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
12 * AFS vnodeops. The "NOTUSED" #define is used to indicate routines and
13 * calling sequences present in an ops table that we don't actually use.
14 * They are present solely for documentation purposes.
16 * So far the only truly scary part is that Linux relies on the inode cache
17 * to be up to date. Don't you dare break a callback and expect an fstat
18 * to give you meaningful information. This appears to be fixed in the 2.1
19 * development kernels. As it is we can fix this now by intercepting the
23 #include <afsconfig.h>
24 #include "../afs/param.h"
28 #include "../afs/sysincludes.h"
29 #include "../afs/afsincludes.h"
30 #include "../afs/afs_stats.h"
31 #include "../afs/afs_osidnlc.h"
33 #include "../h/pagemap.h"
34 #if defined(AFS_LINUX24_ENV)
35 #include "../h/smp_lock.h"
39 #define pageoff(pp) pgoff2loff((pp)->index)
41 #define pageoff(pp) pp->offset
44 extern struct vcache *afs_globalVp;
45 extern afs_rwlock_t afs_xvcache;
47 extern struct dentry_operations *afs_dops;
48 #if defined(AFS_LINUX24_ENV)
49 extern struct inode_operations afs_file_iops;
50 extern struct address_space_operations afs_file_aops;
51 struct address_space_operations afs_symlink_aops;
53 extern struct inode_operations afs_dir_iops;
54 extern struct inode_operations afs_symlink_iops;
58 static int afs_linux_lseek(struct inode *ip, struct file *fp, off_t, int) {}
61 static ssize_t afs_linux_read(struct file *fp, char *buf, size_t count,
65 struct vcache *vcp = ITOAFS(fp->f_dentry->d_inode);
66 cred_t *credp = crref();
70 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
71 ICL_TYPE_OFFSET, offp,
72 ICL_TYPE_INT32, count,
73 ICL_TYPE_INT32, 99999);
75 /* get a validated vcache entry */
76 code = afs_InitReq(&treq, credp);
78 code = afs_VerifyVCache(vcp, &treq);
83 #ifdef AFS_64BIT_CLIENT
84 if (*offp + count > afs_vmMappingEnd) {
87 afs_size_t oldOffset = *offp;
90 if (*offp < afs_vmMappingEnd) {
91 /* special case of a buffer crossing the VM mapping end */
92 afs_int32 tcount = afs_vmMappingEnd - *offp;
94 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
96 code = generic_file_read(fp, buf, tcount, offp);
103 setup_uio(&tuio, &iov, buf + xfered, (afs_offs_t) *offp, count,
104 UIO_READ, AFS_UIOSYS);
105 code = afs_read(vcp, &tuio, credp, 0, 0, 0);
106 xfered += count - tuio.uio_resid;
108 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
109 ICL_TYPE_OFFSET, offp,
111 ICL_TYPE_INT32, code);
113 *offp += count - tuio.uio_resid;
120 #endif /* AFS_64BIT_CLIENT */
121 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
123 code = generic_file_read(fp, buf, count, offp);
125 #ifdef AFS_64BIT_CLIENT
127 #endif /* AFS_64BIT_CLIENT */
130 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
131 ICL_TYPE_OFFSET, offp,
132 ICL_TYPE_INT32, count,
133 ICL_TYPE_INT32, code);
141 /* Now we have integrated VM for writes as well as reads. generic_file_write
142 * also takes care of re-positioning the pointer if file is open in append
143 * mode. Call fake open/close to ensure we do writes of core dumps.
145 static ssize_t afs_linux_write(struct file *fp, const char *buf, size_t count,
150 struct vcache *vcp = ITOAFS(fp->f_dentry->d_inode);
151 struct vrequest treq;
152 cred_t *credp = crref();
156 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
157 ICL_TYPE_OFFSET, offp,
158 ICL_TYPE_INT32, count,
159 ICL_TYPE_INT32, (fp->f_flags & O_APPEND) ? 99998 : 99999);
162 /* get a validated vcache entry */
163 code = (ssize_t)afs_InitReq(&treq, credp);
165 code = (ssize_t)afs_VerifyVCache(vcp, &treq);
167 ObtainWriteLock(&vcp->lock, 529);
169 ReleaseWriteLock(&vcp->lock);
173 #ifdef AFS_64BIT_CLIENT
174 if (*offp + count > afs_vmMappingEnd) {
177 afs_size_t oldOffset = *offp;
178 afs_int32 xfered = 0;
180 if (*offp < afs_vmMappingEnd) {
181 /* special case of a buffer crossing the VM mapping end */
182 afs_int32 tcount = afs_vmMappingEnd - *offp;
185 code = generic_file_write(fp, buf, tcount, offp);
187 if (code != tcount) {
192 setup_uio(&tuio, &iov, buf + xfered, (afs_offs_t) *offp, count,
193 UIO_WRITE, AFS_UIOSYS);
194 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
195 xfered += count - tuio.uio_resid;
198 *offp += count - tuio.uio_resid;
200 /* Purge dirty chunks of file if there are too many dirty chunks.
201 * Inside the write loop, we only do this at a chunk boundary.
202 * Clean up partial chunk if necessary at end of loop.
204 if (AFS_CHUNKBASE(tuio.afsio_offset) != AFS_CHUNKBASE(oldOffset)) {
205 ObtainWriteLock(&vcp->lock,402);
206 code = afs_DoPartialWrite(vcp, &treq);
207 vcp->states |= CDirty;
208 ReleaseWriteLock(&vcp->lock);
212 ObtainWriteLock(&vcp->lock,400);
213 vcp->m.Date = osi_Time(); /* Set file date (for ranlib) */
215 if (*offp > vcp->m.Length) {
216 vcp->m.Length = *offp;
218 ReleaseWriteLock(&vcp->lock);
222 #endif /* AFS_64BIT_CLIENT */
224 code = generic_file_write(fp, buf, count, offp);
226 #ifdef AFS_64BIT_CLIENT
228 #endif /* AFS_64BIT_CLIENT */
231 ObtainWriteLock(&vcp->lock, 530);
232 vcp->m.Date = osi_Time(); /* set modification time */
233 afs_FakeClose(vcp, credp);
235 code2 = afs_DoPartialWrite(vcp, &treq);
236 if (code2 && code >=0)
237 code = (ssize_t) -code2;
238 ReleaseWriteLock(&vcp->lock);
240 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
241 ICL_TYPE_OFFSET, offp,
242 ICL_TYPE_INT32, count,
243 ICL_TYPE_INT32, code);
250 /* This is a complete rewrite of afs_readdir, since we can make use of
251 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
252 * handling and use of bulkstats will need to be reflected here as well.
254 static int afs_linux_readdir(struct file *fp,
255 void *dirbuf, filldir_t filldir)
257 extern struct DirEntry * afs_dir_GetBlob();
258 struct vcache *avc = ITOAFS(FILE_INODE(fp));
259 struct vrequest treq;
260 register struct dcache *tdc;
267 afs_size_t origOffset, tlen;
268 cred_t *credp = crref();
269 struct afs_fakestat_state fakestat;
272 AFS_STATCNT(afs_readdir);
274 code = afs_InitReq(&treq, credp);
281 afs_InitFakeStat(&fakestat);
282 code = afs_EvalFakeStat(&avc, &fakestat, &treq);
284 afs_PutFakeStat(&fakestat);
289 /* update the cache entry */
291 code = afs_VerifyVCache(avc, &treq);
293 afs_PutFakeStat(&fakestat);
298 /* get a reference to the entire directory */
299 tdc = afs_GetDCache(avc, (afs_size_t) 0, &treq, &origOffset, &tlen, 1);
302 afs_PutFakeStat(&fakestat);
306 ObtainReadLock(&avc->lock);
307 ObtainReadLock(&tdc->lock);
309 * Make sure that the data in the cache is current. There are two
310 * cases we need to worry about:
311 * 1. The cache data is being fetched by another process.
312 * 2. The cache data is no longer valid
314 while ((avc->states & CStatd)
315 && (tdc->dflags & DFFetching)
316 && hsame(avc->m.DataVersion, tdc->f.versionNo)) {
317 ReleaseReadLock(&tdc->lock);
318 ReleaseReadLock(&avc->lock);
319 afs_osi_Sleep(&tdc->validPos);
320 ObtainReadLock(&avc->lock);
321 ObtainReadLock(&tdc->lock);
323 if (!(avc->states & CStatd)
324 || !hsame(avc->m.DataVersion, tdc->f.versionNo)) {
325 ReleaseReadLock(&tdc->lock);
326 ReleaseReadLock(&avc->lock);
331 /* Fill in until we get an error or we're done. This implementation
332 * takes an offset in units of blobs, rather than bytes.
335 offset = (int)fp->f_pos;
337 dirpos = BlobScan(&tdc->f.inode, offset);
341 de = afs_dir_GetBlob(&tdc->f.inode, dirpos);
345 ino = (avc->fid.Fid.Volume << 16) + ntohl(de->fid.vnode);
346 ino &= 0x7fffffff; /* Assumes 32 bit ino_t ..... */
347 len = strlen(de->name);
349 /* filldir returns -EINVAL when the buffer is full. */
350 #if (defined(AFS_LINUX24_ENV) || defined(pgoff2loff)) && defined(DECLARE_FSTYPE)
352 unsigned int type=DT_UNKNOWN;
353 struct VenusFid afid;
356 afid.Cell=avc->fid.Cell;
357 afid.Fid.Volume=avc->fid.Fid.Volume;
358 afid.Fid.Vnode=ntohl(de->fid.vnode);
359 afid.Fid.Unique=ntohl(de->fid.vunique);
360 if ((avc->states & CForeign) == 0 &&
361 (ntohl(de->fid.vnode) & 1)) {
363 } else if ((tvc=afs_FindVCache(&afid,0,0))) {
366 } else if (((tvc->states) & (CStatd|CTruth))) {
367 /* CTruth will be set if the object has
372 else if (vtype == VREG)
374 /* Don't do this until we're sure it can't be a mtpt */
375 /* else if (vtype == VLNK)
377 /* what other types does AFS support? */
379 /* clean up from afs_FindVCache */
382 code = (*filldir)(dirbuf, de->name, len, offset, ino, type);
385 code = (*filldir)(dirbuf, de->name, len, offset, ino);
390 offset = dirpos + 1 + ((len+16)>>5);
392 /* If filldir didn't fill in the last one this is still pointing to that
395 fp->f_pos = (loff_t)offset;
397 ReleaseReadLock(&tdc->lock);
399 ReleaseReadLock(&avc->lock);
400 afs_PutFakeStat(&fakestat);
406 int afs_linux_select(struct inode *ip, struct file *fp, int, select_table *);
409 /* in afs_pioctl.c */
410 extern int afs_xioctl(struct inode *ip, struct file *fp,
411 unsigned int com, unsigned long arg);
414 /* We need to detect unmap's after close. To do that, we need our own
415 * vm_operations_struct's. And we need to set them up for both the
416 * private and shared mappings. The fun part is that these are all static
417 * so we'll have to initialize on the fly!
419 static struct vm_operations_struct afs_private_mmap_ops;
420 static int afs_private_mmap_ops_inited = 0;
421 static struct vm_operations_struct afs_shared_mmap_ops;
422 static int afs_shared_mmap_ops_inited = 0;
424 void afs_linux_vma_close(struct vm_area_struct *vmap)
432 vcp = ITOAFS(FILE_INODE(vmap->vm_file));
437 afs_Trace4(afs_iclSetp, CM_TRACE_VM_CLOSE,
438 ICL_TYPE_POINTER, vcp,
439 ICL_TYPE_INT32, vcp->mapcnt,
440 ICL_TYPE_INT32, vcp->opens,
441 ICL_TYPE_INT32, vcp->execsOrWriters);
442 ObtainWriteLock(&vcp->lock, 532);
445 ReleaseWriteLock(&vcp->lock);
448 (void) afs_close(vcp, vmap->vm_file->f_flags, credp);
449 /* only decrement the execsOrWriters flag if this is not a writable
451 if (! (vmap->vm_file->f_flags & (FWRITE | FTRUNC)))
452 vcp->execsOrWriters--;
454 vcp->states &= ~CMAPPED;
459 ReleaseWriteLock(&vcp->lock);
466 static int afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
468 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
469 cred_t *credp = crref();
470 struct vrequest treq;
474 #if defined(AFS_LINUX24_ENV)
475 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
476 ICL_TYPE_POINTER, vmap->vm_start,
477 ICL_TYPE_INT32, vmap->vm_end - vmap->vm_start);
479 afs_Trace4(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
480 ICL_TYPE_POINTER, vmap->vm_start,
481 ICL_TYPE_INT32, vmap->vm_end - vmap->vm_start,
482 ICL_TYPE_INT32, vmap->vm_offset);
485 /* get a validated vcache entry */
486 code = afs_InitReq(&treq, credp);
488 code = afs_VerifyVCache(vcp, &treq);
494 osi_FlushPages(vcp, credp); /* ensure stale pages are gone */
497 code = generic_file_mmap(fp, vmap);
502 ObtainWriteLock(&vcp->lock,531);
503 /* Set out vma ops so we catch the close. The following test should be
504 * the same as used in generic_file_mmap.
506 if ((vmap->vm_flags & VM_SHARED) && (vmap->vm_flags & VM_MAYWRITE)) {
507 if (!afs_shared_mmap_ops_inited) {
508 afs_shared_mmap_ops_inited = 1;
509 afs_shared_mmap_ops = *vmap->vm_ops;
510 afs_shared_mmap_ops.close = afs_linux_vma_close;
512 vmap->vm_ops = &afs_shared_mmap_ops;
515 if (!afs_private_mmap_ops_inited) {
516 afs_private_mmap_ops_inited = 1;
517 afs_private_mmap_ops = *vmap->vm_ops;
518 afs_private_mmap_ops.close = afs_linux_vma_close;
520 vmap->vm_ops = &afs_private_mmap_ops;
524 /* Add an open reference on the first mapping. */
525 if (vcp->mapcnt == 0) {
526 vcp->execsOrWriters++;
528 vcp->states |= CMAPPED;
530 ReleaseWriteLock(&vcp->lock);
539 int afs_linux_open(struct inode *ip, struct file *fp)
542 cred_t *credp = crref();
545 #ifdef AFS_LINUX24_ENV
548 code = afs_open((struct vcache**)&ip, fp->f_flags, credp);
549 #ifdef AFS_LINUX24_ENV
558 /* afs_Close is called from release, since release is used to handle all
559 * file closings. In addition afs_linux_flush is called from sys_close to
560 * handle flushing the data back to the server. The kicker is that we could
561 * ignore flush completely if only sys_close took it's return value from
562 * fput. See afs_linux_flush for notes on interactions between release and
565 static int afs_linux_release(struct inode *ip, struct file *fp)
568 cred_t *credp = crref();
569 struct vcache *vcp = ITOAFS(ip);
572 #ifdef AFS_LINUX24_ENV
576 vcp->flushcnt--; /* protected by AFS global lock. */
579 code = afs_close(vcp, fp->f_flags, credp);
581 #ifdef AFS_LINUX24_ENV
590 #if defined(AFS_LINUX24_ENV)
591 static int afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
593 static int afs_linux_fsync(struct file *fp, struct dentry *dp)
597 struct inode *ip = FILE_INODE(fp);
598 cred_t *credp = crref();
601 #ifdef AFS_LINUX24_ENV
604 code = afs_fsync(ITOAFS(ip), credp);
605 #ifdef AFS_LINUX24_ENV
615 /* No support for async i/o */
616 int afs_linux_fasync(struct inode *ip, struct file *fp, int);
618 /* I don't think it will, at least not as can be detected here. */
619 int afs_linux_check_media_change(kdev_t dev);
621 /* Revalidate media and file system. */
622 int afs_linux_file_revalidate(kdev_t dev);
625 static int afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
628 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
629 cred_t *credp = crref();
630 #ifdef AFS_LINUX24_ENV
631 struct flock64 flock;
636 /* Convert to a lock format afs_lockctl understands. */
637 memset((char*)&flock, 0, sizeof(flock));
638 flock.l_type = flp->fl_type;
639 flock.l_pid = flp->fl_pid;
641 flock.l_start = flp->fl_start;
642 flock.l_len = flp->fl_end - flp->fl_start;
645 code = afs_lockctl(vcp, &flock, cmd, credp);
648 /* Convert flock back to Linux's file_lock */
649 flp->fl_type = flock.l_type;
650 flp->fl_pid = flock.l_pid;
651 flp->fl_start = flock.l_start;
652 flp->fl_end = flock.l_start + flock.l_len;
660 * flush is called from sys_close. We could ignore it, but sys_close return
661 * code comes from flush, not release. We need to use release to keep
662 * the vcache open count correct. Note that flush is called before release
663 * (via fput) in sys_close. vcp->flushcnt is a bit of ugliness to avoid
664 * races and also avoid calling afs_close twice when closing the file.
665 * If we merely checked for opens > 0 in afs_linux_release, then if an
666 * new open occurred when storing back the file, afs_linux_release would
667 * incorrectly close the file and decrement the opens count. Calling afs_close
668 * on the just flushed file is wasteful, since the background daemon will
669 * execute the code that finally decides there is nothing to do.
671 int afs_linux_flush(struct file *fp)
673 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
677 /* Only do this on the last close of the file pointer. */
678 #if defined(AFS_LINUX24_ENV)
679 if (atomic_read(&fp->f_count) > 1)
688 code = afs_close(vcp, fp->f_flags, credp);
689 vcp->flushcnt++; /* protected by AFS global lock. */
696 /* Not allowed to directly read a directory. */
697 ssize_t afs_linux_dir_read(struct file *fp, char *buf, size_t count, loff_t *ppos)
704 #if defined(AFS_LINUX24_ENV)
705 struct file_operations afs_dir_fops = {
706 read: generic_read_dir,
707 readdir: afs_linux_readdir,
709 open: afs_linux_open,
710 release: afs_linux_release,
713 struct file_operations afs_dir_fops = {
714 NULL, /* afs_linux_lseek */
716 NULL, /* afs_linux_write */
718 NULL, /* afs_linux_select */
719 afs_xioctl, /* close enough to use the ported AFS one */
720 NULL, /* afs_linux_mmap */
722 NULL, /* afs_linux_flush */
725 NULL, /* afs_linux_fasync */
726 NULL, /* afs_linux_check_media_change */
727 NULL, /* afs_linux_file_revalidate */
732 #if defined(AFS_LINUX24_ENV)
733 struct file_operations afs_file_fops = {
734 read: afs_linux_read,
735 write: afs_linux_write,
737 mmap: afs_linux_mmap,
738 open: afs_linux_open,
739 flush: afs_linux_flush,
740 release: afs_linux_release,
741 fsync: afs_linux_fsync,
742 lock: afs_linux_lock,
745 struct file_operations afs_file_fops = {
746 NULL, /* afs_linux_lseek */
749 NULL, /* afs_linux_readdir */
750 NULL, /* afs_linux_select */
751 afs_xioctl, /* close enough to use the ported AFS one */
757 NULL, /* afs_linux_fasync */
758 NULL, /* afs_linux_check_media_change */
759 NULL, /* afs_linux_file_revalidate */
765 /**********************************************************************
766 * AFS Linux dentry operations
767 **********************************************************************/
769 /* afs_linux_revalidate
770 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
772 static int afs_linux_revalidate(struct dentry *dp)
776 struct vrequest treq;
777 struct vcache *vcp = ITOAFS(dp->d_inode);
778 struct vcache *rootvp = NULL;
782 if (afs_fakestat_enable && vcp->mvstat == 1 && vcp->mvid &&
783 (vcp->states & CMValid) && (vcp->states & CStatd)) {
784 ObtainSharedLock(&afs_xvcache, 680);
785 rootvp = afs_FindVCache(vcp->mvid, 0, 0);
786 ReleaseSharedLock(&afs_xvcache);
789 #ifdef AFS_LINUX24_ENV
793 /* Make this a fast path (no crref), since it's called so often. */
794 if (vcp->states & CStatd) {
795 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
796 check_bad_parent(dp); /* check and correct mvid */
798 vcache2fakeinode(rootvp, vcp);
801 #ifdef AFS_LINUX24_ENV
804 if (rootvp) afs_PutVCache(rootvp);
810 code = afs_InitReq(&treq, credp);
812 code = afs_VerifyVCache(vcp, &treq);
814 #ifdef AFS_LINUX24_ENV
824 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
825 * In kernels 2.2.10 and above, we are passed an additional flags var which
826 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
827 * we are advised to follow the entry if it is a link or to make sure that
828 * it is a directory. But since the kernel itself checks these possibilities
829 * later on, we shouldn't have to do it until later. Perhaps in the future..
831 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
832 static int afs_linux_dentry_revalidate(struct dentry *dp, int flags)
834 static int afs_linux_dentry_revalidate(struct dentry *dp)
838 cred_t *credp = crref();
839 struct vrequest treq;
840 struct vcache *lookupvcp = NULL;
841 int code, bad_dentry = 1;
842 struct sysname_info sysState;
843 struct vcache *vcp = ITOAFS(dp->d_inode);
844 struct vcache *parentvcp = ITOAFS(dp->d_parent->d_inode);
849 sysState.allocked = 0;
851 /* If it's a negative dentry, then there's nothing to do. */
852 if (!vcp || !parentvcp)
855 /* If it is the AFS root, then there's no chance it needs
857 if (vcp == afs_globalVp) {
862 if ((code = afs_InitReq(&treq, credp)))
865 Check_AtSys(parentvcp, dp->d_name.name, &sysState, &treq);
866 name = sysState.name;
868 /* First try looking up the DNLC */
869 if ((lookupvcp = osi_dnlc_lookup(parentvcp, name, WRITE_LOCK))) {
870 /* Verify that the dentry does not point to an old inode */
871 if (vcp != lookupvcp)
873 /* Check and correct mvid */
874 if (*name != '/' && vcp->mvstat == 2)
875 check_bad_parent(dp);
881 /* A DNLC lookup failure cannot be trusted. Try a real lookup */
882 code = afs_lookup(parentvcp, name, &lookupvcp, credp);
884 /* Verify that the dentry does not point to an old inode */
885 if (vcp != lookupvcp)
893 afs_PutVCache(lookupvcp);
894 if (sysState.allocked)
895 osi_FreeLargeSpace(name);
901 shrink_dcache_parent(dp);
910 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
911 static int afs_linux_dentry_revalidate(struct dentry *dp, int flags)
913 static int afs_linux_dentry_revalidate(struct dentry *dp)
918 struct vrequest treq;
919 struct inode *ip = AFSTOI(dp->d_inode);
921 unsigned long timeout = 3*HZ; /* 3 seconds */
924 printk("negative dentry: %s\n", dp->d_name.name);
926 if (!(flags & LOOKUP_CONTINUE)) {
927 long diff = CURRENT_TIME - dp->d_parent->d_inode->i_mtime;
933 if (time_after(jiffies, dp->d_time + timeout))
944 /* afs_dentry_iput */
945 static void afs_dentry_iput(struct dentry *dp, struct inode *ip)
947 if (ICL_SETACTIVE(afs_iclSetp)) {
949 afs_Trace3(afs_iclSetp, CM_TRACE_DENTRYIPUT,
950 ICL_TYPE_POINTER, ip,
951 ICL_TYPE_STRING, dp->d_parent->d_name.name,
952 ICL_TYPE_STRING, dp->d_name.name);
959 static int afs_dentry_delete(struct dentry *dp)
961 if (ICL_SETACTIVE(afs_iclSetp)) {
963 afs_Trace3(afs_iclSetp, CM_TRACE_DENTRYDELETE, ICL_TYPE_POINTER,
964 dp->d_inode, ICL_TYPE_STRING, dp->d_parent->d_name.name,
965 ICL_TYPE_STRING, dp->d_name.name);
969 if (dp->d_inode && (ITOAFS(dp->d_inode)->states & CUnlinked))
970 return 1; /* bad inode? */
975 #if defined(AFS_LINUX24_ENV)
976 struct dentry_operations afs_dentry_operations = {
977 d_revalidate: afs_linux_dentry_revalidate,
978 d_iput: afs_dentry_iput,
979 d_delete: afs_dentry_delete,
981 struct dentry_operations *afs_dops = &afs_dentry_operations;
983 struct dentry_operations afs_dentry_operations = {
984 afs_linux_dentry_revalidate, /* d_validate(struct dentry *) */
986 NULL, /* d_compare */
987 afs_dentry_delete, /* d_delete(struct dentry *) */
988 NULL, /* d_release(struct dentry *) */
989 afs_dentry_iput /* d_iput(struct dentry *, struct inode *) */
991 struct dentry_operations *afs_dops = &afs_dentry_operations;
994 /**********************************************************************
995 * AFS Linux inode operations
996 **********************************************************************/
1000 * Merely need to set enough of vattr to get us through the create. Note
1001 * that the higher level code (open_namei) will take care of any tuncation
1002 * explicitly. Exclusive open is also taken care of in open_namei.
1004 * name is in kernel space at this point.
1006 int afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
1009 cred_t *credp = crref();
1012 const char *name = dp->d_name.name;
1016 vattr.va_mode = mode;
1019 code = afs_create(ITOAFS(dip), name, &vattr, NONEXCL, mode,
1020 (struct vcache**)&ip, credp);
1023 vattr2inode(ip, &vattr);
1024 /* Reset ops if symlink or directory. */
1025 #if defined(AFS_LINUX24_ENV)
1026 if (S_ISREG(ip->i_mode)) {
1027 ip->i_op = &afs_file_iops;
1028 ip->i_fop = &afs_file_fops;
1029 ip->i_data.a_ops = &afs_file_aops;
1030 } else if (S_ISDIR(ip->i_mode)) {
1031 ip->i_op = &afs_dir_iops;
1032 ip->i_fop = &afs_dir_fops;
1033 } else if (S_ISLNK(ip->i_mode)) {
1034 ip->i_op = &afs_symlink_iops;
1035 ip->i_data.a_ops = &afs_symlink_aops;
1036 ip->i_mapping = &ip->i_data;
1038 printk("afs_linux_create: FIXME\n");
1040 if (S_ISDIR(ip->i_mode))
1041 ip->i_op = &afs_dir_iops;
1042 else if (S_ISLNK(ip->i_mode))
1043 ip->i_op = &afs_symlink_iops;
1046 dp->d_op = afs_dops;
1047 dp->d_time = jiffies;
1048 d_instantiate(dp, ip);
1056 /* afs_linux_lookup */
1057 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1058 struct dentry *afs_linux_lookup(struct inode *dip, struct dentry *dp)
1060 int afs_linux_lookup(struct inode *dip, struct dentry *dp)
1064 cred_t *credp = crref();
1065 struct vcache *vcp=NULL;
1066 const char *comp = dp->d_name.name;
1068 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;
1086 printk("afs_linux_lookup: ip->i_mode 0x%x dp->d_name.name %s code %d\n", ip->i_mode, dp->d_name.name, code);
1088 if (S_ISDIR(ip->i_mode))
1089 ip->i_op = &afs_dir_iops;
1090 else if (S_ISLNK(ip->i_mode))
1091 ip->i_op = &afs_symlink_iops;
1094 dp->d_time = jiffies;
1095 dp->d_op = afs_dops;
1096 d_add(dp, AFSTOI(vcp));
1101 /* It's ok for the file to not be found. That's noted by the caller by
1102 * seeing that the dp->d_inode field is NULL.
1104 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1108 return ERR_PTR(-code);
1116 int afs_linux_link(struct dentry *olddp, struct inode *dip,
1117 struct dentry *newdp)
1120 cred_t *credp = crref();
1121 const char *name = newdp->d_name.name;
1122 struct inode *oldip = olddp->d_inode;
1124 /* If afs_link returned the vnode, we could instantiate the
1125 * dentry. Since it's not, we drop this one and do a new lookup.
1130 code = afs_link(ITOAFS(oldip), ITOAFS(dip), name, credp);
1137 int afs_linux_unlink(struct inode *dip, struct dentry *dp)
1140 cred_t *credp = crref();
1141 const char *name = dp->d_name.name;
1144 code = afs_remove(ITOAFS(dip), name, credp);
1153 int afs_linux_symlink(struct inode *dip, struct dentry *dp,
1157 cred_t *credp = crref();
1159 const char *name = dp->d_name.name;
1161 /* If afs_symlink returned the vnode, we could instantiate the
1162 * dentry. Since it's not, we drop this one and do a new lookup.
1168 code = afs_symlink(ITOAFS(dip), name, &vattr, target, credp);
1174 int afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1177 cred_t *credp = crref();
1178 struct vcache *tvcp = NULL;
1180 const char *name = dp->d_name.name;
1184 vattr.va_mask = ATTR_MODE;
1185 vattr.va_mode = mode;
1186 code = afs_mkdir(ITOAFS(dip), name, &vattr, &tvcp, credp);
1189 tvcp->v.v_op = &afs_dir_iops;
1190 #if defined(AFS_LINUX24_ENV)
1191 tvcp->v.v_fop = &afs_dir_fops;
1193 dp->d_op = afs_dops;
1194 dp->d_time = jiffies;
1195 d_instantiate(dp, AFSTOI(tvcp));
1203 int afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1206 cred_t *credp = crref();
1207 const char *name = dp->d_name.name;
1210 code = afs_rmdir(ITOAFS(dip), name, credp);
1212 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1213 * that failed because a directory is not empty. So, we map
1214 * EEXIST to ENOTEMPTY on linux.
1216 if (code == EEXIST) {
1231 int afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1232 struct inode *newip, struct dentry *newdp)
1235 cred_t *credp = crref();
1236 const char *oldname = olddp->d_name.name;
1237 const char *newname = newdp->d_name.name;
1239 /* Remove old and new entries from name hash. New one will change below.
1240 * While it's optimal to catch failures and re-insert newdp into hash,
1241 * it's also error prone and in that case we're already dealing with error
1242 * cases. Let another lookup put things right, if need be.
1244 if (!list_empty(&olddp->d_hash)) {
1247 if (!list_empty(&newdp->d_hash)) {
1251 code = afs_rename(ITOAFS(oldip), oldname, ITOAFS(newip),
1256 /* update time so it doesn't expire immediately */
1257 newdp->d_time = jiffies;
1258 d_move(olddp, newdp);
1266 /* afs_linux_ireadlink
1267 * Internal readlink which can return link contents to user or kernel space.
1268 * Note that the buffer is NOT supposed to be null-terminated.
1270 static int afs_linux_ireadlink(struct inode *ip, char *target, int maxlen,
1274 cred_t *credp = crref();
1278 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1279 code = afs_readlink(ITOAFS(ip), &tuio, credp);
1283 return maxlen - tuio.uio_resid;
1288 #if !defined(AFS_LINUX24_ENV)
1289 /* afs_linux_readlink
1290 * Fill target (which is in user space) with contents of symlink.
1292 int afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1295 struct inode *ip = dp->d_inode;
1298 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1304 /* afs_linux_follow_link
1305 * a file system dependent link following routine.
1307 struct dentry * afs_linux_follow_link(struct dentry *dp,
1308 struct dentry *basep,
1309 unsigned int follow)
1317 name = osi_Alloc(PATH_MAX+1);
1321 return ERR_PTR(-EIO);
1324 code = afs_linux_ireadlink(dp->d_inode, name, PATH_MAX, AFS_UIOSYS);
1329 res = ERR_PTR(code);
1333 res = lookup_dentry(name, basep, follow);
1337 osi_Free(name, PATH_MAX+1);
1343 /* afs_linux_readpage
1344 * all reads come through here. A strategy-like read call.
1346 int afs_linux_readpage(struct file *fp, struct page *pp)
1349 cred_t *credp = crref();
1350 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1352 afs_offs_t offset = pp->index << PAGE_CACHE_SHIFT;
1354 ulong address = afs_linux_page_address(pp);
1355 afs_offs_t offset = pageoff(pp);
1359 struct inode *ip = FILE_INODE(fp);
1360 int cnt = atomic_read(&pp->count);
1361 struct vcache *avc = ITOAFS(ip);
1364 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE,
1365 ICL_TYPE_POINTER, ip,
1366 ICL_TYPE_POINTER, pp,
1367 ICL_TYPE_INT32, cnt,
1368 ICL_TYPE_INT32, 99999); /* not a possible code value */
1369 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1375 atomic_add(1, &pp->count);
1376 set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
1377 clear_bit(PG_error, &pp->flags);
1380 setup_uio(&tuio, &iovec, (char*)address, offset, PAGESIZE,
1381 UIO_READ, AFS_UIOSYS);
1382 code = afs_rdwr(avc, &tuio, UIO_READ, 0, credp);
1383 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1388 if (tuio.uio_resid) /* zero remainder of page */
1389 memset((void*)(address+(PAGESIZE-tuio.uio_resid)), 0,
1391 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1392 flush_dcache_page(pp);
1393 SetPageUptodate(pp);
1395 set_bit(PG_uptodate, &pp->flags);
1399 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1403 clear_bit(PG_locked, &pp->flags);
1408 if (!code && AFS_CHUNKOFFSET(offset) == 0) {
1410 struct vrequest treq;
1412 code = afs_InitReq(&treq, credp);
1413 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
1414 tdc = afs_FindDCache(avc, offset);
1416 if (!(tdc->mflags & DFNextStarted))
1417 afs_PrefetchChunk(avc, tdc, credp, &treq);
1420 ReleaseWriteLock(&avc->lock);
1425 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE,
1426 ICL_TYPE_POINTER, ip,
1427 ICL_TYPE_POINTER, pp,
1428 ICL_TYPE_INT32, cnt,
1429 ICL_TYPE_INT32, code);
1434 #if defined(AFS_LINUX24_ENV)
1435 int afs_linux_writepage(struct page *pp)
1437 struct address_space *mapping = pp->mapping;
1438 struct inode *inode;
1439 unsigned long end_index;
1440 unsigned offset = PAGE_CACHE_SIZE;
1443 inode = (struct inode *) mapping->host;
1444 end_index = inode->i_size >> PAGE_CACHE_SHIFT;
1447 if (pp->index < end_index)
1449 /* things got complicated... */
1450 offset = inode->i_size & (PAGE_CACHE_SIZE-1);
1451 /* OK, are we completely out? */
1452 if (pp->index >= end_index+1 || !offset)
1456 status = afs_linux_writepage_sync(inode, pp, 0, offset);
1458 SetPageUptodate(pp);
1460 if (status == offset)
1468 /* afs_linux_bmap - supports generic_readpage, but we roll our own. */
1469 int afs_linux_bmap(struct inode *ip, int) { return -EINVAL; }
1471 /* afs_linux_truncate
1472 * Handles discarding disk blocks if this were a device. ext2 indicates we
1473 * may need to zero partial last pages of memory mapped files.
1475 void afs_linux_truncate(struct inode *ip)
1480 /* afs_linux_permission
1481 * Check access rights - returns error if can't check or permission denied.
1483 int afs_linux_permission(struct inode *ip, int mode)
1486 cred_t *credp = crref();
1490 if (mode & MAY_EXEC) tmp |= VEXEC;
1491 if (mode & MAY_READ) tmp |= VREAD;
1492 if (mode & MAY_WRITE) tmp |= VWRITE;
1493 code = afs_access(ITOAFS(ip), tmp, credp);
1502 /* msdos sector mapping hack for memory mapping. */
1503 int afs_linux_smap(struct inode *ip, int) { return -EINVAL; }
1506 #if defined(AFS_LINUX24_ENV)
1507 int afs_linux_writepage_sync(struct inode *ip, struct page *pp,
1508 unsigned long offset,
1511 struct vcache *vcp = ITOAFS(ip);
1520 buffer = kmap(pp) + offset;
1521 base = (pp->index << PAGE_CACHE_SHIFT) + offset;
1524 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1525 ICL_TYPE_POINTER, pp,
1526 ICL_TYPE_INT32, atomic_read(&pp->count),
1527 ICL_TYPE_INT32, 99999);
1529 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
1531 code = afs_write(vcp, &tuio, f_flags, credp, 0);
1535 if (!code && afs_stats_cmperf.cacheCurrDirtyChunks >
1536 afs_stats_cmperf.cacheMaxDirtyChunks) {
1537 struct vrequest treq;
1539 ObtainWriteLock(&vcp->lock, 533);
1540 if (!afs_InitReq(&treq, credp))
1541 code = afs_DoPartialWrite(vcp, &treq);
1542 ReleaseWriteLock(&vcp->lock);
1544 code = code ? -code : count - tuio.uio_resid;
1546 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1547 ICL_TYPE_POINTER, pp,
1548 ICL_TYPE_INT32, atomic_read(&pp->count),
1549 ICL_TYPE_INT32, code);
1558 afs_linux_updatepage(struct file *file, struct page *page,
1559 unsigned long offset, unsigned int count)
1561 struct dentry *dentry = file->f_dentry;
1563 return afs_linux_writepage_sync(dentry->d_inode, page, offset, count);
1566 /* afs_linux_updatepage
1567 * What one would have thought was writepage - write dirty page to file.
1568 * Called from generic_file_write. buffer is still in user space. pagep
1569 * has been filled in with old data if we're updating less than a page.
1571 int afs_linux_updatepage(struct file *fp, struct page *pp,
1572 unsigned long offset,
1573 unsigned int count, int sync)
1575 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
1576 u8 *page_addr = (u8*) afs_linux_page_address(pp);
1582 set_bit(PG_locked, &pp->flags);
1586 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1587 ICL_TYPE_POINTER, pp,
1588 ICL_TYPE_INT32, atomic_read(&pp->count),
1589 ICL_TYPE_INT32, 99999);
1590 setup_uio(&tuio, &iovec, page_addr + offset, (afs_offs_t)(pageoff(pp) + offset),
1591 count, UIO_WRITE, AFS_UIOSYS);
1593 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
1597 code = code ? -code : count - tuio.uio_resid;
1598 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1599 ICL_TYPE_POINTER, pp,
1600 ICL_TYPE_INT32, atomic_read(&pp->count),
1601 ICL_TYPE_INT32, code);
1606 clear_bit(PG_locked, &pp->flags);
1611 #if defined(AFS_LINUX24_ENV)
1612 static int afs_linux_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1618 code = afs_linux_updatepage(file, page, offset, to-offset);
1626 static int afs_linux_prepare_write(struct file *file, struct page *page,
1627 unsigned from, unsigned to)
1633 extern int afs_notify_change(struct dentry *dp, struct iattr* iattrp);
1636 #if defined(AFS_LINUX24_ENV)
1637 struct inode_operations afs_file_iops = {
1638 revalidate: afs_linux_revalidate,
1639 setattr: afs_notify_change,
1640 permission: afs_linux_permission,
1642 struct address_space_operations afs_file_aops = {
1643 readpage: afs_linux_readpage,
1644 writepage: afs_linux_writepage,
1645 commit_write: afs_linux_commit_write,
1646 prepare_write: afs_linux_prepare_write,
1649 struct inode_operations *afs_ops = &afs_file_iops;
1651 struct inode_operations afs_iops = {
1652 &afs_file_fops, /* file operations */
1653 NULL, /* afs_linux_create */
1654 NULL, /* afs_linux_lookup */
1655 NULL, /* afs_linux_link */
1656 NULL, /* afs_linux_unlink */
1657 NULL, /* afs_linux_symlink */
1658 NULL, /* afs_linux_mkdir */
1659 NULL, /* afs_linux_rmdir */
1660 NULL, /* afs_linux_mknod */
1661 NULL, /* afs_linux_rename */
1662 NULL, /* afs_linux_readlink */
1663 NULL, /* afs_linux_follow_link */
1665 NULL, /* afs_linux_writepage */
1666 NULL, /* afs_linux_bmap */
1667 NULL, /* afs_linux_truncate */
1668 afs_linux_permission,
1669 NULL, /* afs_linux_smap */
1670 afs_linux_updatepage,
1671 afs_linux_revalidate,
1674 struct inode_operations *afs_ops = &afs_iops;
1677 /* Separate ops vector for directories. Linux 2.2 tests type of inode
1678 * by what sort of operation is allowed.....
1680 #if defined(AFS_LINUX24_ENV)
1681 struct inode_operations afs_dir_iops = {
1682 create: afs_linux_create,
1683 lookup: afs_linux_lookup,
1684 link: afs_linux_link,
1685 unlink: afs_linux_unlink,
1686 symlink: afs_linux_symlink,
1687 mkdir: afs_linux_mkdir,
1688 rmdir: afs_linux_rmdir,
1689 rename: afs_linux_rename,
1690 revalidate: afs_linux_revalidate,
1691 setattr: afs_notify_change,
1692 permission: afs_linux_permission,
1695 struct inode_operations afs_dir_iops = {
1696 &afs_dir_fops, /* file operations for directories */
1704 NULL, /* afs_linux_mknod */
1706 NULL, /* afs_linux_readlink */
1707 NULL, /* afs_linux_follow_link */
1708 NULL, /* afs_linux_readpage */
1709 NULL, /* afs_linux_writepage */
1710 NULL, /* afs_linux_bmap */
1711 NULL, /* afs_linux_truncate */
1712 afs_linux_permission,
1713 NULL, /* afs_linux_smap */
1714 NULL, /* afs_linux_updatepage */
1715 afs_linux_revalidate,
1719 /* We really need a separate symlink set of ops, since do_follow_link()
1720 * determines if it _is_ a link by checking if the follow_link op is set.
1722 #if defined(AFS_LINUX24_ENV)
1723 static int afs_symlink_filler(struct file *file, struct page *page)
1725 struct inode *ip = (struct inode *) page->mapping->host;
1726 char *p = (char *)kmap(page);
1731 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
1735 p[code] = '\0'; /* null terminate? */
1739 SetPageUptodate(page);
1754 struct address_space_operations afs_symlink_aops = {
1755 readpage: afs_symlink_filler
1758 struct inode_operations afs_symlink_iops = {
1759 readlink: page_readlink,
1760 follow_link: page_follow_link,
1761 setattr: afs_notify_change,
1764 struct inode_operations afs_symlink_iops = {
1765 NULL, /* file operations */
1773 NULL, /* afs_linux_mknod */
1776 afs_linux_follow_link,
1777 NULL, /* readpage */
1778 NULL, /* afs_linux_writepage */
1779 NULL, /* afs_linux_bmap */
1780 NULL, /* afs_linux_truncate */
1781 afs_linux_permission, /* tho the code appears to indicate not used? */
1782 NULL, /* afs_linux_smap */
1783 NULL, /* updatepage */
1784 afs_linux_revalidate, /* tho the code appears to indicate not used? */