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"
27 #include "afs/sysincludes.h"
28 #include "afsincludes.h"
29 #include "afs/afs_stats.h"
30 #include "afs/afs_osidnlc.h"
32 #ifdef HAVE_MM_INLINE_H
33 #include "h/mm_inline.h"
35 #include "h/pagemap.h"
36 #if defined(AFS_LINUX24_ENV)
37 #include "h/smp_lock.h"
39 #if defined(AFS_LINUX26_ENV)
40 #include "h/writeback.h"
44 #define pageoff(pp) pgoff2loff((pp)->index)
46 #define pageoff(pp) pp->offset
49 #if defined(AFS_LINUX26_ENV)
50 #define UnlockPage(pp) unlock_page(pp)
53 extern struct vcache *afs_globalVp;
54 extern afs_rwlock_t afs_xvcache;
56 #if defined(AFS_LINUX24_ENV)
57 extern struct inode_operations afs_file_iops;
58 extern struct address_space_operations afs_file_aops;
59 struct address_space_operations afs_symlink_aops;
61 extern struct inode_operations afs_dir_iops;
62 extern struct inode_operations afs_symlink_iops;
66 afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
69 struct vcache *vcp = ITOAFS(fp->f_dentry->d_inode);
70 cred_t *credp = crref();
74 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
75 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
78 /* get a validated vcache entry */
79 code = afs_InitReq(&treq, credp);
81 code = afs_VerifyVCache(vcp, &treq);
86 #ifdef AFS_64BIT_CLIENT
87 if (*offp + count > afs_vmMappingEnd) {
92 if (*offp < afs_vmMappingEnd) {
93 /* special case of a buffer crossing the VM mapping end */
94 afs_int32 tcount = afs_vmMappingEnd - *offp;
96 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
98 code = generic_file_read(fp, buf, tcount, offp);
100 if (code != tcount) {
105 setup_uio(&tuio, &iov, buf + xfered, (afs_offs_t) * offp, count,
106 UIO_READ, AFS_UIOSYS);
107 code = afs_read(vcp, &tuio, credp, 0, 0, 0);
108 xfered += count - tuio.uio_resid;
110 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER,
111 vcp, ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, -1,
112 ICL_TYPE_INT32, code);
114 *offp += count - tuio.uio_resid;
122 #endif /* AFS_64BIT_CLIENT */
123 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
125 code = generic_file_read(fp, buf, count, offp);
127 #ifdef AFS_64BIT_CLIENT
129 #endif /* AFS_64BIT_CLIENT */
132 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
133 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
142 /* Now we have integrated VM for writes as well as reads. generic_file_write
143 * also takes care of re-positioning the pointer if file is open in append
144 * mode. Call fake open/close to ensure we do writes of core dumps.
147 afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
151 struct vcache *vcp = ITOAFS(fp->f_dentry->d_inode);
152 struct vrequest treq;
153 cred_t *credp = crref();
158 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
159 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
160 (fp->f_flags & O_APPEND) ? 99998 : 99999);
163 /* get a validated vcache entry */
164 code = (ssize_t) afs_InitReq(&treq, credp);
166 code = (ssize_t) afs_VerifyVCache(vcp, &treq);
168 ObtainWriteLock(&vcp->lock, 529);
170 ReleaseWriteLock(&vcp->lock);
174 #ifdef AFS_64BIT_CLIENT
176 if (fp->f_flags & O_APPEND)
177 toffs += vcp->m.Length;
178 if (toffs + count > afs_vmMappingEnd) {
181 afs_size_t oldOffset = *offp;
182 afs_int32 xfered = 0;
184 if (toffs < afs_vmMappingEnd) {
185 /* special case of a buffer crossing the VM mapping end */
186 afs_int32 tcount = afs_vmMappingEnd - *offp;
189 code = generic_file_write(fp, buf, tcount, offp);
191 if (code != tcount) {
197 setup_uio(&tuio, &iov, buf + xfered, (afs_offs_t) toffs, count,
198 UIO_WRITE, AFS_UIOSYS);
199 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
200 xfered += count - tuio.uio_resid;
203 *offp += count - tuio.uio_resid;
205 /* Purge dirty chunks of file if there are too many dirty chunks.
206 * Inside the write loop, we only do this at a chunk boundary.
207 * Clean up partial chunk if necessary at end of loop.
209 if (AFS_CHUNKBASE(tuio.afsio_offset) !=
210 AFS_CHUNKBASE(oldOffset)) {
211 ObtainWriteLock(&vcp->lock, 402);
212 code = afs_DoPartialWrite(vcp, &treq);
213 vcp->states |= CDirty;
214 ReleaseWriteLock(&vcp->lock);
219 ObtainWriteLock(&vcp->lock, 400);
220 vcp->m.Date = osi_Time(); /* Set file date (for ranlib) */
222 if (!(fp->f_flags & O_APPEND) && toffs > vcp->m.Length) {
223 vcp->m.Length = toffs;
225 ReleaseWriteLock(&vcp->lock);
230 #endif /* AFS_64BIT_CLIENT */
232 code = generic_file_write(fp, buf, count, offp);
234 #ifdef AFS_64BIT_CLIENT
236 #endif /* AFS_64BIT_CLIENT */
239 ObtainWriteLock(&vcp->lock, 530);
240 vcp->m.Date = osi_Time(); /* set modification time */
241 afs_FakeClose(vcp, credp);
243 code2 = afs_DoPartialWrite(vcp, &treq);
244 if (code2 && code >= 0)
245 code = (ssize_t) - code2;
246 ReleaseWriteLock(&vcp->lock);
248 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
249 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
257 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
259 /* This is a complete rewrite of afs_readdir, since we can make use of
260 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
261 * handling and use of bulkstats will need to be reflected here as well.
264 afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
266 extern struct DirEntry *afs_dir_GetBlob();
267 struct vcache *avc = ITOAFS(FILE_INODE(fp));
268 struct vrequest treq;
269 register struct dcache *tdc;
276 afs_size_t origOffset, tlen;
277 cred_t *credp = crref();
278 struct afs_fakestat_state fakestat;
280 #if defined(AFS_LINUX26_ENV)
284 AFS_STATCNT(afs_readdir);
286 code = afs_InitReq(&treq, credp);
291 afs_InitFakeStat(&fakestat);
292 code = afs_EvalFakeStat(&avc, &fakestat, &treq);
296 /* update the cache entry */
298 code = afs_VerifyVCache(avc, &treq);
302 /* get a reference to the entire directory */
303 tdc = afs_GetDCache(avc, (afs_size_t) 0, &treq, &origOffset, &tlen, 1);
309 ObtainReadLock(&avc->lock);
310 ObtainReadLock(&tdc->lock);
312 * Make sure that the data in the cache is current. There are two
313 * cases we need to worry about:
314 * 1. The cache data is being fetched by another process.
315 * 2. The cache data is no longer valid
317 while ((avc->states & CStatd)
318 && (tdc->dflags & DFFetching)
319 && hsame(avc->m.DataVersion, tdc->f.versionNo)) {
320 ReleaseReadLock(&tdc->lock);
321 ReleaseReadLock(&avc->lock);
322 afs_osi_Sleep(&tdc->validPos);
323 ObtainReadLock(&avc->lock);
324 ObtainReadLock(&tdc->lock);
326 if (!(avc->states & CStatd)
327 || !hsame(avc->m.DataVersion, tdc->f.versionNo)) {
328 ReleaseReadLock(&tdc->lock);
329 ReleaseReadLock(&avc->lock);
334 /* Fill in until we get an error or we're done. This implementation
335 * takes an offset in units of blobs, rather than bytes.
338 offset = (int) fp->f_pos;
340 dirpos = BlobScan(tdc, offset);
344 de = afs_dir_GetBlob(tdc, dirpos);
348 ino = (avc->fid.Fid.Volume << 16) + ntohl(de->fid.vnode);
349 ino &= 0x7fffffff; /* Assumes 32 bit ino_t ..... */
351 len = strlen(de->name);
353 printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
354 (unsigned long)&tdc->f.inode, dirpos);
355 DRelease((struct buffer *) de, 0);
357 ReleaseReadLock(&avc->lock);
362 /* filldir returns -EINVAL when the buffer is full. */
363 #if defined(AFS_LINUX26_ENV) || ((defined(AFS_LINUX24_ENV) || defined(pgoff2loff)) && defined(DECLARE_FSTYPE))
365 unsigned int type = DT_UNKNOWN;
366 struct VenusFid afid;
369 afid.Cell = avc->fid.Cell;
370 afid.Fid.Volume = avc->fid.Fid.Volume;
371 afid.Fid.Vnode = ntohl(de->fid.vnode);
372 afid.Fid.Unique = ntohl(de->fid.vunique);
373 if ((avc->states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
375 } else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
378 } else if (((tvc->states) & (CStatd | CTruth))) {
379 /* CTruth will be set if the object has
384 else if (vtype == VREG)
386 /* Don't do this until we're sure it can't be a mtpt */
387 /* else if (vtype == VLNK)
389 /* what other types does AFS support? */
391 /* clean up from afs_FindVCache */
394 code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
397 code = (*filldir) (dirbuf, de->name, len, offset, ino);
399 DRelease((struct buffer *)de, 0);
402 offset = dirpos + 1 + ((len + 16) >> 5);
404 /* If filldir didn't fill in the last one this is still pointing to that
407 fp->f_pos = (loff_t) offset;
409 ReleaseReadLock(&tdc->lock);
411 ReleaseReadLock(&avc->lock);
415 afs_PutFakeStat(&fakestat);
418 #if defined(AFS_LINUX26_ENV)
425 /* in afs_pioctl.c */
426 extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
429 #if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
430 static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
432 return afs_xioctl(FILE_INODE(fp), fp, com, arg);
437 /* We need to detect unmap's after close. To do that, we need our own
438 * vm_operations_struct's. And we need to set them up for both the
439 * private and shared mappings. The fun part is that these are all static
440 * so we'll have to initialize on the fly!
442 static struct vm_operations_struct afs_private_mmap_ops;
443 static int afs_private_mmap_ops_inited = 0;
444 static struct vm_operations_struct afs_shared_mmap_ops;
445 static int afs_shared_mmap_ops_inited = 0;
448 afs_linux_vma_close(struct vm_area_struct *vmap)
457 vcp = ITOAFS(FILE_INODE(vmap->vm_file));
462 afs_Trace4(afs_iclSetp, CM_TRACE_VM_CLOSE, ICL_TYPE_POINTER, vcp,
463 ICL_TYPE_INT32, vcp->mapcnt, ICL_TYPE_INT32, vcp->opens,
464 ICL_TYPE_INT32, vcp->execsOrWriters);
465 if ((&vcp->lock)->excl_locked == 0 || (&vcp->lock)->pid_writer == MyPidxx) {
466 ObtainWriteLock(&vcp->lock, 532);
469 printk("AFS_VMA_CLOSE(%d): Skipping Already locked vcp=%p vmap=%p\n",
470 MyPidxx, &vcp, &vmap);
474 ReleaseWriteLock(&vcp->lock);
476 if (need_unlock && vcp->execsOrWriters < 2) {
478 (void)afs_close(vcp, vmap->vm_file->f_flags, credp);
479 /* only decrement the execsOrWriters flag if this is not a
481 if (!(vcp->states & CRO) )
482 if (! (vmap->vm_file->f_flags & (FWRITE | FTRUNC)))
483 vcp->execsOrWriters--;
484 vcp->states &= ~CMAPPED;
486 } else if ((vmap->vm_file->f_flags & (FWRITE | FTRUNC)))
487 vcp->execsOrWriters--;
488 /* If we did not have the lock */
491 if (!vcp->execsOrWriters)
492 vcp->execsOrWriters = 1;
497 ReleaseWriteLock(&vcp->lock);
504 afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
506 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
507 cred_t *credp = crref();
508 struct vrequest treq;
512 #if defined(AFS_LINUX24_ENV)
513 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
514 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
515 vmap->vm_end - vmap->vm_start);
517 afs_Trace4(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
518 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
519 vmap->vm_end - vmap->vm_start, ICL_TYPE_INT32,
523 /* get a validated vcache entry */
524 code = afs_InitReq(&treq, credp);
526 code = afs_VerifyVCache(vcp, &treq);
528 if (!code && (vcp->states & CRO) &&
529 (vmap->vm_file->f_flags & (FWRITE | FTRUNC)))
535 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
538 code = generic_file_mmap(fp, vmap);
543 ObtainWriteLock(&vcp->lock, 531);
544 /* Set out vma ops so we catch the close. The following test should be
545 * the same as used in generic_file_mmap.
547 if ((vmap->vm_flags & VM_SHARED) && (vmap->vm_flags & VM_MAYWRITE)) {
548 if (!afs_shared_mmap_ops_inited) {
549 afs_shared_mmap_ops_inited = 1;
550 afs_shared_mmap_ops = *vmap->vm_ops;
551 afs_shared_mmap_ops.close = afs_linux_vma_close;
553 vmap->vm_ops = &afs_shared_mmap_ops;
555 if (!afs_private_mmap_ops_inited) {
556 afs_private_mmap_ops_inited = 1;
557 afs_private_mmap_ops = *vmap->vm_ops;
558 afs_private_mmap_ops.close = afs_linux_vma_close;
560 vmap->vm_ops = &afs_private_mmap_ops;
564 /* Add an open reference on the first mapping. */
565 if (vcp->mapcnt == 0) {
566 if (!(vcp->states & CRO))
567 vcp->execsOrWriters++;
569 vcp->states |= CMAPPED;
571 ReleaseWriteLock(&vcp->lock);
581 afs_linux_open(struct inode *ip, struct file *fp)
584 cred_t *credp = crref();
586 #ifdef AFS_LINUX24_ENV
590 code = afs_open((struct vcache **)&ip, fp->f_flags, credp);
592 #ifdef AFS_LINUX24_ENV
601 afs_linux_release(struct inode *ip, struct file *fp)
603 struct vcache *vcp = ITOAFS(ip);
604 cred_t *credp = crref();
607 #ifdef AFS_LINUX24_ENV
611 code = afs_close(vcp, fp->f_flags, credp);
613 #ifdef AFS_LINUX24_ENV
621 #if defined(AFS_LINUX24_ENV)
623 afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
626 afs_linux_fsync(struct file *fp, struct dentry *dp)
630 struct inode *ip = FILE_INODE(fp);
631 cred_t *credp = crref();
633 #ifdef AFS_LINUX24_ENV
637 code = afs_fsync(ITOAFS(ip), credp);
639 #ifdef AFS_LINUX24_ENV
649 afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
652 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
653 cred_t *credp = crref();
654 struct AFS_FLOCK flock;
655 /* Convert to a lock format afs_lockctl understands. */
656 memset((char *)&flock, 0, sizeof(flock));
657 flock.l_type = flp->fl_type;
658 flock.l_pid = flp->fl_pid;
660 flock.l_start = flp->fl_start;
661 flock.l_len = flp->fl_end - flp->fl_start;
663 /* Safe because there are no large files, yet */
664 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
665 if (cmd == F_GETLK64)
667 else if (cmd == F_SETLK64)
669 else if (cmd == F_SETLKW64)
671 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
674 code = afs_lockctl(vcp, &flock, cmd, credp);
677 /* Convert flock back to Linux's file_lock */
678 flp->fl_type = flock.l_type;
679 flp->fl_pid = flock.l_pid;
680 flp->fl_start = flock.l_start;
681 flp->fl_end = flock.l_start + flock.l_len;
689 * essentially the same as afs_fsync() but we need to get the return
690 * code for the sys_close() here, not afs_linux_release(), so call
691 * afs_StoreAllSegments() with AFS_LASTSTORE
694 afs_linux_flush(struct file *fp)
696 struct vrequest treq;
697 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
698 cred_t *credp = crref();
703 code = afs_InitReq(&treq, credp);
707 ObtainSharedLock(&vcp->lock, 535);
708 if (vcp->execsOrWriters > 0) {
709 UpgradeSToWLock(&vcp->lock, 536);
710 code = afs_StoreAllSegments(vcp, &treq, AFS_SYNC | AFS_LASTSTORE);
711 ConvertWToSLock(&vcp->lock);
713 code = afs_CheckCode(code, &treq, 54);
714 ReleaseSharedLock(&vcp->lock);
723 #if !defined(AFS_LINUX24_ENV)
724 /* Not allowed to directly read a directory. */
726 afs_linux_dir_read(struct file * fp, char *buf, size_t count, loff_t * ppos)
734 struct file_operations afs_dir_fops = {
735 #if !defined(AFS_LINUX24_ENV)
736 .read = afs_linux_dir_read,
737 .lock = afs_linux_lock,
738 .fsync = afs_linux_fsync,
740 .read = generic_read_dir,
742 .readdir = afs_linux_readdir,
743 #ifdef HAVE_UNLOCKED_IOCTL
744 .unlocked_ioctl = afs_unlocked_xioctl,
748 #ifdef HAVE_COMPAT_IOCTL
749 .compat_ioctl = afs_unlocked_xioctl,
751 .open = afs_linux_open,
752 .release = afs_linux_release,
755 struct file_operations afs_file_fops = {
756 .read = afs_linux_read,
757 .write = afs_linux_write,
758 #ifdef HAVE_UNLOCKED_IOCTL
759 .unlocked_ioctl = afs_unlocked_xioctl,
763 #ifdef HAVE_COMPAT_IOCTL
764 .compat_ioctl = afs_unlocked_xioctl,
766 .mmap = afs_linux_mmap,
767 .open = afs_linux_open,
768 .flush = afs_linux_flush,
769 #ifdef AFS_LINUX26_ENV
770 .sendfile = generic_file_sendfile,
772 .release = afs_linux_release,
773 .fsync = afs_linux_fsync,
774 .lock = afs_linux_lock,
778 /**********************************************************************
779 * AFS Linux dentry operations
780 **********************************************************************/
782 /* afs_linux_revalidate
783 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
786 afs_linux_revalidate(struct dentry *dp)
790 struct vrequest treq;
791 struct vcache *vcp = ITOAFS(dp->d_inode);
792 struct vcache *rootvp = NULL;
794 #ifdef AFS_LINUX24_ENV
799 if (afs_fakestat_enable && vcp->mvstat == 1 && vcp->mvid
800 && (vcp->states & CMValid) && (vcp->states & CStatd)) {
801 ObtainSharedLock(&afs_xvcache, 680);
802 rootvp = afs_FindVCache(vcp->mvid, 0, 0);
803 ReleaseSharedLock(&afs_xvcache);
806 /* Make this a fast path (no crref), since it's called so often. */
807 if (vcp->states & CStatd) {
808 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
809 check_bad_parent(dp); /* check and correct mvid */
811 vcache2fakeinode(rootvp, vcp);
815 afs_PutVCache(rootvp);
817 #ifdef AFS_LINUX24_ENV
824 code = afs_InitReq(&treq, credp);
826 code = afs_VerifyVCache(vcp, &treq);
829 #ifdef AFS_LINUX24_ENV
837 #if defined(AFS_LINUX26_ENV)
839 afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
841 int err = afs_linux_revalidate(dentry);
843 generic_fillattr(dentry->d_inode, stat);
848 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
849 * In kernels 2.2.10 and above, we are passed an additional flags var which
850 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
851 * we are advised to follow the entry if it is a link or to make sure that
852 * it is a directory. But since the kernel itself checks these possibilities
853 * later on, we shouldn't have to do it until later. Perhaps in the future..
855 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
856 #ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
858 afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
861 afs_linux_dentry_revalidate(struct dentry *dp, int flags)
865 afs_linux_dentry_revalidate(struct dentry *dp)
868 cred_t *credp = NULL;
869 struct vrequest treq;
870 struct vcache *lookupvcp = NULL;
871 int code, bad_dentry = 1;
872 struct vcache *vcp, *parentvcp;
874 #ifdef AFS_LINUX24_ENV
879 vcp = ITOAFS(dp->d_inode);
880 parentvcp = ITOAFS(dp->d_parent->d_inode);
882 /* If it's a negative dentry, then there's nothing to do. */
883 if (!vcp || !parentvcp)
886 /* If it is the AFS root, then there's no chance it needs
888 if (vcp == afs_globalVp) {
893 /* Make this a fast path (no crref), since it's called so often. */
894 if (vcp->states & CStatd) {
895 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
896 check_bad_parent(dp); /* check and correct mvid */
904 /* get a validated vcache entry */
905 code = afs_InitReq(&treq, credp);
908 code = afs_VerifyVCache(vcp, &treq);
914 code = afs_lookup(parentvcp, dp->d_name.name, &lookupvcp, credp);
916 /* Verify that the dentry does not point to an old inode */
917 if (vcp != lookupvcp)
925 afs_PutVCache(lookupvcp);
930 shrink_dcache_parent(dp);
934 #ifdef AFS_LINUX24_ENV
943 #if !defined(AFS_LINUX26_ENV)
944 /* afs_dentry_iput */
946 afs_dentry_iput(struct dentry *dp, struct inode *ip)
953 afs_dentry_delete(struct dentry *dp)
955 if (dp->d_inode && (ITOAFS(dp->d_inode)->states & CUnlinked))
956 return 1; /* bad inode? */
961 struct dentry_operations afs_dentry_operations = {
962 .d_revalidate = afs_linux_dentry_revalidate,
963 .d_delete = afs_dentry_delete,
964 #if !defined(AFS_LINUX26_ENV)
965 .d_iput = afs_dentry_iput,
969 /**********************************************************************
970 * AFS Linux inode operations
971 **********************************************************************/
975 * Merely need to set enough of vattr to get us through the create. Note
976 * that the higher level code (open_namei) will take care of any tuncation
977 * explicitly. Exclusive open is also taken care of in open_namei.
979 * name is in kernel space at this point.
981 #ifdef IOP_CREATE_TAKES_NAMEIDATA
983 afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
984 struct nameidata *nd)
987 afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
991 cred_t *credp = crref();
993 const char *name = dp->d_name.name;
997 vattr.va_mode = mode;
999 #if defined(AFS_LINUX26_ENV)
1004 afs_create(ITOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
1005 (struct vcache **)&ip, credp);
1008 vattr2inode(ip, &vattr);
1009 /* Reset ops if symlink or directory. */
1010 #if defined(AFS_LINUX24_ENV)
1011 if (S_ISREG(ip->i_mode)) {
1012 ip->i_op = &afs_file_iops;
1013 ip->i_fop = &afs_file_fops;
1014 ip->i_data.a_ops = &afs_file_aops;
1015 } else if (S_ISDIR(ip->i_mode)) {
1016 ip->i_op = &afs_dir_iops;
1017 ip->i_fop = &afs_dir_fops;
1018 } else if (S_ISLNK(ip->i_mode)) {
1019 ip->i_op = &afs_symlink_iops;
1020 ip->i_data.a_ops = &afs_symlink_aops;
1021 ip->i_mapping = &ip->i_data;
1023 printk("afs_linux_create: FIXME\n");
1025 if (S_ISDIR(ip->i_mode))
1026 ip->i_op = &afs_dir_iops;
1027 else if (S_ISLNK(ip->i_mode))
1028 ip->i_op = &afs_symlink_iops;
1031 dp->d_op = &afs_dentry_operations;
1032 d_instantiate(dp, ip);
1036 #if defined(AFS_LINUX26_ENV)
1043 /* afs_linux_lookup */
1044 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1045 #ifdef IOP_LOOKUP_TAKES_NAMEIDATA
1047 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1048 struct nameidata *nd)
1051 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1055 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1059 cred_t *credp = crref();
1060 struct vcache *vcp = NULL;
1061 const char *comp = dp->d_name.name;
1063 #if defined(AFS_LINUX26_ENV)
1067 code = afs_lookup(ITOAFS(dip), comp, &vcp, credp);
1071 struct inode *ip = AFSTOI(vcp);
1072 /* Reset ops if symlink or directory. */
1073 #if defined(AFS_LINUX24_ENV)
1074 if (S_ISREG(ip->i_mode)) {
1075 ip->i_op = &afs_file_iops;
1076 ip->i_fop = &afs_file_fops;
1077 ip->i_data.a_ops = &afs_file_aops;
1078 } else if (S_ISDIR(ip->i_mode)) {
1079 ip->i_op = &afs_dir_iops;
1080 ip->i_fop = &afs_dir_fops;
1081 } else if (S_ISLNK(ip->i_mode)) {
1082 ip->i_op = &afs_symlink_iops;
1083 ip->i_data.a_ops = &afs_symlink_aops;
1084 ip->i_mapping = &ip->i_data;
1087 ("afs_linux_lookup: ip->i_mode 0x%x dp->d_name.name %s code %d\n",
1088 ip->i_mode, dp->d_name.name, code);
1089 #ifdef STRUCT_INODE_HAS_I_SECURITY
1090 if (ip->i_security == NULL) {
1091 if (security_inode_alloc(ip))
1092 panic("afs_linux_lookup: Cannot allocate inode security");
1096 if (S_ISDIR(ip->i_mode))
1097 ip->i_op = &afs_dir_iops;
1098 else if (S_ISLNK(ip->i_mode))
1099 ip->i_op = &afs_symlink_iops;
1102 dp->d_op = &afs_dentry_operations;
1103 d_add(dp, AFSTOI(vcp));
1105 #if defined(AFS_LINUX26_ENV)
1110 /* It's ok for the file to not be found. That's noted by the caller by
1111 * seeing that the dp->d_inode field is NULL.
1113 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1117 return ERR_PTR(-code);
1126 afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
1129 cred_t *credp = crref();
1130 const char *name = newdp->d_name.name;
1131 struct inode *oldip = olddp->d_inode;
1133 /* If afs_link returned the vnode, we could instantiate the
1134 * dentry. Since it's not, we drop this one and do a new lookup.
1139 code = afs_link(ITOAFS(oldip), ITOAFS(dip), name, credp);
1147 afs_linux_unlink(struct inode *dip, struct dentry *dp)
1150 cred_t *credp = crref();
1151 const char *name = dp->d_name.name;
1152 struct vcache *tvc = ITOAFS(dp->d_inode);
1154 #if defined(AFS_LINUX26_ENV)
1157 if (((VREFCOUNT(tvc) > 0) && tvc->opens > 0)
1158 && !(tvc->states & CUnlinked)) {
1159 struct dentry *__dp;
1161 extern char *afs_newname();
1170 osi_FreeSmallSpace(__name);
1171 __name = afs_newname();
1174 __dp = lookup_one_len(__name, dp->d_parent, strlen(__name));
1178 } while (__dp->d_inode != NULL);
1181 code = afs_rename(ITOAFS(dip), dp->d_name.name, ITOAFS(dip), __dp->d_name.name, credp);
1186 crfree(tvc->uncred);
1188 tvc->uncred = credp;
1189 tvc->states |= CUnlinked;
1201 code = afs_remove(ITOAFS(dip), name, credp);
1206 #if defined(AFS_LINUX26_ENV)
1215 afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
1218 cred_t *credp = crref();
1220 const char *name = dp->d_name.name;
1222 /* If afs_symlink returned the vnode, we could instantiate the
1223 * dentry. Since it's not, we drop this one and do a new lookup.
1229 code = afs_symlink(ITOAFS(dip), name, &vattr, target, credp);
1236 afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1239 cred_t *credp = crref();
1240 struct vcache *tvcp = NULL;
1242 const char *name = dp->d_name.name;
1244 #if defined(AFS_LINUX26_ENV)
1249 vattr.va_mask = ATTR_MODE;
1250 vattr.va_mode = mode;
1251 code = afs_mkdir(ITOAFS(dip), name, &vattr, &tvcp, credp);
1255 tvcp->v.v_op = &afs_dir_iops;
1256 #if defined(AFS_LINUX24_ENV)
1257 tvcp->v.v_fop = &afs_dir_fops;
1259 dp->d_op = &afs_dentry_operations;
1260 d_instantiate(dp, AFSTOI(tvcp));
1263 #if defined(AFS_LINUX26_ENV)
1271 afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1274 cred_t *credp = crref();
1275 const char *name = dp->d_name.name;
1277 #if defined(AFS_LINUX26_ENV)
1281 code = afs_rmdir(ITOAFS(dip), name, credp);
1284 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1285 * that failed because a directory is not empty. So, we map
1286 * EEXIST to ENOTEMPTY on linux.
1288 if (code == EEXIST) {
1296 #if defined(AFS_LINUX26_ENV)
1306 afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1307 struct inode *newip, struct dentry *newdp)
1310 cred_t *credp = crref();
1311 const char *oldname = olddp->d_name.name;
1312 const char *newname = newdp->d_name.name;
1313 struct dentry *rehash = NULL;
1315 #if defined(AFS_LINUX26_ENV)
1316 /* Prevent any new references during rename operation. */
1319 /* Remove old and new entries from name hash. New one will change below.
1320 * While it's optimal to catch failures and re-insert newdp into hash,
1321 * it's also error prone and in that case we're already dealing with error
1322 * cases. Let another lookup put things right, if need be.
1324 #if defined(AFS_LINUX26_ENV)
1325 if (!d_unhashed(newdp)) {
1330 if (!list_empty(&newdp->d_hash)) {
1336 #if defined(AFS_LINUX24_ENV)
1337 if (atomic_read(&olddp->d_count) > 1)
1338 shrink_dcache_parent(olddp);
1342 code = afs_rename(ITOAFS(oldip), oldname, ITOAFS(newip), newname, credp);
1348 #if defined(AFS_LINUX26_ENV)
1357 /* afs_linux_ireadlink
1358 * Internal readlink which can return link contents to user or kernel space.
1359 * Note that the buffer is NOT supposed to be null-terminated.
1362 afs_linux_ireadlink(struct inode *ip, char *target, int maxlen, uio_seg_t seg)
1365 cred_t *credp = crref();
1369 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1370 code = afs_readlink(ITOAFS(ip), &tuio, credp);
1374 return maxlen - tuio.uio_resid;
1379 #if !defined(AFS_LINUX24_ENV)
1380 /* afs_linux_readlink
1381 * Fill target (which is in user space) with contents of symlink.
1384 afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1387 struct inode *ip = dp->d_inode;
1390 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1396 /* afs_linux_follow_link
1397 * a file system dependent link following routine.
1400 afs_linux_follow_link(struct dentry *dp, struct dentry *basep,
1401 unsigned int follow)
1409 name = osi_Alloc(PATH_MAX + 1);
1413 return ERR_PTR(-EIO);
1416 code = afs_linux_ireadlink(dp->d_inode, name, PATH_MAX, AFS_UIOSYS);
1421 res = ERR_PTR(code);
1424 res = lookup_dentry(name, basep, follow);
1428 osi_Free(name, PATH_MAX + 1);
1434 /* afs_linux_readpage
1435 * all reads come through here. A strategy-like read call.
1438 afs_linux_readpage(struct file *fp, struct page *pp)
1441 cred_t *credp = crref();
1442 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1444 afs_offs_t offset = pp->index << PAGE_CACHE_SHIFT;
1446 ulong address = afs_linux_page_address(pp);
1447 afs_offs_t offset = pageoff(pp);
1451 struct inode *ip = FILE_INODE(fp);
1452 int cnt = page_count(pp);
1453 struct vcache *avc = ITOAFS(ip);
1456 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1460 atomic_add(1, &pp->count);
1461 set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
1462 clear_bit(PG_error, &pp->flags);
1465 setup_uio(&tuio, &iovec, (char *)address, offset, PAGESIZE, UIO_READ,
1467 #ifdef AFS_LINUX24_ENV
1471 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip, ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32, 99999); /* not a possible code value */
1472 code = afs_rdwr(avc, &tuio, UIO_READ, 0, credp);
1473 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
1474 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
1477 #ifdef AFS_LINUX24_ENV
1482 if (tuio.uio_resid) /* zero remainder of page */
1483 memset((void *)(address + (PAGESIZE - tuio.uio_resid)), 0,
1485 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1486 flush_dcache_page(pp);
1487 SetPageUptodate(pp);
1489 set_bit(PG_uptodate, &pp->flags);
1493 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1497 clear_bit(PG_locked, &pp->flags);
1502 if (!code && AFS_CHUNKOFFSET(offset) == 0) {
1504 struct vrequest treq;
1507 code = afs_InitReq(&treq, credp);
1508 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
1509 tdc = afs_FindDCache(avc, offset);
1511 if (!(tdc->mflags & DFNextStarted))
1512 afs_PrefetchChunk(avc, tdc, credp, &treq);
1515 ReleaseWriteLock(&avc->lock);
1524 #if defined(AFS_LINUX24_ENV)
1525 #ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
1527 afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
1530 afs_linux_writepage(struct page *pp)
1533 struct address_space *mapping = pp->mapping;
1534 struct inode *inode;
1535 unsigned long end_index;
1536 unsigned offset = PAGE_CACHE_SIZE;
1539 #if defined(AFS_LINUX26_ENV)
1540 if (PageReclaim(pp)) {
1541 return WRITEPAGE_ACTIVATE;
1544 if (PageLaunder(pp)) {
1545 return(fail_writepage(pp));
1549 inode = (struct inode *)mapping->host;
1550 end_index = inode->i_size >> PAGE_CACHE_SHIFT;
1553 if (pp->index < end_index)
1555 /* things got complicated... */
1556 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
1557 /* OK, are we completely out? */
1558 if (pp->index >= end_index + 1 || !offset)
1561 status = afs_linux_writepage_sync(inode, pp, 0, offset);
1562 SetPageUptodate(pp);
1564 if (status == offset)
1571 /* afs_linux_permission
1572 * Check access rights - returns error if can't check or permission denied.
1574 #ifdef IOP_PERMISSION_TAKES_NAMEIDATA
1576 afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
1579 afs_linux_permission(struct inode *ip, int mode)
1583 cred_t *credp = crref();
1587 if (mode & MAY_EXEC)
1589 if (mode & MAY_READ)
1591 if (mode & MAY_WRITE)
1593 code = afs_access(ITOAFS(ip), tmp, credp);
1601 #if defined(AFS_LINUX24_ENV)
1603 afs_linux_writepage_sync(struct inode *ip, struct page *pp,
1604 unsigned long offset, unsigned int count)
1606 struct vcache *vcp = ITOAFS(ip);
1615 buffer = kmap(pp) + offset;
1616 base = (pp->index << PAGE_CACHE_SHIFT) + offset;
1621 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1622 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1623 ICL_TYPE_INT32, 99999);
1625 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
1627 code = afs_write(vcp, &tuio, f_flags, credp, 0);
1632 && afs_stats_cmperf.cacheCurrDirtyChunks >
1633 afs_stats_cmperf.cacheMaxDirtyChunks) {
1634 struct vrequest treq;
1636 ObtainWriteLock(&vcp->lock, 533);
1637 if (!afs_InitReq(&treq, credp))
1638 code = afs_DoPartialWrite(vcp, &treq);
1639 ReleaseWriteLock(&vcp->lock);
1641 code = code ? -code : count - tuio.uio_resid;
1643 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1644 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1645 ICL_TYPE_INT32, code);
1656 /* afs_linux_updatepage
1657 * What one would have thought was writepage - write dirty page to file.
1658 * Called from generic_file_write. buffer is still in user space. pagep
1659 * has been filled in with old data if we're updating less than a page.
1662 afs_linux_updatepage(struct file *fp, struct page *pp, unsigned long offset,
1663 unsigned int count, int sync)
1665 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
1666 u8 *page_addr = (u8 *) afs_linux_page_address(pp);
1672 set_bit(PG_locked, &pp->flags);
1676 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1677 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1678 ICL_TYPE_INT32, 99999);
1679 setup_uio(&tuio, &iovec, page_addr + offset,
1680 (afs_offs_t) (pageoff(pp) + offset), count, UIO_WRITE,
1683 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
1687 code = code ? -code : count - tuio.uio_resid;
1688 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1689 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1690 ICL_TYPE_INT32, code);
1695 clear_bit(PG_locked, &pp->flags);
1700 #if defined(AFS_LINUX24_ENV)
1702 afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
1707 code = afs_linux_writepage_sync(file->f_dentry->d_inode, page,
1708 offset, to - offset);
1709 #if !defined(AFS_LINUX26_ENV)
1717 afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
1720 /* sometime between 2.4.0 and 2.4.19, the callers of prepare_write began to
1721 call kmap directly instead of relying on us to do it */
1722 #if !defined(AFS_LINUX26_ENV)
1728 extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
1731 struct inode_operations afs_file_iops = {
1732 #if defined(AFS_LINUX26_ENV)
1733 .permission = afs_linux_permission,
1734 .getattr = afs_linux_getattr,
1735 .setattr = afs_notify_change,
1736 #elif defined(AFS_LINUX24_ENV)
1737 .permission = afs_linux_permission,
1738 .revalidate = afs_linux_revalidate,
1739 .setattr = afs_notify_change,
1741 .default_file_ops = &afs_file_fops,
1742 .readpage = afs_linux_readpage,
1743 .revalidate = afs_linux_revalidate,
1744 .updatepage = afs_linux_updatepage,
1748 #if defined(AFS_LINUX24_ENV)
1749 struct address_space_operations afs_file_aops = {
1750 .readpage = afs_linux_readpage,
1751 .writepage = afs_linux_writepage,
1752 .commit_write = afs_linux_commit_write,
1753 .prepare_write = afs_linux_prepare_write,
1758 /* Separate ops vector for directories. Linux 2.2 tests type of inode
1759 * by what sort of operation is allowed.....
1762 struct inode_operations afs_dir_iops = {
1763 #if !defined(AFS_LINUX24_ENV)
1764 .default_file_ops = &afs_dir_fops,
1766 .setattr = afs_notify_change,
1768 .create = afs_linux_create,
1769 .lookup = afs_linux_lookup,
1770 .link = afs_linux_link,
1771 .unlink = afs_linux_unlink,
1772 .symlink = afs_linux_symlink,
1773 .mkdir = afs_linux_mkdir,
1774 .rmdir = afs_linux_rmdir,
1775 .rename = afs_linux_rename,
1776 #if defined(AFS_LINUX26_ENV)
1777 .getattr = afs_linux_getattr,
1779 .revalidate = afs_linux_revalidate,
1781 .permission = afs_linux_permission,
1784 /* We really need a separate symlink set of ops, since do_follow_link()
1785 * determines if it _is_ a link by checking if the follow_link op is set.
1787 #if defined(AFS_LINUX24_ENV)
1789 afs_symlink_filler(struct file *file, struct page *page)
1791 struct inode *ip = (struct inode *)page->mapping->host;
1792 char *p = (char *)kmap(page);
1797 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
1802 p[code] = '\0'; /* null terminate? */
1805 SetPageUptodate(page);
1819 struct address_space_operations afs_symlink_aops = {
1820 .readpage = afs_symlink_filler
1824 struct inode_operations afs_symlink_iops = {
1825 #if defined(AFS_LINUX24_ENV)
1826 .readlink = page_readlink,
1827 #if defined(HAVE_KERNEL_PAGE_FOLLOW_LINK)
1828 .follow_link = page_follow_link,
1830 .follow_link = page_follow_link_light,
1831 .put_link = page_put_link,
1833 .setattr = afs_notify_change,
1835 .readlink = afs_linux_readlink,
1836 .follow_link = afs_linux_follow_link,
1837 .permission = afs_linux_permission,
1838 .revalidate = afs_linux_revalidate,