X-Git-Url: https://git.camperquake.de/gitweb.cgi?a=blobdiff_plain;f=module%2Fzfs%2Fzfs_vnops.c;h=3c3e8db4b05829ad702b8a4a7c881a62ba50581c;hb=802e7b5feb0135483de119eac1da192404eb5bb7;hp=21397131025ceb27d92ffa6601e2aabb0412a08c;hpb=e2e7aa2df88708cdcae87688a69ba3f5d1cc43ed;p=zfs.git

diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
index 2139713..3c3e8db 100644
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 /* Portions Copyright 2007 Jeremy Teo */
@@ -63,6 +64,7 @@
 #include <sys/sid.h>
 #include <sys/mode.h>
 #include "fs/fs_subr.h"
+#include <sys/zfs_ctldir.h>
 #include <sys/zfs_fuid.h>
 #include <sys/zfs_sa.h>
 #include <sys/zfs_vnops.h>
@@ -238,6 +240,68 @@ zfs_close(struct inode *ip, int flag, cred_t *cr)
 }
 EXPORT_SYMBOL(zfs_close);
 
+#if defined(SEEK_HOLE) && defined(SEEK_DATA)
+/*
+ * Lseek support for finding holes (cmd == SEEK_HOLE) and
+ * data (cmd == SEEK_DATA). "off" is an in/out parameter.
+ */
+static int
+zfs_holey_common(struct inode *ip, int cmd, loff_t *off)
+{
+	znode_t	*zp = ITOZ(ip);
+	uint64_t noff = (uint64_t)*off; /* new offset */
+	uint64_t file_sz;
+	int error;
+	boolean_t hole;
+
+	file_sz = zp->z_size;
+	if (noff >= file_sz)  {
+		return (ENXIO);
+	}
+
+	if (cmd == SEEK_HOLE)
+		hole = B_TRUE;
+	else
+		hole = B_FALSE;
+
+	error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff);
+
+	/* end of file? */
+	if ((error == ESRCH) || (noff > file_sz)) {
+		/*
+		 * Handle the virtual hole at the end of file.
+		 */
+		if (hole) {
+			*off = file_sz;
+			return (0);
+		}
+		return (ENXIO);
+	}
+
+	if (noff < *off)
+		return (error);
+	*off = noff;
+	return (error);
+}
+
+int
+zfs_holey(struct inode *ip, int cmd, loff_t *off)
+{
+	znode_t	*zp = ITOZ(ip);
+	zfs_sb_t *zsb = ITOZSB(ip);
+	int error;
+
+	ZFS_ENTER(zsb);
+	ZFS_VERIFY_ZP(zp);
+
+	error = zfs_holey_common(ip, cmd, off);
+
+	ZFS_EXIT(zsb);
+	return (error);
+}
+EXPORT_SYMBOL(zfs_holey);
+#endif /* SEEK_HOLE && SEEK_DATA */
+
 #if defined(_KERNEL)
 /*
  * When a file is memory mapped, we must keep the IO data synchronized
@@ -399,18 +463,14 @@ zfs_read(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
 		return (0);
 	}
 
-#ifdef HAVE_MANDLOCKS
 	/*
 	 * Check for mandatory locks
 	 */
-	if (MANDMODE(zp->z_mode)) {
-		if (error = chklock(ip, FREAD,
-		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
-			ZFS_EXIT(zsb);
-			return (error);
-		}
+	if (mandatory_lock(ip) &&
+	    !lock_may_read(ip, uio->uio_loffset, uio->uio_resid)) {
+		ZFS_EXIT(zsb);
+		return (EAGAIN);
 	}
-#endif /* HAVE_MANDLOCK */
 
 	/*
 	 * If we're in FRSYNC mode, sync out this znode before reading it.
@@ -581,17 +641,14 @@ zfs_write(struct inode *ip, uio_t *uio, int ioflag, cred_t *cr)
 		return (EINVAL);
 	}
 
-#ifdef HAVE_MANDLOCKS
 	/*
 	 * Check for mandatory locks before calling zfs_range_lock()
 	 * in order to prevent a deadlock with locks set via fcntl().
 	 */
-	if (MANDMODE((mode_t)zp->z_mode) &&
-	    (error = chklock(ip, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
+	if (mandatory_lock(ip) && !lock_may_write(ip, woff, n)) {
 		ZFS_EXIT(zsb);
-		return (error);
+		return (EAGAIN);
 	}
-#endif /* HAVE_MANDLOCKS */
 
 #ifdef HAVE_UIO_ZEROCOPY
 	/*
@@ -874,7 +931,7 @@ iput_async(struct inode *ip, taskq_t *taskq)
 {
 	ASSERT(atomic_read(&ip->i_count) > 0);
 	if (atomic_read(&ip->i_count) == 1)
-		taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_SLEEP);
+		taskq_dispatch(taskq, (task_func_t *)iput, ip, TQ_PUSHPAGE);
 	else
 		iput(ip);
 }
@@ -940,7 +997,7 @@ zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
 		return (ENOENT);
 	}
 
-	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
+	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_PUSHPAGE);
 	zgd->zgd_zilog = zsb->z_log;
 	zgd->zgd_private = zp;
 
@@ -1120,14 +1177,6 @@ zfs_lookup(struct inode *dip, char *nm, struct inode **ipp, int flags,
 
 	if (flags & LOOKUP_XATTR) {
 		/*
-		 * If the xattr property is off, refuse the lookup request.
-		 */
-		if (!(zsb->z_flags & ZSB_XATTR_USER)) {
-			ZFS_EXIT(zsb);
-			return (EINVAL);
-		}
-
-		/*
 		 * We don't allow recursive attributes..
 		 * Maybe someday we will.
 		 */
@@ -1537,7 +1586,7 @@ top:
 	    &xattr_obj, sizeof (xattr_obj));
 	if (error == 0 && xattr_obj) {
 		error = zfs_zget(zsb, xattr_obj, &xzp);
-		ASSERT3U(error, ==, 0);
+		ASSERT0(error);
 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
 	}
@@ -1914,13 +1963,13 @@ top:
 out:
 	zfs_dirent_unlock(dl);
 
+	zfs_inode_update(dzp);
+	zfs_inode_update(zp);
 	iput(ip);
 
 	if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
 		zil_commit(zilog, 0);
 
-	zfs_inode_update(dzp);
-	zfs_inode_update(zp);
 	ZFS_EXIT(zsb);
 	return (error);
 }
@@ -2027,12 +2076,22 @@ zfs_readdir(struct inode *ip, void *dirent, filldir_t filldir,
 					goto update;
 			}
 
+			/*
+			 * Allow multiple entries provided the first entry is
+			 * the object id.  Non-zpl consumers may safely make
+			 * use of the additional space.
+			 *
+			 * XXX: This should be a feature flag for compatibility
+			 */
 			if (zap.za_integer_length != 8 ||
-			    zap.za_num_integers != 1) {
+			    zap.za_num_integers == 0) {
 				cmn_err(CE_WARN, "zap_readdir: bad directory "
-				    "entry, obj = %lld, offset = %lld\n",
+				    "entry, obj = %lld, offset = %lld, "
+				    "length = %d, num = %lld\n",
 				    (u_longlong_t)zp->z_id,
-				    (u_longlong_t)*pos);
+				    (u_longlong_t)*pos,
+				    zap.za_integer_length,
+				    (u_longlong_t)zap.za_num_integers);
 				error = ENXIO;
 				goto update;
 			}
@@ -2050,7 +2109,7 @@ zfs_readdir(struct inode *ip, void *dirent, filldir_t filldir,
 			dmu_prefetch(os, objnum, 0, 0);
 		}
 
-		if (*pos >= 2) {
+		if (*pos > 2 || (*pos == 2 && !zfs_show_ctldir(zp))) {
 			zap_cursor_advance(&zc);
 			*pos = zap_cursor_serialize(&zc);
 		} else {
@@ -2298,6 +2357,49 @@ zfs_getattr(struct inode *ip, vattr_t *vap, int flags, cred_t *cr)
 EXPORT_SYMBOL(zfs_getattr);
 
 /*
+ * Get the basic file attributes and place them in the provided kstat
+ * structure.  The inode is assumed to be the authoritative source
+ * for most of the attributes.  However, the znode currently has the
+ * authoritative atime, blksize, and block count.
+ *
+ *	IN:	ip	- inode of file.
+ *
+ *	OUT:	sp	- kstat values.
+ *
+ *	RETURN:	0 (always succeeds)
+ */
+/* ARGSUSED */
+int
+zfs_getattr_fast(struct inode *ip, struct kstat *sp)
+{
+	znode_t *zp = ITOZ(ip);
+	zfs_sb_t *zsb = ITOZSB(ip);
+
+	ZFS_ENTER(zsb);
+	ZFS_VERIFY_ZP(zp);
+
+	mutex_enter(&zp->z_lock);
+
+	generic_fillattr(ip, sp);
+	ZFS_TIME_DECODE(&sp->atime, zp->z_atime);
+
+	sa_object_size(zp->z_sa_hdl, (uint32_t *)&sp->blksize, &sp->blocks);
+	if (unlikely(zp->z_blksz == 0)) {
+		/*
+		 * Block size hasn't been set; suggest maximal I/O transfers.
+		 */
+		sp->blksize = zsb->z_max_blksz;
+	}
+
+	mutex_exit(&zp->z_lock);
+
+	ZFS_EXIT(zsb);
+
+	return (0);
+}
+EXPORT_SYMBOL(zfs_getattr_fast);
+
+/*
  * Set the file attributes to the values contained in the
  * vattr structure.
  *
@@ -2420,7 +2522,7 @@ top:
 	aclp = NULL;
 
 	/* Can this be moved to before the top label? */
-	if (zsb->z_vfs->mnt_flags & MNT_READONLY) {
+	if (zfs_is_readonly(zsb)) {
 		err = EROFS;
 		goto out3;
 	}
@@ -2798,7 +2900,7 @@ top:
 		zp->z_mode = new_mode;
 		ASSERT3P(aclp, !=, NULL);
 		err = zfs_aclset_common(zp, aclp, cr, tx);
-		ASSERT3U(err, ==, 0);
+		ASSERT0(err);
 		if (zp->z_acl_cached)
 			zfs_acl_free(zp->z_acl_cached);
 		zp->z_acl_cached = aclp;
@@ -3307,7 +3409,7 @@ top:
 
 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
-			ASSERT3U(error, ==, 0);
+			ASSERT0(error);
 
 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
 			if (error == 0) {
@@ -3712,141 +3814,203 @@ top:
 }
 EXPORT_SYMBOL(zfs_link);
 
+static void
+zfs_putpage_commit_cb(void *arg, int error)
+{
+	struct page *pp = arg;
+
+	if (error) {
+		__set_page_dirty_nobuffers(pp);
+
+		if (error != ECANCELED)
+			SetPageError(pp);
+	} else {
+		ClearPageError(pp);
+	}
+
+	end_page_writeback(pp);
+}
+
 /*
- * Push a page out to disk
- *
- *	IN:	vp	- file to push page to.
- *		pp	- page to push.
- *		off	- start of range pushed.
- *		len	- len of range pushed.
+ * Push a page out to disk, once the page is on stable storage the
+ * registered commit callback will be run as notification of completion.
  *
+ *	IN:	ip	- page mapped for inode.
+ *		pp	- page to push (page is locked)
+ *		wbc	- writeback control data
  *
  *	RETURN:	0 if success
  *		error code if failure
  *
- * NOTE: callers must have locked the page to be pushed.
+ * Timestamps:
+ *	ip - ctime|mtime updated
  */
 /* ARGSUSED */
-static int
-zfs_putapage(struct inode *ip, struct page *pp, u_offset_t off, size_t len)
+int
+zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
 {
-	znode_t    *zp  = ITOZ(ip);
-	zfs_sb_t   *zsb = ITOZSB(ip);
-	dmu_tx_t   *tx;
-	caddr_t	   va;
-	int        err;
+	znode_t		*zp = ITOZ(ip);
+	zfs_sb_t	*zsb = ITOZSB(ip);
+	loff_t		offset;
+	loff_t		pgoff;
+	unsigned int	pglen;
+	rl_t		*rl;
+	dmu_tx_t	*tx;
+	caddr_t		va;
+	int		err = 0;
+	uint64_t	mtime[2], ctime[2];
+	sa_bulk_attr_t	bulk[3];
+	int		cnt = 0;
+	int		sync;
 
+	ZFS_ENTER(zsb);
+	ZFS_VERIFY_ZP(zp);
+
+	ASSERT(PageLocked(pp));
+
+	pgoff = page_offset(pp);     /* Page byte-offset in file */
+	offset = i_size_read(ip);    /* File length in bytes */
+	pglen = MIN(PAGE_CACHE_SIZE, /* Page length in bytes */
+	    P2ROUNDUP(offset, PAGE_CACHE_SIZE)-pgoff);
+
+	/* Page is beyond end of file */
+	if (pgoff >= offset) {
+		unlock_page(pp);
+		ZFS_EXIT(zsb);
+		return (0);
+	}
+
+	/* Truncate page length to end of file */
+	if (pgoff + pglen > offset)
+		pglen = offset - pgoff;
+
+#if 0
 	/*
-	 * Can't push pages past end-of-file.
+	 * FIXME: Allow mmap writes past its quota.  The correct fix
+	 * is to register a page_mkwrite() handler to count the page
+	 * against its quota when it is about to be dirtied.
 	 */
-	if (off >= zp->z_size) {
-		/* ignore all pages */
-		err = 0;
-		goto out;
-	} else if (off + len > zp->z_size)
-		len = zp->z_size - off;
-
 	if (zfs_owner_overquota(zsb, zp, B_FALSE) ||
 	    zfs_owner_overquota(zsb, zp, B_TRUE)) {
 		err = EDQUOT;
-		goto out;
 	}
-top:
+#endif
+
+	set_page_writeback(pp);
+	unlock_page(pp);
+
+	rl = zfs_range_lock(zp, pgoff, pglen, RL_WRITER);
 	tx = dmu_tx_create(zsb->z_os);
-	dmu_tx_hold_write(tx, zp->z_id, off, len);
+
+	sync = ((zsb->z_os->os_sync == ZFS_SYNC_ALWAYS) ||
+	        (wbc->sync_mode == WB_SYNC_ALL));
+	if (!sync)
+		dmu_tx_callback_register(tx, zfs_putpage_commit_cb, pp);
+
+	dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
 
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
 	err = dmu_tx_assign(tx, TXG_NOWAIT);
 	if (err != 0) {
-		if (err == ERESTART) {
+		if (err == ERESTART)
 			dmu_tx_wait(tx);
-			dmu_tx_abort(tx);
-			goto top;
-		}
+
+		/* Will call all registered commit callbacks */
 		dmu_tx_abort(tx);
-		goto out;
+
+		/*
+		 * For the synchronous case the commit callback must be
+		 * explicitly called because there is no registered callback.
+		 */
+		if (sync)
+			zfs_putpage_commit_cb(pp, ECANCELED);
+
+		zfs_range_unlock(rl);
+		ZFS_EXIT(zsb);
+		return (err);
 	}
 
 	va = kmap(pp);
-	ASSERT3U(len, <=, PAGESIZE);
-	dmu_write(zsb->z_os, zp->z_id, off, len, va, tx);
+	ASSERT3U(pglen, <=, PAGE_CACHE_SIZE);
+	dmu_write(zsb->z_os, zp->z_id, pgoff, pglen, va, tx);
 	kunmap(pp);
 
-	if (err == 0) {
-		uint64_t mtime[2], ctime[2];
-		sa_bulk_attr_t bulk[3];
-		int count = 0;
+	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
+	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
+	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zsb), NULL, &zp->z_pflags, 8);
 
-		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL,
-		    &mtime, 16);
-		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL,
-		    &ctime, 16);
-		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
-		    &zp->z_pflags, 8);
-		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
-		    B_TRUE);
-		zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, off, len, 0);
-	}
+	/* Preserve the mtime and ctime provided by the inode */
+	ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
+	ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
+	zp->z_atime_dirty = 0;
+	zp->z_seq++;
+
+	err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
+
+	zfs_log_write(zsb->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0);
 	dmu_tx_commit(tx);
 
-out:
+	zfs_range_unlock(rl);
+
+	if (sync) {
+		zil_commit(zsb->z_log, zp->z_id);
+		zfs_putpage_commit_cb(pp, err);
+	}
+
+	ZFS_EXIT(zsb);
 	return (err);
 }
 
 /*
- * Copy the portion of the file indicated from page into the file.
- *
- *	IN:	ip	- inode of file to push page data to.
- *		wbc	- Unused parameter
- *		data	- pointer to address_space
- *
- *	RETURN:	0 if success
- *		error code if failure
- *
- * Timestamps:
- *	vp - ctime|mtime updated
+ * Update the system attributes when the inode has been dirtied.  For the
+ * moment we're conservative and only update the atime, mtime, and ctime.
  */
-/*ARGSUSED*/
 int
-zfs_putpage(struct page *page, struct writeback_control *wbc, void *data)
+zfs_dirty_inode(struct inode *ip, int flags)
 {
-	struct address_space *mapping = data;
-	struct inode         *ip      = mapping->host;
-	znode_t              *zp      = ITOZ(ip);
-	zfs_sb_t             *zsb     = ITOZSB(ip);
-	rl_t		     *rl;
-	u_offset_t	     io_off;
-	size_t		     io_len;
-	size_t		     len;
-	int		     error;
-
-	io_off = page_offset(page);
-	io_len = PAGESIZE;
+	znode_t		*zp = ITOZ(ip);
+	zfs_sb_t	*zsb = ITOZSB(ip);
+	dmu_tx_t	*tx;
+	uint64_t	atime[2], mtime[2], ctime[2];
+	sa_bulk_attr_t	bulk[3];
+	int		error;
+	int		cnt = 0;
 
 	ZFS_ENTER(zsb);
 	ZFS_VERIFY_ZP(zp);
 
-	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
+	tx = dmu_tx_create(zsb->z_os);
 
-	if (io_off > zp->z_size) {
-		/* past end of file */
-		zfs_range_unlock(rl);
-		ZFS_EXIT(zsb);
-		return (0);
+	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
+	zfs_sa_upgrade_txholds(tx, zp);
+
+	error = dmu_tx_assign(tx, TXG_WAIT);
+	if (error) {
+		dmu_tx_abort(tx);
+		goto out;
 	}
 
-	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
+	mutex_enter(&zp->z_lock);
+	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16);
+	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
+	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
 
-	error = zfs_putapage(ip, page, io_off, len);
-	zfs_range_unlock(rl);
+	/* Preserve the mtime and ctime provided by the inode */
+	ZFS_TIME_ENCODE(&ip->i_atime, atime);
+	ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
+	ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
+	zp->z_atime_dirty = 0;
 
-	if (zsb->z_os->os_sync == ZFS_SYNC_ALWAYS)
-		zil_commit(zsb->z_log, zp->z_id);
+	error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
+	mutex_exit(&zp->z_lock);
+
+	dmu_tx_commit(tx);
+out:
 	ZFS_EXIT(zsb);
 	return (error);
 }
-EXPORT_SYMBOL(zfs_putpage);
+EXPORT_SYMBOL(zfs_dirty_inode);
 
 /*ARGSUSED*/
 void
@@ -3856,9 +4020,10 @@ zfs_inactive(struct inode *ip)
 	zfs_sb_t *zsb = ITOZSB(ip);
 	int error;
 
-#ifdef HAVE_SNAPSHOT
-	/* Early return for snapshot inode? */
-#endif /* HAVE_SNAPSHOT */
+	if (zfsctl_is_node(ip)) {
+		zfsctl_inode_inactive(ip);
+		return;
+	}
 
 	rw_enter(&zsb->z_teardown_inactive_lock, RW_READER);
 	if (zp->z_sa_hdl == NULL) {
@@ -4135,6 +4300,17 @@ zfs_space(struct inode *ip, int cmd, flock64_t *bfp, int flag,
 		return (EINVAL);
 	}
 
+	/*
+	 * Permissions aren't checked on Solaris because on this OS
+	 * zfs_space() can only be called with an opened file handle.
+	 * On Linux we can get here through truncate_range() which
+	 * operates directly on inodes, so we need to check access rights.
+	 */
+	if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
+		ZFS_EXIT(zsb);
+		return (error);
+	}
+
 	off = bfp->l_start;
 	len = bfp->l_len; /* 0 means from off to end of file */