)]}' { "log": [ { "commit": "dea3667bc3c2a0521e8d8855e407a49d9d70028c", "tree": "bda1a624616f2e1bf9d3ec38b30ab002379a8e6a", "parents": [ "b07ad9967f40b164af77205027352ba53729cf5a" ], "author": { "name": "Linus Torvalds", "email": "torvalds@linux-foundation.org", "time": "Sun Apr 24 07:58:46 2011 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@linux-foundation.org", "time": "Sun Apr 24 07:58:46 2011 -0700" }, "message": "vfs: get rid of insane dentry hashing rules\n\nThe dentry hashing rules have been really quite complicated for a long\nwhile, in odd ways. That made functions like __d_drop() very fragile\nand non-obvious.\n\nIn particular, whether a dentry was hashed or not was indicated with an\nexplicit DCACHE_UNHASHED bit. That\u0027s despite the fact that the hash\nabstraction that the dentries use actually have a \u0027is this entry hashed\nor not\u0027 model (which is a simple test of the \u0027pprev\u0027 pointer).\n\nThe reason that was done is because we used the normal \u0027is this entry\nunhashed\u0027 model to mark whether the dentry had _ever_ been hashed in the\ndentry hash tables, and that logic goes back many years (commit\nb3423415fbc2: \"dcache: avoid RCU for never-hashed dentries\").\n\nThat, in turn, meant that __d_drop had totally different unhashing logic\nfor the dentry hash table case and for the anonymous dcache case,\nbecause in order to use the \"is this dentry hashed\" logic as a flag for\nwhether it had ever been on the RCU hash table, we had to unhash such a\ndentry differently so that we\u0027d never think that it wasn\u0027t \u0027unhashed\u0027\nand wouldn\u0027t be free\u0027d correctly.\n\nThat\u0027s just insane. It made the logic really hard to follow, when there\nwere two different kinds of \"unhashed\" states, and one of them (the one\nthat used \"list_bl_unhashed()\") really had nothing at all to do with\nbeing unhashed per se, but with a very subtle lifetime rule instead.\n\nSo turn all of it around, and make it logical.\n\nInstead of having a DENTRY_UNHASHED bit in d_flags to indicate whether\nthe dentry is on the hash chains or not, use the hash chain unhashed\nlogic for that. Suddenly \"d_unhashed()\" just uses \"list_bl_unhashed()\",\nand everything makes sense.\n\nAnd for the lifetime rule, just use an explicit DENTRY_RCUACCEES bit.\nIf we ever insert the dentry into the dentry hash table so that it is\nvisible to RCU lookup, we mark it DENTRY_RCUACCESS to show that it now\nneeds the RCU lifetime rules. Now suddently that test at dentry free\ntime makes sense too.\n\nAnd because unhashing now is sane and doesn\u0027t depend on where the dentry\ngot unhashed from (because the dentry hash chain details doesn\u0027t have\nsome subtle side effects), we can re-unify the __d_drop() logic and use\ncommon code for the unhashing.\n\nAlso fix one more open-coded hash chain bit_spin_lock() that I missed in\nthe previous chain locking cleanup commit.\n\nSigned-off-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\n" }, { "commit": "0f60f240d522772467c7d2cebedb910748c78ed4", "tree": "235d147a07888a97af0149304c3830631e4622e3", "parents": [ "69b195be51620d72956acbf3029adad5765695dc" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Mon Mar 21 14:28:58 2011 +0000" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Mon Mar 21 12:13:10 2011 -0400" }, "message": "FS: lookup_mnt() is only used in the core fs routines now\n\nlookup_mnt() is only used in the core fs routines now, so it doesn\u0027t need to\nbe globally declared anymore. It isn\u0027t exported to modules at the moment, so\nnothing that can be modularised seems to be using it.\n\nSigned-off-by: David Howells \u003cdhowells@redhat.com\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "1aed3e4204dd787d53b3cd6363eb63bb4900c38e", "tree": "503582fa9f7fa68a1da7af3770d3a627d041fb52", "parents": [ "7cc90cc3ffe22a0d81b8d605b20a82ec7911012d" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Fri Mar 18 09:09:02 2011 -0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Fri Mar 18 10:01:59 2011 -0400" }, "message": "lose \u0027mounting_here\u0027 argument in -\u003ed_manage()\n\nit\u0027s always false...\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "ab90911ff90cdab59b31c045c3f0ae480d14f29d", "tree": "683450a66eb9dc6bf053e38d63f4740bb53a7b6e", "parents": [ "87556ef19926e97464e0163a7840140527ae6615" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Fri Jan 14 18:46:51 2011 +0000" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sat Jan 15 20:07:47 2011 -0500" }, "message": "Allow d_manage() to be used in RCU-walk mode\n\nAllow d_manage() to be called from pathwalk when it is in RCU-walk mode as well\nas when it is in Ref-walk mode. This permits __follow_mount_rcu() to call\nd_manage() directly. d_manage() needs a parameter to indicate that it is in\nRCU-walk mode as it isn\u0027t allowed to sleep if in that mode (but should return\n-ECHILD instead).\n\nautofs4_d_manage() can then be set to retain RCU-walk mode if the daemon\naccesses it and otherwise request dropping back to ref-walk mode.\n\nSigned-off-by: David Howells \u003cdhowells@redhat.com\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "cc53ce53c86924bfe98a12ea20b7465038a08792", "tree": "3c9a4923dd9f413c46bfa83a20cb579446df6deb", "parents": [ "9875cf806403fae66b2410a3c2cc820d97731e04" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Fri Jan 14 18:45:26 2011 +0000" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sat Jan 15 20:07:31 2011 -0500" }, "message": "Add a dentry op to allow processes to be held during pathwalk transit\n\nAdd a dentry op (d_manage) to permit a filesystem to hold a process and make it\nsleep when it tries to transit away from one of that filesystem\u0027s directories\nduring a pathwalk. The operation is keyed off a new dentry flag\n(DCACHE_MANAGE_TRANSIT).\n\nThe filesystem is allowed to be selective about which processes it holds and\nwhich it permits to continue on or prohibits from transiting from each flagged\ndirectory. This will allow autofs to hold up client processes whilst letting\nits userspace daemon through to maintain the directory or the stuff behind it\nor mounted upon it.\n\nThe -\u003ed_manage() dentry operation:\n\n\tint (*d_manage)(struct path *path, bool mounting_here);\n\ntakes a pointer to the directory about to be transited away from and a flag\nindicating whether the transit is undertaken by do_add_mount() or\ndo_move_mount() skipping through a pile of filesystems mounted on a mountpoint.\n\nIt should return 0 if successful and to let the process continue on its way;\n-EISDIR to prohibit the caller from skipping to overmounted filesystems or\nautomounting, and to use this directory; or some other error code to return to\nthe user.\n\n-\u003ed_manage() is called with namespace_sem writelocked if mounting_here is true\nand no other locks held, so it may sleep. However, if mounting_here is true,\nit may not initiate or wait for a mount or unmount upon the parameter\ndirectory, even if the act is actually performed by userspace.\n\nWithin fs/namei.c, follow_managed() is extended to check with d_manage() first\non each managed directory, before transiting away from it or attempting to\nautomount upon it.\n\nfollow_down() is renamed follow_down_one() and should only be used where the\nfilesystem deliberately intends to avoid management steps (e.g. autofs).\n\nA new follow_down() is added that incorporates the loop done by all other\ncallers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS\nand CIFS do use it, their use is removed by converting them to use\nd_automount()). The new follow_down() calls d_manage() as appropriate. It\nalso takes an extra parameter to indicate if it is being called from mount code\n(with namespace_sem writelocked) which it passes to d_manage(). follow_down()\nignores automount points so that it can be used to mount on them.\n\n__follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with\nDCACHE_MANAGE_TRANSIT set on the basis that we\u0027re probably going to have to\nsleep. It would be possible to enter d_manage() in rcu-walk mode too, and have\nthat determine whether to abort or not itself. That would allow the autofs\ndaemon to continue on in rcu-walk mode.\n\nNote that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn\u0027t\nrequired as every tranist from that directory will cause d_manage() to be\ninvoked. It can always be set again when necessary.\n\n\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\nWHAT THIS MEANS FOR AUTOFS\n\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\n\nAutofs currently uses the lookup() inode op and the d_revalidate() dentry op to\ntrigger the automounting of indirect mounts, and both of these can be called\nwith i_mutex held.\n\nautofs knows that the i_mutex will be held by the caller in lookup(), and so\ncan drop it before invoking the daemon - but this isn\u0027t so for d_revalidate(),\nsince the lock is only held on _some_ of the code paths that call it. This\nmeans that autofs can\u0027t risk dropping i_mutex from its d_revalidate() function\nbefore it calls the daemon.\n\nThe bug could manifest itself as, for example, a process that\u0027s trying to\nvalidate an automount dentry that gets made to wait because that dentry is\nexpired and needs cleaning up:\n\n\tmkdir S ffffffff8014e05a 0 32580 24956\n\tCall Trace:\n\t [\u003cffffffff885371fd\u003e] :autofs4:autofs4_wait+0x674/0x897\n\t [\u003cffffffff80127f7d\u003e] avc_has_perm+0x46/0x58\n\t [\u003cffffffff8009fdcf\u003e] autoremove_wake_function+0x0/0x2e\n\t [\u003cffffffff88537be6\u003e] :autofs4:autofs4_expire_wait+0x41/0x6b\n\t [\u003cffffffff88535cfc\u003e] :autofs4:autofs4_revalidate+0x91/0x149\n\t [\u003cffffffff80036d96\u003e] __lookup_hash+0xa0/0x12f\n\t [\u003cffffffff80057a2f\u003e] lookup_create+0x46/0x80\n\t [\u003cffffffff800e6e31\u003e] sys_mkdirat+0x56/0xe4\n\nversus the automount daemon which wants to remove that dentry, but can\u0027t\nbecause the normal process is holding the i_mutex lock:\n\n\tautomount D ffffffff8014e05a 0 32581 1 32561\n\tCall Trace:\n\t [\u003cffffffff80063c3f\u003e] __mutex_lock_slowpath+0x60/0x9b\n\t [\u003cffffffff8000ccf1\u003e] do_path_lookup+0x2ca/0x2f1\n\t [\u003cffffffff80063c89\u003e] .text.lock.mutex+0xf/0x14\n\t [\u003cffffffff800e6d55\u003e] do_rmdir+0x77/0xde\n\t [\u003cffffffff8005d229\u003e] tracesys+0x71/0xe0\n\t [\u003cffffffff8005d28d\u003e] tracesys+0xd5/0xe0\n\nwhich means that the system is deadlocked.\n\nThis patch allows autofs to hold up normal processes whilst the daemon goes\nahead and does things to the dentry tree behind the automouter point without\nrisking a deadlock as almost no locks are held in d_manage() and none in\nd_automount().\n\nSigned-off-by: David Howells \u003cdhowells@redhat.com\u003e\nWas-Acked-by: Ian Kent \u003craven@themaw.net\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "9875cf806403fae66b2410a3c2cc820d97731e04", "tree": "6f9546b400716766af95e0f78e3d600e765b2b51", "parents": [ "1a8edf40e7c3eee955e0dd0316a7c9d85e36f597" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Fri Jan 14 18:45:21 2011 +0000" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sat Jan 15 20:05:03 2011 -0500" }, "message": "Add a dentry op to handle automounting rather than abusing follow_link()\n\nAdd a dentry op (d_automount) to handle automounting directories rather than\nabusing the follow_link() inode operation. The operation is keyed off a new\ndentry flag (DCACHE_NEED_AUTOMOUNT).\n\nThis also makes it easier to add an AT_ flag to suppress terminal segment\nautomount during pathwalk and removes the need for the kludge code in the\npathwalk algorithm to handle directories with follow_link() semantics.\n\nThe -\u003ed_automount() dentry operation:\n\n\tstruct vfsmount *(*d_automount)(struct path *mountpoint);\n\ntakes a pointer to the directory to be mounted upon, which is expected to\nprovide sufficient data to determine what should be mounted. If successful, it\nshould return the vfsmount struct it creates (which it should also have added\nto the namespace using do_add_mount() or similar). If there\u0027s a collision with\nanother automount attempt, NULL should be returned. If the directory specified\nby the parameter should be used directly rather than being mounted upon,\n-EISDIR should be returned. In any other case, an error code should be\nreturned.\n\nThe -\u003ed_automount() operation is called with no locks held and may sleep. At\nthis point the pathwalk algorithm will be in ref-walk mode.\n\nWithin fs/namei.c itself, a new pathwalk subroutine (follow_automount()) is\nadded to handle mountpoints. It will return -EREMOTE if the automount flag was\nset, but no d_automount() op was supplied, -ELOOP if we\u0027ve encountered too many\nsymlinks or mountpoints, -EISDIR if the walk point should be used without\nmounting and 0 if successful. The path will be updated to point to the mounted\nfilesystem if a successful automount took place.\n\n__follow_mount() is replaced by follow_managed() which is more generic\n(especially with the patch that adds -\u003ed_manage()). This handles transits from\ndirectories during pathwalk, including automounting and skipping over\nmountpoints (and holding processes with the next patch).\n\n__follow_mount_rcu() will jump out of RCU-walk mode if it encounters an\nautomount point with nothing mounted on it.\n\nfollow_dotdot*() does not handle automounts as you don\u0027t want to trigger them\nwhilst following \"..\".\n\nI\u0027ve also extracted the mount/don\u0027t-mount logic from autofs4 and included it\nhere. It makes the mount go ahead anyway if someone calls open() or creat(),\ntries to traverse the directory, tries to chdir/chroot/etc. into the directory,\nor sticks a \u0027/\u0027 on the end of the pathname. If they do a stat(), however,\nthey\u0027ll only trigger the automount if they didn\u0027t also say O_NOFOLLOW.\n\nI\u0027ve also added an inode flag (S_AUTOMOUNT) so that filesystems can mark their\ninodes as automount points. This flag is automatically propagated to the\ndentry as DCACHE_NEED_AUTOMOUNT by __d_instantiate(). This saves NFS and could\nsave AFS a private flag bit apiece, but is not strictly necessary. It would be\npreferable to do the propagation in d_set_d_op(), but that doesn\u0027t normally\nhave access to the inode.\n\n[AV: fixed breakage in case if __follow_mount_rcu() fails and nameidata_drop_rcu()\nsucceeds in RCU case of do_lookup(); we need to fall through to non-RCU case after\nthat, rather than just returning with ungrabbed *path]\n\nSigned-off-by: David Howells \u003cdhowells@redhat.com\u003e\nWas-Acked-by: Ian Kent \u003craven@themaw.net\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "2a8c0c68487a68441e701f493f43fd547d87c8df", "tree": "0731938d08b2d171aa40cff6c24f6183e06a9d01", "parents": [ "39191628ed169510db2f3f472e1ec14e08f9690f" ], "author": { "name": "Randy Dunlap", "email": "randy.dunlap@oracle.com", "time": "Sat Jan 08 19:37:20 2011 -0800" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@linux-foundation.org", "time": "Mon Jan 10 07:38:53 2011 -0800" }, "message": "fs: fix dcache.h kernel-doc notation\n\nFix new kernel-doc notation warning in dcache.h:\n\n Warning(include/linux/dcache.h:316): Excess function parameter \u0027Returns\u0027 description in \u0027__d_rcu_to_refcount\u0027\n\nSigned-off-by: Randy Dunlap \u003crandy.dunlap@oracle.com\u003e\nCc:\tNick Piggin \u003cnpiggin@kernel.dk\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\n" }, { "commit": "9d55c369bb5e695e629bc35cba2ef607755b3bee", "tree": "d97cd498a67f1f7987d4403ef90e399516a3b8c9", "parents": [ "e1bb57826381199cc79fbf44e9dfeee58fc7b339" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:09 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:32 2011 +1100" }, "message": "fs: implement faster dentry memcmp\n\nThe standard memcmp function on a Westmere system shows up hot in\nprofiles in the `git diff` workload (both parallel and single threaded),\nand it is likely due to the costs associated with trapping into\nmicrocode, and little opportunity to improve memory access (dentry\nname is not likely to take up more than a cacheline).\n\nSo replace it with an open-coded byte comparison. This increases code\nsize by 8 bytes in the critical __d_lookup_rcu function, but the\nspeedup is huge, averaging 10 runs of each:\n\ngit diff st user sys elapsed CPU\nbefore 1.15 2.57 3.82 97.1\nafter 1.14 2.35 3.61 96.8\n\ngit diff mt user sys elapsed CPU\nbefore 1.27 3.85 1.46 349\nafter 1.26 3.54 1.43 333\n\nElapsed time for single threaded git diff at 95.0% confidence:\n -0.21 +/- 0.01\n -5.45% +/- 0.24%\n\nIt\u0027s -0.66% +/- 0.06% elapsed time on my Opteron, so rep cmp costs on the\nfam10h seem to be relatively smaller, but there is still a win.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "4b936885ab04dc6e0bb0ef35e0e23c1a7364d9e5", "tree": "a0173d27c1ce39f173be404d269c2f15144072ab", "parents": [ "873feea09ebc980cbd3631b767356ce1eee65ec1" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:07 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:32 2011 +1100" }, "message": "fs: improve scalability of pseudo filesystems\n\nRegardless of how much we possibly try to scale dcache, there is likely\nalways going to be some fundamental contention when adding or removing children\nunder the same parent. Pseudo filesystems do not seem need to have connected\ndentries because by definition they are disconnected.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "873feea09ebc980cbd3631b767356ce1eee65ec1", "tree": "59a8fce9b138086abee7cf845f62ff70a390cf81", "parents": [ "ceb5bdc2d246f6d81cf61ed70f325308a11821d2" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:06 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:31 2011 +1100" }, "message": "fs: dcache per-inode inode alias locking\n\ndcache_inode_lock can be replaced with per-inode locking. Use existing\ninode-\u003ei_lock for this. This is slightly non-trivial because we sometimes\nneed to find the inode from the dentry, which requires d_inode to be\nstabilised (either with refcount or d_lock).\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "ceb5bdc2d246f6d81cf61ed70f325308a11821d2", "tree": "65fc9e0227994d4ffc80530dd15bb5a9672a295e", "parents": [ "626d607435617cc0f033522083e2bb195b81813c" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:05 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:31 2011 +1100" }, "message": "fs: dcache per-bucket dcache hash locking\n\nWe can turn the dcache hash locking from a global dcache_hash_lock into\nper-bucket locking.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "34286d6662308d82aed891852d04c7c3a2649b16", "tree": "c4b7311404d302e7cb94df7a4690298e1059910a", "parents": [ "44a7d7a878c9cbb74f236ea755b25b6b2e26a9a9" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:57 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:29 2011 +1100" }, "message": "fs: rcu-walk aware d_revalidate method\n\nRequire filesystems be aware of .d_revalidate being called in rcu-walk\nmode (nd-\u003eflags \u0026 LOOKUP_RCU). For now do a simple push down, returning\n-ECHILD from all implementations.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "44a7d7a878c9cbb74f236ea755b25b6b2e26a9a9", "tree": "d4630a38c0d683a7e1b8823d7971753719b8a54d", "parents": [ "fb045adb99d9b7c562dc7fef834857f78249daa1" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:56 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:28 2011 +1100" }, "message": "fs: cache optimise dentry and inode for rcu-walk\n\nPut dentry and inode fields into top of data structure. This allows RCU path\ntraversal to perform an RCU dentry lookup in a path walk by touching only the\nfirst 56 bytes of the dentry.\n\nWe also fit in 8 bytes of inline name in the first 64 bytes, so for short\nnames, only 64 bytes needs to be touched to perform the lookup. We should\nget rid of the hash-\u003eprev pointer from the first 64 bytes, and fit 16 bytes\nof name in there, which will take care of 81% rather than 32% of the kernel\ntree.\n\ninode is also rearranged so that RCU lookup will only touch a single cacheline\nin the inode, plus one in the i_ops structure.\n\nThis is important for directory component lookups in RCU path walking. In the\nkernel source, directory names average is around 6 chars, so this works.\n\nWhen we reach the last element of the lookup, we need to lock it and take its\nrefcount which requires another cacheline access.\n\nAlign dentry and inode operations structs, so members will be at predictable\noffsets and we can group common operations into head of structure.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "fb045adb99d9b7c562dc7fef834857f78249daa1", "tree": "1fd6a4024fffeec568abe100d730589bfdb81c38", "parents": [ "5f57cbcc02cf18f6b22ef4066bb10afeb8f930ff" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:55 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:28 2011 +1100" }, "message": "fs: dcache reduce branches in lookup path\n\nReduce some branches and memory accesses in dcache lookup by adding dentry\nflags to indicate common d_ops are set, rather than having to check them.\nThis saves a pointer memory access (dentry-\u003ed_op) in common path lookup\nsituations, and saves another pointer load and branch in cases where we\nhave d_op but not the particular operation.\n\nPatched with:\n\ngit grep -E \u0027[.\u003e]([[:space:]])*d_op([[:space:]])*\u003d\u0027 | xargs sed -e \u0027s/\\([^\\t ]*\\)-\u003ed_op \u003d \\(.*\\);/d_set_d_op(\\1, \\2);/\u0027 -e \u0027s/\\([^\\t ]*\\)\\.d_op \u003d \\(.*\\);/d_set_d_op(\\\u0026\\1, \\2);/\u0027 -i\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "5f57cbcc02cf18f6b22ef4066bb10afeb8f930ff", "tree": "f02e7ee57e6060f0af1bcda281baf2972d2da72f", "parents": [ "c28cc36469554dc55540f059fbdc7fa22a2c31fc" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:54 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:28 2011 +1100" }, "message": "fs: dcache remove d_mounted\n\nRather than keep a d_mounted count in the dentry, set a dentry flag instead.\nThe flag can be cleared by checking the hash table to see if there are any\nmounts left, which is not time critical because it is performed at detach time.\n\nThe mounted state of a dentry is only used to speculatively take a look in the\nmount hash table if it is set -- before following the mount, vfsmount lock is\ntaken and mount re-checked without races.\n\nThis saves 4 bytes on 32-bit, nothing on 64-bit but it does provide a hole I\nmight use later (and some configs have larger than 32-bit spinlocks which might\nmake use of the hole).\n\nAutofs4 conversion and changelog by Ian Kent \u003craven@themaw.net\u003e:\nIn autofs4, when expring direct (or offset) mounts we need to ensure that we\nblock user path walks into the autofs mount, which is covered by another mount.\nTo do this we clear the mounted status so that follows stop before walking into\nthe mount and are essentially blocked until the expire is completed. The\nautomount daemon still finds the correct dentry for the umount due to the\nfollow mount logic in fs/autofs4/root.c:autofs4_follow_link(), which is set as\nan inode operation for direct and offset mounts only and is called following\nthe lookup that stopped at the covered mount.\n\nAt the end of the expire the covering mount probably has gone away so the\nmounted status need not be restored. But we need to check this and only restore\nthe mounted status if the expire failed.\n\nXXX: autofs may not work right if we have other mounts go over the top of it?\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "31e6b01f4183ff419a6d1f86177cbf4662347cec", "tree": "e215ec9af88352c55e024f784f3d9f8eb13fab85", "parents": [ "3c22cd5709e8143444a6d08682a87f4c57902df3" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:52 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:27 2011 +1100" }, "message": "fs: rcu-walk for path lookup\n\nPerform common cases of path lookups without any stores or locking in the\nancestor dentry elements. This is called rcu-walk, as opposed to the current\nalgorithm which is a refcount based walk, or ref-walk.\n\nThis results in far fewer atomic operations on every path element,\nsignificantly improving path lookup performance. It also avoids cacheline\nbouncing on common dentries, significantly improving scalability.\n\nThe overall design is like this:\n* LOOKUP_RCU is set in nd-\u003eflags, which distinguishes rcu-walk from ref-walk.\n* Take the RCU lock for the entire path walk, starting with the acquiring\n of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are\n not required for dentry persistence.\n* synchronize_rcu is called when unregistering a filesystem, so we can\n access d_ops and i_ops during rcu-walk.\n* Similarly take the vfsmount lock for the entire path walk. So now mnt\n refcounts are not required for persistence. Also we are free to perform mount\n lookups, and to assume dentry mount points and mount roots are stable up and\n down the path.\n* Have a per-dentry seqlock to protect the dentry name, parent, and inode,\n so we can load this tuple atomically, and also check whether any of its\n members have changed.\n* Dentry lookups (based on parent, candidate string tuple) recheck the parent\n sequence after the child is found in case anything changed in the parent\n during the path walk.\n* inode is also RCU protected so we can load d_inode and use the inode for\n limited things.\n* i_mode, i_uid, i_gid can be tested for exec permissions during path walk.\n* i_op can be loaded.\n\nWhen we reach the destination dentry, we lock it, recheck lookup sequence,\nand increment its refcount and mountpoint refcount. RCU and vfsmount locks\nare dropped. This is termed \"dropping rcu-walk\". If the dentry refcount does\nnot match, we can not drop rcu-walk gracefully at the current point in the\nlokup, so instead return -ECHILD (for want of a better errno). This signals the\npath walking code to re-do the entire lookup with a ref-walk.\n\nAside from the final dentry, there are other situations that may be encounted\nwhere we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take\na reference on the last good dentry) and continue with a ref-walk. Again, if\nwe can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup\nusing ref-walk. But it is very important that we can continue with ref-walk\nfor most cases, particularly to avoid the overhead of double lookups, and to\ngain the scalability advantages on common path elements (like cwd and root).\n\nThe cases where rcu-walk cannot continue are:\n* NULL dentry (ie. any uncached path element)\n* parent with d_inode-\u003ei_op-\u003epermission or ACLs\n* dentries with d_revalidate\n* Following links\n\nIn future patches, permission checks and d_revalidate become rcu-walk aware. It\nmay be possible eventually to make following links rcu-walk aware.\n\nUncached path elements will always require dropping to ref-walk mode, at the\nvery least because i_mutex needs to be grabbed, and objects allocated.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "dc0474be3e27463d4d4a2793f82366eed906f223", "tree": "41f75e638442cb343bacdcfbabb17ffc3bd5b4ce", "parents": [ "357f8e658bba8a085c4a5d4331e30894be8096b8" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:43 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:24 2011 +1100" }, "message": "fs: dcache rationalise dget variants\n\ndget_locked was a shortcut to avoid the lazy lru manipulation when we already\nheld dcache_lock (lru manipulation was relatively cheap at that point).\nHowever, how that the lru lock is an innermost one, we never hold it at any\ncaller, so the lock cost can now be avoided. We already have well working lazy\ndcache LRU, so it should be fine to defer LRU manipulations to scan time.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "b5c84bf6f6fa3a7dfdcb556023a62953574b60ee", "tree": "7a2c299a180713e21d5cb653cb933121adf53c31", "parents": [ "949854d02455080d20cd3e1db28a3a18daf7599d" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:38 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:23 2011 +1100" }, "message": "fs: dcache remove dcache_lock\n\ndcache_lock no longer protects anything. remove it.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "949854d02455080d20cd3e1db28a3a18daf7599d", "tree": "9b13a6f86c1d0b91e462a471e53b0e717036b18e", "parents": [ "9abca36087288fe28de4749c71ca003d4b9e3ed0" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:37 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:22 2011 +1100" }, "message": "fs: Use rename lock and RCU for multi-step operations\n\nThe remaining usages for dcache_lock is to allow atomic, multi-step read-side\noperations over the directory tree by excluding modifications to the tree.\nAlso, to walk in the leaf-\u003eroot direction in the tree where we don\u0027t have\na natural d_lock ordering.\n\nThis could be accomplished by taking every d_lock, but this would mean a\nhuge number of locks and actually gets very tricky.\n\nSolve this instead by using the rename seqlock for multi-step read-side\noperations, retry in case of a rename so we don\u0027t walk up the wrong parent.\nConcurrent dentry insertions are not serialised against. Concurrent deletes\nare tricky when walking up the directory: our parent might have been deleted\nwhen dropping locks so also need to check and retry for that.\n\nWe can also use the rename lock in cases where livelock is a worry (and it\nis introduced in subsequent patch).\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "b23fb0a60379a95e10c671f646b259ea2558421e", "tree": "7c3644b91241d32fda502a7be0b78e4c225f8091", "parents": [ "2fd6b7f50797f2e993eea59e0a0b8c6399c811dc" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:35 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:22 2011 +1100" }, "message": "fs: scale inode alias list\n\nAdd a new lock, dcache_inode_lock, to protect the inode\u0027s i_dentry list\nfrom concurrent modification. d_alias is also protected by d_lock.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "2fd6b7f50797f2e993eea59e0a0b8c6399c811dc", "tree": "ce33b94b34844c09103836cf4cfa4364b742f217", "parents": [ "da5029563a0a026c64821b09e8e7b4fd81d3fe1b" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:34 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:21 2011 +1100" }, "message": "fs: dcache scale subdirs\n\nProtect d_subdirs and d_child with d_lock, except in filesystems that aren\u0027t\nusing dcache_lock for these anyway (eg. using i_mutex).\n\nNote: if we change the locking rule in future so that -\u003ed_child protection is\nprovided only with -\u003ed_parent-\u003ed_lock, it may allow us to reduce some locking.\nBut it would be an exception to an otherwise regular locking scheme, so we\u0027d\nhave to see some good results. Probably not worthwhile.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "b7ab39f631f505edc2bbdb86620d5493f995c9da", "tree": "62be97ebc7fc69ceb601f23312d335ebb8038ee7", "parents": [ "2304450783dfde7b0b94ae234edd0dbffa865073" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:32 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:21 2011 +1100" }, "message": "fs: dcache scale dentry refcount\n\nMake d_count non-atomic and protect it with d_lock. This allows us to ensure a\n0 refcount dentry remains 0 without dcache_lock. It is also fairly natural when\nwe start protecting many other dentry members with d_lock.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "789680d1ee9311cdf095241dc02bd9784d799cd1", "tree": "e6a984b0aa4791918f1b665f45210c2ab762969c", "parents": [ "ec2447c278ee973d35f38e53ca16ba7f965ae33d" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:30 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:20 2011 +1100" }, "message": "fs: dcache scale hash\n\nAdd a new lock, dcache_hash_lock, to protect the dcache hash table from\nconcurrent modification. d_hash is also protected by d_lock.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "ec2447c278ee973d35f38e53ca16ba7f965ae33d", "tree": "5d17a0523c301b8c71c0f198ffe7782c5e9c0ea9", "parents": [ "b1e6a015a580ad145689ad1d6b4aa0e03e6c868b" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:29 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:20 2011 +1100" }, "message": "hostfs: simplify locking\n\nRemove dcache_lock locking from hostfs filesystem, and move it into dcache\nhelpers. All that is required is a coherent path name. Protection from\nconcurrent modification of the namespace after path name generation is not\nprovided in current code, because dcache_lock is dropped before the path is\nused.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "b1e6a015a580ad145689ad1d6b4aa0e03e6c868b", "tree": "57a10ef164e4d2f798d9b832dbeaf973aca2ab83", "parents": [ "621e155a3591962420eacdd39f6f0aa29ceb221e" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:28 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:20 2011 +1100" }, "message": "fs: change d_hash for rcu-walk\n\nChange d_hash so it may be called from lock-free RCU lookups. See similar\npatch for d_compare for details.\n\nFor in-tree filesystems, this is just a mechanical change.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "621e155a3591962420eacdd39f6f0aa29ceb221e", "tree": "387a9fb396f1bf24514b712c294182e36ba51076", "parents": [ "fb2d5b86aff355a27ebfc132d3c99f4a940cc3fe" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:27 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:19 2011 +1100" }, "message": "fs: change d_compare for rcu-walk\n\nChange d_compare so it may be called from lock-free RCU lookups. This\ndoes put significant restrictions on what may be done from the callback,\nhowever there don\u0027t seem to have been any problems with in-tree fses.\nIf some strange use case pops up that _really_ cannot cope with the\nrcu-walk rules, we can just add new rcu-unaware callbacks, which would\ncause name lookup to drop out of rcu-walk mode.\n\nFor in-tree filesystems, this is just a mechanical change.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "fb2d5b86aff355a27ebfc132d3c99f4a940cc3fe", "tree": "7fed12adf54473131e8b86c0c302c443b1d6a846", "parents": [ "2bc334dcc7c77be3700dd443d92a78603f76976b" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:26 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:19 2011 +1100" }, "message": "fs: name case update method\n\nsmpfs and ncpfs want to update a live dentry name in-place. Rather than\nhave them open code the locking, provide a documented dcache API.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "fe15ce446beb3a33583af81ffe6c9d01a75314ed", "tree": "bc8af66b6dd2d0f21a2a3f48a19975ae2cdbae4e", "parents": [ "5eef7fa905c814826f518aca2d414ca77508ce30" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:23 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:18 2011 +1100" }, "message": "fs: change d_delete semantics\n\nChange d_delete from a dentry deletion notification to a dentry caching\nadvise, more like -\u003edrop_inode. Require it to be constant and idempotent,\nand not take d_lock. This is how all existing filesystems use the callback\nanyway.\n\nThis makes fine grained dentry locking of dput and dentry lru scanning\nmuch simpler.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "5eef7fa905c814826f518aca2d414ca77508ce30", "tree": "53c40fdbe92c30910befbad14ebace3ab9a342c0", "parents": [ "fbc8d4c04626e015b18cc61199f505920abb48f0" ], "author": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:49:22 2011 +1100" }, "committer": { "name": "Nick Piggin", "email": "npiggin@kernel.dk", "time": "Fri Jan 07 17:50:18 2011 +1100" }, "message": "fs: dcache documentation cleanup\n\nRemove redundant (and incorrect, since dcache RCU lookup) dentry locking\ndocumentation and point to the canonical document.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@kernel.dk\u003e\n" }, { "commit": "8df9d1a4142311c084ffeeacb67cd34d190eff74", "tree": "512e018114ea506659fac73d838bfca0fb97a5a4", "parents": [ "ffd1f4ed5bddccf2277e3d8525bcedf1983319f8" ], "author": { "name": "Miklos Szeredi", "email": "mszeredi@suse.cz", "time": "Tue Aug 10 11:41:41 2010 +0200" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Wed Aug 11 00:29:47 2010 -0400" }, "message": "vfs: show unreachable paths in getcwd and proc\n\nPrepend \"(unreachable)\" to path strings if the path is not reachable\nfrom the current root.\n\nTwo places updated are\n - the return string from getcwd()\n - and symlinks under /proc/$PID.\n\nOther uses of d_path() are left unchanged (we know that some old\nsoftware crashes if /proc/mounts is changed).\n\nSigned-off-by: Miklos Szeredi \u003cmszeredi@suse.cz\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "c103135c14e03fc9a9e5f0adc01df9ad272cf2a1", "tree": "0c75bd6a56385697cd0e79fb695a833c5225e45e", "parents": [ "d0352d3ed722b134dacc21836c1763e7e3523662" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sun Jun 06 22:31:14 2010 -0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Mon Aug 09 16:48:13 2010 -0400" }, "message": "new helper: __dentry_path()\n\nbuilds path relative to fs root, called under dcache_lock,\ndoesn\u0027t append any nonsense to unlinked ones.\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "d83c49f3e36cecd2e8823b6c48ffba083b8a5704", "tree": "a304de4eb43652d2a9528d4b43f798ab821c8b93", "parents": [ "6a251b0ab67989f468f4cb65179e0cf40cf8c295" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Fri Apr 30 17:17:09 2010 -0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sat May 15 07:16:33 2010 -0400" }, "message": "Fix the regression created by \"set S_DEAD on unlink()...\" commit\n\n1) i_flags simply doesn\u0027t work for mount/unlink race prevention;\nwe may have many links to file and rm on one of those obviously\nshouldn\u0027t prevent bind on top of another later on. To fix it\nright way we need to mark _dentry_ as unsuitable for mounting\nupon; new flag (DCACHE_CANT_MOUNT) is protected by d_flags and\ni_mutex on the inode in question. Set it (with dont_mount(dentry))\nin unlink/rmdir/etc., check (with cant_mount(dentry)) in places\nin namespace.c that used to check for S_DEAD. Setting S_DEAD\nis still needed in places where we used to set it (for directories\ngetting killed), since we rely on it for readdir/rmdir race\nprevention.\n\n2) rename()/mount() protection has another bogosity - we unhash\nthe target before we\u0027d checked that it\u0027s not a mountpoint. Fixed.\n\n3) ancient bogosity in pivot_root() - we locked i_mutex on the\nright directory, but checked S_DEAD on the different (and wrong)\none. Noticed and fixed.\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "f3da392e9ff14b9f388e74319e6d195848991c07", "tree": "d9e7660e5c0b4524aeaafcc5b89a9d68a7ca3817", "parents": [ "8c85e125124a473d6f3e9bb187b0b84207f81d91" ], "author": { "name": "Alexey Dobriyan", "email": "adobriyan@gmail.com", "time": "Mon May 04 03:32:03 2009 +0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Thu Jun 11 21:36:06 2009 -0400" }, "message": "dcache: extrace and use d_unlinked()\n\nd_unlinked() will be used in middle-term to ban checkpointing when opened\nbut unlinked file is detected, and in long term, to detect such situation\nand special case on it.\n\nSigned-off-by: Alexey Dobriyan \u003cadobriyan@gmail.com\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "1c755af4df75996b0dd4b7e6cacaf9d57a6ef2ef", "tree": "0a75e35164912f53b21ca48b829cabdfa6ea507d", "parents": [ "79ed0226198c628133530b179a90dbf42b1c2eba" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sat Apr 18 14:06:57 2009 -0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Thu Jun 11 21:36:01 2009 -0400" }, "message": "switch lookup_mnt()\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "c28f7e56e9d95fb531dc3be8df2e7f52bee76d21", "tree": "efce13ed8c4f5b312ef0b77950489d52ef5a039a", "parents": [ "3be25f49b9d6a97eae9bcb96d3292072b7658bd8" ], "author": { "name": "Eric Paris", "email": "eparis@redhat.com", "time": "Thu May 21 17:01:29 2009 -0400" }, "committer": { "name": "Eric Paris", "email": "eparis@redhat.com", "time": "Thu Jun 11 14:57:53 2009 -0400" }, "message": "fsnotify: parent event notification\n\ninotify and dnotify both use a similar parent notification mechanism. We\nadd a generic parent notification mechanism to fsnotify for both of these\nto use. This new machanism also adds the dentry flag optimization which\nexists for inotify to dnotify.\n\nSigned-off-by: Eric Paris \u003ceparis@redhat.com\u003e\nAcked-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\nCc: Christoph Hellwig \u003chch@lst.de\u003e\n" }, { "commit": "e56980d451904b623573ef4966cbab768e433c79", "tree": "2345c1fa43785095bf4e8224a2c90af0d1998f42", "parents": [ "a9f184f02aa49d46c4c35311d93cbcd1c61149df" ], "author": { "name": "Jan Engelhardt", "email": "jengelh@medozas.de", "time": "Wed Feb 11 13:14:54 2009 -0800" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Fri Mar 27 14:43:59 2009 -0400" }, "message": "fs: make struct dentry-\u003ed_op const\n\nThis change will allow for tagging many dentry_operations const in the\nsource tree.\n\nSigned-off-by: Jan Engelhardt \u003cjengelh@medozas.de\u003e\nCc: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "c2452f32786159ed85f0e4b21fec09258f822fc8", "tree": "50d93df47f4547a5699c87a608e85596e4c6165f", "parents": [ "e2b689d82c0394e5239a3557a217f19e2f47f1be" ], "author": { "name": "Nick Piggin", "email": "npiggin@suse.de", "time": "Mon Dec 01 09:33:43 2008 +0100" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Wed Dec 31 18:07:38 2008 -0500" }, "message": "shrink struct dentry\n\nstruct dentry is one of the most critical structures in the kernel. So it\u0027s\nsad to see it going neglected.\n\nWith CONFIG_PROFILING turned on (which is probably the common case at least\nfor distros and kernel developers), sizeof(struct dcache) \u003d\u003d 208 here\n(64-bit). This gives 19 objects per slab.\n\nI packed d_mounted into a hole, and took another 4 bytes off the inline\nname length to take the padding out from the end of the structure. This\nshinks it to 200 bytes. I could have gone the other way and increased the\nlength to 40, but I\u0027m aiming for a magic number, read on...\n\nI then got rid of the d_cookie pointer. This shrinks it to 192 bytes. Rant:\nwhy was this ever a good idea? The cookie system should increase its hash\nsize or use a tree or something if lookups are a problem. Also the \"fast\ndcookie lookups\" in oprofile should be moved into the dcookie code -- how\ncan oprofile possibly care about the dcookie_mutex? It gets dropped after\nget_dcookie() returns so it can\u0027t be providing any sort of protection.\n\nAt 192 bytes, 21 objects fit into a 4K page, saving about 3MB on my system\nwith ~140 000 entries allocated. 192 is also a multiple of 64, so we get\nnice cacheline alignment on 64 and 32 byte line systems -- any given dentry\nwill now require 3 cachelines to touch all fields wheras previously it\nwould require 4.\n\nI know the inline name size was chosen quite carefully, however with the\nreduction in cacheline footprint, it should actually be just about as fast\nto do a name lookup for a 36 character name as it was before the patch (and\nfaster for other sizes). The memory footprint savings for names which are\n\u003c\u003d 32 or \u003e 36 bytes long should more than make up for the memory cost for\n33-36 byte names.\n\nPerformance is a feature...\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "e2761a1167633ed943fea29002f990194923d060", "tree": "e91816e85c7721fcff72e5f84336b1dde9d1ac17", "parents": [ "871c0067d53ba2dc35897c7da1da675bf4c70511" ], "author": { "name": "OGAWA Hirofumi", "email": "hirofumi@mail.parknet.co.jp", "time": "Thu Oct 16 07:50:28 2008 +0900" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Thu Oct 23 05:13:16 2008 -0400" }, "message": "[PATCH vfs-2.6 2/6] vfs: add d_ancestor()\n\nThis adds d_ancestor() instead of d_isparent(), then use it.\n\nIf new_dentry \u003d\u003d old_dentry, is_subdir() returns 1, looks strange.\n\"new_dentry \u003d\u003d old_dentry\" is not subdir obviously. But I\u0027m not\nchecking callers for now, so this keeps current behavior.\n\nSigned-off-by: OGAWA Hirofumi \u003chirofumi@mail.parknet.co.jp\u003e\n" }, { "commit": "9308a6128d9074e348d9f9b5822546fe12a794a9", "tree": "3b67f8b6381162b9ff0800eee8722081ef9ed039", "parents": [ "440037287c5ebb07033ab927ca16bb68c291d309" ], "author": { "name": "Christoph Hellwig", "email": "hch@lst.de", "time": "Mon Aug 11 15:49:12 2008 +0200" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Thu Oct 23 05:13:02 2008 -0400" }, "message": "[PATCH] kill d_alloc_anon\n\nRemove d_alloc_anon now that no users are left.\n\nSigned-off-by: Christoph Hellwig \u003chch@lst.de\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "4ea3ada2955e4519befa98ff55dd62d6dfbd1705", "tree": "b473278d0df8d869d183fbc0e4a1a07585de240a", "parents": [ "3a8cff4f026c0b98bee6291eb28d4df42feb76dc" ], "author": { "name": "Christoph Hellwig", "email": "hch@lst.de", "time": "Mon Aug 11 15:48:57 2008 +0200" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Thu Oct 23 05:13:00 2008 -0400" }, "message": "[PATCH] new helper: d_obtain_alias\n\nThe calling conventions of d_alloc_anon are rather unfortunate for all\nusers, and it\u0027s name is not very descriptive either.\n\nAdd d_obtain_alias as a new exported helper that drops the inode\nreference in the failure case, too and allows to pass-through NULL\npointers and inodes to allow for tail-calls in the export operations.\n\nIncidentally this helper already existed as a private function in\nlibfs.c as exportfs_d_alloc so kill that one and switch the callers\nto d_obtain_alias.\n\nSigned-off-by: Christoph Hellwig \u003chch@lst.de\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "e45b590b976465c258f3e2a6cc84573fc19e16d3", "tree": "54b795a1acd985b9bcc5adbe057706209a2612d3", "parents": [ "2d8a10cd1760e7ecc07a21e409485947c68a3291" ], "author": { "name": "Christoph Hellwig", "email": "hch@lst.de", "time": "Thu Aug 07 23:49:07 2008 +0200" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Mon Aug 25 01:18:05 2008 -0400" }, "message": "[PATCH] change d_add_ci argument ordering\n\nAs pointed out during review d_add_ci argument order should match d_add,\nso switch the dentry and inode arguments.\n\nSigned-off-by: Christoph Hellwig \u003chch@lst.de\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "9403540c0653122ca34884a180439ddbfcbcb524", "tree": "40b5fb192900c61a1c91be6ae8ea7432b61f18aa", "parents": [ "6a178100abf01282eb697ab62b6086b2886dfc00" ], "author": { "name": "Barry Naujok", "email": "bnaujok@sgi.com", "time": "Wed May 21 16:50:46 2008 +1000" }, "committer": { "name": "Niv Sardi", "email": "xaiki@debian.org", "time": "Mon Jul 28 16:58:39 2008 +1000" }, "message": "dcache: Add case-insensitive support d_ci_add() routine\n\nThis add a dcache entry to the dcache for lookup, but changing the name\nthat is associated with the entry rather than the one passed in to the\nlookup routine.\n\nFirst, it sees if the case-exact match already exists in the dcache and\nuses it if one exists. Otherwise, it allocates a new node with the new\nname and splices it into the dcache.\n\nOriginal code from ntfs_lookup in fs/ntfs/namei.c by Anton Altaparmakov.\n\nSigned-off-by: Barry Naujok \u003cbnaujok@sgi.com\u003e\nSigned-off-by: Anton Altaparmakov \u003caia21@cantab.net\u003e\nAcked-by: Christoph Hellwig \u003chch@infradead.org\u003e\n" }, { "commit": "0c81b2a1448bc6a2a9b2d6469fb0669fb4b25e5b", "tree": "6f82579cae6d6e39fa9f837a3c349ded51e19d14", "parents": [ "0729fbf3bc70870370b4f43d652f05a468dc68b8", "70ff05554f91a1edda1f11684da1dbde09e2feea" ], "author": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Fri Jul 11 10:46:50 2008 +0200" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Fri Jul 11 10:46:50 2008 +0200" }, "message": "Merge branch \u0027linus\u0027 into core/rcu\n\nConflicts:\n\n\tinclude/linux/rculist.h\n\tkernel/rcupreempt.c\n\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" }, { "commit": "20d4fdc1a788e4ca0aaf2422772ba668e7e10839", "tree": "cd75fce48616b364ddb968d2085205fbc9f4aca1", "parents": [ "c8e7f449b225ee6c87454ac069f0a041035c5140" ], "author": { "name": "Jan Engelhardt", "email": "jengelh@medozas.de", "time": "Mon Jun 09 16:40:36 2008 -0700" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Mon Jun 23 11:52:30 2008 -0400" }, "message": "[patch 2/4] fs: make struct file arg to d_path const\n\nSigned-off-by: Jan Engelhardt \u003cjengelh@medozas.de\u003e\nCc: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "82524746c27fa418c250a56dd7606b9d3fc79826", "tree": "1801230b8fc2e436e722ac6f54fc53f1c112c310", "parents": [ "32300751b4079cb5688453baa94711579d4285d5" ], "author": { "name": "Franck Bui-Huu", "email": "fbuihuu@gmail.com", "time": "Mon May 12 21:21:05 2008 +0200" }, "committer": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Mon May 19 10:01:37 2008 +0200" }, "message": "rcu: split list.h and move rcu-protected lists into rculist.h\n\nMove rcu-protected lists from list.h into a new header file rculist.h.\n\nThis is done because list are a very used primitive structure all over the\nkernel and it\u0027s currently impossible to include other header files in this\nlist.h without creating some circular dependencies.\n\nFor example, list.h implements rcu-protected list and uses rcu_dereference()\nwithout including rcupdate.h. It actually compiles because users of\nrcu_dereference() are macros. Others RCU functions could be used too but\naren\u0027t probably because of this.\n\nTherefore this patch creates rculist.h which includes rcupdates without to\nmany changes/troubles.\n\nSigned-off-by: Franck Bui-Huu \u003cfbuihuu@gmail.com\u003e\nAcked-by: Paul E. McKenney \u003cpaulmck@linux.vnet.ibm.com\u003e\nAcked-by: Josh Triplett \u003cjosh@kernel.org\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\n" }, { "commit": "735643ee6cc5249bfac07fcad0946a5e7aff4423", "tree": "e725df246f4a3cf88b6b42a28d859ab969acf81c", "parents": [ "71cc2c2152170b8166f59abb0604dc62073aeb92" ], "author": { "name": "Robert P. J. Day", "email": "rpjday@crashcourse.ca", "time": "Wed Apr 30 00:55:12 2008 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@linux-foundation.org", "time": "Wed Apr 30 08:29:54 2008 -0700" }, "message": "Remove \"#ifdef __KERNEL__\" checks from unexported headers\n\nRemove the \"#ifdef __KERNEL__\" tests from unexported header files in\nlinux/include whose entire contents are wrapped in that preprocessor\ntest.\n\nSigned-off-by: Robert P. J. Day \u003crpjday@crashcourse.ca\u003e\nCc: David Woodhouse \u003cdwmw2@infradead.org\u003e\nCc: Sam Ravnborg \u003csam@ravnborg.org\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\n" }, { "commit": "9d1bc60138977d9c79471b344a64f2df13b2ccef", "tree": "84ed2e1920a1a88ae84515bd9d90240faed5805c", "parents": [ "6092d048183b76bfa3f84b32f8158dd8d10bd811" ], "author": { "name": "Miklos Szeredi", "email": "mszeredi@suse.cz", "time": "Thu Mar 27 13:06:21 2008 +0100" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Wed Apr 23 00:04:38 2008 -0400" }, "message": "[patch 2/7] vfs: mountinfo: add seq_file_root()\n\nAdd a new function:\n\n seq_file_root()\n\nThis is similar to seq_path(), but calculates the path relative to the\ngiven root, instead of current-\u003efs-\u003eroot. If the path was unreachable\nfrom root, then modify the root parameter to reflect this.\n\nSigned-off-by: Miklos Szeredi \u003cmszeredi@suse.cz\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "6092d048183b76bfa3f84b32f8158dd8d10bd811", "tree": "2d38af32961eb52b36daca2f29ce69e80f6e7283", "parents": [ "934b25c597c0e98304a7eaec198a87e4633a42bb" ], "author": { "name": "Ram Pai", "email": "linuxram@us.ibm.com", "time": "Thu Mar 27 13:06:20 2008 +0100" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Wed Apr 23 00:04:32 2008 -0400" }, "message": "[patch 1/7] vfs: mountinfo: add dentry_path()\n\n[mszeredi@suse.cz] split big patch into managable chunks\n\nAdd the following functions:\n\n dentry_path()\n seq_dentry()\n\nThese are similar to d_path() and seq_path(). But instead of\ncalculating the path within a mount namespace, they calculate the path\nfrom the root of the filesystem to a given dentry, ignoring mounts\ncompletely.\n\nSigned-off-by: Ram Pai \u003clinuxram@us.ibm.com\u003e\nSigned-off-by: Miklos Szeredi \u003cmszeredi@suse.cz\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "6d59e7f582ef1c1988542d0fc3b36d0087b757ce", "tree": "2e6e2e7d42af8f89f7b87aeaf5261d4cfbd9b625", "parents": [ "807501475fce0ebe68baedf87f202c3e4ee0d12c" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sat Mar 22 15:48:17 2008 -0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Mon Apr 21 23:11:01 2008 -0400" }, "message": "[PATCH] move a bunch of declarations to fs/internal.h\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "cf28b4863f9ee8f122e8ff3ac0d403e07ba9c6d9", "tree": "65c91f6911b34c32e517938289621ce0e7baeaf3", "parents": [ "c32c2f63a9d6c953aaf168c0b2551da9734f76d2" ], "author": { "name": "Jan Blunck", "email": "jblunck@suse.de", "time": "Thu Feb 14 19:38:44 2008 -0800" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@woody.linux-foundation.org", "time": "Thu Feb 14 21:17:09 2008 -0800" }, "message": "d_path: Make d_path() use a struct path\n\nd_path() is used on a \u003cdentry,vfsmount\u003e pair. Lets use a struct path to\nreflect this.\n\n[akpm@linux-foundation.org: fix build in mm/memory.c]\nSigned-off-by: Jan Blunck \u003cjblunck@suse.de\u003e\nAcked-by: Bryan Wu \u003cbryan.wu@analog.com\u003e\nAcked-by: Christoph Hellwig \u003chch@infradead.org\u003e\nCc: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\nCc: \"J. Bruce Fields\" \u003cbfields@fieldses.org\u003e\nCc: Neil Brown \u003cneilb@suse.de\u003e\nCc: Michael Halcrow \u003cmhalcrow@us.ibm.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\n" }, { "commit": "74c3cbe33bc077ac1159cadfea608b501e100344", "tree": "4c4023caa4e15d19780255fa5880df3d36eb292c", "parents": [ "455434d450a358ac5bcf3fc58f8913d13c544622" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sun Jul 22 08:04:18 2007 -0400" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Sun Oct 21 02:37:45 2007 -0400" }, "message": "[PATCH] audit: watching subtrees\n\nNew kind of audit rule predicates: \"object is visible in given subtree\".\nThe part that can be sanely implemented, that is. Limitations:\n\t* if you have hardlink from outside of tree, you\u0027d better watch\nit too (or just watch the object itself, obviously)\n\t* if you mount something under a watched tree, tell audit\nthat new chunk should be added to watched subtrees\n\t* if you umount something in a watched tree and it\u0027s still mounted\nelsewhere, you will get matches on events happening there. New command\ntells audit to recalculate the trees, trimming such sources of false\npositives.\n\nNote that it\u0027s _not_ about path - if something mounted in several places\n(multiple mount, bindings, different namespaces, etc.), the match does\n_not_ depend on which one we are using for access.\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "c23fbb6bcb3eb9cdf39a103edadf57bde8ce309c", "tree": "d79ab2278774de2c1a8061aa948ed068902e87b4", "parents": [ "2793274298c4423d79701e9a8190f2940bf3c785" ], "author": { "name": "Eric Dumazet", "email": "dada1@cosmosbay.com", "time": "Tue May 08 00:26:18 2007 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@woody.linux-foundation.org", "time": "Tue May 08 11:15:03 2007 -0700" }, "message": "VFS: delay the dentry name generation on sockets and pipes\n\n1) Introduces a new method in \u0027struct dentry_operations\u0027. This method\n called d_dname() might be called from d_path() to build a pathname for\n special filesystems. It is called without locks.\n\n Future patches (if we succeed in having one common dentry for all\n pipes/sockets) may need to change prototype of this method, but we now\n use : char *d_dname(struct dentry *dentry, char *buffer, int buflen);\n\n2) Adds a dynamic_dname() helper function that eases d_dname() implementations\n\n3) Defines d_dname method for sockets : No more sprintf() at socket\n creation. This is delayed up to the moment someone does an access to\n /proc/pid/fd/...\n\n4) Defines d_dname method for pipes : No more sprintf() at pipe\n creation. This is delayed up to the moment someone does an access to\n /proc/pid/fd/...\n\nA benchmark consisting of 1.000.000 calls to pipe()/close()/close() gives a\n*nice* speedup on my Pentium(M) 1.6 Ghz :\n\n3.090 s instead of 3.450 s\n\nSigned-off-by: Eric Dumazet \u003cdada1@cosmosbay.com\u003e\nAcked-by: Christoph Hellwig \u003chch@infradead.org\u003e\nAcked-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\n" }, { "commit": "c636ebdb186bf37f98d3839f69293597723edb36", "tree": "db1d54d355cd8030bf5d1645ce6bac8842f32690", "parents": [ "6ce315234aefcbc599dea390c15672156ebf9e7b" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Wed Oct 11 01:22:19 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Wed Oct 11 11:14:25 2006 -0700" }, "message": "[PATCH] VFS: Destroy the dentries contributed by a superblock on unmounting\n\nThe attached patch destroys all the dentries attached to a superblock in one go\nby:\n\n (1) Destroying the tree rooted at s_root.\n\n (2) Destroying every entry in the anon list, one at a time.\n\n (3) Each entry in the anon list has its subtree consumed from the leaves\n inwards.\n\nThis reduces the amount of work generic_shutdown_super() does, and avoids\niterating through the dentry_unused list.\n\nNote that locking is almost entirely absent in the shrink_dcache_for_umount*()\nfunctions added by this patch. This is because:\n\n (1) at the point the filesystem calls generic_shutdown_super(), it is not\n permitted to further touch the superblock\u0027s set of dentries, and nor may\n it remove aliases from inodes;\n\n (2) the dcache memory shrinker now skips dentries that are being unmounted;\n and\n\n (3) the superblock no longer has any external references through which the VFS\n can reach it.\n\nGiven these points, the only locking we need to do is when we remove dentries\nfrom the unused list and the name hashes, which we do a directory\u0027s worth at a\ntime.\n\nWe also don\u0027t need to guard against reference counts going to zero unexpectedly\nand removing bits of the tree we\u0027re working on as nothing else can call dput().\n\nA cut down version of dentry_iput() has been folded into\nshrink_dcache_for_umount_subtree() function. Apart from not needing to unlock\nthings, it also doesn\u0027t need to check for inotify watches.\n\nIn this version of the patch, the complaint about a dentry still being in use\nhas been expanded from a single BUG_ON() and now gives much more information.\n\nSigned-off-by: David Howells \u003cdhowells@redhat.com\u003e\nAcked-by: NeilBrown \u003cneilb@suse.de\u003e\nAcked-by: Ian Kent \u003craven@themaw.net\u003e\nCc: Trond Myklebust \u003ctrond.myklebust@fys.uio.no\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "770bfad846ab6628444428467b11fa6773ae9ea1", "tree": "5fd803319fe718367dca081b23abd36bfc85f78a", "parents": [ "979df72e6f963b42ee484f2eca049c3344da0ba7" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Tue Aug 22 20:06:07 2006 -0400" }, "committer": { "name": "Trond Myklebust", "email": "Trond.Myklebust@netapp.com", "time": "Fri Sep 22 23:24:30 2006 -0400" }, "message": "NFS: Add dentry materialisation op\n\nThe attached patch adds a new directory cache management function that prepares\na disconnected anonymous function to be connected into the dentry tree. The\nanonymous dentry is transferred the name and parentage from another dentry.\n\nThe following changes were made in [try #2]:\n\n (*) d_materialise_dentry() now switches the parentage of the two nodes around\n correctly when one or other of them is self-referential.\n\nThe following changes were made in [try #7]:\n\n (*) d_instantiate_unique() has had the interior part split out as function\n __d_instantiate_unique(). Callers of this latter function must be holding\n the appropriate locks.\n\n (*) _d_rehash() has been added as a wrapper around __d_rehash() to call it\n with the most obvious hash list (the one from the name). d_rehash() now\n calls _d_rehash().\n\n (*) d_materialise_dentry() is now __d_materialise_dentry() and is static.\n\n (*) d_materialise_unique() added to perform the combination of d_find_alias(),\n d_materialise_dentry() and d_add_unique() that the NFS client was doing\n twice, all within a single dcache_lock critical section. This reduces the\n number of times two different spinlocks were being accessed.\n\nThe following further changes were made:\n\n (*) Add the dentries onto their parents d_subdirs lists.\n\nSigned-Off-By: David Howells \u003cdhowells@redhat.com\u003e\nSigned-off-by: Trond Myklebust \u003cTrond.Myklebust@netapp.com\u003e\n" }, { "commit": "a90b9c05df3c1e58eaedc28795d0f5abd896c098", "tree": "bc9c1233bad94175abca731daa9120151e0d9c7e", "parents": [ "13e83599d282ddfd544600df9db5ab343ac4662f" ], "author": { "name": "Ingo Molnar", "email": "mingo@elte.hu", "time": "Mon Jul 03 00:25:04 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Mon Jul 03 15:27:06 2006 -0700" }, "message": "[PATCH] lockdep: annotate dcache\n\nTeach special (recursive) locking code to the lock validator. Has no effect\non non-lockdep kernels.\n\nSigned-off-by: Ingo Molnar \u003cmingo@elte.hu\u003e\nSigned-off-by: Arjan van de Ven \u003carjan@linux.intel.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "454e2398be9b9fa30433fccc548db34d19aa9958", "tree": "1f61cb0c3716a33b661cfc8977e9beeb480a322c", "parents": [ "1ad5544098a69d7dc1fa508cbb17e13a7a952fd8" ], "author": { "name": "David Howells", "email": "dhowells@redhat.com", "time": "Fri Jun 23 02:02:57 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Fri Jun 23 07:42:45 2006 -0700" }, "message": "[PATCH] VFS: Permit filesystem to override root dentry on mount\n\nExtend the get_sb() filesystem operation to take an extra argument that\npermits the VFS to pass in the target vfsmount that defines the mountpoint.\n\nThe filesystem is then required to manually set the superblock and root dentry\npointers. For most filesystems, this should be done with simple_set_mnt()\nwhich will set the superblock pointer and then set the root dentry to the\nsuperblock\u0027s s_root (as per the old default behaviour).\n\nThe get_sb() op now returns an integer as there\u0027s now no need to return the\nsuperblock pointer.\n\nThis patch permits a superblock to be implicitly shared amongst several mount\npoints, such as can be done with NFS to avoid potential inode aliasing. In\nsuch a case, simple_set_mnt() would not be called, and instead the mnt_root\nand mnt_sb would be set directly.\n\nThe patch also makes the following changes:\n\n (*) the get_sb_*() convenience functions in the core kernel now take a vfsmount\n pointer argument and return an integer, so most filesystems have to change\n very little.\n\n (*) If one of the convenience function is not used, then get_sb() should\n normally call simple_set_mnt() to instantiate the vfsmount. This will\n always return 0, and so can be tail-called from get_sb().\n\n (*) generic_shutdown_super() now calls shrink_dcache_sb() to clean up the\n dcache upon superblock destruction rather than shrink_dcache_anon().\n\n This is required because the superblock may now have multiple trees that\n aren\u0027t actually bound to s_root, but that still need to be cleaned up. The\n currently called functions assume that the whole tree is rooted at s_root,\n and that anonymous dentries are not the roots of trees which results in\n dentries being left unculled.\n\n However, with the way NFS superblock sharing are currently set to be\n implemented, these assumptions are violated: the root of the filesystem is\n simply a dummy dentry and inode (the real inode for \u0027/\u0027 may well be\n inaccessible), and all the vfsmounts are rooted on anonymous[*] dentries\n with child trees.\n\n [*] Anonymous until discovered from another tree.\n\n (*) The documentation has been adjusted, including the additional bit of\n changing ext2_* into foo_* in the documentation.\n\n[akpm@osdl.org: convert ipath_fs, do other stuff]\nSigned-off-by: David Howells \u003cdhowells@redhat.com\u003e\nAcked-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\nCc: Nathan Scott \u003cnathans@sgi.com\u003e\nCc: Roland Dreier \u003crolandd@cisco.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "0feae5c47aabdde59cbbec32d150e17102de37f0", "tree": "244f742d943a0516921180b840419fdc329075f0", "parents": [ "de047c1bcd7f7bcfbdc29eb5b439fb332594da3f" ], "author": { "name": "NeilBrown", "email": "neilb@suse.de", "time": "Thu Jun 22 14:47:28 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Thu Jun 22 15:05:57 2006 -0700" }, "message": "[PATCH] Fix dcache race during umount\n\nThe race is that the shrink_dcache_memory shrinker could get called while a\nfilesystem is being unmounted, and could try to prune a dentry belonging to\nthat filesystem.\n\nIf it does, then it will call in to iput on the inode while the dentry is\nno longer able to be found by the umounting process. If iput takes a\nwhile, generic_shutdown_super could get all the way though\nshrink_dcache_parent and shrink_dcache_anon and invalidate_inodes without\never waiting on this particular inode.\n\nEventually the superblock gets freed anyway and if the iput tried to touch\nit (which some filesystems certainly do), it will lose. The promised\n\"Self-destruct in 5 seconds\" doesn\u0027t lead to a nice day.\n\nThe race is closed by holding s_umount while calling prune_one_dentry on\nsomeone else\u0027s dentry. As a down_read_trylock is used,\nshrink_dcache_memory will no longer try to prune the dentry of a filesystem\nthat is being unmounted, and unmount will not be able to start until any\nsuch active prune_one_dentry completes.\n\nThis requires that prune_dcache *knows* which filesystem (if any) it is\ndoing the prune on behalf of so that it can be careful of other\nfilesystems. shrink_dcache_memory isn\u0027t called it on behalf of any\nfilesystem, and so is careful of everything.\n\nshrink_dcache_anon is now passed a super_block rather than the s_anon list\nout of the superblock, so it can get the s_anon list itself, and can pass\nthe superblock down to prune_dcache.\n\nIf prune_dcache finds a dentry that it cannot free, it leaves it where it\nis (at the tail of the list) and exits, on the assumption that some other\nthread will be removing that dentry soon. To try to make sure that some\nwork gets done, a limited number of dnetries which are untouchable are\nskipped over while choosing the dentry to work on.\n\nI believe this race was first found by Kirill Korotaev.\n\nCc: Jan Blunck \u003cjblunck@suse.de\u003e\nAcked-by: Kirill Korotaev \u003cdev@openvz.org\u003e\nCc: Olaf Hering \u003colh@suse.de\u003e\nAcked-by: Balbir Singh \u003cbalbir@in.ibm.com\u003e\nSigned-off-by: Neil Brown \u003cneilb@suse.de\u003e\nSigned-off-by: Balbir Singh \u003cbalbir@in.ibm.com\u003e\nAcked-by: David Howells \u003cdhowells@redhat.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "3e7e241f8c5c87cc3685364feface081c9fa3648", "tree": "43ea46cfec4da31a505e93875feeac945bf906ce", "parents": [ "92476d7fc0326a409ab1d3864a04093a6be9aca7" ], "author": { "name": "Eric W. Biederman", "email": "ebiederm@xmission.com", "time": "Fri Mar 31 02:31:43 2006 -0800" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Fri Mar 31 12:19:00 2006 -0800" }, "message": "[PATCH] dcache: Add helper d_hash_and_lookup\n\nIt is very common to hash a dentry and then to call lookup. If we take fs\nspecific hash functions into account the full hash logic can get ugly.\nFurther full_name_hash as an inline function is almost 100 bytes on x86 so\nhaving a non-inline choice in some cases can measurably decrease code size.\n\nSigned-off-by: Eric W. Biederman \u003cebiederm@xmission.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "c32ccd87bfd1414b0aabfcd8dbc7539ad23bcbaa", "tree": "612dc637976cbe36e8b72924a1f7bd76e75463fd", "parents": [ "bf36b9011e3c5b2739f9da2f6de8a6fa3edded32" ], "author": { "name": "Nick Piggin", "email": "nickpiggin@yahoo.com.au", "time": "Sat Mar 25 03:07:09 2006 -0800" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Sat Mar 25 08:22:53 2006 -0800" }, "message": "[PATCH] inotify: lock avoidance with parent watch status in dentry\n\nPrevious inotify work avoidance is good when inotify is completely unused,\nbut it breaks down if even a single watch is in place anywhere in the\nsystem. Robin Holt notices that udev is one such culprit - it slows down a\n512-thread application on a 512 CPU system from 6 seconds to 22 minutes.\n\nSolve this by adding a flag in the dentry that tells inotify whether or not\nits parent inode has a watch on it. Event queueing to parent will skip\ntaking locks if this flag is cleared. Setting and clearing of this flag on\nall child dentries versus event delivery: this is no in terms of race\ncases, and that was shown to be equivalent to always performing the check.\n\nThe essential behaviour is that activity occuring _after_ a watch has been\nadded and _before_ it has been removed, will generate events.\n\nSigned-off-by: Nick Piggin \u003cnpiggin@suse.de\u003e\nCc: Robert Love \u003crml@novell.com\u003e\nCc: John McCutchan \u003cttb@tentacle.dhs.org\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "1b8623545b42c03eb92e51b28c84acf4b8ba00a3", "tree": "071045ad9c60d2697292c523c77322a70a248fb9", "parents": [ "92118c739df879497b8cc5a2eb3a9dc255f01b20" ], "author": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Thu Dec 15 01:07:03 2005 -0500" }, "committer": { "name": "Al Viro", "email": "viro@zeniv.linux.org.uk", "time": "Tue Feb 07 20:56:35 2006 -0500" }, "message": "[PATCH] remove bogus asm/bug.h includes.\n\nA bunch of asm/bug.h includes are both not needed (since it will get\npulled anyway) and bogus (since they are done too early). Removed.\n\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\n" }, { "commit": "47ba87e0b1269698801310bfd1716b0538282405", "tree": "be429e822401e9f1a213b008b5e6e2f538754888", "parents": [ "09114eb8c53d2d3b2ff9523e011cb68b2e245dce" ], "author": { "name": "Marcelo Tosatti", "email": "marcelo.tosatti@cyclades.com", "time": "Fri Feb 03 03:04:06 2006 -0800" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Fri Feb 03 08:32:04 2006 -0800" }, "message": "[PATCH] make \"struct d_cookie\" depend on CONFIG_PROFILING\n\nShrinks \"struct dentry\" from 128 bytes to 124 on x86, allowing 31 objects\nper slab instead of 30.\n\nCc: John Levon \u003clevon@movementarian.org\u003e\nCc: Philippe Elie \u003cphil.el@wanadoo.fr\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "5160ee6fc891a9ca114be0e90fa6655647bb64b2", "tree": "35d3740a777935582af1b78238f20d2c2971ed55", "parents": [ "21b6bf143d05d77c350d9c6764ae090a877b66ea" ], "author": { "name": "Eric Dumazet", "email": "dada1@cosmosbay.com", "time": "Sun Jan 08 01:03:32 2006 -0800" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Sun Jan 08 20:13:58 2006 -0800" }, "message": "[PATCH] shrink dentry struct\n\nSome long time ago, dentry struct was carefully tuned so that on 32 bits\nUP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple\nof memory cache lines.\n\nThen RCU was added and dentry struct enlarged by two pointers, with nice\nresults for SMP, but not so good on UP, because breaking the above tuning\n(128 + 8 \u003d 136 bytes)\n\nThis patch reverts this unwanted side effect, by using an union (d_u),\nwhere d_rcu and d_child are placed so that these two fields can share their\nmemory needs.\n\nAt the time d_free() is called (and d_rcu is really used), d_child is known\nto be empty and not touched by the dentry freeing.\n\nLockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so\nthe previous content of d_child is not needed if said dentry was unhashed\nbut still accessed by a CPU because of RCU constraints)\n\nAs dentry cache easily contains millions of entries, a size reduction is\nworth the extra complexity of the ugly C union.\n\nSigned-off-by: Eric Dumazet \u003cdada1@cosmosbay.com\u003e\nCc: Dipankar Sarma \u003cdipankar@in.ibm.com\u003e\nCc: Maneesh Soni \u003cmaneesh@in.ibm.com\u003e\nCc: Miklos Szeredi \u003cmiklos@szeredi.hu\u003e\nCc: \"Paul E. McKenney\" \u003cpaulmck@us.ibm.com\u003e\nCc: Ian Kent \u003craven@themaw.net\u003e\nCc: Paul Jackson \u003cpj@sgi.com\u003e\nCc: Al Viro \u003cviro@ftp.linux.org.uk\u003e\nCc: Christoph Hellwig \u003chch@lst.de\u003e\nCc: Trond Myklebust \u003ctrond.myklebust@fys.uio.no\u003e\nCc: Neil Brown \u003cneilb@cse.unsw.edu.au\u003e\nCc: James Morris \u003cjmorris@namei.org\u003e\nCc: Stephen Smalley \u003csds@epoch.ncsc.mil\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "a05964f3917c7c55368c229d7985f8e7c9977e97", "tree": "7ec25550267ef050572c00ba2f37d6a4ff9c469e", "parents": [ "2144440327fa01b2f3f65e355120a78211685702" ], "author": { "name": "Ram Pai", "email": "linuxram@us.ibm.com", "time": "Mon Nov 07 17:20:17 2005 -0500" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Mon Nov 07 18:18:11 2005 -0800" }, "message": "[PATCH] shared mounts handling: umount\n\nAn unmount of a mount creates a umount event on the parent. If the\nparent is a shared mount, it gets propagated to all mounts in the peer\ngroup.\n\nSigned-off-by: Ram Pai \u003clinuxram@us.ibm.com\u003e\nSigned-off-by: Al Viro \u003cviro@zeniv.linux.org.uk\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "3f4bb1f4199b7dc0c958447b1e4898980013b884", "tree": "3d8aacfc6b7bbf7b2d224ae10ca693c55c64739e", "parents": [ "0811bab24ff1eecab38110eda7ea7847db95c64e" ], "author": { "name": "Eric Dumazet", "email": "dada1@cosmosbay.com", "time": "Tue Sep 06 15:18:16 2005 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Wed Sep 07 16:57:41 2005 -0700" }, "message": "[PATCH] struct dentry: place d_hash close to d_parent and d_name to speedup lookups\n\ndentry cache uses sophisticated RCU technology (and prefetching if\navailable) but touches 2 cache lines per dentry during hlist lookup.\n\nThis patch moves d_hash in the same cache line than d_parent and d_name\nfields so that :\n\n1) One cache line is needed instead of two.\n\n2) the hlist_for_each_rcu() prefetching has a chance to bring all the\n needed data in advance, not only the part that includes d_hash.next.\n\nI also changed one old comment that was wrong for 64bits.\n\nA further optimisation would be to separate dentry in two parts, one that\nis mostly read, and one writen (d_count/d_lock) to avoid false sharing on\nSMP/NUMA but this would need different field placement depending on 32bits\nor 64bits platform.\n\nSigned-off-by: Eric Dumazet \u003cdada1@cosmosbay.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "1da177e4c3f41524e886b7f1b8a0c1fc7321cac2", "tree": "0bba044c4ce775e45a88a51686b5d9f90697ea9d", "parents": [], "author": { "name": "Linus Torvalds", "email": "torvalds@ppc970.osdl.org", "time": "Sat Apr 16 15:20:36 2005 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@ppc970.osdl.org", "time": "Sat Apr 16 15:20:36 2005 -0700" }, "message": "Linux-2.6.12-rc2\n\nInitial git repository build. I\u0027m not bothering with the full history,\neven though we have it. We can create a separate \"historical\" git\narchive of that later if we want to, and in the meantime it\u0027s about\n3.2GB when imported into git - space that would just make the early\ngit days unnecessarily complicated, when we don\u0027t have a lot of good\ninfrastructure for it.\n\nLet it rip!\n" } ] }