Index: head/sys/kern/kern_jail.c =================================================================== --- head/sys/kern/kern_jail.c (revision 364768) +++ head/sys/kern/kern_jail.c (revision 364769) @@ -1,4260 +1,4260 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1999 Poul-Henning Kamp. * Copyright (c) 2008 Bjoern A. Zeeb. * Copyright (c) 2009 James Gritton. * All rights reserved. * * 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. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif /* DDB */ #include #define DEFAULT_HOSTUUID "00000000-0000-0000-0000-000000000000" #define PRISON0_HOSTUUID_MODULE "hostuuid" MALLOC_DEFINE(M_PRISON, "prison", "Prison structures"); static MALLOC_DEFINE(M_PRISON_RACCT, "prison_racct", "Prison racct structures"); /* Keep struct prison prison0 and some code in kern_jail_set() readable. */ #ifdef INET #ifdef INET6 #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL|PR_IP6_SADDRSEL #else #define _PR_IP_SADDRSEL PR_IP4_SADDRSEL #endif #else /* !INET */ #ifdef INET6 #define _PR_IP_SADDRSEL PR_IP6_SADDRSEL #else #define _PR_IP_SADDRSEL 0 #endif #endif /* prison0 describes what is "real" about the system. */ struct prison prison0 = { .pr_id = 0, .pr_name = "0", .pr_ref = 1, .pr_uref = 1, .pr_path = "/", .pr_securelevel = -1, .pr_devfs_rsnum = 0, .pr_childmax = JAIL_MAX, .pr_hostuuid = DEFAULT_HOSTUUID, .pr_children = LIST_HEAD_INITIALIZER(prison0.pr_children), #ifdef VIMAGE .pr_flags = PR_HOST|PR_VNET|_PR_IP_SADDRSEL, #else .pr_flags = PR_HOST|_PR_IP_SADDRSEL, #endif .pr_allow = PR_ALLOW_ALL_STATIC, }; MTX_SYSINIT(prison0, &prison0.pr_mtx, "jail mutex", MTX_DEF); struct bool_flags { const char *name; const char *noname; unsigned flag; }; struct jailsys_flags { const char *name; unsigned disable; unsigned new; }; /* allprison, allprison_racct and lastprid are protected by allprison_lock. */ struct sx allprison_lock; SX_SYSINIT(allprison_lock, &allprison_lock, "allprison"); struct prisonlist allprison = TAILQ_HEAD_INITIALIZER(allprison); LIST_HEAD(, prison_racct) allprison_racct; int lastprid = 0; static int do_jail_attach(struct thread *td, struct prison *pr); static void prison_complete(void *context, int pending); static void prison_deref(struct prison *pr, int flags); static char *prison_path(struct prison *pr1, struct prison *pr2); static void prison_remove_one(struct prison *pr); #ifdef RACCT static void prison_racct_attach(struct prison *pr); static void prison_racct_modify(struct prison *pr); static void prison_racct_detach(struct prison *pr); #endif /* Flags for prison_deref */ #define PD_DEREF 0x01 #define PD_DEUREF 0x02 #define PD_LOCKED 0x04 #define PD_LIST_SLOCKED 0x08 #define PD_LIST_XLOCKED 0x10 /* * Parameter names corresponding to PR_* flag values. Size values are for kvm * as we cannot figure out the size of a sparse array, or an array without a * terminating entry. */ static struct bool_flags pr_flag_bool[] = { {"persist", "nopersist", PR_PERSIST}, #ifdef INET {"ip4.saddrsel", "ip4.nosaddrsel", PR_IP4_SADDRSEL}, #endif #ifdef INET6 {"ip6.saddrsel", "ip6.nosaddrsel", PR_IP6_SADDRSEL}, #endif }; const size_t pr_flag_bool_size = sizeof(pr_flag_bool); static struct jailsys_flags pr_flag_jailsys[] = { {"host", 0, PR_HOST}, #ifdef VIMAGE {"vnet", 0, PR_VNET}, #endif #ifdef INET {"ip4", PR_IP4_USER, PR_IP4_USER}, #endif #ifdef INET6 {"ip6", PR_IP6_USER, PR_IP6_USER}, #endif }; const size_t pr_flag_jailsys_size = sizeof(pr_flag_jailsys); /* Make this array full-size so dynamic parameters can be added. */ static struct bool_flags pr_flag_allow[NBBY * NBPW] = { {"allow.set_hostname", "allow.noset_hostname", PR_ALLOW_SET_HOSTNAME}, {"allow.sysvipc", "allow.nosysvipc", PR_ALLOW_SYSVIPC}, {"allow.raw_sockets", "allow.noraw_sockets", PR_ALLOW_RAW_SOCKETS}, {"allow.chflags", "allow.nochflags", PR_ALLOW_CHFLAGS}, {"allow.mount", "allow.nomount", PR_ALLOW_MOUNT}, {"allow.quotas", "allow.noquotas", PR_ALLOW_QUOTAS}, {"allow.socket_af", "allow.nosocket_af", PR_ALLOW_SOCKET_AF}, {"allow.mlock", "allow.nomlock", PR_ALLOW_MLOCK}, {"allow.reserved_ports", "allow.noreserved_ports", PR_ALLOW_RESERVED_PORTS}, {"allow.read_msgbuf", "allow.noread_msgbuf", PR_ALLOW_READ_MSGBUF}, {"allow.unprivileged_proc_debug", "allow.nounprivileged_proc_debug", PR_ALLOW_UNPRIV_DEBUG}, }; const size_t pr_flag_allow_size = sizeof(pr_flag_allow); #define JAIL_DEFAULT_ALLOW (PR_ALLOW_SET_HOSTNAME | \ PR_ALLOW_RESERVED_PORTS | \ PR_ALLOW_UNPRIV_DEBUG) #define JAIL_DEFAULT_ENFORCE_STATFS 2 #define JAIL_DEFAULT_DEVFS_RSNUM 0 static unsigned jail_default_allow = JAIL_DEFAULT_ALLOW; static int jail_default_enforce_statfs = JAIL_DEFAULT_ENFORCE_STATFS; static int jail_default_devfs_rsnum = JAIL_DEFAULT_DEVFS_RSNUM; #if defined(INET) || defined(INET6) static unsigned jail_max_af_ips = 255; #endif /* * Initialize the parts of prison0 that can't be static-initialized with * constants. This is called from proc0_init() after creating thread0 cpuset. */ void prison0_init(void) { uint8_t *file, *data; size_t size; prison0.pr_cpuset = cpuset_ref(thread0.td_cpuset); prison0.pr_osreldate = osreldate; strlcpy(prison0.pr_osrelease, osrelease, sizeof(prison0.pr_osrelease)); /* If we have a preloaded hostuuid, use it. */ file = preload_search_by_type(PRISON0_HOSTUUID_MODULE); if (file != NULL) { data = preload_fetch_addr(file); size = preload_fetch_size(file); if (data != NULL) { /* * The preloaded data may include trailing whitespace, almost * certainly a newline; skip over any whitespace or * non-printable characters to be safe. */ while (size > 0 && data[size - 1] <= 0x20) { data[size--] = '\0'; } if (validate_uuid(data, size, NULL, 0) == 0) { (void)strlcpy(prison0.pr_hostuuid, data, size + 1); } else if (bootverbose) { printf("hostuuid: preload data malformed: '%s'", data); } } } if (bootverbose) printf("hostuuid: using %s\n", prison0.pr_hostuuid); } /* * struct jail_args { * struct jail *jail; * }; */ int sys_jail(struct thread *td, struct jail_args *uap) { uint32_t version; int error; struct jail j; error = copyin(uap->jail, &version, sizeof(uint32_t)); if (error) return (error); switch (version) { case 0: { struct jail_v0 j0; /* FreeBSD single IPv4 jails. */ bzero(&j, sizeof(struct jail)); error = copyin(uap->jail, &j0, sizeof(struct jail_v0)); if (error) return (error); j.version = j0.version; j.path = j0.path; j.hostname = j0.hostname; j.ip4s = htonl(j0.ip_number); /* jail_v0 is host order */ break; } case 1: /* * Version 1 was used by multi-IPv4 jail implementations * that never made it into the official kernel. */ return (EINVAL); case 2: /* JAIL_API_VERSION */ /* FreeBSD multi-IPv4/IPv6,noIP jails. */ error = copyin(uap->jail, &j, sizeof(struct jail)); if (error) return (error); break; default: /* Sci-Fi jails are not supported, sorry. */ return (EINVAL); } return (kern_jail(td, &j)); } int kern_jail(struct thread *td, struct jail *j) { struct iovec optiov[2 * (4 + nitems(pr_flag_allow) #ifdef INET + 1 #endif #ifdef INET6 + 1 #endif )]; struct uio opt; char *u_path, *u_hostname, *u_name; struct bool_flags *bf; #ifdef INET uint32_t ip4s; struct in_addr *u_ip4; #endif #ifdef INET6 struct in6_addr *u_ip6; #endif size_t tmplen; int error, enforce_statfs; bzero(&optiov, sizeof(optiov)); opt.uio_iov = optiov; opt.uio_iovcnt = 0; opt.uio_offset = -1; opt.uio_resid = -1; opt.uio_segflg = UIO_SYSSPACE; opt.uio_rw = UIO_READ; opt.uio_td = td; /* Set permissions for top-level jails from sysctls. */ if (!jailed(td->td_ucred)) { for (bf = pr_flag_allow; bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0; bf++) { optiov[opt.uio_iovcnt].iov_base = __DECONST(char *, (jail_default_allow & bf->flag) ? bf->name : bf->noname); optiov[opt.uio_iovcnt].iov_len = strlen(optiov[opt.uio_iovcnt].iov_base) + 1; opt.uio_iovcnt += 2; } optiov[opt.uio_iovcnt].iov_base = "enforce_statfs"; optiov[opt.uio_iovcnt].iov_len = sizeof("enforce_statfs"); opt.uio_iovcnt++; enforce_statfs = jail_default_enforce_statfs; optiov[opt.uio_iovcnt].iov_base = &enforce_statfs; optiov[opt.uio_iovcnt].iov_len = sizeof(enforce_statfs); opt.uio_iovcnt++; } tmplen = MAXPATHLEN + MAXHOSTNAMELEN + MAXHOSTNAMELEN; #ifdef INET ip4s = (j->version == 0) ? 1 : j->ip4s; if (ip4s > jail_max_af_ips) return (EINVAL); tmplen += ip4s * sizeof(struct in_addr); #else if (j->ip4s > 0) return (EINVAL); #endif #ifdef INET6 if (j->ip6s > jail_max_af_ips) return (EINVAL); tmplen += j->ip6s * sizeof(struct in6_addr); #else if (j->ip6s > 0) return (EINVAL); #endif u_path = malloc(tmplen, M_TEMP, M_WAITOK); u_hostname = u_path + MAXPATHLEN; u_name = u_hostname + MAXHOSTNAMELEN; #ifdef INET u_ip4 = (struct in_addr *)(u_name + MAXHOSTNAMELEN); #endif #ifdef INET6 #ifdef INET u_ip6 = (struct in6_addr *)(u_ip4 + ip4s); #else u_ip6 = (struct in6_addr *)(u_name + MAXHOSTNAMELEN); #endif #endif optiov[opt.uio_iovcnt].iov_base = "path"; optiov[opt.uio_iovcnt].iov_len = sizeof("path"); opt.uio_iovcnt++; optiov[opt.uio_iovcnt].iov_base = u_path; error = copyinstr(j->path, u_path, MAXPATHLEN, &optiov[opt.uio_iovcnt].iov_len); if (error) { free(u_path, M_TEMP); return (error); } opt.uio_iovcnt++; optiov[opt.uio_iovcnt].iov_base = "host.hostname"; optiov[opt.uio_iovcnt].iov_len = sizeof("host.hostname"); opt.uio_iovcnt++; optiov[opt.uio_iovcnt].iov_base = u_hostname; error = copyinstr(j->hostname, u_hostname, MAXHOSTNAMELEN, &optiov[opt.uio_iovcnt].iov_len); if (error) { free(u_path, M_TEMP); return (error); } opt.uio_iovcnt++; if (j->jailname != NULL) { optiov[opt.uio_iovcnt].iov_base = "name"; optiov[opt.uio_iovcnt].iov_len = sizeof("name"); opt.uio_iovcnt++; optiov[opt.uio_iovcnt].iov_base = u_name; error = copyinstr(j->jailname, u_name, MAXHOSTNAMELEN, &optiov[opt.uio_iovcnt].iov_len); if (error) { free(u_path, M_TEMP); return (error); } opt.uio_iovcnt++; } #ifdef INET optiov[opt.uio_iovcnt].iov_base = "ip4.addr"; optiov[opt.uio_iovcnt].iov_len = sizeof("ip4.addr"); opt.uio_iovcnt++; optiov[opt.uio_iovcnt].iov_base = u_ip4; optiov[opt.uio_iovcnt].iov_len = ip4s * sizeof(struct in_addr); if (j->version == 0) u_ip4->s_addr = j->ip4s; else { error = copyin(j->ip4, u_ip4, optiov[opt.uio_iovcnt].iov_len); if (error) { free(u_path, M_TEMP); return (error); } } opt.uio_iovcnt++; #endif #ifdef INET6 optiov[opt.uio_iovcnt].iov_base = "ip6.addr"; optiov[opt.uio_iovcnt].iov_len = sizeof("ip6.addr"); opt.uio_iovcnt++; optiov[opt.uio_iovcnt].iov_base = u_ip6; optiov[opt.uio_iovcnt].iov_len = j->ip6s * sizeof(struct in6_addr); error = copyin(j->ip6, u_ip6, optiov[opt.uio_iovcnt].iov_len); if (error) { free(u_path, M_TEMP); return (error); } opt.uio_iovcnt++; #endif KASSERT(opt.uio_iovcnt <= nitems(optiov), ("kern_jail: too many iovecs (%d)", opt.uio_iovcnt)); error = kern_jail_set(td, &opt, JAIL_CREATE | JAIL_ATTACH); free(u_path, M_TEMP); return (error); } /* * struct jail_set_args { * struct iovec *iovp; * unsigned int iovcnt; * int flags; * }; */ int sys_jail_set(struct thread *td, struct jail_set_args *uap) { struct uio *auio; int error; /* Check that we have an even number of iovecs. */ if (uap->iovcnt & 1) return (EINVAL); error = copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_jail_set(td, auio, uap->flags); free(auio, M_IOV); return (error); } int kern_jail_set(struct thread *td, struct uio *optuio, int flags) { struct nameidata nd; #ifdef INET struct in_addr *ip4; #endif #ifdef INET6 struct in6_addr *ip6; #endif struct vfsopt *opt; struct vfsoptlist *opts; struct prison *pr, *deadpr, *mypr, *ppr, *tpr; struct vnode *root; char *domain, *errmsg, *host, *name, *namelc, *p, *path, *uuid; char *g_path, *osrelstr; struct bool_flags *bf; struct jailsys_flags *jsf; #if defined(INET) || defined(INET6) struct prison *tppr; void *op; #endif unsigned long hid; size_t namelen, onamelen, pnamelen; int born, created, cuflags, descend, enforce; int error, errmsg_len, errmsg_pos; int gotchildmax, gotenforce, gothid, gotrsnum, gotslevel; int jid, jsys, len, level; int childmax, osreldt, rsnum, slevel; #if defined(INET) || defined(INET6) int ii, ij; #endif #ifdef INET int ip4s, redo_ip4; #endif #ifdef INET6 int ip6s, redo_ip6; #endif uint64_t pr_allow, ch_allow, pr_flags, ch_flags; uint64_t pr_allow_diff; unsigned tallow; char numbuf[12]; error = priv_check(td, PRIV_JAIL_SET); if (!error && (flags & JAIL_ATTACH)) error = priv_check(td, PRIV_JAIL_ATTACH); if (error) return (error); mypr = td->td_ucred->cr_prison; if ((flags & JAIL_CREATE) && mypr->pr_childmax == 0) return (EPERM); if (flags & ~JAIL_SET_MASK) return (EINVAL); /* * Check all the parameters before committing to anything. Not all * errors can be caught early, but we may as well try. Also, this * takes care of some expensive stuff (path lookup) before getting * the allprison lock. * * XXX Jails are not filesystems, and jail parameters are not mount * options. But it makes more sense to re-use the vfsopt code * than duplicate it under a different name. */ error = vfs_buildopts(optuio, &opts); if (error) return (error); #ifdef INET ip4 = NULL; #endif #ifdef INET6 ip6 = NULL; #endif g_path = NULL; cuflags = flags & (JAIL_CREATE | JAIL_UPDATE); if (!cuflags) { error = EINVAL; vfs_opterror(opts, "no valid operation (create or update)"); goto done_errmsg; } error = vfs_copyopt(opts, "jid", &jid, sizeof(jid)); if (error == ENOENT) jid = 0; else if (error != 0) goto done_free; error = vfs_copyopt(opts, "securelevel", &slevel, sizeof(slevel)); if (error == ENOENT) gotslevel = 0; else if (error != 0) goto done_free; else gotslevel = 1; error = vfs_copyopt(opts, "children.max", &childmax, sizeof(childmax)); if (error == ENOENT) gotchildmax = 0; else if (error != 0) goto done_free; else gotchildmax = 1; error = vfs_copyopt(opts, "enforce_statfs", &enforce, sizeof(enforce)); if (error == ENOENT) gotenforce = 0; else if (error != 0) goto done_free; else if (enforce < 0 || enforce > 2) { error = EINVAL; goto done_free; } else gotenforce = 1; error = vfs_copyopt(opts, "devfs_ruleset", &rsnum, sizeof(rsnum)); if (error == ENOENT) gotrsnum = 0; else if (error != 0) goto done_free; else gotrsnum = 1; pr_flags = ch_flags = 0; for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++) { vfs_flagopt(opts, bf->name, &pr_flags, bf->flag); vfs_flagopt(opts, bf->noname, &ch_flags, bf->flag); } ch_flags |= pr_flags; for (jsf = pr_flag_jailsys; jsf < pr_flag_jailsys + nitems(pr_flag_jailsys); jsf++) { error = vfs_copyopt(opts, jsf->name, &jsys, sizeof(jsys)); if (error == ENOENT) continue; if (error != 0) goto done_free; switch (jsys) { case JAIL_SYS_DISABLE: if (!jsf->disable) { error = EINVAL; goto done_free; } pr_flags |= jsf->disable; break; case JAIL_SYS_NEW: pr_flags |= jsf->new; break; case JAIL_SYS_INHERIT: break; default: error = EINVAL; goto done_free; } ch_flags |= jsf->new | jsf->disable; } if ((flags & (JAIL_CREATE | JAIL_UPDATE | JAIL_ATTACH)) == JAIL_CREATE && !(pr_flags & PR_PERSIST)) { error = EINVAL; vfs_opterror(opts, "new jail must persist or attach"); goto done_errmsg; } #ifdef VIMAGE if ((flags & JAIL_UPDATE) && (ch_flags & PR_VNET)) { error = EINVAL; vfs_opterror(opts, "vnet cannot be changed after creation"); goto done_errmsg; } #endif #ifdef INET if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP4_USER)) { error = EINVAL; vfs_opterror(opts, "ip4 cannot be changed after creation"); goto done_errmsg; } #endif #ifdef INET6 if ((flags & JAIL_UPDATE) && (ch_flags & PR_IP6_USER)) { error = EINVAL; vfs_opterror(opts, "ip6 cannot be changed after creation"); goto done_errmsg; } #endif pr_allow = ch_allow = 0; for (bf = pr_flag_allow; bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0; bf++) { vfs_flagopt(opts, bf->name, &pr_allow, bf->flag); vfs_flagopt(opts, bf->noname, &ch_allow, bf->flag); } ch_allow |= pr_allow; error = vfs_getopt(opts, "name", (void **)&name, &len); if (error == ENOENT) name = NULL; else if (error != 0) goto done_free; else { if (len == 0 || name[len - 1] != '\0') { error = EINVAL; goto done_free; } if (len > MAXHOSTNAMELEN) { error = ENAMETOOLONG; goto done_free; } } error = vfs_getopt(opts, "host.hostname", (void **)&host, &len); if (error == ENOENT) host = NULL; else if (error != 0) goto done_free; else { ch_flags |= PR_HOST; pr_flags |= PR_HOST; if (len == 0 || host[len - 1] != '\0') { error = EINVAL; goto done_free; } if (len > MAXHOSTNAMELEN) { error = ENAMETOOLONG; goto done_free; } } error = vfs_getopt(opts, "host.domainname", (void **)&domain, &len); if (error == ENOENT) domain = NULL; else if (error != 0) goto done_free; else { ch_flags |= PR_HOST; pr_flags |= PR_HOST; if (len == 0 || domain[len - 1] != '\0') { error = EINVAL; goto done_free; } if (len > MAXHOSTNAMELEN) { error = ENAMETOOLONG; goto done_free; } } error = vfs_getopt(opts, "host.hostuuid", (void **)&uuid, &len); if (error == ENOENT) uuid = NULL; else if (error != 0) goto done_free; else { ch_flags |= PR_HOST; pr_flags |= PR_HOST; if (len == 0 || uuid[len - 1] != '\0') { error = EINVAL; goto done_free; } if (len > HOSTUUIDLEN) { error = ENAMETOOLONG; goto done_free; } } #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { uint32_t hid32; error = vfs_copyopt(opts, "host.hostid", &hid32, sizeof(hid32)); hid = hid32; } else #endif error = vfs_copyopt(opts, "host.hostid", &hid, sizeof(hid)); if (error == ENOENT) gothid = 0; else if (error != 0) goto done_free; else { gothid = 1; ch_flags |= PR_HOST; pr_flags |= PR_HOST; } #ifdef INET error = vfs_getopt(opts, "ip4.addr", &op, &ip4s); if (error == ENOENT) ip4s = 0; else if (error != 0) goto done_free; else if (ip4s & (sizeof(*ip4) - 1)) { error = EINVAL; goto done_free; } else { ch_flags |= PR_IP4_USER; pr_flags |= PR_IP4_USER; if (ip4s > 0) { ip4s /= sizeof(*ip4); if (ip4s > jail_max_af_ips) { error = EINVAL; vfs_opterror(opts, "too many IPv4 addresses"); goto done_errmsg; } ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK); bcopy(op, ip4, ip4s * sizeof(*ip4)); /* * IP addresses are all sorted but ip[0] to preserve * the primary IP address as given from userland. * This special IP is used for unbound outgoing * connections as well for "loopback" traffic in case * source address selection cannot find any more fitting * address to connect from. */ if (ip4s > 1) qsort(ip4 + 1, ip4s - 1, sizeof(*ip4), prison_qcmp_v4); /* * Check for duplicate addresses and do some simple * zero and broadcast checks. If users give other bogus * addresses it is their problem. * * We do not have to care about byte order for these * checks so we will do them in NBO. */ for (ii = 0; ii < ip4s; ii++) { if (ip4[ii].s_addr == INADDR_ANY || ip4[ii].s_addr == INADDR_BROADCAST) { error = EINVAL; goto done_free; } if ((ii+1) < ip4s && (ip4[0].s_addr == ip4[ii+1].s_addr || ip4[ii].s_addr == ip4[ii+1].s_addr)) { error = EINVAL; goto done_free; } } } } #endif #ifdef INET6 error = vfs_getopt(opts, "ip6.addr", &op, &ip6s); if (error == ENOENT) ip6s = 0; else if (error != 0) goto done_free; else if (ip6s & (sizeof(*ip6) - 1)) { error = EINVAL; goto done_free; } else { ch_flags |= PR_IP6_USER; pr_flags |= PR_IP6_USER; if (ip6s > 0) { ip6s /= sizeof(*ip6); if (ip6s > jail_max_af_ips) { error = EINVAL; vfs_opterror(opts, "too many IPv6 addresses"); goto done_errmsg; } ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK); bcopy(op, ip6, ip6s * sizeof(*ip6)); if (ip6s > 1) qsort(ip6 + 1, ip6s - 1, sizeof(*ip6), prison_qcmp_v6); for (ii = 0; ii < ip6s; ii++) { if (IN6_IS_ADDR_UNSPECIFIED(&ip6[ii])) { error = EINVAL; goto done_free; } if ((ii+1) < ip6s && (IN6_ARE_ADDR_EQUAL(&ip6[0], &ip6[ii+1]) || IN6_ARE_ADDR_EQUAL(&ip6[ii], &ip6[ii+1]))) { error = EINVAL; goto done_free; } } } } #endif #if defined(VIMAGE) && (defined(INET) || defined(INET6)) if ((ch_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) { error = EINVAL; vfs_opterror(opts, "vnet jails cannot have IP address restrictions"); goto done_errmsg; } #endif error = vfs_getopt(opts, "osrelease", (void **)&osrelstr, &len); if (error == ENOENT) osrelstr = NULL; else if (error != 0) goto done_free; else { if (flags & JAIL_UPDATE) { error = EINVAL; vfs_opterror(opts, "osrelease cannot be changed after creation"); goto done_errmsg; } if (len == 0 || osrelstr[len - 1] != '\0') { error = EINVAL; goto done_free; } if (len >= OSRELEASELEN) { error = ENAMETOOLONG; vfs_opterror(opts, "osrelease string must be 1-%d bytes long", OSRELEASELEN - 1); goto done_errmsg; } } error = vfs_copyopt(opts, "osreldate", &osreldt, sizeof(osreldt)); if (error == ENOENT) osreldt = 0; else if (error != 0) goto done_free; else { if (flags & JAIL_UPDATE) { error = EINVAL; vfs_opterror(opts, "osreldate cannot be changed after creation"); goto done_errmsg; } if (osreldt == 0) { error = EINVAL; vfs_opterror(opts, "osreldate cannot be 0"); goto done_errmsg; } } root = NULL; error = vfs_getopt(opts, "path", (void **)&path, &len); if (error == ENOENT) path = NULL; else if (error != 0) goto done_free; else { if (flags & JAIL_UPDATE) { error = EINVAL; vfs_opterror(opts, "path cannot be changed after creation"); goto done_errmsg; } if (len == 0 || path[len - 1] != '\0') { error = EINVAL; goto done_free; } NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, path, td); error = namei(&nd); if (error) goto done_free; root = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); g_path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); strlcpy(g_path, path, MAXPATHLEN); error = vn_path_to_global_path(td, root, g_path, MAXPATHLEN); if (error == 0) { path = g_path; } else { /* exit on other errors */ goto done_free; } if (root->v_type != VDIR) { error = ENOTDIR; vput(root); goto done_free; } VOP_UNLOCK(root); } /* * Find the specified jail, or at least its parent. * This abuses the file error codes ENOENT and EEXIST. */ pr = NULL; ppr = mypr; if (cuflags == JAIL_CREATE && jid == 0 && name != NULL) { namelc = strrchr(name, '.'); jid = strtoul(namelc != NULL ? namelc + 1 : name, &p, 10); if (*p != '\0') jid = 0; } sx_xlock(&allprison_lock); if (jid != 0) { /* * See if a requested jid already exists. There is an * information leak here if the jid exists but is not within * the caller's jail hierarchy. Jail creators will get EEXIST * even though they cannot see the jail, and CREATE | UPDATE * will return ENOENT which is not normally a valid error. */ if (jid < 0) { error = EINVAL; vfs_opterror(opts, "negative jid"); goto done_unlock_list; } pr = prison_find(jid); if (pr != NULL) { ppr = pr->pr_parent; /* Create: jid must not exist. */ if (cuflags == JAIL_CREATE) { mtx_unlock(&pr->pr_mtx); error = EEXIST; vfs_opterror(opts, "jail %d already exists", jid); goto done_unlock_list; } if (!prison_ischild(mypr, pr)) { mtx_unlock(&pr->pr_mtx); pr = NULL; } else if (pr->pr_uref == 0) { if (!(flags & JAIL_DYING)) { mtx_unlock(&pr->pr_mtx); error = ENOENT; vfs_opterror(opts, "jail %d is dying", jid); goto done_unlock_list; } else if ((flags & JAIL_ATTACH) || (pr_flags & PR_PERSIST)) { /* * A dying jail might be resurrected * (via attach or persist), but first * it must determine if another jail * has claimed its name. Accomplish * this by implicitly re-setting the * name. */ if (name == NULL) name = prison_name(mypr, pr); } } } if (pr == NULL) { /* Update: jid must exist. */ if (cuflags == JAIL_UPDATE) { error = ENOENT; vfs_opterror(opts, "jail %d not found", jid); goto done_unlock_list; } } } /* * If the caller provided a name, look for a jail by that name. * This has different semantics for creates and updates keyed by jid * (where the name must not already exist in a different jail), * and updates keyed by the name itself (where the name must exist * because that is the jail being updated). */ namelc = NULL; if (name != NULL) { namelc = strrchr(name, '.'); if (namelc == NULL) namelc = name; else { /* * This is a hierarchical name. Split it into the * parent and child names, and make sure the parent * exists or matches an already found jail. */ if (pr != NULL) { if (strncmp(name, ppr->pr_name, namelc - name) || ppr->pr_name[namelc - name] != '\0') { mtx_unlock(&pr->pr_mtx); error = EINVAL; vfs_opterror(opts, "cannot change jail's parent"); goto done_unlock_list; } } else { *namelc = '\0'; ppr = prison_find_name(mypr, name); if (ppr == NULL) { error = ENOENT; vfs_opterror(opts, "jail \"%s\" not found", name); goto done_unlock_list; } mtx_unlock(&ppr->pr_mtx); *namelc = '.'; } namelc++; } if (namelc[0] != '\0') { pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1; name_again: deadpr = NULL; FOREACH_PRISON_CHILD(ppr, tpr) { if (tpr != pr && tpr->pr_ref > 0 && !strcmp(tpr->pr_name + pnamelen, namelc)) { if (pr == NULL && cuflags != JAIL_CREATE) { mtx_lock(&tpr->pr_mtx); if (tpr->pr_ref > 0) { /* * Use this jail * for updates. */ if (tpr->pr_uref > 0) { pr = tpr; break; } deadpr = tpr; } mtx_unlock(&tpr->pr_mtx); } else if (tpr->pr_uref > 0) { /* * Create, or update(jid): * name must not exist in an * active sibling jail. */ error = EEXIST; if (pr != NULL) mtx_unlock(&pr->pr_mtx); vfs_opterror(opts, "jail \"%s\" already exists", name); goto done_unlock_list; } } } /* If no active jail is found, use a dying one. */ if (deadpr != NULL && pr == NULL) { if (flags & JAIL_DYING) { mtx_lock(&deadpr->pr_mtx); if (deadpr->pr_ref == 0) { mtx_unlock(&deadpr->pr_mtx); goto name_again; } pr = deadpr; } else if (cuflags == JAIL_UPDATE) { error = ENOENT; vfs_opterror(opts, "jail \"%s\" is dying", name); goto done_unlock_list; } } /* Update: name must exist if no jid. */ else if (cuflags == JAIL_UPDATE && pr == NULL) { error = ENOENT; vfs_opterror(opts, "jail \"%s\" not found", name); goto done_unlock_list; } } } /* Update: must provide a jid or name. */ else if (cuflags == JAIL_UPDATE && pr == NULL) { error = ENOENT; vfs_opterror(opts, "update specified no jail"); goto done_unlock_list; } /* If there's no prison to update, create a new one and link it in. */ if (pr == NULL) { for (tpr = mypr; tpr != NULL; tpr = tpr->pr_parent) if (tpr->pr_childcount >= tpr->pr_childmax) { error = EPERM; vfs_opterror(opts, "prison limit exceeded"); goto done_unlock_list; } created = 1; mtx_lock(&ppr->pr_mtx); if (ppr->pr_ref == 0) { mtx_unlock(&ppr->pr_mtx); error = ENOENT; vfs_opterror(opts, "jail \"%s\" not found", prison_name(mypr, ppr)); goto done_unlock_list; } ppr->pr_ref++; ppr->pr_uref++; mtx_unlock(&ppr->pr_mtx); pr = malloc(sizeof(*pr), M_PRISON, M_WAITOK | M_ZERO); if (jid == 0) { /* Find the next free jid. */ jid = lastprid + 1; findnext: if (jid == JAIL_MAX) jid = 1; TAILQ_FOREACH(tpr, &allprison, pr_list) { if (tpr->pr_id < jid) continue; if (tpr->pr_id > jid || tpr->pr_ref == 0) { TAILQ_INSERT_BEFORE(tpr, pr, pr_list); break; } if (jid == lastprid) { error = EAGAIN; vfs_opterror(opts, "no available jail IDs"); free(pr, M_PRISON); prison_deref(ppr, PD_DEREF | PD_DEUREF | PD_LIST_XLOCKED); goto done_releroot; } jid++; goto findnext; } lastprid = jid; } else { /* * The jail already has a jid (that did not yet exist), * so just find where to insert it. */ TAILQ_FOREACH(tpr, &allprison, pr_list) if (tpr->pr_id >= jid) { TAILQ_INSERT_BEFORE(tpr, pr, pr_list); break; } } if (tpr == NULL) TAILQ_INSERT_TAIL(&allprison, pr, pr_list); LIST_INSERT_HEAD(&ppr->pr_children, pr, pr_sibling); for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent) tpr->pr_childcount++; pr->pr_parent = ppr; pr->pr_id = jid; /* Set some default values, and inherit some from the parent. */ if (namelc == NULL) namelc = ""; if (path == NULL) { path = "/"; root = mypr->pr_root; vref(root); } strlcpy(pr->pr_hostuuid, DEFAULT_HOSTUUID, HOSTUUIDLEN); pr->pr_flags |= PR_HOST; #if defined(INET) || defined(INET6) #ifdef VIMAGE if (!(pr_flags & PR_VNET)) #endif { #ifdef INET if (!(ch_flags & PR_IP4_USER)) pr->pr_flags |= PR_IP4 | PR_IP4_USER; else if (!(pr_flags & PR_IP4_USER)) { pr->pr_flags |= ppr->pr_flags & PR_IP4; if (ppr->pr_ip4 != NULL) { pr->pr_ip4s = ppr->pr_ip4s; pr->pr_ip4 = malloc(pr->pr_ip4s * sizeof(struct in_addr), M_PRISON, M_WAITOK); bcopy(ppr->pr_ip4, pr->pr_ip4, pr->pr_ip4s * sizeof(*pr->pr_ip4)); } } #endif #ifdef INET6 if (!(ch_flags & PR_IP6_USER)) pr->pr_flags |= PR_IP6 | PR_IP6_USER; else if (!(pr_flags & PR_IP6_USER)) { pr->pr_flags |= ppr->pr_flags & PR_IP6; if (ppr->pr_ip6 != NULL) { pr->pr_ip6s = ppr->pr_ip6s; pr->pr_ip6 = malloc(pr->pr_ip6s * sizeof(struct in6_addr), M_PRISON, M_WAITOK); bcopy(ppr->pr_ip6, pr->pr_ip6, pr->pr_ip6s * sizeof(*pr->pr_ip6)); } } #endif } #endif /* Source address selection is always on by default. */ pr->pr_flags |= _PR_IP_SADDRSEL; pr->pr_securelevel = ppr->pr_securelevel; pr->pr_allow = JAIL_DEFAULT_ALLOW & ppr->pr_allow; pr->pr_enforce_statfs = jail_default_enforce_statfs; pr->pr_devfs_rsnum = ppr->pr_devfs_rsnum; pr->pr_osreldate = osreldt ? osreldt : ppr->pr_osreldate; if (osrelstr == NULL) strlcpy(pr->pr_osrelease, ppr->pr_osrelease, sizeof(pr->pr_osrelease)); else strlcpy(pr->pr_osrelease, osrelstr, sizeof(pr->pr_osrelease)); LIST_INIT(&pr->pr_children); mtx_init(&pr->pr_mtx, "jail mutex", NULL, MTX_DEF | MTX_DUPOK); TASK_INIT(&pr->pr_task, 0, prison_complete, pr); #ifdef VIMAGE /* Allocate a new vnet if specified. */ pr->pr_vnet = (pr_flags & PR_VNET) ? vnet_alloc() : ppr->pr_vnet; #endif /* * Allocate a dedicated cpuset for each jail. * Unlike other initial settings, this may return an erorr. */ error = cpuset_create_root(ppr, &pr->pr_cpuset); if (error) { prison_deref(pr, PD_LIST_XLOCKED); goto done_releroot; } mtx_lock(&pr->pr_mtx); /* * New prisons do not yet have a reference, because we do not * want others to see the incomplete prison once the * allprison_lock is downgraded. */ } else { created = 0; /* * Grab a reference for existing prisons, to ensure they * continue to exist for the duration of the call. */ pr->pr_ref++; #if defined(VIMAGE) && (defined(INET) || defined(INET6)) if ((pr->pr_flags & PR_VNET) && (ch_flags & (PR_IP4_USER | PR_IP6_USER))) { error = EINVAL; vfs_opterror(opts, "vnet jails cannot have IP address restrictions"); goto done_deref_locked; } #endif #ifdef INET if (PR_IP4_USER & ch_flags & (pr_flags ^ pr->pr_flags)) { error = EINVAL; vfs_opterror(opts, "ip4 cannot be changed after creation"); goto done_deref_locked; } #endif #ifdef INET6 if (PR_IP6_USER & ch_flags & (pr_flags ^ pr->pr_flags)) { error = EINVAL; vfs_opterror(opts, "ip6 cannot be changed after creation"); goto done_deref_locked; } #endif } /* Do final error checking before setting anything. */ if (gotslevel) { if (slevel < ppr->pr_securelevel) { error = EPERM; goto done_deref_locked; } } if (gotchildmax) { if (childmax >= ppr->pr_childmax) { error = EPERM; goto done_deref_locked; } } if (gotenforce) { if (enforce < ppr->pr_enforce_statfs) { error = EPERM; goto done_deref_locked; } } if (gotrsnum) { /* * devfs_rsnum is a uint16_t */ if (rsnum < 0 || rsnum > 65535) { error = EINVAL; goto done_deref_locked; } /* * Nested jails always inherit parent's devfs ruleset */ if (jailed(td->td_ucred)) { if (rsnum > 0 && rsnum != ppr->pr_devfs_rsnum) { error = EPERM; goto done_deref_locked; } else rsnum = ppr->pr_devfs_rsnum; } } #ifdef INET if (ip4s > 0) { if (ppr->pr_flags & PR_IP4) { /* * Make sure the new set of IP addresses is a * subset of the parent's list. Don't worry * about the parent being unlocked, as any * setting is done with allprison_lock held. */ for (ij = 0; ij < ppr->pr_ip4s; ij++) if (ip4[0].s_addr == ppr->pr_ip4[ij].s_addr) break; if (ij == ppr->pr_ip4s) { error = EPERM; goto done_deref_locked; } if (ip4s > 1) { for (ii = ij = 1; ii < ip4s; ii++) { if (ip4[ii].s_addr == ppr->pr_ip4[0].s_addr) continue; for (; ij < ppr->pr_ip4s; ij++) if (ip4[ii].s_addr == ppr->pr_ip4[ij].s_addr) break; if (ij == ppr->pr_ip4s) break; } if (ij == ppr->pr_ip4s) { error = EPERM; goto done_deref_locked; } } } /* * Check for conflicting IP addresses. We permit them * if there is no more than one IP on each jail. If * there is a duplicate on a jail with more than one * IP stop checking and return error. */ #ifdef VIMAGE for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent) if (tppr->pr_flags & PR_VNET) break; #else tppr = &prison0; #endif FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) { if (tpr == pr || #ifdef VIMAGE (tpr != tppr && (tpr->pr_flags & PR_VNET)) || #endif tpr->pr_uref == 0) { descend = 0; continue; } if (!(tpr->pr_flags & PR_IP4_USER)) continue; descend = 0; if (tpr->pr_ip4 == NULL || (ip4s == 1 && tpr->pr_ip4s == 1)) continue; for (ii = 0; ii < ip4s; ii++) { if (prison_check_ip4_locked(tpr, &ip4[ii]) == 0) { error = EADDRINUSE; vfs_opterror(opts, "IPv4 addresses clash"); goto done_deref_locked; } } } } #endif #ifdef INET6 if (ip6s > 0) { if (ppr->pr_flags & PR_IP6) { /* * Make sure the new set of IP addresses is a * subset of the parent's list. */ for (ij = 0; ij < ppr->pr_ip6s; ij++) if (IN6_ARE_ADDR_EQUAL(&ip6[0], &ppr->pr_ip6[ij])) break; if (ij == ppr->pr_ip6s) { error = EPERM; goto done_deref_locked; } if (ip6s > 1) { for (ii = ij = 1; ii < ip6s; ii++) { if (IN6_ARE_ADDR_EQUAL(&ip6[ii], &ppr->pr_ip6[0])) continue; for (; ij < ppr->pr_ip6s; ij++) if (IN6_ARE_ADDR_EQUAL( &ip6[ii], &ppr->pr_ip6[ij])) break; if (ij == ppr->pr_ip6s) break; } if (ij == ppr->pr_ip6s) { error = EPERM; goto done_deref_locked; } } } /* Check for conflicting IP addresses. */ #ifdef VIMAGE for (tppr = ppr; tppr != &prison0; tppr = tppr->pr_parent) if (tppr->pr_flags & PR_VNET) break; #else tppr = &prison0; #endif FOREACH_PRISON_DESCENDANT(tppr, tpr, descend) { if (tpr == pr || #ifdef VIMAGE (tpr != tppr && (tpr->pr_flags & PR_VNET)) || #endif tpr->pr_uref == 0) { descend = 0; continue; } if (!(tpr->pr_flags & PR_IP6_USER)) continue; descend = 0; if (tpr->pr_ip6 == NULL || (ip6s == 1 && tpr->pr_ip6s == 1)) continue; for (ii = 0; ii < ip6s; ii++) { if (prison_check_ip6_locked(tpr, &ip6[ii]) == 0) { error = EADDRINUSE; vfs_opterror(opts, "IPv6 addresses clash"); goto done_deref_locked; } } } } #endif onamelen = namelen = 0; if (namelc != NULL) { /* Give a default name of the jid. Also allow the name to be * explicitly the jid - but not any other number, and only in * normal form (no leading zero/etc). */ if (namelc[0] == '\0') snprintf(namelc = numbuf, sizeof(numbuf), "%d", jid); else if ((strtoul(namelc, &p, 10) != jid || namelc[0] < '1' || namelc[0] > '9') && *p == '\0') { error = EINVAL; vfs_opterror(opts, "name cannot be numeric (unless it is the jid)"); goto done_deref_locked; } /* * Make sure the name isn't too long for the prison or its * children. */ pnamelen = (ppr == &prison0) ? 0 : strlen(ppr->pr_name) + 1; onamelen = strlen(pr->pr_name + pnamelen); namelen = strlen(namelc); if (pnamelen + namelen + 1 > sizeof(pr->pr_name)) { error = ENAMETOOLONG; goto done_deref_locked; } FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { if (strlen(tpr->pr_name) + (namelen - onamelen) >= sizeof(pr->pr_name)) { error = ENAMETOOLONG; goto done_deref_locked; } } } pr_allow_diff = pr_allow & ~ppr->pr_allow; if (pr_allow_diff & ~PR_ALLOW_DIFFERENCES) { error = EPERM; goto done_deref_locked; } /* * Let modules check their parameters. This requires unlocking and * then re-locking the prison, but this is still a valid state as long * as allprison_lock remains xlocked. */ mtx_unlock(&pr->pr_mtx); error = osd_jail_call(pr, PR_METHOD_CHECK, opts); if (error != 0) { prison_deref(pr, created ? PD_LIST_XLOCKED : PD_DEREF | PD_LIST_XLOCKED); goto done_releroot; } mtx_lock(&pr->pr_mtx); /* At this point, all valid parameters should have been noted. */ TAILQ_FOREACH(opt, opts, link) { if (!opt->seen && strcmp(opt->name, "errmsg")) { error = EINVAL; vfs_opterror(opts, "unknown parameter: %s", opt->name); goto done_deref_locked; } } /* Set the parameters of the prison. */ #ifdef INET redo_ip4 = 0; if (pr_flags & PR_IP4_USER) { pr->pr_flags |= PR_IP4; free(pr->pr_ip4, M_PRISON); pr->pr_ip4s = ip4s; pr->pr_ip4 = ip4; ip4 = NULL; FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { #ifdef VIMAGE if (tpr->pr_flags & PR_VNET) { descend = 0; continue; } #endif if (prison_restrict_ip4(tpr, NULL)) { redo_ip4 = 1; descend = 0; } } } #endif #ifdef INET6 redo_ip6 = 0; if (pr_flags & PR_IP6_USER) { pr->pr_flags |= PR_IP6; free(pr->pr_ip6, M_PRISON); pr->pr_ip6s = ip6s; pr->pr_ip6 = ip6; ip6 = NULL; FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { #ifdef VIMAGE if (tpr->pr_flags & PR_VNET) { descend = 0; continue; } #endif if (prison_restrict_ip6(tpr, NULL)) { redo_ip6 = 1; descend = 0; } } } #endif if (gotslevel) { pr->pr_securelevel = slevel; /* Set all child jails to be at least this level. */ FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) if (tpr->pr_securelevel < slevel) tpr->pr_securelevel = slevel; } if (gotchildmax) { pr->pr_childmax = childmax; /* Set all child jails to under this limit. */ FOREACH_PRISON_DESCENDANT_LOCKED_LEVEL(pr, tpr, descend, level) if (tpr->pr_childmax > childmax - level) tpr->pr_childmax = childmax > level ? childmax - level : 0; } if (gotenforce) { pr->pr_enforce_statfs = enforce; /* Pass this restriction on to the children. */ FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) if (tpr->pr_enforce_statfs < enforce) tpr->pr_enforce_statfs = enforce; } if (gotrsnum) { pr->pr_devfs_rsnum = rsnum; /* Pass this restriction on to the children. */ FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) tpr->pr_devfs_rsnum = rsnum; } if (namelc != NULL) { if (ppr == &prison0) strlcpy(pr->pr_name, namelc, sizeof(pr->pr_name)); else snprintf(pr->pr_name, sizeof(pr->pr_name), "%s.%s", ppr->pr_name, namelc); /* Change this component of child names. */ FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { bcopy(tpr->pr_name + onamelen, tpr->pr_name + namelen, strlen(tpr->pr_name + onamelen) + 1); bcopy(pr->pr_name, tpr->pr_name, namelen); } } if (path != NULL) { /* Try to keep a real-rooted full pathname. */ strlcpy(pr->pr_path, path, sizeof(pr->pr_path)); pr->pr_root = root; } if (PR_HOST & ch_flags & ~pr_flags) { if (pr->pr_flags & PR_HOST) { /* * Copy the parent's host info. As with pr_ip4 above, * the lack of a lock on the parent is not a problem; * it is always set with allprison_lock at least * shared, and is held exclusively here. */ strlcpy(pr->pr_hostname, pr->pr_parent->pr_hostname, sizeof(pr->pr_hostname)); strlcpy(pr->pr_domainname, pr->pr_parent->pr_domainname, sizeof(pr->pr_domainname)); strlcpy(pr->pr_hostuuid, pr->pr_parent->pr_hostuuid, sizeof(pr->pr_hostuuid)); pr->pr_hostid = pr->pr_parent->pr_hostid; } } else if (host != NULL || domain != NULL || uuid != NULL || gothid) { /* Set this prison, and any descendants without PR_HOST. */ if (host != NULL) strlcpy(pr->pr_hostname, host, sizeof(pr->pr_hostname)); if (domain != NULL) strlcpy(pr->pr_domainname, domain, sizeof(pr->pr_domainname)); if (uuid != NULL) strlcpy(pr->pr_hostuuid, uuid, sizeof(pr->pr_hostuuid)); if (gothid) pr->pr_hostid = hid; FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { if (tpr->pr_flags & PR_HOST) descend = 0; else { if (host != NULL) strlcpy(tpr->pr_hostname, pr->pr_hostname, sizeof(tpr->pr_hostname)); if (domain != NULL) strlcpy(tpr->pr_domainname, pr->pr_domainname, sizeof(tpr->pr_domainname)); if (uuid != NULL) strlcpy(tpr->pr_hostuuid, pr->pr_hostuuid, sizeof(tpr->pr_hostuuid)); if (gothid) tpr->pr_hostid = hid; } } } if ((tallow = ch_allow & ~pr_allow)) { /* Clear allow bits in all children. */ FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) tpr->pr_allow &= ~tallow; } pr->pr_allow = (pr->pr_allow & ~ch_allow) | pr_allow; /* * Persistent prisons get an extra reference, and prisons losing their * persist flag lose that reference. Only do this for existing prisons * for now, so new ones will remain unseen until after the module * handlers have completed. */ born = pr->pr_uref == 0; if (!created && (ch_flags & PR_PERSIST & (pr_flags ^ pr->pr_flags))) { if (pr_flags & PR_PERSIST) { pr->pr_ref++; pr->pr_uref++; } else { pr->pr_ref--; pr->pr_uref--; } } pr->pr_flags = (pr->pr_flags & ~ch_flags) | pr_flags; mtx_unlock(&pr->pr_mtx); #ifdef RACCT if (racct_enable && created) prison_racct_attach(pr); #endif /* Locks may have prevented a complete restriction of child IP * addresses. If so, allocate some more memory and try again. */ #ifdef INET while (redo_ip4) { ip4s = pr->pr_ip4s; ip4 = malloc(ip4s * sizeof(*ip4), M_PRISON, M_WAITOK); mtx_lock(&pr->pr_mtx); redo_ip4 = 0; FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { #ifdef VIMAGE if (tpr->pr_flags & PR_VNET) { descend = 0; continue; } #endif if (prison_restrict_ip4(tpr, ip4)) { if (ip4 != NULL) ip4 = NULL; else redo_ip4 = 1; } } mtx_unlock(&pr->pr_mtx); } #endif #ifdef INET6 while (redo_ip6) { ip6s = pr->pr_ip6s; ip6 = malloc(ip6s * sizeof(*ip6), M_PRISON, M_WAITOK); mtx_lock(&pr->pr_mtx); redo_ip6 = 0; FOREACH_PRISON_DESCENDANT_LOCKED(pr, tpr, descend) { #ifdef VIMAGE if (tpr->pr_flags & PR_VNET) { descend = 0; continue; } #endif if (prison_restrict_ip6(tpr, ip6)) { if (ip6 != NULL) ip6 = NULL; else redo_ip6 = 1; } } mtx_unlock(&pr->pr_mtx); } #endif /* Let the modules do their work. */ sx_downgrade(&allprison_lock); if (born) { error = osd_jail_call(pr, PR_METHOD_CREATE, opts); if (error) { (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); prison_deref(pr, created ? PD_LIST_SLOCKED : PD_DEREF | PD_LIST_SLOCKED); goto done_errmsg; } } error = osd_jail_call(pr, PR_METHOD_SET, opts); if (error) { if (born) (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); prison_deref(pr, created ? PD_LIST_SLOCKED : PD_DEREF | PD_LIST_SLOCKED); goto done_errmsg; } /* Attach this process to the prison if requested. */ if (flags & JAIL_ATTACH) { mtx_lock(&pr->pr_mtx); error = do_jail_attach(td, pr); if (error) { vfs_opterror(opts, "attach failed"); if (!created) prison_deref(pr, PD_DEREF); goto done_errmsg; } } #ifdef RACCT if (racct_enable && !created) { if (!(flags & JAIL_ATTACH)) sx_sunlock(&allprison_lock); prison_racct_modify(pr); if (!(flags & JAIL_ATTACH)) sx_slock(&allprison_lock); } #endif td->td_retval[0] = pr->pr_id; /* * Now that it is all there, drop the temporary reference from existing * prisons. Or add a reference to newly created persistent prisons * (which was not done earlier so that the prison would not be publicly * visible). */ if (!created) { prison_deref(pr, (flags & JAIL_ATTACH) ? PD_DEREF : PD_DEREF | PD_LIST_SLOCKED); } else { if (pr_flags & PR_PERSIST) { mtx_lock(&pr->pr_mtx); pr->pr_ref++; pr->pr_uref++; mtx_unlock(&pr->pr_mtx); } if (!(flags & JAIL_ATTACH)) sx_sunlock(&allprison_lock); } goto done_free; done_deref_locked: prison_deref(pr, created ? PD_LOCKED | PD_LIST_XLOCKED : PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED); goto done_releroot; done_unlock_list: sx_xunlock(&allprison_lock); done_releroot: if (root != NULL) vrele(root); done_errmsg: if (error) { if (vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len) == 0 && errmsg_len > 0) { errmsg_pos = 2 * vfs_getopt_pos(opts, "errmsg") + 1; if (optuio->uio_segflg == UIO_SYSSPACE) bcopy(errmsg, optuio->uio_iov[errmsg_pos].iov_base, errmsg_len); else copyout(errmsg, optuio->uio_iov[errmsg_pos].iov_base, errmsg_len); } } done_free: #ifdef INET free(ip4, M_PRISON); #endif #ifdef INET6 free(ip6, M_PRISON); #endif if (g_path != NULL) free(g_path, M_TEMP); vfs_freeopts(opts); return (error); } /* * struct jail_get_args { * struct iovec *iovp; * unsigned int iovcnt; * int flags; * }; */ int sys_jail_get(struct thread *td, struct jail_get_args *uap) { struct uio *auio; int error; /* Check that we have an even number of iovecs. */ if (uap->iovcnt & 1) return (EINVAL); error = copyinuio(uap->iovp, uap->iovcnt, &auio); if (error) return (error); error = kern_jail_get(td, auio, uap->flags); if (error == 0) error = copyout(auio->uio_iov, uap->iovp, uap->iovcnt * sizeof (struct iovec)); free(auio, M_IOV); return (error); } int kern_jail_get(struct thread *td, struct uio *optuio, int flags) { struct bool_flags *bf; struct jailsys_flags *jsf; struct prison *pr, *mypr; struct vfsopt *opt; struct vfsoptlist *opts; char *errmsg, *name; int error, errmsg_len, errmsg_pos, i, jid, len, locked, pos; unsigned f; if (flags & ~JAIL_GET_MASK) return (EINVAL); /* Get the parameter list. */ error = vfs_buildopts(optuio, &opts); if (error) return (error); errmsg_pos = vfs_getopt_pos(opts, "errmsg"); mypr = td->td_ucred->cr_prison; /* * Find the prison specified by one of: lastjid, jid, name. */ sx_slock(&allprison_lock); error = vfs_copyopt(opts, "lastjid", &jid, sizeof(jid)); if (error == 0) { TAILQ_FOREACH(pr, &allprison, pr_list) { if (pr->pr_id > jid && prison_ischild(mypr, pr)) { mtx_lock(&pr->pr_mtx); if (pr->pr_ref > 0 && (pr->pr_uref > 0 || (flags & JAIL_DYING))) break; mtx_unlock(&pr->pr_mtx); } } if (pr != NULL) goto found_prison; error = ENOENT; vfs_opterror(opts, "no jail after %d", jid); goto done_unlock_list; } else if (error != ENOENT) goto done_unlock_list; error = vfs_copyopt(opts, "jid", &jid, sizeof(jid)); if (error == 0) { if (jid != 0) { pr = prison_find_child(mypr, jid); if (pr != NULL) { if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) { mtx_unlock(&pr->pr_mtx); error = ENOENT; vfs_opterror(opts, "jail %d is dying", jid); goto done_unlock_list; } goto found_prison; } error = ENOENT; vfs_opterror(opts, "jail %d not found", jid); goto done_unlock_list; } } else if (error != ENOENT) goto done_unlock_list; error = vfs_getopt(opts, "name", (void **)&name, &len); if (error == 0) { if (len == 0 || name[len - 1] != '\0') { error = EINVAL; goto done_unlock_list; } pr = prison_find_name(mypr, name); if (pr != NULL) { if (pr->pr_uref == 0 && !(flags & JAIL_DYING)) { mtx_unlock(&pr->pr_mtx); error = ENOENT; vfs_opterror(opts, "jail \"%s\" is dying", name); goto done_unlock_list; } goto found_prison; } error = ENOENT; vfs_opterror(opts, "jail \"%s\" not found", name); goto done_unlock_list; } else if (error != ENOENT) goto done_unlock_list; vfs_opterror(opts, "no jail specified"); error = ENOENT; goto done_unlock_list; found_prison: /* Get the parameters of the prison. */ pr->pr_ref++; locked = PD_LOCKED; td->td_retval[0] = pr->pr_id; error = vfs_setopt(opts, "jid", &pr->pr_id, sizeof(pr->pr_id)); if (error != 0 && error != ENOENT) goto done_deref; i = (pr->pr_parent == mypr) ? 0 : pr->pr_parent->pr_id; error = vfs_setopt(opts, "parent", &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopts(opts, "name", prison_name(mypr, pr)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopt(opts, "cpuset.id", &pr->pr_cpuset->cs_id, sizeof(pr->pr_cpuset->cs_id)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopts(opts, "path", prison_path(mypr, pr)); if (error != 0 && error != ENOENT) goto done_deref; #ifdef INET error = vfs_setopt_part(opts, "ip4.addr", pr->pr_ip4, pr->pr_ip4s * sizeof(*pr->pr_ip4)); if (error != 0 && error != ENOENT) goto done_deref; #endif #ifdef INET6 error = vfs_setopt_part(opts, "ip6.addr", pr->pr_ip6, pr->pr_ip6s * sizeof(*pr->pr_ip6)); if (error != 0 && error != ENOENT) goto done_deref; #endif error = vfs_setopt(opts, "securelevel", &pr->pr_securelevel, sizeof(pr->pr_securelevel)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopt(opts, "children.cur", &pr->pr_childcount, sizeof(pr->pr_childcount)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopt(opts, "children.max", &pr->pr_childmax, sizeof(pr->pr_childmax)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopts(opts, "host.hostname", pr->pr_hostname); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopts(opts, "host.domainname", pr->pr_domainname); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopts(opts, "host.hostuuid", pr->pr_hostuuid); if (error != 0 && error != ENOENT) goto done_deref; #ifdef COMPAT_FREEBSD32 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { uint32_t hid32 = pr->pr_hostid; error = vfs_setopt(opts, "host.hostid", &hid32, sizeof(hid32)); } else #endif error = vfs_setopt(opts, "host.hostid", &pr->pr_hostid, sizeof(pr->pr_hostid)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopt(opts, "enforce_statfs", &pr->pr_enforce_statfs, sizeof(pr->pr_enforce_statfs)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopt(opts, "devfs_ruleset", &pr->pr_devfs_rsnum, sizeof(pr->pr_devfs_rsnum)); if (error != 0 && error != ENOENT) goto done_deref; for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++) { i = (pr->pr_flags & bf->flag) ? 1 : 0; error = vfs_setopt(opts, bf->name, &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; i = !i; error = vfs_setopt(opts, bf->noname, &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; } for (jsf = pr_flag_jailsys; jsf < pr_flag_jailsys + nitems(pr_flag_jailsys); jsf++) { f = pr->pr_flags & (jsf->disable | jsf->new); i = (f != 0 && f == jsf->disable) ? JAIL_SYS_DISABLE : (f == jsf->new) ? JAIL_SYS_NEW : JAIL_SYS_INHERIT; error = vfs_setopt(opts, jsf->name, &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; } for (bf = pr_flag_allow; bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0; bf++) { i = (pr->pr_allow & bf->flag) ? 1 : 0; error = vfs_setopt(opts, bf->name, &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; i = !i; error = vfs_setopt(opts, bf->noname, &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; } i = (pr->pr_uref == 0); error = vfs_setopt(opts, "dying", &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; i = !i; error = vfs_setopt(opts, "nodying", &i, sizeof(i)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopt(opts, "osreldate", &pr->pr_osreldate, sizeof(pr->pr_osreldate)); if (error != 0 && error != ENOENT) goto done_deref; error = vfs_setopts(opts, "osrelease", pr->pr_osrelease); if (error != 0 && error != ENOENT) goto done_deref; /* Get the module parameters. */ mtx_unlock(&pr->pr_mtx); locked = 0; error = osd_jail_call(pr, PR_METHOD_GET, opts); if (error) goto done_deref; prison_deref(pr, PD_DEREF | PD_LIST_SLOCKED); /* By now, all parameters should have been noted. */ TAILQ_FOREACH(opt, opts, link) { if (!opt->seen && strcmp(opt->name, "errmsg")) { error = EINVAL; vfs_opterror(opts, "unknown parameter: %s", opt->name); goto done_errmsg; } } /* Write the fetched parameters back to userspace. */ error = 0; TAILQ_FOREACH(opt, opts, link) { if (opt->pos >= 0 && opt->pos != errmsg_pos) { pos = 2 * opt->pos + 1; optuio->uio_iov[pos].iov_len = opt->len; if (opt->value != NULL) { if (optuio->uio_segflg == UIO_SYSSPACE) { bcopy(opt->value, optuio->uio_iov[pos].iov_base, opt->len); } else { error = copyout(opt->value, optuio->uio_iov[pos].iov_base, opt->len); if (error) break; } } } } goto done_errmsg; done_deref: prison_deref(pr, locked | PD_DEREF | PD_LIST_SLOCKED); goto done_errmsg; done_unlock_list: sx_sunlock(&allprison_lock); done_errmsg: if (error && errmsg_pos >= 0) { vfs_getopt(opts, "errmsg", (void **)&errmsg, &errmsg_len); errmsg_pos = 2 * errmsg_pos + 1; if (errmsg_len > 0) { if (optuio->uio_segflg == UIO_SYSSPACE) bcopy(errmsg, optuio->uio_iov[errmsg_pos].iov_base, errmsg_len); else copyout(errmsg, optuio->uio_iov[errmsg_pos].iov_base, errmsg_len); } } vfs_freeopts(opts); return (error); } /* * struct jail_remove_args { * int jid; * }; */ int sys_jail_remove(struct thread *td, struct jail_remove_args *uap) { struct prison *pr, *cpr, *lpr, *tpr; int descend, error; error = priv_check(td, PRIV_JAIL_REMOVE); if (error) return (error); sx_xlock(&allprison_lock); pr = prison_find_child(td->td_ucred->cr_prison, uap->jid); if (pr == NULL) { sx_xunlock(&allprison_lock); return (EINVAL); } /* Remove all descendants of this prison, then remove this prison. */ pr->pr_ref++; if (!LIST_EMPTY(&pr->pr_children)) { mtx_unlock(&pr->pr_mtx); lpr = NULL; FOREACH_PRISON_DESCENDANT(pr, cpr, descend) { mtx_lock(&cpr->pr_mtx); if (cpr->pr_ref > 0) { tpr = cpr; cpr->pr_ref++; } else { /* Already removed - do not do it again. */ tpr = NULL; } mtx_unlock(&cpr->pr_mtx); if (lpr != NULL) { mtx_lock(&lpr->pr_mtx); prison_remove_one(lpr); sx_xlock(&allprison_lock); } lpr = tpr; } if (lpr != NULL) { mtx_lock(&lpr->pr_mtx); prison_remove_one(lpr); sx_xlock(&allprison_lock); } mtx_lock(&pr->pr_mtx); } prison_remove_one(pr); return (0); } static void prison_remove_one(struct prison *pr) { struct proc *p; int deuref; /* If the prison was persistent, it is not anymore. */ deuref = 0; if (pr->pr_flags & PR_PERSIST) { pr->pr_ref--; deuref = PD_DEUREF; pr->pr_flags &= ~PR_PERSIST; } /* * jail_remove added a reference. If that's the only one, remove * the prison now. */ KASSERT(pr->pr_ref > 0, ("prison_remove_one removing a dead prison (jid=%d)", pr->pr_id)); if (pr->pr_ref == 1) { prison_deref(pr, deuref | PD_DEREF | PD_LOCKED | PD_LIST_XLOCKED); return; } mtx_unlock(&pr->pr_mtx); sx_xunlock(&allprison_lock); /* * Kill all processes unfortunate enough to be attached to this prison. */ sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state != PRS_NEW && p->p_ucred && p->p_ucred->cr_prison == pr) kern_psignal(p, SIGKILL); PROC_UNLOCK(p); } sx_sunlock(&allproc_lock); /* Remove the temporary reference added by jail_remove. */ prison_deref(pr, deuref | PD_DEREF); } /* * struct jail_attach_args { * int jid; * }; */ int sys_jail_attach(struct thread *td, struct jail_attach_args *uap) { struct prison *pr; int error; error = priv_check(td, PRIV_JAIL_ATTACH); if (error) return (error); /* * Start with exclusive hold on allprison_lock to ensure that a possible * PR_METHOD_REMOVE call isn't concurrent with jail_set or jail_remove. * But then immediately downgrade it since we don't need to stop * readers. */ sx_xlock(&allprison_lock); sx_downgrade(&allprison_lock); pr = prison_find_child(td->td_ucred->cr_prison, uap->jid); if (pr == NULL) { sx_sunlock(&allprison_lock); return (EINVAL); } /* * Do not allow a process to attach to a prison that is not * considered to be "alive". */ if (pr->pr_uref == 0) { mtx_unlock(&pr->pr_mtx); sx_sunlock(&allprison_lock); return (EINVAL); } return (do_jail_attach(td, pr)); } static int do_jail_attach(struct thread *td, struct prison *pr) { struct proc *p; struct ucred *newcred, *oldcred; int error; /* * XXX: Note that there is a slight race here if two threads * in the same privileged process attempt to attach to two * different jails at the same time. It is important for * user processes not to do this, or they might end up with * a process root from one prison, but attached to the jail * of another. */ pr->pr_ref++; pr->pr_uref++; mtx_unlock(&pr->pr_mtx); /* Let modules do whatever they need to prepare for attaching. */ error = osd_jail_call(pr, PR_METHOD_ATTACH, td); if (error) { prison_deref(pr, PD_DEREF | PD_DEUREF | PD_LIST_SLOCKED); return (error); } sx_sunlock(&allprison_lock); /* * Reparent the newly attached process to this jail. */ p = td->td_proc; error = cpuset_setproc_update_set(p, pr->pr_cpuset); if (error) goto e_revert_osd; vn_lock(pr->pr_root, LK_EXCLUSIVE | LK_RETRY); if ((error = change_dir(pr->pr_root, td)) != 0) goto e_unlock; #ifdef MAC if ((error = mac_vnode_check_chroot(td->td_ucred, pr->pr_root))) goto e_unlock; #endif VOP_UNLOCK(pr->pr_root); if ((error = pwd_chroot(td, pr->pr_root))) goto e_revert_osd; newcred = crget(); PROC_LOCK(p); oldcred = crcopysafe(p, newcred); newcred->cr_prison = pr; proc_set_cred(p, newcred); setsugid(p); #ifdef RACCT racct_proc_ucred_changed(p, oldcred, newcred); crhold(newcred); #endif PROC_UNLOCK(p); #ifdef RCTL rctl_proc_ucred_changed(p, newcred); crfree(newcred); #endif prison_deref(oldcred->cr_prison, PD_DEREF | PD_DEUREF); crfree(oldcred); return (0); e_unlock: VOP_UNLOCK(pr->pr_root); e_revert_osd: /* Tell modules this thread is still in its old jail after all. */ (void)osd_jail_call(td->td_ucred->cr_prison, PR_METHOD_ATTACH, td); prison_deref(pr, PD_DEREF | PD_DEUREF); return (error); } /* * Returns a locked prison instance, or NULL on failure. */ struct prison * prison_find(int prid) { struct prison *pr; sx_assert(&allprison_lock, SX_LOCKED); TAILQ_FOREACH(pr, &allprison, pr_list) { if (pr->pr_id == prid) { mtx_lock(&pr->pr_mtx); if (pr->pr_ref > 0) return (pr); mtx_unlock(&pr->pr_mtx); } } return (NULL); } /* * Find a prison that is a descendant of mypr. Returns a locked prison or NULL. */ struct prison * prison_find_child(struct prison *mypr, int prid) { struct prison *pr; int descend; sx_assert(&allprison_lock, SX_LOCKED); FOREACH_PRISON_DESCENDANT(mypr, pr, descend) { if (pr->pr_id == prid) { mtx_lock(&pr->pr_mtx); if (pr->pr_ref > 0) return (pr); mtx_unlock(&pr->pr_mtx); } } return (NULL); } /* * Look for the name relative to mypr. Returns a locked prison or NULL. */ struct prison * prison_find_name(struct prison *mypr, const char *name) { struct prison *pr, *deadpr; size_t mylen; int descend; sx_assert(&allprison_lock, SX_LOCKED); mylen = (mypr == &prison0) ? 0 : strlen(mypr->pr_name) + 1; again: deadpr = NULL; FOREACH_PRISON_DESCENDANT(mypr, pr, descend) { if (!strcmp(pr->pr_name + mylen, name)) { mtx_lock(&pr->pr_mtx); if (pr->pr_ref > 0) { if (pr->pr_uref > 0) return (pr); deadpr = pr; } mtx_unlock(&pr->pr_mtx); } } /* There was no valid prison - perhaps there was a dying one. */ if (deadpr != NULL) { mtx_lock(&deadpr->pr_mtx); if (deadpr->pr_ref == 0) { mtx_unlock(&deadpr->pr_mtx); goto again; } } return (deadpr); } /* * See if a prison has the specific flag set. */ int prison_flag(struct ucred *cred, unsigned flag) { /* This is an atomic read, so no locking is necessary. */ return (cred->cr_prison->pr_flags & flag); } int prison_allow(struct ucred *cred, unsigned flag) { /* This is an atomic read, so no locking is necessary. */ return (cred->cr_prison->pr_allow & flag); } /* * Remove a prison reference. If that was the last reference, remove the * prison itself - but not in this context in case there are locks held. */ void prison_free_locked(struct prison *pr) { int ref; mtx_assert(&pr->pr_mtx, MA_OWNED); ref = --pr->pr_ref; mtx_unlock(&pr->pr_mtx); if (ref == 0) taskqueue_enqueue(taskqueue_thread, &pr->pr_task); } void prison_free(struct prison *pr) { mtx_lock(&pr->pr_mtx); prison_free_locked(pr); } /* * Complete a call to either prison_free or prison_proc_free. */ static void prison_complete(void *context, int pending) { struct prison *pr = context; sx_xlock(&allprison_lock); mtx_lock(&pr->pr_mtx); prison_deref(pr, pr->pr_uref ? PD_DEREF | PD_DEUREF | PD_LOCKED | PD_LIST_XLOCKED : PD_LOCKED | PD_LIST_XLOCKED); } /* * Remove a prison reference (usually). This internal version assumes no * mutexes are held, except perhaps the prison itself. If there are no more * references, release and delist the prison. On completion, the prison lock * and the allprison lock are both unlocked. */ static void prison_deref(struct prison *pr, int flags) { struct prison *ppr, *tpr; int ref, lasturef; if (!(flags & PD_LOCKED)) mtx_lock(&pr->pr_mtx); for (;;) { if (flags & PD_DEUREF) { KASSERT(pr->pr_uref > 0, ("prison_deref PD_DEUREF on a dead prison (jid=%d)", pr->pr_id)); pr->pr_uref--; lasturef = pr->pr_uref == 0; if (lasturef) pr->pr_ref++; KASSERT(prison0.pr_uref != 0, ("prison0 pr_uref=0")); } else lasturef = 0; if (flags & PD_DEREF) { KASSERT(pr->pr_ref > 0, ("prison_deref PD_DEREF on a dead prison (jid=%d)", pr->pr_id)); pr->pr_ref--; } ref = pr->pr_ref; mtx_unlock(&pr->pr_mtx); /* * Tell the modules if the last user reference was removed * (even it sticks around in dying state). */ if (lasturef) { if (!(flags & (PD_LIST_SLOCKED | PD_LIST_XLOCKED))) { sx_xlock(&allprison_lock); flags |= PD_LIST_XLOCKED; } (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); mtx_lock(&pr->pr_mtx); ref = --pr->pr_ref; mtx_unlock(&pr->pr_mtx); } /* If the prison still has references, nothing else to do. */ if (ref > 0) { if (flags & PD_LIST_SLOCKED) sx_sunlock(&allprison_lock); else if (flags & PD_LIST_XLOCKED) sx_xunlock(&allprison_lock); return; } if (flags & PD_LIST_SLOCKED) { if (!sx_try_upgrade(&allprison_lock)) { sx_sunlock(&allprison_lock); sx_xlock(&allprison_lock); } } else if (!(flags & PD_LIST_XLOCKED)) sx_xlock(&allprison_lock); TAILQ_REMOVE(&allprison, pr, pr_list); LIST_REMOVE(pr, pr_sibling); ppr = pr->pr_parent; for (tpr = ppr; tpr != NULL; tpr = tpr->pr_parent) tpr->pr_childcount--; sx_xunlock(&allprison_lock); #ifdef VIMAGE if (pr->pr_vnet != ppr->pr_vnet) vnet_destroy(pr->pr_vnet); #endif if (pr->pr_root != NULL) vrele(pr->pr_root); mtx_destroy(&pr->pr_mtx); #ifdef INET free(pr->pr_ip4, M_PRISON); #endif #ifdef INET6 free(pr->pr_ip6, M_PRISON); #endif if (pr->pr_cpuset != NULL) cpuset_rel(pr->pr_cpuset); osd_jail_exit(pr); #ifdef RACCT if (racct_enable) prison_racct_detach(pr); #endif free(pr, M_PRISON); /* Removing a prison frees a reference on its parent. */ pr = ppr; mtx_lock(&pr->pr_mtx); flags = PD_DEREF | PD_DEUREF; } } void prison_hold_locked(struct prison *pr) { mtx_assert(&pr->pr_mtx, MA_OWNED); KASSERT(pr->pr_ref > 0, ("Trying to hold dead prison %p (jid=%d).", pr, pr->pr_id)); pr->pr_ref++; } void prison_hold(struct prison *pr) { mtx_lock(&pr->pr_mtx); prison_hold_locked(pr); mtx_unlock(&pr->pr_mtx); } void prison_proc_hold(struct prison *pr) { mtx_lock(&pr->pr_mtx); KASSERT(pr->pr_uref > 0, ("Cannot add a process to a non-alive prison (jid=%d)", pr->pr_id)); pr->pr_uref++; mtx_unlock(&pr->pr_mtx); } void prison_proc_free(struct prison *pr) { mtx_lock(&pr->pr_mtx); KASSERT(pr->pr_uref > 0, ("Trying to kill a process in a dead prison (jid=%d)", pr->pr_id)); if (pr->pr_uref > 1) pr->pr_uref--; else { /* * Don't remove the last user reference in this context, which * is expected to be a process that is not only locked, but * also half dead. */ pr->pr_ref++; mtx_unlock(&pr->pr_mtx); taskqueue_enqueue(taskqueue_thread, &pr->pr_task); return; } mtx_unlock(&pr->pr_mtx); } /* * Check if a jail supports the given address family. * * Returns 0 if not jailed or the address family is supported, EAFNOSUPPORT * if not. */ int prison_check_af(struct ucred *cred, int af) { struct prison *pr; int error; KASSERT(cred != NULL, ("%s: cred is NULL", __func__)); pr = cred->cr_prison; #ifdef VIMAGE /* Prisons with their own network stack are not limited. */ if (prison_owns_vnet(cred)) return (0); #endif error = 0; switch (af) { #ifdef INET case AF_INET: if (pr->pr_flags & PR_IP4) { mtx_lock(&pr->pr_mtx); if ((pr->pr_flags & PR_IP4) && pr->pr_ip4 == NULL) error = EAFNOSUPPORT; mtx_unlock(&pr->pr_mtx); } break; #endif #ifdef INET6 case AF_INET6: if (pr->pr_flags & PR_IP6) { mtx_lock(&pr->pr_mtx); if ((pr->pr_flags & PR_IP6) && pr->pr_ip6 == NULL) error = EAFNOSUPPORT; mtx_unlock(&pr->pr_mtx); } break; #endif case AF_LOCAL: case AF_ROUTE: break; default: if (!(pr->pr_allow & PR_ALLOW_SOCKET_AF)) error = EAFNOSUPPORT; } return (error); } /* * Check if given address belongs to the jail referenced by cred (wrapper to * prison_check_ip[46]). * * Returns 0 if jail doesn't restrict the address family or if address belongs * to jail, EADDRNOTAVAIL if the address doesn't belong, or EAFNOSUPPORT if * the jail doesn't allow the address family. IPv4 Address passed in in NBO. */ int prison_if(struct ucred *cred, const struct sockaddr *sa) { #ifdef INET const struct sockaddr_in *sai; #endif #ifdef INET6 const struct sockaddr_in6 *sai6; #endif int error; KASSERT(cred != NULL, ("%s: cred is NULL", __func__)); KASSERT(sa != NULL, ("%s: sa is NULL", __func__)); #ifdef VIMAGE if (prison_owns_vnet(cred)) return (0); #endif error = 0; switch (sa->sa_family) { #ifdef INET case AF_INET: sai = (const struct sockaddr_in *)sa; error = prison_check_ip4(cred, &sai->sin_addr); break; #endif #ifdef INET6 case AF_INET6: sai6 = (const struct sockaddr_in6 *)sa; error = prison_check_ip6(cred, &sai6->sin6_addr); break; #endif default: if (!(cred->cr_prison->pr_allow & PR_ALLOW_SOCKET_AF)) error = EAFNOSUPPORT; } return (error); } /* * Return 0 if jails permit p1 to frob p2, otherwise ESRCH. */ int prison_check(struct ucred *cred1, struct ucred *cred2) { return ((cred1->cr_prison == cred2->cr_prison || prison_ischild(cred1->cr_prison, cred2->cr_prison)) ? 0 : ESRCH); } /* * Return 1 if p2 is a child of p1, otherwise 0. */ int prison_ischild(struct prison *pr1, struct prison *pr2) { for (pr2 = pr2->pr_parent; pr2 != NULL; pr2 = pr2->pr_parent) if (pr1 == pr2) return (1); return (0); } /* * Return 1 if the passed credential is in a jail and that jail does not * have its own virtual network stack, otherwise 0. */ int jailed_without_vnet(struct ucred *cred) { if (!jailed(cred)) return (0); #ifdef VIMAGE if (prison_owns_vnet(cred)) return (0); #endif return (1); } /* * Return the correct hostname (domainname, et al) for the passed credential. */ void getcredhostname(struct ucred *cred, char *buf, size_t size) { struct prison *pr; /* * A NULL credential can be used to shortcut to the physical * system's hostname. */ pr = (cred != NULL) ? cred->cr_prison : &prison0; mtx_lock(&pr->pr_mtx); strlcpy(buf, pr->pr_hostname, size); mtx_unlock(&pr->pr_mtx); } void getcreddomainname(struct ucred *cred, char *buf, size_t size) { mtx_lock(&cred->cr_prison->pr_mtx); strlcpy(buf, cred->cr_prison->pr_domainname, size); mtx_unlock(&cred->cr_prison->pr_mtx); } void getcredhostuuid(struct ucred *cred, char *buf, size_t size) { mtx_lock(&cred->cr_prison->pr_mtx); strlcpy(buf, cred->cr_prison->pr_hostuuid, size); mtx_unlock(&cred->cr_prison->pr_mtx); } void getcredhostid(struct ucred *cred, unsigned long *hostid) { mtx_lock(&cred->cr_prison->pr_mtx); *hostid = cred->cr_prison->pr_hostid; mtx_unlock(&cred->cr_prison->pr_mtx); } void getjailname(struct ucred *cred, char *name, size_t len) { mtx_lock(&cred->cr_prison->pr_mtx); strlcpy(name, cred->cr_prison->pr_name, len); mtx_unlock(&cred->cr_prison->pr_mtx); } #ifdef VIMAGE /* * Determine whether the prison represented by cred owns * its vnet rather than having it inherited. * * Returns 1 in case the prison owns the vnet, 0 otherwise. */ int prison_owns_vnet(struct ucred *cred) { /* * vnets cannot be added/removed after jail creation, * so no need to lock here. */ return (cred->cr_prison->pr_flags & PR_VNET ? 1 : 0); } #endif /* * Determine whether the subject represented by cred can "see" * status of a mount point. * Returns: 0 for permitted, ENOENT otherwise. * XXX: This function should be called cr_canseemount() and should be * placed in kern_prot.c. */ int prison_canseemount(struct ucred *cred, struct mount *mp) { struct prison *pr; struct statfs *sp; size_t len; pr = cred->cr_prison; if (pr->pr_enforce_statfs == 0) return (0); if (pr->pr_root->v_mount == mp) return (0); if (pr->pr_enforce_statfs == 2) return (ENOENT); /* * If jail's chroot directory is set to "/" we should be able to see * all mount-points from inside a jail. * This is ugly check, but this is the only situation when jail's * directory ends with '/'. */ if (strcmp(pr->pr_path, "/") == 0) return (0); len = strlen(pr->pr_path); sp = &mp->mnt_stat; if (strncmp(pr->pr_path, sp->f_mntonname, len) != 0) return (ENOENT); /* * Be sure that we don't have situation where jail's root directory * is "/some/path" and mount point is "/some/pathpath". */ if (sp->f_mntonname[len] != '\0' && sp->f_mntonname[len] != '/') return (ENOENT); return (0); } void prison_enforce_statfs(struct ucred *cred, struct mount *mp, struct statfs *sp) { char jpath[MAXPATHLEN]; struct prison *pr; size_t len; pr = cred->cr_prison; if (pr->pr_enforce_statfs == 0) return; if (prison_canseemount(cred, mp) != 0) { bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); strlcpy(sp->f_mntonname, "[restricted]", sizeof(sp->f_mntonname)); return; } if (pr->pr_root->v_mount == mp) { /* * Clear current buffer data, so we are sure nothing from * the valid path left there. */ bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); *sp->f_mntonname = '/'; return; } /* * If jail's chroot directory is set to "/" we should be able to see * all mount-points from inside a jail. */ if (strcmp(pr->pr_path, "/") == 0) return; len = strlen(pr->pr_path); strlcpy(jpath, sp->f_mntonname + len, sizeof(jpath)); /* * Clear current buffer data, so we are sure nothing from * the valid path left there. */ bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); if (*jpath == '\0') { /* Should never happen. */ *sp->f_mntonname = '/'; } else { strlcpy(sp->f_mntonname, jpath, sizeof(sp->f_mntonname)); } } /* * Check with permission for a specific privilege is granted within jail. We * have a specific list of accepted privileges; the rest are denied. */ int prison_priv_check(struct ucred *cred, int priv) { /* * Some policies have custom handlers. This routine should not be * called for them. See priv_check_cred(). */ switch (priv) { + case PRIV_VFS_LOOKUP: case PRIV_VFS_GENERATION: KASSERT(0, ("prison_priv_check instead of a custom handler " "called for %d\n", priv)); } if (!jailed(cred)) return (0); #ifdef VIMAGE /* * Privileges specific to prisons with a virtual network stack. * There might be a duplicate entry here in case the privilege * is only granted conditionally in the legacy jail case. */ switch (priv) { #ifdef notyet /* * NFS-specific privileges. */ case PRIV_NFS_DAEMON: case PRIV_NFS_LOCKD: #endif /* * Network stack privileges. */ case PRIV_NET_BRIDGE: case PRIV_NET_GRE: case PRIV_NET_BPF: case PRIV_NET_RAW: /* Dup, cond. in legacy jail case. */ case PRIV_NET_ROUTE: case PRIV_NET_TAP: case PRIV_NET_SETIFMTU: case PRIV_NET_SETIFFLAGS: case PRIV_NET_SETIFCAP: case PRIV_NET_SETIFDESCR: case PRIV_NET_SETIFNAME : case PRIV_NET_SETIFMETRIC: case PRIV_NET_SETIFPHYS: case PRIV_NET_SETIFMAC: case PRIV_NET_SETLANPCP: case PRIV_NET_ADDMULTI: case PRIV_NET_DELMULTI: case PRIV_NET_HWIOCTL: case PRIV_NET_SETLLADDR: case PRIV_NET_ADDIFGROUP: case PRIV_NET_DELIFGROUP: case PRIV_NET_IFCREATE: case PRIV_NET_IFDESTROY: case PRIV_NET_ADDIFADDR: case PRIV_NET_DELIFADDR: case PRIV_NET_LAGG: case PRIV_NET_GIF: case PRIV_NET_SETIFVNET: case PRIV_NET_SETIFFIB: /* * 802.11-related privileges. */ case PRIV_NET80211_VAP_GETKEY: case PRIV_NET80211_VAP_MANAGE: #ifdef notyet /* * ATM privileges. */ case PRIV_NETATM_CFG: case PRIV_NETATM_ADD: case PRIV_NETATM_DEL: case PRIV_NETATM_SET: /* * Bluetooth privileges. */ case PRIV_NETBLUETOOTH_RAW: #endif /* * Netgraph and netgraph module privileges. */ case PRIV_NETGRAPH_CONTROL: #ifdef notyet case PRIV_NETGRAPH_TTY: #endif /* * IPv4 and IPv6 privileges. */ case PRIV_NETINET_IPFW: case PRIV_NETINET_DIVERT: case PRIV_NETINET_PF: case PRIV_NETINET_DUMMYNET: case PRIV_NETINET_CARP: case PRIV_NETINET_MROUTE: case PRIV_NETINET_RAW: case PRIV_NETINET_ADDRCTRL6: case PRIV_NETINET_ND6: case PRIV_NETINET_SCOPE6: case PRIV_NETINET_ALIFETIME6: case PRIV_NETINET_IPSEC: case PRIV_NETINET_BINDANY: #ifdef notyet /* * NCP privileges. */ case PRIV_NETNCP: /* * SMB privileges. */ case PRIV_NETSMB: #endif /* * No default: or deny here. * In case of no permit fall through to next switch(). */ if (cred->cr_prison->pr_flags & PR_VNET) return (0); } #endif /* VIMAGE */ switch (priv) { /* * Allow ktrace privileges for root in jail. */ case PRIV_KTRACE: #if 0 /* * Allow jailed processes to configure audit identity and * submit audit records (login, etc). In the future we may * want to further refine the relationship between audit and * jail. */ case PRIV_AUDIT_GETAUDIT: case PRIV_AUDIT_SETAUDIT: case PRIV_AUDIT_SUBMIT: #endif /* * Allow jailed processes to manipulate process UNIX * credentials in any way they see fit. */ case PRIV_CRED_SETUID: case PRIV_CRED_SETEUID: case PRIV_CRED_SETGID: case PRIV_CRED_SETEGID: case PRIV_CRED_SETGROUPS: case PRIV_CRED_SETREUID: case PRIV_CRED_SETREGID: case PRIV_CRED_SETRESUID: case PRIV_CRED_SETRESGID: /* * Jail implements visibility constraints already, so allow * jailed root to override uid/gid-based constraints. */ case PRIV_SEEOTHERGIDS: case PRIV_SEEOTHERUIDS: /* * Jail implements inter-process debugging limits already, so * allow jailed root various debugging privileges. */ case PRIV_DEBUG_DIFFCRED: case PRIV_DEBUG_SUGID: case PRIV_DEBUG_UNPRIV: /* * Allow jail to set various resource limits and login * properties, and for now, exceed process resource limits. */ case PRIV_PROC_LIMIT: case PRIV_PROC_SETLOGIN: case PRIV_PROC_SETRLIMIT: /* * System V and POSIX IPC privileges are granted in jail. */ case PRIV_IPC_READ: case PRIV_IPC_WRITE: case PRIV_IPC_ADMIN: case PRIV_IPC_MSGSIZE: case PRIV_MQ_ADMIN: /* * Jail operations within a jail work on child jails. */ case PRIV_JAIL_ATTACH: case PRIV_JAIL_SET: case PRIV_JAIL_REMOVE: /* * Jail implements its own inter-process limits, so allow * root processes in jail to change scheduling on other * processes in the same jail. Likewise for signalling. */ case PRIV_SCHED_DIFFCRED: case PRIV_SCHED_CPUSET: case PRIV_SIGNAL_DIFFCRED: case PRIV_SIGNAL_SUGID: /* * Allow jailed processes to write to sysctls marked as jail * writable. */ case PRIV_SYSCTL_WRITEJAIL: /* * Allow root in jail to manage a variety of quota * properties. These should likely be conditional on a * configuration option. */ case PRIV_VFS_GETQUOTA: case PRIV_VFS_SETQUOTA: /* * Since Jail relies on chroot() to implement file system * protections, grant many VFS privileges to root in jail. * Be careful to exclude mount-related and NFS-related * privileges. */ case PRIV_VFS_READ: case PRIV_VFS_WRITE: case PRIV_VFS_ADMIN: case PRIV_VFS_EXEC: - case PRIV_VFS_LOOKUP: case PRIV_VFS_BLOCKRESERVE: /* XXXRW: Slightly surprising. */ case PRIV_VFS_CHFLAGS_DEV: case PRIV_VFS_CHOWN: case PRIV_VFS_CHROOT: case PRIV_VFS_RETAINSUGID: case PRIV_VFS_FCHROOT: case PRIV_VFS_LINK: case PRIV_VFS_SETGID: case PRIV_VFS_STAT: case PRIV_VFS_STICKYFILE: /* * As in the non-jail case, non-root users are expected to be * able to read kernel/phyiscal memory (provided /dev/[k]mem * exists in the jail and they have permission to access it). */ case PRIV_KMEM_READ: return (0); /* * Depending on the global setting, allow privilege of * setting system flags. */ case PRIV_VFS_SYSFLAGS: if (cred->cr_prison->pr_allow & PR_ALLOW_CHFLAGS) return (0); else return (EPERM); /* * Depending on the global setting, allow privilege of * mounting/unmounting file systems. */ case PRIV_VFS_MOUNT: case PRIV_VFS_UNMOUNT: case PRIV_VFS_MOUNT_NONUSER: case PRIV_VFS_MOUNT_OWNER: if (cred->cr_prison->pr_allow & PR_ALLOW_MOUNT && cred->cr_prison->pr_enforce_statfs < 2) return (0); else return (EPERM); /* * Jails should hold no disposition on the PRIV_VFS_READ_DIR * policy. priv_check_cred will not specifically allow it, and * we may want a MAC policy to allow it. */ case PRIV_VFS_READ_DIR: return (0); /* * Conditionnaly allow locking (unlocking) physical pages * in memory. */ case PRIV_VM_MLOCK: case PRIV_VM_MUNLOCK: if (cred->cr_prison->pr_allow & PR_ALLOW_MLOCK) return (0); else return (EPERM); /* * Conditionally allow jailed root to bind reserved ports. */ case PRIV_NETINET_RESERVEDPORT: if (cred->cr_prison->pr_allow & PR_ALLOW_RESERVED_PORTS) return (0); else return (EPERM); /* * Allow jailed root to reuse in-use ports. */ case PRIV_NETINET_REUSEPORT: return (0); /* * Allow jailed root to set certain IPv4/6 (option) headers. */ case PRIV_NETINET_SETHDROPTS: return (0); /* * Conditionally allow creating raw sockets in jail. */ case PRIV_NETINET_RAW: if (cred->cr_prison->pr_allow & PR_ALLOW_RAW_SOCKETS) return (0); else return (EPERM); /* * Since jail implements its own visibility limits on netstat * sysctls, allow getcred. This allows identd to work in * jail. */ case PRIV_NETINET_GETCRED: return (0); /* * Allow jailed root to set loginclass. */ case PRIV_PROC_SETLOGINCLASS: return (0); /* * Do not allow a process inside a jail to read the kernel * message buffer unless explicitly permitted. */ case PRIV_MSGBUF: if (cred->cr_prison->pr_allow & PR_ALLOW_READ_MSGBUF) return (0); return (EPERM); default: /* * In all remaining cases, deny the privilege request. This * includes almost all network privileges, many system * configuration privileges. */ return (EPERM); } } /* * Return the part of pr2's name that is relative to pr1, or the whole name * if it does not directly follow. */ char * prison_name(struct prison *pr1, struct prison *pr2) { char *name; /* Jails see themselves as "0" (if they see themselves at all). */ if (pr1 == pr2) return "0"; name = pr2->pr_name; if (prison_ischild(pr1, pr2)) { /* * pr1 isn't locked (and allprison_lock may not be either) * so its length can't be counted on. But the number of dots * can be counted on - and counted. */ for (; pr1 != &prison0; pr1 = pr1->pr_parent) name = strchr(name, '.') + 1; } return (name); } /* * Return the part of pr2's path that is relative to pr1, or the whole path * if it does not directly follow. */ static char * prison_path(struct prison *pr1, struct prison *pr2) { char *path1, *path2; int len1; path1 = pr1->pr_path; path2 = pr2->pr_path; if (!strcmp(path1, "/")) return (path2); len1 = strlen(path1); if (strncmp(path1, path2, len1)) return (path2); if (path2[len1] == '\0') return "/"; if (path2[len1] == '/') return (path2 + len1); return (path2); } /* * Jail-related sysctls. */ static SYSCTL_NODE(_security, OID_AUTO, jail, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Jails"); static int sysctl_jail_list(SYSCTL_HANDLER_ARGS) { struct xprison *xp; struct prison *pr, *cpr; #ifdef INET struct in_addr *ip4 = NULL; int ip4s = 0; #endif #ifdef INET6 struct in6_addr *ip6 = NULL; int ip6s = 0; #endif int descend, error; xp = malloc(sizeof(*xp), M_TEMP, M_WAITOK); pr = req->td->td_ucred->cr_prison; error = 0; sx_slock(&allprison_lock); FOREACH_PRISON_DESCENDANT(pr, cpr, descend) { #if defined(INET) || defined(INET6) again: #endif mtx_lock(&cpr->pr_mtx); #ifdef INET if (cpr->pr_ip4s > 0) { if (ip4s < cpr->pr_ip4s) { ip4s = cpr->pr_ip4s; mtx_unlock(&cpr->pr_mtx); ip4 = realloc(ip4, ip4s * sizeof(struct in_addr), M_TEMP, M_WAITOK); goto again; } bcopy(cpr->pr_ip4, ip4, cpr->pr_ip4s * sizeof(struct in_addr)); } #endif #ifdef INET6 if (cpr->pr_ip6s > 0) { if (ip6s < cpr->pr_ip6s) { ip6s = cpr->pr_ip6s; mtx_unlock(&cpr->pr_mtx); ip6 = realloc(ip6, ip6s * sizeof(struct in6_addr), M_TEMP, M_WAITOK); goto again; } bcopy(cpr->pr_ip6, ip6, cpr->pr_ip6s * sizeof(struct in6_addr)); } #endif if (cpr->pr_ref == 0) { mtx_unlock(&cpr->pr_mtx); continue; } bzero(xp, sizeof(*xp)); xp->pr_version = XPRISON_VERSION; xp->pr_id = cpr->pr_id; xp->pr_state = cpr->pr_uref > 0 ? PRISON_STATE_ALIVE : PRISON_STATE_DYING; strlcpy(xp->pr_path, prison_path(pr, cpr), sizeof(xp->pr_path)); strlcpy(xp->pr_host, cpr->pr_hostname, sizeof(xp->pr_host)); strlcpy(xp->pr_name, prison_name(pr, cpr), sizeof(xp->pr_name)); #ifdef INET xp->pr_ip4s = cpr->pr_ip4s; #endif #ifdef INET6 xp->pr_ip6s = cpr->pr_ip6s; #endif mtx_unlock(&cpr->pr_mtx); error = SYSCTL_OUT(req, xp, sizeof(*xp)); if (error) break; #ifdef INET if (xp->pr_ip4s > 0) { error = SYSCTL_OUT(req, ip4, xp->pr_ip4s * sizeof(struct in_addr)); if (error) break; } #endif #ifdef INET6 if (xp->pr_ip6s > 0) { error = SYSCTL_OUT(req, ip6, xp->pr_ip6s * sizeof(struct in6_addr)); if (error) break; } #endif } sx_sunlock(&allprison_lock); free(xp, M_TEMP); #ifdef INET free(ip4, M_TEMP); #endif #ifdef INET6 free(ip6, M_TEMP); #endif return (error); } SYSCTL_OID(_security_jail, OID_AUTO, list, CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_jail_list, "S", "List of active jails"); static int sysctl_jail_jailed(SYSCTL_HANDLER_ARGS) { int error, injail; injail = jailed(req->td->td_ucred); error = SYSCTL_OUT(req, &injail, sizeof(injail)); return (error); } SYSCTL_PROC(_security_jail, OID_AUTO, jailed, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_jail_jailed, "I", "Process in jail?"); static int sysctl_jail_vnet(SYSCTL_HANDLER_ARGS) { int error, havevnet; #ifdef VIMAGE struct ucred *cred = req->td->td_ucred; havevnet = jailed(cred) && prison_owns_vnet(cred); #else havevnet = 0; #endif error = SYSCTL_OUT(req, &havevnet, sizeof(havevnet)); return (error); } SYSCTL_PROC(_security_jail, OID_AUTO, vnet, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_jail_vnet, "I", "Jail owns vnet?"); #if defined(INET) || defined(INET6) SYSCTL_UINT(_security_jail, OID_AUTO, jail_max_af_ips, CTLFLAG_RW, &jail_max_af_ips, 0, "Number of IP addresses a jail may have at most per address family (deprecated)"); #endif /* * Default parameters for jail(2) compatibility. For historical reasons, * the sysctl names have varying similarity to the parameter names. Prisons * just see their own parameters, and can't change them. */ static int sysctl_jail_default_allow(SYSCTL_HANDLER_ARGS) { struct prison *pr; int allow, error, i; pr = req->td->td_ucred->cr_prison; allow = (pr == &prison0) ? jail_default_allow : pr->pr_allow; /* Get the current flag value, and convert it to a boolean. */ i = (allow & arg2) ? 1 : 0; if (arg1 != NULL) i = !i; error = sysctl_handle_int(oidp, &i, 0, req); if (error || !req->newptr) return (error); i = i ? arg2 : 0; if (arg1 != NULL) i ^= arg2; /* * The sysctls don't have CTLFLAGS_PRISON, so assume prison0 * for writing. */ mtx_lock(&prison0.pr_mtx); jail_default_allow = (jail_default_allow & ~arg2) | i; mtx_unlock(&prison0.pr_mtx); return (0); } SYSCTL_PROC(_security_jail, OID_AUTO, set_hostname_allowed, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, PR_ALLOW_SET_HOSTNAME, sysctl_jail_default_allow, "I", "Processes in jail can set their hostnames (deprecated)"); SYSCTL_PROC(_security_jail, OID_AUTO, socket_unixiproute_only, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, (void *)1, PR_ALLOW_SOCKET_AF, sysctl_jail_default_allow, "I", "Processes in jail are limited to creating UNIX/IP/route sockets only (deprecated)"); SYSCTL_PROC(_security_jail, OID_AUTO, sysvipc_allowed, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, PR_ALLOW_SYSVIPC, sysctl_jail_default_allow, "I", "Processes in jail can use System V IPC primitives (deprecated)"); SYSCTL_PROC(_security_jail, OID_AUTO, allow_raw_sockets, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, PR_ALLOW_RAW_SOCKETS, sysctl_jail_default_allow, "I", "Prison root can create raw sockets (deprecated)"); SYSCTL_PROC(_security_jail, OID_AUTO, chflags_allowed, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, PR_ALLOW_CHFLAGS, sysctl_jail_default_allow, "I", "Processes in jail can alter system file flags (deprecated)"); SYSCTL_PROC(_security_jail, OID_AUTO, mount_allowed, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, PR_ALLOW_MOUNT, sysctl_jail_default_allow, "I", "Processes in jail can mount/unmount jail-friendly file systems (deprecated)"); static int sysctl_jail_default_level(SYSCTL_HANDLER_ARGS) { struct prison *pr; int level, error; pr = req->td->td_ucred->cr_prison; level = (pr == &prison0) ? *(int *)arg1 : *(int *)((char *)pr + arg2); error = sysctl_handle_int(oidp, &level, 0, req); if (error || !req->newptr) return (error); *(int *)arg1 = level; return (0); } SYSCTL_PROC(_security_jail, OID_AUTO, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &jail_default_enforce_statfs, offsetof(struct prison, pr_enforce_statfs), sysctl_jail_default_level, "I", "Processes in jail cannot see all mounted file systems (deprecated)"); SYSCTL_PROC(_security_jail, OID_AUTO, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, &jail_default_devfs_rsnum, offsetof(struct prison, pr_devfs_rsnum), sysctl_jail_default_level, "I", "Ruleset for the devfs filesystem in jail (deprecated)"); /* * Nodes to describe jail parameters. Maximum length of string parameters * is returned in the string itself, and the other parameters exist merely * to make themselves and their types known. */ SYSCTL_NODE(_security_jail, OID_AUTO, param, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Jail parameters"); int sysctl_jail_param(SYSCTL_HANDLER_ARGS) { int i; long l; size_t s; char numbuf[12]; switch (oidp->oid_kind & CTLTYPE) { case CTLTYPE_LONG: case CTLTYPE_ULONG: l = 0; #ifdef SCTL_MASK32 if (!(req->flags & SCTL_MASK32)) #endif return (SYSCTL_OUT(req, &l, sizeof(l))); case CTLTYPE_INT: case CTLTYPE_UINT: i = 0; return (SYSCTL_OUT(req, &i, sizeof(i))); case CTLTYPE_STRING: snprintf(numbuf, sizeof(numbuf), "%jd", (intmax_t)arg2); return (sysctl_handle_string(oidp, numbuf, sizeof(numbuf), req)); case CTLTYPE_STRUCT: s = (size_t)arg2; return (SYSCTL_OUT(req, &s, sizeof(s))); } return (0); } /* * CTLFLAG_RDTUN in the following indicates jail parameters that can be set at * jail creation time but cannot be changed in an existing jail. */ SYSCTL_JAIL_PARAM(, jid, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail ID"); SYSCTL_JAIL_PARAM(, parent, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail parent ID"); SYSCTL_JAIL_PARAM_STRING(, name, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail name"); SYSCTL_JAIL_PARAM_STRING(, path, CTLFLAG_RDTUN, MAXPATHLEN, "Jail root path"); SYSCTL_JAIL_PARAM(, securelevel, CTLTYPE_INT | CTLFLAG_RW, "I", "Jail secure level"); SYSCTL_JAIL_PARAM(, osreldate, CTLTYPE_INT | CTLFLAG_RDTUN, "I", "Jail value for kern.osreldate and uname -K"); SYSCTL_JAIL_PARAM_STRING(, osrelease, CTLFLAG_RDTUN, OSRELEASELEN, "Jail value for kern.osrelease and uname -r"); SYSCTL_JAIL_PARAM(, enforce_statfs, CTLTYPE_INT | CTLFLAG_RW, "I", "Jail cannot see all mounted file systems"); SYSCTL_JAIL_PARAM(, devfs_ruleset, CTLTYPE_INT | CTLFLAG_RW, "I", "Ruleset for in-jail devfs mounts"); SYSCTL_JAIL_PARAM(, persist, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail persistence"); #ifdef VIMAGE SYSCTL_JAIL_PARAM(, vnet, CTLTYPE_INT | CTLFLAG_RDTUN, "E,jailsys", "Virtual network stack"); #endif SYSCTL_JAIL_PARAM(, dying, CTLTYPE_INT | CTLFLAG_RD, "B", "Jail is in the process of shutting down"); SYSCTL_JAIL_PARAM_NODE(children, "Number of child jails"); SYSCTL_JAIL_PARAM(_children, cur, CTLTYPE_INT | CTLFLAG_RD, "I", "Current number of child jails"); SYSCTL_JAIL_PARAM(_children, max, CTLTYPE_INT | CTLFLAG_RW, "I", "Maximum number of child jails"); SYSCTL_JAIL_PARAM_SYS_NODE(host, CTLFLAG_RW, "Jail host info"); SYSCTL_JAIL_PARAM_STRING(_host, hostname, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail hostname"); SYSCTL_JAIL_PARAM_STRING(_host, domainname, CTLFLAG_RW, MAXHOSTNAMELEN, "Jail NIS domainname"); SYSCTL_JAIL_PARAM_STRING(_host, hostuuid, CTLFLAG_RW, HOSTUUIDLEN, "Jail host UUID"); SYSCTL_JAIL_PARAM(_host, hostid, CTLTYPE_ULONG | CTLFLAG_RW, "LU", "Jail host ID"); SYSCTL_JAIL_PARAM_NODE(cpuset, "Jail cpuset"); SYSCTL_JAIL_PARAM(_cpuset, id, CTLTYPE_INT | CTLFLAG_RD, "I", "Jail cpuset ID"); #ifdef INET SYSCTL_JAIL_PARAM_SYS_NODE(ip4, CTLFLAG_RDTUN, "Jail IPv4 address virtualization"); SYSCTL_JAIL_PARAM_STRUCT(_ip4, addr, CTLFLAG_RW, sizeof(struct in_addr), "S,in_addr,a", "Jail IPv4 addresses"); SYSCTL_JAIL_PARAM(_ip4, saddrsel, CTLTYPE_INT | CTLFLAG_RW, "B", "Do (not) use IPv4 source address selection rather than the " "primary jail IPv4 address."); #endif #ifdef INET6 SYSCTL_JAIL_PARAM_SYS_NODE(ip6, CTLFLAG_RDTUN, "Jail IPv6 address virtualization"); SYSCTL_JAIL_PARAM_STRUCT(_ip6, addr, CTLFLAG_RW, sizeof(struct in6_addr), "S,in6_addr,a", "Jail IPv6 addresses"); SYSCTL_JAIL_PARAM(_ip6, saddrsel, CTLTYPE_INT | CTLFLAG_RW, "B", "Do (not) use IPv6 source address selection rather than the " "primary jail IPv6 address."); #endif SYSCTL_JAIL_PARAM_NODE(allow, "Jail permission flags"); SYSCTL_JAIL_PARAM(_allow, set_hostname, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may set hostname"); SYSCTL_JAIL_PARAM(_allow, sysvipc, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may use SYSV IPC"); SYSCTL_JAIL_PARAM(_allow, raw_sockets, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may create raw sockets"); SYSCTL_JAIL_PARAM(_allow, chflags, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may alter system file flags"); SYSCTL_JAIL_PARAM(_allow, quotas, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may set file quotas"); SYSCTL_JAIL_PARAM(_allow, socket_af, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may create sockets other than just UNIX/IPv4/IPv6/route"); SYSCTL_JAIL_PARAM(_allow, mlock, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may lock (unlock) physical pages in memory"); SYSCTL_JAIL_PARAM(_allow, reserved_ports, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may bind sockets to reserved ports"); SYSCTL_JAIL_PARAM(_allow, read_msgbuf, CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may read the kernel message buffer"); SYSCTL_JAIL_PARAM(_allow, unprivileged_proc_debug, CTLTYPE_INT | CTLFLAG_RW, "B", "Unprivileged processes may use process debugging facilities"); SYSCTL_JAIL_PARAM_SUBNODE(allow, mount, "Jail mount/unmount permission flags"); SYSCTL_JAIL_PARAM(_allow_mount, , CTLTYPE_INT | CTLFLAG_RW, "B", "Jail may mount/unmount jail-friendly file systems in general"); /* * Add a dynamic parameter allow., or allow... Return * its associated bit in the pr_allow bitmask, or zero if the parameter was * not created. */ unsigned prison_add_allow(const char *prefix, const char *name, const char *prefix_descr, const char *descr) { struct bool_flags *bf; struct sysctl_oid *parent; char *allow_name, *allow_noname, *allowed; #ifndef NO_SYSCTL_DESCR char *descr_deprecated; #endif unsigned allow_flag; if (prefix ? asprintf(&allow_name, M_PRISON, "allow.%s.%s", prefix, name) < 0 || asprintf(&allow_noname, M_PRISON, "allow.%s.no%s", prefix, name) < 0 : asprintf(&allow_name, M_PRISON, "allow.%s", name) < 0 || asprintf(&allow_noname, M_PRISON, "allow.no%s", name) < 0) { free(allow_name, M_PRISON); return 0; } /* * See if this parameter has already beed added, i.e. a module was * previously loaded/unloaded. */ mtx_lock(&prison0.pr_mtx); for (bf = pr_flag_allow; bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0; bf++) { if (strcmp(bf->name, allow_name) == 0) { allow_flag = bf->flag; goto no_add; } } /* * Find a free bit in prison0's pr_allow, failing if there are none * (which shouldn't happen as long as we keep track of how many * potential dynamic flags exist). * * Due to per-jail unprivileged process debugging support * using pr_allow, also verify against PR_ALLOW_ALL_STATIC. * prison0 may have unprivileged process debugging unset. */ for (allow_flag = 1;; allow_flag <<= 1) { if (allow_flag == 0) goto no_add; if (allow_flag & PR_ALLOW_ALL_STATIC) continue; if ((prison0.pr_allow & allow_flag) == 0) break; } /* * Note the parameter in the next open slot in pr_flag_allow. * Set the flag last so code that checks pr_flag_allow can do so * without locking. */ for (bf = pr_flag_allow; bf->flag != 0; bf++) if (bf == pr_flag_allow + nitems(pr_flag_allow)) { /* This should never happen, but is not fatal. */ allow_flag = 0; goto no_add; } prison0.pr_allow |= allow_flag; bf->name = allow_name; bf->noname = allow_noname; bf->flag = allow_flag; mtx_unlock(&prison0.pr_mtx); /* * Create sysctls for the paramter, and the back-compat global * permission. */ parent = prefix ? SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(&sysctl___security_jail_param_allow), OID_AUTO, prefix, CTLFLAG_MPSAFE, 0, prefix_descr) : &sysctl___security_jail_param_allow; (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(parent), OID_AUTO, name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, sysctl_jail_param, "B", descr); if ((prefix ? asprintf(&allowed, M_TEMP, "%s_%s_allowed", prefix, name) : asprintf(&allowed, M_TEMP, "%s_allowed", name)) >= 0) { #ifndef NO_SYSCTL_DESCR (void)asprintf(&descr_deprecated, M_TEMP, "%s (deprecated)", descr); #endif (void)SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(&sysctl___security_jail), OID_AUTO, allowed, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, allow_flag, sysctl_jail_default_allow, "I", descr_deprecated); #ifndef NO_SYSCTL_DESCR free(descr_deprecated, M_TEMP); #endif free(allowed, M_TEMP); } return allow_flag; no_add: mtx_unlock(&prison0.pr_mtx); free(allow_name, M_PRISON); free(allow_noname, M_PRISON); return allow_flag; } /* * The VFS system will register jail-aware filesystems here. They each get * a parameter allow.mount.xxxfs and a flag to check when a jailed user * attempts to mount. */ void prison_add_vfs(struct vfsconf *vfsp) { #ifdef NO_SYSCTL_DESCR vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name, NULL, NULL); #else char *descr; (void)asprintf(&descr, M_TEMP, "Jail may mount the %s file system", vfsp->vfc_name); vfsp->vfc_prison_flag = prison_add_allow("mount", vfsp->vfc_name, NULL, descr); free(descr, M_TEMP); #endif } #ifdef RACCT void prison_racct_foreach(void (*callback)(struct racct *racct, void *arg2, void *arg3), void (*pre)(void), void (*post)(void), void *arg2, void *arg3) { struct prison_racct *prr; ASSERT_RACCT_ENABLED(); sx_slock(&allprison_lock); if (pre != NULL) (pre)(); LIST_FOREACH(prr, &allprison_racct, prr_next) (callback)(prr->prr_racct, arg2, arg3); if (post != NULL) (post)(); sx_sunlock(&allprison_lock); } static struct prison_racct * prison_racct_find_locked(const char *name) { struct prison_racct *prr; ASSERT_RACCT_ENABLED(); sx_assert(&allprison_lock, SA_XLOCKED); if (name[0] == '\0' || strlen(name) >= MAXHOSTNAMELEN) return (NULL); LIST_FOREACH(prr, &allprison_racct, prr_next) { if (strcmp(name, prr->prr_name) != 0) continue; /* Found prison_racct with a matching name? */ prison_racct_hold(prr); return (prr); } /* Add new prison_racct. */ prr = malloc(sizeof(*prr), M_PRISON_RACCT, M_ZERO | M_WAITOK); racct_create(&prr->prr_racct); strcpy(prr->prr_name, name); refcount_init(&prr->prr_refcount, 1); LIST_INSERT_HEAD(&allprison_racct, prr, prr_next); return (prr); } struct prison_racct * prison_racct_find(const char *name) { struct prison_racct *prr; ASSERT_RACCT_ENABLED(); sx_xlock(&allprison_lock); prr = prison_racct_find_locked(name); sx_xunlock(&allprison_lock); return (prr); } void prison_racct_hold(struct prison_racct *prr) { ASSERT_RACCT_ENABLED(); refcount_acquire(&prr->prr_refcount); } static void prison_racct_free_locked(struct prison_racct *prr) { ASSERT_RACCT_ENABLED(); sx_assert(&allprison_lock, SA_XLOCKED); if (refcount_release(&prr->prr_refcount)) { racct_destroy(&prr->prr_racct); LIST_REMOVE(prr, prr_next); free(prr, M_PRISON_RACCT); } } void prison_racct_free(struct prison_racct *prr) { ASSERT_RACCT_ENABLED(); sx_assert(&allprison_lock, SA_UNLOCKED); if (refcount_release_if_not_last(&prr->prr_refcount)) return; sx_xlock(&allprison_lock); prison_racct_free_locked(prr); sx_xunlock(&allprison_lock); } static void prison_racct_attach(struct prison *pr) { struct prison_racct *prr; ASSERT_RACCT_ENABLED(); sx_assert(&allprison_lock, SA_XLOCKED); prr = prison_racct_find_locked(pr->pr_name); KASSERT(prr != NULL, ("cannot find prison_racct")); pr->pr_prison_racct = prr; } /* * Handle jail renaming. From the racct point of view, renaming means * moving from one prison_racct to another. */ static void prison_racct_modify(struct prison *pr) { #ifdef RCTL struct proc *p; struct ucred *cred; #endif struct prison_racct *oldprr; ASSERT_RACCT_ENABLED(); sx_slock(&allproc_lock); sx_xlock(&allprison_lock); if (strcmp(pr->pr_name, pr->pr_prison_racct->prr_name) == 0) { sx_xunlock(&allprison_lock); sx_sunlock(&allproc_lock); return; } oldprr = pr->pr_prison_racct; pr->pr_prison_racct = NULL; prison_racct_attach(pr); /* * Move resource utilisation records. */ racct_move(pr->pr_prison_racct->prr_racct, oldprr->prr_racct); #ifdef RCTL /* * Force rctl to reattach rules to processes. */ FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); cred = crhold(p->p_ucred); PROC_UNLOCK(p); rctl_proc_ucred_changed(p, cred); crfree(cred); } #endif sx_sunlock(&allproc_lock); prison_racct_free_locked(oldprr); sx_xunlock(&allprison_lock); } static void prison_racct_detach(struct prison *pr) { ASSERT_RACCT_ENABLED(); sx_assert(&allprison_lock, SA_UNLOCKED); if (pr->pr_prison_racct == NULL) return; prison_racct_free(pr->pr_prison_racct); pr->pr_prison_racct = NULL; } #endif /* RACCT */ #ifdef DDB static void db_show_prison(struct prison *pr) { struct bool_flags *bf; struct jailsys_flags *jsf; #if defined(INET) || defined(INET6) int ii; #endif unsigned f; #ifdef INET char ip4buf[INET_ADDRSTRLEN]; #endif #ifdef INET6 char ip6buf[INET6_ADDRSTRLEN]; #endif db_printf("prison %p:\n", pr); db_printf(" jid = %d\n", pr->pr_id); db_printf(" name = %s\n", pr->pr_name); db_printf(" parent = %p\n", pr->pr_parent); db_printf(" ref = %d\n", pr->pr_ref); db_printf(" uref = %d\n", pr->pr_uref); db_printf(" path = %s\n", pr->pr_path); db_printf(" cpuset = %d\n", pr->pr_cpuset ? pr->pr_cpuset->cs_id : -1); #ifdef VIMAGE db_printf(" vnet = %p\n", pr->pr_vnet); #endif db_printf(" root = %p\n", pr->pr_root); db_printf(" securelevel = %d\n", pr->pr_securelevel); db_printf(" devfs_rsnum = %d\n", pr->pr_devfs_rsnum); db_printf(" children.max = %d\n", pr->pr_childmax); db_printf(" children.cur = %d\n", pr->pr_childcount); db_printf(" child = %p\n", LIST_FIRST(&pr->pr_children)); db_printf(" sibling = %p\n", LIST_NEXT(pr, pr_sibling)); db_printf(" flags = 0x%x", pr->pr_flags); for (bf = pr_flag_bool; bf < pr_flag_bool + nitems(pr_flag_bool); bf++) if (pr->pr_flags & bf->flag) db_printf(" %s", bf->name); for (jsf = pr_flag_jailsys; jsf < pr_flag_jailsys + nitems(pr_flag_jailsys); jsf++) { f = pr->pr_flags & (jsf->disable | jsf->new); db_printf(" %-16s= %s\n", jsf->name, (f != 0 && f == jsf->disable) ? "disable" : (f == jsf->new) ? "new" : "inherit"); } db_printf(" allow = 0x%x", pr->pr_allow); for (bf = pr_flag_allow; bf < pr_flag_allow + nitems(pr_flag_allow) && bf->flag != 0; bf++) if (pr->pr_allow & bf->flag) db_printf(" %s", bf->name); db_printf("\n"); db_printf(" enforce_statfs = %d\n", pr->pr_enforce_statfs); db_printf(" host.hostname = %s\n", pr->pr_hostname); db_printf(" host.domainname = %s\n", pr->pr_domainname); db_printf(" host.hostuuid = %s\n", pr->pr_hostuuid); db_printf(" host.hostid = %lu\n", pr->pr_hostid); #ifdef INET db_printf(" ip4s = %d\n", pr->pr_ip4s); for (ii = 0; ii < pr->pr_ip4s; ii++) db_printf(" %s %s\n", ii == 0 ? "ip4.addr =" : " ", inet_ntoa_r(pr->pr_ip4[ii], ip4buf)); #endif #ifdef INET6 db_printf(" ip6s = %d\n", pr->pr_ip6s); for (ii = 0; ii < pr->pr_ip6s; ii++) db_printf(" %s %s\n", ii == 0 ? "ip6.addr =" : " ", ip6_sprintf(ip6buf, &pr->pr_ip6[ii])); #endif } DB_SHOW_COMMAND(prison, db_show_prison_command) { struct prison *pr; if (!have_addr) { /* * Show all prisons in the list, and prison0 which is not * listed. */ db_show_prison(&prison0); if (!db_pager_quit) { TAILQ_FOREACH(pr, &allprison, pr_list) { db_show_prison(pr); if (db_pager_quit) break; } } return; } if (addr == 0) pr = &prison0; else { /* Look for a prison with the ID and with references. */ TAILQ_FOREACH(pr, &allprison, pr_list) if (pr->pr_id == addr && pr->pr_ref > 0) break; if (pr == NULL) /* Look again, without requiring a reference. */ TAILQ_FOREACH(pr, &allprison, pr_list) if (pr->pr_id == addr) break; if (pr == NULL) /* Assume address points to a valid prison. */ pr = (struct prison *)addr; } db_show_prison(pr); } #endif /* DDB */ Index: head/sys/kern/kern_priv.c =================================================================== --- head/sys/kern/kern_priv.c (revision 364768) +++ head/sys/kern/kern_priv.c (revision 364769) @@ -1,288 +1,340 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2006 nCircle Network Security, Inc. * Copyright (c) 2009 Robert N. M. Watson * All rights reserved. * * This software was developed by Robert N. M. Watson for the TrustedBSD * Project under contract to nCircle Network Security, Inc. * * 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. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR, NCIRCLE NETWORK SECURITY, * INC., 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. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include /* * `suser_enabled' (which can be set by the security.bsd.suser_enabled * sysctl) determines whether the system 'super-user' policy is in effect. If * it is nonzero, an effective uid of 0 connotes special privilege, * overriding many mandatory and discretionary protections. If it is zero, * uid 0 is offered no special privilege in the kernel security policy. * Setting it to zero may seriously impact the functionality of many existing * userland programs, and should not be done without careful consideration of * the consequences. */ static int __read_mostly suser_enabled = 1; SYSCTL_INT(_security_bsd, OID_AUTO, suser_enabled, CTLFLAG_RWTUN, &suser_enabled, 0, "processes with uid 0 have privilege"); static int unprivileged_mlock = 1; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_mlock, CTLFLAG_RWTUN, &unprivileged_mlock, 0, "Allow non-root users to call mlock(2)"); static int unprivileged_read_msgbuf = 1; SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_read_msgbuf, CTLFLAG_RW, &unprivileged_read_msgbuf, 0, "Unprivileged processes may read the kernel message buffer"); SDT_PROVIDER_DEFINE(priv); SDT_PROBE_DEFINE1(priv, kernel, priv_check, priv__ok, "int"); SDT_PROBE_DEFINE1(priv, kernel, priv_check, priv__err, "int"); static __always_inline int priv_check_cred_pre(struct ucred *cred, int priv) { int error; #ifdef MAC error = mac_priv_check(cred, priv); #else error = 0; #endif return (error); } static __always_inline int priv_check_cred_post(struct ucred *cred, int priv, int error, bool handled) { if (__predict_true(handled)) goto out; /* * Now check with MAC, if enabled, to see if a policy module grants * privilege. */ #ifdef MAC if (mac_priv_grant(cred, priv) == 0) { error = 0; goto out; } #endif /* * The default is deny, so if no policies have granted it, reject * with a privilege error here. */ error = EPERM; out: if (SDT_PROBES_ENABLED()) { if (error) SDT_PROBE1(priv, kernel, priv_check, priv__err, priv); else SDT_PROBE1(priv, kernel, priv_check, priv__ok, priv); } return (error); } /* * Check a credential for privilege. Lots of good reasons to deny privilege; * only a few to grant it. */ int priv_check_cred(struct ucred *cred, int priv) { int error; KASSERT(PRIV_VALID(priv), ("priv_check_cred: invalid privilege %d", priv)); switch (priv) { + case PRIV_VFS_LOOKUP: + return (priv_check_cred_vfs_lookup(cred)); case PRIV_VFS_GENERATION: return (priv_check_cred_vfs_generation(cred)); } /* * We first evaluate policies that may deny the granting of * privilege unilaterally. */ error = priv_check_cred_pre(cred, priv); if (error) goto out; /* * Jail policy will restrict certain privileges that may otherwise be * be granted. */ error = prison_priv_check(cred, priv); if (error) goto out; if (unprivileged_mlock) { /* * Allow unprivileged users to call mlock(2)/munlock(2) and * mlockall(2)/munlockall(2). */ switch (priv) { case PRIV_VM_MLOCK: case PRIV_VM_MUNLOCK: error = 0; goto out; } } if (unprivileged_read_msgbuf) { /* * Allow an unprivileged user to read the kernel message * buffer. */ if (priv == PRIV_MSGBUF) { error = 0; goto out; } } /* * Having determined if privilege is restricted by various policies, * now determine if privilege is granted. At this point, any policy * may grant privilege. For now, we allow short-circuit boolean * evaluation, so may not call all policies. Perhaps we should. * * Superuser policy grants privilege based on the effective (or in * the case of specific privileges, real) uid being 0. We allow the * superuser policy to be globally disabled, although this is * currenty of limited utility. */ if (suser_enabled) { switch (priv) { case PRIV_MAXFILES: case PRIV_MAXPROC: case PRIV_PROC_LIMIT: if (cred->cr_ruid == 0) { error = 0; goto out; } break; case PRIV_VFS_READ_DIR: /* * Allow PRIV_VFS_READ_DIR for root if we're not in a * jail, otherwise deny unless a MAC policy grants it. */ if (jailed(cred)) break; /* FALLTHROUGH */ default: if (cred->cr_uid == 0) { error = 0; goto out; } break; } } /* * Writes to kernel/physical memory are a typical root-only operation, * but non-root users are expected to be able to read it (provided they * have permission to access /dev/[k]mem). */ if (priv == PRIV_KMEM_READ) { error = 0; goto out; } /* * Allow unprivileged process debugging on a per-jail basis. * Do this here instead of prison_priv_check(), so it can also * apply to prison0. */ if (priv == PRIV_DEBUG_UNPRIV) { if (prison_allow(cred, PR_ALLOW_UNPRIV_DEBUG)) { error = 0; goto out; } } return (priv_check_cred_post(cred, priv, error, false)); out: return (priv_check_cred_post(cred, priv, error, true)); } int priv_check(struct thread *td, int priv) { KASSERT(td == curthread, ("priv_check: td != curthread")); return (priv_check_cred(td->td_ucred, priv)); +} + +static int __noinline +priv_check_cred_vfs_lookup_slow(struct ucred *cred) +{ + int error; + + error = priv_check_cred_pre(cred, PRIV_VFS_LOOKUP); + if (error) + goto out; + + if (cred->cr_uid == 0 && suser_enabled) { + error = 0; + goto out; + } + + return (priv_check_cred_post(cred, PRIV_VFS_LOOKUP, error, false)); +out: + return (priv_check_cred_post(cred, PRIV_VFS_LOOKUP, error, true)); + +} + +int +priv_check_cred_vfs_lookup(struct ucred *cred) +{ + int error; + + if (__predict_false(mac_priv_check_fp_flag || + mac_priv_grant_fp_flag || SDT_PROBES_ENABLED())) + return (priv_check_cred_vfs_lookup_slow(cred)); + + error = EPERM; + if (cred->cr_uid == 0 && suser_enabled) + error = 0; + return (error); +} + +int +priv_check_cred_vfs_lookup_nomac(struct ucred *cred) +{ + int error; + + if (__predict_false(mac_priv_check_fp_flag || + mac_priv_grant_fp_flag || SDT_PROBES_ENABLED())) + return (EAGAIN); + + error = EPERM; + if (cred->cr_uid == 0 && suser_enabled) + error = 0; + return (error); } static int __noinline priv_check_cred_vfs_generation_slow(struct ucred *cred) { int error; error = priv_check_cred_pre(cred, PRIV_VFS_GENERATION); if (error) goto out; if (jailed(cred)) { error = EPERM; goto out; } if (cred->cr_uid == 0 && suser_enabled) { error = 0; goto out; } return (priv_check_cred_post(cred, PRIV_VFS_GENERATION, error, false)); out: return (priv_check_cred_post(cred, PRIV_VFS_GENERATION, error, true)); } int priv_check_cred_vfs_generation(struct ucred *cred) { int error; if (__predict_false(mac_priv_check_fp_flag || mac_priv_grant_fp_flag || SDT_PROBES_ENABLED())) return (priv_check_cred_vfs_generation_slow(cred)); error = EPERM; if (!jailed(cred) && cred->cr_uid == 0 && suser_enabled) error = 0; return (error); } Index: head/sys/kern/vfs_subr.c =================================================================== --- head/sys/kern/vfs_subr.c (revision 364768) +++ head/sys/kern/vfs_subr.c (revision 364769) @@ -1,6719 +1,6733 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * 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. * 3. 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. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void vgonel(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_locked(void *arg); static void vfs_knl_assert_unlocked(void *arg); static void destroy_vpollinfo(struct vpollinfo *vi); static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn); static void vnlru_recalc(void); /* * These fences are intended for cases where some synchronization is * needed between access of v_iflags and lockless vnode refcount (v_holdcnt * and v_usecount) updates. Access to v_iflags is generally synchronized * by the interlock, but we have some internal assertions that check vnode * flags without acquiring the lock. Thus, these fences are INVARIANTS-only * for now. */ #ifdef INVARIANTS #define VNODE_REFCOUNT_FENCE_ACQ() atomic_thread_fence_acq() #define VNODE_REFCOUNT_FENCE_REL() atomic_thread_fence_rel() #else #define VNODE_REFCOUNT_FENCE_ACQ() #define VNODE_REFCOUNT_FENCE_REL() #endif /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VIRF_DOOMED vnode. */ static u_long __exclusive_cache_line numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static counter_u64_t vnodes_created; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, "Number of vnodes created by getnewvnode"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of allocates vnodes in the system. */ static TAILQ_HEAD(freelst, vnode) vnode_list; static struct vnode *vnode_list_free_marker; static struct vnode *vnode_list_reclaim_marker; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static long wantfreevnodes; static long __exclusive_cache_line freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static long freevnodes_old; static counter_u64_t recycles_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, "Number of vnodes recycled to meet vnode cache targets"); static counter_u64_t recycles_free_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles_free, CTLFLAG_RD, &recycles_free_count, "Number of free vnodes recycled to meet vnode cache targets"); static counter_u64_t deferred_inact; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, deferred_inact, CTLFLAG_RD, &deferred_inact, "Number of times inactive processing was deferred"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_list * numvnodes * freevnodes */ static struct mtx __exclusive_cache_line vnode_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; static smr_t buf_trie_smr; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; static uma_zone_t vnodepoll_zone; __read_frequently smr_t vfs_smr; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); #define VDBATCH_SIZE 8 struct vdbatch { u_int index; long freevnodes; struct mtx lock; struct vnode *tab[VDBATCH_SIZE]; }; DPCPU_DEFINE_STATIC(struct vdbatch, vd); static void vdbatch_dequeue(struct vnode *vp); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ u_long desiredvnodes; static u_long gapvnodes; /* gap between wanted and desired */ static u_long vhiwat; /* enough extras after expansion */ static u_long vlowat; /* minimal extras before expansion */ static u_long vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static u_long vnlru_read_freevnodes(void); /* * Note that no attempt is made to sanitize these parameters. */ static int sysctl_maxvnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = desiredvnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == desiredvnodes) return (0); mtx_lock(&vnode_list_mtx); desiredvnodes = val; wantfreevnodes = desiredvnodes / 4; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); /* * XXX There is no protection against multiple threads changing * desiredvnodes at the same time. Locking above only helps vnlru and * getnewvnode. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_maxvnodes, "LU", "Target for maximum number of vnodes"); static int sysctl_wantfreevnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = wantfreevnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == wantfreevnodes) return (0); mtx_lock(&vnode_list_mtx); wantfreevnodes = val; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); return (0); } SYSCTL_PROC(_vfs, OID_AUTO, wantfreevnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_wantfreevnodes, "LU", "Target for minimum number of \"free\" vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); static int sysctl_try_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct vnode *vp; struct nameidata nd; char *buf; unsigned long ndflags; int error; if (req->newptr == NULL) return (EINVAL); if (req->newlen >= PATH_MAX) return (E2BIG); buf = malloc(PATH_MAX, M_TEMP, M_WAITOK); error = SYSCTL_IN(req, buf, req->newlen); if (error != 0) goto out; buf[req->newlen] = '\0'; ndflags = LOCKLEAF | NOFOLLOW | AUDITVNODE1 | SAVENAME; NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, buf, curthread); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; if (VN_IS_DOOMED(vp)) { /* * This vnode is being recycled. Return != 0 to let the caller * know that the sysctl had no effect. Return EAGAIN because a * subsequent call will likely succeed (since namei will create * a new vnode if necessary) */ error = EAGAIN; goto putvnode; } counter_u64_add(recycles_count, 1); vgone(vp); putvnode: NDFREE(&nd, 0); out: free(buf, M_TEMP); return (error); } static int sysctl_ftry_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct vnode *vp; struct file *fp; int error; int fd; if (req->newptr == NULL) return (EBADF); error = sysctl_handle_int(oidp, &fd, 0, req); if (error != 0) return (error); error = getvnode(curthread, fd, &cap_fcntl_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) goto drop; counter_u64_add(recycles_count, 1); vgone(vp); VOP_UNLOCK(vp); drop: fdrop(fp, td); return (error); } SYSCTL_PROC(_debug, OID_AUTO, try_reclaim_vnode, CTLTYPE_STRING | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_try_reclaim_vnode, "A", "Try to reclaim a vnode by its pathname"); SYSCTL_PROC(_debug, OID_AUTO, ftry_reclaim_vnode, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_ftry_reclaim_vnode, "I", "Try to reclaim a vnode by its file descriptor"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return (uma_zalloc_smr(buf_trie_zone, M_NOWAIT)); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree_smr(buf_trie_zone, node); } PCTRIE_DEFINE_SMR(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free, buf_trie_smr); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512UL * 1024 * 1024 / 64) /* 8M */ #endif static MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); static struct vnode * vn_alloc_marker(struct mount *mp) { struct vnode *vp; vp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); vp->v_type = VMARKER; vp->v_mount = mp; return (vp); } static void vn_free_marker(struct vnode *vp) { MPASS(vp->v_type == VMARKER); free(vp, M_VNODE_MARKER); } /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bufobj_init(&vp->v_bufobj, vp); /* * Initialize namecache. */ cache_vnode_init(vp); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); vp->v_dbatchcpu = NOCPU; mtx_lock(&vnode_list_mtx); TAILQ_INSERT_BEFORE(vnode_list_free_marker, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; vdbatch_dequeue(vp); mtx_lock(&vnode_list_mtx); TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { struct vdbatch *vd; int cpu, physvnodes, virtvnodes; u_int i; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %lu -> %lu\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_list); mtx_init(&vnode_list_mtx, "vnode_list", NULL, MTX_DEF); /* * The lock is taken to appease WITNESS. */ mtx_lock(&vnode_list_mtx); vnlru_recalc(); mtx_unlock(&vnode_list_mtx); vnode_list_free_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_free_marker, v_vnodelist); vnode_list_reclaim_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_reclaim_marker, v_vnodelist); vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, vnode_init, vnode_fini, UMA_ALIGN_PTR, 0); uma_zone_set_smr(vnode_zone, vfs_smr); vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_SMR); buf_trie_smr = uma_zone_get_smr(buf_trie_zone); uma_prealloc(buf_trie_zone, nbuf); vnodes_created = counter_u64_alloc(M_WAITOK); recycles_count = counter_u64_alloc(M_WAITOK); recycles_free_count = counter_u64_alloc(M_WAITOK); deferred_inact = counter_u64_alloc(M_WAITOK); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); bzero(vd, sizeof(*vd)); mtx_init(&vd->lock, "vdbatch", NULL, MTX_DEF); } } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); if (vfs_op_thread_enter(mp)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MPASS((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0); MPASS((mp->mnt_kern_flag & MNTK_REFEXPIRE) == 0); vfs_mp_count_add_pcpu(mp, ref, 1); vfs_mp_count_add_pcpu(mp, lockref, 1); vfs_op_thread_exit(mp); if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); return (0); } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { int c; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if (vfs_op_thread_enter(mp)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); vfs_mp_count_sub_pcpu(mp, lockref, 1); vfs_mp_count_sub_pcpu(mp, ref, 1); vfs_op_thread_exit(mp); return; } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REL(mp); c = --mp->mnt_lockref; if (mp->mnt_vfs_ops == 0) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MNT_IUNLOCK(mp); return; } if (c < 0) vfs_dump_mount_counters(mp); if (c == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || fsidcmp(&mp->mnt_stat.f_fsid, fsid) != 0) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; if (jailed(td->td_ucred)) { /* * If the jail of the calling thread lacks permission for * this type of file system, deny immediately. */ if (!prison_allow(td->td_ucred, mp->mnt_vfc->vfc_prison_flag)) return (EPERM); /* * If the file system was mounted outside the jail of the * calling thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); } /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * Try to reduce the total number of vnodes. * * This routine (and its user) are buggy in at least the following ways: * - all parameters were picked years ago when RAM sizes were significantly * smaller * - it can pick vnodes based on pages used by the vm object, but filesystems * like ZFS don't use it making the pick broken * - since ZFS has its own aging policy it gets partially combated by this one * - a dedicated method should be provided for filesystems to let them decide * whether the vnode should be recycled * * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. * * @param reclaim_nc_src Only reclaim directories with outgoing namecache * entries if this argument is strue * @param trigger Only reclaim vnodes with fewer than this many resident * pages. * @param target How many vnodes to reclaim. * @return The number of vnodes that were reclaimed. */ static int vlrureclaim(bool reclaim_nc_src, int trigger, u_long target) { struct vnode *vp, *mvp; struct mount *mp; struct vm_object *object; u_long done; bool retried; mtx_assert(&vnode_list_mtx, MA_OWNED); retried = false; done = 0; mvp = vnode_list_reclaim_marker; restart: vp = mvp; while (done < target) { vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) break; if (__predict_false(vp->v_type == VMARKER)) continue; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount > 0 || vp->v_holdcnt == 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src))) goto next_iter; if (vp->v_type == VBAD || vp->v_type == VNON) goto next_iter; if (!VI_TRYLOCK(vp)) goto next_iter; if (vp->v_usecount > 0 || vp->v_holdcnt == 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || VN_IS_DOOMED(vp) || vp->v_type == VNON) { VI_UNLOCK(vp); goto next_iter; } object = atomic_load_ptr(&vp->v_object); if (object == NULL || object->resident_page_count > trigger) { VI_UNLOCK(vp); goto next_iter; } vholdl(vp); VI_UNLOCK(vp); TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); goto next_iter_unlocked; } if (VOP_LOCK(vp, LK_EXCLUSIVE|LK_NOWAIT) != 0) { vdrop(vp); vn_finished_write(mp); goto next_iter_unlocked; } VI_LOCK(vp); if (vp->v_usecount > 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); goto next_iter_unlocked; } counter_u64_add(recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); done++; next_iter_unlocked: if (should_yield()) kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; next_iter: MPASS(vp->v_type != VMARKER); if (!should_yield()) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; } if (done == 0 && !retried) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_HEAD(&vnode_list, mvp, v_vnodelist); retried = true; goto restart; } return (done); } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static int vnlru_free_locked(int count, struct vfsops *mnt_op) { struct vnode *vp, *mvp; struct mount *mp; int ocount; mtx_assert(&vnode_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; ocount = count; mvp = vnode_list_free_marker; restart: vp = mvp; while (count > 0) { vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, mvp, v_vnodelist); break; } if (__predict_false(vp->v_type == VMARKER)) continue; /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. * Don't recycle if we can't get the interlock without * blocking. */ if (vp->v_holdcnt > 0 || (mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) || !VI_TRYLOCK(vp)) { continue; } TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { VI_UNLOCK(vp); continue; } vholdl(vp); count--; mtx_unlock(&vnode_list_mtx); VI_UNLOCK(vp); vtryrecycle(vp); vdrop(vp); mtx_lock(&vnode_list_mtx); goto restart; } return (ocount - count); } void vnlru_free(int count, struct vfsops *mnt_op) { mtx_lock(&vnode_list_mtx); vnlru_free_locked(count, mnt_op); mtx_unlock(&vnode_list_mtx); } static void vnlru_recalc(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; /* * The main freevnodes counter is only updated when threads requeue their vnode * batches. CPUs are conditionally walked to compute a more accurate total. * * Limit how much of a slop are we willing to tolerate. Note: the actual value * at any given moment can still exceed slop, but it should not be by significant * margin in practice. */ #define VNLRU_FREEVNODES_SLOP 128 static __inline void vn_freevnodes_inc(void) { struct vdbatch *vd; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes++; critical_exit(); } static __inline void vn_freevnodes_dec(void) { struct vdbatch *vd; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes--; critical_exit(); } static u_long vnlru_read_freevnodes(void) { struct vdbatch *vd; long slop; int cpu; mtx_assert(&vnode_list_mtx, MA_OWNED); if (freevnodes > freevnodes_old) slop = freevnodes - freevnodes_old; else slop = freevnodes_old - freevnodes; if (slop < VNLRU_FREEVNODES_SLOP) return (freevnodes >= 0 ? freevnodes : 0); freevnodes_old = freevnodes; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); freevnodes_old += vd->freevnodes; } return (freevnodes_old >= 0 ? freevnodes_old : 0); } static bool vnlru_under(u_long rnumvnodes, u_long limit) { u_long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = vnlru_read_freevnodes(); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static bool vnlru_under_unlocked(u_long rnumvnodes, u_long limit) { long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = atomic_load_long(&freevnodes); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static void vnlru_kick(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); if (vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } static void vnlru_proc(void) { u_long rnumvnodes, rfreevnodes, target; unsigned long onumvnodes; int done, force, trigger, usevnodes; bool reclaim_nc_src, want_reread; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; want_reread = false; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (want_reread) { force = vnlru_under(numvnodes, vhiwat) ? 1 : 0; want_reread = false; } /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (rnumvnodes > desiredvnodes) { vnlru_free_locked(rnumvnodes - desiredvnodes, NULL); rnumvnodes = atomic_load_long(&numvnodes); } /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (force == 0 && !vnlru_under(rnumvnodes, vlowat)) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } rfreevnodes = vnlru_read_freevnodes(); onumvnodes = rnumvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (rnumvnodes <= desiredvnodes) usevnodes = rnumvnodes - rfreevnodes; else usevnodes = rnumvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; target = rnumvnodes * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; done = vlrureclaim(reclaim_nc_src, trigger, target); mtx_unlock(&vnode_list_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(UMA_RECLAIM_DRAIN); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } want_reread = true; force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else { want_reread = true; kern_yield(PRI_USER); } } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp); VI_UNLOCK(vp); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if (!VN_IS_DOOMED(vp)) { counter_u64_add(recycles_free_count, 1); vgonel(vp); } VOP_UNLOCK(vp); VI_UNLOCK(vp); vn_finished_write(vnmp); return (0); } /* * Allocate a new vnode. * * The operation never returns an error. Returning an error was disabled * in r145385 (dated 2005) with the following comment: * * XXX Not all VFS_VGET/ffs_vget callers check returns. * * Given the age of this commit (almost 15 years at the time of writing this * comment) restoring the ability to fail requires a significant audit of * all codepaths. * * The routine can try to free a vnode or stall for up to 1 second waiting for * vnlru to clear things up, but ultimately always performs a M_WAITOK allocation. */ static u_long vn_alloc_cyclecount; static struct vnode * __noinline vn_alloc_hard(struct mount *mp) { u_long rnumvnodes, rfreevnodes; mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (rnumvnodes + 1 < desiredvnodes) { vn_alloc_cyclecount = 0; goto alloc; } rfreevnodes = vnlru_read_freevnodes(); if (vn_alloc_cyclecount++ >= rfreevnodes) { vn_alloc_cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (vnlru_free_locked(1, NULL) > 0) goto alloc; if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPEND) == 0) { /* * Wait for space for a new vnode. */ vnlru_kick(); msleep(&vnlruproc_sig, &vnode_list_mtx, PVFS, "vlruwk", hz); if (atomic_load_long(&numvnodes) + 1 > desiredvnodes && vnlru_read_freevnodes() > 1) vnlru_free_locked(1, NULL); } alloc: rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (vnlru_under(rnumvnodes, vlowat)) vnlru_kick(); mtx_unlock(&vnode_list_mtx); return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } static struct vnode * vn_alloc(struct mount *mp) { u_long rnumvnodes; if (__predict_false(vn_alloc_cyclecount != 0)) return (vn_alloc_hard(mp)); rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (__predict_false(vnlru_under_unlocked(rnumvnodes, vlowat))) { atomic_subtract_long(&numvnodes, 1); return (vn_alloc_hard(mp)); } return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } static void vn_free(struct vnode *vp) { atomic_subtract_long(&numvnodes, 1); uma_zfree_smr(vnode_zone, vp); } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); KASSERT(vops->registered, ("%s: not registered vector op %p\n", __func__, vops)); td = curthread; if (td->td_vp_reserved != NULL) { vp = td->td_vp_reserved; td->td_vp_reserved = NULL; } else { vp = vn_alloc(mp); } counter_u64_add(vnodes_created, 1); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; #ifdef WITNESS if (lo->lo_name != tag) { #endif lo->lo_name = tag; #ifdef WITNESS WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } #endif /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_op = vops; v_init_counters(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef DIAGNOSTIC if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode(9), tag %s\n", tag); #endif #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } void getnewvnode_reserve(void) { struct thread *td; td = curthread; MPASS(td->td_vp_reserved == NULL); td->td_vp_reserved = vn_alloc(NULL); } void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; if (td->td_vp_reserved != NULL) { vn_free(td->td_vp_reserved); td->td_vp_reserved = NULL; } } static void __noinline freevnode(struct vnode *vp) { struct bufobj *bo; /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); /* * Paired with vgone. */ vn_seqc_write_end_locked(vp); VNPASS(vp->v_seqc_users == 0, vp); bo = &vp->v_bufobj; VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNPASS(vp->v_holdcnt == VHOLD_NO_SMR, vp); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } #ifdef INVARIANTS /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ vp->v_op = NULL; #endif vp->v_mountedhere = NULL; vp->v_unpcb = NULL; vp->v_rdev = NULL; vp->v_fifoinfo = NULL; vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; vp->v_irflag = 0; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; vn_free(vp); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) return (error); /* * XXX We could save a lock/unlock if this was only * enabled under INVARIANTS */ BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: dirty bufs"); } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); if ((flags & V_VMIO) == 0 && bo->bo_object != NULL) { BO_UNLOCK(bo); vm_object_pip_wait_unlocked(bo->bo_object, "bovlbx"); BO_LOCK(bo); } } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO | V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 && bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: flush dirty failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { /* * If we are flushing both V_NORMAL and V_ALT buffers then * do not skip any buffers. If we are flushing only V_NORMAL * buffers then skip buffers marked as BX_ALTDATA. If we are * flushing only V_ALT buffers then skip buffers not marked * as BX_ALTDATA. */ if (((flags & (V_NORMAL | V_ALT)) != (V_NORMAL | V_ALT)) && (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA) != 0) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0))) { continue; } if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); if (nbp == NULL) break; nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if ((bp->b_flags & B_VMIO) != 0) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, off_t length, int blksize) { struct buf *bp, *nbp; struct bufobj *bo; daddr_t startlbn; CTR4(KTR_VFS, "%s: vp %p with block %d:%ju", __func__, vp, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ startlbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); bo = &vp->v_bufobj; restart_unlocked: BO_LOCK(bo); while (v_inval_buf_range_locked(vp, bo, startlbn, INT64_MAX) == EAGAIN) ; if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart_unlocked; VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } /* * Invalidate the cached pages of a file's buffer within the range of block * numbers [startlbn, endlbn). */ void v_inval_buf_range(struct vnode *vp, daddr_t startlbn, daddr_t endlbn, int blksize) { struct bufobj *bo; off_t start, end; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range"); start = blksize * startlbn; end = blksize * endlbn; bo = &vp->v_bufobj; BO_LOCK(bo); MPASS(blksize == bo->bo_bsize); while (v_inval_buf_range_locked(vp, bo, startlbn, endlbn) == EAGAIN) ; BO_UNLOCK(bo); vn_pages_remove(vp, OFF_TO_IDX(start), OFF_TO_IDX(end + PAGE_SIZE - 1)); } static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn) { struct buf *bp, *nbp; bool anyfreed; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range_locked"); ASSERT_BO_LOCKED(bo); do { anyfreed = false; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || nbp->b_vp != vp || (nbp->b_flags & B_DELWRI) != 0)) return (EAGAIN); } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) return (EAGAIN); } } while (anyfreed); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; b_xflags_t flags; flags = bp->b_xflags; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((flags & (BX_VNDIRTY | BX_VNCLEAN)) != 0 && (flags & (BX_VNDIRTY | BX_VNCLEAN)) != (BX_VNDIRTY | BX_VNCLEAN), ("%s: buffer %p has invalid queue state", __func__, bp)); if ((flags & BX_VNDIRTY) != 0) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); KASSERT((bo->bo_flag & BO_NOBUFS) == 0, ("buf_vlist_add: bo %p does not allow bufs", bo)); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return (BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno)); } /* * Look up a buf using the buffer tries, without the bufobj lock. This relies * on SMR for safe lookup, and bufs being in a no-free zone to provide type * stability of the result. Like other lockless lookups, the found buf may * already be invalid by the time this function returns. */ struct buf * gbincore_unlocked(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_UNLOCKED(bo); bp = BUF_PCTRIE_LOOKUP_UNLOCKED(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return (BUF_PCTRIE_LOOKUP_UNLOCKED(&bo->bo_dirty.bv_root, lblkno)); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); buf_vlist_remove(bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_MPSAFE| CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = bo2vnode(*bo); if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Move the buffer between the clean and dirty lists of its vnode. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; KASSERT((bp->b_flags & B_PAGING) == 0, ("%s: cannot reassign paging buffer %p", __func__, bp)); CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); BO_LOCK(bo); buf_vlist_remove(bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VIRF_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * usecount is manipulated using atomics without holding any locks. * * holdcnt can be manipulated using atomics without holding any locks, * except when transitioning 1<->0, in which case the interlock is held. * * Consumers which don't guarantee liveness of the vnode can use SMR to * try to get a reference. Note this operation can fail since the vnode * may be awaiting getting freed by the time they get to it. */ enum vgetstate vget_prep_smr(struct vnode *vp) { enum vgetstate vs; VFS_SMR_ASSERT_ENTERED(); if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { if (vhold_smr(vp)) vs = VGET_HOLDCNT; else vs = VGET_NONE; } return (vs); } enum vgetstate vget_prep(struct vnode *vp) { enum vgetstate vs; if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { vhold(vp); vs = VGET_HOLDCNT; } return (vs); } void vget_abort(struct vnode *vp, enum vgetstate vs) { switch (vs) { case VGET_USECOUNT: vrele(vp); break; case VGET_HOLDCNT: vdrop(vp); break; default: __assert_unreachable(); } } int vget(struct vnode *vp, int flags) { enum vgetstate vs; vs = vget_prep(vp); return (vget_finish(vp, flags, vs)); } int vget_finish(struct vnode *vp, int flags, enum vgetstate vs) { int error; if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); VNPASS(vs == VGET_HOLDCNT || vs == VGET_USECOUNT, vp); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); error = vn_lock(vp, flags); if (__predict_false(error != 0)) { vget_abort(vp, vs); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } vget_finish_ref(vp, vs); return (0); } void vget_finish_ref(struct vnode *vp, enum vgetstate vs) { int old; VNPASS(vs == VGET_HOLDCNT || vs == VGET_USECOUNT, vp); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); if (vs == VGET_USECOUNT) return; /* * We hold the vnode. If the usecount is 0 it will be utilized to keep * the vnode around. Otherwise someone else lended their hold count and * we have to drop ours. */ old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old >= 0, vp, ("%s: wrong use count %d", __func__, old)); if (old != 0) { #ifdef INVARIANTS old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 1, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif } } void vref(struct vnode *vp) { enum vgetstate vs; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vs = vget_prep(vp); vget_finish_ref(vp, vs); } void vrefact(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old > 0, vp, ("%s: wrong use count %d", __func__, old)); #else refcount_acquire(&vp->v_usecount); #endif } void vlazy(struct vnode *vp) { struct mount *mp; VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode not held", __func__)); if ((vp->v_mflag & VMP_LAZYLIST) != 0) return; /* * We may get here for inactive routines after the vnode got doomed. */ if (VN_IS_DOOMED(vp)) return; mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); if ((vp->v_mflag & VMP_LAZYLIST) == 0) { vp->v_mflag |= VMP_LAZYLIST; TAILQ_INSERT_TAIL(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize++; } mtx_unlock(&mp->mnt_listmtx); } /* * This routine is only meant to be called from vgonel prior to dooming * the vnode. */ static void vunlazy_gone(struct vnode *vp) { struct mount *mp; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); VNPASS(!VN_IS_DOOMED(vp), vp); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; mtx_unlock(&mp->mnt_listmtx); } } static void vdefer_inactive(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode without hold count", __func__)); if (VN_IS_DOOMED(vp)) { vdropl(vp); return; } if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vdropl(vp); return; } if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; vdropl(vp); return; } vlazy(vp); vp->v_iflag |= VI_DEFINACT; VI_UNLOCK(vp); counter_u64_add(deferred_inact, 1); } static void vdefer_inactive_unlocked(struct vnode *vp) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } vdefer_inactive(vp); } enum vput_op { VRELE, VPUT, VUNREF }; /* * Handle ->v_usecount transitioning to 0. * * By releasing the last usecount we take ownership of the hold count which * provides liveness of the vnode, meaning we have to vdrop. * * For all vnodes we may need to perform inactive processing. It requires an * exclusive lock on the vnode, while it is legal to call here with only a * shared lock (or no locks). If locking the vnode in an expected manner fails, * inactive processing gets deferred to the syncer. * * XXX Some filesystems pass in an exclusively locked vnode and strongly depend * on the lock being held all the way until VOP_INACTIVE. This in particular * happens with UFS which adds half-constructed vnodes to the hash, where they * can be found by other code. */ static void vput_final(struct vnode *vp, enum vput_op func) { int error; bool want_unlock; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNPASS(vp->v_holdcnt > 0, vp); VI_LOCK(vp); /* * By the time we got here someone else might have transitioned * the count back to > 0. */ if (vp->v_usecount > 0) goto out; /* * If the vnode is doomed vgone already performed inactive processing * (if needed). */ if (VN_IS_DOOMED(vp)) goto out; if (__predict_true(VOP_NEED_INACTIVE(vp) == 0)) goto out; if (vp->v_iflag & VI_DOINGINACT) goto out; /* * Locking operations here will drop the interlock and possibly the * vnode lock, opening a window where the vnode can get doomed all the * while ->v_usecount is 0. Set VI_OWEINACT to let vgone know to * perform inactive. */ vp->v_iflag |= VI_OWEINACT; want_unlock = false; error = 0; switch (func) { case VRELE: switch (VOP_ISLOCKED(vp)) { case LK_EXCLUSIVE: break; case LK_EXCLOTHER: case 0: want_unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; default: /* * The lock has at least one sharer, but we have no way * to conclude whether this is us. Play it safe and * defer processing. */ error = EAGAIN; break; } break; case VPUT: want_unlock = true; if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } if (error == 0) { vinactive(vp); if (want_unlock) VOP_UNLOCK(vp); vdropl(vp); } else { vdefer_inactive(vp); } return; out: if (func == VPUT) VOP_UNLOCK(vp); vdropl(vp); } /* * Decrement ->v_usecount for a vnode. * * Releasing the last use count requires additional processing, see vput_final * above for details. * * Comment above each variant denotes lock state on entry and exit. */ /* * in: any * out: same as passed in */ void vrele(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VRELE); } /* * in: locked * out: unlocked */ void vput(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) { VOP_UNLOCK(vp); return; } vput_final(vp, VPUT); } /* * in: locked * out: locked */ void vunref(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VUNREF); } void vhold(struct vnode *vp) { int old; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old >= 0 && (old & VHOLD_ALL_FLAGS) == 0, vp, ("%s: wrong hold count %d", __func__, old)); if (old == 0) vn_freevnodes_dec(); } void vholdl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vhold(vp); } void vholdnz(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old > 0 && (old & VHOLD_ALL_FLAGS) == 0, vp, ("%s: wrong hold count %d", __func__, old)); #else atomic_add_int(&vp->v_holdcnt, 1); #endif } /* * Grab a hold count unless the vnode is freed. * * Only use this routine if vfs smr is the only protection you have against * freeing the vnode. * * The code loops trying to add a hold count as long as the VHOLD_NO_SMR flag * is not set. After the flag is set the vnode becomes immutable to anyone but * the thread which managed to set the flag. * * It may be tempting to replace the loop with: * count = atomic_fetchadd_int(&vp->v_holdcnt, 1); * if (count & VHOLD_NO_SMR) { * backpedal and error out; * } * * However, while this is more performant, it hinders debugging by eliminating * the previously mentioned invariant. */ bool vhold_smr(struct vnode *vp) { int count; VFS_SMR_ASSERT_ENTERED(); count = atomic_load_int(&vp->v_holdcnt); for (;;) { if (count & VHOLD_NO_SMR) { VNASSERT((count & ~VHOLD_NO_SMR) == 0, vp, ("non-zero hold count with flags %d\n", count)); return (false); } VNASSERT(count >= 0, vp, ("invalid hold count %d\n", count)); if (atomic_fcmpset_int(&vp->v_holdcnt, &count, count + 1)) { if (count == 0) vn_freevnodes_dec(); return (true); } } } static void __noinline vdbatch_process(struct vdbatch *vd) { struct vnode *vp; int i; mtx_assert(&vd->lock, MA_OWNED); MPASS(curthread->td_pinned > 0); MPASS(vd->index == VDBATCH_SIZE); mtx_lock(&vnode_list_mtx); critical_enter(); freevnodes += vd->freevnodes; for (i = 0; i < VDBATCH_SIZE; i++) { vp = vd->tab[i]; TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, vp, v_vnodelist); MPASS(vp->v_dbatchcpu != NOCPU); vp->v_dbatchcpu = NOCPU; } mtx_unlock(&vnode_list_mtx); vd->freevnodes = 0; bzero(vd->tab, sizeof(vd->tab)); vd->index = 0; critical_exit(); } static void vdbatch_enqueue(struct vnode *vp) { struct vdbatch *vd; ASSERT_VI_LOCKED(vp, __func__); VNASSERT(!VN_IS_DOOMED(vp), vp, ("%s: deferring requeue of a doomed vnode", __func__)); if (vp->v_dbatchcpu != NOCPU) { VI_UNLOCK(vp); return; } sched_pin(); vd = DPCPU_PTR(vd); mtx_lock(&vd->lock); MPASS(vd->index < VDBATCH_SIZE); MPASS(vd->tab[vd->index] == NULL); /* * A hack: we depend on being pinned so that we know what to put in * ->v_dbatchcpu. */ vp->v_dbatchcpu = curcpu; vd->tab[vd->index] = vp; vd->index++; VI_UNLOCK(vp); if (vd->index == VDBATCH_SIZE) vdbatch_process(vd); mtx_unlock(&vd->lock); sched_unpin(); } /* * This routine must only be called for vnodes which are about to be * deallocated. Supporting dequeue for arbitrary vndoes would require * validating that the locked batch matches. */ static void vdbatch_dequeue(struct vnode *vp) { struct vdbatch *vd; int i; short cpu; VNASSERT(vp->v_type == VBAD || vp->v_type == VNON, vp, ("%s: called for a used vnode\n", __func__)); cpu = vp->v_dbatchcpu; if (cpu == NOCPU) return; vd = DPCPU_ID_PTR(cpu, vd); mtx_lock(&vd->lock); for (i = 0; i < vd->index; i++) { if (vd->tab[i] != vp) continue; vp->v_dbatchcpu = NOCPU; vd->index--; vd->tab[i] = vd->tab[vd->index]; vd->tab[vd->index] = NULL; break; } mtx_unlock(&vd->lock); /* * Either we dequeued the vnode above or the target CPU beat us to it. */ MPASS(vp->v_dbatchcpu == NOCPU); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VIRF_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ static void vdrop_deactivate(struct vnode *vp) { struct mount *mp; ASSERT_VI_LOCKED(vp, __func__); /* * Mark a vnode as free: remove it from its active list * and put it up for recycling on the freelist. */ VNASSERT(!VN_IS_DOOMED(vp), vp, ("vdrop: returning doomed vnode")); VNASSERT(vp->v_op != NULL, vp, ("vdrop: vnode already reclaimed.")); VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with VI_OWEINACT set")); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("vnode with VI_DEFINACT set")); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNASSERT(vp->v_mflag & VMP_LAZYLIST, vp, ("lost VMP_LAZYLIST")); /* * Don't remove the vnode from the lazy list if another thread * has increased the hold count. It may have re-enqueued the * vnode to the lazy list and is now responsible for its * removal. */ if (vp->v_holdcnt == 0) { vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; } mtx_unlock(&mp->mnt_listmtx); } vdbatch_enqueue(vp); } static void __noinline vdropl_final(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(VN_IS_DOOMED(vp), vp); /* * Set the VHOLD_NO_SMR flag. * * We may be racing against vhold_smr. If they win we can just pretend * we never got this far, they will vdrop later. */ if (__predict_false(!atomic_cmpset_int(&vp->v_holdcnt, 0, VHOLD_NO_SMR))) { vn_freevnodes_inc(); VI_UNLOCK(vp); /* * We lost the aforementioned race. Any subsequent access is * invalid as they might have managed to vdropl on their own. */ return; } /* * Don't bump freevnodes as this one is going away. */ freevnode(vp); } void vdrop(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (refcount_release_if_not_last(&vp->v_holdcnt)) return; VI_LOCK(vp); vdropl(vp); } void vdropl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!refcount_release(&vp->v_holdcnt)) { VI_UNLOCK(vp); return; } if (!VN_IS_DOOMED(vp)) { vn_freevnodes_inc(); vdrop_deactivate(vp); /* * Also unlocks the interlock. We can't assert on it as we * released our hold and by now the vnode might have been * freed. */ return; } vdropl_final(vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. */ static void vinactivef(struct vnode *vp) { struct vm_object *obj; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } VOP_INACTIVE(vp, curthread); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; } void vinactive(struct vnode *vp) { ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((vp->v_iflag & VI_OWEINACT) == 0) return; if (vp->v_iflag & VI_DOINGINACT) return; if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; return; } vinactivef(vp); } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); if (error != 0) { VOP_UNLOCK(vp); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount <= 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; VI_LOCK(vp); recycled = vrecyclel(vp); VI_UNLOCK(vp); return (recycled); } /* * vrecycle, with the vp interlock held. */ int vrecyclel(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } static void notify_lowervp_vfs_dummy(struct mount *mp __unused, struct vnode *lowervp __unused) { } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { static struct vfsops vgonel_vfsops = { .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, }; struct mount *mp, *ump, *mmp; mp = vp->v_mount; if (mp == NULL) return; if (TAILQ_EMPTY(&mp->mnt_uppers)) return; mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); mmp->mnt_op = &vgonel_vfsops; mmp->mnt_kern_flag |= MNTK_MARKER; MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_VGONE_UPPER; for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { ump = TAILQ_NEXT(ump, mnt_upper_link); continue; } TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); ump = TAILQ_NEXT(mmp, mnt_upper_link); TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); } free(mmp, M_TEMP); mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; wakeup(&mp->mnt_uppers); } MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; struct mount *mp; vm_object_t object; bool active, oweinact; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (vp->v_irflag & VIRF_DOOMED) return; /* * Paired with freevnode. */ vn_seqc_write_begin_locked(vp); vunlazy_gone(vp); vp->v_irflag |= VIRF_DOOMED; /* * Check to see if the vnode is in use. If so, we have to call * VOP_CLOSE() and VOP_INACTIVE(). */ active = vp->v_usecount > 0; oweinact = (vp->v_iflag & VI_OWEINACT) != 0; /* * If we need to do inactive VI_OWEINACT will be set. */ if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vp->v_iflag &= ~VI_DEFINACT; vdropl(vp); } else { VNASSERT(vp->v_holdcnt > 0, vp, ("vnode without hold count")); VI_UNLOCK(vp); } vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if (oweinact || active) { VI_LOCK(vp); vinactivef(vp); VI_UNLOCK(vp); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set later, or in * vm_object_terminate() after the object's page queue is * flushed. */ object = vp->v_bufobj.bo_object; if (object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Handle the VM part. Tmpfs handles v_object on its own (the * OBJT_VNODE check). Nullfs or other bypassing filesystems * should not touch the object borrowed from the lower vnode * (the handle check). */ if (object != NULL && object->type == OBJT_VNODE && object->handle == vp) vnode_destroy_vobject(vp); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p", vp)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); cache_purge_vgone(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_type = VBAD; } /* * Print out a description of a vnode. */ static const char * const typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; _Static_assert((VHOLD_ALL_FLAGS & ~VHOLD_NO_SMR) == 0, "new hold count flag not added to vn_printf"); void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; u_int holdcnt; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("type %s\n", typename[vp->v_type]); holdcnt = atomic_load_int(&vp->v_holdcnt); printf(" usecount %d, writecount %d, refcount %d seqc users %d", vp->v_usecount, vp->v_writecount, holdcnt & ~VHOLD_ALL_FLAGS, vp->v_seqc_users); switch (vp->v_type) { case VDIR: printf(" mountedhere %p\n", vp->v_mountedhere); break; case VCHR: printf(" rdev %p\n", vp->v_rdev); break; case VSOCK: printf(" socket %p\n", vp->v_unpcb); break; case VFIFO: printf(" fifoinfo %p\n", vp->v_fifoinfo); break; default: printf("\n"); break; } buf[0] = '\0'; buf[1] = '\0'; if (holdcnt & VHOLD_NO_SMR) strlcat(buf, "|VHOLD_NO_SMR", sizeof(buf)); printf(" hold count flags (%s)\n", buf + 1); buf[0] = '\0'; buf[1] = '\0'; if (vp->v_irflag & VIRF_DOOMED) strlcat(buf, "|VIRF_DOOMED", sizeof(buf)); if (vp->v_irflag & VIRF_PGREAD) strlcat(buf, "|VIRF_PGREAD", sizeof(buf)); flags = vp->v_irflag & ~(VIRF_DOOMED | VIRF_PGREAD); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VIRF(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_VMSIZEVNLOCK) strlcat(buf, "|VV_VMSIZEVNLOCK", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); if (vp->v_vflag & VV_READLINK) strlcat(buf, "|VV_READLINK", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_TEXT_REF) strlcat(buf, "|VI_TEXT_REF", sizeof(buf)); if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); if (vp->v_iflag & VI_DEFINACT) strlcat(buf, "|VI_DEFINACT", sizeof(buf)); flags = vp->v_iflag & ~(VI_TEXT_REF | VI_MOUNT | VI_DOINGINACT | VI_OWEINACT | VI_DEFINACT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_mflag & VMP_LAZYLIST) strlcat(buf, "|VMP_LAZYLIST", sizeof(buf)); flags = vp->v_mflag & ~(VMP_LAZYLIST); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VMP(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)\n", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_VGONE_UPPER); MNT_KERN_FLAG(MNTK_VGONE_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_MARKER); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_FPLOOKUP); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_REF), mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_lazyvnodelistsize = %d\n", mp->mnt_lazyvnodelistsize); db_printf(" mnt_writeopcount = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_WRITEOPCOUNT), mp->mnt_writeopcount); db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_LOCKREF), mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf(" mnt_vfs_ops = %d\n", mp->mnt_vfs_ops); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt > 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } static void vfs_deferred_inactive(struct vnode *vp, int lkflags) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("VI_DEFINACT still set")); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } if (vn_lock(vp, lkflags) == 0) { VI_LOCK(vp); vinactive(vp); VOP_UNLOCK(vp); vdropl(vp); return; } vdefer_inactive_unlocked(vp); } static int vfs_periodic_inactive_filter(struct vnode *vp, void *arg) { return (vp->v_iflag & VI_DEFINACT); } static void __noinline vfs_periodic_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; int lkflags; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) lkflags |= LK_NOWAIT; MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_inactive_filter, NULL) { if ((vp->v_iflag & VI_DEFINACT) == 0) { VI_UNLOCK(vp); continue; } vp->v_iflag &= ~VI_DEFINACT; vfs_deferred_inactive(vp, lkflags); } } static inline bool vfs_want_msync(struct vnode *vp) { struct vm_object *obj; /* * This test may be performed without any locks held. * We rely on vm_object's type stability. */ if (vp->v_vflag & VV_NOSYNC) return (false); obj = vp->v_object; return (obj != NULL && vm_object_mightbedirty(obj)); } static int vfs_periodic_msync_inactive_filter(struct vnode *vp, void *arg __unused) { if (vp->v_vflag & VV_NOSYNC) return (false); if (vp->v_iflag & VI_DEFINACT) return (true); return (vfs_want_msync(vp)); } static void __noinline vfs_periodic_msync_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; struct thread *td; int lkflags, objflags; bool seen_defer; td = curthread; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) { lkflags |= LK_NOWAIT; objflags = OBJPC_NOSYNC; } else { objflags = OBJPC_SYNC; } MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_msync_inactive_filter, NULL) { seen_defer = false; if (vp->v_iflag & VI_DEFINACT) { vp->v_iflag &= ~VI_DEFINACT; seen_defer = true; } if (!vfs_want_msync(vp)) { if (seen_defer) vfs_deferred_inactive(vp, lkflags); else VI_UNLOCK(vp); continue; } if (vget(vp, lkflags) == 0) { obj = vp->v_object; if (obj != NULL && (vp->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, objflags); VM_OBJECT_WUNLOCK(obj); } vput(vp); if (seen_defer) vdrop(vp); } else { if (seen_defer) vdefer_inactive_unlocked(vp); } } } void vfs_periodic(struct mount *mp, int flags) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if ((mp->mnt_kern_flag & MNTK_NOMSYNC) != 0) vfs_periodic_inactive(mp, flags); else vfs_periodic_msync_inactive(mp, flags); } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); uma_zfree(vnodepoll_zone, vi); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_need_inactive = vop_stdneed_inactive, /* need_inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; VFS_VOP_VECTOR_REGISTER(sync_vnodeops); /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); /* * The filesystem at hand may be idle with free vnodes stored in the * batch. Return them instead of letting them stay there indefinitely. */ vfs_periodic(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } int vn_need_pageq_flush(struct vnode *vp) { struct vm_object *obj; int need; MPASS(mtx_owned(VI_MTX(vp))); need = 0; if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) need = 1; return (need); } /* * Check if vnode represents a disk device */ bool vn_isdisk_error(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: *errp = error; return (error == 0); } bool vn_isdisk(struct vnode *vp) { int error; return (vn_isdisk_error(vp, &error)); } /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ int vaccess_vexec_smr(mode_t file_mode, uid_t file_uid, gid_t file_gid, struct ucred *cred) { + int error; VFS_SMR_ASSERT_ENTERED(); /* Check the owner. */ if (cred->cr_uid == file_uid) { if (file_mode & S_IXUSR) return (0); goto out_error; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) return (0); goto out_error; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) return (0); out_error: /* - * Permission check failed. + * Permission check failed, but it is possible denial will get overwritten + * (e.g., when root is traversing through a 700 directory owned by someone + * else). * * vaccess() calls priv_check_cred which in turn can descent into MAC * modules overriding this result. It's quite unclear what semantics * are allowed for them to operate, thus for safety we don't call them * from within the SMR section. This also means if any such modules * are present, we have to let the regular lookup decide. */ - if (__predict_false(mac_priv_check_fp_flag || mac_priv_grant_fp_flag)) + error = priv_check_cred_vfs_lookup_nomac(cred); + switch (error) { + case 0: + return (0); + case EAGAIN: + /* + * MAC modules present. + */ return (EAGAIN); - return (EACCES); + case EPERM: + return (EACCES); + default: + return (error); + } } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, and credentials. * Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (KERNEL_PANICKED() || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif /* * It may be tempting to add vn_seqc_write_begin/end calls here and * in vop_rename_post but that's not going to work out since some * filesystems relookup vnodes mid-rename. This is probably a bug. * * For now filesystems are expected to do the relevant calls after they * decide what vnodes to operate on. */ if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_fplookup_vexec_debugpre(void *ap __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_fplookup_vexec_debugpost(void *ap __unused, int rc __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_strategy_debugpre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (!KERNEL_PANICKED() && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_debugpre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_debugpost(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_debugpre(void *ap) { struct vop_unlock_args *a = ap; ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_need_inactive_debugpre(void *ap) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } void vop_need_inactive_debugpost(void *ap, int rc) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } #endif void vop_create_pre(void *ap) { struct vop_create_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_create_post(void *ap, int rc) { struct vop_create_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_whiteout_pre(void *ap) { struct vop_whiteout_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_whiteout_post(void *ap, int rc) { struct vop_whiteout_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); } void vop_deleteextattr_pre(void *ap) { struct vop_deleteextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_pre(void *ap) { struct vop_link_args *a; struct vnode *vp, *tdvp; a = ap; vp = a->a_vp; tdvp = a->a_tdvp; vn_seqc_write_begin(vp); vn_seqc_write_begin(tdvp); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a; struct vnode *vp, *tdvp; a = ap; vp = a->a_vp; tdvp = a->a_tdvp; vn_seqc_write_end(vp); vn_seqc_write_end(tdvp); if (!rc) { VFS_KNOTE_LOCKED(vp, NOTE_LINK); VFS_KNOTE_LOCKED(tdvp, NOTE_WRITE); } } void vop_mkdir_pre(void *ap) { struct vop_mkdir_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE | NOTE_LINK); } void vop_mknod_pre(void *ap) { struct vop_mknod_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a; struct vnode *vp; a = ap; vp = a->a_vp; ASSERT_VOP_IN_SEQC(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_REVOKE); } void vop_remove_pre(void *ap) { struct vop_remove_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_begin(dvp); vn_seqc_write_begin(vp); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_end(dvp); vn_seqc_write_end(vp); if (!rc) { VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_pre(void *ap) { struct vop_rmdir_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_begin(dvp); vn_seqc_write_begin(vp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_end(dvp); vn_seqc_write_end(vp); if (!rc) { VFS_KNOTE_LOCKED(dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(vp, NOTE_DELETE); } } void vop_setattr_pre(void *ap) { struct vop_setattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_ATTRIB); } void vop_setacl_pre(void *ap) { struct vop_setacl_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setacl_post(void *ap, int rc __unused) { struct vop_setacl_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); } void vop_setextattr_pre(void *ap) { struct vop_setextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_ATTRIB); } void vop_symlink_pre(void *ap) { struct vop_symlink_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ !VN_IS_DOOMED(a->a_vp))) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp); } static void vfs_knl_assert_locked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); #endif } static void vfs_knl_assert_unlocked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } /* * Returns whether the directory is empty or not. * If it is empty, the return value is 0; otherwise * the return value is an error value (which may * be ENOTEMPTY). */ int vfs_emptydir(struct vnode *vp) { struct uio uio; struct iovec iov; struct dirent *dirent, *dp, *endp; int error, eof; error = 0; eof = 0; ASSERT_VOP_LOCKED(vp, "vfs_emptydir"); dirent = malloc(sizeof(struct dirent), M_TEMP, M_WAITOK); iov.iov_base = dirent; iov.iov_len = sizeof(struct dirent); uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = sizeof(struct dirent); uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = curthread; while (eof == 0 && error == 0) { error = VOP_READDIR(vp, &uio, curthread->td_ucred, &eof, NULL, NULL); if (error != 0) break; endp = (void *)((uint8_t *)dirent + sizeof(struct dirent) - uio.uio_resid); for (dp = dirent; dp < endp; dp = (void *)((uint8_t *)dp + GENERIC_DIRSIZ(dp))) { if (dp->d_type == DT_WHT) continue; if (dp->d_namlen == 0) continue; if (dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN) { error = ENOTEMPTY; break; } if (dp->d_namlen > 2) { error = ENOTEMPTY; break; } if (dp->d_namlen == 1 && dp->d_name[0] != '.') { error = ENOTEMPTY; break; } if (dp->d_namlen == 2 && dp->d_name[1] != '.') { error = ENOTEMPTY; break; } uio.uio_resid = sizeof(struct dirent); } } free(dirent, M_TEMP); return (error); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * Clear out a doomed vnode (if any) and replace it with a new one as long * as the fs is not being unmounted. Return the root vnode to the caller. */ static int __noinline vfs_cache_root_fallback(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; restart: if (mp->mnt_rootvnode != NULL) { MNT_ILOCK(mp); vp = mp->mnt_rootvnode; if (vp != NULL) { if (!VN_IS_DOOMED(vp)) { vrefact(vp); MNT_IUNLOCK(mp); error = vn_lock(vp, flags); if (error == 0) { *vpp = vp; return (0); } vrele(vp); goto restart; } /* * Clear the old one. */ mp->mnt_rootvnode = NULL; } MNT_IUNLOCK(mp); if (vp != NULL) { vfs_op_barrier_wait(mp); vrele(vp); } } error = VFS_CACHEDROOT(mp, flags, vpp); if (error != 0) return (error); if (mp->mnt_vfs_ops == 0) { MNT_ILOCK(mp); if (mp->mnt_vfs_ops != 0) { MNT_IUNLOCK(mp); return (0); } if (mp->mnt_rootvnode == NULL) { vrefact(*vpp); mp->mnt_rootvnode = *vpp; } else { if (mp->mnt_rootvnode != *vpp) { if (!VN_IS_DOOMED(mp->mnt_rootvnode)) { panic("%s: mismatch between vnode returned " " by VFS_CACHEDROOT and the one cached " " (%p != %p)", __func__, *vpp, mp->mnt_rootvnode); } } } MNT_IUNLOCK(mp); } return (0); } int vfs_cache_root(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; if (!vfs_op_thread_enter(mp)) return (vfs_cache_root_fallback(mp, flags, vpp)); vp = atomic_load_ptr(&mp->mnt_rootvnode); if (vp == NULL || VN_IS_DOOMED(vp)) { vfs_op_thread_exit(mp); return (vfs_cache_root_fallback(mp, flags, vpp)); } vrefact(vp); vfs_op_thread_exit(mp); error = vn_lock(vp, flags); if (error != 0) { vrele(vp); return (vfs_cache_root_fallback(mp, flags, vpp)); } *vpp = vp; return (0); } struct vnode * vfs_cache_root_clear(struct mount *mp) { struct vnode *vp; /* * ops > 0 guarantees there is nobody who can see this vnode */ MPASS(mp->mnt_vfs_ops > 0); vp = mp->mnt_rootvnode; if (vp != NULL) vn_seqc_write_begin(vp); mp->mnt_rootvnode = NULL; return (vp); } void vfs_cache_root_set(struct mount *mp, struct vnode *vp) { MPASS(mp->mnt_vfs_ops > 0); vrefact(vp); mp->mnt_rootvnode = vp; } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL; vp = TAILQ_NEXT(vp, v_nmntvnodes)) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * lazy vnodes. See MNT_VNODE_FOREACH_LAZY() in sys/mount.h */ static void mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * Relock the mp mount vnode list lock with the vp vnode interlock in the * conventional lock order during mnt_vnode_next_lazy iteration. * * On entry, the mount vnode list lock is held and the vnode interlock is not. * The list lock is dropped and reacquired. On success, both locks are held. * On failure, the mount vnode list lock is held but the vnode interlock is * not, and the procedure may have yielded. */ static bool mnt_vnode_next_lazy_relock(struct vnode *mvp, struct mount *mp, struct vnode *vp) { VNASSERT(mvp->v_mount == mp && mvp->v_type == VMARKER && TAILQ_NEXT(mvp, v_lazylist) != NULL, mvp, ("%s: bad marker", __func__)); VNASSERT(vp->v_mount == mp && vp->v_type != VMARKER, vp, ("%s: inappropriate vnode", __func__)); ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, mvp, v_lazylist); TAILQ_INSERT_BEFORE(vp, mvp, v_lazylist); /* * Note we may be racing against vdrop which transitioned the hold * count to 0 and now waits for the ->mnt_listmtx lock. This is fine, * if we are the only user after we get the interlock we will just * vdrop. */ vhold(vp); mtx_unlock(&mp->mnt_listmtx); VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); goto out_lost; } VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); /* * There is nothing to do if we are the last user. */ if (!refcount_release_if_not_last(&vp->v_holdcnt)) goto out_lost; mtx_lock(&mp->mnt_listmtx); return (true); out_lost: vdropl(vp); maybe_yield(); mtx_lock(&mp->mnt_listmtx); return (false); } static struct vnode * mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_lazylist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } /* * See if we want to process the vnode. Note we may encounter a * long string of vnodes we don't care about and hog the list * as a result. Check for it and requeue the marker. */ VNPASS(!VN_IS_DOOMED(vp), vp); if (!cb(vp, cbarg)) { if (!should_yield()) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); goto restart; } /* * Try-lock because this is the wrong lock order. */ if (!VI_TRYLOCK(vp) && !mnt_vnode_next_lazy_relock(*mvp, mp, vp)) goto restart; KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the lazy list %p %p", vp, mp)); VNPASS(vp->v_mount == mp, vp); VNPASS(!VN_IS_DOOMED(vp), vp); break; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); /* Check if we are done */ if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); ASSERT_VI_LOCKED(vp, "lazy iter"); return (vp); } struct vnode * __mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } struct vnode * __mnt_vnode_first_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; if (TAILQ_EMPTY(&mp->mnt_lazyvnodelist)) return (NULL); *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); mtx_lock(&mp->mnt_listmtx); vp = TAILQ_FIRST(&mp->mnt_lazyvnodelist); if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_lazylist); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } void __mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); } int vn_dir_check_exec(struct vnode *vp, struct componentname *cnp) { if ((cnp->cn_flags & NOEXECCHECK) != 0) { cnp->cn_flags &= ~NOEXECCHECK; return (0); } return (VOP_ACCESS(vp, VEXEC, cnp->cn_cred, cnp->cn_thread)); } /* * Do not use this variant unless you have means other than the hold count * to prevent the vnode from getting freed. */ void vn_seqc_write_begin_unheld_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_seqc_users >= 0, vp); vp->v_seqc_users++; if (vp->v_seqc_users == 1) seqc_sleepable_write_begin(&vp->v_seqc); } void vn_seqc_write_begin_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_holdcnt > 0, vp); vn_seqc_write_begin_unheld_locked(vp); } void vn_seqc_write_begin(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_begin_locked(vp); VI_UNLOCK(vp); } void vn_seqc_write_begin_unheld(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_begin_unheld_locked(vp); VI_UNLOCK(vp); } void vn_seqc_write_end_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_seqc_users > 0, vp); vp->v_seqc_users--; if (vp->v_seqc_users == 0) seqc_sleepable_write_end(&vp->v_seqc); } void vn_seqc_write_end(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_end_locked(vp); VI_UNLOCK(vp); } Index: head/sys/sys/priv.h =================================================================== --- head/sys/sys/priv.h (revision 364768) +++ head/sys/sys/priv.h (revision 364769) @@ -1,542 +1,544 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2006 nCircle Network Security, Inc. * All rights reserved. * * This software was developed by Robert N. M. Watson for the TrustedBSD * Project under contract to nCircle Network Security, Inc. * * 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. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR, NCIRCLE NETWORK SECURITY, * INC., 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. * * $FreeBSD$ */ /* * Privilege checking interface for BSD kernel. */ #ifndef _SYS_PRIV_H_ #define _SYS_PRIV_H_ /* * Privilege list, sorted loosely by kernel subsystem. * * Think carefully before adding or reusing one of these privileges -- are * there existing instances referring to the same privilege? Third party * vendors may request the assignment of privileges to be used in loadable * modules. Particular numeric privilege assignments are part of the * loadable kernel module ABI, and should not be changed across minor * releases. * * When adding a new privilege, remember to determine if it's appropriate * for use in jail, and update the privilege switch in prison_priv_check() * in kern_jail.c as necessary. */ /* * Track beginning of privilege list. */ #define _PRIV_LOWEST 1 /* * The remaining privileges typically correspond to one or a small * number of specific privilege checks, and have (relatively) precise * meanings. They are loosely sorted into a set of base system * privileges, such as the ability to reboot, and then loosely by * subsystem, indicated by a subsystem name. */ #define _PRIV_ROOT 1 /* Removed. */ #define PRIV_ACCT 2 /* Manage process accounting. */ #define PRIV_MAXFILES 3 /* Exceed system open files limit. */ #define PRIV_MAXPROC 4 /* Exceed system processes limit. */ #define PRIV_KTRACE 5 /* Set/clear KTRFAC_ROOT on ktrace. */ #define PRIV_SETDUMPER 6 /* Configure dump device. */ #define PRIV_REBOOT 8 /* Can reboot system. */ #define PRIV_SWAPON 9 /* Can swapon(). */ #define PRIV_SWAPOFF 10 /* Can swapoff(). */ #define PRIV_MSGBUF 11 /* Can read kernel message buffer. */ #define PRIV_IO 12 /* Can perform low-level I/O. */ #define PRIV_KEYBOARD 13 /* Reprogram keyboard. */ #define PRIV_DRIVER 14 /* Low-level driver privilege. */ #define PRIV_ADJTIME 15 /* Set time adjustment. */ #define PRIV_NTP_ADJTIME 16 /* Set NTP time adjustment. */ #define PRIV_CLOCK_SETTIME 17 /* Can call clock_settime. */ #define PRIV_SETTIMEOFDAY 18 /* Can call settimeofday. */ #define _PRIV_SETHOSTID 19 /* Removed. */ #define _PRIV_SETDOMAINNAME 20 /* Removed. */ /* * Audit subsystem privileges. */ #define PRIV_AUDIT_CONTROL 40 /* Can configure audit. */ #define PRIV_AUDIT_FAILSTOP 41 /* Can run during audit fail stop. */ #define PRIV_AUDIT_GETAUDIT 42 /* Can get proc audit properties. */ #define PRIV_AUDIT_SETAUDIT 43 /* Can set proc audit properties. */ #define PRIV_AUDIT_SUBMIT 44 /* Can submit an audit record. */ /* * Credential management privileges. */ #define PRIV_CRED_SETUID 50 /* setuid. */ #define PRIV_CRED_SETEUID 51 /* seteuid to !ruid and !svuid. */ #define PRIV_CRED_SETGID 52 /* setgid. */ #define PRIV_CRED_SETEGID 53 /* setgid to !rgid and !svgid. */ #define PRIV_CRED_SETGROUPS 54 /* Set process additional groups. */ #define PRIV_CRED_SETREUID 55 /* setreuid. */ #define PRIV_CRED_SETREGID 56 /* setregid. */ #define PRIV_CRED_SETRESUID 57 /* setresuid. */ #define PRIV_CRED_SETRESGID 58 /* setresgid. */ #define PRIV_SEEOTHERGIDS 59 /* Exempt bsd.seeothergids. */ #define PRIV_SEEOTHERUIDS 60 /* Exempt bsd.seeotheruids. */ /* * Debugging privileges. */ #define PRIV_DEBUG_DIFFCRED 80 /* Exempt debugging other users. */ #define PRIV_DEBUG_SUGID 81 /* Exempt debugging setuid proc. */ #define PRIV_DEBUG_UNPRIV 82 /* Exempt unprivileged debug limit. */ #define PRIV_DEBUG_DENIED 83 /* Exempt P2_NOTRACE. */ /* * Dtrace privileges. */ #define PRIV_DTRACE_KERNEL 90 /* Allow use of DTrace on the kernel. */ #define PRIV_DTRACE_PROC 91 /* Allow attaching DTrace to process. */ #define PRIV_DTRACE_USER 92 /* Process may submit DTrace events. */ /* * Firmware privilegs. */ #define PRIV_FIRMWARE_LOAD 100 /* Can load firmware. */ /* * Jail privileges. */ #define PRIV_JAIL_ATTACH 110 /* Attach to a jail. */ #define PRIV_JAIL_SET 111 /* Set jail parameters. */ #define PRIV_JAIL_REMOVE 112 /* Remove a jail. */ /* * Kernel environment privileges. */ #define PRIV_KENV_SET 120 /* Set kernel env. variables. */ #define PRIV_KENV_UNSET 121 /* Unset kernel env. variables. */ /* * Loadable kernel module privileges. */ #define PRIV_KLD_LOAD 130 /* Load a kernel module. */ #define PRIV_KLD_UNLOAD 131 /* Unload a kernel module. */ /* * Privileges associated with the MAC Framework and specific MAC policy * modules. */ #define PRIV_MAC_PARTITION 140 /* Privilege in mac_partition policy. */ #define PRIV_MAC_PRIVS 141 /* Privilege in the mac_privs policy. */ /* * Process-related privileges. */ #define PRIV_PROC_LIMIT 160 /* Exceed user process limit. */ #define PRIV_PROC_SETLOGIN 161 /* Can call setlogin. */ #define PRIV_PROC_SETRLIMIT 162 /* Can raise resources limits. */ #define PRIV_PROC_SETLOGINCLASS 163 /* Can call setloginclass(2). */ /* * System V IPC privileges. */ #define PRIV_IPC_READ 170 /* Can override IPC read perm. */ #define PRIV_IPC_WRITE 171 /* Can override IPC write perm. */ #define PRIV_IPC_ADMIN 172 /* Can override IPC owner-only perm. */ #define PRIV_IPC_MSGSIZE 173 /* Exempt IPC message queue limit. */ /* * POSIX message queue privileges. */ #define PRIV_MQ_ADMIN 180 /* Can override msgq owner-only perm. */ /* * Performance monitoring counter privileges. */ #define PRIV_PMC_MANAGE 190 /* Can administer PMC. */ #define PRIV_PMC_SYSTEM 191 /* Can allocate a system-wide PMC. */ /* * Scheduling privileges. */ #define PRIV_SCHED_DIFFCRED 200 /* Exempt scheduling other users. */ #define PRIV_SCHED_SETPRIORITY 201 /* Can set lower nice value for proc. */ #define PRIV_SCHED_RTPRIO 202 /* Can set real time scheduling. */ #define PRIV_SCHED_SETPOLICY 203 /* Can set scheduler policy. */ #define PRIV_SCHED_SET 204 /* Can set thread scheduler. */ #define PRIV_SCHED_SETPARAM 205 /* Can set thread scheduler params. */ #define PRIV_SCHED_CPUSET 206 /* Can manipulate cpusets. */ #define PRIV_SCHED_CPUSET_INTR 207 /* Can adjust IRQ to CPU binding. */ /* * POSIX semaphore privileges. */ #define PRIV_SEM_WRITE 220 /* Can override sem write perm. */ /* * Signal privileges. */ #define PRIV_SIGNAL_DIFFCRED 230 /* Exempt signalling other users. */ #define PRIV_SIGNAL_SUGID 231 /* Non-conserv signal setuid proc. */ /* * Sysctl privileges. */ #define PRIV_SYSCTL_DEBUG 240 /* Can invoke sysctl.debug. */ #define PRIV_SYSCTL_WRITE 241 /* Can write sysctls. */ #define PRIV_SYSCTL_WRITEJAIL 242 /* Can write sysctls, jail permitted. */ /* * TTY privileges. */ #define PRIV_TTY_CONSOLE 250 /* Set console to tty. */ #define PRIV_TTY_DRAINWAIT 251 /* Set tty drain wait time. */ #define PRIV_TTY_DTRWAIT 252 /* Set DTR wait on tty. */ #define PRIV_TTY_EXCLUSIVE 253 /* Override tty exclusive flag. */ #define _PRIV_TTY_PRISON 254 /* Removed. */ #define PRIV_TTY_STI 255 /* Simulate input on another tty. */ #define PRIV_TTY_SETA 256 /* Set tty termios structure. */ /* * UFS-specific privileges. */ #define PRIV_UFS_EXTATTRCTL 270 /* Can configure EAs on UFS1. */ #define PRIV_UFS_QUOTAOFF 271 /* quotaoff(). */ #define PRIV_UFS_QUOTAON 272 /* quotaon(). */ #define PRIV_UFS_SETUSE 273 /* setuse(). */ /* * ZFS-specific privileges. */ #define PRIV_ZFS_POOL_CONFIG 280 /* Can configure ZFS pools. */ #define PRIV_ZFS_INJECT 281 /* Can inject faults in the ZFS fault injection framework. */ #define PRIV_ZFS_JAIL 282 /* Can attach/detach ZFS file systems to/from jails. */ /* * NFS-specific privileges. */ #define PRIV_NFS_DAEMON 290 /* Can become the NFS daemon. */ #define PRIV_NFS_LOCKD 291 /* Can become NFS lock daemon. */ /* * VFS privileges. */ #define PRIV_VFS_READ 310 /* Override vnode DAC read perm. */ #define PRIV_VFS_WRITE 311 /* Override vnode DAC write perm. */ #define PRIV_VFS_ADMIN 312 /* Override vnode DAC admin perm. */ #define PRIV_VFS_EXEC 313 /* Override vnode DAC exec perm. */ #define PRIV_VFS_LOOKUP 314 /* Override vnode DAC lookup perm. */ #define PRIV_VFS_BLOCKRESERVE 315 /* Can use free block reserve. */ #define PRIV_VFS_CHFLAGS_DEV 316 /* Can chflags() a device node. */ #define PRIV_VFS_CHOWN 317 /* Can set user; group to non-member. */ #define PRIV_VFS_CHROOT 318 /* chroot(). */ #define PRIV_VFS_RETAINSUGID 319 /* Can retain sugid bits on change. */ #define PRIV_VFS_EXCEEDQUOTA 320 /* Exempt from quota restrictions. */ #define PRIV_VFS_EXTATTR_SYSTEM 321 /* Operate on system EA namespace. */ #define PRIV_VFS_FCHROOT 322 /* fchroot(). */ #define PRIV_VFS_FHOPEN 323 /* Can fhopen(). */ #define PRIV_VFS_FHSTAT 324 /* Can fhstat(). */ #define PRIV_VFS_FHSTATFS 325 /* Can fhstatfs(). */ #define PRIV_VFS_GENERATION 326 /* stat() returns generation number. */ #define PRIV_VFS_GETFH 327 /* Can retrieve file handles. */ #define PRIV_VFS_GETQUOTA 328 /* getquota(). */ #define PRIV_VFS_LINK 329 /* bsd.hardlink_check_uid */ #define PRIV_VFS_MKNOD_BAD 330 /* Was: mknod() can mark bad inodes. */ #define PRIV_VFS_MKNOD_DEV 331 /* Can mknod() to create dev nodes. */ #define PRIV_VFS_MKNOD_WHT 332 /* Can mknod() to create whiteout. */ #define PRIV_VFS_MOUNT 333 /* Can mount(). */ #define PRIV_VFS_MOUNT_OWNER 334 /* Can manage other users' file systems. */ #define PRIV_VFS_MOUNT_EXPORTED 335 /* Can set MNT_EXPORTED on mount. */ #define PRIV_VFS_MOUNT_PERM 336 /* Override dev node perms at mount. */ #define PRIV_VFS_MOUNT_SUIDDIR 337 /* Can set MNT_SUIDDIR on mount. */ #define PRIV_VFS_MOUNT_NONUSER 338 /* Can perform a non-user mount. */ #define PRIV_VFS_SETGID 339 /* Can setgid if not in group. */ #define PRIV_VFS_SETQUOTA 340 /* setquota(). */ #define PRIV_VFS_STICKYFILE 341 /* Can set sticky bit on file. */ #define PRIV_VFS_SYSFLAGS 342 /* Can modify system flags. */ #define PRIV_VFS_UNMOUNT 343 /* Can unmount(). */ #define PRIV_VFS_STAT 344 /* Override vnode MAC stat perm. */ #define PRIV_VFS_READ_DIR 345 /* Can read(2) a dirfd, needs sysctl. */ /* * Virtual memory privileges. */ #define PRIV_VM_MADV_PROTECT 360 /* Can set MADV_PROTECT. */ #define PRIV_VM_MLOCK 361 /* Can mlock(), mlockall(). */ #define PRIV_VM_MUNLOCK 362 /* Can munlock(), munlockall(). */ #define PRIV_VM_SWAP_NOQUOTA 363 /* * Can override the global * swap reservation limits. */ #define PRIV_VM_SWAP_NORLIMIT 364 /* * Can override the per-uid * swap reservation limits. */ /* * Device file system privileges. */ #define PRIV_DEVFS_RULE 370 /* Can manage devfs rules. */ #define PRIV_DEVFS_SYMLINK 371 /* Can create symlinks in devfs. */ /* * Random number generator privileges. */ #define PRIV_RANDOM_RESEED 380 /* Closing /dev/random reseeds. */ /* * Network stack privileges. */ #define PRIV_NET_BRIDGE 390 /* Administer bridge. */ #define PRIV_NET_GRE 391 /* Administer GRE. */ #define _PRIV_NET_PPP 392 /* Removed. */ #define _PRIV_NET_SLIP 393 /* Removed. */ #define PRIV_NET_BPF 394 /* Monitor BPF. */ #define PRIV_NET_RAW 395 /* Open raw socket. */ #define PRIV_NET_ROUTE 396 /* Administer routing. */ #define PRIV_NET_TAP 397 /* Can open tap device. */ #define PRIV_NET_SETIFMTU 398 /* Set interface MTU. */ #define PRIV_NET_SETIFFLAGS 399 /* Set interface flags. */ #define PRIV_NET_SETIFCAP 400 /* Set interface capabilities. */ #define PRIV_NET_SETIFNAME 401 /* Set interface name. */ #define PRIV_NET_SETIFMETRIC 402 /* Set interface metrics. */ #define PRIV_NET_SETIFPHYS 403 /* Set interface physical layer prop. */ #define PRIV_NET_SETIFMAC 404 /* Set interface MAC label. */ #define PRIV_NET_ADDMULTI 405 /* Add multicast addr. to ifnet. */ #define PRIV_NET_DELMULTI 406 /* Delete multicast addr. from ifnet. */ #define PRIV_NET_HWIOCTL 407 /* Issue hardware ioctl on ifnet. */ #define PRIV_NET_SETLLADDR 408 /* Set interface link-level address. */ #define PRIV_NET_ADDIFGROUP 409 /* Add new interface group. */ #define PRIV_NET_DELIFGROUP 410 /* Delete interface group. */ #define PRIV_NET_IFCREATE 411 /* Create cloned interface. */ #define PRIV_NET_IFDESTROY 412 /* Destroy cloned interface. */ #define PRIV_NET_ADDIFADDR 413 /* Add protocol addr to interface. */ #define PRIV_NET_DELIFADDR 414 /* Delete protocol addr on interface. */ #define PRIV_NET_LAGG 415 /* Administer lagg interface. */ #define PRIV_NET_GIF 416 /* Administer gif interface. */ #define PRIV_NET_SETIFVNET 417 /* Move interface to vnet. */ #define PRIV_NET_SETIFDESCR 418 /* Set interface description. */ #define PRIV_NET_SETIFFIB 419 /* Set interface fib. */ #define PRIV_NET_VXLAN 420 /* Administer vxlan. */ #define PRIV_NET_SETLANPCP 421 /* Set LAN priority. */ #define PRIV_NET_SETVLANPCP PRIV_NET_SETLANPCP /* Alias Set VLAN priority */ /* * 802.11-related privileges. */ #define PRIV_NET80211_VAP_GETKEY 440 /* Query VAP 802.11 keys. */ #define PRIV_NET80211_VAP_MANAGE 441 /* Administer 802.11 VAP */ #define PRIV_NET80211_VAP_SETMAC 442 /* Set VAP MAC address */ #define PRIV_NET80211_CREATE_VAP 443 /* Create a new VAP */ /* * Placeholder for AppleTalk privileges, not supported anymore. */ #define _PRIV_NETATALK_RESERVEDPORT 450 /* Bind low port number. */ /* * ATM privileges. */ #define PRIV_NETATM_CFG 460 #define PRIV_NETATM_ADD 461 #define PRIV_NETATM_DEL 462 #define PRIV_NETATM_SET 463 /* * Bluetooth privileges. */ #define PRIV_NETBLUETOOTH_RAW 470 /* Open raw bluetooth socket. */ /* * Netgraph and netgraph module privileges. */ #define PRIV_NETGRAPH_CONTROL 480 /* Open netgraph control socket. */ #define PRIV_NETGRAPH_TTY 481 /* Configure tty for netgraph. */ /* * IPv4 and IPv6 privileges. */ #define PRIV_NETINET_RESERVEDPORT 490 /* Bind low port number. */ #define PRIV_NETINET_IPFW 491 /* Administer IPFW firewall. */ #define PRIV_NETINET_DIVERT 492 /* Open IP divert socket. */ #define PRIV_NETINET_PF 493 /* Administer pf firewall. */ #define PRIV_NETINET_DUMMYNET 494 /* Administer DUMMYNET. */ #define PRIV_NETINET_CARP 495 /* Administer CARP. */ #define PRIV_NETINET_MROUTE 496 /* Administer multicast routing. */ #define PRIV_NETINET_RAW 497 /* Open netinet raw socket. */ #define PRIV_NETINET_GETCRED 498 /* Query netinet pcb credentials. */ #define PRIV_NETINET_ADDRCTRL6 499 /* Administer IPv6 address scopes. */ #define PRIV_NETINET_ND6 500 /* Administer IPv6 neighbor disc. */ #define PRIV_NETINET_SCOPE6 501 /* Administer IPv6 address scopes. */ #define PRIV_NETINET_ALIFETIME6 502 /* Administer IPv6 address lifetimes. */ #define PRIV_NETINET_IPSEC 503 /* Administer IPSEC. */ #define PRIV_NETINET_REUSEPORT 504 /* Allow [rapid] port/address reuse. */ #define PRIV_NETINET_SETHDROPTS 505 /* Set certain IPv4/6 header options. */ #define PRIV_NETINET_BINDANY 506 /* Allow bind to any address. */ #define PRIV_NETINET_HASHKEY 507 /* Get and set hash keys for IPv4/6. */ /* * Placeholders for IPX/SPX privileges, not supported any more. */ #define _PRIV_NETIPX_RESERVEDPORT 520 /* Bind low port number. */ #define _PRIV_NETIPX_RAW 521 /* Open netipx raw socket. */ /* * NCP privileges. */ #define PRIV_NETNCP 530 /* Use another user's connection. */ /* * SMB privileges. */ #define PRIV_NETSMB 540 /* Use another user's connection. */ /* * VM86 privileges. */ #define PRIV_VM86_INTCALL 550 /* Allow invoking vm86 int handlers. */ /* * Set of reserved privilege values, which will be allocated to code as * needed, in order to avoid renumbering later privileges due to insertion. */ #define _PRIV_RESERVED0 560 #define _PRIV_RESERVED1 561 #define _PRIV_RESERVED2 562 #define _PRIV_RESERVED3 563 #define _PRIV_RESERVED4 564 #define _PRIV_RESERVED5 565 #define _PRIV_RESERVED6 566 #define _PRIV_RESERVED7 567 #define _PRIV_RESERVED8 568 #define _PRIV_RESERVED9 569 #define _PRIV_RESERVED10 570 #define _PRIV_RESERVED11 571 #define _PRIV_RESERVED12 572 #define _PRIV_RESERVED13 573 #define _PRIV_RESERVED14 574 #define _PRIV_RESERVED15 575 /* * Define a set of valid privilege numbers that can be used by loadable * modules that don't yet have privilege reservations. Ideally, these should * not be used, since their meaning is opaque to any policies that are aware * of specific privileges, such as jail, and as such may be arbitrarily * denied. */ #define PRIV_MODULE0 600 #define PRIV_MODULE1 601 #define PRIV_MODULE2 602 #define PRIV_MODULE3 603 #define PRIV_MODULE4 604 #define PRIV_MODULE5 605 #define PRIV_MODULE6 606 #define PRIV_MODULE7 607 #define PRIV_MODULE8 608 #define PRIV_MODULE9 609 #define PRIV_MODULE10 610 #define PRIV_MODULE11 611 #define PRIV_MODULE12 612 #define PRIV_MODULE13 613 #define PRIV_MODULE14 614 #define PRIV_MODULE15 615 /* * DDB(4) privileges. */ #define PRIV_DDB_CAPTURE 620 /* Allow reading of DDB capture log. */ /* * Arla/nnpfs privileges. */ #define PRIV_NNPFS_DEBUG 630 /* Perforn ARLA_VIOC_NNPFSDEBUG. */ /* * cpuctl(4) privileges. */ #define PRIV_CPUCTL_WRMSR 640 /* Write model-specific register. */ #define PRIV_CPUCTL_UPDATE 641 /* Update cpu microcode. */ /* * Capi4BSD privileges. */ #define PRIV_C4B_RESET_CTLR 650 /* Load firmware, reset controller. */ #define PRIV_C4B_TRACE 651 /* Unrestricted CAPI message tracing. */ /* * OpenAFS privileges. */ #define PRIV_AFS_ADMIN 660 /* Can change AFS client settings. */ #define PRIV_AFS_DAEMON 661 /* Can become the AFS daemon. */ /* * Resource Limits privileges. */ #define PRIV_RCTL_GET_RACCT 670 #define PRIV_RCTL_GET_RULES 671 #define PRIV_RCTL_GET_LIMITS 672 #define PRIV_RCTL_ADD_RULE 673 #define PRIV_RCTL_REMOVE_RULE 674 /* * mem(4) privileges. */ #define PRIV_KMEM_READ 680 /* Open mem/kmem for reading. */ #define PRIV_KMEM_WRITE 681 /* Open mem/kmem for writing. */ /* * Track end of privilege list. */ #define _PRIV_HIGHEST 682 /* * Validate that a named privilege is known by the privilege system. Invalid * privileges presented to the privilege system by a priv_check interface * will result in a panic. This is only approximate due to sparse allocation * of the privilege space. */ #define PRIV_VALID(x) ((x) > _PRIV_LOWEST && (x) < _PRIV_HIGHEST) #ifdef _KERNEL /* * Privilege check interfaces, modeled after historic suser() interfaces, but * with the addition of a specific privilege name. No flags are currently * defined for the API. Historically, flags specified using the real uid * instead of the effective uid, and whether or not the check should be * allowed in jail. */ struct thread; struct ucred; int priv_check(struct thread *td, int priv); int priv_check_cred(struct ucred *cred, int priv); +int priv_check_cred_vfs_lookup(struct ucred *cred); +int priv_check_cred_vfs_lookup_nomac(struct ucred *cred); int priv_check_cred_vfs_generation(struct ucred *cred); #endif #endif /* !_SYS_PRIV_H_ */