d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
a664da774a
freebsd-skq/sys/kern/vfs_cache.c

872 lines
22 KiB
C
Raw Normal View History

/* -> /*- for copyright notices, minor format tweaks as necessary
2005-01-06 23:35:40 +00:00
/*-
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
* Copyright (c) 1989, 1993, 1995
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
* The Regents of the University of California. All rights reserved.
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
*
* This code is derived from software contributed to Berkeley by
* Poul-Henning Kamp of the FreeBSD Project.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
* @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
#include <sys/kernel.h>
Undo part of the tangle of having sys/lock.h and sys/mutex.h included in other "system" header files. Also help the deprecation of lockmgr.h by making it a sub-include of sys/lock.h and removing sys/lockmgr.h form kernel .c files. Sort sys/*.h includes where possible in affected files. OK'ed by: bde (with reservations)
2001-05-01 08:13:21 +00:00
#include <sys/lock.h>
Include sys/_lock.h and sys/_mutex.h to reduce namespace pollution. Requested by: jhb
2002-01-13 21:37:49 +00:00
#include <sys/mutex.h>
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
#include <sys/sysctl.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/mount.h>
#include <sys/vnode.h>
#include <sys/namei.h>
#include <sys/malloc.h>
Split up __getcwd so that kernel callers of the internal version can specify whether the buffer is in user or system space.
2002-09-02 22:40:30 +00:00
#include <sys/syscallsubr.h>
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
#include <sys/sysproto.h>
#include <sys/proc.h>
#include <sys/filedesc.h>
Use a generic implementation of the Fowler/Noll/Vo hash (FNV hash). Make the name cache hash as well as the nfsnode hash use it. As a special tweak, create an unsigned version of register_t. This allows us to use a special tweak for the 64 bit versions that significantly speeds up the i386 version (ie: int64 XOR int64 is slower than int64 XOR int32). The code layout is a little strange for the string function, but I was able to get between 5 to 10% improvement over the original version I started with. The layout affects gcc code generation choices and this way was fastest on x86 and alpha. Note that 'CPUTYPE=p3' etc makes a fair difference to this. It is around 45% faster with -march=pentiumpro on a p6 cpu.
2001-03-17 09:31:06 +00:00
#include <sys/fnv_hash.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Make the VFS cache use zones instead of malloc(9). This results in a small but noticeable increase in performance for name lookup operations. The code uses two zones, one for short names (less than 32 characters) and one for long names (up to NAME_MAX). Since most file names are fairly short, this saves a considerable amount of space that would otherwise be wasted if we always allocated NAME_MAX bytes. The cutoff value of 32 characters was picked arbitrarily and may benefit from some tweaking; it could also be made into a tunable. Submitted by: hmp
2003-06-13 08:46:13 +00:00
#include <vm/uma.h>
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
/*
* This structure describes the elements in the cache of recent
* names looked up by namei.
*/
struct namecache {
Back out the previous change to the queue(3) interface. It was not discussed and should probably not happen. Requested by: msmith and others
2000-05-26 02:09:24 +00:00
LIST_ENTRY(namecache) nc_hash; /* hash chain */
LIST_ENTRY(namecache) nc_src; /* source vnode list */
TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
struct vnode *nc_dvp; /* vnode of parent of name */
struct vnode *nc_vp; /* vnode the name refers to */
u_char nc_flag; /* flag bits */
u_char nc_nlen; /* length of name */
char nc_name[0]; /* segment name */
};
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
/*
* Name caching works as follows:
*
* Names found by directory scans are retained in a cache
* for future reference. It is managed LRU, so frequently
* used names will hang around. Cache is indexed by hash value
* obtained from (vp, name) where vp refers to the directory
* containing name.
*
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
* If it is a "negative" entry, (i.e. for a name that is known NOT to
* exist) the vnode pointer will be NULL.
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
*
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
* Upon reaching the last segment of a path, if the reference
* is for DELETE, or NOCACHE is set (rewrite), and the
* name is located in the cache, it will be dropped.
*/
/*
* Structures associated with name cacheing.
*/
Use the same API as the example code. Allow the initial hash value to be passed in, as the examples do. Incrementally hash in the dvp->v_id (using the official api) rather than add it. This seems to help power-of-two predictable filename trees where the filenames repeat on a power-of-two cycle and the directory trees have power-of-two components in it. The simple add then mask was causing things like 12000+ entry collision chains while most other entries have between 0 and 3 entries each. This way seems to improve things.
2001-03-20 02:10:18 +00:00
#define NCHHASH(hash) \
(&nchashtbl[(hash) & nchash])
Back out the previous change to the queue(3) interface. It was not discussed and should probably not happen. Requested by: msmith and others
2000-05-26 02:09:24 +00:00
static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
static u_long nchash; /* size of hash table */
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
static u_long ncnegfactor = 16; /* ratio of negative entries */
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
static u_long numneg; /* number of cache entries allocated */
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
static u_long numcache; /* number of cache entries allocated */
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
Reclaim directory vnodes held in namecache if few free vnodes are available. Only directory vnodes holding no child directory vnodes held in v_cache_src are recycled, so that directory vnodes near the root of the filesystem hierarchy remain in namecache and directory vnodes are not reclaimed in cascade. The period of vnode reclaiming attempt and the number of vnodes attempted to reclaim can be tuned via sysctl(2). Suggested by: tegge Approved by: phk
2001-04-18 11:19:50 +00:00
static u_long numcachehv; /* number of cache entries with vnodes held */
SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "");
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
#if 0
Reclaim directory vnodes held in namecache if few free vnodes are available. Only directory vnodes holding no child directory vnodes held in v_cache_src are recycled, so that directory vnodes near the root of the filesystem hierarchy remain in namecache and directory vnodes are not reclaimed in cascade. The period of vnode reclaiming attempt and the number of vnodes attempted to reclaim can be tuned via sysctl(2). Suggested by: tegge Approved by: phk
2001-04-18 11:19:50 +00:00
static u_long numcachepl; /* number of cache purge for leaf entries */
SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, "");
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
#endif
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
struct nchstats nchstats; /* cache effectiveness statistics */
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Make a SYSCTL_NODE and a mutex static
2005-02-10 12:16:42 +00:00
static struct mtx cache_lock;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
MTX_SYSINIT(vfscache, &cache_lock, "Name Cache", MTX_DEF);
#define CACHE_LOCK() mtx_lock(&cache_lock)
#define CACHE_UNLOCK() mtx_unlock(&cache_lock)
Make the VFS cache use zones instead of malloc(9). This results in a small but noticeable increase in performance for name lookup operations. The code uses two zones, one for short names (less than 32 characters) and one for long names (up to NAME_MAX). Since most file names are fairly short, this saves a considerable amount of space that would otherwise be wasted if we always allocated NAME_MAX bytes. The cutoff value of 32 characters was picked arbitrarily and may benefit from some tweaking; it could also be made into a tunable. Submitted by: hmp
2003-06-13 08:46:13 +00:00
/*
* UMA zones for the VFS cache.
*
* The small cache is used for entries with short names, which are the
* most common. The large cache is used for entries which are too big to
* fit in the small cache.
*/
static uma_zone_t cache_zone_small;
static uma_zone_t cache_zone_large;
#define CACHE_PATH_CUTOFF 32
#define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF)
#define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX)
#define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
cache_zone_small : cache_zone_large, M_WAITOK)
#define cache_free(ncp) do { \
if (ncp != NULL) \
uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
cache_zone_small : cache_zone_large, (ncp)); \
} while (0)
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
static int doingcache = 1; /* 1 => enable the cache */
SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/* Export size information to userland */
remove duplicate sizeof vnode entry (debug.sizeof.vnode already existed)... move ncsize into debug.sizeof and rename to namecache...
2006-04-16 18:38:30 +00:00
SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
sizeof(struct namecache), "");
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
/*
* The new name cache statistics
*/
Make a SYSCTL_NODE and a mutex static
2005-02-10 12:16:42 +00:00
static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
#define STATNODE(mode, name, var) \
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
STATNODE(CTLFLAG_RD, numneg, &numneg);
STATNODE(CTLFLAG_RD, numcache, &numcache);
static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
Add one more counter so we can truly find out how good our name cache is. If we don't find something and don't what to have found something, it's actually a success.
1997-09-24 15:54:10 +00:00
static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
o Export nchstats ("VFS cache effectiveness statistics") using SYSCTL_OPAQUE. This removes a reason that systat requires setgid kmem. More to come.
2000-11-20 00:41:11 +00:00
SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD, &nchstats,
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
o Export nchstats ("VFS cache effectiveness statistics") using SYSCTL_OPAQUE. This removes a reason that systat requires setgid kmem. More to come.
2000-11-20 00:41:11 +00:00
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
static void cache_zap(struct namecache *ncp);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
char *buf, char **retbuf, u_int buflen);
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
Staticize some malloc M_ instances.
2000-12-08 20:09:00 +00:00
static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
/*
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
* Flags in namecache.nc_flag
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
*/
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
#define NCF_WHITE 1
Create debug.hashstat.[raw]nchash and debug.hashstat.[raw]nfsnode to enable easy access to the hash chain stats. The raw prefixed versions dump an integer array to userland with the chain lengths. This cheats and calls it an array of 'struct int' rather than 'int' or sysctl -a faithfully dumps out the 128K array on an average machine. The non-raw versions return 4 integers: count, number of chains used, maximum chain length, and percentage utilization (fixed point, multiplied by 100). The raw forms are more useful for analyzing the hash distribution, while the other form can be read easily by humans and stats loggers.
2001-04-11 00:39:20 +00:00
/*
* Grab an atomic snapshot of the name cache hash chain lengths
*/
SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
static int
sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
{
int error;
struct nchashhead *ncpp;
struct namecache *ncp;
int n_nchash;
int count;
n_nchash = nchash + 1; /* nchash is max index, not count */
if (!req->oldptr)
return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
/* Scan hash tables for applicable entries */
for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
count = 0;
LIST_FOREACH(ncp, ncpp, nc_hash) {
count++;
}
nuke caddr_t.
2002-06-28 23:17:36 +00:00
error = SYSCTL_OUT(req, &count, sizeof(count));
Create debug.hashstat.[raw]nchash and debug.hashstat.[raw]nfsnode to enable easy access to the hash chain stats. The raw prefixed versions dump an integer array to userland with the chain lengths. This cheats and calls it an array of 'struct int' rather than 'int' or sysctl -a faithfully dumps out the 128K array on an average machine. The non-raw versions return 4 integers: count, number of chains used, maximum chain length, and percentage utilization (fixed point, multiplied by 100). The raw forms are more useful for analyzing the hash distribution, while the other form can be read easily by humans and stats loggers.
2001-04-11 00:39:20 +00:00
if (error)
return (error);
}
return (0);
}
SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD,
0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths");
static int
sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
{
int error;
struct nchashhead *ncpp;
struct namecache *ncp;
int n_nchash;
int count, maxlength, used, pct;
if (!req->oldptr)
return SYSCTL_OUT(req, 0, 4 * sizeof(int));
n_nchash = nchash + 1; /* nchash is max index, not count */
used = 0;
maxlength = 0;
/* Scan hash tables for applicable entries */
for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
count = 0;
LIST_FOREACH(ncp, ncpp, nc_hash) {
count++;
}
if (count)
used++;
if (maxlength < count)
maxlength = count;
}
n_nchash = nchash + 1;
pct = (used * 100 * 100) / n_nchash;
nuke caddr_t.
2002-06-28 23:17:36 +00:00
error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
Create debug.hashstat.[raw]nchash and debug.hashstat.[raw]nfsnode to enable easy access to the hash chain stats. The raw prefixed versions dump an integer array to userland with the chain lengths. This cheats and calls it an array of 'struct int' rather than 'int' or sysctl -a faithfully dumps out the 128K array on an average machine. The non-raw versions return 4 integers: count, number of chains used, maximum chain length, and percentage utilization (fixed point, multiplied by 100). The raw forms are more useful for analyzing the hash distribution, while the other form can be read easily by humans and stats loggers.
2001-04-11 00:39:20 +00:00
if (error)
return (error);
nuke caddr_t.
2002-06-28 23:17:36 +00:00
error = SYSCTL_OUT(req, &used, sizeof(used));
Create debug.hashstat.[raw]nchash and debug.hashstat.[raw]nfsnode to enable easy access to the hash chain stats. The raw prefixed versions dump an integer array to userland with the chain lengths. This cheats and calls it an array of 'struct int' rather than 'int' or sysctl -a faithfully dumps out the 128K array on an average machine. The non-raw versions return 4 integers: count, number of chains used, maximum chain length, and percentage utilization (fixed point, multiplied by 100). The raw forms are more useful for analyzing the hash distribution, while the other form can be read easily by humans and stats loggers.
2001-04-11 00:39:20 +00:00
if (error)
return (error);
nuke caddr_t.
2002-06-28 23:17:36 +00:00
error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
Create debug.hashstat.[raw]nchash and debug.hashstat.[raw]nfsnode to enable easy access to the hash chain stats. The raw prefixed versions dump an integer array to userland with the chain lengths. This cheats and calls it an array of 'struct int' rather than 'int' or sysctl -a faithfully dumps out the 128K array on an average machine. The non-raw versions return 4 integers: count, number of chains used, maximum chain length, and percentage utilization (fixed point, multiplied by 100). The raw forms are more useful for analyzing the hash distribution, while the other form can be read easily by humans and stats loggers.
2001-04-11 00:39:20 +00:00
if (error)
return (error);
nuke caddr_t.
2002-06-28 23:17:36 +00:00
error = SYSCTL_OUT(req, &pct, sizeof(pct));
Create debug.hashstat.[raw]nchash and debug.hashstat.[raw]nfsnode to enable easy access to the hash chain stats. The raw prefixed versions dump an integer array to userland with the chain lengths. This cheats and calls it an array of 'struct int' rather than 'int' or sysctl -a faithfully dumps out the 128K array on an average machine. The non-raw versions return 4 integers: count, number of chains used, maximum chain length, and percentage utilization (fixed point, multiplied by 100). The raw forms are more useful for analyzing the hash distribution, while the other form can be read easily by humans and stats loggers.
2001-04-11 00:39:20 +00:00
if (error)
return (error);
return (0);
}
SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
/*
- Remove old comment for PURGE() as it no longer exists and implied it was a comment to cache_zap(). - Add a comment to quickly state what cache_zap() does. Reviewed by: phk, mux
2003-02-15 18:58:06 +00:00
* cache_zap():
*
* Removes a namecache entry from cache, whether it contains an actual
* pointer to a vnode or if it is just a negative cache entry.
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
*/
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
static void
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
cache_zap(ncp)
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
struct namecache *ncp;
{
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
struct vnode *vp;
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
mtx_assert(&cache_lock, MA_OWNED);
- Add KTR_VFS messages for various name cache related events. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:46:03 +00:00
CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
vp = NULL;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
LIST_REMOVE(ncp, nc_hash);
LIST_REMOVE(ncp, nc_src);
Reclaim directory vnodes held in namecache if few free vnodes are available. Only directory vnodes holding no child directory vnodes held in v_cache_src are recycled, so that directory vnodes near the root of the filesystem hierarchy remain in namecache and directory vnodes are not reclaimed in cascade. The period of vnode reclaiming attempt and the number of vnodes attempted to reclaim can be tuned via sysctl(2). Suggested by: tegge Approved by: phk
2001-04-18 11:19:50 +00:00
if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
vp = ncp->nc_dvp;
Reclaim directory vnodes held in namecache if few free vnodes are available. Only directory vnodes holding no child directory vnodes held in v_cache_src are recycled, so that directory vnodes near the root of the filesystem hierarchy remain in namecache and directory vnodes are not reclaimed in cascade. The period of vnode reclaiming attempt and the number of vnodes attempted to reclaim can be tuned via sysctl(2). Suggested by: tegge Approved by: phk
2001-04-18 11:19:50 +00:00
numcachehv--;
}
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (ncp->nc_vp) {
TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
- Clear v_dd in cache_zap() instead of cache_purge() as cache_purge() may not be called in all cases where we free the cnp. Sponsored by: Isilon Systems, Inc.
2005-06-13 05:59:59 +00:00
ncp->nc_vp->v_dd = NULL;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
} else {
TAILQ_REMOVE(&ncneg, ncp, nc_dst);
numneg--;
}
numcache--;
Make the VFS cache use zones instead of malloc(9). This results in a small but noticeable increase in performance for name lookup operations. The code uses two zones, one for short names (less than 32 characters) and one for long names (up to NAME_MAX). Since most file names are fairly short, this saves a considerable amount of space that would otherwise be wasted if we always allocated NAME_MAX bytes. The cutoff value of 32 characters was picked arbitrarily and may benefit from some tweaking; it could also be made into a tunable. Submitted by: hmp
2003-06-13 08:46:13 +00:00
cache_free(ncp);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
if (vp)
vdrop(vp);
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
}
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
/*
* cache_leaf_test()
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
*
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
* Test whether this (directory) vnode's namei cache entry contains
* subdirectories or not. Used to determine whether the directory is
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
* a leaf in the namei cache or not. Note: the directory may still
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
* contain files in the namei cache.
*
* Returns 0 if the directory is a leaf, -1 if it isn't.
*/
int
cache_leaf_test(struct vnode *vp)
{
struct namecache *ncpc;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
int leaf;
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
leaf = 0;
CACHE_LOCK();
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
for (ncpc = LIST_FIRST(&vp->v_cache_src);
ncpc != NULL;
ncpc = LIST_NEXT(ncpc, nc_src)
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
) {
if (ncpc->nc_vp != NULL && ncpc->nc_vp->v_type == VDIR) {
leaf = -1;
break;
}
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
}
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
return (leaf);
After extensive testing it has been determined that adding complexity to avoid removing higher level directory vnodes from the namecache has no perceivable effect and will be removed. This is especially true when vmiodirenable is turned on, which it is by default now. ( vmiodirenable makes a huge difference in directory caching ). The vfs.vmiodirenable and vfs.nameileafonly sysctls have been left in to allow further testing, but I expect to rip out vfs.nameileafonly soon too. I have also determined through testing that the real problem with numvnodes getting too large is due to the VM Page cache preventing the vnode from being reclaimed. The directory stuff made only a tiny dent relative to Poul's original code, enough so that some tests succeeded. But tests with several million small files show that the bigger problem is the VM Page cache. This will have to be addressed by a future commit. MFC after: 3 days
2001-10-01 04:33:35 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
/*
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
* Lookup an entry in the cache
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
*
* Lookup is called with dvp pointing to the directory to search,
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
* cnp pointing to the name of the entry being sought. If the lookup
* succeeds, the vnode is returned in *vpp, and a status of -1 is
* returned. If the lookup determines that the name does not exist
* (negative cacheing), a status of ENOENT is returned. If the lookup
* fails, a status of zero is returned.
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
*
* vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
* unlocked. If we're looking up . an extra ref is taken, but the lock is
* not recursively acquired.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
int
cache_lookup(dvp, vpp, cnp)
struct vnode *dvp;
struct vnode **vpp;
struct componentname *cnp;
{
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
struct namecache *ncp;
Use a generic implementation of the Fowler/Noll/Vo hash (FNV hash). Make the name cache hash as well as the nfsnode hash use it. As a special tweak, create an unsigned version of register_t. This allows us to use a special tweak for the 64 bit versions that significantly speeds up the i386 version (ie: int64 XOR int64 is slower than int64 XOR int32). The code layout is a little strange for the string function, but I was able to get between 5 to 10% improvement over the original version I started with. The layout affects gcc code generation choices and this way was fastest on x86 and alpha. Note that 'CPUTYPE=p3' etc makes a fair difference to this. It is around 45% faster with -march=pentiumpro on a p6 cpu.
2001-03-17 09:31:06 +00:00
u_int32_t hash;
- Change all filesystems and vfs_cache to relock the dvp once the child is locked in the ISDOTDOT case. Se vfs_lookup.c r1.79 for details. Sponsored by: Isilon Systems, Inc.
2005-04-13 10:59:09 +00:00
int error;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Update vfs_cache.c to use the <sys/queue.h> macros. This makes it easier to read, but doesn't change the speed. Reviewed by: phk Obtained from: via NetBSD
1995-03-06 06:45:52 +00:00
if (!doingcache) {
cnp->cn_flags &= ~MAKEENTRY;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
return (0);
Update vfs_cache.c to use the <sys/queue.h> macros. This makes it easier to read, but doesn't change the speed. Reviewed by: phk Obtained from: via NetBSD
1995-03-06 06:45:52 +00:00
}
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
retry:
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_LOCK();
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
numcalls++;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (cnp->cn_nameptr[0] == '.') {
if (cnp->cn_namelen == 1) {
*vpp = dvp;
- Add KTR_VFS messages for various name cache related events. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:46:03 +00:00
CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
dvp, cnp->cn_nameptr);
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
dothits++;
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
goto success;
Update vfs_cache.c to use the <sys/queue.h> macros. This makes it easier to read, but doesn't change the speed. Reviewed by: phk Obtained from: via NetBSD
1995-03-06 06:45:52 +00:00
}
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
dotdothits++;
Eliminate v_id and v_ddid. The name cache now holds references to vnodes whose names it caches, so we no longer need a `generation number' to tell us if a referenced vnode is invalid. Replace the use of the parent's v_id in the hash function with the address of the parent vnode. Tested by: Peter Holm Glanced at by: jeff, phk
2005-03-30 03:01:36 +00:00
if (dvp->v_dd == NULL ||
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
(cnp->cn_flags & MAKEENTRY) == 0) {
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
return (0);
}
*vpp = dvp->v_dd;
- Add KTR_VFS messages for various name cache related events. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:46:03 +00:00
CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
dvp, cnp->cn_nameptr, *vpp);
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
goto success;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
}
}
Use the same API as the example code. Allow the initial hash value to be passed in, as the examples do. Incrementally hash in the dvp->v_id (using the official api) rather than add it. This seems to help power-of-two predictable filename trees where the filenames repeat on a power-of-two cycle and the directory trees have power-of-two components in it. The simple add then mask was causing things like 12000+ entry collision chains while most other entries have between 0 and 3 entries each. This way seems to improve things.
2001-03-20 02:10:18 +00:00
hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
Eliminate v_id and v_ddid. The name cache now holds references to vnodes whose names it caches, so we no longer need a `generation number' to tell us if a referenced vnode is invalid. Replace the use of the parent's v_id in the hash function with the address of the parent vnode. Tested by: Peter Holm Glanced at by: jeff, phk
2005-03-30 03:01:36 +00:00
hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
Use the same API as the example code. Allow the initial hash value to be passed in, as the examples do. Incrementally hash in the dvp->v_id (using the official api) rather than add it. This seems to help power-of-two predictable filename trees where the filenames repeat on a power-of-two cycle and the directory trees have power-of-two components in it. The simple add then mask was causing things like 12000+ entry collision chains while most other entries have between 0 and 3 entries each. This way seems to improve things.
2001-03-20 02:10:18 +00:00
LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
numchecks++;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
Removed some bogus casts.
1997-12-19 23:18:37 +00:00
!bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
break;
}
/* We failed to find an entry */
if (ncp == 0) {
Add one more counter so we can truly find out how good our name cache is. If we don't find something and don't what to have found something, it's actually a success.
1997-09-24 15:54:10 +00:00
if ((cnp->cn_flags & MAKEENTRY) == 0) {
nummisszap++;
} else {
nummiss++;
}
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
nchstats.ncs_miss++;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
return (0);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
/* We don't want to have an entry, so dump it */
Update vfs_cache.c to use the <sys/queue.h> macros. This makes it easier to read, but doesn't change the speed. Reviewed by: phk Obtained from: via NetBSD
1995-03-06 06:45:52 +00:00
if ((cnp->cn_flags & MAKEENTRY) == 0) {
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
numposzaps++;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
nchstats.ncs_badhits++;
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
cache_zap(ncp);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
return (0);
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
}
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
/* We found a "positive" match, return the vnode */
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
if (ncp->nc_vp) {
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
numposhits++;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
nchstats.ncs_goodhits++;
*vpp = ncp->nc_vp;
- Add KTR_VFS messages for various name cache related events. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:46:03 +00:00
CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
dvp, cnp->cn_nameptr, *vpp, ncp);
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
goto success;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
/* We found a negative match, and want to create it, so purge */
if (cnp->cn_nameiop == CREATE) {
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
numnegzaps++;
Update a couple of counters.
1995-03-12 02:01:20 +00:00
nchstats.ncs_badhits++;
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
cache_zap(ncp);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
return (0);
}
A couple of handles to tweak, more statistics.
1997-09-24 07:46:54 +00:00
numneghits++;
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
/*
- Update a couple of comments to make sense with what today's code is doing (stale comments make arr something something ;)).
2003-02-15 23:25:12 +00:00
* We found a "negative" match, so we shift it to the end of
* the "negative" cache entries queue to satisfy LRU. Also,
* check to see if the entry is a whiteout; indicate this to
* the componentname, if so.
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
*/
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
TAILQ_REMOVE(&ncneg, ncp, nc_dst);
TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
nchstats.ncs_neghits++;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (ncp->nc_flag & NCF_WHITE)
cnp->cn_flags |= ISWHITEOUT;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
return (ENOENT);
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
success:
/*
* On success we return a locked and ref'd vnode as per the lookup
* protocol.
*/
if (dvp == *vpp) { /* lookup on "." */
VREF(*vpp);
CACHE_UNLOCK();
return (-1);
}
if (cnp->cn_flags & ISDOTDOT)
VOP_UNLOCK(dvp, 0, cnp->cn_thread);
VI_LOCK(*vpp);
CACHE_UNLOCK();
- Change all filesystems and vfs_cache to relock the dvp once the child is locked in the ISDOTDOT case. Se vfs_lookup.c r1.79 for details. Sponsored by: Isilon Systems, Inc.
2005-04-13 10:59:09 +00:00
error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
if (cnp->cn_flags & ISDOTDOT)
vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, cnp->cn_thread);
if (error) {
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
*vpp = NULL;
goto retry;
}
return (-1);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
/*
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
* Add an entry to the cache.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
void
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
cache_enter(dvp, vp, cnp)
struct vnode *dvp;
struct vnode *vp;
struct componentname *cnp;
{
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
struct namecache *ncp;
struct nchashhead *ncpp;
Use a generic implementation of the Fowler/Noll/Vo hash (FNV hash). Make the name cache hash as well as the nfsnode hash use it. As a special tweak, create an unsigned version of register_t. This allows us to use a special tweak for the 64 bit versions that significantly speeds up the i386 version (ie: int64 XOR int64 is slower than int64 XOR int32). The code layout is a little strange for the string function, but I was able to get between 5 to 10% improvement over the original version I started with. The layout affects gcc code generation choices and this way was fastest on x86 and alpha. Note that 'CPUTYPE=p3' etc makes a fair difference to this. It is around 45% faster with -march=pentiumpro on a p6 cpu.
2001-03-17 09:31:06 +00:00
u_int32_t hash;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
int hold;
int zap;
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
int len;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
- Add KTR_VFS messages for various name cache related events. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:46:03 +00:00
CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
- Assert that we're not adding a doomed vnode to the name cache. Sponsored by: Isilon Systems, Inc.
2005-06-11 08:47:30 +00:00
VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
("cahe_enter: Adding a doomed vnode"));
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
if (!doingcache)
return;
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (cnp->cn_nameptr[0] == '.') {
if (cnp->cn_namelen == 1) {
return;
}
- Fix a leaked reference to a vnode via v_dd. We rely on cache_purge() and cache_zap() to clear the v_dd pointers when a directory vnode is forcibly discarded. For this to work, all vnodes with v_dd pointers to a directory must also have name cache entries linked via v_cache_dst to that dvp otherwise we could not find them at cache_purge() time. The following code snipit could break this guarantee by unlinking a directory before fetching it's dotdot. The dotdot lookup would initialize the v_dd field of the unlinked directory which could never be cleared. To fix this we don't initialize v_dd for orphaned vnodes. printf("rmdir: %d\n", rmdir("../foo")); /* foo is cwd */ printf("chdir: %d\n", chdir("..")); printf("%s\n", getwd(NULL)); Sponsored by: Isilon Systems, Inc. Discovered by: kkenn Approved by: re (blanket vfs)
2005-06-17 01:05:13 +00:00
/*
* For dotdot lookups only cache the v_dd pointer if the
* directory has a link back to its parent via v_cache_dst.
* Without this an unlinked directory would keep a soft
* reference to its parent which could not be NULLd at
* cache_purge() time.
*/
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
- Fix a leaked reference to a vnode via v_dd. We rely on cache_purge() and cache_zap() to clear the v_dd pointers when a directory vnode is forcibly discarded. For this to work, all vnodes with v_dd pointers to a directory must also have name cache entries linked via v_cache_dst to that dvp otherwise we could not find them at cache_purge() time. The following code snipit could break this guarantee by unlinking a directory before fetching it's dotdot. The dotdot lookup would initialize the v_dd field of the unlinked directory which could never be cleared. To fix this we don't initialize v_dd for orphaned vnodes. printf("rmdir: %d\n", rmdir("../foo")); /* foo is cwd */ printf("chdir: %d\n", chdir("..")); printf("%s\n", getwd(NULL)); Sponsored by: Isilon Systems, Inc. Discovered by: kkenn Approved by: re (blanket vfs)
2005-06-17 01:05:13 +00:00
CACHE_LOCK();
if (!TAILQ_EMPTY(&dvp->v_cache_dst))
dvp->v_dd = vp;
CACHE_UNLOCK();
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
return;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
}
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
hold = 0;
zap = 0;
Make the VFS cache use zones instead of malloc(9). This results in a small but noticeable increase in performance for name lookup operations. The code uses two zones, one for short names (less than 32 characters) and one for long names (up to NAME_MAX). Since most file names are fairly short, this saves a considerable amount of space that would otherwise be wasted if we always allocated NAME_MAX bytes. The cutoff value of 32 characters was picked arbitrarily and may benefit from some tweaking; it could also be made into a tunable. Submitted by: hmp
2003-06-13 08:46:13 +00:00
ncp = cache_alloc(cnp->cn_namelen);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_LOCK();
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
numcache++;
Change the 0xdeadb hack to a flag called VDOOMED. Introduce VFREE which indicates that vnode is on freelist. Rename vholdrele() to vdrop(). Create vfree() and vbusy() to add/delete vnode from freelist. Add vfree()/vbusy() to keep (v_holdcnt != 0 || v_usecount != 0) vnodes off the freelist. Generalize vhold()/v_holdcnt to mean "do not recycle". Fix reassignbuf()s lack of use of vhold(). Use vhold() instead of checking v_cache_src list. Remove vtouch(), the vnodes are always vget'ed soon enough after for it to have any measuable effect. Add sysctl debug.freevnodes to keep track of things. Move cache_purge() up in getnewvnodes to avoid race. Decrement v_usecount after VOP_INACTIVE(), put a vhold() on it during VOP_INACTIVE() Unmacroize vhold()/vdrop() Print out VDOOMED and VFREE flags (XXX: should use %b) Reviewed by: dyson
1997-08-31 07:32:39 +00:00
if (!vp) {
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
numneg++;
Change the 0xdeadb hack to a flag called VDOOMED. Introduce VFREE which indicates that vnode is on freelist. Rename vholdrele() to vdrop(). Create vfree() and vbusy() to add/delete vnode from freelist. Add vfree()/vbusy() to keep (v_holdcnt != 0 || v_usecount != 0) vnodes off the freelist. Generalize vhold()/v_holdcnt to mean "do not recycle". Fix reassignbuf()s lack of use of vhold(). Use vhold() instead of checking v_cache_src list. Remove vtouch(), the vnodes are always vget'ed soon enough after for it to have any measuable effect. Add sysctl debug.freevnodes to keep track of things. Move cache_purge() up in getnewvnodes to avoid race. Decrement v_usecount after VOP_INACTIVE(), put a vhold() on it during VOP_INACTIVE() Unmacroize vhold()/vdrop() Print out VDOOMED and VFREE flags (XXX: should use %b) Reviewed by: dyson
1997-08-31 07:32:39 +00:00
ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
Use 2^N hash sizes rather than primesize, this replaces a division with an and. (Submitted by davidg) Preemptively record ".." values. Reviewed by: phk
1997-09-03 09:20:17 +00:00
} else if (vp->v_type == VDIR) {
vp->v_dd = dvp;
Eliminate v_id and v_ddid. The name cache now holds references to vnodes whose names it caches, so we no longer need a `generation number' to tell us if a referenced vnode is invalid. Replace the use of the parent's v_id in the hash function with the address of the parent vnode. Tested by: Peter Holm Glanced at by: jeff, phk
2005-03-30 03:01:36 +00:00
} else {
vp->v_dd = NULL;
Change the 0xdeadb hack to a flag called VDOOMED. Introduce VFREE which indicates that vnode is on freelist. Rename vholdrele() to vdrop(). Create vfree() and vbusy() to add/delete vnode from freelist. Add vfree()/vbusy() to keep (v_holdcnt != 0 || v_usecount != 0) vnodes off the freelist. Generalize vhold()/v_holdcnt to mean "do not recycle". Fix reassignbuf()s lack of use of vhold(). Use vhold() instead of checking v_cache_src list. Remove vtouch(), the vnodes are always vget'ed soon enough after for it to have any measuable effect. Add sysctl debug.freevnodes to keep track of things. Move cache_purge() up in getnewvnodes to avoid race. Decrement v_usecount after VOP_INACTIVE(), put a vhold() on it during VOP_INACTIVE() Unmacroize vhold()/vdrop() Print out VDOOMED and VFREE flags (XXX: should use %b) Reviewed by: dyson
1997-08-31 07:32:39 +00:00
}
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
/*
- Update a couple of comments to make sense with what today's code is doing (stale comments make arr something something ;)).
2003-02-15 23:25:12 +00:00
* Set the rest of the namecache entry elements, calculate it's
* hash key and insert it into the appropriate chain within
* the cache entries table.
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
*/
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
ncp->nc_vp = vp;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
ncp->nc_dvp = dvp;
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
len = ncp->nc_nlen = cnp->cn_namelen;
Use the same API as the example code. Allow the initial hash value to be passed in, as the examples do. Incrementally hash in the dvp->v_id (using the official api) rather than add it. This seems to help power-of-two predictable filename trees where the filenames repeat on a power-of-two cycle and the directory trees have power-of-two components in it. The simple add then mask was causing things like 12000+ entry collision chains while most other entries have between 0 and 3 entries each. This way seems to improve things.
2001-03-20 02:10:18 +00:00
hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
Use a generic implementation of the Fowler/Noll/Vo hash (FNV hash). Make the name cache hash as well as the nfsnode hash use it. As a special tweak, create an unsigned version of register_t. This allows us to use a special tweak for the 64 bit versions that significantly speeds up the i386 version (ie: int64 XOR int64 is slower than int64 XOR int32). The code layout is a little strange for the string function, but I was able to get between 5 to 10% improvement over the original version I started with. The layout affects gcc code generation choices and this way was fastest on x86 and alpha. Note that 'CPUTYPE=p3' etc makes a fair difference to this. It is around 45% faster with -march=pentiumpro on a p6 cpu.
2001-03-17 09:31:06 +00:00
bcopy(cnp->cn_nameptr, ncp->nc_name, len);
Eliminate v_id and v_ddid. The name cache now holds references to vnodes whose names it caches, so we no longer need a `generation number' to tell us if a referenced vnode is invalid. Replace the use of the parent's v_id in the hash function with the address of the parent vnode. Tested by: Peter Holm Glanced at by: jeff, phk
2005-03-30 03:01:36 +00:00
hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
Use the same API as the example code. Allow the initial hash value to be passed in, as the examples do. Incrementally hash in the dvp->v_id (using the official api) rather than add it. This seems to help power-of-two predictable filename trees where the filenames repeat on a power-of-two cycle and the directory trees have power-of-two components in it. The simple add then mask was causing things like 12000+ entry collision chains while most other entries have between 0 and 3 entries each. This way seems to improve things.
2001-03-20 02:10:18 +00:00
ncpp = NCHHASH(hash);
Update vfs_cache.c to use the <sys/queue.h> macros. This makes it easier to read, but doesn't change the speed. Reviewed by: phk Obtained from: via NetBSD
1995-03-06 06:45:52 +00:00
LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
Reclaim directory vnodes held in namecache if few free vnodes are available. Only directory vnodes holding no child directory vnodes held in v_cache_src are recycled, so that directory vnodes near the root of the filesystem hierarchy remain in namecache and directory vnodes are not reclaimed in cascade. The period of vnode reclaiming attempt and the number of vnodes attempted to reclaim can be tuned via sysctl(2). Suggested by: tegge Approved by: phk
2001-04-18 11:19:50 +00:00
if (LIST_EMPTY(&dvp->v_cache_src)) {
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
hold = 1;
Reclaim directory vnodes held in namecache if few free vnodes are available. Only directory vnodes holding no child directory vnodes held in v_cache_src are recycled, so that directory vnodes near the root of the filesystem hierarchy remain in namecache and directory vnodes are not reclaimed in cascade. The period of vnode reclaiming attempt and the number of vnodes attempted to reclaim can be tuned via sysctl(2). Suggested by: tegge Approved by: phk
2001-04-18 11:19:50 +00:00
numcachehv++;
}
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
- Update a couple of comments to make sense with what today's code is doing (stale comments make arr something something ;)).
2003-02-15 23:25:12 +00:00
/*
* If the entry is "negative", we place it into the
* "negative" cache queue, otherwise, we place it into the
* destination vnode's cache entries queue.
*/
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
if (vp) {
TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
} else {
TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
}
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
if (numneg * ncnegfactor > numcache) {
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
ncp = TAILQ_FIRST(&ncneg);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
zap = 1;
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
}
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
if (hold)
vhold(dvp);
if (zap)
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
cache_zap(ncp);
CACHE_UNLOCK();
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
/*
* Name cache initialization, from vfs_init() when we are booting
*/
Untangle vfsinit() a bit. Use seperate sysinit functions rather than having a super-function calling bits all over the place.
2000-12-06 07:09:08 +00:00
static void
nchinit(void *dummy __unused)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Fixed the hash formula. Lite2 doesn't have phashinit(), so Lite2's hash formula uses `& nchash'. This is very broken when nchash is a prime number instead of 1 less than a power of 2, but the Lite2 formula was merged in. Merged some cosmetic changes from Lite2, rev.1.21 and Lite1. The merge was difficult because the Lite2 code is essentially ours (phk's) except where Lite2 improved or broke it. Summary of the Lite2 changes: - in the copyright, phk's rights have been transferred to the Regents. This change should be reviewed. - nchENOENT went away; the "no" vnode is now simply 0. - comments were improved. - style was "improved". - goto instead of Fanatism (sic) was considered bad :-). - there are some small changes to support whiteouts. - new cache entries are added in more cases. More work is required near here to change the hash table size if kern.desiredvnodes is changed using sysctl. - rescanning of the hash bucket in cache_purgevfs() was removed. This change should be reviewed.
1997-03-08 15:22:14 +00:00
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
TAILQ_INIT(&ncneg);
Make the VFS cache use zones instead of malloc(9). This results in a small but noticeable increase in performance for name lookup operations. The code uses two zones, one for short names (less than 32 characters) and one for long names (up to NAME_MAX). Since most file names are fairly short, this saves a considerable amount of space that would otherwise be wasted if we always allocated NAME_MAX bytes. The cutoff value of 32 characters was picked arbitrarily and may benefit from some tweaking; it could also be made into a tunable. Submitted by: hmp
2003-06-13 08:46:13 +00:00
cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
Untangle vfsinit() a bit. Use seperate sysinit functions rather than having a super-function calling bits all over the place.
2000-12-06 07:09:08 +00:00
nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Untangle vfsinit() a bit. Use seperate sysinit functions rather than having a super-function calling bits all over the place.
2000-12-06 07:09:08 +00:00
SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
/*
Fix a braino in the v_id wraparound code. Give more (current) details in comment. PR: 11307 Spotted by: Ville-Pertti Keinonen <will@iki.fi>
1999-04-24 17:58:14 +00:00
* Invalidate all entries to a particular vnode.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
void
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
cache_purge(vp)
struct vnode *vp;
{
- Add KTR_VFS messages for various name cache related events. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:46:03 +00:00
CTR1(KTR_VFS, "cache_purge(%p)", vp);
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_LOCK();
- Clear v_dd in cache_zap() instead of cache_purge() as cache_purge() may not be called in all cases where we free the cnp. Sponsored by: Isilon Systems, Inc.
2005-06-13 05:59:59 +00:00
while (!LIST_EMPTY(&vp->v_cache_src))
cache_zap(LIST_FIRST(&vp->v_cache_src));
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
while (!TAILQ_EMPTY(&vp->v_cache_dst))
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
Eliminate v_id and v_ddid. The name cache now holds references to vnodes whose names it caches, so we no longer need a `generation number' to tell us if a referenced vnode is invalid. Replace the use of the parent's v_id in the hash function with the address of the parent vnode. Tested by: Peter Holm Glanced at by: jeff, phk
2005-03-30 03:01:36 +00:00
vp->v_dd = NULL;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
/*
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
* Flush all entries referencing a particular filesystem.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
* Since we need to check it anyway, we will flush all the invalid
kern_conf.c: remove a now unused variable. vfs_cache.c: Fix a very rare probelm in the vnode-cache. Submitted by: Terry Lambert <terry@lambert.org>
1995-12-22 15:56:35 +00:00
* entries at the same time.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
void
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
cache_purgevfs(mp)
struct mount *mp;
{
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
struct nchashhead *ncpp;
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
struct namecache *ncp, *nnp;
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
struct nchashhead mplist;
LIST_INIT(&mplist);
ncp = NULL;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
/* Scan hash tables for applicable entries */
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_LOCK();
Use 2^N hash sizes rather than primesize, this replaces a division with an and. (Submitted by davidg) Preemptively record ".." values. Reviewed by: phk
1997-09-03 09:20:17 +00:00
for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
1. Add a {pointer, v_id} pair to the vnode to store the reference to the ".." vnode. This is cheaper storagewise than keeping it in the namecache, and it makes more sense since it's a 1:1 mapping. 2. Also handle the case of "." more intelligently rather than stuff the namecache with pointless entries. 3. Add two lists to the vnode and hang namecache entries which go from or to this vnode. When cleaning a vnode, delete all namecache entries it invalidates. 4. Never reuse namecache enties, malloc new ones when we need it, free old ones when they die. No longer a hard limit on how many we can have. 5. Remove the upper limit on namelength of namecache entries. 6. Make a global list for negative namecache entries, limit their number to a sysctl'able (debug.ncnegfactor) fraction of the total namecache. Currently the default fraction is 1/16th. (Suggestions for better default wanted!) 7. Assign v_id correctly in the face of 32bit rollover. 8. Remove the LRU list for namecache entries, not needed. Remove the #ifdef NCH_STATISTICS stuff, it's not needed either. 9. Use the vnode freelist as a true LRU list, also for namecache accesses. 10. Reuse vnodes more aggresively but also more selectively, if we can't reuse, malloc a new one. There is no longer a hard limit on their number, they grow to the point where we don't reuse potentially usable vnodes. A vnode will not get recycled if still has pages in core or if it is the source of namecache entries (Yes, this does indeed work :-) "." and ".." are not namecache entries any longer...) 11. Do not overload the v_id field in namecache entries with whiteout information, use a char sized flags field instead, so we can get rid of the vpid and v_id fields from the namecache struct. Since we're linked to the vnodes and purged when they're cleaned, we don't have to check the v_id any more. 12. NFS knew about the limitation on name length in the namecache, it shouldn't and doesn't now. Bugs: The namecache statistics no longer includes the hits for ".." and "." hits. Performance impact: Generally in the +/- 0.5% for "normal" workstations, but I hope this will allow the system to be selftuning over a bigger range of "special" applications. The case where RAM is available but unused for cache because we don't have any vnodes should be gone. Future work: Straighten out the namecache statistics. "desiredvnodes" is still used to (bogusly ?) size hash tables in the filesystems. I have still to find a way to safely free unused vnodes back so their number can shrink when not needed. There is a few uses of the v_id field left in the filesystems, scheduled for demolition at a later time. Maybe a one slot cache for unused namecache entries should be implemented to decrease the malloc/free frequency.
1997-05-04 09:17:38 +00:00
for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
nnp = LIST_NEXT(ncp, nc_hash);
if (ncp->nc_dvp->v_mount == mp) {
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
LIST_REMOVE(ncp, nc_hash);
LIST_INSERT_HEAD(&mplist, ncp, nc_hash);
Clean up and improve the namecache. 1. We always keep one 16th of the vnodes on the freelist, so that the namecache doesn't get trashed. It used to be that it wasn't a problem, but the only vnodes getting released these days are directories and things which gets forced out of the VM/cache. The latter is not numerous enough to keep the pool of vnodes needed for the namecache sufficiently big. 2. Purge invalid entries in the namecache as soon as we notice them. This avoids a stale entry pushing out a valid entry on the LRU list. 3. Speed up the lookup in the namecache by avoid a special case branch. 4. Make the cache purge routines do the thing they're supposed to, and in a decently efficient manner. 5. Make the size of the namecache follow the number of vnodes, so that we can always point to all the vnodes we have in core. 6. Readability has gone way up. 7. Added a "options NCH_STATISTICS" feature that will gather more detailed statistics on the performance of the namecache. Reviewed by: davidg
1995-03-09 20:23:45 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
}
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
while (!LIST_EMPTY(&mplist))
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
cache_zap(LIST_FIRST(&mplist));
CACHE_UNLOCK();
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
/*
* Perform canonical checks and cache lookup and pass on to filesystem
* through the vop_cachedlookup only if needed.
*/
int
vfs_cache_lookup(ap)
struct vop_lookup_args /* {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
} */ *ap;
{
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
struct vnode *dvp;
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
int error;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
struct ucred *cred = cnp->cn_cred;
int flags = cnp->cn_flags;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td = cnp->cn_thread;
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
*vpp = NULL;
Change variable naming to be consistent with the rest of VFS code. Reduce number of indirections by using already fetched values.
2000-09-10 03:46:12 +00:00
dvp = ap->a_dvp;
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
Change variable naming to be consistent with the rest of VFS code. Reduce number of indirections by using already fetched values.
2000-09-10 03:46:12 +00:00
if (dvp->v_type != VDIR)
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
return (ENOTDIR);
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
Change variable naming to be consistent with the rest of VFS code. Reduce number of indirections by using already fetched values.
2000-09-10 03:46:12 +00:00
if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
error = VOP_ACCESS(dvp, VEXEC, cred, td);
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
if (error)
return (error);
Change variable naming to be consistent with the rest of VFS code. Reduce number of indirections by using already fetched values.
2000-09-10 03:46:12 +00:00
error = cache_lookup(dvp, vpp, cnp);
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
if (error == 0)
- Get rid of the old LOOKUP_SHARED code. namei() now supplies the proper lock flags via cn_lkflag. Sponsored by: Isilon Systems, Inc.
2005-03-29 10:08:23 +00:00
return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
if (error == ENOENT)
return (error);
- Move the logic that locks and refs the new vnode from vfs_cache_lookup() to cache_lookup(). This allows us to acquire the vnode interlock before dropping the cache lock. This protects the vnodes identity until we have locked it. Sponsored by: Isilon Systems, Inc.
2005-03-29 12:59:06 +00:00
return (0);
Uncut&paste cache_lookup(). This unifies several times in theory indentical 50 lines of code. The filesystems have a new method: vop_cachedlookup, which is the meat of the lookup, and use vfs_cache_lookup() for their vop_lookup method. vfs_cache_lookup() will check the namecache and pass on to the vop_cachedlookup method in case of a miss. It's still the task of the individual filesystems to populate the namecache with cache_enter(). Filesystems that do not use the namecache will just provide the vop_lookup method as usual.
1997-08-26 07:32:51 +00:00
}
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
#ifndef _SYS_SYSPROTO_H_
struct __getcwd_args {
u_char *buf;
u_int buflen;
};
#endif
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/*
* XXX All of these sysctls would probably be more productive dead.
*/
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
static int disablecwd;
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
"Disable the getcwd syscall");
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
Further system call comment cleanup: - Remove also "MP SAFE" after prior "MPSAFE" pass. (suggested by bde) - Remove extra blank lines in some cases. - Add extra blank lines in some cases. - Remove no-op comments consisting solely of the function name, the word "syscall", or the system call name. - Add punctuation. - Re-wrap some comments.
2007-03-05 13:10:58 +00:00
/* Implementation of the getcwd syscall. */
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
int
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
__getcwd(td, uap)
struct thread *td;
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
struct __getcwd_args *uap;
{
(This commit certainly increases the need for a wash&clean of vfs_cache.c, but I decided that it was important for this patch to not bit-rot, and since it is mainly moving code around, the total amount of entropy is epsilon /phk) This is a patch to move the common parts of linux_getcwd() back into kern/vfs_cache.c so that the standard FreeBSD libc getcwd() can use it's extended functionality. The linux syscall linux_getcwd() in compat/linux/linux_getcwd.c has been rewritten to use it too. It should be possible to simplify libc's getcwd() after this. No doubt this code needs some cleaning up, since I've left in the sysctl variables I used for debugging. PR: 48169 Submitted by: James Whitwell <abacau@yahoo.com.au>
2003-03-17 12:21:08 +00:00
Backout the getcwd changes, a more comprehensive effort will be needed.
2003-03-20 10:40:45 +00:00
return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
(This commit certainly increases the need for a wash&clean of vfs_cache.c, but I decided that it was important for this patch to not bit-rot, and since it is mainly moving code around, the total amount of entropy is epsilon /phk) This is a patch to move the common parts of linux_getcwd() back into kern/vfs_cache.c so that the standard FreeBSD libc getcwd() can use it's extended functionality. The linux syscall linux_getcwd() in compat/linux/linux_getcwd.c has been rewritten to use it too. It should be possible to simplify libc's getcwd() after this. No doubt this code needs some cleaning up, since I've left in the sysctl variables I used for debugging. PR: 48169 Submitted by: James Whitwell <abacau@yahoo.com.au>
2003-03-17 12:21:08 +00:00
}
Split up __getcwd so that kernel callers of the internal version can specify whether the buffer is in user or system space.
2002-09-02 22:40:30 +00:00
int
Backout the getcwd changes, a more comprehensive effort will be needed.
2003-03-20 10:40:45 +00:00
kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
Split up __getcwd so that kernel callers of the internal version can specify whether the buffer is in user or system space.
2002-09-02 22:40:30 +00:00
{
char *bp, *tmpbuf;
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
struct filedesc *fdp;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
int error;
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
Backout the getcwd changes, a more comprehensive effort will be needed.
2003-03-20 10:40:45 +00:00
if (disablecwd)
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
return (ENODEV);
Split up __getcwd so that kernel callers of the internal version can specify whether the buffer is in user or system space.
2002-09-02 22:40:30 +00:00
if (buflen < 2)
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
return (EINVAL);
Split up __getcwd so that kernel callers of the internal version can specify whether the buffer is in user or system space.
2002-09-02 22:40:30 +00:00
if (buflen > MAXPATHLEN)
buflen = MAXPATHLEN;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
fdp = td->td_proc->p_fd;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
mtx_lock(&Giant);
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SLOCK(fdp);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
error = vn_fullpath1(td, fdp->fd_cdir, fdp->fd_rdir, tmpbuf,
&bp, buflen);
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SUNLOCK(fdp);
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
mtx_unlock(&Giant);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
if (!error) {
if (bufseg == UIO_SYSSPACE)
bcopy(bp, buf, strlen(bp) + 1);
else
error = copyout(bp, buf, strlen(bp) + 1);
}
- Simplify the cache locking. The lock order relationship with the vnode lock is much simpler than I originally thought it would be. Now, the cache lock is always acquired before the vnode lock. - Provide some gotos in __getcwd() to simplify the unlocking a bit. - Move Giant acquisition down into __getcwd(). Sponsored By: Isilon Systems, Inc.
2005-01-24 10:24:12 +00:00
free(tmpbuf, M_TEMP);
return (error);
Before we start to mess with the VFS name-cache clean things up a little bit: Isolate the namecache in its own file, and give it a dedicated malloc type.
1999-10-03 12:18:29 +00:00
}
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
/*
* Thus begins the fullpath magic.
*/
#undef STATNODE
#define STATNODE(name) \
static u_int name; \
Support for unsigned integer and long sysctl variables. Update the SYSCTL_LONG macro to be consistent with other integer sysctl variables and require an initial value instead of assuming 0. Update several sysctl variables to use the unsigned types. PR: 15251 Submitted by: Kelly Yancey <kbyanc@posi.net>
2000-07-05 07:46:41 +00:00
SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
static int disablefullpath;
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
"Disable the vn_fullpath function");
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
/* These count for kern___getcwd(), too. */
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
STATNODE(numfullpathcalls);
STATNODE(numfullpathfail1);
STATNODE(numfullpathfail2);
STATNODE(numfullpathfail4);
STATNODE(numfullpathfound);
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/*
* Retrieve the full filesystem path that correspond to a vnode from the name
* cache (if available)
*/
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
int
Convert textvp_fullpath() into the more generic vn_fullpath() which takes a struct thread * and a struct vnode * instead of a struct proc *. Temporarily add a textvp_fullpath macro for compatibility.
2001-10-21 15:52:51 +00:00
vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
{
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
char *buf;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
struct filedesc *fdp;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
int error;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
if (disablefullpath)
return (ENODEV);
Convert textvp_fullpath() into the more generic vn_fullpath() which takes a struct thread * and a struct vnode * instead of a struct proc *. Temporarily add a textvp_fullpath macro for compatibility.
2001-10-21 15:52:51 +00:00
if (vn == NULL)
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
return (EINVAL);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
Back out M_* changes, per decision of the TRB. Approved by: trb
2003-02-19 05:47:46 +00:00
buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
Convert textvp_fullpath() into the more generic vn_fullpath() which takes a struct thread * and a struct vnode * instead of a struct proc *. Temporarily add a textvp_fullpath macro for compatibility.
2001-10-21 15:52:51 +00:00
fdp = td->td_proc->p_fd;
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SLOCK(fdp);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
error = vn_fullpath1(td, vn, fdp->fd_rdir, buf, retbuf, MAXPATHLEN);
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SUNLOCK(fdp);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
if (!error)
*freebuf = buf;
else
free(buf, M_TEMP);
return (error);
}
/*
* The magic behind kern___getcwd() and vn_fullpath().
*/
static int
vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
char *buf, char **retbuf, u_int buflen)
{
char *bp;
int error, i, slash_prefixed;
struct namecache *ncp;
bp = buf + buflen - 1;
*bp = '\0';
error = 0;
slash_prefixed = 0;
CACHE_LOCK();
numfullpathcalls++;
if (vp->v_type != VDIR) {
ncp = TAILQ_FIRST(&vp->v_cache_dst);
if (!ncp) {
numfullpathfail2++;
CACHE_UNLOCK();
return (ENOENT);
}
for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--)
*--bp = ncp->nc_name[i];
if (bp == buf) {
numfullpathfail4++;
CACHE_UNLOCK();
return (ENOMEM);
}
*--bp = '/';
slash_prefixed = 1;
vp = ncp->nc_dvp;
}
while (vp != rdir && vp != rootvnode) {
- Replace v_flag with v_iflag and v_vflag - v_vflag is protected by the vnode lock and is used when synchronization with VOP calls is needed. - v_iflag is protected by interlock and is used for dealing with vnode management issues. These flags include X/O LOCK, FREE, DOOMED, etc. - All accesses to v_iflag and v_vflag have either been locked or marked with mp_fixme's. - Many ASSERT_VOP_LOCKED calls have been added where the locking was not clear. - Many functions in vfs_subr.c were restructured to provide for stronger locking. Idea stolen from: BSD/OS
2002-08-04 10:29:36 +00:00
if (vp->v_vflag & VV_ROOT) {
- Don't check v_mount for NULL to determine if a vnode has been recycled. Use the more appropriate VI_DOOMED flag instead. Sponsored by: Isilon Systems, Inc. MFC After: 1 week
2006-02-06 10:15:27 +00:00
if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
error = EBADF;
break;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
}
vp = vp->v_mount->mnt_vnodecovered;
continue;
}
Eliminate v_id and v_ddid. The name cache now holds references to vnodes whose names it caches, so we no longer need a `generation number' to tell us if a referenced vnode is invalid. Replace the use of the parent's v_id in the hash function with the address of the parent vnode. Tested by: Peter Holm Glanced at by: jeff, phk
2005-03-30 03:01:36 +00:00
if (vp->v_dd == NULL) {
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
numfullpathfail1++;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
error = ENOTDIR;
break;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
}
ncp = TAILQ_FIRST(&vp->v_cache_dst);
if (!ncp) {
numfullpathfail2++;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
error = ENOENT;
break;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
}
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
MPASS(ncp->nc_dvp == vp->v_dd);
for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--)
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
*--bp = ncp->nc_name[i];
if (bp == buf) {
numfullpathfail4++;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
error = ENOMEM;
break;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
}
*--bp = '/';
slash_prefixed = 1;
vp = ncp->nc_dvp;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
}
if (error) {
- Apply a big giant lock around the namecache. This has been sitting in my tree since BSDcon.
2003-10-05 07:13:50 +00:00
CACHE_UNLOCK();
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
return (error);
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
}
if (!slash_prefixed) {
if (bp == buf) {
numfullpathfail4++;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
CACHE_UNLOCK();
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
return (ENOMEM);
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
} else {
*--bp = '/';
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
}
}
numfullpathfound++;
Merge kern___cwd() and vn_fullpath(), which were virtually identical, except for places where people forget to update one of them. We now collect only one set of stats for both of these routines. Other changes in this commit include: - Start acquiring Giant again in vn_fullpath(), since it is required when crossing a mount point. - Expand the scope of the cache lock to avoid dropping it and picking it up again for every pathname component. This also makes it trivial to avoid races in stats collection. - Assert that nc_dvp == v_dd for directories instead of returning an error to userland when this is not true. AFAIK, it should always be true when v_dd is non-null. - For vn_fullpath(), handle the first (non-directory) vnode separately. Glanced at by: jeff, phk
2005-03-30 02:59:32 +00:00
CACHE_UNLOCK();
Whitespace cleanup.
2003-06-11 07:35:56 +00:00
*retbuf = bp;
Move procfs_fullpath() to vfs_cache.c, with a rename to textvp_fullpath(). There's no excuse to have code in synthetic filestores that allows direct references to the textvp anymore. Feature requested by: msmith Feature agreed to by: warner Move requested by: phk Move agreed to by: bde
2000-04-26 11:57:45 +00:00
return (0);
}
Reference in New Issue Copy Permalink