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 == -EWOULDBLOCK)
309 if (err) /* ENOMEM or EWOULDBLOCK */
312 err = get_write_access(inode);
317 err = locks_verify_truncate(inode, NULL, iap->ia_size);
319 put_write_access(inode);
325 imode = inode->i_mode;
326 if (iap->ia_valid & ATTR_MODE) {
327 iap->ia_mode &= S_IALLUGO;
328 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
331 /* Revoke setuid/setgid bit on chown/chgrp */
332 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
333 iap->ia_valid |= ATTR_KILL_SUID;
334 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
335 iap->ia_valid |= ATTR_KILL_SGID;
337 /* Change the attributes. */
339 iap->ia_valid |= ATTR_CTIME;
341 err = nfserr_notsync;
342 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
344 err = notify_change(dentry, iap);
349 put_write_access(inode);
351 if (EX_ISSYNC(fhp->fh_export))
352 write_inode_now(inode, 1);
361 #if defined(CONFIG_NFSD_V4)
364 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
370 struct inode *inode = dentry->d_inode;
372 buflen = posix_acl_xattr_size(pacl->a_count);
373 buf = kmalloc(buflen, GFP_KERNEL);
378 len = posix_acl_to_xattr(pacl, buf, buflen);
385 if (inode->i_op && inode->i_op->setxattr) {
387 security_inode_setxattr(dentry, key, buf, len, 0);
388 error = inode->i_op->setxattr(dentry, key, buf, len, 0);
390 security_inode_post_setxattr(dentry, key, buf, len, 0);
399 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
400 struct nfs4_acl *acl)
403 struct dentry *dentry;
405 struct posix_acl *pacl = NULL, *dpacl = NULL;
406 unsigned int flags = 0;
409 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
413 dentry = fhp->fh_dentry;
414 inode = dentry->d_inode;
415 if (S_ISDIR(inode->i_mode))
416 flags = NFS4_ACL_DIR;
418 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
423 error = set_nfsv4_acl_one(dentry, pacl, XATTR_NAME_ACL_ACCESS);
429 error = set_nfsv4_acl_one(dentry, dpacl, XATTR_NAME_ACL_DEFAULT);
437 posix_acl_release(pacl);
438 posix_acl_release(dpacl);
441 error = nfserrno(error);
445 static struct posix_acl *
446 _get_posix_acl(struct dentry *dentry, char *key)
448 struct inode *inode = dentry->d_inode;
450 int buflen, error = 0;
451 struct posix_acl *pacl = NULL;
454 if (inode->i_op == NULL)
456 if (inode->i_op->getxattr == NULL)
459 error = security_inode_getxattr(dentry, key);
463 buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
465 error = buflen < 0 ? buflen : -ENODATA;
469 buf = kmalloc(buflen, GFP_KERNEL);
475 error = inode->i_op->getxattr(dentry, key, buf, buflen);
479 pacl = posix_acl_from_xattr(buf, buflen);
484 pacl = ERR_PTR(error);
489 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
491 struct inode *inode = dentry->d_inode;
493 struct posix_acl *pacl = NULL, *dpacl = NULL;
494 unsigned int flags = 0;
496 pacl = _get_posix_acl(dentry, XATTR_NAME_ACL_ACCESS);
497 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
498 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
500 error = PTR_ERR(pacl);
505 if (S_ISDIR(inode->i_mode)) {
506 dpacl = _get_posix_acl(dentry, XATTR_NAME_ACL_DEFAULT);
507 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
509 else if (IS_ERR(dpacl)) {
510 error = PTR_ERR(dpacl);
514 flags = NFS4_ACL_DIR;
517 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
519 error = PTR_ERR(*acl);
523 posix_acl_release(pacl);
524 posix_acl_release(dpacl);
528 #endif /* defined(CONFIG_NFS_V4) */
530 #ifdef CONFIG_NFSD_V3
532 * Check server access rights to a file system object
538 static struct accessmap nfs3_regaccess[] = {
539 { NFS3_ACCESS_READ, MAY_READ },
540 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
541 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
542 { NFS3_ACCESS_EXTEND, MAY_WRITE },
547 static struct accessmap nfs3_diraccess[] = {
548 { NFS3_ACCESS_READ, MAY_READ },
549 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
550 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
551 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
552 { NFS3_ACCESS_DELETE, MAY_REMOVE },
557 static struct accessmap nfs3_anyaccess[] = {
558 /* Some clients - Solaris 2.6 at least, make an access call
559 * to the server to check for access for things like /dev/null
560 * (which really, the server doesn't care about). So
561 * We provide simple access checking for them, looking
562 * mainly at mode bits, and we make sure to ignore read-only
565 { NFS3_ACCESS_READ, MAY_READ },
566 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
567 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
568 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
574 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
576 struct accessmap *map;
577 struct svc_export *export;
578 struct dentry *dentry;
579 u32 query, result = 0, sresult = 0;
582 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
586 export = fhp->fh_export;
587 dentry = fhp->fh_dentry;
589 if (S_ISREG(dentry->d_inode->i_mode))
590 map = nfs3_regaccess;
591 else if (S_ISDIR(dentry->d_inode->i_mode))
592 map = nfs3_diraccess;
594 map = nfs3_anyaccess;
598 for (; map->access; map++) {
599 if (map->access & query) {
602 sresult |= map->access;
604 err2 = nfsd_permission(export, dentry, map->how);
607 result |= map->access;
610 /* the following error codes just mean the access was not allowed,
611 * rather than an error occurred */
615 /* simply don't "or" in the access bit. */
625 *supported = sresult;
630 #endif /* CONFIG_NFSD_V3 */
635 * Open an existing file or directory.
636 * The access argument indicates the type of open (read/write/lock)
637 * N.B. After this call fhp needs an fh_put
640 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
641 int access, struct file **filp)
643 struct dentry *dentry;
645 int flags = O_RDONLY|O_LARGEFILE, err;
648 * If we get here, then the client has already done an "open",
649 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
650 * in case a chmod has now revoked permission.
652 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
656 dentry = fhp->fh_dentry;
657 inode = dentry->d_inode;
659 /* Disallow write access to files with the append-only bit set
660 * or any access when mandatory locking enabled
663 if (IS_APPEND(inode) && (access & MAY_WRITE))
665 if (IS_ISMNDLK(inode))
672 * Check to see if there are any leases on this file.
673 * This may block while leases are broken.
675 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
676 if (err == -EWOULDBLOCK)
678 if (err) /* NOMEM or WOULDBLOCK */
681 if (access & MAY_WRITE) {
682 flags = O_WRONLY|O_LARGEFILE;
686 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
688 err = PTR_ERR(*filp);
700 nfsd_close(struct file *filp)
707 * As this calls fsync (not fdatasync) there is no need for a write_inode
710 inline void nfsd_dosync(struct file *filp, struct dentry *dp,
711 struct file_operations *fop)
713 struct inode *inode = dp->d_inode;
714 int (*fsync) (struct file *, struct dentry *, int);
716 filemap_fdatawrite(inode->i_mapping);
717 if (fop && (fsync = fop->fsync))
719 filemap_fdatawait(inode->i_mapping);
724 nfsd_sync(struct file *filp)
726 struct inode *inode = filp->f_dentry->d_inode;
727 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
729 nfsd_dosync(filp, filp->f_dentry, filp->f_op);
734 nfsd_sync_dir(struct dentry *dp)
736 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
740 * Obtain the readahead parameters for the file
741 * specified by (dev, ino).
743 static DEFINE_SPINLOCK(ra_lock);
745 static inline struct raparms *
746 nfsd_get_raparms(dev_t dev, ino_t ino)
748 struct raparms *ra, **rap, **frap = NULL;
752 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
753 if (ra->p_ino == ino && ra->p_dev == dev)
756 if (ra->p_count == 0)
759 depth = nfsdstats.ra_size*11/10;
761 spin_unlock(&ra_lock);
770 if (rap != &raparm_cache) {
772 ra->p_next = raparm_cache;
776 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
777 spin_unlock(&ra_lock);
782 * Grab and keep cached pages assosiated with a file in the svc_rqst
783 * so that they can be passed to the netowork sendmsg/sendpage routines
784 * directrly. They will be released after the sending has completed.
787 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
789 unsigned long count = desc->count;
790 struct svc_rqst *rqstp = desc->arg.data;
795 if (rqstp->rq_res.page_len == 0) {
797 rqstp->rq_respages[rqstp->rq_resused++] = page;
798 rqstp->rq_res.page_base = offset;
799 rqstp->rq_res.page_len = size;
800 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
802 rqstp->rq_respages[rqstp->rq_resused++] = page;
803 rqstp->rq_res.page_len += size;
805 rqstp->rq_res.page_len += size;
808 desc->count = count - size;
809 desc->written += size;
814 * Read data from a file. count must contain the requested read count
815 * on entry. On return, *count contains the number of bytes actually read.
816 * N.B. After this call fhp needs an fh_put
819 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
820 struct kvec *vec, int vlen, unsigned long *count)
828 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
832 inode = file->f_dentry->d_inode;
834 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
835 (!lock_may_read(inode, offset, *count)))
839 /* Get readahead parameters */
840 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
843 file->f_ra = ra->p_ra;
845 if (file->f_op->sendfile) {
846 svc_pushback_unused_pages(rqstp);
847 err = file->f_op->sendfile(file, &offset, *count,
848 nfsd_read_actor, rqstp);
852 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
856 /* Write back readahead params */
859 ra->p_ra = file->f_ra;
862 spin_unlock(&ra_lock);
866 nfsdstats.io_read += err;
869 dnotify_parent(file->f_dentry, DN_ACCESS);
879 * Write data to a file.
880 * The stable flag requests synchronous writes.
881 * N.B. After this call fhp needs an fh_put
884 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
885 struct kvec *vec, int vlen,
886 unsigned long cnt, int *stablep)
888 struct svc_export *exp;
890 struct dentry *dentry;
894 int stable = *stablep;
896 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
904 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
905 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
909 dentry = file->f_dentry;
910 inode = dentry->d_inode;
911 exp = fhp->fh_export;
914 * Request sync writes if
915 * - the sync export option has been set, or
916 * - the client requested O_SYNC behavior (NFSv3 feature).
917 * - The file system doesn't support fsync().
918 * When gathered writes have been configured for this volume,
919 * flushing the data to disk is handled separately below.
922 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
924 *stablep = 2; /* FILE_SYNC */
929 if (stable && !EX_WGATHER(exp))
930 file->f_flags |= O_SYNC;
932 /* Write the data. */
933 oldfs = get_fs(); set_fs(KERNEL_DS);
934 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
937 nfsdstats.io_write += cnt;
938 dnotify_parent(file->f_dentry, DN_MODIFY);
941 /* clear setuid/setgid flag after write */
942 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
944 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
947 notify_change(dentry, &ia);
951 if (err >= 0 && stable) {
952 static ino_t last_ino;
953 static dev_t last_dev;
956 * Gathered writes: If another process is currently
957 * writing to the file, there's a high chance
958 * this is another nfsd (triggered by a bulk write
959 * from a client's biod). Rather than syncing the
960 * file with each write request, we sleep for 10 msec.
962 * I don't know if this roughly approximates
963 * C. Juszak's idea of gathered writes, but it's a
964 * nice and simple solution (IMHO), and it seems to
967 if (EX_WGATHER(exp)) {
968 if (atomic_read(&inode->i_writecount) > 1
969 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
970 dprintk("nfsd: write defer %d\n", current->pid);
971 set_current_state(TASK_UNINTERRUPTIBLE);
972 schedule_timeout((HZ+99)/100);
973 dprintk("nfsd: write resume %d\n", current->pid);
976 if (inode->i_state & I_DIRTY) {
977 dprintk("nfsd: write sync %d\n", current->pid);
981 wake_up(&inode->i_wait);
984 last_ino = inode->i_ino;
985 last_dev = inode->i_sb->s_dev;
988 dprintk("nfsd: write complete err=%d\n", err);
1000 #ifdef CONFIG_NFSD_V3
1002 * Commit all pending writes to stable storage.
1003 * Strictly speaking, we could sync just the indicated file region here,
1004 * but there's currently no way we can ask the VFS to do so.
1006 * Unfortunately we cannot lock the file to make sure we return full WCC
1007 * data to the client, as locking happens lower down in the filesystem.
1010 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1011 loff_t offset, unsigned long count)
1016 if ((u64)count > ~(u64)offset)
1017 return nfserr_inval;
1019 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1021 if (EX_ISSYNC(fhp->fh_export)) {
1022 if (file->f_op && file->f_op->fsync) {
1025 err = nfserr_notsupp;
1032 #endif /* CONFIG_NFSD_V3 */
1035 * Create a file (regular, directory, device, fifo); UNIX sockets
1036 * not yet implemented.
1037 * If the response fh has been verified, the parent directory should
1038 * already be locked. Note that the parent directory is left locked.
1040 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1043 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1044 char *fname, int flen, struct iattr *iap,
1045 int type, dev_t rdev, struct svc_fh *resfhp)
1047 struct dentry *dentry, *dchild = NULL;
1055 if (isdotent(fname, flen))
1058 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1062 dentry = fhp->fh_dentry;
1063 dirp = dentry->d_inode;
1065 err = nfserr_notdir;
1066 if(!dirp->i_op || !dirp->i_op->lookup)
1069 * Check whether the response file handle has been verified yet.
1070 * If it has, the parent directory should already be locked.
1072 if (!resfhp->fh_dentry) {
1073 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1075 dchild = lookup_one_len(fname, dentry, flen);
1076 err = PTR_ERR(dchild);
1079 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1083 /* called from nfsd_proc_create */
1084 dchild = dget(resfhp->fh_dentry);
1085 if (!fhp->fh_locked) {
1086 /* not actually possible */
1088 "nfsd_create: parent %s/%s not locked!\n",
1089 dentry->d_parent->d_name.name,
1090 dentry->d_name.name);
1096 * Make sure the child dentry is still negative ...
1099 if (dchild->d_inode) {
1100 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1101 dentry->d_name.name, dchild->d_name.name);
1105 if (!(iap->ia_valid & ATTR_MODE))
1107 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1110 * Get the dir op function pointer.
1115 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1118 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1124 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1127 printk("nfsd: bad file type %o in nfsd_create\n", type);
1133 if (EX_ISSYNC(fhp->fh_export)) {
1134 nfsd_sync_dir(dentry);
1135 write_inode_now(dchild->d_inode, 1);
1139 /* Set file attributes. Mode has already been set and
1140 * setting uid/gid works only for root. Irix appears to
1141 * send along the gid when it tries to implement setgid
1142 * directories via NFS.
1145 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1146 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1148 * Update the file handle to get the new inode info.
1151 err = fh_update(resfhp);
1153 if (dchild && !IS_ERR(dchild))
1158 err = nfserrno(err);
1162 #ifdef CONFIG_NFSD_V3
1164 * NFSv3 version of nfsd_create
1167 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1168 char *fname, int flen, struct iattr *iap,
1169 struct svc_fh *resfhp, int createmode, u32 *verifier,
1172 struct dentry *dentry, *dchild = NULL;
1175 __u32 v_mtime=0, v_atime=0;
1182 if (isdotent(fname, flen))
1184 if (!(iap->ia_valid & ATTR_MODE))
1186 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1190 dentry = fhp->fh_dentry;
1191 dirp = dentry->d_inode;
1193 /* Get all the sanity checks out of the way before
1194 * we lock the parent. */
1195 err = nfserr_notdir;
1196 if(!dirp->i_op || !dirp->i_op->lookup)
1201 * Compose the response file handle.
1203 dchild = lookup_one_len(fname, dentry, flen);
1204 err = PTR_ERR(dchild);
1208 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1212 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1213 /* while the verifier would fit in mtime+atime,
1214 * solaris7 gets confused (bugid 4218508) if these have
1215 * the high bit set, so we use the mode as well
1217 v_mtime = verifier[0]&0x7fffffff;
1218 v_atime = verifier[1]&0x7fffffff;
1220 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1221 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1225 if (dchild->d_inode) {
1228 switch (createmode) {
1229 case NFS3_CREATE_UNCHECKED:
1230 if (! S_ISREG(dchild->d_inode->i_mode))
1233 /* in nfsv4, we need to treat this case a little
1234 * differently. we don't want to truncate the
1235 * file now; this would be wrong if the OPEN
1236 * fails for some other reason. furthermore,
1237 * if the size is nonzero, we should ignore it
1238 * according to spec!
1240 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1243 iap->ia_valid &= ATTR_SIZE;
1247 case NFS3_CREATE_EXCLUSIVE:
1248 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1249 && dchild->d_inode->i_atime.tv_sec == v_atime
1250 && dchild->d_inode->i_mode == v_mode
1251 && dchild->d_inode->i_size == 0 )
1254 case NFS3_CREATE_GUARDED:
1260 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1264 if (EX_ISSYNC(fhp->fh_export)) {
1265 nfsd_sync_dir(dentry);
1266 /* setattr will sync the child (or not) */
1270 * Update the filehandle to get the new inode info.
1272 err = fh_update(resfhp);
1276 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1277 /* Cram the verifier into atime/mtime/mode */
1278 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1279 | ATTR_MTIME_SET|ATTR_ATIME_SET
1281 /* XXX someone who knows this better please fix it for nsec */
1282 iap->ia_mtime.tv_sec = v_mtime;
1283 iap->ia_atime.tv_sec = v_atime;
1284 iap->ia_mtime.tv_nsec = 0;
1285 iap->ia_atime.tv_nsec = 0;
1286 iap->ia_mode = v_mode;
1289 /* Set file attributes.
1290 * Mode has already been set but we might need to reset it
1291 * for CREATE_EXCLUSIVE
1292 * Irix appears to send along the gid when it tries to
1293 * implement setgid directories via NFS. Clear out all that cruft.
1296 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1297 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1301 if (dchild && !IS_ERR(dchild))
1306 err = nfserrno(err);
1309 #endif /* CONFIG_NFSD_V3 */
1312 * Read a symlink. On entry, *lenp must contain the maximum path length that
1313 * fits into the buffer. On return, it contains the true length.
1314 * N.B. After this call fhp needs an fh_put
1317 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1319 struct dentry *dentry;
1320 struct inode *inode;
1324 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1328 dentry = fhp->fh_dentry;
1329 inode = dentry->d_inode;
1332 if (!inode->i_op || !inode->i_op->readlink)
1335 touch_atime(fhp->fh_export->ex_mnt, dentry);
1336 /* N.B. Why does this call need a get_fs()??
1337 * Remove the set_fs and watch the fireworks:-) --okir
1340 oldfs = get_fs(); set_fs(KERNEL_DS);
1341 err = inode->i_op->readlink(dentry, buf, *lenp);
1352 err = nfserrno(err);
1357 * Create a symlink and look up its inode
1358 * N.B. After this call _both_ fhp and resfhp need an fh_put
1361 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1362 char *fname, int flen,
1363 char *path, int plen,
1364 struct svc_fh *resfhp,
1367 struct dentry *dentry, *dnew;
1375 if (isdotent(fname, flen))
1378 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1382 dentry = fhp->fh_dentry;
1383 dnew = lookup_one_len(fname, dentry, flen);
1384 err = PTR_ERR(dnew);
1389 /* Only the MODE ATTRibute is even vaguely meaningful */
1390 if (iap && (iap->ia_valid & ATTR_MODE))
1391 mode = iap->ia_mode & S_IALLUGO;
1393 if (unlikely(path[plen] != 0)) {
1394 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1395 if (path_alloced == NULL)
1398 strncpy(path_alloced, path, plen);
1399 path_alloced[plen] = 0;
1400 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1401 kfree(path_alloced);
1404 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1407 if (EX_ISSYNC(fhp->fh_export))
1408 nfsd_sync_dir(dentry);
1410 err = nfserrno(err);
1413 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1415 if (err==0) err = cerr;
1420 err = nfserrno(err);
1426 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1429 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1430 char *name, int len, struct svc_fh *tfhp)
1432 struct dentry *ddir, *dnew, *dold;
1433 struct inode *dirp, *dest;
1436 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1439 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1447 if (isdotent(name, len))
1451 ddir = ffhp->fh_dentry;
1452 dirp = ddir->d_inode;
1454 dnew = lookup_one_len(name, ddir, len);
1455 err = PTR_ERR(dnew);
1459 dold = tfhp->fh_dentry;
1460 dest = dold->d_inode;
1462 err = vfs_link(dold, dirp, dnew);
1464 if (EX_ISSYNC(ffhp->fh_export)) {
1465 nfsd_sync_dir(ddir);
1466 write_inode_now(dest, 1);
1469 if (err == -EXDEV && rqstp->rq_vers == 2)
1472 err = nfserrno(err);
1481 err = nfserrno(err);
1487 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1490 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1491 struct svc_fh *tfhp, char *tname, int tlen)
1493 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1494 struct inode *fdir, *tdir;
1497 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1500 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1504 fdentry = ffhp->fh_dentry;
1505 fdir = fdentry->d_inode;
1507 tdentry = tfhp->fh_dentry;
1508 tdir = tdentry->d_inode;
1510 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1511 if (fdir->i_sb != tdir->i_sb)
1515 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1518 /* cannot use fh_lock as we need deadlock protective ordering
1519 * so do it by hand */
1520 trap = lock_rename(tdentry, fdentry);
1521 ffhp->fh_locked = tfhp->fh_locked = 1;
1525 odentry = lookup_one_len(fname, fdentry, flen);
1526 err = PTR_ERR(odentry);
1527 if (IS_ERR(odentry))
1531 if (!odentry->d_inode)
1534 if (odentry == trap)
1537 ndentry = lookup_one_len(tname, tdentry, tlen);
1538 err = PTR_ERR(ndentry);
1539 if (IS_ERR(ndentry))
1542 if (ndentry == trap)
1546 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1547 ((atomic_read(&odentry->d_count) > 1)
1548 || (atomic_read(&ndentry->d_count) > 1))) {
1552 err = vfs_rename(fdir, odentry, tdir, ndentry);
1553 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1554 nfsd_sync_dir(tdentry);
1555 nfsd_sync_dir(fdentry);
1564 err = nfserrno(err);
1566 /* we cannot reply on fh_unlock on the two filehandles,
1567 * as that would do the wrong thing if the two directories
1568 * were the same, so again we do it by hand
1570 fill_post_wcc(ffhp);
1571 fill_post_wcc(tfhp);
1572 unlock_rename(tdentry, fdentry);
1573 ffhp->fh_locked = tfhp->fh_locked = 0;
1580 * Unlink a file or directory
1581 * N.B. After this call fhp needs an fh_put
1584 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1585 char *fname, int flen)
1587 struct dentry *dentry, *rdentry;
1592 if (!flen || isdotent(fname, flen))
1594 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1599 dentry = fhp->fh_dentry;
1600 dirp = dentry->d_inode;
1602 rdentry = lookup_one_len(fname, dentry, flen);
1603 err = PTR_ERR(rdentry);
1604 if (IS_ERR(rdentry))
1607 if (!rdentry->d_inode) {
1614 type = rdentry->d_inode->i_mode & S_IFMT;
1616 if (type != S_IFDIR) { /* It's UNLINK */
1618 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1619 (atomic_read(&rdentry->d_count) > 1)) {
1623 err = vfs_unlink(dirp, rdentry);
1624 } else { /* It's RMDIR */
1625 err = vfs_rmdir(dirp, rdentry);
1632 if (EX_ISSYNC(fhp->fh_export))
1633 nfsd_sync_dir(dentry);
1639 err = nfserrno(err);
1644 * Read entries from a directory.
1645 * The NFSv3/4 verifier we ignore for now.
1648 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1649 struct readdir_cd *cdp, encode_dent_fn func)
1653 loff_t offset = *offsetp;
1655 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1659 offset = vfs_llseek(file, offset, 0);
1661 err = nfserrno((int)offset);
1666 * Read the directory entries. This silly loop is necessary because
1667 * readdir() is not guaranteed to fill up the entire buffer, but
1668 * may choose to do less.
1672 cdp->err = nfserr_eof; /* will be cleared on successful read */
1673 err = vfs_readdir(file, (filldir_t) func, cdp);
1674 } while (err >=0 && cdp->err == nfs_ok);
1676 err = nfserrno(err);
1679 *offsetp = vfs_llseek(file, 0, 1);
1681 if (err == nfserr_eof || err == nfserr_toosmall)
1682 err = nfs_ok; /* can still be found in ->err */
1690 * Get file system stats
1691 * N.B. After this call fhp needs an fh_put
1694 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1696 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1697 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1703 * Check for a user's access permissions to this inode.
1706 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1708 struct inode *inode = dentry->d_inode;
1714 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1716 (acc & MAY_READ)? " read" : "",
1717 (acc & MAY_WRITE)? " write" : "",
1718 (acc & MAY_EXEC)? " exec" : "",
1719 (acc & MAY_SATTR)? " sattr" : "",
1720 (acc & MAY_TRUNC)? " trunc" : "",
1721 (acc & MAY_LOCK)? " lock" : "",
1722 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1724 IS_IMMUTABLE(inode)? " immut" : "",
1725 IS_APPEND(inode)? " append" : "",
1726 IS_RDONLY(inode)? " ro" : "");
1727 dprintk(" owner %d/%d user %d/%d\n",
1728 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1731 /* Normally we reject any write/sattr etc access on a read-only file
1732 * system. But if it is IRIX doing check on write-access for a
1733 * device special file, we ignore rofs.
1735 if (!(acc & MAY_LOCAL_ACCESS))
1736 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1737 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1739 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1742 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1745 if (acc & MAY_LOCK) {
1746 /* If we cannot rely on authentication in NLM requests,
1747 * just allow locks, otherwise require read permission, or
1750 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1753 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1756 * The file owner always gets access permission for accesses that
1757 * would normally be checked at open time. This is to make
1758 * file access work even when the client has done a fchmod(fd, 0).
1760 * However, `cp foo bar' should fail nevertheless when bar is
1761 * readonly. A sensible way to do this might be to reject all
1762 * attempts to truncate a read-only file, because a creat() call
1763 * always implies file truncation.
1764 * ... but this isn't really fair. A process may reasonably call
1765 * ftruncate on an open file descriptor on a file with perm 000.
1766 * We must trust the client to do permission checking - using "ACCESS"
1769 if ((acc & MAY_OWNER_OVERRIDE) &&
1770 inode->i_uid == current->fsuid)
1773 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1775 /* Allow read access to binaries even when mode 111 */
1776 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1777 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1778 err = permission(inode, MAY_EXEC, NULL);
1780 return err? nfserrno(err) : 0;
1784 nfsd_racache_shutdown(void)
1788 dprintk("nfsd: freeing readahead buffers.\n");
1790 raparm_cache = raparml = NULL;
1793 * Initialize readahead param cache
1796 nfsd_racache_init(int cache_size)
1802 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1804 if (raparml != NULL) {
1805 dprintk("nfsd: allocating %d readahead buffers.\n",
1807 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1808 for (i = 0; i < cache_size - 1; i++) {
1809 raparml[i].p_next = raparml + i + 1;
1811 raparm_cache = raparml;
1814 "nfsd: Could not allocate memory read-ahead cache.\n");
1817 nfsdstats.ra_size = cache_size;