aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorDavid Chinner <dgc@sgi.com>2007-07-10 21:09:12 -0400
committerTim Shimmin <tes@chook.melbourne.sgi.com>2007-07-14 01:40:53 -0400
commit2a82b8be8a8dacb48cb7371449a7a9daa558b4a8 (patch)
tree44e6a81dd0e7d7dc634e04b9230b5262a254c5ee
parent0892ccd6fe13e08ad9e57007afbb78fe02d66005 (diff)
[XFS] Concurrent Multi-File Data Streams
In media spaces, video is often stored in a frame-per-file format. When dealing with uncompressed realtime HD video streams in this format, it is crucial that files do not get fragmented and that multiple files a placed contiguously on disk. When multiple streams are being ingested and played out at the same time, it is critical that the filesystem does not cross the streams and interleave them together as this creates seek and readahead cache miss latency and prevents both ingest and playout from meeting frame rate targets. This patch set creates a "stream of files" concept into the allocator to place all the data from a single stream contiguously on disk so that RAID array readahead can be used effectively. Each additional stream gets placed in different allocation groups within the filesystem, thereby ensuring that we don't cross any streams. When an AG fills up, we select a new AG for the stream that is not in use. The core of the functionality is the stream tracking - each inode that we create in a directory needs to be associated with the directories' stream. Hence every time we create a file, we look up the directories' stream object and associate the new file with that object. Once we have a stream object for a file, we use the AG that the stream object point to for allocations. If we can't allocate in that AG (e.g. it is full) we move the entire stream to another AG. Other inodes in the same stream are moved to the new AG on their next allocation (i.e. lazy update). Stream objects are kept in a cache and hold a reference on the inode. Hence the inode cannot be reclaimed while there is an outstanding stream reference. This means that on unlink we need to remove the stream association and we also need to flush all the associations on certain events that want to reclaim all unreferenced inodes (e.g. filesystem freeze). SGI-PV: 964469 SGI-Modid: xfs-linux-melb:xfs-kern:29096a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Barry Naujok <bnaujok@sgi.com> Signed-off-by: Donald Douwsma <donaldd@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com> Signed-off-by: Vlad Apostolov <vapo@sgi.com>
Diffstat
-rw-r--r--fs/xfs/Makefile-linux-2.62
-rw-r--r--fs/xfs/linux-2.6/xfs_globals.c1
-rw-r--r--fs/xfs/linux-2.6/xfs_linux.h1
-rw-r--r--fs/xfs/linux-2.6/xfs_sysctl.c11
-rw-r--r--fs/xfs/linux-2.6/xfs_sysctl.h2
-rw-r--r--fs/xfs/xfs.h1
-rw-r--r--fs/xfs/xfs_ag.h1
-rw-r--r--fs/xfs/xfs_bmap.c69
-rw-r--r--fs/xfs/xfs_clnt.h2
-rw-r--r--fs/xfs/xfs_dinode.h4
-rw-r--r--fs/xfs/xfs_filestream.c771
-rw-r--r--fs/xfs/xfs_filestream.h136
-rw-r--r--fs/xfs/xfs_fs.h1
-rw-r--r--fs/xfs/xfs_fsops.c2
-rw-r--r--fs/xfs/xfs_inode.c17
-rw-r--r--fs/xfs/xfs_inode.h1
-rw-r--r--fs/xfs/xfs_mount.h4
-rw-r--r--fs/xfs/xfs_mru_cache.c608
-rw-r--r--fs/xfs/xfs_mru_cache.h57
-rw-r--r--fs/xfs/xfs_vfsops.c26
-rw-r--r--fs/xfs/xfs_vnodeops.c25
21 files changed, 1730 insertions, 12 deletions
diff --git a/fs/xfs/Makefile-linux-2.6 b/fs/xfs/Makefile-linux-2.6
index b49989bb89ad..e7a9a83f0087 100644
--- a/fs/xfs/Makefile-linux-2.6
+++ b/fs/xfs/Makefile-linux-2.6
@@ -64,6 +64,7 @@ xfs-y += xfs_alloc.o \
64 xfs_dir2_sf.o \ 64 xfs_dir2_sf.o \
65 xfs_error.o \ 65 xfs_error.o \
66 xfs_extfree_item.o \ 66 xfs_extfree_item.o \
67 xfs_filestream.o \
67 xfs_fsops.o \ 68 xfs_fsops.o \
68 xfs_ialloc.o \ 69 xfs_ialloc.o \
69 xfs_ialloc_btree.o \ 70 xfs_ialloc_btree.o \
@@ -77,6 +78,7 @@ xfs-y += xfs_alloc.o \
77 xfs_log.o \ 78 xfs_log.o \
78 xfs_log_recover.o \ 79 xfs_log_recover.o \
79 xfs_mount.o \ 80 xfs_mount.o \
81 xfs_mru_cache.o \
80 xfs_rename.o \ 82 xfs_rename.o \
81 xfs_trans.o \ 83 xfs_trans.o \
82 xfs_trans_ail.o \ 84 xfs_trans_ail.o \
diff --git a/fs/xfs/linux-2.6/xfs_globals.c b/fs/xfs/linux-2.6/xfs_globals.c
index ed3a5e1b4b67..bb72c3d4141f 100644
--- a/fs/xfs/linux-2.6/xfs_globals.c
+++ b/fs/xfs/linux-2.6/xfs_globals.c
@@ -46,6 +46,7 @@ xfs_param_t xfs_params = {
46 .inherit_nosym = { 0, 0, 1 }, 46 .inherit_nosym = { 0, 0, 1 },
47 .rotorstep = { 1, 1, 255 }, 47 .rotorstep = { 1, 1, 255 },
48 .inherit_nodfrg = { 0, 1, 1 }, 48 .inherit_nodfrg = { 0, 1, 1 },
49 .fstrm_timer = { 1, 50, 3600*100},
49}; 50};
50 51
51/* 52/*
diff --git a/fs/xfs/linux-2.6/xfs_linux.h b/fs/xfs/linux-2.6/xfs_linux.h
index af24a457d3a3..330c4ba9d404 100644
--- a/fs/xfs/linux-2.6/xfs_linux.h
+++ b/fs/xfs/linux-2.6/xfs_linux.h
@@ -123,6 +123,7 @@
123#define xfs_inherit_nosymlinks xfs_params.inherit_nosym.val 123#define xfs_inherit_nosymlinks xfs_params.inherit_nosym.val
124#define xfs_rotorstep xfs_params.rotorstep.val 124#define xfs_rotorstep xfs_params.rotorstep.val
125#define xfs_inherit_nodefrag xfs_params.inherit_nodfrg.val 125#define xfs_inherit_nodefrag xfs_params.inherit_nodfrg.val
126#define xfs_fstrm_centisecs xfs_params.fstrm_timer.val
126 127
127#define current_cpu() (raw_smp_processor_id()) 128#define current_cpu() (raw_smp_processor_id())
128#define current_pid() (current->pid) 129#define current_pid() (current->pid)
diff --git a/fs/xfs/linux-2.6/xfs_sysctl.c b/fs/xfs/linux-2.6/xfs_sysctl.c
index cd6eaa44aa2b..bb997d75c05c 100644
--- a/fs/xfs/linux-2.6/xfs_sysctl.c
+++ b/fs/xfs/linux-2.6/xfs_sysctl.c
@@ -210,6 +210,17 @@ static ctl_table xfs_table[] = {
210 .extra1 = &xfs_params.inherit_nodfrg.min, 210 .extra1 = &xfs_params.inherit_nodfrg.min,
211 .extra2 = &xfs_params.inherit_nodfrg.max 211 .extra2 = &xfs_params.inherit_nodfrg.max
212 }, 212 },
213 {
214 .ctl_name = XFS_FILESTREAM_TIMER,
215 .procname = "filestream_centisecs",
216 .data = &xfs_params.fstrm_timer.val,
217 .maxlen = sizeof(int),
218 .mode = 0644,
219 .proc_handler = &proc_dointvec_minmax,
220 .strategy = &sysctl_intvec,
221 .extra1 = &xfs_params.fstrm_timer.min,
222 .extra2 = &xfs_params.fstrm_timer.max,
223 },
213 /* please keep this the last entry */ 224 /* please keep this the last entry */
214#ifdef CONFIG_PROC_FS 225#ifdef CONFIG_PROC_FS
215 { 226 {
diff --git a/fs/xfs/linux-2.6/xfs_sysctl.h b/fs/xfs/linux-2.6/xfs_sysctl.h
index a631fb8cc5ac..98b97e399d6f 100644
--- a/fs/xfs/linux-2.6/xfs_sysctl.h
+++ b/fs/xfs/linux-2.6/xfs_sysctl.h
@@ -47,6 +47,7 @@ typedef struct xfs_param {
47 xfs_sysctl_val_t inherit_nosym; /* Inherit the "nosymlinks" flag. */ 47 xfs_sysctl_val_t inherit_nosym; /* Inherit the "nosymlinks" flag. */
48 xfs_sysctl_val_t rotorstep; /* inode32 AG rotoring control knob */ 48 xfs_sysctl_val_t rotorstep; /* inode32 AG rotoring control knob */
49 xfs_sysctl_val_t inherit_nodfrg;/* Inherit the "nodefrag" inode flag. */ 49 xfs_sysctl_val_t inherit_nodfrg;/* Inherit the "nodefrag" inode flag. */
50 xfs_sysctl_val_t fstrm_timer; /* Filestream dir-AG assoc'n timeout. */
50} xfs_param_t; 51} xfs_param_t;
51 52
52/* 53/*
@@ -86,6 +87,7 @@ enum {
86 XFS_INHERIT_NOSYM = 19, 87 XFS_INHERIT_NOSYM = 19,
87 XFS_ROTORSTEP = 20, 88 XFS_ROTORSTEP = 20,
88 XFS_INHERIT_NODFRG = 21, 89 XFS_INHERIT_NODFRG = 21,
90 XFS_FILESTREAM_TIMER = 22,
89}; 91};
90 92
91extern xfs_param_t xfs_params; 93extern xfs_param_t xfs_params;
diff --git a/fs/xfs/xfs.h b/fs/xfs/xfs.h
index bf0a12040b13..b5a7d92c6843 100644
--- a/fs/xfs/xfs.h
+++ b/fs/xfs/xfs.h
@@ -38,6 +38,7 @@
38#define XFS_RW_TRACE 1 38#define XFS_RW_TRACE 1
39#define XFS_BUF_TRACE 1 39#define XFS_BUF_TRACE 1
40#define XFS_VNODE_TRACE 1 40#define XFS_VNODE_TRACE 1
41#define XFS_FILESTREAMS_TRACE 1
41#endif 42#endif
42 43
43#include <linux-2.6/xfs_linux.h> 44#include <linux-2.6/xfs_linux.h>
diff --git a/fs/xfs/xfs_ag.h b/fs/xfs/xfs_ag.h
index b1dd0029c60e..51c09c114a20 100644
--- a/fs/xfs/xfs_ag.h
+++ b/fs/xfs/xfs_ag.h
@@ -196,6 +196,7 @@ typedef struct xfs_perag
196 lock_t pagb_lock; /* lock for pagb_list */ 196 lock_t pagb_lock; /* lock for pagb_list */
197#endif 197#endif
198 xfs_perag_busy_t *pagb_list; /* unstable blocks */ 198 xfs_perag_busy_t *pagb_list; /* unstable blocks */
199 atomic_t pagf_fstrms; /* # of filestreams active in this AG */
199} xfs_perag_t; 200} xfs_perag_t;
200 201
201#define XFS_AG_MAXLEVELS(mp) ((mp)->m_ag_maxlevels) 202#define XFS_AG_MAXLEVELS(mp) ((mp)->m_ag_maxlevels)
diff --git a/fs/xfs/xfs_bmap.c b/fs/xfs/xfs_bmap.c
index 09d86388bb71..51ba689a4552 100644
--- a/fs/xfs/xfs_bmap.c
+++ b/fs/xfs/xfs_bmap.c
@@ -52,6 +52,7 @@
52#include "xfs_quota.h" 52#include "xfs_quota.h"
53#include "xfs_trans_space.h" 53#include "xfs_trans_space.h"
54#include "xfs_buf_item.h" 54#include "xfs_buf_item.h"
55#include "xfs_filestream.h"
55 56
56 57
57#ifdef DEBUG 58#ifdef DEBUG
@@ -2725,9 +2726,15 @@ xfs_bmap_btalloc(
2725 } 2726 }
2726 nullfb = ap->firstblock == NULLFSBLOCK; 2727 nullfb = ap->firstblock == NULLFSBLOCK;
2727 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock); 2728 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2728 if (nullfb) 2729 if (nullfb) {
2729 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino); 2730 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2730 else 2731 ag = xfs_filestream_lookup_ag(ap->ip);
2732 ag = (ag != NULLAGNUMBER) ? ag : 0;
2733 ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2734 } else {
2735 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2736 }
2737 } else
2731 ap->rval = ap->firstblock; 2738 ap->rval = ap->firstblock;
2732 2739
2733 xfs_bmap_adjacent(ap); 2740 xfs_bmap_adjacent(ap);
@@ -2751,13 +2758,22 @@ xfs_bmap_btalloc(
2751 args.firstblock = ap->firstblock; 2758 args.firstblock = ap->firstblock;
2752 blen = 0; 2759 blen = 0;
2753 if (nullfb) { 2760 if (nullfb) {
2754 args.type = XFS_ALLOCTYPE_START_BNO; 2761 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2762 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2763 else
2764 args.type = XFS_ALLOCTYPE_START_BNO;
2755 args.total = ap->total; 2765 args.total = ap->total;
2766
2756 /* 2767 /*
2757 * Find the longest available space. 2768 * Search for an allocation group with a single extent
2758 * We're going to try for the whole allocation at once. 2769 * large enough for the request.
2770 *
2771 * If one isn't found, then adjust the minimum allocation
2772 * size to the largest space found.
2759 */ 2773 */
2760 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno); 2774 startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2775 if (startag == NULLAGNUMBER)
2776 startag = ag = 0;
2761 notinit = 0; 2777 notinit = 0;
2762 down_read(&mp->m_peraglock); 2778 down_read(&mp->m_peraglock);
2763 while (blen < ap->alen) { 2779 while (blen < ap->alen) {
@@ -2783,6 +2799,35 @@ xfs_bmap_btalloc(
2783 blen = longest; 2799 blen = longest;
2784 } else 2800 } else
2785 notinit = 1; 2801 notinit = 1;
2802
2803 if (xfs_inode_is_filestream(ap->ip)) {
2804 if (blen >= ap->alen)
2805 break;
2806
2807 if (ap->userdata) {
2808 /*
2809 * If startag is an invalid AG, we've
2810 * come here once before and
2811 * xfs_filestream_new_ag picked the
2812 * best currently available.
2813 *
2814 * Don't continue looping, since we
2815 * could loop forever.
2816 */
2817 if (startag == NULLAGNUMBER)
2818 break;
2819
2820 error = xfs_filestream_new_ag(ap, &ag);
2821 if (error) {
2822 up_read(&mp->m_peraglock);
2823 return error;
2824 }
2825
2826 /* loop again to set 'blen'*/
2827 startag = NULLAGNUMBER;
2828 continue;
2829 }
2830 }
2786 if (++ag == mp->m_sb.sb_agcount) 2831 if (++ag == mp->m_sb.sb_agcount)
2787 ag = 0; 2832 ag = 0;
2788 if (ag == startag) 2833 if (ag == startag)
@@ -2807,8 +2852,18 @@ xfs_bmap_btalloc(
2807 */ 2852 */
2808 else 2853 else
2809 args.minlen = ap->alen; 2854 args.minlen = ap->alen;
2855
2856 /*
2857 * set the failure fallback case to look in the selected
2858 * AG as the stream may have moved.
2859 */
2860 if (xfs_inode_is_filestream(ap->ip))
2861 ap->rval = args.fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2810 } else if (ap->low) { 2862 } else if (ap->low) {
2811 args.type = XFS_ALLOCTYPE_START_BNO; 2863 if (xfs_inode_is_filestream(ap->ip))
2864 args.type = XFS_ALLOCTYPE_FIRST_AG;
2865 else
2866 args.type = XFS_ALLOCTYPE_START_BNO;
2812 args.total = args.minlen = ap->minlen; 2867 args.total = args.minlen = ap->minlen;
2813 } else { 2868 } else {
2814 args.type = XFS_ALLOCTYPE_NEAR_BNO; 2869 args.type = XFS_ALLOCTYPE_NEAR_BNO;
diff --git a/fs/xfs/xfs_clnt.h b/fs/xfs/xfs_clnt.h
index 5b7eb81453be..f89196cb08d2 100644
--- a/fs/xfs/xfs_clnt.h
+++ b/fs/xfs/xfs_clnt.h
@@ -99,5 +99,7 @@ struct xfs_mount_args {
99 */ 99 */
100#define XFSMNT2_COMPAT_IOSIZE 0x00000001 /* don't report large preferred 100#define XFSMNT2_COMPAT_IOSIZE 0x00000001 /* don't report large preferred
101 * I/O size in stat(2) */ 101 * I/O size in stat(2) */
102#define XFSMNT2_FILESTREAMS 0x00000002 /* enable the filestreams
103 * allocator */
102 104
103#endif /* __XFS_CLNT_H__ */ 105#endif /* __XFS_CLNT_H__ */
diff --git a/fs/xfs/xfs_dinode.h b/fs/xfs/xfs_dinode.h
index b33826961c45..fefd0116bac9 100644
--- a/fs/xfs/xfs_dinode.h
+++ b/fs/xfs/xfs_dinode.h
@@ -257,6 +257,7 @@ typedef enum xfs_dinode_fmt
257#define XFS_DIFLAG_EXTSIZE_BIT 11 /* inode extent size allocator hint */ 257#define XFS_DIFLAG_EXTSIZE_BIT 11 /* inode extent size allocator hint */
258#define XFS_DIFLAG_EXTSZINHERIT_BIT 12 /* inherit inode extent size */ 258#define XFS_DIFLAG_EXTSZINHERIT_BIT 12 /* inherit inode extent size */
259#define XFS_DIFLAG_NODEFRAG_BIT 13 /* do not reorganize/defragment */ 259#define XFS_DIFLAG_NODEFRAG_BIT 13 /* do not reorganize/defragment */
260#define XFS_DIFLAG_FILESTREAM_BIT 14 /* use filestream allocator */
260#define XFS_DIFLAG_REALTIME (1 << XFS_DIFLAG_REALTIME_BIT) 261#define XFS_DIFLAG_REALTIME (1 << XFS_DIFLAG_REALTIME_BIT)
261#define XFS_DIFLAG_PREALLOC (1 << XFS_DIFLAG_PREALLOC_BIT) 262#define XFS_DIFLAG_PREALLOC (1 << XFS_DIFLAG_PREALLOC_BIT)
262#define XFS_DIFLAG_NEWRTBM (1 << XFS_DIFLAG_NEWRTBM_BIT) 263#define XFS_DIFLAG_NEWRTBM (1 << XFS_DIFLAG_NEWRTBM_BIT)
@@ -271,12 +272,13 @@ typedef enum xfs_dinode_fmt
271#define XFS_DIFLAG_EXTSIZE (1 << XFS_DIFLAG_EXTSIZE_BIT) 272#define XFS_DIFLAG_EXTSIZE (1 << XFS_DIFLAG_EXTSIZE_BIT)
272#define XFS_DIFLAG_EXTSZINHERIT (1 << XFS_DIFLAG_EXTSZINHERIT_BIT) 273#define XFS_DIFLAG_EXTSZINHERIT (1 << XFS_DIFLAG_EXTSZINHERIT_BIT)
273#define XFS_DIFLAG_NODEFRAG (1 << XFS_DIFLAG_NODEFRAG_BIT) 274#define XFS_DIFLAG_NODEFRAG (1 << XFS_DIFLAG_NODEFRAG_BIT)
275#define XFS_DIFLAG_FILESTREAM (1 << XFS_DIFLAG_FILESTREAM_BIT)
274 276
275#define XFS_DIFLAG_ANY \ 277#define XFS_DIFLAG_ANY \
276 (XFS_DIFLAG_REALTIME | XFS_DIFLAG_PREALLOC | XFS_DIFLAG_NEWRTBM | \ 278 (XFS_DIFLAG_REALTIME | XFS_DIFLAG_PREALLOC | XFS_DIFLAG_NEWRTBM | \
277 XFS_DIFLAG_IMMUTABLE | XFS_DIFLAG_APPEND | XFS_DIFLAG_SYNC | \ 279 XFS_DIFLAG_IMMUTABLE | XFS_DIFLAG_APPEND | XFS_DIFLAG_SYNC | \
278 XFS_DIFLAG_NOATIME | XFS_DIFLAG_NODUMP | XFS_DIFLAG_RTINHERIT | \ 280 XFS_DIFLAG_NOATIME | XFS_DIFLAG_NODUMP | XFS_DIFLAG_RTINHERIT | \
279 XFS_DIFLAG_PROJINHERIT | XFS_DIFLAG_NOSYMLINKS | XFS_DIFLAG_EXTSIZE | \ 281 XFS_DIFLAG_PROJINHERIT | XFS_DIFLAG_NOSYMLINKS | XFS_DIFLAG_EXTSIZE | \
280 XFS_DIFLAG_EXTSZINHERIT | XFS_DIFLAG_NODEFRAG) 282 XFS_DIFLAG_EXTSZINHERIT | XFS_DIFLAG_NODEFRAG | XFS_DIFLAG_FILESTREAM)
281 283
282#endif /* __XFS_DINODE_H__ */ 284#endif /* __XFS_DINODE_H__ */
diff --git a/fs/xfs/xfs_filestream.c b/fs/xfs/xfs_filestream.c
new file mode 100644
index 000000000000..ce2278611bb7
--- /dev/null
+++ b/fs/xfs/xfs_filestream.c
@@ -0,0 +1,771 @@
1/*
2 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_bmap_btree.h"
20#include "xfs_inum.h"
21#include "xfs_dir2.h"
22#include "xfs_dir2_sf.h"
23#include "xfs_attr_sf.h"
24#include "xfs_dinode.h"
25#include "xfs_inode.h"
26#include "xfs_ag.h"
27#include "xfs_dmapi.h"
28#include "xfs_log.h"
29#include "xfs_trans.h"
30#include "xfs_sb.h"
31#include "xfs_mount.h"
32#include "xfs_bmap.h"
33#include "xfs_alloc.h"
34#include "xfs_utils.h"
35#include "xfs_mru_cache.h"
36#include "xfs_filestream.h"
37
38#ifdef XFS_FILESTREAMS_TRACE
39
40ktrace_t *xfs_filestreams_trace_buf;
41
42STATIC void
43xfs_filestreams_trace(
44 xfs_mount_t *mp, /* mount point */
45 int type, /* type of trace */
46 const char *func, /* source function */
47 int line, /* source line number */
48 __psunsigned_t arg0,
49 __psunsigned_t arg1,
50 __psunsigned_t arg2,
51 __psunsigned_t arg3,
52 __psunsigned_t arg4,
53 __psunsigned_t arg5)
54{
55 ktrace_enter(xfs_filestreams_trace_buf,
56 (void *)(__psint_t)(type | (line << 16)),
57 (void *)func,
58 (void *)(__psunsigned_t)current_pid(),
59 (void *)mp,
60 (void *)(__psunsigned_t)arg0,
61 (void *)(__psunsigned_t)arg1,
62 (void *)(__psunsigned_t)arg2,
63 (void *)(__psunsigned_t)arg3,
64 (void *)(__psunsigned_t)arg4,
65 (void *)(__psunsigned_t)arg5,
66 NULL, NULL, NULL, NULL, NULL, NULL);
67}
68
69#define TRACE0(mp,t) TRACE6(mp,t,0,0,0,0,0,0)
70#define TRACE1(mp,t,a0) TRACE6(mp,t,a0,0,0,0,0,0)
71#define TRACE2(mp,t,a0,a1) TRACE6(mp,t,a0,a1,0,0,0,0)
72#define TRACE3(mp,t,a0,a1,a2) TRACE6(mp,t,a0,a1,a2,0,0,0)
73#define TRACE4(mp,t,a0,a1,a2,a3) TRACE6(mp,t,a0,a1,a2,a3,0,0)
74#define TRACE5(mp,t,a0,a1,a2,a3,a4) TRACE6(mp,t,a0,a1,a2,a3,a4,0)
75#define TRACE6(mp,t,a0,a1,a2,a3,a4,a5) \
76 xfs_filestreams_trace(mp, t, __FUNCTION__, __LINE__, \
77 (__psunsigned_t)a0, (__psunsigned_t)a1, \
78 (__psunsigned_t)a2, (__psunsigned_t)a3, \
79 (__psunsigned_t)a4, (__psunsigned_t)a5)
80
81#define TRACE_AG_SCAN(mp, ag, ag2) \
82 TRACE2(mp, XFS_FSTRM_KTRACE_AGSCAN, ag, ag2);
83#define TRACE_AG_PICK1(mp, max_ag, maxfree) \
84 TRACE2(mp, XFS_FSTRM_KTRACE_AGPICK1, max_ag, maxfree);
85#define TRACE_AG_PICK2(mp, ag, ag2, cnt, free, scan, flag) \
86 TRACE6(mp, XFS_FSTRM_KTRACE_AGPICK2, ag, ag2, \
87 cnt, free, scan, flag)
88#define TRACE_UPDATE(mp, ip, ag, cnt, ag2, cnt2) \
89 TRACE5(mp, XFS_FSTRM_KTRACE_UPDATE, ip, ag, cnt, ag2, cnt2)
90#define TRACE_FREE(mp, ip, pip, ag, cnt) \
91 TRACE4(mp, XFS_FSTRM_KTRACE_FREE, ip, pip, ag, cnt)
92#define TRACE_LOOKUP(mp, ip, pip, ag, cnt) \
93 TRACE4(mp, XFS_FSTRM_KTRACE_ITEM_LOOKUP, ip, pip, ag, cnt)
94#define TRACE_ASSOCIATE(mp, ip, pip, ag, cnt) \
95 TRACE4(mp, XFS_FSTRM_KTRACE_ASSOCIATE, ip, pip, ag, cnt)
96#define TRACE_MOVEAG(mp, ip, pip, oag, ocnt, nag, ncnt) \
97 TRACE6(mp, XFS_FSTRM_KTRACE_MOVEAG, ip, pip, oag, ocnt, nag, ncnt)
98#define TRACE_ORPHAN(mp, ip, ag) \
99 TRACE2(mp, XFS_FSTRM_KTRACE_ORPHAN, ip, ag);
100
101
102#else
103#define TRACE_AG_SCAN(mp, ag, ag2)
104#define TRACE_AG_PICK1(mp, max_ag, maxfree)
105#define TRACE_AG_PICK2(mp, ag, ag2, cnt, free, scan, flag)
106#define TRACE_UPDATE(mp, ip, ag, cnt, ag2, cnt2)
107#define TRACE_FREE(mp, ip, pip, ag, cnt)
108#define TRACE_LOOKUP(mp, ip, pip, ag, cnt)
109#define TRACE_ASSOCIATE(mp, ip, pip, ag, cnt)
110#define TRACE_MOVEAG(mp, ip, pip, oag, ocnt, nag, ncnt)
111#define TRACE_ORPHAN(mp, ip, ag)
112#endif
113
114static kmem_zone_t *item_zone;
115
116/*
117 * Structure for associating a file or a directory with an allocation group.
118 * The parent directory pointer is only needed for files, but since there will
119 * generally be vastly more files than directories in the cache, using the same
120 * data structure simplifies the code with very little memory overhead.
121 */
122typedef struct fstrm_item
123{
124 xfs_agnumber_t ag; /* AG currently in use for the file/directory. */
125 xfs_inode_t *ip; /* inode self-pointer. */
126 xfs_inode_t *pip; /* Parent directory inode pointer. */
127} fstrm_item_t;
128
129
130/*
131 * Scan the AGs starting at startag looking for an AG that isn't in use and has
132 * at least minlen blocks free.
133 */
134static int
135_xfs_filestream_pick_ag(
136 xfs_mount_t *mp,
137 xfs_agnumber_t startag,
138 xfs_agnumber_t *agp,
139 int flags,
140 xfs_extlen_t minlen)
141{
142 int err, trylock, nscan;
143 xfs_extlen_t delta, longest, need, free, minfree, maxfree = 0;
144 xfs_agnumber_t ag, max_ag = NULLAGNUMBER;
145 struct xfs_perag *pag;
146
147 /* 2% of an AG's blocks must be free for it to be chosen. */
148 minfree = mp->m_sb.sb_agblocks / 50;
149
150 ag = startag;
151 *agp = NULLAGNUMBER;
152
153 /* For the first pass, don't sleep trying to init the per-AG. */
154 trylock = XFS_ALLOC_FLAG_TRYLOCK;
155
156 for (nscan = 0; 1; nscan++) {
157
158 TRACE_AG_SCAN(mp, ag, xfs_filestream_peek_ag(mp, ag));
159
160 pag = mp->m_perag + ag;
161
162 if (!pag->pagf_init) {
163 err = xfs_alloc_pagf_init(mp, NULL, ag, trylock);
164 if (err && !trylock)
165 return err;
166 }
167
168 /* Might fail sometimes during the 1st pass with trylock set. */
169 if (!pag->pagf_init)
170 goto next_ag;
171
172 /* Keep track of the AG with the most free blocks. */
173 if (pag->pagf_freeblks > maxfree) {
174 maxfree = pag->pagf_freeblks;
175 max_ag = ag;
176 }
177
178 /*
179 * The AG reference count does two things: it enforces mutual
180 * exclusion when examining the suitability of an AG in this
181 * loop, and it guards against two filestreams being established
182 * in the same AG as each other.
183 */
184 if (xfs_filestream_get_ag(mp, ag) > 1) {
185 xfs_filestream_put_ag(mp, ag);
186 goto next_ag;
187 }
188
189 need = XFS_MIN_FREELIST_PAG(pag, mp);
190 delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0;
191 longest = (pag->pagf_longest > delta) ?
192 (pag->pagf_longest - delta) :
193 (pag->pagf_flcount > 0 || pag->pagf_longest > 0);
194
195 if (((minlen && longest >= minlen) ||
196 (!minlen && pag->pagf_freeblks >= minfree)) &&
197 (!pag->pagf_metadata || !(flags & XFS_PICK_USERDATA) ||
198 (flags & XFS_PICK_LOWSPACE))) {
199
200 /* Break out, retaining the reference on the AG. */
201 free = pag->pagf_freeblks;
202 *agp = ag;
203 break;
204 }
205
206 /* Drop the reference on this AG, it's not usable. */
207 xfs_filestream_put_ag(mp, ag);
208next_ag:
209 /* Move to the next AG, wrapping to AG 0 if necessary. */
210 if (++ag >= mp->m_sb.sb_agcount)
211 ag = 0;
212
213 /* If a full pass of the AGs hasn't been done yet, continue. */
214 if (ag != startag)
215 continue;
216
217 /* Allow sleeping in xfs_alloc_pagf_init() on the 2nd pass. */
218 if (trylock != 0) {
219 trylock = 0;
220 continue;
221 }
222
223 /* Finally, if lowspace wasn't set, set it for the 3rd pass. */
224 if (!(flags & XFS_PICK_LOWSPACE)) {
225 flags |= XFS_PICK_LOWSPACE;
226 continue;
227 }
228
229 /*
230 * Take the AG with the most free space, regardless of whether
231 * it's already in use by another filestream.
232 */
233 if (max_ag != NULLAGNUMBER) {
234 xfs_filestream_get_ag(mp, max_ag);
235 TRACE_AG_PICK1(mp, max_ag, maxfree);
236 free = maxfree;
237 *agp = max_ag;
238 break;
239 }
240
241 /* take AG 0 if none matched */
242 TRACE_AG_PICK1(mp, max_ag, maxfree);
243 *agp = 0;
244 return 0;
245 }
246
247 TRACE_AG_PICK2(mp, startag, *agp, xfs_filestream_peek_ag(mp, *agp),
248 free, nscan, flags);
249
250 return 0;
251}
252
253/*
254 * Set the allocation group number for a file or a directory, updating inode
255 * references and per-AG references as appropriate. Must be called with the
256 * m_peraglock held in read mode.
257 */
258static int
259_xfs_filestream_update_ag(
260 xfs_inode_t *ip,
261 xfs_inode_t *pip,
262 xfs_agnumber_t ag)
263{
264 int err = 0;
265 xfs_mount_t *mp;
266 xfs_mru_cache_t *cache;
267 fstrm_item_t *item;
268 xfs_agnumber_t old_ag;
269 xfs_inode_t *old_pip;
270
271 /*
272 * Either ip is a regular file and pip is a directory, or ip is a
273 * directory and pip is NULL.
274 */
275 ASSERT(ip && (((ip->i_d.di_mode & S_IFREG) && pip &&
276 (pip->i_d.di_mode & S_IFDIR)) ||
277 ((ip->i_d.di_mode & S_IFDIR) && !pip)));
278
279 mp = ip->i_mount;
280 cache = mp->m_filestream;
281
282 item = xfs_mru_cache_lookup(cache, ip->i_ino);
283 if (item) {
284 ASSERT(item->ip == ip);
285 old_ag = item->ag;
286 item->ag = ag;
287 old_pip = item->pip;
288 item->pip = pip;
289 xfs_mru_cache_done(cache);
290
291 /*
292 * If the AG has changed, drop the old ref and take a new one,
293 * effectively transferring the reference from old to new AG.
294 */
295 if (ag != old_ag) {
296 xfs_filestream_put_ag(mp, old_ag);
297 xfs_filestream_get_ag(mp, ag);
298 }
299
300 /*
301 * If ip is a file and its pip has changed, drop the old ref and
302 * take a new one.
303 */
304 if (pip && pip != old_pip) {
305 IRELE(old_pip);
306 IHOLD(pip);
307 }
308
309 TRACE_UPDATE(mp, ip, old_ag, xfs_filestream_peek_ag(mp, old_ag),
310 ag, xfs_filestream_peek_ag(mp, ag));
311 return 0;
312 }
313
314 item = kmem_zone_zalloc(item_zone, KM_MAYFAIL);
315 if (!item)
316 return ENOMEM;
317
318 item->ag = ag;
319 item->ip = ip;
320 item->pip = pip;
321
322 err = xfs_mru_cache_insert(cache, ip->i_ino, item);
323 if (err) {
324 kmem_zone_free(item_zone, item);
325 return err;
326 }
327
328 /* Take a reference on the AG. */
329 xfs_filestream_get_ag(mp, ag);
330
331 /*
332 * Take a reference on the inode itself regardless of whether it's a
333 * regular file or a directory.
334 */
335 IHOLD(ip);
336
337 /*
338 * In the case of a regular file, take a reference on the parent inode
339 * as well to ensure it remains in-core.
340 */
341 if (pip)
342 IHOLD(pip);
343
344 TRACE_UPDATE(mp, ip, ag, xfs_filestream_peek_ag(mp, ag),
345 ag, xfs_filestream_peek_ag(mp, ag));
346
347 return 0;
348}
349
350/* xfs_fstrm_free_func(): callback for freeing cached stream items. */
351void
352xfs_fstrm_free_func(
353 xfs_ino_t ino,
354 fstrm_item_t *item)
355{
356 xfs_inode_t *ip = item->ip;
357 int ref;
358
359 ASSERT(ip->i_ino == ino);
360
361 xfs_iflags_clear(ip, XFS_IFILESTREAM);
362
363 /* Drop the reference taken on the AG when the item was added. */
364 ref = xfs_filestream_put_ag(ip->i_mount, item->ag);
365
366 ASSERT(ref >= 0);
367 TRACE_FREE(ip->i_mount, ip, item->pip, item->ag,
368 xfs_filestream_peek_ag(ip->i_mount, item->ag));
369
370 /*
371 * _xfs_filestream_update_ag() always takes a reference on the inode
372 * itself, whether it's a file or a directory. Release it here.
373 * This can result in the inode being freed and so we must
374 * not hold any inode locks when freeing filesstreams objects
375 * otherwise we can deadlock here.
376 */
377 IRELE(ip);
378
379 /*
380 * In the case of a regular file, _xfs_filestream_update_ag() also
381 * takes a ref on the parent inode to keep it in-core. Release that
382 * too.
383 */
384 if (item->pip)
385 IRELE(item->pip);
386
387 /* Finally, free the memory allocated for the item. */
388 kmem_zone_free(item_zone, item);
389}
390
391/*
392 * xfs_filestream_init() is called at xfs initialisation time to set up the
393 * memory zone that will be used for filestream data structure allocation.
394 */
395int
396xfs_filestream_init(void)
397{
398 item_zone = kmem_zone_init(sizeof(fstrm_item_t), "fstrm_item");
399#ifdef XFS_FILESTREAMS_TRACE
400 xfs_filestreams_trace_buf = ktrace_alloc(XFS_FSTRM_KTRACE_SIZE, KM_SLEEP);
401#endif
402 return item_zone ? 0 : -ENOMEM;
403}
404
405/*
406 * xfs_filestream_uninit() is called at xfs termination time to destroy the
407 * memory zone that was used for filestream data structure allocation.
408 */
409void
410xfs_filestream_uninit(void)
411{
412#ifdef XFS_FILESTREAMS_TRACE
413 ktrace_free(xfs_filestreams_trace_buf);
414#endif
415 kmem_zone_destroy(item_zone);
416}
417
418/*
419 * xfs_filestream_mount() is called when a file system is mounted with the
420 * filestream option. It is responsible for allocating the data structures
421 * needed to track the new file system's file streams.
422 */
423int
424xfs_filestream_mount(
425 xfs_mount_t *mp)
426{
427 int err;
428 unsigned int lifetime, grp_count;
429
430 /*
431 * The filestream timer tunable is currently fixed within the range of
432 * one second to four minutes, with five seconds being the default. The
433 * group count is somewhat arbitrary, but it'd be nice to adhere to the
434 * timer tunable to within about 10 percent. This requires at least 10
435 * groups.
436 */
437 lifetime = xfs_fstrm_centisecs * 10;
438 grp_count = 10;
439
440 err = xfs_mru_cache_create(&mp->m_filestream, lifetime, grp_count,
441 (xfs_mru_cache_free_func_t)xfs_fstrm_free_func);
442
443 return err;
444}
445
446/*
447 * xfs_filestream_unmount() is called when a file system that was mounted with
448 * the filestream option is unmounted. It drains the data structures created
449 * to track the file system's file streams and frees all the memory that was
450 * allocated.
451 */
452void
453xfs_filestream_unmount(
454 xfs_mount_t *mp)
455{
456 xfs_mru_cache_destroy(mp->m_filestream);
457}
458
459/*
460 * If the mount point's m_perag array is going to be reallocated, all
461 * outstanding cache entries must be flushed to avoid accessing reference count
462 * addresses that have been freed. The call to xfs_filestream_flush() must be
463 * made inside the block that holds the m_peraglock in write mode to do the
464 * reallocation.
465 */
466void
467xfs_filestream_flush(
468 xfs_mount_t *mp)
469{
470 /* point in time flush, so keep the reaper running */
471 xfs_mru_cache_flush(mp->m_filestream, 1);
472}
473
474/*
475 * Return the AG of the filestream the file or directory belongs to, or
476 * NULLAGNUMBER otherwise.
477 */
478xfs_agnumber_t
479xfs_filestream_lookup_ag(
480 xfs_inode_t *ip)
481{
482 xfs_mru_cache_t *cache;
483 fstrm_item_t *item;
484 xfs_agnumber_t ag;
485 int ref;
486
487 if (!(ip->i_d.di_mode & (S_IFREG | S_IFDIR))) {
488 ASSERT(0);
489 return NULLAGNUMBER;
490 }
491
492 cache = ip->i_mount->m_filestream;
493 item = xfs_mru_cache_lookup(cache, ip->i_ino);
494 if (!item) {
495 TRACE_LOOKUP(ip->i_mount, ip, NULL, NULLAGNUMBER, 0);
496 return NULLAGNUMBER;
497 }
498
499 ASSERT(ip == item->ip);
500 ag = item->ag;
501 ref = xfs_filestream_peek_ag(ip->i_mount, ag);
502 xfs_mru_cache_done(cache);
503
504 TRACE_LOOKUP(ip->i_mount, ip, item->pip, ag, ref);
505 return ag;
506}
507
508/*
509 * xfs_filestream_associate() should only be called to associate a regular file
510 * with its parent directory. Calling it with a child directory isn't
511 * appropriate because filestreams don't apply to entire directory hierarchies.
512 * Creating a file in a child directory of an existing filestream directory
513 * starts a new filestream with its own allocation group association.
514 *
515 * Returns < 0 on error, 0 if successful association occurred, > 0 if
516 * we failed to get an association because of locking issues.
517 */
518int
519xfs_filestream_associate(
520 xfs_inode_t *pip,
521 xfs_inode_t *ip)
522{
523 xfs_mount_t *mp;
524 xfs_mru_cache_t *cache;
525 fstrm_item_t *item;
526 xfs_agnumber_t ag, rotorstep, startag;
527 int err = 0;
528
529 ASSERT(pip->i_d.di_mode & S_IFDIR);
530 ASSERT(ip->i_d.di_mode & S_IFREG);
531 if (!(pip->i_d.di_mode & S_IFDIR) || !(ip->i_d.di_mode & S_IFREG))
532 return -EINVAL;
533
534 mp = pip->i_mount;
535 cache = mp->m_filestream;
536 down_read(&mp->m_peraglock);
537
538 /*
539 * We have a problem, Houston.
540 *
541 * Taking the iolock here violates inode locking order - we already
542 * hold the ilock. Hence if we block getting this lock we may never
543 * wake. Unfortunately, that means if we can't get the lock, we're
544 * screwed in terms of getting a stream association - we can't spin
545 * waiting for the lock because someone else is waiting on the lock we
546 * hold and we cannot drop that as we are in a transaction here.
547 *
548 * Lucky for us, this inversion is rarely a problem because it's a
549 * directory inode that we are trying to lock here and that means the
550 * only place that matters is xfs_sync_inodes() and SYNC_DELWRI is
551 * used. i.e. freeze, remount-ro, quotasync or unmount.
552 *
553 * So, if we can't get the iolock without sleeping then just give up
554 */
555 if (!xfs_ilock_nowait(pip, XFS_IOLOCK_EXCL)) {
556 up_read(&mp->m_peraglock);
557 return 1;
558 }
559
560 /* If the parent directory is already in the cache, use its AG. */
561 item = xfs_mru_cache_lookup(cache, pip->i_ino);
562 if (item) {
563 ASSERT(item->ip == pip);
564 ag = item->ag;
565 xfs_mru_cache_done(cache);
566
567 TRACE_LOOKUP(mp, pip, pip, ag, xfs_filestream_peek_ag(mp, ag));
568 err = _xfs_filestream_update_ag(ip, pip, ag);
569
570 goto exit;
571 }
572
573 /*
574 * Set the starting AG using the rotor for inode32, otherwise
575 * use the directory inode's AG.
576 */
577 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
578 rotorstep = xfs_rotorstep;
579 startag = (mp->m_agfrotor / rotorstep) % mp->m_sb.sb_agcount;
580 mp->m_agfrotor = (mp->m_agfrotor + 1) %
581 (mp->m_sb.sb_agcount * rotorstep);
582 } else
583 startag = XFS_INO_TO_AGNO(mp, pip->i_ino);
584
585 /* Pick a new AG for the parent inode starting at startag. */
586 err = _xfs_filestream_pick_ag(mp, startag, &ag, 0, 0);
587 if (err || ag == NULLAGNUMBER)
588 goto exit_did_pick;
589
590 /* Associate the parent inode with the AG. */
591 err = _xfs_filestream_update_ag(pip, NULL, ag);
592 if (err)
593 goto exit_did_pick;
594
595 /* Associate the file inode with the AG. */
596 err = _xfs_filestream_update_ag(ip, pip, ag);
597 if (err)
598 goto exit_did_pick;
599
600 TRACE_ASSOCIATE(mp, ip, pip, ag, xfs_filestream_peek_ag(mp, ag));
601
602exit_did_pick:
603 /*
604 * If _xfs_filestream_pick_ag() returned a valid AG, remove the
605 * reference it took on it, since the file and directory will have taken
606 * their own now if they were successfully cached.
607 */
608 if (ag != NULLAGNUMBER)
609 xfs_filestream_put_ag(mp, ag);
610
611exit:
612 xfs_iunlock(pip, XFS_IOLOCK_EXCL);
613 up_read(&mp->m_peraglock);
614 return -err;
615}
616
617/*
618 * Pick a new allocation group for the current file and its file stream. This
619 * function is called by xfs_bmap_filestreams() with the mount point's per-ag
620 * lock held.
621 */
622int
623xfs_filestream_new_ag(
624 xfs_bmalloca_t *ap,
625 xfs_agnumber_t *agp)
626{
627 int flags, err;
628 xfs_inode_t *ip, *pip = NULL;
629 xfs_mount_t *mp;
630 xfs_mru_cache_t *cache;
631 xfs_extlen_t minlen;
632 fstrm_item_t *dir, *file;
633 xfs_agnumber_t ag = NULLAGNUMBER;
634
635 ip = ap->ip;
636 mp = ip->i_mount;
637 cache = mp->m_filestream;
638 minlen = ap->alen;
639 *agp = NULLAGNUMBER;
640
641 /*
642 * Look for the file in the cache, removing it if it's found. Doing
643 * this allows it to be held across the dir lookup that follows.
644 */
645 file = xfs_mru_cache_remove(cache, ip->i_ino);
646 if (file) {
647 ASSERT(ip == file->ip);
648
649 /* Save the file's parent inode and old AG number for later. */
650 pip = file->pip;
651 ag = file->ag;
652
653 /* Look for the file's directory in the cache. */
654 dir = xfs_mru_cache_lookup(cache, pip->i_ino);
655 if (dir) {
656 ASSERT(pip == dir->ip);
657
658 /*
659 * If the directory has already moved on to a new AG,
660 * use that AG as the new AG for the file. Don't
661 * forget to twiddle the AG refcounts to match the
662 * movement.
663 */
664 if (dir->ag != file->ag) {
665 xfs_filestream_put_ag(mp, file->ag);
666 xfs_filestream_get_ag(mp, dir->ag);
667 *agp = file->ag = dir->ag;
668 }
669
670 xfs_mru_cache_done(cache);
671 }
672
673 /*
674 * Put the file back in the cache. If this fails, the free
675 * function needs to be called to tidy up in the same way as if
676 * the item had simply expired from the cache.
677 */
678 err = xfs_mru_cache_insert(cache, ip->i_ino, file);
679 if (err) {
680 xfs_fstrm_free_func(ip->i_ino, file);
681 return err;
682 }
683
684 /*
685 * If the file's AG was moved to the directory's new AG, there's
686 * nothing more to be done.
687 */
688 if (*agp != NULLAGNUMBER) {
689 TRACE_MOVEAG(mp, ip, pip,
690 ag, xfs_filestream_peek_ag(mp, ag),
691 *agp, xfs_filestream_peek_ag(mp, *agp));
692 return 0;
693 }
694 }
695
696 /*
697 * If the file's parent directory is known, take its iolock in exclusive
698 * mode to prevent two sibling files from racing each other to migrate
699 * themselves and their parent to different AGs.
700 */
701 if (pip)
702 xfs_ilock(pip, XFS_IOLOCK_EXCL);
703
704 /*
705 * A new AG needs to be found for the file. If the file's parent
706 * directory is also known, it will be moved to the new AG as well to
707 * ensure that files created inside it in future use the new AG.
708 */
709 ag = (ag == NULLAGNUMBER) ? 0 : (ag + 1) % mp->m_sb.sb_agcount;
710 flags = (ap->userdata ? XFS_PICK_USERDATA : 0) |
711 (ap->low ? XFS_PICK_LOWSPACE : 0);
712
713 err = _xfs_filestream_pick_ag(mp, ag, agp, flags, minlen);
714 if (err || *agp == NULLAGNUMBER)
715 goto exit;
716
717 /*
718 * If the file wasn't found in the file cache, then its parent directory
719 * inode isn't known. For this to have happened, the file must either
720 * be pre-existing, or it was created long enough ago that its cache
721 * entry has expired. This isn't the sort of usage that the filestreams
722 * allocator is trying to optimise, so there's no point trying to track
723 * its new AG somehow in the filestream data structures.
724 */
725 if (!pip) {
726 TRACE_ORPHAN(mp, ip, *agp);
727 goto exit;
728 }
729
730 /* Associate the parent inode with the AG. */
731 err = _xfs_filestream_update_ag(pip, NULL, *agp);
732 if (err)
733 goto exit;
734
735 /* Associate the file inode with the AG. */
736 err = _xfs_filestream_update_ag(ip, pip, *agp);
737 if (err)
738 goto exit;
739
740 TRACE_MOVEAG(mp, ip, pip, NULLAGNUMBER, 0,
741 *agp, xfs_filestream_peek_ag(mp, *agp));
742
743exit:
744 /*
745 * If _xfs_filestream_pick_ag() returned a valid AG, remove the
746 * reference it took on it, since the file and directory will have taken
747 * their own now if they were successfully cached.
748 */
749 if (*agp != NULLAGNUMBER)
750 xfs_filestream_put_ag(mp, *agp);
751 else
752 *agp = 0;
753
754 if (pip)
755 xfs_iunlock(pip, XFS_IOLOCK_EXCL);
756
757 return err;
758}
759
760/*
761 * Remove an association between an inode and a filestream object.
762 * Typically this is done on last close of an unlinked file.
763 */
764void
765xfs_filestream_deassociate(
766 xfs_inode_t *ip)
767{
768 xfs_mru_cache_t *cache = ip->i_mount->m_filestream;
769
770 xfs_mru_cache_delete(cache, ip->i_ino);
771}
diff --git a/fs/xfs/xfs_filestream.h b/fs/xfs/xfs_filestream.h
new file mode 100644
index 000000000000..f655f7dc334c
--- /dev/null
+++ b/fs/xfs/xfs_filestream.h
@@ -0,0 +1,136 @@
1/*
2 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#ifndef __XFS_FILESTREAM_H__
19#define __XFS_FILESTREAM_H__
20
21#ifdef __KERNEL__
22
23struct xfs_mount;
24struct xfs_inode;
25struct xfs_perag;
26struct xfs_bmalloca;
27
28#ifdef XFS_FILESTREAMS_TRACE
29#define XFS_FSTRM_KTRACE_INFO 1
30#define XFS_FSTRM_KTRACE_AGSCAN 2
31#define XFS_FSTRM_KTRACE_AGPICK1 3
32#define XFS_FSTRM_KTRACE_AGPICK2 4
33#define XFS_FSTRM_KTRACE_UPDATE 5
34#define XFS_FSTRM_KTRACE_FREE 6
35#define XFS_FSTRM_KTRACE_ITEM_LOOKUP 7
36#define XFS_FSTRM_KTRACE_ASSOCIATE 8
37#define XFS_FSTRM_KTRACE_MOVEAG 9
38#define XFS_FSTRM_KTRACE_ORPHAN 10
39
40#define XFS_FSTRM_KTRACE_SIZE 16384
41extern ktrace_t *xfs_filestreams_trace_buf;
42
43#endif
44
45/*
46 * Allocation group filestream associations are tracked with per-ag atomic
47 * counters. These counters allow _xfs_filestream_pick_ag() to tell whether a
48 * particular AG already has active filestreams associated with it. The mount
49 * point's m_peraglock is used to protect these counters from per-ag array
50 * re-allocation during a growfs operation. When xfs_growfs_data_private() is
51 * about to reallocate the array, it calls xfs_filestream_flush() with the
52 * m_peraglock held in write mode.
53 *
54 * Since xfs_mru_cache_flush() guarantees that all the free functions for all
55 * the cache elements have finished executing before it returns, it's safe for
56 * the free functions to use the atomic counters without m_peraglock protection.
57 * This allows the implementation of xfs_fstrm_free_func() to be agnostic about
58 * whether it was called with the m_peraglock held in read mode, write mode or
59 * not held at all. The race condition this addresses is the following:
60 *
61 * - The work queue scheduler fires and pulls a filestream directory cache
62 * element off the LRU end of the cache for deletion, then gets pre-empted.
63 * - A growfs operation grabs the m_peraglock in write mode, flushes all the
64 * remaining items from the cache and reallocates the mount point's per-ag
65 * array, resetting all the counters to zero.
66 * - The work queue thread resumes and calls the free function for the element
67 * it started cleaning up earlier. In the process it decrements the
68 * filestreams counter for an AG that now has no references.
69 *
70 * With a shrinkfs feature, the above scenario could panic the system.
71 *
72 * All other uses of the following macros should be protected by either the
73 * m_peraglock held in read mode, or the cache's internal locking exposed by the
74 * interval between a call to xfs_mru_cache_lookup() and a call to
75 * xfs_mru_cache_done(). In addition, the m_peraglock must be held in read mode
76 * when new elements are added to the cache.
77 *
78 * Combined, these locking rules ensure that no associations will ever exist in
79 * the cache that reference per-ag array elements that have since been
80 * reallocated.
81 */
82STATIC_INLINE int
83xfs_filestream_peek_ag(
84 xfs_mount_t *mp,
85 xfs_agnumber_t agno)
86{
87 return atomic_read(&mp->m_perag[agno].pagf_fstrms);
88}
89
90STATIC_INLINE int
91xfs_filestream_get_ag(
92 xfs_mount_t *mp,
93 xfs_agnumber_t agno)
94{
95 return atomic_inc_return(&mp->m_perag[agno].pagf_fstrms);
96}
97
98STATIC_INLINE int
99xfs_filestream_put_ag(
100 xfs_mount_t *mp,
101 xfs_agnumber_t agno)
102{
103 return atomic_dec_return(&mp->m_perag[agno].pagf_fstrms);
104}
105
106/* allocation selection flags */
107typedef enum xfs_fstrm_alloc {
108 XFS_PICK_USERDATA = 1,
109 XFS_PICK_LOWSPACE = 2,
110} xfs_fstrm_alloc_t;
111
112/* prototypes for filestream.c */
113int xfs_filestream_init(void);
114void xfs_filestream_uninit(void);
115int xfs_filestream_mount(struct xfs_mount *mp);
116void xfs_filestream_unmount(struct xfs_mount *mp);
117void xfs_filestream_flush(struct xfs_mount *mp);
118xfs_agnumber_t xfs_filestream_lookup_ag(struct xfs_inode *ip);
119int xfs_filestream_associate(struct xfs_inode *dip, struct xfs_inode *ip);
120void xfs_filestream_deassociate(struct xfs_inode *ip);
121int xfs_filestream_new_ag(struct xfs_bmalloca *ap, xfs_agnumber_t *agp);
122
123
124/* filestreams for the inode? */
125STATIC_INLINE int
126xfs_inode_is_filestream(
127 struct xfs_inode *ip)
128{
129 return (ip->i_mount->m_flags & XFS_MOUNT_FILESTREAMS) ||
130 xfs_iflags_test(ip, XFS_IFILESTREAM) ||
131 (ip->i_d.di_flags & XFS_DIFLAG_FILESTREAM);
132}
133
134#endif /* __KERNEL__ */
135
136#endif /* __XFS_FILESTREAM_H__ */
diff --git a/fs/xfs/xfs_fs.h b/fs/xfs/xfs_fs.h
index 1b60cfc28be5..ec3c9c27e0de 100644
--- a/fs/xfs/xfs_fs.h
+++ b/fs/xfs/xfs_fs.h
@@ -66,6 +66,7 @@ struct fsxattr {
66#define XFS_XFLAG_EXTSIZE 0x00000800 /* extent size allocator hint */ 66#define XFS_XFLAG_EXTSIZE 0x00000800 /* extent size allocator hint */
67#define XFS_XFLAG_EXTSZINHERIT 0x00001000 /* inherit inode extent size */ 67#define XFS_XFLAG_EXTSZINHERIT 0x00001000 /* inherit inode extent size */
68#define XFS_XFLAG_NODEFRAG 0x00002000 /* do not defragment */ 68#define XFS_XFLAG_NODEFRAG 0x00002000 /* do not defragment */
69#define XFS_XFLAG_FILESTREAM 0x00004000 /* use filestream allocator */
69#define XFS_XFLAG_HASATTR 0x80000000 /* no DIFLAG for this */ 70#define XFS_XFLAG_HASATTR 0x80000000 /* no DIFLAG for this */
70 71
71/* 72/*
diff --git a/fs/xfs/xfs_fsops.c b/fs/xfs/xfs_fsops.c
index 2251a49f3e17..432e82347ed6 100644
--- a/fs/xfs/xfs_fsops.c
+++ b/fs/xfs/xfs_fsops.c
@@ -44,6 +44,7 @@
44#include "xfs_trans_space.h" 44#include "xfs_trans_space.h"
45#include "xfs_rtalloc.h" 45#include "xfs_rtalloc.h"
46#include "xfs_rw.h" 46#include "xfs_rw.h"
47#include "xfs_filestream.h"
47 48
48/* 49/*
49 * File system operations 50 * File system operations
@@ -165,6 +166,7 @@ xfs_growfs_data_private(
165 new = nb - mp->m_sb.sb_dblocks; 166 new = nb - mp->m_sb.sb_dblocks;
166 oagcount = mp->m_sb.sb_agcount; 167 oagcount = mp->m_sb.sb_agcount;
167 if (nagcount > oagcount) { 168 if (nagcount > oagcount) {
169 xfs_filestream_flush(mp);
168 down_write(&mp->m_peraglock); 170 down_write(&mp->m_peraglock);
169 mp->m_perag = kmem_realloc(mp->m_perag, 171 mp->m_perag = kmem_realloc(mp->m_perag,
170 sizeof(xfs_perag_t) * nagcount, 172 sizeof(xfs_perag_t) * nagcount,
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index 8fdd30d9ba56..2ef100be6c4f 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -48,6 +48,7 @@
48#include "xfs_dir2_trace.h" 48#include "xfs_dir2_trace.h"
49#include "xfs_quota.h" 49#include "xfs_quota.h"
50#include "xfs_acl.h" 50#include "xfs_acl.h"
51#include "xfs_filestream.h"
51 52
52#include <linux/log2.h> 53#include <linux/log2.h>
53 54
@@ -818,6 +819,8 @@ _xfs_dic2xflags(
818 flags |= XFS_XFLAG_EXTSZINHERIT; 819 flags |= XFS_XFLAG_EXTSZINHERIT;
819 if (di_flags & XFS_DIFLAG_NODEFRAG) 820 if (di_flags & XFS_DIFLAG_NODEFRAG)
820 flags |= XFS_XFLAG_NODEFRAG; 821 flags |= XFS_XFLAG_NODEFRAG;
822 if (di_flags & XFS_DIFLAG_FILESTREAM)
823 flags |= XFS_XFLAG_FILESTREAM;
821 } 824 }
822 825
823 return flags; 826 return flags;
@@ -1151,7 +1154,7 @@ xfs_ialloc(
1151 /* 1154 /*
1152 * Project ids won't be stored on disk if we are using a version 1 inode. 1155 * Project ids won't be stored on disk if we are using a version 1 inode.
1153 */ 1156 */
1154 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1)) 1157 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1155 xfs_bump_ino_vers2(tp, ip); 1158 xfs_bump_ino_vers2(tp, ip);
1156 1159
1157 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) { 1160 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
@@ -1196,8 +1199,16 @@ xfs_ialloc(
1196 flags |= XFS_ILOG_DEV; 1199 flags |= XFS_ILOG_DEV;
1197 break; 1200 break;
1198 case S_IFREG: 1201 case S_IFREG:
1202 if (xfs_inode_is_filestream(pip)) {
1203 error = xfs_filestream_associate(pip, ip);
1204 if (error < 0)
1205 return -error;
1206 if (!error)
1207 xfs_iflags_set(ip, XFS_IFILESTREAM);
1208 }
1209 /* fall through */
1199 case S_IFDIR: 1210 case S_IFDIR:
1200 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 1211 if (pip->i_d.di_flags & XFS_DIFLAG_ANY) {
1201 uint di_flags = 0; 1212 uint di_flags = 0;
1202 1213
1203 if ((mode & S_IFMT) == S_IFDIR) { 1214 if ((mode & S_IFMT) == S_IFDIR) {
@@ -1234,6 +1245,8 @@ xfs_ialloc(
1234 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 1245 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1235 xfs_inherit_nodefrag) 1246 xfs_inherit_nodefrag)
1236 di_flags |= XFS_DIFLAG_NODEFRAG; 1247 di_flags |= XFS_DIFLAG_NODEFRAG;
1248 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1249 di_flags |= XFS_DIFLAG_FILESTREAM;
1237 ip->i_d.di_flags |= di_flags; 1250 ip->i_d.di_flags |= di_flags;
1238 } 1251 }
1239 /* FALLTHROUGH */ 1252 /* FALLTHROUGH */
diff --git a/fs/xfs/xfs_inode.h b/fs/xfs/xfs_inode.h
index f75afecef8e7..d418eeed4ebd 100644
--- a/fs/xfs/xfs_inode.h
+++ b/fs/xfs/xfs_inode.h
@@ -379,6 +379,7 @@ xfs_iflags_test(xfs_inode_t *ip, unsigned short flags)
379#define XFS_ISTALE 0x0010 /* inode has been staled */ 379#define XFS_ISTALE 0x0010 /* inode has been staled */
380#define XFS_IRECLAIMABLE 0x0020 /* inode can be reclaimed */ 380#define XFS_IRECLAIMABLE 0x0020 /* inode can be reclaimed */
381#define XFS_INEW 0x0040 381#define XFS_INEW 0x0040
382#define XFS_IFILESTREAM 0x0080 /* inode is in a filestream directory */
382 383
383/* 384/*
384 * Flags for inode locking. 385 * Flags for inode locking.
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index 0bca2d422719..76ad74758696 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -66,6 +66,7 @@ struct xfs_bmbt_irec;
66struct xfs_bmap_free; 66struct xfs_bmap_free;
67struct xfs_extdelta; 67struct xfs_extdelta;
68struct xfs_swapext; 68struct xfs_swapext;
69struct xfs_mru_cache;
69 70
70extern struct bhv_vfsops xfs_vfsops; 71extern struct bhv_vfsops xfs_vfsops;
71extern struct bhv_vnodeops xfs_vnodeops; 72extern struct bhv_vnodeops xfs_vnodeops;
@@ -424,6 +425,7 @@ typedef struct xfs_mount {
424 struct notifier_block m_icsb_notifier; /* hotplug cpu notifier */ 425 struct notifier_block m_icsb_notifier; /* hotplug cpu notifier */
425 struct mutex m_icsb_mutex; /* balancer sync lock */ 426 struct mutex m_icsb_mutex; /* balancer sync lock */
426#endif 427#endif
428 struct xfs_mru_cache *m_filestream; /* per-mount filestream data */
427} xfs_mount_t; 429} xfs_mount_t;
428 430
429/* 431/*
@@ -463,6 +465,8 @@ typedef struct xfs_mount {
463 * I/O size in stat() */ 465 * I/O size in stat() */
464#define XFS_MOUNT_NO_PERCPU_SB (1ULL << 23) /* don't use per-cpu superblock 466#define XFS_MOUNT_NO_PERCPU_SB (1ULL << 23) /* don't use per-cpu superblock
465 counters */ 467 counters */
468#define XFS_MOUNT_FILESTREAMS (1ULL << 24) /* enable the filestreams
469 allocator */
466 470
467 471
468/* 472/*
diff --git a/fs/xfs/xfs_mru_cache.c b/fs/xfs/xfs_mru_cache.c
new file mode 100644
index 000000000000..7deb9e3cbbd3
--- /dev/null
+++ b/fs/xfs/xfs_mru_cache.c
@@ -0,0 +1,608 @@
1/*
2 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_mru_cache.h"
20
21/*
22 * The MRU Cache data structure consists of a data store, an array of lists and
23 * a lock to protect its internal state. At initialisation time, the client
24 * supplies an element lifetime in milliseconds and a group count, as well as a
25 * function pointer to call when deleting elements. A data structure for
26 * queueing up work in the form of timed callbacks is also included.
27 *
28 * The group count controls how many lists are created, and thereby how finely
29 * the elements are grouped in time. When reaping occurs, all the elements in
30 * all the lists whose time has expired are deleted.
31 *
32 * To give an example of how this works in practice, consider a client that
33 * initialises an MRU Cache with a lifetime of ten seconds and a group count of
34 * five. Five internal lists will be created, each representing a two second
35 * period in time. When the first element is added, time zero for the data
36 * structure is initialised to the current time.
37 *
38 * All the elements added in the first two seconds are appended to the first
39 * list. Elements added in the third second go into the second list, and so on.
40 * If an element is accessed at any point, it is removed from its list and
41 * inserted at the head of the current most-recently-used list.
42 *
43 * The reaper function will have nothing to do until at least twelve seconds
44 * have elapsed since the first element was added. The reason for this is that
45 * if it were called at t=11s, there could be elements in the first list that
46 * have only been inactive for nine seconds, so it still does nothing. If it is
47 * called anywhere between t=12 and t=14 seconds, it will delete all the
48 * elements that remain in the first list. It's therefore possible for elements
49 * to remain in the data store even after they've been inactive for up to
50 * (t + t/g) seconds, where t is the inactive element lifetime and g is the
51 * number of groups.
52 *
53 * The above example assumes that the reaper function gets called at least once
54 * every (t/g) seconds. If it is called less frequently, unused elements will
55 * accumulate in the reap list until the reaper function is eventually called.
56 * The current implementation uses work queue callbacks to carefully time the
57 * reaper function calls, so this should happen rarely, if at all.
58 *
59 * From a design perspective, the primary reason for the choice of a list array
60 * representing discrete time intervals is that it's only practical to reap
61 * expired elements in groups of some appreciable size. This automatically
62 * introduces a granularity to element lifetimes, so there's no point storing an
63 * individual timeout with each element that specifies a more precise reap time.
64 * The bonus is a saving of sizeof(long) bytes of memory per element stored.
65 *
66 * The elements could have been stored in just one list, but an array of
67 * counters or pointers would need to be maintained to allow them to be divided
68 * up into discrete time groups. More critically, the process of touching or
69 * removing an element would involve walking large portions of the entire list,
70 * which would have a detrimental effect on performance. The additional memory
71 * requirement for the array of list heads is minimal.
72 *
73 * When an element is touched or deleted, it needs to be removed from its
74 * current list. Doubly linked lists are used to make the list maintenance
75 * portion of these operations O(1). Since reaper timing can be imprecise,
76 * inserts and lookups can occur when there are no free lists available. When
77 * this happens, all the elements on the LRU list need to be migrated to the end
78 * of the reap list. To keep the list maintenance portion of these operations
79 * O(1) also, list tails need to be accessible without walking the entire list.
80 * This is the reason why doubly linked list heads are used.
81 */
82
83/*
84 * An MRU Cache is a dynamic data structure that stores its elements in a way
85 * that allows efficient lookups, but also groups them into discrete time
86 * intervals based on insertion time. This allows elements to be efficiently
87 * and automatically reaped after a fixed period of inactivity.
88 *
89 * When a client data pointer is stored in the MRU Cache it needs to be added to
90 * both the data store and to one of the lists. It must also be possible to
91 * access each of these entries via the other, i.e. to:
92 *
93 * a) Walk a list, removing the corresponding data store entry for each item.
94 * b) Look up a data store entry, then access its list entry directly.
95 *
96 * To achieve both of these goals, each entry must contain both a list entry and
97 * a key, in addition to the user's data pointer. Note that it's not a good
98 * idea to have the client embed one of these structures at the top of their own
99 * data structure, because inserting the same item more than once would most
100 * likely result in a loop in one of the lists. That's a sure-fire recipe for
101 * an infinite loop in the code.
102 */
103typedef struct xfs_mru_cache_elem
104{
105 struct list_head list_node;
106 unsigned long key;
107 void *value;
108} xfs_mru_cache_elem_t;
109
110static kmem_zone_t *xfs_mru_elem_zone;
111static struct workqueue_struct *xfs_mru_reap_wq;
112
113/*
114 * When inserting, destroying or reaping, it's first necessary to update the
115 * lists relative to a particular time. In the case of destroying, that time
116 * will be well in the future to ensure that all items are moved to the reap
117 * list. In all other cases though, the time will be the current time.
118 *
119 * This function enters a loop, moving the contents of the LRU list to the reap
120 * list again and again until either a) the lists are all empty, or b) time zero
121 * has been advanced sufficiently to be within the immediate element lifetime.
122 *
123 * Case a) above is detected by counting how many groups are migrated and
124 * stopping when they've all been moved. Case b) is detected by monitoring the
125 * time_zero field, which is updated as each group is migrated.
126 *
127 * The return value is the earliest time that more migration could be needed, or
128 * zero if there's no need to schedule more work because the lists are empty.
129 */
130STATIC unsigned long
131_xfs_mru_cache_migrate(
132 xfs_mru_cache_t *mru,
133 unsigned long now)
134{
135 unsigned int grp;
136 unsigned int migrated = 0;
137 struct list_head *lru_list;
138
139 /* Nothing to do if the data store is empty. */
140 if (!mru->time_zero)
141 return 0;
142
143 /* While time zero is older than the time spanned by all the lists. */
144 while (mru->time_zero <= now - mru->grp_count * mru->grp_time) {
145
146 /*
147 * If the LRU list isn't empty, migrate its elements to the tail
148 * of the reap list.
149 */
150 lru_list = mru->lists + mru->lru_grp;
151 if (!list_empty(lru_list))
152 list_splice_init(lru_list, mru->reap_list.prev);
153
154 /*
155 * Advance the LRU group number, freeing the old LRU list to
156 * become the new MRU list; advance time zero accordingly.
157 */
158 mru->lru_grp = (mru->lru_grp + 1) % mru->grp_count;
159 mru->time_zero += mru->grp_time;
160
161 /*
162 * If reaping is so far behind that all the elements on all the
163 * lists have been migrated to the reap list, it's now empty.
164 */
165 if (++migrated == mru->grp_count) {
166 mru->lru_grp = 0;
167 mru->time_zero = 0;
168 return 0;
169 }
170 }
171
172 /* Find the first non-empty list from the LRU end. */
173 for (grp = 0; grp < mru->grp_count; grp++) {
174
175 /* Check the grp'th list from the LRU end. */
176 lru_list = mru->lists + ((mru->lru_grp + grp) % mru->grp_count);
177 if (!list_empty(lru_list))
178 return mru->time_zero +
179 (mru->grp_count + grp) * mru->grp_time;
180 }
181
182 /* All the lists must be empty. */
183 mru->lru_grp = 0;
184 mru->time_zero = 0;
185 return 0;
186}
187
188/*
189 * When inserting or doing a lookup, an element needs to be inserted into the
190 * MRU list. The lists must be migrated first to ensure that they're
191 * up-to-date, otherwise the new element could be given a shorter lifetime in
192 * the cache than it should.
193 */
194STATIC void
195_xfs_mru_cache_list_insert(
196 xfs_mru_cache_t *mru,
197 xfs_mru_cache_elem_t *elem)
198{
199 unsigned int grp = 0;
200 unsigned long now = jiffies;
201
202 /*
203 * If the data store is empty, initialise time zero, leave grp set to
204 * zero and start the work queue timer if necessary. Otherwise, set grp
205 * to the number of group times that have elapsed since time zero.
206 */
207 if (!_xfs_mru_cache_migrate(mru, now)) {
208 mru->time_zero = now;
209 if (!mru->next_reap)
210 mru->next_reap = mru->grp_count * mru->grp_time;
211 } else {
212 grp = (now - mru->time_zero) / mru->grp_time;
213 grp = (mru->lru_grp + grp) % mru->grp_count;
214 }
215
216 /* Insert the element at the tail of the corresponding list. */
217 list_add_tail(&elem->list_node, mru->lists + grp);
218}
219
220/*
221 * When destroying or reaping, all the elements that were migrated to the reap
222 * list need to be deleted. For each element this involves removing it from the
223 * data store, removing it from the reap list, calling the client's free
224 * function and deleting the element from the element zone.
225 */
226STATIC void
227_xfs_mru_cache_clear_reap_list(
228 xfs_mru_cache_t *mru)
229{
230 xfs_mru_cache_elem_t *elem, *next;
231 struct list_head tmp;
232
233 INIT_LIST_HEAD(&tmp);
234 list_for_each_entry_safe(elem, next, &mru->reap_list, list_node) {
235
236 /* Remove the element from the data store. */
237 radix_tree_delete(&mru->store, elem->key);
238
239 /*
240 * remove to temp list so it can be freed without
241 * needing to hold the lock
242 */
243 list_move(&elem->list_node, &tmp);
244 }
245 mutex_spinunlock(&mru->lock, 0);
246
247 list_for_each_entry_safe(elem, next, &tmp, list_node) {
248
249 /* Remove the element from the reap list. */
250 list_del_init(&elem->list_node);
251
252 /* Call the client's free function with the key and value pointer. */
253 mru->free_func(elem->key, elem->value);
254
255 /* Free the element structure. */
256 kmem_zone_free(xfs_mru_elem_zone, elem);
257 }
258
259 mutex_spinlock(&mru->lock);
260}
261
262/*
263 * We fire the reap timer every group expiry interval so
264 * we always have a reaper ready to run. This makes shutdown
265 * and flushing of the reaper easy to do. Hence we need to
266 * keep when the next reap must occur so we can determine
267 * at each interval whether there is anything we need to do.
268 */
269STATIC void
270_xfs_mru_cache_reap(
271 struct work_struct *work)
272{
273 xfs_mru_cache_t *mru = container_of(work, xfs_mru_cache_t, work.work);
274 unsigned long now;
275
276 ASSERT(mru && mru->lists);
277 if (!mru || !mru->lists)
278 return;
279
280 mutex_spinlock(&mru->lock);
281 now = jiffies;
282 if (mru->reap_all ||
283 (mru->next_reap && time_after(now, mru->next_reap))) {
284 if (mru->reap_all)
285 now += mru->grp_count * mru->grp_time * 2;
286 mru->next_reap = _xfs_mru_cache_migrate(mru, now);
287 _xfs_mru_cache_clear_reap_list(mru);
288 }
289
290 /*
291 * the process that triggered the reap_all is responsible
292 * for restating the periodic reap if it is required.
293 */
294 if (!mru->reap_all)
295 queue_delayed_work(xfs_mru_reap_wq, &mru->work, mru->grp_time);
296 mru->reap_all = 0;
297 mutex_spinunlock(&mru->lock, 0);
298}
299
300int
301xfs_mru_cache_init(void)
302{
303 xfs_mru_elem_zone = kmem_zone_init(sizeof(xfs_mru_cache_elem_t),
304 "xfs_mru_cache_elem");
305 if (!xfs_mru_elem_zone)
306 return ENOMEM;
307
308 xfs_mru_reap_wq = create_singlethread_workqueue("xfs_mru_cache");
309 if (!xfs_mru_reap_wq) {
310 kmem_zone_destroy(xfs_mru_elem_zone);
311 return ENOMEM;
312 }
313
314 return 0;
315}
316
317void
318xfs_mru_cache_uninit(void)
319{
320 destroy_workqueue(xfs_mru_reap_wq);
321 kmem_zone_destroy(xfs_mru_elem_zone);
322}
323
324/*
325 * To initialise a struct xfs_mru_cache pointer, call xfs_mru_cache_create()
326 * with the address of the pointer, a lifetime value in milliseconds, a group
327 * count and a free function to use when deleting elements. This function
328 * returns 0 if the initialisation was successful.
329 */
330int
331xfs_mru_cache_create(
332 xfs_mru_cache_t **mrup,
333 unsigned int lifetime_ms,
334 unsigned int grp_count,
335 xfs_mru_cache_free_func_t free_func)
336{
337 xfs_mru_cache_t *mru = NULL;
338 int err = 0, grp;
339 unsigned int grp_time;
340
341 if (mrup)
342 *mrup = NULL;
343
344 if (!mrup || !grp_count || !lifetime_ms || !free_func)
345 return EINVAL;
346
347 if (!(grp_time = msecs_to_jiffies(lifetime_ms) / grp_count))
348 return EINVAL;
349
350 if (!(mru = kmem_zalloc(sizeof(*mru), KM_SLEEP)))
351 return ENOMEM;
352
353 /* An extra list is needed to avoid reaping up to a grp_time early. */
354 mru->grp_count = grp_count + 1;
355 mru->lists = kmem_alloc(mru->grp_count * sizeof(*mru->lists), KM_SLEEP);
356
357 if (!mru->lists) {
358 err = ENOMEM;
359 goto exit;
360 }
361
362 for (grp = 0; grp < mru->grp_count; grp++)
363 INIT_LIST_HEAD(mru->lists + grp);
364
365 /*
366 * We use GFP_KERNEL radix tree preload and do inserts under a
367 * spinlock so GFP_ATOMIC is appropriate for the radix tree itself.
368 */
369 INIT_RADIX_TREE(&mru->store, GFP_ATOMIC);
370 INIT_LIST_HEAD(&mru->reap_list);
371 spinlock_init(&mru->lock, "xfs_mru_cache");
372 INIT_DELAYED_WORK(&mru->work, _xfs_mru_cache_reap);
373
374 mru->grp_time = grp_time;
375 mru->free_func = free_func;
376
377 /* start up the reaper event */
378 mru->next_reap = 0;
379 mru->reap_all = 0;
380 queue_delayed_work(xfs_mru_reap_wq, &mru->work, mru->grp_time);
381
382 *mrup = mru;
383
384exit:
385 if (err && mru && mru->lists)
386 kmem_free(mru->lists, mru->grp_count * sizeof(*mru->lists));
387 if (err && mru)
388 kmem_free(mru, sizeof(*mru));
389
390 return err;
391}
392
393/*
394 * Call xfs_mru_cache_flush() to flush out all cached entries, calling their
395 * free functions as they're deleted. When this function returns, the caller is
396 * guaranteed that all the free functions for all the elements have finished
397 * executing.
398 *
399 * While we are flushing, we stop the periodic reaper event from triggering.
400 * Normally, we want to restart this periodic event, but if we are shutting
401 * down the cache we do not want it restarted. hence the restart parameter
402 * where 0 = do not restart reaper and 1 = restart reaper.
403 */
404void
405xfs_mru_cache_flush(
406 xfs_mru_cache_t *mru,
407 int restart)
408{
409 if (!mru || !mru->lists)
410 return;
411
412 cancel_rearming_delayed_workqueue(xfs_mru_reap_wq, &mru->work);
413
414 mutex_spinlock(&mru->lock);
415 mru->reap_all = 1;
416 mutex_spinunlock(&mru->lock, 0);
417
418 queue_work(xfs_mru_reap_wq, &mru->work.work);
419 flush_workqueue(xfs_mru_reap_wq);
420
421 mutex_spinlock(&mru->lock);
422 WARN_ON_ONCE(mru->reap_all != 0);
423 mru->reap_all = 0;
424 if (restart)
425 queue_delayed_work(xfs_mru_reap_wq, &mru->work, mru->grp_time);
426 mutex_spinunlock(&mru->lock, 0);
427}
428
429void
430xfs_mru_cache_destroy(
431 xfs_mru_cache_t *mru)
432{
433 if (!mru || !mru->lists)
434 return;
435
436 /* we don't want the reaper to restart here */
437 xfs_mru_cache_flush(mru, 0);
438
439 kmem_free(mru->lists, mru->grp_count * sizeof(*mru->lists));
440 kmem_free(mru, sizeof(*mru));
441}
442
443/*
444 * To insert an element, call xfs_mru_cache_insert() with the data store, the
445 * element's key and the client data pointer. This function returns 0 on
446 * success or ENOMEM if memory for the data element couldn't be allocated.
447 */
448int
449xfs_mru_cache_insert(
450 xfs_mru_cache_t *mru,
451 unsigned long key,
452 void *value)
453{
454 xfs_mru_cache_elem_t *elem;
455
456 ASSERT(mru && mru->lists);
457 if (!mru || !mru->lists)
458 return EINVAL;
459
460 elem = kmem_zone_zalloc(xfs_mru_elem_zone, KM_SLEEP);
461 if (!elem)
462 return ENOMEM;
463
464 if (radix_tree_preload(GFP_KERNEL)) {
465 kmem_zone_free(xfs_mru_elem_zone, elem);
466 return ENOMEM;
467 }
468
469 INIT_LIST_HEAD(&elem->list_node);
470 elem->key = key;
471 elem->value = value;
472
473 mutex_spinlock(&mru->lock);
474
475 radix_tree_insert(&mru->store, key, elem);
476 radix_tree_preload_end();
477 _xfs_mru_cache_list_insert(mru, elem);
478
479 mutex_spinunlock(&mru->lock, 0);
480
481 return 0;
482}
483
484/*
485 * To remove an element without calling the free function, call
486 * xfs_mru_cache_remove() with the data store and the element's key. On success
487 * the client data pointer for the removed element is returned, otherwise this
488 * function will return a NULL pointer.
489 */
490void *
491xfs_mru_cache_remove(
492 xfs_mru_cache_t *mru,
493 unsigned long key)
494{
495 xfs_mru_cache_elem_t *elem;
496 void *value = NULL;
497
498 ASSERT(mru && mru->lists);
499 if (!mru || !mru->lists)
500 return NULL;
501
502 mutex_spinlock(&mru->lock);
503 elem = radix_tree_delete(&mru->store, key);
504 if (elem) {
505 value = elem->value;
506 list_del(&elem->list_node);
507 }
508
509 mutex_spinunlock(&mru->lock, 0);
510
511 if (elem)
512 kmem_zone_free(xfs_mru_elem_zone, elem);
513
514 return value;
515}
516
517/*
518 * To remove and element and call the free function, call xfs_mru_cache_delete()
519 * with the data store and the element's key.
520 */
521void
522xfs_mru_cache_delete(
523 xfs_mru_cache_t *mru,
524 unsigned long key)
525{
526 void *value = xfs_mru_cache_remove(mru, key);
527
528 if (value)
529 mru->free_func(key, value);
530}
531
532/*
533 * To look up an element using its key, call xfs_mru_cache_lookup() with the
534 * data store and the element's key. If found, the element will be moved to the
535 * head of the MRU list to indicate that it's been touched.
536 *
537 * The internal data structures are protected by a spinlock that is STILL HELD
538 * when this function returns. Call xfs_mru_cache_done() to release it. Note
539 * that it is not safe to call any function that might sleep in the interim.
540 *
541 * The implementation could have used reference counting to avoid this
542 * restriction, but since most clients simply want to get, set or test a member
543 * of the returned data structure, the extra per-element memory isn't warranted.
544 *
545 * If the element isn't found, this function returns NULL and the spinlock is
546 * released. xfs_mru_cache_done() should NOT be called when this occurs.
547 */
548void *
549xfs_mru_cache_lookup(
550 xfs_mru_cache_t *mru,
551 unsigned long key)
552{
553 xfs_mru_cache_elem_t *elem;
554
555 ASSERT(mru && mru->lists);
556 if (!mru || !mru->lists)
557 return NULL;
558
559 mutex_spinlock(&mru->lock);
560 elem = radix_tree_lookup(&mru->store, key);
561 if (elem) {
562 list_del(&elem->list_node);
563 _xfs_mru_cache_list_insert(mru, elem);
564 }
565 else
566 mutex_spinunlock(&mru->lock, 0);
567
568 return elem ? elem->value : NULL;
569}
570
571/*
572 * To look up an element using its key, but leave its location in the internal
573 * lists alone, call xfs_mru_cache_peek(). If the element isn't found, this
574 * function returns NULL.
575 *
576 * See the comments above the declaration of the xfs_mru_cache_lookup() function
577 * for important locking information pertaining to this call.
578 */
579void *
580xfs_mru_cache_peek(
581 xfs_mru_cache_t *mru,
582 unsigned long key)
583{
584 xfs_mru_cache_elem_t *elem;
585
586 ASSERT(mru && mru->lists);
587 if (!mru || !mru->lists)
588 return NULL;
589
590 mutex_spinlock(&mru->lock);
591 elem = radix_tree_lookup(&mru->store, key);
592 if (!elem)
593 mutex_spinunlock(&mru->lock, 0);
594
595 return elem ? elem->value : NULL;
596}
597
598/*
599 * To release the internal data structure spinlock after having performed an
600 * xfs_mru_cache_lookup() or an xfs_mru_cache_peek(), call xfs_mru_cache_done()
601 * with the data store pointer.
602 */
603void
604xfs_mru_cache_done(
605 xfs_mru_cache_t *mru)
606{
607 mutex_spinunlock(&mru->lock, 0);
608}
diff --git a/fs/xfs/xfs_mru_cache.h b/fs/xfs/xfs_mru_cache.h
new file mode 100644
index 000000000000..624fd10ee8e5
--- /dev/null
+++ b/fs/xfs/xfs_mru_cache.h
@@ -0,0 +1,57 @@
1/*
2 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#ifndef __XFS_MRU_CACHE_H__
19#define __XFS_MRU_CACHE_H__
20
21
22/* Function pointer type for callback to free a client's data pointer. */
23typedef void (*xfs_mru_cache_free_func_t)(unsigned long, void*);
24
25typedef struct xfs_mru_cache
26{
27 struct radix_tree_root store; /* Core storage data structure. */
28 struct list_head *lists; /* Array of lists, one per grp. */
29 struct list_head reap_list; /* Elements overdue for reaping. */
30 spinlock_t lock; /* Lock to protect this struct. */
31 unsigned int grp_count; /* Number of discrete groups. */
32 unsigned int grp_time; /* Time period spanned by grps. */
33 unsigned int lru_grp; /* Group containing time zero. */
34 unsigned long time_zero; /* Time first element was added. */
35 unsigned long next_reap; /* Time that the reaper should
36 next do something. */
37 unsigned int reap_all; /* if set, reap all lists */
38 xfs_mru_cache_free_func_t free_func; /* Function pointer for freeing. */
39 struct delayed_work work; /* Workqueue data for reaping. */
40} xfs_mru_cache_t;
41
42int xfs_mru_cache_init(void);
43void xfs_mru_cache_uninit(void);
44int xfs_mru_cache_create(struct xfs_mru_cache **mrup, unsigned int lifetime_ms,
45 unsigned int grp_count,
46 xfs_mru_cache_free_func_t free_func);
47void xfs_mru_cache_flush(xfs_mru_cache_t *mru, int restart);
48void xfs_mru_cache_destroy(struct xfs_mru_cache *mru);
49int xfs_mru_cache_insert(struct xfs_mru_cache *mru, unsigned long key,
50 void *value);
51void * xfs_mru_cache_remove(struct xfs_mru_cache *mru, unsigned long key);
52void xfs_mru_cache_delete(struct xfs_mru_cache *mru, unsigned long key);
53void *xfs_mru_cache_lookup(struct xfs_mru_cache *mru, unsigned long key);
54void *xfs_mru_cache_peek(struct xfs_mru_cache *mru, unsigned long key);
55void xfs_mru_cache_done(struct xfs_mru_cache *mru);
56
57#endif /* __XFS_MRU_CACHE_H__ */
diff --git a/fs/xfs/xfs_vfsops.c b/fs/xfs/xfs_vfsops.c
index c343fde10ef9..11f5ea29a038 100644
--- a/fs/xfs/xfs_vfsops.c
+++ b/fs/xfs/xfs_vfsops.c
@@ -51,6 +51,8 @@
51#include "xfs_acl.h" 51#include "xfs_acl.h"
52#include "xfs_attr.h" 52#include "xfs_attr.h"
53#include "xfs_clnt.h" 53#include "xfs_clnt.h"
54#include "xfs_mru_cache.h"
55#include "xfs_filestream.h"
54#include "xfs_fsops.h" 56#include "xfs_fsops.h"
55 57
56STATIC int xfs_sync(bhv_desc_t *, int, cred_t *); 58STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
@@ -81,6 +83,8 @@ xfs_init(void)
81 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf"); 83 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
82 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork"); 84 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
83 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl"); 85 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
86 xfs_mru_cache_init();
87 xfs_filestream_init();
84 88
85 /* 89 /*
86 * The size of the zone allocated buf log item is the maximum 90 * The size of the zone allocated buf log item is the maximum
@@ -164,6 +168,8 @@ xfs_cleanup(void)
164 xfs_cleanup_procfs(); 168 xfs_cleanup_procfs();
165 xfs_sysctl_unregister(); 169 xfs_sysctl_unregister();
166 xfs_refcache_destroy(); 170 xfs_refcache_destroy();
171 xfs_filestream_uninit();
172 xfs_mru_cache_uninit();
167 xfs_acl_zone_destroy(xfs_acl_zone); 173 xfs_acl_zone_destroy(xfs_acl_zone);
168 174
169#ifdef XFS_DIR2_TRACE 175#ifdef XFS_DIR2_TRACE
@@ -320,6 +326,9 @@ xfs_start_flags(
320 else 326 else
321 mp->m_flags &= ~XFS_MOUNT_BARRIER; 327 mp->m_flags &= ~XFS_MOUNT_BARRIER;
322 328
329 if (ap->flags2 & XFSMNT2_FILESTREAMS)
330 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
331
323 return 0; 332 return 0;
324} 333}
325 334
@@ -518,6 +527,9 @@ xfs_mount(
518 if (mp->m_flags & XFS_MOUNT_BARRIER) 527 if (mp->m_flags & XFS_MOUNT_BARRIER)
519 xfs_mountfs_check_barriers(mp); 528 xfs_mountfs_check_barriers(mp);
520 529
530 if ((error = xfs_filestream_mount(mp)))
531 goto error2;
532
521 error = XFS_IOINIT(vfsp, args, flags); 533 error = XFS_IOINIT(vfsp, args, flags);
522 if (error) 534 if (error)
523 goto error2; 535 goto error2;
@@ -575,6 +587,13 @@ xfs_unmount(
575 */ 587 */
576 xfs_refcache_purge_mp(mp); 588 xfs_refcache_purge_mp(mp);
577 589
590 /*
591 * Blow away any referenced inode in the filestreams cache.
592 * This can and will cause log traffic as inodes go inactive
593 * here.
594 */
595 xfs_filestream_unmount(mp);
596
578 XFS_bflush(mp->m_ddev_targp); 597 XFS_bflush(mp->m_ddev_targp);
579 error = xfs_unmount_flush(mp, 0); 598 error = xfs_unmount_flush(mp, 0);
580 if (error) 599 if (error)
@@ -694,6 +713,7 @@ xfs_mntupdate(
694 mp->m_flags &= ~XFS_MOUNT_BARRIER; 713 mp->m_flags &= ~XFS_MOUNT_BARRIER;
695 } 714 }
696 } else if (!(vfsp->vfs_flag & VFS_RDONLY)) { /* rw -> ro */ 715 } else if (!(vfsp->vfs_flag & VFS_RDONLY)) { /* rw -> ro */
716 xfs_filestream_flush(mp);
697 bhv_vfs_sync(vfsp, SYNC_DATA_QUIESCE, NULL); 717 bhv_vfs_sync(vfsp, SYNC_DATA_QUIESCE, NULL);
698 xfs_attr_quiesce(mp); 718 xfs_attr_quiesce(mp);
699 vfsp->vfs_flag |= VFS_RDONLY; 719 vfsp->vfs_flag |= VFS_RDONLY;
@@ -909,6 +929,9 @@ xfs_sync(
909{ 929{
910 xfs_mount_t *mp = XFS_BHVTOM(bdp); 930 xfs_mount_t *mp = XFS_BHVTOM(bdp);
911 931
932 if (flags & SYNC_IOWAIT)
933 xfs_filestream_flush(mp);
934
912 return xfs_syncsub(mp, flags, NULL); 935 return xfs_syncsub(mp, flags, NULL);
913} 936}
914 937
@@ -1659,6 +1682,7 @@ xfs_vget(
1659 * in stat(). */ 1682 * in stat(). */
1660#define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */ 1683#define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
1661#define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */ 1684#define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
1685#define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
1662 1686
1663STATIC unsigned long 1687STATIC unsigned long
1664suffix_strtoul(char *s, char **endp, unsigned int base) 1688suffix_strtoul(char *s, char **endp, unsigned int base)
@@ -1845,6 +1869,8 @@ xfs_parseargs(
1845 args->flags |= XFSMNT_ATTR2; 1869 args->flags |= XFSMNT_ATTR2;
1846 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) { 1870 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1847 args->flags &= ~XFSMNT_ATTR2; 1871 args->flags &= ~XFSMNT_ATTR2;
1872 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
1873 args->flags2 |= XFSMNT2_FILESTREAMS;
1848 } else if (!strcmp(this_char, "osyncisdsync")) { 1874 } else if (!strcmp(this_char, "osyncisdsync")) {
1849 /* no-op, this is now the default */ 1875 /* no-op, this is now the default */
1850 cmn_err(CE_WARN, 1876 cmn_err(CE_WARN,
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index 2067d0b0a10e..60fd0be90a16 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -51,6 +51,7 @@
51#include "xfs_refcache.h" 51#include "xfs_refcache.h"
52#include "xfs_trans_space.h" 52#include "xfs_trans_space.h"
53#include "xfs_log_priv.h" 53#include "xfs_log_priv.h"
54#include "xfs_filestream.h"
54 55
55STATIC int 56STATIC int
56xfs_open( 57xfs_open(
@@ -783,6 +784,8 @@ xfs_setattr(
783 di_flags |= XFS_DIFLAG_PROJINHERIT; 784 di_flags |= XFS_DIFLAG_PROJINHERIT;
784 if (vap->va_xflags & XFS_XFLAG_NODEFRAG) 785 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
785 di_flags |= XFS_DIFLAG_NODEFRAG; 786 di_flags |= XFS_DIFLAG_NODEFRAG;
787 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
788 di_flags |= XFS_DIFLAG_FILESTREAM;
786 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { 789 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
787 if (vap->va_xflags & XFS_XFLAG_RTINHERIT) 790 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
788 di_flags |= XFS_DIFLAG_RTINHERIT; 791 di_flags |= XFS_DIFLAG_RTINHERIT;
@@ -1536,7 +1539,17 @@ xfs_release(
1536 if (vp->v_vfsp->vfs_flag & VFS_RDONLY) 1539 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1537 return 0; 1540 return 0;
1538 1541
1539 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { 1542 if (!XFS_FORCED_SHUTDOWN(mp)) {
1543 /*
1544 * If we are using filestreams, and we have an unlinked
1545 * file that we are processing the last close on, then nothing
1546 * will be able to reopen and write to this file. Purge this
1547 * inode from the filestreams cache so that it doesn't delay
1548 * teardown of the inode.
1549 */
1550 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1551 xfs_filestream_deassociate(ip);
1552
1540 /* 1553 /*
1541 * If we previously truncated this file and removed old data 1554 * If we previously truncated this file and removed old data
1542 * in the process, we want to initiate "early" writeout on 1555 * in the process, we want to initiate "early" writeout on
@@ -1551,7 +1564,6 @@ xfs_release(
1551 bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE); 1564 bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
1552 } 1565 }
1553 1566
1554
1555#ifdef HAVE_REFCACHE 1567#ifdef HAVE_REFCACHE
1556 /* If we are in the NFS reference cache then don't do this now */ 1568 /* If we are in the NFS reference cache then don't do this now */
1557 if (ip->i_refcache) 1569 if (ip->i_refcache)
@@ -2541,6 +2553,15 @@ xfs_remove(
2541 */ 2553 */
2542 xfs_refcache_purge_ip(ip); 2554 xfs_refcache_purge_ip(ip);
2543 2555
2556 /*
2557 * If we are using filestreams, kill the stream association.
2558 * If the file is still open it may get a new one but that
2559 * will get killed on last close in xfs_close() so we don't
2560 * have to worry about that.
2561 */
2562 if (link_zero && xfs_inode_is_filestream(ip))
2563 xfs_filestream_deassociate(ip);
2564
2544 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); 2565 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2545 2566
2546 /* 2567 /*
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-07 14:28:02 -0400