1 #define MSNFS /* HACK HACK */
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
19 #include <linux/config.h>
20 #include <linux/string.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
24 #include <linux/file.h>
25 #include <linux/mount.h>
26 #include <linux/major.h>
27 #include <linux/ext2_fs.h>
28 #include <linux/proc_fs.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/net.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/pagemap.h>
36 #include <linux/module.h>
37 #include <linux/namei.h>
38 #include <linux/vfs.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/dnotify.h>
49 #include <linux/posix_acl.h>
50 #include <linux/posix_acl_xattr.h>
51 #include <linux/xattr_acl.h>
52 #include <linux/xattr.h>
53 #include <linux/nfs4.h>
54 #include <linux/nfs4_acl.h>
55 #include <linux/nfsd_idmap.h>
56 #include <linux/security.h>
57 #endif /* CONFIG_NFSD_V4 */
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
65 /* We must ignore files (but only files) which might have mandatory
66 * locks on them because there is no way to know if the accesser has
69 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
72 * This is a cache of readahead params that help us choose the proper
73 * readahead strategy. Initially, we set all readahead parameters to 0
74 * and let the VFS handle things.
75 * If you increase the number of cached files very much, you'll need to
76 * add a hash table here.
79 struct raparms *p_next;
84 struct file_ra_state p_ra;
87 static struct raparms * raparml;
88 static struct raparms * raparm_cache;
91 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
93 * Returns -EAGAIN leaving *dpp and *expp unchanged,
94 * or nfs_ok having possibly changed *dpp and *expp
97 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
98 struct svc_export **expp)
100 struct svc_export *exp = *expp, *exp2 = NULL;
101 struct dentry *dentry = *dpp;
102 struct vfsmount *mnt = mntget(exp->ex_mnt);
103 struct dentry *mounts = dget(dentry);
106 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
108 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
115 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
116 /* successfully crossed mount point */
122 if (exp2) exp_put(exp2);
131 * Look up one component of a pathname.
132 * N.B. After this call _both_ fhp and resfh need an fh_put
134 * If the lookup would cross a mountpoint, and the mounted filesystem
135 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
136 * accepted as it stands and the mounted directory is
137 * returned. Otherwise the covered directory is returned.
138 * NOTE: this mountpoint crossing is not supported properly by all
139 * clients and is explicitly disallowed for NFSv3
140 * NeilBrown <neilb@cse.unsw.edu.au>
143 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
144 int len, struct svc_fh *resfh)
146 struct svc_export *exp;
147 struct dentry *dparent;
148 struct dentry *dentry;
151 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
153 /* Obtain dentry and export. */
154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
158 dparent = fhp->fh_dentry;
159 exp = fhp->fh_export;
164 /* Lookup the name, but don't follow links */
165 if (isdotent(name, len)) {
167 dentry = dget(dparent);
168 else if (dparent != exp->ex_dentry) {
169 dentry = dget_parent(dparent);
170 } else if (!EX_NOHIDE(exp))
171 dentry = dget(dparent); /* .. == . just like at / */
173 /* checking mountpoint crossing is very different when stepping up */
174 struct svc_export *exp2 = NULL;
176 struct vfsmount *mnt = mntget(exp->ex_mnt);
177 dentry = dget(dparent);
178 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
180 dp = dget_parent(dentry);
184 exp2 = exp_parent(exp->ex_client, mnt, dentry,
194 dentry = dget(dparent);
203 dentry = lookup_one_len(name, dparent, len);
204 err = PTR_ERR(dentry);
208 * check if we have crossed a mount point ...
210 if (d_mountpoint(dentry)) {
211 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
218 * Note: we compose the file handle now, but as the
219 * dentry may be negative, it may need to be updated.
221 err = fh_compose(resfh, exp, dentry, fhp);
222 if (!err && !dentry->d_inode)
235 * Set various file attributes.
236 * N.B. After this call fhp needs an fh_put
239 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
240 int check_guard, time_t guardtime)
242 struct dentry *dentry;
244 int accmode = MAY_SATTR;
250 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
251 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
252 if (iap->ia_valid & ATTR_SIZE)
256 err = fh_verify(rqstp, fhp, ftype, accmode);
257 if (err || !iap->ia_valid)
260 dentry = fhp->fh_dentry;
261 inode = dentry->d_inode;
263 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
264 * which only requires access, and "set-[ac]time-to-X" which
265 * requires ownership.
266 * So if it looks like it might be "set both to the same time which
267 * is close to now", and if inode_change_ok fails, then we
268 * convert to "set to now" instead of "set to explicit time"
270 * We only call inode_change_ok as the last test as technically
271 * it is not an interface that we should be using. It is only
272 * valid if the filesystem does not define it's own i_op->setattr.
274 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
275 #define MAX_TOUCH_TIME_ERROR (30*60)
276 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
277 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
279 /* Looks probable. Now just make sure time is in the right ballpark.
280 * Solaris, at least, doesn't seem to care what the time request is.
281 * We require it be within 30 minutes of now.
283 time_t delta = iap->ia_atime.tv_sec - get_seconds();
284 if (delta<0) delta = -delta;
285 if (delta < MAX_TOUCH_TIME_ERROR &&
286 inode_change_ok(inode, iap) != 0) {
287 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
288 * this will cause notify_change to set these times to "now"
290 iap->ia_valid &= ~BOTH_TIME_SET;
294 /* The size case is special. It changes the file as well as the attributes. */
295 if (iap->ia_valid & ATTR_SIZE) {
296 if (iap->ia_size < inode->i_size) {
297 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
303 * If we are changing the size of the file, then
304 * we need to break all leases.
306 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
307 if (err) /* ENOMEM or EWOULDBLOCK */
310 err = get_write_access(inode);
315 err = locks_verify_truncate(inode, NULL, iap->ia_size);
317 put_write_access(inode);
323 imode = inode->i_mode;
324 if (iap->ia_valid & ATTR_MODE) {
325 iap->ia_mode &= S_IALLUGO;
326 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
329 /* Revoke setuid/setgid bit on chown/chgrp */
330 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
331 iap->ia_valid |= ATTR_KILL_SUID;
332 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
333 iap->ia_valid |= ATTR_KILL_SGID;
335 /* Change the attributes. */
337 iap->ia_valid |= ATTR_CTIME;
339 err = nfserr_notsync;
340 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
342 err = notify_change(dentry, iap);
347 put_write_access(inode);
349 if (EX_ISSYNC(fhp->fh_export))
350 write_inode_now(inode, 1);
359 #if defined(CONFIG_NFSD_V4)
362 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
368 struct inode *inode = dentry->d_inode;
370 buflen = posix_acl_xattr_size(pacl->a_count);
371 buf = kmalloc(buflen, GFP_KERNEL);
376 len = posix_acl_to_xattr(pacl, buf, buflen);
383 if (inode->i_op && inode->i_op->setxattr) {
385 security_inode_setxattr(dentry, key, buf, len, 0);
386 error = inode->i_op->setxattr(dentry, key, buf, len, 0);
388 security_inode_post_setxattr(dentry, key, buf, len, 0);
397 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
398 struct nfs4_acl *acl)
401 struct dentry *dentry;
403 struct posix_acl *pacl = NULL, *dpacl = NULL;
404 unsigned int flags = 0;
407 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
411 dentry = fhp->fh_dentry;
412 inode = dentry->d_inode;
413 if (S_ISDIR(inode->i_mode))
414 flags = NFS4_ACL_DIR;
416 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
421 error = set_nfsv4_acl_one(dentry, pacl, XATTR_NAME_ACL_ACCESS);
427 error = set_nfsv4_acl_one(dentry, dpacl, XATTR_NAME_ACL_DEFAULT);
435 posix_acl_release(pacl);
436 posix_acl_release(dpacl);
439 error = nfserrno(error);
443 static struct posix_acl *
444 _get_posix_acl(struct dentry *dentry, char *key)
446 struct inode *inode = dentry->d_inode;
448 int buflen, error = 0;
449 struct posix_acl *pacl = NULL;
452 if (inode->i_op == NULL)
454 if (inode->i_op->getxattr == NULL)
457 error = security_inode_getxattr(dentry, key);
462 if (inode->i_op == NULL)
464 if (inode->i_op->getxattr == NULL)
467 error = security_inode_getxattr(dentry, key);
471 buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
473 error = buflen < 0 ? buflen : -ENODATA;
477 buf = kmalloc(buflen, GFP_KERNEL);
483 error = inode->i_op->getxattr(dentry, key, buf, buflen);
487 pacl = posix_acl_from_xattr(buf, buflen);
492 pacl = ERR_PTR(error);
497 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
499 struct inode *inode = dentry->d_inode;
501 struct posix_acl *pacl = NULL, *dpacl = NULL;
502 unsigned int flags = 0;
504 pacl = _get_posix_acl(dentry, XATTR_NAME_ACL_ACCESS);
505 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
506 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
508 error = PTR_ERR(pacl);
513 if (S_ISDIR(inode->i_mode)) {
514 dpacl = _get_posix_acl(dentry, XATTR_NAME_ACL_DEFAULT);
515 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
517 else if (IS_ERR(dpacl)) {
518 error = PTR_ERR(dpacl);
522 flags = NFS4_ACL_DIR;
525 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
527 error = PTR_ERR(*acl);
531 posix_acl_release(pacl);
532 posix_acl_release(dpacl);
536 #endif /* defined(CONFIG_NFS_V4) */
538 #ifdef CONFIG_NFSD_V3
540 * Check server access rights to a file system object
546 static struct accessmap nfs3_regaccess[] = {
547 { NFS3_ACCESS_READ, MAY_READ },
548 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
549 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
550 { NFS3_ACCESS_EXTEND, MAY_WRITE },
555 static struct accessmap nfs3_diraccess[] = {
556 { NFS3_ACCESS_READ, MAY_READ },
557 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
558 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
559 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
560 { NFS3_ACCESS_DELETE, MAY_REMOVE },
565 static struct accessmap nfs3_anyaccess[] = {
566 /* Some clients - Solaris 2.6 at least, make an access call
567 * to the server to check for access for things like /dev/null
568 * (which really, the server doesn't care about). So
569 * We provide simple access checking for them, looking
570 * mainly at mode bits, and we make sure to ignore read-only
573 { NFS3_ACCESS_READ, MAY_READ },
574 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
575 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
576 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
582 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
584 struct accessmap *map;
585 struct svc_export *export;
586 struct dentry *dentry;
587 u32 query, result = 0, sresult = 0;
590 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
594 export = fhp->fh_export;
595 dentry = fhp->fh_dentry;
597 if (S_ISREG(dentry->d_inode->i_mode))
598 map = nfs3_regaccess;
599 else if (S_ISDIR(dentry->d_inode->i_mode))
600 map = nfs3_diraccess;
602 map = nfs3_anyaccess;
606 for (; map->access; map++) {
607 if (map->access & query) {
610 sresult |= map->access;
612 err2 = nfsd_permission(export, dentry, map->how);
615 result |= map->access;
618 /* the following error codes just mean the access was not allowed,
619 * rather than an error occurred */
623 /* simply don't "or" in the access bit. */
633 *supported = sresult;
638 #endif /* CONFIG_NFSD_V3 */
643 * Open an existing file or directory.
644 * The access argument indicates the type of open (read/write/lock)
645 * N.B. After this call fhp needs an fh_put
648 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
649 int access, struct file **filp)
651 struct dentry *dentry;
653 int flags = O_RDONLY|O_LARGEFILE, err;
656 * If we get here, then the client has already done an "open",
657 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
658 * in case a chmod has now revoked permission.
660 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
664 dentry = fhp->fh_dentry;
665 inode = dentry->d_inode;
667 /* Disallow access to files with the append-only bit set or
668 * with mandatory locking enabled
671 if (IS_APPEND(inode) || IS_ISMNDLK(inode))
677 * Check to see if there are any leases on this file.
678 * This may block while leases are broken.
680 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
681 if (err) /* NOMEM or WOULDBLOCK */
684 if (access & MAY_WRITE) {
685 flags = O_WRONLY|O_LARGEFILE;
689 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
691 err = PTR_ERR(*filp);
703 nfsd_close(struct file *filp)
710 * As this calls fsync (not fdatasync) there is no need for a write_inode
713 inline void nfsd_dosync(struct file *filp, struct dentry *dp,
714 struct file_operations *fop)
716 struct inode *inode = dp->d_inode;
717 int (*fsync) (struct file *, struct dentry *, int);
719 filemap_fdatawrite(inode->i_mapping);
720 if (fop && (fsync = fop->fsync))
722 filemap_fdatawait(inode->i_mapping);
727 nfsd_sync(struct file *filp)
729 struct inode *inode = filp->f_dentry->d_inode;
730 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
732 nfsd_dosync(filp, filp->f_dentry, filp->f_op);
737 nfsd_sync_dir(struct dentry *dp)
739 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
743 * Obtain the readahead parameters for the file
744 * specified by (dev, ino).
746 static spinlock_t ra_lock = SPIN_LOCK_UNLOCKED;
748 static inline struct raparms *
749 nfsd_get_raparms(dev_t dev, ino_t ino)
751 struct raparms *ra, **rap, **frap = NULL;
755 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
756 if (ra->p_ino == ino && ra->p_dev == dev)
759 if (ra->p_count == 0)
762 depth = nfsdstats.ra_size*11/10;
764 spin_unlock(&ra_lock);
773 if (rap != &raparm_cache) {
775 ra->p_next = raparm_cache;
779 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
780 spin_unlock(&ra_lock);
785 * Grab and keep cached pages assosiated with a file in the svc_rqst
786 * so that they can be passed to the netowork sendmsg/sendpage routines
787 * directrly. They will be released after the sending has completed.
790 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
792 unsigned long count = desc->count;
793 struct svc_rqst *rqstp = desc->arg.data;
798 if (rqstp->rq_res.page_len == 0) {
800 rqstp->rq_respages[rqstp->rq_resused++] = page;
801 rqstp->rq_res.page_base = offset;
802 rqstp->rq_res.page_len = size;
803 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
805 rqstp->rq_respages[rqstp->rq_resused++] = page;
806 rqstp->rq_res.page_len += size;
808 rqstp->rq_res.page_len += size;
811 desc->count = count - size;
812 desc->written += size;
817 * Read data from a file. count must contain the requested read count
818 * on entry. On return, *count contains the number of bytes actually read.
819 * N.B. After this call fhp needs an fh_put
822 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
823 struct kvec *vec, int vlen, unsigned long *count)
831 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
835 inode = file->f_dentry->d_inode;
837 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
838 (!lock_may_read(inode, offset, *count)))
842 /* Get readahead parameters */
843 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
846 file->f_ra = ra->p_ra;
848 if (file->f_op->sendfile) {
849 svc_pushback_unused_pages(rqstp);
850 err = file->f_op->sendfile(file, &offset, *count,
851 nfsd_read_actor, rqstp);
855 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
859 /* Write back readahead params */
862 ra->p_ra = file->f_ra;
865 spin_unlock(&ra_lock);
869 nfsdstats.io_read += err;
872 dnotify_parent(file->f_dentry, DN_ACCESS);
882 * Write data to a file.
883 * The stable flag requests synchronous writes.
884 * N.B. After this call fhp needs an fh_put
887 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
888 struct kvec *vec, int vlen,
889 unsigned long cnt, int *stablep)
891 struct svc_export *exp;
893 struct dentry *dentry;
897 int stable = *stablep;
899 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
907 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
908 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
912 dentry = file->f_dentry;
913 inode = dentry->d_inode;
914 exp = fhp->fh_export;
917 * Request sync writes if
918 * - the sync export option has been set, or
919 * - the client requested O_SYNC behavior (NFSv3 feature).
920 * - The file system doesn't support fsync().
921 * When gathered writes have been configured for this volume,
922 * flushing the data to disk is handled separately below.
925 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
927 *stablep = 2; /* FILE_SYNC */
932 if (stable && !EX_WGATHER(exp))
933 file->f_flags |= O_SYNC;
935 /* Write the data. */
936 oldfs = get_fs(); set_fs(KERNEL_DS);
937 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
940 nfsdstats.io_write += cnt;
941 dnotify_parent(file->f_dentry, DN_MODIFY);
944 /* clear setuid/setgid flag after write */
945 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
947 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
950 notify_change(dentry, &ia);
954 if (err >= 0 && stable) {
955 static ino_t last_ino;
956 static dev_t last_dev;
959 * Gathered writes: If another process is currently
960 * writing to the file, there's a high chance
961 * this is another nfsd (triggered by a bulk write
962 * from a client's biod). Rather than syncing the
963 * file with each write request, we sleep for 10 msec.
965 * I don't know if this roughly approximates
966 * C. Juszak's idea of gathered writes, but it's a
967 * nice and simple solution (IMHO), and it seems to
970 if (EX_WGATHER(exp)) {
971 if (atomic_read(&inode->i_writecount) > 1
972 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
973 dprintk("nfsd: write defer %d\n", current->pid);
974 set_current_state(TASK_UNINTERRUPTIBLE);
975 schedule_timeout((HZ+99)/100);
976 dprintk("nfsd: write resume %d\n", current->pid);
979 if (inode->i_state & I_DIRTY) {
980 dprintk("nfsd: write sync %d\n", current->pid);
984 wake_up(&inode->i_wait);
987 last_ino = inode->i_ino;
988 last_dev = inode->i_sb->s_dev;
991 dprintk("nfsd: write complete err=%d\n", err);
1003 #ifdef CONFIG_NFSD_V3
1005 * Commit all pending writes to stable storage.
1006 * Strictly speaking, we could sync just the indicated file region here,
1007 * but there's currently no way we can ask the VFS to do so.
1009 * Unfortunately we cannot lock the file to make sure we return full WCC
1010 * data to the client, as locking happens lower down in the filesystem.
1013 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1014 loff_t offset, unsigned long count)
1019 if ((u64)count > ~(u64)offset)
1020 return nfserr_inval;
1022 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1024 if (EX_ISSYNC(fhp->fh_export)) {
1025 if (file->f_op && file->f_op->fsync) {
1028 err = nfserr_notsupp;
1035 #endif /* CONFIG_NFSD_V3 */
1038 * Create a file (regular, directory, device, fifo); UNIX sockets
1039 * not yet implemented.
1040 * If the response fh has been verified, the parent directory should
1041 * already be locked. Note that the parent directory is left locked.
1043 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1046 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1047 char *fname, int flen, struct iattr *iap,
1048 int type, dev_t rdev, struct svc_fh *resfhp)
1050 struct dentry *dentry, *dchild = NULL;
1058 if (isdotent(fname, flen))
1061 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1065 dentry = fhp->fh_dentry;
1066 dirp = dentry->d_inode;
1068 err = nfserr_notdir;
1069 if(!dirp->i_op || !dirp->i_op->lookup)
1072 * Check whether the response file handle has been verified yet.
1073 * If it has, the parent directory should already be locked.
1075 if (!resfhp->fh_dentry) {
1076 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1078 dchild = lookup_one_len(fname, dentry, flen);
1079 err = PTR_ERR(dchild);
1082 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1086 /* called from nfsd_proc_create */
1087 dchild = dget(resfhp->fh_dentry);
1088 if (!fhp->fh_locked) {
1089 /* not actually possible */
1091 "nfsd_create: parent %s/%s not locked!\n",
1092 dentry->d_parent->d_name.name,
1093 dentry->d_name.name);
1099 * Make sure the child dentry is still negative ...
1102 if (dchild->d_inode) {
1103 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1104 dentry->d_name.name, dchild->d_name.name);
1108 if (!(iap->ia_valid & ATTR_MODE))
1110 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1113 * Get the dir op function pointer.
1118 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1121 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1127 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1130 printk("nfsd: bad file type %o in nfsd_create\n", type);
1136 if (EX_ISSYNC(fhp->fh_export)) {
1137 nfsd_sync_dir(dentry);
1138 write_inode_now(dchild->d_inode, 1);
1142 /* Set file attributes. Mode has already been set and
1143 * setting uid/gid works only for root. Irix appears to
1144 * send along the gid when it tries to implement setgid
1145 * directories via NFS.
1148 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1149 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1151 * Update the file handle to get the new inode info.
1154 err = fh_update(resfhp);
1156 if (dchild && !IS_ERR(dchild))
1161 err = nfserrno(err);
1165 #ifdef CONFIG_NFSD_V3
1167 * NFSv3 version of nfsd_create
1170 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1171 char *fname, int flen, struct iattr *iap,
1172 struct svc_fh *resfhp, int createmode, u32 *verifier,
1175 struct dentry *dentry, *dchild = NULL;
1178 __u32 v_mtime=0, v_atime=0;
1185 if (isdotent(fname, flen))
1187 if (!(iap->ia_valid & ATTR_MODE))
1189 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1193 dentry = fhp->fh_dentry;
1194 dirp = dentry->d_inode;
1196 /* Get all the sanity checks out of the way before
1197 * we lock the parent. */
1198 err = nfserr_notdir;
1199 if(!dirp->i_op || !dirp->i_op->lookup)
1204 * Compose the response file handle.
1206 dchild = lookup_one_len(fname, dentry, flen);
1207 err = PTR_ERR(dchild);
1211 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1215 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1216 /* while the verifier would fit in mtime+atime,
1217 * solaris7 gets confused (bugid 4218508) if these have
1218 * the high bit set, so we use the mode as well
1220 v_mtime = verifier[0]&0x7fffffff;
1221 v_atime = verifier[1]&0x7fffffff;
1223 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1224 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1228 if (dchild->d_inode) {
1231 switch (createmode) {
1232 case NFS3_CREATE_UNCHECKED:
1233 if (! S_ISREG(dchild->d_inode->i_mode))
1236 /* in nfsv4, we need to treat this case a little
1237 * differently. we don't want to truncate the
1238 * file now; this would be wrong if the OPEN
1239 * fails for some other reason. furthermore,
1240 * if the size is nonzero, we should ignore it
1241 * according to spec!
1243 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1246 iap->ia_valid &= ATTR_SIZE;
1250 case NFS3_CREATE_EXCLUSIVE:
1251 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1252 && dchild->d_inode->i_atime.tv_sec == v_atime
1253 && dchild->d_inode->i_mode == v_mode
1254 && dchild->d_inode->i_size == 0 )
1257 case NFS3_CREATE_GUARDED:
1263 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1267 if (EX_ISSYNC(fhp->fh_export)) {
1268 nfsd_sync_dir(dentry);
1269 /* setattr will sync the child (or not) */
1273 * Update the filehandle to get the new inode info.
1275 err = fh_update(resfhp);
1279 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1280 /* Cram the verifier into atime/mtime/mode */
1281 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1282 | ATTR_MTIME_SET|ATTR_ATIME_SET
1284 /* XXX someone who knows this better please fix it for nsec */
1285 iap->ia_mtime.tv_sec = v_mtime;
1286 iap->ia_atime.tv_sec = v_atime;
1287 iap->ia_mtime.tv_nsec = 0;
1288 iap->ia_atime.tv_nsec = 0;
1289 iap->ia_mode = v_mode;
1292 /* Set file attributes.
1293 * Mode has already been set but we might need to reset it
1294 * for CREATE_EXCLUSIVE
1295 * Irix appears to send along the gid when it tries to
1296 * implement setgid directories via NFS. Clear out all that cruft.
1299 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1300 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1304 if (dchild && !IS_ERR(dchild))
1309 err = nfserrno(err);
1312 #endif /* CONFIG_NFSD_V3 */
1315 * Read a symlink. On entry, *lenp must contain the maximum path length that
1316 * fits into the buffer. On return, it contains the true length.
1317 * N.B. After this call fhp needs an fh_put
1320 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1322 struct dentry *dentry;
1323 struct inode *inode;
1327 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1331 dentry = fhp->fh_dentry;
1332 inode = dentry->d_inode;
1335 if (!inode->i_op || !inode->i_op->readlink)
1338 touch_atime(fhp->fh_export->ex_mnt, dentry);
1339 /* N.B. Why does this call need a get_fs()??
1340 * Remove the set_fs and watch the fireworks:-) --okir
1343 oldfs = get_fs(); set_fs(KERNEL_DS);
1344 err = inode->i_op->readlink(dentry, buf, *lenp);
1355 err = nfserrno(err);
1360 * Create a symlink and look up its inode
1361 * N.B. After this call _both_ fhp and resfhp need an fh_put
1364 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1365 char *fname, int flen,
1366 char *path, int plen,
1367 struct svc_fh *resfhp,
1370 struct dentry *dentry, *dnew;
1378 if (isdotent(fname, flen))
1381 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1385 dentry = fhp->fh_dentry;
1386 dnew = lookup_one_len(fname, dentry, flen);
1387 err = PTR_ERR(dnew);
1392 /* Only the MODE ATTRibute is even vaguely meaningful */
1393 if (iap && (iap->ia_valid & ATTR_MODE))
1394 mode = iap->ia_mode & S_IALLUGO;
1396 if (unlikely(path[plen] != 0)) {
1397 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1398 if (path_alloced == NULL)
1401 strncpy(path_alloced, path, plen);
1402 path_alloced[plen] = 0;
1403 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1404 kfree(path_alloced);
1407 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1410 if (EX_ISSYNC(fhp->fh_export))
1411 nfsd_sync_dir(dentry);
1413 err = nfserrno(err);
1416 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1418 if (err==0) err = cerr;
1423 err = nfserrno(err);
1429 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1432 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1433 char *name, int len, struct svc_fh *tfhp)
1435 struct dentry *ddir, *dnew, *dold;
1436 struct inode *dirp, *dest;
1439 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1442 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1450 if (isdotent(name, len))
1454 ddir = ffhp->fh_dentry;
1455 dirp = ddir->d_inode;
1457 dnew = lookup_one_len(name, ddir, len);
1458 err = PTR_ERR(dnew);
1462 dold = tfhp->fh_dentry;
1463 dest = dold->d_inode;
1465 err = vfs_link(dold, dirp, dnew);
1467 if (EX_ISSYNC(ffhp->fh_export)) {
1468 nfsd_sync_dir(ddir);
1469 write_inode_now(dest, 1);
1472 if (err == -EXDEV && rqstp->rq_vers == 2)
1475 err = nfserrno(err);
1484 err = nfserrno(err);
1490 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1493 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1494 struct svc_fh *tfhp, char *tname, int tlen)
1496 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1497 struct inode *fdir, *tdir;
1500 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1503 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1507 fdentry = ffhp->fh_dentry;
1508 fdir = fdentry->d_inode;
1510 tdentry = tfhp->fh_dentry;
1511 tdir = tdentry->d_inode;
1513 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1514 if (fdir->i_sb != tdir->i_sb)
1518 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1521 /* cannot use fh_lock as we need deadlock protective ordering
1522 * so do it by hand */
1523 trap = lock_rename(tdentry, fdentry);
1524 ffhp->fh_locked = tfhp->fh_locked = 1;
1528 odentry = lookup_one_len(fname, fdentry, flen);
1529 err = PTR_ERR(odentry);
1530 if (IS_ERR(odentry))
1534 if (!odentry->d_inode)
1537 if (odentry == trap)
1540 ndentry = lookup_one_len(tname, tdentry, tlen);
1541 err = PTR_ERR(ndentry);
1542 if (IS_ERR(ndentry))
1545 if (ndentry == trap)
1549 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1550 ((atomic_read(&odentry->d_count) > 1)
1551 || (atomic_read(&ndentry->d_count) > 1))) {
1555 err = vfs_rename(fdir, odentry, tdir, ndentry);
1556 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1557 nfsd_sync_dir(tdentry);
1558 nfsd_sync_dir(fdentry);
1567 err = nfserrno(err);
1569 /* we cannot reply on fh_unlock on the two filehandles,
1570 * as that would do the wrong thing if the two directories
1571 * were the same, so again we do it by hand
1573 fill_post_wcc(ffhp);
1574 fill_post_wcc(tfhp);
1575 unlock_rename(tdentry, fdentry);
1576 ffhp->fh_locked = tfhp->fh_locked = 0;
1583 * Unlink a file or directory
1584 * N.B. After this call fhp needs an fh_put
1587 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1588 char *fname, int flen)
1590 struct dentry *dentry, *rdentry;
1595 if (!flen || isdotent(fname, flen))
1597 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1602 dentry = fhp->fh_dentry;
1603 dirp = dentry->d_inode;
1605 rdentry = lookup_one_len(fname, dentry, flen);
1606 err = PTR_ERR(rdentry);
1607 if (IS_ERR(rdentry))
1610 if (!rdentry->d_inode) {
1617 type = rdentry->d_inode->i_mode & S_IFMT;
1619 if (type != S_IFDIR) { /* It's UNLINK */
1621 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1622 (atomic_read(&rdentry->d_count) > 1)) {
1626 err = vfs_unlink(dirp, rdentry);
1627 } else { /* It's RMDIR */
1628 err = vfs_rmdir(dirp, rdentry);
1635 if (EX_ISSYNC(fhp->fh_export))
1636 nfsd_sync_dir(dentry);
1642 err = nfserrno(err);
1647 * Read entries from a directory.
1648 * The NFSv3/4 verifier we ignore for now.
1651 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1652 struct readdir_cd *cdp, encode_dent_fn func)
1656 loff_t offset = *offsetp;
1658 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1662 offset = vfs_llseek(file, offset, 0);
1664 err = nfserrno((int)offset);
1669 * Read the directory entries. This silly loop is necessary because
1670 * readdir() is not guaranteed to fill up the entire buffer, but
1671 * may choose to do less.
1675 cdp->err = nfserr_eof; /* will be cleared on successful read */
1676 err = vfs_readdir(file, (filldir_t) func, cdp);
1677 } while (err >=0 && cdp->err == nfs_ok);
1679 err = nfserrno(err);
1682 *offsetp = vfs_llseek(file, 0, 1);
1684 if (err == nfserr_eof || err == nfserr_toosmall)
1685 err = nfs_ok; /* can still be found in ->err */
1693 * Get file system stats
1694 * N.B. After this call fhp needs an fh_put
1697 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1699 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1700 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1706 * Check for a user's access permissions to this inode.
1709 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1711 struct inode *inode = dentry->d_inode;
1717 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1719 (acc & MAY_READ)? " read" : "",
1720 (acc & MAY_WRITE)? " write" : "",
1721 (acc & MAY_EXEC)? " exec" : "",
1722 (acc & MAY_SATTR)? " sattr" : "",
1723 (acc & MAY_TRUNC)? " trunc" : "",
1724 (acc & MAY_LOCK)? " lock" : "",
1725 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1727 IS_IMMUTABLE(inode)? " immut" : "",
1728 IS_APPEND(inode)? " append" : "",
1729 IS_RDONLY(inode)? " ro" : "");
1730 dprintk(" owner %d/%d user %d/%d\n",
1731 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1734 /* Normally we reject any write/sattr etc access on a read-only file
1735 * system. But if it is IRIX doing check on write-access for a
1736 * device special file, we ignore rofs.
1738 if (!(acc & MAY_LOCAL_ACCESS))
1739 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1740 if (EX_RDONLY(exp) || IS_RDONLY(inode)
1741 || (exp && MNT_IS_RDONLY(exp->ex_mnt)))
1743 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1746 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1749 if (acc & MAY_LOCK) {
1750 /* If we cannot rely on authentication in NLM requests,
1751 * just allow locks, otherwise require read permission, or
1754 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1757 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1760 * The file owner always gets access permission for accesses that
1761 * would normally be checked at open time. This is to make
1762 * file access work even when the client has done a fchmod(fd, 0).
1764 * However, `cp foo bar' should fail nevertheless when bar is
1765 * readonly. A sensible way to do this might be to reject all
1766 * attempts to truncate a read-only file, because a creat() call
1767 * always implies file truncation.
1768 * ... but this isn't really fair. A process may reasonably call
1769 * ftruncate on an open file descriptor on a file with perm 000.
1770 * We must trust the client to do permission checking - using "ACCESS"
1773 if ((acc & MAY_OWNER_OVERRIDE) &&
1774 inode->i_uid == current->fsuid)
1777 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1779 /* Allow read access to binaries even when mode 111 */
1780 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1781 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1782 err = permission(inode, MAY_EXEC, NULL);
1784 return err? nfserrno(err) : 0;
1788 nfsd_racache_shutdown(void)
1792 dprintk("nfsd: freeing readahead buffers.\n");
1794 raparm_cache = raparml = NULL;
1797 * Initialize readahead param cache
1800 nfsd_racache_init(int cache_size)
1806 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1808 if (raparml != NULL) {
1809 dprintk("nfsd: allocating %d readahead buffers.\n",
1811 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1812 for (i = 0; i < cache_size - 1; i++) {
1813 raparml[i].p_next = raparml + i + 1;
1815 raparm_cache = raparml;
1818 "nfsd: Could not allocate memory read-ahead cache.\n");
1821 nfsdstats.ra_size = cache_size;