forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2023-12-11 072de836f53be56a70cecf70b43ae43b7ce17376 
 kernel/fs/xfs/xfs_iops.c
....@@ -10,38 +10,29 @@
1010 #include "xfs_log_format.h"
1111 #include "xfs_trans_resv.h"
1212 #include "xfs_mount.h"
13
-#include "xfs_da_format.h"
1413 #include "xfs_inode.h"
15
-#include "xfs_bmap.h"
16
-#include "xfs_bmap_util.h"
1714 #include "xfs_acl.h"
1815 #include "xfs_quota.h"
19
-#include "xfs_error.h"
2016 #include "xfs_attr.h"
2117 #include "xfs_trans.h"
2218 #include "xfs_trace.h"
2319 #include "xfs_icache.h"
2420 #include "xfs_symlink.h"
25
-#include "xfs_da_btree.h"
2621 #include "xfs_dir2.h"
27
-#include "xfs_trans_space.h"
2822 #include "xfs_iomap.h"
29
-#include "xfs_defer.h"
23
+#include "xfs_error.h"
3024 
31
-#include <linux/capability.h>
32
-#include <linux/xattr.h>
3325 #include <linux/posix_acl.h>
3426 #include <linux/security.h>
35
-#include <linux/iomap.h>
36
-#include <linux/slab.h>
3727 #include <linux/iversion.h>
28
+#include <linux/fiemap.h>
3829 
3930 /*
40
- * Directories have different lock order w.r.t. mmap_sem compared to regular
31
+ * Directories have different lock order w.r.t. mmap_lock compared to regular
4132  * files. This is due to readdir potentially triggering page faults on a user
4233  * buffer inside filldir(), and this happens with the ilock on the directory
4334  * held. For regular files, the lock order is the other way around - the
44
- * mmap_sem is taken during the page fault, and then we lock the ilock to do
35
+ * mmap_lock is taken during the page fault, and then we lock the ilock to do
4536  * block mapping. Hence we need a different class for the directory ilock so
4637  * that lockdep can tell them apart.
4738  */
....@@ -59,8 +50,15 @@
5950 	int			error = 0;
6051 
6152 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
62
-		error = xfs_attr_set(ip, xattr->name, xattr->value,
63
-				      xattr->value_len, ATTR_SECURE);
53
+		struct xfs_da_args	args = {
54
+			.dp		= ip,
55
+			.attr_filter	= XFS_ATTR_SECURE,
56
+			.name		= xattr->name,
57
+			.namelen	= strlen(xattr->name),
58
+			.value		= xattr->value,
59
+			.valuelen	= xattr->value_len,
60
+		};
61
+		error = xfs_attr_set(&args);
6462 		if (error < 0)
6563 			break;
6664 	}
....@@ -239,7 +237,7 @@
239237 	umode_t		mode,
240238 	bool		flags)
241239 {
242
-	return xfs_vn_mknod(dir, dentry, mode, 0);
240
+	return xfs_generic_create(dir, dentry, mode, 0, false);
243241 }
244242 
245243 STATIC int
....@@ -248,7 +246,7 @@
248246 	struct dentry	*dentry,
249247 	umode_t		mode)
250248 {
251
-	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
249
+	return xfs_generic_create(dir, dentry, mode | S_IFDIR, 0, false);
252250 }
253251 
254252 STATIC struct dentry *
....@@ -480,18 +478,55 @@
480478 	struct inode		*inode,
481479 	struct delayed_call	*done)
482480 {
481
+	struct xfs_inode	*ip = XFS_I(inode);
483482 	char			*link;
484483 
485
-	ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
484
+	ASSERT(ip->i_df.if_flags & XFS_IFINLINE);
486485 
487486 	/*
488487 	 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
489488 	 * if_data is junk.
490489 	 */
491
-	link = XFS_I(inode)->i_df.if_u1.if_data;
492
-	if (!link)
490
+	link = ip->i_df.if_u1.if_data;
491
+	if (XFS_IS_CORRUPT(ip->i_mount, !link))
493492 		return ERR_PTR(-EFSCORRUPTED);
494493 	return link;
494
+}
495
+
496
+static uint32_t
497
+xfs_stat_blksize(
498
+	struct xfs_inode	*ip)
499
+{
500
+	struct xfs_mount	*mp = ip->i_mount;
501
+
502
+	/*
503
+	 * If the file blocks are being allocated from a realtime volume, then
504
+	 * always return the realtime extent size.
505
+	 */
506
+	if (XFS_IS_REALTIME_INODE(ip))
507
+		return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
508
+
509
+	/*
510
+	 * Allow large block sizes to be reported to userspace programs if the
511
+	 * "largeio" mount option is used.
512
+	 *
513
+	 * If compatibility mode is specified, simply return the basic unit of
514
+	 * caching so that we don't get inefficient read/modify/write I/O from
515
+	 * user apps. Otherwise....
516
+	 *
517
+	 * If the underlying volume is a stripe, then return the stripe width in
518
+	 * bytes as the recommended I/O size. It is not a stripe and we've set a
519
+	 * default buffered I/O size, return that, otherwise return the compat
520
+	 * default.
521
+	 */
522
+	if (mp->m_flags & XFS_MOUNT_LARGEIO) {
523
+		if (mp->m_swidth)
524
+			return mp->m_swidth << mp->m_sb.sb_blocklog;
525
+		if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
526
+			return 1U << mp->m_allocsize_log;
527
+	}
528
+
529
+	return PAGE_SIZE;
495530 }
496531 
497532 STATIC int
....@@ -523,11 +558,10 @@
523558 	stat->blocks =
524559 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
525560 
526
-	if (ip->i_d.di_version == 3) {
561
+	if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
527562 		if (request_mask & STATX_BTIME) {
528563 			stat->result_mask |= STATX_BTIME;
529
-			stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
530
-			stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
564
+			stat->btime = ip->i_d.di_crtime;
531565 		}
532566 	}
533567 
....@@ -553,16 +587,7 @@
553587 		stat->rdev = inode->i_rdev;
554588 		break;
555589 	default:
556
-		if (XFS_IS_REALTIME_INODE(ip)) {
557
-			/*
558
-			 * If the file blocks are being allocated from a
559
-			 * realtime volume, then return the inode's realtime
560
-			 * extent size or the realtime volume's extent size.
561
-			 */
562
-			stat->blksize =
563
-				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
564
-		} else
565
-			stat->blksize = xfs_preferred_iosize(mp);
590
+		stat->blksize = xfs_stat_blksize(ip);
566591 		stat->rdev = 0;
567592 		break;
568593 	}
....@@ -672,9 +697,7 @@
672697 		 */
673698 		ASSERT(udqp == NULL);
674699 		ASSERT(gdqp == NULL);
675
-		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
676
-					   xfs_kgid_to_gid(gid),
677
-					   xfs_get_projid(ip),
700
+		error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
678701 					   qflags, &udqp, &gdqp, NULL);
679702 		if (error)
680703 			return error;
....@@ -716,12 +739,7 @@
716739 			if (error)	/* out of quota */
717740 				goto out_cancel;
718741 		}
719
-	}
720742 
721
-	/*
722
-	 * Change file ownership.  Must be the owner or privileged.
723
-	 */
724
-	if (mask & (ATTR_UID|ATTR_GID)) {
725743 		/*
726744 		 * CAP_FSETID overrides the following restrictions:
727745 		 *
....@@ -743,7 +761,6 @@
743761 				olddquot1 = xfs_qm_vop_chown(tp, ip,
744762 							&ip->i_udquot, udqp);
745763 			}
746
-			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
747764 			inode->i_uid = uid;
748765 		}
749766 		if (!gid_eq(igid, gid)) {
....@@ -755,7 +772,6 @@
755772 				olddquot2 = xfs_qm_vop_chown(tp, ip,
756773 							&ip->i_gdquot, gdqp);
757774 			}
758
-			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
759775 			inode->i_gid = gid;
760776 		}
761777 	}
....@@ -857,7 +873,7 @@
857873 	/*
858874 	 * Short circuit the truncate case for zero length files.
859875 	 */
860
-	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
876
+	if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
861877 		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
862878 			return 0;
863879 
....@@ -893,7 +909,7 @@
893909 	if (newsize > oldsize) {
894910 		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
895911 		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
896
-				&did_zeroing, &xfs_iomap_ops);
912
+				&did_zeroing, &xfs_buffered_write_iomap_ops);
897913 	} else {
898914 		/*
899915 		 * iomap won't detect a dirty page over an unwritten block (or a
....@@ -906,7 +922,7 @@
906922 		if (error)
907923 			return error;
908924 		error = iomap_truncate_page(inode, newsize, &did_zeroing,
909
-				&xfs_iomap_ops);
925
+				&xfs_buffered_write_iomap_ops);
910926 	}
911927 
912928 	if (error)
....@@ -1134,7 +1150,7 @@
11341150 				&xfs_xattr_iomap_ops);
11351151 	} else {
11361152 		error = iomap_fiemap(inode, fieinfo, start, length,
1137
-				&xfs_iomap_ops);
1153
+				&xfs_read_iomap_ops);
11381154 	}
11391155 	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
11401156 
....@@ -1233,13 +1249,12 @@
12331249 {
12341250 	struct xfs_mount	*mp = ip->i_mount;
12351251 
1236
-	/* Only supported on non-reflinked files. */
1237
-	if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1252
+	/* Only supported on regular files. */
1253
+	if (!S_ISREG(VFS_I(ip)->i_mode))
12381254 		return false;
12391255 
1240
-	/* DAX mount option or DAX iflag must be set. */
1241
-	if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1242
-	    !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1256
+	/* Only supported on non-reflinked files. */
1257
+	if (xfs_is_reflink_inode(ip))
12431258 		return false;
12441259 
12451260 	/* Block size must match page size */
....@@ -1247,29 +1262,54 @@
12471262 		return false;
12481263 
12491264 	/* Device has to support DAX too. */
1250
-	return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL;
1265
+	return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
12511266 }
12521267 
1253
-STATIC void
1254
-xfs_diflags_to_iflags(
1255
-	struct inode		*inode,
1256
-	struct xfs_inode	*ip)
1268
+static bool
1269
+xfs_inode_should_enable_dax(
1270
+	struct xfs_inode *ip)
12571271 {
1258
-	uint16_t		flags = ip->i_d.di_flags;
1272
+	if (!IS_ENABLED(CONFIG_FS_DAX))
1273
+		return false;
1274
+	if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1275
+		return false;
1276
+	if (!xfs_inode_supports_dax(ip))
1277
+		return false;
1278
+	if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1279
+		return true;
1280
+	if (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
1281
+		return true;
1282
+	return false;
1283
+}
12591284 
1260
-	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1261
-			    S_NOATIME | S_DAX);
1285
+void
1286
+xfs_diflags_to_iflags(
1287
+	struct xfs_inode	*ip,
1288
+	bool init)
1289
+{
1290
+	struct inode            *inode = VFS_I(ip);
1291
+	unsigned int            xflags = xfs_ip2xflags(ip);
1292
+	unsigned int            flags = 0;
12621293 
1263
-	if (flags & XFS_DIFLAG_IMMUTABLE)
1264
-		inode->i_flags |= S_IMMUTABLE;
1265
-	if (flags & XFS_DIFLAG_APPEND)
1266
-		inode->i_flags |= S_APPEND;
1267
-	if (flags & XFS_DIFLAG_SYNC)
1268
-		inode->i_flags |= S_SYNC;
1269
-	if (flags & XFS_DIFLAG_NOATIME)
1270
-		inode->i_flags |= S_NOATIME;
1271
-	if (xfs_inode_supports_dax(ip))
1272
-		inode->i_flags |= S_DAX;
1294
+	ASSERT(!(IS_DAX(inode) && init));
1295
+
1296
+	if (xflags & FS_XFLAG_IMMUTABLE)
1297
+		flags |= S_IMMUTABLE;
1298
+	if (xflags & FS_XFLAG_APPEND)
1299
+		flags |= S_APPEND;
1300
+	if (xflags & FS_XFLAG_SYNC)
1301
+		flags |= S_SYNC;
1302
+	if (xflags & FS_XFLAG_NOATIME)
1303
+		flags |= S_NOATIME;
1304
+	if (init && xfs_inode_should_enable_dax(ip))
1305
+		flags |= S_DAX;
1306
+
1307
+	/*
1308
+	 * S_DAX can only be set during inode initialization and is never set by
1309
+	 * the VFS, so we cannot mask off S_DAX in i_flags.
1310
+	 */
1311
+	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1312
+	inode->i_flags |= flags;
12731313 }
12741314 
12751315 /*
....@@ -1288,17 +1328,14 @@
12881328 	gfp_t			gfp_mask;
12891329 
12901330 	inode->i_ino = ip->i_ino;
1291
-	inode->i_state = I_NEW;
1331
+	inode->i_state |= I_NEW;
12921332 
12931333 	inode_sb_list_add(inode);
12941334 	/* make the inode look hashed for the writeback code */
12951335 	inode_fake_hash(inode);
12961336 
1297
-	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1298
-	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1299
-
13001337 	i_size_write(inode, ip->i_d.di_size);
1301
-	xfs_diflags_to_iflags(inode, ip);
1338
+	xfs_diflags_to_iflags(ip, true);
13021339 
13031340 	if (S_ISDIR(inode->i_mode)) {
13041341 		/*
....@@ -1310,9 +1347,7 @@
13101347 		lockdep_set_class(&inode->i_rwsem,
13111348 				  &inode->i_sb->s_type->i_mutex_dir_key);
13121349 		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1313
-		ip->d_ops = ip->i_mount->m_dir_inode_ops;
13141350 	} else {
1315
-		ip->d_ops = ip->i_mount->m_nondir_inode_ops;
13161351 		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
13171352 	}
13181353