2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
11 * Linux specific vnodeops. Also includes the glue routines required to call
14 * So far the only truly scary part is that Linux relies on the inode cache
15 * to be up to date. Don't you dare break a callback and expect an fstat
16 * to give you meaningful information. This appears to be fixed in the 2.1
17 * development kernels. As it is we can fix this now by intercepting the
21 #include <afsconfig.h>
22 #include "afs/param.h"
25 #include "afs/sysincludes.h"
26 #include "afsincludes.h"
27 #include "afs/afs_stats.h"
29 #ifdef HAVE_MM_INLINE_H
30 #include <linux/mm_inline.h>
32 #include <linux/pagemap.h>
33 #include <linux/smp_lock.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
37 #include "afs/afs_bypasscache.h"
39 #include "osi_compat.h"
40 #include "osi_pagecopy.h"
42 #ifndef HAVE_LINUX_PAGEVEC_LRU_ADD_FILE
43 #define __pagevec_lru_add_file __pagevec_lru_add
47 #define MAX_ERRNO 1000L
50 extern struct backing_dev_info *afs_backing_dev_info;
52 extern struct vcache *afs_globalVp;
53 extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
55 /* This function converts a positive error code from AFS into a negative
56 * code suitable for passing into the Linux VFS layer. It checks that the
57 * error code is within the permissable bounds for the ERR_PTR mechanism.
59 * _All_ error codes which come from the AFS layer should be passed through
60 * this function before being returned to the kernel.
64 afs_convert_code(int code) {
65 if ((code >= 0) && (code <= MAX_ERRNO))
71 /* Linux doesn't require a credp for many functions, and crref is an expensive
72 * operation. This helper function avoids obtaining it for VerifyVCache calls
76 afs_linux_VerifyVCache(struct vcache *avc, cred_t **retcred) {
81 if (avc->f.states & CStatd) {
89 code = afs_InitReq(&treq, credp);
91 code = afs_VerifyVCache2(avc, &treq);
98 return afs_convert_code(code);
102 afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
105 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
108 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
109 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
111 code = afs_linux_VerifyVCache(vcp, NULL);
114 /* Linux's FlushPages implementation doesn't ever use credp,
115 * so we optimise by not using it */
116 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
118 code = do_sync_read(fp, buf, count, offp);
122 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
123 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
130 /* Now we have integrated VM for writes as well as reads. generic_file_write
131 * also takes care of re-positioning the pointer if file is open in append
132 * mode. Call fake open/close to ensure we do writes of core dumps.
135 afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
138 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
143 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
144 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
145 (fp->f_flags & O_APPEND) ? 99998 : 99999);
147 code = afs_linux_VerifyVCache(vcp, &credp);
149 ObtainWriteLock(&vcp->lock, 529);
151 ReleaseWriteLock(&vcp->lock);
154 code = do_sync_write(fp, buf, count, offp);
158 ObtainWriteLock(&vcp->lock, 530);
160 if (vcp->execsOrWriters == 1 && !credp)
163 afs_FakeClose(vcp, credp);
164 ReleaseWriteLock(&vcp->lock);
166 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
167 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
176 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
178 /* This is a complete rewrite of afs_readdir, since we can make use of
179 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
180 * handling and use of bulkstats will need to be reflected here as well.
183 afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
185 struct vcache *avc = VTOAFS(FILE_INODE(fp));
186 struct vrequest treq;
194 afs_size_t origOffset, tlen;
195 cred_t *credp = crref();
196 struct afs_fakestat_state fakestat;
199 AFS_STATCNT(afs_readdir);
201 code = afs_convert_code(afs_InitReq(&treq, credp));
206 afs_InitFakeStat(&fakestat);
207 code = afs_convert_code(afs_EvalFakeStat(&avc, &fakestat, &treq));
211 /* update the cache entry */
213 code = afs_convert_code(afs_VerifyVCache2(avc, &treq));
217 /* get a reference to the entire directory */
218 tdc = afs_GetDCache(avc, (afs_size_t) 0, &treq, &origOffset, &tlen, 1);
224 ObtainSharedLock(&avc->lock, 810);
225 UpgradeSToWLock(&avc->lock, 811);
226 ObtainReadLock(&tdc->lock);
228 * Make sure that the data in the cache is current. There are two
229 * cases we need to worry about:
230 * 1. The cache data is being fetched by another process.
231 * 2. The cache data is no longer valid
233 while ((avc->f.states & CStatd)
234 && (tdc->dflags & DFFetching)
235 && hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
236 ReleaseReadLock(&tdc->lock);
237 ReleaseSharedLock(&avc->lock);
238 afs_osi_Sleep(&tdc->validPos);
239 ObtainSharedLock(&avc->lock, 812);
240 ObtainReadLock(&tdc->lock);
242 if (!(avc->f.states & CStatd)
243 || !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
244 ReleaseReadLock(&tdc->lock);
245 ReleaseSharedLock(&avc->lock);
250 /* Set the readdir-in-progress flag, and downgrade the lock
251 * to shared so others will be able to acquire a read lock.
253 avc->f.states |= CReadDir;
254 avc->dcreaddir = tdc;
255 avc->readdir_pid = MyPidxx2Pid(MyPidxx);
256 ConvertWToSLock(&avc->lock);
258 /* Fill in until we get an error or we're done. This implementation
259 * takes an offset in units of blobs, rather than bytes.
262 offset = (int) fp->f_pos;
264 dirpos = BlobScan(tdc, offset);
268 de = afs_dir_GetBlob(tdc, dirpos);
272 ino = afs_calc_inum (avc->f.fid.Fid.Volume, ntohl(de->fid.vnode));
275 len = strlen(de->name);
277 printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
278 (unsigned long)&tdc->f.inode, dirpos);
280 ReleaseSharedLock(&avc->lock);
286 /* filldir returns -EINVAL when the buffer is full. */
288 unsigned int type = DT_UNKNOWN;
289 struct VenusFid afid;
292 afid.Cell = avc->f.fid.Cell;
293 afid.Fid.Volume = avc->f.fid.Fid.Volume;
294 afid.Fid.Vnode = ntohl(de->fid.vnode);
295 afid.Fid.Unique = ntohl(de->fid.vunique);
296 if ((avc->f.states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
298 } else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
301 } else if (((tvc->f.states) & (CStatd | CTruth))) {
302 /* CTruth will be set if the object has
307 else if (vtype == VREG)
309 /* Don't do this until we're sure it can't be a mtpt */
310 /* else if (vtype == VLNK)
312 /* what other types does AFS support? */
314 /* clean up from afs_FindVCache */
318 * If this is NFS readdirplus, then the filler is going to
319 * call getattr on this inode, which will deadlock if we're
323 code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
329 offset = dirpos + 1 + ((len + 16) >> 5);
331 /* If filldir didn't fill in the last one this is still pointing to that
334 fp->f_pos = (loff_t) offset;
336 ReleaseReadLock(&tdc->lock);
338 UpgradeSToWLock(&avc->lock, 813);
339 avc->f.states &= ~CReadDir;
341 avc->readdir_pid = 0;
342 ReleaseSharedLock(&avc->lock);
346 afs_PutFakeStat(&fakestat);
353 /* in afs_pioctl.c */
354 extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
357 #if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
358 static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
360 return afs_xioctl(FILE_INODE(fp), fp, com, arg);
367 afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
369 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
373 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
374 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
375 vmap->vm_end - vmap->vm_start);
377 /* get a validated vcache entry */
378 code = afs_linux_VerifyVCache(vcp, NULL);
381 /* Linux's Flushpage implementation doesn't use credp, so optimise
382 * our code to not need to crref() it */
383 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
385 code = generic_file_mmap(fp, vmap);
388 vcp->f.states |= CMAPPED;
396 afs_linux_open(struct inode *ip, struct file *fp)
398 struct vcache *vcp = VTOAFS(ip);
399 cred_t *credp = crref();
403 code = afs_open(&vcp, fp->f_flags, credp);
407 return afs_convert_code(code);
411 afs_linux_release(struct inode *ip, struct file *fp)
413 struct vcache *vcp = VTOAFS(ip);
414 cred_t *credp = crref();
418 code = afs_close(vcp, fp->f_flags, credp);
419 ObtainWriteLock(&vcp->lock, 807);
424 ReleaseWriteLock(&vcp->lock);
428 return afs_convert_code(code);
432 #if defined(FOP_FSYNC_TAKES_DENTRY)
433 afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
435 afs_linux_fsync(struct file *fp, int datasync)
439 struct inode *ip = FILE_INODE(fp);
440 cred_t *credp = crref();
443 code = afs_fsync(VTOAFS(ip), credp);
446 return afs_convert_code(code);
452 afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
455 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
456 cred_t *credp = crref();
457 struct AFS_FLOCK flock;
459 /* Convert to a lock format afs_lockctl understands. */
460 memset(&flock, 0, sizeof(flock));
461 flock.l_type = flp->fl_type;
462 flock.l_pid = flp->fl_pid;
464 flock.l_start = flp->fl_start;
465 if (flp->fl_end == OFFSET_MAX)
466 flock.l_len = 0; /* Lock to end of file */
468 flock.l_len = flp->fl_end - flp->fl_start + 1;
470 /* Safe because there are no large files, yet */
471 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
472 if (cmd == F_GETLK64)
474 else if (cmd == F_SETLK64)
476 else if (cmd == F_SETLKW64)
478 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
481 code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
484 if ((code == 0 || flp->fl_type == F_UNLCK) &&
485 (cmd == F_SETLK || cmd == F_SETLKW)) {
486 code = afs_posix_lock_file(fp, flp);
487 if (code && flp->fl_type != F_UNLCK) {
488 struct AFS_FLOCK flock2;
490 flock2.l_type = F_UNLCK;
492 afs_lockctl(vcp, &flock2, F_SETLK, credp);
496 /* If lockctl says there are no conflicting locks, then also check with the
497 * kernel, as lockctl knows nothing about byte range locks
499 if (code == 0 && cmd == F_GETLK && flock.l_type == F_UNLCK) {
500 afs_posix_test_lock(fp, flp);
501 /* If we found a lock in the kernel's structure, return it */
502 if (flp->fl_type != F_UNLCK) {
508 /* Convert flock back to Linux's file_lock */
509 flp->fl_type = flock.l_type;
510 flp->fl_pid = flock.l_pid;
511 flp->fl_start = flock.l_start;
512 if (flock.l_len == 0)
513 flp->fl_end = OFFSET_MAX; /* Lock to end of file */
515 flp->fl_end = flock.l_start + flock.l_len - 1;
521 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
523 afs_linux_flock(struct file *fp, int cmd, struct file_lock *flp) {
525 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
526 cred_t *credp = crref();
527 struct AFS_FLOCK flock;
528 /* Convert to a lock format afs_lockctl understands. */
529 memset(&flock, 0, sizeof(flock));
530 flock.l_type = flp->fl_type;
531 flock.l_pid = flp->fl_pid;
536 /* Safe because there are no large files, yet */
537 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
538 if (cmd == F_GETLK64)
540 else if (cmd == F_SETLK64)
542 else if (cmd == F_SETLKW64)
544 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
547 code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
550 if ((code == 0 || flp->fl_type == F_UNLCK) &&
551 (cmd == F_SETLK || cmd == F_SETLKW)) {
552 flp->fl_flags &=~ FL_SLEEP;
553 code = flock_lock_file_wait(fp, flp);
554 if (code && flp->fl_type != F_UNLCK) {
555 struct AFS_FLOCK flock2;
557 flock2.l_type = F_UNLCK;
559 afs_lockctl(vcp, &flock2, F_SETLK, credp);
563 /* Convert flock back to Linux's file_lock */
564 flp->fl_type = flock.l_type;
565 flp->fl_pid = flock.l_pid;
573 * essentially the same as afs_fsync() but we need to get the return
574 * code for the sys_close() here, not afs_linux_release(), so call
575 * afs_StoreAllSegments() with AFS_LASTSTORE
578 #if defined(FOP_FLUSH_TAKES_FL_OWNER_T)
579 afs_linux_flush(struct file *fp, fl_owner_t id)
581 afs_linux_flush(struct file *fp)
584 struct vrequest treq;
592 if ((fp->f_flags & O_ACCMODE) == O_RDONLY) { /* readers dont flush */
600 vcp = VTOAFS(FILE_INODE(fp));
602 code = afs_InitReq(&treq, credp);
605 /* If caching is bypassed for this file, or globally, just return 0 */
606 if (cache_bypass_strategy == ALWAYS_BYPASS_CACHE)
609 ObtainReadLock(&vcp->lock);
610 if (vcp->cachingStates & FCSBypass)
612 ReleaseReadLock(&vcp->lock);
615 /* future proof: don't rely on 0 return from afs_InitReq */
620 ObtainSharedLock(&vcp->lock, 535);
621 if ((vcp->execsOrWriters > 0) && (file_count(fp) == 1)) {
622 UpgradeSToWLock(&vcp->lock, 536);
623 if (!AFS_IS_DISCONNECTED) {
624 code = afs_StoreAllSegments(vcp,
626 AFS_SYNC | AFS_LASTSTORE);
628 afs_DisconAddDirty(vcp, VDisconWriteOsiFlush, 1);
630 ConvertWToSLock(&vcp->lock);
632 code = afs_CheckCode(code, &treq, 54);
633 ReleaseSharedLock(&vcp->lock);
640 return afs_convert_code(code);
643 struct file_operations afs_dir_fops = {
644 .read = generic_read_dir,
645 .readdir = afs_linux_readdir,
646 #ifdef HAVE_UNLOCKED_IOCTL
647 .unlocked_ioctl = afs_unlocked_xioctl,
651 #ifdef HAVE_COMPAT_IOCTL
652 .compat_ioctl = afs_unlocked_xioctl,
654 .open = afs_linux_open,
655 .release = afs_linux_release,
656 .llseek = default_llseek,
659 struct file_operations afs_file_fops = {
660 .read = afs_linux_read,
661 .write = afs_linux_write,
662 #ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
663 .aio_read = generic_file_aio_read,
664 .aio_write = generic_file_aio_write,
666 #ifdef HAVE_UNLOCKED_IOCTL
667 .unlocked_ioctl = afs_unlocked_xioctl,
671 #ifdef HAVE_COMPAT_IOCTL
672 .compat_ioctl = afs_unlocked_xioctl,
674 .mmap = afs_linux_mmap,
675 .open = afs_linux_open,
676 .flush = afs_linux_flush,
677 #if defined(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
678 .sendfile = generic_file_sendfile,
680 #if defined(STRUCT_FILE_OPERATIONS_HAS_SPLICE)
681 .splice_write = generic_file_splice_write,
682 .splice_read = generic_file_splice_read,
684 .release = afs_linux_release,
685 .fsync = afs_linux_fsync,
686 .lock = afs_linux_lock,
687 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
688 .flock = afs_linux_flock,
690 .llseek = default_llseek,
694 /**********************************************************************
695 * AFS Linux dentry operations
696 **********************************************************************/
698 /* fix_bad_parent() : called if this dentry's vcache is a root vcache
699 * that has its mvid (parent dir's fid) pointer set to the wrong directory
700 * due to being mounted in multiple points at once. fix_bad_parent()
701 * calls afs_lookup() to correct the vcache's mvid, as well as the volume's
702 * dotdotfid and mtpoint fid members.
704 * dp - dentry to be checked.
705 * credp - credentials
706 * vcp, pvc - item's and parent's vcache pointer
710 * This dentry's vcache's mvid will be set to the correct parent directory's
712 * This root vnode's volume will have its dotdotfid and mtpoint fids set
713 * to the correct parent and mountpoint fids.
717 fix_bad_parent(struct dentry *dp, cred_t *credp, struct vcache *vcp, struct vcache *pvc)
719 struct vcache *avc = NULL;
721 /* force a lookup, so vcp->mvid is fixed up */
722 afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
723 if (!avc || vcp != avc) { /* bad, very bad.. */
724 afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
725 "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
726 ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
727 ICL_TYPE_POINTER, dp);
730 AFS_RELE(AFSTOV(avc));
735 /* afs_linux_revalidate
736 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
739 afs_linux_revalidate(struct dentry *dp)
742 struct vcache *vcp = VTOAFS(dp->d_inode);
746 if (afs_shuttingdown)
752 /* Make this a fast path (no crref), since it's called so often. */
753 if (vcp->states & CStatd) {
754 struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
756 if (*dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
757 if (vcp->mvid->Fid.Volume != pvc->fid.Fid.Volume) { /* bad parent */
760 fix_bad_parent(dp); /* check and correct mvid */
769 /* This avoids the crref when we don't have to do it. Watch for
770 * changes in afs_getattr that don't get replicated here!
772 if (vcp->f.states & CStatd &&
773 (!afs_fakestat_enable || vcp->mvstat != 1) &&
775 (vType(vcp) == VDIR || vType(vcp) == VLNK)) {
776 code = afs_CopyOutAttrs(vcp, &vattr);
779 code = afs_getattr(vcp, &vattr, credp);
784 afs_fill_inode(AFSTOV(vcp), &vattr);
788 return afs_convert_code(code);
792 afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
794 int err = afs_linux_revalidate(dentry);
796 generic_fillattr(dentry->d_inode, stat);
801 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
802 * In kernels 2.2.10 and above, we are passed an additional flags var which
803 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
804 * we are advised to follow the entry if it is a link or to make sure that
805 * it is a directory. But since the kernel itself checks these possibilities
806 * later on, we shouldn't have to do it until later. Perhaps in the future..
808 * The code here assumes that on entry the global lock is not held
811 #ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
812 afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
814 afs_linux_dentry_revalidate(struct dentry *dp, int flags)
818 cred_t *credp = NULL;
819 struct vcache *vcp, *pvcp, *tvc = NULL;
820 struct dentry *parent;
822 struct afs_fakestat_state fakestate;
825 afs_InitFakeStat(&fakestate);
828 vcp = VTOAFS(dp->d_inode);
830 if (vcp == afs_globalVp)
833 parent = dget_parent(dp);
834 pvcp = VTOAFS(parent->d_inode);
836 if ((vcp->mvstat == 1) || (vcp->mvstat == 2)) { /* need to lock */
842 if (locked && vcp->mvstat == 1) { /* mount point */
843 if (vcp->mvid && (vcp->f.states & CMValid)) {
846 struct vrequest treq;
848 code = afs_InitReq(&treq, credp);
850 (strcmp(dp->d_name.name, ".directory") == 0)) {
854 code = afs_TryEvalFakeStat(&vcp, &fakestate, &treq);
856 code = afs_EvalFakeStat(&vcp, &fakestate, &treq);
857 if ((tryEvalOnly && vcp->mvstat == 1) || code) {
858 /* a mount point, not yet replaced by its directory */
863 if (locked && *dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
864 if (vcp->mvid->Fid.Volume != pvcp->f.fid.Fid.Volume) { /* bad parent */
865 fix_bad_parent(dp, credp, vcp, pvcp); /* check and correct mvid */
870 /* If the last looker changes, we should make sure the current
871 * looker still has permission to examine this file. This would
872 * always require a crref() which would be "slow".
874 if (vcp->last_looker != treq.uid) {
875 if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS))
878 vcp->last_looker = treq.uid;
883 /* If the parent's DataVersion has changed or the vnode
884 * is longer valid, we need to do a full lookup. VerifyVCache
885 * isn't enough since the vnode may have been renamed.
888 if ((!locked) && (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd)) ) {
894 if (locked && (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd))) {
895 afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
896 if (!tvc || tvc != vcp) {
901 if (afs_getattr(vcp, &vattr, credp)) {
906 vattr2inode(AFSTOV(vcp), &vattr);
907 dp->d_time = hgetlo(pvcp->f.m.DataVersion);
910 /* should we always update the attributes at this point? */
911 /* unlikely--the vcache entry hasn't changed */
916 /* If this code is ever enabled, we should use dget_parent to handle
917 * getting the parent, and dput() to dispose of it. See above for an
919 pvcp = VTOAFS(dp->d_parent->d_inode);
920 if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time)
924 /* No change in parent's DataVersion so this negative
925 * lookup is still valid. BUT, if a server is down a
926 * negative lookup can result so there should be a
927 * liftime as well. For now, always expire.
940 afs_PutFakeStat(&fakestate); /* from here on vcp may be no longer valid */
942 /* we hold the global lock if we evaluated a mount point */
949 shrink_dcache_parent(dp);
955 if (have_submounts(dp))
963 afs_dentry_iput(struct dentry *dp, struct inode *ip)
965 struct vcache *vcp = VTOAFS(ip);
968 if (!AFS_IS_DISCONNECTED || (vcp->f.states & CUnlinked)) {
969 (void) afs_InactiveVCache(vcp, NULL);
972 afs_linux_clear_nfsfs_renamed(dp);
978 afs_dentry_delete(struct dentry *dp)
980 if (dp->d_inode && (VTOAFS(dp->d_inode)->f.states & CUnlinked))
981 return 1; /* bad inode? */
986 struct dentry_operations afs_dentry_operations = {
987 .d_revalidate = afs_linux_dentry_revalidate,
988 .d_delete = afs_dentry_delete,
989 .d_iput = afs_dentry_iput,
992 /**********************************************************************
993 * AFS Linux inode operations
994 **********************************************************************/
998 * Merely need to set enough of vattr to get us through the create. Note
999 * that the higher level code (open_namei) will take care of any tuncation
1000 * explicitly. Exclusive open is also taken care of in open_namei.
1002 * name is in kernel space at this point.
1005 #ifdef IOP_CREATE_TAKES_NAMEIDATA
1006 afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
1007 struct nameidata *nd)
1009 afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
1013 cred_t *credp = crref();
1014 const char *name = dp->d_name.name;
1019 vattr.va_mode = mode;
1020 vattr.va_type = mode & S_IFMT;
1023 code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
1027 struct inode *ip = AFSTOV(vcp);
1029 afs_getattr(vcp, &vattr, credp);
1030 afs_fill_inode(ip, &vattr);
1031 insert_inode_hash(ip);
1032 dp->d_op = &afs_dentry_operations;
1033 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1034 d_instantiate(dp, ip);
1039 return afs_convert_code(code);
1042 /* afs_linux_lookup */
1043 static struct dentry *
1044 #ifdef IOP_LOOKUP_TAKES_NAMEIDATA
1045 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1046 struct nameidata *nd)
1048 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1051 cred_t *credp = crref();
1052 struct vcache *vcp = NULL;
1053 const char *comp = dp->d_name.name;
1054 struct inode *ip = NULL;
1055 struct dentry *newdp = NULL;
1059 code = afs_lookup(VTOAFS(dip), (char *)comp, &vcp, credp);
1065 afs_getattr(vcp, &vattr, credp);
1066 afs_fill_inode(ip, &vattr);
1067 if (hlist_unhashed(&ip->i_hash))
1068 insert_inode_hash(ip);
1070 dp->d_op = &afs_dentry_operations;
1071 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1074 if (ip && S_ISDIR(ip->i_mode)) {
1075 struct dentry *alias;
1077 /* Try to invalidate an existing alias in favor of our new one */
1078 alias = d_find_alias(ip);
1079 /* But not if it's disconnected; then we want d_splice_alias below */
1080 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
1081 if (d_invalidate(alias) == 0) {
1090 newdp = d_splice_alias(ip, dp);
1094 /* It's ok for the file to not be found. That's noted by the caller by
1095 * seeing that the dp->d_inode field is NULL.
1097 if (!code || code == ENOENT)
1100 return ERR_PTR(afs_convert_code(code));
1104 afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
1107 cred_t *credp = crref();
1108 const char *name = newdp->d_name.name;
1109 struct inode *oldip = olddp->d_inode;
1111 /* If afs_link returned the vnode, we could instantiate the
1112 * dentry. Since it's not, we drop this one and do a new lookup.
1117 code = afs_link(VTOAFS(oldip), VTOAFS(dip), (char *)name, credp);
1121 return afs_convert_code(code);
1124 /* We have to have a Linux specific sillyrename function, because we
1125 * also have to keep the dcache up to date when we're doing a silly
1126 * rename - so we don't want the generic vnodeops doing this behind our
1131 afs_linux_sillyrename(struct inode *dir, struct dentry *dentry,
1134 struct vcache *tvc = VTOAFS(dentry->d_inode);
1135 struct dentry *__dp = NULL;
1136 char *__name = NULL;
1139 if (afs_linux_nfsfs_renamed(dentry))
1147 osi_FreeSmallSpace(__name);
1148 __name = afs_newname();
1151 __dp = lookup_one_len(__name, dentry->d_parent, strlen(__name));
1154 osi_FreeSmallSpace(__name);
1157 } while (__dp->d_inode != NULL);
1160 code = afs_rename(VTOAFS(dir), (char *)dentry->d_name.name,
1161 VTOAFS(dir), (char *)__dp->d_name.name,
1164 tvc->mvid = (void *) __name;
1167 crfree(tvc->uncred);
1169 tvc->uncred = credp;
1170 tvc->f.states |= CUnlinked;
1171 afs_linux_set_nfsfs_renamed(dentry);
1173 osi_FreeSmallSpace(__name);
1178 __dp->d_time = hgetlo(VTOAFS(dir)->f.m.DataVersion);
1179 d_move(dentry, __dp);
1188 afs_linux_unlink(struct inode *dip, struct dentry *dp)
1191 cred_t *credp = crref();
1192 const char *name = dp->d_name.name;
1193 struct vcache *tvc = VTOAFS(dp->d_inode);
1195 if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
1196 && !(tvc->f.states & CUnlinked)) {
1198 code = afs_linux_sillyrename(dip, dp, credp);
1201 code = afs_remove(VTOAFS(dip), (char *)name, credp);
1208 return afs_convert_code(code);
1213 afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
1216 cred_t *credp = crref();
1218 const char *name = dp->d_name.name;
1220 /* If afs_symlink returned the vnode, we could instantiate the
1221 * dentry. Since it's not, we drop this one and do a new lookup.
1227 code = afs_symlink(VTOAFS(dip), (char *)name, &vattr, (char *)target, credp);
1230 return afs_convert_code(code);
1234 afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1237 cred_t *credp = crref();
1238 struct vcache *tvcp = NULL;
1240 const char *name = dp->d_name.name;
1243 vattr.va_mask = ATTR_MODE;
1244 vattr.va_mode = mode;
1246 code = afs_mkdir(VTOAFS(dip), (char *)name, &vattr, &tvcp, credp);
1249 struct inode *ip = AFSTOV(tvcp);
1251 afs_getattr(tvcp, &vattr, credp);
1252 afs_fill_inode(ip, &vattr);
1254 dp->d_op = &afs_dentry_operations;
1255 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1256 d_instantiate(dp, ip);
1261 return afs_convert_code(code);
1265 afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1268 cred_t *credp = crref();
1269 const char *name = dp->d_name.name;
1271 /* locking kernel conflicts with glock? */
1274 code = afs_rmdir(VTOAFS(dip), (char *)name, credp);
1277 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1278 * that failed because a directory is not empty. So, we map
1279 * EEXIST to ENOTEMPTY on linux.
1281 if (code == EEXIST) {
1290 return afs_convert_code(code);
1295 afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1296 struct inode *newip, struct dentry *newdp)
1299 cred_t *credp = crref();
1300 const char *oldname = olddp->d_name.name;
1301 const char *newname = newdp->d_name.name;
1302 struct dentry *rehash = NULL;
1304 /* Prevent any new references during rename operation. */
1306 if (!d_unhashed(newdp)) {
1311 if (atomic_read(&olddp->d_count) > 1)
1312 shrink_dcache_parent(olddp);
1315 code = afs_rename(VTOAFS(oldip), (char *)oldname, VTOAFS(newip), (char *)newname, credp);
1319 olddp->d_time = 0; /* force to revalidate */
1325 return afs_convert_code(code);
1329 /* afs_linux_ireadlink
1330 * Internal readlink which can return link contents to user or kernel space.
1331 * Note that the buffer is NOT supposed to be null-terminated.
1334 afs_linux_ireadlink(struct inode *ip, char *target, int maxlen, uio_seg_t seg)
1337 cred_t *credp = crref();
1341 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1342 code = afs_readlink(VTOAFS(ip), &tuio, credp);
1346 return maxlen - tuio.uio_resid;
1348 return afs_convert_code(code);
1351 #if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1352 /* afs_linux_readlink
1353 * Fill target (which is in user space) with contents of symlink.
1356 afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1359 struct inode *ip = dp->d_inode;
1362 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1368 /* afs_linux_follow_link
1369 * a file system dependent link following routine.
1371 static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
1376 name = osi_Alloc(PATH_MAX);
1382 code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
1390 code = vfs_follow_link(nd, name);
1393 osi_Free(name, PATH_MAX);
1398 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1400 /* Populate a page by filling it from the cache file pointed at by cachefp
1401 * (which contains indicated chunk)
1402 * If task is NULL, the page copy occurs syncronously, and the routine
1403 * returns with page still locked. If task is non-NULL, then page copies
1404 * may occur in the background, and the page will be unlocked when it is
1408 afs_linux_read_cache(struct file *cachefp, struct page *page,
1409 int chunk, struct pagevec *lrupv,
1410 struct afs_pagecopy_task *task) {
1411 loff_t offset = page_offset(page);
1412 struct page *newpage, *cachepage;
1413 struct address_space *cachemapping;
1417 cachemapping = cachefp->f_dentry->d_inode->i_mapping;
1421 /* From our offset, we now need to work out which page in the disk
1422 * file it corresponds to. This will be fun ... */
1423 pageindex = (offset - AFS_CHUNKTOBASE(chunk)) >> PAGE_CACHE_SHIFT;
1425 while (cachepage == NULL) {
1426 cachepage = find_get_page(cachemapping, pageindex);
1429 newpage = page_cache_alloc_cold(cachemapping);
1435 code = add_to_page_cache(newpage, cachemapping,
1436 pageindex, GFP_KERNEL);
1438 cachepage = newpage;
1441 page_cache_get(cachepage);
1442 if (!pagevec_add(lrupv, cachepage))
1443 __pagevec_lru_add_file(lrupv);
1446 page_cache_release(newpage);
1448 if (code != -EEXIST)
1452 lock_page(cachepage);
1456 if (!PageUptodate(cachepage)) {
1457 ClearPageError(cachepage);
1458 code = cachemapping->a_ops->readpage(NULL, cachepage);
1459 if (!code && !task) {
1460 wait_on_page_locked(cachepage);
1463 unlock_page(cachepage);
1467 if (PageUptodate(cachepage)) {
1468 copy_highpage(page, cachepage);
1469 flush_dcache_page(page);
1470 SetPageUptodate(page);
1475 afs_pagecopy_queue_page(task, cachepage, page);
1487 page_cache_release(cachepage);
1493 afs_linux_readpage_fastpath(struct file *fp, struct page *pp, int *codep)
1495 loff_t offset = page_offset(pp);
1496 struct inode *ip = FILE_INODE(fp);
1497 struct vcache *avc = VTOAFS(ip);
1499 struct file *cacheFp = NULL;
1502 struct pagevec lrupv;
1504 /* Not a UFS cache, don't do anything */
1505 if (cacheDiskType != AFS_FCACHE_TYPE_UFS)
1508 /* Can't do anything if the vcache isn't statd , or if the read
1509 * crosses a chunk boundary.
1511 if (!(avc->f.states & CStatd) ||
1512 AFS_CHUNK(offset) != AFS_CHUNK(offset + PAGE_SIZE)) {
1516 ObtainWriteLock(&avc->lock, 911);
1518 /* XXX - See if hinting actually makes things faster !!! */
1520 /* See if we have a suitable entry already cached */
1524 /* We need to lock xdcache, then dcache, to handle situations where
1525 * the hint is on the free list. However, we can't safely do this
1526 * according to the locking hierarchy. So, use a non blocking lock.
1528 ObtainReadLock(&afs_xdcache);
1529 dcLocked = ( 0 == NBObtainReadLock(&tdc->lock));
1531 if (dcLocked && (tdc->index != NULLIDX)
1532 && !FidCmp(&tdc->f.fid, &avc->f.fid)
1533 && tdc->f.chunk == AFS_CHUNK(offset)
1534 && !(afs_indexFlags[tdc->index] & (IFFree | IFDiscarded))) {
1535 /* Bonus - the hint was correct */
1538 /* Only destroy the hint if its actually invalid, not if there's
1539 * just been a locking failure */
1541 ReleaseReadLock(&tdc->lock);
1548 ReleaseReadLock(&afs_xdcache);
1551 /* No hint, or hint is no longer valid - see if we can get something
1552 * directly from the dcache
1555 tdc = afs_FindDCache(avc, offset);
1558 ReleaseWriteLock(&avc->lock);
1563 ObtainReadLock(&tdc->lock);
1565 /* Is the dcache we've been given currently up to date */
1566 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
1567 (tdc->dflags & DFFetching)) {
1568 ReleaseWriteLock(&avc->lock);
1569 ReleaseReadLock(&tdc->lock);
1574 /* Update our hint for future abuse */
1577 /* Okay, so we've now got a cache file that is up to date */
1579 /* XXX - I suspect we should be locking the inodes before we use them! */
1581 cacheFp = afs_linux_raw_open(&tdc->f.inode);
1582 pagevec_init(&lrupv, 0);
1584 code = afs_linux_read_cache(cacheFp, pp, tdc->f.chunk, &lrupv, NULL);
1586 if (pagevec_count(&lrupv))
1587 __pagevec_lru_add_file(&lrupv);
1589 filp_close(cacheFp, NULL);
1592 ReleaseReadLock(&tdc->lock);
1593 ReleaseWriteLock(&avc->lock);
1600 /* afs_linux_readpage
1602 * This function is split into two, because prepare_write/begin_write
1603 * require a readpage call which doesn't unlock the resulting page upon
1607 afs_linux_fillpage(struct file *fp, struct page *pp)
1612 struct iovec *iovecp;
1613 struct inode *ip = FILE_INODE(fp);
1614 afs_int32 cnt = page_count(pp);
1615 struct vcache *avc = VTOAFS(ip);
1616 afs_offs_t offset = page_offset(pp);
1620 if (afs_linux_readpage_fastpath(fp, pp, &code)) {
1630 auio = osi_Alloc(sizeof(uio_t));
1631 iovecp = osi_Alloc(sizeof(struct iovec));
1633 setup_uio(auio, iovecp, (char *)address, offset, PAGE_SIZE, UIO_READ,
1638 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
1639 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
1640 99999); /* not a possible code value */
1642 code = afs_rdwr(avc, auio, UIO_READ, 0, credp);
1644 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
1645 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
1647 AFS_DISCON_UNLOCK();
1650 /* XXX valid for no-cache also? Check last bits of files... :)
1651 * Cognate code goes in afs_NoCacheFetchProc. */
1652 if (auio->uio_resid) /* zero remainder of page */
1653 memset((void *)(address + (PAGE_SIZE - auio->uio_resid)), 0,
1656 flush_dcache_page(pp);
1657 SetPageUptodate(pp);
1662 osi_Free(auio, sizeof(uio_t));
1663 osi_Free(iovecp, sizeof(struct iovec));
1666 return afs_convert_code(code);
1670 afs_linux_prefetch(struct file *fp, struct page *pp)
1673 struct vcache *avc = VTOAFS(FILE_INODE(fp));
1674 afs_offs_t offset = page_offset(pp);
1676 if (AFS_CHUNKOFFSET(offset) == 0) {
1678 struct vrequest treq;
1683 code = afs_InitReq(&treq, credp);
1684 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
1685 tdc = afs_FindDCache(avc, offset);
1687 if (!(tdc->mflags & DFNextStarted))
1688 afs_PrefetchChunk(avc, tdc, credp, &treq);
1691 ReleaseWriteLock(&avc->lock);
1696 return afs_convert_code(code);
1701 afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
1702 struct list_head *page_list, unsigned num_pages)
1707 struct iovec* iovecp;
1708 struct nocache_read_request *ancr;
1710 struct pagevec lrupv;
1714 struct inode *ip = FILE_INODE(fp);
1715 struct vcache *avc = VTOAFS(ip);
1716 afs_int32 base_index = 0;
1717 afs_int32 page_count = 0;
1720 /* background thread must free: iovecp, auio, ancr */
1721 iovecp = osi_Alloc(num_pages * sizeof(struct iovec));
1723 auio = osi_Alloc(sizeof(uio_t));
1724 auio->uio_iov = iovecp;
1725 auio->uio_iovcnt = num_pages;
1726 auio->uio_flag = UIO_READ;
1727 auio->uio_seg = AFS_UIOSYS;
1728 auio->uio_resid = num_pages * PAGE_SIZE;
1730 ancr = osi_Alloc(sizeof(struct nocache_read_request));
1732 ancr->offset = auio->uio_offset;
1733 ancr->length = auio->uio_resid;
1735 pagevec_init(&lrupv, 0);
1737 for(page_ix = 0; page_ix < num_pages; ++page_ix) {
1739 if(list_empty(page_list))
1742 pp = list_entry(page_list->prev, struct page, lru);
1743 /* If we allocate a page and don't remove it from page_list,
1744 * the page cache gets upset. */
1746 isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
1747 if(pp->index > isize) {
1754 offset = page_offset(pp);
1755 auio->uio_offset = offset;
1756 base_index = pp->index;
1758 iovecp[page_ix].iov_len = PAGE_SIZE;
1759 code = add_to_page_cache(pp, mapping, pp->index, GFP_KERNEL);
1760 if(base_index != pp->index) {
1763 page_cache_release(pp);
1764 iovecp[page_ix].iov_base = (void *) 0;
1766 ancr->length -= PAGE_SIZE;
1773 page_cache_release(pp);
1774 iovecp[page_ix].iov_base = (void *) 0;
1777 if(!PageLocked(pp)) {
1781 /* increment page refcount--our original design assumed
1782 * that locking it would effectively pin it; protect
1783 * ourselves from the possiblity that this assumption is
1784 * is faulty, at low cost (provided we do not fail to
1785 * do the corresponding decref on the other side) */
1788 /* save the page for background map */
1789 iovecp[page_ix].iov_base = (void*) pp;
1791 /* and put it on the LRU cache */
1792 if (!pagevec_add(&lrupv, pp))
1793 __pagevec_lru_add_file(&lrupv);
1797 /* If there were useful pages in the page list, make sure all pages
1798 * are in the LRU cache, then schedule the read */
1800 if (pagevec_count(&lrupv))
1801 __pagevec_lru_add_file(&lrupv);
1803 code = afs_ReadNoCache(avc, ancr, credp);
1806 /* If there is nothing for the background thread to handle,
1807 * it won't be freeing the things that we never gave it */
1808 osi_Free(iovecp, num_pages * sizeof(struct iovec));
1809 osi_Free(auio, sizeof(uio_t));
1810 osi_Free(ancr, sizeof(struct nocache_read_request));
1812 /* we do not flush, release, or unmap pages--that will be
1813 * done for us by the background thread as each page comes in
1814 * from the fileserver */
1815 return afs_convert_code(code);
1820 afs_linux_bypass_readpage(struct file *fp, struct page *pp)
1822 cred_t *credp = NULL;
1824 struct iovec *iovecp;
1825 struct nocache_read_request *ancr;
1829 * Special case: if page is at or past end of file, just zero it and set
1832 if (page_offset(pp) >= i_size_read(fp->f_mapping->host)) {
1833 zero_user_segment(pp, 0, PAGE_CACHE_SIZE);
1834 SetPageUptodate(pp);
1841 /* receiver frees */
1842 auio = osi_Alloc(sizeof(uio_t));
1843 iovecp = osi_Alloc(sizeof(struct iovec));
1845 /* address can be NULL, because we overwrite it with 'pp', below */
1846 setup_uio(auio, iovecp, NULL, page_offset(pp),
1847 PAGE_SIZE, UIO_READ, AFS_UIOSYS);
1849 /* save the page for background map */
1850 get_page(pp); /* see above */
1851 auio->uio_iov->iov_base = (void*) pp;
1852 /* the background thread will free this */
1853 ancr = osi_Alloc(sizeof(struct nocache_read_request));
1855 ancr->offset = page_offset(pp);
1856 ancr->length = PAGE_SIZE;
1859 code = afs_ReadNoCache(VTOAFS(FILE_INODE(fp)), ancr, credp);
1862 return afs_convert_code(code);
1866 afs_linux_can_bypass(struct inode *ip) {
1867 switch(cache_bypass_strategy) {
1868 case NEVER_BYPASS_CACHE:
1870 case ALWAYS_BYPASS_CACHE:
1872 case LARGE_FILES_BYPASS_CACHE:
1873 if(i_size_read(ip) > cache_bypass_threshold)
1880 /* Check if a file is permitted to bypass the cache by policy, and modify
1881 * the cache bypass state recorded for that file */
1884 afs_linux_bypass_check(struct inode *ip) {
1887 int bypass = afs_linux_can_bypass(ip);
1890 trydo_cache_transition(VTOAFS(ip), credp, bypass);
1898 afs_linux_readpage(struct file *fp, struct page *pp)
1902 if (afs_linux_bypass_check(FILE_INODE(fp))) {
1903 code = afs_linux_bypass_readpage(fp, pp);
1905 code = afs_linux_fillpage(fp, pp);
1907 code = afs_linux_prefetch(fp, pp);
1914 /* Readpages reads a number of pages for a particular file. We use
1915 * this to optimise the reading, by limiting the number of times upon which
1916 * we have to lookup, lock and open vcaches and dcaches
1920 afs_linux_readpages(struct file *fp, struct address_space *mapping,
1921 struct list_head *page_list, unsigned int num_pages)
1923 struct inode *inode = mapping->host;
1924 struct vcache *avc = VTOAFS(inode);
1926 struct file *cacheFp = NULL;
1928 unsigned int page_idx;
1930 struct pagevec lrupv;
1931 struct afs_pagecopy_task *task;
1933 if (afs_linux_bypass_check(inode))
1934 return afs_linux_bypass_readpages(fp, mapping, page_list, num_pages);
1936 if (cacheDiskType == AFS_FCACHE_TYPE_MEM)
1940 if ((code = afs_linux_VerifyVCache(avc, NULL))) {
1945 ObtainWriteLock(&avc->lock, 912);
1948 task = afs_pagecopy_init_task();
1951 pagevec_init(&lrupv, 0);
1952 for (page_idx = 0; page_idx < num_pages; page_idx++) {
1953 struct page *page = list_entry(page_list->prev, struct page, lru);
1954 list_del(&page->lru);
1955 offset = page_offset(page);
1957 if (tdc && tdc->f.chunk != AFS_CHUNK(offset)) {
1959 ReleaseReadLock(&tdc->lock);
1964 filp_close(cacheFp, NULL);
1969 if ((tdc = afs_FindDCache(avc, offset))) {
1970 ObtainReadLock(&tdc->lock);
1971 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
1972 (tdc->dflags & DFFetching)) {
1973 ReleaseReadLock(&tdc->lock);
1980 cacheFp = afs_linux_raw_open(&tdc->f.inode);
1983 if (tdc && !add_to_page_cache(page, mapping, page->index,
1985 page_cache_get(page);
1986 if (!pagevec_add(&lrupv, page))
1987 __pagevec_lru_add_file(&lrupv);
1989 afs_linux_read_cache(cacheFp, page, tdc->f.chunk, &lrupv, task);
1991 page_cache_release(page);
1993 if (pagevec_count(&lrupv))
1994 __pagevec_lru_add_file(&lrupv);
1997 filp_close(cacheFp, NULL);
1999 afs_pagecopy_put_task(task);
2003 ReleaseReadLock(&tdc->lock);
2007 ReleaseWriteLock(&avc->lock);
2012 /* Prepare an AFS vcache for writeback. Should be called with the vcache
2015 afs_linux_prepare_writeback(struct vcache *avc) {
2016 if (avc->f.states & CPageWrite) {
2017 return AOP_WRITEPAGE_ACTIVATE;
2019 avc->f.states |= CPageWrite;
2024 afs_linux_dopartialwrite(struct vcache *avc, cred_t *credp) {
2025 struct vrequest treq;
2028 if (!afs_InitReq(&treq, credp))
2029 code = afs_DoPartialWrite(avc, &treq);
2031 return afs_convert_code(code);
2035 afs_linux_complete_writeback(struct vcache *avc) {
2036 avc->f.states &= ~CPageWrite;
2039 /* Writeback a given page syncronously. Called with no AFS locks held */
2041 afs_linux_page_writeback(struct inode *ip, struct page *pp,
2042 unsigned long offset, unsigned int count,
2045 struct vcache *vcp = VTOAFS(ip);
2053 buffer = kmap(pp) + offset;
2054 base = page_offset(pp) + offset;
2057 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2058 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2059 ICL_TYPE_INT32, 99999);
2061 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
2063 code = afs_write(vcp, &tuio, f_flags, credp, 0);
2065 i_size_write(ip, vcp->f.m.Length);
2066 ip->i_blocks = ((vcp->f.m.Length + 1023) >> 10) << 1;
2068 code = code ? afs_convert_code(code) : count - tuio.uio_resid;
2070 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2071 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2072 ICL_TYPE_INT32, code);
2081 afs_linux_writepage_sync(struct inode *ip, struct page *pp,
2082 unsigned long offset, unsigned int count)
2086 struct vcache *vcp = VTOAFS(ip);
2089 /* Catch recursive writeback. This occurs if the kernel decides
2090 * writeback is required whilst we are writing to the cache, or
2091 * flushing to the server. When we're running syncronously (as
2092 * opposed to from writepage) we can't actually do anything about
2093 * this case - as we can't return AOP_WRITEPAGE_ACTIVATE to write()
2096 ObtainWriteLock(&vcp->lock, 532);
2097 afs_linux_prepare_writeback(vcp);
2098 ReleaseWriteLock(&vcp->lock);
2102 code = afs_linux_page_writeback(ip, pp, offset, count, credp);
2105 ObtainWriteLock(&vcp->lock, 533);
2107 code1 = afs_linux_dopartialwrite(vcp, credp);
2108 afs_linux_complete_writeback(vcp);
2109 ReleaseWriteLock(&vcp->lock);
2120 #ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
2121 afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
2123 afs_linux_writepage(struct page *pp)
2126 struct address_space *mapping = pp->mapping;
2127 struct inode *inode;
2130 unsigned int to = PAGE_CACHE_SIZE;
2135 if (PageReclaim(pp)) {
2136 return AOP_WRITEPAGE_ACTIVATE;
2137 /* XXX - Do we need to redirty the page here? */
2142 inode = mapping->host;
2143 vcp = VTOAFS(inode);
2144 isize = i_size_read(inode);
2146 /* Don't defeat an earlier truncate */
2147 if (page_offset(pp) > isize) {
2148 set_page_writeback(pp);
2154 ObtainWriteLock(&vcp->lock, 537);
2155 code = afs_linux_prepare_writeback(vcp);
2156 if (code == AOP_WRITEPAGE_ACTIVATE) {
2157 /* WRITEPAGE_ACTIVATE is the only return value that permits us
2158 * to return with the page still locked */
2159 ReleaseWriteLock(&vcp->lock);
2164 /* Grab the creds structure currently held in the vnode, and
2165 * get a reference to it, in case it goes away ... */
2171 ReleaseWriteLock(&vcp->lock);
2174 set_page_writeback(pp);
2176 SetPageUptodate(pp);
2178 /* We can unlock the page here, because it's protected by the
2179 * page_writeback flag. This should make us less vulnerable to
2180 * deadlocking in afs_write and afs_DoPartialWrite
2184 /* If this is the final page, then just write the number of bytes that
2185 * are actually in it */
2186 if ((isize - page_offset(pp)) < to )
2187 to = isize - page_offset(pp);
2189 code = afs_linux_page_writeback(inode, pp, 0, to, credp);
2192 ObtainWriteLock(&vcp->lock, 538);
2194 /* As much as we might like to ignore a file server error here,
2195 * and just try again when we close(), unfortunately StoreAllSegments
2196 * will invalidate our chunks if the server returns a permanent error,
2197 * so we need to at least try and get that error back to the user
2200 code1 = afs_linux_dopartialwrite(vcp, credp);
2202 afs_linux_complete_writeback(vcp);
2203 ReleaseWriteLock(&vcp->lock);
2208 end_page_writeback(pp);
2209 page_cache_release(pp);
2220 /* afs_linux_permission
2221 * Check access rights - returns error if can't check or permission denied.
2224 #ifdef IOP_PERMISSION_TAKES_NAMEIDATA
2225 afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
2227 afs_linux_permission(struct inode *ip, int mode)
2231 cred_t *credp = crref();
2235 if (mode & MAY_EXEC)
2237 if (mode & MAY_READ)
2239 if (mode & MAY_WRITE)
2241 code = afs_access(VTOAFS(ip), tmp, credp);
2245 return afs_convert_code(code);
2249 afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
2253 struct inode *inode = FILE_INODE(file);
2254 loff_t pagebase = page_offset(page);
2256 if (i_size_read(inode) < (pagebase + offset))
2257 i_size_write(inode, pagebase + offset);
2259 if (PageChecked(page)) {
2260 SetPageUptodate(page);
2261 ClearPageChecked(page);
2264 code = afs_linux_writepage_sync(inode, page, offset, to - offset);
2270 afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
2274 /* http://kerneltrap.org/node/4941 details the expected behaviour of
2275 * prepare_write. Essentially, if the page exists within the file,
2276 * and is not being fully written, then we should populate it.
2279 if (!PageUptodate(page)) {
2280 loff_t pagebase = page_offset(page);
2281 loff_t isize = i_size_read(page->mapping->host);
2283 /* Is the location we are writing to beyond the end of the file? */
2284 if (pagebase >= isize ||
2285 ((from == 0) && (pagebase + to) >= isize)) {
2286 zero_user_segments(page, 0, from, to, PAGE_CACHE_SIZE);
2287 SetPageChecked(page);
2288 /* Are we we writing a full page */
2289 } else if (from == 0 && to == PAGE_CACHE_SIZE) {
2290 SetPageChecked(page);
2291 /* Is the page readable, if it's wronly, we don't care, because we're
2292 * not actually going to read from it ... */
2293 } else if ((file->f_flags && O_ACCMODE) != O_WRONLY) {
2294 /* We don't care if fillpage fails, because if it does the page
2295 * won't be marked as up to date
2297 afs_linux_fillpage(file, page);
2303 #if defined(STRUCT_ADDRESS_SPACE_OPERATIONS_HAS_WRITE_BEGIN)
2305 afs_linux_write_end(struct file *file, struct address_space *mapping,
2306 loff_t pos, unsigned len, unsigned copied,
2307 struct page *page, void *fsdata)
2310 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
2312 code = afs_linux_commit_write(file, page, from, from + len);
2315 page_cache_release(page);
2320 afs_linux_write_begin(struct file *file, struct address_space *mapping,
2321 loff_t pos, unsigned len, unsigned flags,
2322 struct page **pagep, void **fsdata)
2325 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2326 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
2329 page = grab_cache_page_write_begin(mapping, index, flags);
2332 code = afs_linux_prepare_write(file, page, from, from + len);
2335 page_cache_release(page);
2343 static struct inode_operations afs_file_iops = {
2344 .permission = afs_linux_permission,
2345 .getattr = afs_linux_getattr,
2346 .setattr = afs_notify_change,
2349 static struct address_space_operations afs_file_aops = {
2350 .readpage = afs_linux_readpage,
2351 .readpages = afs_linux_readpages,
2352 .writepage = afs_linux_writepage,
2353 #if defined (STRUCT_ADDRESS_SPACE_OPERATIONS_HAS_WRITE_BEGIN)
2354 .write_begin = afs_linux_write_begin,
2355 .write_end = afs_linux_write_end,
2357 .commit_write = afs_linux_commit_write,
2358 .prepare_write = afs_linux_prepare_write,
2363 /* Separate ops vector for directories. Linux 2.2 tests type of inode
2364 * by what sort of operation is allowed.....
2367 static struct inode_operations afs_dir_iops = {
2368 .setattr = afs_notify_change,
2369 .create = afs_linux_create,
2370 .lookup = afs_linux_lookup,
2371 .link = afs_linux_link,
2372 .unlink = afs_linux_unlink,
2373 .symlink = afs_linux_symlink,
2374 .mkdir = afs_linux_mkdir,
2375 .rmdir = afs_linux_rmdir,
2376 .rename = afs_linux_rename,
2377 .getattr = afs_linux_getattr,
2378 .permission = afs_linux_permission,
2381 /* We really need a separate symlink set of ops, since do_follow_link()
2382 * determines if it _is_ a link by checking if the follow_link op is set.
2384 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2386 afs_symlink_filler(struct file *file, struct page *page)
2388 struct inode *ip = (struct inode *)page->mapping->host;
2389 char *p = (char *)kmap(page);
2393 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
2398 p[code] = '\0'; /* null terminate? */
2400 SetPageUptodate(page);
2412 static struct address_space_operations afs_symlink_aops = {
2413 .readpage = afs_symlink_filler
2415 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
2417 static struct inode_operations afs_symlink_iops = {
2418 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2419 .readlink = page_readlink,
2420 # if defined(HAVE_LINUX_PAGE_FOLLOW_LINK)
2421 .follow_link = page_follow_link,
2423 .follow_link = page_follow_link_light,
2424 .put_link = page_put_link,
2426 #else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
2427 .readlink = afs_linux_readlink,
2428 .follow_link = afs_linux_follow_link,
2429 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
2430 .setattr = afs_notify_change,
2434 afs_fill_inode(struct inode *ip, struct vattr *vattr)
2438 vattr2inode(ip, vattr);
2440 ip->i_mapping->backing_dev_info = afs_backing_dev_info;
2441 /* Reset ops if symlink or directory. */
2442 if (S_ISREG(ip->i_mode)) {
2443 ip->i_op = &afs_file_iops;
2444 ip->i_fop = &afs_file_fops;
2445 ip->i_data.a_ops = &afs_file_aops;
2447 } else if (S_ISDIR(ip->i_mode)) {
2448 ip->i_op = &afs_dir_iops;
2449 ip->i_fop = &afs_dir_fops;
2451 } else if (S_ISLNK(ip->i_mode)) {
2452 ip->i_op = &afs_symlink_iops;
2453 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2454 ip->i_data.a_ops = &afs_symlink_aops;
2455 ip->i_mapping = &ip->i_data;