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) {
79 struct vrequest *treq = NULL;
82 if (avc->f.states & CStatd) {
90 code = afs_CreateReq(&treq, credp);
92 code = afs_VerifyVCache2(avc, treq);
101 return afs_convert_code(code);
104 #if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER) || defined(HAVE_LINUX_GENERIC_FILE_AIO_READ)
105 # if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER)
107 afs_linux_read_iter(struct kiocb *iocb, struct iov_iter *iter)
108 # elif defined(LINUX_HAS_NONVECTOR_AIO)
110 afs_linux_aio_read(struct kiocb *iocb, char __user *buf, size_t bufsize,
114 afs_linux_aio_read(struct kiocb *iocb, const struct iovec *buf,
115 unsigned long bufsize, loff_t pos)
118 struct file *fp = iocb->ki_filp;
120 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
121 # if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER)
122 loff_t pos = iocb->ki_pos;
123 unsigned long bufsize = iter->nr_segs;
128 afs_Trace4(afs_iclSetp, CM_TRACE_AIOREADOP, ICL_TYPE_POINTER, vcp,
129 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
130 (afs_int32)bufsize, ICL_TYPE_INT32, 99999);
131 code = afs_linux_VerifyVCache(vcp, NULL);
134 /* Linux's FlushPages implementation doesn't ever use credp,
135 * so we optimise by not using it */
136 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
138 # if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER)
139 code = generic_file_read_iter(iocb, iter);
141 code = generic_file_aio_read(iocb, buf, bufsize, pos);
146 afs_Trace4(afs_iclSetp, CM_TRACE_AIOREADOP, ICL_TYPE_POINTER, vcp,
147 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
148 (afs_int32)bufsize, ICL_TYPE_INT32, code);
154 afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
157 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
160 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
161 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
163 code = afs_linux_VerifyVCache(vcp, NULL);
166 /* Linux's FlushPages implementation doesn't ever use credp,
167 * so we optimise by not using it */
168 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
170 code = do_sync_read(fp, buf, count, offp);
174 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
175 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
183 /* Now we have integrated VM for writes as well as reads. the generic write operations
184 * also take care of re-positioning the pointer if file is open in append
185 * mode. Call fake open/close to ensure we do writes of core dumps.
187 #if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER) || defined(HAVE_LINUX_GENERIC_FILE_AIO_READ)
188 # if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER)
190 afs_linux_write_iter(struct kiocb *iocb, struct iov_iter *iter)
191 # elif defined(LINUX_HAS_NONVECTOR_AIO)
193 afs_linux_aio_write(struct kiocb *iocb, const char __user *buf, size_t bufsize,
197 afs_linux_aio_write(struct kiocb *iocb, const struct iovec *buf,
198 unsigned long bufsize, loff_t pos)
202 struct vcache *vcp = VTOAFS(iocb->ki_filp->f_dentry->d_inode);
204 # if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER)
205 loff_t pos = iocb->ki_pos;
206 unsigned long bufsize = iter->nr_segs;
211 afs_Trace4(afs_iclSetp, CM_TRACE_AIOWRITEOP, ICL_TYPE_POINTER, vcp,
212 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
213 (afs_int32)bufsize, ICL_TYPE_INT32,
214 (iocb->ki_filp->f_flags & O_APPEND) ? 99998 : 99999);
216 code = afs_linux_VerifyVCache(vcp, &credp);
218 ObtainWriteLock(&vcp->lock, 529);
220 ReleaseWriteLock(&vcp->lock);
223 # if defined(STRUCT_FILE_OPERATIONS_HAS_READ_ITER)
224 code = generic_file_write_iter(iocb, iter);
226 code = generic_file_aio_write(iocb, buf, bufsize, pos);
231 ObtainWriteLock(&vcp->lock, 530);
233 if (vcp->execsOrWriters == 1 && !credp)
236 afs_FakeClose(vcp, credp);
237 ReleaseWriteLock(&vcp->lock);
239 afs_Trace4(afs_iclSetp, CM_TRACE_AIOWRITEOP, ICL_TYPE_POINTER, vcp,
240 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(pos), ICL_TYPE_INT32,
241 (afs_int32)bufsize, ICL_TYPE_INT32, code);
250 afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
253 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
258 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
259 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
260 (fp->f_flags & O_APPEND) ? 99998 : 99999);
262 code = afs_linux_VerifyVCache(vcp, &credp);
264 ObtainWriteLock(&vcp->lock, 529);
266 ReleaseWriteLock(&vcp->lock);
269 code = do_sync_write(fp, buf, count, offp);
273 ObtainWriteLock(&vcp->lock, 530);
275 if (vcp->execsOrWriters == 1 && !credp)
278 afs_FakeClose(vcp, credp);
279 ReleaseWriteLock(&vcp->lock);
281 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
282 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
292 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
294 /* This is a complete rewrite of afs_readdir, since we can make use of
295 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
296 * handling and use of bulkstats will need to be reflected here as well.
299 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
300 afs_linux_readdir(struct file *fp, struct dir_context *ctx)
302 afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
305 struct vcache *avc = VTOAFS(FILE_INODE(fp));
306 struct vrequest *treq = NULL;
312 struct DirBuffer entry;
315 afs_size_t origOffset, tlen;
316 cred_t *credp = crref();
317 struct afs_fakestat_state fakestat;
320 AFS_STATCNT(afs_readdir);
322 code = afs_convert_code(afs_CreateReq(&treq, credp));
327 afs_InitFakeStat(&fakestat);
328 code = afs_convert_code(afs_EvalFakeStat(&avc, &fakestat, treq));
332 /* update the cache entry */
334 code = afs_convert_code(afs_VerifyVCache2(avc, treq));
338 /* get a reference to the entire directory */
339 tdc = afs_GetDCache(avc, (afs_size_t) 0, treq, &origOffset, &tlen, 1);
345 ObtainWriteLock(&avc->lock, 811);
346 ObtainReadLock(&tdc->lock);
348 * Make sure that the data in the cache is current. There are two
349 * cases we need to worry about:
350 * 1. The cache data is being fetched by another process.
351 * 2. The cache data is no longer valid
353 while ((avc->f.states & CStatd)
354 && (tdc->dflags & DFFetching)
355 && hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
356 ReleaseReadLock(&tdc->lock);
357 ReleaseWriteLock(&avc->lock);
358 afs_osi_Sleep(&tdc->validPos);
359 ObtainWriteLock(&avc->lock, 812);
360 ObtainReadLock(&tdc->lock);
362 if (!(avc->f.states & CStatd)
363 || !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
364 ReleaseReadLock(&tdc->lock);
365 ReleaseWriteLock(&avc->lock);
370 /* Set the readdir-in-progress flag, and downgrade the lock
371 * to shared so others will be able to acquire a read lock.
373 avc->f.states |= CReadDir;
374 avc->dcreaddir = tdc;
375 avc->readdir_pid = MyPidxx2Pid(MyPidxx);
376 ConvertWToSLock(&avc->lock);
378 /* Fill in until we get an error or we're done. This implementation
379 * takes an offset in units of blobs, rather than bytes.
382 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
385 offset = (int) fp->f_pos;
388 dirpos = BlobScan(tdc, offset);
392 code = afs_dir_GetVerifiedBlob(tdc, dirpos, &entry);
394 if (!(avc->f.states & CCorrupt)) {
395 struct cell *tc = afs_GetCellStale(avc->f.fid.Cell, READ_LOCK);
396 afs_warn("Corrupt directory (%d.%d.%d.%d [%s] @%lx, pos %d)",
397 avc->f.fid.Cell, avc->f.fid.Fid.Volume,
398 avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique,
399 tc ? tc->cellName : "",
400 (unsigned long)&tdc->f.inode, dirpos);
402 afs_PutCell(tc, READ_LOCK);
403 UpgradeSToWLock(&avc->lock, 814);
404 avc->f.states |= CCorrupt;
410 de = (struct DirEntry *)entry.data;
411 ino = afs_calc_inum (avc->f.fid.Cell, avc->f.fid.Fid.Volume,
412 ntohl(de->fid.vnode));
413 len = strlen(de->name);
415 /* filldir returns -EINVAL when the buffer is full. */
417 unsigned int type = DT_UNKNOWN;
418 struct VenusFid afid;
421 afid.Cell = avc->f.fid.Cell;
422 afid.Fid.Volume = avc->f.fid.Fid.Volume;
423 afid.Fid.Vnode = ntohl(de->fid.vnode);
424 afid.Fid.Unique = ntohl(de->fid.vunique);
425 if ((avc->f.states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
427 } else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
430 } else if (((tvc->f.states) & (CStatd | CTruth))) {
431 /* CTruth will be set if the object has
436 else if (vtype == VREG)
438 /* Don't do this until we're sure it can't be a mtpt */
439 /* else if (vtype == VLNK)
441 /* what other types does AFS support? */
443 /* clean up from afs_FindVCache */
447 * If this is NFS readdirplus, then the filler is going to
448 * call getattr on this inode, which will deadlock if we're
452 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
453 /* dir_emit returns a bool - true when it succeeds.
454 * Inverse the result to fit with how we check "code" */
455 code = !dir_emit(ctx, de->name, len, ino, type);
457 code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
464 offset = dirpos + 1 + ((len + 16) >> 5);
466 /* If filldir didn't fill in the last one this is still pointing to that
472 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
473 ctx->pos = (loff_t) offset;
475 fp->f_pos = (loff_t) offset;
477 ReleaseReadLock(&tdc->lock);
479 UpgradeSToWLock(&avc->lock, 813);
480 avc->f.states &= ~CReadDir;
482 avc->readdir_pid = 0;
483 ReleaseSharedLock(&avc->lock);
486 afs_PutFakeStat(&fakestat);
487 afs_DestroyReq(treq);
494 /* in afs_pioctl.c */
495 extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
498 #if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
499 static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
501 return afs_xioctl(FILE_INODE(fp), fp, com, arg);
508 afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
510 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
514 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
515 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
516 vmap->vm_end - vmap->vm_start);
518 /* get a validated vcache entry */
519 code = afs_linux_VerifyVCache(vcp, NULL);
522 /* Linux's Flushpage implementation doesn't use credp, so optimise
523 * our code to not need to crref() it */
524 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
526 code = generic_file_mmap(fp, vmap);
529 vcp->f.states |= CMAPPED;
537 afs_linux_open(struct inode *ip, struct file *fp)
539 struct vcache *vcp = VTOAFS(ip);
540 cred_t *credp = crref();
544 code = afs_open(&vcp, fp->f_flags, credp);
548 return afs_convert_code(code);
552 afs_linux_release(struct inode *ip, struct file *fp)
554 struct vcache *vcp = VTOAFS(ip);
555 cred_t *credp = crref();
559 code = afs_close(vcp, fp->f_flags, credp);
560 ObtainWriteLock(&vcp->lock, 807);
565 ReleaseWriteLock(&vcp->lock);
569 return afs_convert_code(code);
573 #if defined(FOP_FSYNC_TAKES_DENTRY)
574 afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
575 #elif defined(FOP_FSYNC_TAKES_RANGE)
576 afs_linux_fsync(struct file *fp, loff_t start, loff_t end, int datasync)
578 afs_linux_fsync(struct file *fp, int datasync)
582 struct inode *ip = FILE_INODE(fp);
583 cred_t *credp = crref();
585 #if defined(FOP_FSYNC_TAKES_RANGE)
586 mutex_lock(&ip->i_mutex);
589 code = afs_fsync(VTOAFS(ip), credp);
591 #if defined(FOP_FSYNC_TAKES_RANGE)
592 mutex_unlock(&ip->i_mutex);
595 return afs_convert_code(code);
601 afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
604 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
605 cred_t *credp = crref();
606 struct AFS_FLOCK flock;
608 /* Convert to a lock format afs_lockctl understands. */
609 memset(&flock, 0, sizeof(flock));
610 flock.l_type = flp->fl_type;
611 flock.l_pid = flp->fl_pid;
613 flock.l_start = flp->fl_start;
614 if (flp->fl_end == OFFSET_MAX)
615 flock.l_len = 0; /* Lock to end of file */
617 flock.l_len = flp->fl_end - flp->fl_start + 1;
619 /* Safe because there are no large files, yet */
620 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
621 if (cmd == F_GETLK64)
623 else if (cmd == F_SETLK64)
625 else if (cmd == F_SETLKW64)
627 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
630 code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
633 if ((code == 0 || flp->fl_type == F_UNLCK) &&
634 (cmd == F_SETLK || cmd == F_SETLKW)) {
635 code = afs_posix_lock_file(fp, flp);
636 if (code && flp->fl_type != F_UNLCK) {
637 struct AFS_FLOCK flock2;
639 flock2.l_type = F_UNLCK;
641 afs_lockctl(vcp, &flock2, F_SETLK, credp);
645 /* If lockctl says there are no conflicting locks, then also check with the
646 * kernel, as lockctl knows nothing about byte range locks
648 if (code == 0 && cmd == F_GETLK && flock.l_type == F_UNLCK) {
649 afs_posix_test_lock(fp, flp);
650 /* If we found a lock in the kernel's structure, return it */
651 if (flp->fl_type != F_UNLCK) {
657 /* Convert flock back to Linux's file_lock */
658 flp->fl_type = flock.l_type;
659 flp->fl_pid = flock.l_pid;
660 flp->fl_start = flock.l_start;
661 if (flock.l_len == 0)
662 flp->fl_end = OFFSET_MAX; /* Lock to end of file */
664 flp->fl_end = flock.l_start + flock.l_len - 1;
670 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
672 afs_linux_flock(struct file *fp, int cmd, struct file_lock *flp) {
674 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
675 cred_t *credp = crref();
676 struct AFS_FLOCK flock;
677 /* Convert to a lock format afs_lockctl understands. */
678 memset(&flock, 0, sizeof(flock));
679 flock.l_type = flp->fl_type;
680 flock.l_pid = flp->fl_pid;
685 /* Safe because there are no large files, yet */
686 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
687 if (cmd == F_GETLK64)
689 else if (cmd == F_SETLK64)
691 else if (cmd == F_SETLKW64)
693 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
696 code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
699 if ((code == 0 || flp->fl_type == F_UNLCK) &&
700 (cmd == F_SETLK || cmd == F_SETLKW)) {
701 flp->fl_flags &=~ FL_SLEEP;
702 code = flock_lock_file_wait(fp, flp);
703 if (code && flp->fl_type != F_UNLCK) {
704 struct AFS_FLOCK flock2;
706 flock2.l_type = F_UNLCK;
708 afs_lockctl(vcp, &flock2, F_SETLK, credp);
712 /* Convert flock back to Linux's file_lock */
713 flp->fl_type = flock.l_type;
714 flp->fl_pid = flock.l_pid;
722 * essentially the same as afs_fsync() but we need to get the return
723 * code for the sys_close() here, not afs_linux_release(), so call
724 * afs_StoreAllSegments() with AFS_LASTSTORE
727 #if defined(FOP_FLUSH_TAKES_FL_OWNER_T)
728 afs_linux_flush(struct file *fp, fl_owner_t id)
730 afs_linux_flush(struct file *fp)
733 struct vrequest *treq = NULL;
741 if ((fp->f_flags & O_ACCMODE) == O_RDONLY) { /* readers dont flush */
749 vcp = VTOAFS(FILE_INODE(fp));
751 code = afs_CreateReq(&treq, credp);
754 /* If caching is bypassed for this file, or globally, just return 0 */
755 if (cache_bypass_strategy == ALWAYS_BYPASS_CACHE)
758 ObtainReadLock(&vcp->lock);
759 if (vcp->cachingStates & FCSBypass)
761 ReleaseReadLock(&vcp->lock);
764 /* future proof: don't rely on 0 return from afs_InitReq */
769 ObtainSharedLock(&vcp->lock, 535);
770 if ((vcp->execsOrWriters > 0) && (file_count(fp) == 1)) {
771 UpgradeSToWLock(&vcp->lock, 536);
772 if (!AFS_IS_DISCONNECTED) {
773 code = afs_StoreAllSegments(vcp,
775 AFS_SYNC | AFS_LASTSTORE);
777 afs_DisconAddDirty(vcp, VDisconWriteOsiFlush, 1);
779 ConvertWToSLock(&vcp->lock);
781 code = afs_CheckCode(code, treq, 54);
782 ReleaseSharedLock(&vcp->lock);
785 afs_DestroyReq(treq);
790 return afs_convert_code(code);
793 struct file_operations afs_dir_fops = {
794 .read = generic_read_dir,
795 #if defined(STRUCT_FILE_OPERATIONS_HAS_ITERATE)
796 .iterate = afs_linux_readdir,
798 .readdir = afs_linux_readdir,
800 #ifdef HAVE_UNLOCKED_IOCTL
801 .unlocked_ioctl = afs_unlocked_xioctl,
805 #ifdef HAVE_COMPAT_IOCTL
806 .compat_ioctl = afs_unlocked_xioctl,
808 .open = afs_linux_open,
809 .release = afs_linux_release,
810 .llseek = default_llseek,
811 #ifdef HAVE_LINUX_NOOP_FSYNC
814 .fsync = simple_sync_file,
818 struct file_operations afs_file_fops = {
819 #ifdef STRUCT_FILE_OPERATIONS_HAS_READ_ITER
820 .read_iter = afs_linux_read_iter,
821 .write_iter = afs_linux_write_iter,
822 .read = new_sync_read,
823 .write = new_sync_write,
824 #elif defined(HAVE_LINUX_GENERIC_FILE_AIO_READ)
825 .aio_read = afs_linux_aio_read,
826 .aio_write = afs_linux_aio_write,
827 .read = do_sync_read,
828 .write = do_sync_write,
830 .read = afs_linux_read,
831 .write = afs_linux_write,
833 #ifdef HAVE_UNLOCKED_IOCTL
834 .unlocked_ioctl = afs_unlocked_xioctl,
838 #ifdef HAVE_COMPAT_IOCTL
839 .compat_ioctl = afs_unlocked_xioctl,
841 .mmap = afs_linux_mmap,
842 .open = afs_linux_open,
843 .flush = afs_linux_flush,
844 #if defined(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
845 .sendfile = generic_file_sendfile,
847 #if defined(STRUCT_FILE_OPERATIONS_HAS_SPLICE)
848 # if defined(HAVE_LINUX_ITER_FILE_SPLICE_WRITE)
849 .splice_write = iter_file_splice_write,
851 .splice_write = generic_file_splice_write,
853 .splice_read = generic_file_splice_read,
855 .release = afs_linux_release,
856 .fsync = afs_linux_fsync,
857 .lock = afs_linux_lock,
858 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
859 .flock = afs_linux_flock,
861 .llseek = default_llseek,
864 static struct dentry *
865 canonical_dentry(struct inode *ip)
867 struct vcache *vcp = VTOAFS(ip);
868 struct dentry *first = NULL, *ret = NULL, *cur;
869 #if defined(D_ALIAS_IS_HLIST) && !defined(HLIST_ITERATOR_NO_NODE)
870 struct hlist_node *p;
874 * if vcp->target_link is set, and can be found in ip->i_dentry, use that.
875 * otherwise, use the first dentry in ip->i_dentry.
876 * if ip->i_dentry is empty, use the 'dentry' argument we were given.
878 /* note that vcp->target_link specifies which dentry to use, but we have
879 * no reference held on that dentry. so, we cannot use or dereference
880 * vcp->target_link itself, since it may have been freed. instead, we only
881 * use it to compare to pointers in the ip->i_dentry list. */
885 # ifdef HAVE_DCACHE_LOCK
886 spin_lock(&dcache_lock);
888 spin_lock(&ip->i_lock);
891 #if defined(D_ALIAS_IS_HLIST)
892 # if defined(HLIST_ITERATOR_NO_NODE)
893 hlist_for_each_entry(cur, &ip->i_dentry, d_alias) {
895 hlist_for_each_entry(cur, p, &ip->i_dentry, d_alias) {
898 list_for_each_entry_reverse(cur, &ip->i_dentry, d_alias) {
901 if (!vcp->target_link || cur == vcp->target_link) {
914 vcp->target_link = ret;
916 # ifdef HAVE_DCACHE_LOCK
920 spin_unlock(&dcache_lock);
925 spin_unlock(&ip->i_lock);
931 /**********************************************************************
932 * AFS Linux dentry operations
933 **********************************************************************/
935 /* afs_linux_revalidate
936 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
939 afs_linux_revalidate(struct dentry *dp)
941 struct vattr *vattr = NULL;
942 struct vcache *vcp = VTOAFS(dp->d_inode);
946 if (afs_shuttingdown)
951 code = afs_CreateAttr(&vattr);
956 /* This avoids the crref when we don't have to do it. Watch for
957 * changes in afs_getattr that don't get replicated here!
959 if (vcp->f.states & CStatd &&
960 (!afs_fakestat_enable || vcp->mvstat != 1) &&
962 (vType(vcp) == VDIR || vType(vcp) == VLNK)) {
963 code = afs_CopyOutAttrs(vcp, vattr);
966 code = afs_getattr(vcp, vattr, credp);
971 afs_fill_inode(AFSTOV(vcp), vattr);
973 afs_DestroyAttr(vattr);
978 return afs_convert_code(code);
982 * Set iattr data into vattr. Assume vattr cleared before call.
985 iattr2vattr(struct vattr *vattrp, struct iattr *iattrp)
987 vattrp->va_mask = iattrp->ia_valid;
988 if (iattrp->ia_valid & ATTR_MODE)
989 vattrp->va_mode = iattrp->ia_mode;
990 if (iattrp->ia_valid & ATTR_UID)
991 vattrp->va_uid = afs_from_kuid(iattrp->ia_uid);
992 if (iattrp->ia_valid & ATTR_GID)
993 vattrp->va_gid = afs_from_kgid(iattrp->ia_gid);
994 if (iattrp->ia_valid & ATTR_SIZE)
995 vattrp->va_size = iattrp->ia_size;
996 if (iattrp->ia_valid & ATTR_ATIME) {
997 vattrp->va_atime.tv_sec = iattrp->ia_atime.tv_sec;
998 vattrp->va_atime.tv_usec = 0;
1000 if (iattrp->ia_valid & ATTR_MTIME) {
1001 vattrp->va_mtime.tv_sec = iattrp->ia_mtime.tv_sec;
1002 vattrp->va_mtime.tv_usec = 0;
1004 if (iattrp->ia_valid & ATTR_CTIME) {
1005 vattrp->va_ctime.tv_sec = iattrp->ia_ctime.tv_sec;
1006 vattrp->va_ctime.tv_usec = 0;
1011 * Rewrite the inode cache from the attr. Assumes all vattr fields are valid.
1014 vattr2inode(struct inode *ip, struct vattr *vp)
1016 ip->i_ino = vp->va_nodeid;
1017 #ifdef HAVE_LINUX_SET_NLINK
1018 set_nlink(ip, vp->va_nlink);
1020 ip->i_nlink = vp->va_nlink;
1022 ip->i_blocks = vp->va_blocks;
1023 #ifdef STRUCT_INODE_HAS_I_BLKBITS
1024 ip->i_blkbits = AFS_BLKBITS;
1026 #ifdef STRUCT_INODE_HAS_I_BLKSIZE
1027 ip->i_blksize = vp->va_blocksize;
1029 ip->i_rdev = vp->va_rdev;
1030 ip->i_mode = vp->va_mode;
1031 ip->i_uid = afs_make_kuid(vp->va_uid);
1032 ip->i_gid = afs_make_kgid(vp->va_gid);
1033 i_size_write(ip, vp->va_size);
1034 ip->i_atime.tv_sec = vp->va_atime.tv_sec;
1035 ip->i_atime.tv_nsec = 0;
1036 ip->i_mtime.tv_sec = vp->va_mtime.tv_sec;
1037 /* Set the mtime nanoseconds to the sysname generation number.
1038 * This convinces NFS clients that all directories have changed
1039 * any time the sysname list changes.
1041 ip->i_mtime.tv_nsec = afs_sysnamegen;
1042 ip->i_ctime.tv_sec = vp->va_ctime.tv_sec;
1043 ip->i_ctime.tv_nsec = 0;
1046 /* afs_notify_change
1047 * Linux version of setattr call. What to change is in the iattr struct.
1048 * We need to set bits in both the Linux inode as well as the vcache.
1051 afs_notify_change(struct dentry *dp, struct iattr *iattrp)
1053 struct vattr *vattr = NULL;
1054 cred_t *credp = crref();
1055 struct inode *ip = dp->d_inode;
1059 code = afs_CreateAttr(&vattr);
1064 iattr2vattr(vattr, iattrp); /* Convert for AFS vnodeops call. */
1066 code = afs_setattr(VTOAFS(ip), vattr, credp);
1068 afs_getattr(VTOAFS(ip), vattr, credp);
1069 vattr2inode(ip, vattr);
1071 afs_DestroyAttr(vattr);
1076 return afs_convert_code(code);
1080 afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1082 int err = afs_linux_revalidate(dentry);
1084 generic_fillattr(dentry->d_inode, stat);
1090 parent_vcache_dv(struct inode *inode, cred_t *credp)
1093 struct vcache *pvcp;
1096 * If parent is a mount point and we are using fakestat, we may need
1097 * to look at the fake vcache entry instead of what the vfs is giving
1098 * us. The fake entry is the one with the useful DataVersion.
1100 pvcp = VTOAFS(inode);
1101 if (pvcp->mvstat == 1 && afs_fakestat_enable) {
1102 struct vrequest treq;
1103 struct afs_fakestat_state fakestate;
1109 afs_InitReq(&treq, credp);
1110 afs_InitFakeStat(&fakestate);
1111 afs_TryEvalFakeStat(&pvcp, &fakestate, &treq);
1114 afs_PutFakeStat(&fakestate);
1116 return hgetlo(pvcp->f.m.DataVersion);
1119 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
1120 * In kernels 2.2.10 and above, we are passed an additional flags var which
1121 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
1122 * we are advised to follow the entry if it is a link or to make sure that
1123 * it is a directory. But since the kernel itself checks these possibilities
1124 * later on, we shouldn't have to do it until later. Perhaps in the future..
1126 * The code here assumes that on entry the global lock is not held
1129 #if defined(DOP_REVALIDATE_TAKES_UNSIGNED)
1130 afs_linux_dentry_revalidate(struct dentry *dp, unsigned int flags)
1131 #elif defined(DOP_REVALIDATE_TAKES_NAMEIDATA)
1132 afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
1134 afs_linux_dentry_revalidate(struct dentry *dp, int flags)
1137 cred_t *credp = NULL;
1138 struct vcache *vcp, *pvcp, *tvc = NULL;
1139 struct dentry *parent;
1141 struct afs_fakestat_state fakestate;
1144 afs_uint32 parent_dv;
1147 /* We don't support RCU path walking */
1148 # if defined(DOP_REVALIDATE_TAKES_UNSIGNED)
1149 if (flags & LOOKUP_RCU)
1151 if (nd->flags & LOOKUP_RCU)
1156 afs_InitFakeStat(&fakestate);
1159 vcp = VTOAFS(dp->d_inode);
1161 if (vcp == afs_globalVp)
1164 parent = dget_parent(dp);
1165 pvcp = VTOAFS(parent->d_inode);
1167 if ((vcp->mvstat == 1) || (vcp->mvstat == 2) ||
1168 (pvcp->mvstat == 1 && afs_fakestat_enable)) { /* need to lock */
1175 if (vcp->mvstat == 1) { /* mount point */
1176 if (vcp->mvid && (vcp->f.states & CMValid)) {
1177 int tryEvalOnly = 0;
1179 struct vrequest *treq = NULL;
1181 code = afs_CreateReq(&treq, credp);
1186 if ((strcmp(dp->d_name.name, ".directory") == 0)) {
1190 code = afs_TryEvalFakeStat(&vcp, &fakestate, treq);
1192 code = afs_EvalFakeStat(&vcp, &fakestate, treq);
1193 afs_DestroyReq(treq);
1194 if ((tryEvalOnly && vcp->mvstat == 1) || code) {
1195 /* a mount point, not yet replaced by its directory */
1200 } else if (vcp->mvstat == 2 && *dp->d_name.name != '/') {
1201 osi_Assert(vcp->mvid != NULL);
1206 /* If the last looker changes, we should make sure the current
1207 * looker still has permission to examine this file. This would
1208 * always require a crref() which would be "slow".
1210 if (vcp->last_looker != treq.uid) {
1211 if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
1216 vcp->last_looker = treq.uid;
1220 parent_dv = parent_vcache_dv(parent->d_inode, credp);
1222 /* If the parent's DataVersion has changed or the vnode
1223 * is longer valid, we need to do a full lookup. VerifyVCache
1224 * isn't enough since the vnode may have been renamed.
1227 if ((!locked) && (parent_dv > dp->d_time || !(vcp->f.states & CStatd)) ) {
1233 if (locked && (parent_dv > dp->d_time || !(vcp->f.states & CStatd))) {
1234 struct vattr *vattr = NULL;
1238 code = afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
1241 /* We couldn't perform the lookup, so we're not okay. */
1244 } else if (tvc == vcp) {
1245 /* We got back the same vcache, so we're good. */
1248 } else if (tvc == VTOAFS(dp->d_inode)) {
1249 /* We got back the same vcache, so we're good. This is
1250 * different from the above case, because sometimes 'vcp' is
1251 * not the same as the vcache for dp->d_inode, if 'vcp' was a
1252 * mtpt and we evaluated it to a root dir. In rare cases,
1253 * afs_lookup might not evalute the mtpt when we do, or vice
1254 * versa, so the previous case will not succeed. But this is
1255 * still 'correct', so make sure not to mark the dentry as
1256 * invalid; it still points to the same thing! */
1260 /* We got back a different file, so we're definitely not
1267 /* Force unhash; the name doesn't point to this file
1270 if (code && code != ENOENT) {
1271 /* ...except if we couldn't perform the actual lookup,
1272 * we don't know if the name points to this file or not. */
1278 code = afs_CreateAttr(&vattr);
1284 if (afs_getattr(vcp, vattr, credp)) {
1286 afs_DestroyAttr(vattr);
1290 vattr2inode(AFSTOV(vcp), vattr);
1291 dp->d_time = parent_dv;
1293 afs_DestroyAttr(vattr);
1296 /* should we always update the attributes at this point? */
1297 /* unlikely--the vcache entry hasn't changed */
1302 /* If this code is ever enabled, we should use dget_parent to handle
1303 * getting the parent, and dput() to dispose of it. See above for an
1305 pvcp = VTOAFS(dp->d_parent->d_inode);
1306 if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time)
1310 /* No change in parent's DataVersion so this negative
1311 * lookup is still valid. BUT, if a server is down a
1312 * negative lookup can result so there should be a
1313 * liftime as well. For now, always expire.
1326 afs_PutFakeStat(&fakestate); /* from here on vcp may be no longer valid */
1328 /* we hold the global lock if we evaluated a mount point */
1336 * If we had a negative lookup for the name we want to forcibly
1337 * unhash the dentry.
1338 * Otherwise use d_invalidate which will not unhash it if still in use.
1341 shrink_dcache_parent(dp);
1350 if (have_submounts(dp))
1358 afs_dentry_iput(struct dentry *dp, struct inode *ip)
1360 struct vcache *vcp = VTOAFS(ip);
1363 if (!AFS_IS_DISCONNECTED || (vcp->f.states & CUnlinked)) {
1364 (void) afs_InactiveVCache(vcp, NULL);
1367 afs_linux_clear_nfsfs_renamed(dp);
1373 #if defined(DOP_D_DELETE_TAKES_CONST)
1374 afs_dentry_delete(const struct dentry *dp)
1376 afs_dentry_delete(struct dentry *dp)
1379 if (dp->d_inode && (VTOAFS(dp->d_inode)->f.states & CUnlinked))
1380 return 1; /* bad inode? */
1385 #ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
1386 static struct vfsmount *
1387 afs_dentry_automount(afs_linux_path_t *path)
1389 struct dentry *target;
1391 /* avoid symlink resolution limits when resolving; we cannot contribute to
1392 * an infinite symlink loop */
1393 current->total_link_count--;
1395 target = canonical_dentry(path->dentry->d_inode);
1397 if (target == path->dentry) {
1404 path->dentry = target;
1407 spin_lock(&path->dentry->d_lock);
1408 path->dentry->d_flags &= ~DCACHE_NEED_AUTOMOUNT;
1409 spin_unlock(&path->dentry->d_lock);
1414 #endif /* STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
1416 struct dentry_operations afs_dentry_operations = {
1417 .d_revalidate = afs_linux_dentry_revalidate,
1418 .d_delete = afs_dentry_delete,
1419 .d_iput = afs_dentry_iput,
1420 #ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
1421 .d_automount = afs_dentry_automount,
1422 #endif /* STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
1425 /**********************************************************************
1426 * AFS Linux inode operations
1427 **********************************************************************/
1431 * Merely need to set enough of vattr to get us through the create. Note
1432 * that the higher level code (open_namei) will take care of any tuncation
1433 * explicitly. Exclusive open is also taken care of in open_namei.
1435 * name is in kernel space at this point.
1438 #if defined(IOP_CREATE_TAKES_BOOL)
1439 afs_linux_create(struct inode *dip, struct dentry *dp, umode_t mode,
1441 #elif defined(IOP_CREATE_TAKES_UMODE_T)
1442 afs_linux_create(struct inode *dip, struct dentry *dp, umode_t mode,
1443 struct nameidata *nd)
1444 #elif defined(IOP_CREATE_TAKES_NAMEIDATA)
1445 afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
1446 struct nameidata *nd)
1448 afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
1451 struct vattr *vattr = NULL;
1452 cred_t *credp = crref();
1453 const char *name = dp->d_name.name;
1459 code = afs_CreateAttr(&vattr);
1463 vattr->va_mode = mode;
1464 vattr->va_type = mode & S_IFMT;
1466 code = afs_create(VTOAFS(dip), (char *)name, vattr, NONEXCL, mode,
1470 struct inode *ip = AFSTOV(vcp);
1472 afs_getattr(vcp, vattr, credp);
1473 afs_fill_inode(ip, vattr);
1474 insert_inode_hash(ip);
1475 #if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
1476 dp->d_op = &afs_dentry_operations;
1478 dp->d_time = parent_vcache_dv(dip, credp);
1479 d_instantiate(dp, ip);
1482 afs_DestroyAttr(vattr);
1488 return afs_convert_code(code);
1491 /* afs_linux_lookup */
1492 static struct dentry *
1493 #if defined(IOP_LOOKUP_TAKES_UNSIGNED)
1494 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1496 #elif defined(IOP_LOOKUP_TAKES_NAMEIDATA)
1497 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1498 struct nameidata *nd)
1500 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1503 cred_t *credp = crref();
1504 struct vcache *vcp = NULL;
1505 const char *comp = dp->d_name.name;
1506 struct inode *ip = NULL;
1507 struct dentry *newdp = NULL;
1511 code = afs_lookup(VTOAFS(dip), (char *)comp, &vcp, credp);
1514 struct vattr *vattr = NULL;
1515 struct vcache *parent_vc = VTOAFS(dip);
1517 if (parent_vc == vcp) {
1518 /* This is possible if the parent dir is a mountpoint to a volume,
1519 * and the dir entry we looked up is a mountpoint to the same
1520 * volume. Linux cannot cope with this, so return an error instead
1521 * of risking a deadlock or panic. */
1528 code = afs_CreateAttr(&vattr);
1536 afs_getattr(vcp, vattr, credp);
1537 afs_fill_inode(ip, vattr);
1538 if (hlist_unhashed(&ip->i_hash))
1539 insert_inode_hash(ip);
1541 afs_DestroyAttr(vattr);
1543 #if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
1544 dp->d_op = &afs_dentry_operations;
1546 dp->d_time = parent_vcache_dv(dip, credp);
1550 if (ip && S_ISDIR(ip->i_mode)) {
1551 d_prune_aliases(ip);
1553 #ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
1554 ip->i_flags |= S_AUTOMOUNT;
1558 * Take an extra reference so the inode doesn't go away if
1559 * d_splice_alias drops our reference on error.
1562 #ifdef HAVE_LINUX_IHOLD
1568 newdp = d_splice_alias(ip, dp);
1573 /* It's ok for the file to not be found. That's noted by the caller by
1574 * seeing that the dp->d_inode field is NULL.
1576 if (!code || code == ENOENT) {
1578 * d_splice_alias can return an error (EIO) if there is an existing
1579 * connected directory alias for this dentry.
1581 if (!IS_ERR(newdp)) {
1587 * Depending on the kernel version, d_splice_alias may or may
1588 * not drop the inode reference on error. If it didn't, do it
1591 #if defined(D_SPLICE_ALIAS_LEAK_ON_ERROR)
1599 return ERR_PTR(afs_convert_code(code));
1604 afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
1607 cred_t *credp = crref();
1608 const char *name = newdp->d_name.name;
1609 struct inode *oldip = olddp->d_inode;
1611 /* If afs_link returned the vnode, we could instantiate the
1612 * dentry. Since it's not, we drop this one and do a new lookup.
1617 code = afs_link(VTOAFS(oldip), VTOAFS(dip), (char *)name, credp);
1621 return afs_convert_code(code);
1624 /* We have to have a Linux specific sillyrename function, because we
1625 * also have to keep the dcache up to date when we're doing a silly
1626 * rename - so we don't want the generic vnodeops doing this behind our
1631 afs_linux_sillyrename(struct inode *dir, struct dentry *dentry,
1634 struct vcache *tvc = VTOAFS(dentry->d_inode);
1635 struct dentry *__dp = NULL;
1636 char *__name = NULL;
1639 if (afs_linux_nfsfs_renamed(dentry))
1647 osi_FreeSmallSpace(__name);
1648 __name = afs_newname();
1651 __dp = lookup_one_len(__name, dentry->d_parent, strlen(__name));
1654 osi_FreeSmallSpace(__name);
1657 } while (__dp->d_inode != NULL);
1660 code = afs_rename(VTOAFS(dir), (char *)dentry->d_name.name,
1661 VTOAFS(dir), (char *)__dp->d_name.name,
1664 tvc->mvid = (void *) __name;
1667 crfree(tvc->uncred);
1669 tvc->uncred = credp;
1670 tvc->f.states |= CUnlinked;
1671 afs_linux_set_nfsfs_renamed(dentry);
1673 osi_FreeSmallSpace(__name);
1678 __dp->d_time = hgetlo(VTOAFS(dir)->f.m.DataVersion);
1679 d_move(dentry, __dp);
1688 afs_linux_unlink(struct inode *dip, struct dentry *dp)
1691 cred_t *credp = crref();
1692 const char *name = dp->d_name.name;
1693 struct vcache *tvc = VTOAFS(dp->d_inode);
1695 if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
1696 && !(tvc->f.states & CUnlinked)) {
1698 code = afs_linux_sillyrename(dip, dp, credp);
1701 code = afs_remove(VTOAFS(dip), (char *)name, credp);
1708 return afs_convert_code(code);
1713 afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
1716 cred_t *credp = crref();
1717 struct vattr *vattr = NULL;
1718 const char *name = dp->d_name.name;
1720 /* If afs_symlink returned the vnode, we could instantiate the
1721 * dentry. Since it's not, we drop this one and do a new lookup.
1726 code = afs_CreateAttr(&vattr);
1731 code = afs_symlink(VTOAFS(dip), (char *)name, vattr, (char *)target, NULL,
1733 afs_DestroyAttr(vattr);
1738 return afs_convert_code(code);
1742 #if defined(IOP_MKDIR_TAKES_UMODE_T)
1743 afs_linux_mkdir(struct inode *dip, struct dentry *dp, umode_t mode)
1745 afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1749 cred_t *credp = crref();
1750 struct vcache *tvcp = NULL;
1751 struct vattr *vattr = NULL;
1752 const char *name = dp->d_name.name;
1755 code = afs_CreateAttr(&vattr);
1760 vattr->va_mask = ATTR_MODE;
1761 vattr->va_mode = mode;
1763 code = afs_mkdir(VTOAFS(dip), (char *)name, vattr, &tvcp, credp);
1766 struct inode *ip = AFSTOV(tvcp);
1768 afs_getattr(tvcp, vattr, credp);
1769 afs_fill_inode(ip, vattr);
1771 #if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
1772 dp->d_op = &afs_dentry_operations;
1774 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1775 d_instantiate(dp, ip);
1777 afs_DestroyAttr(vattr);
1783 return afs_convert_code(code);
1787 afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1790 cred_t *credp = crref();
1791 const char *name = dp->d_name.name;
1793 /* locking kernel conflicts with glock? */
1796 code = afs_rmdir(VTOAFS(dip), (char *)name, credp);
1799 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1800 * that failed because a directory is not empty. So, we map
1801 * EEXIST to ENOTEMPTY on linux.
1803 if (code == EEXIST) {
1812 return afs_convert_code(code);
1817 afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1818 struct inode *newip, struct dentry *newdp)
1821 cred_t *credp = crref();
1822 const char *oldname = olddp->d_name.name;
1823 const char *newname = newdp->d_name.name;
1824 struct dentry *rehash = NULL;
1826 /* Prevent any new references during rename operation. */
1828 if (!d_unhashed(newdp)) {
1833 afs_maybe_shrink_dcache(olddp);
1836 code = afs_rename(VTOAFS(oldip), (char *)oldname, VTOAFS(newip), (char *)newname, credp);
1840 olddp->d_time = 0; /* force to revalidate */
1846 return afs_convert_code(code);
1850 /* afs_linux_ireadlink
1851 * Internal readlink which can return link contents to user or kernel space.
1852 * Note that the buffer is NOT supposed to be null-terminated.
1855 afs_linux_ireadlink(struct inode *ip, char *target, int maxlen, uio_seg_t seg)
1858 cred_t *credp = crref();
1862 memset(&tuio, 0, sizeof(tuio));
1863 memset(&iov, 0, sizeof(iov));
1865 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1866 code = afs_readlink(VTOAFS(ip), &tuio, credp);
1870 return maxlen - tuio.uio_resid;
1872 return afs_convert_code(code);
1875 #if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1876 /* afs_linux_readlink
1877 * Fill target (which is in user space) with contents of symlink.
1880 afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1883 struct inode *ip = dp->d_inode;
1886 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1892 /* afs_linux_follow_link
1893 * a file system dependent link following routine.
1895 static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
1900 name = kmalloc(PATH_MAX, GFP_NOFS);
1906 code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
1914 nd_set_link(nd, name);
1919 afs_linux_put_link(struct dentry *dentry, struct nameidata *nd)
1921 char *name = nd_get_link(nd);
1923 if (name && !IS_ERR(name))
1927 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1929 /* Populate a page by filling it from the cache file pointed at by cachefp
1930 * (which contains indicated chunk)
1931 * If task is NULL, the page copy occurs syncronously, and the routine
1932 * returns with page still locked. If task is non-NULL, then page copies
1933 * may occur in the background, and the page will be unlocked when it is
1937 afs_linux_read_cache(struct file *cachefp, struct page *page,
1938 int chunk, struct pagevec *lrupv,
1939 struct afs_pagecopy_task *task) {
1940 loff_t offset = page_offset(page);
1941 struct inode *cacheinode = cachefp->f_dentry->d_inode;
1942 struct page *newpage, *cachepage;
1943 struct address_space *cachemapping;
1947 cachemapping = cacheinode->i_mapping;
1951 /* If we're trying to read a page that's past the end of the disk
1952 * cache file, then just return a zeroed page */
1953 if (AFS_CHUNKOFFSET(offset) >= i_size_read(cacheinode)) {
1954 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
1955 SetPageUptodate(page);
1961 /* From our offset, we now need to work out which page in the disk
1962 * file it corresponds to. This will be fun ... */
1963 pageindex = (offset - AFS_CHUNKTOBASE(chunk)) >> PAGE_CACHE_SHIFT;
1965 while (cachepage == NULL) {
1966 cachepage = find_get_page(cachemapping, pageindex);
1969 newpage = page_cache_alloc_cold(cachemapping);
1975 code = add_to_page_cache(newpage, cachemapping,
1976 pageindex, GFP_KERNEL);
1978 cachepage = newpage;
1981 page_cache_get(cachepage);
1982 if (!pagevec_add(lrupv, cachepage))
1983 __pagevec_lru_add_file(lrupv);
1986 page_cache_release(newpage);
1988 if (code != -EEXIST)
1992 lock_page(cachepage);
1996 if (!PageUptodate(cachepage)) {
1997 ClearPageError(cachepage);
1998 code = cachemapping->a_ops->readpage(NULL, cachepage);
1999 if (!code && !task) {
2000 wait_on_page_locked(cachepage);
2003 unlock_page(cachepage);
2007 if (PageUptodate(cachepage)) {
2008 copy_highpage(page, cachepage);
2009 flush_dcache_page(page);
2010 SetPageUptodate(page);
2015 afs_pagecopy_queue_page(task, cachepage, page);
2027 page_cache_release(cachepage);
2033 afs_linux_readpage_fastpath(struct file *fp, struct page *pp, int *codep)
2035 loff_t offset = page_offset(pp);
2036 struct inode *ip = FILE_INODE(fp);
2037 struct vcache *avc = VTOAFS(ip);
2039 struct file *cacheFp = NULL;
2042 struct pagevec lrupv;
2044 /* Not a UFS cache, don't do anything */
2045 if (cacheDiskType != AFS_FCACHE_TYPE_UFS)
2048 /* No readpage (ex: tmpfs) , skip */
2049 if (cachefs_noreadpage)
2052 /* Can't do anything if the vcache isn't statd , or if the read
2053 * crosses a chunk boundary.
2055 if (!(avc->f.states & CStatd) ||
2056 AFS_CHUNK(offset) != AFS_CHUNK(offset + PAGE_SIZE)) {
2060 ObtainWriteLock(&avc->lock, 911);
2062 /* XXX - See if hinting actually makes things faster !!! */
2064 /* See if we have a suitable entry already cached */
2068 /* We need to lock xdcache, then dcache, to handle situations where
2069 * the hint is on the free list. However, we can't safely do this
2070 * according to the locking hierarchy. So, use a non blocking lock.
2072 ObtainReadLock(&afs_xdcache);
2073 dcLocked = ( 0 == NBObtainReadLock(&tdc->lock));
2075 if (dcLocked && (tdc->index != NULLIDX)
2076 && !FidCmp(&tdc->f.fid, &avc->f.fid)
2077 && tdc->f.chunk == AFS_CHUNK(offset)
2078 && !(afs_indexFlags[tdc->index] & (IFFree | IFDiscarded))) {
2079 /* Bonus - the hint was correct */
2082 /* Only destroy the hint if its actually invalid, not if there's
2083 * just been a locking failure */
2085 ReleaseReadLock(&tdc->lock);
2092 ReleaseReadLock(&afs_xdcache);
2095 /* No hint, or hint is no longer valid - see if we can get something
2096 * directly from the dcache
2099 tdc = afs_FindDCache(avc, offset);
2102 ReleaseWriteLock(&avc->lock);
2107 ObtainReadLock(&tdc->lock);
2109 /* Is the dcache we've been given currently up to date */
2110 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
2111 (tdc->dflags & DFFetching))
2114 /* Update our hint for future abuse */
2117 /* Okay, so we've now got a cache file that is up to date */
2119 /* XXX - I suspect we should be locking the inodes before we use them! */
2121 cacheFp = afs_linux_raw_open(&tdc->f.inode);
2122 if (!cacheFp->f_dentry->d_inode->i_mapping->a_ops->readpage) {
2123 cachefs_noreadpage = 1;
2127 pagevec_init(&lrupv, 0);
2129 code = afs_linux_read_cache(cacheFp, pp, tdc->f.chunk, &lrupv, NULL);
2131 if (pagevec_count(&lrupv))
2132 __pagevec_lru_add_file(&lrupv);
2134 filp_close(cacheFp, NULL);
2137 ReleaseReadLock(&tdc->lock);
2138 ReleaseWriteLock(&avc->lock);
2145 ReleaseWriteLock(&avc->lock);
2146 ReleaseReadLock(&tdc->lock);
2151 /* afs_linux_readpage
2153 * This function is split into two, because prepare_write/begin_write
2154 * require a readpage call which doesn't unlock the resulting page upon
2158 afs_linux_fillpage(struct file *fp, struct page *pp)
2163 struct iovec *iovecp;
2164 struct inode *ip = FILE_INODE(fp);
2165 afs_int32 cnt = page_count(pp);
2166 struct vcache *avc = VTOAFS(ip);
2167 afs_offs_t offset = page_offset(pp);
2171 if (afs_linux_readpage_fastpath(fp, pp, &code)) {
2181 auio = kmalloc(sizeof(struct uio), GFP_NOFS);
2182 iovecp = kmalloc(sizeof(struct iovec), GFP_NOFS);
2184 setup_uio(auio, iovecp, (char *)address, offset, PAGE_SIZE, UIO_READ,
2189 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
2190 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
2191 99999); /* not a possible code value */
2193 code = afs_rdwr(avc, auio, UIO_READ, 0, credp);
2195 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
2196 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
2198 AFS_DISCON_UNLOCK();
2201 /* XXX valid for no-cache also? Check last bits of files... :)
2202 * Cognate code goes in afs_NoCacheFetchProc. */
2203 if (auio->uio_resid) /* zero remainder of page */
2204 memset((void *)(address + (PAGE_SIZE - auio->uio_resid)), 0,
2207 flush_dcache_page(pp);
2208 SetPageUptodate(pp);
2217 return afs_convert_code(code);
2221 afs_linux_prefetch(struct file *fp, struct page *pp)
2224 struct vcache *avc = VTOAFS(FILE_INODE(fp));
2225 afs_offs_t offset = page_offset(pp);
2227 if (AFS_CHUNKOFFSET(offset) == 0) {
2229 struct vrequest *treq = NULL;
2234 code = afs_CreateReq(&treq, credp);
2235 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
2236 tdc = afs_FindDCache(avc, offset);
2238 if (!(tdc->mflags & DFNextStarted))
2239 afs_PrefetchChunk(avc, tdc, credp, treq);
2242 ReleaseWriteLock(&avc->lock);
2244 afs_DestroyReq(treq);
2248 return afs_convert_code(code);
2253 afs_linux_bypass_readpages(struct file *fp, struct address_space *mapping,
2254 struct list_head *page_list, unsigned num_pages)
2259 struct iovec* iovecp;
2260 struct nocache_read_request *ancr;
2262 struct pagevec lrupv;
2266 struct inode *ip = FILE_INODE(fp);
2267 struct vcache *avc = VTOAFS(ip);
2268 afs_int32 base_index = 0;
2269 afs_int32 page_count = 0;
2272 /* background thread must free: iovecp, auio, ancr */
2273 iovecp = osi_Alloc(num_pages * sizeof(struct iovec));
2275 auio = osi_Alloc(sizeof(struct uio));
2276 auio->uio_iov = iovecp;
2277 auio->uio_iovcnt = num_pages;
2278 auio->uio_flag = UIO_READ;
2279 auio->uio_seg = AFS_UIOSYS;
2280 auio->uio_resid = num_pages * PAGE_SIZE;
2282 ancr = osi_Alloc(sizeof(struct nocache_read_request));
2284 ancr->offset = auio->uio_offset;
2285 ancr->length = auio->uio_resid;
2287 pagevec_init(&lrupv, 0);
2289 for(page_ix = 0; page_ix < num_pages; ++page_ix) {
2291 if(list_empty(page_list))
2294 pp = list_entry(page_list->prev, struct page, lru);
2295 /* If we allocate a page and don't remove it from page_list,
2296 * the page cache gets upset. */
2298 isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
2299 if(pp->index > isize) {
2306 offset = page_offset(pp);
2307 ancr->offset = auio->uio_offset = offset;
2308 base_index = pp->index;
2310 iovecp[page_ix].iov_len = PAGE_SIZE;
2311 code = add_to_page_cache(pp, mapping, pp->index, GFP_KERNEL);
2312 if(base_index != pp->index) {
2315 page_cache_release(pp);
2316 iovecp[page_ix].iov_base = (void *) 0;
2318 ancr->length -= PAGE_SIZE;
2325 page_cache_release(pp);
2326 iovecp[page_ix].iov_base = (void *) 0;
2329 if(!PageLocked(pp)) {
2333 /* increment page refcount--our original design assumed
2334 * that locking it would effectively pin it; protect
2335 * ourselves from the possiblity that this assumption is
2336 * is faulty, at low cost (provided we do not fail to
2337 * do the corresponding decref on the other side) */
2340 /* save the page for background map */
2341 iovecp[page_ix].iov_base = (void*) pp;
2343 /* and put it on the LRU cache */
2344 if (!pagevec_add(&lrupv, pp))
2345 __pagevec_lru_add_file(&lrupv);
2349 /* If there were useful pages in the page list, make sure all pages
2350 * are in the LRU cache, then schedule the read */
2352 if (pagevec_count(&lrupv))
2353 __pagevec_lru_add_file(&lrupv);
2355 code = afs_ReadNoCache(avc, ancr, credp);
2358 /* If there is nothing for the background thread to handle,
2359 * it won't be freeing the things that we never gave it */
2360 osi_Free(iovecp, num_pages * sizeof(struct iovec));
2361 osi_Free(auio, sizeof(struct uio));
2362 osi_Free(ancr, sizeof(struct nocache_read_request));
2364 /* we do not flush, release, or unmap pages--that will be
2365 * done for us by the background thread as each page comes in
2366 * from the fileserver */
2367 return afs_convert_code(code);
2372 afs_linux_bypass_readpage(struct file *fp, struct page *pp)
2374 cred_t *credp = NULL;
2376 struct iovec *iovecp;
2377 struct nocache_read_request *ancr;
2381 * Special case: if page is at or past end of file, just zero it and set
2384 if (page_offset(pp) >= i_size_read(fp->f_mapping->host)) {
2385 zero_user_segment(pp, 0, PAGE_CACHE_SIZE);
2386 SetPageUptodate(pp);
2393 /* receiver frees */
2394 auio = osi_Alloc(sizeof(struct uio));
2395 iovecp = osi_Alloc(sizeof(struct iovec));
2397 /* address can be NULL, because we overwrite it with 'pp', below */
2398 setup_uio(auio, iovecp, NULL, page_offset(pp),
2399 PAGE_SIZE, UIO_READ, AFS_UIOSYS);
2401 /* save the page for background map */
2402 get_page(pp); /* see above */
2403 auio->uio_iov->iov_base = (void*) pp;
2404 /* the background thread will free this */
2405 ancr = osi_Alloc(sizeof(struct nocache_read_request));
2407 ancr->offset = page_offset(pp);
2408 ancr->length = PAGE_SIZE;
2411 code = afs_ReadNoCache(VTOAFS(FILE_INODE(fp)), ancr, credp);
2414 return afs_convert_code(code);
2418 afs_linux_can_bypass(struct inode *ip) {
2420 switch(cache_bypass_strategy) {
2421 case NEVER_BYPASS_CACHE:
2423 case ALWAYS_BYPASS_CACHE:
2425 case LARGE_FILES_BYPASS_CACHE:
2426 if (i_size_read(ip) > cache_bypass_threshold)
2433 /* Check if a file is permitted to bypass the cache by policy, and modify
2434 * the cache bypass state recorded for that file */
2437 afs_linux_bypass_check(struct inode *ip) {
2440 int bypass = afs_linux_can_bypass(ip);
2443 trydo_cache_transition(VTOAFS(ip), credp, bypass);
2451 afs_linux_readpage(struct file *fp, struct page *pp)
2455 if (afs_linux_bypass_check(FILE_INODE(fp))) {
2456 code = afs_linux_bypass_readpage(fp, pp);
2458 code = afs_linux_fillpage(fp, pp);
2460 code = afs_linux_prefetch(fp, pp);
2467 /* Readpages reads a number of pages for a particular file. We use
2468 * this to optimise the reading, by limiting the number of times upon which
2469 * we have to lookup, lock and open vcaches and dcaches
2473 afs_linux_readpages(struct file *fp, struct address_space *mapping,
2474 struct list_head *page_list, unsigned int num_pages)
2476 struct inode *inode = mapping->host;
2477 struct vcache *avc = VTOAFS(inode);
2479 struct file *cacheFp = NULL;
2481 unsigned int page_idx;
2483 struct pagevec lrupv;
2484 struct afs_pagecopy_task *task;
2486 if (afs_linux_bypass_check(inode))
2487 return afs_linux_bypass_readpages(fp, mapping, page_list, num_pages);
2489 if (cacheDiskType == AFS_FCACHE_TYPE_MEM)
2492 /* No readpage (ex: tmpfs) , skip */
2493 if (cachefs_noreadpage)
2497 if ((code = afs_linux_VerifyVCache(avc, NULL))) {
2502 ObtainWriteLock(&avc->lock, 912);
2505 task = afs_pagecopy_init_task();
2508 pagevec_init(&lrupv, 0);
2509 for (page_idx = 0; page_idx < num_pages; page_idx++) {
2510 struct page *page = list_entry(page_list->prev, struct page, lru);
2511 list_del(&page->lru);
2512 offset = page_offset(page);
2514 if (tdc && tdc->f.chunk != AFS_CHUNK(offset)) {
2516 ReleaseReadLock(&tdc->lock);
2521 filp_close(cacheFp, NULL);
2526 if ((tdc = afs_FindDCache(avc, offset))) {
2527 ObtainReadLock(&tdc->lock);
2528 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
2529 (tdc->dflags & DFFetching)) {
2530 ReleaseReadLock(&tdc->lock);
2537 cacheFp = afs_linux_raw_open(&tdc->f.inode);
2538 if (!cacheFp->f_dentry->d_inode->i_mapping->a_ops->readpage) {
2539 cachefs_noreadpage = 1;
2545 if (tdc && !add_to_page_cache(page, mapping, page->index,
2547 page_cache_get(page);
2548 if (!pagevec_add(&lrupv, page))
2549 __pagevec_lru_add_file(&lrupv);
2551 afs_linux_read_cache(cacheFp, page, tdc->f.chunk, &lrupv, task);
2553 page_cache_release(page);
2555 if (pagevec_count(&lrupv))
2556 __pagevec_lru_add_file(&lrupv);
2560 filp_close(cacheFp, NULL);
2562 afs_pagecopy_put_task(task);
2566 ReleaseReadLock(&tdc->lock);
2570 ReleaseWriteLock(&avc->lock);
2575 /* Prepare an AFS vcache for writeback. Should be called with the vcache
2578 afs_linux_prepare_writeback(struct vcache *avc) {
2579 if (avc->f.states & CPageWrite) {
2580 return AOP_WRITEPAGE_ACTIVATE;
2582 avc->f.states |= CPageWrite;
2587 afs_linux_dopartialwrite(struct vcache *avc, cred_t *credp) {
2588 struct vrequest *treq = NULL;
2591 if (!afs_CreateReq(&treq, credp)) {
2592 code = afs_DoPartialWrite(avc, treq);
2593 afs_DestroyReq(treq);
2596 return afs_convert_code(code);
2600 afs_linux_complete_writeback(struct vcache *avc) {
2601 avc->f.states &= ~CPageWrite;
2604 /* Writeback a given page syncronously. Called with no AFS locks held */
2606 afs_linux_page_writeback(struct inode *ip, struct page *pp,
2607 unsigned long offset, unsigned int count,
2610 struct vcache *vcp = VTOAFS(ip);
2618 memset(&tuio, 0, sizeof(tuio));
2619 memset(&iovec, 0, sizeof(iovec));
2621 buffer = kmap(pp) + offset;
2622 base = page_offset(pp) + offset;
2625 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2626 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2627 ICL_TYPE_INT32, 99999);
2629 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
2631 code = afs_write(vcp, &tuio, f_flags, credp, 0);
2633 i_size_write(ip, vcp->f.m.Length);
2634 ip->i_blocks = ((vcp->f.m.Length + 1023) >> 10) << 1;
2636 code = code ? afs_convert_code(code) : count - tuio.uio_resid;
2638 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2639 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2640 ICL_TYPE_INT32, code);
2649 afs_linux_writepage_sync(struct inode *ip, struct page *pp,
2650 unsigned long offset, unsigned int count)
2654 struct vcache *vcp = VTOAFS(ip);
2657 /* Catch recursive writeback. This occurs if the kernel decides
2658 * writeback is required whilst we are writing to the cache, or
2659 * flushing to the server. When we're running syncronously (as
2660 * opposed to from writepage) we can't actually do anything about
2661 * this case - as we can't return AOP_WRITEPAGE_ACTIVATE to write()
2664 ObtainWriteLock(&vcp->lock, 532);
2665 afs_linux_prepare_writeback(vcp);
2666 ReleaseWriteLock(&vcp->lock);
2670 code = afs_linux_page_writeback(ip, pp, offset, count, credp);
2673 ObtainWriteLock(&vcp->lock, 533);
2675 code1 = afs_linux_dopartialwrite(vcp, credp);
2676 afs_linux_complete_writeback(vcp);
2677 ReleaseWriteLock(&vcp->lock);
2688 #ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
2689 afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
2691 afs_linux_writepage(struct page *pp)
2694 struct address_space *mapping = pp->mapping;
2695 struct inode *inode;
2698 unsigned int to = PAGE_CACHE_SIZE;
2705 inode = mapping->host;
2706 vcp = VTOAFS(inode);
2707 isize = i_size_read(inode);
2709 /* Don't defeat an earlier truncate */
2710 if (page_offset(pp) > isize) {
2711 set_page_writeback(pp);
2717 ObtainWriteLock(&vcp->lock, 537);
2718 code = afs_linux_prepare_writeback(vcp);
2719 if (code == AOP_WRITEPAGE_ACTIVATE) {
2720 /* WRITEPAGE_ACTIVATE is the only return value that permits us
2721 * to return with the page still locked */
2722 ReleaseWriteLock(&vcp->lock);
2727 /* Grab the creds structure currently held in the vnode, and
2728 * get a reference to it, in case it goes away ... */
2734 ReleaseWriteLock(&vcp->lock);
2737 set_page_writeback(pp);
2739 SetPageUptodate(pp);
2741 /* We can unlock the page here, because it's protected by the
2742 * page_writeback flag. This should make us less vulnerable to
2743 * deadlocking in afs_write and afs_DoPartialWrite
2747 /* If this is the final page, then just write the number of bytes that
2748 * are actually in it */
2749 if ((isize - page_offset(pp)) < to )
2750 to = isize - page_offset(pp);
2752 code = afs_linux_page_writeback(inode, pp, 0, to, credp);
2755 ObtainWriteLock(&vcp->lock, 538);
2757 /* As much as we might like to ignore a file server error here,
2758 * and just try again when we close(), unfortunately StoreAllSegments
2759 * will invalidate our chunks if the server returns a permanent error,
2760 * so we need to at least try and get that error back to the user
2763 code1 = afs_linux_dopartialwrite(vcp, credp);
2765 afs_linux_complete_writeback(vcp);
2766 ReleaseWriteLock(&vcp->lock);
2771 end_page_writeback(pp);
2772 page_cache_release(pp);
2783 /* afs_linux_permission
2784 * Check access rights - returns error if can't check or permission denied.
2787 #if defined(IOP_PERMISSION_TAKES_FLAGS)
2788 afs_linux_permission(struct inode *ip, int mode, unsigned int flags)
2789 #elif defined(IOP_PERMISSION_TAKES_NAMEIDATA)
2790 afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
2792 afs_linux_permission(struct inode *ip, int mode)
2799 /* Check for RCU path walking */
2800 #if defined(IOP_PERMISSION_TAKES_FLAGS)
2801 if (flags & IPERM_FLAG_RCU)
2803 #elif defined(MAY_NOT_BLOCK)
2804 if (mode & MAY_NOT_BLOCK)
2810 if (mode & MAY_EXEC)
2812 if (mode & MAY_READ)
2814 if (mode & MAY_WRITE)
2816 code = afs_access(VTOAFS(ip), tmp, credp);
2820 return afs_convert_code(code);
2824 afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
2828 struct inode *inode = FILE_INODE(file);
2829 loff_t pagebase = page_offset(page);
2831 if (i_size_read(inode) < (pagebase + offset))
2832 i_size_write(inode, pagebase + offset);
2834 if (PageChecked(page)) {
2835 SetPageUptodate(page);
2836 ClearPageChecked(page);
2839 code = afs_linux_writepage_sync(inode, page, offset, to - offset);
2845 afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
2849 /* http://kerneltrap.org/node/4941 details the expected behaviour of
2850 * prepare_write. Essentially, if the page exists within the file,
2851 * and is not being fully written, then we should populate it.
2854 if (!PageUptodate(page)) {
2855 loff_t pagebase = page_offset(page);
2856 loff_t isize = i_size_read(page->mapping->host);
2858 /* Is the location we are writing to beyond the end of the file? */
2859 if (pagebase >= isize ||
2860 ((from == 0) && (pagebase + to) >= isize)) {
2861 zero_user_segments(page, 0, from, to, PAGE_CACHE_SIZE);
2862 SetPageChecked(page);
2863 /* Are we we writing a full page */
2864 } else if (from == 0 && to == PAGE_CACHE_SIZE) {
2865 SetPageChecked(page);
2866 /* Is the page readable, if it's wronly, we don't care, because we're
2867 * not actually going to read from it ... */
2868 } else if ((file->f_flags && O_ACCMODE) != O_WRONLY) {
2869 /* We don't care if fillpage fails, because if it does the page
2870 * won't be marked as up to date
2872 afs_linux_fillpage(file, page);
2878 #if defined(STRUCT_ADDRESS_SPACE_OPERATIONS_HAS_WRITE_BEGIN)
2880 afs_linux_write_end(struct file *file, struct address_space *mapping,
2881 loff_t pos, unsigned len, unsigned copied,
2882 struct page *page, void *fsdata)
2885 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
2887 code = afs_linux_commit_write(file, page, from, from + len);
2890 page_cache_release(page);
2895 afs_linux_write_begin(struct file *file, struct address_space *mapping,
2896 loff_t pos, unsigned len, unsigned flags,
2897 struct page **pagep, void **fsdata)
2900 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2901 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
2904 page = grab_cache_page_write_begin(mapping, index, flags);
2907 code = afs_linux_prepare_write(file, page, from, from + len);
2910 page_cache_release(page);
2917 #ifndef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
2919 afs_linux_dir_follow_link(struct dentry *dentry, struct nameidata *nd)
2921 struct dentry **dpp;
2922 struct dentry *target;
2924 if (current->total_link_count > 0) {
2925 /* avoid symlink resolution limits when resolving; we cannot contribute to
2926 * an infinite symlink loop */
2927 /* only do this for follow_link when total_link_count is positive to be
2928 * on the safe side; there is at least one code path in the Linux
2929 * kernel where it seems like it may be possible to get here without
2930 * total_link_count getting incremented. it is not clear on how that
2931 * path is actually reached, but guard against it just to be safe */
2932 current->total_link_count--;
2935 target = canonical_dentry(dentry->d_inode);
2937 # ifdef STRUCT_NAMEIDATA_HAS_PATH
2938 dpp = &nd->path.dentry;
2948 *dpp = dget(dentry);
2951 nd->last_type = LAST_BIND;
2955 #endif /* !STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT */
2958 static struct inode_operations afs_file_iops = {
2959 .permission = afs_linux_permission,
2960 .getattr = afs_linux_getattr,
2961 .setattr = afs_notify_change,
2964 static struct address_space_operations afs_file_aops = {
2965 .readpage = afs_linux_readpage,
2966 .readpages = afs_linux_readpages,
2967 .writepage = afs_linux_writepage,
2968 #if defined (STRUCT_ADDRESS_SPACE_OPERATIONS_HAS_WRITE_BEGIN)
2969 .write_begin = afs_linux_write_begin,
2970 .write_end = afs_linux_write_end,
2972 .commit_write = afs_linux_commit_write,
2973 .prepare_write = afs_linux_prepare_write,
2978 /* Separate ops vector for directories. Linux 2.2 tests type of inode
2979 * by what sort of operation is allowed.....
2982 static struct inode_operations afs_dir_iops = {
2983 .setattr = afs_notify_change,
2984 .create = afs_linux_create,
2985 .lookup = afs_linux_lookup,
2986 .link = afs_linux_link,
2987 .unlink = afs_linux_unlink,
2988 .symlink = afs_linux_symlink,
2989 .mkdir = afs_linux_mkdir,
2990 .rmdir = afs_linux_rmdir,
2991 .rename = afs_linux_rename,
2992 .getattr = afs_linux_getattr,
2993 .permission = afs_linux_permission,
2994 #ifndef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
2995 .follow_link = afs_linux_dir_follow_link,
2999 /* We really need a separate symlink set of ops, since do_follow_link()
3000 * determines if it _is_ a link by checking if the follow_link op is set.
3002 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
3004 afs_symlink_filler(struct file *file, struct page *page)
3006 struct inode *ip = (struct inode *)page->mapping->host;
3007 char *p = (char *)kmap(page);
3011 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
3016 p[code] = '\0'; /* null terminate? */
3018 SetPageUptodate(page);
3030 static struct address_space_operations afs_symlink_aops = {
3031 .readpage = afs_symlink_filler
3033 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
3035 static struct inode_operations afs_symlink_iops = {
3036 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
3037 .readlink = page_readlink,
3038 # if defined(HAVE_LINUX_PAGE_FOLLOW_LINK)
3039 .follow_link = page_follow_link,
3041 .follow_link = page_follow_link_light,
3042 .put_link = page_put_link,
3044 #else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
3045 .readlink = afs_linux_readlink,
3046 .follow_link = afs_linux_follow_link,
3047 .put_link = afs_linux_put_link,
3048 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
3049 .setattr = afs_notify_change,
3053 afs_fill_inode(struct inode *ip, struct vattr *vattr)
3057 vattr2inode(ip, vattr);
3059 ip->i_mapping->backing_dev_info = afs_backing_dev_info;
3060 /* Reset ops if symlink or directory. */
3061 if (S_ISREG(ip->i_mode)) {
3062 ip->i_op = &afs_file_iops;
3063 ip->i_fop = &afs_file_fops;
3064 ip->i_data.a_ops = &afs_file_aops;
3066 } else if (S_ISDIR(ip->i_mode)) {
3067 ip->i_op = &afs_dir_iops;
3068 ip->i_fop = &afs_dir_fops;
3070 } else if (S_ISLNK(ip->i_mode)) {
3071 ip->i_op = &afs_symlink_iops;
3072 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
3073 ip->i_data.a_ops = &afs_symlink_aops;
3074 ip->i_mapping = &ip->i_data;