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"
31 #ifdef HAVE_MM_INLINE_H
32 #include "h/mm_inline.h"
34 #include "h/pagemap.h"
35 #if defined(AFS_LINUX24_ENV)
36 #include "h/smp_lock.h"
38 #if defined(AFS_LINUX26_ENV)
39 #include "h/writeback.h"
43 #define pageoff(pp) pgoff2loff((pp)->index)
45 #define pageoff(pp) pp->offset
48 #if defined(AFS_LINUX26_ENV)
49 #define UnlockPage(pp) unlock_page(pp)
52 extern struct vcache *afs_globalVp;
54 afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
57 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
58 cred_t *credp = crref();
62 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
63 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
66 /* get a validated vcache entry */
67 code = afs_InitReq(&treq, credp);
69 code = afs_VerifyVCache(vcp, &treq);
74 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
77 code = do_sync_read(fp, buf, count, offp);
79 code = generic_file_read(fp, buf, count, offp);
84 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
85 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
94 /* Now we have integrated VM for writes as well as reads. generic_file_write
95 * also takes care of re-positioning the pointer if file is open in append
96 * mode. Call fake open/close to ensure we do writes of core dumps.
99 afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
102 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
103 struct vrequest treq;
104 cred_t *credp = crref();
108 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
109 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
110 (fp->f_flags & O_APPEND) ? 99998 : 99999);
113 /* get a validated vcache entry */
114 code = (ssize_t) afs_InitReq(&treq, credp);
116 code = (ssize_t) afs_VerifyVCache(vcp, &treq);
118 ObtainWriteLock(&vcp->lock, 529);
120 ReleaseWriteLock(&vcp->lock);
126 code = do_sync_write(fp, buf, count, offp);
128 code = generic_file_write(fp, buf, count, offp);
133 ObtainWriteLock(&vcp->lock, 530);
134 afs_FakeClose(vcp, credp);
135 ReleaseWriteLock(&vcp->lock);
137 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
138 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
146 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
148 /* This is a complete rewrite of afs_readdir, since we can make use of
149 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
150 * handling and use of bulkstats will need to be reflected here as well.
153 afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
155 extern struct DirEntry *afs_dir_GetBlob();
156 struct vcache *avc = VTOAFS(FILE_INODE(fp));
157 struct vrequest treq;
158 register struct dcache *tdc;
165 afs_size_t origOffset, tlen;
166 cred_t *credp = crref();
167 struct afs_fakestat_state fakestat;
169 #if defined(AFS_LINUX26_ENV)
173 AFS_STATCNT(afs_readdir);
175 code = afs_InitReq(&treq, credp);
180 afs_InitFakeStat(&fakestat);
181 code = afs_EvalFakeStat(&avc, &fakestat, &treq);
185 /* update the cache entry */
187 code = afs_VerifyVCache(avc, &treq);
191 /* get a reference to the entire directory */
192 tdc = afs_GetDCache(avc, (afs_size_t) 0, &treq, &origOffset, &tlen, 1);
198 ObtainSharedLock(&avc->lock, 810);
199 UpgradeSToWLock(&avc->lock, 811);
200 ObtainReadLock(&tdc->lock);
202 * Make sure that the data in the cache is current. There are two
203 * cases we need to worry about:
204 * 1. The cache data is being fetched by another process.
205 * 2. The cache data is no longer valid
207 while ((avc->states & CStatd)
208 && (tdc->dflags & DFFetching)
209 && hsame(avc->m.DataVersion, tdc->f.versionNo)) {
210 ReleaseReadLock(&tdc->lock);
211 ReleaseSharedLock(&avc->lock);
212 afs_osi_Sleep(&tdc->validPos);
213 ObtainSharedLock(&avc->lock, 812);
214 ObtainReadLock(&tdc->lock);
216 if (!(avc->states & CStatd)
217 || !hsame(avc->m.DataVersion, tdc->f.versionNo)) {
218 ReleaseReadLock(&tdc->lock);
219 ReleaseSharedLock(&avc->lock);
224 /* Set the readdir-in-progress flag, and downgrade the lock
225 * to shared so others will be able to acquire a read lock.
227 avc->states |= CReadDir;
228 avc->dcreaddir = tdc;
229 avc->readdir_pid = MyPidxx;
230 ConvertWToSLock(&avc->lock);
232 /* Fill in until we get an error or we're done. This implementation
233 * takes an offset in units of blobs, rather than bytes.
236 offset = (int) fp->f_pos;
238 dirpos = BlobScan(tdc, offset);
242 de = afs_dir_GetBlob(tdc, dirpos);
246 ino = afs_calc_inum (avc->fid.Fid.Volume, ntohl(de->fid.vnode));
249 len = strlen(de->name);
251 printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
252 (unsigned long)&tdc->f.inode, dirpos);
253 DRelease((struct buffer *) de, 0);
254 ReleaseSharedLock(&avc->lock);
260 /* filldir returns -EINVAL when the buffer is full. */
261 #if defined(AFS_LINUX26_ENV) || ((defined(AFS_LINUX24_ENV) || defined(pgoff2loff)) && defined(DECLARE_FSTYPE))
263 unsigned int type = DT_UNKNOWN;
264 struct VenusFid afid;
267 afid.Cell = avc->fid.Cell;
268 afid.Fid.Volume = avc->fid.Fid.Volume;
269 afid.Fid.Vnode = ntohl(de->fid.vnode);
270 afid.Fid.Unique = ntohl(de->fid.vunique);
271 if ((avc->states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
273 } else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
276 } else if (((tvc->states) & (CStatd | CTruth))) {
277 /* CTruth will be set if the object has
282 else if (vtype == VREG)
284 /* Don't do this until we're sure it can't be a mtpt */
285 /* else if (vtype == VLNK)
287 /* what other types does AFS support? */
289 /* clean up from afs_FindVCache */
293 * If this is NFS readdirplus, then the filler is going to
294 * call getattr on this inode, which will deadlock if we're
298 code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
302 code = (*filldir) (dirbuf, de->name, len, offset, ino);
304 DRelease((struct buffer *)de, 0);
307 offset = dirpos + 1 + ((len + 16) >> 5);
309 /* If filldir didn't fill in the last one this is still pointing to that
312 fp->f_pos = (loff_t) offset;
314 ReleaseReadLock(&tdc->lock);
316 UpgradeSToWLock(&avc->lock, 813);
317 avc->states &= ~CReadDir;
319 avc->readdir_pid = 0;
320 ReleaseSharedLock(&avc->lock);
324 afs_PutFakeStat(&fakestat);
327 #if defined(AFS_LINUX26_ENV)
334 /* in afs_pioctl.c */
335 extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
338 #if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
339 static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
341 return afs_xioctl(FILE_INODE(fp), fp, com, arg);
348 afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
350 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
351 cred_t *credp = crref();
352 struct vrequest treq;
356 #if defined(AFS_LINUX24_ENV)
357 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
358 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
359 vmap->vm_end - vmap->vm_start);
361 afs_Trace4(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
362 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
363 vmap->vm_end - vmap->vm_start, ICL_TYPE_INT32,
367 /* get a validated vcache entry */
368 code = afs_InitReq(&treq, credp);
372 code = afs_VerifyVCache(vcp, &treq);
376 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
379 code = generic_file_mmap(fp, vmap);
382 vcp->states |= CMAPPED;
395 afs_linux_open(struct inode *ip, struct file *fp)
397 struct vcache *vcp = VTOAFS(ip);
398 cred_t *credp = crref();
401 #ifdef AFS_LINUX24_ENV
405 code = afs_open(&vcp, fp->f_flags, credp);
407 #ifdef AFS_LINUX24_ENV
416 afs_linux_release(struct inode *ip, struct file *fp)
418 struct vcache *vcp = VTOAFS(ip);
419 cred_t *credp = crref();
422 #ifdef AFS_LINUX24_ENV
426 code = afs_close(vcp, fp->f_flags, credp);
428 #ifdef AFS_LINUX24_ENV
437 #if defined(AFS_LINUX24_ENV)
438 afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
440 afs_linux_fsync(struct file *fp, struct dentry *dp)
444 struct inode *ip = FILE_INODE(fp);
445 cred_t *credp = crref();
447 #ifdef AFS_LINUX24_ENV
451 code = afs_fsync(VTOAFS(ip), credp);
453 #ifdef AFS_LINUX24_ENV
463 afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
466 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
467 cred_t *credp = crref();
468 struct AFS_FLOCK flock;
469 /* Convert to a lock format afs_lockctl understands. */
470 memset((char *)&flock, 0, sizeof(flock));
471 flock.l_type = flp->fl_type;
472 flock.l_pid = flp->fl_pid;
474 flock.l_start = flp->fl_start;
475 flock.l_len = flp->fl_end - flp->fl_start;
477 /* Safe because there are no large files, yet */
478 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
479 if (cmd == F_GETLK64)
481 else if (cmd == F_SETLK64)
483 else if (cmd == F_SETLKW64)
485 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
488 code = afs_lockctl(vcp, &flock, cmd, credp);
491 #ifdef AFS_LINUX24_ENV
492 if ((code == 0 || flp->fl_type == F_UNLCK) &&
493 (cmd == F_SETLK || cmd == F_SETLKW)) {
494 #ifdef POSIX_LOCK_FILE_WAIT_ARG
495 code = posix_lock_file(fp, flp, 0);
497 flp->fl_flags &=~ FL_SLEEP;
498 code = posix_lock_file(fp, flp);
500 if (code && flp->fl_type != F_UNLCK) {
501 struct AFS_FLOCK flock2;
503 flock2.l_type = F_UNLCK;
505 afs_lockctl(vcp, &flock2, F_SETLK, credp);
510 /* Convert flock back to Linux's file_lock */
511 flp->fl_type = flock.l_type;
512 flp->fl_pid = flock.l_pid;
513 flp->fl_start = flock.l_start;
514 flp->fl_end = flock.l_start + flock.l_len;
521 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
523 afs_linux_flock(struct file *fp, int cmd, struct file_lock *flp) {
525 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
526 cred_t *credp = crref();
527 struct AFS_FLOCK flock;
528 /* Convert to a lock format afs_lockctl understands. */
529 memset((char *)&flock, 0, sizeof(flock));
530 flock.l_type = flp->fl_type;
531 flock.l_pid = flp->fl_pid;
534 flock.l_len = OFFSET_MAX;
536 /* Safe because there are no large files, yet */
537 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
538 if (cmd == F_GETLK64)
540 else if (cmd == F_SETLK64)
542 else if (cmd == F_SETLKW64)
544 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
547 code = afs_lockctl(vcp, &flock, cmd, credp);
550 if ((code == 0 || flp->fl_type == F_UNLCK) &&
551 (cmd == F_SETLK || cmd == F_SETLKW)) {
552 flp->fl_flags &=~ FL_SLEEP;
553 code = flock_lock_file_wait(fp, flp);
554 if (code && flp->fl_type != F_UNLCK) {
555 struct AFS_FLOCK flock2;
557 flock2.l_type = F_UNLCK;
559 afs_lockctl(vcp, &flock2, F_SETLK, credp);
563 /* Convert flock back to Linux's file_lock */
564 flp->fl_type = flock.l_type;
565 flp->fl_pid = flock.l_pid;
573 * essentially the same as afs_fsync() but we need to get the return
574 * code for the sys_close() here, not afs_linux_release(), so call
575 * afs_StoreAllSegments() with AFS_LASTSTORE
578 #if defined(FOP_FLUSH_TAKES_FL_OWNER_T)
579 afs_linux_flush(struct file *fp, fl_owner_t id)
581 afs_linux_flush(struct file *fp)
584 struct vrequest treq;
585 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
586 cred_t *credp = crref();
591 code = afs_InitReq(&treq, credp);
595 ObtainSharedLock(&vcp->lock, 535);
596 if (vcp->execsOrWriters > 0) {
597 UpgradeSToWLock(&vcp->lock, 536);
598 code = afs_StoreAllSegments(vcp, &treq, AFS_SYNC | AFS_LASTSTORE);
599 ConvertWToSLock(&vcp->lock);
601 code = afs_CheckCode(code, &treq, 54);
602 ReleaseSharedLock(&vcp->lock);
611 #if !defined(AFS_LINUX24_ENV)
612 /* Not allowed to directly read a directory. */
614 afs_linux_dir_read(struct file * fp, char *buf, size_t count, loff_t * ppos)
622 struct file_operations afs_dir_fops = {
623 #if !defined(AFS_LINUX24_ENV)
624 .read = afs_linux_dir_read,
625 .lock = afs_linux_lock,
626 .fsync = afs_linux_fsync,
628 .read = generic_read_dir,
630 .readdir = afs_linux_readdir,
631 #ifdef HAVE_UNLOCKED_IOCTL
632 .unlocked_ioctl = afs_unlocked_xioctl,
636 #ifdef HAVE_COMPAT_IOCTL
637 .compat_ioctl = afs_unlocked_xioctl,
639 .open = afs_linux_open,
640 .release = afs_linux_release,
643 struct file_operations afs_file_fops = {
644 .read = afs_linux_read,
645 .write = afs_linux_write,
646 #ifdef GENERIC_FILE_AIO_READ
647 .aio_read = generic_file_aio_read,
648 .aio_write = generic_file_aio_write,
650 #ifdef HAVE_UNLOCKED_IOCTL
651 .unlocked_ioctl = afs_unlocked_xioctl,
655 #ifdef HAVE_COMPAT_IOCTL
656 .compat_ioctl = afs_unlocked_xioctl,
658 .mmap = afs_linux_mmap,
659 .open = afs_linux_open,
660 .flush = afs_linux_flush,
661 #if defined(AFS_LINUX26_ENV) && defined(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
662 .sendfile = generic_file_sendfile,
664 .release = afs_linux_release,
665 .fsync = afs_linux_fsync,
666 .lock = afs_linux_lock,
667 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
668 .flock = afs_linux_flock,
673 /**********************************************************************
674 * AFS Linux dentry operations
675 **********************************************************************/
677 /* check_bad_parent() : Checks if this dentry's vcache is a root vcache
678 * that has its mvid (parent dir's fid) pointer set to the wrong directory
679 * due to being mounted in multiple points at once. If so, check_bad_parent()
680 * calls afs_lookup() to correct the vcache's mvid, as well as the volume's
681 * dotdotfid and mtpoint fid members.
683 * dp - dentry to be checked.
687 * This dentry's vcache's mvid will be set to the correct parent directory's
689 * This root vnode's volume will have its dotdotfid and mtpoint fids set
690 * to the correct parent and mountpoint fids.
694 check_bad_parent(struct dentry *dp)
697 struct vcache *vcp = VTOAFS(dp->d_inode), *avc = NULL;
698 struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
700 if (vcp->mvid->Fid.Volume != pvc->fid.Fid.Volume) { /* bad parent */
703 /* force a lookup, so vcp->mvid is fixed up */
704 afs_lookup(pvc, dp->d_name.name, &avc, credp);
705 if (!avc || vcp != avc) { /* bad, very bad.. */
706 afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
707 "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
708 ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
709 ICL_TYPE_POINTER, dp);
712 AFS_RELE(AFSTOV(avc));
719 /* afs_linux_revalidate
720 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
723 afs_linux_revalidate(struct dentry *dp)
726 struct vcache *vcp = VTOAFS(dp->d_inode);
730 #ifdef AFS_LINUX24_ENV
736 /* Make this a fast path (no crref), since it's called so often. */
737 if (vcp->states & CStatd) {
739 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
740 check_bad_parent(dp); /* check and correct mvid */
743 #ifdef AFS_LINUX24_ENV
751 code = afs_getattr(vcp, &vattr, credp);
753 vattr2inode(AFSTOV(vcp), &vattr);
756 #ifdef AFS_LINUX24_ENV
764 #if defined(AFS_LINUX26_ENV)
766 afs_linux_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
768 int err = afs_linux_revalidate(dentry);
770 generic_fillattr(dentry->d_inode, stat);
776 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
777 * In kernels 2.2.10 and above, we are passed an additional flags var which
778 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
779 * we are advised to follow the entry if it is a link or to make sure that
780 * it is a directory. But since the kernel itself checks these possibilities
781 * later on, we shouldn't have to do it until later. Perhaps in the future..
784 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
785 #ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
786 afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
788 afs_linux_dentry_revalidate(struct dentry *dp, int flags)
791 afs_linux_dentry_revalidate(struct dentry *dp)
795 cred_t *credp = NULL;
796 struct vcache *vcp, *pvcp, *tvc = NULL;
799 #ifdef AFS_LINUX24_ENV
806 vcp = VTOAFS(dp->d_inode);
807 pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
809 if (vcp == afs_globalVp)
812 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
813 check_bad_parent(dp); /* check and correct mvid */
816 /* If the last looker changes, we should make sure the current
817 * looker still has permission to examine this file. This would
818 * always require a crref() which would be "slow".
820 if (vcp->last_looker != treq.uid) {
821 if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS))
824 vcp->last_looker = treq.uid;
828 /* If the parent's DataVersion has changed or the vnode
829 * is longer valid, we need to do a full lookup. VerifyVCache
830 * isn't enough since the vnode may have been renamed.
833 if (hgetlo(pvcp->m.DataVersion) > dp->d_time || !(vcp->states & CStatd)) {
836 afs_lookup(pvcp, dp->d_name.name, &tvc, credp);
837 if (!tvc || tvc != vcp)
840 if (afs_getattr(vcp, &vattr, credp))
843 vattr2inode(AFSTOV(vcp), &vattr);
844 dp->d_time = hgetlo(pvcp->m.DataVersion);
847 /* should we always update the attributes at this point? */
848 /* unlikely--the vcache entry hasn't changed */
852 pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
853 if (hgetlo(pvcp->m.DataVersion) > dp->d_time)
857 /* No change in parent's DataVersion so this negative
858 * lookup is still valid. BUT, if a server is down a
859 * negative lookup can result so there should be a
860 * liftime as well. For now, always expire.
878 shrink_dcache_parent(dp);
881 #ifdef AFS_LINUX24_ENV
892 afs_dentry_iput(struct dentry *dp, struct inode *ip)
894 struct vcache *vcp = VTOAFS(ip);
897 (void) afs_InactiveVCache(vcp, NULL);
904 afs_dentry_delete(struct dentry *dp)
906 if (dp->d_inode && (VTOAFS(dp->d_inode)->states & CUnlinked))
907 return 1; /* bad inode? */
912 struct dentry_operations afs_dentry_operations = {
913 .d_revalidate = afs_linux_dentry_revalidate,
914 .d_delete = afs_dentry_delete,
915 .d_iput = afs_dentry_iput,
918 /**********************************************************************
919 * AFS Linux inode operations
920 **********************************************************************/
924 * Merely need to set enough of vattr to get us through the create. Note
925 * that the higher level code (open_namei) will take care of any tuncation
926 * explicitly. Exclusive open is also taken care of in open_namei.
928 * name is in kernel space at this point.
931 #ifdef IOP_CREATE_TAKES_NAMEIDATA
932 afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
933 struct nameidata *nd)
935 afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
939 cred_t *credp = crref();
940 const char *name = dp->d_name.name;
945 vattr.va_mode = mode;
946 vattr.va_type = mode & S_IFMT;
948 #if defined(AFS_LINUX26_ENV)
952 code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
956 struct inode *ip = AFSTOV(vcp);
958 afs_getattr(vcp, &vattr, credp);
959 afs_fill_inode(ip, &vattr);
960 dp->d_op = &afs_dentry_operations;
961 dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
962 d_instantiate(dp, ip);
966 #if defined(AFS_LINUX26_ENV)
973 /* afs_linux_lookup */
974 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
975 static struct dentry *
976 #ifdef IOP_LOOKUP_TAKES_NAMEIDATA
977 afs_linux_lookup(struct inode *dip, struct dentry *dp,
978 struct nameidata *nd)
980 afs_linux_lookup(struct inode *dip, struct dentry *dp)
984 afs_linux_lookup(struct inode *dip, struct dentry *dp)
987 cred_t *credp = crref();
988 struct vcache *vcp = NULL;
989 const char *comp = dp->d_name.name;
990 struct inode *ip = NULL;
991 #if defined(AFS_LINUX26_ENV)
992 struct dentry *newdp = NULL;
996 #if defined(AFS_LINUX26_ENV)
1000 code = afs_lookup(VTOAFS(dip), comp, &vcp, credp);
1006 afs_getattr(vcp, &vattr, credp);
1007 afs_fill_inode(ip, &vattr);
1009 dp->d_op = &afs_dentry_operations;
1010 dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
1013 #if defined(AFS_LINUX24_ENV)
1014 if (ip && S_ISDIR(ip->i_mode)) {
1015 struct dentry *alias;
1017 /* Try to invalidate an existing alias in favor of our new one */
1018 alias = d_find_alias(ip);
1019 #if defined(AFS_LINUX26_ENV)
1020 /* But not if it's disconnected; then we want d_splice_alias below */
1021 if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
1025 if (d_invalidate(alias) == 0) {
1029 #if defined(AFS_LINUX26_ENV)
1038 #if defined(AFS_LINUX26_ENV)
1039 newdp = d_splice_alias(ip, dp);
1044 #if defined(AFS_LINUX26_ENV)
1049 /* It's ok for the file to not be found. That's noted by the caller by
1050 * seeing that the dp->d_inode field is NULL.
1052 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1053 #if defined(AFS_LINUX26_ENV)
1054 if (!code || code == ENOENT)
1061 return ERR_PTR(-code);
1070 afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
1073 cred_t *credp = crref();
1074 const char *name = newdp->d_name.name;
1075 struct inode *oldip = olddp->d_inode;
1077 /* If afs_link returned the vnode, we could instantiate the
1078 * dentry. Since it's not, we drop this one and do a new lookup.
1083 code = afs_link(VTOAFS(oldip), VTOAFS(dip), name, credp);
1091 afs_linux_unlink(struct inode *dip, struct dentry *dp)
1094 cred_t *credp = crref();
1095 const char *name = dp->d_name.name;
1096 struct vcache *tvc = VTOAFS(dp->d_inode);
1098 #if defined(AFS_LINUX26_ENV)
1101 if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
1102 && !(tvc->states & CUnlinked)) {
1103 struct dentry *__dp;
1105 extern char *afs_newname();
1114 osi_FreeSmallSpace(__name);
1115 __name = afs_newname();
1118 __dp = lookup_one_len(__name, dp->d_parent, strlen(__name));
1122 } while (__dp->d_inode != NULL);
1125 code = afs_rename(VTOAFS(dip), dp->d_name.name, VTOAFS(dip), __dp->d_name.name, credp);
1127 tvc->mvid = (void *) __name;
1130 crfree(tvc->uncred);
1132 tvc->uncred = credp;
1133 tvc->states |= CUnlinked;
1135 osi_FreeSmallSpace(__name);
1140 __dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
1149 code = afs_remove(VTOAFS(dip), name, credp);
1154 #if defined(AFS_LINUX26_ENV)
1163 afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
1166 cred_t *credp = crref();
1168 const char *name = dp->d_name.name;
1170 /* If afs_symlink returned the vnode, we could instantiate the
1171 * dentry. Since it's not, we drop this one and do a new lookup.
1177 code = afs_symlink(VTOAFS(dip), name, &vattr, target, credp);
1184 afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1187 cred_t *credp = crref();
1188 struct vcache *tvcp = NULL;
1190 const char *name = dp->d_name.name;
1192 #if defined(AFS_LINUX26_ENV)
1196 vattr.va_mask = ATTR_MODE;
1197 vattr.va_mode = mode;
1199 code = afs_mkdir(VTOAFS(dip), name, &vattr, &tvcp, credp);
1202 struct inode *ip = AFSTOV(tvcp);
1204 afs_getattr(tvcp, &vattr, credp);
1205 afs_fill_inode(ip, &vattr);
1207 dp->d_op = &afs_dentry_operations;
1208 dp->d_time = hgetlo(VTOAFS(dip)->m.DataVersion);
1209 d_instantiate(dp, ip);
1213 #if defined(AFS_LINUX26_ENV)
1221 afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1224 cred_t *credp = crref();
1225 const char *name = dp->d_name.name;
1227 /* locking kernel conflicts with glock? */
1230 code = afs_rmdir(VTOAFS(dip), name, credp);
1233 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1234 * that failed because a directory is not empty. So, we map
1235 * EEXIST to ENOTEMPTY on linux.
1237 if (code == EEXIST) {
1251 afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1252 struct inode *newip, struct dentry *newdp)
1255 cred_t *credp = crref();
1256 const char *oldname = olddp->d_name.name;
1257 const char *newname = newdp->d_name.name;
1258 struct dentry *rehash = NULL;
1260 #if defined(AFS_LINUX26_ENV)
1261 /* Prevent any new references during rename operation. */
1264 /* Remove old and new entries from name hash. New one will change below.
1265 * While it's optimal to catch failures and re-insert newdp into hash,
1266 * it's also error prone and in that case we're already dealing with error
1267 * cases. Let another lookup put things right, if need be.
1269 #if defined(AFS_LINUX26_ENV)
1270 if (!d_unhashed(newdp)) {
1275 if (!list_empty(&newdp->d_hash)) {
1281 #if defined(AFS_LINUX24_ENV)
1282 if (atomic_read(&olddp->d_count) > 1)
1283 shrink_dcache_parent(olddp);
1287 code = afs_rename(VTOAFS(oldip), oldname, VTOAFS(newip), newname, credp);
1293 #if defined(AFS_LINUX26_ENV)
1302 /* afs_linux_ireadlink
1303 * Internal readlink which can return link contents to user or kernel space.
1304 * Note that the buffer is NOT supposed to be null-terminated.
1307 afs_linux_ireadlink(struct inode *ip, char *target, int maxlen, uio_seg_t seg)
1310 cred_t *credp = crref();
1314 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1315 code = afs_readlink(VTOAFS(ip), &tuio, credp);
1319 return maxlen - tuio.uio_resid;
1324 #if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1325 /* afs_linux_readlink
1326 * Fill target (which is in user space) with contents of symlink.
1329 afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1332 struct inode *ip = dp->d_inode;
1335 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1341 /* afs_linux_follow_link
1342 * a file system dependent link following routine.
1344 #if defined(AFS_LINUX24_ENV)
1345 static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
1350 name = osi_Alloc(PATH_MAX);
1356 code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
1364 code = vfs_follow_link(nd, name);
1367 osi_Free(name, PATH_MAX);
1372 #else /* !defined(AFS_LINUX24_ENV) */
1374 static struct dentry *
1375 afs_linux_follow_link(struct dentry *dp, struct dentry *basep,
1376 unsigned int follow)
1384 name = osi_Alloc(PATH_MAX + 1);
1388 return ERR_PTR(-EIO);
1391 code = afs_linux_ireadlink(dp->d_inode, name, PATH_MAX, AFS_UIOSYS);
1396 res = ERR_PTR(code);
1399 res = lookup_dentry(name, basep, follow);
1403 osi_Free(name, PATH_MAX + 1);
1407 #endif /* AFS_LINUX24_ENV */
1408 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1410 /* afs_linux_readpage
1411 * all reads come through here. A strategy-like read call.
1414 afs_linux_readpage(struct file *fp, struct page *pp)
1417 cred_t *credp = crref();
1418 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1420 afs_offs_t offset = ((loff_t) pp->index) << PAGE_CACHE_SHIFT;
1422 ulong address = afs_linux_page_address(pp);
1423 afs_offs_t offset = pageoff(pp);
1427 struct inode *ip = FILE_INODE(fp);
1428 int cnt = page_count(pp);
1429 struct vcache *avc = VTOAFS(ip);
1432 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1436 atomic_add(1, &pp->count);
1437 set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
1438 clear_bit(PG_error, &pp->flags);
1441 setup_uio(&tuio, &iovec, (char *)address, offset, PAGE_SIZE, UIO_READ,
1443 #ifdef AFS_LINUX24_ENV
1447 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 */
1448 code = afs_rdwr(avc, &tuio, UIO_READ, 0, credp);
1449 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
1450 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
1453 #ifdef AFS_LINUX24_ENV
1458 if (tuio.uio_resid) /* zero remainder of page */
1459 memset((void *)(address + (PAGE_SIZE - tuio.uio_resid)), 0,
1461 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1462 flush_dcache_page(pp);
1463 SetPageUptodate(pp);
1465 set_bit(PG_uptodate, &pp->flags);
1469 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1473 clear_bit(PG_locked, &pp->flags);
1478 if (!code && AFS_CHUNKOFFSET(offset) == 0) {
1480 struct vrequest treq;
1483 code = afs_InitReq(&treq, credp);
1484 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
1485 tdc = afs_FindDCache(avc, offset);
1487 if (!(tdc->mflags & DFNextStarted))
1488 afs_PrefetchChunk(avc, tdc, credp, &treq);
1491 ReleaseWriteLock(&avc->lock);
1501 #if defined(AFS_LINUX24_ENV)
1503 afs_linux_writepage_sync(struct inode *ip, struct page *pp,
1504 unsigned long offset, unsigned int count)
1506 struct vcache *vcp = VTOAFS(ip);
1515 buffer = kmap(pp) + offset;
1516 base = (((loff_t) pp->index) << PAGE_CACHE_SHIFT) + offset;
1521 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1522 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1523 ICL_TYPE_INT32, 99999);
1525 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
1527 code = afs_write(vcp, &tuio, f_flags, credp, 0);
1529 ip->i_size = vcp->m.Length;
1530 ip->i_blocks = ((vcp->m.Length + 1023) >> 10) << 1;
1533 struct vrequest treq;
1535 ObtainWriteLock(&vcp->lock, 533);
1536 if (!afs_InitReq(&treq, credp))
1537 code = afs_DoPartialWrite(vcp, &treq);
1538 ReleaseWriteLock(&vcp->lock);
1540 code = code ? -code : count - tuio.uio_resid;
1542 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1543 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1544 ICL_TYPE_INT32, code);
1556 #ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
1557 afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
1559 afs_linux_writepage(struct page *pp)
1562 struct address_space *mapping = pp->mapping;
1563 struct inode *inode;
1564 unsigned long end_index;
1565 unsigned offset = PAGE_CACHE_SIZE;
1568 #if defined(AFS_LINUX26_ENV)
1569 if (PageReclaim(pp)) {
1570 # if defined(WRITEPAGE_ACTIVATE)
1571 return WRITEPAGE_ACTIVATE;
1573 return AOP_WRITEPAGE_ACTIVATE;
1577 if (PageLaunder(pp)) {
1578 return(fail_writepage(pp));
1582 inode = (struct inode *)mapping->host;
1583 end_index = inode->i_size >> PAGE_CACHE_SHIFT;
1586 if (pp->index < end_index)
1588 /* things got complicated... */
1589 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
1590 /* OK, are we completely out? */
1591 if (pp->index >= end_index + 1 || !offset)
1594 status = afs_linux_writepage_sync(inode, pp, 0, offset);
1595 SetPageUptodate(pp);
1597 if (status == offset)
1603 /* afs_linux_updatepage
1604 * What one would have thought was writepage - write dirty page to file.
1605 * Called from generic_file_write. buffer is still in user space. pagep
1606 * has been filled in with old data if we're updating less than a page.
1609 afs_linux_updatepage(struct file *fp, struct page *pp, unsigned long offset,
1610 unsigned int count, int sync)
1612 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
1613 u8 *page_addr = (u8 *) afs_linux_page_address(pp);
1619 set_bit(PG_locked, &pp->flags);
1623 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1624 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1625 ICL_TYPE_INT32, 99999);
1626 setup_uio(&tuio, &iovec, page_addr + offset,
1627 (afs_offs_t) (pageoff(pp) + offset), count, UIO_WRITE,
1630 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
1632 ip->i_size = vcp->m.Length;
1633 ip->i_blocks = ((vcp->m.Length + 1023) >> 10) << 1;
1636 struct vrequest treq;
1638 ObtainWriteLock(&vcp->lock, 533);
1639 vcp->m.Date = osi_Time(); /* set modification time */
1640 if (!afs_InitReq(&treq, credp))
1641 code = afs_DoPartialWrite(vcp, &treq);
1642 ReleaseWriteLock(&vcp->lock);
1645 code = code ? -code : count - tuio.uio_resid;
1646 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1647 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
1648 ICL_TYPE_INT32, code);
1653 clear_bit(PG_locked, &pp->flags);
1658 /* afs_linux_permission
1659 * Check access rights - returns error if can't check or permission denied.
1662 #ifdef IOP_PERMISSION_TAKES_NAMEIDATA
1663 afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
1665 afs_linux_permission(struct inode *ip, int mode)
1669 cred_t *credp = crref();
1673 if (mode & MAY_EXEC)
1675 if (mode & MAY_READ)
1677 if (mode & MAY_WRITE)
1679 code = afs_access(VTOAFS(ip), tmp, credp);
1686 #if defined(AFS_LINUX24_ENV)
1688 afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
1693 code = afs_linux_writepage_sync(file->f_dentry->d_inode, page,
1694 offset, to - offset);
1695 #if !defined(AFS_LINUX26_ENV)
1703 afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
1706 /* sometime between 2.4.0 and 2.4.19, the callers of prepare_write began to
1707 call kmap directly instead of relying on us to do it */
1708 #if !defined(AFS_LINUX26_ENV)
1714 extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
1717 static struct inode_operations afs_file_iops = {
1718 #if defined(AFS_LINUX26_ENV)
1719 .permission = afs_linux_permission,
1720 .getattr = afs_linux_getattr,
1721 .setattr = afs_notify_change,
1722 #elif defined(AFS_LINUX24_ENV)
1723 .permission = afs_linux_permission,
1724 .revalidate = afs_linux_revalidate,
1725 .setattr = afs_notify_change,
1727 .default_file_ops = &afs_file_fops,
1728 .readpage = afs_linux_readpage,
1729 .revalidate = afs_linux_revalidate,
1730 .updatepage = afs_linux_updatepage,
1734 #if defined(AFS_LINUX24_ENV)
1735 static struct address_space_operations afs_file_aops = {
1736 .readpage = afs_linux_readpage,
1737 .writepage = afs_linux_writepage,
1738 .commit_write = afs_linux_commit_write,
1739 .prepare_write = afs_linux_prepare_write,
1744 /* Separate ops vector for directories. Linux 2.2 tests type of inode
1745 * by what sort of operation is allowed.....
1748 static struct inode_operations afs_dir_iops = {
1749 #if !defined(AFS_LINUX24_ENV)
1750 .default_file_ops = &afs_dir_fops,
1752 .setattr = afs_notify_change,
1754 .create = afs_linux_create,
1755 .lookup = afs_linux_lookup,
1756 .link = afs_linux_link,
1757 .unlink = afs_linux_unlink,
1758 .symlink = afs_linux_symlink,
1759 .mkdir = afs_linux_mkdir,
1760 .rmdir = afs_linux_rmdir,
1761 .rename = afs_linux_rename,
1762 #if defined(AFS_LINUX26_ENV)
1763 .getattr = afs_linux_getattr,
1765 .revalidate = afs_linux_revalidate,
1767 .permission = afs_linux_permission,
1770 /* We really need a separate symlink set of ops, since do_follow_link()
1771 * determines if it _is_ a link by checking if the follow_link op is set.
1773 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1775 afs_symlink_filler(struct file *file, struct page *page)
1777 struct inode *ip = (struct inode *)page->mapping->host;
1778 char *p = (char *)kmap(page);
1783 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
1788 p[code] = '\0'; /* null terminate? */
1791 SetPageUptodate(page);
1805 static struct address_space_operations afs_symlink_aops = {
1806 .readpage = afs_symlink_filler
1808 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1810 static struct inode_operations afs_symlink_iops = {
1811 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1812 .readlink = page_readlink,
1813 #if defined(HAVE_KERNEL_PAGE_FOLLOW_LINK)
1814 .follow_link = page_follow_link,
1816 .follow_link = page_follow_link_light,
1817 .put_link = page_put_link,
1819 #else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
1820 .readlink = afs_linux_readlink,
1821 .follow_link = afs_linux_follow_link,
1822 #if !defined(AFS_LINUX24_ENV)
1823 .permission = afs_linux_permission,
1824 .revalidate = afs_linux_revalidate,
1826 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1827 #if defined(AFS_LINUX24_ENV)
1828 .setattr = afs_notify_change,
1833 afs_fill_inode(struct inode *ip, struct vattr *vattr)
1837 vattr2inode(ip, vattr);
1839 /* Reset ops if symlink or directory. */
1840 if (S_ISREG(ip->i_mode)) {
1841 ip->i_op = &afs_file_iops;
1842 #if defined(AFS_LINUX24_ENV)
1843 ip->i_fop = &afs_file_fops;
1844 ip->i_data.a_ops = &afs_file_aops;
1847 } else if (S_ISDIR(ip->i_mode)) {
1848 ip->i_op = &afs_dir_iops;
1849 #if defined(AFS_LINUX24_ENV)
1850 ip->i_fop = &afs_dir_fops;
1853 } else if (S_ISLNK(ip->i_mode)) {
1854 ip->i_op = &afs_symlink_iops;
1855 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1856 ip->i_data.a_ops = &afs_symlink_aops;
1857 ip->i_mapping = &ip->i_data;
1861 /* insert_inode_hash(ip); -- this would make iget() work (if we used it) */