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/writeback.h>
34 #include <linux/pagevec.h>
35 #include <linux/aio.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 int cachefs_noreadpage = 0;
52 extern struct backing_dev_info *afs_backing_dev_info;
54 extern struct vcache *afs_globalVp;
56 /* This function converts a positive error code from AFS into a negative
57 * code suitable for passing into the Linux VFS layer. It checks that the
58 * error code is within the permissable bounds for the ERR_PTR mechanism.
60 * _All_ error codes which come from the AFS layer should be passed through
61 * this function before being returned to the kernel.
65 afs_convert_code(int code) {
66 if ((code >= 0) && (code <= MAX_ERRNO))
72 /* Linux doesn't require a credp for many functions, and crref is an expensive
73 * operation. This helper function avoids obtaining it for VerifyVCache calls
77 afs_linux_VerifyVCache(struct vcache *avc, cred_t **retcred) {
82 if (avc->f.states & CStatd) {
90 code = afs_InitReq(&treq, credp);
92 code = afs_VerifyVCache2(avc, &treq);
99 return afs_convert_code(code);
102 #ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
103 # ifdef LINUX_HAS_NONVECTOR_AIO
105 afs_linux_aio_read(struct kiocb *iocb, char __user *buf, size_t bufsize,
109 afs_linux_aio_read(struct kiocb *iocb, const struct iovec *buf,
110 unsigned long bufsize, loff_t pos)
113 struct file *fp = iocb->ki_filp;
115 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
118 afs_Trace4(afs_iclSetp, CM_TRACE_AIOREADOP, ICL_TYPE_POINTER, vcp,
119 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
120 (afs_int32)bufsize, ICL_TYPE_INT32, 99999);
121 code = afs_linux_VerifyVCache(vcp, NULL);
124 /* Linux's FlushPages implementation doesn't ever use credp,
125 * so we optimise by not using it */
126 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
128 code = generic_file_aio_read(iocb, buf, bufsize, pos);
132 afs_Trace4(afs_iclSetp, CM_TRACE_AIOREADOP, ICL_TYPE_POINTER, vcp,
133 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
134 (afs_int32)bufsize, ICL_TYPE_INT32, code);
140 afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
143 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
146 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
147 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
149 code = afs_linux_VerifyVCache(vcp, NULL);
152 /* Linux's FlushPages implementation doesn't ever use credp,
153 * so we optimise by not using it */
154 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
156 code = do_sync_read(fp, buf, count, offp);
160 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
161 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
169 /* Now we have integrated VM for writes as well as reads. the generic write operations
170 * also take care of re-positioning the pointer if file is open in append
171 * mode. Call fake open/close to ensure we do writes of core dumps.
173 #ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
174 # ifdef LINUX_HAS_NONVECTOR_AIO
176 afs_linux_aio_write(struct kiocb *iocb, const char __user *buf, size_t bufsize,
180 afs_linux_aio_write(struct kiocb *iocb, const struct iovec *buf,
181 unsigned long bufsize, loff_t pos)
185 struct vcache *vcp = VTOAFS(iocb->ki_filp->f_dentry->d_inode);
190 afs_Trace4(afs_iclSetp, CM_TRACE_AIOWRITEOP, ICL_TYPE_POINTER, vcp,
191 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
192 (afs_int32)bufsize, ICL_TYPE_INT32,
193 (iocb->ki_filp->f_flags & O_APPEND) ? 99998 : 99999);
195 code = afs_linux_VerifyVCache(vcp, &credp);
197 ObtainWriteLock(&vcp->lock, 529);
199 ReleaseWriteLock(&vcp->lock);
202 code = generic_file_aio_write(iocb, buf, bufsize, pos);
206 ObtainWriteLock(&vcp->lock, 530);
208 if (vcp->execsOrWriters == 1 && !credp)
211 afs_FakeClose(vcp, credp);
212 ReleaseWriteLock(&vcp->lock);
214 afs_Trace4(afs_iclSetp, CM_TRACE_AIOWRITEOP, ICL_TYPE_POINTER, vcp,
215 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
216 (afs_int32)bufsize, ICL_TYPE_INT32, code);
225 afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
228 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
233 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
234 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
235 (fp->f_flags & O_APPEND) ? 99998 : 99999);
237 code = afs_linux_VerifyVCache(vcp, &credp);
239 ObtainWriteLock(&vcp->lock, 529);
241 ReleaseWriteLock(&vcp->lock);
244 code = do_sync_write(fp, buf, count, offp);
248 ObtainWriteLock(&vcp->lock, 530);
250 if (vcp->execsOrWriters == 1 && !credp)
253 afs_FakeClose(vcp, credp);
254 ReleaseWriteLock(&vcp->lock);
256 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
257 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
267 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
269 /* This is a complete rewrite of afs_readdir, since we can make use of
270 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
271 * handling and use of bulkstats will need to be reflected here as well.
274 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
275 afs_linux_readdir(struct file *fp, struct dir_context *ctx)
277 afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
280 struct vcache *avc = VTOAFS(FILE_INODE(fp));
281 struct vrequest treq;
287 struct DirBuffer entry;
290 afs_size_t origOffset, tlen;
291 cred_t *credp = crref();
292 struct afs_fakestat_state fakestat;
295 AFS_STATCNT(afs_readdir);
297 code = afs_convert_code(afs_InitReq(&treq, credp));
302 afs_InitFakeStat(&fakestat);
303 code = afs_convert_code(afs_EvalFakeStat(&avc, &fakestat, &treq));
307 /* update the cache entry */
309 code = afs_convert_code(afs_VerifyVCache2(avc, &treq));
313 /* get a reference to the entire directory */
314 tdc = afs_GetDCache(avc, (afs_size_t) 0, &treq, &origOffset, &tlen, 1);
320 ObtainWriteLock(&avc->lock, 811);
321 ObtainReadLock(&tdc->lock);
323 * Make sure that the data in the cache is current. There are two
324 * cases we need to worry about:
325 * 1. The cache data is being fetched by another process.
326 * 2. The cache data is no longer valid
328 while ((avc->f.states & CStatd)
329 && (tdc->dflags & DFFetching)
330 && hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
331 ReleaseReadLock(&tdc->lock);
332 ReleaseWriteLock(&avc->lock);
333 afs_osi_Sleep(&tdc->validPos);
334 ObtainWriteLock(&avc->lock, 812);
335 ObtainReadLock(&tdc->lock);
337 if (!(avc->f.states & CStatd)
338 || !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
339 ReleaseReadLock(&tdc->lock);
340 ReleaseWriteLock(&avc->lock);
345 /* Set the readdir-in-progress flag, and downgrade the lock
346 * to shared so others will be able to acquire a read lock.
348 avc->f.states |= CReadDir;
349 avc->dcreaddir = tdc;
350 avc->readdir_pid = MyPidxx2Pid(MyPidxx);
351 ConvertWToSLock(&avc->lock);
353 /* Fill in until we get an error or we're done. This implementation
354 * takes an offset in units of blobs, rather than bytes.
357 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
360 offset = (int) fp->f_pos;
363 dirpos = BlobScan(tdc, offset);
367 code = afs_dir_GetVerifiedBlob(tdc, dirpos, &entry);
369 if (!(avc->f.states & CCorrupt)) {
370 struct cell *tc = afs_GetCellStale(avc->f.fid.Cell, READ_LOCK);
371 afs_warn("Corrupt directory (%d.%d.%d.%d [%s] @%lx, pos %d)",
372 avc->f.fid.Cell, avc->f.fid.Fid.Volume,
373 avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique,
374 tc ? tc->cellName : "",
375 (unsigned long)&tdc->f.inode, dirpos);
377 afs_PutCell(tc, READ_LOCK);
378 UpgradeSToWLock(&avc->lock, 814);
379 avc->f.states |= CCorrupt;
385 de = (struct DirEntry *)entry.data;
386 ino = afs_calc_inum (avc->f.fid.Cell, avc->f.fid.Fid.Volume,
387 ntohl(de->fid.vnode));
388 len = strlen(de->name);
390 /* filldir returns -EINVAL when the buffer is full. */
392 unsigned int type = DT_UNKNOWN;
393 struct VenusFid afid;
396 afid.Cell = avc->f.fid.Cell;
397 afid.Fid.Volume = avc->f.fid.Fid.Volume;
398 afid.Fid.Vnode = ntohl(de->fid.vnode);
399 afid.Fid.Unique = ntohl(de->fid.vunique);
400 if ((avc->f.states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
402 } else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
405 } else if (((tvc->f.states) & (CStatd | CTruth))) {
406 /* CTruth will be set if the object has
411 else if (vtype == VREG)
413 /* Don't do this until we're sure it can't be a mtpt */
414 /* else if (vtype == VLNK)
416 /* what other types does AFS support? */
418 /* clean up from afs_FindVCache */
422 * If this is NFS readdirplus, then the filler is going to
423 * call getattr on this inode, which will deadlock if we're
427 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
428 /* dir_emit returns a bool - true when it succeeds.
429 * Inverse the result to fit with how we check "code" */
430 code = !dir_emit(ctx, de->name, len, ino, type);
432 code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
439 offset = dirpos + 1 + ((len + 16) >> 5);
441 /* If filldir didn't fill in the last one this is still pointing to that
447 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
448 ctx->pos = (loff_t) offset;
450 fp->f_pos = (loff_t) offset;
452 ReleaseReadLock(&tdc->lock);
454 UpgradeSToWLock(&avc->lock, 813);
455 avc->f.states &= ~CReadDir;
457 avc->readdir_pid = 0;
458 ReleaseSharedLock(&avc->lock);
461 afs_PutFakeStat(&fakestat);
468 /* in afs_pioctl.c */
469 extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
472 #if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
473 static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
475 return afs_xioctl(FILE_INODE(fp), fp, com, arg);
482 afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
484 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
488 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
489 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
490 vmap->vm_end - vmap->vm_start);
492 /* get a validated vcache entry */
493 code = afs_linux_VerifyVCache(vcp, NULL);
496 /* Linux's Flushpage implementation doesn't use credp, so optimise
497 * our code to not need to crref() it */
498 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
500 code = generic_file_mmap(fp, vmap);
503 vcp->f.states |= CMAPPED;
511 afs_linux_open(struct inode *ip, struct file *fp)
513 struct vcache *vcp = VTOAFS(ip);
514 cred_t *credp = crref();
518 code = afs_open(&vcp, fp->f_flags, credp);
522 return afs_convert_code(code);
526 afs_linux_release(struct inode *ip, struct file *fp)
528 struct vcache *vcp = VTOAFS(ip);
529 cred_t *credp = crref();
533 code = afs_close(vcp, fp->f_flags, credp);
534 ObtainWriteLock(&vcp->lock, 807);
539 ReleaseWriteLock(&vcp->lock);
543 return afs_convert_code(code);
547 #if defined(FOP_FSYNC_TAKES_DENTRY)
548 afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
549 #elif defined(FOP_FSYNC_TAKES_RANGE)
550 afs_linux_fsync(struct file *fp, loff_t start, loff_t end, int datasync)
552 afs_linux_fsync(struct file *fp, int datasync)
556 struct inode *ip = FILE_INODE(fp);
557 cred_t *credp = crref();
559 #if defined(FOP_FSYNC_TAKES_RANGE)
560 mutex_lock(&ip->i_mutex);
563 code = afs_fsync(VTOAFS(ip), credp);
565 #if defined(FOP_FSYNC_TAKES_RANGE)
566 mutex_unlock(&ip->i_mutex);
569 return afs_convert_code(code);
575 afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
578 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
579 cred_t *credp = crref();
580 struct AFS_FLOCK flock;
582 /* Convert to a lock format afs_lockctl understands. */
583 memset(&flock, 0, sizeof(flock));
584 flock.l_type = flp->fl_type;
585 flock.l_pid = flp->fl_pid;
587 flock.l_start = flp->fl_start;
588 if (flp->fl_end == OFFSET_MAX)
589 flock.l_len = 0; /* Lock to end of file */
591 flock.l_len = flp->fl_end - flp->fl_start + 1;
593 /* Safe because there are no large files, yet */
594 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
595 if (cmd == F_GETLK64)
597 else if (cmd == F_SETLK64)
599 else if (cmd == F_SETLKW64)
601 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
604 if ((vcp->f.states & CRO)) {
605 if (flp->fl_type == F_WRLCK) {
614 code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
617 if ((code == 0 || flp->fl_type == F_UNLCK) &&
618 (cmd == F_SETLK || cmd == F_SETLKW)) {
619 code = afs_posix_lock_file(fp, flp);
620 if (code && flp->fl_type != F_UNLCK) {
621 struct AFS_FLOCK flock2;
623 flock2.l_type = F_UNLCK;
625 afs_lockctl(vcp, &flock2, F_SETLK, credp);
629 /* If lockctl says there are no conflicting locks, then also check with the
630 * kernel, as lockctl knows nothing about byte range locks
632 if (code == 0 && cmd == F_GETLK && flock.l_type == F_UNLCK) {
633 afs_posix_test_lock(fp, flp);
634 /* If we found a lock in the kernel's structure, return it */
635 if (flp->fl_type != F_UNLCK) {
641 /* Convert flock back to Linux's file_lock */
642 flp->fl_type = flock.l_type;
643 flp->fl_pid = flock.l_pid;
644 flp->fl_start = flock.l_start;
645 if (flock.l_len == 0)
646 flp->fl_end = OFFSET_MAX; /* Lock to end of file */
648 flp->fl_end = flock.l_start + flock.l_len - 1;
654 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
656 afs_linux_flock(struct file *fp, int cmd, struct file_lock *flp) {
658 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
659 cred_t *credp = crref();
660 struct AFS_FLOCK flock;
661 /* Convert to a lock format afs_lockctl understands. */
662 memset(&flock, 0, sizeof(flock));
663 flock.l_type = flp->fl_type;
664 flock.l_pid = flp->fl_pid;
669 /* Safe because there are no large files, yet */
670 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
671 if (cmd == F_GETLK64)
673 else if (cmd == F_SETLK64)
675 else if (cmd == F_SETLKW64)
677 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
680 code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
683 if ((code == 0 || flp->fl_type == F_UNLCK) &&
684 (cmd == F_SETLK || cmd == F_SETLKW)) {
685 flp->fl_flags &=~ FL_SLEEP;
686 code = flock_lock_file_wait(fp, flp);
687 if (code && flp->fl_type != F_UNLCK) {
688 struct AFS_FLOCK flock2;
690 flock2.l_type = F_UNLCK;
692 afs_lockctl(vcp, &flock2, F_SETLK, credp);
696 /* Convert flock back to Linux's file_lock */
697 flp->fl_type = flock.l_type;
698 flp->fl_pid = flock.l_pid;
706 * essentially the same as afs_fsync() but we need to get the return
707 * code for the sys_close() here, not afs_linux_release(), so call
708 * afs_StoreAllSegments() with AFS_LASTSTORE
711 #if defined(FOP_FLUSH_TAKES_FL_OWNER_T)
712 afs_linux_flush(struct file *fp, fl_owner_t id)
714 afs_linux_flush(struct file *fp)
717 struct vrequest treq;
725 if ((fp->f_flags & O_ACCMODE) == O_RDONLY) { /* readers dont flush */
733 vcp = VTOAFS(FILE_INODE(fp));
735 code = afs_InitReq(&treq, credp);
738 /* If caching is bypassed for this file, or globally, just return 0 */
739 if (cache_bypass_strategy == ALWAYS_BYPASS_CACHE)
742 ObtainReadLock(&vcp->lock);
743 if (vcp->cachingStates & FCSBypass)
745 ReleaseReadLock(&vcp->lock);
748 /* future proof: don't rely on 0 return from afs_InitReq */
753 ObtainSharedLock(&vcp->lock, 535);
754 if ((vcp->execsOrWriters > 0) && (file_count(fp) == 1)) {
755 UpgradeSToWLock(&vcp->lock, 536);
756 if (!AFS_IS_DISCONNECTED) {
757 code = afs_StoreAllSegments(vcp,
759 AFS_SYNC | AFS_LASTSTORE);
761 afs_DisconAddDirty(vcp, VDisconWriteOsiFlush, 1);
763 ConvertWToSLock(&vcp->lock);
765 code = afs_CheckCode(code, &treq, 54);
766 ReleaseSharedLock(&vcp->lock);
773 return afs_convert_code(code);
776 struct file_operations afs_dir_fops = {
777 .read = generic_read_dir,
778 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
779 .iterate = afs_linux_readdir,
781 .readdir = afs_linux_readdir,
783 #ifdef HAVE_UNLOCKED_IOCTL
784 .unlocked_ioctl = afs_unlocked_xioctl,
788 #ifdef HAVE_COMPAT_IOCTL
789 .compat_ioctl = afs_unlocked_xioctl,
791 .open = afs_linux_open,
792 .release = afs_linux_release,
793 .llseek = default_llseek,
794 #ifdef HAVE_LINUX_NOOP_FSYNC
797 .fsync = simple_sync_file,
801 struct file_operations afs_file_fops = {
802 #ifdef HAVE_LINUX_GENERIC_FILE_AIO_READ
803 .aio_read = afs_linux_aio_read,
804 .aio_write = afs_linux_aio_write,
805 .read = do_sync_read,
806 .write = do_sync_write,
808 .read = afs_linux_read,
809 .write = afs_linux_write,
811 #ifdef HAVE_UNLOCKED_IOCTL
812 .unlocked_ioctl = afs_unlocked_xioctl,
816 #ifdef HAVE_COMPAT_IOCTL
817 .compat_ioctl = afs_unlocked_xioctl,
819 .mmap = afs_linux_mmap,
820 .open = afs_linux_open,
821 .flush = afs_linux_flush,
822 #if defined(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
823 .sendfile = generic_file_sendfile,
825 #if defined(STRUCT_FILE_OPERATIONS_HAS_SPLICE)
826 .splice_write = generic_file_splice_write,
827 .splice_read = generic_file_splice_read,
829 .release = afs_linux_release,
830 .fsync = afs_linux_fsync,
831 .lock = afs_linux_lock,
832 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
833 .flock = afs_linux_flock,
835 .llseek = default_llseek,
838 static struct dentry *
839 canonical_dentry(struct inode *ip)
841 struct vcache *vcp = VTOAFS(ip);
842 struct dentry *first = NULL, *ret = NULL, *cur;
843 #if defined(D_ALIAS_IS_HLIST) && !defined(HLIST_ITERATOR_NO_NODE)
844 struct hlist_node *p;
848 * if vcp->target_link is set, and can be found in ip->i_dentry, use that.
849 * otherwise, use the first dentry in ip->i_dentry.
850 * if ip->i_dentry is empty, use the 'dentry' argument we were given.
852 /* note that vcp->target_link specifies which dentry to use, but we have
853 * no reference held on that dentry. so, we cannot use or dereference
854 * vcp->target_link itself, since it may have been freed. instead, we only
855 * use it to compare to pointers in the ip->i_dentry list. */
859 # ifdef HAVE_DCACHE_LOCK
860 spin_lock(&dcache_lock);
862 spin_lock(&ip->i_lock);
865 #if defined(D_ALIAS_IS_HLIST)
866 # if defined(HLIST_ITERATOR_NO_NODE)
867 hlist_for_each_entry(cur, &ip->i_dentry, d_alias) {
869 hlist_for_each_entry(cur, p, &ip->i_dentry, d_alias) {
872 list_for_each_entry_reverse(cur, &ip->i_dentry, d_alias) {
875 if (!vcp->target_link || cur == vcp->target_link) {
888 vcp->target_link = ret;
890 # ifdef HAVE_DCACHE_LOCK
894 spin_unlock(&dcache_lock);
899 spin_unlock(&ip->i_lock);
905 /**********************************************************************
906 * AFS Linux dentry operations
907 **********************************************************************/
909 /* fix_bad_parent() : called if this dentry's vcache is a root vcache
910 * that has its mvid (parent dir's fid) pointer set to the wrong directory
911 * due to being mounted in multiple points at once. fix_bad_parent()
912 * calls afs_lookup() to correct the vcache's mvid, as well as the volume's
913 * dotdotfid and mtpoint fid members.
915 * dp - dentry to be checked.
916 * credp - credentials
917 * vcp, pvc - item's and parent's vcache pointer
921 * This dentry's vcache's mvid will be set to the correct parent directory's
923 * This root vnode's volume will have its dotdotfid and mtpoint fids set
924 * to the correct parent and mountpoint fids.
928 fix_bad_parent(struct dentry *dp, cred_t *credp, struct vcache *vcp, struct vcache *pvc)
930 struct vcache *avc = NULL;
932 /* force a lookup, so vcp->mvid is fixed up */
933 afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
934 if (!avc || vcp != avc) { /* bad, very bad.. */
935 afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
936 "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
937 ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
938 ICL_TYPE_POINTER, dp);
941 AFS_RELE(AFSTOV(avc));
946 /* afs_linux_revalidate
947 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
950 afs_linux_revalidate(struct dentry *dp)
953 struct vcache *vcp = VTOAFS(dp->d_inode);
957 if (afs_shuttingdown)
963 /* Make this a fast path (no crref), since it's called so often. */
964 if (vcp->states & CStatd) {
965 struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
967 if (*dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
968 if (vcp->mvid->Fid.Volume != pvc->fid.Fid.Volume) { /* bad parent */
971 fix_bad_parent(dp); /* check and correct mvid */
980 /* This avoids the crref when we don't have to do it. Watch for
981 * changes in afs_getattr that don't get replicated here!
983 if (vcp->f.states & CStatd &&
984 (!afs_fakestat_enable || vcp->mvstat != 1) &&
986 (vType(vcp) == VDIR || vType(vcp) == VLNK)) {
987 code = afs_CopyOutAttrs(vcp, &vattr);
990 code = afs_getattr(vcp, &vattr, credp);
995 afs_fill_inode(AFSTOV(vcp), &vattr);
999 return afs_convert_code(code);
1003 * Set iattr data into vattr. Assume vattr cleared before call.
1006 iattr2vattr(struct vattr *vattrp, struct iattr *iattrp)
1008 vattrp->va_mask = iattrp->ia_valid;
1009 if (iattrp->ia_valid & ATTR_MODE)
1010 vattrp->va_mode = iattrp->ia_mode;
1011 if (iattrp->ia_valid & ATTR_UID)
1012 vattrp->va_uid = afs_from_kuid(iattrp->ia_uid);
1013 if (iattrp->ia_valid & ATTR_GID)
1014 vattrp->va_gid = afs_from_kgid(iattrp->ia_gid);
1015 if (iattrp->ia_valid & ATTR_SIZE)
1016 vattrp->va_size = iattrp->ia_size;
1017 if (iattrp->ia_valid & ATTR_ATIME) {
1018 vattrp->va_atime.tv_sec = iattrp->ia_atime.tv_sec;
1019 vattrp->va_atime.tv_usec = 0;
1021 if (iattrp->ia_valid & ATTR_MTIME) {
1022 vattrp->va_mtime.tv_sec = iattrp->ia_mtime.tv_sec;
1023 vattrp->va_mtime.tv_usec = 0;
1025 if (iattrp->ia_valid & ATTR_CTIME) {
1026 vattrp->va_ctime.tv_sec = iattrp->ia_ctime.tv_sec;
1027 vattrp->va_ctime.tv_usec = 0;
1032 * Rewrite the inode cache from the attr. Assumes all vattr fields are valid.
1035 vattr2inode(struct inode *ip, struct vattr *vp)
1037 ip->i_ino = vp->va_nodeid;
1038 #ifdef HAVE_LINUX_SET_NLINK
1039 set_nlink(ip, vp->va_nlink);
1041 ip->i_nlink = vp->va_nlink;
1043 ip->i_blocks = vp->va_blocks;
1044 #ifdef STRUCT_INODE_HAS_I_BLKBITS
1045 ip->i_blkbits = AFS_BLKBITS;
1047 #ifdef STRUCT_INODE_HAS_I_BLKSIZE
1048 ip->i_blksize = vp->va_blocksize;
1050 ip->i_rdev = vp->va_rdev;
1051 ip->i_mode = vp->va_mode;
1052 ip->i_uid = afs_make_kuid(vp->va_uid);
1053 ip->i_gid = afs_make_kgid(vp->va_gid);
1054 i_size_write(ip, vp->va_size);
1055 ip->i_atime.tv_sec = vp->va_atime.tv_sec;
1056 ip->i_atime.tv_nsec = 0;
1057 ip->i_mtime.tv_sec = vp->va_mtime.tv_sec;
1058 /* Set the mtime nanoseconds to the sysname generation number.
1059 * This convinces NFS clients that all directories have changed
1060 * any time the sysname list changes.
1062 ip->i_mtime.tv_nsec = afs_sysnamegen;
1063 ip->i_ctime.tv_sec = vp->va_ctime.tv_sec;
1064 ip->i_ctime.tv_nsec = 0;
1067 /* afs_notify_change
1068 * Linux version of setattr call. What to change is in the iattr struct.
1069 * We need to set bits in both the Linux inode as well as the vcache.
1072 afs_notify_change(struct dentry *dp, struct iattr *iattrp)
1075 cred_t *credp = crref();
1076 struct inode *ip = dp->d_inode;
1080 iattr2vattr(&vattr, iattrp); /* Convert for AFS vnodeops call. */
1083 code = afs_setattr(VTOAFS(ip), &vattr, credp);
1085 afs_getattr(VTOAFS(ip), &vattr, credp);
1086 vattr2inode(ip, &vattr);
1090 return afs_convert_code(code);
1094 afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1096 int err = afs_linux_revalidate(dentry);
1098 generic_fillattr(dentry->d_inode, stat);
1103 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
1104 * In kernels 2.2.10 and above, we are passed an additional flags var which
1105 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
1106 * we are advised to follow the entry if it is a link or to make sure that
1107 * it is a directory. But since the kernel itself checks these possibilities
1108 * later on, we shouldn't have to do it until later. Perhaps in the future..
1110 * The code here assumes that on entry the global lock is not held
1113 #if defined(DOP_REVALIDATE_TAKES_UNSIGNED)
1114 afs_linux_dentry_revalidate(struct dentry *dp, unsigned int flags)
1115 #elif defined(DOP_REVALIDATE_TAKES_NAMEIDATA)
1116 afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
1118 afs_linux_dentry_revalidate(struct dentry *dp, int flags)
1122 cred_t *credp = NULL;
1123 struct vcache *vcp, *pvcp, *tvc = NULL;
1124 struct dentry *parent;
1126 struct afs_fakestat_state fakestate;
1131 /* We don't support RCU path walking */
1132 # if defined(DOP_REVALIDATE_TAKES_UNSIGNED)
1133 if (flags & LOOKUP_RCU)
1135 if (nd->flags & LOOKUP_RCU)
1140 afs_InitFakeStat(&fakestate);
1143 vcp = VTOAFS(dp->d_inode);
1145 if (vcp == afs_globalVp)
1148 parent = dget_parent(dp);
1149 pvcp = VTOAFS(parent->d_inode);
1151 if ((vcp->mvstat == 1) || (vcp->mvstat == 2)) { /* need to lock */
1157 if (locked && vcp->mvstat == 1) { /* mount point */
1158 if (vcp->mvid && (vcp->f.states & CMValid)) {
1159 int tryEvalOnly = 0;
1161 struct vrequest treq;
1163 code = afs_InitReq(&treq, credp);
1165 (strcmp(dp->d_name.name, ".directory") == 0)) {
1169 code = afs_TryEvalFakeStat(&vcp, &fakestate, &treq);
1171 code = afs_EvalFakeStat(&vcp, &fakestate, &treq);
1172 if ((tryEvalOnly && vcp->mvstat == 1) || code) {
1173 /* a mount point, not yet replaced by its directory */
1179 if (locked && *dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
1180 if (vcp->mvid->Fid.Volume != pvcp->f.fid.Fid.Volume) { /* bad parent */
1181 fix_bad_parent(dp, credp, vcp, pvcp); /* check and correct mvid */
1186 /* If the last looker changes, we should make sure the current
1187 * looker still has permission to examine this file. This would
1188 * always require a crref() which would be "slow".
1190 if (vcp->last_looker != treq.uid) {
1191 if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
1196 vcp->last_looker = treq.uid;
1201 /* If the parent's DataVersion has changed or the vnode
1202 * is longer valid, we need to do a full lookup. VerifyVCache
1203 * isn't enough since the vnode may have been renamed.
1206 if ((!locked) && (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd)) ) {
1212 if (locked && (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd))) {
1215 code = afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
1216 if (!tvc || tvc != vcp) {
1218 /* Force unhash if name is known not to exist. */
1224 if (afs_getattr(vcp, &vattr, credp)) {
1229 vattr2inode(AFSTOV(vcp), &vattr);
1230 dp->d_time = hgetlo(pvcp->f.m.DataVersion);
1233 /* should we always update the attributes at this point? */
1234 /* unlikely--the vcache entry hasn't changed */
1239 /* If this code is ever enabled, we should use dget_parent to handle
1240 * getting the parent, and dput() to dispose of it. See above for an
1242 pvcp = VTOAFS(dp->d_parent->d_inode);
1243 if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time)
1247 /* No change in parent's DataVersion so this negative
1248 * lookup is still valid. BUT, if a server is down a
1249 * negative lookup can result so there should be a
1250 * liftime as well. For now, always expire.
1263 afs_PutFakeStat(&fakestate); /* from here on vcp may be no longer valid */
1265 /* we hold the global lock if we evaluated a mount point */
1273 * If we had a negative lookup for the name we want to forcibly
1274 * unhash the dentry.
1275 * Otherwise use d_invalidate which will not unhash it if still in use.
1278 shrink_dcache_parent(dp);
1287 if (have_submounts(dp))
1295 afs_dentry_iput(struct dentry *dp, struct inode *ip)
1297 struct vcache *vcp = VTOAFS(ip);
1300 if (!AFS_IS_DISCONNECTED || (vcp->f.states & CUnlinked)) {
1301 (void) afs_InactiveVCache(vcp, NULL);
1304 afs_linux_clear_nfsfs_renamed(dp);
1310 #if defined(DOP_D_DELETE_TAKES_CONST)
1311 afs_dentry_delete(const struct dentry *dp)
1313 afs_dentry_delete(struct dentry *dp)
1316 if (dp->d_inode && (VTOAFS(dp->d_inode)->f.states & CUnlinked))
1317 return 1; /* bad inode? */
1322 #ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
1323 static struct vfsmount *
1324 afs_dentry_automount(afs_linux_path_t *path)
1326 struct dentry *target;
1328 /* avoid symlink resolution limits when resolving; we cannot contribute to
1329 * an infinite symlink loop */
1330 current->total_link_count--;
1332 target = canonical_dentry(path->dentry->d_inode);
1334 if (target == path->dentry) {
1341 path->dentry = target;
1344 spin_lock(&path->dentry->d_lock);
1345 path->dentry->d_flags &= ~DCACHE_NEED_AUTOMOUNT;
1346 spin_unlock(&path->dentry->d_lock);
1351 #endif /* STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
1353 struct dentry_operations afs_dentry_operations = {
1354 .d_revalidate = afs_linux_dentry_revalidate,
1355 .d_delete = afs_dentry_delete,
1356 .d_iput = afs_dentry_iput,
1357 #ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
1358 .d_automount = afs_dentry_automount,
1359 #endif /* STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
1362 /**********************************************************************
1363 * AFS Linux inode operations
1364 **********************************************************************/
1368 * Merely need to set enough of vattr to get us through the create. Note
1369 * that the higher level code (open_namei) will take care of any tuncation
1370 * explicitly. Exclusive open is also taken care of in open_namei.
1372 * name is in kernel space at this point.
1375 #if defined(IOP_CREATE_TAKES_BOOL)
1376 afs_linux_create(struct inode *dip, struct dentry *dp, umode_t mode,
1378 #elif defined(IOP_CREATE_TAKES_UMODE_T)
1379 afs_linux_create(struct inode *dip, struct dentry *dp, umode_t mode,
1380 struct nameidata *nd)
1381 #elif defined(IOP_CREATE_TAKES_NAMEIDATA)
1382 afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
1383 struct nameidata *nd)
1385 afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
1389 cred_t *credp = crref();
1390 const char *name = dp->d_name.name;
1395 vattr.va_mode = mode;
1396 vattr.va_type = mode & S_IFMT;
1399 code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
1403 struct inode *ip = AFSTOV(vcp);
1405 afs_getattr(vcp, &vattr, credp);
1406 afs_fill_inode(ip, &vattr);
1407 insert_inode_hash(ip);
1408 #if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
1409 dp->d_op = &afs_dentry_operations;
1411 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1412 d_instantiate(dp, ip);
1417 return afs_convert_code(code);
1420 /* afs_linux_lookup */
1421 static struct dentry *
1422 #if defined(IOP_LOOKUP_TAKES_UNSIGNED)
1423 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1425 #elif defined(IOP_LOOKUP_TAKES_NAMEIDATA)
1426 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1427 struct nameidata *nd)
1429 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1432 cred_t *credp = crref();
1433 struct vcache *vcp = NULL;
1434 const char *comp = dp->d_name.name;
1435 struct inode *ip = NULL;
1436 struct dentry *newdp = NULL;
1440 code = afs_lookup(VTOAFS(dip), (char *)comp, &vcp, credp);
1444 struct vcache *parent_vc = VTOAFS(dip);
1446 if (parent_vc == vcp) {
1447 /* This is possible if the parent dir is a mountpoint to a volume,
1448 * and the dir entry we looked up is a mountpoint to the same
1449 * volume. Linux cannot cope with this, so return an error instead
1450 * of risking a deadlock or panic. */
1458 afs_getattr(vcp, &vattr, credp);
1459 afs_fill_inode(ip, &vattr);
1460 if (hlist_unhashed(&ip->i_hash))
1461 insert_inode_hash(ip);
1463 #if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
1464 dp->d_op = &afs_dentry_operations;
1466 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1469 if (ip && S_ISDIR(ip->i_mode)) {
1470 d_prune_aliases(ip);
1472 #ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
1473 ip->i_flags |= S_AUTOMOUNT;
1476 newdp = d_splice_alias(ip, dp);
1481 /* It's ok for the file to not be found. That's noted by the caller by
1482 * seeing that the dp->d_inode field is NULL.
1484 if (!code || code == ENOENT)
1487 return ERR_PTR(afs_convert_code(code));
1491 afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
1494 cred_t *credp = crref();
1495 const char *name = newdp->d_name.name;
1496 struct inode *oldip = olddp->d_inode;
1498 /* If afs_link returned the vnode, we could instantiate the
1499 * dentry. Since it's not, we drop this one and do a new lookup.
1504 code = afs_link(VTOAFS(oldip), VTOAFS(dip), (char *)name, credp);
1508 return afs_convert_code(code);
1511 /* We have to have a Linux specific sillyrename function, because we
1512 * also have to keep the dcache up to date when we're doing a silly
1513 * rename - so we don't want the generic vnodeops doing this behind our
1518 afs_linux_sillyrename(struct inode *dir, struct dentry *dentry,
1521 struct vcache *tvc = VTOAFS(dentry->d_inode);
1522 struct dentry *__dp = NULL;
1523 char *__name = NULL;
1526 if (afs_linux_nfsfs_renamed(dentry))
1534 osi_FreeSmallSpace(__name);
1535 __name = afs_newname();
1538 __dp = lookup_one_len(__name, dentry->d_parent, strlen(__name));
1541 osi_FreeSmallSpace(__name);
1544 } while (__dp->d_inode != NULL);
1547 code = afs_rename(VTOAFS(dir), (char *)dentry->d_name.name,
1548 VTOAFS(dir), (char *)__dp->d_name.name,
1551 tvc->mvid = (void *) __name;
1554 crfree(tvc->uncred);
1556 tvc->uncred = credp;
1557 tvc->f.states |= CUnlinked;
1558 afs_linux_set_nfsfs_renamed(dentry);
1560 osi_FreeSmallSpace(__name);
1565 __dp->d_time = hgetlo(VTOAFS(dir)->f.m.DataVersion);
1566 d_move(dentry, __dp);
1575 afs_linux_unlink(struct inode *dip, struct dentry *dp)
1578 cred_t *credp = crref();
1579 const char *name = dp->d_name.name;
1580 struct vcache *tvc = VTOAFS(dp->d_inode);
1582 if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
1583 && !(tvc->f.states & CUnlinked)) {
1585 code = afs_linux_sillyrename(dip, dp, credp);
1588 code = afs_remove(VTOAFS(dip), (char *)name, credp);
1595 return afs_convert_code(code);
1600 afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
1603 cred_t *credp = crref();
1605 const char *name = dp->d_name.name;
1607 /* If afs_symlink returned the vnode, we could instantiate the
1608 * dentry. Since it's not, we drop this one and do a new lookup.
1614 code = afs_symlink(VTOAFS(dip), (char *)name, &vattr, (char *)target, NULL,
1618 return afs_convert_code(code);
1622 #if defined(IOP_MKDIR_TAKES_UMODE_T)
1623 afs_linux_mkdir(struct inode *dip, struct dentry *dp, umode_t mode)
1625 afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1629 cred_t *credp = crref();
1630 struct vcache *tvcp = NULL;
1632 const char *name = dp->d_name.name;
1635 vattr.va_mask = ATTR_MODE;
1636 vattr.va_mode = mode;
1638 code = afs_mkdir(VTOAFS(dip), (char *)name, &vattr, &tvcp, credp);
1641 struct inode *ip = AFSTOV(tvcp);
1643 afs_getattr(tvcp, &vattr, credp);
1644 afs_fill_inode(ip, &vattr);
1646 #if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
1647 dp->d_op = &afs_dentry_operations;
1649 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1650 d_instantiate(dp, ip);
1655 return afs_convert_code(code);
1659 afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1662 cred_t *credp = crref();
1663 const char *name = dp->d_name.name;
1665 /* locking kernel conflicts with glock? */
1668 code = afs_rmdir(VTOAFS(dip), (char *)name, credp);
1671 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1672 * that failed because a directory is not empty. So, we map
1673 * EEXIST to ENOTEMPTY on linux.
1675 if (code == EEXIST) {
1684 return afs_convert_code(code);
1689 afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1690 struct inode *newip, struct dentry *newdp)
1693 cred_t *credp = crref();
1694 const char *oldname = olddp->d_name.name;
1695 const char *newname = newdp->d_name.name;
1696 struct dentry *rehash = NULL;
1698 /* Prevent any new references during rename operation. */
1700 if (!d_unhashed(newdp)) {
1705 afs_maybe_shrink_dcache(olddp);
1708 code = afs_rename(VTOAFS(oldip), (char *)oldname, VTOAFS(newip), (char *)newname, credp);
1712 olddp->d_time = 0; /* force to revalidate */
1718 return afs_convert_code(code);
1722 /* afs_linux_ireadlink
1723 * Internal readlink which can return link contents to user or kernel space.
1724 * Note that the buffer is NOT supposed to be null-terminated.
1727 afs_linux_ireadlink(struct inode *ip, char *target, int maxlen, uio_seg_t seg)
1730 cred_t *credp = crref();
1734 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1735 code = afs_readlink(VTOAFS(ip), &tuio, credp);
1739 return maxlen - tuio.uio_resid;
1741 return afs_convert_code(code);
1744 #if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1745 /* afs_linux_readlink
1746 * Fill target (which is in user space) with contents of symlink.
1749 afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1752 struct inode *ip = dp->d_inode;
1755 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1761 /* afs_linux_follow_link
1762 * a file system dependent link following routine.
1764 static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
1769 name = kmalloc(PATH_MAX, GFP_NOFS);
1775 code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
1783 nd_set_link(nd, name);
1788 afs_linux_put_link(struct dentry *dentry, struct nameidata *nd)
1790 char *name = nd_get_link(nd);
1792 if (name && !IS_ERR(name))
1796 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1798 /* Populate a page by filling it from the cache file pointed at by cachefp
1799 * (which contains indicated chunk)
1800 * If task is NULL, the page copy occurs syncronously, and the routine
1801 * returns with page still locked. If task is non-NULL, then page copies
1802 * may occur in the background, and the page will be unlocked when it is
1806 afs_linux_read_cache(struct file *cachefp, struct page *page,
1807 int chunk, struct pagevec *lrupv,
1808 struct afs_pagecopy_task *task) {
1809 loff_t offset = page_offset(page);
1810 struct inode *cacheinode = cachefp->f_dentry->d_inode;
1811 struct page *newpage, *cachepage;
1812 struct address_space *cachemapping;
1816 cachemapping = cacheinode->i_mapping;
1820 /* If we're trying to read a page that's past the end of the disk
1821 * cache file, then just return a zeroed page */
1822 if (AFS_CHUNKOFFSET(offset) >= i_size_read(cacheinode)) {
1823 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1824 SetPageUptodate(page);
1830 /* From our offset, we now need to work out which page in the disk
1831 * file it corresponds to. This will be fun ... */
1832 pageindex = (offset - AFS_CHUNKTOBASE(chunk)) >> PAGE_CACHE_SHIFT;
1834 while (cachepage == NULL) {
1835 cachepage = find_get_page(cachemapping, pageindex);
1838 newpage = page_cache_alloc_cold(cachemapping);
1844 code = add_to_page_cache(newpage, cachemapping,
1845 pageindex, GFP_KERNEL);
1847 cachepage = newpage;
1850 page_cache_get(cachepage);
1851 if (!pagevec_add(lrupv, cachepage))
1852 __pagevec_lru_add_file(lrupv);
1855 page_cache_release(newpage);
1857 if (code != -EEXIST)
1861 lock_page(cachepage);
1865 if (!PageUptodate(cachepage)) {
1866 ClearPageError(cachepage);
1867 code = cachemapping->a_ops->readpage(NULL, cachepage);
1868 if (!code && !task) {
1869 wait_on_page_locked(cachepage);
1872 unlock_page(cachepage);
1876 if (PageUptodate(cachepage)) {
1877 copy_highpage(page, cachepage);
1878 flush_dcache_page(page);
1879 SetPageUptodate(page);
1884 afs_pagecopy_queue_page(task, cachepage, page);
1896 page_cache_release(cachepage);
1902 afs_linux_readpage_fastpath(struct file *fp, struct page *pp, int *codep)
1904 loff_t offset = page_offset(pp);
1905 struct inode *ip = FILE_INODE(fp);
1906 struct vcache *avc = VTOAFS(ip);
1908 struct file *cacheFp = NULL;
1911 struct pagevec lrupv;
1913 /* Not a UFS cache, don't do anything */
1914 if (cacheDiskType != AFS_FCACHE_TYPE_UFS)
1917 /* No readpage (ex: tmpfs) , skip */
1918 if (cachefs_noreadpage)
1921 /* Can't do anything if the vcache isn't statd , or if the read
1922 * crosses a chunk boundary.
1924 if (!(avc->f.states & CStatd) ||
1925 AFS_CHUNK(offset) != AFS_CHUNK(offset + PAGE_SIZE)) {
1929 ObtainWriteLock(&avc->lock, 911);
1931 /* XXX - See if hinting actually makes things faster !!! */
1933 /* See if we have a suitable entry already cached */
1937 /* We need to lock xdcache, then dcache, to handle situations where
1938 * the hint is on the free list. However, we can't safely do this
1939 * according to the locking hierarchy. So, use a non blocking lock.
1941 ObtainReadLock(&afs_xdcache);
1942 dcLocked = ( 0 == NBObtainReadLock(&tdc->lock));
1944 if (dcLocked && (tdc->index != NULLIDX)
1945 && !FidCmp(&tdc->f.fid, &avc->f.fid)
1946 && tdc->f.chunk == AFS_CHUNK(offset)
1947 && !(afs_indexFlags[tdc->index] & (IFFree | IFDiscarded))) {
1948 /* Bonus - the hint was correct */
1951 /* Only destroy the hint if its actually invalid, not if there's
1952 * just been a locking failure */
1954 ReleaseReadLock(&tdc->lock);
1961 ReleaseReadLock(&afs_xdcache);
1964 /* No hint, or hint is no longer valid - see if we can get something
1965 * directly from the dcache
1968 tdc = afs_FindDCache(avc, offset);
1971 ReleaseWriteLock(&avc->lock);
1976 ObtainReadLock(&tdc->lock);
1978 /* Is the dcache we've been given currently up to date */
1979 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
1980 (tdc->dflags & DFFetching))
1983 /* Update our hint for future abuse */
1986 /* Okay, so we've now got a cache file that is up to date */
1988 /* XXX - I suspect we should be locking the inodes before we use them! */
1990 cacheFp = afs_linux_raw_open(&tdc->f.inode);
1991 if (!cacheFp->f_dentry->d_inode->i_mapping->a_ops->readpage) {
1992 cachefs_noreadpage = 1;
1996 pagevec_init(&lrupv, 0);
1998 code = afs_linux_read_cache(cacheFp, pp, tdc->f.chunk, &lrupv, NULL);
2000 if (pagevec_count(&lrupv))
2001 __pagevec_lru_add_file(&lrupv);
2003 filp_close(cacheFp, NULL);
2006 ReleaseReadLock(&tdc->lock);
2007 ReleaseWriteLock(&avc->lock);
2014 ReleaseWriteLock(&avc->lock);
2015 ReleaseReadLock(&tdc->lock);
2020 /* afs_linux_readpage
2022 * This function is split into two, because prepare_write/begin_write
2023 * require a readpage call which doesn't unlock the resulting page upon
2027 afs_linux_fillpage(struct file *fp, struct page *pp)
2032 struct iovec *iovecp;
2033 struct inode *ip = FILE_INODE(fp);
2034 afs_int32 cnt = page_count(pp);
2035 struct vcache *avc = VTOAFS(ip);
2036 afs_offs_t offset = page_offset(pp);
2040 if (afs_linux_readpage_fastpath(fp, pp, &code)) {
2050 auio = kmalloc(sizeof(struct uio), GFP_NOFS);
2051 iovecp = kmalloc(sizeof(struct iovec), GFP_NOFS);
2053 setup_uio(auio, iovecp, (char *)address, offset, PAGE_SIZE, UIO_READ,
2058 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
2059 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
2060 99999); /* not a possible code value */
2062 code = afs_rdwr(avc, auio, UIO_READ, 0, credp);
2064 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
2065 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
2067 AFS_DISCON_UNLOCK();
2070 /* XXX valid for no-cache also? Check last bits of files... :)
2071 * Cognate code goes in afs_NoCacheFetchProc. */
2072 if (auio->uio_resid) /* zero remainder of page */
2073 memset((void *)(address + (PAGE_SIZE - auio->uio_resid)), 0,
2076 flush_dcache_page(pp);
2077 SetPageUptodate(pp);
2086 return afs_convert_code(code);
2090 afs_linux_prefetch(struct file *fp, struct page *pp)
2093 struct vcache *avc = VTOAFS(FILE_INODE(fp));
2094 afs_offs_t offset = page_offset(pp);
2096 if (AFS_CHUNKOFFSET(offset) == 0) {
2098 struct vrequest treq;
2103 code = afs_InitReq(&treq, credp);
2104 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
2105 tdc = afs_FindDCache(avc, offset);
2107 if (!(tdc->mflags & DFNextStarted))
2108 afs_PrefetchChunk(avc, tdc, credp, &treq);
2111 ReleaseWriteLock(&avc->lock);
2116 return afs_convert_code(code);
2121 afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
2122 struct list_head *page_list, unsigned num_pages)
2127 struct iovec* iovecp;
2128 struct nocache_read_request *ancr;
2130 struct pagevec lrupv;
2134 struct inode *ip = FILE_INODE(fp);
2135 struct vcache *avc = VTOAFS(ip);
2136 afs_int32 base_index = 0;
2137 afs_int32 page_count = 0;
2140 /* background thread must free: iovecp, auio, ancr */
2141 iovecp = osi_Alloc(num_pages * sizeof(struct iovec));
2143 auio = osi_Alloc(sizeof(struct uio));
2144 auio->uio_iov = iovecp;
2145 auio->uio_iovcnt = num_pages;
2146 auio->uio_flag = UIO_READ;
2147 auio->uio_seg = AFS_UIOSYS;
2148 auio->uio_resid = num_pages * PAGE_SIZE;
2150 ancr = osi_Alloc(sizeof(struct nocache_read_request));
2152 ancr->offset = auio->uio_offset;
2153 ancr->length = auio->uio_resid;
2155 pagevec_init(&lrupv, 0);
2157 for(page_ix = 0; page_ix < num_pages; ++page_ix) {
2159 if(list_empty(page_list))
2162 pp = list_entry(page_list->prev, struct page, lru);
2163 /* If we allocate a page and don't remove it from page_list,
2164 * the page cache gets upset. */
2166 isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
2167 if(pp->index > isize) {
2174 offset = page_offset(pp);
2175 ancr->offset = auio->uio_offset = offset;
2176 base_index = pp->index;
2178 iovecp[page_ix].iov_len = PAGE_SIZE;
2179 code = add_to_page_cache(pp, mapping, pp->index, GFP_KERNEL);
2180 if(base_index != pp->index) {
2183 page_cache_release(pp);
2184 iovecp[page_ix].iov_base = (void *) 0;
2186 ancr->length -= PAGE_SIZE;
2193 page_cache_release(pp);
2194 iovecp[page_ix].iov_base = (void *) 0;
2197 if(!PageLocked(pp)) {
2201 /* increment page refcount--our original design assumed
2202 * that locking it would effectively pin it; protect
2203 * ourselves from the possiblity that this assumption is
2204 * is faulty, at low cost (provided we do not fail to
2205 * do the corresponding decref on the other side) */
2208 /* save the page for background map */
2209 iovecp[page_ix].iov_base = (void*) pp;
2211 /* and put it on the LRU cache */
2212 if (!pagevec_add(&lrupv, pp))
2213 __pagevec_lru_add_file(&lrupv);
2217 /* If there were useful pages in the page list, make sure all pages
2218 * are in the LRU cache, then schedule the read */
2220 if (pagevec_count(&lrupv))
2221 __pagevec_lru_add_file(&lrupv);
2223 code = afs_ReadNoCache(avc, ancr, credp);
2226 /* If there is nothing for the background thread to handle,
2227 * it won't be freeing the things that we never gave it */
2228 osi_Free(iovecp, num_pages * sizeof(struct iovec));
2229 osi_Free(auio, sizeof(struct uio));
2230 osi_Free(ancr, sizeof(struct nocache_read_request));
2232 /* we do not flush, release, or unmap pages--that will be
2233 * done for us by the background thread as each page comes in
2234 * from the fileserver */
2235 return afs_convert_code(code);
2240 afs_linux_bypass_readpage(struct file *fp, struct page *pp)
2242 cred_t *credp = NULL;
2244 struct iovec *iovecp;
2245 struct nocache_read_request *ancr;
2249 * Special case: if page is at or past end of file, just zero it and set
2252 if (page_offset(pp) >= i_size_read(fp->f_mapping->host)) {
2253 zero_user_segment(pp, 0, PAGE_CACHE_SIZE);
2254 SetPageUptodate(pp);
2261 /* receiver frees */
2262 auio = osi_Alloc(sizeof(struct uio));
2263 iovecp = osi_Alloc(sizeof(struct iovec));
2265 /* address can be NULL, because we overwrite it with 'pp', below */
2266 setup_uio(auio, iovecp, NULL, page_offset(pp),
2267 PAGE_SIZE, UIO_READ, AFS_UIOSYS);
2269 /* save the page for background map */
2270 get_page(pp); /* see above */
2271 auio->uio_iov->iov_base = (void*) pp;
2272 /* the background thread will free this */
2273 ancr = osi_Alloc(sizeof(struct nocache_read_request));
2275 ancr->offset = page_offset(pp);
2276 ancr->length = PAGE_SIZE;
2279 code = afs_ReadNoCache(VTOAFS(FILE_INODE(fp)), ancr, credp);
2282 return afs_convert_code(code);
2286 afs_linux_can_bypass(struct inode *ip) {
2288 switch(cache_bypass_strategy) {
2289 case NEVER_BYPASS_CACHE:
2291 case ALWAYS_BYPASS_CACHE:
2293 case LARGE_FILES_BYPASS_CACHE:
2294 if (i_size_read(ip) > cache_bypass_threshold)
2301 /* Check if a file is permitted to bypass the cache by policy, and modify
2302 * the cache bypass state recorded for that file */
2305 afs_linux_bypass_check(struct inode *ip) {
2308 int bypass = afs_linux_can_bypass(ip);
2311 trydo_cache_transition(VTOAFS(ip), credp, bypass);
2319 afs_linux_readpage(struct file *fp, struct page *pp)
2323 if (afs_linux_bypass_check(FILE_INODE(fp))) {
2324 code = afs_linux_bypass_readpage(fp, pp);
2326 code = afs_linux_fillpage(fp, pp);
2328 code = afs_linux_prefetch(fp, pp);
2335 /* Readpages reads a number of pages for a particular file. We use
2336 * this to optimise the reading, by limiting the number of times upon which
2337 * we have to lookup, lock and open vcaches and dcaches
2341 afs_linux_readpages(struct file *fp, struct address_space *mapping,
2342 struct list_head *page_list, unsigned int num_pages)
2344 struct inode *inode = mapping->host;
2345 struct vcache *avc = VTOAFS(inode);
2347 struct file *cacheFp = NULL;
2349 unsigned int page_idx;
2351 struct pagevec lrupv;
2352 struct afs_pagecopy_task *task;
2354 if (afs_linux_bypass_check(inode))
2355 return afs_linux_bypass_readpages(fp, mapping, page_list, num_pages);
2357 if (cacheDiskType == AFS_FCACHE_TYPE_MEM)
2360 /* No readpage (ex: tmpfs) , skip */
2361 if (cachefs_noreadpage)
2365 if ((code = afs_linux_VerifyVCache(avc, NULL))) {
2370 ObtainWriteLock(&avc->lock, 912);
2373 task = afs_pagecopy_init_task();
2376 pagevec_init(&lrupv, 0);
2377 for (page_idx = 0; page_idx < num_pages; page_idx++) {
2378 struct page *page = list_entry(page_list->prev, struct page, lru);
2379 list_del(&page->lru);
2380 offset = page_offset(page);
2382 if (tdc && tdc->f.chunk != AFS_CHUNK(offset)) {
2384 ReleaseReadLock(&tdc->lock);
2389 filp_close(cacheFp, NULL);
2394 if ((tdc = afs_FindDCache(avc, offset))) {
2395 ObtainReadLock(&tdc->lock);
2396 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
2397 (tdc->dflags & DFFetching)) {
2398 ReleaseReadLock(&tdc->lock);
2405 cacheFp = afs_linux_raw_open(&tdc->f.inode);
2406 if (!cacheFp->f_dentry->d_inode->i_mapping->a_ops->readpage) {
2407 cachefs_noreadpage = 1;
2413 if (tdc && !add_to_page_cache(page, mapping, page->index,
2415 page_cache_get(page);
2416 if (!pagevec_add(&lrupv, page))
2417 __pagevec_lru_add_file(&lrupv);
2419 afs_linux_read_cache(cacheFp, page, tdc->f.chunk, &lrupv, task);
2421 page_cache_release(page);
2423 if (pagevec_count(&lrupv))
2424 __pagevec_lru_add_file(&lrupv);
2428 filp_close(cacheFp, NULL);
2430 afs_pagecopy_put_task(task);
2434 ReleaseReadLock(&tdc->lock);
2438 ReleaseWriteLock(&avc->lock);
2443 /* Prepare an AFS vcache for writeback. Should be called with the vcache
2446 afs_linux_prepare_writeback(struct vcache *avc) {
2447 if (avc->f.states & CPageWrite) {
2448 return AOP_WRITEPAGE_ACTIVATE;
2450 avc->f.states |= CPageWrite;
2455 afs_linux_dopartialwrite(struct vcache *avc, cred_t *credp) {
2456 struct vrequest treq;
2459 if (!afs_InitReq(&treq, credp))
2460 code = afs_DoPartialWrite(avc, &treq);
2462 return afs_convert_code(code);
2466 afs_linux_complete_writeback(struct vcache *avc) {
2467 avc->f.states &= ~CPageWrite;
2470 /* Writeback a given page syncronously. Called with no AFS locks held */
2472 afs_linux_page_writeback(struct inode *ip, struct page *pp,
2473 unsigned long offset, unsigned int count,
2476 struct vcache *vcp = VTOAFS(ip);
2484 buffer = kmap(pp) + offset;
2485 base = page_offset(pp) + offset;
2488 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2489 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2490 ICL_TYPE_INT32, 99999);
2492 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
2494 code = afs_write(vcp, &tuio, f_flags, credp, 0);
2496 i_size_write(ip, vcp->f.m.Length);
2497 ip->i_blocks = ((vcp->f.m.Length + 1023) >> 10) << 1;
2499 code = code ? afs_convert_code(code) : count - tuio.uio_resid;
2501 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2502 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2503 ICL_TYPE_INT32, code);
2512 afs_linux_writepage_sync(struct inode *ip, struct page *pp,
2513 unsigned long offset, unsigned int count)
2517 struct vcache *vcp = VTOAFS(ip);
2520 /* Catch recursive writeback. This occurs if the kernel decides
2521 * writeback is required whilst we are writing to the cache, or
2522 * flushing to the server. When we're running syncronously (as
2523 * opposed to from writepage) we can't actually do anything about
2524 * this case - as we can't return AOP_WRITEPAGE_ACTIVATE to write()
2527 ObtainWriteLock(&vcp->lock, 532);
2528 afs_linux_prepare_writeback(vcp);
2529 ReleaseWriteLock(&vcp->lock);
2533 code = afs_linux_page_writeback(ip, pp, offset, count, credp);
2536 ObtainWriteLock(&vcp->lock, 533);
2538 code1 = afs_linux_dopartialwrite(vcp, credp);
2539 afs_linux_complete_writeback(vcp);
2540 ReleaseWriteLock(&vcp->lock);
2551 #ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
2552 afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
2554 afs_linux_writepage(struct page *pp)
2557 struct address_space *mapping = pp->mapping;
2558 struct inode *inode;
2561 unsigned int to = PAGE_CACHE_SIZE;
2566 if (PageReclaim(pp)) {
2567 return AOP_WRITEPAGE_ACTIVATE;
2568 /* XXX - Do we need to redirty the page here? */
2573 inode = mapping->host;
2574 vcp = VTOAFS(inode);
2575 isize = i_size_read(inode);
2577 /* Don't defeat an earlier truncate */
2578 if (page_offset(pp) > isize) {
2579 set_page_writeback(pp);
2585 ObtainWriteLock(&vcp->lock, 537);
2586 code = afs_linux_prepare_writeback(vcp);
2587 if (code == AOP_WRITEPAGE_ACTIVATE) {
2588 /* WRITEPAGE_ACTIVATE is the only return value that permits us
2589 * to return with the page still locked */
2590 ReleaseWriteLock(&vcp->lock);
2595 /* Grab the creds structure currently held in the vnode, and
2596 * get a reference to it, in case it goes away ... */
2602 ReleaseWriteLock(&vcp->lock);
2605 set_page_writeback(pp);
2607 SetPageUptodate(pp);
2609 /* We can unlock the page here, because it's protected by the
2610 * page_writeback flag. This should make us less vulnerable to
2611 * deadlocking in afs_write and afs_DoPartialWrite
2615 /* If this is the final page, then just write the number of bytes that
2616 * are actually in it */
2617 if ((isize - page_offset(pp)) < to )
2618 to = isize - page_offset(pp);
2620 code = afs_linux_page_writeback(inode, pp, 0, to, credp);
2623 ObtainWriteLock(&vcp->lock, 538);
2625 /* As much as we might like to ignore a file server error here,
2626 * and just try again when we close(), unfortunately StoreAllSegments
2627 * will invalidate our chunks if the server returns a permanent error,
2628 * so we need to at least try and get that error back to the user
2631 code1 = afs_linux_dopartialwrite(vcp, credp);
2633 afs_linux_complete_writeback(vcp);
2634 ReleaseWriteLock(&vcp->lock);
2639 end_page_writeback(pp);
2640 page_cache_release(pp);
2651 /* afs_linux_permission
2652 * Check access rights - returns error if can't check or permission denied.
2655 #if defined(IOP_PERMISSION_TAKES_FLAGS)
2656 afs_linux_permission(struct inode *ip, int mode, unsigned int flags)
2657 #elif defined(IOP_PERMISSION_TAKES_NAMEIDATA)
2658 afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
2660 afs_linux_permission(struct inode *ip, int mode)
2667 /* Check for RCU path walking */
2668 #if defined(IOP_PERMISSION_TAKES_FLAGS)
2669 if (flags & IPERM_FLAG_RCU)
2671 #elif defined(MAY_NOT_BLOCK)
2672 if (mode & MAY_NOT_BLOCK)
2678 if (mode & MAY_EXEC)
2680 if (mode & MAY_READ)
2682 if (mode & MAY_WRITE)
2684 code = afs_access(VTOAFS(ip), tmp, credp);
2688 return afs_convert_code(code);
2692 afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
2696 struct inode *inode = FILE_INODE(file);
2697 loff_t pagebase = page_offset(page);
2699 if (i_size_read(inode) < (pagebase + offset))
2700 i_size_write(inode, pagebase + offset);
2702 if (PageChecked(page)) {
2703 SetPageUptodate(page);
2704 ClearPageChecked(page);
2707 code = afs_linux_writepage_sync(inode, page, offset, to - offset);
2713 afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
2717 /* http://kerneltrap.org/node/4941 details the expected behaviour of
2718 * prepare_write. Essentially, if the page exists within the file,
2719 * and is not being fully written, then we should populate it.
2722 if (!PageUptodate(page)) {
2723 loff_t pagebase = page_offset(page);
2724 loff_t isize = i_size_read(page->mapping->host);
2726 /* Is the location we are writing to beyond the end of the file? */
2727 if (pagebase >= isize ||
2728 ((from == 0) && (pagebase + to) >= isize)) {
2729 zero_user_segments(page, 0, from, to, PAGE_CACHE_SIZE);
2730 SetPageChecked(page);
2731 /* Are we we writing a full page */
2732 } else if (from == 0 && to == PAGE_CACHE_SIZE) {
2733 SetPageChecked(page);
2734 /* Is the page readable, if it's wronly, we don't care, because we're
2735 * not actually going to read from it ... */
2736 } else if ((file->f_flags && O_ACCMODE) != O_WRONLY) {
2737 /* We don't care if fillpage fails, because if it does the page
2738 * won't be marked as up to date
2740 afs_linux_fillpage(file, page);
2746 #if defined(STRUCT_ADDRESS_SPACE_OPERATIONS_HAS_WRITE_BEGIN)
2748 afs_linux_write_end(struct file *file, struct address_space *mapping,
2749 loff_t pos, unsigned len, unsigned copied,
2750 struct page *page, void *fsdata)
2753 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
2755 code = afs_linux_commit_write(file, page, from, from + len);
2758 page_cache_release(page);
2763 afs_linux_write_begin(struct file *file, struct address_space *mapping,
2764 loff_t pos, unsigned len, unsigned flags,
2765 struct page **pagep, void **fsdata)
2768 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2769 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
2772 page = grab_cache_page_write_begin(mapping, index, flags);
2775 code = afs_linux_prepare_write(file, page, from, from + len);
2778 page_cache_release(page);
2785 #ifndef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
2787 afs_linux_dir_follow_link(struct dentry *dentry, struct nameidata *nd)
2789 struct dentry **dpp;
2790 struct dentry *target;
2792 if (current->total_link_count > 0) {
2793 /* avoid symlink resolution limits when resolving; we cannot contribute to
2794 * an infinite symlink loop */
2795 /* only do this for follow_link when total_link_count is positive to be
2796 * on the safe side; there is at least one code path in the Linux
2797 * kernel where it seems like it may be possible to get here without
2798 * total_link_count getting incremented. it is not clear on how that
2799 * path is actually reached, but guard against it just to be safe */
2800 current->total_link_count--;
2803 target = canonical_dentry(dentry->d_inode);
2805 # ifdef STRUCT_NAMEIDATA_HAS_PATH
2806 dpp = &nd->path.dentry;
2816 *dpp = dget(dentry);
2819 nd->last_type = LAST_BIND;
2823 #endif /* !STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
2826 static struct inode_operations afs_file_iops = {
2827 .permission = afs_linux_permission,
2828 .getattr = afs_linux_getattr,
2829 .setattr = afs_notify_change,
2832 static struct address_space_operations afs_file_aops = {
2833 .readpage = afs_linux_readpage,
2834 .readpages = afs_linux_readpages,
2835 .writepage = afs_linux_writepage,
2836 #if defined (STRUCT_ADDRESS_SPACE_OPERATIONS_HAS_WRITE_BEGIN)
2837 .write_begin = afs_linux_write_begin,
2838 .write_end = afs_linux_write_end,
2840 .commit_write = afs_linux_commit_write,
2841 .prepare_write = afs_linux_prepare_write,
2846 /* Separate ops vector for directories. Linux 2.2 tests type of inode
2847 * by what sort of operation is allowed.....
2850 static struct inode_operations afs_dir_iops = {
2851 .setattr = afs_notify_change,
2852 .create = afs_linux_create,
2853 .lookup = afs_linux_lookup,
2854 .link = afs_linux_link,
2855 .unlink = afs_linux_unlink,
2856 .symlink = afs_linux_symlink,
2857 .mkdir = afs_linux_mkdir,
2858 .rmdir = afs_linux_rmdir,
2859 .rename = afs_linux_rename,
2860 .getattr = afs_linux_getattr,
2861 .permission = afs_linux_permission,
2862 #ifndef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
2863 .follow_link = afs_linux_dir_follow_link,
2867 /* We really need a separate symlink set of ops, since do_follow_link()
2868 * determines if it _is_ a link by checking if the follow_link op is set.
2870 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2872 afs_symlink_filler(struct file *file, struct page *page)
2874 struct inode *ip = (struct inode *)page->mapping->host;
2875 char *p = (char *)kmap(page);
2879 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
2884 p[code] = '\0'; /* null terminate? */
2886 SetPageUptodate(page);
2898 static struct address_space_operations afs_symlink_aops = {
2899 .readpage = afs_symlink_filler
2901 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
2903 static struct inode_operations afs_symlink_iops = {
2904 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2905 .readlink = page_readlink,
2906 # if defined(HAVE_LINUX_PAGE_FOLLOW_LINK)
2907 .follow_link = page_follow_link,
2909 .follow_link = page_follow_link_light,
2910 .put_link = page_put_link,
2912 #else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
2913 .readlink = afs_linux_readlink,
2914 .follow_link = afs_linux_follow_link,
2915 .put_link = afs_linux_put_link,
2916 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
2917 .setattr = afs_notify_change,
2921 afs_fill_inode(struct inode *ip, struct vattr *vattr)
2925 vattr2inode(ip, vattr);
2927 ip->i_mapping->backing_dev_info = afs_backing_dev_info;
2928 /* Reset ops if symlink or directory. */
2929 if (S_ISREG(ip->i_mode)) {
2930 ip->i_op = &afs_file_iops;
2931 ip->i_fop = &afs_file_fops;
2932 ip->i_data.a_ops = &afs_file_aops;
2934 } else if (S_ISDIR(ip->i_mode)) {
2935 ip->i_op = &afs_dir_iops;
2936 ip->i_fop = &afs_dir_fops;
2938 } else if (S_ISLNK(ip->i_mode)) {
2939 ip->i_op = &afs_symlink_iops;
2940 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2941 ip->i_data.a_ops = &afs_symlink_aops;
2942 ip->i_mapping = &ip->i_data;