Date: Tue, 22 Oct 2019 22:52:53 +0000 (UTC) From: Mateusz Guzik <mjg@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r353904 - head/sys/fs/pseudofs Message-ID: <201910222252.x9MMqrq6095522@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: mjg Date: Tue Oct 22 22:52:53 2019 New Revision: 353904 URL: https://svnweb.freebsd.org/changeset/base/353904 Log: pseudofs: hashed vncache Vast majority of uses the cache are just checking if there is an entry present on process exit (and evicting it if so). Both checking and eviction process are very expensive and put the lock protecting it high up on the profile during poudriere -j 104. Convert the linked list into a hash. This allows to almost always avoid taking the lock in the first place (and consequently almost removes it from the profile). Note only one lock is preserved as a split did not meaningfully impact contention. Should the cache be used for something it will still run into contention issues. The code needs a rewrite, but should someone want to tidy it up further the following can be done: 1) per-chain locks (or at least an array) 2) hashing by something else than just pid Sponsored by: The FreeBSD Foundation Modified: head/sys/fs/pseudofs/pseudofs_internal.h head/sys/fs/pseudofs/pseudofs_vncache.c Modified: head/sys/fs/pseudofs/pseudofs_internal.h ============================================================================== --- head/sys/fs/pseudofs/pseudofs_internal.h Tue Oct 22 22:23:59 2019 (r353903) +++ head/sys/fs/pseudofs/pseudofs_internal.h Tue Oct 22 22:52:53 2019 (r353904) @@ -45,8 +45,7 @@ struct pfs_vdata { struct pfs_node *pvd_pn; pid_t pvd_pid; struct vnode *pvd_vnode; - struct pfs_vdata*pvd_prev, *pvd_next; - int pvd_dead:1; + SLIST_ENTRY(pfs_vdata) pvd_hash; }; /* Modified: head/sys/fs/pseudofs/pseudofs_vncache.c ============================================================================== --- head/sys/fs/pseudofs/pseudofs_vncache.c Tue Oct 22 22:23:59 2019 (r353903) +++ head/sys/fs/pseudofs/pseudofs_vncache.c Tue Oct 22 22:52:53 2019 (r353904) @@ -50,10 +50,9 @@ __FBSDID("$FreeBSD$"); static MALLOC_DEFINE(M_PFSVNCACHE, "pfs_vncache", "pseudofs vnode cache"); static struct mtx pfs_vncache_mutex; -static struct pfs_vdata *pfs_vncache; static eventhandler_tag pfs_exit_tag; static void pfs_exit(void *arg, struct proc *p); -static void pfs_purge_locked(struct pfs_node *pn, bool force); +static void pfs_purge_all(void); static SYSCTL_NODE(_vfs_pfs, OID_AUTO, vncache, CTLFLAG_RW, 0, "pseudofs vnode cache"); @@ -80,6 +79,10 @@ SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, misses, CTLFLAG extern struct vop_vector pfs_vnodeops; /* XXX -> .h file */ +static SLIST_HEAD(pfs_vncache_head, pfs_vdata) *pfs_vncache_hashtbl; +static u_long pfs_vncache_hash; +#define PFS_VNCACHE_HASH(pid) (&pfs_vncache_hashtbl[(pid) & pfs_vncache_hash]) + /* * Initialize vnode cache */ @@ -88,6 +91,7 @@ pfs_vncache_load(void) { mtx_init(&pfs_vncache_mutex, "pfs_vncache", NULL, MTX_DEF); + pfs_vncache_hashtbl = hashinit(maxproc / 4, M_PFSVNCACHE, &pfs_vncache_hash); pfs_exit_tag = EVENTHANDLER_REGISTER(process_exit, pfs_exit, NULL, EVENTHANDLER_PRI_ANY); } @@ -100,9 +104,7 @@ pfs_vncache_unload(void) { EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag); - mtx_lock(&pfs_vncache_mutex); - pfs_purge_locked(NULL, true); - mtx_unlock(&pfs_vncache_mutex); + pfs_purge_all(); KASSERT(pfs_vncache_entries == 0, ("%d vncache entries remaining", pfs_vncache_entries)); mtx_destroy(&pfs_vncache_mutex); @@ -115,17 +117,20 @@ int pfs_vncache_alloc(struct mount *mp, struct vnode **vpp, struct pfs_node *pn, pid_t pid) { + struct pfs_vncache_head *hash; struct pfs_vdata *pvd, *pvd2; struct vnode *vp; int error; /* * See if the vnode is in the cache. - * XXX linear search is not very efficient. */ + hash = PFS_VNCACHE_HASH(pid); + if (SLIST_EMPTY(hash)) + goto alloc; retry: mtx_lock(&pfs_vncache_mutex); - for (pvd = pfs_vncache; pvd; pvd = pvd->pvd_next) { + SLIST_FOREACH(pvd, hash, pvd_hash) { if (pvd->pvd_pn == pn && pvd->pvd_pid == pid && pvd->pvd_vnode->v_mount == mp) { vp = pvd->pvd_vnode; @@ -150,10 +155,9 @@ retry: } } mtx_unlock(&pfs_vncache_mutex); - +alloc: /* nope, get a new one */ pvd = malloc(sizeof *pvd, M_PFSVNCACHE, M_WAITOK); - pvd->pvd_next = pvd->pvd_prev = NULL; error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp); if (error) { free(pvd, M_PFSVNCACHE); @@ -208,7 +212,7 @@ retry2: * going to insert into the cache. Recheck after * pfs_vncache_mutex is reacquired. */ - for (pvd2 = pfs_vncache; pvd2; pvd2 = pvd2->pvd_next) { + SLIST_FOREACH(pvd2, hash, pvd_hash) { if (pvd2->pvd_pn == pn && pvd2->pvd_pid == pid && pvd2->pvd_vnode->v_mount == mp) { vp = pvd2->pvd_vnode; @@ -228,11 +232,7 @@ retry2: ++pfs_vncache_misses; if (++pfs_vncache_entries > pfs_vncache_maxentries) pfs_vncache_maxentries = pfs_vncache_entries; - pvd->pvd_prev = NULL; - pvd->pvd_next = pfs_vncache; - if (pvd->pvd_next) - pvd->pvd_next->pvd_prev = pvd; - pfs_vncache = pvd; + SLIST_INSERT_HEAD(hash, pvd, pvd_hash); mtx_unlock(&pfs_vncache_mutex); return (0); } @@ -243,19 +243,17 @@ retry2: int pfs_vncache_free(struct vnode *vp) { - struct pfs_vdata *pvd; + struct pfs_vdata *pvd, *pvd2; mtx_lock(&pfs_vncache_mutex); pvd = (struct pfs_vdata *)vp->v_data; KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n")); - if (pvd->pvd_next) - pvd->pvd_next->pvd_prev = pvd->pvd_prev; - if (pvd->pvd_prev) { - pvd->pvd_prev->pvd_next = pvd->pvd_next; + SLIST_FOREACH(pvd2, PFS_VNCACHE_HASH(pvd->pvd_pid), pvd_hash) { + if (pvd2 != pvd) + continue; + SLIST_REMOVE(PFS_VNCACHE_HASH(pvd->pvd_pid), pvd, pfs_vdata, pvd_hash); --pfs_vncache_entries; - } else if (pfs_vncache == pvd) { - pfs_vncache = pvd->pvd_next; - --pfs_vncache_entries; + break; } mtx_unlock(&pfs_vncache_mutex); @@ -267,6 +265,15 @@ pfs_vncache_free(struct vnode *vp) /* * Purge the cache of dead entries * + * The code is not very efficient and this perhaps can be addressed without + * a complete rewrite. Previous iteration was walking a linked list from + * scratch every time. This code only walks the relevant hash chain (if pid + * is provided), but still resorts to scanning the entire cache at least twice + * if a specific component is to be removed which is slower. This can be + * augmented with resizing the hash. + * + * Explanation of the previous state: + * * This is extremely inefficient due to the fact that vgone() not only * indirectly modifies the vnode cache, but may also sleep. We can * neither hold pfs_vncache_mutex across a vgone() call, nor make any @@ -277,39 +284,51 @@ pfs_vncache_free(struct vnode *vp) * The only way to improve this situation is to change the data structure * used to implement the cache. */ + static void -pfs_purge_locked(struct pfs_node *pn, bool force) +pfs_purge_one(struct vnode *vnp) { + + VOP_LOCK(vnp, LK_EXCLUSIVE); + vgone(vnp); + VOP_UNLOCK(vnp, 0); + vdrop(vnp); +} + +void +pfs_purge(struct pfs_node *pn) +{ struct pfs_vdata *pvd; struct vnode *vnp; + u_long i, removed; - mtx_assert(&pfs_vncache_mutex, MA_OWNED); - pvd = pfs_vncache; - while (pvd != NULL) { - if (force || pvd->pvd_dead || - (pn != NULL && pvd->pvd_pn == pn)) { + mtx_lock(&pfs_vncache_mutex); +restart: + removed = 0; + for (i = 0; i < pfs_vncache_hash; i++) { +restart_chain: + SLIST_FOREACH(pvd, &pfs_vncache_hashtbl[i], pvd_hash) { + if (pn != NULL && pvd->pvd_pn != pn) + continue; vnp = pvd->pvd_vnode; vhold(vnp); mtx_unlock(&pfs_vncache_mutex); - VOP_LOCK(vnp, LK_EXCLUSIVE); - vgone(vnp); - VOP_UNLOCK(vnp, 0); + pfs_purge_one(vnp); + removed++; mtx_lock(&pfs_vncache_mutex); - vdrop(vnp); - pvd = pfs_vncache; - } else { - pvd = pvd->pvd_next; + goto restart_chain; } } + if (removed > 0) + goto restart; + mtx_unlock(&pfs_vncache_mutex); } -void -pfs_purge(struct pfs_node *pn) +static void +pfs_purge_all(void) { - mtx_lock(&pfs_vncache_mutex); - pfs_purge_locked(pn, false); - mtx_unlock(&pfs_vncache_mutex); + pfs_purge(NULL); } /* @@ -318,16 +337,25 @@ pfs_purge(struct pfs_node *pn) static void pfs_exit(void *arg, struct proc *p) { + struct pfs_vncache_head *hash; struct pfs_vdata *pvd; - int dead; + struct vnode *vnp; + int pid; - if (pfs_vncache == NULL) + pid = p->p_pid; + hash = PFS_VNCACHE_HASH(pid); + if (SLIST_EMPTY(hash)) return; +restart: mtx_lock(&pfs_vncache_mutex); - for (pvd = pfs_vncache, dead = 0; pvd != NULL; pvd = pvd->pvd_next) - if (pvd->pvd_pid == p->p_pid) - dead = pvd->pvd_dead = 1; - if (dead) - pfs_purge_locked(NULL, false); + SLIST_FOREACH(pvd, hash, pvd_hash) { + if (pvd->pvd_pid != pid) + continue; + vnp = pvd->pvd_vnode; + vhold(vnp); + mtx_unlock(&pfs_vncache_mutex); + pfs_purge_one(vnp); + goto restart; + } mtx_unlock(&pfs_vncache_mutex); }
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201910222252.x9MMqrq6095522>