2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
11 * Linux specific vnodeops. Also includes the glue routines required to call
14 * So far the only truly scary part is that Linux relies on the inode cache
15 * to be up to date. Don't you dare break a callback and expect an fstat
16 * to give you meaningful information. This appears to be fixed in the 2.1
17 * development kernels. As it is we can fix this now by intercepting the
21 #include <afsconfig.h>
22 #include "afs/param.h"
25 #include "afs/sysincludes.h"
26 #include "afsincludes.h"
27 #include "afs/afs_stats.h"
29 #ifdef HAVE_MM_INLINE_H
30 #include "h/mm_inline.h"
32 #include "h/pagemap.h"
33 #if defined(AFS_LINUX24_ENV)
34 #include "h/smp_lock.h"
36 #if defined(AFS_CACHE_BYPASS)
38 #include "afs/afs_bypasscache.h"
41 #include "osi_pagecopy.h"
44 #define pageoff(pp) pgoff2loff((pp)->index)
46 #define pageoff(pp) pp->offset
49 #ifndef HAVE_PAGEVEC_LRU_ADD_FILE
50 #define __pagevec_lru_add_file __pagevec_lru_add
54 #define MAX_ERRNO 1000L
57 extern struct vcache *afs_globalVp;
58 extern int afs_notify_change(struct dentry *dp, struct iattr *iattrp);
59 #if defined(AFS_LINUX24_ENV)
60 /* Some uses of BKL are perhaps not needed for bypass or memcache--
61 * why don't we try it out? */
62 extern struct afs_cacheOps afs_UfsCacheOps;
63 #define maybe_lock_kernel() \
65 if(afs_cacheType == &afs_UfsCacheOps) \
70 #define maybe_unlock_kernel() \
72 if(afs_cacheType == &afs_UfsCacheOps) \
75 #endif /* AFS_LINUX24_ENV */
78 /* This function converts a positive error code from AFS into a negative
79 * code suitable for passing into the Linux VFS layer. It checks that the
80 * error code is within the permissable bounds for the ERR_PTR mechanism.
82 * _All_ error codes which come from the AFS layer should be passed through
83 * this function before being returned to the kernel.
86 static inline int afs_convert_code(int code) {
87 if ((code >= 0) && (code <= MAX_ERRNO))
93 /* Linux doesn't require a credp for many functions, and crref is an expensive
94 * operation. This helper function avoids obtaining it for VerifyVCache calls
97 static inline int afs_linux_VerifyVCache(struct vcache *avc, cred_t **retcred) {
102 if (avc->f.states & CStatd) {
110 code = afs_InitReq(&treq, credp);
112 code = afs_VerifyVCache2(avc, &treq);
119 return afs_convert_code(code);
123 afs_linux_read(struct file *fp, char *buf, size_t count, loff_t * offp)
126 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
127 #if defined(AFS_CACHE_BYPASS) && LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
128 afs_size_t isize, offindex;
132 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
133 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
135 code = afs_linux_VerifyVCache(vcp, NULL);
138 #if defined(AFS_CACHE_BYPASS) && LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
139 isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
140 offindex = *offp >> PAGE_CACHE_SHIFT;
141 if(offindex > isize) {
146 /* Linux's FlushPages implementation doesn't ever use credp,
147 * so we optimise by not using it */
148 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
151 code = do_sync_read(fp, buf, count, offp);
153 code = generic_file_read(fp, buf, count, offp);
158 afs_Trace4(afs_iclSetp, CM_TRACE_READOP, ICL_TYPE_POINTER, vcp,
159 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
161 #if defined(AFS_CACHE_BYPASS) && LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
169 /* Now we have integrated VM for writes as well as reads. generic_file_write
170 * also takes care of re-positioning the pointer if file is open in append
171 * mode. Call fake open/close to ensure we do writes of core dumps.
174 afs_linux_write(struct file *fp, const char *buf, size_t count, loff_t * offp)
177 struct vcache *vcp = VTOAFS(fp->f_dentry->d_inode);
182 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
183 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
184 (fp->f_flags & O_APPEND) ? 99998 : 99999);
186 code = afs_linux_VerifyVCache(vcp, &credp);
188 ObtainWriteLock(&vcp->lock, 529);
190 ReleaseWriteLock(&vcp->lock);
194 code = do_sync_write(fp, buf, count, offp);
196 code = generic_file_write(fp, buf, count, offp);
201 ObtainWriteLock(&vcp->lock, 530);
203 if (vcp->execsOrWriters == 1 && !credp)
206 afs_FakeClose(vcp, credp);
207 ReleaseWriteLock(&vcp->lock);
209 afs_Trace4(afs_iclSetp, CM_TRACE_WRITEOP, ICL_TYPE_POINTER, vcp,
210 ICL_TYPE_OFFSET, offp, ICL_TYPE_INT32, count, ICL_TYPE_INT32,
219 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
221 /* This is a complete rewrite of afs_readdir, since we can make use of
222 * filldir instead of afs_readdir_move. Note that changes to vcache/dcache
223 * handling and use of bulkstats will need to be reflected here as well.
226 afs_linux_readdir(struct file *fp, void *dirbuf, filldir_t filldir)
228 struct vcache *avc = VTOAFS(FILE_INODE(fp));
229 struct vrequest treq;
230 register struct dcache *tdc;
237 afs_size_t origOffset, tlen;
238 cred_t *credp = crref();
239 struct afs_fakestat_state fakestat;
242 AFS_STATCNT(afs_readdir);
244 code = afs_convert_code(afs_InitReq(&treq, credp));
249 afs_InitFakeStat(&fakestat);
250 code = afs_convert_code(afs_EvalFakeStat(&avc, &fakestat, &treq));
254 /* update the cache entry */
256 code = afs_convert_code(afs_VerifyVCache2(avc, &treq));
260 /* get a reference to the entire directory */
261 tdc = afs_GetDCache(avc, (afs_size_t) 0, &treq, &origOffset, &tlen, 1);
267 ObtainSharedLock(&avc->lock, 810);
268 UpgradeSToWLock(&avc->lock, 811);
269 ObtainReadLock(&tdc->lock);
271 * Make sure that the data in the cache is current. There are two
272 * cases we need to worry about:
273 * 1. The cache data is being fetched by another process.
274 * 2. The cache data is no longer valid
276 while ((avc->f.states & CStatd)
277 && (tdc->dflags & DFFetching)
278 && hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
279 ReleaseReadLock(&tdc->lock);
280 ReleaseSharedLock(&avc->lock);
281 afs_osi_Sleep(&tdc->validPos);
282 ObtainSharedLock(&avc->lock, 812);
283 ObtainReadLock(&tdc->lock);
285 if (!(avc->f.states & CStatd)
286 || !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
287 ReleaseReadLock(&tdc->lock);
288 ReleaseSharedLock(&avc->lock);
293 /* Set the readdir-in-progress flag, and downgrade the lock
294 * to shared so others will be able to acquire a read lock.
296 avc->f.states |= CReadDir;
297 avc->dcreaddir = tdc;
298 avc->readdir_pid = MyPidxx2Pid(MyPidxx);
299 ConvertWToSLock(&avc->lock);
301 /* Fill in until we get an error or we're done. This implementation
302 * takes an offset in units of blobs, rather than bytes.
305 offset = (int) fp->f_pos;
307 dirpos = BlobScan(tdc, offset);
311 de = afs_dir_GetBlob(tdc, dirpos);
315 ino = afs_calc_inum (avc->f.fid.Fid.Volume, ntohl(de->fid.vnode));
318 len = strlen(de->name);
320 printf("afs_linux_readdir: afs_dir_GetBlob failed, null name (inode %lx, dirpos %d)\n",
321 (unsigned long)&tdc->f.inode, dirpos);
323 ReleaseSharedLock(&avc->lock);
329 /* filldir returns -EINVAL when the buffer is full. */
330 #if (defined(AFS_LINUX24_ENV) || defined(pgoff2loff)) && defined(DECLARE_FSTYPE)
332 unsigned int type = DT_UNKNOWN;
333 struct VenusFid afid;
336 afid.Cell = avc->f.fid.Cell;
337 afid.Fid.Volume = avc->f.fid.Fid.Volume;
338 afid.Fid.Vnode = ntohl(de->fid.vnode);
339 afid.Fid.Unique = ntohl(de->fid.vunique);
340 if ((avc->f.states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
342 } else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
345 } else if (((tvc->f.states) & (CStatd | CTruth))) {
346 /* CTruth will be set if the object has
351 else if (vtype == VREG)
353 /* Don't do this until we're sure it can't be a mtpt */
354 /* else if (vtype == VLNK)
356 /* what other types does AFS support? */
358 /* clean up from afs_FindVCache */
362 * If this is NFS readdirplus, then the filler is going to
363 * call getattr on this inode, which will deadlock if we're
367 code = (*filldir) (dirbuf, de->name, len, offset, ino, type);
371 code = (*filldir) (dirbuf, de->name, len, offset, ino);
376 offset = dirpos + 1 + ((len + 16) >> 5);
378 /* If filldir didn't fill in the last one this is still pointing to that
381 fp->f_pos = (loff_t) offset;
383 ReleaseReadLock(&tdc->lock);
385 UpgradeSToWLock(&avc->lock, 813);
386 avc->f.states &= ~CReadDir;
388 avc->readdir_pid = 0;
389 ReleaseSharedLock(&avc->lock);
393 afs_PutFakeStat(&fakestat);
400 /* in afs_pioctl.c */
401 extern int afs_xioctl(struct inode *ip, struct file *fp, unsigned int com,
404 #if defined(HAVE_UNLOCKED_IOCTL) || defined(HAVE_COMPAT_IOCTL)
405 static long afs_unlocked_xioctl(struct file *fp, unsigned int com,
407 return afs_xioctl(FILE_INODE(fp), fp, com, arg);
414 afs_linux_mmap(struct file *fp, struct vm_area_struct *vmap)
416 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
420 #if defined(AFS_LINUX24_ENV)
421 afs_Trace3(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
422 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
423 vmap->vm_end - vmap->vm_start);
425 afs_Trace4(afs_iclSetp, CM_TRACE_GMAP, ICL_TYPE_POINTER, vcp,
426 ICL_TYPE_POINTER, vmap->vm_start, ICL_TYPE_INT32,
427 vmap->vm_end - vmap->vm_start, ICL_TYPE_INT32,
431 /* get a validated vcache entry */
432 code = afs_linux_VerifyVCache(vcp, NULL);
434 /* Linux's Flushpage implementation doesn't use credp, so optimise
435 * our code to not need to crref() it */
436 osi_FlushPages(vcp, NULL); /* ensure stale pages are gone */
438 code = generic_file_mmap(fp, vmap);
441 vcp->f.states |= CMAPPED;
448 afs_linux_open(struct inode *ip, struct file *fp)
450 struct vcache *vcp = VTOAFS(ip);
451 cred_t *credp = crref();
454 #ifdef AFS_LINUX24_ENV
458 code = afs_open(&vcp, fp->f_flags, credp);
460 #ifdef AFS_LINUX24_ENV
461 maybe_unlock_kernel();
465 return afs_convert_code(code);
469 afs_linux_release(struct inode *ip, struct file *fp)
471 struct vcache *vcp = VTOAFS(ip);
472 cred_t *credp = crref();
475 #ifdef AFS_LINUX24_ENV
479 code = afs_close(vcp, fp->f_flags, credp);
481 #ifdef AFS_LINUX24_ENV
482 maybe_unlock_kernel();
486 return afs_convert_code(code);
490 #if defined(AFS_LINUX24_ENV)
491 afs_linux_fsync(struct file *fp, struct dentry *dp, int datasync)
493 afs_linux_fsync(struct file *fp, struct dentry *dp)
497 struct inode *ip = FILE_INODE(fp);
498 cred_t *credp = crref();
500 #ifdef AFS_LINUX24_ENV
504 code = afs_fsync(VTOAFS(ip), credp);
506 #ifdef AFS_LINUX24_ENV
507 maybe_unlock_kernel();
510 return afs_convert_code(code);
516 afs_linux_lock(struct file *fp, int cmd, struct file_lock *flp)
519 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
520 cred_t *credp = crref();
521 struct AFS_FLOCK flock;
522 #if defined(POSIX_TEST_LOCK_CONFLICT_ARG)
523 struct file_lock conflict;
524 #elif defined(POSIX_TEST_LOCK_RETURNS_CONFLICT)
525 struct file_lock *conflict;
528 /* Convert to a lock format afs_lockctl understands. */
529 memset(&flock, 0, sizeof(flock));
530 flock.l_type = flp->fl_type;
531 flock.l_pid = flp->fl_pid;
533 flock.l_start = flp->fl_start;
534 flock.l_len = flp->fl_end - flp->fl_start + 1;
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 #ifdef AFS_LINUX24_ENV
551 if ((code == 0 || flp->fl_type == F_UNLCK) &&
552 (cmd == F_SETLK || cmd == F_SETLKW)) {
553 # ifdef POSIX_LOCK_FILE_WAIT_ARG
554 code = posix_lock_file(fp, flp, 0);
556 flp->fl_flags &=~ FL_SLEEP;
557 code = posix_lock_file(fp, flp);
559 if (code && flp->fl_type != F_UNLCK) {
560 struct AFS_FLOCK flock2;
562 flock2.l_type = F_UNLCK;
564 afs_lockctl(vcp, &flock2, F_SETLK, credp);
568 /* If lockctl says there are no conflicting locks, then also check with the
569 * kernel, as lockctl knows nothing about byte range locks
571 if (code == 0 && cmd == F_GETLK && flock.l_type == F_UNLCK) {
572 # if defined(POSIX_TEST_LOCK_CONFLICT_ARG)
573 if (posix_test_lock(fp, flp, &conflict)) {
574 locks_copy_lock(flp, &conflict);
575 flp->fl_type = F_UNLCK;
579 # elif defined(POSIX_TEST_LOCK_RETURNS_CONFLICT)
580 if ((conflict = posix_test_lock(fp, flp))) {
581 locks_copy_lock(flp, conflict);
582 flp->fl_type = F_UNLCK;
587 posix_test_lock(fp, flp);
588 /* If we found a lock in the kernel's structure, return it */
589 if (flp->fl_type != F_UNLCK) {
597 /* Convert flock back to Linux's file_lock */
598 flp->fl_type = flock.l_type;
599 flp->fl_pid = flock.l_pid;
600 flp->fl_start = flock.l_start;
601 flp->fl_end = flock.l_start + flock.l_len - 1;
604 return afs_convert_code(code);
607 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
609 afs_linux_flock(struct file *fp, int cmd, struct file_lock *flp) {
611 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
612 cred_t *credp = crref();
613 struct AFS_FLOCK flock;
614 /* Convert to a lock format afs_lockctl understands. */
615 memset(&flock, 0, sizeof(flock));
616 flock.l_type = flp->fl_type;
617 flock.l_pid = flp->fl_pid;
620 flock.l_len = OFFSET_MAX;
622 /* Safe because there are no large files, yet */
623 #if defined(F_GETLK64) && (F_GETLK != F_GETLK64)
624 if (cmd == F_GETLK64)
626 else if (cmd == F_SETLK64)
628 else if (cmd == F_SETLKW64)
630 #endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
633 code = afs_lockctl(vcp, &flock, cmd, credp);
636 if ((code == 0 || flp->fl_type == F_UNLCK) &&
637 (cmd == F_SETLK || cmd == F_SETLKW)) {
638 flp->fl_flags &=~ FL_SLEEP;
639 code = flock_lock_file_wait(fp, flp);
640 if (code && flp->fl_type != F_UNLCK) {
641 struct AFS_FLOCK flock2;
643 flock2.l_type = F_UNLCK;
645 afs_lockctl(vcp, &flock2, F_SETLK, credp);
649 /* Convert flock back to Linux's file_lock */
650 flp->fl_type = flock.l_type;
651 flp->fl_pid = flock.l_pid;
654 return afs_convert_code(code);
659 * essentially the same as afs_fsync() but we need to get the return
660 * code for the sys_close() here, not afs_linux_release(), so call
661 * afs_StoreAllSegments() with AFS_LASTSTORE
664 #if defined(FOP_FLUSH_TAKES_FL_OWNER_T)
665 afs_linux_flush(struct file *fp, fl_owner_t id)
667 afs_linux_flush(struct file *fp)
670 struct vrequest treq;
674 #if defined(AFS_CACHE_BYPASS)
680 if ((fp->f_flags & O_ACCMODE) == O_RDONLY) { /* readers dont flush */
688 vcp = VTOAFS(FILE_INODE(fp));
690 code = afs_InitReq(&treq, credp);
693 #if defined(AFS_CACHE_BYPASS)
694 /* If caching is bypassed for this file, or globally, just return 0 */
695 if(cache_bypass_strategy == ALWAYS_BYPASS_CACHE)
698 ObtainReadLock(&vcp->lock);
699 if(vcp->cachingStates & FCSBypass)
701 ReleaseReadLock(&vcp->lock);
704 /* future proof: don't rely on 0 return from afs_InitReq */
709 ObtainSharedLock(&vcp->lock, 535);
710 if ((vcp->execsOrWriters > 0) && (file_count(fp) == 1)) {
711 UpgradeSToWLock(&vcp->lock, 536);
712 if (!AFS_IS_DISCONNECTED) {
713 code = afs_StoreAllSegments(vcp,
715 AFS_SYNC | AFS_LASTSTORE);
717 afs_DisconAddDirty(vcp, VDisconWriteOsiFlush, 1);
719 ConvertWToSLock(&vcp->lock);
721 code = afs_CheckCode(code, &treq, 54);
722 ReleaseSharedLock(&vcp->lock);
729 return afs_convert_code(code);
732 #if !defined(AFS_LINUX24_ENV)
733 /* Not allowed to directly read a directory. */
735 afs_linux_dir_read(struct file * fp, char *buf, size_t count, loff_t * ppos)
743 struct file_operations afs_dir_fops = {
744 #if !defined(AFS_LINUX24_ENV)
745 .read = afs_linux_dir_read,
746 .lock = afs_linux_lock,
747 .fsync = afs_linux_fsync,
749 .read = generic_read_dir,
751 .readdir = afs_linux_readdir,
752 #ifdef HAVE_UNLOCKED_IOCTL
753 .unlocked_ioctl = afs_unlocked_xioctl,
757 #ifdef HAVE_COMPAT_IOCTL
758 .compat_ioctl = afs_unlocked_xioctl,
760 .open = afs_linux_open,
761 .release = afs_linux_release,
764 struct file_operations afs_file_fops = {
765 .read = afs_linux_read,
766 .write = afs_linux_write,
767 #ifdef GENERIC_FILE_AIO_READ
768 .aio_read = generic_file_aio_read,
769 .aio_write = generic_file_aio_write,
771 #ifdef HAVE_UNLOCKED_IOCTL
772 .unlocked_ioctl = afs_unlocked_xioctl,
776 #ifdef HAVE_COMPAT_IOCTL
777 .compat_ioctl = afs_unlocked_xioctl,
779 .mmap = afs_linux_mmap,
780 .open = afs_linux_open,
781 .flush = afs_linux_flush,
782 .release = afs_linux_release,
783 .fsync = afs_linux_fsync,
784 .lock = afs_linux_lock,
785 #ifdef STRUCT_FILE_OPERATIONS_HAS_FLOCK
786 .flock = afs_linux_flock,
791 /**********************************************************************
792 * AFS Linux dentry operations
793 **********************************************************************/
795 /* check_bad_parent() : Checks if this dentry's vcache is a root vcache
796 * that has its mvid (parent dir's fid) pointer set to the wrong directory
797 * due to being mounted in multiple points at once. If so, check_bad_parent()
798 * calls afs_lookup() to correct the vcache's mvid, as well as the volume's
799 * dotdotfid and mtpoint fid members.
801 * dp - dentry to be checked.
805 * This dentry's vcache's mvid will be set to the correct parent directory's
807 * This root vnode's volume will have its dotdotfid and mtpoint fids set
808 * to the correct parent and mountpoint fids.
812 check_bad_parent(struct dentry *dp)
815 struct vcache *vcp = VTOAFS(dp->d_inode), *avc = NULL;
816 struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
818 if (vcp->mvid->Fid.Volume != pvc->f.fid.Fid.Volume) { /* bad parent */
821 /* force a lookup, so vcp->mvid is fixed up */
822 afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
823 if (!avc || vcp != avc) { /* bad, very bad.. */
824 afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
825 "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
826 ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
827 ICL_TYPE_POINTER, dp);
830 AFS_RELE(AFSTOV(avc));
837 /* afs_linux_revalidate
838 * Ensure vcache is stat'd before use. Return 0 if entry is valid.
841 afs_linux_revalidate(struct dentry *dp)
844 struct vcache *vcp = VTOAFS(dp->d_inode);
848 if (afs_shuttingdown)
851 #ifdef AFS_LINUX24_ENV
857 /* Make this a fast path (no crref), since it's called so often. */
858 if (vcp->f.states & CStatd) {
860 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
861 check_bad_parent(dp); /* check and correct mvid */
864 #ifdef AFS_LINUX24_ENV
871 /* This avoids the crref when we don't have to do it. Watch for
872 * changes in afs_getattr that don't get replicated here!
874 if (vcp->f.states & CStatd &&
875 (!afs_fakestat_enable || vcp->mvstat != 1) &&
877 (vType(vcp) == VDIR || vType(vcp) == VLNK)) {
878 code = afs_CopyOutAttrs(vcp, &vattr);
881 code = afs_getattr(vcp, &vattr, credp);
885 afs_fill_inode(AFSTOV(vcp), &vattr);
888 #ifdef AFS_LINUX24_ENV
889 maybe_unlock_kernel();
892 return afs_convert_code(code);
895 /* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
896 * In kernels 2.2.10 and above, we are passed an additional flags var which
897 * may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
898 * we are advised to follow the entry if it is a link or to make sure that
899 * it is a directory. But since the kernel itself checks these possibilities
900 * later on, we shouldn't have to do it until later. Perhaps in the future..
903 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
904 #ifdef DOP_REVALIDATE_TAKES_NAMEIDATA
905 afs_linux_dentry_revalidate(struct dentry *dp, struct nameidata *nd)
907 afs_linux_dentry_revalidate(struct dentry *dp, int flags)
910 afs_linux_dentry_revalidate(struct dentry *dp)
914 cred_t *credp = NULL;
915 struct vcache *vcp, *pvcp, *tvc = NULL;
917 struct afs_fakestat_state fakestate;
919 #ifdef AFS_LINUX24_ENV
923 afs_InitFakeStat(&fakestate);
927 vcp = VTOAFS(dp->d_inode);
928 pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
930 if (vcp == afs_globalVp)
933 if (vcp->mvstat == 1) { /* mount point */
934 if (vcp->mvid && (vcp->f.states & CMValid)) {
937 struct vrequest treq;
940 code = afs_InitReq(&treq, credp);
942 #ifdef AFS_DARWIN_ENV
943 (strcmp(dp->d_name.name, ".DS_Store") == 0) ||
944 (strcmp(dp->d_name.name, "Contents") == 0) ||
946 (strcmp(dp->d_name.name, ".directory") == 0)) {
950 code = afs_TryEvalFakeStat(&vcp, &fakestate, &treq);
952 code = afs_EvalFakeStat(&vcp, &fakestate, &treq);
953 if ((tryEvalOnly && vcp->mvstat == 1) || code) {
954 /* a mount point, not yet replaced by its directory */
959 if (*dp->d_name.name != '/' && vcp->mvstat == 2) /* root vnode */
960 check_bad_parent(dp); /* check and correct mvid */
963 /* If the last looker changes, we should make sure the current
964 * looker still has permission to examine this file. This would
965 * always require a crref() which would be "slow".
967 if (vcp->last_looker != treq.uid) {
968 if (!afs_AccessOK(vcp, (vType(vcp) == VREG) ? PRSFS_READ : PRSFS_LOOKUP, &treq, CHECK_MODE_BITS))
971 vcp->last_looker = treq.uid;
975 /* If the parent's DataVersion has changed or the vnode
976 * is longer valid, we need to do a full lookup. VerifyVCache
977 * isn't enough since the vnode may have been renamed.
980 if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time || !(vcp->f.states & CStatd)) {
983 afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
984 if (!tvc || tvc != vcp)
987 if (afs_getattr(vcp, &vattr, credp))
990 vattr2inode(AFSTOV(vcp), &vattr);
991 dp->d_time = hgetlo(pvcp->f.m.DataVersion);
994 /* should we always update the attributes at this point? */
995 /* unlikely--the vcache entry hasn't changed */
999 pvcp = VTOAFS(dp->d_parent->d_inode); /* dget_parent()? */
1000 if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time)
1004 /* No change in parent's DataVersion so this negative
1005 * lookup is still valid. BUT, if a server is down a
1006 * negative lookup can result so there should be a
1007 * liftime as well. For now, always expire.
1020 afs_PutFakeStat(&fakestate);
1026 shrink_dcache_parent(dp);
1029 #ifdef AFS_LINUX24_ENV
1030 maybe_unlock_kernel();
1035 if (have_submounts(dp))
1043 afs_dentry_iput(struct dentry *dp, struct inode *ip)
1045 struct vcache *vcp = VTOAFS(ip);
1048 if (!AFS_IS_DISCONNECTED || (vcp->f.states & CUnlinked)) {
1049 (void) afs_InactiveVCache(vcp, NULL);
1052 #ifdef DCACHE_NFSFS_RENAMED
1053 dp->d_flags &= ~DCACHE_NFSFS_RENAMED;
1060 afs_dentry_delete(struct dentry *dp)
1062 if (dp->d_inode && (VTOAFS(dp->d_inode)->f.states & CUnlinked))
1063 return 1; /* bad inode? */
1068 struct dentry_operations afs_dentry_operations = {
1069 .d_revalidate = afs_linux_dentry_revalidate,
1070 .d_delete = afs_dentry_delete,
1071 .d_iput = afs_dentry_iput,
1074 /**********************************************************************
1075 * AFS Linux inode operations
1076 **********************************************************************/
1080 * Merely need to set enough of vattr to get us through the create. Note
1081 * that the higher level code (open_namei) will take care of any tuncation
1082 * explicitly. Exclusive open is also taken care of in open_namei.
1084 * name is in kernel space at this point.
1087 #ifdef IOP_CREATE_TAKES_NAMEIDATA
1088 afs_linux_create(struct inode *dip, struct dentry *dp, int mode,
1089 struct nameidata *nd)
1091 afs_linux_create(struct inode *dip, struct dentry *dp, int mode)
1095 cred_t *credp = crref();
1096 const char *name = dp->d_name.name;
1101 vattr.va_mode = mode;
1102 vattr.va_type = mode & S_IFMT;
1105 code = afs_create(VTOAFS(dip), (char *)name, &vattr, NONEXCL, mode,
1109 struct inode *ip = AFSTOV(vcp);
1111 afs_getattr(vcp, &vattr, credp);
1112 afs_fill_inode(ip, &vattr);
1113 insert_inode_hash(ip);
1114 dp->d_op = &afs_dentry_operations;
1115 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1116 d_instantiate(dp, ip);
1121 return afs_convert_code(code);
1124 /* afs_linux_lookup */
1125 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1126 static struct dentry *
1127 #ifdef IOP_LOOKUP_TAKES_NAMEIDATA
1128 afs_linux_lookup(struct inode *dip, struct dentry *dp,
1129 struct nameidata *nd)
1131 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1135 afs_linux_lookup(struct inode *dip, struct dentry *dp)
1138 cred_t *credp = crref();
1139 struct vcache *vcp = NULL;
1140 const char *comp = dp->d_name.name;
1141 struct inode *ip = NULL;
1145 code = afs_lookup(VTOAFS(dip), (char *)comp, &vcp, credp);
1151 afs_getattr(vcp, &vattr, credp);
1152 afs_fill_inode(ip, &vattr);
1154 #ifdef HAVE_KERNEL_HLIST_UNHASHED
1155 hlist_unhashed(&ip->i_hash)
1157 ip->i_hash.prev == NULL
1160 insert_inode_hash(ip);
1162 dp->d_op = &afs_dentry_operations;
1163 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1166 #if defined(AFS_LINUX24_ENV)
1167 if (ip && S_ISDIR(ip->i_mode)) {
1168 struct dentry *alias;
1170 /* Try to invalidate an existing alias in favor of our new one */
1171 alias = d_find_alias(ip);
1173 if (d_invalidate(alias) == 0) {
1187 /* It's ok for the file to not be found. That's noted by the caller by
1188 * seeing that the dp->d_inode field is NULL.
1190 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,2,10)
1194 return ERR_PTR(afs_convert_code(code));
1198 return afs_convert_code(code);
1203 afs_linux_link(struct dentry *olddp, struct inode *dip, struct dentry *newdp)
1206 cred_t *credp = crref();
1207 const char *name = newdp->d_name.name;
1208 struct inode *oldip = olddp->d_inode;
1210 /* If afs_link returned the vnode, we could instantiate the
1211 * dentry. Since it's not, we drop this one and do a new lookup.
1216 code = afs_link(VTOAFS(oldip), VTOAFS(dip), (char *)name, credp);
1220 return afs_convert_code(code);
1224 afs_linux_unlink(struct inode *dip, struct dentry *dp)
1227 cred_t *credp = crref();
1228 const char *name = dp->d_name.name;
1229 struct vcache *tvc = VTOAFS(dp->d_inode);
1231 if (VREFCOUNT(tvc) > 1 && tvc->opens > 0
1232 && !(tvc->f.states & CUnlinked)) {
1233 struct dentry *__dp;
1243 osi_FreeSmallSpace(__name);
1244 __name = afs_newname();
1247 __dp = lookup_one_len(__name, dp->d_parent, strlen(__name));
1251 } while (__dp->d_inode != NULL);
1254 code = afs_rename(VTOAFS(dip), (char *)dp->d_name.name, VTOAFS(dip), (char *)__dp->d_name.name, credp);
1256 tvc->mvid = (void *) __name;
1259 crfree(tvc->uncred);
1261 tvc->uncred = credp;
1262 tvc->f.states |= CUnlinked;
1263 #ifdef DCACHE_NFSFS_RENAMED
1264 dp->d_flags |= DCACHE_NFSFS_RENAMED;
1267 osi_FreeSmallSpace(__name);
1272 __dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1281 code = afs_remove(VTOAFS(dip), (char *)name, credp);
1287 return afs_convert_code(code);
1292 afs_linux_symlink(struct inode *dip, struct dentry *dp, const char *target)
1295 cred_t *credp = crref();
1297 const char *name = dp->d_name.name;
1299 /* If afs_symlink returned the vnode, we could instantiate the
1300 * dentry. Since it's not, we drop this one and do a new lookup.
1306 code = afs_symlink(VTOAFS(dip), (char *)name, &vattr, (char *)target, credp);
1309 return afs_convert_code(code);
1313 afs_linux_mkdir(struct inode *dip, struct dentry *dp, int mode)
1316 cred_t *credp = crref();
1317 struct vcache *tvcp = NULL;
1319 const char *name = dp->d_name.name;
1322 vattr.va_mask = ATTR_MODE;
1323 vattr.va_mode = mode;
1325 code = afs_mkdir(VTOAFS(dip), (char *)name, &vattr, &tvcp, credp);
1328 struct inode *ip = AFSTOV(tvcp);
1330 afs_getattr(tvcp, &vattr, credp);
1331 afs_fill_inode(ip, &vattr);
1333 dp->d_op = &afs_dentry_operations;
1334 dp->d_time = hgetlo(VTOAFS(dip)->f.m.DataVersion);
1335 d_instantiate(dp, ip);
1340 return afs_convert_code(code);
1344 afs_linux_rmdir(struct inode *dip, struct dentry *dp)
1347 cred_t *credp = crref();
1348 const char *name = dp->d_name.name;
1350 /* locking kernel conflicts with glock? */
1353 code = afs_rmdir(VTOAFS(dip), (char *)name, credp);
1356 /* Linux likes to see ENOTEMPTY returned from an rmdir() syscall
1357 * that failed because a directory is not empty. So, we map
1358 * EEXIST to ENOTEMPTY on linux.
1360 if (code == EEXIST) {
1369 return afs_convert_code(code);
1374 afs_linux_rename(struct inode *oldip, struct dentry *olddp,
1375 struct inode *newip, struct dentry *newdp)
1378 cred_t *credp = crref();
1379 const char *oldname = olddp->d_name.name;
1380 const char *newname = newdp->d_name.name;
1381 struct dentry *rehash = NULL;
1383 if (!list_empty(&newdp->d_hash)) {
1388 #if defined(AFS_LINUX24_ENV)
1389 if (atomic_read(&olddp->d_count) > 1)
1390 shrink_dcache_parent(olddp);
1394 code = afs_rename(VTOAFS(oldip), (char *)oldname, VTOAFS(newip), (char *)newname, credp);
1398 olddp->d_time = 0; /* force to revalidate */
1404 return afs_convert_code(code);
1408 /* afs_linux_ireadlink
1409 * Internal readlink which can return link contents to user or kernel space.
1410 * Note that the buffer is NOT supposed to be null-terminated.
1413 afs_linux_ireadlink(struct inode *ip, char *target, int maxlen, uio_seg_t seg)
1416 cred_t *credp = crref();
1420 setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
1421 code = afs_readlink(VTOAFS(ip), &tuio, credp);
1425 return maxlen - tuio.uio_resid;
1427 return afs_convert_code(code);
1430 #if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
1431 /* afs_linux_readlink
1432 * Fill target (which is in user space) with contents of symlink.
1435 afs_linux_readlink(struct dentry *dp, char *target, int maxlen)
1438 struct inode *ip = dp->d_inode;
1441 code = afs_linux_ireadlink(ip, target, maxlen, AFS_UIOUSER);
1447 /* afs_linux_follow_link
1448 * a file system dependent link following routine.
1450 #if defined(AFS_LINUX24_ENV)
1451 static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
1456 name = osi_Alloc(PATH_MAX);
1462 code = afs_linux_ireadlink(dentry->d_inode, name, PATH_MAX - 1, AFS_UIOSYS);
1470 code = vfs_follow_link(nd, name);
1473 osi_Free(name, PATH_MAX);
1478 #else /* !defined(AFS_LINUX24_ENV) */
1480 static struct dentry *
1481 afs_linux_follow_link(struct dentry *dp, struct dentry *basep,
1482 unsigned int follow)
1490 name = osi_Alloc(PATH_MAX + 1);
1494 return ERR_PTR(-EIO);
1497 code = afs_linux_ireadlink(dp->d_inode, name, PATH_MAX, AFS_UIOSYS);
1502 if (code < -MAX_ERRNO)
1503 res = ERR_PTR(-EIO);
1505 res = ERR_PTR(code);
1508 res = lookup_dentry(name, basep, follow);
1512 osi_Free(name, PATH_MAX + 1);
1516 #endif /* AFS_LINUX24_ENV */
1517 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
1519 #if defined(AFS_CACHE_BYPASS)
1521 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
1524 afs_linux_can_bypass(struct inode *ip) {
1525 switch(cache_bypass_strategy) {
1526 case NEVER_BYPASS_CACHE:
1528 case ALWAYS_BYPASS_CACHE:
1530 case LARGE_FILES_BYPASS_CACHE:
1531 if(i_size_read(ip) > cache_bypass_threshold)
1539 afs_linux_cache_bypass_read(struct file *fp, struct address_space *mapping,
1540 struct list_head *page_list, unsigned num_pages)
1545 struct iovec* iovecp;
1546 struct nocache_read_request *ancr;
1547 struct page *pp, *ppt;
1548 struct pagevec lrupv;
1552 struct inode *ip = FILE_INODE(fp);
1553 struct vcache *avc = VTOAFS(ip);
1554 afs_int32 bypasscache = 0; /* bypass for this read */
1555 afs_int32 base_index = 0;
1556 afs_int32 page_count = 0;
1559 /* background thread must free: iovecp, auio, ancr */
1560 iovecp = osi_Alloc(num_pages * sizeof(struct iovec));
1562 auio = osi_Alloc(sizeof(uio_t));
1563 auio->uio_iov = iovecp;
1564 auio->uio_iovcnt = num_pages;
1565 auio->uio_flag = UIO_READ;
1566 auio->uio_seg = AFS_UIOSYS;
1567 auio->uio_resid = num_pages * PAGE_SIZE;
1569 ancr = osi_Alloc(sizeof(struct nocache_read_request));
1571 ancr->offset = auio->uio_offset;
1572 ancr->length = auio->uio_resid;
1574 pagevec_init(&lrupv, 0);
1576 for(page_ix = 0; page_ix < num_pages; ++page_ix) {
1578 if(list_empty(page_list))
1581 pp = list_entry(page_list->prev, struct page, lru);
1582 /* If we allocate a page and don't remove it from page_list,
1583 * the page cache gets upset. */
1585 isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
1586 if(pp->index > isize) {
1593 offset = ((loff_t) pp->index) << PAGE_CACHE_SHIFT;
1594 auio->uio_offset = offset;
1595 base_index = pp->index;
1597 iovecp[page_ix].iov_len = PAGE_SIZE;
1598 code = add_to_page_cache(pp, mapping, pp->index, GFP_KERNEL);
1599 if(base_index != pp->index) {
1602 page_cache_release(pp);
1603 iovecp[page_ix].iov_base = (void *) 0;
1611 page_cache_release(pp);
1612 iovecp[page_ix].iov_base = (void *) 0;
1615 if(!PageLocked(pp)) {
1619 /* save the page for background map */
1620 iovecp[page_ix].iov_base = (void*) pp;
1622 /* and put it on the LRU cache */
1623 if (!pagevec_add(&lrupv, pp))
1624 __pagevec_lru_add(&lrupv);
1628 /* If there were useful pages in the page list, make sure all pages
1629 * are in the LRU cache, then schedule the read */
1631 pagevec_lru_add(&lrupv);
1633 code = afs_ReadNoCache(avc, ancr, credp);
1636 /* If there is nothing for the background thread to handle,
1637 * it won't be freeing the things that we never gave it */
1638 osi_Free(iovecp, num_pages * sizeof(struct iovec));
1639 osi_Free(auio, sizeof(uio_t));
1640 osi_Free(ancr, sizeof(struct nocache_read_request));
1642 /* we do not flush, release, or unmap pages--that will be
1643 * done for us by the background thread as each page comes in
1644 * from the fileserver */
1646 return afs_convert_code(code);
1649 #endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0) */
1650 #endif /* defined(AFS_CACHE_BYPASS */
1653 afs_linux_read_cache(struct file *cachefp, struct page *page,
1654 int chunk, struct pagevec *lrupv,
1655 struct afs_pagecopy_task *task) {
1656 loff_t offset = page_offset(page);
1657 struct page *newpage, *cachepage;
1658 struct address_space *cachemapping;
1662 cachemapping = cachefp->f_dentry->d_inode->i_mapping;
1666 /* From our offset, we now need to work out which page in the disk
1667 * file it corresponds to. This will be fun ... */
1668 pageindex = (offset - AFS_CHUNKTOBASE(chunk)) >> PAGE_CACHE_SHIFT;
1670 while (cachepage == NULL) {
1671 cachepage = find_get_page(cachemapping, pageindex);
1674 newpage = page_cache_alloc_cold(cachemapping);
1680 code = add_to_page_cache(newpage, cachemapping,
1681 pageindex, GFP_KERNEL);
1683 cachepage = newpage;
1686 page_cache_get(cachepage);
1687 if (!pagevec_add(lrupv, cachepage))
1688 __pagevec_lru_add_file(lrupv);
1691 page_cache_release(newpage);
1693 if (code != -EEXIST)
1697 lock_page(cachepage);
1701 if (!PageUptodate(cachepage)) {
1702 ClearPageError(cachepage);
1703 code = cachemapping->a_ops->readpage(NULL, cachepage);
1704 if (!code && !task) {
1705 wait_on_page_locked(cachepage);
1708 unlock_page(cachepage);
1712 if (PageUptodate(cachepage)) {
1713 copy_highpage(page, cachepage);
1714 flush_dcache_page(page);
1715 SetPageUptodate(page);
1718 afs_pagecopy_queue_page(task, cachepage, page);
1730 page_cache_release(cachepage);
1736 afs_linux_readpage_fastpath(struct file *fp, struct page *pp, int *codep)
1738 loff_t offset = page_offset(pp);
1739 struct inode *ip = FILE_INODE(fp);
1740 struct vcache *avc = VTOAFS(ip);
1742 struct file *cacheFp = NULL;
1745 struct pagevec lrupv;
1747 /* Not a UFS cache, don't do anything */
1748 if (cacheDiskType != AFS_FCACHE_TYPE_UFS)
1751 /* Can't do anything if the vcache isn't statd , or if the read
1752 * crosses a chunk boundary.
1754 if (!(avc->f.states & CStatd) ||
1755 AFS_CHUNK(offset) != AFS_CHUNK(offset + PAGE_SIZE)) {
1759 ObtainWriteLock(&avc->lock, 911);
1761 /* XXX - See if hinting actually makes things faster !!! */
1763 /* See if we have a suitable entry already cached */
1767 /* We need to lock xdcache, then dcache, to handle situations where
1768 * the hint is on the free list. However, we can't safely do this
1769 * according to the locking hierarchy. So, use a non blocking lock.
1771 ObtainReadLock(&afs_xdcache);
1772 dcLocked = ( 0 == NBObtainReadLock(&tdc->lock));
1774 if (dcLocked && (tdc->index != NULLIDX)
1775 && !FidCmp(&tdc->f.fid, &avc->f.fid)
1776 && tdc->f.chunk == AFS_CHUNK(offset)
1777 && !(afs_indexFlags[tdc->index] & (IFFree | IFDiscarded))) {
1778 /* Bonus - the hint was correct */
1781 /* Only destroy the hint if its actually invalid, not if there's
1782 * just been a locking failure */
1784 ReleaseReadLock(&tdc->lock);
1791 ReleaseReadLock(&afs_xdcache);
1794 /* No hint, or hint is no longer valid - see if we can get something
1795 * directly from the dcache
1798 tdc = afs_FindDCache(avc, offset);
1801 ReleaseWriteLock(&avc->lock);
1806 ObtainReadLock(&tdc->lock);
1808 /* Is the dcache we've been given currently up to date */
1809 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
1810 (tdc->dflags & DFFetching)) {
1811 ReleaseWriteLock(&avc->lock);
1812 ReleaseReadLock(&tdc->lock);
1817 /* Update our hint for future abuse */
1820 /* Okay, so we've now got a cache file that is up to date */
1822 /* XXX - I suspect we should be locking the inodes before we use them! */
1824 cacheFp = afs_linux_raw_open(&tdc->f.inode, NULL);
1825 pagevec_init(&lrupv, 0);
1827 code = afs_linux_read_cache(cacheFp, pp, tdc->f.chunk, &lrupv, NULL);
1829 if (pagevec_count(&lrupv))
1830 __pagevec_lru_add_file(&lrupv);
1832 filp_close(cacheFp, NULL);
1835 ReleaseReadLock(&tdc->lock);
1836 ReleaseWriteLock(&avc->lock);
1843 /* afs_linux_readpage
1844 * all reads come through here. A strategy-like read call.
1847 afs_linux_readpage(struct file *fp, struct page *pp)
1850 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1852 afs_offs_t offset = ((loff_t) pp->index) << PAGE_CACHE_SHIFT;
1854 ulong address = afs_linux_page_address(pp);
1855 afs_offs_t offset = pageoff(pp);
1857 #if defined(AFS_CACHE_BYPASS)
1858 afs_int32 bypasscache = 0; /* bypass for this read */
1859 struct nocache_read_request *ancr;
1860 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
1865 struct iovec *iovecp;
1866 struct inode *ip = FILE_INODE(fp);
1867 afs_int32 cnt = page_count(pp);
1868 struct vcache *avc = VTOAFS(ip);
1871 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
1873 if (afs_linux_readpage_fastpath(fp, pp, &code)) {
1881 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1885 atomic_add(1, &pp->count);
1886 set_bit(PG_locked, &pp->flags); /* other bits? See mm.h */
1887 clear_bit(PG_error, &pp->flags);
1889 #if defined(AFS_CACHE_BYPASS)
1890 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
1891 /* If the page is past the end of the file, skip it */
1892 isize = (i_size_read(fp->f_mapping->host) - 1) >> PAGE_CACHE_SHIFT;
1893 if(pp->index > isize) {
1900 /* if bypasscache, receiver frees, else we do */
1901 auio = osi_Alloc(sizeof(uio_t));
1902 iovecp = osi_Alloc(sizeof(struct iovec));
1904 setup_uio(auio, iovecp, (char *)address, offset, PAGE_SIZE, UIO_READ,
1907 #if defined(AFS_CACHE_BYPASS)
1908 bypasscache = afs_linux_can_bypass(ip);
1910 /* In the new incarnation of selective caching, a file's caching policy
1911 * can change, eg because file size exceeds threshold, etc. */
1912 trydo_cache_transition(avc, credp, bypasscache);
1917 /* save the page for background map */
1918 auio->uio_iov->iov_base = (void*) pp;
1919 /* the background thread will free this */
1920 ancr = osi_Alloc(sizeof(struct nocache_read_request));
1922 ancr->offset = offset;
1923 ancr->length = PAGE_SIZE;
1925 maybe_lock_kernel();
1926 code = afs_ReadNoCache(avc, ancr, credp);
1927 maybe_unlock_kernel();
1929 goto done; /* skips release page, doing it in bg thread */
1933 #ifdef AFS_LINUX24_ENV
1934 maybe_lock_kernel();
1938 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
1939 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
1940 99999); /* not a possible code value */
1942 code = afs_rdwr(avc, auio, UIO_READ, 0, credp);
1944 afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
1945 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
1947 AFS_DISCON_UNLOCK();
1949 #ifdef AFS_LINUX24_ENV
1950 maybe_unlock_kernel();
1953 /* XXX valid for no-cache also? Check last bits of files... :)
1954 * Cognate code goes in afs_NoCacheFetchProc. */
1955 if (auio->uio_resid) /* zero remainder of page */
1956 memset((void *)(address + (PAGE_SIZE - auio->uio_resid)), 0,
1959 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1960 flush_dcache_page(pp);
1961 SetPageUptodate(pp);
1963 set_bit(PG_uptodate, &pp->flags);
1967 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
1971 clear_bit(PG_locked, &pp->flags);
1976 #if defined(AFS_CACHE_BYPASS)
1977 /* do not call afs_GetDCache if cache is bypassed */
1982 /* free if not bypassing cache */
1983 osi_Free(auio, sizeof(uio_t));
1984 osi_Free(iovecp, sizeof(struct iovec));
1986 if (!code && AFS_CHUNKOFFSET(offset) == 0) {
1988 struct vrequest treq;
1991 code = afs_InitReq(&treq, credp);
1992 if (!code && !NBObtainWriteLock(&avc->lock, 534)) {
1993 tdc = afs_FindDCache(avc, offset);
1995 if (!(tdc->mflags & DFNextStarted))
1996 afs_PrefetchChunk(avc, tdc, credp, &treq);
1999 ReleaseWriteLock(&avc->lock);
2004 #if defined(AFS_CACHE_BYPASS)
2008 return afs_convert_code(code);
2011 /* Readpages reads a number of pages for a particular file. We use
2012 * this to optimise the reading, by limiting the number of times upon which
2013 * we have to lookup, lock and open vcaches and dcaches
2017 afs_linux_readpages(struct file *fp, struct address_space *mapping,
2018 struct list_head *page_list, unsigned int num_pages)
2020 struct inode *inode = mapping->host;
2021 struct vcache *avc = VTOAFS(inode);
2023 struct file *cacheFp = NULL;
2025 unsigned int page_idx;
2027 struct pagevec lrupv;
2028 struct afs_pagecopy_task *task;
2030 #if defined(AFS_CACHE_BYPASS)
2031 bypasscache = afs_linux_can_bypass(ip);
2033 /* In the new incarnation of selective caching, a file's caching policy
2034 * can change, eg because file size exceeds threshold, etc. */
2035 trydo_cache_transition(avc, credp, bypasscache);
2038 return afs_linux_cache_bypass_read(ip, mapping, page_list, num_pages);
2042 if ((code = afs_linux_VerifyVCache(avc, NULL))) {
2047 ObtainWriteLock(&avc->lock, 912);
2050 task = afs_pagecopy_init_task();
2053 pagevec_init(&lrupv, 0);
2054 for (page_idx = 0; page_idx < num_pages; page_idx++) {
2055 struct page *page = list_entry(page_list->prev, struct page, lru);
2056 list_del(&page->lru);
2057 offset = page_offset(page);
2059 if (tdc && tdc->f.chunk != AFS_CHUNK(offset)) {
2061 ReleaseReadLock(&tdc->lock);
2066 filp_close(cacheFp, NULL);
2071 if ((tdc = afs_FindDCache(avc, offset))) {
2072 ObtainReadLock(&tdc->lock);
2073 if (!hsame(avc->f.m.DataVersion, tdc->f.versionNo) ||
2074 (tdc->dflags & DFFetching)) {
2075 ReleaseReadLock(&tdc->lock);
2082 cacheFp = afs_linux_raw_open(&tdc->f.inode, NULL);
2085 if (tdc && !add_to_page_cache(page, mapping, page->index,
2087 page_cache_get(page);
2088 if (!pagevec_add(&lrupv, page))
2089 __pagevec_lru_add_file(&lrupv);
2091 afs_linux_read_cache(cacheFp, page, tdc->f.chunk, &lrupv, task);
2093 page_cache_release(page);
2095 if (pagevec_count(&lrupv))
2096 __pagevec_lru_add_file(&lrupv);
2099 filp_close(cacheFp, NULL);
2101 afs_pagecopy_put_task(task);
2105 ReleaseReadLock(&tdc->lock);
2109 ReleaseWriteLock(&avc->lock);
2114 #if defined(AFS_LINUX24_ENV)
2116 afs_linux_writepage_sync(struct inode *ip, struct page *pp,
2117 unsigned long offset, unsigned int count)
2119 struct vcache *vcp = VTOAFS(ip);
2128 buffer = kmap(pp) + offset;
2129 base = (((loff_t) pp->index) << PAGE_CACHE_SHIFT) + offset;
2132 maybe_lock_kernel();
2134 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2135 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2136 ICL_TYPE_INT32, 99999);
2138 ObtainWriteLock(&vcp->lock, 532);
2139 if (vcp->f.states & CPageWrite) {
2140 ReleaseWriteLock(&vcp->lock);
2142 maybe_unlock_kernel();
2145 /* should mark it dirty? */
2148 vcp->f.states |= CPageWrite;
2149 ReleaseWriteLock(&vcp->lock);
2151 setup_uio(&tuio, &iovec, buffer, base, count, UIO_WRITE, AFS_UIOSYS);
2153 code = afs_write(vcp, &tuio, f_flags, credp, 0);
2155 i_size_write(ip, vcp->f.m.Length);
2156 ip->i_blocks = ((vcp->f.m.Length + 1023) >> 10) << 1;
2158 ObtainWriteLock(&vcp->lock, 533);
2160 struct vrequest treq;
2162 if (!afs_InitReq(&treq, credp))
2163 code = afs_DoPartialWrite(vcp, &treq);
2165 code = code ? afs_convert_code(code) : count - tuio.uio_resid;
2167 vcp->f.states &= ~CPageWrite;
2168 ReleaseWriteLock(&vcp->lock);
2170 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2171 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2172 ICL_TYPE_INT32, code);
2175 maybe_unlock_kernel();
2184 #ifdef AOP_WRITEPAGE_TAKES_WRITEBACK_CONTROL
2185 afs_linux_writepage(struct page *pp, struct writeback_control *wbc)
2187 afs_linux_writepage(struct page *pp)
2190 struct address_space *mapping = pp->mapping;
2191 struct inode *inode;
2192 unsigned long end_index;
2193 unsigned offset = PAGE_CACHE_SIZE;
2196 if (PageLaunder(pp)) {
2197 return(fail_writepage(pp));
2200 inode = (struct inode *)mapping->host;
2201 end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
2204 if (pp->index < end_index)
2206 /* things got complicated... */
2207 offset = i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
2208 /* OK, are we completely out? */
2209 if (pp->index >= end_index + 1 || !offset)
2212 status = afs_linux_writepage_sync(inode, pp, 0, offset);
2213 SetPageUptodate(pp);
2215 if (status == offset)
2221 /* afs_linux_updatepage
2222 * What one would have thought was writepage - write dirty page to file.
2223 * Called from generic_file_write. buffer is still in user space. pagep
2224 * has been filled in with old data if we're updating less than a page.
2227 afs_linux_updatepage(struct file *fp, struct page *pp, unsigned long offset,
2228 unsigned int count, int sync)
2230 struct vcache *vcp = VTOAFS(FILE_INODE(fp));
2231 u8 *page_addr = (u8 *) afs_linux_page_address(pp);
2237 set_bit(PG_locked, &pp->flags);
2242 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2243 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2244 ICL_TYPE_INT32, 99999);
2245 setup_uio(&tuio, &iovec, page_addr + offset,
2246 (afs_offs_t) (pageoff(pp) + offset), count, UIO_WRITE,
2249 code = afs_write(vcp, &tuio, fp->f_flags, credp, 0);
2251 i_size_write(ip, vcp->f.m.Length);
2252 ip->i_blocks = ((vcp->f.m.Length + 1023) >> 10) << 1;
2255 struct vrequest treq;
2257 ObtainWriteLock(&vcp->lock, 533);
2258 vcp->f.m.Date = osi_Time(); /* set modification time */
2259 if (!afs_InitReq(&treq, credp))
2260 code = afs_DoPartialWrite(vcp, &treq);
2261 ReleaseWriteLock(&vcp->lock);
2264 code = code ? afs_convert_code(code) : count - tuio.uio_resid;
2265 afs_Trace4(afs_iclSetp, CM_TRACE_UPDATEPAGE, ICL_TYPE_POINTER, vcp,
2266 ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, page_count(pp),
2267 ICL_TYPE_INT32, code);
2269 AFS_DISCON_UNLOCK();
2273 clear_bit(PG_locked, &pp->flags);
2278 /* afs_linux_permission
2279 * Check access rights - returns error if can't check or permission denied.
2282 #ifdef IOP_PERMISSION_TAKES_NAMEIDATA
2283 afs_linux_permission(struct inode *ip, int mode, struct nameidata *nd)
2285 afs_linux_permission(struct inode *ip, int mode)
2289 cred_t *credp = crref();
2293 if (mode & MAY_EXEC)
2295 if (mode & MAY_READ)
2297 if (mode & MAY_WRITE)
2299 code = afs_access(VTOAFS(ip), tmp, credp);
2303 return afs_convert_code(code);
2306 #if defined(AFS_LINUX24_ENV) && !defined(HAVE_WRITE_BEGIN)
2308 afs_linux_commit_write(struct file *file, struct page *page, unsigned offset,
2313 code = afs_linux_writepage_sync(file->f_dentry->d_inode, page,
2314 offset, to - offset);
2321 afs_linux_prepare_write(struct file *file, struct page *page, unsigned from,
2324 /* sometime between 2.4.0 and 2.4.19, the callers of prepare_write began to
2325 call kmap directly instead of relying on us to do it */
2331 #if defined(HAVE_WRITE_BEGIN)
2333 afs_linux_write_end(struct file *file, struct address_space *mapping,
2334 loff_t pos, unsigned len, unsigned copied,
2335 struct page *page, void *fsdata)
2338 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
2340 code = afs_linux_writepage_sync(file->f_dentry->d_inode, page,
2343 page_cache_release(page);
2348 afs_linux_write_begin(struct file *file, struct address_space *mapping,
2349 loff_t pos, unsigned len, unsigned flags,
2350 struct page **pagep, void **fsdata)
2353 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2354 #if defined(HAVE_GRAB_CACHE_PAGE_WRITE_BEGIN)
2355 page = grab_cache_page_write_begin(mapping, index, flags);
2357 page = __grab_cache_page(mapping, index);
2366 static struct inode_operations afs_file_iops = {
2367 #if defined(AFS_LINUX24_ENV)
2368 .permission = afs_linux_permission,
2369 .revalidate = afs_linux_revalidate,
2370 .setattr = afs_notify_change,
2372 .default_file_ops = &afs_file_fops,
2373 .readpage = afs_linux_readpage,
2374 .revalidate = afs_linux_revalidate,
2375 .updatepage = afs_linux_updatepage,
2379 #if defined(AFS_LINUX24_ENV)
2380 static struct address_space_operations afs_file_aops = {
2381 .readpage = afs_linux_readpage,
2382 .readpages = afs_linux_readpages,
2383 .writepage = afs_linux_writepage,
2384 #if defined (HAVE_WRITE_BEGIN)
2385 .write_begin = afs_linux_write_begin,
2386 .write_end = afs_linux_write_end,
2388 .commit_write = afs_linux_commit_write,
2389 .prepare_write = afs_linux_prepare_write,
2395 /* Separate ops vector for directories. Linux 2.2 tests type of inode
2396 * by what sort of operation is allowed.....
2399 static struct inode_operations afs_dir_iops = {
2400 #if !defined(AFS_LINUX24_ENV)
2401 .default_file_ops = &afs_dir_fops,
2403 .setattr = afs_notify_change,
2405 .create = afs_linux_create,
2406 .lookup = afs_linux_lookup,
2407 .link = afs_linux_link,
2408 .unlink = afs_linux_unlink,
2409 .symlink = afs_linux_symlink,
2410 .mkdir = afs_linux_mkdir,
2411 .rmdir = afs_linux_rmdir,
2412 .rename = afs_linux_rename,
2413 .revalidate = afs_linux_revalidate,
2414 .permission = afs_linux_permission,
2417 /* We really need a separate symlink set of ops, since do_follow_link()
2418 * determines if it _is_ a link by checking if the follow_link op is set.
2420 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2422 afs_symlink_filler(struct file *file, struct page *page)
2424 struct inode *ip = (struct inode *)page->mapping->host;
2425 char *p = (char *)kmap(page);
2428 maybe_lock_kernel();
2430 code = afs_linux_ireadlink(ip, p, PAGE_SIZE, AFS_UIOSYS);
2435 p[code] = '\0'; /* null terminate? */
2436 maybe_unlock_kernel();
2438 SetPageUptodate(page);
2444 maybe_unlock_kernel();
2452 static struct address_space_operations afs_symlink_aops = {
2453 .readpage = afs_symlink_filler
2455 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
2457 static struct inode_operations afs_symlink_iops = {
2458 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2459 .readlink = page_readlink,
2460 #if defined(HAVE_KERNEL_PAGE_FOLLOW_LINK)
2461 .follow_link = page_follow_link,
2463 .follow_link = page_follow_link_light,
2464 .put_link = page_put_link,
2466 #else /* !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE) */
2467 .readlink = afs_linux_readlink,
2468 .follow_link = afs_linux_follow_link,
2469 #if !defined(AFS_LINUX24_ENV)
2470 .permission = afs_linux_permission,
2471 .revalidate = afs_linux_revalidate,
2473 #endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
2474 #if defined(AFS_LINUX24_ENV)
2475 .setattr = afs_notify_change,
2480 afs_fill_inode(struct inode *ip, struct vattr *vattr)
2484 vattr2inode(ip, vattr);
2486 /* Reset ops if symlink or directory. */
2487 if (S_ISREG(ip->i_mode)) {
2488 ip->i_op = &afs_file_iops;
2489 #if defined(AFS_LINUX24_ENV)
2490 ip->i_fop = &afs_file_fops;
2491 ip->i_data.a_ops = &afs_file_aops;
2494 } else if (S_ISDIR(ip->i_mode)) {
2495 ip->i_op = &afs_dir_iops;
2496 #if defined(AFS_LINUX24_ENV)
2497 ip->i_fop = &afs_dir_fops;
2500 } else if (S_ISLNK(ip->i_mode)) {
2501 ip->i_op = &afs_symlink_iops;
2502 #if defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
2503 ip->i_data.a_ops = &afs_symlink_aops;
2504 ip->i_mapping = &ip->i_data;