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 "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;
644 /* Safe because there are no large files, yet */
645 #if F_GETLK != F_GETLK64
648 else if (cmd = F_SETLK64)
650 else if (cmd = F_SETLKW64)
652 #endif /* F_GETLK != F_GETLK64 */
655 code = afs_lockctl(vcp, &flock, cmd, credp);
658 /* Convert flock back to Linux's file_lock */
659 flp->fl_type = flock.l_type;
660 flp->fl_pid = flock.l_pid;
661 flp->fl_start = flock.l_start;
662 flp->fl_end = flock.l_start + flock.l_len;
670 * flush is called from sys_close. We could ignore it, but sys_close return
671 * code comes from flush, not release. We need to use release to keep
672 * the vcache open count correct. Note that flush is called before release
673 * (via fput) in sys_close. vcp->flushcnt is a bit of ugliness to avoid
674 * races and also avoid calling afs_close twice when closing the file.
675 * If we merely checked for opens > 0 in afs_linux_release, then if an
676 * new open occurred when storing back the file, afs_linux_release would
677 * incorrectly close the file and decrement the opens count. Calling afs_close
678 * on the just flushed file is wasteful, since the background daemon will
679 * execute the code that finally decides there is nothing to do.
681 int afs_linux_flush(struct file *fp)
683 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
687 /* Only do this on the last close of the file pointer. */
688 #if defined(AFS_LINUX24_ENV)
689 if (atomic_read(&fp->f_count) > 1)
698 code = afs_close(vcp, fp->f_flags, credp);
699 vcp->flushcnt++; /* protected by AFS global lock. */
706 /* Not allowed to directly read a directory. */
707 ssize_t afs_linux_dir_read(struct file *fp, char *buf, size_t count, loff_t *ppos)
714 #if defined(AFS_LINUX24_ENV)
715 struct file_operations afs_dir_fops = {
716 read: generic_read_dir,
717 readdir: afs_linux_readdir,
719 open: afs_linux_open,
720 release: afs_linux_release,
723 struct file_operations afs_dir_fops = {
724 NULL, /* afs_linux_lseek */
726 NULL, /* afs_linux_write */
728 NULL, /* afs_linux_select */
729 afs_xioctl, /* close enough to use the ported AFS one */
730 NULL, /* afs_linux_mmap */
732 NULL, /* afs_linux_flush */
735 NULL, /* afs_linux_fasync */
736 NULL, /* afs_linux_check_media_change */
737 NULL, /* afs_linux_file_revalidate */
742 #if defined(AFS_LINUX24_ENV)
743 struct file_operations afs_file_fops = {
744 read: afs_linux_read,
745 write: afs_linux_write,
747 mmap: afs_linux_mmap,
748 open: afs_linux_open,
749 flush: afs_linux_flush,
750 release: afs_linux_release,
751 fsync: afs_linux_fsync,
752 lock: afs_linux_lock,
755 struct file_operations afs_file_fops = {
756 NULL, /* afs_linux_lseek */
759 NULL, /* afs_linux_readdir */
760 NULL, /* afs_linux_select */
761 afs_xioctl, /* close enough to use the ported AFS one */
767 NULL, /* afs_linux_fasync */
768 NULL, /* afs_linux_check_media_change */
769 NULL, /* afs_linux_file_revalidate */
775 /**********************************************************************
776 * AFS Linux dentry operations
777 **********************************************************************/
779 /* afs_linux_revalidate
780 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
782 static int afs_linux_revalidate(struct dentry *dp)
786 struct vrequest treq;
787 struct vcache *vcp = ITOAFS(dp->d_inode);
788 struct vcache *rootvp = NULL;
792 if (afs_fakestat_enable && vcp->mvstat == 1 && vcp->mvid &&
793 (vcp->states & CMValid) && (vcp->states & CStatd)) {
794 ObtainSharedLock(&afs_xvcache, 680);
795 rootvp = afs_FindVCache(vcp->mvid, 0, 0);
796 ReleaseSharedLock(&afs_xvcache);
799 #ifdef AFS_LINUX24_ENV
803 /* Make this a fast path (no crref), since it's called so often. */
804 if (vcp->states & CStatd) {
805 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
806 check_bad_parent(dp); /* check and correct mvid */
808 vcache2fakeinode(rootvp, vcp);
811 #ifdef AFS_LINUX24_ENV
814 if (rootvp) afs_PutVCache(rootvp);
820 code = afs_InitReq(&treq, credp);
822 code = afs_VerifyVCache(vcp, &treq);
824 #ifdef AFS_LINUX24_ENV
834 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
835 * In kernels 2.2.10 and above, we are passed an additional flags var which
836 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
837 * we are advised to follow the entry if it is a link or to make sure that
838 * it is a directory. But since the kernel itself checks these possibilities
839 * later on, we shouldn't have to do it until later. Perhaps in the future..
841 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
842 static int afs_linux_dentry_revalidate(struct dentry *dp, int flags)
844 static int afs_linux_dentry_revalidate(struct dentry *dp)
848 cred_t *credp = crref();
849 struct vrequest treq;
850 struct vcache *lookupvcp = NULL;
851 int code, bad_dentry = 1;
852 struct sysname_info sysState;
853 struct vcache *vcp = ITOAFS(dp->d_inode);
854 struct vcache *parentvcp = ITOAFS(dp->d_parent->d_inode);
859 sysState.allocked = 0;
861 /* If it's a negative dentry, then there's nothing to do. */
862 if (!vcp || !parentvcp)
865 /* If it is the AFS root, then there's no chance it needs
867 if (vcp == afs_globalVp) {
872 if ((code = afs_InitReq(&treq, credp)))
875 Check_AtSys(parentvcp, dp->d_name.name, &sysState, &treq);
876 name = sysState.name;
878 /* First try looking up the DNLC */
879 if ((lookupvcp = osi_dnlc_lookup(parentvcp, name, WRITE_LOCK))) {
880 /* Verify that the dentry does not point to an old inode */
881 if (vcp != lookupvcp)
883 /* Check and correct mvid */
884 if (*name != '/' && vcp->mvstat == 2)
885 check_bad_parent(dp);
891 /* A DNLC lookup failure cannot be trusted. Try a real lookup */
892 code = afs_lookup(parentvcp, name, &lookupvcp, credp);
894 /* Verify that the dentry does not point to an old inode */
895 if (vcp != lookupvcp)
903 afs_PutVCache(lookupvcp);
904 if (sysState.allocked)
905 osi_FreeLargeSpace(name);
911 shrink_dcache_parent(dp);
920 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
921 static int afs_linux_dentry_revalidate(struct dentry *dp, int flags)
923 static int afs_linux_dentry_revalidate(struct dentry *dp)
928 struct vrequest treq;
929 struct inode *ip = AFSTOI(dp->d_inode);
931 unsigned long timeout = 3*HZ; /* 3 seconds */
934 printk("negative dentry: %s\n", dp->d_name.name);
936 if (!(flags & LOOKUP_CONTINUE)) {
937 long diff = CURRENT_TIME - dp->d_parent->d_inode->i_mtime;
943 if (time_after(jiffies, dp->d_time + timeout))
954 /* afs_dentry_iput */
955 static void afs_dentry_iput(struct dentry *dp, struct inode *ip)
957 if (ICL_SETACTIVE(afs_iclSetp)) {
959 afs_Trace3(afs_iclSetp, CM_TRACE_DENTRYIPUT,
960 ICL_TYPE_POINTER, ip,
961 ICL_TYPE_STRING, dp->d_parent->d_name.name,
962 ICL_TYPE_STRING, dp->d_name.name);
969 static int afs_dentry_delete(struct dentry *dp)
971 if (ICL_SETACTIVE(afs_iclSetp)) {
973 afs_Trace3(afs_iclSetp, CM_TRACE_DENTRYDELETE, ICL_TYPE_POINTER,
974 dp->d_inode, ICL_TYPE_STRING, dp->d_parent->d_name.name,
975 ICL_TYPE_STRING, dp->d_name.name);
979 if (dp->d_inode && (ITOAFS(dp->d_inode)->states & CUnlinked))
980 return 1; /* bad inode? */
985 #if defined(AFS_LINUX24_ENV)
986 struct dentry_operations afs_dentry_operations = {
987 d_revalidate: afs_linux_dentry_revalidate,
988 d_iput: afs_dentry_iput,
989 d_delete: afs_dentry_delete,
991 struct dentry_operations *afs_dops = &afs_dentry_operations;
993 struct dentry_operations afs_dentry_operations = {
994 afs_linux_dentry_revalidate, /* d_validate(struct dentry *) */
996 NULL, /* d_compare */
997 afs_dentry_delete, /* d_delete(struct dentry *) */
998 NULL, /* d_release(struct dentry *) */
999 afs_dentry_iput /* d_iput(struct dentry *, struct inode *) */
1001 struct dentry_operations *afs_dops = &afs_dentry_operations;
1004 /**********************************************************************
1005 * AFS Linux inode operations
1006 **********************************************************************/
1010 * Merely need to set enough of vattr to get us through the create. Note
1011 * that the higher level code (open_namei) will take care of any tuncation
1012 * explicitly. Exclusive open is also taken care of in open_namei.
1014 * name is in kernel space at this point.
1016 int afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
1019 cred_t *credp = crref();
1022 const char *name = dp->d_name.name;
1026 vattr.va_mode = mode;
1029 code = afs_create(ITOAFS(dip), name, &vattr, NONEXCL, mode,
1030 (struct vcache**)&ip, credp);
1033 vattr2inode(ip, &vattr);
1034 /* Reset ops if symlink or directory. */
1035 #if defined(AFS_LINUX24_ENV)
1036 if (S_ISREG(ip->i_mode)) {
1037 ip->i_op = &afs_file_iops;
1038 ip->i_fop = &afs_file_fops;
1039 ip->i_data.a_ops = &afs_file_aops;
1040 } else if (S_ISDIR(ip->i_mode)) {
1041 ip->i_op = &afs_dir_iops;
1042 ip->i_fop = &afs_dir_fops;
1043 } else if (S_ISLNK(ip->i_mode)) {
1044 ip->i_op = &afs_symlink_iops;
1045 ip->i_data.a_ops = &afs_symlink_aops;
1046 ip->i_mapping = &ip->i_data;
1048 printk("afs_linux_create: FIXME\n");
1050 if (S_ISDIR(ip->i_mode))
1051 ip->i_op = &afs_dir_iops;
1052 else if (S_ISLNK(ip->i_mode))
1053 ip->i_op = &afs_symlink_iops;
1056 dp->d_op = afs_dops;
1057 dp->d_time = jiffies;
1058 d_instantiate(dp, ip);
1066 /* afs_linux_lookup */
1067 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1068 struct dentry *afs_linux_lookup(struct inode *dip, struct dentry *dp)
1070 int afs_linux_lookup(struct inode *dip, struct dentry *dp)
1074 cred_t *credp = crref();
1075 struct vcache *vcp=NULL;
1076 const char *comp = dp->d_name.name;
1078 code = afs_lookup(ITOAFS(dip), comp, &vcp, credp);
1081 struct inode *ip = AFSTOI(vcp);
1082 /* Reset ops if symlink or directory. */
1083 #if defined(AFS_LINUX24_ENV)
1084 if (S_ISREG(ip->i_mode)) {
1085 ip->i_op = &afs_file_iops;
1086 ip->i_fop = &afs_file_fops;
1087 ip->i_data.a_ops = &afs_file_aops;
1088 } else if (S_ISDIR(ip->i_mode)) {
1089 ip->i_op = &afs_dir_iops;
1090 ip->i_fop = &afs_dir_fops;
1091 } else if (S_ISLNK(ip->i_mode)) {
1092 ip->i_op = &afs_symlink_iops;
1093 ip->i_data.a_ops = &afs_symlink_aops;
1094 ip->i_mapping = &ip->i_data;
1096 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);
1098 if (S_ISDIR(ip->i_mode))
1099 ip->i_op = &afs_dir_iops;
1100 else if (S_ISLNK(ip->i_mode))
1101 ip->i_op = &afs_symlink_iops;
1104 dp->d_time = jiffies;
1105 dp->d_op = afs_dops;
1106 d_add(dp, AFSTOI(vcp));
1111 /* It's ok for the file to not be found. That's noted by the caller by
1112 * seeing that the dp->d_inode field is NULL.
1114 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1118 return ERR_PTR(-code);
1126 int afs_linux_link(struct dentry *olddp, struct inode *dip,
1127 struct dentry *newdp)
1130 cred_t *credp = crref();
1131 const char *name = newdp->d_name.name;
1132 struct inode *oldip = olddp->d_inode;
1134 /* If afs_link returned the vnode, we could instantiate the
1135 * dentry. Since it's not, we drop this one and do a new lookup.
1140 code = afs_link(ITOAFS(oldip), ITOAFS(dip), name, credp);
1147 int afs_linux_unlink(struct inode *dip, struct dentry *dp)
1150 cred_t *credp = crref();
1151 const char *name = dp->d_name.name;
1154 code = afs_remove(ITOAFS(dip), name, credp);
1163 int afs_linux_symlink(struct inode *dip, struct dentry *dp,
1167 cred_t *credp = crref();
1169 const char *name = dp->d_name.name;
1171 /* If afs_symlink returned the vnode, we could instantiate the
1172 * dentry. Since it's not, we drop this one and do a new lookup.
1178 code = afs_symlink(ITOAFS(dip), name, &vattr, target, credp);
1184 int 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;
1194 vattr.va_mask = ATTR_MODE;
1195 vattr.va_mode = mode;
1196 code = afs_mkdir(ITOAFS(dip), name, &vattr, &tvcp, credp);
1199 tvcp->v.v_op = &afs_dir_iops;
1200 #if defined(AFS_LINUX24_ENV)
1201 tvcp->v.v_fop = &afs_dir_fops;
1203 dp->d_op = afs_dops;
1204 dp->d_time = jiffies;
1205 d_instantiate(dp, AFSTOI(tvcp));
1213 int afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1216 cred_t *credp = crref();
1217 const char *name = dp->d_name.name;
1220 code = afs_rmdir(ITOAFS(dip), name, credp);
1222 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1223 * that failed because a directory is not empty. So, we map
1224 * EEXIST to ENOTEMPTY on linux.
1226 if (code == EEXIST) {
1241 int afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1242 struct inode *newip, struct dentry *newdp)
1245 cred_t *credp = crref();
1246 const char *oldname = olddp->d_name.name;
1247 const char *newname = newdp->d_name.name;
1249 /* Remove old and new entries from name hash. New one will change below.
1250 * While it's optimal to catch failures and re-insert newdp into hash,
1251 * it's also error prone and in that case we're already dealing with error
1252 * cases. Let another lookup put things right, if need be.
1254 if (!list_empty(&olddp->d_hash)) {
1257 if (!list_empty(&newdp->d_hash)) {
1261 code = afs_rename(ITOAFS(oldip), oldname, ITOAFS(newip),
1266 /* update time so it doesn't expire immediately */
1267 newdp->d_time = jiffies;
1268 d_move(olddp, newdp);
1276 /* afs_linux_ireadlink
1277 * Internal readlink which can return link contents to user or kernel space.
1278 * Note that the buffer is NOT supposed to be null-terminated.
1280 static int afs_linux_ireadlink(struct inode *ip, char *target, int maxlen,
1284 cred_t *credp = crref();
1288 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1289 code = afs_readlink(ITOAFS(ip), &tuio, credp);
1293 return maxlen - tuio.uio_resid;
1298 #if !defined(AFS_LINUX24_ENV)
1299 /* afs_linux_readlink
1300 * Fill target (which is in user space) with contents of symlink.
1302 int afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1305 struct inode *ip = dp->d_inode;
1308 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1314 /* afs_linux_follow_link
1315 * a file system dependent link following routine.
1317 struct dentry * afs_linux_follow_link(struct dentry *dp,
1318 struct dentry *basep,
1319 unsigned int follow)
1327 name = osi_Alloc(PATH_MAX+1);
1331 return ERR_PTR(-EIO);
1334 code = afs_linux_ireadlink(dp->d_inode, name, PATH_MAX, AFS_UIOSYS);
1339 res = ERR_PTR(code);
1343 res = lookup_dentry(name, basep, follow);
1347 osi_Free(name, PATH_MAX+1);
1353 /* afs_linux_readpage
1354 * all reads come through here. A strategy-like read call.
1356 int afs_linux_readpage(struct file *fp, struct page *pp)
1359 cred_t *credp = crref();
1360 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1362 afs_offs_t offset = pp->index << PAGE_CACHE_SHIFT;
1364 ulong address = afs_linux_page_address(pp);
1365 afs_offs_t offset = pageoff(pp);
1369 struct inode *ip = FILE_INODE(fp);
1370 int cnt = atomic_read(&pp->count);
1371 struct vcache *avc = ITOAFS(ip);
1374 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE,
1375 ICL_TYPE_POINTER, ip,
1376 ICL_TYPE_POINTER, pp,
1377 ICL_TYPE_INT32, cnt,
1378 ICL_TYPE_INT32, 99999); /* not a possible code value */
1379 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1385 atomic_add(1, &pp->count);
1386 set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
1387 clear_bit(PG_error, &pp->flags);
1390 setup_uio(&tuio, &iovec, (char*)address, offset, PAGESIZE,
1391 UIO_READ, AFS_UIOSYS);
1392 code = afs_rdwr(avc, &tuio, UIO_READ, 0, credp);
1393 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1398 if (tuio.uio_resid) /* zero remainder of page */
1399 memset((void*)(address+(PAGESIZE-tuio.uio_resid)), 0,
1401 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1402 flush_dcache_page(pp);
1403 SetPageUptodate(pp);
1405 set_bit(PG_uptodate, &pp->flags);
1409 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1413 clear_bit(PG_locked, &pp->flags);
1418 if (!code && AFS_CHUNKOFFSET(offset) == 0) {
1420 struct vrequest treq;
1422 code = afs_InitReq(&treq, credp);
1423 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
1424 tdc = afs_FindDCache(avc, offset);
1426 if (!(tdc->mflags & DFNextStarted))
1427 afs_PrefetchChunk(avc, tdc, credp, &treq);
1430 ReleaseWriteLock(&avc->lock);
1435 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE,
1436 ICL_TYPE_POINTER, ip,
1437 ICL_TYPE_POINTER, pp,
1438 ICL_TYPE_INT32, cnt,
1439 ICL_TYPE_INT32, code);
1444 #if defined(AFS_LINUX24_ENV)
1445 int afs_linux_writepage(struct page *pp)
1447 struct address_space *mapping = pp->mapping;
1448 struct inode *inode;
1449 unsigned long end_index;
1450 unsigned offset = PAGE_CACHE_SIZE;
1453 inode = (struct inode *) mapping->host;
1454 end_index = inode->i_size >> PAGE_CACHE_SHIFT;
1457 if (pp->index < end_index)
1459 /* things got complicated... */
1460 offset = inode->i_size & (PAGE_CACHE_SIZE-1);
1461 /* OK, are we completely out? */
1462 if (pp->index >= end_index+1 || !offset)
1466 status = afs_linux_writepage_sync(inode, pp, 0, offset);
1468 SetPageUptodate(pp);
1470 if (status == offset)
1478 /* afs_linux_bmap - supports generic_readpage, but we roll our own. */
1479 int afs_linux_bmap(struct inode *ip, int) { return -EINVAL; }
1481 /* afs_linux_truncate
1482 * Handles discarding disk blocks if this were a device. ext2 indicates we
1483 * may need to zero partial last pages of memory mapped files.
1485 void afs_linux_truncate(struct inode *ip)
1490 /* afs_linux_permission
1491 * Check access rights - returns error if can't check or permission denied.
1493 int afs_linux_permission(struct inode *ip, int mode)
1496 cred_t *credp = crref();
1500 if (mode & MAY_EXEC) tmp |= VEXEC;
1501 if (mode & MAY_READ) tmp |= VREAD;
1502 if (mode & MAY_WRITE) tmp |= VWRITE;
1503 code = afs_access(ITOAFS(ip), tmp, credp);
1512 /* msdos sector mapping hack for memory mapping. */
1513 int afs_linux_smap(struct inode *ip, int) { return -EINVAL; }
1516 #if defined(AFS_LINUX24_ENV)
1517 int afs_linux_writepage_sync(struct inode *ip, struct page *pp,
1518 unsigned long offset,
1521 struct vcache *vcp = ITOAFS(ip);
1530 buffer = kmap(pp) + offset;
1531 base = (pp->index << PAGE_CACHE_SHIFT) + offset;
1534 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1535 ICL_TYPE_POINTER, pp,
1536 ICL_TYPE_INT32, atomic_read(&pp->count),
1537 ICL_TYPE_INT32, 99999);
1539 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
1541 code = afs_write(vcp, &tuio, f_flags, credp, 0);
1545 if (!code && afs_stats_cmperf.cacheCurrDirtyChunks >
1546 afs_stats_cmperf.cacheMaxDirtyChunks) {
1547 struct vrequest treq;
1549 ObtainWriteLock(&vcp->lock, 533);
1550 if (!afs_InitReq(&treq, credp))
1551 code = afs_DoPartialWrite(vcp, &treq);
1552 ReleaseWriteLock(&vcp->lock);
1554 code = code ? -code : count - tuio.uio_resid;
1556 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1557 ICL_TYPE_POINTER, pp,
1558 ICL_TYPE_INT32, atomic_read(&pp->count),
1559 ICL_TYPE_INT32, code);
1568 afs_linux_updatepage(struct file *file, struct page *page,
1569 unsigned long offset, unsigned int count)
1571 struct dentry *dentry = file->f_dentry;
1573 return afs_linux_writepage_sync(dentry->d_inode, page, offset, count);
1576 /* afs_linux_updatepage
1577 * What one would have thought was writepage - write dirty page to file.
1578 * Called from generic_file_write. buffer is still in user space. pagep
1579 * has been filled in with old data if we're updating less than a page.
1581 int afs_linux_updatepage(struct file *fp, struct page *pp,
1582 unsigned long offset,
1583 unsigned int count, int sync)
1585 struct vcache *vcp = ITOAFS(FILE_INODE(fp));
1586 u8 *page_addr = (u8*) afs_linux_page_address(pp);
1592 set_bit(PG_locked, &pp->flags);
1596 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1597 ICL_TYPE_POINTER, pp,
1598 ICL_TYPE_INT32, atomic_read(&pp->count),
1599 ICL_TYPE_INT32, 99999);
1600 setup_uio(&tuio, &iovec, page_addr + offset, (afs_offs_t)(pageoff(pp) + offset),
1601 count, UIO_WRITE, AFS_UIOSYS);
1603 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
1607 code = code ? -code : count - tuio.uio_resid;
1608 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
1609 ICL_TYPE_POINTER, pp,
1610 ICL_TYPE_INT32, atomic_read(&pp->count),
1611 ICL_TYPE_INT32, code);
1616 clear_bit(PG_locked, &pp->flags);
1621 #if defined(AFS_LINUX24_ENV)
1622 static int afs_linux_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1628 code = afs_linux_updatepage(file, page, offset, to-offset);
1636 static int afs_linux_prepare_write(struct file *file, struct page *page,
1637 unsigned from, unsigned to)
1643 extern int afs_notify_change(struct dentry *dp, struct iattr* iattrp);
1646 #if defined(AFS_LINUX24_ENV)
1647 struct inode_operations afs_file_iops = {
1648 revalidate: afs_linux_revalidate,
1649 setattr: afs_notify_change,
1650 permission: afs_linux_permission,
1652 struct address_space_operations afs_file_aops = {
1653 readpage: afs_linux_readpage,
1654 writepage: afs_linux_writepage,
1655 commit_write: afs_linux_commit_write,
1656 prepare_write: afs_linux_prepare_write,
1659 struct inode_operations *afs_ops = &afs_file_iops;
1661 struct inode_operations afs_iops = {
1662 &afs_file_fops, /* file operations */
1663 NULL, /* afs_linux_create */
1664 NULL, /* afs_linux_lookup */
1665 NULL, /* afs_linux_link */
1666 NULL, /* afs_linux_unlink */
1667 NULL, /* afs_linux_symlink */
1668 NULL, /* afs_linux_mkdir */
1669 NULL, /* afs_linux_rmdir */
1670 NULL, /* afs_linux_mknod */
1671 NULL, /* afs_linux_rename */
1672 NULL, /* afs_linux_readlink */
1673 NULL, /* afs_linux_follow_link */
1675 NULL, /* afs_linux_writepage */
1676 NULL, /* afs_linux_bmap */
1677 NULL, /* afs_linux_truncate */
1678 afs_linux_permission,
1679 NULL, /* afs_linux_smap */
1680 afs_linux_updatepage,
1681 afs_linux_revalidate,
1684 struct inode_operations *afs_ops = &afs_iops;
1687 /* Separate ops vector for directories. Linux 2.2 tests type of inode
1688 * by what sort of operation is allowed.....
1690 #if defined(AFS_LINUX24_ENV)
1691 struct inode_operations afs_dir_iops = {
1692 create: afs_linux_create,
1693 lookup: afs_linux_lookup,
1694 link: afs_linux_link,
1695 unlink: afs_linux_unlink,
1696 symlink: afs_linux_symlink,
1697 mkdir: afs_linux_mkdir,
1698 rmdir: afs_linux_rmdir,
1699 rename: afs_linux_rename,
1700 revalidate: afs_linux_revalidate,
1701 setattr: afs_notify_change,
1702 permission: afs_linux_permission,
1705 struct inode_operations afs_dir_iops = {
1706 &afs_dir_fops, /* file operations for directories */
1714 NULL, /* afs_linux_mknod */
1716 NULL, /* afs_linux_readlink */
1717 NULL, /* afs_linux_follow_link */
1718 NULL, /* afs_linux_readpage */
1719 NULL, /* afs_linux_writepage */
1720 NULL, /* afs_linux_bmap */
1721 NULL, /* afs_linux_truncate */
1722 afs_linux_permission,
1723 NULL, /* afs_linux_smap */
1724 NULL, /* afs_linux_updatepage */
1725 afs_linux_revalidate,
1729 /* We really need a separate symlink set of ops, since do_follow_link()
1730 * determines if it _is_ a link by checking if the follow_link op is set.
1732 #if defined(AFS_LINUX24_ENV)
1733 static int afs_symlink_filler(struct file *file, struct page *page)
1735 struct inode *ip = (struct inode *) page->mapping->host;
1736 char *p = (char *)kmap(page);
1741 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
1745 p[code] = '\0'; /* null terminate? */
1749 SetPageUptodate(page);
1764 struct address_space_operations afs_symlink_aops = {
1765 readpage: afs_symlink_filler
1768 struct inode_operations afs_symlink_iops = {
1769 readlink: page_readlink,
1770 follow_link: page_follow_link,
1771 setattr: afs_notify_change,
1774 struct inode_operations afs_symlink_iops = {
1775 NULL, /* file operations */
1783 NULL, /* afs_linux_mknod */
1786 afs_linux_follow_link,
1787 NULL, /* readpage */
1788 NULL, /* afs_linux_writepage */
1789 NULL, /* afs_linux_bmap */
1790 NULL, /* afs_linux_truncate */
1791 afs_linux_permission, /* tho the code appears to indicate not used? */
1792 NULL, /* afs_linux_smap */
1793 NULL, /* updatepage */
1794 afs_linux_revalidate, /* tho the code appears to indicate not used? */