->fl_type == F_UNLCK) &&
+ (cmd == F_SETLK || cmd == F_SETLKW)) {
+ code = afs_posix_lock_file(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();
+ }
+ }
+ /* If lockctl says there are no conflicting locks, then also check with the
+ * kernel, as lockctl knows nothing about byte range locks
+ */
+ if (code == 0 && cmd == F_GETLK && flock.l_type == F_UNLCK) {
+ afs_posix_test_lock(fp, flp);
+ /* If we found a lock in the kernel's structure, return it */
+ if (flp->fl_type != F_UNLCK) {
+ crfree(credp);
+ return 0;
+ }
+ }
+
/* Convert flock back to Linux's file_lock */
flp->fl_type = flock.l_type;
flp->fl_pid = flock.l_pid;
flp->fl_start = flock.l_start;
- flp->fl_end = flock.l_start + flock.l_len;
+ flp->fl_end = flock.l_start + flock.l_len - 1;
crfree(credp);
- return -code;
+ return afs_convert_code(code);
+}
+
+#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(&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 afs_convert_code(code);
}
+#endif
/* afs_linux_flush
* essentially the same as afs_fsync() but we need to get the return
* 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 {
+ afs_DisconAddDirty(vcp, VDisconWriteOsiFlush, 1);
+ }
ConvertWToSLock(&vcp->lock);
}
code = afs_CheckCode(code, &treq, 54);
ReleaseSharedLock(&vcp->lock);
out:
+ AFS_DISCON_UNLOCK();
AFS_GUNLOCK();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
-#if !defined(AFS_LINUX24_ENV)
-/* Not allowed to directly read a directory. */
-ssize_t
-afs_linux_dir_read(struct file * fp, char *buf, size_t count, loff_t * ppos)
-{
- return -EISDIR;
-}
-#endif
-
-
-
struct file_operations afs_dir_fops = {
-#if !defined(AFS_LINUX24_ENV)
- .read = afs_linux_dir_read,
- .lock = afs_linux_lock,
- .fsync = afs_linux_fsync,
-#else
.read = generic_read_dir,
-#endif
.readdir = afs_linux_readdir,
#ifdef HAVE_UNLOCKED_IOCTL
.unlocked_ioctl = afs_unlocked_xioctl,
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(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
.sendfile = generic_file_sendfile,
#endif
+#if 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
};
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 */
+ if (vcp->mvid->Fid.Volume != pvc->f.fid.Fid.Volume) { /* bad parent */
credp = crref();
/* force a lookup, so vcp->mvid is fixed up */
- afs_lookup(pvc, dp->d_name.name, &avc, credp);
+ afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
if (!avc || vcp != avc) { /* bad, very bad.. */
afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
"check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
cred_t *credp;
int code;
-#ifdef AFS_LINUX24_ENV
- lock_kernel();
-#endif
+ if (afs_shuttingdown)
+ return EIO;
+
+ afs_maybe_lock_kernel();
AFS_GLOCK();
+#ifdef notyet
/* Make this a fast path (no crref), since it's called so often. */
- if (vcp->states & CStatd) {
+ if (vcp->f.states & CStatd) {
if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
check_bad_parent(dp); /* check and correct mvid */
AFS_GUNLOCK();
-#ifdef AFS_LINUX24_ENV
unlock_kernel();
-#endif
return 0;
}
+#endif
- credp = crref();
- code = afs_getattr(vcp, &vattr, credp);
+ /* This avoids the crref when we don't have to do it. Watch for
+ * changes in afs_getattr that don't get replicated here!
+ */
+ if (vcp->f.states & CStatd &&
+ (!afs_fakestat_enable || vcp->mvstat != 1) &&
+ !afs_nfsexporter) {
+ code = afs_CopyOutAttrs(vcp, &vattr);
+ } else {
+ credp = crref();
+ code = afs_getattr(vcp, &vattr, credp);
+ crfree(credp);
+ }
if (!code)
- vattr2inode(AFSTOV(vcp), &vattr);
+ afs_fill_inode(AFSTOV(vcp), &vattr);
AFS_GUNLOCK();
-#ifdef AFS_LINUX24_ENV
- unlock_kernel();
-#endif
- crfree(credp);
+ afs_maybe_unlock_kernel();
- return -code;
+ return afs_convert_code(code);
}
-#if defined(AFS_LINUX26_ENV)
static int
afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
{
}
return err;
}
-#endif
/* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
* In kernels 2.2.10 and above, we are passed an additional flags var which
* later on, we shouldn't have to do it until later. Perhaps in the future..
*/
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
-afs_linux_dentry_revalidate(struct dentry *dp)
-#endif
{
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();
-#endif
+ afs_maybe_lock_kernel();
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->f.states & CMValid)) {
+ int tryEvalOnly = 0;
+ int code = 0;
+ struct vrequest treq;
+
+ credp = crref();
+ code = afs_InitReq(&treq, credp);
+ if (
+ (strcmp(dp->d_name.name, ".directory") == 0)) {
+ 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
* isn't enough since the vnode may have been renamed.
*/
- if (hgetlo(pvcp->m.DataVersion) > dp->d_time || !(vcp->states & CStatd)) {
+ if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd)) {
credp = crref();
- afs_lookup(pvcp, dp->d_name.name, &tvc, credp);
+ afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
if (!tvc || tvc != vcp)
goto bad_dentry;
goto bad_dentry;
vattr2inode(AFSTOV(vcp), &vattr);
- dp->d_time = hgetlo(pvcp->m.DataVersion);
+ dp->d_time = hgetlo(pvcp->f.m.DataVersion);
}
/* should we always update the attributes at this point? */
} else {
#ifdef notyet
pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
- if (hgetlo(pvcp->m.DataVersion) > dp->d_time)
+ if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time)
goto bad_dentry;
#endif
/* Clean up */
if (tvc)
afs_PutVCache(tvc);
+ afs_PutFakeStat(&fakestate);
AFS_GUNLOCK();
if (credp)
crfree(credp);
shrink_dcache_parent(dp);
d_drop(dp);
}
-#ifdef AFS_LINUX24_ENV
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
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_remunlink(vcp, 1); /* perhaps afs_InactiveVCache() instead */
+ if (!AFS_IS_DISCONNECTED || (vcp->f.states & CUnlinked)) {
+ (void) afs_InactiveVCache(vcp, NULL);
+ }
AFS_GUNLOCK();
+ afs_linux_clear_nfsfs_renamed(dp);
iput(ip);
}
static int
afs_dentry_delete(struct dentry *dp)
{
- if (dp->d_inode && (VTOAFS(dp->d_inode)->states & CUnlinked))
+ if (dp->d_inode && (VTOAFS(dp->d_inode)->f.states & CUnlinked))
return 1; /* bad inode? */
return 0;
vattr.va_mode = mode;
vattr.va_type = mode & S_IFMT;
-#if defined(AFS_LINUX26_ENV)
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
AFS_GLOCK();
code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
&vcp, credp);
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);
+ dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
d_instantiate(dp, ip);
}
AFS_GUNLOCK();
-#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
/* afs_linux_lookup */
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
static struct dentry *
#ifdef IOP_LOOKUP_TAKES_NAMEIDATA
afs_linux_lookup(struct inode *dip, struct dentry *dp,
#else
afs_linux_lookup(struct inode *dip, struct dentry *dp)
#endif
-#else
-static int
-afs_linux_lookup(struct inode *dip, struct dentry *dp)
-#endif
{
- struct vattr vattr;
cred_t *credp = crref();
struct vcache *vcp = NULL;
const char *comp = dp->d_name.name;
- struct dentry *res = NULL;
struct inode *ip = NULL;
+ struct dentry *newdp = NULL;
int code;
-#if defined(AFS_LINUX26_ENV)
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
AFS_GLOCK();
- code = afs_lookup(VTOAFS(dip), comp, &vcp, credp);
+ code = afs_lookup(VTOAFS(dip), (char *)comp, &vcp, credp);
if (vcp) {
- ip = AFSTOV(vcp);
+ struct vattr vattr;
+ ip = AFSTOV(vcp);
afs_getattr(vcp, &vattr, credp);
afs_fill_inode(ip, &vattr);
+ if (hlist_unhashed(&ip->i_hash))
+ insert_inode_hash(ip);
}
dp->d_op = &afs_dentry_operations;
- dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
+ dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
AFS_GUNLOCK();
-#if defined(AFS_LINUX24_ENV)
+
if (ip && S_ISDIR(ip->i_mode)) {
- d_prune_aliases(ip);
- res = d_find_alias(ip);
+ struct dentry *alias;
+
+ /* Try to invalidate an existing alias in favor of our new one */
+ alias = d_find_alias(ip);
+ /* But not if it's disconnected; then we want d_splice_alias below */
+ if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
+ if (d_invalidate(alias) == 0) {
+ dput(alias);
+ } else {
+ iput(ip);
+ afs_maybe_unlock_kernel();
+ crfree(credp);
+ return alias;
+ }
+ }
}
- if (res) {
- if (d_unhashed(res))
- d_rehash(res);
- iput(ip);
- } else
-#endif
- d_add(dp, ip);
+ newdp = d_splice_alias(ip, dp);
-#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
/* It's ok for the file to not be found. That's noted by the caller by
* seeing that the dp->d_inode field is NULL.
*/
-#if defined(AFS_LINUX24_ENV)
- if (code == 0)
- return res;
-#endif
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
- if (code == ENOENT)
- return ERR_PTR(0);
- else
- return ERR_PTR(-code);
-#else
- if (code == ENOENT)
- code = 0;
- return -code;
-#endif
+ if (!code || code == ENOENT)
+ return newdp;
+ else
+ return ERR_PTR(afs_convert_code(code));
}
static int
d_drop(newdp);
AFS_GLOCK();
- code = afs_link(VTOAFS(oldip), VTOAFS(dip), name, credp);
+ code = afs_link(VTOAFS(oldip), VTOAFS(dip), (char *)name, credp);
AFS_GUNLOCK();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
static int
const char *name = dp->d_name.name;
struct vcache *tvc = VTOAFS(dp->d_inode);
-#if defined(AFS_LINUX26_ENV)
- lock_kernel();
-#endif
- if (((VREFCOUNT(tvc) > 0) && tvc->opens > 0)
- && !(tvc->states & CUnlinked)) {
+ afs_maybe_lock_kernel();
+ if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
+ && !(tvc->f.states & CUnlinked)) {
struct dentry *__dp;
char *__name;
- extern char *afs_newname();
__dp = NULL;
__name = NULL;
} while (__dp->d_inode != NULL);
AFS_GLOCK();
- code = afs_rename(VTOAFS(dip), dp->d_name.name, VTOAFS(dip), __dp->d_name.name, credp);
+ code = afs_rename(VTOAFS(dip), (char *)dp->d_name.name, VTOAFS(dip), (char *)__dp->d_name.name, credp);
if (!code) {
tvc->mvid = (void *) __name;
crhold(credp);
crfree(tvc->uncred);
}
tvc->uncred = credp;
- tvc->states |= CUnlinked;
+ tvc->f.states |= CUnlinked;
+ afs_linux_set_nfsfs_renamed(dp);
+ } else {
+ osi_FreeSmallSpace(__name);
}
AFS_GUNLOCK();
if (!code) {
- __dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
+ __dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
d_move(dp, __dp);
}
dput(__dp);
}
AFS_GLOCK();
- code = afs_remove(VTOAFS(dip), name, credp);
+ code = afs_remove(VTOAFS(dip), (char *)name, credp);
AFS_GUNLOCK();
if (!code)
d_drop(dp);
out:
-#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
VATTR_NULL(&vattr);
AFS_GLOCK();
- code = afs_symlink(VTOAFS(dip), name, &vattr, target, credp);
+ code = afs_symlink(VTOAFS(dip), (char *)name, &vattr, (char *)target, credp);
AFS_GUNLOCK();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
static int
struct vattr vattr;
const char *name = dp->d_name.name;
-#if defined(AFS_LINUX26_ENV)
- lock_kernel();
-#endif
+ afs_maybe_lock_kernel();
VATTR_NULL(&vattr);
vattr.va_mask = ATTR_MODE;
vattr.va_mode = mode;
AFS_GLOCK();
- code = afs_mkdir(VTOAFS(dip), name, &vattr, &tvcp, credp);
+ code = afs_mkdir(VTOAFS(dip), (char *)name, &vattr, &tvcp, credp);
if (tvcp) {
struct inode *ip = AFSTOV(tvcp);
afs_fill_inode(ip, &vattr);
dp->d_op = &afs_dentry_operations;
- dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
+ dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
d_instantiate(dp, ip);
}
AFS_GUNLOCK();
-#if defined(AFS_LINUX26_ENV)
- unlock_kernel();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
static int
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(VTOAFS(dip), name, credp);
+ code = afs_rmdir(VTOAFS(dip), (char *)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;
+ return afs_convert_code(code);
}
const char *newname = newdp->d_name.name;
struct dentry *rehash = NULL;
-#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)
+ afs_maybe_lock_kernel();
+
if (!d_unhashed(newdp)) {
d_drop(newdp);
rehash = newdp;
}
-#else
- 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(VTOAFS(oldip), oldname, VTOAFS(newip), newname, credp);
+ code = afs_rename(VTOAFS(oldip), (char *)oldname, VTOAFS(newip), (char *)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();
-#endif
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
if (!code)
return maxlen - tuio.uio_resid;
else
- return -code;
+ return afs_convert_code(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.
*/
-static struct dentry *
-afs_linux_follow_link(struct dentry *dp, struct dentry *basep,
- unsigned int follow)
+static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
{
- int code = 0;
+ int code;
char *name;
- struct dentry *res;
-
- AFS_GLOCK();
- name = osi_Alloc(PATH_MAX + 1);
+ name = osi_Alloc(PATH_MAX);
if (!name) {
- AFS_GUNLOCK();
- dput(basep);
- return ERR_PTR(-EIO);
+ return -EIO;
}
- code = afs_linux_ireadlink(dp->d_inode, name, PATH_MAX, AFS_UIOSYS);
+ AFS_GLOCK();
+ code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
AFS_GUNLOCK();
if (code < 0) {
- dput(basep);
- res = ERR_PTR(code);
- } else {
- name[code] = '\0';
- res = lookup_dentry(name, basep, follow);
+ goto out;
}
- AFS_GLOCK();
- osi_Free(name, PATH_MAX + 1);
- AFS_GUNLOCK();
- return res;
+ name[code] = '\0';
+ code = vfs_follow_link(nd, name);
+
+out:
+ osi_Free(name, PATH_MAX);
+
+ return code;
}
-#endif
-/* afs_linux_readpage
- * all reads come through here. A strategy-like read call.
+#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
+
+#if defined(AFS_CACHE_BYPASS)
+#endif /* defined(AFS_CACHE_BYPASS */
+
+/* Populate a page by filling it from the cache file pointed at by cachefp
+ * (which contains indicated chunk)
+ * If task is NULL, the page copy occurs syncronously, and the routine
+ * returns with page still locked. If task is non-NULL, then page copies
+ * may occur in the background, and the page will be unlocked when it is
+ * ready for use.
*/
static int
-afs_linux_readpage(struct file *fp, struct page *pp)
-{
- int code;
- cred_t *credp = crref();
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
+afs_linux_read_cache(struct file *cachefp, struct page *page,
+ int chunk, struct pagevec *lrupv,
+ struct afs_pagecopy_task *task) {
+ loff_t offset = page_offset(page);
+ struct page *newpage, *cachepage;
+ struct address_space *cachemapping;
+ int pageindex;
+ int code = 0;
+
+ cachemapping = cachefp->f_dentry->d_inode->i_mapping;
+ newpage = NULL;
+ cachepage = NULL;
+
+ /* From our offset, we now need to work out which page in the disk
+ * file it corresponds to. This will be fun ... */
+ pageindex = (offset - AFS_CHUNKTOBASE(chunk)) >> PAGE_CACHE_SHIFT;
+
+ while (cachepage == NULL) {
+ cachepage = find_get_page(cachemapping, pageindex);
+ if (!cachepage) {
+ if (!newpage)
+ newpage = page_cache_alloc_cold(cachemapping);
+ if (!newpage) {
+ code = -ENOMEM;
+ goto out;
+ }
+
+ code = add_to_page_cache(newpage, cachemapping,
+ pageindex, GFP_KERNEL);
+ if (code == 0) {
+ cachepage = newpage;
+ newpage = NULL;
+
+ page_cache_get(cachepage);
+ if (!pagevec_add(lrupv, cachepage))
+ __pagevec_lru_add_file(lrupv);
+
+ } else {
+ page_cache_release(newpage);
+ newpage = NULL;
+ if (code != -EEXIST)
+ goto out;
+ }
+ } else {
+ lock_page(cachepage);
+ }
+ }
+
+ if (!PageUptodate(cachepage)) {
+ ClearPageError(cachepage);
+ code = cachemapping->a_ops->readpage(NULL, cachepage);
+ if (!code && !task) {
+ wait_on_page_locked(cachepage);
+ }
+ } else {
+ unlock_page(cachepage);
+ }
+
+ if (!code) {
+ if (PageUptodate(cachepage)) {
+ copy_highpage(page, cachepage);
+ flush_dcache_page(page);
+ SetPageUptodate(page);
+
+ if (task)
+ unlock_page(page);
+ } else if (task) {
+ afs_pagecopy_queue_page(task, cachepage, page);
+ } else {
+ code = -EIO;
+ }
+ }
+
+ if (code && task) {
+ unlock_page(page);
+ }
+
+out:
+ if (cachepage)
+ page_cache_release(cachepage);
+
+ return code;
+}
+
+static int inline
+afs_linux_readpage_fastpath(struct file *fp, struct page *pp, int *codep)
+{
+ loff_t offset = page_offset(pp);
+ struct inode *ip = FILE_INODE(fp);
+ struct vcache *avc = VTOAFS(ip);
+ struct dcache *tdc;
+ struct file *cacheFp = NULL;
+ int code;
+ int dcLocked = 0;
+ struct pagevec lrupv;
+
+ /* Not a UFS cache, don't do anything */
+ if (cacheDiskType != AFS_FCACHE_TYPE_UFS)
+ return 0;
+
+ /* Can't do anything if the vcache isn't statd , or if the read
+ * crosses a chunk boundary.
+ */
+ if (!(avc->f.states & CStatd) ||
+ AFS_CHUNK(offset) != AFS_CHUNK(offset + PAGE_SIZE)) {
+ return 0;
+ }
+
+ ObtainWriteLock(&avc->lock, 911);
+
+ /* XXX - See if hinting actually makes things faster !!! */
+
+ /* See if we have a suitable entry already cached */
+ tdc = avc->dchint;
+
+ if (tdc) {
+ /* We need to lock xdcache, then dcache, to handle situations where
+ * the hint is on the free list. However, we can't safely do this
+ * according to the locking hierarchy. So, use a non blocking lock.
+ */
+ ObtainReadLock(&afs_xdcache);
+ dcLocked = ( 0 == NBObtainReadLock(&tdc->lock));
+
+ if (dcLocked && (tdc->index != NULLIDX)
+ && !FidCmp(&tdc->f.fid, &avc->f.fid)
+ && tdc->f.chunk == AFS_CHUNK(offset)
+ && !(afs_indexFlags[tdc->index] & (IFFree | IFDiscarded))) {
+ /* Bonus - the hint was correct */
+ afs_RefDCache(tdc);
+ } else {
+ /* Only destroy the hint if its actually invalid, not if there's
+ * just been a locking failure */
+ if (dcLocked) {
+ ReleaseReadLock(&tdc->lock);
+ avc->dchint = NULL;
+ }
+
+ tdc = NULL;
+ dcLocked = 0;
+ }
+ ReleaseReadLock(&afs_xdcache);
+ }
+
+ /* No hint, or hint is no longer valid - see if we can get something
+ * directly from the dcache
+ */
+ if (!tdc)
+ tdc = afs_FindDCache(avc, offset);
+
+ if (!tdc) {
+ ReleaseWriteLock(&avc->lock);
+ return 0;
+ }
+
+ if (!dcLocked)
+ ObtainReadLock(&tdc->lock);
+
+ /* Is the dcache we've been given currently up to date */
+ if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
+ (tdc->dflags & DFFetching)) {
+ ReleaseWriteLock(&avc->lock);
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ return 0;
+ }
+
+ /* Update our hint for future abuse */
+ avc->dchint = tdc;
+
+ /* Okay, so we've now got a cache file that is up to date */
+
+ /* XXX - I suspect we should be locking the inodes before we use them! */
+ AFS_GUNLOCK();
+ cacheFp = afs_linux_raw_open(&tdc->f.inode, NULL);
+ pagevec_init(&lrupv, 0);
+
+ code = afs_linux_read_cache(cacheFp, pp, tdc->f.chunk, &lrupv, NULL);
+
+ if (pagevec_count(&lrupv))
+ __pagevec_lru_add_file(&lrupv);
+
+ filp_close(cacheFp, NULL);
+ AFS_GLOCK();
+
+ ReleaseReadLock(&tdc->lock);
+ ReleaseWriteLock(&avc->lock);
+ afs_PutDCache(tdc);
+
+ *codep = code;
+ return 1;
+}
+
+/* afs_linux_readpage
+ *
+ * This function is split into two, because prepare_write/begin_write
+ * require a readpage call which doesn't unlock the resulting page upon
+ * success.
+ */
+static int
+afs_linux_fillpage(struct file *fp, struct page *pp)
+{
+ afs_int32 code;
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);
-#endif
- uio_t tuio;
- struct iovec iovec;
+ uio_t *auio;
+ struct iovec *iovecp;
struct inode *ip = FILE_INODE(fp);
- int cnt = page_count(pp);
+ afs_int32 cnt = page_count(pp);
struct vcache *avc = VTOAFS(ip);
+ afs_offs_t offset = page_offset(pp);
+ cred_t *credp;
+ AFS_GLOCK();
+ if (afs_linux_readpage_fastpath(fp, pp, &code)) {
+ AFS_GUNLOCK();
+ return code;
+ }
+ AFS_GUNLOCK();
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
+ credp = crref();
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);
-#ifdef AFS_LINUX24_ENV
- lock_kernel();
-#endif
+ 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);
+
+ afs_maybe_lock_kernel();
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_DISCON_LOCK();
+ 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_DISCON_UNLOCK();
AFS_GUNLOCK();
-#ifdef AFS_LINUX24_ENV
- unlock_kernel();
-#endif
-
+ afs_maybe_unlock_kernel();
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)
+ /* 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);
+
flush_dcache_page(pp);
SetPageUptodate(pp);
-#else
- set_bit(PG_uptodate, &pp->flags);
-#endif
- }
+ } /* !code */
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
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) {
+ osi_Free(auio, sizeof(uio_t));
+ osi_Free(iovecp, sizeof(struct iovec));
+
+ crfree(credp);
+ return afs_convert_code(code);
+}
+
+static int
+afs_linux_prefetch(struct file *fp, struct page *pp)
+{
+ int code = 0;
+ struct vcache *avc = VTOAFS(FILE_INODE(fp));
+ afs_offs_t offset = page_offset(pp);
+
+ if (AFS_CHUNKOFFSET(offset) == 0) {
struct dcache *tdc;
struct vrequest treq;
+ cred_t *credp;
+ credp = crref();
AFS_GLOCK();
code = afs_InitReq(&treq, credp);
if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
if (tdc) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, credp, &treq);
- afs_PutDCache(tdc);
+ afs_PutDCache(tdc);
}
ReleaseWriteLock(&avc->lock);
}
AFS_GUNLOCK();
+ crfree(credp);
}
+ return afs_convert_code(code);
+
+}
+
+#if defined(AFS_CACHE_BYPASS)
+
+static int
+afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
+ struct list_head *page_list, unsigned num_pages)
+{
+ afs_int32 page_ix;
+ uio_t *auio;
+ afs_offs_t offset;
+ struct iovec* iovecp;
+ struct nocache_read_request *ancr;
+ struct page *pp;
+ struct pagevec lrupv;
+ afs_int32 code = 0;
+
+ cred_t *credp;
+ struct inode *ip = FILE_INODE(fp);
+ struct vcache *avc = VTOAFS(ip);
+ afs_int32 base_index = 0;
+ afs_int32 page_count = 0;
+ afs_int32 isize;
+
+ /* 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))
+ unlock_page(pp);
+ continue;
+ }
+
+ if(page_ix == 0) {
+ offset = page_offset(pp);
+ 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))
+ unlock_page(pp);
+ page_cache_release(pp);
+ iovecp[page_ix].iov_base = (void *) 0;
+ base_index++;
+ continue;
+ }
+ base_index++;
+ if(code) {
+ if(PageLocked(pp))
+ unlock_page(pp);
+ page_cache_release(pp);
+ iovecp[page_ix].iov_base = (void *) 0;
+ } else {
+ page_count++;
+ if(!PageLocked(pp)) {
+ lock_page(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);
+ credp = crref();
+ code = afs_ReadNoCache(avc, ancr, credp);
+ crfree(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 */
+ return afs_convert_code(code);
+}
+
+
+static int
+afs_linux_bypass_readpage(struct file *fp, struct page *pp)
+{
+ cred_t *credp = NULL;
+ uio_t *auio;
+ struct iovec *iovecp;
+ struct nocache_read_request *ancr;
+ int code;
+
+ ClearPageError(pp);
+
+ /* receiver frees */
+ auio = osi_Alloc(sizeof(uio_t));
+ iovecp = osi_Alloc(sizeof(struct iovec));
+
+ /* address can be NULL, because we overwrite it with 'pp', below */
+ setup_uio(auio, iovecp, NULL, page_offset(pp),
+ PAGE_SIZE, UIO_READ, AFS_UIOSYS);
+
+ /* save the page for background map */
+ /* XXX - Shouldn't we get a reference count here? */
+ 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 = page_offset(pp);
+ ancr->length = PAGE_SIZE;
+
+ credp = crref();
+ afs_maybe_lock_kernel();
+ code = afs_ReadNoCache(VTOAFS(FILE_INODE(fp)), ancr, credp);
+ afs_maybe_unlock_kernel();
crfree(credp);
- return -code;
+
+ return afs_convert_code(code);
}
+static inline int
+afs_linux_can_bypass(struct inode *ip) {
+ switch(cache_bypass_strategy) {
+ case NEVER_BYPASS_CACHE:
+ return 0;
+ case ALWAYS_BYPASS_CACHE:
+ return 1;
+ case LARGE_FILES_BYPASS_CACHE:
+ if(i_size_read(ip) > cache_bypass_threshold)
+ return 1;
+ default:
+ return 0;
+ }
+}
+
+/* Check if a file is permitted to bypass the cache by policy, and modify
+ * the cache bypass state recorded for that file */
+
+static inline int
+afs_linux_bypass_check(struct inode *ip) {
+ cred_t* credp;
+
+ int bypass = afs_linux_can_bypass(ip);
+
+ credp = crref();
+ trydo_cache_transition(VTOAFS(ip), credp, bypass);
+ crfree(credp);
+
+ return bypass;
+}
+
+#else
+static inline int
+afs_linux_bypass_check(struct inode *ip) {
+ return 0;
+}
+static inline int
+afs_linux_bypass_readpage(struct file *fp, struct page *pp) {
+ return 0;
+}
+static inline int
+afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
+ struct list_head *page_list, unsigned int num_pages) {
+ return 0;
+}
+#endif
-#if defined(AFS_LINUX24_ENV)
static int
-afs_linux_writepage_sync(struct inode *ip, struct page *pp,
- unsigned long offset, unsigned int count)
+afs_linux_readpage(struct file *fp, struct page *pp)
+{
+ int code;
+
+ if (afs_linux_bypass_check(FILE_INODE(fp))) {
+ code = afs_linux_bypass_readpage(fp, pp);
+ } else {
+ code = afs_linux_fillpage(fp, pp);
+ if (!code)
+ code = afs_linux_prefetch(fp, pp);
+ unlock_page(pp);
+ }
+
+ return code;
+}
+
+/* Readpages reads a number of pages for a particular file. We use
+ * this to optimise the reading, by limiting the number of times upon which
+ * we have to lookup, lock and open vcaches and dcaches
+ */
+
+static int
+afs_linux_readpages(struct file *fp, struct address_space *mapping,
+ struct list_head *page_list, unsigned int num_pages)
+{
+ struct inode *inode = mapping->host;
+ struct vcache *avc = VTOAFS(inode);
+ struct dcache *tdc;
+ struct file *cacheFp = NULL;
+ int code;
+ unsigned int page_idx;
+ loff_t offset;
+ struct pagevec lrupv;
+ struct afs_pagecopy_task *task;
+
+ if (afs_linux_bypass_check(inode))
+ return afs_linux_bypass_readpages(fp, mapping, page_list, num_pages);
+
+ AFS_GLOCK();
+ if ((code = afs_linux_VerifyVCache(avc, NULL))) {
+ AFS_GUNLOCK();
+ return code;
+ }
+
+ ObtainWriteLock(&avc->lock, 912);
+ AFS_GUNLOCK();
+
+ task = afs_pagecopy_init_task();
+
+ tdc = NULL;
+ pagevec_init(&lrupv, 0);
+ for (page_idx = 0; page_idx < num_pages; page_idx++) {
+ struct page *page = list_entry(page_list->prev, struct page, lru);
+ list_del(&page->lru);
+ offset = page_offset(page);
+
+ if (tdc && tdc->f.chunk != AFS_CHUNK(offset)) {
+ AFS_GLOCK();
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ AFS_GUNLOCK();
+ tdc = NULL;
+ if (cacheFp)
+ filp_close(cacheFp, NULL);
+ }
+
+ if (!tdc) {
+ AFS_GLOCK();
+ if ((tdc = afs_FindDCache(avc, offset))) {
+ ObtainReadLock(&tdc->lock);
+ if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
+ (tdc->dflags & DFFetching)) {
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ tdc = NULL;
+ }
+ }
+ AFS_GUNLOCK();
+ if (tdc)
+ cacheFp = afs_linux_raw_open(&tdc->f.inode, NULL);
+ }
+
+ if (tdc && !add_to_page_cache(page, mapping, page->index,
+ GFP_KERNEL)) {
+ page_cache_get(page);
+ if (!pagevec_add(&lrupv, page))
+ __pagevec_lru_add_file(&lrupv);
+
+ afs_linux_read_cache(cacheFp, page, tdc->f.chunk, &lrupv, task);
+ }
+ page_cache_release(page);
+ }
+ if (pagevec_count(&lrupv))
+ __pagevec_lru_add_file(&lrupv);
+
+ if (tdc)
+ filp_close(cacheFp, NULL);
+
+ afs_pagecopy_put_task(task);
+
+ AFS_GLOCK();
+ if (tdc) {
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ }
+
+ ReleaseWriteLock(&avc->lock);
+ AFS_GUNLOCK();
+ return 0;
+}
+
+/* Prepare an AFS vcache for writeback. Should be called with the vcache
+ * locked */
+static inline int
+afs_linux_prepare_writeback(struct vcache *avc) {
+ if (avc->f.states & CPageWrite) {
+ return AOP_WRITEPAGE_ACTIVATE;
+ }
+ avc->f.states |= CPageWrite;
+ return 0;
+}
+
+static inline int
+afs_linux_dopartialwrite(struct vcache *avc, cred_t *credp) {
+ struct vrequest treq;
+ int code = 0;
+
+ if (!afs_InitReq(&treq, credp))
+ code = afs_DoPartialWrite(avc, &treq);
+
+ return afs_convert_code(code);
+}
+
+static inline void
+afs_linux_complete_writeback(struct vcache *avc) {
+ avc->f.states &= ~CPageWrite;
+}
+
+/* Writeback a given page syncronously. Called with no AFS locks held */
+static int
+afs_linux_page_writeback(struct inode *ip, struct page *pp,
+ unsigned long offset, unsigned int count,
+ cred_t *credp)
{
struct vcache *vcp = VTOAFS(ip);
char *buffer;
afs_offs_t base;
int code = 0;
- cred_t *credp;
uio_t tuio;
struct iovec iovec;
int f_flags = 0;
buffer = kmap(pp) + offset;
- base = (((loff_t) pp->index) << PAGE_CACHE_SHIFT) + offset;
+ base = page_offset(pp) + offset;
- credp = crref();
- lock_kernel();
+ afs_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),
code = afs_write(vcp, &tuio, f_flags, credp, 0);
- ip->i_size = vcp->m.Length;
- ip->i_blocks = ((vcp->m.Length + 1023) >> 10) << 1;
+ i_size_write(ip, vcp->f.m.Length);
+ ip->i_blocks = ((vcp->f.m.Length + 1023) >> 10) << 1;
- code = code ? -code : count - tuio.uio_resid;
+ code = code ? afs_convert_code(code) : count - tuio.uio_resid;
afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
ICL_TYPE_INT32, code);
AFS_GUNLOCK();
- unlock_kernel();
- crfree(credp);
+ afs_maybe_unlock_kernel();
kunmap(pp);
return code;
}
+static int
+afs_linux_writepage_sync(struct inode *ip, struct page *pp,
+ unsigned long offset, unsigned int count)
+{
+ int code;
+ int code1 = 0;
+ struct vcache *vcp = VTOAFS(ip);
+ cred_t *credp;
+
+ /* Catch recursive writeback. This occurs if the kernel decides
+ * writeback is required whilst we are writing to the cache, or
+ * flushing to the server. When we're running syncronously (as
+ * opposed to from writepage) we can't actually do anything about
+ * this case - as we can't return AOP_WRITEPAGE_ACTIVATE to write()
+ */
+ AFS_GLOCK();
+ ObtainWriteLock(&vcp->lock, 532);
+ afs_linux_prepare_writeback(vcp);
+ ReleaseWriteLock(&vcp->lock);
+ AFS_GUNLOCK();
+
+ credp = crref();
+ code = afs_linux_page_writeback(ip, pp, offset, count, credp);
+
+ afs_maybe_lock_kernel();
+ AFS_GLOCK();
+ ObtainWriteLock(&vcp->lock, 533);
+ if (code > 0)
+ code1 = afs_linux_dopartialwrite(vcp, credp);
+ afs_linux_complete_writeback(vcp);
+ ReleaseWriteLock(&vcp->lock);
+ AFS_GUNLOCK();
+ afs_maybe_unlock_kernel();
+ crfree(credp);
+
+ if (code1)
+ return code1;
+
+ return code;
+}
static int
#ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
{
struct address_space *mapping = pp->mapping;
struct inode *inode;
- unsigned long end_index;
- unsigned offset = PAGE_CACHE_SIZE;
- long status;
+ struct vcache *vcp;
+ cred_t *credp;
+ unsigned int to = PAGE_CACHE_SIZE;
+ loff_t isize;
+ int code = 0;
+ int code1 = 0;
-#if defined(AFS_LINUX26_ENV)
if (PageReclaim(pp)) {
- return WRITEPAGE_ACTIVATE;
- }
-#else
- if (PageLaunder(pp)) {
- return(fail_writepage(pp));
+ return AOP_WRITEPAGE_ACTIVATE;
+ /* XXX - Do we need to redirty the page here? */
}
-#endif
- inode = (struct inode *)mapping->host;
- end_index = inode->i_size >> PAGE_CACHE_SHIFT;
+ page_cache_get(pp);
- /* 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
- * What one would have thought was writepage - write dirty page to file.
- * 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.
- */
-static int
-afs_linux_updatepage(struct file *fp, struct page *pp, unsigned long offset,
- unsigned int count, int sync)
-{
- struct vcache *vcp = VTOAFS(FILE_INODE(fp));
- u8 *page_addr = (u8 *) afs_linux_page_address(pp);
- int code = 0;
- cred_t *credp;
- uio_t tuio;
- struct iovec iovec;
+ inode = mapping->host;
+ vcp = VTOAFS(inode);
+ isize = i_size_read(inode);
- set_bit(PG_locked, &pp->flags);
+ /* Don't defeat an earlier truncate */
+ if (page_offset(pp) > isize)
+ goto done;
- credp = crref();
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);
- setup_uio(&tuio, &iovec, page_addr + offset,
- (afs_offs_t) (pageoff(pp) + offset), count, UIO_WRITE,
- AFS_UIOSYS);
+ ObtainWriteLock(&vcp->lock, 537);
+ code = afs_linux_prepare_writeback(vcp);
+ if (code) {
+ ReleaseWriteLock(&vcp->lock);
+ AFS_GUNLOCK();
+ return code;
+ }
+ /* Grab the creds structure currently held in the vnode, and
+ * get a reference to it, in case it goes away ... */
+ credp = vcp->cred;
+ crhold(credp);
+ ReleaseWriteLock(&vcp->lock);
+ AFS_GUNLOCK();
- code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
+ /* If this is the final page, then just write the number of bytes that
+ * are actually in it */
+ if ((isize - page_offset(pp)) < to )
+ to = isize - page_offset(pp);
- ip->i_size = vcp->m.Length;
- ip->i_blocks = ((vcp->m.Length + 1023) >> 10) << 1;
+ code = afs_linux_page_writeback(inode, pp, 0, to, credp);
- code = code ? -code : count - tuio.uio_resid;
- afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
- ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
- ICL_TYPE_INT32, code);
+ afs_maybe_lock_kernel();
+ AFS_GLOCK();
+ ObtainWriteLock(&vcp->lock, 538);
- AFS_GUNLOCK();
+ /* As much as we might like to ignore a file server error here,
+ * and just try again when we close(), unfortunately StoreAllSegments
+ * will invalidate our chunks if the server returns a permanent error,
+ * so we need to at least try and get that error back to the user
+ */
+ if (code == to)
+ code1 = afs_linux_dopartialwrite(vcp, credp);
+
+ afs_linux_complete_writeback(vcp);
+ ReleaseWriteLock(&vcp->lock);
crfree(credp);
+ AFS_GUNLOCK();
+ afs_maybe_unlock_kernel();
+
+done:
+ SetPageUptodate(pp);
+ if ( code != AOP_WRITEPAGE_ACTIVATE ) {
+ /* XXX - do we need to redirty the page here? */
+ unlock_page(pp);
+ }
+
+ page_cache_release(pp);
+
+ if (code1)
+ return code1;
+
+ if (code == to)
+ return 0;
- clear_bit(PG_locked, &pp->flags);
return code;
}
-#endif
/* afs_linux_permission
* Check access rights - returns error if can't check or permission denied.
AFS_GUNLOCK();
crfree(credp);
- return -code;
+ return afs_convert_code(code);
}
-#if defined(AFS_LINUX24_ENV)
+#if !defined(HAVE_WRITE_BEGIN)
static int
afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
unsigned to)
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;
}
+#else
-extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
+static int
+afs_linux_write_end(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned len, unsigned copied,
+ struct page *page, void *fsdata)
+{
+ int code;
+ unsigned from = pos & (PAGE_CACHE_SIZE - 1);
+
+ code = afs_linux_writepage_sync(file->f_dentry->d_inode, page,
+ from, copied);
+ unlock_page(page);
+ page_cache_release(page);
+ return code;
+}
+
+static int
+afs_linux_write_begin(struct file *file, struct address_space *mapping,
+ loff_t pos, unsigned len, unsigned flags,
+ struct page **pagep, void **fsdata)
+{
+ struct page *page;
+ pgoff_t index = pos >> PAGE_CACHE_SHIFT;
+ page = grab_cache_page_write_begin(mapping, index, flags);
+ *pagep = page;
+
+ return 0;
+}
#endif
+
static struct inode_operations afs_file_iops = {
-#if defined(AFS_LINUX26_ENV)
.permission = afs_linux_permission,
.getattr = afs_linux_getattr,
.setattr = afs_notify_change,
-#elif defined(AFS_LINUX24_ENV)
- .permission = afs_linux_permission,
- .revalidate = afs_linux_revalidate,
- .setattr = afs_notify_change,
-#else
- .default_file_ops = &afs_file_fops,
- .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,
+ .readpages = afs_linux_readpages,
.writepage = afs_linux_writepage,
- .commit_write = afs_linux_commit_write,
- .prepare_write = afs_linux_prepare_write,
-};
+#if defined (HAVE_WRITE_BEGIN)
+ .write_begin = afs_linux_write_begin,
+ .write_end = afs_linux_write_end,
+#else
+ .commit_write = afs_linux_commit_write,
+ .prepare_write = afs_linux_prepare_write,
#endif
+};
/* Separate ops vector for directories. Linux 2.2 tests type of inode
*/
static struct inode_operations afs_dir_iops = {
-#if !defined(AFS_LINUX24_ENV)
- .default_file_ops = &afs_dir_fops,
-#else
.setattr = afs_notify_change,
-#endif
.create = afs_linux_create,
.lookup = afs_linux_lookup,
.link = afs_linux_link,
.mkdir = afs_linux_mkdir,
.rmdir = afs_linux_rmdir,
.rename = afs_linux_rename,
-#if defined(AFS_LINUX26_ENV)
.getattr = afs_linux_getattr,
-#else
- .revalidate = afs_linux_revalidate,
-#endif
.permission = afs_linux_permission,
};
/* 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();
+ afs_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();
+ afs_maybe_unlock_kernel();
SetPageUptodate(page);
kunmap(page);
- UnlockPage(page);
+ unlock_page(page);
return 0;
fail:
- unlock_kernel();
+ afs_maybe_unlock_kernel();
SetPageError(page);
kunmap(page);
- UnlockPage(page);
+ unlock_page(page);
return code;
}
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)
+# if defined(HAVE_KERNEL_PAGE_FOLLOW_LINK)
.follow_link = page_follow_link,
-#else
+# else
.follow_link = page_follow_link_light,
.put_link = page_put_link,
-#endif
- .setattr = afs_notify_change,
-#else
+# endif
+#else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
.readlink = afs_linux_readlink,
.follow_link = afs_linux_follow_link,
- .permission = afs_linux_permission,
- .revalidate = afs_linux_revalidate,
-#endif
+#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
+ .setattr = afs_notify_change,
};
void
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(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) */
}
git://git.openafs.org / openafs.git / blobdiff