2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
31 #include <linux/security.h>
35 #endif /* CONFIG_NFSD_V3 */
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms *p_next;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
68 } ____cacheline_aligned_in_smp;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
87 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
91 err = follow_down(&path);
95 exp2 = rqst_exp_get_by_name(rqstp, &path);
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
110 if (nfsd_v4client(rqstp) ||
111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
112 /* successfully crossed mount point */
114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
120 path.dentry = dentry;
130 static void follow_to_parent(struct path *path)
134 while (path->dentry == path->mnt->mnt_root && follow_up(path))
136 dp = dget_parent(path->dentry);
141 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
143 struct svc_export *exp2;
144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
145 .dentry = dget(dparent)};
147 follow_to_parent(&path);
149 exp2 = rqst_exp_parent(rqstp, &path);
150 if (PTR_ERR(exp2) == -ENOENT) {
151 *dentryp = dget(dparent);
152 } else if (IS_ERR(exp2)) {
154 return PTR_ERR(exp2);
156 *dentryp = dget(path.dentry);
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
168 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
170 if (d_mountpoint(dentry))
172 if (nfsd4_is_junction(dentry))
174 if (!(exp->ex_flags & NFSEXP_V4ROOT))
176 return dentry->d_inode != NULL;
180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
181 const char *name, unsigned int len,
182 struct svc_export **exp_ret, struct dentry **dentry_ret)
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
191 dparent = fhp->fh_dentry;
192 exp = exp_get(fhp->fh_export);
194 /* Lookup the name, but don't follow links */
195 if (isdotent(name, len)) {
197 dentry = dget(dparent);
198 else if (dparent != exp->ex_path.dentry)
199 dentry = dget_parent(dparent);
200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
201 dentry = dget(dparent); /* .. == . just like at / */
203 /* checking mountpoint crossing is very different when stepping up */
204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
210 * In the nfsd4_open() case, this may be held across
211 * subsequent open and delegation acquisition which may
212 * need to take the child's i_mutex:
214 fh_lock_nested(fhp, I_MUTEX_PARENT);
215 dentry = lookup_one_len(name, dparent, len);
216 host_err = PTR_ERR(dentry);
220 * check if we have crossed a mount point ...
222 if (nfsd_mountpoint(dentry, exp)) {
223 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
229 *dentry_ret = dentry;
235 return nfserrno(host_err);
239 * Look up one component of a pathname.
240 * N.B. After this call _both_ fhp and resfh need an fh_put
242 * If the lookup would cross a mountpoint, and the mounted filesystem
243 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
244 * accepted as it stands and the mounted directory is
245 * returned. Otherwise the covered directory is returned.
246 * NOTE: this mountpoint crossing is not supported properly by all
247 * clients and is explicitly disallowed for NFSv3
248 * NeilBrown <neilb@cse.unsw.edu.au>
251 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
252 unsigned int len, struct svc_fh *resfh)
254 struct svc_export *exp;
255 struct dentry *dentry;
258 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
261 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
264 err = check_nfsd_access(exp, rqstp);
268 * Note: we compose the file handle now, but as the
269 * dentry may be negative, it may need to be updated.
271 err = fh_compose(resfh, exp, dentry, fhp);
272 if (!err && !dentry->d_inode)
281 * Commit metadata changes to stable storage.
284 commit_metadata(struct svc_fh *fhp)
286 struct inode *inode = fhp->fh_dentry->d_inode;
287 const struct export_operations *export_ops = inode->i_sb->s_export_op;
289 if (!EX_ISSYNC(fhp->fh_export))
292 if (export_ops->commit_metadata)
293 return export_ops->commit_metadata(inode);
294 return sync_inode_metadata(inode, 1);
298 * Go over the attributes and take care of the small differences between
299 * NFS semantics and what Linux expects.
302 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
304 /* sanitize the mode change */
305 if (iap->ia_valid & ATTR_MODE) {
306 iap->ia_mode &= S_IALLUGO;
307 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
310 /* Revoke setuid/setgid on chown */
311 if (!S_ISDIR(inode->i_mode) &&
312 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
313 iap->ia_valid |= ATTR_KILL_PRIV;
314 if (iap->ia_valid & ATTR_MODE) {
315 /* we're setting mode too, just clear the s*id bits */
316 iap->ia_mode &= ~S_ISUID;
317 if (iap->ia_mode & S_IXGRP)
318 iap->ia_mode &= ~S_ISGID;
320 /* set ATTR_KILL_* bits and let VFS handle it */
321 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
327 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
330 struct inode *inode = fhp->fh_dentry->d_inode;
333 if (iap->ia_size < inode->i_size) {
336 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
337 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
342 host_err = get_write_access(inode);
346 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
348 goto out_put_write_access;
351 out_put_write_access:
352 put_write_access(inode);
354 return nfserrno(host_err);
358 * Set various file attributes. After this call fhp needs an fh_put.
361 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
362 int check_guard, time_t guardtime)
364 struct dentry *dentry;
366 int accmode = NFSD_MAY_SATTR;
370 bool get_write_count;
373 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
374 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
375 if (iap->ia_valid & ATTR_SIZE)
378 /* Callers that do fh_verify should do the fh_want_write: */
379 get_write_count = !fhp->fh_dentry;
382 err = fh_verify(rqstp, fhp, ftype, accmode);
385 if (get_write_count) {
386 host_err = fh_want_write(fhp);
388 return nfserrno(host_err);
391 dentry = fhp->fh_dentry;
392 inode = dentry->d_inode;
394 /* Ignore any mode updates on symlinks */
395 if (S_ISLNK(inode->i_mode))
396 iap->ia_valid &= ~ATTR_MODE;
401 nfsd_sanitize_attrs(inode, iap);
404 * The size case is special, it changes the file in addition to the
407 if (iap->ia_valid & ATTR_SIZE) {
408 err = nfsd_get_write_access(rqstp, fhp, iap);
414 * RFC5661, Section 18.30.4:
415 * Changing the size of a file with SETATTR indirectly
416 * changes the time_modify and change attributes.
418 * (and similar for the older RFCs)
420 if (iap->ia_size != i_size_read(inode))
421 iap->ia_valid |= ATTR_MTIME;
424 iap->ia_valid |= ATTR_CTIME;
426 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
427 err = nfserr_notsync;
428 goto out_put_write_access;
432 host_err = notify_change(dentry, iap, NULL);
434 err = nfserrno(host_err);
436 out_put_write_access:
438 put_write_access(inode);
440 err = nfserrno(commit_metadata(fhp));
445 #if defined(CONFIG_NFSD_V4)
447 * NFS junction information is stored in an extended attribute.
449 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
452 * nfsd4_is_junction - Test if an object could be an NFS junction
454 * @dentry: object to test
456 * Returns 1 if "dentry" appears to contain NFS junction information.
457 * Otherwise 0 is returned.
459 int nfsd4_is_junction(struct dentry *dentry)
461 struct inode *inode = dentry->d_inode;
465 if (inode->i_mode & S_IXUGO)
467 if (!(inode->i_mode & S_ISVTX))
469 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
473 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
474 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
475 struct xdr_netobj *label)
479 struct dentry *dentry;
481 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
485 dentry = fhp->fh_dentry;
487 mutex_lock(&dentry->d_inode->i_mutex);
488 host_error = security_inode_setsecctx(dentry, label->data, label->len);
489 mutex_unlock(&dentry->d_inode->i_mutex);
490 return nfserrno(host_error);
493 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
494 struct xdr_netobj *label)
496 return nfserr_notsupp;
500 #endif /* defined(CONFIG_NFSD_V4) */
502 #ifdef CONFIG_NFSD_V3
504 * Check server access rights to a file system object
510 static struct accessmap nfs3_regaccess[] = {
511 { NFS3_ACCESS_READ, NFSD_MAY_READ },
512 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
513 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
514 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
519 static struct accessmap nfs3_diraccess[] = {
520 { NFS3_ACCESS_READ, NFSD_MAY_READ },
521 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
522 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
523 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
524 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
529 static struct accessmap nfs3_anyaccess[] = {
530 /* Some clients - Solaris 2.6 at least, make an access call
531 * to the server to check for access for things like /dev/null
532 * (which really, the server doesn't care about). So
533 * We provide simple access checking for them, looking
534 * mainly at mode bits, and we make sure to ignore read-only
537 { NFS3_ACCESS_READ, NFSD_MAY_READ },
538 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
539 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
540 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
546 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
548 struct accessmap *map;
549 struct svc_export *export;
550 struct dentry *dentry;
551 u32 query, result = 0, sresult = 0;
554 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
558 export = fhp->fh_export;
559 dentry = fhp->fh_dentry;
561 if (S_ISREG(dentry->d_inode->i_mode))
562 map = nfs3_regaccess;
563 else if (S_ISDIR(dentry->d_inode->i_mode))
564 map = nfs3_diraccess;
566 map = nfs3_anyaccess;
570 for (; map->access; map++) {
571 if (map->access & query) {
574 sresult |= map->access;
576 err2 = nfsd_permission(rqstp, export, dentry, map->how);
579 result |= map->access;
582 /* the following error codes just mean the access was not allowed,
583 * rather than an error occurred */
587 /* simply don't "or" in the access bit. */
597 *supported = sresult;
602 #endif /* CONFIG_NFSD_V3 */
604 static int nfsd_open_break_lease(struct inode *inode, int access)
608 if (access & NFSD_MAY_NOT_BREAK_LEASE)
610 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
611 return break_lease(inode, mode | O_NONBLOCK);
615 * Open an existing file or directory.
616 * The may_flags argument indicates the type of open (read/write/lock)
617 * and additional flags.
618 * N.B. After this call fhp needs an fh_put
621 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
622 int may_flags, struct file **filp)
627 int flags = O_RDONLY|O_LARGEFILE;
631 validate_process_creds();
634 * If we get here, then the client has already done an "open",
635 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
636 * in case a chmod has now revoked permission.
638 * Arguably we should also allow the owner override for
639 * directories, but we never have and it doesn't seem to have
640 * caused anyone a problem. If we were to change this, note
641 * also that our filldir callbacks would need a variant of
642 * lookup_one_len that doesn't check permissions.
645 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
646 err = fh_verify(rqstp, fhp, type, may_flags);
650 path.mnt = fhp->fh_export->ex_path.mnt;
651 path.dentry = fhp->fh_dentry;
652 inode = path.dentry->d_inode;
654 /* Disallow write access to files with the append-only bit set
655 * or any access when mandatory locking enabled
658 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
661 * We must ignore files (but only files) which might have mandatory
662 * locks on them because there is no way to know if the accesser has
665 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
671 host_err = nfsd_open_break_lease(inode, may_flags);
672 if (host_err) /* NOMEM or WOULDBLOCK */
675 if (may_flags & NFSD_MAY_WRITE) {
676 if (may_flags & NFSD_MAY_READ)
677 flags = O_RDWR|O_LARGEFILE;
679 flags = O_WRONLY|O_LARGEFILE;
682 file = dentry_open(&path, flags, current_cred());
684 host_err = PTR_ERR(file);
688 host_err = ima_file_check(file, may_flags, 0);
694 if (may_flags & NFSD_MAY_64BIT_COOKIE)
695 file->f_mode |= FMODE_64BITHASH;
697 file->f_mode |= FMODE_32BITHASH;
701 err = nfserrno(host_err);
703 validate_process_creds();
711 nfsd_close(struct file *filp)
717 * Obtain the readahead parameters for the file
718 * specified by (dev, ino).
721 static inline struct raparms *
722 nfsd_get_raparms(dev_t dev, ino_t ino)
724 struct raparms *ra, **rap, **frap = NULL;
727 struct raparm_hbucket *rab;
729 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
730 rab = &raparm_hash[hash];
732 spin_lock(&rab->pb_lock);
733 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
734 if (ra->p_ino == ino && ra->p_dev == dev)
737 if (ra->p_count == 0)
740 depth = nfsdstats.ra_size;
742 spin_unlock(&rab->pb_lock);
752 if (rap != &rab->pb_head) {
754 ra->p_next = rab->pb_head;
758 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
759 spin_unlock(&rab->pb_lock);
764 * Grab and keep cached pages associated with a file in the svc_rqst
765 * so that they can be passed to the network sendmsg/sendpage routines
766 * directly. They will be released after the sending has completed.
769 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
770 struct splice_desc *sd)
772 struct svc_rqst *rqstp = sd->u.data;
773 struct page **pp = rqstp->rq_next_page;
774 struct page *page = buf->page;
779 if (rqstp->rq_res.page_len == 0) {
781 put_page(*rqstp->rq_next_page);
782 *(rqstp->rq_next_page++) = page;
783 rqstp->rq_res.page_base = buf->offset;
784 rqstp->rq_res.page_len = size;
785 } else if (page != pp[-1]) {
787 if (*rqstp->rq_next_page)
788 put_page(*rqstp->rq_next_page);
789 *(rqstp->rq_next_page++) = page;
790 rqstp->rq_res.page_len += size;
792 rqstp->rq_res.page_len += size;
797 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
798 struct splice_desc *sd)
800 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
804 nfsd_finish_read(struct file *file, unsigned long *count, int host_err)
807 nfsdstats.io_read += host_err;
809 fsnotify_access(file);
812 return nfserrno(host_err);
815 __be32 nfsd_splice_read(struct svc_rqst *rqstp,
816 struct file *file, loff_t offset, unsigned long *count)
818 struct splice_desc sd = {
826 rqstp->rq_next_page = rqstp->rq_respages + 1;
827 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
828 return nfsd_finish_read(file, count, host_err);
831 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen,
832 unsigned long *count)
839 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
841 return nfsd_finish_read(file, count, host_err);
845 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file,
846 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
848 if (file->f_op->splice_read && rqstp->rq_splice_ok)
849 return nfsd_splice_read(rqstp, file, offset, count);
851 return nfsd_readv(file, offset, vec, vlen, count);
855 * Gathered writes: If another process is currently writing to the file,
856 * there's a high chance this is another nfsd (triggered by a bulk write
857 * from a client's biod). Rather than syncing the file with each write
858 * request, we sleep for 10 msec.
860 * I don't know if this roughly approximates C. Juszak's idea of
861 * gathered writes, but it's a nice and simple solution (IMHO), and it
864 * Note: we do this only in the NFSv2 case, since v3 and higher have a
865 * better tool (separate unstable writes and commits) for solving this
868 static int wait_for_concurrent_writes(struct file *file)
870 struct inode *inode = file_inode(file);
871 static ino_t last_ino;
872 static dev_t last_dev;
875 if (atomic_read(&inode->i_writecount) > 1
876 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
877 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
879 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
882 if (inode->i_state & I_DIRTY) {
883 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
884 err = vfs_fsync(file, 0);
886 last_ino = inode->i_ino;
887 last_dev = inode->i_sb->s_dev;
892 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
893 loff_t offset, struct kvec *vec, int vlen,
894 unsigned long *cnt, int *stablep)
896 struct svc_export *exp;
897 struct dentry *dentry;
902 int stable = *stablep;
905 unsigned int pflags = current->flags;
909 * We want less throttling in balance_dirty_pages()
910 * and shrink_inactive_list() so that nfs to
911 * localhost doesn't cause nfsd to lock up due to all
912 * the client's dirty pages or its congested queue.
914 current->flags |= PF_LESS_THROTTLE;
916 dentry = file->f_path.dentry;
917 inode = dentry->d_inode;
918 exp = fhp->fh_export;
920 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
925 /* Write the data. */
926 oldfs = get_fs(); set_fs(KERNEL_DS);
927 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
932 nfsdstats.io_write += host_err;
933 fsnotify_modify(file);
937 host_err = wait_for_concurrent_writes(file);
939 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
943 dprintk("nfsd: write complete host_err=%d\n", host_err);
947 err = nfserrno(host_err);
949 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE);
953 __be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp,
954 struct file **file, struct raparms **ra)
959 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, file);
963 inode = file_inode(*file);
965 /* Get readahead parameters */
966 *ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
968 if (*ra && (*ra)->p_set)
969 (*file)->f_ra = (*ra)->p_ra;
973 void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra)
975 /* Write back readahead params */
977 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
978 spin_lock(&rab->pb_lock);
979 ra->p_ra = file->f_ra;
982 spin_unlock(&rab->pb_lock);
988 * Read data from a file. count must contain the requested read count
989 * on entry. On return, *count contains the number of bytes actually read.
990 * N.B. After this call fhp needs an fh_put
992 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
993 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
999 err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra);
1003 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count);
1005 nfsd_put_tmp_read_open(file, ra);
1011 * Write data to a file.
1012 * The stable flag requests synchronous writes.
1013 * N.B. After this call fhp needs an fh_put
1016 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1017 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1023 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1024 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1027 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1030 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1035 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1043 #ifdef CONFIG_NFSD_V3
1045 * Commit all pending writes to stable storage.
1047 * Note: we only guarantee that data that lies within the range specified
1048 * by the 'offset' and 'count' parameters will be synced.
1050 * Unfortunately we cannot lock the file to make sure we return full WCC
1051 * data to the client, as locking happens lower down in the filesystem.
1054 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1055 loff_t offset, unsigned long count)
1058 loff_t end = LLONG_MAX;
1059 __be32 err = nfserr_inval;
1064 end = offset + (loff_t)count - 1;
1069 err = nfsd_open(rqstp, fhp, S_IFREG,
1070 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1073 if (EX_ISSYNC(fhp->fh_export)) {
1074 int err2 = vfs_fsync_range(file, offset, end, 0);
1076 if (err2 != -EINVAL)
1077 err = nfserrno(err2);
1079 err = nfserr_notsupp;
1086 #endif /* CONFIG_NFSD_V3 */
1089 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1093 * Mode has already been set earlier in create:
1095 iap->ia_valid &= ~ATTR_MODE;
1097 * Setting uid/gid works only for root. Irix appears to
1098 * send along the gid on create when it tries to implement
1099 * setgid directories via NFS:
1101 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1102 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1104 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1105 /* Callers expect file metadata to be committed here */
1106 return nfserrno(commit_metadata(resfhp));
1109 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1110 * setting size to 0 may fail for some specific file systems by the permission
1111 * checking which requires WRITE permission but the mode is 000.
1112 * we ignore the resizing(to 0) on the just new created file, since the size is
1113 * 0 after file created.
1115 * call this only after vfs_create() is called.
1118 nfsd_check_ignore_resizing(struct iattr *iap)
1120 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1121 iap->ia_valid &= ~ATTR_SIZE;
1125 * Create a file (regular, directory, device, fifo); UNIX sockets
1126 * not yet implemented.
1127 * If the response fh has been verified, the parent directory should
1128 * already be locked. Note that the parent directory is left locked.
1130 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1133 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1134 char *fname, int flen, struct iattr *iap,
1135 int type, dev_t rdev, struct svc_fh *resfhp)
1137 struct dentry *dentry, *dchild = NULL;
1147 if (isdotent(fname, flen))
1150 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1154 dentry = fhp->fh_dentry;
1155 dirp = dentry->d_inode;
1157 err = nfserr_notdir;
1158 if (!dirp->i_op->lookup)
1161 * Check whether the response file handle has been verified yet.
1162 * If it has, the parent directory should already be locked.
1164 if (!resfhp->fh_dentry) {
1165 host_err = fh_want_write(fhp);
1169 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1170 fh_lock_nested(fhp, I_MUTEX_PARENT);
1171 dchild = lookup_one_len(fname, dentry, flen);
1172 host_err = PTR_ERR(dchild);
1175 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1179 /* called from nfsd_proc_create */
1180 dchild = dget(resfhp->fh_dentry);
1181 if (!fhp->fh_locked) {
1182 /* not actually possible */
1184 "nfsd_create: parent %pd2 not locked!\n",
1191 * Make sure the child dentry is still negative ...
1194 if (dchild->d_inode) {
1195 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1200 if (!(iap->ia_valid & ATTR_MODE))
1202 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1205 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1206 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1212 * Get the dir op function pointer.
1218 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1220 nfsd_check_ignore_resizing(iap);
1223 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1229 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1235 err = nfsd_create_setattr(rqstp, resfhp, iap);
1238 * nfsd_create_setattr already committed the child. Transactional
1239 * filesystems had a chance to commit changes for both parent and
1240 * child * simultaneously making the following commit_metadata a
1243 err2 = nfserrno(commit_metadata(fhp));
1247 * Update the file handle to get the new inode info.
1250 err = fh_update(resfhp);
1252 if (dchild && !IS_ERR(dchild))
1257 err = nfserrno(host_err);
1261 #ifdef CONFIG_NFSD_V3
1263 static inline int nfsd_create_is_exclusive(int createmode)
1265 return createmode == NFS3_CREATE_EXCLUSIVE
1266 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1270 * NFSv3 and NFSv4 version of nfsd_create
1273 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1274 char *fname, int flen, struct iattr *iap,
1275 struct svc_fh *resfhp, int createmode, u32 *verifier,
1276 bool *truncp, bool *created)
1278 struct dentry *dentry, *dchild = NULL;
1282 __u32 v_mtime=0, v_atime=0;
1288 if (isdotent(fname, flen))
1290 if (!(iap->ia_valid & ATTR_MODE))
1292 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1296 dentry = fhp->fh_dentry;
1297 dirp = dentry->d_inode;
1299 /* Get all the sanity checks out of the way before
1300 * we lock the parent. */
1301 err = nfserr_notdir;
1302 if (!dirp->i_op->lookup)
1305 host_err = fh_want_write(fhp);
1309 fh_lock_nested(fhp, I_MUTEX_PARENT);
1312 * Compose the response file handle.
1314 dchild = lookup_one_len(fname, dentry, flen);
1315 host_err = PTR_ERR(dchild);
1319 /* If file doesn't exist, check for permissions to create one */
1320 if (!dchild->d_inode) {
1321 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1326 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1330 if (nfsd_create_is_exclusive(createmode)) {
1331 /* solaris7 gets confused (bugid 4218508) if these have
1332 * the high bit set, so just clear the high bits. If this is
1333 * ever changed to use different attrs for storing the
1334 * verifier, then do_open_lookup() will also need to be fixed
1337 v_mtime = verifier[0]&0x7fffffff;
1338 v_atime = verifier[1]&0x7fffffff;
1341 if (dchild->d_inode) {
1344 switch (createmode) {
1345 case NFS3_CREATE_UNCHECKED:
1346 if (! S_ISREG(dchild->d_inode->i_mode))
1349 /* in nfsv4, we need to treat this case a little
1350 * differently. we don't want to truncate the
1351 * file now; this would be wrong if the OPEN
1352 * fails for some other reason. furthermore,
1353 * if the size is nonzero, we should ignore it
1354 * according to spec!
1356 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1359 iap->ia_valid &= ATTR_SIZE;
1363 case NFS3_CREATE_EXCLUSIVE:
1364 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1365 && dchild->d_inode->i_atime.tv_sec == v_atime
1366 && dchild->d_inode->i_size == 0 ) {
1371 case NFS4_CREATE_EXCLUSIVE4_1:
1372 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1373 && dchild->d_inode->i_atime.tv_sec == v_atime
1374 && dchild->d_inode->i_size == 0 ) {
1380 case NFS3_CREATE_GUARDED:
1387 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1395 nfsd_check_ignore_resizing(iap);
1397 if (nfsd_create_is_exclusive(createmode)) {
1398 /* Cram the verifier into atime/mtime */
1399 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1400 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1401 /* XXX someone who knows this better please fix it for nsec */
1402 iap->ia_mtime.tv_sec = v_mtime;
1403 iap->ia_atime.tv_sec = v_atime;
1404 iap->ia_mtime.tv_nsec = 0;
1405 iap->ia_atime.tv_nsec = 0;
1409 err = nfsd_create_setattr(rqstp, resfhp, iap);
1412 * nfsd_create_setattr already committed the child
1413 * (and possibly also the parent).
1416 err = nfserrno(commit_metadata(fhp));
1419 * Update the filehandle to get the new inode info.
1422 err = fh_update(resfhp);
1426 if (dchild && !IS_ERR(dchild))
1432 err = nfserrno(host_err);
1435 #endif /* CONFIG_NFSD_V3 */
1438 * Read a symlink. On entry, *lenp must contain the maximum path length that
1439 * fits into the buffer. On return, it contains the true length.
1440 * N.B. After this call fhp needs an fh_put
1443 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1445 struct inode *inode;
1451 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1455 path.mnt = fhp->fh_export->ex_path.mnt;
1456 path.dentry = fhp->fh_dentry;
1457 inode = path.dentry->d_inode;
1460 if (!inode->i_op->readlink)
1464 /* N.B. Why does this call need a get_fs()??
1465 * Remove the set_fs and watch the fireworks:-) --okir
1468 oldfs = get_fs(); set_fs(KERNEL_DS);
1469 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1480 err = nfserrno(host_err);
1485 * Create a symlink and look up its inode
1486 * N.B. After this call _both_ fhp and resfhp need an fh_put
1489 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1490 char *fname, int flen,
1492 struct svc_fh *resfhp)
1494 struct dentry *dentry, *dnew;
1499 if (!flen || path[0] == '\0')
1502 if (isdotent(fname, flen))
1505 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1509 host_err = fh_want_write(fhp);
1514 dentry = fhp->fh_dentry;
1515 dnew = lookup_one_len(fname, dentry, flen);
1516 host_err = PTR_ERR(dnew);
1520 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1521 err = nfserrno(host_err);
1523 err = nfserrno(commit_metadata(fhp));
1528 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1530 if (err==0) err = cerr;
1535 err = nfserrno(host_err);
1541 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1544 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1545 char *name, int len, struct svc_fh *tfhp)
1547 struct dentry *ddir, *dnew, *dold;
1552 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1555 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1559 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1565 if (isdotent(name, len))
1568 host_err = fh_want_write(tfhp);
1570 err = nfserrno(host_err);
1574 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1575 ddir = ffhp->fh_dentry;
1576 dirp = ddir->d_inode;
1578 dnew = lookup_one_len(name, ddir, len);
1579 host_err = PTR_ERR(dnew);
1583 dold = tfhp->fh_dentry;
1588 host_err = vfs_link(dold, dirp, dnew, NULL);
1590 err = nfserrno(commit_metadata(ffhp));
1592 err = nfserrno(commit_metadata(tfhp));
1594 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1597 err = nfserrno(host_err);
1603 fh_drop_write(tfhp);
1608 err = nfserrno(host_err);
1614 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1617 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1618 struct svc_fh *tfhp, char *tname, int tlen)
1620 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1621 struct inode *fdir, *tdir;
1625 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1628 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1632 fdentry = ffhp->fh_dentry;
1633 fdir = fdentry->d_inode;
1635 tdentry = tfhp->fh_dentry;
1636 tdir = tdentry->d_inode;
1639 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1642 host_err = fh_want_write(ffhp);
1644 err = nfserrno(host_err);
1648 /* cannot use fh_lock as we need deadlock protective ordering
1649 * so do it by hand */
1650 trap = lock_rename(tdentry, fdentry);
1651 ffhp->fh_locked = tfhp->fh_locked = 1;
1655 odentry = lookup_one_len(fname, fdentry, flen);
1656 host_err = PTR_ERR(odentry);
1657 if (IS_ERR(odentry))
1661 if (!odentry->d_inode)
1664 if (odentry == trap)
1667 ndentry = lookup_one_len(tname, tdentry, tlen);
1668 host_err = PTR_ERR(ndentry);
1669 if (IS_ERR(ndentry))
1671 host_err = -ENOTEMPTY;
1672 if (ndentry == trap)
1676 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1678 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1681 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
1683 host_err = commit_metadata(tfhp);
1685 host_err = commit_metadata(ffhp);
1692 err = nfserrno(host_err);
1694 * We cannot rely on fh_unlock on the two filehandles,
1695 * as that would do the wrong thing if the two directories
1696 * were the same, so again we do it by hand.
1698 fill_post_wcc(ffhp);
1699 fill_post_wcc(tfhp);
1700 unlock_rename(tdentry, fdentry);
1701 ffhp->fh_locked = tfhp->fh_locked = 0;
1702 fh_drop_write(ffhp);
1709 * Unlink a file or directory
1710 * N.B. After this call fhp needs an fh_put
1713 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1714 char *fname, int flen)
1716 struct dentry *dentry, *rdentry;
1722 if (!flen || isdotent(fname, flen))
1724 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1728 host_err = fh_want_write(fhp);
1732 fh_lock_nested(fhp, I_MUTEX_PARENT);
1733 dentry = fhp->fh_dentry;
1734 dirp = dentry->d_inode;
1736 rdentry = lookup_one_len(fname, dentry, flen);
1737 host_err = PTR_ERR(rdentry);
1738 if (IS_ERR(rdentry))
1741 if (!rdentry->d_inode) {
1748 type = rdentry->d_inode->i_mode & S_IFMT;
1750 if (type != S_IFDIR)
1751 host_err = vfs_unlink(dirp, rdentry, NULL);
1753 host_err = vfs_rmdir(dirp, rdentry);
1755 host_err = commit_metadata(fhp);
1759 err = nfserrno(host_err);
1765 * We do this buffering because we must not call back into the file
1766 * system's ->lookup() method from the filldir callback. That may well
1767 * deadlock a number of file systems.
1769 * This is based heavily on the implementation of same in XFS.
1771 struct buffered_dirent {
1775 unsigned int d_type;
1779 struct readdir_data {
1780 struct dir_context ctx;
1786 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1787 loff_t offset, u64 ino, unsigned int d_type)
1789 struct readdir_data *buf = __buf;
1790 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1791 unsigned int reclen;
1793 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1794 if (buf->used + reclen > PAGE_SIZE) {
1799 de->namlen = namlen;
1800 de->offset = offset;
1802 de->d_type = d_type;
1803 memcpy(de->name, name, namlen);
1804 buf->used += reclen;
1809 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1810 struct readdir_cd *cdp, loff_t *offsetp)
1812 struct buffered_dirent *de;
1816 struct readdir_data buf = {
1817 .ctx.actor = nfsd_buffered_filldir,
1818 .dirent = (void *)__get_free_page(GFP_KERNEL)
1822 return nfserrno(-ENOMEM);
1827 struct inode *dir_inode = file_inode(file);
1828 unsigned int reclen;
1830 cdp->err = nfserr_eof; /* will be cleared on successful read */
1834 host_err = iterate_dir(file, &buf.ctx);
1847 * Various filldir functions may end up calling back into
1848 * lookup_one_len() and the file system's ->lookup() method.
1849 * These expect i_mutex to be held, as it would within readdir.
1851 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1855 de = (struct buffered_dirent *)buf.dirent;
1857 offset = de->offset;
1859 if (func(cdp, de->name, de->namlen, de->offset,
1860 de->ino, de->d_type))
1863 if (cdp->err != nfs_ok)
1866 reclen = ALIGN(sizeof(*de) + de->namlen,
1869 de = (struct buffered_dirent *)((char *)de + reclen);
1871 mutex_unlock(&dir_inode->i_mutex);
1872 if (size > 0) /* We bailed out early */
1875 offset = vfs_llseek(file, 0, SEEK_CUR);
1878 free_page((unsigned long)(buf.dirent));
1881 return nfserrno(host_err);
1888 * Read entries from a directory.
1889 * The NFSv3/4 verifier we ignore for now.
1892 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1893 struct readdir_cd *cdp, filldir_t func)
1897 loff_t offset = *offsetp;
1898 int may_flags = NFSD_MAY_READ;
1900 /* NFSv2 only supports 32 bit cookies */
1901 if (rqstp->rq_vers > 2)
1902 may_flags |= NFSD_MAY_64BIT_COOKIE;
1904 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1908 offset = vfs_llseek(file, offset, SEEK_SET);
1910 err = nfserrno((int)offset);
1914 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1916 if (err == nfserr_eof || err == nfserr_toosmall)
1917 err = nfs_ok; /* can still be found in ->err */
1925 * Get file system stats
1926 * N.B. After this call fhp needs an fh_put
1929 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1933 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1935 struct path path = {
1936 .mnt = fhp->fh_export->ex_path.mnt,
1937 .dentry = fhp->fh_dentry,
1939 if (vfs_statfs(&path, stat))
1945 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1947 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1951 * Check for a user's access permissions to this inode.
1954 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1955 struct dentry *dentry, int acc)
1957 struct inode *inode = dentry->d_inode;
1960 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
1963 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1965 (acc & NFSD_MAY_READ)? " read" : "",
1966 (acc & NFSD_MAY_WRITE)? " write" : "",
1967 (acc & NFSD_MAY_EXEC)? " exec" : "",
1968 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1969 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1970 (acc & NFSD_MAY_LOCK)? " lock" : "",
1971 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1973 IS_IMMUTABLE(inode)? " immut" : "",
1974 IS_APPEND(inode)? " append" : "",
1975 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1976 dprintk(" owner %d/%d user %d/%d\n",
1977 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
1980 /* Normally we reject any write/sattr etc access on a read-only file
1981 * system. But if it is IRIX doing check on write-access for a
1982 * device special file, we ignore rofs.
1984 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
1985 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
1986 if (exp_rdonly(rqstp, exp) ||
1987 __mnt_is_readonly(exp->ex_path.mnt))
1989 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1992 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1995 if (acc & NFSD_MAY_LOCK) {
1996 /* If we cannot rely on authentication in NLM requests,
1997 * just allow locks, otherwise require read permission, or
2000 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2003 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2006 * The file owner always gets access permission for accesses that
2007 * would normally be checked at open time. This is to make
2008 * file access work even when the client has done a fchmod(fd, 0).
2010 * However, `cp foo bar' should fail nevertheless when bar is
2011 * readonly. A sensible way to do this might be to reject all
2012 * attempts to truncate a read-only file, because a creat() call
2013 * always implies file truncation.
2014 * ... but this isn't really fair. A process may reasonably call
2015 * ftruncate on an open file descriptor on a file with perm 000.
2016 * We must trust the client to do permission checking - using "ACCESS"
2019 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2020 uid_eq(inode->i_uid, current_fsuid()))
2023 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2024 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2026 /* Allow read access to binaries even when mode 111 */
2027 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2028 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2029 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2030 err = inode_permission(inode, MAY_EXEC);
2032 return err? nfserrno(err) : 0;
2036 nfsd_racache_shutdown(void)
2038 struct raparms *raparm, *last_raparm;
2041 dprintk("nfsd: freeing readahead buffers.\n");
2043 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2044 raparm = raparm_hash[i].pb_head;
2046 last_raparm = raparm;
2047 raparm = raparm->p_next;
2050 raparm_hash[i].pb_head = NULL;
2054 * Initialize readahead param cache
2057 nfsd_racache_init(int cache_size)
2062 struct raparms **raparm = NULL;
2065 if (raparm_hash[0].pb_head)
2067 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2068 nperbucket = max(2, nperbucket);
2069 cache_size = nperbucket * RAPARM_HASH_SIZE;
2071 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2073 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2074 spin_lock_init(&raparm_hash[i].pb_lock);
2076 raparm = &raparm_hash[i].pb_head;
2077 for (j = 0; j < nperbucket; j++) {
2078 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2081 raparm = &(*raparm)->p_next;
2086 nfsdstats.ra_size = cache_size;
2090 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2091 nfsd_racache_shutdown();